more information - www.cambridge.

org/9780521881685
ESSENTIALS OF MICRO- AND NANOFLUIDICS

This book introduces students to the basic physical principles needed to analyze
fluid flow in micro- and nanosize devices. This is the first book to unify the thermal
sciences with electrostatics, electrokinetics, and colloid science; electrochemistry;
and molecular biology. The author discusses key concepts and principles, such as
the essentials of viscous flows, electrochemistry, heat and mass transfer phenomena,
elements of molecular and cell biology, and much more. This textbook presents state-
of-the-art analytical and computational approaches to problems in all these areas,
especially in electrokinetic flows, and gives examples of the use of these disciplines to
design devices used for rapid molecular analysis, biochemical sensing, drug delivery,
DNA analysis, the design of an artificial kidney, and other transport phenomena. This
textbook includes exercise problems, modern examples of the applications of these
sciences, and a solutions manual available to qualified instructors.

A. Terrence Conlisk is Professor of mechanical and aerospace engineering at The

Ohio State University. He is an internationally recognized expert in the areas of micro-
and nanofluidics, helicopter aerodynamics, and complex flows driven by vortices. He
is the author of numerous publications and hundreds of technical presentations and
seminars delivered throughout the world. After his PhD thesis (Purdue, 1978) on the
prediction of the fluid dynamics and separation of isotopes in a gas centrifuge, he
began his work on various aspects of the dynamics of two- and three-dimensional
vortices, with a focus on helicopter aerodynamics. Since 1999, he has been involved
in modeling ionic and biomolecular transport through micro- and nanochannels for
the design of devices used for rapid molecular analysis, sensing, drug delivery, and
other applications. Professor Conlisk’s wide spectrum of research interests makes
him uniquely qualified to write on the thoroughly interdisciplinary fields of micro-
and nanofluidics.
ESSENTIALS OF MICRO-
AND NANOFLUIDICS
With Applications to the
Biological and Chemical
Sciences

A. Terrence Conlisk
The Ohio State University
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, São Paulo, Delhi, Mexico City
Cambridge University Press
32 Avenue of the Americas, New York, NY 10013-2473, USA
www.cambridge.org
Information on this title: www.cambridge.org/9780521881685


C A. Terrence Conlisk 2013

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.

First published 2013

Printed in the United States of America

A catalog record for this publication is available from the British Library.

Library of Congress Cataloging in Publication Data

Conlisk, A. Terrence, 1950–
Essentials of micro- and nanofluidics : with applications to the biological and chemical sciences /
A. Terrence Conlisk.
p. ; cm.
Includes bibliographical references and index.
ISBN 978-0-521-88168-5 (hardback)
I. Title.
[DNLM: 1. Thermodynamics. 2. Biomedical Technology. 3. Hydrodynamics.
4. Micro-Electrical-Mechanical Systems. 5. Nanostructures – therapeutic use. 6. Static Electricity.
QU 34]
572 .436–dc23 2011033658

ISBN 978-0-521-88168-5 Hardback

Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or
third-party Internet Web sites referred to in this publication and does not guarantee that any content on
such Web sites is, or will remain, accurate or appropriate.
To my mother and father, Ginny and Terry, who first taught me the
value of education.

To my brother and sisters, Virginia, Bill, Mary, and Elizabeth for

friendship, love and support, and the many good times that so few
families experience.

And to my wife, Paulette, and children, Terry and Katie, for their
love and understanding; and for putting up with me over these many
years. Without you this book would not have been possible.
 Contents

Preface page xv

1 Introduction and Overview 1

1.1 Micro- and nanofluidics 1
1.2 Some micro- and nanofluidic devices 3
1.3 What is it about the nanoscale? 7
1.4 Nanotechnology 11
1.5 What is a fluid? 13
1.6 Historical perspectives 14
1.6.1 Fluid mechanics 15
1.6.2 Heat and mass transfer 17
1.6.3 Electrokinetic phenomena 19
1.7 The thermal sciences 20
1.8 Electrostatics 23
1.9 Electrolyte solutions 25
1.10 The electrical double layer 26
1.11 Colloidal systems 29
1.12 Molecular biology 32
1.13 The convergence of molecular biology and engineering 34
1.14 Design of micro- and nanofluidic devices 35
1.15 Unit systems 37
1.16 A word about notation 37
1.17 Chapter summary 38

2 Preparatory Concepts 40
2.1 Introduction 40
2.2 Important constitutive laws 41
2.3 Determining transport properties 45
2.3.1 Viscosity 45
2.3.2 Diffusion coefficient 48
2.3.3 Thermal conductivity 52

vii
viii Contents

2.3.4 Electrical permittivity 54

2.3.5 Surface tension and wettability 55
2.4 Classification of fluid flows 59
2.5 Elements of thermodynamics 62
2.6 The nature of frictional losses in channels and pipes 68
2.7 Chapter summary 70

3 The Governing Equations for an Electrically Conducting Fluid 74

3.1 Introduction 74
3.2 The continuum approximation and its limitations 75
3.3 Kinematics 77
3.4 Surface and body forces 83
3.5 The continuity equation 87
3.6 The Navier–Stokes equations 88
3.7 Mass transport 93
3.7.1 Definitions 93
3.7.2 Governing equation 97
3.8 Electrostatics 100
3.9 Energy transport 102
3.10 Two-dimensional, steady, and incompressible flow 106
3.11 Boundary and initial conditions 106
3.11.1 Velocity boundary conditions 107
3.11.2 Mass transfer boundary conditions 113
3.11.3 Electrostatics boundary conditions 114
3.11.4 Temperature boundary conditions 116
3.11.5 Other boundary conditions 117
3.12 Dimensional analysis and similarity 117
3.13 Fluid, electrostatics, and heat and mass transfer analogies 123
3.13.1 Mole fraction and temperature similarity 123
3.13.2 Velocity and electrical potential similarity 125
3.14 Other stress–strain relationships 126
3.15 Mathematical character of partial differential equations 128
3.15.1 Introduction 128
3.15.2 Mathematical classification of second-order partial
differential equations 128
3.15.3 Characteristic curves 129
3.15.4 Boundary and initial conditions 130
3.15.5 Classification of the governing equations of micro-
and nanofluidics 131
3.16 Well-posed problems 131
3.17 The role of fabrication, experiments, and theory in micro-
and nanofluidics 132
3.18 Chapter summary 134
ix Contents

4 The Essentials of Viscous Flow 140

4.1 Introduction 140
4.2 The structure of flow in a pipe or channel 141
4.3 Poiseuille flow in a pipe or channel 143
4.4 The velocity in slip flow 146
4.4.1 Gases 146
4.4.2 Liquids 147
4.5 Flow in a thin film under gravity 148
4.6 The boundary layer on a flat plate 150
4.7 Fully developed suction flows 155
4.8 Developing suction flows 158
4.9 The lubrication approximation 162
4.10 A surface tension–driven flow 166
4.11 Stokes flow past a sphere 169
4.12 Sedimentation of a solid particle 172
4.13 A simple model for blood flow 173
4.14 Chapter summary 174
5 Heat and Mass Transfer Phenomena in Channels and Tubes 180
5.1 Introduction 180
5.2 One-dimensional temperature distributions in channel flow 181
5.3 Thermal and mass transfer entrance regions 184
5.4 The temperature distribution in fully developed tube flow 189
5.5 The Graetz problem for a channel 189
5.6 Mass transfer in thin films 192
5.7 Classical Taylor–Aris dispersion 194
5.8 The stochastic nature of diffusion: Brownian motion 199
5.9 Unsteady mass transport in uncharged membranes 201
5.10 Temperature and concentration boundary layers 205
5.11 Chapter summary 207
6 Introduction to Electrostatics 213
6.1 Introduction 213
6.2 Coulomb’s law: The electric field 214
6.3 The electric field due to an isolated large flat plate 216
6.4 Gauss’s law 218
6.5 The electric potential 219
6.6 The electric dipole and polar molecules 221
6.7 Poisson’s equation 222
6.8 Current and current density 225
6.9 Maxwell’s equations 226
6.10 Chapter summary 227
x Contents

7 Elements of Electrochemistry and the Electrical Double Layer 230

7.1 Introduction 230
7.2 The structure of water and ionic species 231
7.3 Chemical bonds in biology and chemistry 233
7.4 Hydration of ions 234
7.5 Chemical potential 236
7.6 The Gibbs function and chemical equilibrium 240
7.7 Electrochemical potential 243
7.8 Acids, bases, and electrolytes 244
7.9 Site-binding models of the silica surface 246
7.10 Polymer surfaces 249
7.11 Qualitative description of the electrical double layer 251
7.12 Electrolyte and potential distribution in the electrical double layer 253
7.13 Multivalent asymmetric mixtures 259
7.14 The ζ potential and surface charge density: Putting it all together 260
7.14.1 The classical liquid-side view for a symmetric electrolyte 260
7.14.2 The solid-side view and connection to the liquid side 262
7.15 The electrical double layer on a cylinder 265
7.16 The electrical double layer on a sphere 266
7.17 Electrical conductivity in an electrolyte solution 267
7.18 Semi-permeable membranes 270
7.19 The Derjaguin approximation 275
7.20 Chapter summary 278
8 Elements of Molecular and Cell Biology 283
8.1 Introduction 283
8.2 Nucleic acids and polysaccharides 285
8.3 Proteins 287
8.3.1 Protein function 288
8.3.2 Protein structure 289
8.3.3 Some common proteins 292
8.3.4 Few polypeptide chains are useful 295
8.4 Protein binding 295
8.5 Cells 298
8.6 The cell membrane 300
8.7 Membrane transport and ion channels 301
8.8 Chapter summary 304
9 Electrokinetic Phenomena 306
9.1 Introduction 306
9.2 Electro-osmosis 307
9.2.1 The relationship between velocity and potential 307
9.2.2 The Debye–Hückel approximation reviewed 312
9.2.3 Another similarity revealed 312
xi Contents

9.2.4 Asymptotic solution for binary electrolytes of arbitrary

valence 313
9.2.5 Walls with different ζ potentials 316
9.2.6 Species velocities in electro-osmotic flow: Electromigration 318
9.2.7 Current and current density in electro-osmotic flow 320
9.2.8 Electro-osmotic flow in an annulus 322
9.2.9 Electro-osmotic flow in nozzles and diffusers 324
9.2.10 Dispersion in electro-osmotic flow 328
9.3 Electrophoresis: Single particles 331
9.3.1 Introduction 331
9.3.2 Electrophoretic mobility 332
9.3.3 Henry’s solution 334
9.3.4 The full nonlinear problem 336
9.4 Streaming potential 338
9.5 Sedimentation potential 341
9.6 Joule heating 342
9.7 Chapter summary 344
10 Essential Numerical Methods 348
10.1 Introduction 348
10.2 Types of errors 350
10.3 Taylor series 351
10.4 Zeros of functions 353
10.4.1 Numerical methods 353
10.4.2 Polynomials 358
10.5 Interpolation 359
10.5.1 Linear interpolation 360
10.5.2 The difference table 361
10.5.3 Lagrangian polynomial interpolation 362
10.5.4 Newton interpolation formulas 363
10.5.5 Matlab interpolation functions 365
10.5.6 Cubic spline interpolation 366
10.6 Curve fitting 370
10.7 Numerical differentiation 373
10.7.1 Derivatives from Taylor series 373
10.7.2 A more accurate forward formula for the first derivative 375
10.8 Numerical integration 376
10.8.1 The trapezoidal rule 377
10.8.2 Simpson’s rules 380
10.8.3 Matlab integration functions 382
10.8.4 The indefinite integral 382
10.8.5 Other formulas 383
10.8.6 Grid (mesh) size 383
10.8.7 Singularities 384
xii Contents

10.9 Solution of linear systems 386

10.9.1 Solving sets of linear equations in Matlab 389
10.9.2 Iterative solution to linear systems 390
10.9.3 Tridiagonal systems 393
10.9.4 Ill-conditioning and stability 396
10.10 Solution of boundary value problems 398
10.10.1 Introduction 398
10.10.2 Linear equations 399
10.10.3 Nonlinear equations 403
10.10.4 Systems of ordinary differential equations 405
10.10.5 Derivative boundary conditions 407
10.10.6 Convergence tests and Richardson extrapolation 409
10.10.7 Solving boundary value problems with Matlab functions 410
10.11 Solution of initial value problems 411
10.11.1 Introduction 411
10.11.2 Taylor series method 413
10.11.3 Euler methods 414
10.11.4 Runge-Kutta methods 416
10.11.5 Adams–Moulton methods 419
10.11.6 Symplectic integrators 419
10.11.7 Stiff equations and stability 424
10.11.8 Solving initial value problems using Matlab functions 428
10.12 Numerical solution of the PNP system 428
10.13 Partial differential equations 430
10.13.1 Elliptic equations 431
10.13.2 Parabolic equations 432
10.13.3 The Matlab PDE solver 435
10.14 Verification and validation of numerical solutions 435
10.15 Chapter summary 438
11 Molecular Simulations 447
11.1 Introduction 447
11.2 The molecular world 449
11.3 Ensembles 451
11.4 The potentials 451
11.5 Using the Lennard–Jones potential 453
11.6 Molecular models for water 456
11.7 Periodic boundary conditions 457
11.8 The Ewald sum 460
11.9 Numerical issues 463
11.9.1 Time integration 463
11.9.2 Truncation of interactions 464
11.9.3 Boundary conditions 465
xiii Contents

11.10 Postprocessing 465

11.11 Nonequilibrium molecular dynamics 467
11.11.1 Introduction 467
11.11.2 Poiseuille flow 468
11.11.3 Electro-osmotic flow 469
11.12 Molecular dynamics packages 471
11.12.1 Introduction 471
11.12.2 What MD/NEMD simulators do 471
11.13 Summary 472
12 Applications 475
12.1 Introduction 475
12.2 DNA transport 476
12.2.1 How does DNA move? 477
12.2.2 Mathematical model 479
12.2.3 Results 481
12.2.4 DNA current 482
12.2.5 Comparison with experiment 483
12.3 Development of an artificial kidney 484
12.3.1 Background 484
12.3.2 The nanopore membrane for filtration 486
12.3.3 Hindered transport 487
12.4 Biochemical sensing 491
12.4.1 Introduction 491
12.4.2 What is a biosensor? 492
12.4.3 Receptor-based classification of biosensors 493
12.4.4 Transducer-based classification of biosensors 494
12.4.5 Evaluation of biosensor performance 495
12.4.6 Nanopores and nanopore membranes for biochemical
sensing 496
12.5 Chapter summary 498
Appendix A Matched Asymptotic Expansions 501
A.1 Introduction 501
A.2 Terminology 501
A.3 Asymptotic sequences and expansions 502
A.4 Regular perturbations 503
A.5 Singular perturbations 504
Appendix B Vector Operations in Curvilinear Coordinates 508
B.1 Cylindrical coordinates 508
B.2 Spherical coordinates 508
B.3 Rectangular coordinates 509
xiv Contents

Appendix C Web Sites 510

C.1 Fluid dynamics and micro- and nanofluidics 510
C.2 General nanotechnology 511
C.3 Wikipedia 511

Appendix D A Semester Course Syllabus 512

Bibliography 515
Index 533
 Preface

The book is meant to be used as a text for an interdisciplinary course in micro-

and nanofluidics that includes the study of ionic and biomolecular transport at
the advanced undergraduate and beginning graduate levels. The rationale for this
book is that most, if not all, problems in the twenty-first century are interdisci-
plinary in nature, yet no textbooks address the topics required for investigating
problems that cut across disciplines in engineering, the physical sciences, and
mathematics. The closest approach to this concept is in the several texts that
address the thermal sciences at a strictly undergraduate level (Moran et al.,
2003). Another set of texts addresses problems in applied mathematics appli-
cable to engineering problems generally at the advanced graduate level (Bird
et al., 2002). Still another set of texts under the general area of biophysics links
the mathematics and biological sciences, again most often at the advanced grad-
uate level (Murray, 2001, 2003). In contrast, this book aims at the advanced
undergraduate and beginning graduate student pool.
A number of other related books are on the market, but all are monographs
directed at the senior graduate student (Karniadakis et al., 2005; Masliyah and
Bhattacharjee, 2006; Tabeling, 2005; Liou & Fang, 2006; Nguyen & Wereley,
2002; Bruus, 2008; Kirby, 2010; Chang and Yeo, 2010). All these texts empha-
size the unique features of transport at the micro- and nanoscale, of which there
are many. In contrast, while the reader will be exposed to many of these unique
features, it is my contention that transport at the micro- and nanoscale actually
unifies all the thermal sciences, fluid dynamics, heat and mass transfer, and ther-
modynamics; it also, sometimes by necessity, unifies the thermal sciences with
electrostatics, electrokinetics, and colloid science; electrochemistry; and molec-
ular biology. This book is the first to show how all these fields are interrelated
at the micro- and nanoscale and show how it is essential for a researcher, stu-
dent, or faculty to acquire an understanding at some level of all these fields.
The fundamental concepts within these fields are supported by addressing the
continuum and molecular computation techniques that may be employed to solve
these problems.
The objective of this book is to introduce students in the physical and mathe-
matical sciences and engineering to the basic physical principles appropriate to
analyzing fluid flow in micro- and nanoscale devices. The book will emphasize
xv
xvi Preface

the fundamental principles involved in the formulation and solution of problems

in fluid mechanics and mass transfer for pressure-driven and electrically driven
motion of biofluids and electrolyte solutions at the micro- and nanoscale. It will
introduce the student to a variety of subject matter spanning the physical sci-
ences, thermal engineering, and applied numerical methods to enable the student
to solve problems of an interdisciplinary nature. On completion of the book, the
student should be able to extract from a raw physical situation the essential prin-
ciples from which a useful model for thermal, ionic, and biomolecular transport
may be developed.
The primary target audience of this text is the advanced undergraduate and
beginning graduate engineering student; however, it is hoped that the book will be
accessible to some advanced undergraduate students in physics and the chemical
and biological sciences with the appropriate mathematics background.
In writing this book, I have been greatly influenced by the style and format of
White (2006), which is aimed at a similar audience and contains exercises at the
end of each chapter. White uses canonical problems in the field to illustrate basic
fluid dynamic phenomena, and I have followed this style, expanding into heat
and mass transfer, electrostatics, electrochemistry, electrokinetics, and molecular
biology.
As with any book project, many people have contributed. I am thankful for
Dr. David Mott, who read, with a keen eye, an advanced draft of the book, and to
Professor Shaurya Prakash, who read a nearly final version of the manuscript. I
am very thankful to Professor Susan Olesik, who read an early draft of the elec-
trochemistry chapter, and to Dr. Arfaan Rampersaud, who reviewed the chapter
on molecular biology. I am also thankful to Professor Minami Yoda, with whom
I have worked over the past seven or so years. We have cut our teeth on micro-
and nanofluidics together over that time. I am grateful to my colleagues Professor
Shuvo Roy, Dr. Bill Fissell, and Professor Andrew Zydney, who introduced me
to the fluid dynamics of the kidney. Professor Sherwin Singer read the molecular
simulation chapter and made many suggestions and corrections that have been
incorporated. Dr. Harvey Zambrano also helped me greatly with that chapter.
I am also grateful to Professors Narayan Aluru and Ron Larson and Dr. Dirk
Gillespie for their contributions.
And thanks to those researchers who have contributed the boxed vignettes that
are about their work or some aspect of micro- and nanofluidics for which I did
not have room. They are acknowledged at the end of the presentation.
I am particularly grateful for all the discussions I have had with faculty at the
nanoscale science and engineering center, called the Center for the Affordable
Nanoengineering of Polymeric Biomedical Devices (CANPBD).
Thanks go to my present and former students, Prashanth Ramesh, Ankan
Kumar, Pradeep Gnanaprakasam, and Devi Pulla, some of whose work appears
in the book; to Mike Stubblebine, who helped catalog the figures; and to Pro-
fessor Subhra Datta and Dr. Lei Chen, who both have done much in the way
of research that appears in this book. Subhra wrote first drafts of several sec-
tions, and Lei produced many of the figures that appear in the book. Both have
xvii Preface

read the manuscript, portions more than once. Thanks also to Dan Hoying,
an undergraduate physics student who read a nearly final version of the book,
and Zhizi Peng and Cong Zhang, who produced several figures; Cong was a
significant contributor to the solutions manual; and to Harvey Zambrano who
helped me write a section in the molecular dynamics chapter; and to Kevin
Disotell who read the final proofs; and to Jim Marcicki who taught me about
batteries.
I am grateful to my “Introduction to Micro- and Nanofluidics” class in the
Autumn quarter of 2010, who used the book and had many suggestions that were
heartily received and implemented. Thanks also go to graduate student Martin
Kearney-Fisher, who read the entire manuscript and gave me pages of corrections
and suggestions, almost all of which I have incorporated. Martin’s suggestions
have made the manuscript much better.
Thanks also to my editor, Peter Gordon, who kept me on task and made a
number of suggestions on how to write the book, especially the early chapters.
He also provided me with additional resources that allowed a more thorough
treatment of this rapidly expanding field. And thanks go to Peggy Rote for her
diligence in managing the production process.
If I have forgotten to thank someone, I apologize.
The book begins with an introduction and overview of micro- and nanofluidics
in Chapter 1, followed by Chapter 2, “Preparatory Concepts.” Chapter 2 is meant
to unify concepts on two levels: discussing the fundamental roles of transport
coefficients in fluid mechanics and heat and mass transfer and the relationship
between thermodynamics, the equilibrium science, and heat transfer and fluid
mechanics, the nonequilibrium thermal sciences. These two initial chapters are
followed by a discussion of the governing equations and boundary conditions
associated with micro- and nanofluidics. At the micro and nano levels, several
new phenomena come into play:1

1. Because of the large surface-to-volume ratio, the characteristics of surfaces

play a major role in fluid and mass transport.
2. Classical means of transporting fluids, such as pressure drop, may not be
possible.
3. Noncontinuum effects arise when the length scale associated with the fluid
transport becomes less than 10 nm.

All these issues are discussed throughout the book, and the second point is the
reason that electrokinetic transport methods become important.
The next three chapters cover the fundamentals of viscous flow, heat and mass
transport, and electrostatics. While writing this book, I was astonished at how
similar these fields are in the way of expressing basic transport phenomena. Many
analogies between these three (or four, if mass transfer is considered separately)
disciplines are discussed throughout.
1
We are speaking here primarily of liquid flows. Gas flows are treated extensively by Liou and
Fang (2006) and Karniadakis et al. (2005).
xviii Preface

Following these three chapters are two chapters covering the fundamentals of
electrochemistry and molecular and cell biology. These two chapters are meant to
reintroduce engineering students to material they may have had in their first-year
course work, although parts of each chapter are written at a higher level.
Following these two chapters, electrokinetic phenomena are discussed in Chap-
ter 9. The two most important of these phenomena, electro-osmosis and elec-
trophoresis, are discussed in great detail, and canonical problems of electro-
osmosis are described in the spirit of the style of White (2006).
The next chapter covers the basics of numerical methods, from zero finding to
the numerical solution of partial differential equations. The primary role of this
chapter is to introduce the student to basic numerical methods that can be used to
solve problems in micro- and nanofluidics. For some engineering students, this
chapter is likely to be a review; however, physics and chemistry students may
find this chapter valuable.
Next, the fundamental concepts involved in performing molecular simula-
tions, specifically equilibrium molecular dynamics and nonequilibrium molecu-
lar dynamics, are presented along with examples of Poiseuille flow and electro-
osmotic flow. It is surprising how different the philosophy and expectations of
what is achievable in a simulation on the molecular level are from the continuum
perspective. The reader need only compare the presentation of the numerical
methods in this chapter with those presented in the previous chapter to see this.
Note the differences in how the continuum and molecular results are verified and
validated.
The book ends with a chapter devoted exclusively to applications. Applications
are too numerous to mention, but I have chosen those applications with which
I am familiar. Thus a simple model for DNA transport is presented, along with
a section on biochemical sensing and the fluid mechanics and mass transfer
involved in the design of a renal assist device.
Each chapter is followed by a set of exercises that range from simple calcu-
lations, such as determining the Debye length for a given set of parameters, to
finding the solution of viscous flow through an annulus to the calculation of the
numerical solution of the Poisson equation to completely open-ended exercises
that require a written report. The exercises after Chapter 12 are all open ended. In
these open-ended exercises, other applications not included in the book are intro-
duced such as the use of nanoparticles to treat cancer. These exercises have been
designed to make maximum use of the Web and emphasize the development of
the technical writing skill of the student. A short introduction to technical writing
is available from the author, on request.
Several appendices, giving a short introduction to the method of matched
asymptotic expansions, the governing equations in cylindrical and spherical coor-
dinates, a list of interesting and useful Web sites, and a prospective syllabus, are
also included. Writing this book has been a tremendous learning experience, and
I have bought several chemistry and biology dictionaries. I have also acquired
more biology and chemistry textbooks in six years than I have in my entire life
xix Preface

(I have Cambridge University Press to thank for some of these). While all this
about learning from books is true, I cannot tell you how many times I have been
to Wikipedia or used the other Web sites that appear in Appendix B.
Much of the material in the book is gleaned from research papers, from those
that are very old and classical to those published very recently. I have tried to
be judicious in my choice of references, and to those whom I have overlooked,
I apologize. I would be happy to be informed of the omission of a major paper
that would contribute to a future manuscript. This has been quite a task, and I
and my students, and several faculty, have read parts of the book, as noted earlier.
Nevertheless, errors are inevitable, and I would be grateful if I could be informed
of any errors that do appear in the book. For these errors, I take full responsibility.
The emphasis in this book has been the interdisciplinary nature of micro- and
nanofluidics. This book is me speaking about what I think is important to know
in micro- and nanofluidics. Thus I take responsibility for those many topics that
are left out. For this, I do not apologize but merely say that tough choices were
made. I hope that this book will be read by students with diverse backgrounds
and that they will benefit from what I hope is a lucid presentation.
 1 Introduction and Overview

1.1 Micro- and nanofluidics

Analyzing and computing fluid flow at small scales is becoming increasingly

important because of the emergence of new technologies such as the ability to
construct microelectromechanical systems (MEMS). These systems may be used
for drug delivery and its control; DNA and protein manipulation and transport;
and the desire to manufacture laboratories on a microchip for rapid molecular
analysis, requiring the modeling of flows on a length scale approaching molecular
dimensions. On these small scales, new flow features appear that are not seen in
macroscale flows.
Because of the large surface-to-volume ratio in nanochannels, surface proper-
ties become enormously important. Because the pressure drop p ∼ 1/h 3 , it is
prohibitively large for a nanoscale channel. Thus fluid, biomaterials such as pro-
teins, and other colloidal particles are most often transported electrokinetically,
and the art of designing micro- and nanodevices requires a significant amount of
knowledge of fluid flow and mass transfer (biofluids are multicomponent mix-
tures) and often heat transfer, electrokinetics, electrochemistry, and molecular
biology. To efficiently manufacture laboratories on a microchip, the analysis and
computation of flows on a length scale approaching molecular dimensions, the
nanoscale, are required.
The common thread is micro- and nanofluidics. Thus micro- and nanofluidics
play the role of unifying the fields of fluid mechanics, heat and mass transfer,
electrostatics and electrodynamics, electrochemistry, and molecular biology. In
particular, nanofluidics opens the door to uncovering the structure and confor-
mation of biomaterials, such as proteins, through molecular simulation.
The objective of this book is to introduce the reader to micro- and nanoflu-
idics, the basic mechanics of modeling fluid flows and heat and mass transfer,
that is, at very small scales. The emphasis is on those systems that have biolog-
ical and chemical applications. These systems are commonly at the microscale,
with length scales ∼100 µm or 100 × 10−6 meters, and the 100 times smaller
nanoscale, at length scales ∼100 nm or 100 × 10−9 meters. In many of these sys-
tems, transport is from the microscale to the nanoscale and back to the microscale.

1
2 Essentials of Micro- and Nanofluidics

Microfluidics in general consists of three distinct components:

1. Modeling: computational and theoretical

2. Fabrication
3. Experimental methods

The emphasis will be on modeling because of the limitations of experimental

methods at the micro- and nanoscale, although experimental methods are also
discussed, where appropriate; Bohn (2009) discusses some of the experimental
methods used to probe single molecules. Because this is primarily a book about
modeling, fabrication methods are not discussed.
Government research programs, such as the Defense Advanced Projects
Research Agency (DARPA) and Simulation of Biological Systems (Simbiosys),
(which ran from 2002 to 2005), deal with the development of microsystems that
can be used to identify many types of molecules through their transport charac-
teristics. The ultimate goal of the program was to develop computer-aided design
tools (CAD) for applications to chemical–biological warfare defense, infectious
disease monitoring, and drug delivery.
The governing equations of fluid flow on a length scale orders of magnitude
greater than a molecular diameter are well known to be the Navier–Stokes equa-
tions, which are a statement of Newton’s law for a fluid. Along with conservation
of mass and appropriate boundary and initial conditions in the case of unsteady
flow, these equations form a well-posed problem from which, for an incompress-
ible flow (constant density), the velocity field and pressure may be obtained.
Applications in the biomedical field involve, for the most part, internal flows in
micro- and nanochannels and tubes. The fluids are generally electrolyte mixtures,
with, perhaps, a biomolecular component, usually some protein (say, albumin).
Thus mass transfer occurs, and because many biomolecules (e.g., most proteins)
are charged, there is an electric field as well. The determination of the identity
and rates of transport of ionic and biomolecular species is one of the purposes of
many micro–nanoscale devices.
Because micro- and nanofluidics usually involves the transport of charged
species, its study requires a multidisciplinary approach. Thus the study of fluid
flows at the microscale and nanoscale most often requires expertise in elec-
trochemistry, surface chemistry, electrostatics and electrokinetics, molecular
biology, heat and mass transfer, and macro-scale fluid mechanics. To design
devices having micro- and nanoscale features requires a team approach involving
chemists, biologists, medical practitioners, engineers, and systems analysts. As
one might guess, each of these disciplines speaks a somewhat different language,
and it is only with some effort that these technical language barriers can be
overcome.
There are a number of textbooks on various aspects of electrokinetic phenom-
ena, including the fluid mechanics of electrokinetics; for more details, see Chang
and Yeo (2010), Masliyah and Bhattacharjee (2006), Karniadakis et al. (2005),
 3 Introduction and Overview

Bruus (2008), Kirby (2010), Tabeling (2005) and Li (2004). All the aforemen-
tioned books are essentially monographs written primarily for advanced graduate
students beginning their research careers.
It is with this interdisciplinary view of micro- and nanofluidics that our journey
begins. We start in this chapter by presenting some examples of micro- and
nanofluidic devices, followed by a working definition and a bit of the history of
nanotechnology. Then a broad discussion of the fields of fluid mechanics and heat
and mass transfer is presented. This is followed by a discussion of electrostatics
and the character of electrolytes (charged fluid mixtures). Micro- and nanofluidics
often deals with mixtures of liquids and solid particles. If the particles are less
than a micron (10−6 m) but larger than about 1 nm, the mixture is termed a
colloidal mixture. Finally, we introduce some of the basic concepts of molecular
biology. One of the major objectives of this book is to show how micro- and
nanofluidics unifies the fields of the thermal sciences, electrochemical systems,
and molecular biology. The unifying features of these fields are described next,
followed by a discussion of a typical design procedure. The chapter concludes
with two short sections on unit systems and notation, the latter being a very
important section. In short, all notation is local.

1.2 Some micro- and nanofluidic devices

In this section, we qualitatively describe several micro- and nanofluidic devices

and their applications. An interesting application of microtechnology is small
drug delivery devices. These devices can deliver very small and precise doses of
medicines quickly and efficiently. Some devices may also be used as biomolecular
separators because different species (often electrically charged) and biomolecules
travel at different speeds in these channels, due primarily to their differences in
size, charge, and shape. Devices of this sort perform analyses faster and are
more efficient in a wide variety of applications, including water quality, medical
diagnostics, and applications associated with national security such as sensing
chemical and biological toxins.
As you reflect on the art of designing micro- and nanodevices, think of the
fact that this activity requires a significant understanding of fluid flow and mass
transfer and, often, heat transfer and electrokinetics. Mass transfer is especially
relevant to biofluids because they are multicomponent mixtures.
A specific example of this sort is an electro-osmotic pump, or sometimes
nanopump, which is used to induce transport of charged molecules. Such a
device is depicted in Figure 1.1(a). This biomedical device is useful for the
delivery of various types of proteins, such as albumin and immunoglobulin, both
of which have many biomedical uses (Peters, 1996). Nine channels are shown
in the depiction, but an actual device may employ 20,000–40,000 channels. In
general, in biomolecular transport and analysis systems, it is essential to have lots
of little channels to distinguish the very small molecules that are being analyzed.
4 Essentials of Micro- and Nanofluidics

The device depicted in Figure 1.1(a) is often called a synthetic nanopore

membrane. In biology and chemistry, the term membrane is used to describe a
thin sheet of porous material that can be either natural (the outer skin of a cell is
a membrane) or synthetic.

What is Lab-on-a-chip?
The ability to fabricate devices on the micro and nanoscale has led to the devel-
opment of devices that can identify different molecules, separate these molecules,
manipulate them, and transport these molecules. What had once required a labo-
ratory and large samples can now be done at very small scales.
The generic name given to these types of devices is “Lab-on-a-chip (LOC),”
which refers to a type of processing. The key feature of LOC is that the various
steps in a diagnostic procedure “the laboratory” are integrated on to one small
devices “the chip.” These devices have also been called micro total analysis
systems (μ TAS) and the two terms are most often taken to be equivalent. Most
LOC applications have a biochemical component to them.
Such systems work because in a large or small sample the following general
principles regarding the molecules under consideration here apply: Molecules can
be identified by properties, such as how they appear under the action of a laser.
Molecules of different size and electric charge characteristics move at different
speeds allowing them to be separated based on a simple criteria. Molecules may
be modified by inducing chemical reactions with other molecules. And finally
molecules can be moved from one place to another by the bulk motion of the fluid.
Lab-on-a-chip devices are being developed for bacteria screening, cancer detec-
tion, and unicellular exploration.
Such LOC systems have several advantages that make them attractive in
biomedical applications:
1. Lower equipment costs and power requirements because of the small length
scales,
2. reduced separation and reaction times again because of the small length
scales,
3. LOC typically require nano- to picoliter (vs. mL for comparable macroscale
analyses) volumes of analyte and reagents, reducing chemical costs and
biochemical hazard and waste disposal problems;
4. integrating sophisticated chemistry procedures within a single system, LOC
can be used by nonspecialists to perform complex analyses.

The devices from within the chemistry and biochemistry industries described
earlier were created using MEMS technology. The broad range of micro- and
nanodevices of this sort are commonly referred to as a lab on a chip: a device that
incorporates all the steps of chemical–biochemical analysis to perform a given
measurement. Sometimes these systems are called micro total analysis systems,
or µTAS.
 5 Introduction and Overview

Inlet and Outlet

ports

Electrode
Nanopore
Membrane

Donor Receiver
(a) (b)

(a) Drawing of a nanopump containing a nanopore membrane. (b) Scanning electron microscope (SEM) image
Figure 1.1
of a synthetic nanopore membrane. From Conlisk et al. (2009).

As an example of a lab on a chip, Sandia National Laboratory has developed

a fully integrated chemical analysis system that has the ability to determine con-
stituents of gas and liquid samples within 1 min (Figure 1.2) through integration
of sample collection, separation, and detection steps. It is made up of a com-
pact power source, lasers and photodiodes, a microprocessor, and micro-sized
injection and separation channels. In contrast, much of the current chemistry–
biochemistry analysis labs consist of equipment as large as a microwave oven for
each of the analysis steps.
The use of lab-on-a-chip devices in a chemical–biochemical analysis lab
reduces equipment costs but can also reduce the cost of other resources. The
advantages of a lab-on-a-chip device are comparable to the advantages of minia-
turization seen in the computer industry. For example, when reducing the size of a
computer chip, the distance electrons need to travel is much shorter, reducing the
processing time. This same principle applies to microfluidic devices: reducing
the size of a channel reduces the distance molecules need to travel, therefore

(a) (b)

Sandia National Laboratory fully integrated chemical analysis system that can be held in one hand. (a) Final
Figure 1.2
product. (b) Cutaway image of the device, showing many of the components.
 6 Essentials of Micro- and Nanofluidics

objective
argon ion laser
488-nm dichroic mirror

pinhole
bandpass filter
shutter
Detector

(a) (b)

(a) Chip used for attomole level chemical detection courtesy of Professor Paul Bohn. (b) The detection system
Figure 1.3
from Kuo et al. (2003).

reducing processing time. These devices are highly portable and also require
much less reagent, or sample, the chemical compound used to detect and identify
an analyte. For a standard analysis experiment, microliters or larger of reagents
are used for each experiment; however, with lab-on-a-chip devices, only nanoliter
or picoliter volumes may be required for each experiment. Think of the savings
in chemical use and the environmental benefits of reduced chemical and biotoxic
waste.
A number of security applications are associated with these devices. Nanocap-
illary array membranes of approximately circular cross section (Kemery et al.,
1998) are being used to detect biological warfare agents in concentrations at
the attomole (10−18 mole) level. Such a system is depicted in Figure 1.3. Chan-
nels on the order of 10–100 nm are employed to manipulate these biochemicals,
which must be handled in very low concentrations. A molecule is identified by a
laser-induced fluorescence signal, and the device can identify molecular size and
charge based on its transfer characteristics through the channel. A similar device
can be employed with an enzyme sample (Gong et al., 2008).
The nanocapillary array membranes (NCAMs) used in the 3-D configuration
shown in Figure 1.3 function in the same manner in which individual transistors
function in integrated circuits: by controlling the temporal and spatial delivery
of ultralow-volume fluid packets. Achieving an all-electronic fluidic switching
network with no moving parts is an enabling development for 3-D integrated
microfluidic devices. In nearly all cases of multidimensional chemical analysis,
the chemical sample needs to be processed through multiple sequential chemical
unit operations. These might include separation of a desired component from
a raw mixture, subsequent chemical processing (derivatization) to visualize the
compound, and placing it in the right spatial location for detection and/or further
characterization. In simple 2-D (planar) structures, the management of chip real
estate soon becomes a design challenge. Going into the third dimension makes
 7 Introduction and Overview

it possible to save real estate and achieve highly compact and efficient designs.
Furthermore, there is the possibility of using the individual nanopores to carry
out enhanced chemical reactions by taking advantage of the small distances to
improve the efficiency of chemical turnover. In this sense, the NCAMs are more
than just simple fluidic switching elements; they can be thought of as attoliter-
scale chemical reactors with on-demand delivery (or generation) of reagents and
removal or collection of products.

Ion Channels
The basic units of all living organisms are cells. In order to keep cells functioning
properly there must be a continuous flux of ions in and out of the cell and its
components. The cell and many of its components are surrounded by a plasma
membrane which provides selective transfer of ions through ion channels. The ion
channels are embedded in the cell membrane, are usually negatively charged and
are about 10 angstroms in diameter. The polarity of the plasma membrane makes
it challenging for molecules to move in and out of cells and its components.
Thus ions (Ca2+ , Cl− , K+ , Na+ , H+ , Mg2+ , HCO3− , PO4 2− ) must selectively
move through the membrane via protein channels electrokinetically through a
combination of electro-osmosis and electromigration. Both of these fluid dynamic
phenomena are discussed in Chapter 9 and ion channels are described in more
detail in Chapter 8.
Very recently, on April 26, 2009, new ion channels that govern the function of
the inner ear were found in a very surprising place! (http://medicalnewstoday.com/
articles/147506.php)

In contrast to the synthetic devices described earlier, natural nanochannels exist

in cells for the purpose of providing nutrients and discarding waste; in that sense,
they function as natural pumping systems. That is, these natural nanochannels,
called ion channels, act as electro-osmotic pumps that contain perhaps thousands
of nanopores – a natural nanopore membrane. Peter Agre of Johns Hopkins and
Roderick MacKinnon of Rockefeller University won the 2003 Nobel Prize in
Chemistry for their work on understanding how natural ion channels work.

1.3 What is it about the nanoscale?

Why is everything different at the nanoscale? Or is anything different at all?

As the typical length scale of the channel approaches the microscale level and
beyond to the nanoscale level (Table 1.1), conventional means of moving fluids,
such as with a pressure gradient, become ever more difficult, and the character of
the surfaces bounding a fluid becomes ever more important. Consider the channel
depicted in Figure 1.4(a). In micro- and nanofluidics, the surface-to-volume ratio
is very large, making the nature of the surface (e.g., its charge, roughness, and
 8 Essentials of Micro- and Nanofluidics

Table 1.1. SI units of length measurement

Factor Prefix Symbol
9
10 giga G
6
10 mega M
3
10 kilo k
10 deka da
10−1 deci d
10−2 centi c
10−3 milli m
−6
10 micro μ
−9
10 nano n
−10
10 angstrom A
−12
10 pico p
−15
10 femto f
−18
10 atto a

whether it is hydrophobic or hydrophilic) very important. The surface-to-volume

ratio for a channel having dimensions (L , h, W ) = (1 m, 1 m, 1 m) is
 
S 1 1 1
=2 + + = 6 m−1 (1.1)
V W h L

On the other hand, for a channel having dimensions (L , h, W ) = (3 µm, 1 µm,

40 µm) which is typical of a class of nanopore membranes, the surface-to-volume

Side View
L

Flow
h
y,v
x,u L>>h
z,w
End View
W
r,u

z,w
z,w D
W>>h
(a) Rectangular channel (b) Cylindrical tube

(a) Geometry of a typical channel. In applications, h  W , L , where W is the width of the channel and L
Figure 1.4
is its length in the primary flow direction. Variables u, v, w are the fluid velocities in the x , y, z directions.
(b) Sketch of a cylindrical pore having velocities (u, v, w) in the (r , θ , z) directions.
9 Introduction and Overview

ratio is
S
∼ 2 × 106 m−1 (1.2)
V
For a 20 nm channel, (L , h, W ) = (3 µm, 20 nm, 40 µm), the surface-to-volume
ratio is even higher:
S 2
∼ ∼ 40 × 109 m−1 (1.3)
V h
Because of the large surface-to-volume ratio, a surface roughness, for example,
of 5 nm in a 1 µm channel, is negligible, whereas in a 10 nm channel, that same
roughness can have a profound effect on the flow. The same situation occurs for
a cylindrical tube (Figure 1.4(b)). In this case,
S 1 2
= = (1.4)
V R D
where R is the radius of the tube.
Several comments can be made about fluid flows in channels under 1 µm in
minimum dimension:

1. Surface properties of a channel or tube, such as electrical surface charge

density and roughness, become very important because of the large surface-
to-volume ratio.
2. Significant increases in flow rate may be attained if the surfaces of the
channel are hydrophobic (water hating); that is, significant fluid slip may
occur at the wall. This occurence is termed induced slip or apparent slip.
3. The continuum approximation may break down, especially for gas flows.
4. Pressure-driven flow is only viable at very low flow rates, on the order of
nL/min or 10−9 L/min, in nanoconstrained channels because of the very
large pressure drops required otherwise, on the order of atmospheres.
5. Molecular diffusion, which is very slow at the macroscale, is fast at the
micro- and nanoscale, the time scale being t ∼ L 2 /D AB .

From the second comment, it is thus seen that for liquids, whether there is slip or
no slip at the wall can be a function of surface chemistry, whereas in gases, slip is
entirely controlled by the magnitude of the Knudsen number, the ratio of the mean
free path to the characteristic length scale. However, it should be mentioned that
liquid flows remain in continuum even for channels whose smallest dimensions
approach 10 nm.
As mentioned, it becomes increasingly difficult to pump liquids by pressure
in nanoscale channels. To see this, let us compare the pressure drop for electro-
osmotic flow with the corresponding pressure drop for pressure-driven flow or
Poiseuille flow. The volume flow rate in electro-osmotic flow may be estimated by
Q e = CU0 hW (1.5)
where U0 is the electro-osmotic velocity scale and is independent of h (as we will
see), C is a constant that depends on the concentration of the electrolyte, and W is
 10 Essentials of Micro- and Nanofluidics

3
Q = 1 nL/min
Q = 10 nL/min
Q = 0.1μ L/min
2.5
Q = 1 μ L/min

0.03
2 Q = 1 nL/min
0.025 Q = 10 nL/min
Pressure Drop (Atm)

Pressure Drop (Atm)

0.02
1.5
0.015

1 0.01

0.005

0.5
0
10 20 30 40 50 60 70 80
Channel Height (nm)
0
10 20 30 40 50 60 70 80
Channel Height (nm)

Pressure drop as a function of channel height to achieve a volume flow rate of Q = 10−6 L/min and several
Figure 1.5
values of the flow rate on the μL scale typical of many existing systems. Here L = liter; 1000 L = 1 m3 . The
applied electrical potential for Q = 10−6 L/min is very small.

the width of the channel. The velocity scale U0 turns out to be directly proportional
to the imposed electric field, and thus the flow rate is proportional to h. Conversely,
for Poiseuille flow, the volume flow rate in a parallel plate channel is given by

W h3
Qp = p (1.6)
12μL

where p is the pressure drop. This means that pressure-driven flow requires
large pressure drops, as depicted in Figure 1.5; note that at a channel height of
10 nm, 3 atm of pressure drop is required to drive a flow of Q = 10−6 L/min,
which is a characteristic flow rate in drug delivery applications. However, if the
flow rate is Q = 10−9 L/min, the pressure drop is not nearly as large. Three
atmospheres is a large pressure drop in a liquid, and clearly a relatively large
pump would be required to provide this pressure drop. This is a major consid-
eration when designing a nanopore membrane for the applications discussed
previously.
As has been seen, it is often not feasible to transport fluids in nanochannels
using an imposed pressure drop; electro-osmosis and electrophoresis are often
used for transporting both charged and uncharged species and biomolecules.
That is, electrokinetic phenomena play a crucial role in micro- and nanofluidics.
Moreover, it is important to note that at present, velocity, temperature, and con-
centration profiles across a channel cannot be measured in channels having at
least one dimension under about 1 µm (Sadr et al., 2006; Breuer, 2005). Thus,
to understand the physics of flows at those scales, modeling is not only neces-
sary but also essential in describing the important features of the flow within a
microdevice having nanoscale features such as a nanopore membrane.
 11 Introduction and Overview

1.4 Nanotechnology

I have heard as many definitions of nanotechnology as there are researchers. I

will take a stab at an engineering definition: nanotechnology, or simply nanotech,
is the study of phenomena on the atomic and molecular scale; that is, phenomena
that occur on length scales less than about 100 nm. The tech part implies an
application. Nanotechnology has been used to create solid materials with novel
properties, for the development of medical devices, for rapid molecular analysis,
and, as will be seen, in many other disciplines. For example, though the focus of
this book is on biomedical applications in the biological and chemical sciences,
other applications of micro- and nanotechnology include defense applications
of chemical and biochemical sensing, desalination, water purification, and the
design of batteries and fuel cells.
Much of what you have learned about fluid mechanics involves flows in pipes on
the order of several inches in diameter (heating, ventilating, and air-conditioning),
on the scale of meters (propulsion), on the scale of tens of meters (aerodynamics),
or on the scale of hundreds or even thousands of meters (weather prediction) to
millions and billions of meters flowing ever outward toward the edges of the
universe: the astrophysics length scale.
On a smaller scale – but no less exciting and mysterious – this book is all
about fluid flows, heat and mass transfer, and electrokinetics at very small scales.
Microfluidics is generally viewed as the study of flows whose primary length
scales are below about 100 µm = 10−4 m (Table 1.1). Nanofluidics refers to
flows at length scales below about 100 nm. In recent years, manufacturing tech-
niques have been developed and used to build entire micromachines, includ-
ing thermal actuators, microvalves, micro- and nanopumps, gears, cantilevers,
and other microdevices (Gad-el Hak, 2001). These devices have been used for
biomedical applications such as drug delivery, chemical and biochemical sensors,
micromixers and microseparators, rapid molecular analyzers, development of an
artificial kidney, and DNA analysis and transport, to name only a few applica-
tions. Wouldn’t it be a major advance in chemotherapy if a drug were so specific
to cancer cells that healthy cells and tissues would be left alone? Today, most, if
not all, chemotherapy drugs kill both cancerous and healthy cells.
The number of devices having nanoscale features has exploded in the last
five years, and the number of companies that nanotechnology has spawned has
increased greatly. A Web search on the term “nanotechnology” yields thousands
of hits and nearly a dozen suggested new search terms.
The nanoscale is the scale of biology and chemistry. A glance at Figure 1.6
reveals that deoxyribonucleic acid (DNA) and the protein adenosine triphosphate
(ATP) have at least one of their dimensions at the nanoscale. The diameters
of many important proteins in a globular conformation are within the range of
2–10 nm. Drug delivery systems often transport carrier proteins through a series
of channels in parallel that allow uniform delivery of the drug. Lipid spheres,
 12 Essentials of Micro- and Nanofluidics

How Small is Small?

Water Glucose Antibody Virus Bacterium Cancer cell A period Tennis ball

10–1 1 10 102 103 104 105 106 107 108

Nanometers

Nanodevices
Nanopores
Dendrimers
Nanotubes
Quantum dots
Nanoshells

The scales of things, courtesy of the National Institute of General Medical Sciences of the National Institutes
Figure 1.6
of Health.

termed liposomes, are already on the market; these 100 nm diameter spheres
carry anticancer drugs to a target tumor (Malsch, 2005). Biochemical sensing
applications involve the transport of an analyte to a target, where it is then
identified by an electronic transduction process. In these applications, it is clear
that fluid dynamics is the underlying pillar of each process.
The birth of nanotechnology and micro- and nanofluidics is often attributed to
one man. Nobel Laureate Richard Feynman was a physicist known for his studies
of quantum electrodynamics, the superfluidity of supercooled liquid helium,
and high-energy physics. He won the Nobel Prize for his work on quantum
electrodynamics in 1965 and was known in his later years for popularizing physics
through both books and lectures until his death in 1988.
Many date the drive for miniaturization
to the vision Feynman presented in a paper
titled “There’s Plenty of Room at the Bot-
tom: An Invitation to Enter a New Field of
Physics” (Feynman, 1961), presented at the
American Physical Society annual meeting
on December 29, 1959. In that paper, he
asked the question, Why cannot we write
the entire 24 volumes of the Encyclopaedia
Britanica on the head of a pin? This would
require that each of the many thousands of
pages of this popular printed encyclopedia
be reduced in size by 1/25,000. Feynman’s
vision in 1959, before the advent of hand
calculators, much less laptops and iPads,
foreshadowed the revolution in miniaturization that has exploded during the last
20 years and is increasing in scope and speed daily.
Feynman actually issued two challenges, each with a $1000 prize. While
William McLellan built a 1/64th-scale electric motor (Gribbin, 1997) within a
 13 Introduction and Overview

year, technically meeting the challenge, it took until 1985 for Tom Newman to
reduce a book page to 1/25,000 its size and, in the process, greatly advancing the
science.
In a startling prediction foreshadowing the primary objective of drug delivery
in the twenty-first century, Feynman revealed that

a friend of mine (Albert R. Hibbs) suggests a very interesting possibility for rel-
atively small machines. He says that, although it is a very wild idea, it would be
interesting in surgery if you could swallow the surgeon. You put the mechanical
surgeon inside the blood vessel and it goes into the heart and “looks” around. (Of
course the information has to be fed out.) It finds out which valve is the faulty one
and takes a little knife and slices it out. Other machines might be permanently incor-
porated in the body to assist some inadequately functioning organ. Feynman (1961)

This is a stunning prediction in 1959 that is coming true right before our eyes.
Indeed, artificial organs are now common, and the design of an artificial kidney
is discussed in Chapter 12.
Feynman also suggested the explosion of interest in science and technology at
the length scale of biology, the nanoscale. Since 1989, the Foresight Conference
on Advanced Nanotechnology has been held annually. It is to Feynman that we
owe thanks for the explosion of the fields of micro- and nanotechnology, a branch
of which is micro- and nanofluidics.
Nanotechnology and, in particular, micro- and nanofluidics require a multi-
disciplinary approach. As mentioned earlier, the study of fluid flows at micro-
and nanoscales requires expertise in electrochemistry, but real expertise requires
an understanding of surface chemistry, electrostatics and electrokinetics, elec-
trochemistry, molecular biology, heat and mass transfer, and macroscale fluid
mechanics. To design devices having micro- and nanoscale features requires a
team approach involving chemists, biologists, medical researchers and practition-
ers, engineers, and systems analysts. As one might guess, each of these disciplines
speaks a different language, and it is sometimes only with great effort that these
technical language barriers are overcome.

1.5 What is a fluid?

Most students reading this book have had a standard undergraduate fluid mechan-
ics class. In this section, we review a few of those topics to refresh your memory
and to make sure we are using the same terms and notation you used previously.
You have had the experience of diving into a lake or swimming pool and watching
as the water parts to make way for your body. This experience is a reflection that
unlike a solid, a fluid is a material that continuously deforms under a shear stress,
or force per unit area, no matter how small the shear stress may be (Figure 1.7).
In contrast, a solid will break once the shear stress rises above a certain level.
There are two types of fluids: liquids and gases.
 14 Essentials of Micro- and Nanofluidics

In a solid, the individual molecules are

packed tightly together and will not move
unless a large enough shear stress is applied.
When a large enough shear stress is applied
to a solid, each molecule is displaced in
a uniform way. Conversely, in a liquid,
the molecules are farther apart so that
under an applied stress, different groups of
molecules may respond differently.
A gas will always expand to fill a con-
tainer. In a gas, the molecules are much
The impact of a red-colored drop on a pool farther apart than in a liquid. Conversely,
Figure 1.7
of water. A fluid always deforms continuously a liquid will occupy a finite volume based
when impacted by a liquid or solid. Image sup- on the amount of mass present. Thus, if the
plied by Wim van Hoeve, Tim Segers, Hans
container volume is larger than the volume
Kroes, Detlef Lohse, Michel Versluis, Physics of
Fluids group, University of Twente, Netherlands. of the liquid, a free surface will separate
the liquid from the gas, which will fill the
empty space in the container. A gas flowing at high speed (i.e., near or above the
speed of sound) behaves much differently from a liquid flow. However, at low
speeds (much less than the speed of sound), roughly at speeds less than 100 mph,
gas flows are quantitatively similar to liquid flows.
Much of classical fluid dynamics deals with gases such as air, hydrogen, oxy-
gen, helium, and steam and with liquids such as water, oil, gasoline, and alcohol.
The properties of these fluids are well known, having often been measured by a
number of different research groups and published as tables in multiple appen-
dices in fluid mechanics, heat transfer, and thermodynamics textbooks. In this
text, because many of the most interesting applications are biomedical, we will
also deal with liquid mixtures such as sodium chloride and water and phosphate
buffered saline (PBS), a biofluid that is similar in composition to serum in the
blood. Such fluids are called aqueous solutions and may be considered to have
a constant mass per unit volume, or density, as is the case for an incompress-
ible liquid. In addition, these liquid mixtures contain charged components, or
ions, which are atoms that have lost or gained an electron. Solutions contain-
ing charged species are called electrolytes or electrolyte solutions. Electrolyte
solutions respond to electric fields and provide the physical mechanism for the
occurrence of electrokinetic phenomena. For the case of a dilute mixture, or
low concentrations of the solute (i.e., the sodium chloride), these mixtures have
properties near to those of water.

1.6 Historical perspectives

The history of the thermal sciences, defined as the fields of fluid mechanics, ther-
modynamics, and heat and mass transfer, is fascinating because all the governing
 15 Introduction and Overview

(a) (b)

(a) Da Vinci’s drawing of a helicopter. (b) Self-portrait of Da Vinci from Brookhaven National Laboratory.
Figure 1.8

equations and boundary conditions, except for one, the no-slip condition, were
established within a 50 year period from 1800 to 1850. Moreover, James Clerk
Maxwell was working on the governing equations of electrostatics in the 1860s.
Thus, by around 1870, all the equations and boundary conditions associated with
the thermal sciences and electromagnetics were established, except for the no-slip
condition. It is for this reason that the history of fluid mechanics is particularly
interesting, and we begin there.1

1.6.1 Fluid mechanics

Fluid mechanics is one of the component parts of the thermal sciences, a field
that includes thermodynamics and heat transfer. As will be seen later in this text,
there are many similarities between fluid mechanics and heat and mass transfer
and even between the thermal sciences and electrostatics. This is especially true
of the boundary conditions that accompany the governing equations derived in
rather general form in Chapter 3. Indeed, it will be demonstrated that under rather
general conditions, the electrical potential and the fluid velocity are equivalent.
The formal study of fluid mechanics, defined as the study of both stagnant and
flowing liquids and gases, may have begun with Leonardo Da Vinci around 1500,
when he derived the equation of conservation of mass. He was the epitome of
the Renaissance man, a painter, sculptor, botanist, engineer, and scientist, among
other talents. His papers include sketches of free jets, vortex shedding behind
bluff bodies, and the first outline of a helicopter (Figure 1.8).
In 1687, Isaac Newton recognized the influence of viscosity when he said,
“The resistance which arises from the lack of lubricity in the parts of the fluid –
other things being equal – is proportional to the velocity by which the parts of
the fluid are being separated from each other” (as quoted in White, 2006, p. 2).

1
Much of the discussion here is based on the book by Tokaty (1971), an excellent reference for
the history of fluid mechanics in general.
 16 Essentials of Micro- and Nanofluidics

In today’s terminology, we would say that

stress is proportional to rate of strain; for
this reason, we call such fluids Newtonian.
The shear stress, τ , is defined by2
du
τ =μ (1.7)
dy
where μ is the dynamic viscosity, or simply
viscosity, and u is the velocity; y is a length.
The units of shear stress are N/m2 , or force
Streamlines for Stokes flow past a sphere.
Figure 1.9
per area, the same units as pressure. Water,
air, oil, and PBS (blood serum) are all Newtonian fluids.
The concept of shear stress originated by Newton would eventually become the
central concept in the derivation of the Navier–Stokes equations, the expression of
conservation of linear momentum. But first, in the 1660s, Newton had to invent the
calculus. The invention of the calculus enabled such applied mathematicians as
Daniel Bernoulli and Leonard Euler to approach fluid mechanics problems from
a mathematical perspective, resulting in the derivation of the Bernoulli equation
in 1738 (appearing in Bernoulli’s book Hydrodynamica) in a form essentially
unchanged today but that neglects the effect of viscosity; the equations governing
the flow of an inviscid fluid are called the Euler equations.
To incorporate the influence of viscosity, a frictional resistance term was added
to Euler’s equations by Navier in 1827, by Cauchy in 1828, by Poisson in 1829,
by St. Venant in 1843, and lastly, by Stokes in 1845 (White, 2006). Stokes
was the first to use the term coefficient of viscosity. The resulting equations are
now what we call the Navier–Stokes equations. The Navier–Stokes equations
are similar to the equations of stress in a solid (Dawson, 1976, p. 88), which
are termed Navier’s equations, and the Navier–Stokes equations are nonlinear,
coupled equations encompassing, at a minimum, four dependent variables, three
velocity components, and the pressure for an incompressible fluid.
Stokes flow is the name given to very slow flow of a fluid past a body or flow
past a very small body. Stokes, in fact, solved the Navier–Stokes equations for
flow past a sphere in 1851, thus deriving the drag formula (Figure 1.9)

D = 6πμU∞ a (1.8)

where D is the drag, μ is the viscosity, U∞ is the velocity far from the sphere, and
a is its radius. This formula is among the most famous in all fluid mechanics and
is the basis for the derivation of an estimate for the species diffusion coefficient
in a liquid mixture.
Even before the Navier–Stokes equations were derived, there was interest in
flows within pipes and channels (Tokaty, 1971). It appears as if a Frenchman

2
Actually, there are six distinct components of stress, as we will see in Chapter 3.
17 Introduction and Overview

named Chezy wrote the first equation governing flow in a channel around 1813,
but it was not until Jean Louis Marie Poiseuille performed experiments in 1838
that the relationship for the volume flow rate, Q (in the notation of Poiseuille),
pd 4
Q∝ (1.9)
L
was established (Tokaty, 1971). In this equation, p denotes the pressure drop,
d is the diameter of the tube, and L is its length. Poiseuille was a physiologist
who was interested in the laws governing the flow of blood. Subsequently, this
relationship was also established by Jacobson in 1860 and by Hagen in 1869 in
the form we now call the Hagen–Poiseuille formula. Why Jacobson’s name was
not added before Hagen’s name is unclear. Today, flow in a pipe or channel is
called Poiseuille flow.
In the same paper by Hagen in 1869, he suggested that in watching the flow
of dyed water in a cylinder, the fluid flows in concentric cylinders, which move
at different speeds, depending on their distance from the wall. This type of flow
has been termed laminar. Hagen’s experimental result and the hypothesis of the
no-slip condition (Goldstein, 1965b; Bernoulli, 1738) allowed Poiseuille to prove
theoretically that the velocity across the pipe was parabolic.
Throughout this period, the appropriate boundary conditions to impose on the
the aforementioned dependent variables in the Navier–Stokes equations had not
been fully identified though the no-slip condition had been suggested. Charles
Augustine Coulomb, around 1800, and later Stokes, around 1845, suggested that
the fluid should “stick” to a solid boundary, which we now term the no-slip
condition. It took the pioneering work of Ludwig Prandtl in 1904 to verify the
no-slip condition in his discovery of the boundary layer, that thin layer near a
solid surface in an external flow where the fluid velocity decreases rapidly to
zero.3 The notion of a viscous boundary layer is that viscosity is an important
physical effect only in a very thin layer near a solid surface; away from the
immediate vicinity of the surface, viscosity has no effect on the flow.

1.6.2 Heat and mass transfer

Heat transfer is the study of energy transport due to a temperature difference
(Incropera & Dewitt, 1990). While the basic notions of heat and temperature had
been identified in the 1600s, it was not until 1803 that an applied mathematician
named Joseph Fourier postulated that heat flows according to a temperature
gradient and that at steady state, heat flows from hot to cold, thereby establishing
Fourier’s law (Figure 1.10):
∂T
Q x = −k A (1.10)
∂x

3
Actually, the term boundary layer has been generalized to include any region where the fluid
velocity or any other dependent variable (concentration) rapidly changes.
 18 Essentials of Micro- and Nanofluidics

TC −TH where Q x is the heat transfer rate in the x

Q = −k A >0
L direction in watts (W), A is the area and k is
L a proportionality constant called the thermal
conductivity, which has units of W/m◦ K.
Hot Heat Cold This postulate was developed in the course of
Flow
the development of his famous Fourier series
method of approximating functions. How-
Insulation ever, it was not until the 1850s that the first
and second laws of thermodynamics were
Sketch illustrating Fourier’s law of heat trans-
Figure 1.10 developed by Rudolf Clausius and William
fer. In simple cases, the partial derivative is
constant and so may be approximated by the Thomson (Lord Kelvin).4
finite difference shown. The modern history of mass transfer, the
transfer of mass of a species due to a concen-
tration difference of the species (mass transfer requires that the fluid be a mixture),
appears to have begun with Thomas Graham, who was interested in the diffusion
of gases around 1830 (Cussler, 1997). Graham conducted experiments that took
place at constant pressure, a major result being the demonstration that diffusion
in liquids is orders of magnitude slower than diffusion in gases. Adolf Fick, who
was working in research in the general area of physiology, came across Graham’s
experiments and postulated that “the diffusion of dissolved material . . . is left
completely to the influence of molecular forces basic to the same law . . . for the
spreading of warmth in a conductor and which has already been applied with
great success to the spreading of electricity” (Cussler, 1997, p. 14). In other
words, Fick’s law was to be of the same form as the law that Fourier developed
in 1822 for the heat transfer rate through a body. Thus Fick wrote, for the mass
transport of a given species in the x direction (Figure 1.11),

∂ci
Ji = −D A (1.11)
∂x

where D is the diffusion coefficient, A is the cross-sectional area, ci is the

concentration, and Ji is the flow rate of species i in moles/sec, for example.
Note the similarity to Fourier’s law and to the equation for the viscous force
at a surface (F = Aτ ) defining Newton’s law. Indeed, the viscosity, thermal
conductivity, and thermal diffusivity or diffusion coefficient all play the same
role in these relationships and are called transport properties.
Fick published his first paper on diffusion in 1855, just after the Navier–
Stokes equations were developed, and he was initially unsure of his hypothesis.
He understood the difference between thermodynamic equilibrium and a steady

4
There are other pioneers of heat transfer associated with convection and radiation. For more
on this topic, I found the following Web site interesting: http://www.seas.ucla.edu/jht/pioneers/
pioneers.html. Another good Web site on the history of science and engineering is the Lienhard
engines site: http://www.uh.edu/engines/.
 19 Introduction and Overview

Sketch illustrating Fick’s law of mass transfer. In simple cases, the partial derivative is constant and so may
Figure 1.11
be approximated by the finite difference shown. The presence of large donor and receiver reservoirs ensures
that the concentrations do not depend on time.

state and tried to integrate the unsteady mass transport equation,

 
∂c1 ∂ 2 c1 1 ∂ A ∂c1
=D + (1.12)
∂t ∂x2 A ∂x ∂x

but the numerical effort was too much. He even tried to measure the second
derivative experimentally and finally succeeded in designing an experimental
system consisting of a glass cylinder containing sodium chloride on the bottom
and a larger volume of water on the top. After periodically changing the water in
the top of the container, he was able to establish a steady concentration gradient
(Figure 1.12). There are a great many other pioneers in the history of mass
transfer, but Fick and Graham are certainly the leaders in diffusion of mass.

1.6.3 Electrokinetic phenomena

Researchers were aware of electrokinetic phenomena as far back as 1808, when
F. F. Reuss discovered that an electric field can induce flow through a tube
(Burgeen & Nakache, 1964), which, we
have seen, is called electro-osmotic flow.
After about 50 more years of research, G.
Wiedmann deduced that the volume flow
rate through a tube was proportional to the
applied current. In 1859, Quinke discovered
the phenomenon of streaming potential.
James Clerk Maxwell laid the founda-
tions of the field of electrostatics with his
Figure 1.12
Fick’s experimental data; the data for the tube famous paper that presented what are now
indicates steady state mass transfer, and the called Maxwell’s equations in 1864. Prior
experimental comparison is very good. This was
his third attempt to validate his own law! From
to that, in 1856, he published a paper titled
Cussler (1997, p. 17). The result for the funnel “On Faraday’s Lines of Force” (Maxwell,
shows transient behavior. 1847), in which he suggested that “the
 20 Essentials of Micro- and Nanofluidics

present state of electrical science seems particularly unfavourable to spacula-

tion” (p. 27). Even in that paper, he recognized the link between the electrical
properties of a fluid and electrostatics, discussing, in a section titled “Theory of
the Motion of an Incompressible Fluid,” a version of the continuity equation and
what turns out to be the Navier–Stokes equations!
At about the same time Wiedemann and Quinke were making their discoveries
experimentally, James Clerk Maxwell was in the process of deriving the equations
governing electromagnetic phenomena. Although he was not truly the originator
of the individual equations of Coulomb, Ampere, and Faraday, he was the first to
derive them independently in a unified set (Constant, 1958). His equations include
an important modification to Ampere’s law, the law relating the magnetic field
to the current density; the collective set of equations is known as the Maxwell’s
equations.
In this presentation of the history of the thermal sciences and electrokinetics,
it is difficult to know when to stop. In the late 1800s, Helmholtz was working on
the electrical double layer, and as mentioned earlier, in 1904, Prandtl discovered
the boundary layer, thereby suggesting the last boundary conditions required
in the theoretical study of the thermal sciences and electrokinetic phenomena.
What is clear is that from around 1800 to 1870, the theoretical study of the
thermal sciences and electrokinetic phenomena exploded and set the stage for
the major discoveries of the late nineteenth and early twentieth centuries.

1.7 The thermal sciences

The design of air-conditioning systems, propulsion and power systems, internal

combustion engines, and many other systems involves not only fluid mechanics
but also thermodynamics, heat transfer, and mass transfer. Together these disci-
plines form the core of the subject known as the thermal sciences. Fluid mechanics
is the primary subject of this text, and in this section, the essence of thermodynam-
ics and heat and mass transfer is briefly described. Often heat and mass transfer
are lumped together because the disciplines use similar methods of analysis.
Thermodynamics is the study of

r Energy transformation
r Energy transmission
r Energy conversion

It forms the basis for an engineer’s initial analysis of the generation of power in
power plants and the operation of pumps, turbines, compressors, and fans (Figure
1.13). The science of thermodynamics determines the direction in which pro-
cesses occur. Thermodynamics is an equilibrium science in the sense that it can
predict the final temperature of a system but not the rate at which a system achieves
a final uniform temperature. The dependent variables in a thermodynamic analysis
 21 Introduction and Overview

are uniform in space, with no gradients

being permitted (Figure 1.14).
As you learned in your undergraduate
course work, thermodynamics is character-
ized by two laws: the first law of thermo-
dynamics, which is an equilibrium expres-
sion of conservation of energy, and the
second law of thermodynamics, which
determines the direction in which a pro-
cess will occur. Both these equations are
algebraic equations typically defining the
Picture of the Navajo Power Plant in Arizona, relationship between work and heat trans-
Figure 1.13
courtesy of the United States Geological Sur- fer and entropy changes as a function of
vey. Thermodynamics is used as an engineer’s temperature and pressure.
initial analysis to determine the operating char-
acteristics of a power plant.

Josiah Willard Gibbs

Josiah Willard Gibbs (1839–1903) was a Renaissance man being at once an
engineer, theoretical physicist, chemist and mathematician. Most students meet
Gibbs’s work early in their studies as the father of Vector Analysis. The analysis
of fluid mixtures is emphasized in this book and the importance of Gibbs’s work
cannot be underestimated. Gibbs made major advances in our understanding of
the thermodynamics of fluid mixtures. In mid-career he published his famous
paper Gibbs (1961) “On the Equilibrium of Heterogenous Substances,” a paper
that is viewed as a significant advance in the theory of physical chemistry. We
know Gibbs in thermodynamics for his development of the Gibbs phase rule,
Gibbs free energy, the Gibbs-Duhem equation and for the introduction of the
chemical potential. The recently published book by Kjelstrup & Bedeaux (2008)
is an extension of Gibbs’ work valid at equilibrium to non-equilibrium systems.

The rate at which an energy transfer process takes place can be predicted by
the discipline called heat transfer. Heat transfer describes energy flow as a result
of a temperature difference. The combustion process in a gas turbine involves
very large temperature gradients and many chemical reactions, and so heat and
mass transfer effects must be considered, along with fluid dynamics, in that
problem. Heat transfer occurs in a direction of decreasing temperature and is a
nonequilibrium phenomenon.
Heat transfer in the absence of a phase change can occur in three different
modes:
r Conduction
r Convection
r Radiation
 22 Essentials of Micro- and Nanofluidics

PEM FUEL CELL Conduction is associated with the random

Electrical Current
molecular motion of the molecules making
Excess e– e– Water and
Fuel Heat Out up the material of interest and is described
e– mathematically by Fourier’s law. Convec-
e–
tion is the transfer of energy between a
H– H2 O
solid surface and a moving fluid: forced
H–
H2
convection is the heat transfer within a
H– O2
fluid driven by external means, and free
H–
convection is the heat transfer through
a fluid due to a temperature difference.
Fuel Air
in in Radiation refers to the heat transfer due to
Anode Cathode
Membrane 250 μm electromagnetic waves moving to or from
surfaces. Heat transfer rates are governed
A thermodynamic analysis is used as an engi-
Figure 1.14 by the nonequilibrium form of conservation
neer’s initial attempt to determine the oper-
ating characteristics of a polymer electrolyte of energy in which temperature gradients
membrane (PEM) fuel cell. Image courtesy of in all three spatial dimensions can occur.
the Savannah River National Laboratory of the Analogous to heat transfer, mass transfer
Department of Energy.
describes the flow of a mass of a given
species as a result of a driving mechanism; the type of diffusion that is analogous
to Fourier’s law of heat transfer is Fick’s law, whereby mass transfer occurs as
a result of a concentration difference. As with heat transfer, mass of a given
species will flow from a higher to a lower concentration according to Fick’s law,
and we speak of convective and diffusion mass transfer. When we speak of mass
transfer, it is understood that the fluid in question is a mixture or a combination
of molecules of distinctly different types, in contrast to a pure fluid, which consists
of an ensemble of like molecules (Figure
1.15). Water, for example, is a pure fluid,
and salt water (i.e., salt fully dissolved in
water) is a mixture. Thus we can speak of a
flux of salt (i.e., mass/area/time) and a flux
of water. The total flux of the mixture is
obtained by a suitable averaging of the flux
of water and the salt (Figure 1.16).
The three most common modes of mass
transfer are

r Diffusion
r Convection
r Chemical reaction

Mass transfer occurs in the burning of rocket Convection mass transfer is entirely analo-
Figure 1.15
fuel. In general, the fuel may be a liquid or a
gous to convection heat transfer. In addition
solid, and oxygen or air induces combustion.
The products of the combustion shown are a to mass transfer due to a concentration dif-
mixture of the burned fuel and air. ference, as just described, diffusion mass
 23 Introduction and Overview

Proteins
Red Blood
Cell

Core
P1 P2
Region

Membrane

Mass transfer takes place in the kidney. Salt and other small molecules are filtered out of blood as urine under
Figure 1.16
the combined action of diffusion and convection.

transfer may also occur as a result of a pressure difference or pressure diffusion.

Thermal diffusion occurs as a result of a difference in temperature. These last two
modes of mass transport are usually not important in the applications discussed
here.
One final mode of mass transfer, which is nominally termed diffusion, is mass
transport due to an electrical potential difference. This mode of mass transfer is
the driving mechanism behind electro-osmotic flow.
It is important to note that the term mass transfer is not what is meant by flow
of a pure fluid such as water in a pipe or channel, even though mass transfer
is taking place. Mass transfer in the present context means that a given species
within a fluid mixture is transferred relative to bulk fluid motion.

1.8 Electrostatics

To use an electric field to move a fluid, the fluid must be electrically conduct-
ing. The electrically charged components of such a fluid are called electrolytes;
an example of an electrolyte is blood serum in aqueous solution, whose main
components are dissociated sodium chloride and water. The electrolyte can be
transported through a channel using an electric field, provided that the walls of
the channel are electrically charged; charged solid surfaces are discussed in the
next section.
That materials could be electrically charged was known to the Greeks as far
back as 600 B.C. For example, they found that amber rubbed with wool could
attract light objects. Thus amber could acquire an electric charge. The term
electric is derived from the Greek word elektron, which means “amber” (Burgeen
& Nakache, 1964).
There are two types of electric charge, negative and positive charge, with the
fundamental properties that like charges repel and unlike charges attract. All mate-
rial is made up of subatomic particles, and these fundamental particles are the neg-
atively charged electron, the positively charged proton, and the neutral neutron.
 24 Essentials of Micro- and Nanofluidics

An atom is a collection of these sub-

atomic particles, with protons and neutrons
generally arranged in a sphere called the
nucleus, around which are the electrons;
this atom is electrically neutral in that it has
no net charge. The radius of the nucleus
Direction of the electric field from (left) pos-
of an atom is about 10−14 m (Condon &
Figure 1.17
itively and (right) negatively charged point Morse, 1929). Atoms may gain or lose elec-
spheres. trons, and the process of doing this is called
ionization. Loss of an electron means that
the atom is positively charged, and the resulting atom is called a cation. A gain
of an electron means that the atom is negatively charged, and the resulting atom
is called an anion. Both cations and anions are called ions.
Electrically charged materials are generally grouped into two classes: conduc-
tors and insulators, or dielectrics. Metals are, in general, electrical conductors,
with electrons in their outer shell free to move, the free charge, and thus able to
pass electrical current freely; most other materials are electrical insulators that
do not pass electrical current in nearly the amount as conductors and in general
do not contain free charge. Electrolyte solutions are an exception to the rule
that dielectrics do not pass current. On the contrary, electrolyte solutions contain
dissociated ions that can move freely in solution; thus electrolyte solutions are
ionic conductors.
The essence of electrostatics is embodied in Coulomb’s law. Charles Augustin
de Coulomb (1736–1806) showed experimentally that the force acting between
two charged particles is
ee 
F= (1.13)
4π0r 2
where 0 is the electrical permittivity of a vacuum, 0 = 8.85 × 10−12 C2 /N m2 ,
r is the distance between the two charges, and e and e are the charges of the
particles. For one proton, e = 1.602 × 10−19 C, where C is coulombs. Note that
this law is of the same form as Newton’s law of gravitation,
Gmm 
F= (1.14)
r2
where G = 6.67300 × 10−11 m3 /kg sec2 is the gravitational constant. The unit
of charge is called the coulomb (C), and equation (1.13) is called Coulomb’s law.
The electric field at any point is defined as the force acting on a single charge

e at that point. The electric field generated by a single particle of charge e is
given by
F e
E= = (1.15)
e 4π0r 2
and the units of electric field are newtons per coulomb, or N/C. Note that if e is
positive, the electric field is directed outward in the radial direction (Figure 1.17),
and it is directed radially inward if the charge is negative.
 25 Introduction and Overview

If the medium is not a vacuum, then a relative permittivity, or dielectric

constant, is defined as

e
r = (1.16)
0

The electrical permittivity is a transport property just as is viscosity. Air is a poor

conductor of electricity; water is a much better conductor of electricity than air
but is still a dielectric.

1.9 Electrolyte solutions

We shall be concerned with the motion, the chemical kinetics, and the colloidal
nature of mixtures that involve the study of ions, the solute species, and a solvent,
which is usually water. The field of electrochemistry has traditionally comprised
two rather separate and distinct subjects (Bockris & Reddy, 1998): the study
of ionic solutions and electrodics. Electrodics involves the study of electrodes,
usually metallic, which generate an electric field, thus interacting with the ions
in solution in the interfacial region between the metallic and solution phases.
This portion of electrochemistry is extremely important in understanding the
operation of fuel cells. Ionic solutions are those mixtures that contain ions or
charged species; in this book, most, if not all, the ionic solutions that we consider
are aqueous solutions. The charged nature of ionic solutions means that they
can be transported using electric fields; thus, although rather distinct subjects
in their own right, it is the electrodes which create the electro-osmotic flow of
electrolyte solutions discussed in Chapter 9. There are many applications in which
electrochemistry is an essential tool, and many of them are described by Bockris
and Reddy (1998).
The fluids of interest in this text are aqueous solutions of electrolytes in which
water is the solvent and ions are the solute. Sodium and chloride, when dissolved
in water, dissociate into Na+ and Cl− , and these ions are surrounded by the water
molecules, a process known as hydration. In electrochemistry, a buffer consists
of a weak acid and a related salt so that the pH of the solution does not change;
sodium chloride, sodium phosphate, and potassium phosphate are the buffers in
phosphate buffered saline, a mixture with a
Table 1.2. Composition of PBS. composition close to blood serum and which is
The pH of standard PBS is 7.4
a common biological fliuid. The composition
Symbol Molarity (M) of PBS is given in Table 1.2.
Na+ 0.15420
Notation alert: In this section, we will employ
Cl− 0.14067
the notation used most often in the electrochem-
K+ 0.00414 istry literature, using square brackets to denote
H2 PO−
4 0.00147 activity. This eliminates the need to use long and
HPO2−
4 0.00810 cumbersome subscripts. Activity in chemistry is
defined as a = γ c, where c is concentration in
 26 Essentials of Micro- and Nanofluidics

mole/liter and γ is a dimensionless activity coefficient. On the other hand, engi-

neers define activity in terms of mole fraction, and activity is thus dimensionless:
a = γ X (Moran & Shapiro, 2007; Denbigh, 1971).
A Lewis acid is a proton (H+ ) donor so that in many cases, it is negatively
charged, and a Lewis base is a proton acceptor and so is positively charged. The
pH of a solution is defined as pH = log10 1/[H+ ] or [H+ ] = 10−pH and measures
the acidity of the solution. For pure water, which is electrically neutral [H+ ] =
[OH− ] = 1 × 10−7 or pH = 7. This definition of pH relies on the definition
of activity as [H+ ] = γ cH+ , which means that there is a dimensional quantity
inside the logarithm defining pH, a mathematical difficulty that persists today.
This difficulty can be avoided by defining the activity as a = γ c/c0 where c0 =
1 mole/liter. The definition of activity will be clear from the context.
Acids and bases are linked by the general proton transfer reaction

HA + B A− + BH+ (1.17)

The products of this reaction are termed an acid–base pair. The extent of the
dissociation is measured by an equilibrium constant, defined by

[A− ][BH+ ]
K= (1.18)
[B][HA]

The concept of an equilibrium constant

described here is the same as is custom-
arily used in the thermodynamic analy-
sis of combustion, another example of
a type of chemical reaction. The limit
K → 0 means that little dissociation has
occurred. The valence of an ion is a whole
number and indicates the number of elec-
trons the ion has gained or lost. In aque-
ous solution, for example, calcium has a
Gatorade is an electrolyte mixture that replenishes
valence z Ca = +2. The protein albumin
the electrolytes in the body after intense physical
activity. The salts in the blood serum are removed has a valence estimated to be z Alb = −17
from the body when you sweat. (Peters, 1996).

1.10 The electrical double layer

In dissociated electrolyte mixtures, where a molecule (such as NaCl) has broken

up into its ionic components Na+ and Cl− , even in the absence of an imposed
electric field, an electrical double layer (EDL) will be present near the charged
surfaces of a channel or tube; this situation is depicted in Figure 1.18. For example,
 27 Introduction and Overview

ζ H 2O
SiOH

+
– + Na
SiO Na

–
SiO

Diffuse Layer
–
SiO
H2O

Silica Stern
SiOH
Layer Cl
–

–
SiO

–
SiO
H 2O

–
SiO Na
+

–
SiOH2 + Cl

Graphic representation of the hydration of ions by water molecules near a charged silica surface. The electric
Figure 1.18
double layer (EDL) consists of a layer of counter ions pinned to the wall, the Stern layer, and a diffuse layer
of mobile ions outside that layer. Here the wall is shown as being negatively charged, and the ζ −potential is
defined at the Stern plane.

if a wall is negatively charged, an excess of positive charge will accumulate near

the wall because opposite charges are attractive. The thickness of the layer near
the wall is of the order of the ionic diameter of hydrated counterions, and this
layer is called the Stern layer (Figure 1.18). Any charged molecule sufficiently
far away from the surface will be “screened” from the negatively charged wall
by the presence of the positive ions. The layer outside the Stern layer consists
of mobile ions and is called the diffuse layer, and the total system, including the
Stern layer and the diffuse layer, is called the EDL (Hunter, 1981). The presence
of the charged surface and the ions thus creates an electric field.
The nominal length scale associated with the EDL is the Debye length, defined
by
√
e RT 1
λ= = (1.19)
FI 1/2 κ

where F is Faraday’s constant; e is the electrical permittivity of the medium;


I is the ionic strength, I = i z i2 ci ; ci are the concentrations of the electrolyte
constituents at some reference location; R is the universal gas constant; z i is the
valence of species i, and T is the temperature.

Petrus (Peter) Josephus Wilhelmus Debye

Petrus (Peter) Josephus Wilhelmus Debye (1884–1966) received his first degree
in Electrical Engineering in 1905 and then a PhD in Physics in 1908 under the
tutelage of Arnold Sommerfeld. At Zurich in 1920 as a Professor of Physics
Debye along with his assistant Erich Huckel developed a theory of why concen-
trated electrolytes exhibit behavior different from classical ideal solutions. They
attributed the discrepancy to the presence of electrostatic interactions between
ions; these results are presented in Debye and Huckel (1923).
28 Essentials of Micro- and Nanofluidics

Notation alert: In the chemistry literature, the ionic strength is usually defined

by I = 1/2 i z i2 ci . A one molar solution (1 M = 1 mole/liter) of dissociated
sodium chloride is defined by the concentration of each species being 1 M. With
the current definition, the same mixture has the ionic strength equal to 2 M –
1 M sodium and 1 M chloride, for example. The present definition is to avoid
a 2 floating around in other equations involving the Debye length. It must be
said, however, that the classical chemistry definition works only for monovalent
species for which the concentrations are equal. For multivalent ions, the classi-
cal definition is extremely confusing. Where absolutely necessary, however, the
classical definition will be used.
Internal flows in cylindrical or rectangular channels are the most common form
of transport in micro- and nanofluidics. The material used for the walls of these
channels is very important because of the large surface-to-volume ratio; silica,
when immersed in water, becomes charged and loses a proton, according to the
reaction

SiOH(solid) ↔ SiO− (solid) + H+ (aqueous) (1.20)

Thus the silica walls are negatively charged and will attract the positive ions
which are in solution, forming a thin layer of positively charged ions pinned at
the wall. For a positively charged wall, the roles of the positive and negative
charges are reversed.
It should be pointed out that there are many silica surfaces of different chemical
composition, and no attempt will be made in this text to make this distinction.
The primary difference between silica surfaces is the value of the ζ potential.
The EDL will always be present near a charged wall, even in the absence
of a flow. If electrodes are placed upstream and downstream of a rectangular
channel, then a separate and distinct electric field is set up, and the ions will
be set into motion, dragging the solvent with it and inducing electro-osmosis. The
device depicted in Figure 1.19 is called an electro-osmotic pump. The abundance
of cations over anions within the diffuse layer generates the bulk motion of the
fluid. In this pump, the anions will be attracted to the positive electrode, which is
termed the anode. Likewise, the cations will be attracted to the negative electrode,
or cathode. Thus the cations will move at a speed slightly greater than the bulk
electro-osmotic flow, while the cations will move at a speed slightly slower than
the bulk electro-osmotic flow. The speed of the ions depends on their valence;
the higher the positive valence, the faster the cation speed. The unique feature of
electro-osmotic flow is that for a wide range of parameters away from the walls
of the channel, the velocity will be constant, unlike for Poiseuille flow, which is
parabolic.
The thickness of the EDL is actually an asymptotic property much like the
boundary layer thickness in classical external fluid mechanics (White, 2006),
which is defined as the location where the velocity reaches 1 percent of its value
 29 Introduction and Overview

Electrodes placed upstream and downstream of a channel having charged walls, generating an electro-osmotic
Figure 1.19
flow.

far from the wall. If we define the dimensionless parameter

λ
= (1.21)
h
where h is the channel height, then for   1, the dimensionless thickness of
the EDL, defined here as the location where the electrical potential reaches 1%
of its value far from the wall, is normally δ/ h ∼ 4 − 6, depending on the
ionic strength. This situation is similar to the formula for the thickness of the
Blasius boundary layer on a flat plate, denoted by δ and δ/x = 5 Re−1/2 , where
x measures distance from the leading edge of the plate and Re is the (large)
Reynolds number. The Reynolds number is the main dimensionless parameter
associated with viscous flow, and for the Blasius boundary layer on a flat plate of
length L, it is defined by
ρU∞ L
Re = (1.22)
μ
where ρ is the fluid density, μ is the viscosity, U∞ is the velocity far from the
plate, and L is its length. If  ∼ 1, we say that the EDLs overlap. The ionic
concentration and electrical potential distributions within the EDL are discussed
extensively in Chapter 7.

1.11 Colloidal systems

The solutions considered by electrochemists often contain embedded particles.

The term colloid science refers to the science of transporting particles and
molecules of a size less than 1 µm and greater than about 10 Å. It should be
cautioned that the term colloid is very broad, and the intent here is not to define
specific terms relating to different colloid systems but to give an overview of the
types of colloidal systems that are common in micro- and nanofluidics.
A colloidal dispersion can be a mixture that contains relatively large, soft,
or hard polymeric molecules; solid particles; or small molecules such as salt-
dissociated ionic species. Most ions have a dimension on the order of angstroms,
30 Essentials of Micro- and Nanofluidics

Table 1.3. Examples of two-phase colloidal systems. A gold sol is a mixture

of gold nanoparticles
Continuous Dispersed Name Example
Liquid gas foam beer
Liquid liquid emulsion milk
Liquid solid colloidal suspension gold sol

and the radius of a globular glucose molecule is about 1 nm; the globular protein
albumin has a nominal radius of about 3.5 nm. Examples of colloidal systems
are given in Table 1.3 (Murrell & Jenkins, 1982). Mixtures containing particles
larger than 1 µm have been termed by some researchers as suspensions (Raymond,
2007), and modeling of these types of systems is extremely difficult. A gold sol
is a mixture of gold nanoparticles.
Thus colloidal systems include saltwater mixtures, biofluids such as blood, and
systems including proteins and DNA, which is a biopolymer. Blood is a fluid that
consists of suspended particles (red and white blood cells) in an aqueous solution
of proteins and salts (electrolytes) called plasma. Colloidal systems are, more
often than not, aqueous systems.
Polymers are large-molecular-weight materials made up of a large number of
atoms, linked together in a chain. There may be one or several repeating groups,
depending on the specific polymer. Chemists consider a large-molecular-weight
polymer to be above approximately Mw = 500 g/mol (Murrell & Jenkins, 1982).
For example, polystyrene beads, which are often used to measure fluid velocities
(Sadr et al., 2004, 2006, 2007), have a molecular weight of Mw = 108 g/mol.
Because polymers are so large, researchers have suggested that solutions con-
taining polymers can never be considered ideal solutions. However, this rather
obvious statement is contradicted by evidence that these solutions often behave
as if they were ideal (Murrell & Jenkins, 1982).
Polymers are often soft in the sense that they will change shape in solution
based on the nature of its constitution. It is not hard to imagine that a polymer
may fold in on itself in water if some of its groups are hydrophobic, or “water
hating.” Polymers can also stretch out linearly, as DNA does when it is exposed
to sufficient fluid stress.
Other types of colloid systems are dispersions of soluble and insoluble mate-
rial in water. The former are called lyophilic colloids, and the latter are called
lyophobic colloids. The dispersed phase (i.e., the particles) can actually aggre-
gate and form larger complexes and, in some cases, must be considered two-phase
systems. Lyophilic colloids are commonly treated as a one-phase but perhaps mul-
ticomponent mixture. In this case, the continuous phase is merely the solvent:
water.
Gels are colloidal fluids that contain large macromolecules made up of long
polymer chains. Aqueous solutions containing naturally occurring proteins, such
as collagen or sugars, are in the gel family. The water molecules interact with
31 Introduction and Overview

one or several of these long chain polymers and occupy the spaces between the
polymers. The term emulsion is used to denote an aqueous solution containing
oils or fat (Table 1.3).
In theory, the most important factors influencing the behavior of a colloidal
mixture are particle size, particle shape and rigidity or flexibility, channel and tube
surface properties (i.e., surface charge density), particle–particle interactions, and
particle–solvent interactions. In practice, however, the most important properties
of a dilute electrolyte mixture are the channel and tube surface properties and the
particle solvent interaction through hydration (i.e., solvation) (Shaw, 1969) (Fig-
ure 1.18). For dilute mixtures typical of biological applications, particle–particle
or molecule–molecule interactions should be a second-order effect. Moreover,
particle size only becomes important when the ratio of particle size to the chan-
nel or tube dimension is O(0.1), although the specific limit may depend on the
application.
In colloidal systems, interparticle interactions are usually represented by a
potential that incorporates the attractive van der Waals forces and the repulsive
electrostatic forces. The van der Waals force represents the effect of the finite size
of individual molecules, and the van der Waals equation of state, to be discussed
later, represents a correction to the perfect gas law, for which intermolecular
forces are not negligible.
If electrostatic forces are negligible, then the van der Waals forces will cause
colloidal particles to stick together; this is an indication that the colloidal mixture
is unstable. Van der Waals forces are a class of forces between molecules or parts
of molecules other than those due to covalent bonds (the sharing of pairs of elec-
trons between molecules) or the explicit electrostatic forces. These forces arise
from dipole–dipole interactions, from induced dipole–dipole interactions, and
from dispersion forces due to transient dipoles in atoms. For example, water has
a permanent dipole moment (see Chapter 6 for the definition of dipole moment),
and so there will be a strong van der Waals force existing between two neighbor-
ing water molecules. Van der Waals forces can overwhelm electrostatic forces,
even if the particles are charged.
The study of colloidal stability was first described in separate works by
Derjaguin and Landau (1941) and Verwey and Overbeek (1948), and for this
reason, theoretical work on the stability of colloidal systems is named for these
four authors as DLVO theory.
Colloidal particles can be used as simulants for DNA and other biomolecules.
As will be seen, colloidal particles, such as polystrene beads, can be used to
measure velocities near the wall in a microchannel. Negatively charged and
fluorescence-labeled polystyrene (PS) beads of size 3–40 nm have been used to
examine the transport of particles in micronozzles, as depicted in Figure 1.20
(Wang et al., 2005, 2008). The length of the nozzle is 650 µm, and the left-end
and right-end heights are 20 µm and 130 µm, respectively; the nozzle is 40 µm
deep, and the walls are negatively charged. The particles are immersed in a 0.1 M
NaCl solution, and an 80 V/cm DC electric field was applied across the tapered
 32 Essentials of Micro- and Nanofluidics

Polystyrene beads are sometimes used as biomolecule simulants. The particle donor reservoir is on the right
Figure 1.20
and the particle receiver reservoir is on the left of the diffuser (Wang et al., 2005, 2008). The walls are negatively
charged, and the positive electrode is placed in the particle receiver. The flow is from left to right and thus it
is a diffuser. The polystyrene beads are shown, and their motion is from right to left, in a direction opposite to
the bulk electro-osmotic flow. Picture courtesy of L. J. Lee. Used with permission.

channel, with the positively charged electrode placed in the particle receiver
region (left). Since the walls are negatively charged, cations will always be more
populous than anions and will move to the negative electrode, dragging the bulk
fluid flow from left to right. Thus, in terms of the bulk electro-osmotic velocity,
the device is a diffuser. The negatively charged PS beads move from right to left
in a direction opposite to the bulk flow, so for these particles, the device is a
nozzle. More will be said about such electrokinetic phenomena in Chapter 9.

1.12 Molecular biology

According to Freifelder (1987), the term molecular biology was first used by
William Astbury in 1945, when he was studying the chemical and physical
characteristics of macromolecules. Significant advances were made after that,
when it was recognized that the study of biologically simpler entities, such as
bacteria and bacteriophages (bacterial viruses), was necessary prior to the study
of more complicated entities such as animal cells. This kind of activity led Watson
and Crick to make their discovery of the structure of DNA in the 1950s (Crick and
Watson, 1953) (Figure 1.21). The sizes of some important biological molecules
are depicted in Figure 1.6.

Crick and Watson (1953) revealed the structure of DNA in 1945. Sketch courtesy the National Institute of
Figure 1.21
General Medical Sciences of the National Institutes of Health.
 33 Introduction and Overview

Today, more than 50 years later, rapid advances are being made in determin-
ing the structure and function of single proteins and other biomolecules. Such
studies have application in many areas, including rapid molecular analysis, gene
therapy, biochemical sensing, and drug delivery. These applications require the
development of engineering tools to design entire systems, often specifically for
the study of single biological macromolecules. Since the nanoscale is the scale
of biology, nanotechnology is being used to create a new field called molecular
medicine (Roco, 2005). As Roco points out, significant advances have been made
in “measurements at the molecular and subcellular levels and in understanding
the cell as a highly ordered molecular machine” (Ishijima & Yanagida, 2001,
p. 3). Moreover, the molecular origin of disease can be better understood, leading
to a much better chance for a cure to such diseases as cancer, diabetes, cystic
fibrosis, and neurological disorders, to name only a few.
Cells are populated by macromolecules; in vivo, they are very large molecular
weight (104 –1012 Da) (1 Da = 1 gram/mol) polymers that inhabit a cell which may
contain 104 –105 different types of molecules. Conversely, many of the molecules
that inhabit the cell are very small, having molecular weights no more than several
hundred Da.
Biomacromolecules or biopolymers are of three different types: proteins,
nucleic acids, and polysacchrides (sugars). The three most important properties
of these macromolecules from an engineering perspective are their physical size,
shape, and charge, if any. For example, most proteins are negatively charged,
while glucose is uncharged. As will be seen, this one property of a molecule
significantly affects transport mechanisms in nanopore membranes.
The three-dimensional structure of many proteins has been determined exper-
imentally, usually by protein crystallography, and the data have been stored in the
Protein Data Bank (PDB) by Brookhaven National Laboratory. The Web site is
located at http://www.rcsb.org, and the PDB was cre-
ated in 1971. The Worldwide Protein Data Bank was
created in 2004 to coordinate the worldwide dissemi-
nation of protein structures. Ribbon diagrams of pro-
tein structure, also known as Richardson diagrams
(Richardson, 1973), which are an interpolation of
the smooth curve through the polypeptide chain of
the protein, can be downloaded in a variety of for-
mats. Over 50,000 structures exist in the World-
wide Protein Data Bank site. A detailed discussion
of the structure and function of proteins and other
biomolecules appears in section 8.3. One represen-
Ribbon view of the structure of a tation of the primary protein structure is depicted in
Figure 1.22
typical protein, one type of macro- Figure 1.22.
molecule that resides in a cell.
In addition, there has been much work on
Proteins are discussed in detail
in Chapter 8. Image courtesy of molecular simulation of proteins. These calcula-
Brookhaven National Laboratory. tions can identify the conformation and binding
 34 Essentials of Micro- and Nanofluidics

characteristics of proteins surrounded by water (Becker & Karplus, 2006;

McCammon & Harvey, 1987) and the transport of single- and double-stranded
DNA (Cui, 2004). These simulations provide a molecular picture of biomacro-
molecules and biopolymers not possible with the continuum approach.
Much of the analysis of macromolecules in the balance of the book concerns
proteins. The study of the structure and function of a set of proteins, or proteome,
associated with a given set of genes, or genome, is called proteomics. Increasingly,
micro- and nanofluidic devices are being used to analyze single proteins. For
example, a mass spectrometry device can be used to measure the charge to
mass ratio of various ionic constituents that have been fragmented from specific
proteins (Glazer & Nikaido, 2007; Landers, 1994).

1.13 The convergence of molecular biology and engineering

Roco (2005) has pointed out that nanotechnology is beginning to play a revolu-
tionary role in biomedicine. Indeed, all biological systems originate with entities
that have spatial dimensions on the nanometer scale. Roco (2005) defines nan-
otechnology as “the ability to measure, design and manipulate on the molecular
and supramolecular levels on the scale of about 1 to 100 nm in an effort to
understand, create and use material structures, devices, and systems with fun-
damentally new properties and functions attributable to their small structures”
(p. 69). The goal of nanotechnology, he writes, is “to assemble molecules into
useful objects hierarchically integrated along several length scales and then, after
use, disassemble objects into molecules.”
The convergence of molecular biology and engineering may be illustrated
by the analogy between natural ion channels and the channels that compose a
synthetic nanopore membrane. Natural nanoscale conduits called ion channels
play a crucial role in the transport of biofluids into and out of cells (Figure 1.23).
The basic units of all living organisms are cells. To keep the cells functioning
properly, a continuous flux of ions in and out of the cell and the cell components
is required. The ion channels play an important role in this process. The walls
of the ion channel consist of electrically charged proteins, most often negatively
charged, that control the transfer of ions. The polarity of the membrane makes
it challenging for some ions and molecules to move in and out of cells and
their components. Ion channels are also size selective. Natural ion channels are
roughly circular, although the cross-sectional area varies in the primary flow
direction.
Advances in micro- and nanofabrication techniques resulting in some of the
devices discussed in some of the previous sections have channels with dimensions
approaching molecular scales in one dimension. The individual channels in these
nanopore membranes may thus be termed synthetic ion channels. Moreover,
analysis of the natural ion channel is very often performed by the same methods
used for the synthetic ion channel (Chen et al., 1995; Gillespie, 1999; Gillespie
& Eisenberg, 2001; Hollerbach et al., 2001).
 35 Introduction and Overview

Computational simulation of the potassium (K + ) ion channel showing the proteins in a ribbon view that make
Figure 1.23
up the walls of the channel. Ion channels are discussed in Chapter 8. Image courtesy of the National Center
for Computational Science.

These so-called nanopumps can be designed for a specific task, whether it

be drug delivery, chemical and biochemical sensing and analysis, or molecular
separation. The primary design variables are the dimensions of the channels
in the array, the materials used for the chip, and the size of the upstream and
downstream reservoirs (see Figure 1.1). Often nanopore membranes have been
employed for the transport of globular proteins such as albumin and lysozyme
and for sugars such as glucose. As mentioned before, these molecules have a
typical dimension on the order of ∼3 nm, and transport of such proteins has been
demonstrated in channels having their smallest dimension on the order of 10 nm
(Martin et al., 2005). Having such a small dimension ensures that the amount
of the protein delivered through the membrane increases linearly with time – so
called zeroth-order release in the drug delivery literature.
The design of these devices thus requires a knowledge of the structure and
properties of biomolecules and a good grasp of the basic principles of the ther-
mal sciences, particularly fluid dynamics and mass transfer, and electrostatics
and electrodynamics. While the electrostatics–electrodynamics portion of this
material is not generally taught within the undergraduate mechanical or chemical
engineering curriculum, it is routinely taught within bioengineering graduate pro-
grams. Thus the basic principles of molecular biology and thermal engineering
must combine to produce efficient and useful design tools.

1.14 Design of micro- and nanofluidic devices

The first step in the design of a micro–nano device may be the development of
a theoretical or computational model of the device prior to the fabrication of a
model device. Theoretical and/or computational models are generally much less
expensive than the actual fabrication of the devices.
 36 Essentials of Micro- and Nanofluidics

Market

Initial Need

Prototype

Yes

Simplified Model for No

Device
Successful Design?
(Component and System) Has the need been met?

Solution for
Component and System Build a Model Device
Operation

Simplified view of the design of a micro–nanofluidic device. The actual process may include a theoretical–
Figure 1.24
computational retro design at the prototype stage.

Developing theoretical and/or computational models of flows at the micro- and

nanoscales is especially important because experimental methods are limited to
larger scales. To see this, consider the flow of sodium-chloride-water in a square
channel of the order of 20 µm on a side. Suppose the walls of the channel are
charged and that the motion is driven by an external electric field or electro-
osmotic flow. On the wall, there is the EDL, which in general is of the order of
1–100 nm thick. To probe the EDL, spatial resolution on the order of 1/10th of
the thickness is required (Sadr et al., 2006). Local electrostatic forces and charge
density can be measured using atomic force microscopy (AFM), but this is an
intrusive method that disturbs the region of interest. A surface force apparatus
(SFA) may also be used, but this method is also intrusive. Both methods have
reported spatial resolutions of O(1–10 nm) (Sadr et al., 2006).
In general, AFM and SFA techniques cannot be used to probe EDL structure
in a flowing fluid. Thus some other nonintrusive technique is required. Sadr et al.
(2004, 2006) have developed a technique they call nano-particle image velocime-
try (nPIV), which uses evanescent waves to measure the average velocity within
the first 400 nm next to the wall. Thus velocity, temperature, concentration, and
electric potential profiles cannot be measured in channels having dimension(s)
less than ∼1 µm.
Thus modeling is essential at the micro- and nanoscale. Moreover, using dimen-
sional analysis, as described in Chapter 3, microscale experimental measurements
 37 Introduction and Overview

at small ionic strength can often be made equivalent to nanoscale experimental

measurements at a higher ionic strength. Thus theoretical and computational
results at the nanoscale can validate microscale experiments, and microscale
experiments will validate theoretical and computational results at the nanoscale.
While experimental measurements at the microscale are difficult, they are an
essential tool in the design of microfluidic devices.
It is useful to outline a typical design process in general terms. It is clearly
the case that a given design process will vary widely, depending on the applica-
tion. The development of a microfluidic device and, in particular, a biomedical
microfluidic device first requires that a need be identified. For example, there may
be a need for an artificial kidney, or a device to treat cancer, or a flow-through
biosensor. Such a device could use electro-osmotic pumping, or in the case of the
development of an artificial kidney, the natural pressure drop in the body (Fissell,
2006; Fissell & Humes, 2006; Conlisk et al., 2009). Then a simplified model of
the device should be developed to identify the physical parameters, such as size,
flow rate, and concentrations, required to satisfy the need.
Once the simplified computational model is completed, a model device should
be fabricated and experimental results compared with the solution of the sim-
plified model. Once the model device has been validated (see Section 10.14)
and the experimental results agree to a specified tolerance, a prototype device,
suitable for clinical trials, can be fabricated. Finally, the device will go to market
(of course, if the Food and Drug Administration (FDA) agrees). Note that this
process involves the triad of theory–computation, fabrication, and experiment,
all three of which are essential features in micro- and nanofluidics. A generic
design procedure is depicted in Figure 1.24.

1.15 Unit systems

In this book, the System Internationale (SI) unit system will be used exclusively.
The units for each major quantity of interest are defined in Table 1.4, where,
for example, the unit of force is the newton, N, the unit of mass is the kilogram, kg,
and the unit of length and time are the meter, m, and second, sec, respectively. Thus
m
1 N = 1 kg
sec2
The system of units summarized in Table 1.4 is standard in undergraduate
thermal science textbooks.

1.16 A word about notation

A few words on notation in this book are due. In an interdisciplinary book such
as this, notation can be a nightmare. Across the disciplines of fluid mechanics,
 38 Essentials of Micro- and Nanofluidics

Table 1.4. SI units used for analysis of an electrically

conducting fluid
Quantity Unit Symbol
Mass kilogram kg
Length meter m
Time second sec or s
Force newton N
= 1 kgm
s2

Temperature kelvin K
Energy joule J
Concentration moles/liter M
3
Concentration kgi /m
Charge coulomb C
Potential volt V
= 1 Nm
C

heat and mass transfer, electrostatics, and electrochemistry, the same symbols
are often used. In addition, sometimes it proves useful to work in dimensionless
form, whereas at other times, it does not. The philosophy I have thus taken is
that all notation is local; that is, the meaning of each quantity, where there may
be some question of its meaning, is defined in each section or subsection. For
example, the internal energy in thermodynamics is denoted by u, which is also
most often used for the streamwise fluid velocity.
In the section on electro-osmotic flow in nozzles and diffusers, E x is used
for the dimensionless electric field in the x direction, whereas in the section on
Henry’s solution for the electrophoretic velocity of a charged particle, the same
symbol is used for the dimensional quantity. It simply makes no sense to present
Henry’s equation in dimensionless form, and it is maddening to continually use
a superscript asterisk for dimensional quantities, as is often done.
I considered including a nomenclature so that definitions need not be provided
locally. However, this requires the reader to continually flip back and forth to
check definitions. Finally, to aid the reader, where appropriate, I have included
“notation alerts” to minimize confusion.

1.17 Chapter summary

The purpose of this chapter is to give the reader an overview of the field of
nanotechnology, in general, and micro- and nanofluidics, in particular. Principles
of micro- and nanofluidics are being used to design devices that can be used for
biomedical applications such as drug delivery, chemical and biochemical sensors,
micromixers and microseparators, rapid molecular analyzers, development of an
39 Introduction and Overview

artificial kidney, and DNA analysis and transport. During the course of this
discussion, it becomes evident that the nanoscale is the scale of biology and
chemistry. Most of the concepts introduced in this chapter will reappear later in
the book, some, such as the Debye length, many more times.
Micro- and nanofluidics cuts across a number of disciplines. It has elements of
fluid mechanics, to be sure, but also mass transport, electromechanics, molecular
biology, and electrochemistry. All these fields are introduced in general terms
in this chapter. The historical summary clearly indicates that the fields of fluid
mechanics, heat and mass transfer, and electrostatics all developed simultaneously
in the middle to late nineteenth century.
Many applications to be discussed in this book involve the transport of
biomolecules such as proteins. Indeed, the applications cited here require the
development of engineering tools to design entire systems, often specifically
for the study of single biological macromolecules. A brief introduction to
biomolecules (i.e., DNA and proteins) is presented that will lead into the detailed
discussion of the structure and function of proteins and other biomolecules that
appears in Section 8.3.
One glance at the table of contents and the discussion in this chapter reveals
the central theme of this book: micro- and nanofluidics unifies and integrates the
fields of fluid mechanics, heat and mass transfer, electrostatics and electrokinetics,
and electrochemistry. No other subject can be said to do this.
 2 Preparatory Concepts

2.1 Introduction

All of the physical laws of micro- and nanofluidics, whether fluid mechanics,
heat and mass transfer, or electrokinetics, involve transport coefficients. The
Navier–Stokes equations require viscosity for their application to the viscous flow
phenomena so central to micro- and nanofluidics. Each of the other disciplines
has its own dependence on transport coefficients, and it is surprising how similar
these relationships are. Understanding the similarities and differences in the
roles these transport properties play in each discipline forms the basis of this
chapter.
Accordingly, in this chapter, we present the basic constitutive laws of fluid
mechanics, mass transfer, heat transfer, and electrostatics. These constitutive
laws form the basis for the derivation of
the governing equations in Chapter 3. We
also discuss concepts of surface tension and
wettability that are important because of the
large surface-to-volume ratio in micro- and
nanofluidics. We then discuss how to deter-
mine these transport coefficients for use in
the solution of problems.
Water, paint and oil have very different values Having presented these constitutive laws,
of viscosity. we discuss the various types of fluid flows,
from macroscale down to nanoscale, and
the differences among internal and external flow. This section is followed by
the presentation of the basic elements of thermodynamics and the interrelation-
ship between fluid mechanics, heat transfer, and thermodynamics for pipe and
channel flows, the internal flows that are common in micro- and nanofluidics.
The irreversible conversion of mechanical energy to unusable thermal energy
creates the pressure drop in a pipe that is directly proportional to the viscosity;
inversely proportional to D 4 , the diameter of a pipe; and inversely proportional
to h 3 , the height of a wide channel. It is the last two dependencies that reduce the
effectiveness of fluid transport by pressure at the micro- and nanoscale.

40
 41 Preparatory Concepts

This chapter concludes with a summary and set of exercises designed both
to refresh the engineer’s memory and to introduce potentially new material,
specifically in the area of electrostatics.

2.2 Important constitutive laws

All materials, whether solids or fluids, have physical properties that determine
the materials’ behavior under the influence of an external action. Moreover, there
are certain quantities, which may be termed dependent variables, that depend on
these physical properties and that are important to determine by either experiment
or calculation. These variables are identified as follows:
r Fluid mechanics: velocity field (u, v, w), pressure p
r Single-phase heat transfer: temperature T
r Mass transfer: mass density of species A: ρ or molar density (concentration)
A
of species A: c A
r Electrostatics/dynamics: electric potential φ and/or electric field E

In identifying these dependent variables, the medium has to be assumed incom-

pressible; if the medium is not incompressible, the density must be added to the
dependent variable list. These dependent variables can be measured or calculated
as the solution of the partial differential equations that arise in the development
of conservation laws such as conservation of mass, momentum, energy, and elec-
tric charge. To fully characterize a system, then, a sufficient number of values
of the seven unknowns (for a binary mixture) in the preceding list should be
measured or calculated. For the case of an N -component mixture, then, there
are 6 + N − 1 unknown dependent variables in the general three-dimensional
case.
It should be noted that pressure, density, and temperature are also called
thermodynamic properties in the sense that the change in these properties from
one physical state to another is independent of the process path between the two
states. The quantity entropy, which determines the direction of a process, is also
a thermodynamic property.
The conservation laws mentioned earlier contain parameters that characterize
the material of interest, whether it be a solid or a fluid. These material properties
are generally grouped into two classes:

1. Transport properties, for example, viscosity, thermal conductivity, diffusion

coefficient, electrical permittivity;
2. Thermodynamic properties, for example, density, internal energy, enthalpy,
and specific heat.

Transport properties are usually more difficult to determine than thermodynamic

properties and are associated with the molecular nature of the material; these
 42 Essentials of Micro- and Nanofluidics

properties are generally obtained experimentally, although it should be mentioned

that for a gas, the values of these properties may be obtained using the kinetic
theory of gases, at least for viscosity, thermal conductivity, and the diffusion
coefficient.

Notation alert: In the physical science literature, the viscosity is often denoted
by η. In this text, μ will be used for viscosity, according to conventional notation
in the engineering community.
The transport properties are defined in terms of the dependent variables in
specific ways, and some of these definitions are surprisingly similar in form.
This was evident in the historical development of these formulas, as discussed in
Section 1.6. In one dimension, for which the dependent variables vary only in
the direction normal to a wall such as that in a fully developed channel flow or
boundary layer, the shear stress for a Newtonian fluid is defined as the derivative
of the velocity times the viscosity, or
du
τ =μ (2.1)
dy
where τ is the shear stress, u is the velocity, and y is the coordinate normal to the
wall. The units of shear stress are force per area, or N/m2 , and so the viscosity
has units of N sec/m2 . The derivative of the velocity on the right side of equation
(2.1) is called the rate of strain. The direction of action of the shear stress is
depicted in Figure 2.1. Analogous to the viscosity μ, the thermal conductivity k
is the proportionality constant in Fourier’s law of heat conduction:
dT
q = −k (2.2)
dy

du T
τ=μ dy w2

y q=−k dT
dy

T
w1
σ=ε dφ
c w2 −
e −−−−
− − − − dy − − − −

dc
J=−D dy

− − − − − − − − − − − − −
c
w1 σ=−ε dφ
e dy

Definition of transport properties in one dimension. Note the similarity of the formulas. The surface charge
Figure 2.1
density σ is defined at the charged walls of a channel.
43 Preparatory Concepts

where q is the heat flux in W/m2 , where W denotes wat t = 1 Joule/sec, and T is
the temperature. The unit of thermal conductivity is W/m◦ K. As is the case with
Newton’s law, Fourier’s law is an empirical law.
The diffusion coefficient of species A into B is defined by Fick’s law, and on
a molar basis,
dc A
J A = −D AB (2.3)
dy
where J A is the flux of species A, c A is the local concentration of species A,
and D AB is the diffusion coefficient. The units of J A are mole/m2 sec if the
concentration is given in mole/m3 . Again, as with Fourier’s law, Fick’s law is an
empirical relationship, developed by measuring a known flux as a function of a
known concentration gradient. Fick’s law on a mass basis is given by
dρ A
j A = −D AB (2.4)
dy

where ρ A is the mass density and the units of j A are kg/m2 sec. Generally,
engineers prefer mass units, whereas chemists and chemical engineers prefer
molar units. The units of the diffusion coefficient are m2 /sec.
Specific to the field of mass transfer, several scales are associated with
molar concentration. Typically, chemists work in molar, designated by 1M = 1
mole/liter. The mole fraction , X A = c A /c, is a dimensionless number; here c is
the total concentration of the mixture, defined for a binary mixture as c = c A + c B .
Often the total concentration remains nearly constant, and in this case, Fick’s law
can be written as
dXA
J A = −cD AB (2.5)
dy
The advantage of working with mass or mole fractions is that both quantities
are dimensionless, although equation (2.5) assumes that the total concentration
is approximately constant. Similar comments apply to the definition of the mass
fraction, ω A = ρ A /ρ, where ρ is the total density.

Small uncharged Large polar Pumping molecules

Gases polar molecules molecules lons against their gradient

CO2 ,O2 Ethanol, Urea H2O Glucose Ca2+, Mg2+, Cl-

K+
extracellular

cytosol
ATP ADP + P
Na+

Passive Facilitated Active Transport

Diffusion Diffusion

Mass diffusion takes place across the cell membrane, which is porous to gases and small, uncharged polar
molecules. A molecule is classified as polar if the locations of the positive and negative charges do not coincide.
Water is a polar molecule. From Saltzman (2009).
44 Essentials of Micro- and Nanofluidics

Table 2.1. Units for the transport properties and their corresponding
dependent and independent variables
Property Units DQ Units DV Units
μ Nsec
m2
τ N
m2
u m
sec

W W
wk mK
q m2
T K
m2 mole mole
DAB sec
JA (mole) m2 sec
cA m3

m2 kg
DAB sec
j A (mass) kg
m2 sec
ρA m3

C2
e Nm2
σ C
m2
φ volt = J
Coul

DQ = defined quantity; DV = dependent variable.

Finally, the electrical permittivity is defined in terms of the surface charge

density σ and the electrical potential φ:
dφ
σ = e (2.6)
dy
where the outward unit normal to the surface is in the positive direction. If the
outward unit normal to the surface is in the negative direction, a negative sign is
required.
The surface charge density has units of C/m2 if the potential is in volts. While
the preceding equations serve as definitions of the viscosity, thermal conductivity,
and diffusion coefficient and are empirical laws, equation (2.6) is obtained from
Gauss’s law and the fact that the electric field is irrotational: ∇ × E = 0.
A summary of the one-dimensional relationships complete with units is given in
Table 2.1.
It should be noted here that for multidimensional problems, there are differ-
ences in the forms of the preceding definitions of the transport properties. For
example, because stress is a tensor, there are six distinct components of stress
when symmetry is accounted for. Moreover, the heat and mass fluxes are vectors,
and the surface charge density is a scalar. All of these transport properties are
functions of temperature.
The magnitude of the transport properties for a liquid and a gas indicates the
different transport mechanisms of momentum, heat, mass, and ease of transport-
ing electricity. Table 2.2 indicates that the viscosity is much larger in a liquid
than in a gas. Thus shear forces are expected to be much larger in a liquid than
in a gas for the same velocity gradient. Likewise, liquids conduct energy more
efficiently than gases. Conversely, mass transfer is more efficient in gases with

Table 2.2. Approximate magnitudes of the primary

transport parameters for liquids relative to gases.
μL kL eL DL
μG kG eG DG

10 to 100 10 to 100 10 to 100 10−4

 45 Preparatory Concepts

a similar concentration difference, and the electric field is inversely proportional

to the relative permittivity.

2.3 Determining transport properties

The calculation of transport properties is described in this section. There are

several ways to determine these values: first, they can be measured directly or
indirectly, as in a parallel plate viscometer for measuring viscosity. For a gas,
these transport properties can also be calculated theoretically using the kinetic
theory of gases. However, for liquids, there is no such unified theory, and most
often, the properties are measured; also, molecular simulations have been used
for this purpose. An extensive collection of calculation methods is given by Latini
et al. (2006).
Theories have also been developed for liquid mixtures; however, for the dilute
solutions of interest in this book, the transport properties will be nearly the values
appropriate for the solvent, most often water.
The purpose of this section is to present several of the most important formulas
used to calculate each of the four transport properties, with the focus being on
water. The best source of this kind of information is the open literature in the
form of archival papers in which new results for known and novel materials
are published. Much in the way of transport properties also appears in the CRC
Handbook of Chemistry and Physics (Haynes, 2011–2012).
It should be pointed out that all of the methods for empirically predicting
transport coefficients are only as accurate as the experimental data used. In
particular, all of the schemes used to predict diffusion coefficients in liquids
show large fluctuations in percentage error when compared with actual data
(Reid et al., 1987).

2.3.1 Viscosity
Viscosity is especially important in internal flows considered in this book, and
so it is useful to introduce this transport property here. It is the viscosity that
forces fluid to stick to walls of a channel or tube: the no-slip condition. Viscosity
of corn syrup, oils, and blood is relatively large, while the viscosity of water is
somewhat low, and the viscosity of air is smaller still. Viscosity expresses the
effect of molecular motions and interactions on the macroscale. For a perfect
gas, an equation to predict viscosity exists (Reid et al., 1987) in terms of particle
diameter and molecular weight. The viscosity of air is μ ∼ 10−5 N sec/m2 while
the viscosity of water is μ ∼ 10−3 N sec/m2 at room temperature.
Consider the flow between two plates, and suppose that the upper plate is
moving at a small speed u, as in Figure 2.2, and that the fluid velocity is
independent of x; this fluid dynamics problem is called Couette flow, and such
a device is called a parallel plate viscometer. After a small time, an initially
 46 Essentials of Micro- and Nanofluidics

SHEAR STRESS IS RELATED TO VELOCITY

Δl
FE at t=t 0 ΔF

τ Δu
yx

Y Δy FE at t=t +Δ t
Δα 0

X
Δx

Geometry of Couette flow; this device is called a parallel plate viscometer. FE = fluid element; the solid line
Figure 2.2
indicates the FE at time t0 .

undeformed fluid cube will undergo a shearing force, as shown in Figure 2.2. The
shear stress is the force per unit area on the upper plate and is defined by
F
τ = lim (2.7)
A→0 A

where A is the area over which the force acts. For Newtonian fluids, the shear
stress is proportional to the rate of strain, which is
α
γ = lim (2.8)
t→0 t
To determine γ , note that from the geometry

l = ut (2.9)

and for small angles α,

y sin(α) ∼ yα = l (2.10)

Substituting for l, it is evident that

α u
= (2.11)
t y
Thus in the limit as y → 0 and t → 0
du
τ∝ (2.12)
dy
The proportionality constant is the viscosity μ first proposed by Stokes and leads
to the definition of the shear stress for a Newtonian fluid discussed previously.
The viscosity can be calculated using the solution for the velocity of Couette
flow (Figure 2.3). As will be seen, the solution for the velocity is
y
u(y) = U (2.13)
h
Suppose the velocity of the plate is U = 0.01 m/sec, the channel height h =
0.1 cm, the force on the upper plate is measured to be 10−6 N, and the width
 47 Preparatory Concepts

τ U=0.01 m/sec

u(y)
h=0.1 cm
y

X τ
Δx

Calculation of the viscosity in a parallel plate viscometer.

Figure 2.3

and length of the cell are L = W = 10 cm. To calculate the viscosity given this
information, from the definition of the shear stress
τh
μ= (2.14)
U
the shear stress is force per area so that
10−6 N N
τ= = 10−2 2 (2.15)
0.01 m × 0.01 m m
Thus
10−2 mN2 × 10−3 m N sec
μ= m = 10−3 2 (2.16)
0.01 sec m
The viscosity is sometimes expressed in Poise and 1 N sec/m2 = 10 P, where P
stands for Poise.
The direction of the stress is important. On the upper wall, the force (stress ×
area) on the fluid is causing motion of the fluid, and so the stress is in the positive
x direction. The bottom plate is preventing motion of the fluid, and so the force
on the fluid is in the negative x direction. More will be said about the direction
of stress in Chapter 3.
The viscosity of liquid water is a function of temperature, and White (2006)
suggests that
 2
μ T0 T0
ln =a+b +c (2.17)
μ0 T T
where T0 = 273 K and μ0 = 0.00179 N sec/m2 . The values of the constants are
(White, 2006) a = −2.10, b = −4.45, and c = 6.55. According to White (2006),
the accuracy of this fit is ±1%. A table of values of the viscosity of water as a
function of temperature is presented in Table 2.3. The viscosity of liquids is a
very weak function of pressure, which is almost always neglected. Values for the
viscosities of several common solvents are depicted on Table 2.4.
Using the values in Table 2.4, we note that the stress exerted on the wall of a
channel is about 5 times greater for oil than it is for water and 5000 times greater
48 Essentials of Micro- and Nanofluidics

Table 2.3. Viscosity of liquid water as a

function of temperature
Temperature Viscosity
◦
T ( C) μ ( Nmsec
2 )

0 1.787 × 10−3
20 1.0019 × 10−3
40 0.6530 × 10−3
60 0.4665 × 10−3
80 0.3548 × 10−3
100 0.2821 × 10−3

From Hardy and Cottingham (1949)

than for water vapor. For example, in a h = 10 µm channel for fluid moving
at U = 1 cm/sec, the shear stress exerted by oil on one wall of the channel is
approximately
1 cm/sec
τ = 4.6 × 10−3 × = 4.6 sec−1 (2.18)
10 × 10−6 m
The electrolyte solutions considered here are dilute, containing at most about
1 M of the electrolyte. Thus, in most of the applications described here, use
of the viscosity of water for the viscosity of aqueous solutions does not lead
to significant error; this is most often the case in micro- and nanofluidics. Of
course, for complex fluids, such as blood, which often exhibits non-Newtonian
behavior, this is not the case, and the viscosity will depend, for example, on the
concentration of red and white blood cells. More will be said about this when
non-Newtonian fluids are discussed in Chapter 3.

2.3.2 Diffusion coefficient

The closest approach to a theory of liquid diffusion coefficient was developed by
Stokes in 1850 and Einstein in 1905 (Plawski, 2001). They considered the limiting

Table 2.4. Values of viscosity of selected liquids and water vapor

at the indicated temperatures
Fluid Temperature Viscosity
T (K) μ ( Nmsec
2 )

Water (l) 300 0.855 × 10−3

Water (g) 300 9.09 × 10−6
Seawater 294 1.06 × 10−3
Engine oil (unused) 300 4.86 × 10−3
Methanol 293 0.59 × 10−5
 49 Preparatory Concepts

Model for calculating the diffusivity of a liquid due to Stokes in 1850 and Einstein in 1905. The solute must be
Figure 2.4
much larger than the solvent molecule. Here a human serum albumin molecule whose radius is about 3.5 nm
is depicted surrounded by water molecules of radius about 10 times smaller.

case where a large particle is diffusing through a cloud of smaller particles.

This situation is depicted in Figure 2.4. In the Stokes–Einstein picture, a tagged
molecule A passes through a sea of smaller particles randomly. They assumed
that the force is proportional to the velocity of species A relative to the mass
averaged velocity of the mixture:

F = f (u A − u) (2.19)

with f being a proportionality constant. There are two cases: in one, the diffusing
particle sticks to the fluid molecules, and in the second, it slips. For these cases,
the value of f can be shown to be

f = 6πμa no slip (stick) (2.20)

f = 4πμa slip (2.21)

where a is the radius of the diffusing particle. These values of f represent

a coefficient in the Stokes drag law multiplying the relative velocity. Einstein
defined the force as a gradient in chemical potential; the chemical potential will
be defined in Chapter 7, but for now, we define the chemical potential in terms of
the number density n A (number of molecules of A per volume) as

μ A = μ0A + kb T ln n A (2.22)

where kb = R/N A = 1.38 × 10−23 J/K is Boltzmann’s constant, R is the universal

gas constant, and N A is Avagadro’s number. The chemical potential has units of
energy, and so its gradient will have units of force. Thus Einstein wrote the drag
force F as the gradient in chemical potential:

F = ∇μ A (2.23)

so that in one dimension,

dμ A −kb T dn A
F =− = (2.24)
dx nA dx
50 Essentials of Micro- and Nanofluidics

In Section 3.7, we will see that

D AB dc A D AB dn A
uA − u = − =− (2.25)
cA d x nA dx
where the last equality uses the fact that n A = N A c A . Equating the two expressions
for the force leads to
kb T
D AB = (2.26)
βaμ
where β = 6π for no slip and β = 4π for slip between molecules. Typically, this
equation predicts the diffusion coefficient with an accuracy of 20%.

Albert Einstein and the Diffusion Coefficient

So how did Einstein get his name on the Stokes–Einstein equation? Because he
knew a little fluid mechanics and mass transfer! Einstein’s PhD Dissertation, “A
New Determination of Molecular Dimensions,” was published in the same year,
1905, that he published his two papers on the theory of relativity. Einstein was
concerned with the determination of the volume of sugar that dissolves in water,
so that he could determine if the sugar molecules were hydrated (Stachel, 1998).
While we have viewed the Stokes–Einstein equation as determining the diffusion
coefficient, Einstein thought of the same equation as determining the molecular
size given the diffusion coefficdient! In his dissertation, Einstein’s object was
to show that “. . . sizes of molecules of substances dissolved in an undissociated
dilute solution can be determined from the internal viscosity of the solution and
of the pure solvent, and from the diffusion rate of the solute within the solvent
provided that the volume of the solute molecule is large compared to the solvent
molecule.” And “. . . such a molecule will behave appproximately like a solid body
suspended in the solvent”. Stokes drag law applies! Einstein expressed his result
in the formula
RT 1
NP =
6π k D
in which N is Avagadro’s number, P is the radius of the solute, k is the viscosity,
and D the diffusion coefficient. This part of his dissertation was published in
the paper Einstein (1905b). All five of the papers published in 1905 by Einstein
including the two on the theory of relativity are lucidly discussed in the short
book by Stachel (1998). The quotes above are taken from that book.

For the general case that lies between perfect slip and perfect no-slip, the
diffusion coefficient is given by
 
3μ B + aβ AB kb T
D AB = (2.27)
2μ B + aβ AB 6πμ B a
based on a result for the drag force given by Lamb (1945). Here β AB is the
coefficient of sliding friction of the solute molecule.
The Stokes–Einstein equation has been shown to be fairly accurate for describ-
ing the diffusion of large spherical particles or molecules in cases where the
solvent appears to the diffusing species as a continuum (Bird et al., 2002). The
51 Preparatory Concepts

hydrodynamic theory further suggests that the shape of the diffusing species is
very important. Tyrrell and Harris (1984) note that the validity of equation (2.26)
is also restricted by the fact that the motion of only one diffusing particle is
considered, and so the relationship can strictly only apply in the limit of infinite
dilution, that is, in the limit of zero solute
Table 2.5. Diffusion coefficients at infinite concentration. Values of diffusion coeffi-
dilution (i.e., very low concentration) for
cients of several neutral solutes in water are
some common fluids in water at
◦
T = 25 C. From Cussler (1997) given in Table 2.5.
Solute D A (cm2 /sec)
The diffusion coefficient D AB in gases
D AB ∼ 10−1 cm2 /sec and in liquids D AB ∼
Air 2.00 × 10−5
10−5 cm2 /sec. Biomolecules in aqueous
−5
Nitrogen 1.88 × 10 solution have diffusion coefficients that are
−5
Oxygen 2.10 × 10 smaller due to their relatively large size,
−5 especially in nanoscale channels; typically,
Methanol 0.84 × 10
Ammonia 1.64 × 10−5 D AB ∼ 10−6 − 10−7 cm2 /sec.
While the Stokes–Einstein equation is a
simple and useful tool for estimating dif-
fusion coefficients for very dilute mixtures, more accurate methods have been
established. An empirical modification of the Stokes–Einstein equation is the
Wilke–Chang correlation (Bird et al., 2002)

7.4 × 10−8 (ψ B M B )1/2 T

D AB = (2.28)
μ B V A0.6

valid at very low concentration; here B is the solvent, the species with the highest
concentration. In this equation, ψ B is an association factor, M B is the molecular
weight, V A is the partial molar volume, and μ is the viscosity of the solvent in
centipoise (1 cP = 10−2 P). Wilke and Chang suggest that ψ B = 2.6 for water as
the solvent.
Many other correlations have been developed over the years, with widely
varying accuracy. Several others are discussed in Bird et al. (2002) and Reid
et al. (1987), and many other diffusion coefficients are given in Reid et al. (1987).
While these correlations are often useful, it is often easier to use a measured value
of liquid diffusivity, and diffusivities for several ions and biomolecules in water
are found in Table 2.6. The hydrated radii are usually used in the Stokes–Einstein
relation.
Ficoll and dextran are part of a class of hydrophilic polymers that are often
used to simulate the action of proteins, specifically in the study of transport in
the kidney (Fissell et al., 2007); dextran is a neutral glucose-linked polymer, and
Ficolls are sucrose-linked polymers. Both Ficoll and dextran occur in a variety of
molecular sizes, and the effective radii for Ficoll and dextran often fall between
the Stokes–Einstein value and the hydrated value; the radii can be related to the
molecular weight, as described in Section 3.7.
Biofluids are multicomponent mixtures, and the process of predicting diffu-
sion coefficients becomes much more difficult. When all the components of the
52 Essentials of Micro- and Nanofluidics

Table 2.6. Typical molecular and hydrated radii and diffusion coefficients in bulk water at room
temperature (25◦ C). All diffusion coefficients should be multipled by 10−5
Solute D A (cm2 /sec) Molecular radius (nm) Hydrated radius (nm)
Ions
Li+ 1.03∗ 0.094 0.382
+ ∗
Na 1.33 0.117 0.358
+ ∗
K 1.96 0.149 0.331
+ ∗
Cs 2.06 0.186 0.329
2+ ∗
Mg 0.71 0.072 0.428
2+ ∗
Ca 0.79 0.100 0.412
Cl− 2.032∗ 0.164 0.332
Biomolecules
Albumin 0.061∗∗ – 7.2
∗∗∗
Glucose 0.94 – 1
∗∗
IgG 0.04 – 10
Ficoll 5–10 – 4.81–2.45
Dextran 0.009–0.1 – 27.0–2.36

The asterisk symbol denotes data from Cussler (1997), as calculated from the data of Robinson and
Stokes (1959); double asterisks are from Granicka et al. (2003), and triple asterisks are from the Web site
http://oto.wustl.edu/cochlea/model/diffcoef.htm. Molecular and hydrated radii are from Volkov et al. (1997).

mixture are dilute, a good approximation to the diffusivity is to use the binary
diffusivity of the given component in the solvent.
As an example, using the Stokes–Einstein relation, with the molecular diameter
in Table 2.6, the diffusion coefficient for a sodium ion in water at T = 25 C is
given by D AB = 1.86 × 10−5 cm2 /sec, a full 40 percent higher than in Table 2.6.
However, the molecular radius may not be the appropriate length scale. Using the
hydrated radius a = 0.358 nm gives D AB = 0.6 × 10−5 cm2 /sec, which is much
too low. The average of these two values gives D AB = 1.23 × 10−5 cm2 /sec,
which is close to the experimental value. Thus there is some question as to which
radius to use, and care should be used in the interpretation of the Stokes–Einstein
relation for the diffusion coefficient for ionic species in particular. Note that the
influence of valence is incorporated implicitly through the use of the experimental
value of the radii. More puzzling is the fact that data from Keilland (1937) suggest
that the hydrated radius of sodium be taken as a = 0.225 nm, which, when used,
gives a result, again, close to the experimental value. Other values for the hydrated
and ionic radii are given by Jorgensen (1990).

2.3.3 Thermal conductivity

The thermal conductivity is the proportionality constant in Fourier’s law and has
units of W/m◦ K. In general, the thermal conductivity of gases is much lower
53 Preparatory Concepts

Zinc Silver
Pure metals
Nickel Aluminum
Alloys
Plastics Ice Oxides
Nonmetallic solids
Foams Fibers
Insulation systems
Oils Water Mercury
Carbon Liquids
dioxide Hydrogen
Gases

0.01 0.1 1 10 100 1000

Thermal conductivity (W/m.K)

Typical values of the thermal conductivity for a variety of materials. From Turns (2006)

than that for liquids. Solids have the highest conductivity; quartz has a thermal
conductivity of 7.6 W/m◦ K at T = 20◦ C, while glass has a conductivity of
0.76 W/m◦ K at T = 20◦ C. Homogenous materials have a thermal conductivity
which independent of direction.1
Because of the higher thermal conductivity, solids are better thermal conduc-
tors than liquids, which are better conductors than gases. Note that a crystalline
structured solid such as quartz has a higher conductivity than glass, whose phys-
ical structure is amorphous; that is, it is not densely ordered and behaves more
like a liquid. This is the opposite of the diffusion coefficient, which is highest in
gases and lowest in solids. Metals have the highest thermal conductivity. As with
the diffusion coefficient, there is a well-developed kinetic theory to predict the
thermal conductivity of gases. Because of the magnitudes of the solid thermal
conductivity, solid surfaces control the amount of heat transfer through a given
system.
In general, the thermal conductivity of liquids decreases with temperature,
although water is an exception to this behavior, increasing with temperature.
Water has a relatively high thermal conductivity compared with other liquids,
reaching a maximum of 0.7 W/m◦ K at T = 150◦ C. Typical values of the thermal
conductivity appear in Table 2.7.
The temperature dependence of many liquids can be described by (Reid et al.,
1987)

k(T ) = A + BT + C T 2 (2.29)

where A, B, C are constants. For water, A = −0.383 W/m◦ K, B = 5.254 × 10−3

W/m◦ K2 , and C = −6.369 × 10−6 W/m◦ K3 . These constants have been tabulated
for many other liquids, and an extensive table is given by Reid et al. (1987), along
with a discussion of many other methods of estimation. Latini et al. (2006) also
has a significant amount of information on the calculation of thermal conductivity.

1
However, fibrous materials such as wood and laminates have different conductivities in different
coordinate directions.
54 Essentials of Micro- and Nanofluidics

Table 2.7. Thermal conductivity of liquid

water as a function of temperature
Temperature Thermal Conductivity
◦
W
T ( C) k mK
0 0.5619
20 0.5996
40 0.6286
60 0.6507
80 0.6668
100 0.6775

From Kestin (1978).

For restricted ranges of temperature, the thermal conductivity is often represented

by
k = k0 (1 + bT ) (2.30)
with T in ◦ C and where k0 and b are constants.

2.3.4 Electrical permittivity

The electrical permittivity r is a transport property like viscosity and, for a
vacuum, is defined by
 
−2
−7 N sec 1
2
c0 10 2
= (2.31)
C 4π0

where c0 = 3 × 108 m/sec is the speed of light in a vacuum. Thus the permittivity
in a vacuum is
1 C2
× 10−9
0 = (2.32)
36π Nm2
If the medium is not a vacuum, a relative permittivity or dielectric constant is
defined by
e
r = (2.33)
0
where e is the material permittivity. Materials that have a relative permittivity
greater than one are called dielectrics, and selected permittivity values are given
in Table 2.8.
The relative permittivity is a function of temperature, and the permittivity of
water decreases with increasing temperature in the form
r = Ae− B
T
(2.34)
Knox and McCormack (1994) suggest that A = 305.7 and B = 219, and a table
of values found in the Handbook of Chemistry (Haynes, 2011–2012) appears in
55 Preparatory Concepts

Table 2.8. Relative permittivity or dielectric

constant of some materials at room
temperature (Israelachvili, 1992)
Material r
Vacuum 1.0
Air 1.000549
Water 78.54
Methanol 33
Salt water 81
Ethylene glycol 40.7
Glass 4–10
Quartz (SiO2 ) 4.5
Polystyrene 2.4
Sodium chloride (NaCl) 6.0

High relativity permeability indicates a reduced

electric field strength.

Table 2.9. Using the Knox and McCormack (1994) formula, r = 78.4, which is
consistent with the tabular values.
Another common curve fit for the dielectric constant or relative permittivity
of water is the formula developed by Archer and Wang (1990):

r = 87.86 − 0.3963T + 7.036 × 10−4 T 2 (2.35)

where the temperature is in ◦ C.

2.3.5 Surface tension and wettability

When a liquid interfaces with a gas, say, water with air, a liquid–gas interface is
formed, that acts as if it were a stretched membrane and a net pressure change
across the surface is present owing to the lack of water molecules at the liquid–gas
interface. Consider the simple case of a curved surface depicted in Figure 2.5. The

Table 2.9. Relative permittivity of water

as a function of temperature
(Haynes, 2011–2012)
Temperature (◦ C) r
0 87.74
10 83.83
20 80.10
30 76.55
40 73.15
 56 Essentials of Micro- and Nanofluidics

molecules in the liquid are held together by

van der Waals forces in a nonpolar liquid
and by the short-range hydrogen bonds in
a polar liquid like water. These attractive
forces are of a different nature from the
electrostatic forces that hold the protons,
neutrons, and electrons together to form an
atom. Thus there is a natural tendency for
the surface to contract, and this is done by
the surface tension force pulling down on
Side view of a liquid drop in a gas.
Figure 2.5 the drop, as shown in Figure 2.5. The pres-
sure inside the drop acts over the drop cross-sectional area π R 2 , and the surface
tension pulls the drop in the opposite direction around the circumference. Then
the net pressure change across the surface is balanced by the tension caused by the
significant change in density between the liquid and the gas, and for a spherical
drop, this “surface tension” is defined, in the absence of any flow, by the force
balance
γ
p = 2 (2.36)
R
where γ is the surface tension and has units of N/m and R is the radius of
the drop. At T = 20◦ C, the surface tension coefficient for a clean air–water
interface is γ = 0.073 N/m. For a bubble, a gas inside surrounded by a liquid, the
result is
γ
p = 4 (2.37)
R
since the liquid bubble surface has two interfaces with the gas.
Finally, for a general curved surface with radii of curvature in each of the two
coordinate directions, we have
γ
p = 2 (2.38)
R 1 + R2
These equations relating the pressure to the surface tension are collectively called
the Laplace equations.
Note that surface tension has units of force per length, and so the free energy
associated with a given curved surface is given by E = γ A, where A is the area
of the surface. By the first law of thermodynamics, the energy must also be equal
to the work done by the surface tension force W = E = γ A (Batchelor, 1967).
A free and curved surface is a surface of minimum energy, and the relationships
between the pressure and curvature of the surface shown earlier can be obtained
by energy considerations as well. Consider the spherical drop given previously;
the total energy of the surface is given by the difference in pressure and the
surface tension:

d E = −( pin − pout )d V + γ d A = −4π R 2 ( pin − pout )d R + 8π Rγ d R (2.39)

57 Preparatory Concepts

Table 2.10. Surface tension for water as a

function of temperature (Karniadakis et al.,
2005)
Temperature (◦ C) γ mN
m

10 74.2
20 72.8
30 71.2
40 69.6

The term mN = 10−3 N.

and the minimum energy is obtained by setting d E/d R = 0 so that

γ
pin − pout = 2 (2.40)
R
as obtained earlier.
In the preceding equations, the influence of gravity is neglected. The charac-
√
teristic length scale associated with gravity is l g = γ /ρg, which, for water at
room temperature, is lg = 2.7 mm. The effect of gravity will then be negligible if
the radius of the drop (or bubble) is much less than lg . The dimensionless number
Bo = D/l g is called the Bond number, where D is the diameter of the bubble,
and gravity will be negligible if Bo  1.
Surface tension decreases with increasing temperature, and some common
values for water are shown in Table 2.10. White (2006) suggests that a good
curve fit to these data is
γ (N/m) = 0.076 − 0.00017T (◦ C) (2.41)
The radius of curvature of a sphere is its radius, R. For a general curved surface
of the form
z = h(x, y) (2.42)
there are two principal radii of curvature in each of the coordinate directions;
Wehausen and Laitone (1960) provide the detailed expression for the two principal
radii of curvature, and for a curve of the form y = h(x), the methods of differential
geometry indicate that the radius of curvature is given by
  2 3/2
1 + dh dx
R= d2h
(2.43)
dx2

In particular, if the slope of the surface is small, dh/d x  1, then

1
R= d2h
(2.44)
dx2

In differential form, the equations for the pressure difference become

 
dp d 1 d 3h
=γ =γ 3 (2.45)
dx dx R dx
 58 Essentials of Micro- and Nanofluidics

Side view of a liquid drop in a gas, indicating the contact angle for a wetting surface (top) and a nonwetting
Figure 2.6
surface (bottom).

There are two dimensionless numbers associated with surface tension, γ . The
first is the Weber number, defined by W e = ρU 2 L/γ , and is a ratio between
inertial forces and surface tension forces; here U is a characteristic velocity. The
other is the Capillary number, Ca = μV /γ , and is the ratio between the viscous
forces and the surface tension forces.
The concept of surface tension is intimately connected with and determines a
fluid’s ability to wet a surface. A contact angle is defined by the angle between the
liquid and the surface, as depicted in Figure 2.6. If the contact angle θc < 90◦ , the
surface is said to be partially wetted; conversely, if the contact angle θc > 90◦ ,
the surface is said to be nonwetting. If the liquid is water, a partially wetted or
fully wetted surface (θc = 0) is called hydrophilic, while a nonwetting surface is
called hydrophobic.
Near the contact line, the location where the bubble meets the solid surface,
local equilibrium is characterized by three local surface tension coefficients,
associated with the contact between the solid and liquid (SL), the solid and gas
(SG), and the liquid and gas (LG). A force balance on the contact line in the
horizontal direction under equilibrium conditions reveals that (Figure 2.7)

γSG = γLS + γLG cosθC (2.46)

This is called Young’s law and is a simple first approximation of the macroscopic
view of the microscopic region near the contact line. There are several problems
with this equation since, assuming that the contact angle can be measured by
some means, there are still three unknowns and one equation. Thus at least two
of the surface tension coefficients should be measured; in Young’s equation, γ S L
is often termed the interfacial tension, and values for a number of solvents have
been measured and compiled in Israelachvili (1992).

γ
LG
θc
γ γ SG
SL

Surface tension coefficients for deriving Young’s equation.

Figure 2.7
 59 Preparatory Concepts

Unlike the other properties, such as viscosity and thermal conductivity, the
surface tension of a given liquid can be altered substantially by the addition
of a small amount of surfactant. Surface active agents, or surfactants, are long
molecules having a hydrophilic head and a hydrophobic tail. They are often added
to biofluids to prevent aggregation and may be used to prevent target molecules
from adhering to surfaces. In general, surfactants reduce the surface tension of
water.
Models for the surface tension of liquid mixtures have been proposed over
the years, and a popular formula for binary solutions is given by Shereshefsky
(1967), according to
G s
γ0 X 2 e RT
γ = γ1 −  G s (2.47)
1 + X 2 e RT − 1

where γ1 is the surface tension of the pure solvent, X 2 is the mole fraction of the
solute in the bulk, and G s is the Gibbs free energy change in replacing 1 M
of solvent with 1 M of solute in the interfacial region. Values of G s /RT for a
number of nonelectrolyte mixtures have been compiled by Tahery et al. (2005)
from work published in the open literature. For example, G s /RT = 1.98 for a
water–methanol mixture at T = 25◦ C. A molecular model built specifically for
polar liquid mixtures has been given by Li and Liu (2001), and for example,
the surface tension of a methanol–water mixture for a methanol mole fraction
of X 2 = 0.2 is calculated to be γ = 40 mN/m. The surface tension for aqueous
electrolyte solutions has been presented by Levin and Flores-Mena (2001) based
on the analysis of the Poisson–Boltzman equation of electrostatics; for very low
concentrations, the surface tension can be predicted by the Onsager and Samaras
limiting law (Onsager & Samaras, 1934). In general, electrolyte solutions exhibit
surface tensions higher than pure water.
Surface tension phenomena are essential features in a number of applications,
including inkjet printers, electrowetting on dielectric (EWOD), thermocapillary
pumping, instabilities of falling films, coating flows, and Marangoni phenomena.
The latter refers to the currents set up when the surface tension varies with
temperature. Thermocapillary pumping will be discussed later in the book.

2.4 Classification of fluid flows

The field of macroscale fluid mechanics can be classified in a number of ways.

As has been discussed, the most important property of fluids is their viscosity,
which is characteristic of a fluid’s ability to stick to walls. A fluid in motion away
from the influence of any solid boundaries tends to act as if its viscosity were
zero. In this case, we call the flow inviscid. Conversely, fluids that move near solid
boundaries are flows that are termed viscous, based on the fluid’s finite viscosity.
In general, most flows of interest in micro- and nanofluidics are internal, viscous,
 60 Essentials of Micro- and Nanofluidics

(a) (b)

Examples of macroscale external fluid flows.

Figure 2.8

incompressible laminar flows. The main dimensionless parameter that describes

viscous flow is the Reynolds number, which, for an internal flow, is defined by
ρU d
Re = (2.48)
μ
where ρ is the fluid density, μ is the fluid viscosity, and U and d are typical velocity
and length scales, respectively. For an internal flow or a flow bounded on all sides
by walls, the length scale D is the diameter for a tube, and for a rectangular
channel, its height h or the hydraulic diameter D H = 4hW /(2h + 2W ), where
W is its spanwise width. The velocity scale is usually, but not always, taken to be
the average velocity at a section, defined by
1
V = ud A (2.49)
A A

where A is the cross-sectional area and u is the velocity. The formal definition
of the volume flow rate is Q = U A and has units of Q ∼ m3 /sec; the mass flow
rate is defined by ṁ = ρ Q. For a very wide channel of height h, W  h, which
is common in micro- and nanofluidics; the average velocity is given by
1 h
V = ud A (2.50)
h 0

Fluid mechanics is all around us, from atmospheric flows which involve heat
and mass transfer and phase change to isothermal flows of water and air in pipes,
the basis of the field of plumbing and air-conditioning, to the aerodynamics of
cars and planes, and so on. In Figure 2.8 are two examples of such flows: the flow
past a twin rotor Boeing V22 helicopter and the flow field generated by a tornado.
A tornado is a region of very rapidly swirling fluid, with varying temperature
causing very high winds; such a highly swirling flow is called a vortex. The
prediction of our daily weather is largely a giant fluid dynamics problem coupled
with a variable temperature and mass transfer. There are also electrical effects
that occur in a rain storm: lightning.
 61 Preparatory Concepts

Classification of fluid flow problems. Micro- and nanofluidic flows are almost always laminar, viscous, incom-
Figure 2.9
pressible, internal flows. The symbol L denotes a typical length scale.

In the examples given here, the Reynolds number can be on the order 106 or
higher and is thus very large. Conversely, in many applications of interest in the
biological sciences, such as flows in small capillaries; in small microdevices for
analysis of biomolecules; and in sensing and detection of chemical and biological
species, the Reynolds number is often orders of magnitude less than 1.
In Figure 2.9 is a classification scheme designed for the flows of liquids from
the “small” macroscale down through the nanoscale. As can be seen, as the
length scale decreases from the macro- to the “large” nanoscale, the Navier–
Stokes equations generally apply for a liquid. However, for a gas, this picture is
significantly different, and molecular effects generally appear at the microscale.
Molecular effects in aqueous solutions begin to exert themselves at the “small”
nanoscale, which is about 20 water diameters or 6 nm (Zhu et al., 2005). Of
course, surface effects begin to exert themselves far above this value, and slip
flow may occur at hydrophobic walls.
Below the “small” nanoscale regime, molecular simulations, such as molecular
dynamics (MD) and nonequilibrium molecular dynamics (NEMD) (i.e., electro-
osmotic flow) and, to some extent, Direct Simulation Monte Carlo (DSMC), must
be used to describe the flow field. It is interesting to note that DSMC was first
conceived for rarefied external gas flows (Bird, 1994) but has begun to be used
for selected liquid nanoscale flows.
While the subject of fluid mechanics dominates this text, we will also describe
in detail the fundamentals of heat and mass transfer. Together these subjects
form two of the three disciplines that are termed the thermal sciences, and many
problems span all these individual disciplines. Thermodynamics, unlike fluid
mechanics and heat and mass transfer, is an equilibrium science and the third
rung of the thermal sciences, permeating through fluid mechanics in particular;
a brief introduction to thermodynamics is presented next.
 62 Essentials of Micro- and Nanofluidics

2.5 Elements of thermodynamics

Fluid mechanics, heat and mass transfer, and thermodynamics are inextrica-
bly linked. Fluid mechanics quantities such as pressure and density (or specific
volume v = 1/ρ) and the heat transfer property, temperature, are all thermo-
dynamic properties that describe the state of a system. Other thermodynamic
properties are internal energy, enthalpy, entropy, and the specific heats c p and cv .
In this section, a brief introduction to thermodynamics is presented, and in the
next section, it is shown how the loss of energy, or irreversibilities, affect flows
in pipes and channels.
Pressure is the macroscopic effect of particles within a fluid colliding with
themselves. Pressure has units of N/m2 and is easily measured by using a number
of devices (White, 2003; Munson et al., 2005). We usually speak of absolute and
gage pressure, which are related by pgage = pabsolute − patm . For example, a tire
gage measures gage pressure and adjusts the scale to the atmospheric pressure.
A perfect gas is one that satisfies the relation

p = ρ RT (2.51)

where R is the gas constant, p is pressure, and T is the temperature. A perfect

gas is one for which intermolecular forces (van der Waals forces) are negligible
and the molecules are assumed to have zero mass. The gas constant is gas
specific, and for air, R = 287 m2 /s2 K = 287 kJ/kgK. The universal gas constant
Ru = R/Mw = 8.314 kJ/k mole K, where Mw is the molecular weight; for air,
Mw = 28.97. The density of water is a weak function of temperature, and ρ ∼ =
1000 kg/m3 at normal room temperatures. For example, at 1 atm of pressure,
p = 1.01 × 105 N/m2 , and T = 300◦ K, the density of air is ρ = 1.16 kg/m3 .
The specific heat is defined as the amount of heat per mass needed to raise the
temperature of a material 1◦ C. The specific heat at constant pressure is defined
by
dh
cp = at constant pressure (2.52)
dT
where h = u + p/ρ is the enthalpy, and here, u is the internal energy.
Notation alert: In the equations defining the specific heats, h is the enthalpy,
defined by h = u + pv, and u is the internal energy and not the fluid velocity.
The specific heat at constant volume is defined by
du
cv = at constant volume (2.53)
dT
While the names of these two specific heats imply the opposite, these quantities are
defined for processes in which both the temperature and pressure can vary from
one value to another. This point is discussed in some detail in most undergraduate
 63 Preparatory Concepts

Cold water Hot water

in out

Vapor

Hot water
25 mm
Liquid heater

(a) (b)

Sketch of a closed system (a) and an open system (b).

Figure 2.10

thermodynamics texts (Moran & Shapiro, 2007). For an incompressible liquid,

c p = cv = c so that h = c p T = cv T = cT , and the units of specific heat are
kJ/kgK.
Notation alert: In this section, R denotes the gas constant for a specific gas in
units of kJ/kgK as opposed to the universal gas constant Ru that is in units of
kJ/kmole K. Later, however, because in electrokinetic phenomena, the universal
gas constant is used exclusively. The context should be clear.
Using the definition of the perfect gas, the specific heats can be related to the
gas constant by R = c p − cv so that

kR R
cp = and cv = (2.54)
k−1 k−1
where k = c p /cv and is constant over a wide range of temperatures.
In thermodynamics, several terms have very specific meanings. A system is an
arbitrary collection of matter, and the user defines the system of interest. There
are two types of systems:

1. Closed system: no mass flow is permitted across the boundaries of a closed

system; there may be energy flow across the boundaries of a closed system
(Figure 2.10a).
2. Open System: mass flow naturally flows across the boundaries of an open
system (Figure 2.10b).

In fluid dynamics, open systems are used almost exclusively. A sketch of these
types of systems is depicted in Figure 2.10.
Equilibrium is defined as the condition in which the characteristics or properties
of a system such as pressure and temperature are not rapidly changing with time.
Most often, these properties are independent of time.
 64 Essentials of Micro- and Nanofluidics

. As an example, suppose there is a gas con-

Rapid heating: Q
fined in a closed system. The box is initially at
T = T0 , and then suddenly, one end is heated,
as in Figure 2.11. What is the temperature of
Tair(x,t)
air the gas in the box for time t > 0? The system
in this case is not in equilibrium, and ther-
modynamic methods cannot be used to cal-
Figure 2.11
Illustration of a nonequilibrium process. culate the spatially varying temperature as a
Rapid heating causes the entire process to function of time. Heat transfer is the disci-
be nonequilibrium, nonquasistatic and thus
outside the realm of treatment by thermo-
pline that can answer this question. However,
dynamics. in many situations, if heat is added slowly and
for a long enough time, the temperature will
become independent of time and space, and
thermodynamics can predict the final temperature.

Terminology alert: In classical thermodynamics, the term property refers to

a quantity whose change between two equilibrium states is independent of the
process path. Thus enthalpy, internal energy, and entropy are examples of ther-
modynamic properties. Such terminology is normally not used in fluid mechanics
or heat and mass transfer.

A thermodynamic property is any quantity characteristic of a system, for exam-

ple, volume, mass, pressure, or temperature. Heat and work are not properties
but are processes that act to change the properties of the system. The state of a
system is the condition of a system described by its properties. For any property,
say, enthalpy h, the integral over a process path between states 1 and 2 is

2 1
dh = h 2 − h 1 = h = − dh (2.55)
1 2

However, for heat and work, which are not properties,

2 2
δq = q δw = w (2.56)
1 1

that is, the process of heat transfer and work being transferred to or from a system
depends on the specific way in which it was done.
One might ask the question, How many properties are required to fix the state
of a system? The answer is that it depends on whether the fluid is a gas or a liquid
and on the number of phases present in the system. For example, for a perfect
gas such as air, the pressure is p = ρ RT , where ρ is the density, R is the gas
constant, and T is the temperature so that two properties in general are required
to fix the third. For a liquid in equilibrium with its vapor, that is, in a saturated
state, only one property is required.
 65 Preparatory Concepts

A process is the act of changing a system from

Mvap
one equilibrium condition to another. A property
has no meaning thermodynamically unless equilib-
rium is established. When each stage of the pro-
cess is arbitrarily close to equilibrium, the process
Mliq
is called quasistatic. This is the ideal situation, and
it is noted that quasistatic processes do not often
occur in nature. Nevertheless, calculations are made
to approximate the real processes.
Mmix = Mvap + Mliq The core of thermodynamics are the first law of
thermodynamics or conservation of energy and the
One property, such as temperature, second law of thermodynamics. These two laws are
fixes the the pressure of the system.
used for a first estimate of the energy output (tur-
bine), the energy input (pump, compressor), and the
efficiency of a thermodynamic power or refrigeration cycle.
Consider the steady flow of a fluid through a control volume in which mass can
cross the boundaries: an open system, as shown in Figure 2.12. The first law of
thermodynamics is an expression of the conservation of energy over the control
volume and, in the absence of chemical reactions, is given by

q − w = e (2.57)

where q is the heat transfer from the system, w is the work output, and e is
the change in energy in fluid flowing out of the control volume, e = e2 − e1 =
eo − ei . Here the energy is given by

e = h + K E + P E (2.58)

where K E and P E are the change in kinetic energy and potential energy,
respectively. The units of all the terms in equation (2.57) are kJ/kg = J/g. The
first law of thermodynamics can also be written on a rate basis in the form

Q̇ − Ẇ = ṁe (2.59)

CV q w

1=i
V p
2
p1 h
2=o

L
Conservation of energy for a control volume with one inlet and one outlet. The directions of positive heat q and
Figure 2.12
work w are shown.
66 Essentials of Micro- and Nanofluidics

where here
V22 V2
e = h 2 + + gz 2 − h 1 + 1 + gz 1 (2.60)
2 2
In this formulation, work output from a turbine is a positive number.
The second law of thermodynamics is
δq
s ≥ (2.61)
T
or on a differential basis
δq
ds ≥ (2.62)
T
where s is the entropy in units of kJ/kg K and δq is used to denote that the heat
transfer q is not a property in the thermodynamic sense. In an ideal process with
no energy losses
δq
ds = (2.63)
T
and for no heat losses, ds = 0, and the process is isentropic. Under this condition,
the work output from a turbine, −Ẇ = ṁe, is a maximum. Equation (2.63)
holds for both closed and open systems.

Inlet Steam turbine

1 2 3
Extraction
locations Exit

A turbine is a work-producing device and the

maximum amount of work output is when the
process is isentropic.

Notation alert: In the following analysis of the second law, σ is used to denote
entropy production and not surface charge density.

Integrating equation (2.62),

2 δq
s2 − s1 = +σ (2.64)
1 T
where σ is called the entropy production and σ = 0 for a reversible process and
σ > 0 for an irreversible process. A reversible process is one for which there are
 67 Preparatory Concepts

The initial condition for the subsequent nonequilibrium process.

Figure 2.13

no frictional or other losses in the given system; an irreversible process, such as

steady viscous flow in a pipe or channel, is one for which there are such losses.
A process for which σ < 0 is impossible.
The magnitude of the entropy change determines the efficiency of a pro-
cess such as the power production in a typical power plant. An entropy change
also determines the direction in which a process occurs and the efficiencies of
machines such as pumps, compressors, and fans. For example, it is common
knowledge that if a partition separates two compartments of different pressures
in a closed system (Figure 2.13), for which no mass can pass its boundaries, and
the partition is broken, the final pressure p f satisfies

pL < p f < p H (2.65)

The reverse process is never observed to occur spontaneously because energy is
required to make it happen. In the same situation as with the pressure, the same
is true for heat transfer,
TL < T f < T H (2.66)
if the system is insulated or q = 0. These two common facts are a manifestation
of the second law of thermodynamics. In both cases, if both tanks are insulated,
the entropy is constant: s = 0.
The specific heats and entropy are related through the T ds equations, which
are obtained by combining the first and second laws of thermodynamics. For a
closed system, there are no kinetic and potential energy changes, and the first law
reads
δq − δw = du (2.67)
In a closed system, the only work that can be done is boundary work δw = pdv,
and so combining with the second law
T ds = du + pdv = dh − vd p (2.68)
where the second equals sign is obtained using the definition of enthalpy. The
T ds equations have been derived for a closed system, however, since all of the
 68 Essentials of Micro- and Nanofluidics

CV
Fsx
1=i
V p
2
p1 h
2=o
Fsx

Conservation of linear momentum for a control volume with one inlet and one outlet. The number 1 denotes
Figure 2.14
the inlet and 2 the outlet.

variables in equations (2.68) are thermodynamic variables and describe the state
of the system, they are valid for open systems as well. Equation (2.68) can be
integrated for constant specific volume to obtain the entropy change, and
T2
s = c ln (2.69)
T1
for a liquid undergoing a constant pressure process and where c p = cv = c is the
specific heat. Thus, in a pipe with q = 0, the entropy will rise, resulting in a rise
in temperature down the pipe.
Principles of thermodynamics explain the nature of flow in a pipe or chan-
nel. Because of viscosity, there is a loss of mechanical energy, which emerges
irreversibly as thermal energy due to the rise in entropy. This is discussed next.

2.6 The nature of frictional losses in channels and pipes

Consider the channel of cross-sectional area A depicted in Figure 2.14. Conser-

vation of linear momentum in the streamwise (x) direction over the indicated
control volume at steady state leads to

Fsx + Fbx + p1 A − p2 A = ṁ(V2 − V1 ) (2.70)

where Fsx and Fbx are the surface and body forces acting on the control volume
and V2 and V1 are the average velocities at sections 1 and 2. Equation (2.70) states
that the sum of the forces acting on the control volume is equal to the change
in linear momentum over the control volume. In the fully developed region of
the channel where the velocity is independent of the streamwise direction (see
Chapter 4), V2 = V1 , and we assume Fbx = 0. In this case,
Fsx
p1 − p2 = (2.71)
A
The force Fsx is due to the viscous stress on the wall, and thus Fsx = 2W Lτw if
h  W  L, where W is the width of the channel, as depicted in Figure 2.14
and τw is the shear stress at the wall. Thus
2τw L
p1 − p2 = (2.72)
h
69 Preparatory Concepts

Fluid mechanics, heat transfer, and thermodynamics are intimately linked

through the coupled nature of the momentum and energy equations. The energy
equation, or the first law of thermodynamics, in the absence of any work, results
in (for an incompressible fluid)
p2 − p1
q = u2 − u1 + (2.73)
ρ
where u 2 and u 1 are the internal energies at 1 and 2 and q is the heat transferred
across the control volume boundary in kJ/kg. Thus
p1 − p2 = ρ(u 2 − u 1 − q) = ρ(u − q) > 0 (2.74)
The pressure difference p1 − p2 on the left hand side of equation (2.74) represents
a loss in mechanical energy. Often, the heat transfer q < 0 since heat energy is
usually lost to the surroundings; moreover, u 2 − u 1 > 0. Thus the entire right-
hand side of equation (2.74) represents thermal energy, and equation (2.74) is
thus a statement that mechanical energy is transformed into unrecoverable thermal
energy. This is an example of a thermodynamic irreversibility.
In fluid mechanics, the quantity p1 − p2 is proportional to a “loss.” Thus a
head loss is defined by
p1 − p2 = ρgh L (2.75)
where h L is the head loss and
L V2
hL = f (2.76)
h 2g
where f is called the friction factor and g is the gravitational acceleration; for
laminar flow, the friction factor in the fully developed region of a wide channel is
f = 24/Reh , where the Reynolds number is based on the channel height h and,
for a cylindrical tube, f = 64/ReD . Note that as the Reynolds number approaches
zero, the friction factor becomes infinite; that is, for Poiseuille flow, resistance to
flow in very small channels and tubes is very large.
Note that head loss has units of length and
u2 − u1 − q
hL = >0 (2.77)
g
Using the T ds equation in the form T ds = du + pdv and the expression defining
the entropy production, u − q = T σ for constant temperature. Thus equation
(2.77) can be written in the form
Tσ
hL = >0 (2.78)
g
Equation (2.78) is the link between fluid mechanics and thermodynamics for
internal viscous flow. The same phenomenon occurs in both pipes and channels
of any cross section.
From the definition of internal energy and the definition of the heat transfer at
low fluid speeds of interest, here the head loss can also be related to differences
 70 Essentials of Micro- and Nanofluidics

in temperature. Thus
cT + k ∂∂Ty |w
hL = (2.79)
g
where c is the specific heat, T is the temperature difference down the channel,
and k is the thermal conductivity, and heat transfer at the wall has been assumed
to be by conduction only.
From equation (2.77), it is seen that

q = u 2 − u 1 − gh L (2.80)

Here the internal energy u 2 = cT2 , where c = 1 kJ/kgK is the specific heat, for
example. Typically in liquid flows, the temperature rise between two points in a
channel or tube in the absence of driven heat transfer to the fluid is very small,
perhaps less than 1 K. In very small channels on the order of h = 1 µm or less,
the heat loss q in kJ/kg is almost all due to the head loss. As an example, for water
flowing in a channel of length L = 10 µm and height h = 1 µm, and a velocity
m
of 0.03 sec , the head loss h L = 0.3 m and the resulting heat loss is q = −2.38
kJ/kg for T2 − T1 = 0.5K. What will happen if the channel is smaller?
This analysis links in a precise way the three branches of the thermal sciences.
Given the friction factor of fluid mechanics, the combination of the heat loss
to the surroundings and the increase in internal energy of the fluid in a pipe or
channel can be determined. Likewise, the discipline of heat transfer relates the
temperature gradient at the wall to the thermodynamic heat loss. It should be
mentioned that thermodynamics does not give any information about the internal
energy at points between 1 and 2 in Figure 2.14.

2.7 Chapter summary

Transport coefficients are fundamental to the development of the governing equa-

tions of fluid mechanics and heat and mass transfer and for the conservation of
electrical charge for an electrically conducting fluid. Viscosity, thermal conduc-
tivity, the diffusion coefficient, and the electrical permittivity are the four transport
coefficients that appear in the governing equations of fluid mechanics, heat and
mass transfer, and electrostatics. In addition to these fundamental transport coef-
ficients, the concept of surface tension, which is important to some applications
in micro- and nanofluidics, is also introduced. The primary purpose of this chap-
ter is to introduce the reader to the process of determining these coefficients
from experiment as well as theory. The constitutive laws defining these transport
coefficients have been shown to have distinct similarities when the variation of
the dependent variables is confined to one dimension.
The fluid mixtures considered in this text are dilute enough that there will be
no significant error in assuming that the transport coefficients have the properties
of the solvent. The exception is surface tension, which is considerably altered by
the addition of a small amount of surfactant.
71 Preparatory Concepts

The form of the constitutive laws in the definition of viscosity, thermal conduc-
tivity, and diffusion coefficient were determined empirically from a large number
of experiments. Conversely, one interpretation of the electrical permittivity is that
it defines the surface charge density.
Fluid mechanics, heat and mass transfer, and thermodynamics are inextrica-
bly linked. It is the irreversible conversion of the thermal energy defined by
thermodynamics and heat transfer that causes the pressure drop in a pipe. The
magnitude of the pressure drop is linearly proportional to the entropy production
σ , thereby providing an explicit means of evaluating the entropy production that
is not usually possible in a thermodynamic framework.

EXERCISES
2.1 For Poiseuille flow in a channel, the volume flow rate is defined by
h
Q= uW dy
0

where u is the velocity, h is the height of the channel, and W is its width.
From methods developed in Chapter 4, it can be shown that

W h 3 p
Q=
12μL

Calculate the required pressure drop to achieve a volume flow rate of 10−6
L/min for a channel of width W = 100 µm, length L = 100 µm, and for
channel heights of h = 1 µm, 0.1 µm, and 0.01 µm. Assume the working
fluid is water. Repeat for air. Calculate the Reynolds number for each height.
Repeat the calculation if the flow rate is 10−9 L/min. Put your results in a
table with the channel height h as the independent parameter, and compare
the results for water and air for each flow rate. Use the channel height h
as the length scale in the definition of the Reynolds number. Take the
densities to be 1000 and 1.4 kg/m3 for water and air, respectively, and
.8 × 10−3 and 1.8 × 10−5 N sec/m2 for the viscosities.
2.2 The velocity for Poiseuille flow in a pipe is given by
 2
R2 d p r
w(r ) = − 1−
4μ d x R

Find the average velocity in the pipe, which is defined as

1 R
V = 2πr w dr
A 0

where R is the pipe radius and A = π R 2 . Show that the flow rate is given
by

π R4 d p
Q=−
8μ d x
72 Essentials of Micro- and Nanofluidics

Calculate the pressure drop in water required to produce a flow rate of

Q = 1 µL/min through a tube having a radius R = 10−7 m using the same
values of the length L as in the previous problem.
2.3 Plot the Debye length in water
√
e RT
λ=
F I 1/2

where F is Faraday’s constant, e is the electrical permittivity of the medium,


I is the ionic strength, I = i z i2 ci , ci is the concentrations of the elec-
trolyte constituents at some reference location, R is the universal gas con-
stant, z i is the valence of species i, and T is the temperature, as a function
of ionic strength from c = c1 = c2 = 30 × 10−6 M to T = 1 M, for a pair
of monovalent electrolytes such as sodium chloride in water. For each value
of the ionic strength you choose, assuming that the mixture is electrically
neutral, calculate the number density of the ionic species if n i = ci N a,
where N a is Avagadro’s number. Repeat the calculation for the same values
of the concentration for a pair of +2, −2 ions.
2.4 A sodium ion has an ionic diameter of about 0.2 nm. Assuming that the ion
is a point charge, calculate the electric field and the potential at a distance
of r = 6 nm from its center in an aqueous system. Repeat the calculation if
the solvent is methanol.
2.5 Albumin is a globular protein and so, to a first approximation, it may be
assumed to be spherical and to have an ionic diameter of about 7 nm. Esti-
mate its diffusion coefficient in water using the Stokes–Einstein formula.
2.6 Estimate the diffusion coefficient of sodium in water using the Stokes–
Einstein relation.
2.7 An equation of state for water relates the pressure to the density by
 n
p ρ
= (A + 1) −A
p0 ρ0

where A = 3000, n = 7, p0 = 1 atm, and ρ0 = 998 kg/m3 . Calculate the

pressure required for the density of water to double.
2.8 For thin electric double layers, compared to the smallest channel dimension,
the electro-osmotic velocity is constant across the channel and is given by

u = CU0

where C is a dimensionless constant dependent on the concentration at

some reference location and U0 is the velocity scale. Find a relationship
between U0 and the pressure drop in Poiseuille flow in a wide channel if
the volume flow rate is the same as in the electro-osmotic flow.
2.9 Calculate the heat transfer rate in water if a temperature difference of
T = 3◦ C is measured over a distance of 1 µm at T = 60◦ C.
73 Preparatory Concepts

2.10 Is there a realistic combination of parameters such that the mass transfer
rate has the same numerical value as the heat transfer rate if the fluid is
water?
2.11 Is there a realistic combination of parameters such that the shear stress has
the same numerical value as the heat transfer rate?
2.12 A separation of two species is required based on their valence. Which flow
would yield the most efficient separation: EOF or Poiseuille? In considering
the EOF, assume that the EDL is thin.
 3 The Governing Equations for an
Electrically Conducting Fluid

3.1 Introduction

In the first two chapters, the field of micro- and nanofluidics was introduced in a
general framework, describing multiple applications as well as the scientific issues
associated with fluid flow at small length scales. In Chapter 2, the fundamental
transport properties involved in mathematical description of fluid flow, heat and
mass transfer, and electrostatics were introduced. In this chapter, we use these
concepts to derive, from first principles, the governing equations necessary for
analyzing micro- and nanofluidic phenomena. For the purposes of this chapter, it
will be assumed that the properties of the fluid medium are constant.
Microfluidic devices are being used for rapid and continuous purification of
proteins; a sketch of such a device is shown in Figure 3.1. The device addresses
the need for high-throughput purification of very small amounts of proteins and
enzymes from the carrier fluid. The term protein purification refers to a series of
operations meant to isolate a single protein or enzyme in a complicated mixture.
Here the microfluidic transport processes involve mass transport of a relatively
large number of species with the target molecules present in as little as microgram
per liter concentrations. This device can purify a sample in a short period of time
and does not require a large amount of sample. The means of developing a model
for such a device is discussed in this chapter.
The governing equations of fluid motion on the macroscale are the (incom-
pressible) Navier–Stokes equations. These equations, along with conservation
of mass, or the continuity equation, enable the calculation of, in the general
three-dimensional case, the three velocity components and the pressure. The
equations of electrostatics provide information to calculate the electrical poten-
tial given the distribution of ionic concentrations within the domain of interest.
The energy equation determines the temperature, and finally, conservation of
species determines the local concentration of each of the mixture constituents.
Together, these equations form a set of 6 + N − 1 = 5 + N equations in 5 + N
unknowns, where N is the number of constituents of the fluid mixture. Note that
only N − 1 species concentrations are unknown because the sum of the concen-
trations, the total concentration, is known. The governing equations are presented

74
 75 The Governing Equations for an Electrically Conducting Fluid

in Cartesian coordinates, with the relevant

operators for cylindrical and spherical coor-
dinates appearing in Appendix B.
After the derivation of each of these gov-
erning equations, the appropriate boundary
conditions are presented, and then the sub-
ject of dimensional analysis is discussed.
This issue is an important one, especially
for nanofluidic channels for which velocity,
concentration, electric potential, and tem-
The Sandia Lab-on-a-Chip for the high- perature distributions are extremely diffi-
Figure 3.1
throughput purification of minute amounts of
proteins and enzymes from a carrier fluid
cult to measure.
(Meagher et al., 2008). Reproduced by permis- Next, dimensional analysis is used to
sion of The Royal Society of Chemistry. transform the dimensional governing equa-
tions into dimensionless form. This process
leads to a discussion of several analogies between heat and mass transfer and
velocity and electrical potential. Next, the mathematical character of the gov-
erning equations and well-posed problems are discussed, with emphasis on the
fact that the character of a partial differential equation determines the numerical
method used to solve it. Finally, we discuss the role of fabrication, experiments,
and theory in micro- and nanofluidics.

3.2 The continuum approximation and its limitations

In the study of thermal science and the field of electrostatics and electrodynamics,
it is tacitly assumed that the fluid may be regarded as a continuum. This means
that from a macroscopic viewpoint, the character of the flow, heat and mass
transfer, and the electrical properties of the fluid may be described by averaging
over some small but finite volume. In practice, the size of this small but finite
volume is never specified; however, for the fluid to behave as a continuum, the
averaging volume must be much greater than a molecular diameter in a liquid
and the mean free path in a gas.
To illustrate the continuum hypothesis, consider the case of interest in this
book: a liquid. Consider a small imaginary volume within the domain of flow.
The volume is assumed to be so small relative to the smallest macroscopic space
dimension that the cube defines a point in the space occupied by the fluid. Consider
a region in space as depicted in Figure 3.2, and let us determine the density at the
point (x0 , y0 , z 0 ). In macroscopic terms, the density is defined as
m
ρ= (3.1)
V
but this value may not be the value at (x0 , y0 , z 0 ). To determine that, we must
take the limit
dm
ρ = lim  (3.2)
V →V d V
 76 Essentials of Micro- and Nanofluidics

Illustration of the applicability of the continuum approach for V > d V  .

Figure 3.2

where V  is some (unspecified) critical volume. As V → V  , the cube may

contain only a small number of molecules; that is, the number of individual
molecules in the volume is of the same order as the the number of fluid molecules
passing into and out of the volume. When this occurs, the continuum assumption
breaks down.
The mean free path of a gas is the distance a molecule travels between
collisions. In air, the mean free path is λ p ∼ 60 nm. Thus, as shown in Sec-
tion 3.11, it is expected that the continuum approximation will break down for
flow in a channel whose smallest dimension is about 600 nm = 0.6 µm = 10λ p .
In liquids, it makes no sense to speak of a mean free path because the molecules
are colliding all the time. For liquids, it makes more sense to speak of a mean
molecular spacing. The mean molecular spacing can be taken to be the molecule’s
diameter, which, for water, is D ∼ 0.3 nm = 3 Å.
If a nominal criterion for the critical volume is taken to be 10 times the
molecular diameter, then it is expected that the critical volume for water will be
on the order of V  ∼10 × 4/3π(δ 3 /8) ∼ 140 × 10−30 m3 . Thus the continuum
approximation for the flow of an aqueous electrolyte (i.e., salt water) is expected
to be valid for the smallest channel dimension on the order of 10 nm!
The alternative to continuum approximations of the flow is to treat each
molecule individually and solve Newton’s law in the form

F = ma (3.3)

directly. There are many ways to do this, and these molecular simulation methods
will be discussed later in this book. Two methods often used when the continuum
approximation breaks down are termed molecular dynamics simulations (Fig-
ure 3.3), often used for liquids, and Monte Carlo simulations, most often used for
gases. However, it must be said that these simulations are computationally inten-
sive, often taking a month for a single simulation, and many approximations are
required even to get to that point. It is for this reason that micro- and nanofluidic
 77 The Governing Equations for an Electrically Conducting Fluid

Molecular dynamics simulation of an albumin molecule in a channel; the solvent, water, is not shown.
Figure 3.3
The channel walls are also simulated using MD. Without an MD simulation, the conformation, or shape and
orientation of the albumin molecule, an important protein, cannot be determined. Courtesy of Professor Sherwin
Singer. Used with permission.

devices are difficult to design based solely on molecular simulations. Molecular

dynamics or some other molecular method is required to determine a molecule’s
shape and orientation.

3.3 Kinematics

Kinematics is the study of bodies in motion, without reference to the masses or

forces that cause the motion. There are two ways of describing the motion of a
fluid particle within a given flow field. The motion of a fluid particle could be
considered a rigid body, and its trajectory could be followed instantaneously in

A Lagrangian description follows each fluid element in time like following the motion of molecules inside a
Figure 3.4
carbon nanotube. Image courtesy of Lawrence Livermore National Laboratory.
 78 Essentials of Micro- and Nanofluidics

time. Such a scheme of monitoring the

motion of such a particle is called the
y * t=t+dt Lagrangian description of fluid motion.
dy=vdt Each fluid particle is thus differentiated
by its corresponding starting position, and
* dx=udt the velocity is described by the function
t=t V (x 0 , y0 , z 0 , t), in which the time t is the
only variable. The entire flow field consist-
x
ing of the motion of these fluid particles
Definition of the velocity of a fluid particle. is represented by determining a number of
Figure 3.5
particle paths originating from different ini-
tial positions. This method of describing
fluid motion is important for determining residence times of a given analyte in a
rapid molecular analysis tool.
Now consider a fluid flow through a microfluidic device such as described in
Chapter 1. Instead of tracing a single particle path from an upstream reservoir
through a nanofluidic channel (as in Figure 1.1), suppose attention is focused
on, say, the space within the microfluidic (or nanofluidic) channels or even in
the reservoirs; then the velocity within the given space is a function of position
within that space, or V = V (x, y, z, t). Note that the velocity can vary widely,
depending on the position (x, y, z) within the control space of interest and time.
If the initial position of such a particle of interest falls within the space, then each
value of (x 0 , y0 , z 0 ) corresponds to some value of (x, y, z) at each time t so that

V (x 0 , y0 , z 0 , t) = V (x, y, z, t) (3.4)

and this equation defines the Eulerian description of fluid motion.

Any fluid particle has a given fluid velocity, which may vary in time and/or
space. Consider a general function H = H (x, y, z, t). The differential of this
function is thus

∂H ∂H ∂H ∂H
dH = dx + dy + dz + dt (3.5)
∂x ∂y ∂z ∂t

Dividing by dt, then,

dH ∂ H dx ∂ H dy ∂ H dz ∂H
= + + + (3.6)
dt ∂ x dt ∂ y dt ∂z dt ∂t

It is recognized that the differentials appearing in equation (3.6) are the local fluid
velocities; for example, referring to Figure 3.5, the velocity in the x direction is
defined by

dx
u(x, y, z) = (3.7)
dt
79 The Governing Equations for an Electrically Conducting Fluid

and similarly for the other two directions. Thus the material derivative is given
by

dH DH ∂H ∂H ∂H ∂H
= = +u +v +w (3.8)
dt Dt ∂t ∂x ∂y ∂z

The symbol D/Dt for the material derivative is used to distinguish the material
derivative from the time derivative of a particle trajectory d/dt.
Applying equation (3.8) to the Eulerian velocity vector V (x, y, z, t) =
(u(x, y, z, t), v(x, y, z, t), w(x, y, z, t)), it is found, for example, that

Du ∂u ∂u ∂u ∂u
= +u +v +w (3.9)
Dt ∂t ∂x ∂y ∂z

The first term in equation (3.9) is called the local acceleration and is nonzero in
unsteady or transient situations. The second term is the convective acceleration.
If the velocity is independent of time, the flow is said to be steady. In the
Eulerian description of fluid motion, equation (3.9) shows that a fluid particle
may accelerate even if the flow is steady. The form of the material derivative for
the other two velocity components is easily obtained.
The material derivative for the temperature and concentration of a given species
is required for the energy and species mass balances. These two quantities are
obtained by taking H = T , the temperature, and H = c A , the concentration in
equation (3.8), respectively.
In many cases, the velocity field is independent of one of the coordinate direc-
tions. For example, a velocity field of the form V = (u(x, y, t), v(x, y, t), 0)
is a two-dimensional flow, and if steady, the velocity has the form V =
(u(x, y), v(x, y), 0). Fully developed flow in the x direction in a channel or
pipe is one-dimensional; V = (u(y), 0, 0) for a channel (see Figure 1.4), and
V = (0, 0, w(r)) in the (r, θ, z) directions for a pipe of circular cross section.
The material derivatives in cylindrical and spherical coordinates are presented in
Appendix B.
There are several different ways to visualize fluid flows:

r Velocity vectors in the direction of the velocity at a given point

r Contours of constant velocity
r Streamlines
r Pathlines
r Streaklines

The depiction of streamlines at a fixed time is termed a Eulerian description of

the flow field since the flow field is considered a function of position (Figure 3.6).
Streamlines are defined at each fixed time by the equation

dr × u = 0 (3.10)
 80 Essentials of Micro- and Nanofluidics

v dx
B
Single Stream
Tube u dy

ψ=ψ A
Β y

ψ=ψΑ x
Surface consisting
of Streamlines

A1 A2

Definition of a streamline, a line parallel to the velocity vector.

Figure 3.6

where × denotes the cross-product, and this definition is equivalent to

dx dy dz
= = (3.11)
u v w
in Cartesian coordinates. By their definition, streamlines are lines parallel to the
instantaneous local velocity field.
Another way to define streamlines valid in two-dimensions is to define a stream
function ψ. Consider the continuity equation in Cartesian coordinates, which, in
two dimensions, is

∂u ∂v
+ =0 (3.12)
∂x ∂y

where u and v are the velocities in the x and y directions, respectively. We define
the stream function as that function which satisfies
∂ψ ∂ψ
u= and v=− (3.13)
∂y ∂x

In this case, subject to sufficient restrictions on the mathematical continuity and

differentiability of the stream function, which are satisfied in any reasonable
physical situation, the continuity equation, or conservation of mass equation,
is identically satisfied. It should be mentioned that a stream function cannot
be defined in three dimensions. Note also that if the vorticity vector vanishes,
∇ × V = 0, then it is easy to show that

∇2ψ = 0 (3.14)

The lines ψ = constant are then the streamlines.

 81 The Governing Equations for an Electrically Conducting Fluid

Streakline: Line traced out

by the set of all fluid particles
which passed through release point
at some earlier time τ<t.
τ<

particle 2 t
f
ti

Release
Point
particle 1
tf

Pathline: Line traced

out by a single fluid
particle which Pathline and streakline
passed through must intersect at some
release point at later time
τ<t..
some earlier time τ<

Illustration of the difference between pathlines and streaklines.

Figure 3.7

A pathline is a line traced out in time by a given fluid particle. The set of
equations satisfied by pathlines are
d xi
= u i (xi , t) (3.15)
dt
subject to initial conditions
xi = xi,0 at t = 0 (3.16)
for i = 1, 2, 3. These are ordinary differential equations in the time domain for a
given particle and describe the Lagrangian representation of fluid motion.
A streakline is a line traced out by a neutrally buoyant marker that is con-
tinuously injected at a fixed point in space. Suppose the injection point is given
by (x1,0 , x2,0 , x 3,0 ). A particle at a given point (x 1 , x2 , x 3 ) at time t must have
passed through (x 1,0 , x 2,0 , x3,0 ) at some earlier time τ < t. The governing equa-
tions to determine streaklines are the same as for pathlines, except that the initial
conditions are
xi = xi,0 at t = τ (3.17)
Streaklines are important because it is streaklines that experimentalists visualize
when dye or hydrogen bubbles are used to track fluid particles. For steady flow,
pathlines, streamlines, and streaklines are all equivalent. However, streaklines
are very difficult to calculate because they trace the time history of a potentially
large set of fluid particles, all of which left the same point at different times.
In contrast, pathlines trace the time history of only a single fluid particle. The
difference between streaklines and pathlines is illustrated in Figure 3.7, which
shows two particles that left the release point at different times. Their pathlines
must intersect the streaklines at two points at a given final time t f because the
pathline of a single particle and one point on the streakline coincide at the initial
time at the point of release.
82 Essentials of Micro- and Nanofluidics

Streaklines in History
When Osborne Reynolds was doing his experiments in 1883 to identify that there
are two types of flow, laminar and turbulent, he used streaklines to delineate the
two regimes. At low Reynolds number, dye injected into a tube passed neatly in
a line down the center of the tube. Turbulent flow was identified by the intense
breakup of the dye streak a certain distance downstream of the release point. The
history of fluid mechanics was changed forever.
Dye injection
needle
Dye streak

Flow

Osborne Reynolds’s experiment used streaklines to observe the difference between laminar and turbulent
flow. The eddies downstream on the bottom figure indicate that the flow is turbulent. See Goldstein
(1965a).

As an example, consider the flow field defined by

u = ay, v = −a(x − bt) (3.18)
with a and b constant. This is an unsteady two-dimensional flow that satisfies
conservation of mass equation (3.12). The equation for the streamlines is
dy v
= (3.19)
dx u
Separating variables and integrating, it is easily shown that the streamlines are
circles with their center at y = 0, x = bt, defined by
(x − bt)2
y2 + = const (3.20)
2
The pathlines are obtained by solving the set of equations
dx
= ay (3.21)
dt
dy
= −a(x − bt) (3.22)
dt
The solution is obtained by standard techniques as
x(t) = A cos(at) + B sin(at) + bt (3.23)
y(t) = B cos(at) − A sin(at) + b/a (3.24)
 83 The Governing Equations for an Electrically Conducting Fluid

10

x
5

0
y

−5
x
Streakline
Pathline
−10
−5 0 5 10 15
x

Results for the streaklines and pathlines, showing three intersection points. The calculation was carried out
Figure 3.8
from t = 0 to t = 4, and the results are plotted at t = 4. The plus denotes the starting point and the asterisk
denotes the end point. The cross denotes the crossing points.

To obtain a particular pathline, initial conditions need to be specified, and if, for
example, y = 7 at t = 0, then the solution is
y(t) = (7 − b/a) cos(at) − sin(at) + b/a (3.25)
Likewise, the streaklines are obtained in the same way as the pathlines, except
that the initial condition is expressed as
x = 1, y = 7 at t = τ (3.26)
To determine the constants A and B, we need to solve the system of equations

1 = A cos(aτ ) + B sin(aτ ) + bτ (3.27)

7 = −A sin(aτ ) + B cos(aτ ) + b/a (3.28)

The results for A and B are as follows:

A = (1 − bτ ) cos(aτ ) − (7 − b/a) sin(aτ ) (3.29)

B = (7 − b/a) cos(aτ ) + (1 − bτ ) sin(aτ ) (3.30)

Substitution of the results for A and B into the expressions for x(t; τ ) and y(t; τ )
and evaluation at t = 4 yields the solutions. The results are depicted in Figure 3.8.

3.4 Surface and body forces

The Navier–Stokes equations are the expression of Newton’s law F = m a for a

fluid. Thus the forces on a given fluid particle must be evaluated. Instead of work-
ing with forces directly, it is customary to work with stress or force per unit area.
Two types of stress may act on a fluid particle: surface stress and body stress.
Only those stresses that act in a given direction will cause motion in that direction.
Surface forces act directly on the surface of a fluid volume. Body forces, such as
 84 Essentials of Micro- and Nanofluidics

gravity and an electric field, act on the fluid as a whole.

The surface stresses in Figure 3.9 have two subscripts
corresponding to the surface on which they act and
System
the direction they are pointing. For example, a positive
x surface is defined as a surface whose outward unit
normal is pointing in the positive x direction. A stress
is defined to be positive if it is on a “positive” surface
pointing in the positive coordinate direction or on a
Pressure forces
act inward “negative” surface pointing in the negative coordinate
direction. All the stresses in Figure 3.9 have positive
Pressure is a normal stress.
magnitudes as drawn.

Wind induces a shear stress on water causing waves to form.

A cube of fluid moving in space may undergo four distinct types of changes:

1. Translation
2. Rotation
3. Extensional strain
4. Angular or shear strain

Translation and rotation cannot generate a stress; the extensional and angular
strains are related to gradients of the velocity field and the pressure.

Stresses on a fluid cube. Stress is defined as force per unit area. Adapted from Turns (2006).
Figure 3.9
 85 The Governing Equations for an Electrically Conducting Fluid

Two-dimensional view of (top) extensional and (bottom) shear strain on a fluid cube. Adapted from Currie
Figure 3.10
(2003).

Consider the the extensional and angular strains that may occur to a fluid cube
as depicted in Figure 3.10; strain is defined as the change in a given length in the
fluid cube, and the rate of strain is the quantity of interest.
For the extensional strain, suppose the length A = d x and the length B = dy.
The notation σx x denotes the rate of the change in length of a fluid element in the
x direction, d x due to flow over its original length, and has units of sec−1 . Thus
the fractional change in length in the x direction defines the extensional strain
rate, and
∂u
dx + ∂x
d xdt − dx ∂u
σx x dt = = dt (3.31)
dx ∂x
and in a similar fashion,
∂v
σ yy = (3.32)
∂y
∂w
σzz = (3.33)
∂z
Now consider the angular strain rate. The key to determining this quantity is
to evaluate the angles dα and dβ in Figure 3.10 at time t + t that at time t were
zero. Note that if dα > dβ, then the fluid cube has undergone a counterclockwise
rotation, whereas if the opposite is the case, a clockwise rotation is obtained, and
e = d  tan dβ and f = c tan dα. It is now necessary to determine the lengths
c , d  , e, and f .
The length of the original face d at t = t is d = dy; thus it has changed
as a result of the change in the v velocity so that the length d  at t = t + dt
is d  = dy + (∂v/∂ y) dydt using the Taylor series approximation for small dt.
Similarly, the length c has changed as a result of the change in the u velocity;
86 Essentials of Micro- and Nanofluidics

thus c = d x + (∂u/∂ x) d xdt. The length e was originally zero and has become
nonzero as a result of the change in u velocity with y. Thus e = (∂u/∂ y) dydt,
and finally, for the same reason, f = (∂v/∂ x) d xdt.
Note that in Figure 3.10, the rotation of the fluid element is about the z axis,
the third axis into the page. If counterclockwise rotation is considered positive,
let z denote the average rotation rate of the fluid cube. The angle dβ indicates a
clockwise rotation, and the angle dα indicates a counterclockwise rotation. Then
the average rotation rate associated with the fluid cube is
1
dz = (dα − dβ) (3.34)
2
Note that
dα f ∂v
= lim tan−1  = (3.35)
dt dt→0 c ∂x
dβ e ∂u
= lim tan−1  = (3.36)
dt dt→0 d ∂y
so that
 
dz 1 ∂v ∂u
= − (3.37)
dt 2 ∂x ∂y
A similar analysis shows that the rotation rates about the x and y axes are given
by
 
dx 1 ∂w ∂v
= − (3.38)
dt 2 ∂y ∂z
 
d y 1 ∂u ∂w
= − (3.39)
dt 2 ∂z ∂x
Twice the rotation rate of a fluid particle is an important quantity, and
d
ω=2 =∇×V (3.40)
dt
is called the vorticity. It is the vorticity that determines the character of the
velocity field; note that vorticity is a conserved quantity
∇ •ω =0 (3.41)
If ω = 0, the flow is said to be irrotational.
It remains to define the shear strain rate; in keeping with convention (White,
2006; Currie, 2003), the shear strain rate in the x y plane is defined as the average
decrease in the angle that is initially 90◦ , or
   
1 dα dβ 1 ∂v ∂u
σx y = + = + (3.42)
2 dt dt 2 ∂x ∂y
In similar fashion, it is found that
 
1 ∂w ∂v
σ yz = + (3.43)
2 ∂y ∂z
 
1 ∂u ∂w
σx z = + (3.44)
2 ∂z ∂x
 87 The Governing Equations for an Electrically Conducting Fluid

The shear strain rate is a tensor; note that it is symmetric, σi j = σ ji , so that there
are six distinct values. It is the relationship between stress and rate of strain,
termed a constitutive relation, that is specific to a fluid.
The pressure is a normal stress and will contribute to the extensional strain
rate depicted in Figure 3.10. Thus, if the fluid is Newtonian, the stress tensor is
given by
 
∂u i ∂u j
τi j = − pδi j + 2μσi j = − pδi j + μ + (3.45)
∂x j ∂ xi
for i = 1, 2, 3 in tensor notation, where u i = (u, v, w). The subject of rheology
is concerned with determining the stress–strain rate relationship for a given fluid
or class of fluids. Here δi j is the Dirac delta function:

δi j = 0 i = j (3.46)

δi j = 1 i = j (3.47)

For example, in conventional Cartesian notation for a Newtonian fluid,

   
dα dβ ∂v ∂u
τx y = 2μσx y = μ + =μ + (3.48)
dt dt ∂x ∂y
∂u
τx x = − p + μ (3.49)
∂x
It is now possible to determine the governing equations for an electrically
conducting fluid. A Newtonian fluid having constant properties is assumed.

3.5 The continuity equation

In the sections to follow, the governing equations will be developed in Cartesian

coordinates; typically, channels in a nanopore membrane are long and wide, as in
Figure 3.11, so the governing equations may be considerably simplified. In your
first fluid dynamics course, you studied the integral form of the mass conservation

Side View
L

Flow
h
y,v
x,u L>>h
z,w
End View
W

z,w
W>>h

Geometry of a typical channel; systems of such channels make up a nanopore membrane.

Figure 3.11
 88 Essentials of Micro- and Nanofluidics

equation, which is
∂
ρdV + ρV • d A = 0 (3.50)
∂t V S

where dV denotes that the integral is over a volume and d A denotes that the
integral is over an area. The first term is the time rate of change in mass within
the control volume, and the second term is the net mass passing out of the control
volume. Using the divergence theorem, which is defined for any vector H as

∇ • H dV = H •dA (3.51)

then
∂
ρdV + ∇ • ρ V dV = 0 (3.52)
∂t V
Combining the terms after exchanging the time derivative and the integral, it is
noted that the only way for the integral to be zero is for the integrand to be zero,
or
∂ρ
+ ∇ • ρV = 0 (3.53)
∂t
This is the differential expression of conservation of mass. Expanding the second
term of equation (3.53), we note that equation (3.53) can be written as
Dρ
+ ρ∇ • V = 0 (3.54)
Dt
For incompressible flow, defined as
Dρ
=0 (3.55)
Dt
it is seen that
∂u ∂v ∂w
∇•V = + + =0 (3.56)
∂x ∂y ∂z
This is a single partial differential equation with unknowns (u, v, w). Note that
the simpler but more restrictive definition of incompressible flow ρ = constant
also satisfies equation (3.55).

3.6 The Navier–Stokes equations

The Navier–Stokes equations are a statement of conservation of linear momen-

tum, or equivalently, Newton’s law:
F = ma (3.57)
where F is the sum of the forces exerted on a fluid element, m is its mass, and
a is its acceleration. In fluid mechanics, it is customary to write this equation in
terms of force per unit volume so that
F
= ρa (3.58)
V
 89 The Governing Equations for an Electrically Conducting Fluid

Differential element showing pressure forces and

Figure 3.12
viscous forces acting on the element.

In integral form, the incompressible momentum equation takes the form

F ∂
= ρ V dV + ρV V • d A (3.59)
V ∂t V S

The right-hand side of this equation is equivalent to the material derivative for
an infinitesimal control volume, and in the general three-dimensional case, the
momentum equation takes the form
DV
ρ = F = Fsurface + Fbody (3.60)
Dt
The surface forces correspond to both pressure and viscous forces, and the
body forces correspond to either gravity or electrostatic forces in the case of an
electrically conducting fluid such as an electrolyte solution.
Refering to Figure 3.12, consider the net surface force on the fluid cube. In the
x direction on the front, right, and top faces, the surface force is
d Fx = τx x,front dydz + τ yx,right d xdz + τzx,top d xd y (3.61)
For the back, left, and bottom faces, a Taylor series expansion can be used; for
example, on the back face, the normal stress in the x direction is given by
∂τx x
τx x,back = τx x,front −
dx (3.62)
∂x
The net surface force in the x direction is given by
Fx,net = τx,front − τx,back (3.63)
so that the net surface force in the x direction is given by
∂τx x ∂τ yx ∂τzx
Fx,net = d xd ydz + d yd xdz + dzd xdy (3.64)
∂x ∂y ∂z
In a stagnant fluid, this net force is zero, assuming that there are no body forces in
the x direction. Dividing through by the volume, since the stress is a symmetric
tensor, the net force per unit volume is given by
∂τx x ∂τx y ∂τx z
f x,net = + + (3.65)
∂x ∂y ∂z
90 Essentials of Micro- and Nanofluidics

The procedure is similar for f y and f z . Note that

∂ ∂ ∂
f x,net = (τx x ) + (τ yx ) + (τzx ) (3.66)
∂x ∂y ∂z
so that f x,net is the divergence of the first column of the stress tensor. Thus, from
the x component f x , the net surface force per unit volume can be written in the
form
f surface,net = ∇ • τi j (3.67)
where i j refers to the triad (x, y, z) = (1, 2, 3).
At this point, it is useful to discuss indicial or tensor notation, which is benefi-
cial when the energy equation is derived. The stress τi j is a first-order tensor that
can be defined by
 
τ 
 11 τ12 τ13 
 
 τ21 τ22 τ23  (3.68)
 
 τ31 τ32 τ33 

In indicial notation, (u, v, w) = (u 1 , u 2 , u 3 ), and the continuity equation is

written as
∂u i
=0 (3.69)
∂ xi
with the occurrence of a repeated index defined as a sum from i = 1 − 3. The
vector velocity field is denoted simply by u i in indicial or tensor notation. The
advantage of this notation is that equations may be written in tensor notation
much more compactly.
Thus Newton’s law is given by
DV
ρ = b + ∇ • τi j (3.70)
Dt
where b is the body force per unit volume. Note that fluid will experience a
convective acceleration if the cross-sectional area contracts and a convective
deceleration if the area widens (Figure 3.13).
If the velocity vanishes, the resulting field of fluid mechanics is called hydro-
statics; in this case, there are no shear stresses, and the normal stress is given by
τii = − p.
The last step in the derivation of the Navier–Stokes equations is to specify the
stress–strain relationship, which may be expressed as

τi j = function(σi j ) (3.71)

for each distinct component. Stokes, back in the mid-nineteenth century, postu-
lated three conditions that a deformation law must satisfy (Tokaty, 1971):
r The fluid is a continuum.
r The fluid is isotropic.
r The relationship must approach the hydrostatic case as the strain rates approach
zero.
 91 The Governing Equations for an Electrically Conducting Fluid

u1 u2

1 2

In moving through a channel for which the area

Figure 3.13
widens from point 1 to point 2, a fluid ele-
ment experiences a convective deceleration,
u (∂ u/∂ x ) < 0. If the area contracts, a fluid
element will experience a convective accelera-
tion, and if the area is constant, the convective
term u (∂ u/∂ x ) = 0.

This is the case for a Newtonian fluid for which

τi j = μσi j − δi j p (3.72)

In the course of determining this relationship, a second coefficient of viscosity

emerges that, theoretically, cannot be proven to be zero; however, this term is
multiplied by ∇ • V , which is zero for an incompressible fluid. Thus the second
coefficient of viscosity will not be discussed further.
The Navier–Stokes equations in Cartesian coordinates are obtained by sub-
stituting the stress–strain relationship into equation (3.60), and the momentum
equation in the x direction is given by
Du ∂p
= bx −
ρ + viscous terms (3.73)
Dt ∂x
Using the continuity equation to eliminate the terms involving the velocities v
and w,
∂u ∂u ∂u ∂u 1 ∂p
+u +v +w =− + bx + ν∇ 2 u (3.74)
∂t ∂x ∂y ∂z ρ ∂x
Similarly, in the other two coordinate directions,
∂v ∂v ∂v ∂v 1 ∂p
+u +v +w =− + b y + ν∇ 2 v (3.75)
∂t ∂x ∂y ∂z ρ ∂y
∂w ∂w ∂w ∂w 1 ∂p
+u +v +w =− + bz + ν∇ 2 w (3.76)
∂t ∂x ∂y ∂z ρ ∂z
where ν = μ/ρ is the kinematic viscosity and ∇ 2 is the Laplacian in Cartesian
coordinates
∂2 ∂2 ∂2
∇2 = + + (3.77)
∂x2 ∂ y2 ∂z 2
92 Essentials of Micro- and Nanofluidics

Along with the continuity equation, these are four equations in four unknowns
(u, v, w, p) assuming incompressible flow.
The form of the Navier–Stokes equations depends on the coordinate system
used; the preceding derivation assumes the standard Cartesian coordinate system.
The equations can be put in what is termed invariant form, a form valid for both
spherical and cylindrical coordinate systems. Written in vector form, the Navier–
Stokes equations are
∂V 1 1
+ (V • ∇)V = B − ∇ p + ν∇ 2 V (3.78)
∂t ρ ρ
The vector operators in this equation still depend on the coordinate system
used; the vector operator (V • ∇)V is different in the three coordinate systems.
However, using the vector identities

∇ 2 V = ∇(∇ • V ) − ∇ × ∇ × V (3.79)
 
| V |2
(V • ∇)V = ∇ −V ×∇ ×V (3.80)
2

the Navier–Stokes equations become

 
∂V | V |2 1
+∇ − V × ∇ × V = B − ∇ p − ν∇ × ∇ × V (3.81)
∂t 2 ρ
for an incompressible fluid. All the vector operations required for this invariant
form appear in Appendix B.
Several comments need to be made about these equations. First, they are highly
nonlinear; this is manifest in terms auch as
∂w
u (3.82)
∂x
in which one unknown multiplies another. Second, they are partial differential
equations and can be either elliptic, parabolic, or hyperbolic. Sometimes they
can be of different character in different parts of the flow field, as is the case
when compressibility of the fluid is important. Third, they are second-order
equations because of the second derivatives that appear on the right-hand sides
of the equations. Thus, in each coordinate direction, two boundary conditions are
required to uniquely solve the equations; one initial condition is required in the
case of unsteady flow.
Finally, they are a system of coupled equations. The nature of the fluid velocity
in the x direction depends on the velocities in all the other directions. Clearly this
makes the solution of these equations difficult.
The Navier–Stokes equations are, in general, difficult to solve in their full form.
Flows at small scales are characterized by small velocities and are laminar; that
is, the fluid flows in regular layers with little or no mixing between layers. In this
case, the nonlinear terms, the convective acceleration terms in the Navier–Stokes
equations, can often be neglected. This is the case in micro- and nanofluidics.
 93 The Governing Equations for an Electrically Conducting Fluid

Depicted is a microfluidic chip for performing com-

prehensive two-dimensional separations of proteins
and peptides based on electrophoretic mobility. These
devices are fabricated in glass substrates using pho-
tolithographic patterning and wet chemical etching
methods. The resultant trench features are closed with
a glass coverplate with 2-mm diameter vials that act
as fluid reservoirs. The channel features have widths
ranging from 20 – 75 µm and depths of 10 µm. The first
and second dimension channels are ≈20 and 1.5-cm
long respectively. The serpentine structure includes
asymmetrically tapered turns to control geometrical
band broadening. Mixtures of peptides or proteins are
introduced at the sample reservoir, S. Buffer reser-
voir, B1, contains a buffer appropriate for perform-
ing micellar electrokinetic chromatography (MEKC),
and reservoir B2 contains a buffer for performing
capillary electrophoretic (CE) separations. Reservoirs
SW1, SW2, and W are waste reservoirs for collecting
waste fluids and application of control voltages. Electric potentials are applied
to all six reservoirs in a time-dependent fashion to transport materials through-
out the structure. The four-way intersections, V1 and V2, act as electrokinetic
valves, which are actuated by modulating the voltages applied to reservoirs B1
and B2. The sample mixture is injected through valve V1 with chromatographic
separation unfolding in the serpentine channel. Bands eluting from the chro-
matographic separation channel are sampled with valve V2 and injected into the
CE channel. Electrophoretically separated components are detected at position
D. The CE injections are performed at 1 Hz, allowing multiple sampling of the
chromatographic bands.
Contributed by Professor J. Michael Ramsey.
Reference: Ramsey, J. D., Jacobson, S. C., Culbertson, C. T., Ramsey, J. M.
Analytical Chemistry 2003, 75, 3758–3764.

3.7 Mass transport

As noted previously, there are several ways to describe mass transport phenomena,
depending on whether mass or molar units are used. Thus some definitions are
required.

3.7.1 Definitions
Flows in micro- and nanofluidic geometries inevitably require transport of elec-
trolyte mixtures in which mass transport is a crucial feature. A mixture consists
of two or more distinct chemical species, and the amount of a given species can
94 Essentials of Micro- and Nanofluidics

be represented by its mass density, ρi which has units of kg/m3 , or its molar con-
centration, ci , which has units of moles/m3 . The molarity of a mixture is defined
as 1 M = 1 mole/L (read 1 molar = 1 mole/liter) and is the most common scale
used by chemists.
The mass density and the molar concentration are related by the species molar
mass:1
ρi = Mi ci (3.83)
The units of Mi are kg/kmole, which, in biochemistry, is called a dalton after the
pioneering chemist John Dalton, so that 1 g/mole = 1 kg/kmole = 1 Da.2 The
molar mass is the amount of mass in one mole; the molar mass of a molecule is
the sum of the atomic weights of the individual atoms making up the molecule.
For example, the molar mass of water (H2 O) is 1 × 2 + 1 × 16 = 18 Da
because the molar mass of hydrogen is M H = 1 Da; the molar mass of oxygen
is M O = 16. Often the unit Da is omitted, and it is said that the molar mass
of naturally occurring elements ranges from M = 1 to 238 and that the molar
mass of small and simple chemical compounds ranges from M ∼10 to 1000. The
common protein albumin has a molar mass of M = 66,000. Large polymers can
have a molar mass M ∼ = 4 × 106 . Selected values of molar mass for a number of
compounds are given in Table 3.1.
The molar mass is directly related to the size of the molecule. Venturoli and
Rippe (2005) suggest that for globular proteins, the molecular density is about
1.33 g/cm3 = 1330 kg/m3 , about 30 percent higher than water. On this basis, the
molecular weight for a protein with this density, modeled as a hard sphere, is
given by
4πa 3
M= ρ NA (3.84)
3
Solving for the molecular radius a,
 0.333
3M
a= = AM B (3.85)
4πρ N A
Results for A and B for selected molecules are given in Table 3.2.
The mixture density is then defined as the sum of the constituent densities, or

ρ= ρi (3.86)
i

Similarly, the total molar density or molar concentration is given by


c= ci (3.87)
i

1
The older term that is often used instead of molar mass is molecular weight, which is a dimen-
sionless quantity measured in atomic mass units (1/12 the mass of carbon, C12 ). The two terms
are often used interchangeably. Here we will use the term molar mass.
2
Actually, the dalton is, strictly speaking, a unit of mass: 1 Da = 1.660538782 × 10−27 kg.
95 The Governing Equations for an Electrically Conducting Fluid

Table 3.1. Molar masses of some atoms and molecules

Atom/compound Symbol Molar mass (Da)
Hydrogen H 1
Oxygen O 16
Carbon C 12
Sodium Na 23
Chlorine Cl 35
Water H2 O 18
Sodium chloride NaCl 57
Albumin – 66,000
Glucose C6 H12 O6 180
DNA (ds) – 660 (single base pair)
RNA – 300–500
Polyethylene glycol PEG 1000–35,000
Polymethyl methacrylate PMMA 80,000–170,000
Ficoll – 104 –105
Dextran – 10–106

These numbers are rounded; e.g., the molar mass of hydrogen is M = 1.00794. The chemical
formulas of the molecules are not shown because they are too long. The length of a single double-
stranded (ds) DNA base pair is about 0.34 nm. Ficoll is a sucrose-based polymer and is sometimes
used to investigate the sieving function of the kidney. Dextran is a glucose-based polymer and is
also sometimes used for kidney sieving analysis.

The mixture composition can also be expressed in terms of mass fraction ωi ,

ρi
ωi = (3.88)
ρ

or mole fraction,
ci
Xi = (3.89)
c

Table 3.2. Molar mass and size relationships as

compiled by Venturoli and Rippe (2005)
Molecule A B
Hard sphere 0.67 0.333
Hydrated hard sphere 0.74 0.333
Globular proteins 0.483 0.386
Ficoll 0.421 0.427
Monodisperse dextran 0.488 0.437
Polydisperse dextran 0.33 0.463
96 Essentials of Micro- and Nanofluidics

The mean molar mass of a mixture is the weighted sum of the individual molar
masses, or

M= X i Mi (3.90)
i

Note that the mole fraction and mass fractions sum to 1 by definition and are
dimensionless quantities. For a mixture of ideal gases, the mole fractions are
related to the partial pressure by
pi
Xi = (3.91)
p
where p is the total pressure. This equation is obtained by assuming that each
species exerts its own partial pressure at the mixture temperature
pi
ci = (3.92)
Rui T
where Ru is the universal gas constant in kJ/kmole K.
For a binary mixture of species A and B, the mass fraction is related to mole
fraction by
ωA
MA
XA = ωA ωB (3.93)
MA
+ MB

Note that depending on the value of the molar mass, the mole fraction can be
much less than the mass fraction. The mass fraction in terms of mole fraction can
be obtained by inverting the preceding equation and
X A MA
ωA = (3.94)
X A MA + X B MB
Another mass transfer scale that is often used is the molality, defined as the
number of moles of solute per kilogram of solvent: m i = n i /M0 n 0 , where 0
denotes the solvent. The ratio
mi 1
= (3.95)
Xi M0 X 0
For dilute solutions, X i ≈ n i /n 0 so that in this limit, the molality m i ≈ X i /M0 .
Diffusion mass transport takes place when there is a concentration gradient,
just as heat transfer takes place when there is a temperature gradient. Just as heat
is transferred from a higher to a lower temperature, diffusion takes place from a
higher to a lower concentration. Because gradients of concentration are required
for diffusion to take place, diffusion mass transport is a nonequilibrium and
irreversible process. Note that there are many similarities between heat and mass
transfer, and recall that the two fields developed simultaneously (Figure 3.14).
Thus there are many common solutions of the same form, and these analogies
will be explored later in this chapter.
 97 The Governing Equations for an Electrically Conducting Fluid

Prediction of the weather requires the computation of fluid flow and heat and mass transfer of an air-water
Figure 3.14
vapor mixture, in its simplest form. Image courtesy of the National Oceanic and Atmospheric Administration.

An Artificial Kidney
The function of the kidney is to filter out small ions such as sodium and chloride
from the blood serum, while retaining larger proteins such as albumin. End Stage
Renal Disease (ESRD) is the term given to a dysfunctional kidney that does not
retain enough albumin (over 99%). A nanopore membrane can be used as a Renal
Assist Device (RAD) to partially supplement the native kidney. A sketch of such a
device is depicted on Figure 12.7. Each pore in the membrane is about h ∼ 10nm
with the flow field generated by the natural pressure drop across the native kidney.
Thus the base flow is a Poiseuille flow which must be supplemented by a mass
transfer analysis for the diffusion and convection of albumin through the pore.
See Chapter 12 for more details.

3.7.2 Governing equation

Let us first consider the case of zero bulk motion. Then mass transport takes place
by diffusion only, and the mass flux is given by Fick’s law:

j A = −D AB ∇ρ A (3.96)

where j A is a mass flux and has units of kg/sec m2 . If the density of the mixture
remains constant, the flux can be expressed in terms of the mass fraction as

j A = −ρ D AB ∇ω A (3.97)

On a molar basis,

J A = −D AB ∇c A (3.98)

and J A has units of kmole/sec m2 . In terms of the mole fraction,

J A = −cD AB ∇ X A (3.99)

if the total concentration remains constant.

98 Essentials of Micro- and Nanofluidics

Organic Waste
O2 in Salts
CO2 out Water

Nutrients Unabsorbed
Salts matter
Water and
waste

The body is one big mass transfer machine that involves transport of gases through
liquids, gases through cells, and transport of liquid water everywhere. The body
transports charged liquids, salts in water; this fact means that mass transfer often
takes place in the presence of a local electric field. Most proteins are negatively
charged and are transported through cell membranes by pressure driven flow and
migrate through many regions of the body under a combination of concentration
gradients (diffusion) and electric potential gradients; that is, an electric field. A
general look at mass transport within the body can be found in Saltzman (2009).

The mass flux of species A relative to a fixed coordinate system is

n A = ρAvA (3.100)

where v A is a species velocity with the corresponding definition for species B:

n B = ρB vB (3.101)

The mass averaged velocity vector in the Navier–Stokes equations V of the

mixture is then defined by

ρ V = n A + n B = ρ Av A + ρB vB (3.102)

Dividing through by the density ρ,

V = ω A V A + ω B VB (3.103)

for a binary mixture. The mass flux of species A relative to the mixture mass
averaged velocity is

j A = ρ A (v A − V ) (3.104)

and from the definition of n A , we have

n A = jA + ρA V (3.105)
99 The Governing Equations for an Electrically Conducting Fluid

Substituting from Fick’s law, we have

n A = −ρ D AB ∇ω A + ρω A V (3.106)

Similar manipulations lead to

J A = c A (v A − V ) (3.107)

N A = −cD AB ∇ X A + cX A V (3.108)

when the mass transport is described on a molar basis. The preceding equations
for the fluxes are appropriate for convection and diffusion mechanisms and do
not include an electric component.
As noted previously, an electric field is the most common method of driving the
flow of biological fluids in nanoscale channels. For strong electrolytes, a single
salt component, such as NaCl, will be entirely dissociated so that nominally, the
mixture has three components: undissociated water and positive and negative ions
making up the single salt component. In this case, adding the mass transport due
to an electric field, the molar flux of species A for a dilute electrically conducting
mixture is

N A = −D AB ∇c A + m A z A c A E + c A V (3.109)

Here D AB is the diffusion coefficient, R is the universal gas constant, T is the

temperature, z A m A is called the ionic mobility with m A = F D AB /RT , z A is the
valence, F = 96,500 Coul/mole is Faraday’s constant, and E is the electric field.
Equation (3.109) is called the Nernst–Planck equation. In this form, the term
involving the electric field in the flux equation is called electrical migration.
The mass transport equation is then
∂c A
+ ∇ • NA = 0 (3.110)
∂t
Expanding the divergence operator, the ∇• term, the governing equation for
the concentration of species A is given by
 
Dc A D AB z A F ∂c A E x ∂c A E y ∂c A E z
= D AB ∇ c A +
2
+ + (3.111)
Dt RT ∂x ∂y ∂z
where the diffusion coefficient has been assumed constant. This equation is often
written in terms of the mole fraction X A if the total concentration is approximately
constant, which is often the case.
Note that the material derivative has appeared in a similar fashion to the
momentum equation. By definition of the material derivative, if
Dc A
=0 (3.112)
Dt
then in a steady flow field, provided that the concentration field is also steady,
lines of constant concentration will coincide with streamlines.
 100 Essentials of Micro- and Nanofluidics

3.8 Electrostatics

At steady state, the electric field E must satisfy the reduced set of Maxwell’s
equations since there is no magnetic field, and in this case,

∇×E =0 (3.113)

so that the electric field is solenoidal. The electric field must also satisfy

∇ • e E = ρe (3.114)

where ρe is the volume charge density. As in the case of potential fluid flow, an
electrical potential can be defined as

E = −∇φ (3.115)

and so

∇ • (e E) = −∇ • (e ∇φ) = ρe (3.116)

which is a Poisson equation that determines the electrical potential φ. This equa-
tion is the differential form of Gauss’s law. Here e is the permittivity, and the
charge density is given by
 
ρe = F z i ci = Fc zi X i (3.117)
i i

where ci is the molar concentration of species i, X i is the mole fraction of species

i, zi is its valence, F is Faraday’s constant, and c is the total concentration, which
usually remains constant. If ρe = 0 in a given region, that region is said to be
electrically neutral. For constant permittivity,

e ∇ 2 φ = −Fc zi X i (3.118)
i

It is important to understand how electric fields can be set up. If the walls of
the channel are charged, there is a naturally occuring potential associated with the
electrical double layer. This part of the total potential is illustrated in Figure 1.18.
If, in addition, electrodes are placed upstream and downstream of a channel,
there is an external electric field that causes motion of the fluid, as in Figure 1.19.
In most cases of practical interest, the potential associated with the external
electric field can be decoupled from the potential associated with the electrical
double layer; the volume charge density turns out to be small (see Section 7.18),
and then, to leading order,

∇ 2φE = 0 (3.119)

that amounts to neglecting the effect of the electric double layers on the electrodes.
Thus, if the external electric field is oriented in the x direction only and the
surfaces of the channel are electrically insulating ( ∂φ ∂y
E
= ∂φ
∂z
E
= 0), then the
potential associated with the externally imposed electric field satisfies, again
101 The Governing Equations for an Electrically Conducting Fluid

to leading order,
d 2φE
=0 (3.120)
dx2
and so the external potential is linear with x. In this case, the external electric
field is constant, and the total electric field can be written as

E = E 0 î − ∇φ (3.121)

where φ is the electrical potential associated with the electrical double layer and
E 0 is the imposed electric field.

The Equation of Charge Conservation

The governing equation of mass transport for species A in an electrolyte solution
is given by
∂c A
= ∇ • (−D A ∇c A + m A z A c A E + c A V )
∂t
The current density is given by

J=F z i Ni
i
n
Multiplying the mass transport equation by F i=1 z i results in
 
 ∂ci   
zi F =∇• − F Di z i ∇ci + Fm i zi ci E +
2
Fci z i V
i
∂t i i i
n
But the volume charge density is defined by ρe = F i=1 z i ci so that
 
∂ρe 
+ (V • ∇)ρe = ∇ • F Di z i ∇ci + σe E
∂t i

where σe is the electrical conductivity and is defined by


σe = F m i z i2 ci
i

and is the principle of conservation of charge. Note that this equation is a

convective-diffusion equation for the charge density. Note that the higher the
mobility, m i z i , the higher the conductivity. The equation of conservation of charge
is a confirmation that charge is conducted through an electrolyte by the motion of
ions. If all of the diffusion coefficients are assumed equal and constant, Di = D,
∂ρe
+ (V • ∇)ρe = D∇ 2 ρe + ∇ • (σe E)
∂t
Note that for steady flow,
∇•J =0
The electrical conductivity is normally assumed to be constant, which really
means that the concentrations are constant. In this case, and if the fluid velocity
is zero, this equation reduces to Ohm’s Law J = σe E.
 102 Essentials of Micro- and Nanofluidics

The current density is a very important quantity for the sensing of molecules
in micro- and nanodevices. The dimensional current density on a molar basis is
given by (Newman, 1972)

J=F z i Ni (3.122)

where J is the dimensional current density. Substituting for the flux of species i
in molar form, Ni ,

J=F z i (−cD AB ∇ X i + cm i z i X i E + cX i V ) (3.123)
i

Changes in current have been used to sense and analyze biomolecules in

nanochannels, and this subject will be considered in Chapters 9 and 12.
It should be noted that strictly speaking, the term electrostatics refers to a
situation in which the net current density is zero. In the flows of electrolytes
described in this text, this is not the case. When charges move, an additional
force on the fluid appears in the magnitude V × B, where B is the mag-
netic induction; the magnetic induction arises from motion of the ions in an
electro-osmotic flow, for example. (This phenomenon is discussed in Chapter 9).
However, in the cases of interest here, to leading order, the effect of magnetic
induction can safely be ignored. This point will be discussed in greater detail in
Section 6.9.

3.9 Energy transport

The derivation of the thermal energy equation begins with the consideration of
the first law of thermodynamics for an open system, which is given by (see
Section 2.5)

d Q − dW = d E t (3.124)

where Q, W , and E t are the extensive values of heat transfer, work, and total
energy in the system in joules/m3 or energy per volume, respectively, just as
the momentum equation was derived as a balance of forces per volume.3 Recall
that heat and work are dependent on the process path, whereas the energy of
the system is independent of the path and is hence a property in the ther-
modynamic sense. The work term consists of that done by surface forces and

3
The term energy equation is often given to the equation resulting from multiplying a component
of the Navier–Stokes equations by the velocity in that component equation. The resulting equation
has units of energy per volume, but it is not equivalent to the first law of thermodynamics. Such
an equation is called the mechanical energy equation.
103 The Governing Equations for an Electrically Conducting Fluid

that done by body forces and any external work fields, such as shaft work in a
turbine.

In a turbine, the flow of fluid rotates a shaft, that produces work: shaft work.

The derivation of the energy equation is extremely cumbersome in a Cartesian

coordinate system. Thus a mix of indicial and vector notation is used after the
manner of the presentation by White (2006). Again, we use a differential control
volume, as we did in the derivation of the Navier–Stokes equations.
The total energy per volume in a system is assumed to consist of internal energy,
kinetic and potential energy, and the electrical energy present in a conducting
fluid. Thus the total energy per unit volume in the system can be written as
follows:
 
1 1
E t = ρ e + V 2 + b • r + e E 2 (3.125)
2 2

where E is the magnitude of the electric field, E =| E |, V is the magnitude

of the velocity V =| V |, b is the body force per unit volume, e is the internal
energy per mass, and r is the displacement vector. Note that thermal radiation is
not considered.
Notation alert: Here E t and E are used for total energy and electric field,
respectively. The meaning of each quantity should be clear from the context.
To be consistent with the Navier–Stokes equations, the first law should be
written on a rate basis using the material derivative
DQ DW D Et
− = (3.126)
Dt Dt Dt
where the units in equation (3.126) are joule/m3 sec. Substituting for E t , it follows
that
 
D Et De DV DE
=ρ +V + g • r + e E (3.127)
Dt Dt Dt Dt
Consider the heat transfer term; from Figure 3.15, considering only conduction,
for now, and using Fourier’s law, the net heat transfer by conduction through the
 104 Essentials of Micro- and Nanofluidics

Conduction heat transfer in a slab.

Figure 3.15

slab is given by
    
∂qx ∂qx ∂ ∂T
qx − qx + dydz = − = k dydz (3.128)
∂x ∂x ∂x ∂x
where qx is the heat flux, that is, the net heat transport per unit area, through the
left face of the fluid cube. The other coordinate directions are similar. Thus, after
dividing by the volume element,4
DQ
= −∇ • q = +∇ • (k∇T ) + q̇ (3.129)
Dt
where q̇ is a volumetric heat generation rate.
The rate of work done by the surface forces is defined as force times velocity.
For example, in the x direction,

wsx = −(uτx x + vτx y + wτx z ) (3.130)

Using the same procedure as with the conduction mode of heat transfer,
DWs
= −∇ ws = ∇ • V • τi j (3.131)
Dt
The expression ∇ • V • τi j can be expanded as
∂u i
∇ • V • τi j = V • ∇ • τi j + τi j (3.132)
∂x j
Note from the momentum equation that
 
DV
∇ • τi j = ρ +b (3.133)
Dt

where b is the body force per unit volume vector, and so

 
DV
V • ∇ • τi j = ρ V +V •b (3.134)
Dt
The work done by the body forces per unit area is defined by

wb = V • b (3.135)

4
The local time derivative ∂/∂t appears only in the term D E t /Dt because the time rate of increase
in energy within the volume balances the rest of the terms in equation (3.126).
105 The Governing Equations for an Electrically Conducting Fluid

and we decompose the body force vector as

b = g + be (3.136)

where g is the gravitational body force per unit volume and be is the electrical
body force per unit volume. Note that the two terms of the right-hand side of
equation (3.134) are the kinetic energy and the energy associated with the body
force terms of De/Dt in equation (3.125), and so those two terms cancel with
the corresponding terms in the total energy of the system. Thus, using indicial
notation, where necessary, to simplify the presentation,
 
D 1 ∂u i
ρ e + e E 2 = ∇ • (k∇T ) + τi j + q̇ (3.137)
Dt 2 ∂x j
The units of equation (3.137) are joule/m3 sec.
There are many forms of the energy equation; first consider the case where
there is no electric field and no source term. Then the energy equation takes the
form
De ∂u i
ρ = ∇ • (k∇T ) + τi j (3.138)
Dt ∂x j
The second term can be split into two parts corresponding to the viscous and
pressure components of the stress:
∂u i ∂u i
τi j = τij − p∇ • V (3.139)
∂x j ∂x j
The term
∂u i
τij (3.140)
∂x j
is called the viscous dissipation and is almost always negligible in flows at low
velocities. For incompressible flow, the internal energy e = c p T = cv T = cT ,
and thus, for constant specific heat,
DT ∂u i
ρc = ∇ • (k∇T ) + τij (3.141)
Dt ∂x j
which is the most common form of the energy equation. It is the viscous dissipa-
tion term that is responsible for the entropy production discussed in the previous
chapter.
In addition, for negligible viscous dissipation and for a constant thermal con-
ductivity,
DT
ρc = k∇ 2 T (3.142)
Dt
Note that this energy equation is similar in form to the Navier–Stokes equations
for the individual velocities in the absence of a pressure gradient, and it turns out
that there are analogies that arise in certain cases. These analogies are discussed
in Section 3.13.
 106 Essentials of Micro- and Nanofluidics

3.10 Two-dimensional, steady, and incompressible flow

The governing equations for an electrically conducting fluid in the presence of

fluid flow and heat and mass transfer are complicated; however, fortunately, they
are not often solved in their full form. In the special case of steady, incompressible
and two-dimensional flow (the channels in Figure 3.11 are very wide, leading to
∂/∂z = 0) in a fluid with constant properties, which is by far the most common
situation, the governing equations become
∂u ∂v
+ =0 (3.143)
∂x ∂y
∂u ∂u 1 ∂p
u +v =− + bx + ν∇ 2 u (3.144)
∂x ∂y ρ ∂x
∂v ∂v 1 ∂p
+v
u =− + b y + ν∇ 2 v (3.145)
∂x ∂y ρ ∂y
 
∂c A ∂c A D AB z A F ∂c A E x ∂c A E y
u +v = D AB ∇ c A +
2
+ (3.146)
∂x ∂y RT ∂x ∂y
   
∂T ∂T ∂E ∂E
ρc u +v + e E u +v = k∇ 2 T + q̇ (3.147)
∂x ∂y ∂x ∂y

e ∇ 2 φ = −Fc zi X i (3.148)
i
where
∂2 ∂2
∇2 = + (3.149)
∂ x 2 ∂ y2
Later, the concept of a fully developed regime will be introduced, and in this
situation, in the streamwise direction,
∂
=0
∂x
leading to further simplifications. It will be shown that for most relevant cases, the
length required to become fully developed is shortest for the electrical potential,
being on the order of a Debye length, next shortest for the fluid problem, next
shortest for the thermal entry length, and longest for the mass transfer problem.
Indeed, mass transfer problems are almost never fully developed.

3.11 Boundary and initial conditions

The governing partial differential equations defined so far are subject to boundary
and initial conditions. The nature of these conditions depends on what is known
in the particular problem of interest. There are generally three types of canonical
conditions, although more complicated combinations do occur. These canonical
conditions are as follows:
1. Specified value of an independent variable at the boundary such as the
no-slip condition u = 0 at a solid boundary
107 The Governing Equations for an Electrically Conducting Fluid

2. Specified gradient or flux at the boundary such as at a horizontal or vertical,

constant thickness free surface separating a liquid and a gas or vapor where
the shear stress vanishes, ∂u/∂n = 0 where n denotes the direction normal
to a free surface, or at the boundary between two media, where the heat
flux and electric field normal to the surface are continuous
3. A linear combination of the two types such as in a convection heat transfer
problem, ∂ T /∂ y = h(T − T∞ ), where h is a heat transfer coefficient.

In the following sections, we discuss each of the fluid dynamics, mass transfer,
electrostatics, and heat transfer boundary conditions in some detail.

3.11.1 Velocity boundary conditions

Prandtl, in the early 1900s, performed fundamental experiments demonstrating
that in a fluid with finite viscosity, the velocity component parallel to a solid
surface will be the velocity of that surface. If the surface is not moving, then
the fluid velocity is zero at the surface; that is, the fluid sticks to the surface.
Bernoulli (1738) appears to have been the first to recognize this when he noted
significant differences between his calculations for the case in which the viscosity
μ = 0 and his measurements (Goldstein, 1965b; Lauga et al., 2005). In light of
this, Goldstein (1965b) discusses the possible boundary conditions:
r The fluid molecules at the wall are bound to the solid molecules (Coulomb).
r A thin layer of fluid near the wall is bound to the solid (Girard).
r There is a slip at the wall that, in some limit, results in the no-slip condition
(Navier).

Ludwig Prandtl was born in 1874 in Bavaria to a family that encouraged his
interest in physical phenomena in nature (Anderson, 1982). He received his PhD
degree from the University of Munich in 1900 with a doctoral thesis in solid
mechanics. Having been appointed as Professor of Mechanics at the Technische
Hochschule in Hanover, he continued his new-found interest in fluid mechanics
he developed as a designer of a machine that removed material shavings by
suction. It was at Hanover that he began developing his famous boundary layer
theory, culminating in his famous 1904 paper delivered to the Third Congress of
Mathematicians at Heidelberg. Indeed, his most famous contribution has always
been the demonstration of the no-slip condition in fluid mechanics. However,
he also made significant contributions in other areas of fluid mechanics and heat
transfer. During the period from 1905–1910, he became interested in compressible
flow and his name appears in compressible flow in the Prandtl-Meyer expansion
fan, the flow past a corner whose measure exceeds 90◦ . After this he focused his
attention on low-speed aerodynamics, developing the still-used Prandtl Lifting
Line and Lifting Surface theories for calculating lift and induced drag. Prandtl’s
demonstration of the no-slip condition altered the course of viscous flow research
for an entire century.
 108 Essentials of Micro- and Nanofluidics

y u(y ) There was some debate as to which of the situa-

tions is appropriate, until Prandtl discovered the
boundary layer in 1904. His discovery that the
boundary layer consists of the regime in which
uw the fluid is brought to relative rest at the wall
resolved D’Alembert’s paradox that suggested that
Ls when μ = 0, there is no drag on a flat plate, which
is in conflict with observation. This laid to rest
∂u the debate of the previous 100 years over the exis-
uw = Ls
∂y
tence of the no-slip condition until very recently,
y=0
when the no-slip condition has been questioned
in liquid flows in microfluidic devices near both
Definition of slip length. Sketch by
Figure 3.16
Professor Minami Yoda. hydrophilic and hydrophobic surfaces, with a con-
sensus building that the no-slip condition may not
be appropriate near hydrophobic surfaces.
It should be pointed out that the no-slip condition has been established at the
macroscale using empirical means, the result of a large number of experiments
begun by Prandtl. It also needs noting that the physical mechanisms for liquid
slip are still a matter of some debate, and a detailed discussion of current work is
beyond the scope of this text. An excellent review of some recent experimental
results showing slip, particularly in liquids, has been presented by Lauga et al.
(2005), and an excellent discussion of the historical view of liquid slip is given
by Karniadakis et al. (2005). A short summary of these slip concepts follows.
The amount of fluid slip is quantified by defining a slip length. The slip
length in Figure 3.16 can be inferred by indirect techniques, such as flowrate
measurement, and also by molecular dynamics simulations and atomic force
microscopy.5 There is evidence to suggest that slip can occur at hydrophobic
surfaces, with water molecules being replaced by a gas (e.g., air) layer, inducing a
free-surface boundary condition (see later). To distinguish this phenomenon from
slip near nonhydrophobic surfaces, this type of slip has sometimes been termed
apparent slip (Lauga et al., 2005; Bhushan, 2007; Wang et al., 2009). There have
been thousands of papers on this phenomenon; see, in particular, Schnell (1956)
and Barrat and Bocquet (1999). At hydrophilic surfaces in an aqueous solution,
the no-slip condition is generally thought to hold, although this entire field is still
a matter for continuing research (Koplik et al., 1989; Honig & Ducker, 2007).
The amount of slip, if it occurs, is also a function of the roughness of the surface
(Priezjev & Troian, 2006). If the surface is hydrophilic, roughness decreases slip
(Richardson, 1973), whereas for a hydrophobic surface, roughness allows the
formation of gas pockets that increase slip (de Gennes, 2002). It is important to
note that it has been recognized for a long time that slip can occur in rarefied
gases and in non-Newtonian fluids.

5
The atomic force microscope (AFM) was developed to measure atomic-scale topographical
features on a charged or uncharged surface using a highly flexible cantilever beam, the tip of
the probe. Forces as small as 1 nN can be measured using an AFM. Detailed information on the
operation and applications of the device is given by Bhushan (2007).
109 The Governing Equations for an Electrically Conducting Fluid

Slip Length Increases with Hydrophobicity

Sketches of hydrophilic and hydrophobic surfaces. Hydrophobicity increases from left to right. Courtesy
Bharat Bhushan, used with permission (Wang et al., 2009).

The figure above depicts three surfaces, one a smooth mica surface that
is hydrophillic; a hydrophobic surface created by self assembly of alkane
n-hexatriacontane and lotus wax. The slip-lengths measured using atomic force
microscopy (AFM) are essentially zero for the hydrophilic surface, to 44 nm and
133 nm for the hydrophobic and superhydrophobic surfaces respectively. The rms
roughnesses of the three surfaces is 0.2 nm for the mica, and 11 nm and 178 nm
for the hydrophobic and superhydrophobic surfaces respectively. Contributed by
Professor Bharat Bhushan.

The no-slip condition applies to the vast majority of flows even at the micro-
and nanoscale and simply states that the velocity component tangent to the wall
equals the wall velocity. The solid wall condition requires the velocity component
normal to the wall to be zero. This condition applies to both liquids and gases,
provided that the smallest dimension in the physical system of interest is much
greater than the mean free path in the case of a gas and a typical molecular
diameter in a liquid.
In a channel bounded by two walls at y = 0, h, for example, the no-slip
condition in three dimensions, is expressed as

u = w = 0 y = 0 and y=h (3.150)

whereas the solid wall condition is expressed as v = 0 at y = 0, h.

Many biomaterials are porous, and for a porous wall at y = 0,

u, w = 0 v = vw y = 0 (3.151)

The velocity at the wall, vw , must be calculated from the properties of the porous
material.
Whether slip occurs in a gas is dependent on the Knudsen number, and slip will
occur if K n = λ/ h = O(10−3 ), where λ is the mean free path of the gas. Note
that K n ∼1 if either λ is large, such as in a rarefied gas, or if h, the characteristic
dimension, is small, as in a microdevice. The mean free path of air is around
λ ∼ 60–70 nm.
 110 Essentials of Micro- and Nanofluidics

Kn= 0.0001 0.001 0.01 0.1 1 10 100

Continuum:NS Free−
no−slip. Transition molecule
Continuum:NS regime: non− flow
slip. continuum.

The effect of Knudsen number on the nature of the flow.

Figure 3.17

There is a well-developed theory of the nature of slip flow for gase; the essential
features are depicted in Figure 3.17. For K n < 0.001, the flow is essentially of the
continuum variety, with the no-slip condition holding. For 0.001 < K n < 0.1,
the flow is still continuum, but with the slip condition to be discussed here.
Next is a transition regime for 0.1 < K n < 10, and finally, for K n > 10, free
molecular flow occurs. In the transition regime, a molecular simulation must be
used; direct simulation Monte Carlo (DSMC) simulations (Bird, 1994) are often
used in gases.
The slip condition was apparently first proposed by Navier in 1827, who
suggested that the velocity at a solid surface is proportional to the wall shear
stress: thus the form of the boundary condition is given by
∂u
u fluid = u wall + L s
at a solid boundary (3.152)
∂n
where n is the direction normal to the surface. Here L s is a slip length; for gases,
L s = λ is the mean free path. This equation assumes hard-sphere molecules and
perfectly diffuse reflection. For both diffuse and specular reflection,
2 − σv ∂u
u fluid = u wall + Ls at a solid boundary (3.153)
σv ∂n
where σv is an accomodation coefficient and is the fraction reflected diffu-
sively. Atomically smooth surfaces are more likely to achieve specular reflection,
whereas atomically rough surfaces are more likely to exhibit diffuse reflection,
with fluid molecules rebounding at random angles. The Navier slip condition can
also be used for liquids, with the slip length taken to be a molecular diameter.
Equation (3.153) is first order in the Knudsen number and is valid for an
isothermal flow. This equation can be extended to second order by retaining the
next term in the Taylor series expansion for the velocity near the wall and
 
2 − σv ∂u 1 2 ∂ 2u
u fluid = u wall + Ls + L s 2 + · · · at a solid boundary
σv ∂n 2 ∂n
(3.154)
In dimensionless form, this equation can be written
 
2 − σv ∂u 1 2∂ u
2
u fluid = u wall + Kn + Kn at a solid boundary (3.155)
σv ∂n 2 ∂n 2
111 The Governing Equations for an Electrically Conducting Fluid

where now the variable n has been scaled on L s . Karniadakis et al. (2005) write
this second-order condition in the form
2 − σv K n ∂u
∂n
u fluid = u wall + at a solid boundary (3.156)
σv 1 − B(K n)K n
where, to match the Taylor series,
1 u 
B= (3.157)
2 u
where the prime denotes differentiation with respect to n. They suggest that B
be considered an empirical parameter so as to extend the validity of this slip
boundary condition for larger Knudsen numbers. For compressible gases, these
slip boundary conditions are considerably more complicated and are coupled to
the temperature distribution.
The concept of liquid slip is much less developed than that for gases. For one
thing, there is no such thing as a mean free path. Since liquid molecules are
always colliding with other molecules a molecular diameter away, an argument
can be made that the mean free path is simply a molecular diameter. At higher
shear rates in liquids, Thompson and Troian (1997) have suggested, instead of
the no-slip condition,
∂u
u = Ls (3.158)
∂n
where γ̇c is a slip length that depends on the local shear rate
 −1/2
−1 ∂u
L s = L s 1 − γ̇c
0
(3.159)
∂n
where γc is a critical rate of strain obtained from experiment. L 0s is a scale length,
which is taken to be 17 molecular diameters in water by Thompson and Troian
(1997). On the basis of these results, the Navier no-slip condition can be viewed
as the low strain limit of a more general form that diverges at a critical value of
the strain rate.
A free surface is the boundary between a liquid and a gas or vapor. In this case,
the tangential velocity and the tangential component of shear stress on either side
of the free surface must be continuous; in the simple case where the pressure is
uniform and there is negligible curvature in the surface,
V1 = V2 and τ 1 = τ2 (3.160)
where
∂u 1
τ1 = μ1 (3.161)
∂n
In three dimensions, in the absence of surface tension, the stresses in each of the
coordinate directions are continuous across the free surface. It is to be noted that
the viscosity of a gas is much smaller than that of a liquid. If the liquid is denoted
by 1 and the gas by 2, because viscosity of a gas is much smaller than that of a
liquid, usually, τ1 ∼0.
112 Essentials of Micro- and Nanofluidics

Experimental Data on Slip Length

40 40
Ls [nm]

0 0

–40 –40
(a) (b)
0 1000 2000 0 1000 2000
. [s–1]
. [s–1]

These two figures show the slip length as a function of the shear rate γ̇ for steady,
fully-developed and creeping (Reynolds numbers Re < 0.25) Poiseuille flow of
2 mM (䊊) and 10 mM (䊉) ammonium acetate (CH3 COONH4 ) and 2 mM (䉭) and
10 mM (䉱) ammonium bicarbonate (NH4 HCO3 ) aqueous solutions through 33 µm
deep microchannels (Li & Yoda 2010). The slip lengths were estimated using a
local method, evanescent wave-based multilayer nano-particle image velocimetry
(Li & Yoda 2008), by extrapolating the velocity profile, which is effectively linear
within the first 0.5 µm next to the channel wall, from three independent velocity
measurements in this near-wall region. The plot on the left (a) gives slip lengths
for naturally hydrophilic fused-silica channels, while that on the right (b) gives slip
lengths for identical channels (within fabrication tolerances) made hydrophobic
by a self-assembled monolayer of octadecyl trichlorosilane (OTS). In all cases, the
results, for microchannels with a cross-sectional aspect ratio exceeding 15, give
shear rates within 5% on average of analytical predictions for two-dimensional
Poiseuille flow after the data are corrected for the effects of nonuniform tracer
distribution. The error bars denote the maximum standard deviations in the slip
lengths of (a) 25 nm and (b) 40 nm, based on the uncertainties in the linear
curve-fit and the actual velocity data.
Reference: H. F. Li and M. Yoda (2010) An experimental study of slip considering
the effects of nonuniform colloidal tracer distributions. J. Fluid Mech., vol. 662,
pp. 269–287, 2010.
Contributed by Professor Minami Yoda

In free surface problems, the interface shape z = η(x, y) is an unknown that

must be obtained in the course of the solution. In this case, the kinematic condition
is that the velocity normal to the free surface must be equal to the fluid velocity
at the free surface:

Dη ∂η ∂η ∂η
w(x, y) = = +u +v (3.162)
Dt ∂t ∂x ∂y
113 The Governing Equations for an Electrically Conducting Fluid

These free surface boundary conditions also hold at a liquid-liquid interface

separating two fluids having different properties.
For the case of a curved surface, surface tension effects can be significant.
The effect of surface tension is to produce a discontinuity in the normal stress
that is proportional to the mean curvature of the surface (Wehausen & Laitone,
1960). The expressions for the stresses are presented earlier in this chapter, and
the general conditions for free surfaces in the presence of surface tension are
presented by Wehausen and Laitone (1960) and Levich and Krylov (1969). In
particular, for two-dimensional flow, it can be shown that the two conditions
reduce to
  d 2h
dh dh dh
[ p] − 2[μu x ] − [μ(u y + vx )] = γ  
dx2
3/2
(3.163)
dx dx 1 + dh
2 dx
dx
d h 2
dh
[ p] + [μ(u y + vx )] − 2[μv y ] = γ  
dx2
(3.164)
dx 2 3/2
1 + dh dx

where [ p] = p1 − p2 denotes, for example, the jump in the pressure across the
interface and u x = ∂u/∂ x.

3.11.2 Mass transfer boundary conditions

Most often, these boundary conditions fall into the cases of specified concentra-
tion or specified flux or a combination of these. Thus, on a molar basis,

c A = c A0 at a solid boundary (3.165)

if the specified concentration is known or

Nn A = N A0 at a solid boundary (3.166)

where Nn A is the flux in the direction normal to the boundary. At a solid wall,
N A0 = 0.
Finally, a mass transfer convection coefficient, h̄ m , can be defined in terms of
the gradient of the concentration in situations where flow past a boundary occurs,
and
∂c A
− = h̄ m (c A − c A∞ ) at a solid boundary (3.167)
∂n
where c A∞ is the concentration far from the wall. The mass transfer convection
coefficient is an empirically obtained coefficient, and equation (3.167) serves as
a definition of h̄ m .
In electro-osmotic flow, the flux of a given species is given by

N A = −D AB ∇c A + m A z A c A E + c A V at a solid boundary (3.168)

Normal to the wall, the flux is zero, and so the condition is

∂c A
−D AB + m A z A c A E n = 0 at a solid boundary (3.169)
∂n
 114 Essentials of Micro- and Nanofluidics

Sketch of a biochemical reaction taking place at a surface containing bioreceptors. The ligand is the target
Figure 3.18
analyte; ligands are atoms, molecules, ions, or other structures that bind to other structures to form a complex.
The terms ligand and receptor are generic.

where E n is the electric field component normal to the wall. If a chemical reaction
occurs at a surface, there will be a net flux of a species A at the surface. For a
homogenous reaction, this boundary condition is

N A = K A c A at a solid boundary (3.170)

Note that the rate constant K A has units of m/sec. The form of the boundary
condition depends on the character of the chemical reaction and so depends on
the physical problem.
Biochemical reactions occur when biomolecules attach and/or detach from
surfaces (Figure 3.18). Typically, these interactions are characterized by two rate
constants, which are denoted by K on and K off . The rate constant K on is called
the association rate constant, associated with adsorption to the surface, and K off ,
with the dissociation rate constant associated with desorption from the surface.
There are several forms of these reactions, and one such boundary condition is
of the Michaelis–Menten type (Murray, 2001), given by
∂c A ∂c A
−D AB = = K on c A (cs0 − cs ) − K off cs (3.171)
∂n ∂t
where c A is the solute concentration on the liquid side at the wall and has units of
M = mole/m3 , cs0 is the total number of sites available for binding in mole/m2 ,
and cs is the surface concentration in mole/m2 . Rhee et al. (1989) derives the
form of the rate of adsorption–desorption on the right side of equation (3.171).

3.11.3 Electrostatics boundary conditions

In electrostatics, the curl of the electric field vanishes; that is, ∇ × E = 0. Fol-
lowing the treatment of Masliyah and Bhattacharjee (2006), applying Stokes’s
theorem, it follows that

E • ds = 0 (3.172)
 115 The Governing Equations for an Electrically Conducting Fluid

Material 1
y
A A a
Interface

x
Material 2
A A a

An infinitesimal control volume encompassing an interface between two different materials. Here A is the area
Figure 3.19
of the indicated face. Adapted from Masliyah and Bhattacharjee (2006).


where indicates the integral around the boundary of the control volume in
Figure 3.19 in a counterclockwise manner. Assuming d = a initially, it follows
that
E x,1 = E x,2 (3.173)
or the x component of the electric field in each material is the same. This means
that the x derivative of the potential is the same, and thus integrating along the
interface φ1 = φ2 + C. Letting a → 0 with d fixed, it follows that C = 0, and so
the potential is continuous at the interface.
For constant electrical permittivity, the electric field satisfies the equation
 e ∇ • E = ρe (3.174)
Integrating this equation over the control volume of Figure 3.19, and using the
preceding condition for the x component of the electric field,
∂φ ∂φ 1
e1 E 1y − e2 E 2y = −e1 |1 +e2 |2 = ρe dV (3.175)
∂n ∂n A V

The term on the right-hand side of this equation has units of Coul/m2 and is thus
a surface charge density. If ρe = 0, then
∂φ ∂φ
|1 = e2
e1 |2 (3.176)
∂n ∂n
Note that this condition is of the same form as the continuity of heat flux at a
solid boundary between two materials or the continuity of heat flux between two
materials. The two other components of the electric field that are parallel to the
wall must also be continuous. If material 1 is water and material 2 is silica, then
e1  e2 , and approximately,
∂φ 1
−e1 |1 = ρe dV at a solid boundary (3.177)
∂n A V
116 Essentials of Micro- and Nanofluidics

where the outward normal to the boundary is assumed to be in the positive

direction.
In this formulation, leading to equation (3.175), care must be used in defining
the quantity ρe because its determination may include surface chemistry reactions
on the solid side (say, material 2). Another way to obtain the boundary condition
is to integrate the Poisson equation for the electrical potential equation from the
Stern plane (Section 1.10) to the bulk, where the potential is uniform, also leading
to
∂φ 1
−e1 |1 = ρe dV (3.178)
∂n A V
For electroneutrality to be preserved,
1
ρe dV + σ = 0 (3.179)
A V
leading to the equation
∂φ σ
− = at a solid boundary (3.180)
∂n e
where now σ is the surface charge density on the wall. The sign in front of the
normal derivative depends on the sign of the outward normal, and in general,
σ
n̂ • ∇φ = at a solid boundary (3.181)
e
An alternative to the gradient boundary condition is the specified potential. If
the electrical potential at the Stern plane, termed the ζ potential, is known, then
φ = ζ at a solid boundary (3.182)

3.11.4 Temperature boundary conditions

These boundary conditions mirror the mass transfer boundary conditions and
actually predate their development. Thus
T = T0 at a solid boundary (3.183)
if the specified temperature is known or
∂T
−k = q0 at a solid boundary (3.184)
∂n
At a boundary between two different media,
∂T ∂T
−k1 |1 = −k2 |2 (3.185)
∂n ∂n
that is, the heat flux should be continuous.
Finally, a heat transfer convection coefficient h̄ T can be defined in terms of the
gradient of the concentration in situations where flow past a boundary occurs and
∂T
−k = h̄ T (T − T∞ ) (3.186)
∂n
where T∞ is the temperature far from the wall. As with the mass transfer coeffi-
cient, the h̄ T is an empirically obtained coefficient. At a free surface, the temper-
ature and heat flux normal to the surface should be continuous.
 117 The Governing Equations for an Electrically Conducting Fluid

3.11.5 Other boundary conditions

Other boundary conditions arise in specific situations such as those conditions
on the velocity, temperature, and pressure at the inlet and outlet of tubes and
channels. Often streamwise gradients of velocity are assumed to vanish at the
outlet of a channel or pipe.
When solving the Navier–Stokes equations, boundary conditions on the pres-
sure must be specified. These boundary conditions come from the equations
themselves and involve velocity derivatives both normal and tangential to the
wall. The bottom line is that the boundary conditions depend on what is known
at the ends of the domain for the specific problem of interest.

3.12 Dimensional analysis and similarity

An important question to ask is, Under what conditions are flows of different
fluids flowing in two geometrically similar geometries (same shape), with each
appropriate length scale of the same ratio, equivalent? Such similar flows are
called dynamically similar, and the means to determine whether such flows are
dynamically similar is called dimensional analysis. For two flows of different
fluids, at different velocities and different linear dimensions (but with similar
geometries, e.g., two spheres), to be dynamically similar, it is reasonable to
expect that the ratio of the forces on the two bodies will be fixed at all times.
As you learned in your undergraduate fluid dynamics course, at the macroscale,
the most important function of dimensional analysis is to provide a means of
designing experiments to be performed on the model scale. The model scale can
be an order of magnitude or more smaller than the scale on which information
is sought, the prototype. In micro- and nanofluidics, of course, it is not neces-
sary to scale down the experiments; indeed, it may be desireable to scale up
the experiments though in my experience, I have never seen this done! More-
over, there are often difficulties in precisely determining the very dimensions
of microfluidic devices (see Section 3.17). The function of dimensional analy-
sis in micro- and nanofluidics is to identify the characteristics of the problem,
such as whether it is one-, two-, or three-
dimensional, and to identify analogies
between fluid flow, heat transfer, mass trans-
U
fer, and electrostatics. Both these functions
lead to an increased understanding of flows
a
at small scales.
Consider the case of flow past a smooth,
uncharged sphere, as in Figure 3.20. Then
the drag on the sphere can be a func-
ρ,μ
tion of several parameters, namely, D =
Flow past a spherical particle in a fluid with D(ρ, μ, d, V ), where d is the sphere diame-
Figure 3.20
density ρ and viscosity μ. ter and U is the velocity far from the sphere.
118 Essentials of Micro- and Nanofluidics

These five parameters can be reduced to two parameters using dimensional anal-
ysis. It is easily shown that the dimensionless drag coefficient on a sphere is
defined by
F
cd = = f (Re) (3.187)
ρU 2 d 2
where f is a function to be determined and Re is the Reynolds number. Thus flows
past two spheres are dynamically similar if the Reynolds numbers are the same
in each case. The Reynolds numbers can thus be made equivalent by adjusting
the free stream velocity U or the diameter d, or both. Experiments designed for
water may be performed in air if the Reynolds number is the same in each case.
Dimensional analysis is the art and science of reducing the number of param-
eters that constitute the solution to a given problem. Dimensional analysis recog-
nizes that certain physical parameters, including fluid properties and velocities
and length scales, appear only in certain groupings. These groupings are dimen-
sionless parameters, which are then used to characterize the system. The most
important example in viscous flow is the Reynolds number, and for the sphere,
Re = ρU d/μ.
Three procedures may be employed to determine the relevant dimensionless
groups, first, by inspection, because certain obvious parameters can be determined
by experience such as ratios of length scales and the Reynolds number. The
dimensionless parameters can also be determined explicitly, as suggested by the
Buckingham pi theorem, which is usually done in undergraduate fluid dynamics
courses (Munson et al., 2005). Finally, these parameters may be obtained by
writing the governing equations in dimensionless form, and this is the method
described here.
The governing equations for the flow of an electrically conducting fluid in the
isothermal case indicate that a typical velocity in three dimensions is a function
of a number of variables, and in the general case, for steady flow,

u = u(x, y, z, U, p, d1 , d2 , d3 , μ, ρ, D AB , e ) (3.188)

where di are the typical length scales in the three coordinate directions and U is a
typical velocity scale. In most cases, a single velocity scale is obvious. This is an
overwhelming number of parameters, and dimensional analysis can be employed
to reduce significantly the number of parameters. All the parameters on the right-
hand side of equation (3.188) are dimensional, and (x, y, z) are variables in the
coordinate directions.
Consider first the case of a three dimensional flow in a channel for which the
three length scales di are all of different magnitudes, as in Figure 3.11. This is
the case for the nanopump depicted in Figure 1.1b. In that pump, (d1 , d2 , d3 ) =
(L , h, W ) = (3 µm, 4–50 nm, 44 µm); that is, each channel in a membrane is
nanoconstrained in one dimension, having a (uniform or nearly uniform) pore
size ranging from 4 to 50 nm. The number of independent parameters can be
reduced substantially by nondimensionalizing the governing equations.
119 The Governing Equations for an Electrically Conducting Fluid

The first step in the process is to define a dimensionless length in each of the
coordinate directions; for example, in the streamwise direction of the flow,
x
x∗ = (3.189)
L
Substituting into the Navier–Stokes equations with
∂ dx∗ ∂
= (3.190)
∂x dx ∂x∗
all the terms involving derivatives in the coordinate directions will be partially
nondimensional. Next, a velocity scale must be determined. The velocity scale is
normally fairly obvious but is also particular to the problem. Defining a dimen-
sionless velocity,
u
u∗ = (3.191)
U
where it is supposed that the value (i.e., a number, U = 10−3 m/sec) of U is
known. (In what follows, the asterisk will be dropped, and the presence of dimen-
sionless parameters in a given equation will signal the fact that the equation is
dimensionless.)
Typically, the channels in a nanopore membrane are of the shape depicted in
Figure 3.11. First, the continuity equation becomes
∂u ∂v ∂w
1 + + 2 =0 (3.192)
∂x ∂y ∂z
where all three velocities are scaled on the single velocity scale U and 1 = h/L
and 2 = h/W . In this convention, h is assumed to be the smallest dimension,
and so both 1 and 2 will be small. In this situation,
∂v
=0
∂y
and so since v = 0 at y = 0, 1, according to the solid-wall boundary condition,
v = 0 everywhere. This leads to the definition of fully developed flow, where the
streamwise velocity does not depend on the streamwise coordinate: ∂u/∂ x = 0
(assuming w = 0). This is the case for many micro- and nanoscale membranes
used in biological applications.
Using the same procedure in the Navier–Stokes equations, the x momentum
equation becomes
 
∂u ∂u ∂u ∂u ∂p
+ Re 1 u +v + 2 w =− + bnd x + ∇ 2 u (3.193)
∂t ∂x ∂y ∂z ∂x
where bnd x is the dimensionless body force in the x direction. The fluid time
scale is t f = h 2 /ν, so the dimensionless time variable for the fluid is t ∗ = t/t f .
The dimensional body force due to the presence of a constant electric field in
the x direction is given by

bx = E 0 ρe (3.194)
120 Essentials of Micro- and Nanofluidics

N
where ρe = Fc i=1 z i X i is the volume charge density, z i is the valence of
the ionic species i, and F is Faraday’s constant. The dimensionless body force
emerges from the nondimensionalization process after multiplication of the
quantity h 2 /μU and

FcE 0 h 2  β 
N N
bnd x = zi X i = 2 zi X i (3.195)
μU i=1  i=1

where β = c/I is the ratio of the total concentration to the ionic strength and is
usually large. It is also common to scale the concentrations on the ionic strength,
in which case, β = 1. For electro-osmotic flow, the velocity scale turns out to be
U = e E 0 φ0 /μ, where φ0 = RT /F.
In the preceding equations, the lengths are nondimensionalized on the triad
(L , h, W ) (Figure 3.11), and (u, v, w) are the dimensionless velocities in each of
the coordinate directions (x, y, z); also
∂2 ∂2 2 ∂
2
∇ 2 = 12 + +  (3.196)
∂ x 2 ∂ y2 2
∂z 2
Consider first the case in which the body force term bnd x = 0 and the flow is
thus pressure driven only. Then

u = u(x, y, z, U, p, d1 , d2 , d3 , μ, ρ) (3.197)

After nondimensionalizing the problem, the number of parameters in the list has
been reduced; in the full three dimensional case, u = u(x, y, z, p, 1 , 2 , Re),
where now x, y, z are dimensionless; note that there are three fewer parameters
than in the dimensional specification.
In the case of fully developed flow in a wide and long channel,
∂ ∂
= =0 (3.198)
∂x ∂z
so that u = u(y, p, Re). However, for fully developed flow, the convective terms
in equation (3.193) vanish, and thus u = u(y, p). Actually, it will be shown that
u = u(y, d p/d x) and that d p/d x is constant for incompressible flow of a liquid.
For a gas, if compressibility effects are important, changes in density will cause
the pressure to deviate from the linear behavior of liquids.
The preceding scaling is appropriate for internal flow at low Reynolds numbers.
The essential difference between the large and small Reynolds number cases is
that for small Reynolds number, the pressure is scaled on the viscosity
p∗
p= μU
(3.199)
L

as earlier, whereas in the large Reynolds number case typical of aerodynamic

flows, the pressure is scaled as
p∗
p= (3.200)
ρU 2
121 The Governing Equations for an Electrically Conducting Fluid

Using the same procedure, the mass transfer equation can also be written in
dimensionless form, and
 
∂XA ∂XA ∂XA ∂XA
Sc + ReSc 1 u +v + 2 w
∂t ∂x ∂y ∂z
 
∂ X A Ex ∂ X A Ey ∂ X A Ez
+ z A 1 + + 2 = ∇2 X A (3.201)
∂x ∂y ∂z
where time is scaled on the fluid time scale. The mass transfer time scale is
tm = h 2 /D AB , and the fluid time scale is t f = h 2 /ν so that
tf 1
= 1 (3.202)
tm Sc
for liquids, where Sc = ν/D AB and, for liquids, Sc ∼1000. This means that on
the mass transfer time scale, the velocity is constant, and on the velocity time
scale (much smaller), the mole fraction is constant.
The equation for the electric potential may be nondimensionalized in the same
way. The dimensionless potential φ = φ ∗ /φ0 , and so
Fch 2 
∇ 2φ = − zi X i (3.203)
e φ0 i

where here φ is the dimensionless potential. Taking φ0 = RT /F, it is noted that

Fch 2 F 2 ch 2 h2 c
= = 2 (3.204)
e φ0 e RT λ I
and thus

 2 ∇ 2 φ = −β zi X i (3.205)
i

where  = λ/ h and λ is the Debye length. Note that there is no time derivative
in the potential equation.
Finally, the dimensionless energy equation is given by
 
∂θ ∂θ ∂θ ∂θ
Pr + RePr 1 u +v + 2 w
∂t ∂x ∂y ∂z
 
∂ E2 ∂ E2 ∂ E2 q̇h 2
+ e 1 u +v + 2 w = ∇ 2 θ + Ec + (3.206)
∂x ∂y ∂z kT

where θ is a dimensionless temperature θ = T − T0 /T , where T is a typical

temperature difference in the system; Pr is the Prandtl number, Pr = μc p /k; and
T0 is a reference temperature. The Eckert number is defined by Ec = U 2 /cv T .
The parameter e is defined by

e U E 02 h
e =
kT
and is usually very small.
122 Essentials of Micro- and Nanofluidics

The viscous dissipation function  is obtained by expanding the second term

on the right-hand side of equation (3.137), and
 2  2    
∂u ∂v ∂w 2 ∂v ∂u 2
=2 +2 +2 + +
∂x ∂y ∂z ∂x ∂y
 2  2
∂w ∂v ∂u ∂w
+ + + + (3.207)
∂y ∂z ∂z ∂x
The energy source term q̇ in micro- and nanofluidics is often associated with
what is called Joule heating. When a fluid with a net electric charge experiences
an imposed electric field, there is a rise in temperature due to the resistance caused
by electrical current. Joule heating can cause a temperature gradient across the
channel and a rise in overall temperature The source term due to Joule heating is
given by (Hughes & Gaylord, 1964)

I2 | σe E |2
q̇ = = = σe | E |2 (3.208)
σe σe
where σe is the electrical conductivity of the electrolyte, the proportionality

constant between the current density and the electric field: J = i z i Ni = σe E.
Thus
q̇h 2 h 2 E 02 σe
= | E |2 (3.209)
kT kT

Notation alert: Note that σe in this section is the electrical conductivity defined
previously and not surface tension or surface charge density.

Equation (3.209) may be interpreted physically as

h 2 E 02 σe electrical power density
∼
kT thermal power density
Moreover, the presence of temperature gradients in the system can lead to varia-
tions in properties in one or more directions in the preceding equations in which,
for simplicity, properties have been assumed constant.
The present section has identified the key dimensionless parameters that arise
in the flow of a fluid mixture under electrokinetic effects. These dimension-
less parameters have distinct physical meaning as the ratio of key dimensional
quantities. Thus the Reynolds number
ρU02 h 2 inertial force
Re = =
μU0 h viscous force
and for small Reynolds numbers, it is thus seen that the flow is viscously domi-
nated.
The Schmidt number
μ
viscous diffusion rate
Sc = L2
ρ D AB
=
L2
mass diffusion rate
 123 The Governing Equations for an Electrically Conducting Fluid

and for heat transfer, the Prandtl number

μc p ν viscous diffusion rate
Pr = = =
k α thermal diffusion rate
where
k
α=
ρc p
is the thermal diffusivity. The main dimensionless parameter in electrokinetic
phenomena is the ratio of two lengths
λ
=
h
where λ is the Debye length.

3.13 Fluid, electrostatics, and heat and mass transfer analogies

As noted in the previous section, at the micro- and nanoscale, the process of
dimensional analysis reveals similarities between heat transfer, mass transfer,
velocity, and electrical potential. In this section, three types of similarities between
profiles are discussed:
r Mole fraction and temperature similarity in transient one-dimensional prob-
lems
r Velocity and electrical potential similarity in electro-osmotic flow
r Velocity and temperature at large Reynolds number

In some cases, such as velocity–electrical potential similarity, the two distribu-

tions are identical. Conversely, the distributions may be of the same form and the
problems solved by the same techniques, even if the distributions differ in scale
due to the differing values of the Reynolds, Prandtl, and Schmidt numbers.

3.13.1 Mole fraction and temperature similarity

Suppose that the energy and mass transfer balances are between convection and
diffusion. Suppose also that the pressure gradient vanishes. Then the governing
equations in the case of constant properties for two-dimensional flow are given
by the following:
 
∂u ∂u ∂u
+ Re 1 u +v = ∇ 2u (3.210)
∂t ∂x ∂y
 
∂v ∂v ∂v
+ Re 1 u +v = ∇ 2v (3.211)
∂t ∂x ∂y
 
∂XA ∂XA ∂XA
Sc + Re Sc 1 u +v = ∇2 X A (3.212)
∂t ∂x ∂y
 
∂θ ∂θ ∂θ
Pr + Re Pr 1 u +v = ∇ 2θ (3.213)
∂t ∂x ∂y
124 Essentials of Micro- and Nanofluidics

Note the similarity in all these equations; if the boundary conditions for all the
dependent variables are the same, and Pr = Sc = 1, then u = v = θ = X A ; that
is, all the dependent variables have the same solution.6 The boundary conditions
for internal flow are usually not the same, so these variables will not all be
equivalent. Nevertheless, the equations are analogous in that in many situations,
these equations can be solved numerically by the same procedure. Note that in
liquid flows, 1 < Pr  Sc so that the time scales for each of the velocities and
the mass fraction and temperature are different.
Many heat and mass transfer problems have equivalent solutions, often in
the transient fully developed regime. Consider two-dimensional heat and mass
transfer in a channel in the case that the velocity field (u, v) = (0, 0); then the
governing equations for the temperature and mole fraction are

∂XA ∂2 X A ∂2 X A
Sc = + (3.214)
∂t ∂x2 ∂ y2
∂θ ∂ 2θ ∂ 2θ
Pr = 2+ 2 (3.215)
∂t ∂x ∂y

where we have assumed a single length scale 1 = 1. Suppose that the temperature
satisfies θ = 1 at x = 0 and θ → 0 as x → ∞ and that the sidewalls are insulated:
∂θ /∂ y = 0 at y = 0, 1. Also suppose that θ = 0 at t = 0. Then the heat transfer
is one-dimensional in the x direction, and the solution for the dimensionless
temperature is
 √ 
x Pr
θ(x, t) = erfc √ (3.216)
2 t
Here erfc is the complimentary error function defined by
2 ∞
e−ξ dξ
2
erfc(x) = (3.217)
π x

and, for example, the dimensionless temperature may be defined by

T − Ti
θ(x, t) = (3.218)
T0 − Ti
where the initial temperature is denoted by Ti and T = T0 at x = 0. Similarly,
the mass transfer problem having the same boundary and initial conditions has
the solution
 √ 
x Sc
X A (x, t) = X A0 erfc √ (3.219)
2 t
Note that the independent variable in equation (3.219) is
√
x Sc
η= √ (3.220)
2 t
6
They say “never say never,” but the author cannot think of a case in which this may be true for
liquids since at a minimum, Sc ∼ 1000, and at a maximum, Pr ∼ 50–100 (generous).
125 The Governing Equations for an Electrically Conducting Fluid

a specific combination of the dimensionless variables (x, t); η is called a similarity

variable and indicates the absence of specific length and time scales. Putting this
into dimensional form,
x∗
η= √ (3.221)
2 D AB t ∗
and so the length and time scales cancel. Solutions as in equations (3.216) and
(3.219) are called similarity solutions.
For example, the concentration profile in equation (3.219) corresponds to the
transport of a drug into a membrane at early times. The species flux is given by

∂c A DA
J A (0, t) = −D A = c A0 (3.222)
∂ y y=0 πt
The total amount of species A or the weight gain per unit area delivered to the
half-space y > 0 over time is given by

t DAt
wg (t) = J A (0, t)dt = 2c A0 (3.223)
0 π
Many drugs are bound to proteins, particularly albumin. Albumin is a charged
molecule, and so the problem of the delivery of a charged species to a permeable
membrane is an important one.

3.13.2 Velocity and electrical potential similarity

In the second analogy, recall that in the previous section, it was shown that the
dimensionless electrical body force in the streamwise momentum equation is
given by
FcE 0 h 2  β 
N N
bnd x = zi X i = 2 zi X i (3.224)
μU i=1  i=1
Now assume that there is no pressure gradient and that the Reynolds number is
very small: Re  1. Then, for incompressible flow, the streamwise momentum
equation becomes
β  N
∇2u = − zi X i (3.225)
 2 i=1
Comparing this equation with the equation for the electrical potential, which is
β 
N
∇ 2φ = − zi X i (3.226)
 2 i=1
it is seen that the two equations are identical.
Suppose that the velocity satisfies the no-slip condition at the wall and that
the electric potential satisfies φ(0) = φ(1) = 0. Then both the equations and the
boundary conditions are identical, and on a dimensionless basis, u(y) = φ(y). In
reality, the potential does not vanish at the wall but satisfies φ = ζ at y = 0, 1,
 126 Essentials of Micro- and Nanofluidics

and thus u = φ − ζ . The equivalence of the dimensionless velocity and the

dimensionless potential is important because then measurements of the velocity
are equivalent to measurements of the electric potential. This analogy is also true
for a free surface at y = 1. Can you see why?
Moreover, it is noted that a Debye length of λ = 1 nm in a h = 100 nm channel
gives the same value of  = 0.01 as a λ = 100 nm Debye length in a h = 10 µm
channel so that microscale measurements can be validated by nanoscale
computations, and conversely, nanoscale computations can be validated by
microscale experiments. Note that this similarity analysis does not apply for
unsteady flow because there is no time derivative in the potential equation.

3.14 Other stress–strain relationships

Fluids such as water, air, gasoline, and oil and simple mixtures such as blood
serum (phosphate buffered saline, or PBS), which is essentially a mixture of water
and sodium chloride, are all Newtonian fluids in that the stress varies linearly
with rate of strain. Newton’s law of viscosity is an empirical law, having been
developed by a series of measurements using viscometers and other measurement
techniques. Rheology is the study of the deformation and flow of matter under
an applied stress, and although the Navier–Stokes equations still hold for a non-
Newtonian fluid, it is the stress–strain relationship that changes.
Non-Newtonian behavior is exhibited by more complex mixtures such as sus-
pensions, emulsions, ointments, gels, and polymers. As was noted previously, a
colloid, for example, is a suspension of particles in a liquid such as water, where
the particles range in size from 1 nm to 1 µm. These solutions can be either
dilute, ranging in concentration from 1 µM (micromolar) to 1 mM (millimolar),
or concentrated, which chemists consider to be greater than 1 M. Conversely,
engineers consider mixtures to be concentrated up to the solubility limit, which
can be up to a mole fraction X A = 0.6 in lithium bromide, water mixtures. The
particle phase is called the dispersed phase, and the liquid is the solvent phase.
The simplest forms of the stress–strain relationships for non-Newtonian fluids
are depicted in Figure 3.21. Power law fluids can be either shear thinning (pseu-
doplastic) or shear thickening (dilatant). Power law fluids are described by the
expression
 n
du
τ yx = μa (3.227)
dy
where n is an experimentally determined parameter and μa is termed the apparent
viscosity. This expression is for the case of fully developed flow in a channel. In
the more general case, the full expression for the strain tensor must be used.
There are several objections to the usefulness of the formula for a power law
fluid, one being that n may not be constant over the entire range of interest for a
given fluid. Also, the units of the apparent viscosity depend on the value of n, and
 127 The Governing Equations for an Electrically Conducting Fluid

Some non-Newtonian viscosity curves. Adapted from White (2006).

Figure 3.21

the formula cannot hold for a strain rate less than zero. Some examples of power
law fluids are blood (in some parameter ranges), milk, some colloidal solutions,
and concentrated solutions of sugar and water.
To overcome the last objection, the shear stress is written in a modified form:
 n
du
τ yx = μa α (3.228)
dy

where α is defined by

(du/dy)n
α= (3.229)
| du/dy |n

A better empirical formula that fits most data better than the simple power law
is the Carreau equation (Bird et al., 2002):
μa − μ∞ n−1
= (1 + (λ0 γ̇ )2 ) 2 (3.230)
μ0 − μ∞

where γ̇ = du/dy is the strain rate. Values of the zero shear rate viscosity (μ0 )
and the infinite shear rate viscosity (μ∞ ) and λ for several fluids are given by
Bird et al. (2002).
Many colloidal systems exhibit what is called a Bingham type of behavior, for
which there is a limiting stress below which the material acts as if it were a solid.
Above this stress, the material flows like a fluid. The expression for the shear
stress, in this case, is

du
τ yx − τ B = μa for τ yx > τ B (3.231)
dy

and τ yx = 0 for τ yx < τ B .

The vast majority of the phenomena considered in this text can be analyzed
using the assumption of a Newtonian fluid.
 128 Essentials of Micro- and Nanofluidics

3.15 Mathematical character of partial differential equations

3.15.1 Introduction
The governing equations for an electrically conducting fluid are a complicated
set of nonlinear partial differential equations (PDEs) that are commonly solved
numerically. The mathematical structure of a given PDE is important because it
determines the analytical or computational techniques that are best suited to solve
the problem. It is worthwhile to note that use of the wrong numerical technique
can lead to results that are not only inaccurate but wrong. Moreover, information
concerning existence and uniqueness for nonlinear equations is limited. Gen-
erally, the engineer, in modeling a physical process, assumes that a solution is
possible and develops a model on this basis. If the physical process is properly
modeled, there should be a solution to the mathematical idealization of the pro-
cess. If such a solution cannot be found, either analytically or numerically, then
the assumptions of the physical model should be reexamined.
We begin by describing the different types of linear PDEs and the differences
between each. The nature of each of these equations determines the numerical
method used to solve it, if required.

3.15.2 Mathematical classification of second-order partial

differential equations
Because the most common forms of PDE encountered in the fields described here
are of second order, as exhibited by each of the governing equations discussed in
this chapter, we begin with this type of equation in the classification discussion.
There are three types of partial differential equations: elliptic, parabolic, and
hyperbolic. In an elliptic equation, information, can travel in all directions, and
the solution of the equation at one point affects the solution at all other points
in the domain. The prototype of an elliptic equation is Laplace’s equation. This
equation arises in the slow flow of a highly viscous fluid, in conduction-dominated
heat transfer problems, and in mass diffusion–dominated problems. The electrical
potential is also an elliptic equation.
In a parabolic equation, information can travel in only one direction. Thus the
solution at a given point is dependent only on what the solution is in a given
subset of the domain. The prototype of a parabolic equation is the unsteady heat
or mass equation.
The third type of equation is called hyperbolic, and the prototype is the wave
equation. In a second order wave equation, information is carried in two direc-
tions, called characteristic directions. This type of equation describes the pro-
duction and travel of sound. The occurrence of the second order wave equation
in heat and mass transfer is not common. First order equations arise in the
study of chromatography, the process of separating chemical species by selective
adsorption of the species on a solid substrate. See Rhee et al. (1986, 1989) for
details.
129 The Governing Equations for an Electrically Conducting Fluid

Bearing in mind that the Navier–Stokes equations and the Euler equations are
highly nonlinear, we begin first with a discussion of the linear case. Some of these
ideas can be carried over directly to the nonlinear case, as will be seen. Consider,
then, the most general two-dimensional, linear, second-order equation, defined
by
∂ 2φ ∂ 2φ ∂ 2φ ∂φ ∂φ
a + b + c +d +e + fφ +g = 0 (3.232)
∂x 2 ∂ x∂ y ∂y 2 ∂x ∂y
Then the equation is said to be hyperbolic at a point (x, y) if b2 − 4ac > 0,
parabolic if b2 − 4ac = 0, and elliptic if b2 − 4ac < 0. An equation is hyperbolic,
parabolic, or elliptic in a domain if it is hyperbolic, parabolic, or elliptic at every
point in the domain.
We can define new coordinates (ξ, η) such that equation (3.232) becomes
particularly simple. Let

ξ = ξ (x, y)
η = η(x, y)

define a transformation of coordinates such that

J = ξx η y − ξ y ηx = 0 (3.233)

where J is called the Jacobian of the transformation. The subscripts in this

equation denote differentiation with respect to the indicated variable. Then the
second-order equation becomes

Aφξ ξ + 2Bφξ η + Cφηη + · · · = 0 (3.234)

where the coefficients A, B, and C are related to the transformation by

A = aξx2 + 2bξx ξ y + cξ y2 (3.235)

B = aξx ηx + bξx η y + bξ y ηx + cξ y η y (3.236)

C = aηx2 + 2bηx η y + cη2y (3.237)

The way information is carried by the PDE is embodied in characteristic

curves. A second-order hyperbolic equation has two characteristic curves: a
parabolic equation one characteristic curve, and an elliptic equation no charac-
teristic curves. We discuss the determination of these curves for a hyperbolic
equation next.

3.15.3 Characteristic curves

While hyperbolic equations are rare in micro- and nanofluidics, it is customary
to indicate how to determine the characteristic curves when the equation is
hyperbolic because for a parabolic equation, there is a single characteristic curve
130 Essentials of Micro- and Nanofluidics

that is known (say, time in the unsteady heat equation).7 Then, with a particular
definition of new coordinates ξ and η, we can force A = C = 0, and equation
(3.234) becomes
φξ η = 0 (3.238)
This is called the canonical form of a linear hyperbolic equation. For the wave
equation, A = C = 0 if we choose (ξ, η) such that they satisfy the first-order
equations
ξ x = λ1 ξ y (3.239)

η x = λ2 η y (3.240)

In order for A = 0, we must have

aλ2 + 2bλ + c = 0 (3.241)
This is a quadratic equation that has two distinct roots, say, λ1 (x, y) and λ2 (x, y).
Requiring C = 0 yields the same equation, and the two roots λ1 (x, y) and λ2 (x, y)
are called the characteristic curves of the hyperbolic partial differential equation.
The calculation of the characteristic curves depends on the precise form of the
coefficients of the PDE.
As an example, consider the second-order wave equation in one spatial dimen-
sion, given by
φtt − φx x = 0 (3.242)
where again, the subscripts denote differentiation. In this case, the characteristic
curves are determined by the fact that a = 1 and c = −1 so that
ξ = x +t (3.243)

η = x −t (3.244)
These are the directions along which information propagates in a hyperbolic
equation. The general form of the solution of the wave equation is thus
φ = F(x + t) + G(x − t) (3.245)
which can be obtained directly from the canonical form.

3.15.4 Boundary and initial conditions

In the preceding solution to the wave equation, F and G are arbitrary functions.
To determine these functions, initial and boundary conditions must be given.

7
However, the phenomenon of chemical adsorption discussed in Chapter 7 has been traditionally
analyzed as one or more one-dimensional quasilinear equations. See the two-volume set by Rhee
et al. (1986, 1989).
 131 The Governing Equations for an Electrically Conducting Fluid

For the second-order wave equation in a single spatial dimension, two initial
conditions and two boundary conditions are required; for the heat equation, one
initial condition and two boundary conditions in each direction, and for Laplace’s
equation, two boundary conditions in each direction, are required.

3.15.5 Classification of the governing equations of micro- and nanofluidics

It is well known that the steady Navier–Stokes equations are elliptic because the
viscosity ν > 0; however, the boundary layer equations are parabolic because
a = 0 if x is considered to be the streamwise direction. The steady energy
and mass transfer equations are also elliptic, as is the Poisson equation for the
electrical potential. The unsteady flow and their corresponding analogies in
heat and mass transfer discussed in Section 3.13 are parabolic. The occurrence
of hyperbolic equations in micro- and nanofluidics is somewhat less common
but can occur in unsteady mass transfer when diffusion is minimal. Of course,
first-order equations in the time domain are by their nature hyperbolic and
occur in a variety of problems including chromatography as described by
Aris (1956, 1959).
Most of the situations described in this book are elliptic in nature; ordinary
second-order differential equations of the boundary value type are elliptic by
nature.

3.16 Well-posed problems

In engineering and applied science, mathematical models of physical processes

must be accurate and robust, characteristics of the solution that are normally
termed stable. Moreover, though it is common that there may be many different
mathematical models for a given physical process, when specific assumptions
are made, the solution to the problem should be unique; that is, though differ-
ent assumptions lead to different mathematical models, the solution to a single
mathematical model of a physical process must be unique.
These considerations lead to the definition of a well-posed problem. A mathe-
matical model of a physical process is well posed if

1. A solution exists
2. The solution is unique
3. The solution is continuously dependent on the data

The last requirement is especially important. This means that if we change a

boundary or initial condition by a small amount, the solution changes by a small
amount. Moreover, in engineering problems, this concept should be extended to
small changes in material properties such a viscosity of a fluid, thermal conduc-
tivity of a solid, and Young’s modulus. The concept of continuous dependence on
 132 Essentials of Micro- and Nanofluidics

the data is thus related to the stability of the solution. Stability of mathematical
models is a subject for an entire book.

3.17 The role of fabrication, experiments, and theory in micro- and nanofluidics

The field of micro- and nanofluidics inherently involves the consideration of

length scales on the order of millimeters, the size of reservoirs in drug delivery
systems, at the very least, down to the nanometer scale, the size of the pores in
a nanopore membrane. Because of the small length scales involved, determining
precisely the dimension of the fluid conduits is often difficult. Thus fabrication of
microfluidic devices requires precise control at every step of the fabrication pro-
cess, and even with precise control, fabrication tolerances are much greater, rela-
tively speaking, than in macroscale manufacturing (Gad-el Hak, 2001; Nguyen &
Wereley, 2002; Lee, 2006). Moreover, there are inherent limitations to what can
be measured at the submicron scale. Consequently, research at the microscale on
down requires that the fabricator, experimentalist, and analyst work closely and
in tandem in designing micro- and nanofluidic devices.
Most, if not all, experimental methods appropriate for microfluidic appli-
cations use tracer particles. Many of the experimental methods used today in
the microfluidics area originated in high Reynolds number flows; examples
include laser Doppler velocimetry (LDV), molecular tagging velocimetry (MTV,
Lempert et al., 1995), and particle image velocimetry (PIV, Adrian & Yao, 1985;
Adrian, 1991). These experimental techniques can be distinguished by the type
of tracer used and the spatiotemporal characteristics of the measurement: LDV,
for example, yields a point measurement of velocity with excellent temporal res-
olution (up to MHz), whereas PIV, the most commonly used velocimetry method,
gives two or three velocity components in a 2-D plane (2D-2C or 3D-3C), while
volumetric (3D-3C) PIV is being extended from the macroscale to the microscale
(Yoda, 2006). MTV, conversely, typically gives a measurement of up to two of
the three velocity components over a one-dimensional line or a two-dimensional
plane. An excellent reference book that describes these methods in more detail is
the monograph edited by Breuer (2005).
Given the overall dimensions of micro- and nanofluidic channels, <10–
100 µm, nearly all flows in the micro- and nanofluidic regime are laminar and
often fall within the Stokes flow regime. Moreover, to make accurate flow mea-
surements in such small devices, with their limited space, the method should
also be nonintrusive. Thus intrusive methods such as atomic force microscopy
(AFM), which are useful for probing static surfaces and fluids, are not appropriate
for micro- and nanofluidic applications because the fluid will not remain static,
especially if the AFM tip is vibrating.
The previously mentioned experimental visualization techniques have a mini-
mum spatial resolution of about 50 µm for LDV and 500 µm for PIV in macroscale
applications (Yoda, 2006). Spatial resolutions as fine as ∼2.5 µm have been
 133 The Governing Equations for an Electrically Conducting Fluid

achieved with LDV and PIV for applications at the

microscale (Czarske et al., 2002). However, even
these “best case” limits exceed the typical pore
dimensions in a nanopore membrane by about 4
orders of magnitude. Thus it is impossible with
these techniques to measure velocity, concentra-
tion, and temperature profiles in a channel smaller
than these resolution limits of a few microns. The
Sketch of the field of view in the conclusion must be that these two velocimetry
Figure 3.22
experimental setup used by Yoda and methods should be interpreted as bulk flow mea-
coworkers (Sadr et al., 2004, 2006). surement techniques at the nanoscale. The level
of resolution of these methods at the micro- and
nanoscale means that developing analytical and computational models for micro-
and nanoscale flows is not only necessary, but essential.
Despite this limitation, both PIV and MTV are currently used to probe
microscale flows in gases and liquids, including electro-osmotic flow (Devase-
nathipathy & Santiago, 2005; Sadr et al., 2004, 2006). For example, in µPIV,
an aqueous solution is often seeded with 200 nm–1 µm diameter fluores-
cently dyed polystyrene spheres having a density just slightly greater than water
(ρ ∼1050 kg/m3 ). Illumination by light at a wavelength required to excite the
fluorescent tracer particles is viewed by a microscope and then recorded by
a digital (CCD = charge-coupled device, CMOS = complementary metal oxide
semiconductor) camera.
Typically, the depth of the view (the dimension normal to the field of view) in
µPIV is on the order of 1–2 µm, and the velocity field obtained from a 2-D plane
is averaged over the depth of correlation, which is at least 2 µm. Sadr et al. (2004,
2006) and Wereley and Meinhart (2010) have extended the µPIV technique to
the submicron scale using evanescent-wave illumination so that the depth of
view in their nPIV is ∼200 nm. Nano-PIV is, however, limited to accessing the
region near the wall, whereas µPIV and MTV are more bulk flow techniques. By
comparing pairs of images, all these techniques then yield the average velocity
of the fluid within the field of view. A sketch of the field of view used in Yoda’s
experiments is depicted in Figure 3.22. It should also be mentioned that MTV and
µPIV use volume illumination, whereas nPIV uses evanescent wave illumination
with an intensity that decays exponentially away from the wall.
In all these methods, the velocity along the optical axis should be negligible for
these methods to accurately measure the tracer displacement in the primary flow
direction, which must be at a 90◦ angle to the optical axis. The tracer particles
must be carefully tuned to the dimensions of the interrogation window, the field of
view, and the minimum spatial resolution of the image (Yoda, 2006). Furthermore,
the observation time between the two images must be small enough to permit a
reasonably accurate estimate to the velocity of the particles, which is assumed
to be the local fluid velocity. It should be noted, however, that Brownian motion,
which statistically has a negligible effect on particle motion in the bulk, varies
 134 Essentials of Micro- and Nanofluidics

with distance from the wall and may therefore have a nonstochastic influence on
particle motion near walls (Sadr et al., 2007; Yoda, 2006).

3.18 Chapter summary

In this chapter, the governing equations for mass momentum and heat and mass
transport and for electrostatics have been derived. The equations have been pre-
sented in both dimensional and dimensionless form. Dimensional analysis will
prove to be a powerful tool for the analysis and solution of the wide range of
problems addressed in this text. The main dimensionless parameters associated
with these problems have been identified, and their physical meaning has been
explained.
In presenting the governing equations and their boundary conditions, it is
evident that there are a number of similarities between the fields of heat mass,
momentum, and mass transport in mixtures. All equations but the basic equa-
tion for electrostatic potential are convective–diffusion equations in some form.
Moreover, the boundary conditions, especially those associated with heat and
mass transport in mixtures, are similar, if not identical, in form. In fact, it
will be seen that in fully developed electro-osmotic flow, it is common for
the dimensionless velocity to be equivalent to the electrical potential up to
a constant. This situation is described in Chapter 9 and leads to the conclu-
sion that the electrical double layer structure in terms of the electrical potential
can be determined by measurement of the velocity within the electrical double
layer.
The existence of key dimensionless parameters has been emphasized in this
chapter, and dimensional analysis has revealed several analogies: those situations
that result in similar, if not identical, solutions to problems in the different fields
of fluid mechanics, heat and mass transfer, and electrostatics. This fact inevitably
leads to a comment on notation. In a multidisciplinary book of this type, it is
impossible to standardize the notation completely in that the same symbol is used
for the same quantity throughout. For example, the symbol μ is used for vis-
cosity by mechanical engineers and for electro-osmotic mobility by chemists. In
addition, the symbol u may be a dimensional or dimensionless velocity or, as has
been seen, internal energy. For this reason, I have tried to make clear the meaning
of each symbol as it appears; in addition, I have endeavored, with I think some
success, to make each section of the book as self contained as possible. Finally,
in this regard, I must emphasize that the appearance of dimensionless parameters
in an equation signifies that all quantities in the equation are dimensionless. Con-
versely, the appearance of properties, such as viscosity and electrical permittivity,
in an equation signifies that each quantity in the equation is dimensional, with
each term in the equation having the same units. This philosophy will continue
throughout the text.
135 The Governing Equations for an Electrically Conducting Fluid

The governing equations have been presented in their full three-dimensional

and unsteady form. However, the reader will struggle to find a situation where
these equations are solved in their full form. Where appropriate in the following
chapters, simplifying assumptions are made to reduce the complexity of the
equations while, at the same time, providing a reasonable solution to a given
problem. For example, in the design of nanopore membranes, it is often possible to
achieve a reasonable solution for determining flow rate using the fully developed
form of the equations since the entrance length is very short. Thus Section 3.6
and the sections immediately following may effectively be skipped in many
situations.
The fluid has been assumed to be a continuum, and it will be shown later
in the text that this is the case as long as the smallest dimension of a channel
or membrane is no smaller than ∼20 solvent diameters. Molecular simulation
methods that must be used for smaller channels are described in Chapter 11.
That being said, most nanopore membranes for most applications are bigger than
20 solvent diameters. The exception may be membranes used for desalination,
for which the pore size can be on the order of 1–3 nm. Such a process is called
ultrafiltration (Zeman & Zydney, 1996). In the majority of the remaining chapters
of this text, the important problems in fluid mechanics, heat and mass transfer,
and electrostatics and dynamics are solved using the these governing equations
as a starting point.

EXERCISES
3.1 The velocity field near the core of a tornado is sometimes approximated by
the velocity field

Q k
u=− îr + î θ
2πr 2πr

where (r, θ) are polar coordinates, r 2 = x 2 + y 2 , and k and Q are constants.

a. Write the velocity components in Cartesian coordinates.
b. Determine the equation describing the streamlines for incompressible
flow, and plot several streamlines.
3.2 Sketch the streamlines for the two-dimensional flow

u = αx v = −αy

with α as a positive constant.

3.3 Consider air to be a binary mixture of 79 percent nitrogen and 21 percent
oxygen on a molar basis. Find the mass concentrations.
3.4 Consider a mixture of water and sodium chloride to consist of 95 percent
water and 0.05 percent sodium chloride on a molar basis. Find the mass
concentrations, and determine the ionic strength of the mixture.
136 Essentials of Micro- and Nanofluidics

3.5 Starting with the fully three-dimensional and steady Navier–Stokes equa-
tions in dimensional form, determine the equations governing fully devel-
oped flow in a wide channel. Repeat for a cylindrical tube.
3.6 The velocity distribution in a channel is parabolic, as we know, and the
expression for the velocity is
1 dp 2
u(y) = (y − hy)
2μ d x
where h is the channel height. Write this equation in dimensionless form,
and identify the velocity scale.
3.7 Water flows continuously down a very small crack in cement under the
action of gravity and the equation governing flow is
∂ 2u
μ = −ρg
∂ y2
where μ is the dynamic viscosity, ρ is the density and g is the acceleration
due to gravity. There is no pressure gradient. Assume the crack is rectangular
and its height h is much smaller than the width W .
a. Using the channel height h as the length scale, and V as the velocity
scale, write the equation in non-dimensional form and identify a possible
velocity scale V .
b. Using appropriate conditions at the channel walls (which are solid) at
y = 0, h determine the velocity distribution in m/sec and the volume
flowrate per width of the channel. The channel height h = 0.005 cm.
What is the maximum velocity?
3.8 The vorticity is defined as the curl of the velocity field

ω =∇ ×u

Show by direct calculation that the vorticity satisfies

Dω
= ν∇ 2 ω
Dt
Write the equation in dimensionless form. What are the units of vorti-
city?
3.9 Neglecting internal energy generation and viscous dissipation, show that
the two-dimensional, steady state energy equation in a channel for small e
reduces to
 
∂θ ∂θ
RePr 1 u +v = ∇2θ
∂x ∂y
where 1 is as defined in the text. If the temperature boundary conditions
are T = T0 at x = 0, L and t = 0 at y = 0, with
∂T
−k = q0 at y = h
∂y
137 The Governing Equations for an Electrically Conducting Fluid

and q0 = constant, write the boundary conditions in dimensionless form.

Discuss the qualitative features of the problem based on the relative mag-
nitude of h and L. Would you expect the solution to depend on x? In what
parameter regime? Discuss the choice of the temperature scale.
3.10 Write the corresponding analogue of this problem for the concentration,
assuming the walls at x = 0, 1 are impermeable and discuss its physical
meaning. Would you expect the solution to depend on x? Discuss.
3.11 The governing equation for mass transport in a channel is given by

∂c A ∂c A
u +v = D AB ∇ 2 c A
∂x ∂y

A heterogenous chemical reaction of first order takes place at the surface

y = 0 so that the boundary condition is

∂c A
−D AB = kr c A at y = 0
∂y
∂c A
=0 at y = h
∂y

and at x = 0. Write the equation in dimensionless form, and identify the

Damkohler number as the dimensionless group Da = kr h/D AB . Assume
that the velocity scale is the average velocity in a Poiseuille flow. Can the
concentration distribution be fully developed?
3.12 In some situations, electrical potential and the velocity are equivalent. Con-
sider fully developed channel flow, and begin with the governing equations
and boundary conditions for the potential and the velocity in the form
(assuming no pressure drop)

d 2φ∗
e = ρe
dy ∗2

where ρe = −F i z i ci and

d 2u∗
μ = bx∗
dy ∗2

where bx∗ = ρe E 0 . Define dimensionless variables

y∗
y=
h
u∗
u=
U0
φ∗
φ=
φ0
138 Essentials of Micro- and Nanofluidics

and show explicitly that the equations for the potential and the velocity are
equivalent for fully developed flow if the velocity scale is taken to be
e E 0 φ0
U0 =
μ
where E0 is the applied electric field and φ0 = RT /F, and where the
electrical potential φ represents the perturbation to the ζ potential at the
wall, that is, the electric potential satisfies the boundary conditions φ(0) =
φ(1) = 0.
3.13 Using the fluid–electrostatics analogy, show that the friction coefficient
2
Cf = −
Re
where
σh
=
e φ0
is a dimensionless parameter.
3.14 Suppose there is an alternating electric field E ∗ (t) imposed on an electrolyte
solution in a wide, two-dimensional channel in the x ∗ direction. Show that
on the concentration time scale, the governing equations are

∂XA ∂2 X A β  n
∂φ ∂ X A
− + 2 zAX A zi X i − z A =0
∂t ∂y 2  i=1
∂y ∂y

∂ 2u β E x (t)  ∂u
+ zi X i = α
∂y 2  2 ∂t
∂ 2φ β E x (t) 
+ zi X i = 0
∂y 2 2
where α = 1/Sc. For large Schmidt number, the velocity and potential are
still approximately equal (as long as the boundary conditions are the same
for each) at long times.
3.15 Repeat the previous problem, except scale time on the fluid time scale. What
do you conclude about the similarity of the velocity and electric potential?
3.16 According to White (2006), Prandtl, in 1932, proposed the equation

d 2u du
 + +u =0
dy 2 dy
for   1 as a model for the velocity in a boundary layer. Take the domain
of solution to be y = 0 to y = 5 and the boundary conditions u(0) = 1 and
u(5) = 1. (I have modified the boundary conditions somewhat from what is
in White. ) Solve this equation for  = 1, 0.01, 0.0001 analytically. What
do you find? A numerical solution of this equation would be very difficult
for small . Is there a way to rescale the equation for small  to remove this
difficulty?
139 The Governing Equations for an Electrically Conducting Fluid

3.17 The concentration of a pollutant in a fluid is found to be

c(x, y, t) = βx 2 ye−αt

for the flow field given by

u = αx v = −αy

Calculate Dc/Dt. What do you find? Plot lines of constant concentration

at several times.
 4 The Essentials of Viscous Flow

4.1 Introduction

As discussed in Chapter 1, many of the current or planned microdevices contain

rectangular channels arranged in an array; such channel systems are termed
nanopore membranes. For example, the nanopore membranes depicted in Fig-
ure 1.1 contain up to 30,000 channels, depending on the channel height, which
is the nanoscale dimension. Scanning electron microscope (SEM) images of
nanopore membranes fabricated by Dr. Shuvo Roy’s group at the Cleveland
Clinic1 are depicted in Figure 4.1. You will recognize Figure 4.1(a) as Figure
1.1(b), and is repeated for clarity. The nanopore membrane is fabricated so as
to contain channels of a uniform nanoscale height. They fabricated membranes
with a variety of pore sizes, ranging from h = 6.3 nm to h = 91 nm, and there
are 11, 500 pores per membrane. Similar membranes are now being used in the
development of an artificial kidney (Fissell, 2006). Obviously, it would be difficult
to model the flow in such an array from first principles.
There are at least two ways to overcome this problem: (1) treating the membrane
as a porous media problem, the membrane being characterized by a permeability
and a porosity, and (2) investigating the flow in each channel separately, assuming
that the flow behaves similarly in each individual channel. It is the latter approach
that is explored here.
In the design of devices employing nanopore membranes, there are questions
that must be answered. Consider two limiting cases, in which the membrane is
to be used for delivery of a substance and in which the membrane is used for
filtration of a solute of specified size; that is, some solutes are allowed to pass
through the membrane, whereas others are not, for example,

Design question 1: A device must be designed to deliver a specified amount

of a certain drug imbedded in a particle or protein having a nominal size of
a = 50 nm. What should be the size of the pores in this membrane?

Design question 2: A device needs to filter all particles or molecules below

3 nm nominal size, while retaining all particles or molecules above 3 nm in size.
1
Dr. Roy is now in the Department of Bioengineering and Therapeutic Sciences at the University
of California, San Francisco.

140
 141 The Essentials of Viscous Flow

(a) (b)

Microscopic images of nanopores. (a) Top view of membrane; from Conlisk et al. (2009). (b) Top view of a single
Figure 4.1
nanopore. Approximate length scale is shown. There are 11, 500 pores per membrane. Courtesy of Shuvo Roy
and Cleveland Clinic. Used with permission.

The particles are charged. Should the membrane be charged? How much? What
should be the size of the pores in this membrane?

Nanopore membranes are often designed to carry fluids containing charged

species. Electrolyte solutions have been discussed earlier, and in this chapter, a
number of internal viscous flow problems relevant to microfluidics and nanoflu-
idics are described. In particular, the reason that electrokinetic phenomena are
used so often in micro- and nanofluidics is evident from the calculations described
in this chapter (see also Figure 1.5). It is useful to point out that most channels
of interest have one very small dimension, perhaps as small as tens of nanome-
ters, whereas the other two dimensions are usually on the micron scale. Thus
one-dimensional flow is a good approximation for most membranes.
In this chapter, pressure-driven flow is considered exclusively, and only steady
flows are presented. Once the fundamentals of electrostatics and electrochemistry
are described, electrokinetically driven flow of an electrolyte solution, which also
involves principles of mass transfer and electrostatics, will be combined with
the results of this chapter. In the next section, the structure of steady flow in a
pipe or channel is described, followed by the presentation of Poiseuille flow in a
channel and pipe. This is followed by a discussion of the slip boundary condition
and the velocity profile for both gases and liquids. Next a series of other viscous
flow problems are discussed, including thin film flow, the Blasius boundary layer,
fully developed and developing suction flows, lubrication, flow around a bubble,
Stokes flow past a sphere, and a simple model for blood flow.

4.2 The structure of flow in a pipe or channel

The structure of flow in a pipe or a channel is depicted in Figure 4.2 for a pipe.
For a channel, replace r with y and D with h or 2h. Suppose that there is a
large reservoir upstream of the entrance to the pipe or channel, as is the case
 142 Essentials of Micro- and Nanofluidics

Viscous
Fully
Developed
Inviscid
Developing
Boundary Layers

U r D
0 z

Entrance Length
Laminar

Flow through a pipe, showing the entrance region followed by the fully developed region.
Figure 4.2

in many of the microdevices already described.2 Then the velocity profile at

the inlet will be relatively uniform. To adjust to the presence of the solid walls,
the fluid at the wall must assume the wall velocity and the no-slip condition,
which, in the case of a stationary wall, is zero. This adjustment region, is the
entrance region, and its length, the entrance length, is a function of the Reynolds
number. For a pipe, the entrance length for laminar flow is conventionally taken
to be (Langhaar, 1942)
Le
= 0.06 Re (4.1)
D
for large Re, in laminar flow. Note that L e → 0 as Re → 0. A formula that
includes the necessary short entry region as Re → 0 is given by Chen (1973),
Le 0.6
= + 0.056 Re (4.2)
D 1 + 0.035 Re
for a circular pipe, and in the limit Re → 0, L e /D = 0.6. A similar formula due
to Atkinson et al. (1969) is given by
Le
= 0.59 + 0.056 Re (4.3)
D
In the entrance region, the flow resembles a boundary layer because it is
unbounded on one side, at least for large Reynolds number.
For a rectangular channel, the hydraulic diameter is given by
4hW
Dh = (4.4)
2W + 2h
which is just 4 times the cross-sectional area divided by the wetted perimeter. For
a wide channel, Dh ∼ 2h, and the preceding formulas hold for channels provided
that the hydraulic diameter is used for the length scale.

2
For the purposes of this discussion, the pipe or channel walls are assumed to be uncharged.
 143 The Essentials of Viscous Flow

Side View
L

Flow
h
y,v
x,u L>>h
z,w
End View
W

z,w
W>>h
Geometry for Poiseuille flow in a channel.
Figure 4.3

After the boundary layers meet, the velocity no longer depends on x, and a
simple parabolic profile for the velocity can be obtained for the case of laminar
flow. This region is called the fully developed regime, and many microflows
discussed in this book can be assumed to be fully developed because at low
Reynolds numbers, the entrance region is very short.

4.3 Poiseuille flow in a pipe or channel

Consider the geometry of the channel depicted in Figure 4.3 for a channel so that
r is replaced with y. The region of interest is the fully developed zone in a very
wide channel so that the velocity does not depend on the streamwise variable x
nor on the spanwise variable z because W  h (see Figure 4.3). For the purpose
of a physical interpretation of the relationship between pressure drop, flow rate,
and viscosity, the dimensional form of the equations is used, and incompressible
flow is assumed.
The Navier–Stokes equations reduce to
∂ 2u ∂p
μ = − Bx (4.5)
∂ y2 ∂x
∂p
− By = 0 (4.6)
∂y
Note that the flow does not accelerate because the convective terms vanish iden-
tically. The velocity distribution in the channel is one-dimensional, and the flow
is said to be hydrostatic in the y direction because a velocity does not appear in
the equation (v = 0). In this section, the body force in the streamwise direction
Bx = 0. Later, it will be seen that this body force will be nonzero for electro-
osmotic flow driven by an external electric field.
 144 Essentials of Micro- and Nanofluidics

Average Velocity
Velocity u=u(y)

y,v
x,u
Shear Stress Pressure
v=0!
Velocity, shear stress, and pressure distribution in fully developed flow in a channel. Velocity and shear stress
Figure 4.4
are plotted as functions of y and the pressure as a function of x .

If the body force term is conservative, as in the case of gravity, then

f g = −g∇z, and it can be absorbed into the pressure gradient term; thus, if
the conservative body force is neglected, then p = p(x). A function of x cannot
be equal to a function of y unless the function is constant. Thus we can easily
integrate the governing momentum equation

1 d p 2 C1
u= y + y + C2 (4.7)
2μ d x μ
where C1 and C 2 are constants. Applying the no-slip condition on the two walls
y = 0, h, we can determine C1 and C2 = 0 so that
 2
h2 d p y y
u= − (4.8)
2μ d x h h

and the flow is thus in the direction of decreasing pressure. This Poiseuille flow
velocity distribution is a simple parabolic distribution originally suggested by
Jean L. M. Poiseuille (1799–1869), and thus the shear stress is linear:
 
∂u dp 1
τ (y) = μ = y− h (4.9)
∂y dx 2

The velocity shear stress and pressure gradient are depicted in Figure 4.4.
Note that the pressure is linear: p = p(x) = p0 + ( p L − p0 ) x/L and d p/d x =
( p L − p0 )/L = −p/L = constant.
Jean L. M. Poiseuille was a practicing physician who realized that “progress in
physiology demands a knowledge of the laws of motion of the blood” (p. 87)
Tokaty (1971). Thus one of the most famous velocity distributions in fluid
mechanics was developed by a medical doctor.
The maximum velocity occurs at the centerline and has the magnitude

h2 d p
u max = − (4.10)
8μ d x
145 The Essentials of Viscous Flow

Table 4.1. Results for the calculation of the flow rate through a channel
m m 3
h (µm) D H (µm) V ( sec ) Q ( sec ) Re
−5 −14
1 2 8.3 × 10 8.3 × 10 1.7 × 10−4
23 45 0.04 1.0 × 10−9 2.0
45 86 0.17 7.6 × 10−9 14.5
78 145 0.51 4.0 × 10−8 73.0
100 182 0.83 8.3 × 10−8 151.5

The volume flow rate in m3 /sec is

h h3 W d p
Q= uW dy = − (4.11)
0 12μ d x
and note that flow rate is inversely proportional to viscosity and proportional to
h 3 . In terms of the pressure drop,
Q h 3 p
= (4.12)
W 12μL
12μL Q
p = (4.13)
W h3
If the flow rate is held fixed, the required pressure drop increases dramatically as
h → 0; the preceding formula for p is used to create Figure 1.5. In dimension-
less form,
p 24 L
= (4.14)
ρ U2
2
Re h
where f = 24/Re is called the friction factor. The average velocity is U = Q/A,
where A = W h is the cross-sectional area. As the channel height decreases, the
required pressure drop to get a fixed flow rate increases dramatically. The average
velocity is U = Q/A, where A = W h is the cross-sectional area. It should be
noted that the Reynolds number in this formulation is based on the channel height
h. In channels of different shape, a hydraulic diameter should be used.
As a numerical example, suppose the pressure drop p = 104 N/m2 (0.1 atm)
for water with the width W = 10−3 m and length L = 0.01 m. The volume flow
rates for various channel heights may easily be calculated; these values are shown
in Table 4.1.
The corresponding equations for axisymmetric flow in a tube are (see Appendix
B)
1 ∂r u ∂w
+ =0 (4.15)
r ∂r ∂x
   
1 ∂ ∂w ∂ 2 w 1 ∂w ∂p
μ r =μ + = − Bx (4.16)
r ∂r ∂r ∂r 2 r ∂r ∂x
∂p
− Br = 0 (4.17)
∂r
 146 Essentials of Micro- and Nanofluidics

w (r) = wmax 1 − r 2
2
R

R r

wavg

wmax = 2 wavg

Geometry for flow in a cylindrical pipe or tube.

Figure 4.5

Equation (4.16) may be integrated twice, and apply the boundary condition that
∂u/∂r = 0 at r = 0 to ensure that the velocity is finite at r = 0 and
 2
R2 d p r
w(r ) = − 1− (4.18)
4μ d x R
where R is the radius of the tube and the volume flow rate is
π R4 d p
Q=− (4.19)
8μ d x
and the corresponding pressure drop is
128μL Q
p = (4.20)
π D4
where D is the diameter, and in dimensionless form,
p 64 L
U2
=
ρ 2 Re D
Note that for both the channel and the tube (Figure 4.5), the average velocity
scales on the pressure drop as
dp
. U =C (4.21)
dx
In practice, the constant C can be chosen to coincide with the average velocity
or the maximum velocity in the channel or tube. For example, if the maximum
velocity is chosen, for the channel,
h2
C =− (4.22)
4μ
and it is thus seen that wmax = 2wave .

4.4 The velocity in slip flow

4.4.1 Gases
Fluid slip at a wall arises when the mean free path of the molecules is comparable
to the geometrical length scale. When a gas becomes rarified, that is, when the
147 The Essentials of Viscous Flow

mean free path of the molecules becomes of the same order as a given space
dimension, then
λ
Kn = = O(1) (4.23)
h
Another way for the Knudsen number to become O(1) is for the channel height
h to become small at atmospheric conditions. The mean free path of air is around
λ ∼ 60–70 nm, depending on temperature.
Let us consider the case in which the Knudsen number is in the range such
that the flow may be defined as continuum flow with slip. Let us also assume that
the flow is low speed so that the density remains constant. Then the governing
equation

∂ 2u ∂p
μ = − bx (4.24)
∂y 2 ∂x

but now with the slip condition given by equation (3.153). The solution in the
absence of body forces is, for σv = 1,
 2
h2 d p y y λ λ
u= − − for 1 (4.25)
2μ d x h h h h

and note that the velocity distribution is merely offset by a fixed distance. Note that
to leading order in y, the velocity vanishes at y = −λ, and this is the definition
of slip length.
The shear stress is thus the same as for no slip, and
 
∂u dp 1
τ (y) = μ = y− h (4.26)
∂y dx 2

Note that the velocity is still a maximum at the centerline and that the pressure
drop is linear. The volume flow rate is given by
 
W dp 3 λ
Q=− h 1+6 (4.27)
12μ d x h

Note here that for continuum approximation to hold, K n < 0.1; above this value,
the flow is in the transition regime, where noncontinuum methods must be used,
as depicted in Figure 3.17.

4.4.2 Liquids
Using the boundary condition suggested by Thompson and Troian (1997) in
equation (3.159), the velocity is
 2  
h2 d p y y h d p −1/2 h d p
u= − − L 0s 1 + γ̇c−1 (4.28)
2μ d x h h 2μ d x 2μ d x
 148 Essentials of Micro- and Nanofluidics

h h
x

Vapor Vapor

L
g

Liquid Film

A thin liquid film falling under gravity.

Figure 4.6

The volume flow rate is easily calculated, as we have done before:

⎛  −1/2 ⎞
W h3 d p ⎝ h 2
d p ⎠
Q=− 1 + 6L 0s 1 + γ̇ −1 (4.29)
12μ d x 2μ d x c

Note that as in gases, the flow rate is increased by a constant amount.

The validity of the no-slip condition has been verified at macroscale. In liquids
at microscale and below, recent experimental work has shown that slip can occur
in liquids as a result of surface chemistry, especially at hydrophobic surfaces,
as discussed in Section 3.11. It does need noting that the physical mechanisms
for liquid slip are still a matter of some debate; an excellent discussion of the
historical view of liquid slip is given by Karniadakis et al. (2005).

More on Slip!
According to Karniadakis et al. (2005), evidence of fluid slip between solid
surfaces and water was found as early as the 1860s. However, other experiments
refuted these results and Karniadakis et al. (2005) suggest that experiments by
Whetham around 1890 led to the conclusion that the no-slip condition holds at
hydrophillic surfaces. Note that this is long before Prandtl’s experiments that took
place around 1904. In fact, Day (1990), in his analysis of the validity of the no-slip
condition, does not even mention Prandtl.

4.5 Flow in a thin film under gravity

In the same way we study Poiseuille flow, consider the case of a fluid running
down a solid wall under gravity, as depicted in Figure 4.6. If the entire region is
at constant pressure, then there will be no pressure gradient in the liquid film. For
149 The Essentials of Viscous Flow

fully developed flow, the governing equation for the streamwise velocity is

∂ 2u
μ = −bx (4.30)
∂ y2
∂p
− by = 0 (4.31)
∂y
with bx = ρg and b y = 0. The boundary conditions are

u=0 y=0 (4.32)

and at the free surface,

∂u
= 0 at y = h (4.33)
∂y
As noted in the previous chapter, this second boundary condition indicates that
the shear stress at the free surface is zero and comes from a stress balance at
the interface in the absence of the effects of surface tension. It is an approximate
condition valid when, typically, the vapor or gas has a dynamic viscosity much
less than that of water and the film thickness is constant, rendering surface tension
effects negligible.

A falling liquid film.

The solution for the velocity is obtained in the same way as for Poiseuille flow,
and
   
ρgh 2 y 1 y 2
u= − (4.34)
μ h 2 h

and so the appropriate velocity scale is U = ρgh 2 /μ. In dimensionless form,

1
u = Y − Y2 (4.35)
2
 150 Essentials of Micro- and Nanofluidics

where Y = y/ h. The velocity is sketched in Figure 4.6. It is useful to note that if

the film is thin compared to the diameter of a cylinder, the same solution holds
in cylindrical coordinates.
The flow rate is given by
h ρgh 3 W 1
Q=W udy = = U hW (4.36)
0 3μ 3
where W is the width of the film into the page. Note that the h 3 dependence seen
in Poiseuille flow carries over to this film flow. If the film thickness is h = 1 µm,
then the velocity scale for water is U = 10−5 m/sec.
Care must be used when applying this very simple velocity distribution because
it has been assumed that the film thickness h is a constant or nearly so. However,
at the micro- or nanoscale, if the film surface is even slightly curved, surface
tension may play a role in the fluid dynamics and destroy the one-dimensional
nature of the flow profile.

4.6 The boundary layer on a flat plate

The function of a boundary layer, an example of an external flow that is

unbounded on one side, is to bring the velocity to zero on a solid body. Boundary
layers occur at high Reynolds numbers, and at first glance, it might not appear
appropriate to discuss the Blasius boundary layer in this text on microfluidics,
for which the focus is internal flow. However, given Prandtl’s role in the demon-
stration of the validity of the no-slip condition, it is useful to present. Moreover,
boundary layer–type behavior, in which the velocity varies rapidly in a very thin
region, occurs in electro-osmotic flow for thin electric double layers, to be dis-
cussed in detail in Chapter 9. In addition, the entrance flow in a duct at high
Reynolds number is essentially an external flow governed approximately by the
boundary layer equations, as we have just seen.
Blasius, a student of Prandtl, was the first to solve the problem of laminar flow
past a flat plate. The coordinate system is depicted in Figure 4.7. The Navier–
Stokes equations in two-dimensional steady flow are given by
∂u ∂v
+ =0 (4.37)
∂x ∂y
∂u ∂u ∂p 1 2
u +v =− + ∇ u (4.38)
∂x ∂y ∂x Re
∂v ∂v ∂p 1 2
u +v =− + ∇ v (4.39)
∂x ∂y ∂y Re
The pressure is scaled on the dynamic pressure, or
p∗
p= (4.40)
ρU∞2

and Re = ρU∞ L/μ.

 151 The Essentials of Viscous Flow

Inviscid Flow

Boundary
Layer

y
Uoo −1/2
1/2
O(Re )
δ(x)~x

L
Viscous Flow

Coordinate system for a boundary layer on a flat plate.

Figure 4.7

The boundary layer is very thin, so let us see if we can simplify the Navier–
Stokes equations. From the continuity equation, using dimensional analysis,
∂u U∞
∼ (4.41)
∂x L
∂v v
∼ (4.42)
∂y δ
where δ, the boundary layer thickness, is the scale in the y direction. Thus the v
velocity is
δ
v ∼ U∞ (4.43)
L
and so because δ  L, the velocity v is very small. In fact, because of this, the
second momentum equation reduces to
∂p
=0 (4.44)
∂y
that is, the boundary layer is so thin that the pressure does not vary across it.
By the same arguments, since δ  L,
∂2 ∂2
 (4.45)
∂x2 ∂ y2
and so the streamwise momentum equation becomes
∂u ∂u ∂p 1 ∂ 2u
u +v =− + (4.46)
∂x ∂y ∂x Re ∂ y 2
So now a reduced but still accurate system of equations has been developed, and in
dimensionless form (see Section 3.12 for the mechanics of nondimensionalizing
152 Essentials of Micro- and Nanofluidics

the equation) the boundary layer equations are given by

∂u ∂v
+ =0 (4.47)
∂x ∂y
∂u ∂u ∂p 1 ∂ 2u
u +v =− + (4.48)
∂x ∂y ∂x Re ∂ y 2
∂p
=0 (4.49)
∂y

y δH(x2) δH(x)
δH(x1)
x

x=0 x = x1 x = x2

Sketch of the velocity distribution in a boundary layer on a flat plate.

These equations are much simpler than the two-dimensional Navier–Stokes

equations. Since the pressure does not change across the boundary layer, its value
inside the boundary layer must be the same as its value outside the boundary
layer, where the flow is inviscid. Outside the boundary layer, the fluid acts as if it
were inviscid, and in dimensional terms, the effective kinematic viscosity ν = 0;
in dimensionless terms, this means Re = ∞. Of course, the Reynolds number is
never really ∞, it is just very large. In this case, the momentum equation reduces
to
∂u ∂p dp
u =− = (4.50)
∂x ∂x dx
Integrating this equation between two points results in
2
U∞,1 U2
= p2 + ρ ∞,2
p1 + ρ (4.51)
2 2
for any two points 1, 2. Since U∞ is constant for flow over a flat plate, p1 = p2
for any two points, and thus
dp
=0 (4.52)
dx
This is true for any body that does not have curvature. For example, there will be
a streamwise pressure gradient for flow past a sphere or cylinder. The Blasius
boundary layer equations are the resulting equations:
∂u ∂v
+ =0 (4.53)
∂x ∂y
∂u ∂u 1 ∂ 2u
u +v = (4.54)
∂x ∂y Re ∂ y 2
153 The Essentials of Viscous Flow

and now there are two equations in two unknowns (u, v) because the pressure
has been removed from the problem; in the more general case, the pressure is
known from the inviscid flow outside the boundary layer.
These equations are nonlinear, and the Reynolds number is large. Thus we
will have a tough time solving these equations numerically because a very small
parameter 1/Re multiplies the highest-order term in the equation. This is an
example of a singular perturbation problem, and a short discussion of this type
of problem appears in Appendix A. To remedy this situation, let us scale out the
Reynolds number by writing
y
ŷ = −1/2 (4.55)
Re
v = Re−1/2 v̂ + · · · . (4.56)
Then
∂u ∂ v̂
+ =0 (4.57)
∂x ∂ ŷ
∂u ∂u ∂ 2u
u + v̂ = 2 (4.58)
∂x ∂ ŷ ∂ ŷ
Now far away from the plate, where y ∼ 1, say (y ∗ = L, y ∗ dimensional)
ŷ ∼ y/Re−1/2  1, is very large because the Reynolds number is large. Thus the
boundary conditions are
u=v=0 ŷ = 0 (4.59)
u → 1 as ŷ → ∞ (4.60)
The first boundary condition is the no-slip and solid wall condition, and the
second condition is a matching condition with the free stream. This last condition
means that the dimensionless velocity u approaches the appropriate free stream
value as y ∼ δ, where δ is the boundary layer thickness.
The boundary layer equations can be transformed into an ordinary differential
equation that is much easier to solve. This is done through similarity analysis.
Similarity solutions arise, in general, when there is no physical length scale
associated with the problem. A fluid particle inside the boundary layer does not
“see” the leading or trailing edge of the flat plate. In this case, define a similarity
variable of the form η = ŷ/ξ (x). Defining the fluid velocities in the form
u(η) = f  (4.61)
and from the continuity equation
v̂ = −f ξ  + ηξ  f  (4.62)
and substituting into equation (4.58) leads to
f  + ξ ξ  f f  = 0 (4.63)
The requirement for similarity is that ξ ξ  = C, where C √
is a constant. This
constant is arbitrary, and taking C = 1 leads to η = ŷ/ 2x. This is now
 154 Essentials of Micro- and Nanofluidics

6
u
U0
v
√
5 U0 Rex

3
η

0
0 0.2 0.4 0.6 0.8 1

Solution for the velocity distribution in a Blasius boundary layer. The v velocity is scaled by the Reynolds
Figure 4.8
number so that it fits on the same plot.

a third-order differential equation that can be solved numerically, subject to

the boundary conditions

f  (0) = f (0) = 0 (4.64)


f (∞) = 1

and now the original partial differential equation has been transformed into an
ordinary differential equation. This makes the numerical solution a lot easier, and
it takes mere seconds to solve on a desktop or laptop computer. We will discuss
the methods to solve this problem numerically in Chapter 10.
The solutions for the velocities are shown in Figure 4.8. Here the velocity
normal to the plate is scaled by the Reynolds number to fit on the same plot.
Thus what is plotted is v̂ = −f ξ  + ηξ  f  . Note that both functions approach a
constant as the boundary layer edge is approached.
The wall shear stress is calculated as
du ∂η
τw (x) = at ŷ = 0 (4.65)
dη ∂ ŷ

and using the numerical solution to obtain the value of f  = du/dη at the wall,
in dimensionless form, the skin friction coefficient is given by
τw
Cf = = C f = 0.664 Re−1/2
x (4.66)
1
2
ρU∞2

where Rex = U∞ x/ν. The actual boundary layer thickness is defined as the point
where the velocity profile reaches 99 percent of its free stream value, and the
result is
−1/2
δ = 5x ∗ Rex ∗ (4.67)

Note that δ is dimensional; recall that Rex ∗  1, except very near the leading
edge of the plate.
 155 The Essentials of Viscous Flow

Inviscid Flow: ν=μ/ρ=0

Boundary
Layer
y
Uoo
δ(x)~ const

L
Suction Viscous Flow

Boundary layer over a flat plate with suction.

Figure 4.9

The velocity u in Figure 4.8 looks almost parabolic, and a simple formula that
approximates the numerical solution is

u ∗ = U∞ (2η − η2 ) (4.68)

where now η = y ∗ /δ. The coefficient for the skin friction coefficient has the value
0.730 instead of 0.664, and the coefficient for the boundary layer thickness has
the value 5.477 instead of 5 – not bad for such a simple formula.

4.7 Fully developed suction flows

There is also a class of flows with porous walls for which analytical solutions are
possible. Let us consider an external flow such as a simple boundary layer flow
over a flat plate, as depicted in Figure 4.9. The fluid may be a liquid or a gas, as
long as the velocity of the gas flow is much less than the speed of sound.
As with the boundary layer, the streamwise pressure gradient vanishes, and
the streamwise momentum equation is assumed to be
d 2u du
μ 2
= ρv (4.69)
dy dy
The flow is assumed to be fully developed, which means
∂u
=0 (4.70)
∂x
then ∂v/∂ y = 0 so that at most, v = constant. For this balance to occur,
U∞ ρvU∞
μ ∼ (4.71)
δ2 δ
where δ is the width of the layer. This means that
vδ
Re L ∼1 (4.72)
U∞ L
 156 Essentials of Micro- and Nanofluidics

Sink Fully Developed

y
h
U

L
Source

A duct flow with constant suction.

Figure 4.10

In addition, from the continuity equation, the only solution can be

v = vw = constant (4.73)

and thus the equation for the streamwise velocity is

d 2u du
μ 2
= ρvw (4.74)
dy dy
The solution to this equation subject to the no-slip condition at the wall and the
matching condition at the edge of the boundary layer is
 
u(y) = U∞ 1 − e yvw /ν (4.75)

It is clear that vw must be negative and cannot be positive. The streamwise velocity
is sketched in Figure 4.9. Similar to the Blasius boundary layer, the thickness of
the layer δ is defined as the location where u reaches 99 percent of the value U∞ ,
that is, u = .99U∞ , and this means

1 − eδvw /ν = 0.99 (4.76)

or
ν
δ = −4.6 (4.77)
vw
For U∞ = 0.001 m/sec in water, where the kinematic viscosity ν = μ/ρ ∼
10−6 m2 /sec, a value of vw = −0.00001 m/sec corresponds to δ ∼ 0.05 m. Con-
versely, in air, ν = 2 × 10−5 and δ = 1 m.
Now consider the fully developed flow between two parallel plates: the
Poiseuille flow, only now with a source on one wall and a sink on the other,
as depicted in Figure 4.10. Again, just as before, and for the same reason,

v = vw = constant (4.78)

and the governing equation for the streamwise velocity is

d 2u dp du
μ 2
− = ρvw (4.79)
dy dx dy
 157 The Essentials of Viscous Flow

Re=0.1
0.8
Re=1
Re=10
0.6
Y

0.4

0.2

0
0 0.05 0.1 0.15 0.2 0.25
u

Velocity profile for several Reynolds numbers for the porous duct.
Figure 4.11

where we consider only pressure driven flow.

Recall that the solution for the Poiseuille flow is given by
 2
h2 d p y y
u= − (4.80)
2μ d x h h
It is clear from this equation that the velocity scale can be taken to be
h2 d p
U =− (4.81)
2μ d x
in m/sec if the pressure is given in newtons. The same velocity scale should be
used in the present problem to match up the solutions as vw → 0. In dimensionless
form,
d 2u du
2
= −2 + Re (4.82)
dY dY
where Re = ρvw h/μ is the suction Reynolds number based on the wall velocity
and Y = y/ h. The solution is
2
u = A + BeReY + Y (4.83)
Re

The constants are found by applying the boundary conditions that u = 0 at

Y = 0, 1 and
2 1 − eReY 2
u=− + Y (4.84)
Re 1 − e Re Re
For small suction Re, the solution approaches the Poiseuille flow, whereas for
large Re, the velocity profile is linear over most of the duct and drops rapidly to
zero near the upper wall; moreover, the maximum velocity no longer occurs at
the centerline. In fact, singular perturbation methods can be used to show that the
leading-order term in the boundary layer that forms near Y = 1 is given by
2
uBL = (1 − e− ŷ ) (4.85)
Re
 158 Essentials of Micro- and Nanofluidics

Sink Developing Flow

vw
y

U0 2h

x αvw

L
Source or Sink

A developing suction flow in a duct. The magnitude and direction of suction–injection may be varied. The case
Figure 4.12
α = −1 indicates sinks on both walls.

where ŷ = (1 − Y ) Re. This local solution valid near Y = 1 matches with the
“outer solution” u outer = 2Y /Re, valid in the region away from Y = 1. Note that
as the suction velocity gets large, the axial velocity must get smaller to achieve
mass flow conservation.
The dimensional volume flow rate is given by
 
1 2 1 1 − Re
1 Re
e 1
Q = W hU udY = W hU0 Re − + Re Re (4.86)
0 , 2 1 − eRe 1−e

and the average velocity is

 
1 2 1 1 − Re
1 Re
e 1
Uave = U udY = U0 − + Re
(4.87)
0 Re 2 1 − eRe 1 − eRe

As a numerical example, for a dimensionless pressure gradient d p/d x = −2

at a suction Reynolds number Re = 2, the dimensional volume flow rate Q =
1.56 × 10−6 m3 /sec for a channel height of 0.1 mm and a channel width of
W = 1 µm.

4.8 Developing suction flows

There is an interesting class of problems for which the suction–injection velocity

is not unidirectional and fully developed, as depicted in Figure 4.12. These
problems have been investigated in particular by Berman (1953), White et al.
(1958), Terrill (1964), Terrill and Thomas (1969), Terrill and Shrestha (1965),
and others for both channels and tubes and are termed developing suction flows.
These types of flows occur in the kidney, as discussed in the box in Chapter 2.
For a channel, a mass balance between x = 0 and x = x results in (Figure
4.12)
h
U0 Ac = (1 − α)vw A + W udy (4.88)
−h
159 The Essentials of Viscous Flow

where the wall velocity vw has been assumed constant, Ac = 2hW is the cross-
sectional area, and A = x W . Here it proves useful to use the half height of the
channel for the length scale, and U0 is the (uniform) velocity at the inlet to the
channel. Dividing by W in the two-dimensional case equation (4.88) becomes
h
udy = 2U0 h − (1 − α)vw x (4.89)
−h

so that the flow cannot be fully developed. However, limiting conditions on the
magnitude of the suction velocity allow solutions and, as suggested by equation
(4.89), define

ψ(x, y) = (2hU0 − (1 − α)vw x) f (y) (4.90)

In dimensionless form, the stream function equation becomes

 
1 − α vw
ψ(x, y) = 1 − x f (y) (4.91)
2 U0
where x and y are scaled on h and ψ has been scaled on 2U0 h.
The governing equations are the two-dimensional Navier–Stokes equations,
and substituting equation (4.90) into the x momentum equation, in dimensional
form,
1
ν f  − vw (1 − α) f f  + vw (1 − α) f 2 = px (4.92)
ρ (2hU0 − (1 − α)vw x)
where the differentiation is with respect to y. The dimensionless form of the
equation should be (can you show this?)
1 − α  1 − α 2 px
f  − Rew f f + Rew f = 1−α vw
=K (4.93)
2 2 1 − 2 U0 x
where K is a constant, px = ∂ p/∂ x is dimensionless, and Rew = 2vw h/ν is the
suction–injection Reynolds number and x is scaled on h. The pressure is scaled
on the viscosity so that the pressure is scaled by μU0 /2h. The dimensionless
axial velocity scaled on U0 is given by
 
∂ψ 1 − α vw
u= = 1− x f (4.94)
∂y 2 U0
∂ψ 1 − α vw
v=− = f (4.95)
∂x 2 U0
Differentiating equation (4.93) shows that
∂2 p
=0 (4.96)
∂ x∂ y
The boundary conditions are
2 2α
f (1) = , f (−1) = (4.97)
1−α 1−α
f  (−1) = f  (1) = 0 (4.98)

Note that K = K (Rew ) is unknown.

160 Essentials of Micro- and Nanofluidics

Consider first the solution to this equation for small suction–injection Reynolds
number. For example, if h = 1 µm, vw = 0.01 m/sec, then Rew = 0.01 for water,
and the function f can be expanded in a power series, a regular perturbation
expansion:

f = f 0 + Rew f 1 + Re2w f 2 + · · · (4.99)

Then, differentiating equation (4.93), the leading-order term in the series satisfies

f 0 = 0 (4.100)

The solution is easily shown to be

1+α 3 1
f0 = + y − y3 (4.101)
1−α 2 2
and so the dimensionless stream function is given by
  
1 − α vw 1+α 3 1
ψ0 = 1 − x + y − y3 (4.102)
2 U0 1−α 2 2
The corresponding velocities are
    
1 − α vw  1 − α vw 3 3 2
u0 = 1 − x f = 1− x − y (4.103)
2 U0 2 U0 2 2
 
1 − α vw 1 + α 3 1
v0 = + y − y3 (4.104)
2 U0 1 − α 2 2
The dimensionless pressure can easily be calculated from both momentum
equations. From the x momentum equation,
 
1 − α vw
px = K (Rew ) 1 − x (4.105)
2 U0
and, integrating,
 
K 1 − α vw x 2
p(x, y) = x− + pr (y) (4.106)
Rew 2 U0 2
where
 2
vw  vw f2
pr (y) = f − (4.107)
U0 U0 2
For the suction Reynolds number Rew = O(1), equation (4.93) is a nonlinear
ordinary equation that may be solved numerically; note that the parameter K is
unknown and must be determined in the course of the solution. The numerical
techniques described in Chapter 10 may be employed.
Results are shown for two cases; in Figure 4.13 are the streamline patterns for
two different Reynolds numbers for a sink on the upper wall. Note that the flow
is always developing. In Figure 4.14 is the source flow that is symmetric about
the centerline. As with both these figures, the wall velocity is assumed known;
this, of course, is not true in practice, and vw must be calculated from an analysis
of the region outside the duct.
 161 The Essentials of Viscous Flow

ψ/(2U0 h)
1

0.2
0.7

0.4
0.9

0.6

0.1
0.3
0.8

0.5
0.5
0.7
0.6
0.5 0.4 3
0.

2
0.

0.1
y/h
0 0.4
0.3
0.2

0.2 0.1
−0.5 0.1

−1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
x/h
(a)

ψ/(2U0 h)
1

0.6

0.1
0.8

0.3
0.5
0.7

0.2
0.9

0.4
0.5
0.6
8

0.1
y/h

0.

0.3
0.5
7

0
0.

0.2
0.4
0.6
0.5
−0.5 0.4 0.3
0.2
0.3 0.1
0.2
0.1
0.1

−1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
x/h
(b)

Streamlines for α = 0 and vw /U 0 = 0.4 for a sink at y = 1 and a solid wall at y = −1. (a) Re w = 0.
Figure 4.13
(b) Re w = 5.

There are some anomalous results for this class of problems for channels, and
these issues are discussed extensively by Drazin and Riley (2006). For exam-
ple, multiple solutions have been obtained for Rew > 12.165, with one solution
exhibiting reduced velocity in the core region of the channel; reversed flow may
also occur. There are also multiple solutions for the suction solution in a cylin-
drical tube (White, 2006), and there are also regimes in Reynolds number where
no solutions exist! It is perhaps not surprising that there would be limitations on

ψ/(2U h)
0
1
0.8
0.
5

0.9
0.8 0.6
0.7

0.8
0.6 0.5 0.6 0.7
0.4
0.5 0.6
0.4 0.3 0.4 0.5
0.4
0.3
0.2 0.3
0.2 0.2
0.2
0.1 0.1 0.1
y/h

0 0 0 0 0
−0.1 −0.1
−0.1
−0.2
−0.2
−0.2 −0.2 −0.3
−0.3
−0.4
−0.3 −0.4 −0.5
−0.4 −0.5 −0.6
−0.4
−0.5 −0.6 −0.7
−0.6 −0.8

−0.7
−0.8 −0.6 −0.9
.8
.5

−0
−0

−1
0 0.5 1 1.5 2 2.5
x/h

Streamlines for sources on both walls α = −1 and vw /U 0 = −0.2, Re w = −5.

Figure 4.14
 162 Essentials of Micro- and Nanofluidics

the source and sink magnitudes, although the precise Reynolds numbers at which
the limits occur are not obvious.
The suction velocity at the wall has been assumed to be known, but actually, it
must be calculated from the properties of the porous medium. A porous medium is
characterized by two primary parameters: porosity and permeability. The porosity
is simply the ratio of the void space where fluid can flow to the total volume of
the material and is a property of the porous medium itself. The permeability is
a measure of how easy it is for fluid to pass through the porous medium. The
permeability is defined in one dimension by Darcy’s law:
K0 A d p
Qp = − (4.108)
μ dx
where K 0 is the permeability, and is usually determined empirically, and A is the
area. Note that the permeability has units of velocity.
Of course, the analysis of porous media is very important in hydrologic applica-
tions and in soil science. In the present context, the nanopore membranes depicted
in Figure 4.1 can be considered porous media. The porosity of many membranes
of this type is estimated to be about 1% for pores ranging from 10−100 nm
in the nanoscale dimension and the ratio KL0 ∼ min 10 mL
mmHg m2
(Fissell and Humes,
2006). The streamwise length of the pores is L = 5 µm, the spanwise width is
W = 45 µm, and the overall area of these membranes is typically A ∼ 1 mm2 .

4.9 The lubrication approximation

The classical definition of the term lubrication is the art of reducing frictional
resistance between two surfaces by placing a liquid there. In classical lubrication
problems, the convective terms in the governing equations do not vanish; however,
they are small enough that they can be neglected. This is often the case in micro-
and nanoflows, whether the variation comes from a slow variation in roughness
of the surfaces or slight changes in the surface charge density in electro-osmotic
flow.
In recent years, the term lubrication approximation has come to mean any
flow where the dominant (but not exclusive, as in fully developed flow in a pipe
or channel) variation in the velocity or other variable is in the cross-streamwise
direction, the y direction. In lubrication problems, the streamwise velocity will
vary slowly in x and so a small velocity component must be present in the
y direction. In this section, dimensional variables will be considered initially,
and then important expressions will be nondimensionalized near the end of the
section.
Notation alert: As has been mentioned, an asterisk is often used to denote a
dimensional variable in this text. To do that in this section would be particularly
awkward, and so several times in this section, we will use the same symbol for
both a dimensional variable and a dimensionless variable. The dimensionless
versions are indicated by the presence of dimensionless parameters, while the
 163 The Essentials of Viscous Flow

hi y ho
x
U

A slider bearing to illustrate the lubrication approximation. In recent years, the lubrication approximation has
Figure 4.15
taken on a larger meaning in the sense that it applies to flows having a slight variation in the streamwise
direction, no matter the cause.

dimensional form is indicated by the presence of a physical variable such as

viscosity.

Consider a typical slider bearing, which is depicted in Figure 4.15. Dimensions

may be, for example, for lubricating oil, h i = 1 µm, L = 1 cm, U = 0.25 m/sec,
μ = 0.10 Nsec/m2 , ρ = 1000 kg/m3 , where the properties are those of a 10 W
oil at T = 20◦ C. The ratio of convective terms to the viscous terms is
ρu ∂u
∂x ρ UU ρU h h h
 L
= = Reh (4.109)
∂2u
μ ∂ y2 μ h2
U
μ L L

and Re1 = Reh (h/L) ∼ 2.5 × 10−5 . Thus, even though the convective terms are
not identically zero, they are far smaller than the viscous terms, and so to leading
order, they may be neglected. For example, these same equations apply to the
study of small-amplitude periodic roughness in a channel or tube, provided that
the roughness wave length is much greater than the amplitude of the roughness.
Thus, if Re1  1, then the convective terms may be neglected, even if v = 0
and u = u(x, y). The only restriction is that the v velocity should be small.
Indeed, from the continuity equation,
h
v∼ U U
L
since h/L  1.
The governing equations are the same as the Poiseuille equations with the
boundary conditions
u = U, v = 0, y=0
u = 0, y=h
u = v = 0, y=h
This is a superposition of Poiseuille and Couette flow except that h = h(x) and
the solution for u is
 
1 dp 2 y
u= (y − h(x)y) + U 1 − (4.110)
2μ d x h(x)
The difference between this problem and the classical Poiseuille and Couette
problems is that v = 0 since h = h(x) and the pressure gradient is unkown.
164 Essentials of Micro- and Nanofluidics

To obtain the pressure gradient, note that integration of the continuity equation
leads to
y ∂u
v=− dy (4.111)
0 ∂x

so that
h ∂u
v(x, h) = − dy
0 ∂x
To evaluate the integral the Leibnitz rule is used
F2 (x) ∂
f (x, y)dy
F1 (x) ∂x
d F2 (x)
= f (x, y)dy + F1 (x) f (x, F(x)) − F2 f (x, F2 (x))
dx F1 (x)

Here F1 = 0, F2 = h(x), f = u(x, y). But F1 = 0, u(x, h(x)) = 0 so that both
the last two terms drop out. Thus since v(x, h) = 0 the pressure must satisfy
d h  1 dp 
y

(y − hy) + U 1 −
2
dy = 0 (4.112)
d x 0 2μ d x h
or
 
d dp 3 dh
h − 6μU =0 (4.113)
dx dx dx
and this is the one-dimensional Reynolds equation, named after Osborne
Reynolds, the same Reynolds as in the Reynolds number.
For a linear variation of the slider block,
x
h(x) = h i + (h o − h i ) (4.114)
L
an analytical solution may be obtained. To show this, one integration of the
Reynolds equation yields
dp 3
h = 6μU (h − h m ) (4.115)
dx
where h m is the value of h(x) at which d p/d x = 0, assuming such a point exists.
Integrating again,
x  1 hm

p(x) = 6μU − 3 dx (4.116)
0 h2 h
because dh/d x = const; then
x  
6μU 1 hm
p= dh 2
− 3 dh
dx 0 h h
 
−6μU L 1 hm
= − 2 +A (4.117)
h o − hi h 2h
where A is an integration constant. Now h m and A are constants that must be
determined by applying the boundary conditions; the most common boundary
 165 The Essentials of Viscous Flow

conditions on the pressure in lubrication flows are p = 0 at x = 0, L and at

x = 0, h = h i . Thus, at x = 0,
 
−6μU L 1 hm
0= − 2 +A (4.118)
ho − hi hi 2h i
and at x = L,
 
−6μU L 1 hm
0= − 2 +A (4.119)
ho − hi ho 2h o
Solving this system of equations leads to
6μU L
p(x) =  (h − h i )(h − h o ) (4.120)
h 2 h 2o − h i2
and the maximum pressure is pm = p(h m ), where
1
− 1
2h o h i
h m = 2 h1o hi
=
h 2o
− 1
h i2
hi + ho

The normalized height expression based on h i is given by

h(x) = 1 + mx (4.121)
where here m = h o − h i /h i , and the axial coordinate is scaled on L. Thus the
dimensionless pressure is given by
(h − 1)(h − h o )
p(x) = 6  (4.122)
h 2 h 2o − 1
where the pressure is scaled as
p∗
p= μU L
hi hi

and p∗ is dimensional.
Results for the pressure distribution are shown in Figure 4.16. The pressure
increases rapidly as the slope increases in magnitude and the point of maximum

3
ho
= 0.3
hi
2.5


2
µUL/h2i
p − p∞

0.4


1.5
P ressure

0.5
1
0.6

0.5 0.7

0.9
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
L

Dimensionless pressure as a function of x for a linear variation of the slot. Note that the pressure increases
Figure 4.16
rapidly as the slope increases in magnitude.
 166 Essentials of Micro- and Nanofluidics

pressure moves down the slot. A dimensionless pressure of p = 2 corresponds

to a dimensional pressure of 5 × 1010 N/m2 for an oil having a viscosity of
0.1 Nsec/m2 for the preceding parameters. A large pressure indeed!
The dimensional volume flow rate through the slot is given by
h 1 dp 3 Uh
Q= udy = − h + (4.123)
0 12 d x 2
and solving for the pressure gradient and integrating from 0 to L, the volume
flow rate is given by
hi hoU
Q= (4.124)
hi + ho
per width of the bearing. As an example, for the parameters described earlier,
the flow rate Q = 8.3 × 10−7 m3 /sec and h m /h i = 2/3; that is, the maximum in
the pressure occurs when the height has decreased by 2/3 and the x location is
x/L = 2/3. Since the pressure in the gap p > 0, there is a net force upward on
the surface to support the load.
Note that the solution for the velocity given by equation (4.110) is valid for
any variation of y = h(x), as long as the surface variation satisfies the lubrication
approximation. Thus the solution is valid for a dimensional variation of

h = h 0 + α sin ωx (4.125)

where α  h 0 and αω  1, a formula that corresponds to periodic roughness,

as noted earlier.

4.10 A surface tension–driven flow

At small scales, capillary forces can be significant because the pressure force
exerted on a bubble is proportional to the inverse of its radius of curvature.
Bubbles in microchannels can interfere with the process of filling the channel
with liquid. Suppose a bubble is moving steadily in a tube at constant velocity U ,
small enough that Stokes flow applies. In a reference frame attached to the bubble,
the bubble is stationary, and the tube is moving at velocity −U in the opposite
direction, as shown in Figure 4.17. Note that the bubble is axisymmetric within
the tube and that the gap (H (x)) between the bubble and the tube wall is assumed
to be very small. The interface can be divided into several regions; the front and
rear boundaries of the bubble enclose a region in which the standard lubrication
limit can be used. Within this region are two subregimes: one in which the gap
between the bubble and the wall is constant and equal to δ and the transition
regime from the regions near the bubble ends to the tube.
In the transition regime at the front or rear end, assuming the gap H  a,
Cartesian coordinates can be used, and the governing equation is given by
∂ 2u 1 ∂p
= (4.126)
∂y 2 μ ∂x
 167 The Essentials of Viscous Flow

-U
a

y H
-U x δ

H
y
-U
x
δ

Sketch of a bubble in a tube showing the change in coordinate system for which the flow in the gap is steady.
Figure 4.17
The gap between the bubble and the channel walls is H . The coordinate system is fixed to the bubble so that
relative to the bubble, the tube is moving at −U , where +U is the speed of the moving bubble.

At the channel wall, the velocity boundary condition is u = −U at y = 0, and

at the gas–liquid interface, the tangential stress at the interface is zero, which
gives
∂u
= 0 at y = H (x) (4.127)
∂y
The normal stress condition at the interface is given by the Young–Laplace
equation (Bretherton, 1961):
 
σ Hx x ∂u
p1 + = −2μ 1 +  (4.128)
a 1 + Hx 2 2
∂x

where σ is the interfacial tension. Bretherton (1961) shows that the pressure can
be approximated by
d2 H σ
p1 ≈ −σ 2
− (4.129)
dx a
On the basis of equation (4.126) and the boundary conditions, the solution for
the streamwise velocity, assuming that the lubrication approximation holds, is
 
1 d p y2
u= − Hy − U (4.130)
μ dx 2
in a coordinate system traveling with the bubble. Substituting equation (4.128)
into equation (4.130), the velocity profile is given by
 
σ d 3 H y2
u = −U − − Hy (4.131)
μ dx3 2
and so the volume flow rate through through the gap at the rear of the bubble is
given by
 
H (x) σ H 3 d3 H
Q = −2πa udy = 2πa U H − (4.132)
0 3μ d x 3
168 Essentials of Micro- and Nanofluidics

In the region under the bubble, the thickness of the liquid film is a constant
H (x) = δ, and there is no pressure gradient, so the flow velocity across the film
becomes

u = −U (4.133)

and the corresponding volume flow rate is

Q = −2πaδU (4.134)

From mass conservation, the two flow rates must be equal, and this condition
leads to
d3 H
H3 − Ca H = −Caδ (4.135)
dx3
where Ca = 3μU /σ is the capillary number and, in this analysis, is assumed
small. Defining new dimensionless variables
 
H x 3μU 1/3 x
η= ξ= = Ca 1/3 (4.136)
δ δ σ δ
the equation in dimensionless form becomes

d 3η
η3 =η−1 (4.137)
dξ 3
This equation can be solved numerically or by the method of matched asymp-
totic expansions, as is done by Wilson (1982) for the drag-out problem of film
theory.
Note that the thickness of the layer between the bubble and the tube wall δ
is unknown and must be calculated along with the solution. Three conditions
on the function η are also required. These conditions are discussed in detail
by Probstein (1989) based on physical arguments originally given by Landau
and Levich (1942). In particular, under the bubble in the lubrication region,
H (x) = δ = constant; near the front and rear ends of the bubble, η → ∞, and in
this regime, equation (4.137) indicates that d 3 η/dξ 3 = 0, and so

η  Aξ 2 + Bξ + C (4.138)

where the constants A, B, and C are obtained from the boundary conditions. In
particular, from a numerical solution, Bretherton (1961) gives the result
δ
= 0.643Ca 2/3 (4.139)
a
leading to a parabolic profile for the layer between the bubble and the tube:

x2
η = 0.3215Ca 2/3 (4.140)
δ2
This solution is valid in the transition regime, where both η  1 and H/a  1.
 169 The Essentials of Viscous Flow

Uⴥ
r

Coordinate system for Stokes flow past a sphere.

Figure 4.18

4.11 Stokes flow past a sphere

Many problems in micro- and nanofluidics involve flow past a sphere as depicted
on Figure 4.18. Einstein’s derivation of the diffusion coefficient, or, more pre-
cisely, the molecular dimension of a solute (Einstein, 1905b), is based on the
drag for flow past a sphere derived by Stokes. Assume that the Reynolds number
Re = ρU∞ a/μ ∼ 0 is small, where U is the relative velocity between the sphere
and the velocity far from it, and assume that the flow is axisymmetric about
the transverse axis and is thus independent of the transverse angle φ. The gov-
erning equations are thus two-dimensional and are given in dimensionless form
by
   
| V |2
Re ∇ − V × ∇ × V = −∇ p − ν∇ × ∇ × V (4.141)
2

For Re = 0, these equations reduce to

0 = −∇ p − ν∇ × ∇ × V (4.142)

where here V = (ur , u θ , 0). This is the so-called Stokes flow past a sphere.
It is simplest to determine the stream function first and then determine the
velocity components. The continuity equation in spherical coordinates is given
by
1 ∂ 2 1 ∂
(r u r ) + (u θ sin θ) = 0 (4.143)
r 2 ∂r r sin θ ∂θ
and so the stream function can be defined as
∂ψ
1
ur = (4.144)
r 2 sin θ
∂θ
1 ∂ψ
uθ = − (4.145)
r sin θ ∂r
Eliminating the pressure by cross-differentiating the momentum equations, the
stream function then satisfies
  2
∂2 sin θ ∂ 1 ∂
+ 2 ψ =0 (4.146)
∂r 2 r ∂θ sin θ ∂θ
 170 Essentials of Micro- and Nanofluidics

The boundary conditions are from no slip,

∂ψ ∂ψ
(r = 1, θ) = (r = 1, θ) = 0 (4.147)
∂r ∂θ
and far from the sphere,
1
ψ → r 2 sin2 θ (4.148)
2
because the velocities far from the sphere are u r ∼ cos θ and u θ ∼ −sin θ.
As suggested by the far-field boundary condition, let us try a solution of the
form
 
D
ψ(r, θ) = Ar 4 + Br 2 + Cr + sin2 θ (4.149)
r
The condition u r = 0 at r = a requires, in dimensionless form,

A+ B +C + D =0 (4.150)

and for u θ = 0 at r = a,

4A + 2B + C − D = 0 (4.151)

The far-field condition requires immediately that B = 1/2 and A = 0. Thus

C = −3/4 and D = 1/4 so that, in dimensionless form,
 
1 1
ψ(r, θ) = 2r 2 + − 3r sin2 θ (4.152)
4 r
The dimensionless velocity components are given by the definition of the stream
function as
 
1 1
u r = 2 2r + − 3r cos θ
2
(4.153)
2r r
 
1 1
uθ = − 4r − 2 − 3 sin θ (4.154)
4r r
Note that the velocities are everywhere below the free stream values.
The streamlines are shown in Figure 4.19. Note that they are symmetric about
θ = 90◦ , and the streamlines are remarkably similar to the potential flow past a

Streamlines for Stokes flow past a sphere.

Figure 4.19
171 The Essentials of Viscous Flow

sphere. However, a closer look reveals that the velocities are significantly different
(White, 2006). The pressure is given by

3 cos θ
p = p∞ − μU∞ a 2 (4.155)
2 r
At low Reynolds number, the pressure should scale with the viscosity so that in
dimensionless form (as in the developing suction and lubrication flows),

p − p∞ 3 cos θ
=− (4.156)
μU∞ /a 2 r2

At low Reynolds number, the drag is dominated by viscous effects, as opposed

to form drag, which is dominant at high Reynolds number. The uniform flow
tends to push the sphere to the right and is thus positive, and the drag may
be obtained by integrating the stress components in the direction of motion. It
is the shear stress acting on the radial plane (r) in the positive θ direction that
induces the viscous component of the drag, and this component is, in dimensional
form,
 
∂ uθ μ ∂u r 3μU∞ a 3
τr θ = μr + =− sin θ (4.157)
∂r r r ∂θ 2r 4

The pressure also induces a stress, and this component normal to the sphere is

∂u r
τrr = −p + 2μ (4.158)
∂r
The overall drag is calculated from
π π
D=− τr θ sin θd A − pcos θd A (4.159)
0 0

where d A = 2πa 2 sin θdθ is the area element of integration. Performing the
integration, the result is

D = 4πμaU∞ + 2πμaU∞ = 6πμaU∞ (4.160)

This formula is the basis for the Stokes–Einstein equation used to estimate the
diffusion coefficient of a given species in a liquid mixture. If the sphere is moving
at speed Us in a mean streaming motion of speed U∞ , the drag is given by

D = 6πμa(U∞ − Us ) (4.161)

As noted long ago, there is a problem with this solution for small but non-
zero Reynolds number. To see this, note that the order of magnitude of a typical
convective term is U∞2
/ar 2 and that the order of magnitude of a typical viscous
term is ∼ νU∞ /a 2r 2 so that these terms are the same order of magnitude when
Re(r/a) ∼ 1. Thus, far from the sphere, the solution will always break down. This
 172 Essentials of Micro- and Nanofluidics

fact was noted by Oseen (1910), who suggested that the governing equations be
retained in the form

ReU∞ • ∇ V = −∇ p + ∇ 2 V (4.162)

where now U∞ is the velocity field far from the sphere.3 It can be seen from the
preceding discussion that there are two regions: near the sphere, where r ∼ 1 and
the convective terms are negligible, and far from the sphere, where r Re ∼ 1. The
flow far from the sphere requires a correction of O(Re), and the result for the
stream function was first given by Oseen (1910); further terms in the asymptotic
expansion for small Reynolds number are given by Proudman and Pearson (1957).
The Oseen (1910) result for the first order Reynolds number correction is
  
1 1 3
(1 + cos θ) 1 − e− 2 Rer (1−cosθ )
1
ψ(r, θ) = 2r +
2
sin2 θ −
4 r 2Re
(4.163)
This leads to a correction to the dimensionless drag coefficient in the form of
 
D 6π 3
cD = = 1 + Re (4.164)
ρU∞ 2 a2 Re 8

4.12 Sedimentation of a solid particle

Sedimentation is the process by which particles in a solution settle under gravity

or another force field such as a centrifugal force. The term sedimentation is also
used to describe the changes in concentration of a solute caused in part by a
gravitational field. In this case, the drag force on the particle must balance the net
weight of the particle. For a spherical particle, using Stokes’s drag law, the force
balance in a stagnant fluid becomes
4 3
6πμaU = πa (ρ p − ρ) (4.165)
3
so that the sedimentation velocity is given by
 
2 a2 ρ p
U= −1 g (4.166)
9 ν ρ
The particle Reynolds number is thus
 
ρU a 2 a3 ρp
Re = = −1 g (4.167)
μ 9 ν2 ρ
For small particles apart from the very large particle density limit, the particle
Reynolds number will be much less than 1.

3
It is hard to express the Oseen correction equations in the invariant form.
 173 The Essentials of Viscous Flow

Proteins
Red Blood
Cell

Core
P1 P2
Region

Membrane

Red blood cells and platelets make blood a non-Newtonian fluid in many situations.
Figure 4.20

Relative to the motion of the particle, the local flow field around a particle
is similar to the the flow field past a fixed sphere described in the previous
section.

4.13 A simple model for blood flow

Blood is a fluid that contains a number of components, including water, salts, and
proteins as well as platelets, leukocytes, and erythrocytes. It is the erythrocytes, or
red blood cells, that make blood a non-Newtonian fluid, and the volume fraction
of the red blood cells is called the hematocrit. The plasma portion of blood (water,
salts, and proteins) behaves as a Newtonian fluid (Figure 4.20).
For hematocrit less than 40 percent, blood is usually modeled as a Casson fluid
(Fung, 1981) for which the shear stress in one dimension is given by
√ √ 
τ = τ0 + μ|γ̇ | τ ≥ τ0 (4.168)

and γ̇ = ∂w/∂r = 0 for | τ | ≤ τ0 . Here τ0 is a constant yield stress; the yield

stress is very nearly τ0 = 0.5 N/m2 , which is very small (Fung, 1981).
Physiological data in both arteries and veins suggest that the flow is Newtonian
near the wall and non-Newtonian in the core of the tube. From a shell balance on
an infinitesimal control volume in the fully developed region of the tube,
 2
∂w √ r dp
τ= μ| | + τ0 = − (4.169)
∂r 2 dx
From the definition of the shear stress, the quantity τ0 can be evaluated where
∂w/∂r = 0 at r = rc , and so
rc d p
τ0 = − (4.170)
2 dx
If rc is the radius at which the transition to non-Newtonian behavior takes place,
then for τ > τ0 , the Casson equation can be integrated for constant viscosity to
 174 Essentials of Micro- and Nanofluidics

give
 
1 dp 8
w(r ) = − R − r − rc1/2 (R 3/2 − r 3/2 ) + 2rc (R − r )
2 2
for r > rc
4μ d x 3
(4.171)

w(r) = wc = w(rc ) for r < rc (4.172)

Equation (4.171) can be put in dimensionless form by taking the velocity scale
to be U0 = −(R 2 /4μ)(d p/d x) and
8
w(r ) = 1 − r 2 − rc1/2 (1 − r 3/2 ) + 2rc (1 − r ) (4.173)
3
and now rc is scaled on the radius of the tube R. Note that in dimensionless form,

8 
wc = 1 − rc2 − rc1/2 1 − rc3/2 + 2rc (1 − rc ) (4.174)
3
The velocity profile is depicted in Figure 4.21. The sketch of the velocity profile
in Fung (1981) implies that there is a discontinuity in stress at r = rc . However,
a quick calculation shows that the stress is zero at r = rc when approached from
either side.
The velocity may be integrated to obtain the volume flow rate:
R R4 d p
Q = 2π r wdr = − F(ξ ) (4.175)
0 8μ d x
where
16 1/2 4 1
F(ξ ) = 1 − ξ + ξ − ξ4 (4.176)
7 3 21
and ξ = 2τ0 /R(d p/d x)−1 . Comparing the flow rate result with the Newtonian
value
R4 d p
Q=− (4.177)
8μ d x
it is seen that the Casson flow rate is always less than its Newtonian counter-
part.

4.14 Chapter summary

The objective of the chapter was to introduce the student to a number of vis-
cous flow problems relevant to micro- and nanofluidics. Because micro- and
nanofluidics most often involves internal flows, the vast majority of the solu-
tions presented here are for internal flows. In addition, the structure of the
flow in channels and tubes, including the concept of an entrance length, is also
described.
 175 The Essentials of Viscous Flow

1
r =0.1
c
r =0.2
c
0.8 r =0.3
c
r =0.4
c
r =0.5
c
0.6
r

0.4

0.2

0
0 0.1 0.2 0.3 0.4
u

Velocity profile for blood flow modeled as a Casson fluid.

Figure 4.21

In this chapter, the walls of the channel or tube are assumed to be uncharged.
Solutions are presented for
r Poiseuille flow in a channel and tube
r Fully developed flow in a thin film
r Slip flow in a channel and tube for both liquids and gases
r Blasius boundary layer on a flat plate
r Fully developed suction flows
r Developing suction flows
r Lubrication approximation in a channel
r Surface tension–driven flow due to a moving bubble
r Stokes flow past a sphere
r Sedimentation
r Casson fluid model of blood flow

Additional viscous flow solutions are presented as exercises. All these flows have
the property that the relevant Reynolds number is moderate or small (except for
the Blasius boundary layer) so that the flow is laminar, and in many cases, the
convective terms can be neglected and the flows are at most two-dimensional.
We will revisit some of these solutions later; the presence of Poiseuille flow in
a nanopore membrane is an important feature of the sieving efficiency of the
artificial kidney discussed in Chapter 12.
The focus in this chapter is on steady flows. However, there are many other
applications for which transients are important. Some of these flows are discussed
elsewhere (Acheson, 1990; White, 2006) in great detail, and these and other
references may be consulted for the details.

EXERCISES
4.1 Water enters a channel of height h = 1 µm. Assuming that the boundary
layers on the walls can be approximated by a Blasius boundary layer,
176 Essentials of Micro- and Nanofluidics

determine the entrance length for a flow rate of Q = 1 mm3 /sec. The width
of the channel is W = 40 µm. Compare with the given formulas for the
entrance length described in Section 4.2.
4.2 Repeat the previous problem for air.

CV u(r)

U
r
z

Exercise 4.3

4.3 Using Bernoulli’s equation, estimate the pressure drop in the entrance region
of a pipe. Using the control volume approach, show that the drag in the inlet
region of a pipe is given by
 
1
D = πa 2 p0 − px − ρU 2
3
Is the formula different for a rectangular channel?
4.4 An incompressible fluid flows in a channel with a moving upper wall at
speed Uw and imposed pressure drop d p/d x. Superpose the Couette and
Poiseuille flows and write the velocity in dimensionless form. Calculate the
imposed Couette velocity such that the volume flow rate is zero. Plot the
streamlines for several values of the dimensionless parameter that governs
the motion (i.e., the ratio of the Poiseuille and Couette velocity scales).
film

film

L
cross−section
view U g

Exercise 4.5

4.5 Consider the viscous flow on a wire that is being extruded from a liquid
bath, as in the figure. The diameter of the wire is on the order of the the film
177 The Essentials of Viscous Flow

thickness so that cylindrical coordinates must be used. Outside the film is

air at constant pressure patm .
a. Neglecting the effects of surface tension, find the solution to the problem
and write it in dimensionless form. What is the velocity scale?
b. Plot the velocity distribution across the film in dimensionless form for
the ratio of the film thickness h to the wire radius a of 0.3. Take the ratio
of the velocity U to the velocity scale found in part a to be 1.
c. Calculate the flow rate and the power per unit length required to pull
the wire out of the bath. Use the values U = 1 µm/sec, wire radius
a = 1 µm, film thickness h = 0.3a and the density and viscosity taken
to be that of water.
4.6 Consider the steady laminar flow between two coaxial tubes. The flow
is maintained by a pressure gradient. Calculate the velocity distribution
within the annulus. Calculate the volume flow rate and relate it to the
pressure drop. Write the expression for the pressure drop in dimensionless
form, thus finding an expression for the head loss parameter.
4.7 Repeat the previous problem for the case in which the inner cylinder is
moving at speed U .
4.8 Verify the leading-order expression for the pressure distribution for the
developing suction flow at low Reynolds number, which is
 
K 1 − α vw x 2
p(x, y) = x− + pr (y)
Rew 2 U0 2

where
 2
vw  vw f2
pr (y) = f −
U0 U0 2

4.9 Numerically solve the developing suction problem for a sink at y = 1 and
a solid wall at y = 0, for vw /U0 = 0.2, thus validating the results of Figure
4.14. Such a geometry is a simple model for a device to be used for an
artificial kidney. Calculate the leading-order term for the velocities and the
pressure for Reynolds numbers Rew = 0, 1, and 3. Plot the streamlines for
each case.

2hi x

L
Exercise 4.10
178 Essentials of Micro- and Nanofluidics

4.10 A pressure-driven flow of water passes through a converging nozzle of

length 3 µm. The nozzle is very wide, W = 40 µm, and is 50 nm in height
at the entrance and 40 nm in height at the exit. Calculate the velocity
distribution and volume flow rate under the lubrication approximation. For
a pressure drop of 0.01 atm, what is the volume flow rate and the Reynolds
number? Plot the axial velocity and the pressure at several locations along
the nozzle. How would the results change for a diffuser?
4.11 Repeat the previous problem for a cylindrical nozzle of diameter D =
50 nm.
4.12 Consider the Stokes flow past a circular cylinder. Show that the stream
function satisfies
 2
∂2 1 ∂ 1 ∂2
+ + 2 2 ψ =0
∂r 2 r ∂r r ∂θ
Along with the no-slip condition on the cylinder surface, which is
∂ψ ∂ψ
(a, θ) = (a, θ) = 0
∂r ∂θ
ψ ∼ Ur sin θ as r → ∞

show that the solution of the form ψ = Ur sin θ leads to

 
D
ψ = Ar + Br log r + Cr +
2
sin θ
r
and hence there is no solution to the problem because no choice of the
constants can satisfy all boundary conditions.
Proudman and Pearson (1957) show that using methods of matched
asymptotic expansions that near the cylinder,
 
r log r − 12 r + 2r1
ψ=  8
log Re − γ + 12
where γ is Euler’s constant γ ∼ 0.58. Thus the constants B, C, and D can
be determined, but the constant A cannot.
4.13 Show that the solution for a moving sphere through a motionless liquid is
obtained merely by subtracting the uniform flow term in the Stokes flow
past a fixed sphere.
4.14 The slow flow past a spherical bubble of radius a may be modeled by
assuming that the boundary condition on the bubble is no tangential stress
(∂/∂θ = 0) at the bubble surface. Show that the stream function is given by
U 2
(r − ar )sin2 θ
ψ=
2
and that the drag is given by D = 4πμU a.
4.15 In a rapid molecular analysis system, the channels in the membrane making
up the system are often rectangular. Calculate the solution for the velocity
179 The Essentials of Viscous Flow

field of a Casson fluid in fully developed pressure-driven channel flow of

height h. At what channel height would you expect the influence of the
size of the red and white blood cells to become important? Assume that the
channel is very wide, and thus provide the analogue for a channel of blood
flow through a tube.
 5 Heat and Mass Transfer Phenomena
in Channels and Tubes

5.1 Introduction

In the previous chapter, the types of steady viscous flows common to micro- and
nanofluidics were presented. Of particular importance is the Poiseuille flow and
flow past a sphere. In the present chapter, classical solutions for heat transport
and mass transport of uncharged species that may occur in parallel with the
Poiseuille flow are described. This chapter will lead into a discussion of elec-
trostatics and electrochemistry, concepts that are so important in micro- and
nanofluidics.
Both heat and mass transfer are discussed in this chapter. This is because of
the similarity in the forms of the governing equations and boundary conditions
and the similar nature of the transport properties: the thermal conductivity and
the mass diffusivity or diffusion coefficient. Indeed, heat and mass transfer often
occur simultaneously (Figure 5.1). Both heat and mass transfer can be grouped
into two modes: conduction heat transfer, diffusion mass transfer, and convective
heat and mass transfer. A third mode of heat transfer, radiation heat transfer, is
not discussed.
Fortunately, most liquid flows in micro- and nanochannels do not encounter
large temperature and concentration gradients so that in general, the fluid prop-
erties remain relatively constant. Thus the governing equations can usually be
decoupled to some extent, even at Reynolds numbers not normally viewed as
being small.
Several basic problems of heat transfer in channels are considered before
presenting the formal definition of the fully developed temperature distribution.
Unlike the formal definition of fully developed fluid flow, the solution for the
fully developed temperature distribution depends on the boundary conditions on
the temperature at the walls of a tube or channel. The two cases of constant wall
temperature and constant heat flux are considered, and the length of the thermal
entrance region can be calculated for both these conditions. It should be noted
that because the Schmidt number is so large for liquid flows, mass transfer in
channels and tubes is seldom fully developed.
The presentation in this chapter is similar to the presentation of previous
chapter. After discussion of the entrance regime, several relevant mass transfer
180
 181 Heat and Mass Transfer Phenomena in Channels and Tubes

The National Institute of Standards and Technology (NIST) Lab-on-a-Chip microreactor for analysis of the
Figure 5.1
influence of polymer additives and the synthesis of copolymer surfactants. Go to www.nist.gov and search
polymers for more information.

problems are discussed, including mass transfer in thin films, Taylor–Aris disper-
sion, unsteady mass transport with drug delivery applications, and a mass transfer
view of Brownian motion.
Solutions for steady state thermal and mass transfer boundary layers are pre-
sented, illustrating the influence of the relevant dimensionless parameters, the
Prandtl and Schmidt numbers. It is apparent from the contents of this chapter that
there are a number of analogous problems in the two disciplines of heat and mass
transfer, perhaps a result of their parallel historical development, as discussed in
Chapter 1.

5.2 One-dimensional temperature distributions in channel flow

In this section, simple one-dimensional temperature distributions in channel flow

are described and compared in form to the corresponding velocity distributions.
Here it is assumed that the temperature variations are small so that changes in
properties with temperature can be neglected. Also a constant wall temperature
is assumed, and the temperature T = T (y), where y measures distance from the
wall and 0 ≤ y ≤ h.
Consider the temperature distribution in a channel containing a liquid flowing
at a small Reynolds number where the flow is two-dimensional. The governing
equation at steady state for constant properties is given by

ρc p (u • ∇)T = k∇ 2 T (5.1)

Note that the temperature T and all the other variables are dimensional; the
appropriate temperature scale cannot be chosen until the boundary conditions are
specified.
In the hydrodynamic fully developed region, the transverse velocity v = 0, and
at low Reynolds numbers and moderate Prandtl numbers, convective terms are
negligible so that

∂2T ∂2T
+ =0 (5.2)
∂x2 ∂ y2
182 Essentials of Micro- and Nanofluidics

For example, at physiological temperatures of T = 37◦ C, the Prandtl number of

water is Pr ∼ 4.5. Thus, for the Reynolds number Re ∼ 10−3 , Re Pr = 0.0045,
and the convective terms in the energy equation will be negligible. If the channel is
long so that ∂/∂ x  ∂/∂ y, then we might expect that the temperature distribution
is given by the solution of the one dimensional equation
d2T
=0 (5.3)
dy 2
Assuming that the wall temperatures are known, the boundary conditions can be
taken to be

T = Tw1 at y = 0 (5.4)

T = Tw2 at y = h (5.5)

Integrating twice and applying the boundary conditions,

y
T (y) = (Tw2 − Tw1 ) + Tw1 (5.6)
h
This equation sets the temperature scale as Tscale = T = Tw2 − Tw1 , and the
dimensionless temperature is thus
T − Tw1
θ= =Y (5.7)
T
where Y = y/ h. Clearly if Tw2 = Tw1 , then the temperature is constant across
the channel. This temperature distribution is equivalent to the temperature distri-
bution in a solid slab.

Notation alert: In this section, Q refers to overall heat transfer rate in watts and
not to volume flow rate.

The heat transfer flux in W/m2 K is defined by the equation

∂T dT Q
q = −k = −k = (5.8)
∂y dy A
in this case, and from this simple linear solution, we find that the total heat transfer
rate is
k AT T
Q=− =− (5.9)
h R
where R is the heat transfer resistance. If R is large, the heat transfer rate is
small. In microchannel flows, for a channel with L = 1 µm in the flow direction
and W = 40 µm wide, R ∼ 4.4 × 105◦ C/m for a channel height of h = 100 nm.
The negative sign is to ensure that Q > 0 as it flows from a higher to a lower
temperature. Thus, for Tw2 > Tw1 , the heat flux is negative (in the negative y
direction). This situation is depicted in Figure 5.2. Note also that the heat flux is
constant, ensuring an overall energy balance qw1 = qw2 .
The Nusselt number in this problem is defined by N u = qw b/k(Tw1 − Tw2 ),
where here b is an appropriate length scale. Physically, the Nusselt number is
the ratio of the convective heat transfer rate to the conduction heat transfer rate,
 183 Heat and Mass Transfer Phenomena in Channels and Tubes

T2 1

q q

T1 0

Temperature distribution in a rectangular channel due to a temperature difference between the two walls. The
Figure 5.2
figure on the right depicts the dimensionless distribution from equation (5.7).

and this interpretation is similar to that of the Reynolds number being the ratio
of the convective forces to viscous forces. In this situation, b = h, and using the
expression for the heat flux, it is seen that N u = 1.
Now consider the conduction heat transfer through a duct with a uniform
source of thermal energy; the equation governing the one-dimensional tempera-
ture distribution is
d2T qgen
+ =0 (5.10)
dy 2 k
where again, a constant thermal conductivity has been assumed. Here qgen has
units of W/m3 , and this problem could occur in the flow of an electrolyte solution
where qgen arises due to the presence of an electric field, a phenomenon known
as Joule heating.
For constant temperature boundary conditions, as previously, we can integrate
twice to obtain
y qgen 2
T (y) = (Tw2 − Tw1 ) + Tw1 − (y − yh) (5.11)
h 2k
Note that for Tw2 = Tw1 = Tw , the temperature distribution is of the same form
as the velocity distribution in fully developed flow in a channel, with the quantity
qgen /2k taking the place of −(1/2μ) (d p/d x) and
qgen 2
T (y) − Tw = − (y − hy) (5.12)
2k
or in dimensionless form,
T (y) − Tw
θ(y) = qgen h 2
= Y − Y2 (5.13)
2k

Thus, on a dimensionless basis, θ(y) = u(y), where u(y) is scaled on the

Poiseuille velocity U0 = −(h 2 /2μ) (d p/d x).
In the general case where Tw2 − Tw1 = 0, the solution for the temperature is
a superposition of the linear temperature profile and the parabolic distribution
due to the heat source, much like the superposition of Poiseuille and Couette
flow. For this situation, there are two temperature scales that could be used to
form the dimensionless temperature distribution: Tscale = T = Tw2 − Tw1 and
Tscale = qgen h 2 /k. If both quantities are of the same magnitude, either can be used
to form the dimensionless temperature.
 184 Essentials of Micro- and Nanofluidics

Another problem of interest is the balance between conduction and viscous

dissipation terms in the energy equation. In this case, the governing equation
becomes
 2
d2T du
k 2 =μ (5.14)
dy dy
in the fully developed region. If the velocity is given by the Poiseuille flow
distribution,
1 dp 2
u(y) = (y − yh) (5.15)
2μ d x
then since
du 1 dp
= (2y − h) (5.16)
dy 2μ d x
integrating equation (5.14), the temperature is given by
 2
1 dp y4 2 3 h2 y2 h3
T = Tw + − y h+ − y (5.17)
4μk d x 3 3 2 6
where the wall temperatures are assumed equal and constant: Tw2 = Tw1 = Tw .
Thus the temperature scale can be defined by
 2
dp
dx
h4
Tscale = (5.18)
4μk
It is left as an exercise to write equation (5.14) in dimensionless form and to
estimate the order of magnitude of the resulting dimensionless parameters such
that viscous dissipation is an O(1) effect.
In the present section, the dependence of the temperature on the streamwise
coordinate has been neglected in a manner similar to what is the case in the
hydrodynamically fully developed region, and several canonical problems have
been identified. In the next two sections, the thermal entrance region is formally
defined.

5.3 Thermal and mass transfer entrance regions

Just as for fluid flow, there is a thermally developing region where the temperature
distribution adjusts to a different boundary condition on the walls of the channel
or tube. As with the fluid flow, the temperature initially behaves similarly to
an external flow. The expression for the thermal entrance length is of the same
form as for the hydrodynamic entrance length, and for a constant tube surface
temperature and laminar flow,
L eT
= 0.06 RePr (5.19)
D
for large Reynolds number in the laminar regime Re ≤ 2300.
 185 Heat and Mass Transfer Phenomena in Channels and Tubes

Viscous
Fully
Developed

Developing Thermal
Boundary Layers

U r D
0 x

Thermal
Entrance Length
Fully
Developed
Hydrodynamically Laminar
Developing Flow Pr > 1 Poiseuille
Flow

Flow through a pipe showing the thermal entrance region for Pr > 1. The picture is similar for a channel with
Figure 5.3
r replaced by y.

It is interesting to note that some heat transfer texts (Incropera & Dewitt,
1990; Kays and Crawford, 1980) indicate that the constant in the definition of the
entrance length is 0.05; this may be a function of the definition of the length of the
entrance region as being the length where the temperature approaches to within
95 or 99 percent of the fully developed value. Alternatively, the discrepancy could
be due to the interpolation of the individual numerical result (Kays and Crawford,
1980). In any case, the difference is small and irrelevant for applications because
usually, only an estimate of the entrance region length is required.
In liquid flows, the Prandtl number Pr > 1 and so the thermal entrance length
is somewhat longer than the hydrodynamic entrance length (Figure 5.3). For a
uniform heat flux, the result is virtually no different than the case of constant
surface temperature (Kays & Crawford, 1980).
Similarly, the entrance length for mass transfer in a tube on the macroscopic
scale is
L em
= 0.06 ReSc (5.20)
D
Note that for a Reynolds number of Re = 10 and Sc = 1000, L em /D = 600,
which is extremely large, and for this reason, concentration profiles are rarely
considered to be fully developed in channels of interest in microfluidics and
nanofluidics.
Specifically for microchannels, Lee and Garimella (2006) have performed
a number of numerical computations for different channel aspect ratios for a
constant heat flux and found that the result for a slit pore is
L eT
= 0.057 RePr (5.21)
DH
which agrees with the macroscale computation for the tube at constant sur-
face temperature. Recall that the hydrodynamic entrance length is nonzero for
186 Essentials of Micro- and Nanofluidics

Re → 0, as described in the previous chapter; this corresponds to the entrance

region for the Stokes flow problem in a channel or tube. Based on the analogy
between fluid mechanics and heat transfer and, by extension, mass transfer (see
equation (5.20)), Lee and Garimella (2006) suggest that the Reynolds number
can be replaced by RePr for heat transfer and ReSc for mass transfer in those
formulas.
While the formulas for the thermal entrance length are similar to the hydro-
dynamic entrance length for both constant temperature and constant heat flux
boundary conditions, the formal definition of thermally fully developed flow is
somewhat different, depending on the boundary conditions at the wall. The mag-
nitude of the thermal energy rate transported in the axial direction in kJ/sec is
typically defined in terms of a mean temperature Tm as

Q = ṁcTm = ρ AU cTm = ρwcT d A (5.22)

A

where U is the average axial velocity and A is the cross-sectional area of the
tube. Equation (5.22) defines the area-averaged mean temperature as
1
Tm = wT d A (5.23)
UA A

If Tw denotes the wall temperature, then a dimensionless temperature may be

defined by
Tw − T
θ= (5.24)
Tw − Tm
The flow is said to be thermally fully developed if
∂θ
=0 (5.25)
∂x
This condition is reached if the surface temperature is uniform, as in the previous
section, or if the heat transfer rate in the streamwise direction is constant. Note
that the simple solutions discussed in the previous section in the channel flow
case satisfy these conditions.
For a tube, note that equation (5.25) implies that
∂T
∂θ |r =a
= constant = − ∂r (5.26)
∂r Tw − Tm
which means that the heat flux across the tube is constant. The heat flux normal
to the wall can also be expressed in terms of a heat transfer coefficient as
∂T
q̇r = −k = h T (Tw − Tm ) (5.27)
∂r
and so equation (5.26) implies that the heat transfer coefficient h T is constant.
For a Prandtl number Pr > 1, let us assume that the velocity is fully developed;
if the Prandtl number Pr ∼ 1 or smaller, as is the case for gases, then the flow
field and the temperature distribution are fully coupled. For Pr  1, the velocity
 187 Heat and Mass Transfer Phenomena in Channels and Tubes

xent,T
Tw
Tm,0
r

δT

x=0 x = x1 x = x2 (= xent,T)

T(r, x)

Tw x = x2
x = x1

Tm,0 x=0

r
0 R

Thermal entry region for T w > T m , showing the temperature distribution in the entrance region of a pipe. At
Figure 5.4
x = x 2 , the temperature distribution is fully developed. The picture is similar for a channel with r replaced
by y.

field becomes fully developed sooner than the temperature, and the governing
equation for the temperature in a tube is given by
 
∂T ∂2T 1 ∂T
ρcw =k + (5.28)
∂x ∂r 2 r ∂r

where for a liquid, c p = cv = c. Using the definition of θ, and since ∂θ /∂ x = 0,

 
∂T dTw dTw dTm
= −θ − (5.29)
∂x dx dx dx

If ∂ T /∂ x = constant, then equation (5.28) can be integrated directly. For

constant heat flux, since h T = constant, then Tw − Tm = constant and

∂T dTw dTm
= = = constant (5.30)
∂x dx dx
Conversely, if Tw = constant, then

∂T Tw − T dTm
= (5.31)
∂x Tw − Tm d x

and the solution to equation (5.28) can be calculated numerically or by iteration,

as suggested by Kays and Crawford (1980). The thermal entry region for a tube
at constant wall temperature is depicted in Figure 5.4.
Moreover, because of the similarity of the heat and mass transfer equations,
similar methods may be used for the mass transfer problem. However, unlike
188 Essentials of Micro- and Nanofluidics

temperature, surface concentrations cannot normally be measured, and so the

no-flux condition corresponding to
∂c A
= 0 at the wall (5.32)
∂r
is used, leading to a concentration profile independent of r due exclusively to the
boundary condition. Moreover, since the Schmidt number is so large in liquids,
on the macroscale, as noted above, the concentration distribution is often not fully
developed at all. This will be seen later when Taylor dispersion is considered.
There have been some indications in the literature that the various hydrody-
namic, thermal, and mass transfer entrance lengths in microscale geometries may
be different from those at the macroscale. A careful reading of the literature shows
that this is likely not the case. Reasons for the possible discrepancies include the
difficulties in measuring anything at such small scales. Because the dimensions
of pipes and channels are small, any small error in the determination of the pre-
cise dimensions is magnified, leading to results that conflict with the macroscale
result. Indeed, Judy et al. (2002) suggest that any such differences in the related
problem of determining the pressure drop are within the experimental error of
the measurement of the dimensions. In the same spirit, Bavier and Ayela (2004)
suggest that some experimental methods are poorly adapted to the microscale
regime. The conclusion is that these formulas do indeed apply to the microscale.

Jean-Baptiste Joseph Fourier was born

1768 in Auxerre, France. Fourier became
obsessed with mathematics and heat at
an early age, excelling specifically in
his mathematics studies. During his life
he was a professor of mathematics at
the École Polytechnique for many years.
However, it was during his service to
Napoleon as Prefect of the Department of
Isère in Grenoble, France where he devel-
oped his theory of heat conduction by
1807 despite heavy criticisms by contem-
poraries such as Laplace, Lagrange, and
Poisson. This theory was based upon and
solved using a series of function expan-
sions now known as Fourier series (Churchill, 2005). Fourier developed his famous
mathematics series to describe the heat conduction in solid bodies such as the
earth. His studies on heat conduction were combined with his trigonometric series
and they were presented together in his famous memoir, On the Propagation of
Heat in Solid Bodies. Today, the Fourier series has become an essential tool when
solving many engineering problems. Unfortunately, his contemporaries mixed
reviews and competing theories kept his critical discoveries from being accepted
and recognized by the scientific community with the proper respect during his
lifetime.
 189 Heat and Mass Transfer Phenomena in Channels and Tubes

5.4 The temperature distribution in fully developed tube flow

For the constant heat flux boundary condition, the equation governing the tem-
perature in a tube is given by
   
1 ∂ ∂T 2wave r 2 dTm
r = 1− 2 (5.33)
r ∂r ∂r α a dx
Here wave is the average velocity in the tube and α = k/ρc p is the thermal
diffusivity.
For constant heat flux, dTm /d x is constant, and this equation can be integrated
twice with respect to the radial coordinate r, with the result
 
2wave dTm 3 2 r4 r2
T = Tw + − a − + (5.34)
α dx 16 16a 2 4
The mean temperature is thus given by
11 2wave dTm 2
Tm = Tw − a (5.35)
96 α d x
and the local heat flux rate is given by
11 2wave dTm 2
q  = h T (Tw − Tm ) = h T a = constant (5.36)
96 α d x
Applying conservation of thermal energy and neglecting losses over a differential
control volume in the tube, the heat flux at the wall is also given by
ρcwave a dTm
q  = (5.37)
2 dx
so that the Nusselt number
hT D 48 ∼
Nu = = = 4.364 (5.38)
k 11
The constant temperature solution is a bit more complex, and Kays and Craw-
ford (1980) suggest the use of successive approximations beginning with the
constant heat flux solution. In this case, the Nusselt number in the fully devel-
oped zone is N u = 3.66. Sketches of the axial temperature distribution for both
the constant temperature and constant heat flux are depicted in Figure 5.5.

5.5 The Graetz problem for a channel

For the Peclet number PeT = RePr  1, the hydrodynamic entrance length is
shorter than the thermal entrance length, so the flow can be assumed to be
fully developed, as depicted in Figure 5.6. The governing equation and boundary
conditions in the thermally developing region for a channel having constant wall
temperature in dimensionless form are
∂θ ∂ 2θ
y(1 − y) = 2 (5.39)
∂ξ ∂y
 190 Essentials of Micro- and Nanofluidics

Tw
Tw
T
T m T Tm

x x
(a) (b)

Sketch of the temperature in fully developed laminar flow. (a) Constant heat transfer flux. (b) Constant wall
Figure 5.5
temperature.

where ξ = x/RePr = x/PeT subject to

θ(0, y) = 1 (5.40)

θ(ξ, 0) = θ(ξ, 1) = 0 (5.41)

The variable x = x ∗ / h, and here θ is the scaled temperature, defined by

T − Tw
θ= (5.42)
T0 − Tw
where T0 is the temperature at the entrance to the channel. This is the Graetz
problem, and in this section, a solution to this problem is developed.
The solution may be obtained by separation of variables in the form (Brown,
1960)
∞

ci Yi e−λi x
2
θ= (5.43)
i=1

Substituting into the preceding governing equation, the functions Yi satisfy

Yi + λi2 y(1 − y)Yi = 0 (5.44)

Viscous
Fully
Developed

Developing Thermal
Boundary Layers

U y h
0 x

Thermal
Entrance Length
Domain of
Hydrodynamically solution
Developing Flow Pr > 1 for the Graetz
problem

The domain of solution for the Graetz problem for Pr  1 in a channel.

Figure 5.6
191 Heat and Mass Transfer Phenomena in Channels and Tubes

subject to Y (0) = Y (1) = 0. This is what is called a Sturm–Liouville problem1 so

that the solution may be written formally in terms of an infinite series expansion of
the functions Yi , and λi are the eigenvalues of equation (5.44). These eigenvalues
may be calculated numerically, and this has been done for the first 10 eigenvalues
by Brown (1960).
The heat flux at the wall is defined by
∂T
qw = −k (5.45)
∂y
which must be balanced by that convected through the channel at the mean
temperature

qw = h T (Tm − Tw ) (5.46)

Equating these two quantities defines the local heat transfer coefficient as
k ∂θ
hT = − (5.47)
hθm ∂ y
The Nusselt number is defined by
hh T 1 ∂θ
Nu = = | y=0 (5.48)
k θm ∂ y
Thus all that are needed to determine the Nusselt number are the mean temperature
and the temperature gradient at the wall.
Out of this analysis comes the value of the Nusselt number in the fully devel-
oped regime, which, for a very wide channel, is
hT DH
Nu = = 7.54 (5.49)
k
for constant temperature and N u = 8.24 for the constant heat flux solution. In
this case, the hydraulic diameter is D H ∼ 2h, and from the numerical solution,
the asymptotic Nusselt number is reached when

2ξ = 0.12

leading to
L eT
= 0.06 RePr (5.50)
2h
for the entrance length, as stated in Section 5.3. Detailed numerical data for
channels of various aspect ratios and the solution to the Graetz problem in a tube
are presented by Kays and Crawford (1980).

1
A Sturm–Liouville problem is an ordinary differential equation with boundary conditions that, in
the language of linear algebra, form a complete orthonormal set, and the solution may be written
formally in terms of an infinite series expansion of the functions. There are limitations on the
form of the coefficients in the equation.
 192 Essentials of Micro- and Nanofluidics

5.6 Mass transfer in thin films

A problem similar to the thermal Graetz problem in a channel considered in the

last section is mass transport in a thin film. For simplicity, consider the absorption
of, say, oxygen, O2 , into water. Then the governing equation for mass transfer is
 
∂ω ∂ω ∂ 2ω
ρ u +v = ρ D AB 2 (5.51)
∂x ∂y ∂y
where ω is the mass fraction of the gas in the liquid. The film flow is almost fully
developed because the amount of gas absorbed into the film is very small. Indeed,
the diffusion coefficient of O2 in water is only 2.4 × 10−9 m2 /sec. This means
that the values of the viscosity and density of the liquid film are not significantly
affected so that the velocity is as given in Section 4.5. Because oxygen is being
absorbed into the film, the v velocity will be finite but very small (lubrication
approximation), and thus, to a very good approximation,
∂ω A ∂ 2ωA
u(y) = D AB (5.52)
∂x ∂ y2
where
  2 
ρgh 2 y 1 y
u= − (5.53)
μ h 2 h
The problem is subject to boundary conditions

∂ω A
= 0 at y = 0 (5.54)
∂y
ωA = 0 at x = 0 (5.55)

ω A = ω As at y = h (5.56)

The first boundary condition is appropriate for a solid impermeable wall. This
problem is similar, though not identical, to the Graetz problem and has been
solved by Pigford.2
Because the diffusion coefficient is very small, we expect that the mass transfer
is slow so that the diffusion of the gas takes place only near the liquid–gas surface.
This means that we can replace the velocity with its value at the interface, and
ρgh 2
u(h) = Umax = (5.57)
2μ
approximately, and now y measures distance locally from the vapor–liquid inter-
face. Thus the problem becomes
ρgh 2 ∂ω A ∂ω A ∂ 2ω A
= Umax = D AB (5.58)
2μ ∂ x ∂x ∂ y2
2
R. L. Pigford, PhD thesis, University of Illinois, 1941.
 193 Heat and Mass Transfer Phenomena in Channels and Tubes

subject to
∂ω A
= 0 as y → ∞ (5.59)
∂y

ωA = 0 at x = 0 (5.60)

ω A = ω As at y = 0 (5.61)

where we have replaced y with h − y and made the approximation that mass
transfer takes place only near the surface. The first boundary condition can be
replaced by

ω → 0 as y → ∞ (5.62)

The solution for this problem is given in many texts and can be obtained by
using a Laplace transform in x. The result is
 
ωA y ωA 2 η −ξ 2
= erfc √ = =1− e dξ (5.63)
ω As 4D AB x/Umax ω As π 0
√
where η is a similarity variable and η = y/ 4D AB x/Umax .
The flux of gas into the film is given by
∂ω A 
n Ay = −ρ D AB = −ρω A0 D AB Umax /(π x) (5.64)
∂y
and integrating over a length L, the total amount of mass transferred is
W L 
MA = n Ay dzd x = 2ρW ω A0 D AB Umax L/π (5.65)
0 0

The result for the velocity and the mass fraction is sketched qualitatively in
Figure 5.7.

h h
x

Vapor Vapor

L
g

O O2
2

Liquid Film

A thin liquid film falling under gravity: Velocity and mass fraction of a gas dissolving into a liquid.
Figure 5.7
 194 Essentials of Micro- and Nanofluidics

Sketch of the broadening of the concentration distribution of a solute as it flows through a channel or tube,
Figure 5.8
Taylor dispersion. The solid line is a Gaussian distribution of solute locally around the point z = wt, provided
the dispersion coefficient is used as opposed to the diffusion coefficient.

As an indication of the magnitude of the flux of gas into the film, suppose
that oxygen is diffusing into a film of water of height h = 10−4 m and length
L = 1 cm and width W = 1 cm. Then
ρgh 2 1000 × 9.8 × 10−8 m m
Umax = = = .05 (5.66)
2μ 2 × 10−3 sec sec
Suppose we assume that the amount of O2 dissolved in the water corresponds to
ωs = 0.0013 at the free surface. Then the mass M absorbed is
W L 
M= n Ay dzd x = 2ρW ω As D AB Umax L/π (5.67)
0 0

or

M = 2ρW ω As D AB Umax L/π (5.68)

= 2 × 1000 × 10−2 × 0.0013 2.4 × 10−9 0.05 × 0.01/π = 3 × 10−8 kg O2
(5.69)
a very small number indeed.

5.7 Classical Taylor–Aris dispersion

G. I. Taylor (1886–1975), the “father” of dispersion, was a physicist who made

significant discoveries in a variety of areas including applied mathematics, clas-
sical physics, and fluid mechanics. His fluid mechanics work was very broad,
dealing with applications in meteorology and oceanography in addition to his
work in chemical separation science, his theory of dispersion. In oceanography,
he is the discoverer of the Taylor column, that column of water in a rotating flow
in which the velocity does not vary in a direction parallel to the rotation axis. He
was an avid sailor, having, as a boy, designed and built a sailboat that he eventually
sailed on the River Thames. George Batchelor, his student, was an eminent fluid
mechanician in his own right. See the Cambridge book The Life and Legacy of
G. I. Taylor, by George Batchelor, published in 1994 for more information.

Notation Alert: In keeping with Taylor’s original notation we use z for the axial
coordinate instead of x that heat transfer texts use.
195 Heat and Mass Transfer Phenomena in Channels and Tubes

In micro- and nanofluidics, it is often desired to follow a solute as it is trans-

ported down a tube. A solute’s speed can often be used to determine its identity,
and different species in a mixture can be separated based on their speed of travel
through a tube. In the early 1950s, G.I. Taylor (1953, 1954) considered the trans-
port of a small slug of solute in a tube to investigate the longtime spreading
behavior of the solute slug as it moves under a pressure-driven flow in the tube.
The solute will spread relative to the local flow speed because of the parabolic
variation of the Poiseuille flow, termed convective dispersion (Probstein, 1989)
and also because of solute radial diffusion and axial convection. Such a process
of spreading of the solute is called dispersion and is depicted schematically in
Figure 5.8.
Taylor (1953, 1954) was interested in the long-time behavior of the solute such
that the time scale associated with the convective transport of the solute is much
longer than the time scale associated with diffusion or
L a2
 (5.70)
wmax DA
or, in dimensionless form,
a
Pe =  Pe  1 (5.71)
L
where Pe = wmax a/D A is the mass transfer Peclet number, a is the radius of the
tube, and wmax is the fluid velocity at the centerline of the tube, and D A is the
diffusion coefficient, dropping the solvent subscript, B. Our job here is twofold:
to determine the precise distribution of the radially averaged concentration of
solute as a function of the streamwise variable and to determine how much the
average solute concentration spreads, quantified by the definition of a dispersion
coefficient.
On the basis of this long–time scale analysis, Taylor began with the convective–
diffusion equation for the mass fraction of species A in a fully developed flow,
given by
 
∂ω A ∂ω A 1 ∂ ∂ω A
+ w(r ) = DA r (5.72)
∂t ∂z r ∂r ∂r
with the boundary conditions
∂ω A
= 0 at r = a, 0 (5.73)
∂r
which expresses that no mass can cross the boundary and that the solution is
symmetric about the centerline, respectively. A slug of solute is injected at a
position in the channel in fully developed Poiseuille flow at t = 0, according to

ω A = ω0 δ(z − z 0 ) (5.74)

Typically, z 0 is taken to be zero. Taylor assumed that any mass transport is by

axial convection and radial diffusion and that the mixture is assumed to be dilute
in the species A. The task is to evaluate the axial spreading of the average mass
fraction as a function of axial position relative to the average velocity of the flow.
196 Essentials of Micro- and Nanofluidics

Notation alert: In keeping with Taylor’s original notation,  f  will denote an

average in this section.
In quantifying dispersion by defining a dispersion coefficient, what we will
eventually find is that the radially averaged mass fraction of solute will spread
according to a Gaussian distribution

2
mA0  − (z−wt)
ω A  = √
4K t
e (5.75)
2πa 2 ρ π K t
where K is the dispersion coefficient. This is the same solution as for the analysis
of the diffusion of a δ function source in an unbounded media with the diffusion
coefficient replaced by the dispersion coefficient. Skip the math, if you want to,
and just plot this as a function of z − wt for several times.
Taylor wrote the solution in a coordinate system moving with the average fluid
velocity z̄ = z − wt, where w = wmax /2, the average velocity in the tube.
In this coordinate system, the time derivative in the two coordinate systems are
related by
∂ ∂ ∂ z̄ ∂ ∂ ∂
|z = |z̄ − |z̄ = |z̄ − w |z̄ (5.76)
∂t ∂t ∂t z ∂t ∂t ∂t
After substituting for the average axial velocity and neglecting the axial diffusion
term, the governing equation becomes
   
a 2 wmax 1 2 ∂ω A 1 ∂ ∂ω A
−ξ = ξ (5.77)
DA 2 ∂ z̄ ξ ∂ξ ∂ξ
where ξ = r/a. No time derivative appears in equation (5.77) because the long
time solution is quasi-steady in the coordinate system moving with the average
velocity w.
The Poiseuille velocity expression is the velocity relative to the average velocity
in the tube in dimensionless form, with the axial velocity scaled on the maximum
velocity. Taylor argued that at large times, ∂ω A /∂ z̄ should, to a first approximation,
be independent of ξ ,3 an argument similar to the argument used in fully developed
heat transfer for constant heat flux. Equation (5.77) can then be integrated twice
to obtain
 
a 2 wmax ∂ω A 1 4
ω A (ξ, z̄) = ξ − ξ + ω A (0, z̄)
2
(5.78)
8D A ∂ z̄ 2
The average of a quantity is defined in dimensionless form as
1
ω A ξ dξ 1
ω A  = 0 1 =2 ω A ξ dξ (5.79)
0 ξ dξ 0

Taking the average of equation (5.78),

1 a 2 wmax ∂ω A 
ω A (ξ, z̄) = 2 ω A ξ dξ = + ω A (0, z̄) (5.80)
0 24DA ∂ z̄
3
An asymptotic analysis of the problem confirms this. See Datta et al. (2008).
 197 Heat and Mass Transfer Phenomena in Channels and Tubes

1 Pe=10
Pe=20
0.5 Pe=40

−0.5

−1

−1.5

0 0.2 0.4 0.6 0.8 1

Plot of the function ( P e/8) f (ξ ) from equation (5.81), with z̄ scaled by the radius a. The result indicates that
Figure 5.9
the concentration is lower than the average near the center of the tube and higher than the average near the
tube boundary as it flows through a channel or tube. Here P e = awmax / D A . The broadening is called Taylor
dispersion.

Substituting into equation (5.78), the departure from the average mass fraction is
given by
 
a 2 wmax ∂ω A 1 1
ω A (ξ, z̄) − ω A (ξ, z̄) = ξ2 − ξ4 − (5.81)
8D A ∂ z̄ 2 3
Writing the right side of this equation in dimensionless form, we have
 
 Pe ∂ω A 1 4 1
ω A (ξ, z̄) − ω A (ξ, z̄) = ξ − ξ −
2
(5.82)
8 ∂ ẑ 2 3
where ẑ = z̄/L,  = a/L, and Pe = wmax a/D A . Recalling that from equation
(5.71),  Pe  1, so that the amount of dispersion is small.
A plot of
 
Pe 1 4 1
f (ξ ) = ξ − ξ −
2
(5.83)
8 2 3
shows that it is negative in the center of the channel and positive at the wall,
as shown in Figure 5.9 for ∂ω A /∂ z̄ > 0, so that the mass fraction of the solute
is higher near the tube boundary ξ = 1. Conversely, for ∂ω A /∂ z̄ < 0, the mass
fraction is lower near the wall of the tube. This radial variation in mass fraction
originates from the radial variation in w and gives rise to dispersion or axial
spreading of the solute, as discussed later. The effect is more pronounced the
larger the Peclet number. Thus, near the center at higher velocities, diffusion
is outward toward the wall, whereas at the lower velocities near the wall, the
diffusion is inward.
To determine the dispersion coefficient in the form given by Taylor, we note
that the average mass flux of species A through the pulse relative to the average
velocity is given by
ρ a
n A  = (ω A (ξ, z̄) − ω A (ξ, z̄))(w − w)ξ dξ
πa 2 0
  
ρa 2 w2 ∂ω A  1 1 1 1
= ξ − ξ4 −
2
− ξ ξ dξ
2
(5.84)
DA ∂ z̄ 0 2 3 2
198 Essentials of Micro- and Nanofluidics

or
ρa 2 w2 ∂ω A  ∂ω A 
n A  = − = −Kρ (5.85)
48D A ∂ z̄ ∂ z̄
Here K is the dispersion coefficient
a 2 w2 DA 2
K = = Peave (5.86)
48D A 48
The dispersion coefficient measures the deviation of the average solute flux from
the product of the averaged mass fraction and averaged velocity. Note that in
most practical applications, dispersion should be minimized.
Let us now write a mass balance in terms of the average mass fraction across
the tube and across a plane moving with the average fluid velocity gives
∂ω A  ∂n A 
ρ =− (5.87)
∂t ∂ z̄
Equations (5.85) and (5.86) can now be used to predict the axial spreading origi-
nating from the velocity profile w(r ). Substituting for n A  from equation (5.85),
∂ω A  ∂n A  ∂ 2 ω A 
=− =K (5.88)
∂t ∂ z̄ ∂ z̄ 2
The solution to this equation for the case where the solute is initially at a point
having an averaged mass indicated by m A0  at time t = 0 is given by
 2
m A0  − (z−wt)
ω A  = √
4K t
e (5.89)
2πa 2 ρ π K t
Thus the radially averaged mass fraction of species A decays axially according
to a Gaussian distribution locally about the point z̄ = wt in the same way that a
solute released from a point in free space disperses with the diffusion coefficient
here replaced by the dispersion coefficient.
Strictly speaking, the Taylor solution is only valid for large Peclet number and
for axial diffusion negligible with respect to the dispersion (Peave > 7 approxi-
mately); Aris (1956) removed the assumption of large Peclet number and extended
Taylor’s solution to include axial diffusion. His result for the dispersion coefficient
K is
 
2
Peave
K = DA +1 (5.90)
48
√
Note that for Peave < 48  7, axial diffusion increases axial dispersion, and
√
for Peave > 48, axial diffusion decreases axial dispersion. Bird et al. (2002)
show a figure summarizing the regions of validity of the Taylor and Aris formulas.
They suggest that the long time limit corresponds to a dimensionless time of
DAt
∼1
a2
and so for a diffusion coefficient of D A ∼ 10−9 m2 /sec and a ∼ 1 µm, t ∼ 1 msec
is a small time scale indeed.
 199 Heat and Mass Transfer Phenomena in Channels and Tubes

It is evident that because of the similarity between the heat and mass transfer
equations, there is a similar Taylor–Aris dispersion problem for temperature,
with the mass fraction ω A replaced by a suitably defined average temperature
(Leal, 2007).

5.8 The stochastic nature of diffusion: Brownian motion

It was shown that the concentration distribution of a solute undergoing Taylor–

Aris dispersion from a point source satisfies a Gaussian distribution. This fact
was derived deterministically; however, the Gaussian distribution is most associ-
ated with a statistical approach, and this interpretation is discussed here. In this
statistical analysis, the diffusion coefficient emerges naturally as the variance of
all possible positions that molecules can inhabit.
The molecules in a stagnant gas undergo rapid and very high frequency oscil-
lations arising from the collisions with other molecules; the same is true in a
liquid, although the frequency is somewhat different because of the proximity of
other molecules. This oscillatory motion occurs simultaneously with any other
fluid motion, such as electro-osmosis, and is called Brownian motion after the
botanist Robert Brown (Daune, 1993), who showed that rapid oscillations of
molecules are ubiquitous (Stachel, 1998). Because these oscillations are so rapid,
corresponding to about 1014 collisions per second for a macromolecule in a liquid
(Daune, 1993), the precise trajectory of each molecule cannot be observed.

Robert Brown

Brownian motion is essentially a random process and considers the random

motion of n particles in the x direction. Suppose n(x, t) is the number of molecules
at the given position x at a given time t. Einstein (1905a, 1905b) first exhibited the
stochastic nature of diffusion using a simple approach, as follows, in the second
of five papers that changed the course of science in 1905 (Stachel, 1998). (He
200 Essentials of Micro- and Nanofluidics

eventually wrote a book on the subject (Einstein, 1956).) Fixing x at a later time,
say, t + τ , the total number of particles is given by the sum over all the possible
positions b from which the molecules could move from x − b to x, weighted by
a probability function p(b), or
∞
n(x, t + τ ) = n(x − a, t) p(b)db (5.91)
−∞

where the probability function p satisfies the constraints

∞
p(b)db = 1 (5.92)
−∞
∞
bp(b)db = 0 (5.93)
−∞

Expanding the function n(x, t) in a Taylor series in both space and time, assuming
small displacements, and then substituting into equation (5.91), it follows that
∂n τ 2 ∂ 2n ∞
n(x, t) + τ + = n(x, t) p(b)db
∂t 2 ∂t 2 −∞

∂n ∞ ∂ 2n ∞ b2
− bp(b)db + p(b)db + · · ·
∂x −∞ ∂x2 −∞ 2
(5.94)

The second term on the right-hand side vanishes, and the integral condition on
p results in the first term being just n(x, t). Thus, for small τ and small b, to
leading order,
∂n 1 ∞ ∂ 2n ∂ 2n
= b2 p(b)db = D (5.95)
∂t 2τ −∞ ∂x2 ∂x2
if the coefficient D is taken to be
1 ∞ σ
D= b2 p(b)db = (5.96)
2τ −∞ 2τ
where σ denotes the variance. Equation (5.95) is recognized as the one-
dimensional diffusion equation, and D is the translational diffusion coefficient,
which is linearly related to the variance of b. Note that because n and the con-
centration are proportional, equation (5.95) can also be written in terms of the
concentration.
The deterministic solution to this equation with a number of molecules n 0
initially concentrated at x = 0 is the same as the Gaussian distribution
n0 −x 2
n(x, t) = √ e 4Dt (5.97)
2 π Dt
where n(x, t) is the number of molecules per length and can also be interpreted
as the probability that a particle will reach the point x after N steps:
−x 2
P(x, N ) = (πσ 2 )−1/2 e σ2 (5.98)
√ √
where σ is the variance and σ = 2 Dt = N b.
 201 Heat and Mass Transfer Phenomena in Channels and Tubes

Reservoirs with solute

x=−L−a x=−a x=a x=L+a

No solute
initially

Geometry of the problem considered by Wilson. This problem is a simple model for drug delivery to a membrane,
Figure 5.10
here defined as the region −a ≤ x ≤ a.

The same type of analysis is valid for a set of solute particles in a solvent,
provided that the particles are large compared to the solvent molecules. In that
case, the diffusion coefficient is given by the Stokes–Einstein equation.
The effect of Brownian motion is usually not accounted for in models because
the long time effect on the trajectories of molecules and particles integrates out
because of the random nature of the process. This is certainly true in the absence
of solid surfaces. However, Brownian motion may be a factor when measuring
fluid velocities near a wall using neutrally bouyant particles because these effects
do not integrate out. The particles can bounce off the walls, which will obviate
the random nature of the process.
In the work of Sadr et al. (2007), designed to estimate slip length, the authors
show experimentally and computationally that Brownian motion can lead to an
overestimate of particle velocities near a wall and hence an overestimate for the
slip length. Moreover, the effect of Brownian motion may be even more important
in nanochannels under O(100 nm) in the smallest dimension. At such dimensions,
the particles are always near a wall, making a suitable Brownian correction
essential. Improving the spatial resolution of these velocimetry techniques will
require the use of smaller particles on the order of several nanometers in radius,
such as quantum dots, in contrast to the ∼100 nm particles used by Sadr et al.
(2007). The accuracy of methods to measure slip lengths at small scales is a
matter for future research.

5.9 Unsteady mass transport in uncharged membranes

The delivery of a solute to an uncharged membrane has been analyzed by Wilson

(1948), who considered mass transport to a central membrane, as depicted in
Figure 5.10. The governing equation is given by
∂c A ∂ 2c A
= DA 2 (5.99)
∂t ∂x
where the solute A may be an uncharged biomolecule such as glucose.
202 Essentials of Micro- and Nanofluidics

The unique feature of Wilson’s work is that instead of the concentration, he

works with
x
M(x, t) = c A (x, t)d x (5.100)
0

which is the total amount of the species per unit area of the cross section trans-
ported into the membrane from x = 0 to x = x. As depicted in Figure 5.11,
solute is contained in reservoirs surrounding the membrane, with only solvent
in the membrane. At time t = 0, fluid is allowed to diffuse into the membrane.
The membrane is of finite size compared with the upstream and downstream
reservoirs so that only a fixed amount of solute can diffuse into the membrane.
The concentration of solute in the reservoirs is assumed to be uniform in space
but not in time. Again, referring to Figure 5.11, the concentration at x = ±a will
vary with time.
Substituting equation (5.100) into equation (5.99), we find that

∂M ∂2 M
= DA 2 (5.101)
∂t ∂x
and symmetry of the problem requires

∂2 M
=0 at x = 0 (5.102)
∂x2
The boundary condition at x = a was described by March and Weaver (1928),
who analyzed the corresponding heat transfer problem. Suppose initially that the
concentration in the reservoir is c0 . Then its concentration at some time t later
will be c0 minus what has been diffused away, or
t ∂c A
Lc A (a, t) = Lc0 − D A |x=a dt at x = a (5.103)
0 ∂x

where L is the length scale associated with the reservoirs. Differentiation of this
equation with respect to time leads to

∂c A ∂c A
L = −D A at x = a (5.104)
∂t ∂x
The boundary condition in this form is difficult to implement because it is of
mixed type; that is, it contains both a time and space derivative.
Returning to equation (5.103), Wilson argues that the concentration of free
solute at x = ±a must be continuous so that the concentration of solute in the
bath is given by ∂ M/∂ x at x = a. Using the definition of M, and assuming
equation (5.101) holds at x = a, the condition of the total amount of solute
remaining fixed is

∂M
M+L = Lc0 at x = a (5.105)
∂x
203 Heat and Mass Transfer Phenomena in Channels and Tubes

This condition is recognized as a convection-type boundary condition similar to

that in heat transfer discussed in Chapter 3. The initial condition is
M = 0 for −a ≤ x ≤ a (5.106)
M = c0 L in the reservoirs (5.107)

This completes the specification of the problem.

It is, however, useful to remove the constant in the boundary condition at
x = a. Let M = Bx + f (x, t). Then the problem for f is
∂f ∂2 f
= DA 2 (5.108)
∂t ∂x
with conditions
∂2 f
= 0 at x = 0 (5.109)
∂x2
∂f
L + f =0 at x = a (5.110)
∂x
and initial condition
Lc0 x
f (x, 0) = − at t = 0 (5.111)
a+L
Here B = L/a + L (c0 ).
Note that the form of the solution for M is the superposition of the steady
state and a transient. The problem is solved using Fourier series methods, and the
eigenfunctions given by
qn x
f n = e− pn t sin (5.112)
a
satisfy the partial differential equation if
D A qn2
pn = (5.113)
a2
Thus the general solution is given by
 qn x
f (x, t) = An e− pn t sin (5.114)
n
a
where the symmetry condition at x = 0 rules out the cosine functions. The full
solution for the amount of solute in the membrane per unit area of the sheet is
given by
Lc0 x  qn x
M(x, t) = + An e− pn t sin (5.115)
a+L n
a
The coefficients An , pn , and qn are obtained by application of the boundary
conditions and, for example,
Lqn
tan qn = − (5.116)
a
2a Lc0 a
An = (5.117)
a2 + a L + qn L qn cos qn
2 2

The values of qn must be nonzero and positive.

 204 Essentials of Micro- and Nanofluidics

α= 0.2 α= 5
1 1
2
t D/a = 0.007
2
t D/a = 0.065
2 0.8333
t D/a = 0.327
2
t D/a = 0.654

2
t D/a = 0.016
c(x)/c0

c(x)/c0
2
t D/a = 0.161
0.5 0.5
t D/a2= 0.807
t D/a2= 1.615

0.1667

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/a x/a
(a) (b)

Sketch of the solution to Wilson’s problem for two values of α = L /a. Here c(x ) = c A /c0 is the dimensionless
Figure 5.11
concentration. (a) α = 0.2. (b) α = 5.

The amount of mass of solute absorbed into the membrane is given by M(a, t)
and
∞
M(a, t) 2α(α + 1) −βqn t
=1− e (5.118)
M∞ n=1
1 + α + α 2 qn2

where α = L/a and β = D A /a 2 , and

2a Lc0
M∞ = (5.119)
a+L
obtained by evaluating 2M(a, ∞). Note that the amount of solute absorbed by
the membrane is dependent on the dimensions of the reservoir and sheet and on
the initial concentration. In the limit as a/L → 0, we have

M∞ = 2ac0 (5.120)

and qn = (n + 1/2)π .
Figure 5.11 depicts results for two different values of α = L/a. The value of the
dimensionless steady state concentration is c(x) = L/L + a; at t D/a 2 ∼ 1.6, the
solution for α = 5 has reached steady state, while the result for α = 0.2, steady
state is reached much more quickly.
As an example, consider the diffusion of glucose into a nanopore membrane.
A typical reservoir size may be L = 40 mm and membrane length a = 1 µm.
For a solvent concentration of c0 = 4 mole/L, since L/a  1,
mole
M∞ = 2ac0 = 8 × 10−3
m2
Note that the amount of solute absorbed into the membrane is independent of the
diffusion coefficient; however, the rate at which the process occurs is dependent
on the diffusion coefficient through the characteristic time scale τ = a 2 /D A . For
glucose, D A ∼ 10−9 , and so τ ∼ 1 msec.
 205 Heat and Mass Transfer Phenomena in Channels and Tubes

1 1
y 1 y 1
δ δ
δm δT
x x
u=0 u=0
wall wall
XΑ=0
Sc>>1 Pr>>1
θ=0
(a) (b)

Sketch of the boundary layer thicknesses for the flow and heat and mass transfer from a flat plate; the
Figure 5.12
coordinate system (x , y) is shown. For a liquid δ  δT  δm , and the values of the scaled variables θ and
X A at the wall and the boundary layer edge are shown.

5.10 Temperature and concentration boundary layers

So far, we have considered internal flows exclusively. Yet the flow, heat, and mass
transfer in the entrance to a pipe or channel is similar to, though not quite the
same as, that in an external boundary layer flow. However, the external boundary
layer solution, as with the fluid dynamics problem, is a good first approximation
to the heat and mass transfer problems in the developing region.
Consider the steady flow past a flat plate unbounded on one side so that
the boundary layer on the plate is thin. This means that the Reynolds number
is nominally Re  1. Assuming a single length scale, say, L, the governing
equations for the mole fraction of a species A and the temperature are
∂XA ∂XA 1 ∂2 X A
u +v = (5.121)
∂x ∂y ReSc ∂ y 2

∂θ ∂θ 1 ∂ 2θ
u +v = (5.122)
∂x ∂y RePr ∂ y 2
Assume constant temperature and mass fraction boundary conditions so that,
in dimensionless form,

XA = θ = 0 y = 0 (5.123)

X A = θ = 1 at y = δm , δT (5.124)

XA = θ = 1 x = 0 (5.125)

where δm and δT correspond to the boundary layer edge, as in Figure 5.12. Here,
for example, the dimensionless temperature distribution is defined by
T − Tw
θ= (5.126)
T0 − Tw
with a corresponding expression for X A , where T0 and Tw are the temperature
at x = 0 and at the wall, respectively; here it is assumed that T0 = T∞ , the
206 Essentials of Micro- and Nanofluidics

temperature far from the plate. The velocity is a uniform free stream, and there
is no pressure gradient.
In the problems of interest here for liquids, Re  PeT  Pem , where PeT =
RePr is the thermal Peclet number and Pem = ReSc is the mass transfer Peclet
number. Note that for air and other perfect gases, Pr ∼ 1 and Sc ∼1. The Prandtl
number for water is Pe ∼ 5–15, depending on temperature, and for aqueous
solutions, the Schmidt number is Sc ∼ 1000. Thus the mass transfer boundary
layer is much thinner than the thermal boundary layer, which is just a bit thinner
than the velocity boundary layer. Because the Reynolds and Peclet numbers are
large, the temperature and mass fraction vary rapidly near the wall. On the basis of
this realization, and recalling the Blasius boundary layer problem of the previous
chapter, a boundary layer variable can be defined as ŷ = y/Re−1/2 . Clearly the
same similarity analysis that was done for the Blasius boundary layer problem
applies to the heat and mass transfer problems. Thus, using the same similarity
analysis, it is found that θ and X A satisfy

θ  + Pr f θ  = 0 (5.127)

X A + Sc f X A = 0 (5.128)

√
where the prime refers to differentiation with respect to η = ŷ/ 2x, the Blasius
boundary layer variable, and f is the Blasius boundary layer function, defined in
the previous chapter.
These two equations may be integrated analytically for arbitrary Prandtl and
Schmidt numbers, and with the outer boundary condition applied as η → ∞, the
solution for the temperature is
 η −Pr  η f ds
e 0 dη
θ = 0 ∞ (5.129)
∞ −Pr f ds
0 e 0 dη

and similarly for the mole fraction.

For large values of the Prandtl and Schmidt numbers, the thermal and mass
transfer boundary layers are much thinner than the fluid boundary layer, and so the
value of the function f near the wall may be used in place of the full numerical
function obtained from the Blasius solution. It is shown from the numerical
solution of the Blasius problem that f ∼ 0.664η near the wall with the present
variables, and so, for example,
 η −CPrη3
e dη
θ =  0∞ −CPrη3 (5.130)
0 e dη

where C = 0.332/3. Equation (5.130) indicates that for large Prandtl number, a
new boundary layer variable may be defined as η̂ = η/Pr−1/3 , which will be seen
to be the origin of the Pr1/3 dependence on the Nusselt number.
 207 Heat and Mass Transfer Phenomena in Channels and Tubes

The large Prandtl number dependence of the boundary layer variable would
suggest that the ratio of the boundary layer thicknesses should be
δT
∼ Pr−1/3 (5.131)
δ
with the same dependence of δm on the Schmidt number. However, the structure
of each of the equations is a bit different from the Blasius equation, and numerical
evaluation of the solution to the preceding temperature problem indicates that
δT
∼ Pr−0.4 (5.132)
δ
The heat flux at the wall is defined by
∂T ∂η
qw = −k | y=0 = −k(Tw − T0 )θ  (0) (5.133)
∂y ∂y
√
where η = ŷ/ 2x. The Nusselt number is thus given by
qw x hT x θ  (0)
N ux = = = − √ Re1/2 (5.134)
k(Tw − T0 ) k 2
x

where
−1
θ  (0) =  ∞ (5.135)
η −Pr f ds
0 e 0 dη

The value of θ (0) can be evaluated from the integral numerically in the general
case, and according to White (2006), a good curve fit valid for 0.1 < Pr < 10,000
is
θ  (0)
√ = 0.332Pr1/3 (5.136)
2
so that

N u x = 0.332Re1/2
x Pr
1/3
(5.137)

The Sherwood number is the analogue of the Nusselt number for mass transfer,
and the result follows easily from the result for the Nusselt number and
hm L X A (0) 1/2
Sh x = = √ Rex (5.138)
D AB 2
with the similar result that

Sh x = 0.332Re1/2
x Sc
1/3
(5.139)

5.11 Chapter summary

Heat and mass transfer share a number of common themes, and in this chapter,
this relationship is apparent. First, simple fully developed temperature profiles are
presented and analogies of these simple temperature distributions with Poiseuille
208 Essentials of Micro- and Nanofluidics

and Couette flow are identified. Next, it is shown that the mass transfer and
thermal entry lengths have the same form and are straightforward extensions of
the hydrodynamic entrance length. The definition of a fully developed temperature
distribution in a channel is presented and discussed for the two limiting cases of
constant heat flux and constant temperature at the bounding surface. However,
because of the magnitude of the Schmidt number, internal flow, mass transfer in
liquid flows is almost never fully developed.
What follows next is a survey of several problems that have analytical, exact
or nearly-exact solutions. These problem include:

r The Graetz problem: the axially developing temperature distribution in a fully

developed fluid flow in a rectangular channel, valid at large thermal Peclet
number
r Mass transfer in a thin liquid film for small diffusion coefficient
r Taylor-Aris dispersion, the spreading of a solute in a fully developed pressure
driven flow
r Unsteady mass transport in an uncharged membrane
r Heat and mass transport in high Reynolds number boundary layers.

It is also shown in this chapter that the diffusion coefficient can be interpreted
stochastically as the variance of the probability that a particle could originate
from a position x − b at time t given the position at a later time t + τ . This
interpretation of diffusion reinforces the concept that indeed, mass diffusion is a
random process.
Chapters 4 and 5 provide an overview of the disciplines within the thermal
sciences, called fluid mechanics and heat and mass transfer. In the next three
chapters we leave this discipline to study electrostatics, electrochemistry and
molecular biology all of which are necessary for the study of electrokinetic
phenomena to follow.

EXERCISES
5.1 Determine the solution for the velocity of the superposition of Couette flow
and Poiseuille flow, and identify the parameters for which the resulting
expression is equivalent to the superposition of one dimensional thermal
conduction and a duct with constant heat generation. Is there a reasonable
set of parameters such that the two dimensionless distributions have the
same value?
5.2 The dimensions of a typical channel used in microfluidic applications are
height h = 20 nm, length in the flow direction L = 3 µm, and width
W = 40 µm. Show that the concentration distribution will not be fully
developed for a flow rate of Q = 10−12 sec
m3
.
5.3 Write equation (5.14) in dimensionless form, and identify the appropriate
temperature scale.
209 Heat and Mass Transfer Phenomena in Channels and Tubes

5.4 Integrate the expressions for the Nusselt number and the Sherwood number
for the boundary layer over the length of the plate and thus define averaged
parameters. Calculate the ratio of these parameters for air and water.
5.5 Beginning with the dimensional form of the energy equation, verify by the
proper nondimensionalization the dimensionless equation and boundary
conditions for the Graetz problem. Identify the temperature scale.
5.6 Calculate the solution to the Graetz problem discussed in the text for Pr  1
in a cylindrical tube.
5.7 In electro-osmotic flow, the velocity in a channel can be a plug flow for thin
electrical double layers. Using the boundary conditions appropriate for the
Graetz problem in section 5.5, calculate the solution for the temperature in
dimensionless form for a wide channel, assuming constant properties and
u = U0 = constant. How would the solution for a circular tube be different?

h
x

air
T=T w
L

insulated

Liquid Film
Exercise 5.8

5.8 Water is flowing vertically in the direction of gravity down a plane wall,
as shown. Outside the liquid film is air. The energy balance is between
conduction and convection. Calculate the temperature of the water inside
the liquid film if the free surface is insulated (the thermal conductivity
of air is much smaller than that of water) and the wall temperature is
constant T = Tw at y = 0 and T = Te at the entrance x = 0. Write the
210 Essentials of Micro- and Nanofluidics

expression in dimensionless form, and identify the appropriate temperature

scale. Compare with the associated mass transfer problem.
5.9 The mass transfer Graetz problem for flow in a channel having solid walls
is
∂ω ∂ 2ω 2∂ ω
2
 Pey(1 − y) = + 
∂x ∂ y2 ∂x2
subject to ω = ω0 at x = 0 and
∂ω
= 0 at y = 0, 1
∂y
Here  = h/L and Pe = U h/D A . Show explicitly that the only solution
to this problem for arbitrary Peclet number Pe = Pem = ReSc is ω = ω0
and thus there is no dispersion. Does this result contradict the Taylor-Aris
solution?
5.10 The solution for an initially concentrated pulse of solute having mass m A0 
in an infinite medium is given by
 (z−wt)2
mA0 
ω A  = √ e − 4K t
2πa 2 ρ π K t
Show that this distribution satisfies equation (5.88). Calculate
∂ω A 
∂ z̄
Plot the solution at several fixed points η = z − wt as a function of time.
Write the solution in dimensionless form. Is it easier to plot?
5.11 Repeating the Aris procedure for mass dispersion in a tube, show that the
dispersion coefficient for a wide channel of height 2h is given by
 
2h 2 u2
K = DA 1 +
105D 2A
You might want to solve the problem in dimensionless form. See the dis-
cussion of dispersion in wide channels W/ h → ∞ in the paper by ? the
limit of which does not correspond to this result.
5.12 Extend the Wilson problem to the case where the reservoirs are not the
same size and the symmetry condition at x = 0 is no longer appropriate. Is
an analytical solution possible?
5.13 Certain types of cells are spherical and assume that the concentration of a
solute outside of the cells satisfies
 
∂c A 1 ∂ 2 ∂c A
= DA 2 r
∂t r ∂ ∂r
with boundary conditions

c A (r, t) = c0 for r ≥ a at t = 0
c A (r, t) = at r = a for t > 0
c A (r, t) = c0 for r → ∞ for t > 0
211 Heat and Mass Transfer Phenomena in Channels and Tubes

Find the solution for the concentration c A , and show that the steady state
solution is
cA a
=1−
c0 r
Hint: Make a change of variable C = r c A , and show that C satisfies the
planar equation.
5.14 A solute used as a drug is confined to a thin band of width 2a at time t = 0
so that
∂c A ∂ 2c A
= DA 2
∂t ∂x
This is the opposite of Wilson’s problem. Using Wilson’s procedure, find
the solution for the concentration in terms of the total amount of mass of
solute, M. Here the initial condition is

M = 0 | x |> a
M = c0 a | x |≤ a

How would you find the solution for x ≥ a or a → 0?

5.15 A solute used for drug delivery diffuses through tissue according to the
equation

∂c A ∂ 2cA
= D A 2 − ke c A
∂t ∂x
subject to a maintained concentration at x = 0

c A (x, t) = c0 for x = 0 for t > 0

c A (x, t) = 0 for x → ∞ for t > 0
c A (x, t) = 0 for t = 0

Here ke is an elimination constant and has units of sec−1 . Show that the
steady state solution is
 ke
cA −x
=e DA
c0
Write the solution in dimensionless form. Plot the solution for D A = 1 ×
10−7 cm2 /sec and ke = 10−8 − 10−4 sec−1 . Calculate the effectiveness of
drug delivery defined as
c̄ A
η=
c0
where
L
c A (x)d x
c̄ A = 0
L
is the average concentration in the tissue over a length L.
212 Essentials of Micro- and Nanofluidics

5.16 A solute satisfies

∂ 2c A
DA − ke c A = 0
∂x2
subject to a maintained concentration at x = 0

c A (x) = c0 for x = 0 for t > 0

dc A
N A = −D A = 0 for x = L
dx
Solve this problem, and calculate the average concentration
L
c A (x)d x
c̄ A = 0
L
Find an expression for the flux at x = 0. Take D A = 1 × 10−7 cm2 /sec and
ke = 10−8 − 10−4 sec−1 , L = 1 cm.
 6 Introduction to Electrostatics

6.1 Introduction

Electrostatics and electrodynamics are essential features of micro- and nanoflu-

idics. At the macroscale pumps, compressors and other fluid-handling devices
move fluid using a pressure difference, as was noted in Chapter 4. However,
as was demonstrated in Chapter 1, this is often not feasible at the micro- and
nanoscale because very large pressure drops on the order of atmospheres are
required for flows through channels under about 0.1 µm, channels that are often
used in applications.
The objective of this chapter is to introduce, in a general way, the concept of an
electric field at the advanced undergraduate level, typically associated with the
material presented in an introductory physics class for physics majors; for those
students already familiar with the principles of electrostatics, this chapter may be
scanned or skipped.
In this chapter, the basics of electrostatics are presented, paying particular
attention to the electric field associated with one or more charged spheres, planes,
or cylindrical surfaces. In the course of discussion, the concept of a surface charge
density is defined, which is a concept central to electrochemistry and micro- and
nanofluidics. This is followed by a discussion of the fundamental law of Gauss
relating the electric field to the charge density of a given medium. The concept
of electrical potential is then introduced, and the similarities between the velocity
potential of fluid mechanics are noted.
Next, the concept of an electric dipole is introduced, followed by the derivation
of the Poisson equation of electrostatics, which is solved for several different
situations. Current and current density are then defined, and all these topics
will lead into a detailed discussion of the electric double layer in an electrolyte
mixture, presented in the next chapter.

213
 214 Essentials of Micro- and Nanofluidics

6.2 Coulomb’s law: The electric field

Most of the physical laws bear the name of their originator and Coulomb’s Law is
no different. Charles Augustin de Coulomb (1736–1806) was a French physicist
from a rather aristrocratic family who performed experiments that resulted in the
law that bears his name. Actually while Coloumb’s Law bears his name it was a
contemporary, Joseph Priestley (1733–1804), a chemist and natural philosopher
who anticipated his results in a comprehensive treatise on electricity (Priestley,
1767) twenty years before Coulomb’s demonstration. Priestley is best known as
the discoverer of oxygen. Other of his books include Essay on the First Principles
of Government. A Renaissance man indeed.

Electrically conducting fluids are the norm in micro- and nanofluidics. To begin
to discuss how such electrically conducting fluids may be made to move under
the action of an electric field, it is useful to discuss the underlying theory of elec-
trostatics in some detail. Recall from the introduction of the concept of an electric
field in Chapter 1 that the force acting between two particles of charge q and q  is
qq 
F= (6.1)
4πe r 2
where e is the electrical permittivity of the medium and r is the distance between
the two charges; if the charged particles are atoms or ions, | q| = 1.602 × 10−19 C,
the charge of one proton.
The electric field at any point is defined as the force per unit charge acting on
a single charge at that point, or
F q N
E= 
= 2
(6.2)
q 4πe r C

where r is the distance from the point charge. This equation is known as Coulomb’s
law. Note that q is positive and that the electric field is directed outward for a
positive charge in the radial direction (Figure 6.1) and is directed radially inward
if the charge is negative.
If a number of charges are present, the force at charge qi , due to all the other
charges, is given by
 qi q j (ri − r j )
Fi = (6.3)
j=i
4πe | ri − r j |3

where ri , for example, is the vector distance of the charge qi from the origin of the
coordinate system. The electric field at the point qi , due to all the other charges,
is the vector sum of the field produced by all the charges:
1  q j (ri − r j )
E(ri ) = (6.4)
4πe i= j | ri − r j |3

The electric field lines for two charges are shown in Figure 6.2.
 215 Introduction to Electrostatics

5
4
3
2
1

Y 0 +
−1
−2
−3
−4
−5
−4 −2 0 2 4
X

Lines of constant electric field or electric field lines from a point charge in free space.
Figure 6.1

As an indication of the magnitude of an electric field, suppose a 100 volt

(recall that 1 volt = 1 Nm/Coul) battery consisting of two parallel plates 1 cm
apart generates an electric field of E = 104 N/C, where C stands for Coul. Then
the force on an electron in this field is given by

Fe = q E = 1.6 × 10−19 C × 104 N /C = 1.6 × 10−15 N (6.5)

Compare this value with the gravitational force,

Fg = mg = 9.1 × 10−31 kg × 9.8 N/kg = 8.9 × 10−30 N (6.6)

and note that the gravitational force is much smaller than the force generated by
the electric field.
In reality, electric fields are set up by charges distributed over linear length;
surfaces or volumes of conductors of finite size, that is, along a wire; the walls
of a channel; the surface or volume of a sphere; and the region between the walls
of a channel. In the general case, when charges are distributed over a volume, the
sum in the preceding equation becomes an integral, and

1 ρe (r  )(r − r  ) 
E(r ) = dV (6.7)
4πe V | r − r  |3

5 5
4 4
3 3
2 2
1 1
0 + + 0 + –
Y

−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−4 −2 0 2 4 −4 −2 0 2 4
X X

Electric field lines for two charges in the given configuration.

Figure 6.2
 216 Essentials of Micro- and Nanofluidics

y
dEy dE

dE z
y
r
θ
z

Geometry for calculating the electric field due

Figure 6.3
to a long charged wire.

where ρe is the volume charge density. Equation (6.7) is one of the many forms
of the Bio–Savart law, a form similar to that which appears in fluid mechanics.
Consider the case in which electrical charges are distributed over a surface.
Then, in the limit as q j → 0, the result is just the integral, and writing dq =
σ d A;
1 r̂ dq 1 r̂ σ d A
E= 2
= (6.8)
4πe r 4πe A r2
where σ is the surface charge density having units C/m2 . Here we have written
the result in terms of the unit vector r̂, where
(r − r  )
r̂ =
| r − r |
and r 2 =| r − r  |2 . The surface charge density plays an important role in elec-
trokinetic phenomena, and the electrochemistry of surface charge density will be
discussed in the next chapter.

6.3 The electric field due to an isolated large flat plate

To begin to derive the expression for the electric field due to a large, flat surface,
consider the field induced by a long charged wire, as depicted in Figure 6.3. This
is a classical problem that appears in many textbooks on electrostatics, and the
present treatment closely follows the approach of Sears and Zemansky (1964).
Assume that the wire is very long so that the integral for the electric field is
1 r̂ dq 1 ∞ r̂ λdz
E= 2
= (6.9)
4πe wire r 4πe −∞ r2
where, in this section, we use λ to denote the charge density per unit length. By
symmetry, there are two components to the electric field,
1 ∞ λcos θdz
Ez = − (6.10)
4πe −∞ r2

1 ∞ λsin θdz
Ey = (6.11)
4πe −∞ r2
 217 Introduction to Electrostatics

y
dE

dEy
dEx +
P
+ + + +
+ + r +
+ θ
+ +
+ +
z + dx +
+ +
+
+
x

Geometry for calculating the electric field due to a single charged plate.
Figure 6.4

where we have used the relations r 2 = y 2 + z 2 , z = −r cos θ, and y = −rsin θ

so that r = −ycsc θ. Thus yz = −tan θ so that z = −ycot θ and thus dz =
ycsc2 θd θ. Then, assuming that the line charge density λ is constant,
∞ dz π ycos θcsc2 θ 1 π
cos θ = dθ = cos θd θ = 0 (6.12)
−∞ r2 0 y 2 csc2 θ y 0

Thus E z = 0. Similarly, for E y , we have

λ ∞ dz λ π λ
Ey = sin θ 2
= sin θd θ = 2 (6.13)
4πe r −∞ r 4πe y 0 4πe y
Note that the electric field is oriented in a direction normal to the wire and decays
as 1/y as y → ∞.
The expression for the electric field induced by a charged wire can be used to
calculate the electric field due to a large flat plate, as also discussed in Sears and
Zemansky (1964). Assume that the plate is very large in the x and z directions of
Figure 6.4 and that the distance of the point P to the surface is much smaller than
the plate dimensions so that edge effects can be neglected, and in addition, assume
that the electric field does not vary in the z direction. Then the electric field will
be in a direction perpendicular to the plate, and d A = Ld x and dq = σ Ld x,
where L is the length of the plate in the z direction. Then, in terms of the charge
density per unit area,
dx
d E y = d Esin θ = 2σ sin θ (6.14)
4πe r
where r 2 = x 2 + y 2 and with sin θ = ry . Thus
∞ y dx
E y = 2kσ (6.15)
−∞ r r
Assuming that the surface charge density is independent of position on the plate,
dx σy ∞ dx
Ey = 2σ sin θ =2
plate 4πe r 4πe −∞ y2 + x2
σ y 1 −1 x ∞ σ
=2 tan |−∞ = (6.16)
4πe y y 2e
 218 Essentials of Micro- and Nanofluidics

dA

A charged surface showing outward normal d A

Figure 6.5
and the direction of the electric field.

This result is constant and independent of position. In addition, as long as the

dimensions of the plate are very large compared to the distance of the field point
from the surface, this result is independent of the shape of the surface. Moreover,
near any surface, the electric field is always normal to the surface.
The concept of a charged surface is critical to the transport of fluids at
nanoscale. Charged surfaces provide the driving mechanism for the generation
of electro-osmotic flow discussed in Chapter 9.

6.4 Gauss’s law

Carl Friedrich Gauss was a German mathematician who made significant con-
tributions to number theory. In the spirit of Fourier he could be described as a
physicist and engineer as well as a mathematician. Gauss’s law was formulated
around 1835 but not published until 1867.

Consider a single point charge in a dielectric material; then, taking the integral
of the charge over a surface surrounding it (Figure 6.5),
q
4π e E • d A = cos θd A = q d = 4πq (6.17)
S S r2 S

where d is the solid angle. Thus

e E • d A = q (6.18)
S

for a single charge. For a finite number of discrete charges,


e E • d A = qj (6.19)
S j

and for a continuous distribution of charges over a volume,

e E • d A = ρe dV (6.20)
S V
 219 Introduction to Electrostatics

where ρe is the volume charge density. Equation (6.20) is the integral form of
Gauss’s law for a continuous distribution of charges in a volume.
Using Gauss’s law, the electric field of a uniformly charged sphere of total
charge q over an area A = 4πr 2 at a given radius r is given by

e E • d A = e E A = q (6.21)
S
and so
q
E= (6.22)
4πe r 2
as obtained from the definition of the electric field.
Similarly, for a single charged flat plate, the result is

e S E • d A = σ A (6.23)

2e E A = σ A (6.24)

or
σ
E= (6.25)
2e
also as derived previously.

6.5 The electric potential

Consider the electric field given by a single point charge in a medium of permit-
tivity e at a position r, which we have given as
q
E= (6.26)
4πe r 2
Integrating from a point a to point b,
b bqdr
Edr = 2
a a 4πe r
 
q 1 1
= − (6.27)
4πe ra rb
If the integral is taken from b to a, then the result is the negative of the preceding
result. Thus, around any closed path, the integral of the electric field satisfies

E • ds = 0 (6.28)

This property of the electric field is the same condition that defines a thermo-
dynamic property. This turns out to be a general condition, and using Stokes’s
theorem,

E • ds = ∇ × E • n̂d A (6.29)
A

it is seen that the electric field is irrotational: ∇ × E = 0.

220 Essentials of Micro- and Nanofluidics

We then define the potential difference as

2
φa − φb = E • ds (6.30)
1

The electric potential is defined as the work done in moving a unit charge:
b
φ=− E • ds (6.31)
a

This formula is the analogue of the formula for mechanical work given by
b
W =− F • ds (6.32)
a

The units of the electric potential are Nm/C = 1 volt = 1 V.

The potential for a sphere of a single charge in an unbounded medium is given
by
∞ q
φ= Edr = (6.33)
r 4πe r
It is perhaps surprising that this distribution is virtually identical to the radial
velocity in potential flow for a point source or sink, which is given by
m
ur = (6.34)
r
where m is the strength of the source or sink.
Note that equation (6.33) implies that for this simple geometry,
∂φ
E =− (6.35)
∂r
so that the electric field is the radial derivative of the potential. In general, in
differential form,

dφ = − E • d l (6.36)

Conversely, the directional derivative of the potential along l is

dφ = ∇φ • d l (6.37)

and thus

E = −∇φ (6.38)

and it is noted that this relationship occurs often in mechanics; for inviscid and
irrotational flow and potential flow, the fluid velocity may be defined by a velocity
potential:

u = −∇φ (6.39)

It will turn out that this relationship implies that in the absence of charges in a
medium, u ∝ E; other situations in which the velocity is identical to the electric
potential arise in electro-osmotic flow where there is net charge near a surface,
and this point is discussed in Chapter 9.
 221 Introduction to Electrostatics

Eθ
P

r2
r r1
θ
−q q
d/2 d/2

Geometry for calculating the potential field due

Figure 6.6
to a dipole.

6.6 The electric dipole and polar molecules

An electric dipole is set up as a combination of a positive and negative charge,

and the expression for the potential is a linear superposition of each, according
to
 
q 1 1
φ= − (6.40)
4πe r1 r2
From Figure 6.7, using the law of cosines, r1 = r − (d/2)cos θ and r2 = r +
(d/2)cos θ so that
q dcos θ
φ= (6.41)
4πe r 2 − d42 cos2 θ
Now, for d  r ,
qdcos θ
φ= (6.42)
4πe r 2
The quantity p = qd is called the dipole moment and the unit of dipole moment
is 1 debye = 1 D = 3.336 × 10−30 Cm. As an example, the dipole moment
associated with two charges separated by 0.5 nm is p = 1.602 × 10−19 C × 0.5 ×
10−9 m = 0.8 × 10−28 Cm = 24 D. Small molecules have dipole moments on
the order of 1 D; for water vapor, d = 1.85 D (Israelachvili, 1992). Again, note
the similarity to the radial velocity induced by a potential doublet, the analogue
of the electric dipole in fluid mechanics,
cos θ
u r = −m (6.43)
r2
The electric field lines, or lines of constant electric field due to a dipole, are
shown in Figure 6.6. Note that the lines flow outward from the positive charge
and inward to the negative charge.
Molecules in a dielectric are classified as being polar or nonpolar. A nonpolar
molecule is one in which the center of mass of the positive nuclei and the electrons
coincide, while a polar molecule is one in which they do not. Symmetrical
molecules such as H2 , O2 , and N2 are nonpolar whereas water is polar; the polar
nature of water is illustrated in Figure 6.7. Under the influence of an electric field,
nonpolar molecules can be polarized; in this case, we call them induced dipoles.
 222 Essentials of Micro- and Nanofluidics

The polarization field strength is defined by

P = N ed (6.44)

where N is the number of dipoles per unit volume and d is the vector connecting
the two charges oriented in a direction from the negative charge to the positive
charge. For most materials P, the polarization vector
is in the direction of the electric field and is written in
terms of a susceptibility χ, so P = χe E.
The behavior of polar molecules in the absence of
an electric field and in the presence of an electric field
is depicted in Figures 6.8a and 6.8b, respectively. In
the presence of an electric field, the polar molecules
line up with the electric field lines. The behavior of a
nonpolar molecule in the absence of an electric field is
depicted in Figure 6.9a; in the presence of an electric
field, these molecules will orient in the direction of
A water molecule, illustrating
the polarization field, as in Figure 6.9b, which is very
Figure 6.7
its polar nature. similar to the polar molecule.

6.7 Poisson’s equation

Using Gauss’s law, along with the divergence theorem from calculus, it follows
that

e ∇ • EdV = ρe dV (6.45)
V V

+ − +
−
−+

− +

+
−
(a)
-+
-+
-+
-+
(b)

Polar molecules (a) in the absence of an electric

Figure 6.8
field and (b) in the presence of an electric field.
 223 Introduction to Electrostatics

+
+ -
-+
-
-+
+
+ - -+
- -+

(a) (b)

Nonpolar molecules (a) in the absence of an electric field and (b) in the presence of an electric field: Induced
Figure 6.9
dipoles.

Because the limits of integration are the same for each side of the equation, we
may equate integrands, and so
ρe
∇•E= (6.46)
e
This is the differential form of Gauss’s law. Because the electric field is the
gradient of the potential, we have
ρe
∇2φ = − (6.47)
e
where ρe is the volume charge density and, in general, is a function of (x, y, z).
This equation is Poisson’s equation for the electrical potential, as was derived in
Chapter 3.
As an example, let us calculate the potential field between two large plates
between which is fluid having volume charge density ρe (Figure 6.10). Consider
first the case of ρe = 0, which may be the case for deionized water. If the plates
are very long in both the x and z directions, the potential will satisfy

∂ 2φ d 2φ ρe
= =− (6.48)
∂y 2 dy 2 e
Thus, for ρe = 0,

φ = Ay + B (6.49)

+ E1 −
+ −
+ −
+ −
−
+ E2 −
+ −
+ −
−
+ y −
+ −
+ −
+ −
+ −
h −
+
−
+ −
+ −

Geometry for calculating the potential field between two large plates.
Figure 6.10
224 Essentials of Micro- and Nanofluidics

where A and B are constants. To find these constants, the potential or its normal
derivative at the two walls is required, as discussed in Chapter 3. At the two
surfaces, set φ = ζ0 at y = 0 and φ = ζ1 at y = h. Then
y
φ = (ζ1 − ζ0 ) + ζ0 (6.50)
h
so that the electric field is constant and
dφ (ζ1 − ζ0 ) σ
E =− =− = (6.51)
dy h e
Note that if ζ1 = ζ0 , the electric field vanishes, which is the case for a conductor.
The surface charge densities are thus defined by
dφ (ζ1 − ζ0 )
σ0 = −e | y=0 = −e (6.52)
dy h
dφ (ζ1 − ζ0 )
σ1 =  e | y=h = e (6.53)
dy h
There is a plus sign in the second equation because the outward unit normal to the
surface is in the negative y direction. Note that σ0 + σ1 = 0 to keep the system
electrically neutral. For water with ζ0 = 1 volt, ζ1 = 0, and h = 100 nm,
78.54 C2 Nm 107 C
σ0 = × 10−9 2
× 1 × = .00125 2 (6.54)
36π Nm C m m
Also, if both walls are of the same charge, the potential is constant and the electric
field E = 0.
For nonzero volume but constant charge density, the electric potential is given
by
−ρe0 2
φ= y + Ay + B (6.55)
20
and applying the boundary conditions,
−ρe0 2 ζ 0 − ζ1
φ= (y − yh) + y + ζ0 (6.56)
2e h
Note that for equal potentials on the two walls, which occurs often, the electric
potential is of the same parabolic form as the velocity in Poiseuille flow.
The electric field is
dφ ρe0 ζ0 − ζ 1
E =− = (2y − h) + (6.57)
dy 2e h
Integrating the potential equation once,
dφ dφ −ρe0 h
| y=h − | y=0 = (6.58)
dy dy e
or

σ1 + σ0 + ρe0 h = 0 (6.59)
 225 Introduction to Electrostatics

Current flows through an electric circuit, in this

Figure 6.11
case, a resistor.

This means that to be electrically neutral,

 
ρe0 h ζ1 − ζ0
σ0 = −e + (6.60)
2e h
 
ρe0 h ζ1 − ζ0
σ1 = −e − (6.61)
2e h
The present example may be of theoretical interest only because the volume
charge density is rarely, if ever, a nonzero constant in a given region.

6.8 Current and current density

In an electrical conductor, the charged species will “flow” in the direction of a

decreasing potential. The current density in the y direction is defined by
dφ
J = −σe (6.62)
dy
where σe is termed the electrical conductivity. In terms of the electric field,
J = σe E. Equation (6.62) is Ohm’s law; J is termed the omhic current; and if
the potential is linear in a given direction, as is the case in the previous section
with ρe = 0, then
σe φ
J =− (6.63)
h
where h is the length over which the potential changes (Figure 6.11).
The current is defined as Ie = J A, where A is the cross-sectional area. Defining
the electrical conductance as
σe A
Le = (6.64)
h
the current is given by Ie = L e φ. The units of current density and current
are amp/m2 and amp, respectively. The resistance of a conductor is defined as
 226 Essentials of Micro- and Nanofluidics

R = 1/L e , and so Ohm’s law is written as

φ = Ie R (6.65)
a familiar form in many physics and chemistry texts. Many of these texts use the
form V = Ie R = I R, where V is actually the potential drop across the circuit.
The unit of resistivity is the ohm, and the unit of conductivity is ohm−1 m−1 =
siemens/m.
In the example in the previous section that investigates the electric field between
two large plates,
dφ (ζ1 − ζ0 ) σ
E =− =− = (6.66)
dy h e
so that the current density is
σe σ
J= (6.67)
e

6.9 Maxwell’s equations

When a current is set up in an electrically conducting medium by an electric

field, there will also be a magnetic field, a result of the forces induced by the
moving charges that make up the current. When the electric and magnetic fields
are coupled, and the resulting equations are Maxwell’s equations.
Maxwell’s equations, which describe the flow of charge and the electric and
magnetic fields in moving media, are
∇ • B = 0 Gauss’s law for magnetism (6.68)
∂B
∇×E =− Faraday’s law (6.69)
∂t
 
∂D
∇ × B = μM J + Ampere’s law (6.70)
∂t
∇ • D = ρe Gauss’s law (6.71)

Here H is the intensity of the magnetic field, E is the electric field, D is the
displacement field, B is the magnetic induction field, J is the current density, and
ρe is the charge density.
If the material is an isotropic, permeable dielectric, then
D = e E, B = μ M H (6.72)

and Ampere’s law becomes

 
∂E
∇ × B = μM J + e (6.73)
∂t
where the magnetic permeability μ M is related to the electrical permittivity by
c = (μ M e )−1/2 and c is the speed of light. The unit of the magnetic induction
field is the tesla, 1 tesla = 1 volt sec/m2 . In a vacuum, μ M = 4π × 10−7 H/m,
 227 Introduction to Electrostatics

where H stands for henry. The magnitude of magnetic induction fields that are
measured can be on the order of 1 tesla. Even when the electric field is time
varying, the magnetic field term in Faraday’s law is negligible; for example, an
electric field of 1 V over a 10 µm channel will require a very short time scale
of about 10−10 sec for a magnetic field of B = 1 T to balance Faraday’s law.
Similarly, the left side of Ampere’s law dominates, unless the electric field is
very large. Thus, in most cases of practical interest for this book, the electric and
magnetic fields can be decoupled.

6.10 Chapter summary

In this chapter, the basic principles of electrostatics have been presented. The
two main dependent variables in electrostatics are the electric field, the force per
charge that exists between two charged bodies, and the electrical potential, the
gradient of the electric field. In Chapter 3, it was shown that the fluid velocity is
equivalent to the electric potential under rather general assumptions of the flow
in pipes and channels.
Several canonical solutions for the electric field and potential relevant to micro-
and nanofluidics have been derived, including the electric field due to a long wire
and a large flat plate. The concepts of surface and volume charge density have been
introduced and will be used extensively in the following chapters. The differential
form of Gauss’s law relates the electric field, and hence the potential, to volume
charge density and leads to a Poisson equation for the electrical potential.
The potential between two charged plates has been discussed, and the current
density has been defined. Both concepts will be used extensively in the following
chapters.
Finally, Maxwell’s electromagnetic equations are presented in SI units, and it
is shown that in micro- and nanofluidics, the magnetic and electric fields can be
decoupled – a significant simplification.

EXERCISES
6.1 Calculate the electric field and the electric potential due to a point charge of
magnitude e = 10−10 C (C = coulomb) at a distance r = 10 nm from the
charge in air. Repeat for water.
6.2 A finite-sized sphere has a surface charge density of magnitude σ =
−0.01 C/m2 . Show, using Gauss’s law, that the electric field is given by
Q
E=
4πe r 2

where Q = 4πσ a 2 is the total charge on the sphere in C. Calculate the electric
field at r = 100 nm from the surface of the sphere of radius a = 1 µm for a
charge density σ = −0.01 C/m2 in water and methanol.
228 Essentials of Micro- and Nanofluidics

6.3 Find the electric field using Gauss’s law for a conductor that has a surface
charge density on a wall of σ = σ0 for y = 0 and a volume charge density
ρ = ρ0 e−βy for y > 0. Hint: Note that there is only one component to the
electric field, and this is in a direction normal to the surface. Calculate the
integral

E • nd A
A

over the two appropriate surfaces.

6.4 Calculate the work done in charging a sphere to a surface charge density
σ = −0.05 C/m2 if the sphere diameter is D = 1 µm. Repeat for D = 10
nm.
q

−q
Exercise 6.5

6.5 A charge of magnitude q sits a distance d above a grounded plane wall (i.e.,
a wall having potential φ = 0) as shown. Show that the potential can be
represented as a superposition of the indicated charge and an image charge
placed a distance d below the plane wall. Neglect edge effects. Calculate the
surface charge density on the wall, assuming a two-dimensional distribution.
Show that
∞
σ (x)d x = −q
−∞

Note that the same method of images is used in fluid mechanics to satisfy
the condition that the velocity normal to the surface is zero.
6.6 In the Helmholtz view of the electrical double layer, the volume charge
density near a charged wall is given by
ρe = a + by 0 ≤ y ≤ d
ρe = 0 y ≥ d
Using the boundary conditions
dφ σ
= at y = 0
dy e
φ=0 y=d
integrate the Poisson equation to find the potential for 0 ≤ y ≤ d.
229 Introduction to Electrostatics

ε e1

y=0

ε e2
Exercise 6.7

6.7 A boundary separates two media having different dielectric properties. The
volume charge density is given by

ρei = ki φ

Find the potential in each region, and by integrating the governing equation
for the potential over a suitable region, show that the surface charge density
is given by
dφ dφ
σ = e1 |1 − e2 |2
dy dy
Suppose that region 1 is water and region 2 is silica. Calculate the surface
charge density for k1 = k2 = .01 1/m2 . How far from the wall in k units is
the potential φ = 0.01ζ , where ζ is the potential at y = 0?
 7 Elements of Electrochemistry and the
Electrical Double Layer

7.1 Introduction

A working definition of the field of electrochemistry is the study of charged

chemical and biological material. The field of electrochemistry appears to have
begun over 200 years ago, with the discovery by Galvani that a frog’s limbs
responded to the touching of its nerves by a scalpel with a severe twitch when the
scalpel was electrically charged by a nearby source (Bockris & Reddy, 1998).
Much of electrochemistry involves the study of the properties of ionic solutions
first studied by P. Debye and E. Hückel in the 1920s. The modern field of
electrochemistry is separated into two related but distinct aspects: ionic solutions
and electrode kinetics. Of course, in the problems of interest in this book, the
two are related; however, Bockris and Reddy (1998) point out that the two fields
today are significantly different, with the field of what they call “electrodics”
branching out to applications in the automobile industry, the study of batteries,
and fuel cells.
In this chapter, the essential principles of electrochemistry are presented,
enabling a thorough understanding of the chemical principles involved in the
study of electrokinetic phenomena to follow. Many biofluids are aqueous mixtures
of electrolytes, which are species that are either positively or negatively charged.
The mixtures of interest are aqueous and are usually dilute in the electrolyte
species. An example of a common electrolyte mixture is phosphate buffered
saline (PBS), which contains five different ionic species in an aqueous solution.
The chapter is divided into four parts. It begins with a presentation of the
structure of water and then proceeds to a qualitative discussion of the phenomenon
of hydration, in which water molecules can surround an ion due to the polar nature
of the water molecule.
The second portion of the chapter introduces the concept of the chemical and
electrochemical potential and the Gibbs function and their roles in chemical equi-
librium. In this portion of the chapter, the definition of acids and bases, central to
the field of modern electrochemistry, is discussed, leading to the characterization
of surfaces exposed to an aqueous solution. Many solids acquire a negative charge
when they come in contact with an aqueous medium (Hunter, 1981; Probstein,

230
 231 Elements of Electrochemistry and the Electrical Double Layer

1989). The development of electrical charge on the walls is a combination of

physical and chemical processes. The surface charge is acquired through a vari-
ety of mechanisms such as ion adsorption, exposure to charged crystal surfaces,
and ionization of surface groups (Masliyah and Bhattacharjee, 2006). Among
these, the ionization of surface groups plays a dominant role. For devices with
surfaces made of glass or silicon, surface silanol groups (SiOH) undergo depro-
tonation, which results in development of negative charge on the walls. Thus the
mechanics of how this process occurs is addressed, leading to the definition of
the surface charge density discussed in Chapters 3 and 6, and the relationship
between the surface charge and the ζ potential is derived.
If the surface is negatively charged, the positively charged ions (called the
counterions) in the vicinity of the surface tend to collect near the charged surface,
and an inner layer of counterions is attached to the surface. Outside this pinned
layer, ions float within the electrolyte solution near the surface. The total system,
including the pinned and floating ions or diffuse layer, is called the electrical
double layer (EDL) (Hunter, 1981).
The third portion of the chapter deals with the analysis of the electrical double
layer on the liquid side and determining the potential and species distribution
within the EDL, which is dependent on the value of the surface charge density
and hence the ζ potential on the wall. A relationship between the surface charge
density and ζ potential is derived based on linking the solid-side definition
of these quantities with that derived from the liquid definition. This leads to
the result that the surface charge density σ = σ (ζ, p H, K i ), where K i are the
equilibrium constants associated with the solid-side surface chemical reactions.
In this context, solutions for the electrical potential on spherical and cylindrical
surfaces are also calculated, and the interaction between a solid particle and a
plane wall is also discussed.
The chapter concludes with a discussion of electrical conductivity, semi-
permeable membranes that are characteristic of batteries and a discussion of
the Derjaguin approximation and its extension for the determination of the inter-
action energy between two surfaces.

7.2 The structure of water and ionic species

Water is the most abundant liquid on earth and, from a biological perspective,
makes up about 70 percent of a human cell’s weight (Alberts et al., 1998; Alberts
et al., 1994). A sketch of the structure of water is depicted in Figure 6.7; the angle
between the two hydrogen atoms and the oxygen atom is about 110◦ , and the two
hydrogen atoms are linked to the oxygen atom by a covalent bond in which the
different atoms share pairs of electrons. In the case of water, the oxygen atom is
surrounded by eight electrons. A single water molecule has a nominal diameter of
about 3 Å.
 232 Essentials of Micro- and Nanofluidics

Hydrogen Bond (Water Molecules)

O H
O Oxygen

H H H O H

Hydrogen H

Sketch of a hydrogen bond. Image courtesy of the National Institute of General Medical Sciences of the National
Figure 7.1
Institutes of Health.

Because water is polar, when two water molecules approach each other, the neg-
ative end of the one water molecule may closely approach the positive end of the
other water molecule. This rather weak electrostatic attractive force (Coulombic
interaction, defined by Coulomb’s law) can result in the formation of a hydrogen
bond (Figure 7.1). This type of bond is much weaker than the covalent bond and
can easily be broken by the random motion of the surrounding molecules. Indeed,
hydrogen bonds are extremely short term, forming and breaking very fast on the
order of picoseconds. However, in the bulk, the many hydrogen bonds are the
reason that water is a liquid at room temperature and not a gas (Daune, 1993).
Hydrogen bonds can occur even between differently charged portions of the same
molecule.
In the simple point charge (SPC) model of water, the distance between
the oxygen and hydrogen molecules in the H2 O molecule is 0.1 nm and the
HOH angle is 109.47◦ , and the dipole moment of this SPC model for water is
d = 2.27. There are many other models (Karniadakis et al., 2005; Sadus, 1997;
Becker & Karplus, 2006; Franks, 1972) of the structure of water, and in the
extended SPC model, the charges on the O and H molecules are assumed to
be qO = −0.8476 and qH = 0.4238. The dipole moment for the extended SPC
model is d = 2.35.
The localization of charge, as shown in Figure 6.7 allows the two regions to
bond to ions and other charged molecules, thus making the water molecule highly
polar, with a positive charge located on one end and a negative charge on the
other. In general, polar molecules have a high dielectric constant and are good
electrical conductors. Conversely, hydrocarbon solvents, such as methanol, that
are nonpolar have lower dielectric constants and are poor conductors.
In general, polar molecules that can form hydrogen (anions) and ionic (cations)
bonds dissolve readily in water. These types of molecules are called hydrophilic,
or water loving. Ions are an example of a hydrophilic molecule because they can
easily be surrounded by one or more water molecules; in this case, the ions are said
to be hydrated. More will be said about hydration later. Conversely, uncharged
molecules, such as the hydrocarbon family, that contain many C–H bonds are
water hating or hydrophobic. Cell membranes contain many of these types of
 233 Elements of Electrochemistry and the Electrical Double Layer

bonds, and thus portions of cell membranes are hydrophobic. The structure of
cell membranes is discussed in the next chapter.

7.3 Chemical bonds in biology and chemistry

Molecules, and indeed all matter are made of atoms, which are arranged in a
specific way to form molecules. The individual atoms are held together by chem-
ical bonds. Each atom is composed of a positively charged nucleus consisting of
protons and neutrons surrounded by a cloud of negatively charged electrons. An
atom as a whole must be electrically neutral.
The electrons are arranged in a rigid series of at most five shells, with the
electrons in the innermost shell bound most tightly to the nucleus. An atom is
most stable when each of the possible shells (depending on the atom) is filled.
The innermost shell can hold at most 2 electrons, the second and third 8 electrons,
and the fourth and fifth 18 electrons. In the biological world, atoms with more
than four shells are rare (Alberts et al., 1998).
How atoms combine to form molecules is determined by the number of elec-
trons in the outermost shell. For example, hydrogen has one electron in its only
shell, and thus it can combine with other atoms that can provide the electron to
fill its outer shell; that is, hydrogen is highly reactive. The valence is defined
as the number of electrons an atom must gain or lose to have a fully populated
outer shell. Virtually all the elements in the living world have unfilled outer
shells. Those elements having filled outer shells are chemically unreactive; two
examples are helium and argon.
Atoms with unfilled outer shells exchange electrons. How they do this leads to
the existence of a chemical bond between two or more atoms. When two atoms
share a pair of electrons, the bond between the two atoms is called a covalent
bond. Similarly, an ionic bond is formed when an electron is transferred from
one atom to the other.
There are several different types of covalent bonds. Most covalent bonds share
two electrons and are called single bonds. When four electrons are shared, two
between each atom, the bond is called a double bond. The bond length of a
double bond is shorter, and the double bond is stronger than the single bond.
The van der Waals force is an attractive–repulsive force that is not due to
elecrostatic or covalent bonding. This force in a gas is characterized by the van
der Waals equation of state (Moran & Shapiro, 2007), which is a modification of
the perfect gas law to account for molecular size and intermolecular forces. The
effect of van der Waals forces in a liquid is about 106 times that in a gas because
of the proximity of individual molecules, making the influence of intermolecular
forces that much greater.
Chemical reactions play an important role in determining properties of bound-
ing surfaces in nanopore membranes; for example, the surface charge on a silica
surface is determined by a reaction between the uncharged silica solid and the
 234 Essentials of Micro- and Nanofluidics

hydrogen atom in water. In particular, the nature of the silica surface is discussed
in detail later in this chapter.

7.4 Hydration of ions

As has already been mentioned, ions may become hydrated, that is, surrounded by
water molecules (Figure 7.2). A molecule is a discrete set of atoms held together
by chemical bonds. Inherent in the discussion of hydration phenomena is the
concept of solubility. In general, like dissolves like; that is, polar molecules, such
as ions, are more soluble in a polar solvent than in a nonpolar solvent. Thus the
ion–water interaction is stabilizing; that is, it is more energetically favorable.

Albert Einstein and Hydration

Albert Einstein was writing his dissertation at a time when even the existence of
atoms and molecules was being questioned. He showed his fluid dynamics prowess
with the derivation of the Stokes-Einstein formula for the diffusion coefficient as
discussed in Chapter 2. But also in his dissertation, in talking about the viscosity
of an aqueous sugar solution, he derived the formula for the viscosity of the
solution as
k∗
=1+
k
where  is the volume of the sugar molecules per unit volume of the water. Stachel
(1998) points out an error in this equation that Einstein corrects in a publication
in 1922. The correct equation is
k∗
= 1 + 2.5
k
Einstein quotes a value of  = 0.0245 “from Burkard’s observations (Landolt
and Börnstein’s Tables)”. In a simple calculation, he found that the density of a
1% aqueous sugar solution increased roughly linearly with sugar concentration by
a factor of 1.0061. Thus the increase in viscosity is four times that of the increase
in density! He concludes:

Thus, while the sugar solution behaves like a mixture of water and solid sugar
with respect to its density, the effect on viscosity is four times larger than what
would result from the from the suspension of the same amount of sugar. It
seems to me that, from the point of view of molecular theory, this result can
only be interpreted by assuming that a sugar molecule in solution impedes the
mobility of the water in its immediate vicinity, so that an amount of water that
is three times larger than the volume of the sugar molecule is attached to the
sugar molecule.
Einstein deduced that a sugar molecule must be hydrated! Such a phenomenon
was the subject of heated debate at the time (Stachel, 1998).
 235 Elements of Electrochemistry and the Electrical Double Layer

On the basis of the analysis of the the

electrical double layer (EDL) surround-
ing an ion, as discussed in Chapter 1, the
local electric field can be on the order of
1010 V/m. Under the influence of this large
electric field, the polar ends of water mole-
cules can electrically bond to the ion, creat-
ing a “molecule” whose effective radius is
larger than the unhydrated ion. A sketch of
a hydrated ion is depicted in Figure 7.2. Val-
ues of the hydrated ionic radius can range
from 3 to 5 times the unhydrated ion radius.
Sketch of the hydration of a sodium ion by water
Figure 7.2
molecules. For example, the ratio H = a H /a, where
a is the unhydrated radius and a H is the
hydrated radius, is typically H = 1.83 for
Cl− and H = 3.91 for Na+ (Conway, 1981). Multivalent ions typically have
hydration radii larger than monovalent ions because of the stronger electric fields
around the multivalent ion.
The hydrated radius of ions is often calculated from the Gibbs free energy
(Daune, 1993). One way to analyze the hydration process is to calculate the work
required to bring an ion from a vacuum with dielectric constant e1 = 1 to the
water medium having a dielectric constant of e2 ∼ 78–80. This work is given by
integrating the electrical force over all charges, or

 
q2 1 1
WH = − (7.1)
8π0r e1 e2

where r is the radius of hydration; this formula was first given by Born (Daune,
1993). This work value is equivalent to the Gibbs free energy of hydration W H =
G H , and the radius of hydration can then be calculated from measured values of
this work.
Several radii of hydrated ions are depicted in Table 2.6. Care should be taken
in interpreting the radii depicted in Table 2.6. The ionic radius is that radius of
the ion as a crystalline solid, assuming spherical symmetry. The higher the cation
charge, the larger the hydrated radius, and the radius of a cation is nearly its van
der Waals radius (Daune, 1993), the radius of a molecule modeled as a single hard
sphere. Moreover, owing to Brownian motion, the molecules will be vibrating
at a very high frequency, and as such, the radii depicted in Table 2.6 must be
considered time averages over a large number of ions.
Hydration is also measured by the coordination number, the number of solvent
molecules around an ion, and the solvation number, a measure of the dynamic
state of the ion–solvent complex as they move in solution. This subject is far
beyond the scope of this book, and the details of the definitions of these two
numbers are given in Bockris and Reddy (1998).
 236 Essentials of Micro- and Nanofluidics

For the purposes of this book, the most important property of an ion is its
hydrated size, and in addition to the experimental data depicted in Table 2.6, the
solvated radius can be obtained from ab initio molecular simulations.

7.5 Chemical potential

The chemical potential is an important quantity for establishing under what

conditions equilibrium will exist. The term equilibrium is often used, and it is
useful to discuss what the term means in the context of a chemical reaction.
Suppose a chemical reaction takes place until completion, and no other changes
can be observed. Then the system is said to be in equilibrium if the reverse
reaction takes place and the original starting state is recovered (Butler, 1998).
The extensive Gibbs function in thermodynamics, defined in units of joules
for a pure fluid, is defined as (Moran & Shapiro, 2007)

G = H − T S, (7.2)

where H is the extensive enthalpy in joules = J or kJ, S is the extensive entropy,

and T is temperature. The Gibbs function, or Gibbs free energy, is a property of
the state of the system and taking the differential

dG = d H − T d S − SdT (7.3)

From the first and second laws of thermodynamics, T d S = d H − V d p, and so

dG = V d p − SdT (7.4)

Thus, at constant temperature and pressure, equilibrium requires that dG = 0.

For a fluid mixture, the Gibbs function is also a function of the number of
moles of each species present in the mixture. Thus G = G(T, p, n1 , n 2 , . . .), and
so, in the general case,

m
dG = V d p − SdT + μi dn i (7.5)
i=1

where μi is the chemical potential and n i is the number of moles of species i.

From the definition of the differential of a function, the chemical potential μi is
thus defined as
∂G
μi = (7.6)
∂n i T, p,n j
and so the units of chemical potential are J/mole.
The chemical potential is what is termed a partial molar quantity. For any
extensive property Y = Y (T, P, n i ), a partial molar quantity is defined by taking
the partial derivative of Y with respect to n i , keeping the temperature, pressure,
and n j fixed. For example, the partial molar volume is defined by
∂μi
Vi = |T, p,n j (7.7)
∂n i
237 Elements of Electrochemistry and the Electrical Double Layer

For any extensive property Y = Y (T, p, n i ) and for any constant α,

αY (T, p, n i ) = Y (T, p, αn i ) (7.8)

Differentiating with respect to α, we find that


m
∂Y
Y (T, p, n i ) = ni , (7.9)
i=1
∂(αn i )

which holds, in particular, for α = 1. This means that the extensive Gibbs function
is also defined by

m
G = μN = μi n i (7.10)
i=1

for a mixture of m components, where N is the total number of moles in the

system.
For a single component, in molar units, G = μN or μ = G/N = ḡ, where ḡ
is the specific molar Gibbs function in units of J/mole. For example, assuming a
perfect gas, the chemical potential of air, taken to be a pure fluid, is
G
μ= = ḡ = h̄ − T s̄ = T (c̄ p − s̄) (7.11)
N
in molar units. In mass units,
G
μ= = g = h − T s = T (c p − s) (7.12)
M
Most often, as will be seen, changes in the chemical potential are important
rather than the specific value of the chemical potential. For air, using property val-
ues from thermodynamic tables, the change in chemical potential from T = 290 K
to T = 310 K at constant pressure in mass units is to three digit accuracy

μ = 310(1.005 − 1.735) − 290(1.005 − 1.668)

kJ
= −226.300 + 192.270 = −34.030 (7.13)
kg
To obtain the same value on a molar basis, the mass result is multiplied by the
molecular weight of air Mw = 29 kg/kmole.
As another example, consider the change in chemical potential as liquid water
becomes vapor at T = 100◦ C. Using thermodynamic tables for the entropy val-
ues,
kJ
μ = 2257 − 373 × (7.3549 − 1.3069) = 2257 − 2256 ∼ 0 (7.14)
kg
To the accuracy that the tabulated properties of water are measured, the change
in chemical potential is zero. In fact, this is a statement of phase equilibrium: that
a liquid in equilibrium with its vapor has the same chemical potential, or

μvap = μliq (7.15)

238 Essentials of Micro- and Nanofluidics

This relationship also holds for any component in a mixture:

μi,vap = μi,liq (7.16)

For a mixture, the differential change in Gibbs function from equation (7.10)
is given by

m 
m
dG = μi dn i + n i dμi (7.17)
i=1 i=1

Consequently, equating the two expressions for dG readers


m
n i dμi = V d p − SdT (7.18)
i=1

Equation (7.18) is called the Gibbs–Duhem equation and provides the neces-
sary relationship between temperature, pressure, and chemical potential changes.
Thus, at constant temperature and pressure, the chemical potential of each species
is constant.
Consider now the case of a perfect gas mixture of m components. Then the
chemical potential μi is defined as

μi = μio (T ) + RT ln X i for i = 1, . . . , m, (7.19)

where R is the universal gas constant and X i is the mole fraction. The quantity
μio is the chemical potential of the gas (or vapor) at a suitably defined standard
state. For a gas, the quantity μio is usually independent of pressure, at least if the
gas is perfect.
For example, air, though, for practical purposes, it is considered a pure sub-
stance, is about 79 percent nitrogen on a molar basis, and the quantity
kJ kJ
μN2 − μoN2 = 8.314 × 298 K ln 0.79 = −584.02 (7.20)
kmole K kmole
at T = 298 K. Because the molar mass of nitrogen is MN2 = 28.01, μN2 − μoN2 =
−584.02/28.01 = −20.85 kJ/kg.
For a process in which a given species in a gas changes composition from X i,1
to X i,2 , the change in chemical potential at constant temperature for a perfect gas
is given by
X i,2
μi = RT ln (7.21)
X i,1
where 1 and 2 refer to the initial and final states, respectively. Thus
μ
X i,2 = X i,1 e− RT (7.22)

which is the classical Boltzman distribution.

For a liquid mixture, the chemical potential is defined by (Denbigh, 1971)

μi = μi∗ (T, p) + RT ln ai for i = 1, . . . , m, (7.23)

where now the reference chemical potential μi∗ (T, p) is a function of temperature
and pressure, but is in general unknown, and ai is the activity. The activity
239 Elements of Electrochemistry and the Electrical Double Layer

coefficient is defined as γi = ai /X i , where xi is the mole fraction of species

i in solution. The activity coefficient is, in general, a function of temperature,
pressure, and mole fraction. A liquid mixture is said to be an ideal solution if
γi = 1 so that the activity may be identified with the mole fraction of species i.
Consider now a nonideal or concentrated binary mixture. Then, using the
definition of the intensive Gibbs function in equation (7.10), the intensive Gibbs
function on a molar basis is given by
g(T, p, x A ) = X A μ∗A + (1 − X A )μ∗B
+ RT (X A ln X A + (1 − X A ) ln(1 − X A )) + g E , (7.24)
where g E (T, p, X A ) is called the excess Gibbs function and is nonzero for
nonideal mixtures. Using the definition of the activity coefficient, the excess
Gibbs function is
g E (T, p, X A ) = RT (X A ln γ A + X B ln γ B ). (7.25)
In concentrated, nonideal solutions, models for the excess Gibbs function are
required. A detailed analysis of such models is beyond the scope of this book,
and the solutions of interest in micro- and nanofluidics are most often ideal.
It is worthwhile to mention several of the most popular methods; these are the
Margules, Val Laar, Wilson, and non-random two-liquid (NRTL) models. These
models are discussed in great detail in Taylor and Krishna (1993) and, in general,
are derived from experimental data.
In the case of ionic species, the chemical potential takes the form
μ B = μ∗B + RT ln a± (7.26)
√
where a± = γ± X ± , where X ± is the mole fraction and γ± = γ+ γ− is called
the mean activity coefficient. In the Debye–Hückel limit, the activity coefficient
is given by
√
A D H z+ z− I
− ln γ± = √ (7.27)
1 + BD H a I

where I is the classical ionic strength for a binary mixture, I = (1/2) ci zi2
(Fawcett, 2004), and a is the radius of the ion. The constants A D H and B D H are
defined by
 
N A e2 F 2000 1/2
ADH = (7.28)
8π (e RT )3

 1/2
2000
BD H = F (7.29)
e RT
The factor of 1000 is inserted to convert mole/liter to mole/m3 because Faraday’s
constant and the electrical permittivity include a meter in their units. Fawcett
(2004) provides values for these constants and for a 1 mM solution at 25◦ C, for
univalent electrolytes γ± = 0.96, neglecting the effect of the constant B D H since
a is small.
 240 Essentials of Micro- and Nanofluidics

As pointed out by Denbigh (1971), the definition of the standard state quantities
μi∗and μio is not always standard, and care must be taken in examination of the
experimental data for these quantities. The same is true of the activity coefficient.

7.6 The Gibbs function and chemical equilibrium

As noted in the previous section, the Gibbs function plays a crucial role in the
establishment of chemical equilibrium. Consider the extensive form of the first
law of thermodynamics for an open system:
δ Q − δW = d H = dG + T d S (7.30)
where the process is assumed to occur at constant temperature. From the second
law of thermodynamics,
T d S = δ Q + T dσ (7.31)
where σ ≥ 0 is the entropy production in units of, say, kJ/K. Substituting into the
first law
−δW = dG + T dσ (7.32)
or
dG = −T dσ − δW (7.33)
Thus, whatever the sign of the work done, dG ≤ δWmax .
Now suppose δW = 0. Then it is clear that dG ≤ 0 since dσ ≥ 0. Note that
dG > 0 is impossible, and so this restriction on the Gibbs function follows from
the limitations on the entropy change for a given process. Thus the value of
the Gibbs function must decrease during a process until equilibrium is reached
when dG = 0, as we saw from the Gibbs–Duhem equation. Note that constant
temperature and pressure have been assumed.
Suppose a general reaction is represented by the equation
ν A A + νB B νE E + νF F (7.34)
Then, if the reaction proceeds at constant temperature and pressure, the condition
of equilibrium is given by

m
dG T,P = μi dn i = μ A dn A + μ B dn B + μ E dn E + μ F dn F = 0 (7.35)
i=1

For example, for the forward reaction, dn A = −kν A and dn E = +kν E , with
similar expressions for the other two species. Here k is a proportionality con-
stant, depending on the amount of the reactants, and is normally small. Because
dG T,P = 0, it must be that
νE μE + νF μF − ν A μ A − νB μB = 0 (7.36)
Equation (7.36) is known as the equation of reaction equilibrium.
241 Elements of Electrochemistry and the Electrical Double Layer

If a liquid mixture is ideal, it has been seen that the chemical potential is
defined as

μi = μi∗ (T, p) + RT ln X i for i = 1, . . . , m (7.37)

and substituting into the equation of reaction equilibrium

ν E (μ∗E (T, p) + RT ln X E ) + ν F (μ∗F (T, p) + RT ln X F )

−ν A (μ∗A (T, p) + RT ln X A ) − ν B (μ∗B (T, p) + RT ln X B ) = 0 (7.38)

where it is recognized from the previous section that the chemical potential
μi∗ (T, p) = gi∗ , the intensive Gibbs function. Rearranging the terms of this equa-
tion, it is readily seen that
 
∗ (X E )ν E (X F )ν F
−G (T, p) = RT ln (7.39)
(X A )ν A (X B )ν B
where G ∗ (T, p) is defined by

G ∗ (T, p) = ν E μ∗E (T, p) + ν F μ∗F (T, p) − ν A μ∗A (T, p) − ν B μ∗B (T, p)

(7.40)
Note that the stoichiometric coefficients νi are given from the reaction. The
equilibrium constant is then defined as
 
(X E )ν E (X F )ν F
K = ln (7.41)
(X A )ν A (X B )ν B
and so in terms of the change in Gibbs function for the reaction,
G ∗ (T, p)
K = e− RT (7.42)

where, in all of this analysis, R is the universal gas constant. Of course, this
analysis can be extended for any number of reacting species and for simultaneous
reactions and ionization processes.
The equilibrium constants are, in most cases, determined from experiments.
The equilibrium mole fractions in the case of an ideal liquid mixture are then
determined from the solution of a set of nonlinear equations (Moran & Shapiro,
2007). This procedure also applies to ionic mixtures, as will be described for
reactions that occur on silica surfaces in Section 7.9.
To illustrate the process, consider the reaction of gaseous carbon monoxide
with water vapor. The theoretical chemical reaction, the completed reaction is
given by

CO2 + H2 CO + H2 O (7.43)

In actuality, there may be some reactants left in the products, and so the actual
reaction may be

CO2 + H2 wCO + xCO2 + yH2 O + zH2 (7.44)

242 Essentials of Micro- and Nanofluidics

It is common to use the number of moles of each species in the definition of the
equilibrium constant, which is
 1+1−1−1
wy p
K = (7.45)
x z pr e f N T
where N T = w + x + y + z is the total number of moles in the product gases.
Note that the reverse reaction has an equilibrium constant of 1/K . Equation (7.44)
needs to be balanced by conserving the moles of C, O, and H. These equations
are

2 = w + 2x + y O balance (7.46)

1 = w + x C balance (7.47)

2 = 2y + 2z H balance (7.48)

Equation (7.45) becomes

wy
K = (7.49)
xz
These are four equations in four unknowns for x, z, w, and y that may be solved
in Matlab or some other software package. However, the equilibrium constant
K must be known; these values are given in many different places, including
in undergraduate textbooks on thermodynamics and in the National Institute of
Standards (NIST) database. Its value depends on the temperature of reaction.
At room temperature, T = 298◦ , log10 K ∼ −0.5 or K = 10−5 , and the reac-
tion does not proceed. Conversely, at T = 1600◦ , log10 K = 0.474, K = 2.979,
and for this value of the equilibrium constant, x = 0.367, y = 0.633, z = 0.367,
and w = 0.633. Note that because of the nonlinearity of the equilibrium constant
equation, there may be multiple solutions, with only one being relevant. Note
also that the total number of moles has been conserved. One can plot the yield
of this reaction as a function of temperature by plotting the results for a number
of values of K (Moran & Shapiro, 2007). Can you combine these four equations
into a single nonlinear equation for one of the variables?
Various salts, acids, and bases dissociate when in contact with water. For
example, in aqueous solution, sodium chloride dissociates according to

NaCl(solid) Na+ (aq) + Cl− (aq) (7.50)

The same methodology leads to the definition of the equilibrium constant as

x2 1 x2
K = = (7.51)
1−x 1+x 1 − x2
where, for example, N+ = N− = x is the number of moles of Na+ . Thus

K
x= (7.52)
1+K
 243 Elements of Electrochemistry and the Electrical Double Layer

Inlet and outlet

ports

Nanopore
membrane

Donor Receiver

Depiction of donor and receiver regions separated by a nanopore membrane.

Figure 7.3

In N aCl dissociation, the equilibrium constant for this ionic reaction is large,
and so the dissociation is complete, and x = 1. Multiple reactions may occur
in many situations, and this case will be considered later in this chapter, in the
section on silica surfaces.

7.7 Electrochemical potential

Analgous to the chemical potential, the electrochemical potential in J/mole is

defined by

μi = μi∗ + RT ln ai + z i Fφ (7.53)

where R = N A kb is the universal gas constant; F is Faraday’s constant, F =

N A e = 96,500 C/mole; and φ is the electrical potential. For the dilute solutions
of interest, here the activity is assumed to be equal to the mole fraction, ai = xi .
Thus

μi = μi∗ + RT ln X i + z i Fφ (7.54)

In the same way that the chemical potential is constant as a condition of

equilibrium for uncharged species, the electrochemical potential is constant for
charged species (ions). Note that at T = 300 K, RT = 2494 J/mole for R =
8.314 J/mole K. For a monovalent ion and a potential of φ = 1 V = 1 J/C,
Fφ = 96,500 J/mole and so often at high potentials, the electrostatic term can
dominate.
To illustrate the use of the electrochemical potential, consider a charged species
in an electrolyte solution in an upstream and downstream reservoir separated by
a porous membrane of some type, as depicted in Figure 7.3. At equilibrium
for constant temperature and pressure, the chemical potentials in the donor and
receiver must be equal so that μ D = μ R , where D stands for donor reservoir and
R stands for receiver reservoir. The chemical potential scale μi∗ can be chosen as
 244 Essentials of Micro- and Nanofluidics

equal in the two reservoirs, and so

XR
μ R − μ D = RT ln + z Fφ = 0 (7.55)
XD
where φ = φ R − φ D . Thus the mole fractions in the two regions are related by

X R = X D e−z Fφ/RT (7.56)

which is recognized again as the Boltzmann distribution. Note that X R < X x D if

φ > 0 and X R > X D if φ < 0.
Putting equation (7.56) in terms of the potential leads to the Nernst equation
(Hunter, 1981), which is given by
RT XR
φ = ln (7.57)
zF XD
The fact that at equilibrium, the electrochemical potentials are constant in the
donor and receiver regions is called Donnan equilibrium.

7.8 Acids, bases, and electrolytes

A solution is defined as a mixture of different components of different chemical

structure. Under normal conditions, sodium chloride dissociates into Na+ and Cl−
in an aqueous solution, and because of their charge, these ions are surrounded
by water molecules. This complex thus has a length scale that is larger than
its so-called ionic radius. Blood serum is similar in composition to PBS. In
electrochemistry, a buffer consists of a weak acid and a conjugate base so that
when added to the solution, the pH of the solution does not change; sodium
phosphate and potassium phosphate are the buffers in PBS, both containing a
hydrogen atom. Generally, the biofluids of interest here are dilute in the charged
species, and so the properties of such fluids, density and viscosity, for example,
can be taken to be those of water.

Milk of magnesia is an antacid, a base that acts to neutralize acids in the stomach. The active base is
magnesium hydroxide: Mg(OH)2 .
245 Elements of Electrochemistry and the Electrical Double Layer

An acid is a proton (H+ ) donor, and a base is a proton acceptor, according to

the Bronsted–Lowry terminology (Raymond, 2007). Acids and bases are linked
by the general proton transfer reaction

HA + B A− + BH+ (7.58)

and the extent of the dissociation is measured by an equilibrium constant

defined by
[A− ][BH+ ]
K = (7.59)
[HA]
The limit K → 0 means little dissociation has occurred and defines a weak acid
or base. The double arrow in equation (7.58) means that some of the reactant HA
remains, and the equilibrium constant is the ratio of the constants associated with
the forward and backward reactions: K = k f /kb . The concept of an equilibrium
constant described here is the same as is customarily used in the combustion
literature, another example of a type of chemical reaction (Moran & Shapiro,
2007).

Notation alert: The square brackets denote the activity (which is dimensionless)
of the given species, as is common in the chemistry literature. This notation
eliminates the need for cumbersome subscripts on the concentrations such as
[A− ] = aA− .
Acids and bases can be strong or weak, depending on the degree of ionization.
For example, a strong acid or base completely dissociates; sodium hydroxide
(NaOH), also known as lye, is a strong base such that

NaOH → Na+ + OH− (7.60)

In chemistry, the notation pC is defined as pC = − log10 C. In this context, a

strong base is defined as that having pOH = − log10 [OH− ] ∼ 0 and increasing
pH, while a strong acid is defined as pH = − log10 [H+ ] ∼ 0 and increasing pOH.
For example, ammonia (NH3 ) is a weak base in water because very little
dissociates, having pH ∼ 9. Its reaction with water is governed by the equation

NH3 + H2 O OH− + NH+

4 (7.61)

with the equilibrium constant defined by

[OH− ][NH+4]
K = ∼ 2 × 10−5 (7.62)
[NH3 ]
Water itself is also an acid and base; for the ammonia reaction, the water serves
as an acid because it donates a proton to the ammonia. The equilibrium constants
for many weak acids range from K a ∼ 10−1 to 10−5 and for weak bases from
K b ∼ 10−5 to 10−9 . These values can be found in most undergraduate analytical
chemistry textbooks and in Butler (1998).
Water itself can ionize; it does so according to the reaction

2H2 O H3 O+ + OH− (7.63)

 246 Essentials of Micro- and Nanofluidics

with
[OH− ][H3 O+ ]
K = (7.64)
[H2 O]2
In the chemistry literature, the liquid water reactant is often dropped, and the
equilibrium constant for this reaction is sometimes written
K w = [OH− ][H+ ] (7.65)
for a single H2 O molecule transfering its proton. The equilibrium constant for the
ionization of water at T = 25◦ C is K w = 1 × 10−14 . Because pure water must be
electrically neutral, [H+ ] = [OH− ] = 1 × 10−7 . Thus the pH of neutral water is
pH = 7 and is termed the neutral pH. The value of K in the ionization of water
decreases with decreasing temperature.
In the same way that we speak of strong and weak acids and bases, we speak
of strong and weak electrolytes. For example, as noted earlier, sodium chloride
dissociates in water according to
NaCl (solid) → Na+ (aq) + Cl− (aq) (7.66)
with the equilibrium constant defined by
[Cl− ][Na+ ]
K = (7.67)
[NaCl]
and K → ∞, indicating complete dissociation (no NaCl in the products) at nor-
mal temperatures as described in Section 7.6; thus sodium chloride is a strong
electrolyte and you sense the different approaches taken in the engineering com-
munity, from this, the chemistry presentation. The limit K → 0 defines a weak
electrolyte. Generally speaking, the electrolytes of interest in micro- and nanoflu-
idics are strong. Water itself is a weak electrolyte.

7.9 Site-binding models of the silica surface

In Chapter 3, the surface charge density was defined in terms of the normal
derivative of the electrical potential with the proportionality constant being the
electrical permittivity. It was revisited in Chapter 6 in the definition of the electric
field due to a charged surface.Thus the electrical potential was defined in terms of
the potential in the liquid adjacent to the solid wall. In this section, a complemen-
tary view of the surface charge density based on the properties of the solid silica
(i.e., silicon family) surface is presented; this will lead to a relationship between
the surface charge density and the ζ potential that incorporates the effect of pH.
Recall that the electrical double layer was introduced in Section 1.10.
Silicon (chemical symbol Si) is one of the most abundant elements on earth.
Silicon may bond with oxygen to form silicon dioxide or silica, which has chem-
ical symbol SiO2 , found commonly in the form of a crystalline mineral. Recently
silicon or silica nanoporous membranes have received attention as candidates for
247 Elements of Electrochemistry and the Electrical Double Layer

implantable devices, specifically in the design of artificial kidneys (Fissell, 2006;

Fissell & Humes, 2006) and for glucose sensors. Silicon may be made biocom-
patible by treating the surface with polyethylene glycol (PEG), which also can
reduce the chances of biofouling at the surfaces of the membrane.
The charge on the silica surface is regulated by the availability of potential
determining ions (PDI) (Hunter, 1981), that is, ions that allow the surface to
acquire charge through electrochemical reactions. For a silica surface in the pH
range of interest (pH = 5–9), the hydrogen ion H+ is considered the primary
PDI (Iler, 1979). The pH-dependent surface charge acquired by the surface in
turn affects the distribution of ions in the electrolytes that form the electrical
double layer near the surface (Elimelech, 1998). To understand the effect of pH
on the ζ potential, the predictions of the surface charge from a solid-side model
of adsorption on the silica surface will be combined with a liquid-side model of
the EDL, to be discussed later in this chapter.
At pH ∼ 4, protons desorb from the SiOH to form the SiO− on the surface
so that at pH > 4, the silica surface is negatively charged (Icenhower & Dove,
2000). For pH < 6, the SiOH complexes make up virtually all of the surface
sites on a quartz surface, while even at a pH = 7, they account for ∼98 percent
of the sites. Only at large pH ∼ 12 is there a significant fraction of the complex
SiO–Na+ , on the surface, the fraction being just over 30 percent.
The isoelectric point of the surface, or point of zero charge, occurs at pH = 2.4,
meaning the reaction is in equilibrium and the surface has no charge. If pH > 2.4,
the process of deprotonating SiOH will overwhelm the protonation of SiO, driving
the surface charge density strongly negative. If pH > 4, almost all SiOH molecules
are deprotonated so that the silicon surface is highly negatively charged and the
charge density can be treated as a constant. For pure water, pH = 7, and the pH
of most physiological fluids, such as blood serum and saline, is pH ∼ 7.4.
The family of surfaces made of silicon, Si, and its derivatives, among them
silica, SiOH, and silicon dioxide, SiO2 , can be crystalline (rigidly ordered) or
amorphous (disordered). SiO2 and SiOH are also known as two of many forms
of what is termed the quartz group. As noted in Section 7.1, when exposed to
water, for example, SiOH dissociates according to (Israelachvili, 1992)

SiOH (solid) SiO− (solid) + H+ (aqu) (7.68)

Thus the silica walls having SiO− ions (Figure 7.4) will attract the positive ions
that are in solution, a process termed adsorption, forming the EDL.
The equilibrium constant for the reaction is defined by
[SiO− ][H+ ]s
K1 = (7.69)
[SiOH]
which has the value K 1 = 10−6.8 . The subscript s denotes that the H+ adsorbs to
the surface from the liquid side. In dilute electrolyte solutions, the activity and
concentration are approximately equal. In both these equations, the square brack-
ets denote mole fraction, and the equilibrium constant K 1 is scaled by 1 mole/m3
 248 Essentials of Micro- and Nanofluidics

ζ H2O
SiOH

Na+
SiO– Na+

SiO–

Diffuse Layer
SiO–
H2O

Silica Stern
SiOH
Layer Cl–

SiO–

SiO–
H2O

SiO– Na+

SiOH2+ Cl–

Sketch showing the ions that make up the solid side of the silica surface, predominantly SiO− in the pH range
Figure 7.4
of interest. Adsorption of Cl− in the pH range of interest is rare and the surface consists of predominantly SiO−
in the pH range of interest.

so that, for example, K 1 = K1∗ /1 mole/m3 , so that K 1 is dimensionless. As noted

previously, this point is not often very clear in textbooks and research papers.
The number density of surface silanols (Ns ) in molecules/m2 is constant for a
particular silica surface in its pristine state, meaning that the surface is immersed
in a vacuum. The number of molecules on the surface is given by

Ns = ([SiO− ] + [SiOH])N A (7.70)

where N A is Avogadro’s number. The quantity Ns can be measured using a vari-

ety of experimental methods, for example, chromatography with tracer molecules
followed by detection with accurate weight gain measurements; infrared spec-
troscopy and nuclear magnetic resonance (Iler, 1979; Kirby & Hasselbrink,
2004a); and from experimental data, Ns = 8.3 × 10−6 mole/m2 N A , according
to Papirer (2000).
The surface charge density σ that must neutralize the excess charge in the
diffuse layer of the EDL is then defined by

σ = −[SiO− ]F (7.71)

where F is Faraday’s constant and the units of surface charge density are C/m2 .
As mentioned previously, the first term in equation (7.71) is usually much smaller
than the second term.
It is desireable to express equation (7.71) in terms of the equilibrium constants
because they are usually known from experiments. Using equations (7.70) and
 249 Elements of Electrochemistry and the Electrical Double Layer

1 0
ζ= −10 mV ζ= −10 mV
ζ= −20 mV ζ= −20 mV
0.8 −0.02
ζ= −40 mV ζ= −40 mV

σ (Coulomb/m2)
−σ NA /(F Ns)

0.6 −0.04

0.4 −0.06

0.2 −0.08

0 −0.1
4 5 6 7 8 9 10 4 6 8 10
pH pH
(a) (b)

Surface charge density for various values of the ζ potential using equation (7.72). (a) Dimensionless form.
Figure 7.5
(b) Dimensional form. Here N s = 5 × 1017 /m2 and K 1 = 10−6.8 .

(7.71) and the definitions of the equilibrium constants,

Ns F K1
σ =− (7.72)
N A K 1 + [H+ ]s
The activity or mole fraction of [H+ ]s at the surface is essentially unknown;
however, its value in the bulk can be calculated and related to the surface value
by the Boltzmann distribution, and if the potential at the Stern plane is taken to
be the ζ potential, then (see Section 7.12) [H+ ]s = b[H+ ]b f , where b = e−ζ F/RT
and b f stands for “bulk fluid.” Equation (7.72) is the solid-side definition of the
surface charge density.
The general picture of the relationship between surface charge density and ζ
potential is depicted in Figure 7.5. Note that the dimensionless surface charge
density σ N A /F Ns → 1 as pH → 9–10, and for these values of K 1 and Ns ,
σ → −0.08 C/m2 as p H → 9–10.
From this analysis, it is apparent that the surface charge density σ =
σ (ζ, p H, K i ) and that the chosen value of Ns is a parameter. As is seen here,
the precise form of the expression for the surface charge density (here equation
(7.72)) depends on the specific model of the EDL. These EDL models are sum-
marized by Venema et al. (1996), including the Nernstian–Stern layer model,
which requires no surface binding sites. They specifically considered cadmium
ion adsorption and found that none of the five models they tested, including the
triple layer model, could explain all the data. In that sense, this area is still fertile
ground for research.

7.10 Polymer surfaces

The term polymer is used to describe large molecules that are formed by the chem-
ical union of five or more molecules of a single compound, called a monomer. A
 250 Essentials of Micro- and Nanofluidics

combination of two monomers is called a

dimer, whereas the combination of three
and four monomers are called trimers
and tetramers, respectively. Proteins and
nucleic acids containing carbon are organic
biopolymers and will be discussed in detail
in the next chapter. The reaction mechanism
for the formation of the polymer as a chain
of monomers is called polymerization.
Image of PDMS contributed by Professor Derek
Figure 7.6 Two of the more common polymers that
Hansford.
are often used as materials for nanopore
membranes are polydimethylsiloxane
(PDMS) (Figure 7.6) and polymethyl methacrylate (PMMA). Both materials are
of the plastics family and are used as substitutes for glass and the silica family
of surfaces. PDMS is used in contact lenses and for caulking, to name just
two applications. Both PDMS and PMMA are softer than the silicas, and both
are easily scratched. PMMA is sold under the trade name Plexiglas (Kirby &
Hasselbrink, 2004b).
The organic polymers PDMS and PMMA have lower surface charge density
and are more hydrophobic than other surfaces. The result of lower surface charge
density is a smaller electro-osmotic flow rate; conversely, the hydrophobic nature
of these two polymer surfaces may lead to slip, resulting in a higher electro-
osmotic velocity. The ζ potential of organic polymers is more sensitive to buffer
additives with hydrophobic groups in comparison to silica because of their higher
surface hydrophobicity (Kirby & Hasselbrink, 2004C). The properties of the
substrates summarized in Table 7.1 are obtained from the review by Kirby and
Hasselbrink (2004c), unless otherwise mentioned.
Data for PDMS from a number of different sources suggest that there is a
linear dependence of the ζ potential on pH over a wide range of mole fractions
of the bulk mixture; Kirby and Hasselbrink (2004c) suggest from the data that
ζ ∼
= −(2 + 7(pH − 3)) log10 C (7.73)
where C is a reference counterion concentration.

Table 7.1. Properties of silica, PDMS, and PMMA surfaces relevant to microfluidic applications, based
on Kirby and Hasselbrink (2004c)
Repeating Dissociable Surface Hydrophilic or
Substrate unit surface group charge − d(dζ
pC )
(mV) hydrophobic

Silica Silicon dioxide silanol high 20 hydrophilic

PDMS Dimethylsiloxane silanol low 30 hydrophobic
PMMA Methylmethacrylate carboxyl low 10 hydrophobic

The fifth column tabulates the approximate reduction in the magnitude of zeta potential per 10-fold increase in the bulk solution concentration
C of a 1:1 symmetric salt at near-neutral pH. Here pC= − log10 (C). The PDMS surface has fewer surface silanol and therefore typically
lower surface charge density than silica surfaces under the same conditions (Kirby & Hasselbrink, 2004C).
 251 Elements of Electrochemistry and the Electrical Double Layer

Experimental results for PMMA are less clear, and there is a wide range of
scatter in the ζ potential as a function of pH. In this case, only the dependence on
the concentration is clear, being ζ = a0 − a1 log10 C for univalent electrolytes.
The values for the constants a0 = −4.06 mV and a1 = −12.57 mV are valid over
the range 7 < pH < 8.

7.11 Qualitative description of the electrical double layer

A charged surface in contact with an aqueous electrolyte is shown in Figure 7.9.

If a surface is negatively charged, the positively charged ions in the vicinity
of the surface tend to collect near the charged surface, a process often called
charge polarization, a phenomenon that occurs in several contexts (Chang and
Yeo, 2010). This process forms a thin layer of positively charged ions near the
surface, so-called counterions. The thickness of this layer of “pinned” ions near
the surface is of the order of the ionic diameter of hydrated counterions, and the
edge of this layer is called the Stern layer, and the edge of this layer is the Stern
plane. The electrical potential at the Stern plane is called the ζ potential.
The Stern plane is the plane of closest approach of an ion, and the precise
position is dictated by the size of the ions at the surface. In the classical theory
of the EDL, the ions outside the Stern layer are distributed according to the
equilibrium Boltzman distribution. The layer outside the Stern layer is called the
diffuse layer. The total system, including the Stern layer and the diffuse layer, is
called the EDL. In the continuum model of the EDL, the ions in the diffuse layer
are most often modeled as point charges.
Because the electrical double layer is so thin, being on the order of nanome-
ters to hundreds of nanometers at the most, its structure has not been probed
experimentally. However, the analogy between the velocity and the potential in
electro-osmotic flow discussed in Chapter 3 means that velocity measurements
within the EDL can be used to extract electrical potential values (Sadr et al.,
2006). This point has already been discussed in Sections 3.13 and 3.17. The con-
centration of the cations and anions near a charged surface has traditionally been
obtained by analysis (e.g., molecular dynamics simulations) and/or qualitative
and descriptive models of the EDL near a charged surface.
The ion distribution within the EDL can be described by using the number den-
sity, concentration, or mole fraction. Engineers usually prefer the dimensionless
mole fraction, whereas chemists usually use concentration or number density. In
this section, mole fraction, will be used, and for a binary mixture, these terms
will be denoted by g for the cation and f for the anion.
The simplest model for the EDL was originally given by Helmholtz (1897)
long ago, and he assumed that electrical neutrality was achieved in a layer of fixed
length. For a negatively charged surface, he assumed that the distributions of the
anions and the cations are linear with distance from the surface, as depicted in
Figure 7.7.
 252 Essentials of Micro- and Nanofluidics

Almost 30 years later, the Debye–Hückel

picture of the EDL (Debye & Hückel, 1923)
+ wall − wall
φ g +
was postulated, in which the influences of
f=g=goo the ionic species are equal and opposite
0
y y but nonlinear functions of distance from
f − the surface so that the surface mole frac-
− wall
tions of the coion (anion for a negatively
charged surface), the magnitude of the sur-
Figure 7.7
Potential (left) and mole fractions near a neg- face mole fractions f 0 , and counterion g 0
atively charged surface according to the ideas are symmetric about their asymptotic val-
of Helmholtz (1897). Here g denotes the cation
mole fraction and f denotes the anion mole
ues far from the surface; see Figure 7.8a.
fraction. The Gouy–Chapman (Gouy, 1910;
Chapman, 1913) model of the EDL allows
for more counterions to bind to the surface
charges than coions and is one example of charge polarization. In ion-exchange
membranes, the concentration of the counterion may be driven to zero (see Section
7.18). This means that for a negatively charged surface, g 0 can be much larger
than its asymptotic value, whereas f 0 is not much lower than the asymptotic
value in the core. This situation is depicted in Figure 7.8b. Whether the Debye–
Hückel picture or the Gouy–Chapman model of the EDL is relevant depends on
the surface charge density, with the Debye–Hückel picture occuring at low sur-
face charge densities and the Gouy–Chapman model occuring for higher surface
charge densities.
The double layer model described can be further refined by delineating the
intrinsic surface charges SiOH from the adsorbed ions, say, Na+ and Cl− , leading
to what is now termed the triple layer model (James, 1981; Persello, 2000). The
structure of the triple layer model is depicted in Figure 7.9. On a silica surface,
the protons and the hydroxide ions from the surface charge layer compose the
solid-side, innermost layer, with the adsorbed electrolytes pinned in a second

(a) (b)

(a) Debye–Hückel (Debye & Hückel, 1923) picture of the EDL. Here g denotes the cation mole fraction and f
Figure 7.8
denotes the anion mole fraction. This model occurs for low surface charge densities. (b) Gouy–Chapman model
(Gouy, 1910; Chapman, 1913) of the EDL valid at higher surface charge densities.
 253 Elements of Electrochemistry and the Electrical Double Layer

layer adjacent to the surface layer, called

the inner Helmholtz plane (IHP). In this
model, the H+ and OH− ions react at the
surface of the solid, while the center of the
electrolyte ions resides approximately one
hydrated ion radius from the surface plane.
The remaining electrolyte ions constitute
the diffuse layer.
In the triple layer model, there are sur-
face charge densities associated with the
reaction with a proton or hydroxyl group
Figure 7.9
The triple layer model of the EDL differenti- (the OH group) and associated with the
ates layers at the surface according to hydroxyl reactions involving adsorbed ions. This
group surface charges and charge generated by
forms what is normally called the surface
adsorbed ions. The ζ potential takes its value
at the outer Helmholtz plane (OHP). charge density in the classical EDL picture.
Despite the dichotomy between the surface
and adsorbed layers, the triple layer model leads to a surface charge density
equivalent to that of the double layer model.
Stern (1924) recognized that there are a number of other assumptions embed-
ded in these qualitative and simple models. In the models discussed so far, the
ions have been assumed to be point charges, and the solvent is not modified by
the presence of the charges. He proposed that the finite size of the ions affects the
value of the potential at the surface, and this is called the excluded volume effect.

7.12 Electrolyte and potential distribution in the electrical double layer

In the previous two sections, the electrochemistry of the solid side of a charged
surface, such as silica, was described. In this section, we discuss the nature of the
liquid side of such a surface.
Consider a stagnant electrolyte solution bounded by a single charged flat plate,
as in Figure 1.18. As we know, at equilibrium, the electrochemical potential is
constant. Thus, if y is the distance measured from the wall, assuming that the
wall is long in both the x and z directions, the electrochemical potential of each
species is constant in the y direction, and

dμi
=0 (7.74)
dy

In the general case, where the electrochemical potential may depend on all three
spatial variables, ∇μi = 0. Thus, in one dimension,

1 dci z i F dφ
=− (7.75)
ci dy RT dy
254 Essentials of Micro- and Nanofluidics

and the concentration of each species is obtained by integrating;

−z i Fφ
ci = ci0 e RT (7.76)

the Boltzmann distribution where here ci0 is the mole fraction far from the wall,
where the potential is defined to be zero. The mole fraction is defined as the
concentration of a given species divided by the total concentration, including the
solvent.
Conversely, we could define
−z i F(φ−ζ )
ci = ci0 e RT (7.77)

in which case, ci0 is the concentration of species i at the wall, where φ = ζ .

Because the concentration is given by ci = n i /N A , the Boltzman distribution
could also be expressed as a number density:
−z i Fφ
n i = n i0 e RT (7.78)

Consider the case of N species of valence z i . Note that counterions will collect
on the wall in greater numbers than coions, which are repelled by the wall. Then
the equation for the potential becomes

e  0 −zi Fφ
N
d 2φ ρe
∇ 2φ = = − = − c z i e RT (7.79)
dy 2 e e i=1 i

where ρe is the volume charge density. It is clear from the argument of the
exponential in the volume charge density that the potential scale for nondi-
mensionalization should be taken as φ0 = RT /F so that the potential scale is
φ0 = 26 mV.
For N = 2, and for a pair of monovalent ions of equal but opposite valence,
with φ scaled on φ0 ,
d 2φ 1
∇ 2φ = = 2 sinh φ (7.80)
dy 2 κ
where
2F 2 c0
κ2 = (7.81)
e RT
has units of m−2 and so 1/κ 2 = λ2 has units of m2 and λ is the Debye length.
Here c0 is the concentration of the two electrolytes in the bulk, where their
concentrations are equal. Defining Y = κ y = y/λ, in dimensionless form,
d 2φ
= sinh φ (7.82)
dY 2
This equation is recognized as dimensionless because of the absence of all of the
dimensional parameters present in equation (7.80). For a pair of multivalent ions
of valence z and −z, the coefficient is just multiplied by the absolute value of the
255 Elements of Electrochemistry and the Electrical Double Layer

valence, and in dimensionless form,

d 2φ
= z sinh zφ (7.83)
dY 2
If the boundary condition at the surface is assumed to be given by
φ = φs at Y = 0 (7.84)
and at large distances, φ → 0 as Y → ∞, an analytical solution may be found.
The quantity φs can be identified as the dimensionless ζ potential.
Consider first a pair of monovalent ions. To find the solution, we can integrate
once after multiplying by 2 (dφ/dY ) to obtain

dφ (cosh φ − 1)
= −2 (7.85)
dY 2
Now, using the identity

cosh φ − 1 φ
= sinh (7.86)
2 2
it is possible to integrate again, noting that
 
dφ |φ|
= ln tanh (7.87)
sinh φ2 4
and so
tanh φ4
= e−Y (7.88)
tanh φ4s
The solution is thus
 
−1 φs −Y
φ = 4 tanh tanh e (7.89)
4
For an arbitrary value of the valence, we have
 
4 −1 zφs −zY
φ = tanh tanh e (7.90)
z 4
Since
1 1 + tanh x
tanh−1 x = (7.91)
2 1 − tanh x
this solution can also be expressed as
 
2 1 + e−zY tanh( zφ4 s )
φ = ln (7.92)
z 1 − e−zY tanh( zφ4 s )
The combination of equations (7.76) or (7.78) and (7.92) is called the Poisson–
Boltzmann solution for the classical EDL. Here φs is identified as the dimension-
less ζ –potential.
The solution for the potential is depicted in Figure 7.10; the Debye–Hückel
approximation valid for small potentials is also shown. A negative value of the
 256 Essentials of Micro- and Nanofluidics

0
4
Analytical
Analytical
Debye−Huckel
Debye−Huckel −0.5
3.5
φ =4.0
s
−1 φ = −0.25
3 s

−1.5
2.5
φ =2.0 φ = −1.0

Potential
Potential

s s

−2
2
φ =1.0 φ = −2.0
s s
−2.5
1.5

φ =0.25 −3 φ = −4.0
s s
1

−3.5
0.5

−4
0 0 1 2 3 4 5 6
0 1 2 3 4 5 6
Y Y
(a) φs > 0 (b) φs < 0

Dimensionless potential for several different values of the surface potential for a 1:1 electrolyte (z1 = 1,
Figure 7.10
z2 = −1) at a bulk ionic strength of I = 0.3 M = 0.15 Na + 0.15 Cl.

surface potential implies that the wall is negatively charged, whereas a positive
value implies a positively charged wall. Note that the solutions are mirror images
of each other. The EDL thickness is indicative of the region near the wall over
which there is a surplus of cations over anions for a negatively charged wall or
vice versa for a positively charged wall.
For small dimensionless potentials, equation (7.83) may be linearized because
sinh φ ∼ φ + · · · as φ → 0, and thus for z = 1

d 2φ
= sinh φ = φ + · · · (7.93)
dY 2
The solution to this equation is simply

φ = φs e−Y (7.94)

and the linearization process is called the Debye–Hückel approximation; this

solution is also plotted in Figure 7.10. The thickness of the EDL for the DH
approximation can be determined analytically, and
δe
0.01 = e− λ (7.95)

or
1
δe = λ ln (7.96)
0.01
or δe = 4.6λ, a formula that is similar to the definition of the boundary layer
thickness in fluid mechanics.
In the general case where a mixture contains N ions, the Debye length is given
N 0 2
by λ2 = 1/κ 2 = e RT /F 2 I , where I = i=1 ci z i . For a pair of monovalent
dissociated ions such as NaCl, I = c1 + c2 = 2c0 , where c0 is the concentration
of each species in the region where the mixture is electrically neutral.
 257 Elements of Electrochemistry and the Electrical Double Layer

−3 −3
x 10 x 10
3.5 8
Analytical Na Analytical Na
Analytical Cl 7 Analytical Cl
Debye−Huckel Na Debye−Huckel Na
Debye−Huckel Cl Debye−Huckel Cl
6

3 5
Mole Fraction

Mole Fraction
4

2.5 φs= −0.25 2

φs= −1.0

2 −1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Y Y
(a) φs = −0.25 (b) φ s =−1

Mole fractions for two different values of the surface potential for a 1:1 electrolyte (z1 = 1, z2 = −1) for a
Figure 7.11
negatively charged wall at a bulk ionic strength of I = 0.3 M.

Notation alert: This is just a reminder that chemists define the ionic strength as
N 0 2
I = 1/2 i=1 ci z i so that there will be a 2 in the definition of the Debye length.
This definition is not appropriate for mixtures containing multivalent ions.

As a numerical example, consider an aqueous electrolyte solution with

0.015 M Na+ and 0.015 M Cl− . Then, at T = 300 K,

! 8.315 moleJ K × 300 K × 78.54 × 36π
1
× 10−9 Nm
C2
λ=
2

30 m3 × (96500 moles )
moles C 2

= 2.5 × 10−9 m = 2.5 nm (7.97)

For an ionic strength of 0.3 M, λ = 0.8 nm.
Note in Figure 7.10, the surprising result that the DH approximation gives a
good representation of the full nonlinear solution for values of the potential much
higher than expected. It is expected that the approximation should break down
for potentials φs ∼ 0.2, but this does not occur. However, it is seen that the DH
approximation is much worse for the mole fractions, for which
X i = X i0 e−zi φ ∼ X i0 (1 − z i φ + · · · ) (7.98)
as depicted in Figure 7.11. Adding more terms to the Taylor series approximation
for the exponential only marginally improves the situation.
Let us now consider the EDL on channel walls, as depicted in Figure 7.12.
Now there are two length scales λ and the height of the channel h. Moreover,
if the Debye length is large enough, or the channel height is small enough, the
potential may not be zero over most of the core of the channel. In anticipation of
this possibility, if we nondimensionalize the y coordinate on h, then
d 2φ
2 = z sinh zφ (7.99)
dY 2
 258 Essentials of Micro- and Nanofluidics

− − − − − − − − − − − − − − − − − − − − −
+ + + + + + + + + + + + + + + + + + ++ +
− + + + +
+ + − + − + ++ − + −
+−

+ +
+ + −
+ − +
−
+ + +
+ − − + −
+−
+ + − + −
+ + + ++ − + −
− + +
+ + + + + + + + + + + + + + + + + + ++ +
− − − − − − − − − − − − − − − − − − − − −

h ~ λ

The classical EDL on two plates.

Figure 7.12

where φ is the dimensionless potential and Y = y/ h and  = λ/ h. The boundary

conditions now become

φ = φs at Y = 0, 1 (7.100)

If both of the EDLs are thin, then the solution for the potential is as given earlier,
with Y = y/λ = y/h near y = 0 and Y = h − y/λ = h − y/h near y = h.
This situation is depicted in Figure 7.13 and is the straightforward extension to
the case of a single plate.
If  = O(1), the double layers are said to be overlapping, and the integration
procedure for the overlapped case cannot be performed as before because the
boundary condition must be applied at the walls Y = 0, 1. The integration
Y dφ √
√ = ± 2Y (7.101)
0 cosh zφ − cosh zφs
results in a generalized hypergeometric series, which must be evaluated numeri-
cally. It is thus easier to numerically calculate the result directly, and these results
are shown in Figure 7.13 for several values of the surface potential, the results

0 0

−0.5 φ =−0.25
−0.5 s
φ =−1
−1 φ =−0.25 −1 s
s
φ =−1 −1.5
−1.5 s
Potential

φ =−2
Potential

φ =−2 s
−2 s −2
φs=−4 −2.5 φ =−4
−2.5 s

−3 −3

−3.5 −3.5
DH solution DH solution
−4 Analytical solution −4 Analytical solution
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
y y

(a) h = 10nm (b) h = 30nm

Dimensionless potential for several different values of the surface potential for a 1:1 electrolyte (z1 = 1,
Figure 7.13
z2 = −1) for a positively charged wall in a channel of height h = 10, 30 nm at a bulk ionic strength of 0.1 M
(0.05 + 0.05 M).
 259 Elements of Electrochemistry and the Electrical Double Layer

−3
−3
x 10 x 10
7 2.5
Analytical Na Analytical Na
6 Analytical Cl Analytical Cl
Debye−Huckel Na 2 Debye−Huckel Na
5 Debye−Huckel Cl Debye−Huckel Cl

Mole Fraction
Mole Fraction

4 1.5

3
1 φs=−1
2
φ =−2
s
1 0.5

0
0
−1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
y y
(a) h = 10 nm (b) h =30nm

Mole fractions for two different values of the surface potential for a 1:1 electrolyte (z1 = 1, z2 = −1) for a
Figure 7.14
positively charged wall in a channel of height h = 10, 30 nm at a bulk ionic strength of 0.1 M.

for the mole fractions are depicted in Figure 7.14. In the Debye–Hückel limit, the
solution in the overlapping case is given by
y−1/2
cosh 
φ = φs (7.102)
cosh 21
It is left as an exercise to calculate the solution for the potential when the surface
charge density rather than the ζ potential is known.

7.13 Multivalent asymmetric mixtures

In the previous section, a binary mixture of monovalent ions was considered

exclusively. Grahame (1953) was apparently the first to investigate the case of
asymmetric multivalent mixtures. Using the methods described in the previous
section results in the generalized dimensionless expression after one integration
of the Poisson equation (Y = y/λ) is
   N 
dφ 2  N
−zi φ
= 0
X i zi e − 0
Xi (7.103)
dY i=1 i=1

where X i0 are the dimensionless concentrations scaled on the ionic strength in the
bulk; the potential scale is RT /F, as before.
Unfortunately, the integration can be carried out only for specific values of the
valences. Grahame (1953) considered explicitly the case of a 2:1 (z 1 = 2; z 2 =
−1) electrolyte and a 1:2 electrolyte. In his notation, the Debye length is based
on the bulk concentration of the less abundant species, and for a 2:1 electrolyte,
this results in the dimensionless equation
 
dφ 2
= e−2φ + eφ − 3 (7.104)
dY
 260 Essentials of Micro- and Nanofluidics

Using the identity

(e−φ − 1)(1 + 2eφ )1/2 = (e−2φ + eφ − 3)1/2 (7.105)

equation (7.104) can be integrated again to obtain (φs = ζ )

Y φ dφ
dY = √ (7.106)
0 ζ (e−φ − 1) 1 + 2eφ

This integral can be evaluated by making the transformation ξ = (1 + 2eφ )1/2 or

from integral tables for φ < 0 to obtain

2 1 + 2eφ 1 + 2eζ
Y = √ tanh−1 − tanh−1 (7.107)
3 3 3

which is an equation that gives the potential φ as an implicit function of Y . Here

tanh−1 is the inverse hyperbolic tangent function.
For φ > 0, the result is given by
 √ √ 
1 (1 + 2eφ )1/2 + 3 (1 + 2eζ )1/2 + 3
Y =√ ln √ − ln √ (7.108)
3 (1 + 2eφ )1/2 − 3 (1 + 2eζ )1/2 − 3

It is left to the exercises to evaluate these two equations and plot the potential as
a function of Y .

7.14 The ζ potential and surface charge density: Putting it all together

Having determined the electrical potential and the concentrations on the liquid
side of the surface, we can now develop a relationship that relates the surface
charge density, the ζ potential, and pH.

7.14.1 The classical liquid-side view for a symmetric electrolyte

Recall from Figure 1.18 that the ζ potential is the electrical potential at the Stern
plane and is directly related to the surface charge density. An explicit expression
for surface charge density as a function of ζ potential can be obtained for a
symmetric binary z:z electrolyte using the liquid-side analysis. Then, using the
previous site-binding analysis on the solid side, we can tie it all together.
The derivative of the potential normal to the wall is given by

 
2kb T  0 −zk i Teφ
N
dφ
= ∓! ni e b − 1 (7.109)
dy e i=1

where we choose the negative sign if the potential is positive on the wall and the
positive sign if the potential is negative at the wall. For a symmetrical pair of
261 Elements of Electrochemistry and the Electrical Double Layer

ions, the number density of the ions n 01 = n 02 = n 0 in the bulk, and

         
dφ 4kb T n 0 zeφ 4RT c0 z Fφ
=∓ cosh −1 =∓ cosh −1
dy e kb T Fe RT
(7.110)
Using the identity given by equation (7.86), for a pair of symmetric electrolytes,
  
dφ 8kb T n 0 zeφ
=∓ sinh (7.111)
dy e 2kb T
Specifically, at the wall,
   
dφ  8kb T n 0 zeζ
=∓ sinh (7.112)
dy 0 e 2kb T
where it is noted that here ζ is dimensional.
The total charge of the cations and anions in the gap must be offset by the total
surface charge so that the system is electrically neutral. Consider the case of a
thin double layer  = λ/ h  1 outside of which is a neutral aqueous solution.
Then, if σ0 denotes the surface charge density on the liquid side,
∞ ∞ d 2φ
σ0 + ρe dy = σ0 − e dy = 0 (7.113)
0 0 dy 2
Substituting from earlier,
 
dφ  zeζ
σ0 = −e  = 8e kb T n 0 sinh (7.114)
dy 0 2kb T
and solving for the ζ potential,
 
RT σ0 Fλ
ζ = sinh−1 (7.115)
zF e RT

This is a fundamental relationship between the surface charge density, the

Debye length, λ, and the ζ potential. In the Debye–Hückel limit of small potential,

 
dφ  zeζ
σ0 = e  = 8e kb T n 0 (7.116)
dy 0 2kb T
so that
F
σ0 = ze κζ (7.117)
RT
and this equation is called Grahame’s equation. Note that the quantity
σ0 F
= zζ (7.118)
e κ RT
is dimensionless.
262 Essentials of Micro- and Nanofluidics

For a finite channel with overlapped double layers having the same surface
charge,
1 d 2φ
2σ0 − e dy = 0 (7.119)
0 dy 2
so that
F
σ0 = 2ze κζ tanh κh (7.120)
RT
in the Debye–Hückel limit.

7.14.2 The solid-side view and connection to the liquid side

The preceding simple formula for the relationship between the surface charge
density and ζ potential valid in general for a z:z electrolyte cannot distinguish
the effect of pH since pH defines the adsorption of H+ ions on the surface, as
discussed in Section 7.9. For the case of a simple mixture such as Na+ –Cl− –
water, a straightforward extension of equation (7.72) for the negatively charged
surface leads to
 
Ns F K1
σ0 = −  (7.121)
NA K 1 + b K N a [Na+ ] + [H+ ]

where K 1 ∼ 10−7 is the equilibrium constant of the silanol reaction given by

equation (7.68), K N a is the equilibrium constant of surface adsorption of sodium
ions, and Ns is the total number density of surface silanols. The adsorption of
chloride ions, the coions, has been neglected. Here b is the Boltzmann factor
b = e−ζ F/RT , defined previously. In equation (7.121), it is assumed that the silica
surface has sufficient negative charge (pH  3) so that protonation of neutral
silanol groups and adsorption of chloride ions can be neglected. Note that for
ζ < 0, b > 0 and in equation (7.121), the square brackets denote bulk or reservoir
mole fractions.
To connect the solid side with the liquid side, squaring equation (7.109) leads
to

N
σ02
n i0 (b zi − 1) − =0 (7.122)
i=1
2e RT

To obtain the ζ potential as a function of the concentration of the various species

in the system, including hydrogen ions, which reveal the pH dependence of the
ζ potential, we equate the liquid- and solid-side expressions for the surface charge
density, σ = σ0 , which results in a nonlinear algebraic equation in the Boltzman
factor, b requiring numerical solution.
Consider the situation in which all electrolytes (neutral molecules, such as
NaCl, KNO3 , HCl, and H2 O, capable of dissociation) used to formulate the
solution in the reservoir are symmetric and dissociate into an equal number of z
valent cations and z valent anions, where z is taken as a positive integer. In this
situation, the total number of species N can only be an even number, and there
263 Elements of Electrochemistry and the Electrical Double Layer

are only N /2 distinct n i0 if the ions are modeled as point charges. In this case, the
summation in equation (7.122) for a binary electrolyte is given by

n 0 (b z + b−z − 2) = n 0 (b z/2 − b−z/2 )2 (7.123)

Substituting into equation (7.109) and using the definition of Debye length,
e RT z/2
σ0 = − (b − b−z/2 ) (7.124)
z Fλ
where the minus sign in front of the expression on the left emphasizes that the
ζ potential and surface charge have the same sign. This equation is an alternate
form of equation (7.114), only it is more general in the sense that it is valid for
multicomponent mixtures containing only +z and −z valent species.
In the limit of large ζ potential such that b  1, equation (7.124) becomes
ze RT z/2
σ0 = − b (7.125)
λ F
It is uncertain whether the adsorption of sodium plays a significant role in
the current context when compared with H+ adsorption (Kirby & Hasselbrink,
2004a). Thus the solid-side equation (7.121) is rewritten, neglecting sodium ion
adsorption to the silica surface (setting K N a = 0) as
Ns F K 1
σ0 = − (7.126)
N A (K 1 + b[H+ ])
Recall that K 1 is the equilibrium constant for the SiOH dissociation and is
estimated to be K 1 = 10−6.8 . In most cases, the second term in the denominator
of equation (7.126) is dominant under conditions when −ζ F/RT  1 and [H+ ]
is not too small.
When −ζ F/RT  1, we can combine equation (7.125), calculating σ0 from
the liquid side, with equation (7.126), calculating the same quantity from the
solid side, to obtain
 
3zζ F A
exp − = + (7.127)
2RT [H ]
where the dimensionless number
z F 2 Ns F K 1 λ
A= >0 (7.128)
N A e RT
Taking natural logarithms of both sides of equation (7.127) and rearranging, it
is found that
ζF 2
− = (ln A − ln[H+ ]) (7.129)
RT 3
In the chemistry literature, it is common to convert to logarithms to the base
10 using log 10 = 2.303, and utilizing the definition of pH, the final result for the
ζ potential is (Revil et al., 1999)
ζF
= 1.535(log10 A + pH) (7.130)
RT
where A = A(pH) through the dependence of A on the Debye length.
264 Essentials of Micro- and Nanofluidics

Equation (7.130) suggests that the zeta potential has an approximately linear
dependence with pH at large ζ potential, as discussed elsewhere in the literature
(Kirby & Hasselbrink, 2004a). This equation predicts a change in zeta potential
of about 1.535RT /F = 40 mV per pH unit. In practice, the change in ζ potential
per pH unit depends on the type of surface. Experimental data on porous silicon
substrates suggest an approximate slope of about 30 mV (Thust et al., 1996). For
small ζ potential, b → 1, and combining the liquid and solid sides, the result is
 
Ns λ 10−pH
ζ = 1+ (7.131)
e N A K1

In the general case of symmetric electrolytes discussed earlier, the equation

for the ζ potential becomes
 
e RT z/2 −z/2 Ns F K1
(b − b )=  (7.132)
z Fλ NA K 1 + b K Na [Na+ ] + [H+ ]

which is a nonlinear equation that may be solved using a numerical zero-finding

technique. Obviously, this procedure may be extended to more complex mixtures
of asymmetric electrolytes, as discussed previously.
Consider the calculation of the effect of solution pH = − log10 [H+ ] on the
zeta potential in an aqueous solution of the salt NaCl. The pH of the solution
can be varied by adding small quantities of either the acid hydrogen chloride,
HCl, to obtain pH < 7 or the base sodium hydroxide, NaOH, for pH > 7. The
relative amounts of NaCl and the acid or base for each desired pH value are
so chosen to keep the solution ionic strength I of the solution constant. The
required concentration of NaCl can be calculated for a given pH from the desired
ionic strength I using I = [Na+ ] + [H+ ] = [NaCl] + 10−pH for pH < 7 and
I = [Cl− ] + [OH− ] = [NaCl] + 10−pH+14 for pH > 7. It may be noted, however,
that for the practical pH range of 5 < pH < 10 and NaCl concentrations of 1 mM
and above, [OH− ]  [Cl− ] and/or [H+ ]  [Na+ ]; therefore the ionic strength is
dominated by the sodium chloride, or I  [NaCl].
Because the ionic strength is held constant, the Debye length λ is a constant as
the pH is varied. Noting that z = 1 for the sodium chloride mixture, multiplying
both sides of equation (7.132) by x = b1/2 and rearranging, the following fourth
order polynomial equation in x results:

B(K Na [Na+ ] + [H+ ])x 4 + B(K 1 − K Na [Na+ ] − [H+ ])x 2 − K 1 x − B K 1 = 0

(7.133)

where B =  RT N A /Ns F 2 and is independent of the changing pH of the solution.

Equation (7.133) can be solved for x through analytical and/or numerical root
finding techniques, for example, through the use of the function roots in Matlab.
Then the zeta potential is calculated from ζ = −2RT /F(ln x).
Figure 7.15 shows the effect of pH on the zeta potential for a solution with
I = 0.2 mM (0.1 MNa, 0.1 MCl), K 1 = 10−6.5 , K Na = 10−3.5 , Ns = 4.8 nm−2 ,
 265 Elements of Electrochemistry and the Electrical Double Layer

−40

−50

−60
pK1 = 7.5
ζ (mV)
−70

−80
pK1 = 7
−90

−100 pK1 = 6.5

−110
5 6 7 8 9 10
pH

The pH dependence of ζ potential calculated using the site-binding model (equation (7.133)) for a sodium
Figure 7.15
chloride solution in contact with a fused silica surface. The parameters used are I = 0.2 mM (0.1 MNa,
0.1 MCl), K N a = 10−3.5 , N s = 4.8 nm−2 .

considered to represent a fused silica surface (Revil et al., 1999; Papirer, 2000).
For low (acidic) to near-neutral pH values, the magnitude of the ζ potential
increases almost linearly with pH because, if the number of available H+ ions is
small, the tendency of the charged SiO− groups to recombine with H+ to form
neutral SiOH groups is reduced. However, a plateau in the ζ versus pH curve is
eventually reached, as can be verified by observing that the right-hand side of
equation (7.132) becomes a constant for small values of [H+ ]. A similar behavior
has also been observed in experiments (Kirby & Hasselbrink, 2004a).

7.15 The electrical double layer on a cylinder

The EDL on a cylinder of radius a is also an important problem because tubular

membranes are used in filtration processes, such as water purification and desali-
nation, and other applications. Analogous to the plane wall case, the dimensionless
governing equation for the potential for a z:z electrolyte is given by
 
∂ 2φ 1 ∂φ
2 + = z sinh zφ (7.134)
∂r 2 r ∂r

where  = λ/a and λ is the Debye length.

In general, in cylindrical coordinates, integration of equation (7.134) cannot
be performed analytically in the general nonlinear case. However, under the
Debye–Hückel approximation, an analytical solution may be found. Linearizing
this equation for z = 1 for a pair of symmetric electrolytes yields
 
d 2φ 1 dφ
 2
2
+ =φ (7.135)
dr r dr
 266 Essentials of Micro- and Nanofluidics

This is a Bessel equation of order zero, and the solution for the potential in the
EDL on the inside of the cylinder is given by

I0 r
φ = ζ 1 (7.136)
I0 
Note that this solution is valid for any value of  = λ/a where a is the radius of
the cylinder. For small , the solution reduces to the planar case of a decaying
exponential.

7.16 The electrical double layer on a sphere

A spherical microparticle flowing in a channel is a common process in micro and

nanofluidics. Thus it is often necessary to solve for the electrical potential in the
EDL around the sphere. In this case, the governing equation for the potential is,
in dimensionless form,
 
2 1 d 2 dφ
 2 r = z sinh zφ (7.137)
r dr dr
as in the planar case; the boundary conditions are φ = ζ at r = a and φ → 0 as
r → ∞.
As with the cylindrical geometry, the full nonlinear problem cannot be inte-
grated analytically. Under the Debeye–Hückel approximation for small potential,
we have the reduced equation
 
2 1 d 2 dφ
 2 r = z2φ (7.138)
r dr dr
and the solution is given by
e−z 
r r
ez 
φ(r ) = A +B (7.139)
r r
The constants A and B must be found from the boundary conditions because
the potential must remain bounded at r = ∞, B = 0. Applying the boundary
condition on the sphere, we have
1 z(1−r )/
φ(r ) = ζe (7.140)
r
The condition of charge neutrality is similar to that for the flat plate, and so
the total charge in coulombs on the surface is given by
 
∞ ∞ d 2 ∗
∗ ∗ ∗2 d φ
 = −4πe r 2ρe dr = 4πe r dr ∗ (7.141)
a a dr ∗ dr ∗

Performing the integration,

 
1 1
 = 4πe a ζ + 2
(7.142)
λ a
 267 Elements of Electrochemistry and the Electrical Double Layer

The surface charge density is thus given by

 
 1 1
σ0 = = e ζ + (7.143)
4πa 2 λ a

As an example, consider a sphere having a radius a = 10−7 m = 100 nm in a

mixture where the number density is n 0 = 1025 molecules/m3 . The concentration
is
n0 1025 moles
c= = = 17 = 0.017 M (7.144)
NA 6.02 × 10 23 m3
with a Debye length of
 
RT r 0 8.315 × 300 × 78.54 × 36π
1
× 10−9
λ= = = 2.34 nm (7.145)
2I F 2 2 × 17 × 96,5002

For σ0 = −0.1 C/m 2 , the ζ potential at the surface is

λ
ζ∗ = σ0 = −0.34 V (7.146)
e

7.17 Electrical conductivity in an electrolyte solution

We have already defined the electrical conductivity for a general electrically con-
ducting material in Section 6.8; just as in a solid, electrically conducting media,
electric current in an electrolyte solution or plasma, originates from the motion of
charge carriers such as ions and charged biomolecules. For constant electrolyte
concentration, a given ion in the electrolyte reaches a constant velocity in the
direction of the electric field, the migration velocity, determined by the action
of the electric field and the electrostatic and hydrodynamic interactions with the
surrounding solvent molecules. These two effects act in opposite directions, with
the electric field tending to increase the migration velocity and the interactions
with the surrounding molecules tending to decrease the migration velocity.
We have already seen in Chapter 3 that the flux of a charged species A in an
electrolyte solution, in the absence of a velocity field, is given by

N A = −D A ∇c A + m A z A c A E (7.147)

and the current density is defined by


n
J=F z i Ni (7.148)
i=1

Assuming that the concentration is constant, we have


n
J=F zi m i E (7.149)
i=1
268 Essentials of Micro- and Nanofluidics

where m i = z i F Di /RT is the ionic mobility. Ohm’s law for an electrolyte

solution is

J = σe E (7.150)

so that the electrical conductivity σe is defined by


n
σe = F m i z i2 ci (7.151)
i=1

For solutions of 1:1 electrolytes, specifying the ionic strength is equivalent,

under electrical neutrality situations, to specifying the concentration of cations
and anions. If the ionic strength (without the 1/2, as the chemists define it) is

defined by I = z i2 ci , there are I /2 moles of cations and I /2 moles of anions
per liter of a 1:1 electrolyte solution. If the mobility of the cation is m + and the
mobility of the anion is m − , then equation (7.151) takes the form

(m + + m − )I F
σe = (7.152)
2
According to the classical theory of electrolyte solutions (Fuoss & Onsager,
1955), the mobilities of the ions are dependent on the diffusion coefficients,
which are functions of concentration, and so both m + and m − are dependent on
the concentration of the ions in the electrolyte and therefore on the ionic strength
I . The dependence of ionic mobilities on ionic strength results in the following
expression for the sum of mobilities in equation (7.152) (Fawcett, 2004):
" #
+ −
 0 B2 I 1/2
m + m = m + + m − − B1 m + + m − +
0 0 0
√ (7.153)
F 2(1 + κa)

where m 0+ and m 0− are the mobilities at infinite dilution (in the limit I → 0) for the
cation and anion, B1 and B2 are experimentally determined constants (Fawcett,
2004), and a is an ionic size parameter.
The mobilities at infinite dilution are most often experimentally determined
and often tabulated in standard physical chemistry texts (Castellan, 1983) or
electrochemistry texts (Fawcett, 2004; Bockris & Reddy, 1998). At 25◦ C, B1 =
0.2292 L1/2 mol1/2 and B2 = 60.57 × 10−4 S m2 L1/2 mol−3/2 . The SI unit of
conductivity is siemens/meter = S/m. In the denominator of the second term, κ
is the inverse of the Debye length and is proportional to I 1/2 .
As noted previously in this chapter, a given ion in an electrolyte is always
surrounded by a cloud of other ions, consisting predominantly of ions of opposite
charge and solvated by polar water molecules and thus hydrated. When an ion
moves, its ionic cloud moves in the opposite direction and generates an additional
viscous drag on the ion. The term containing B1 in equation (7.154) represents
this electrophoretic effect, which results in a decrease in the drift velocity of the
ion. A given ion moving together with its ionic cloud is also electrically polarized,
with the effective center of positive charge slightly leading the center of negative
269 Elements of Electrochemistry and the Electrical Double Layer

Table 7.2. The experimental and theoretical conductivities

calculated using equation (7.154) of I0 = 0.001, 0.01 and
0.1 M potassium chloride (KCl) solutions
I0 σe,exp ( mS ) σe,theo ( mS )
0.001 M 0.0147 0.0147
0.01 M 0.1413 0.1413
0.1 M 1.2880 1.2780

Equation (7.154) was used with a = 0.35 nm. The experimental

conductivity values are as quoted for the corresponding K Cl con-
ductivity standards in the product catalog of Aquaspex (http://www
.aquaspex.com.au/products/index.php?class=13).

charge in the direction of its motion. This charge asymmetry originates from the
finite time it takes for the ionic cloud to respond to the motion of the ion and is
known as the relaxation effect. The resultant electric field is oppositely directed
to the applied field and results in a reduction in mobility of the ion. The relaxation
effect is incorporated in the term containing B2 in equation (7.154). Substituting
equation (7.153) into equation (7.152),
 $  %
m 0+ + m 0− m 0 + m 0− I 1/2
σe = I F− B1 + F + B2 √ (7.154)
2 2 2 2(1 + κa)

As an illustration of use of equation (7.154), consider the conductivity of a

potassium chloride or KCl aqueous solution. The mobilities at infinite dilu-
tion for K+ ions is well known – m 0+ = 7.91 × 10−8 m2 /sV – and that for Cl−
ions is m 0− = 7.62 × 10−8 m2 /sV (Chang, 2000). The conductivity of an elec-
trically neutral I0 = 0.001 M KCl is calculated using these mobility values,
σe = F(m 0+ + m 0− )c = 0.0147 S/m, and this value agrees with experiment (see
Table 7.2). The magnitudes of the electrical conductivity for water under several
conditions are shown in Table 7.3.

Table 7.3. Electrical conductivities for

water under several conditions
Fluid σe
µS
Ultra pure water ∼0.06 cm
µS
Distilled water ∼1 cm
µS
Raw water ∼60 cm

0.05% NaCl ∼1 mS
cm

Seawater ∼50 mS
cm

In industrial practice, µS/cm is the unit of

choice for conductivity.
 270 Essentials of Micro- and Nanofluidics

Simple model of a cation exchange membrane.

Figure 7.16

7.18 Semi-permeable membranes

A phenomenon we have mentioned in the context of biological ion channels is

the ability of membranes to selectively allow one species to pass through but not
another. For example, a battery is a cation exchange membrane in the sense that
cations will be allowed to pass from the anode to the cathode freely, while the
anions are prevented from passage into the cathode, as depicted in Figure 7.16.
This is a subject that could be expanded into an entire chapter, and there is a
vast body of literature that I will mention in several open-ended exercises at the
end of this chapter. For now, let us consider a simple one-dimensional model
for such a membrane. The interested reader should consult the series of papers
by Rubinstein referenced in Zaltzman and Rubinstein (2007) and the papers by
Bazant et al. (2005) and Chu and Bazant (2005).
The governing equations for a one-dimensional analysis are
(gx + gφx )x = 0 (7.155)
( f x − f φx )x = 0 (7.156)
 2 φx x = −(g − f ) (7.157)
where  = λ/L and λ is the Debye length. The subscript x denotes a derivative
in the x direction. Here g is the dimensionless cation concentration scaled on
the bulk concentration, and f is the corresponding anion concentration. These
equations are subject to boundary conditions
f x − f φx = 0 x = 0, 1 (7.158)

g = g 0 x = 0, 1 (7.159)

φ=V x =1 (7.160)

φ=0 x =0 (7.161)
The potential has been scaled on RT /F, and g, f are scaled on their bulk
concentrations. With this scaling, these equations and boundary conditions are
271 Elements of Electrochemistry and the Electrical Double Layer

Lithium-Ion Batteries
Batteries are electrochemical cells that produce current by converting stored
chemical energy into electrical energy. The primary chemical reaction that drives
all commonly used batteries is called an oxidation-reduction reaction. The reac-
tions take place at each of two electrodes. The anode is where ions are oxidized
and lose some of their electrons. The cathode is where ions are reduced and gain
electrons. Depending on whether the cell is being charged or discharged, either
electrode can be the anode or cathode at a given instant. In between each half
reaction, the electrons traverse an external circuit and provide useful energy in
the form of electricity. A separator that is permeable to ions but not electrons
lies between the electrodes to prevent an electrical short circuit. A liquid or gel
electrolyte solution enables ion transfer between the electrodes.
As a specific example, consider the lithium-ion battery. This battery chemistry
has gained recent popularity in a variety of portable electronics and automotive
applications. The typical lithium-ion battery features one graphite electrode and
the other made of a metal oxide or phosphate such as LiCoO2 , Li N i O2 , or
Li Fe P O4 . During discharge, lithium ions are transported from the graphite
electrode into the electrolyte, described by the forward path of the half reaction
Li y C ↔ Li + + e− + C
where C represents a generic graphite compound and 0 < y < 1 is the amount of
lithium stored in the graphite electrode divided by its saturation value.
The ions diffuse through the separator to the cathode where they intercalate or
pass into the solid material. This process is described by the forward path of the
cathode half reaction
Li + + e− + M ↔ Li z M
where M represents a generic metal compound and 0 < z < 1 is the quantity
analogous to y in the cathode.
Transport of lithium ions into the cathode solid material completes the
oxidation-reduction reaction. During charging, the same process is repeated in
reverse and the reactions follow the backward paths of the aforementioned reac-
tions. The schematic below illustrates the path of electrons and lithium ions for
charge and discharge.

Contributed by Jim Marcicki.

272 Essentials of Micro- and Nanofluidics

supplenented by the condition that the number of anions remains fixed so that
1
f (x)d x = 1 (7.162)
0

where the concentration f has been scaled on its bulk value, where the anion
concentration is equal to the cation concentration. In a battery, for example, the
current is fixed, and only the cations provide current. Thus the current is given
by

I = N A = Ng = N (7.163)

Thus one integration of the governing equations for g and f yields

gx + gφx = N (7.164)

f x − f φx = 0 (7.165)

Consider first the outer solution away from the electrodes; adding the equations
for g and f results in
N
go = f o = (x + A) (7.166)
2
and subtracting,
1
φx = (7.167)
x+A
so that

φo = ln(x + A) + B (7.168)

To obtain A, we use the integral constraint on the anions and get

2
A= − 1/2 (7.169)
N
Thus, in the outer region, where to leading order the problem is electrically
neutral,
 
N 1
go = f o = x− +1 (7.170)
2 2
This is the Rubinstein result. The constant B is still undetermined.
To find B, we need to match the outer solution to the inner solution near the
electrodes. Near x = 0, we write, in the usual way, ξ = x/, giving a Boltzmann
distribution for both f and, g, and, for example,

g(ξ ) = g 0 e−φ x = 0 (7.171)

g(ξ ) = g 0 e−(φ−V ) x = 1 (7.172)

273 Elements of Electrochemistry and the Electrical Double Layer

where φ is the solution for the potential inside the EDL. Matching in the limit as
ξ → ∞, we obtain
 
N
lim ln(g 0 e− ) = ln g 0 − φ = lim ln g = ln 1 − (7.173)
ξ →∞ x→0 4
so that near x = 0, the outer solution for φ must satisfy
 
N
φ(0) = ln g − ln 1 −
0
(7.174)
4
and similarly, near x = 1,
 
N
ln g − (φ − V ) = ln 1 +
0
(7.175)
4
giving
 
N
φ(1) = ln g 0 + V + ln 1 + (7.176)
4
We now find the constant B by equating the limit of the outer solution for φ as
x → 1 to find
 
N
B = ln g0 + V + ln 1 + − ln(1 + A) (7.177)
4
Thus the outer solution for the potential is given by
 
1 g0
φ = V + ln N (x − 1/2) + 1) + ln (7.178)
2 (1 + N4 )2
Note the interesting result that to leading order,
1 N2
φx x = −   2
= 0 (7.179)
4 1+ 1N x − 1
2 2

so that the volume charge density must be nonzero at O( 2 ); that is, the volume
charge density
 
1 N2
ρe =  2
  (7.180)
4 1+ 1N x − 1 2
2 2

to leading order in  and is induced by the electric field.

The solution for the outer potential needs to be matched with the inner solution
at x = 0. This matching will yield the current–voltage relationship, and near
x = 0,
     
N N N
ln g 0 − ln 1 − = ln 1 − + ln g 0 − 2 ln 1 + (7.181)
4 4 4
leading to

1 − e−V /2
N =4 (7.182)
1 + e−V /2
 274 Essentials of Micro- and Nanofluidics

4 15

3
10
2
5
1
Current

Voltage
0 0

−1
−5
−2
−10
−3

−4 −15
−15 −10 −5 0 5 10 15 −4 −2 0 2 4
Voltage Current
(a) (b)

Plots of the current-voltage relationship for a cation exchange membrane.

Figure 7.17

This is the Rubinstein result. Note that as V → ∞, the current J = N → 4; this

value is termed the limiting current. In terms of the voltage, we have
1 + N /4
V = 2 ln (7.183)
1 − N /4
The result is depicted in Figure 7.17. The limiting current is reached at a dimen-
sionless voltage of about V ∼ 10.
If the permselective membrane is selective to cations, its concentration near the
electrodes will increase substantially, and the anion concentration will decrease
to near zero. A typical result is depicted in Figure 7.18. Note that this result
resembles the Gouy–Chapman distribution, for which the concentrations are
asymmetric around the mean and are discussed in Section 7.11, although this
time, the current is nonzero, resulting in a linear variation of the concentrations.

14
g
f
12 phi

10

8
g, f

0
0 0.5 1 1.5 2
x

Typical result for the scaled concentrations and potential in a cation exchange membrane, near x = 0. As
Figure 7.18
J → 4, the limiting current, the anion concentration at the wall, approaches zero.
 275 Elements of Electrochemistry and the Electrical Double Layer

10

Current
6 III
II
4

2
I
0
0 2 4 6 8
Voltage

Plot of the current-voltage relationship for a cation exchange membrane, illustrating the behavior in the three
Figure 7.19
regions; I, below the limiting current; II, at or near the limiting current; and III, above the limiting current.

The divergence of the cation and anion concentrations near a permselective

membrane is called concentration polarization (Zaltzman & Rubinstein, 2007).
Bazant et al. (2005) and Chu and Bazant (2005) discuss the structure of
boundary layers that appear when the limiting current is approached. If N = J =
4 − O( 2/3 ), a boundary layer of x ∼ O( 2/3 ) appears outside the classical diffuse
layer of O(). Chu and Bazant (2005) also analyze the regime above the limiting
current, as illustrated in Figure 7.19. Above the limiting current, there is a finite
region over which the anion concentration vanishes near the cathode, a case of
extreme concentration polarization. The concept of a limiting current was posed
by Nernst over a century ago (Chu & Bazant, 2005).
The EDL near each wall is fundamentally different from the EDL discussed
in Section 7.12 because the structure is generated not by a native surface charge
and/or ζ potential but by conducting electrodes with imposed potential that creates
a current. No current is passed by silica surfaces and other surfaces that have
been discussed previously in this chapter. Thus the EDLs near x = 0, 1 are said
to be induced by the electric field. The possibility of large electric fields can lead
to extreme polarization, whereby, in this case, the cation is highly concentrated
near x = 0, while the anion is severely depleted. This means that the solution
may not be dilute and that the electrical permittivity and the diffusion coefficient
may depend on the local concentration (Bazant et al., 2005). Moreover, large
concentrations may require the consideration of finite ion volume effects and
ion–solvent interactions, effects that are beyond the scope of this book. In this
problem, no flow field is generated in the x direction because there are no walls
to provide the excess charge required for such a flow.

7.19 The Derjaguin approximation

Notation alert: The symbol E in this section denotes interaction energy and not
electric field.
 276 Essentials of Micro- and Nanofluidics

The Derjaguin approximation uses the

a2 lubrication approximation to determine the
a1
interaction between colloidal particles or
between colloidal particles and walls (Der-
jaguin, 1934; Russel et al., 1991; Bhat-
tacharjee & Elimelech, 1997). Derjaguin
r
(1934) recognized that in many cases, the
interaction between different bodies, two
z
flat surfaces, two particles, or a particle and
a surface is not significant until the parti-
cles come very close together. Consider the
Figure 7.20
The interaction of two speres of radius a1 and interaction between two colloidal spheres
a2 , assuming the interaction is confined to a
shown in Figure 7.20. Then the force nor-
small area near the point of minimum distance
z = h, is called the Derjaguin approximation. mal to the surface of the two particles is
This assumption is equivalent to the lubrication given by an integral of the pressure over the
approximation. surface of the particle:

Fz = p(h)n · ez d S (7.184)
S

where h is the minumum distance between the particles. The energy between the
particle and the surface is the integral of the total force, or

E =− Fz dz (7.185)
z

where z measures the normal distance between the particles. Let

P=− p(h)dz (7.186)

z

so that P has units of N/m. Thus

n · ez
E= P(h) dA (7.187)
A |n · ez |
because d S = 1/|n · ez |d A, where d A = d xd y from standard vector integration
methods; note that the order of the integrations has been switched.
It is difficult to use Cartesian coordinates for the integration, and the Derjaguin
approximation for the interaction of two spheres of radii a1 and a2 relies on the
fact that in a local cylindrical coordinate system as depicted in Figure 7.20 (Russel
et al., 1991),
   
1 2 1 1 r4 r4
z(r ) = h + r + +O , (7.188)
2 a1 a2 a13 a23

where h is the point of minimum distance between the two spheres and thus
dz = r dr (1/a1 + 1/a2 ). The energy is thus
∞ 2π ∞
E = 2π Pr dr = Pdz (7.189)
0
1
a1
+ a12 h
277 Elements of Electrochemistry and the Electrical Double Layer

Note that if a1  a2 , the formula reduces to

∞
E = 2πa2 Pdz = 2πa2 Fz (7.190)
h

which is essentially the Derjaguin formula for the interaction between a plane,
wall, and sphere.
The force between the two bodies is obtained by differentiating equation
(7.189) with respect to h; the result is
a1 a2
Fz = 2π P (7.191)
a 1 + a2
where P is defined earlier.
Bhattacharjee and Elimelech (1997) and Bhattacharjee et al. (1998) have gone
beyond the Derjaguin approximation and calculated the interaction between two
spheres (Bhattacharjee et al., 1998) and a sphere and a plane wall (Bhattacharjee
& Elimelech, 1997) using what they call the surface element integration (SEI).
They relaxed the assumption that all the interactions are localized, integrating
over cylindrical slices on the sphere; for the interaction between a sphere and a
plane wall, they obtained
⎡ ⎛  ⎞ ⎛  ⎞⎤
a a r 2 r 2
E = 2π ⎣ P ⎝h + a − a 1− ⎠ − P ⎝h + a + a 1 − ⎠⎦ r dr
a 2 a 2
0 0

(7.192)

The case of two spheres is of a similar form.

The use of the Derjaguin approximation depends on the nature of the interac-
tions between the two bodies. For example, the form of energy per area, P, for
the plane wall–particle interaction for van der Waals attraction is given by
A
PvdW = − (7.193)
12π z 2
where A ∼ 10−18 J is the Hamaker constant, and so the Derjaguin approximation
gives
Aa
E =− (7.194)
6h
The integral using the SEI method can be integrated analytically; the result is
  
A a a h
E =− + + ln (7.195)
6 h h + 2a h + 2a
It is left as an exercise to plot equations (7.194) and (7.195) to show that the
Derjaguin approximation is inaccurate for h/a > 0.1.
As the particles get smaller, the Derjaguin approximation results in a larger
error because the local curvature is larger. The minimum value of h for which the
Derjaguin approximation is valid drops significantly as the particle size decreases.
Whether the particle can get that close to the wall becomes an issue, especially
 278 Essentials of Micro- and Nanofluidics

if the two surfaces have the same sign charge. Russel et al. (1991) suggests that
the approximation may fail even for micron-size particles.

7.20 Chapter summary

This chapter provides an introduction to the basic concepts of electrochemistry

that are important for the discussion of electrokinetic phenomena to follow. The
basic structure of water and the concept of hydration of ions and the definition
of the various radii that are used in the electrochemistry are presented. The
fundamental definitions of acids and bases are given for the reader to understand
ionization processes.
The chemical and electrochemical potential and the Gibbs function are then
introduced; the invariance of these potentials determines the equilibrium behavior
of the electrical potential and the concentrations of the ionic species within the
EDL on the liquid side of the surface.
These subjects lead to an extensive discussion of the EDL, including the pre-
sentation of the various phenomenological models of the EDL that have appeared,
including the description of the triple layer model of the EDL, a refinement of the
basic two layer model. The surface charge density is defined based on the nature
of site-binding models on the solid side of a surface.
The potential distribution within the electrical double and the Boltzmann dis-
tribution for the ionic species allows the solid-side surface charge density of
a site-binding model and fluid site-binding model surface charge density to be
linked, yielding a fundamental expression for the ζ potential as a function of pH
and the equilibrium constants of the other important adsorption reactions specific
surfaces.
The potential distribution in the EDL is derived for both monovalent symmetric
mixtures and asymmetric mixtures in the case of a plane wall. The potential
distribution on a cylinder or a sphere must be calculated numerically in the
nonlinear regime; analytical solutions are derived in the Debye–Hückel limit of
small potential.
Next is a discussion of the electrical conductivity of an electrolyte solution, a
measure of its ability to conduct electrical current from one electrode to the other.
This occurs in a semipermeable membrane that allows one set of ions to pass
through the solution and into the electrode(s), while preventing the other ions from
doing so. A cation exchange membrane allows cations to pass from the anode to
the cathode, while preventing anions from doing so. The concentrations resemble
the behavior of a Gouy–Chapman picture of the EDL and is one example of
concentration polarization. The chapter ends with an introduction to Derjaguin’s
approximation and an improvement to that approximation for the interaction
between two spheres or a sphere and a plane wall.
Many of these concepts will be revisited in Chapter 9 and Chapter 10 in various
forms. In particular, it will be seen that this EDL analysis leads to the analogy
279 Elements of Electrochemistry and the Electrical Double Layer

between the potential in the EDL and the velocity in electro-osmotic flow, as was
already pointed out in Chapter 3.

EXERCISES
7.1 A 0.2 M sodium–chloride–water solution (0.1 MNa, 0.1 MCl) is contained
in a region upstream of a nanopore membrane. The potential drop across
the membrane is measured to be −5 V. At a temperature of T = 20◦ C,
calculate the activity of sodium in the downstream reservoir.
7.2 Given that the reaction

SiOH + M+ SiOHM + H+

describes the adsorption of a metal cation, such as sodium, on a silica, with

the equilibrium constant defined by
[H+ ][SiOHM]
KM =
[SiOH][M+ ]
show that the surface charge density is given by
Ns F [H+ ]2s − K 1
σ =
N A K 1 + [H+ ]s + K M [M + ]
Calculate the surface charge density for the adsorption of sodium for which
K N A = 10−7.1 , and use the values of K 1 = 10−6.8 and ζ = −20 mV at
pH = 7.
7.3 Given that the reaction

SiOH + Ca2+ + H2 O SiOCaOH + 2H+

describes the adsorption of calcium on a silica surface, find an expression

for the surface charge density in terms of the appropriate activities or mole
fraction in the same form, as in the previous problem for a sodium–chloride–
calcium solution.
7.4 A 0.1 M sodium–chloride–water (0.05 Na, 0.05 Cl) mixture fills a chan-
nel made of silica surfaces. The surface charge density is known to
be σ = −20 mC2 . The ζ potential is unknown. Using the Debye–Hückel
approximation with the Neumann condition at the wall and the proper con-
dition in the bulk, calculate the potential if the EDL is thin. What is the
effective width of the EDL? Plot the potential near the surface y = 0.
7.5 For the previous problem, find the concentrations of the sodium and chloride
at the surface y = 0. Evaluate the validity of the Debye–Hückel approxi-
mation.
7.6 A 0.01 M sodium–chloride–water mixture fills a channel made of silica
surfaces, with a height of h = 100 nm. The surface charge density at y = 0
is known to be σ = −20 mC2 , and at y = h, the surface charge density is σ =
15 mV. Using the Debye–Hückel approximation and the proper condition
280 Essentials of Micro- and Nanofluidics

in the bulk, calculate the potential distribution in the channel. Repeat for
h = 10 nm.
7.7 For the previous problem, find the concentrations of the sodium and chlo-
ride at each of the surfaces. Evaluate the validity of the Debye–Hückel
approximation.
7.8 Consider the case of EDLs near each wall in a channel under the Debye–
Hückel approximation. Compare the dimensionless solutions for the fixed
ζ potential boundary condition and the fixed surface charge distribution at
the wall. Conclude that in dimensionless form, the surface charge density
and the ζ potential are equivalent. Identify the dimensional form of this
equivalence as the Grahame equation.
7.9 Plot the dimensionless potential for a z:z electrolyte of known ζ potential
near y = 0 for z = 1, 3 (equation (7.92)) and dimensionless ζ potential,
ζ = φs = −2. What do you notice about the effective EDL thickness (i.e.,
the number of Debye lengths), defined as when the potential φ < 0.01?
Base the Debye length on a bulk ionic strength of I = 0.3 M. Perform the
calculation at T = 300 K. Compare with the Debye–Hückel result.
7.10 In the text, the solution of Grahame for a 2:1 electrolyte was calculated.
Plot the solution for the potential for ζ = 1, and compare with the solution
for a 1:1 electrolyte. Plot each of the concentrations, and again compare
with the 1:1 case. A numerical solution using a zero-finding technique is
required.
7.11 Following Grahame, calculate the solution for a 1:2 electrolyte solution,
and repeat the steps of the previous problem. In this case, the key identity
is given by
(e φ − 1)(1 + 2e−φ )1/2 = (e2φ + e−φ − 3)1/2
7.12 An electrolyte solution resides between two plates having the same ζ poten-
tial. Show that the pressure between the two plates in a static fluid satisfies
dp dφ
+ ρe =0
dy dy
where ρe is the volume charge density and ρe = −e d 2 φ/dy 2 . For a
1:1 electrolyte, show that the force between the two plates under the
Debye–Hückel approximation due to the presence of the electrolyte solution
satisfies
F 1
= ∓ e ζ 2 κ 2 at y = 0, 1
A 2
for κh  1, and
F 1
= p(h) − p(0) = e ζ 2 κ 2 tanh κh for κh  1
A 2
What do you conclude about the force between the plates in each limit?
Here κ = 1/λ. Note that d p/dy = 0 when ρe = 0. The pressure p in this
problem is one form of osmotic pressure.
281 Elements of Electrochemistry and the Electrical Double Layer

7.13 From the liquid-side point of view, for a z:z electrolyte, the ζ potential
and the surface charge density are related by equation (7.115). Given that
the ζ potential can be measured using the streaming potential, plot the
surface charge density for ζ = 26, 52, 78 mV at a temperature T = 35◦ C
for a 0.15 M solution for z = 1. Keeping the molarity fixed, how does the
solution change for z = 2, 3?
7.14 For a binary symmetric monovalent electrolyte, a Taylor series approxima-
tion to the exponential Boltzmann distribution yields the equation for the
potential as (Oyanader & Arce, 2005)
 
2d φ φ2 φ4 φ6
2
 =φ 1+ + + + ···
dy 2 3 5 7

Solve this equation numerically, and find the potential and mole fraction
profiles for electro-osmotic flow in a 20 nm channel. The electrolyte solution
is 0.1 M NaCl, and the ζ potential on the walls is −40 mV.
7.15 Plot the ζ potential as a function of pH for a 0.015, 0.15 M binary mixture
of monovalent electrolytes.
7.16 In the limit as  → 0, find the solution for the potential on the outside of
the surface of a cylinder for the constant potential boundary condition.
7.17 The governing equations for a cation exchange membrane in the core are

gx + gφx = I = N

f x − f φx = 0

Defining

1
c= (g + f )
2
and
1
ρ= (g − f )
2
show that the equations can be written in the form

cx + ρφx = N /2

ρx + cφx = N /2

and that conservations of anions reduce to

1
(c − ρ)d x = 1
0

Away from the limiting current, ρ  c; using this approximation, deduce

the form of c and ρ. Show, in particular, that using conservation of anions
282 Essentials of Micro- and Nanofluidics

in the outer region with I = N ,

 
I 1
g= f = x− +1
2 2
   
I 1
φ = ln x− +1 + B
2 2
where B is a constant to be determined by matching.
7.18 This is an open-ended exercise. Write a report on the derivation of the
Butler–Volmer law of electrochemistry. Discuss what is it used for, and
relate its use to electrochemical cells.
7.19 This is an open-ended exercise. Write a report on the results of the paper
by Bazant et al. (2005). How is the Butler–Volmer law used? What are the
differences between this paper and the paper by Zaltzman and Rubinstein
(2007)?
7.20 Compare the results using the Derjaguin and SEI methods for the van der
Waals attraction, equations (7.194) and (7.195), as a function of h/a. For
a fluid particle of radius a = 1 µm, at what value of h is the difference
between the two results more than 20 percent?
7.21 Using the methods described by Bhattacharjee et al. (1998), show that the
interaction energy between two spheres of radii a1 and a2 in a van der Waals
attraction is given by

A 2a1 a2 2a1 a2
E vdW = − + 2
6h h + 2a1 h + 2a2 h
2 h + 2a1 h + 2a2 h + 4a1 a2
 
h 2 + 2a1 h + 2a2 h
+ ln
h 2 + 2a1 h + 2a2 h4a1 a2
 8 Elements of Molecular
and Cell Biology

8.1 Introduction

It may be said that at some level cells are the building blocks of all living things.
The most basic of all organisms are made up of single cells, and higher organisms,
such as animals and humans, are made of a large number of cells all arranged in
a specific way.
Because the nanoscale is the scale of biology, there has been an explosion of
new knowledge and new ideas in the biological sciences. This has been the result
of significant advances in static measurement techniques such as the surface
force apparatus (SFA), the atomic force microscope (AFM), and the scanning
tunneling microscope (STM). Moreover, optical techniques can be employed for
single molecule detection and analysis. With explosive improvements in com-
puter architecture, molecular dynamics (MD) and its derivatives (nonequilibrium
molecular dynamics, NEMD) are now being employed to study the motion of
ions and proteins in a flowing solution and the conformation of proteins and other
macromolecules in a static medium.
The links between nanotechnology and biology are numerous. Entire devices
for the rapid analysis of proteins, drug delivery, biochemical sensing, and other
applications are being developed at a rapid pace. Entire books are being published
on biomedical applications at the nanoscale (Malsch, 2005; Ciofalo et al., 1999).
Artificial and implantable pumping systems employing nanopore membranes are
being developed for filtering proteins and ions for use as a renal assist device
(Fissell, 2006) (see Figure 4.1).
The main objective of this chapter is to introduce the student to the most
common biomolecules used in the applications of interest. These biomolecules,
such as proteins and polysaccharides, are soft materials and are deformable when
sheared by a fluid flow. That being said, it is extremely difficult to describe the
nature of the deformation, though some measurements have been reported in the
literature. Indeed, albumin is thought to form the shape of a wedge of cheese as
it enters a nanopore membrane (Ferrer et al., 2001). Although a matter of debate
in the literature, the most important properties of a biomolecule are its overall
size and valence or charge. These two properties determine the primary transport
characteristics of the biomolecule.
283
284 Essentials of Micro- and Nanofluidics

The Atomic Force Microscope (AFM)

The atomic force microscope (AFM), also
known as the scanning probe microscope (SPM),
is a high resolution imaging technique capable
of resolving surface features on the order of frac-
tions of nanometers. To obtain a surface image,
a cantilever with a sharp tip is scanned on the
specimen surface. Interaction forces between the
tip and the surface (on the order of nanonew-
tons), coming from mechanical contact, van der
Waals, and capillary forces, lead to deflection
of the cantilever. This deflection is measured
using optical detection techniques. A piezoelec-
tric tube is responsible for scanning the sample
horizontally and vertically, and is attached either
to the specimen (as shown in the example above)
Example of a commercial atomic force or to the assembly containing the tip, depending
microscope. on the microscope configuration. In the static or
contact mode of operation, the tip is in contact
with the specimen at all times, the tip-surface force is kept constant, and a feed-
back mechanism allows the tip to trace the surface topography during the scan.
In the dynamic or “tapping” mode, the cantilever is vibrated and the oscillation
amplitude is maintained constant during scanning. The AFM can be used to probe
the structure of biological cells.
In addition to surface imaging, the AFM is also a versatile nanoscale char-
acterization technique. Imaging can be performed either in air or while the tip
and specimen are immersed in a liquid medium. Other capabilities of the AFM
include the measurement of mechanical forces (such as lateral forces, viscoelas-
ticity, adhesion etc.), electrical properties (such as capacitance, resistance, surface
potential), thermal properties, and magnetic forces.

Contributed by Manny Palacio

In this chapter, we introduce the basic concepts of molecular and cell biology,
with an emphasis on information that is being exploited by nanotechnology today.
This chapter is not meant to be all-inclusive; much more material is found in the
excellent book by Alberts et al. (1998), which I have found extremely informative
and highly recommend to the interested reader.
The chapter begins with a discussion of the chemical structure of nucleic
acids, and polysaccharides are described, with an emphasis on the structure of
DNA. Proteins and protein binding are described, leading to a discussion of the
operation of a cell and the structure of its outer membrane. The chapter concludes
with a description of the transport properties of ion channels.
 285 Elements of Molecular and Cell Biology

8.2 Nucleic acids and polysaccharides

Nucleic acids are polymers consisting of nucleotides. Those based on a sugar

called ribose are called ribonucleic acids (RNA), and those based on deoxyribose
are called deoxyribonucleic acids (DNA). RNA is single stranded, whereas DNA
is usually double stranded, although single-stranded DNA (ss-DNA) does exist.
These two nucleotides are described in greater detail later.
Early in the 1940s, it was discovered that genetic information is primarily
involved with developing instructions for making proteins. Deoxyribonulceic
acid (DNA) was identified in the 1940s as the likely carrier of this genetic
information (Alberts et al., 1998). Crick and Watson (1953) determined the
structure of DNA as a double helix, which immediately made clear how the DNA
may be copied, or replicated, and began to explain the mechanism by which DNA
encodes instructions for making proteins.

Watson, Crick, Franklin, and Wilkins and the Discovery of DNA

James Watson (1928–) and Francis Crick (1916–2004) were colleagues at the
Cavendish Laboratory in Cambridge when they began their work leading to the
discovery of the primary structure of DNA. Watson’s background was in genetics,
having a PhD from the University of Indiana. Francis Crick was a physicist having
received a bachelor’s degree from University College, London. He was already
at Cavendish when Watson arrived and together they put together the structure of
DNA. Two other colleagues played key roles and the paper Watson and Crick wrote
(Crick and Watson, 1953) relied on their colleagues’ input. Rosalind Franklin had
produced X-ray crystallography images of DNA that Watson and Crick used to
formulate their view of the structure of DNA. Unfortunately Franklin died in
1958 before she could be honored with the Nobel prize that was won by Watson
and Crick in 1962. Maurice Wilkins was a nuclear physicist who changed fields
to work with Franklin in X-ray crystallography. Wilkins had also produced X-
ray images of DNA of which Watson and Crick were aware. In reality, many
believe that both Franklin and Wilkins should have been co-authoers on the paper
describing one of the most important discoveries in the history of the world.

DNA, genes, and chromosomes are interrelated in that based on work in the
early twentieth century, it was discovered that genes are carried by chromosomes,
which consist of DNA and proteins. In a glossary at the end of his excellent book,
Alberts et al. (1998) defines a gene as a “region of DNA that controls a discrete
hereditary characteristic of an organism, usually corresponding to a single protein
or RNA” (p. G-8). In the same way, he defines a chromosome as a “long threadlike
structure composed of DNA and associated proteins that carries part or all of the
genetic information of an organism (p. G-4). Thus these three structures, DNA,
genes, and chromosomes, and their associated proteins work together to keep cells
functioning properly and are crucial elements in the development of new cells.
 286 Essentials of Micro- and Nanofluidics

Base pairs
Adenine Thymine

Guanine Cytosine

Sugar phosphate
backbone

Sketch of double stranded DNA showing the four bases adenine (A), cytosine (C), guanine (G), or thymine (T).
Figure 8.1
Image courtesy of the National Library of Medicine.

DNA is composed of two long nucleotide chains held together by hydrogen

bonds. Nucleotides are structures containing five-carbon sugars attached to one or
more phosphate groups (a phosphorus central atom surrounded by four oxygens)
and a base that can be either adenine (A), cytosine (C), guanine (G), or thymine
(T). A high-resolution image and a sketch of DNA is depicted in Figure 8.1. Two
nucleotides connected by a hydrogen bond is called a base pair (bp). The length
of a DNA molecule is measured by its number of base pairs. One base pair is
approximately 2 Å long and 3 nm in diameter. A DNA strand can unravel and be
as long as several microns.
RNA, or ribonucleic acid, as with DNA, is made up of four nucleotides con-
taining the four bases uracil (U), adenine (A), cytosine (C), and guanine (G).
RNA has two very important features: it carries information based on its individ-
ual sequence and passes it on by replication, and its unique shape determines its
function.
In these DNA–RNA structures, A pairs with T or U and G pairs with C; the
A–T(U) pairing contains two hydrogen bonds, and the G–C pairing contains three
hydrogen bonds. RNA comes in many forms, and messenger RNA (mRNA) is
one of the most important. The production of mRNA from its DNA template is
called transcription, and mRNA is essential in the synthesizing, or coding, of
proteins. The coding process is the initial step in the assembly of amino acids
into a protein.
Polysaccharides consist of a large number of sugars, many composed of glu-
cose, which can be linear as well as globular in shape; glucose is globular in
 287 Elements of Molecular and Cell Biology

solution, with a nominal radius of a = 0.37 nm. The term globular means
that the molecule is roughly in the shape of a sphere; polysaccharides can be
highly deformable. Glucose alone is a monosaccharide with the chemical formula
C6 H12 O6 . The monosaccharides have a chemical formula of the form (CH2 O)n ,
where n is usually n = 3, 4, 5, 6, 7. The simple sugars are the primary energy
sources for cells.
Polysaccharides are held together by covalent bonds and are often very large
and almost macroscopic in size. They occur in bacteria cell walls, and plant cells
are made exclusively of polyaccharides.

8.3 Proteins

Many biologists consider proteins to be the most important biological element

of any living system. All chemical reactions within a living system are catalyzed
by enzymes, which are protein molecules that catalyze, or increase the rate of,
a biochemical reaction by themselves or in complexes with small molecules.
They provide an organism with the essential food elements of carbon, hydrogen,

The Thermodynamics of Molecular Biology

As long ago as the early 1960s, Terrell L. Hill was talking about the Thermody-
namics of Small Systems (Hill, 1963, 1964). The “small systems” he was talking
about at the time were colloidal particles and biomolecules in solution. His Part I
and Part II monographs were part of the, at that time, W. A. Benjamin Inc. series
titled “Frontiers in Chemistry.” He was a Professor of Chemistry at the University
of Oregon and the monograph appears to be the first in the series of short books
on these topics.
Terrell Hill is called “Theoretician Extraordinarius” by Lloyd Ferguson in a
short, two-page historical narrative published in the journal Cell Biochemistry
and Biophysics Vol. 11 in 1987, a volume dedicated to Professor Hill. Hill worked
on the Manhattan Project at UC Berkeley and worked at the National Institutes
of Health as Chief of Theoretical Molecular Biology in the National Institute of
Diabetes, and Digestive and Kidney Diseases. In the preface of Part I, Hill (1963)
says

Small thermodynamic systems should be of interest to three classes of read-

ers (1) experimentalists working with colloidal particles, polymers, or macro-
molecules; (2) theoreticians concerned with the preceding fields or with the
statistical mechanics of any kind of finite system; (3) those with interest in
thermodynamics per se.
He talks about biolecules in solution and his main objective is to to extend
the range of validity of ordinary thermodynamic definitions to include small,
nonmacroscopic systems (e.g., a single macromolecule or colloidal particle.)
288 Essentials of Micro- and Nanofluidics

For example, he writes the Gibbs free energy of a colloidal particle

G = N f ( p, T ) + a( p, T )N 2/3 + b(T )lnN + c( p, T )

Here, N is the number of molecules in solution and, as N → ∞, the macroscopic
limit is attained. Only the first term describes macroscopic thermodynamics and
the last three terms describe the microscopic system. Recall that in the macro-
scopic sense the chemical potential μ is defined by
∂G
μ= | p,T
∂N
Hill appears to be the first researcher to address these problems in a formal way.
As applications he suggests in Volume I

We may anticipate two main classes of applications of small systems thermo-

dynamics: (1) as an aid in analyzing, classifying, and correlating equilibrium
experimental data on “small systems” such as (noninteracting) colloidal parti-
cles, liquid droplets, crystallites, macromolecules, polymers, polyelectrolytes,
nucleic acids, proteins, etc.; and (2) to verify, stimulate, and provide a frame-
work for statistical mechanical analysis of models of finite (i.e. “small”) sys-
tems.
Terrell Hill is the forerunner of researchers in the modern era of the thermody-
namics of molecular biology.

nitrogen, and sulfur. Also, proteins carry out most of the processes in the cell.
These molecules are made up of a combination of 20 different amino acids that
are joined together by a covalent peptide bond.
The production of a protein, or protein synthesis, begins with a gene on a
particular strand of the DNA. A gene governs which amino acids join together;
they do this by expelling a molecule of water, a process called condensation. A
gene is made up of a linear array of bases that contains the specific information
or coding for the synthesis of a protein. Thus a gene contains the information
required to produce a functional RNA. A gene is said to be expressed when its
product, the RNA or a protein, is completed.

8.3.1 Protein function

There are seven basic types of proteins classified according to their function,
although different authors use different terms to describe each class; see, for
example, (Alberts et al. 1998, panel 5-1). Enzymes are catalysts in biological
reactions within the cell. The immune system responds to foreign bacteria and
viruses by producing antibodies that destroy or bind to the antigen, the foreign
agent. The antigen is the catalyst, or reaction enhancer, for inducing the immune
response: the production of the antibodies.
 289 Elements of Molecular and Cell Biology

Polypeptide backbone

Gly Pro Thr Gly Thr Gly Ser Lys

Amino acid
subgroups

Sketch indicating the primary structure of a protein.

Figure 8.2

Some proteins, called transport proteins, carry materials around in the body
(e.g., hemoglobin and oxygen), where that material is used by different cells.
Regulatory proteins control the metabolism and reproduction of the cell by man-
aging aspects such as temperature and pH. Other proteins, structural proteins,
have mechanical properties that give support to bones, tendons, and skin in man
and large animals. The interactions of actin and myosin proteins, which are move-
ment proteins, provides the proper mechanism for the contraction and expansion
of the heart and the other muscles throughout the body.
Nutrient proteins are important to a growing organism because of their abun-
dance of amino acids. Proteins are present in the solid material of the body as
well as in blood serum. For example, actin is one of the three major components
of the cytoskeleton. Proteins are dynamic structures that rotate and deform in
response to random thermal motions (Alberts et al., 1998).
The two most important properties of proteins are their structure and their
function, the latter of which we just described. We now discuss their structure.

8.3.2 Protein structure

Proteins are complex molecules, and as such, the entire structure cannot be
viewed at the same time; moreover, all but the simplest proteins cannot be written
in terms of a simple chemical formula. Thus the biology community has defined
several levels of structure, described here.
The primary structure (or amino acid sequence) is a complete description of
the chemical parts of a protein. The sequence is unique to the particular organ
or tissue and replicates by cell division. Amino acids are bonded together in a
chain by peptide bonds, which are amide bonds formed between the α-carboxylate
group of one amino acid and the α-amino group of another. The primary structure
of a single polypeptide is shown in Figure 8.2.
The main component of a protein is the polypeptide backbone, which is com-
posed of an amino acid sequence to which side chains are attached. The side
chains can be polar or nonpolar and charged or uncharged. The polar side chains
290 Essentials of Micro- and Nanofluidics

Human serum albumin and its cousin bovine serum albumin are proteins of enor-
mous importance in biology. The function of the kidney is to filter out small ions
and other small constituents of blood serum while retaining larger biomolecules
such as HSA and the red and white blood cells. In addition, altered HSA appears
to be an indicator of the progression of heart disease (Sinha et al., 2004). It is well
known that albumin binds to metals at its N-terminus, particularly cobalt. Car-
diac ischemia, or the reduction of the transport of blood and oxygen to the heart
appears to coincide with the reduced ability of albumin to bind to cobalt. Such
a situation will change the electrostatic properties of an albumin molecule and
hence its transport properties in solution. Ischemia Technologies of Denver, CO
commercialized the first device to analyze albumin for indications of cardiac
ischemia in 1997.

Cobalt bound to albumin at the N-terminus. Albumin without bound cobalt.

are often arranged on the outside of the protein, where they can hydrogen-bond
with water molecules and other polar solvents. The nonpolar or hydrophobic side
chains are often arranged so that they do not disrupt the water molecules; these
side chains tend to “hide” from water on the inside of a protein. Amino acids
have an amino group on one end (NH2 ), called the N-terminus, and a carboxyl
group (COOH) on the other end, called the C-terminus.
The polypeptide backbone of a protein folds back on itself into a shape or
conformation based on the criterion that the shape be that of minimum Gibbs
free energy. A protein can be unfolded under the action of a particular solvent,
a state in which the protein is termed denatured, meaning the protein loses its
biological function capability, and this situation is depicted in Figure 8.3. The
difference between the denatured and native state (folded) may be only several
 291 Elements of Molecular and Cell Biology

Sketch showing protein folding and the unfolding or denatured state in the presence of a particular solvent.
Figure 8.3
Many proteins can be denatured by urea.

hundred hydrogen bonds out of many thousands of bonds, depending on the size
of the protein.
The appropriate conformation of a protein in a cell is highly stable and will
change only slightly on interaction with other molecules. The way the polypeptide
backbone folds is determined entirely by its amino acid sequence. The first
protein to have been “sequenced” was insulin, whose sequence was decoded
in 1955 (Alberts et al., 1998). It should be noted, however, that it is still not
possible to predict a protein’s shape from its amino acid sequence, and many
proteins are much more complex than insulin, some even containing more than
one polypeptide backbone.
The secondary structure depicts the long amino acid chain by describing
its folding pattern. While all proteins have their own unique folding patterns,
two of the most common patterns are the α-helix and the β-sheet. Maximizing
the hydrogen bonding between amide hydrogen and carbonyl oxygen of the
peptide bonds causes these two types of constructions. The α-helix is a helical,
right-handed configuration that involves all amide hydrogen (N–H bonds) and
carbonyl oxygen (C=O, double bond) of the peptide backbone. The carbonyl
oxygen interacts with the amide hydrogen of the amino acid four down from the
oxygen.
The nearly completely extended β-sheet also involves the hydrogen bonding
of all the amides and carbonyls. The structure can be further classified as either
a parallel or antiparallel β-sheet. It is the secondary structure that is visualized
in several ways, as depicted in Figure 8.4. Each portion of the protein is color
coded in a specific way by the user. The software package ProteinShop is used to
produce Figure 8.4; another common rendering program is MidasPlus. Proteins
can also be rendered using a wire model and a space-filled model (Alberts et al.,
1998).
The tertiary structure describes the additional folding of the secondary struc-
ture. The three-dimensional twisting and folding of the overall chain is depicted;
 292 Essentials of Micro- and Nanofluidics

(a) (b)

Two means of rendering the secondary structure of a protein. (a) Cartoon rendering showing the α-helix helix
Figure 8.4
tubes and the β-sheet shown by the arrows and the coiled regions. This is called a ribbon model. (b) Same as
(a), using only tubes, called a backbone model. These diagrams were rendered using the software program
ProteinShop. See http://vis.lbl.gov/scrivelli/public/∼silvia page/Pshp.gallery.html.

the forces involved in this further folding are van der Waals forces, hydrogen
bonding, ionic bonding, and even covalent bonding. The van der Waals force
folding is between the side chain R groups of nonpolar amino acids that are
hydrophobic. The hydrogen bonding is between the side chain R groups consist-
ing of one or more organic compounds CH x . Ionic bond folding is the result of
the interaction of a positively charged amino acid and a negatively charged one.
Covalent bonding also occurs between thiol-containing amino acids. A thiol is a
molecule that contains a sulfur hydrogen bond (S-H). The term quartenary struc-
ture is used if the protein contains more than one backbone. Sketches rendering
all these structures are depicted in Figure 8.5.
Bioinformatics is the term given to the creation of databases and computational
algorithms for the advancement of our knowledge of the biological world. The
structures of proteins are most often determined experimentally, generating a very
large amount of information that must be reduced and manipulated to produce
the types of pictures seen in this section. More frequently these days, molecular
dynamics simulations are used for the same purpose (Figure 8.6).

8.3.3 Some common proteins

Proteins occur in the solid material of the body as well as in blood serum. For
example, actin is one of the three major components of the cytoskeleton. The walls
of ion channels, the primary conduits to and from cells, are made of proteins,
which allow certain ions into the cell, depending on the function of the particular
channel, and certain waste materials out.
Albumin is a charged protein that exists in many forms. Human serum albu-
min (HSA) is a plasma protein made in the liver that helps maintain chemical
equilibrium in the bloodstream. HSA can bind fatty acids and other hydrophobic
 293 Elements of Molecular and Cell Biology

Primary protein structure

as sequence of a chain of amino acids
Amino Acids

Pleated sheet Alpha helix

Secondary protein structure

occurs when the sequence of amino acids
are linked by hydrogen

Pleated sheet
Tertiery protein structure
occurs when certain structure are present
between alpha sheet and pleated sheets
Alpha helix

Quaternary protein structure

is a protein consisting of more than one
amino acid

Sketches depicting the various structures of a protein. Image courtesy of the National Library of Medicine of
Figure 8.5
the National Institutes of Health.

molecules and can also be used as a drug carrier; there are six distinct sets of
binding sites on HSA.
Drugs are often classified according to the extent that they bind to plasma
proteins. Calcium is transported by HSA; the maintenance of a stable level of
calcium is important in cardiac tissue. Albumin has an average concentration
of about 40 mg/ml in serum and has a molecular weight of M = 66 kDa. The
diffusion coefficient of albumin in blood serum (an electrolyte mixture) is D =
6.1 × 10−11 m2 /sec, and its valence is z ∼ −19 (Peters, 1996).1

1
Recall that the valence is the number of electrons required to fully populate the outer shell of an
atom. To speak of a valence of a protein, then, may not be appropriate because the surface charge
of a protein usually varies with position on the outer surface of the protein. A protein may also
have a volume charge density. Nevertheless, it is common to speak of the valence of molecules
such as albumin.
 294 Essentials of Micro- and Nanofluidics

Space-filling view of human serum albumin. In a color rendering molecules are coded according to the CPK
Figure 8.6
(Koltun, 1965) color scheme: gray is carbon, oxygen is red, nitrogen is blue, and yellow is sulfur. From the
Protein Data Bank, pdb1bm0.

Human IgG is the most abundant of human immunoglobulins, which are

more commonly known as antibodies. They recognize antigens on the surface
of foreign bodies like bacteria and viruses. Human IgG has a distinctive hinged
shape characteristic of antibodies (see Figure 8.8).
Human insulin triggers cells to absorb excess glucose from the blood when the
glucose level gets too high. Produced in the pancreas, insulin is soluble in serum
at physiological pH ∼ 7.4. Its molecular weight is ∼6 kDa. If the pancreas cannot
produce enough insulin (type 1) or the cells removing glucose casnnot use the
insulin that is produced (type 2), diabetes is the result.
The key role of apolipoproteinE is the metabolism of lipoproteins, which are
a macromolecular complex of lipids and proteins that form the walls of the cell.
Apolipoprotein activates enzymes that are important in the covalent modification
of lipids. They contain receptor ligands that help direct remodeled lipoproteins
to specific tissue sites. Apolipoproteins are also involved in development and
repair of nervous tissue. There are more than 18 variants of apolipoproteinE,
and an imbalance in one of them (called E4) can result in the development of
Alzheimer’s disease. The molecular weight is ∼34.2 kDa.
Lysozyme is a major bacteria-fighting component of the immune system. The
small enzyme destroys the carbohydrate cell wall of bacteria. Lysozyme protects
areas that are abundant in food from bacteria and is found in tears, mucus, saliva,
blood serum, and milk. The molecular weight is around 14.4 kDa.
Human interferon is another major part of the immune defense system. Once
activated, interferon slows the growth of a foreign substance, say, bacteria and
viruses, by slowing or completely blocking its function. There are three classes of
interferon: α, β, and γ . Interferon α–2a may be used to treat a form of leukemia;
interferon α–2b may be used to treat hepatitus C, and the molecular weight is
∼19 kDa.
 295 Elements of Molecular and Cell Biology

Illustration of protein binding, showing a number of weak noncovalent bonds holding the ligand (red) in place.
Figure 8.7

8.3.4 Few polypeptide chains are useful

Because there are 20 distinct amino acids, there are 204 possible combinations
for a polypeptide chain four amino acids long. Thus there are 20n combinations
of a polypeptide chain n amino acids long. This is a vast number, but fortunately,
many fewer than these very large numbers will be important due to the unique
structure and function of each polypeptide chain. Moreover, owing to natural
selection, each protein conformation is extremely stable, although even a change
of a few atoms in the structure can have disastrous consequences (Alberts et al.,
1998).

8.4 Protein binding

One of the consequences of the specificity of a protein’s structure and function is

the ability of the protein to bind to other molecules. This binding can be either
strong and long in duration or weak and short-lived. A particular protein can bind
only one or a small number of molecules, and that molecule is called a ligand.
Where that ligand binds on a given protein is called a binding site. Many proteins
have several binding sites.
Binding sites on a protein are cavities in the surface of the protein; the ligand
fits into the cavity such that a large number of weak noncovalent bonds may form,
as shown in Figure 8.7. Note that the ligand is ideally suited to the binding site, and
unless this is so, many weak noncovalent bonds do not form, and the binding fails.
As has already been mentioned, immunoglobulins, or antibodies, are the body’s
response to a foreign substance. They bind to the target called the antigen with
a remarkable specificity, and there is a one-to-one correspondence between the
number of antibodies and the number of antigens. Antibodies are Y -shaped, with
two antigen binding sites at the tips of the Y , as shown in Figure 8.8. This
 296 Essentials of Micro- and Nanofluidics

Antibody
Heavy chain

Light chain

Antigen-binding
region

Constant region

Assembled antibody molecule

Sketch of an antibody.
Figure 8.8

structure is known to be well suited for grabbing other molecules. Antibody–

antigen binding is extremely tight, so the limiting rate constant is the dissociation
constant; that is, koff ∼ 0 (see the chemical reaction boundary condition discus-
sion in Chapter 3).
Two of these chains are said to be “light,” and two of them are said to be
“heavy,” with the terms referring to their relative molecular weights. IgG also
contains two binding sites for the antigen, as shown in Figure 8.8, and there are
several types of IgG, which are built up from the same basic structure.
IgG occurs naturally in the body but may also be introduced into the body in
many different ways (Saltzman, 2001). The controlled delivery of IgG extends
its life in the body greatly. However, depending on the methods of delivery, early
time results for the fraction of IgG in the plasma (<5 days) can vary widely.
The strength of the binding of a protein to a ligand is characterized by its
equilibrium constant, K . The process of binding is considered a chemical reaction,
and if the chemical equation is defined by

A+B AB (8.1)

the equilibrium constant is defined by

kon [AB]
K = = (8.2)
koff [A][B]

where kon is the association rate of the bound complex [AB] and koff is the disso-
ciation rate of [AB]. The requirement of conservation of species at equilibrium
that dissociation rate = association rate leads to equation (8.2). The greater the
equilibrium constant, the greater the strength of the binding; typically, in a cell,
in dimensional form, K ∼ 108 –1010 L/mole.
297 Elements of Molecular and Cell Biology

At equilibrium, the change in Gibbs function for the reaction dG = 0, where

G is the Gibbs free energy.2 Then dG = 0 requires that G f − G i = 0, where the
subscripts f and i denote the final and initial states, respectively. Initially, the
value of the Gibbs free energy is the sum of that for species A and species B:

G i = G i0 + RT ln[A] + RT ln[B] (8.3)

After the reaction has taken place,

G f = G 0f + RT ln[AB] (8.4)

requiring

G 0 [AB]
− = ln = ln K (8.5)
RT [A][B]

where G 0 is the Gibbs function change for the reaction G 0 = G 0f − G i0 . Thus,

if K = 10 at T = 298 K, then G 0 = −1.36 kcal/mole = −5.7 kJ/mole, and
each 10-fold increase in the equilibrium constant corresponds to an increase in
free energy of about 1.4 kcal/mole.
As a typical example described by Alberts et al. (1998) supposes n molecules of
both A and B are present in a cell; let the concentration initially [ A] = [B] = C.
Let C − x be the concentrations of A and B at equilibrium. Because the reaction is
given by [A] + [B] [AB], the number of moles at equilibrium of the complex
[AB] is x. Then, using the definition of K ,
x
=K (8.6)
(C − x)2
or

K (C − x)2 − x = 0 (8.7)

This is a single quadratic equation for x that can easily be solved by the quadratic
formula, and the result is

2K C + 1 ± (2K C + 1)2 − 4K 2 C 2
x= (8.8)
2K
In this case, only the negative sign is relevant (why?), and for a solution containing
1000 molecules of A and B, which, for a typical cell, corresponds to [ A] = [B] =
10−9 M, and for K = 1010 L/mole at equilibrium, there are 270 molecules of A
and B and 730 molecules of AB. Of course, as the equilibrium constant decreases,
the yield of AB decreases.

2
The mole fractions should be used here to avoid a dimensional quantity within a function; here
square brackets denote concentration, as in Alberts et al. (1998), and the end result will not be
affected.
 298 Essentials of Micro- and Nanofluidics

Sketch of sliced-open animal eukaryotic cell. Image courtesy of the National Institute of General Medical
Figure 8.9
Sciences of the National Institutes of Health.

8.5 Cells

The cell is the basic unit of all living things. Each cell consists of a membrane
surrounding an aqueous solution through which the cell aquires nutrients and
removes waste products. Cells are small, of the order of 5–20 µm, and they can
be visualized with the aid of a microscope. The various types of microscopes
that are used to visualize cells are described by Alberts et al. (1998), and it is the
invention of the microscope in the seventeenth century that led to the discovery
of the cell.
By the nineteenth century, with the aid of research conducted by a number of
biologists, a cell doctrine (Alberts et al., 1998) maintained that

1. All living things are made up of cells and the products formed from cells.
2. Cells are units of structure and function.
3. All cells arise from preexisting cells.

There are two types of cells: eukaryotic cells, which have distinct compartments
and are found in the higher mammals, and prokaryotic cells, such as bacteria,
which have no such compartments (Figure 8.9). Eukaryotic human cells are the
subject of this section. It is important to note that all cells are living entities and
that viruses are not cells because they cannot replicate.
Cells are composed of an outer flexible cell membrane made of lipids called
a lipid bilayer. Most of the cell membrane is impermeable; however, it does
contain the ion channels that are used to transport material into or out of the
cell, depending on its function. This membrane allows the cell to maintain the
appropriate level of pH, electrolytes, and other chemicals. Cell death most often
occurs when this membrane is ruptured.
The cytoskeleton allows the cell to change its shape and move. It consists of a
series of rodlike filaments of various diameters (7–25 nm) that anchor the cell to
its surroundings. These filaments extend throughout the cell, forming a gridlike
299 Elements of Molecular and Cell Biology

Table 8.1. Composition of a typical eukaryotic cell.

Types of
Molecule % Weight each molecule
Water 70.0 1
Inorganic ions 1.0 20
Sugars and precursors 1.0 250
Amino acids and precursors 0.4 100
Nucleotides and precursors 0.4 100
Fatty acids and precursors 1.0 50
Other small molecules 0.2 300
Macromolecules 26.0 3000

From Alberts et al. (1998). Macromolecules include proteins, nucleic acids, and
polysacchrides.

network. The cytoskeleton is made up of microtubules, intermediate filaments,

and actin. The intermediate filaments, in particular, are strong and durable and
allow the cell to withstand mechanical stress.
Inside the cell, there are specific regions that have specific functions required
for the operation of a healthy cell. The mitochondria are porous membranes within
the cell, in which most of the cell’s energy is produced, often from sugars such as
glucose. The so-called outer membrane contains a large number of ion channels,
some called transport proteins, that allow large molecules into the mitochondria.
The nucleus is the “command” center (Ethier & Simmons, 2007) and contains
the genetic material for molecules that are synthesized by the cell. Proteins and
other molecules are synthesized initially by the copying of genetic information
coded in DNA into messenger (mRNA), which then leaves the nucleus and
enters the endoplasmic reticulum, where the proteins and molecules are actually
synthesized.
Once the molecules are synthesized in the endoplasmic reticulum, they are
moved to the Golgi apparatus, where the newly synthesized molecules undergo a
series of modifications before they can be biologically active. Any postsynthesis
mistakes are automatically discarded in the lysozomes.
The cytosol (the light region in Figure 8.9) is what surrounds the structures just
described. The cytosol consists of a water-based gel in which a number of key bio-
chemical reactions take place. The cytosol is the origin of the manufacturing pro-
cess of proteins, which are synthesized by molecular machines called ribosomes.
Most of a cell is water, and other than water, most of the other constituents of
a cell are carbon based. Carbon has four electrons in its outer shell so that carbon
can form covalent bonds with four other atoms. This means that carbon is ideal
for forming large macromolecules. The carbon compounds that compose the cell,
and indeed all molecules that contain carbon, are called organic molecules.
The main constituents of a cell are depicted in Table 8.1. From the table it is seen
that a cell contains four major classes of organic molecules: sugars, nucleotides,
 300 Essentials of Micro- and Nanofluidics

fatty acids, and amino acids. The sugars, especially glucose, act as an energy
source for the cell. Nucleotides act as carriers of chemical energy, an important
example of which is adenosine triphosphate (ATP). A fatty acid is, in general,
composed of a long hydrophobic hydrocarbon chain that is not very reactive. At
the other end is a hydrophilic carboxyl group that is chemically reactive. Fatty
acids act as a food reserve and can produce much more energy than glucose. A
fatty acid belongs to a class of materials called lipids.
Cells communicate with each other through a variety of molecules such as pro-
teins, nucleotides, fatty acids, and several others. Those molecules used to com-
municate with other cells are called hormones. Insulin, cortisone, and estrogen
are examples of hormones. Insulin regulates the blood sugar level, for example,
and estrogen regulates reproductive function.
The destination of the message sent by one cell to another is called the receptor.
Large hydrophilic signaling molecules cannot cross the cell plasma membrane
and so must reside there. Conversely, small hydrophobic molecules, such as
steroid hormones, can cross the plasma membrane into the interior of the target
cell. Perhaps the best known of this class of signaling molecules is testosterone.
A steroid is any one of a class of organic compounds that has a 17 carbon nucleus
(Alberts et al., 1998); cholesterol is a steroid.

8.6 The cell membrane

Each cell is surrounded by a thin, fatty film called the plasma membrane that is
on the order of 10 nm thick. This membrane prevents the contents of the cell from
leaking out but also provides the pathway for the cell to receive the necessary
nutrients to grow and for the cell to rid itself of waste. Thus the cell membrane
contains a network of ion-selective channels and pumps, some of which are made
from selected proteins. As the cell grows, the membrane grows as well; thus the
cell membrane is a thin, deformable shell and behaves much like a fluid in that it
deforms under stress and, if torn, it heals quickly.
The plasma membrane has a two-layer structure comprising lipid molecules
that have a hydrophilic head and a hydrophobic tail. The lipid molecules are
oriented with their hydrophilic heads in contact with the aqueous environment
outside the cell, a configuration that is energetically most favorable (Alberts
et al., 1998). The cell membrane is depicted in Figure 8.10. The lipid bilayer is
about 5 nm thick, and embedded in the lipid bilayer are the membrane protein
channels that facilitate transport to and from the cell. These membrane proteins
are usually negatively charged, and some of these channels make up what are
called ion channels, which regulate the concentration of crucial ions, such as
sodium, chloride, calcium, and magnesium, to name a few, inside and outside the
cell.
Because of the chemical structure of the lipid bilayer, small, nonpolar
molecules, such as gaseous oxygen and carbon dioxide, diffuse through readily.
Small, uncharged polar molecules, such as water, can also pass through the lipid
 301 Elements of Molecular and Cell Biology

Sketch of the structure of a plasma membrane surrounding a cell, including the lipid bilayer. The ion channels
Figure 8.10
are shown as the protein structures they are. Image courtesy of the National Institute of General Medical
Sciences of the National Institutes of Health.

bilayer rather easily, but the larger the molecule, the more difficult transport
becomes. Lipid bilayers prevent passage of ions and other charged molecules.
Because these molecules are attracted to water, they do not want to enter the fatty
or hydrocarbon region of the lipid bilayer.
The transport of nonpolar and polar molecules through the lipid bilayer is
solely by diffusion. However, for cells to grow, that is, to absorb nutrients such
as sugars, ions, and amino acids, among others, and for the cell to release waste
material, much faster transport is necessary. The cell provides this faster transport
through the use of ion channels; these channels are described in detail in the next
section.

8.7 Membrane transport and ion channels

Ion channels play a crucial role in the transport of biofluids to and from cells. To
keep the cells functioning properly, there needs to be a continuous flux of ions in
and out of the cell and the cell components. In this section, a qualitative picture
of ionchannels is described, in keeping with the dominant theme of this chapter.
Approximate concentrations of selected ions inside and outside a cell are given
in Table 8.2.
An ion channel is a narrow, water-filled tube, permeable to the few ions and
molecules small enough to fit through the tube (approximately 10 Å in diameter);
their walls are made of proteins, most of which are negatively charged. They have
wide entrances and a cross section that varies in the transport, or streamwise,
direction (Figure 1.23). Ion channels are responsible for electrical signaling in
302 Essentials of Micro- and Nanofluidics

Table 8.2. Concentrations of ions inside and outside the cell

Ion Concentration Concentration
Inside cell (mM) Outside cell (mM)
+
Na 5–15 145
+
K 140 5
Mg2+ 0.5 1–2
Ca2+ 10−7 1–2
H+ 7 × 10−5 4 × 10−5
Cl− 5–15 110
Fixed anions high 0

The fixed anions are those molecules that cannot leave the cell. The divalent
cation concentrations are for the free ions in the cytosol; large parts of both
Ca 2+ and Mg 2+ are bound to proteins and other molecules. From Alberts
et al. (1998).

nerves and muscles.3 The ions responsible for the majority of nervous system
signaling are Ca2+ , Cl− , K+ , and Na+ , and these four ions are transported by
most ion channels. Ion channels are responsive to different stimuli: a membrane
potential change or a ligand, a molecule that attaches itself to the channel, causing
it to open.
Ion channels respond to the stimuli in a process called gating, the opening
or closing of the pore. The open pore has a selective permeability to ions, only
allowing certain ions to flow through their electrochemical potential gradients.
The ions flow at a very high rate, greater than 106 ions per second, with a single
ion taking on the order of 1 ms to traverse the channel. This rate of transport is
much greater than that of a carrier protein, a protein to which an ion is bound,
which performs a similar function – see Table 8.3 (Alberts et al., 1998). This rush
of ions through the channel changes the electrical potential across the membrane,
the transmembrane potential, which in turn changes the electrical body force
driving the ions through the channel. Thus the operation of an ion channel is
much like the electrokinetic phenomenon of electro-osmotic flow, which will be
discussed in Chapter 9.
As mentioned earlier, the two main characteristics of ion channels are that
they are selective and gated. Selectivity depends on the relative size of the ion
channel and the ion that it is supposed to pass. Large ions will not move through
small channels owing to both steric (size) and electrostatic effects. As has been
mentioned, in general, ions are hydrated by water molecules, which makes the
ions appear bigger than they really are. For the highly selective ion channels, it
is generally assumed that these hydrated water molecules must be released from
the ions so that they pass through the ion channel in single file.

3
In this text, we have used the term channel for a passage that is rectangular in cross section. Ion
channels are approximately circular; nevertheless, we use the accepted term ion channel rather
than the clumsy ion tube.
303 Elements of Molecular and Cell Biology

Table 8.3. Gating mechanism and location of the four most common ion
channels
Channel Gate Location
+
Na voltage plasma membrane nerve cell axon
+
K voltage plasma membrane nerve cell axon
2+
Ca voltage plasma membrane nerve cell axon
−
Cl ligand plasma membrane neurons

From Alberts et al. (1998).

Ion channels are gated, as mentioned earlier. Voltage gated channels are gen-
erally thought to operate based on a potential drop across the channel, the trans-
membrane potential, allowing the ion concentrations upstream and downstream,
according to the Donnan potential discussed in the previous chapter. Thus ion
channels open and close when needed, and it is generally thought that the open-
ing or closing of the channel is random. The transmembrane potential affects the
probability of an ion channel being open or closed. According to the Donnan
potential, the transmembrane potential is related to the concentrations inside and
outside the cell by
RT Co
φ = ln (8.9)
zF Ci
where Co and Ci are the concentrations outside and inside the cell and z is the ion
valence. Given the concentrations in Table 8.2, the appropriate transmembrane
potential can be determined. From the Nernst equation (8.9), the transmembrane
potential for ion channels is of the order of φ = RT F
. Note that the Donnan
potential assumes electrical neutrality everywhere in the system.
The state of an ion channel, that is, determining whether it is open or closed,
is determined by measuring the electrical current through the channel. Today,
this can be done using what is called a patch clamp device, described in detail
by Alberts et al. (1998) and in many other places. The patch clamp itself is
just a glass tube that is attached to the cell enclosing the target ion channel. A
constant voltage source is passed through the solution that houses the cell, with
the electrical circuit closed via the insertion of a metal wire into the solution in
the glass tube.
The transmembrane potential can be held at a desired value (“clamped”) with
the current measured at this given transmembrane potential. The transmembrane
potential can be varied to assess its influence on the gating process. Thus the
current–voltage relationship is a useful measure of the behavior of ion channels.
Generally, the results of the patch clamp recording is a time trace of the current,
on the order of p A, showing sudden changes in the current as the channel opens
or closes.
Much more detail about ion channels is available in a number of texts, including
Alberts et al. (1994), Alberts et al. (1998), Friedman (2008), and Hille (2001).
It should also be mentioned that ion channels can be modeled as perm-selective
channels (Section 7.18), and the analysis is much the same as in that section (see
 304 Essentials of Micro- and Nanofluidics

Gillespie, 1999; Gillespie & Eisenberg, 2001; Hollerbach et al., 2001; Barcilon
et al., 1997; Barcilon et al., 1992). In most cases, ion channels do not seem to
operate near the limiting current.

8.8 Chapter summary

The objective of this chapter is to introduce the reader to the fundamentals of

molecular and cell biology that are important for understanding biomolecular
transport. Many nanoscale devices are designed to transport proteins, DNA,
and polysaccharides. Thus knowing the basic structure and properties of these
molecules will aid in understanding how these devices should be designed.
In particular, it has been seen that the two most important characteristics of
a protein are its structure and function. Because of the inherent complexity of
proteins, biologists have defined four levels of structure for analysis, beginning
with the primary structure, the amino acid sequence specific to the protein.
Determining a protein’s conformation is an important exercise, and this is often
done either by experiment or increasingly by molecular simulation.
Different proteins perform different tasks, and in general, there are seven
distinct types of proteins, though authors may differ on the terminology used.
Each of these seven types of proteins has been discussed in this chapter, and
examples have been presented for each type of protein.
One of the consequences of a protein’s structure and function is its ability to
bind to other molecules. The nature of protein binding is discussed in this chapter,
specifically antigen–antibody binding, which is so important in fighting disease.
Protein binding is viewed as a biochemical reaction similar to those reactions
discussed in the previous chapter.
The last three sections discuss the processes that occur at the cellular and
subcellular levels. The living cell is the foundation of life itself and comprises
mainly water and other macromolecules. Ions and other small molecules pass
into and out of cells through ion channels specific to the passage of a given ion.
The walls of an ion channel consist of proteins, most often negatively charged,
and so electrostatic and electrodynamic interactions are extremely important in
their operation.
Many of the principles presented in this chapter will be revisited when biomed-
ical applications are considered in Chapter 12.

EXERCISES
About these exercises: Several of these problems can be used as writing exercises
and some are meant to be open-ended.

8.1 Given that a sodium atom has one electron in its outer shell, and a chlorine
atom has seven, sketch the the ionic bond that is formed between Na+ and
Cl− .
305 Elements of Molecular and Cell Biology

8.2 Discuss qualitatively how the conformation of a protein may affect its
transport in a synthetic nanopore membrane. Identify the main dimensional
parameters that govern its transport.
8.3 Transcription is the process by which a cell uses DNA to make RNA.
Discuss the role of transcription within the context of the entire process of
the synthesis of proteins.
8.4 Discuss in detail how a protein is synthesized.
8.5 Plot the yield of a complex [AB] given by the solution of equation (8.6),
either in concentration or number of molecules, as a function of the equi-
librium constant K .
8.6 Solve the previous problem in dimensionless form, that is, nondimension-
alize equation (8.6), and show that each term in the equation is of the same
order of magnitude.
8.7 Choosing a single common protein, such as albumin, using the Web or
another appropriate source, map the charge on the surface, and calculate
the total surface charge. What does the valence of a protein mean?
8.8 A globular protein of diameter 3 nm that is negatively charged approaches
a nanopore membrane having 10 nm pores. The membrane is negatively
charged. Discuss the ability of the globular protein to pass through the
membrane.
8.9 An ion channel is gated, that is, it opens and closes. Suggest possibilities
for the time scale of the opening and closing procedure. What is the time
scale during the period that the ion channel is opened? For the calcium
channel, use the Web or another appropriate source to determine the flow
rate of calcium through the channel.
8.10 Cystic fibrosis is a disease caused by a malfunctioning of ion channels.
Discuss the specifics of what goes wrong and the current treatment, if any,
for the disease.
8.11 The term lipid raft has recently been used to describe the structure of the
cell membrane. Precisely what is a lipid raft?
 9 Electrokinetic Phenomena

9.1 Introduction

It was shown in Chapter 2 that as the length scale of microchannels approaches

the nanoscale, pressure-driven flow becomes increasingly difficult to achieve.
The precise cutoff depends on the desired flow rate, but for a volume flow rate of
Q = 1 µL/min, electro-osmotic flow (EOF) becomes significantly more efficient
around an individual channel height of h = 40 nm. Using the material studied in
Chapters 4–8, we can now investigate flows of ionic and biomolecular species in
micro- and nanochannels.
Electrokinetic phenomena are generally grouped into four classes:

1. Electro-osmosis (EOF): the bulk motion of a fluid caused by an electric

field
2. Electrophoresis: the motion of a charged particle in an otherwise motionless
fluid or the motion of a charged particle relative to a bulk motion
3. Streaming potential or streaming current: the potential induced by a pres-
sure gradient at zero current flow of an electrolyte mixture
4. Sedimentation potential: the electric field induced when charged particles
move relative to a liquid under a gravitational or centrifugal or other force
field

Note that electrophoresis and sedimentation potential refer to particles and

electro-osmosis and streaming potential refer to electrolyte solutions. The first
three of these are by far the most important for our applications, and in this chap-
ter, we discuss the phenomenon of electro-osmosis first, followed by a discussion
of electrophoresis and streaming potential.
In this chapter, electro-osmosis and electrophoresis will receive the most atten-
tion. It is important to note that for channels with charged walls, the electrical
double layer (EDL) will be present independent of whether there is electro-
osmotic flow. Elements of electrochemistry will permeate through this chapter,
along with the discussion of electrokinetic phenomena. This is especially true
of the asymptotic analysis of mixtures of arbitrary valence that could have been
discussed in Chapter 7. In the end, I thought that that section goes better in this
chapter.
306
 307 Electrokinetic Phenomena

Reservoirs with solute

Channels w/−
charged walls

+ −
x

Nanopore
Donor Membrane
Region

Receiver
Region

(a) (b)

(a) Geometry of a typical nanopore membrane, showing electrodes in the donor and receiver regions (not to
Figure 9.1
scale). The walls in the channels are usually negatively charged, but positively charged walls have also been
used. (b) Side view of a typical channel in the nanopore membrane, showing a sketch of the EDL. The channel
is usually long and wide, as in Figure 1.4.

We first discuss several aspects of electro-osmotic flow in a wide channel,

the most common geometry in the field of micro- and nanofluidics, and show
that asymptotic solutions may be obtained for species of arbitrary valence in a
binary solution. We discuss walls having dissimilar wall charge or ζ potential
and follow that with a study of species transport and the channel current and
current density. These sections are followed by analysis of electro-osmosis in
several canonical geometries as was done for viscous flow in Chapter 4; these
canonical flows include the annulus and nozzles and diffusers; nozzles are often
used in electroporation, the opening of a cell for injection of DNA or genes (Chen
& Conlisk, 2008; Wang et al., 2008). The section on electro-osmosis concludes
with a discussion of dispersion, the spreading of a solute in channel flows.
Next electrophoresis, here defined as the relative motion of particles in a fluid,
is described; the models range from a simple force balance for thin and thick
EDLs, yielding a simple algebraic expression for the electrophoretic velocity,
to a full numercial solution valid for the entire range of EDL thicknesses. The
streaming potential, the potential that is induced by an electrically conducting
fluid in a pressure-driven flow, is presented next. The streaming potential is often
used to measure the average ζ potential in a channel, and it is shown how this
can be done. The chapter concludes with a brief description of the sedimentation
potential and Joule heating.

9.2 Electro-osmosis
9.2.1 The relationship between velocity and potential
A biomedical application device, described previously, is illustrated in Fig-
ure 9.1a. The electro-osmotic flow is driven by electrodes placed in upstream
308 Essentials of Micro- and Nanofluidics

and downstream reservoirs. This geometry may be useful as a drug delivery vehi-
cle. Fluid flows from the upstream reservoir into the mouth of the channel(s) and
very quickly reaches a fully developed condition. The electrical potential is com-
posed of two parts corresponding to the presence of the EDL and the externally
imposed potential

ψ = φ + φi (9.1)

where the subscript i denotes the imposed field. Here we assume that the imposed
field is characterized by a constant electric field E 0 . In this section, we will
work with dimensionless variables almost exclusively. The interested reader is
encouraged to determine the appropriate length and velocity scales that are used
to obtain the dimensionless equations and solutions for the potential and velocity.
The dimensionless form of the streamwise momentum equation in the fully
developed flow region is thus given by (Chapter 3)
∂ 2u ∂p 
2 = 2 −β zi X i (9.2)
∂y 2 ∂x i

and the Poisson equation for the potential is

∂ 2φ 
2 = −β zi X i (9.3)
∂ y2 i

where the partial derivatives in this one-dimensional fully developed analysis are
really total derivatives  = λ/ h and β = c/I , as in Chapter 3. Here X i is the
mole fraction, but as mentioned previously, we could scale the concentrations on
the ionic strength, in which case, β = 1. Equating the right-hand side of equation
(9.3) in equation (9.2) gives
∂ 2u ∂ p ∂ 2φ
= + 2 (9.4)
∂ y2 ∂x ∂y
Note that the pressure is scaled on the dimensional electro-osmotic velocity
U0 = e φ0 E 0 /μ, where E 0 is the imposed electric field in the streamwise direction
and the dimensional potential is scaled on φ0 = RT /F. This equation may be
integrated twice, and for a potential specified at each wall, y = 0, 1,
1 dp 2
u(y) = φ(y) + (y − y) + (ζ0 − ζ1 )y − ζ0 (9.5)
2 dx
where here ζ0 and ζ1 are the dimensionless (scaled on RT /F) ζ potentials at each
wall y = 0, 1. Note that if both walls have the same ζ potential and there is no
axial pressure gradient, then u(y) = φ(y) − ζ0 = φ(y) − ζ , and thus the velocity
is equivalent to the potential perturbation from the ζ potential, as discussed in
Chapter 3. The potential φ = φ(y) is the dimensionless potential due only to
the presence of the EDLs. Thus, if the walls of the channel are not charged,
there is no EDL, and even with the presence of the electrodes, there is no bulk
electro-osmotic flow.
 309 Electrokinetic Phenomena

1.6
4.5 ε=0.01
1.4 ε=0.05
4 ε=0.1

0
Rescaled Mole Fractions Xi/Xi
1.2 ε=0.5
3.5

Potential and Velocity

1 3

0.8 2.5

2
0.6
1.5
0.4 ε=0.01
ε=0.05 1
0.2 ε=0.1
ε=0.5 0.5
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
y y

(a) potential and velocity (b) mole fractions

Potential, velocity, and rescaled mole fractions for a 1:1 electrolyte for various values of . Here the dimensional
Figure 9.2
potential on both walls is ζ ∗ = −40 mV. In (b), the mole fractions are rescaled based on the upstream reservoir
mole fractions as X i /X 0i (note that the values are O (1) – always a good thing). The cations are plotted in
black lines, and the anions are plotted in gray lines.

Any of the potential distributions described in Chapter 7 may be used to

describe the velocity. For a pair of monovalent ions in channels for which the
double layers are overlapping, under the Debye–Hückel approximation,
y−1/2
cosh 
φ(y) = ζ (9.6)
cosh 21
In the general nonlinear case, the solution must be calculated numerically. For a
thin double layer  → 0, the cosh solution reduces to
y
φ = ζ e−  (9.7)

as has been indicated before. Note that in this case, as y → ∞, u → −ζ .

For the case of a thin double layer,  = λ/ h  1, recall that the solution to the
nonlinear equation for the potential for a pair of symmetric electrolytes of valence
z which satisfies φ = ζ at y = 0 and φ → 0 as y → ∞ (the dimensionless y
coordinate is scaled on ) is
 
4 −1 −z y zζ
φ = tanh e tanh (9.8)
z 4
The electro-osmotic part of the velocity field near the wall at y = 0 is given by
 
4 y zζ
u(y) = φ(y) − ζ = tanh−1 e−z  tanh −ζ (9.9)
z 4
where ζ is the dimensionless ζ potential at y = 0. For negatively charged walls,
the ζ potential will be negative, and thus the velocity in the outer region will
be positive. This means that the cathode is located downstream and the anode is
located upstream. A similar expression for the velocity holds near y = 1.
Figure 9.2 depicts the solution for the velocity and potential and mole fractions
for several values of  in the absence of a pressure gradient. Note that both
overlapped and thin EDL cases are shown. As  → 0, the core of the channel
becomes electrically neutral, whereas for  = 0.5, it is not. Moreover, for a thin
310 Essentials of Micro- and Nanofluidics

EDL, the velocity is constant, which is advantageous for reducing dispersion.

Note that the mole fractions are rescaled on the reservoir mole fraction for each
species X i0 , which makes the value of the rescaled concentration O(1) – a good
thing.
It is useful to discuss the scaling of a typical concentration. Equations (9.2)
and (9.3) contain the dimensionless parameter β = c/I , where c is the total
concentration, including the solvent, and I is the reference ionic strength. This
is a direct result of using mole fractions as the dependent variable, a scaling that
is typical in engineering and chemistry. Because the total concentration includes
the solvent, the mole fractions of the electrolytes are usually small, with the
combined parameter β X i0 = ci /I = O(1), where X i0 is the cation mole fraction
in the upstream reservoir.
The dimensionless volume flow rate through the channel is given by
1
Q= udy (9.10)
0

where here y = 0 at the bottom wall of the channel and y = 1 at the top wall of
the channel. For the cosh potential profile, the overlapped double layer situation,
the dimensionless flow rate is given by
 
1 1 dp
Q = ζ  tanh −1 − (9.11)
2 12 d x
Even though, strictly speaking,  = O(1) if the double layers overlap, one can
take the limit  → 0, as we have seen in equation (9.6), to show that
1 dp
Q = −ζ − (9.12)
12 d x
The effect of a pressure gradient is depicted in Figure 9.3. For a negatively charged
wall, a favorable pressure gradient d p/d x < 0 increases the flow rate, while an
adverse pressure gradient d p/d x > 0 reduces the flow rate. Note that eventually,
the flow will reverse beginning in the center of the channel, where the Poiseuille
velocity is a maximum.
Because the boundary conditions for the cosh potential distribution are constant
potential at the walls, the surface charge density must be calculated from the
potential solution. In dimensionless form,
∂φ 1
σ =− = ζ tanh (9.13)
∂y 2
at y = 0 with a similar expression at y = 1. Here the dimensionless surface
charge density is scaled on e φ0 / h.
Notation alert: Here come the asterisks again! We need to use Q ∗ to distinguish
it from its dimensionless counterpart Q.
Looking only at the electro-osmotic part of the flow, the electro-osmotic pump-
ing, for thin EDLs, the velocity in the core of the channel far from the wall(s) is
 311 Electrokinetic Phenomena

0.9 dpdx=−1
dpdx=0
0.8 dpdx=1
dpdx=2
0.7

y 0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.1 0.2 0.3 0.4 0.5
u

Dimensionless velocity distribution for both electro-osmotic and pressue gradient effects. Here the electric
Figure 9.3
field corresponds to 0.06 V over a channel of length L = 3.5 µm; the channel height h = 20 nm and the
electrolyte solution concentration is 0.1 M in the reservoir. The ζ potential is −10 mV on both walls. The
dimensionless pressure gradient ∂ p/∂ x is −1, 0, 1, and 2, respectively.

constant, and U∞ = −ζ . In dimensional form, this equation is given by

∗ e φ 0 E x ζ ∗ e E x ζ ∗
U∞ =− =− (9.14)
μ φ0 μ
where ζ ∗ = ζ RT /F is the dimensional ζ potential; this is the Smolukowski
formula for the velocity in the bulk of the channel. The dimensional flow rate is
thus given by
e E x ζ ∗
Q ∗ = −ρ A (9.15)
μ
where A is the cross-sectional area. The electro-osmotic mobility is defined by
∗
U∞ −ζ ∗ e
μeo = = (9.16)
Ex μ
For constant potential boundary conditions, the velocity and the potential
perturbation from the ζ potential are equivalent; as in classical boundary layer
theory, the width of the EDL can be defined as that location where the velocity
and potential reach 99 percent of their free stream value. In the Debye–Hückel
approximation for   1, the dimensionless width is given by δe = 4.61, where
δe is nondimensional, scaled on the channel height. Thus the dimensional width of
the EDL is given by δe∗ = 4.61λ, a result that was discussed in Chapter 7. A fully
numerical result in this case also results in δe∗ = 4.6λ. Note that the Debye length
is not a good measure of the asymptotic extent of the EDL and that the constant
4.6 is similar to the result obtained from the Blasius boundary layer in fluid
mechanics, for which the boundary layer thickness is given by δ B L = 5Re−1/2
for large Reynolds number.
 312 Essentials of Micro- and Nanofluidics

−0.2 2

1.8
−0.4
1.6

Dimensionless Potential
−0.6

Dimensionless Velocity
−0.8 1.4

−1 1.2

−1.2 1

−1.4 0.8

−1.6 0.6

−1.8 0.4
Poisson−Boltzmann 0.2 Poisson−Boltzmann
−2
Debye−Huckel Debye−Huckel
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x
(a) (b)

Comparison of the results for the dimensionless potential and velocity calculated from the Poisson–Boltzmann
Figure 9.4
equations and the Debye–Hückel approximation, using the fixed surface charge boundary condition. The
height of the channel is h = 20 nm, and the electrolyte solution is 0.1 M Na+ and 0.1 M Cl− . (a) Dimensionless
potential for a fixed surface charge density σ = −0.04 C/m2 . The corresponding dimensionless wall potential
is ζ = −1.86 (−48 mV in dimensional form) based on the Poisson–Boltzmann equations and −2.15 (−55.6 mV
in dimensional form) based on the Debye–Hückel approximation. (b) Dimensionless velocity.

9.2.2 The Debye–Hückel approximation reviewed

The Debye–Hückel (DH) approximation is a powerful tool enabling an analytical
solution not only for the potential but for the velocity as well. It is thus of
interest to inquire as to what the error is in making this approximation; thus we
compare the full nonlinear solution for the potential and velocity with the DH
approximation. In Figure 9.4 are results comparing the nonlinear solution for a 1:1
electrolyte with the DH approximation, using the fixed surface charge condition.
The calculations from Figure 9.4 assumes concentrations of c A0 = 0.1 M and
c B0 = 0.1 M in the bulk. Note in Figure 9.4a, that the ζ potential on the wall is
different for the DH approximation. This results in a DH approximation for the
velocity that is larger than the full nonlinear solution.
The results presented on Figure 9.4 indicate that for a given surface charge
density, the Debye Huckel approximation overestimates the magnitude of the zeta
potential, and hence the flow rate in a channel with thin EDLs. This is because the
dimensionless Poisson-Boltzmann (PB) estimate ζ P B = 2 sinh−1 σ/2 is always
smaller than that of the Debye–Hückel estimate ζ D H A = σ for any given value
of (positive/negative) σ (see for example, Figure 9.5). On the other hand, if the
constant potential boundary condition is used, the DH approximation will under-
predict the PB result since the approximation is only valid for small potential.

9.2.3 Another similarity revealed

Using the similarity of the velocity and potential, the skin friction coefficient, c f ,
can be related to the surface charge density. At low Reynolds number, the skin
friction can be defined in the same way as for high Reynolds numbers, and since
u = φ − ζ,
2 ∂u 2 ∂φ
cf = = (9.17)
Re ∂ y Re ∂ y
 313 Electrokinetic Phenomena

−10

−20

−30

φ(mV) −40

−50

−60

−70

−80

−90 Poisson−Boltzmann
Debye−Huckel
−0.07 −0.06 −0.05 −0.04 −0.03 −0.02 −0.01 0
σ(C/m2)

Comparison of ζ potential for different surface charge density calculated from Poisson–Boltzmann and Debye–
Figure 9.5
Hückel approximations. The concentration in the bulk is 0.1 M Na+ and 0.1 M Cl− .

Now, in this dimensionless form, if the dimensional surface charge density is

scaled on e φ0 /λ, then

∂φ
σ = − (9.18)
∂y

if the outward unit normal is in the negative coordinate direction, for example, at
y = 0. Thus, in terms of surface charge density,

2σ 2ζ
cf = − =− (9.19)
Re Re
for a negatively charged wall. Note that the first equality holds independent of
the DH approximation and that the second equality holds in the limit of a thin
double layer. Both relationships suggest that experimental measurements of the
shear stress can lead to a significant amount of information about the ζ potential
and surface charge density.
It is not the convention to do so, but for very small Reynolds numbers,
conceptually, it is more reasonable to scale the shear stress on the viscous
scale μU0 / h, where U0 is the electro-osmotic velocity scale, and in this case,
c f = ∂u/∂ y = ∂φ/∂ y.

9.2.4 Asymptotic solution for binary electrolytes of arbitrary valence

By far the most common situation in microchannel flows is the case where the
EDL is thin, as depicted in Figure 9.6. In Figure 9.2 it is seen that for this case,
the velocity is a constant in the core of the channel u = −ζ for a pair of monova-
lent ionic species; in this section, we show how this constant can be obtained
directly by the method of matched asymptotic expansions without the need
for an analytical solution for the potential and for arbitrary values of the valence
of the species.
 314 Essentials of Micro- and Nanofluidics

If the EDL is thin, the other channel wall is not visible. The flow direction is left to right.
Figure 9.6

We use here terminology that is introduced in Appendix A. For   1 and thus

as  → 0, the highest order differentiated term in the Poisson equation is small,
except in the region immediately next to the wall. This means that away from
the wall, the boundary condition at the wall cannot be satisfied. This is what is
known as a singular perturbation problem (VanDyke, 1975). For many problems
in fluid mechanics and heat and mass transfer, these methods can greatly simplify
the solution process. Of course, the Blasius boundary layer problem of fluid
mechanics is an example of a singular perturbation problem.
Consider the case of two species of arbitrary valence, with the cation (we
assume a negatively charged wall, although this assumption is not necessary)
X A = g and the anion X B = f . It is easier to start in the region near the walls. For
example, near y = 0, as we have done before, let Y = y/, and this transformation
blows up the EDL so that we can see inside it. Then, because of the no-flux
condition at the walls,
∂g ∂φ
+ zg g =0 (9.20)
∂Y ∂Y
This is simply the one dimensional Boltzmann equation, which has the well-
known solution

g = g R e−z g φ (9.21)

where g R is the cation mole fraction in the upstream reservoir, where the potential
due to the EDLs on the sidewalls of the membrane is taken to be zero. The solution
for f follows in the same way:

f = f R e−z f φ (9.22)

For thin EDLs, the quantities f R and g R are equivalent to the mole fractions in
the bulk.
In the method of matched asymptotic expansions, this inner solution for the
mole fractions of the species (if that is taken as the scale) must be matched with
an outer solution valid away from the wall.1 To obtain the matching conditions,

1
The reader may want to refer to Appendix A for some simple examples of the appropriate
matching procedure.
315 Electrokinetic Phenomena

the outer solution for the electrolyte of positive valence, go (o for “outer”), must
be

go = lim g = g R e−z g φo (9.23)

Y →∞

Similarly,

f o = lim f = f R e−z f φo (9.24)

Y →∞

Since the core region must be electrically neutral with the volume charge density
being zero,
zg
fo = − go (9.25)
zf
Thus, from the limit of the inner solution,

z g g R e−z g φo = −z f f R e−z f φo , (9.26)

and solving for the outer solution φo ,

1 −z g g R
φo = ln (9.27)
zg − z f z f fR
and u o = φo − ζ . Note that for z g = −z f , φo = 0; that is, the potential has the
same value as in the reservoir. The anion mole fraction in the core of the channel
is thus

zg
f o = − g R f R e−(z g +z f )φo (9.28)
zf

with a similar solution for go . For a (1, −1) electrolyte,

f o = go = g R (9.29)

Evaluating the Boltzmann distributions at the wall and dividing the two equa-
tions, the dimensionless ζ potential is given by
1 gw
ζ = ln (9.30)
z f − zg fw
so that the dimensionless flow rate u o = Q = −ζ , as in the previous section. The
dimensional volume flow rate is thus given by
 
e φ0 E x 1 gw
Q∗ = ρ A ln (9.31)
μ zg − z f fw
where A is the cross-sectional area of the channel and φ0 = RT /F. For a nega-
tively charged wall, gw > f w .
Note that since the EDL is so thin, the flow rate to leading order does not
require the solution in the inner region near the wall. This is a powerful result
that is independent of the potential distribution within the EDL.
316 Essentials of Micro- and Nanofluidics

9.2.5 Walls with different ζ potentials

Fabrication procedures for microchannels often involve injection molding
(Devasenathipathy et al., 1998), laser ablation (Bianchi et al., 1993), and etch-
ing of an open groove on a substrate followed by sealing of the groove by a
coverslide of another material. For example, Bianchi et al. (1993) use a molded
Polydimethylsiloxane (PDMS) channel sealed by a cover made of glass in their
experiments. In such situations, the ζ potential on the sealing wall (say, top)
differs substantially from that on the side and bottom walls.
When the ζ potentials of the bottom and top wall are not the same, the axial
electro-osmotic velocity varies linearly across the channel and the familiar plug
profile of EOF no longer occurs. One serious consequence of this nonuniform
distribution is increased Taylor dispersion of samples that reduce the efficiency of
microfluidic separations such as capillary electrophoresis. However, walls having
different ζ potentials are used in what is termed an asymmetric clamping cell to
calculate the average ζ potential of the two walls. The procedure is described in
Masliyah and Bhattacharjee (2006).
Consider the case of monovalent electrolytes. For the case where the walls have
differing ζ potentials and hence differing wall mole fractions, the dimensionless
velocity in the outer region satisfies

u o = Ay + B (9.32)

where E and F are constants. As in the previous section, φo = 0. Because of the

no-flux condition, the outer solution for the mole fractions is a constant

f o = go = f R = g R (9.33)

In the inner region near y = 1, for example,

u i = φi + D (9.34)

and near y = 0, we have

u i = φi + E (9.35)

Clearly to satisfy the no-slip condition, D = −ζ1 and E = −ζ0 . The constants A
and B are found by the matching condition

lim u i = lim u o (9.36)

Y →∞ y→0

where again Y = y/ and thus B = −ζ0 and A = ζ0 − ζ1 . This procedure can
easily be extended to multivalent species.
Integrating the solution for the velocity in the outer region, the dimensionless
flow rate is given by
1
Q = − (ζ0 + ζ1 ) + O() (9.37)
2
and the flow rate is proportional to the average ζ potential.
 317 Electrokinetic Phenomena

−3
x 10
0.8
Cations
0.7 3.5 Anions

0.6
Potential and Velocity

Mole Fractions
0.5
2.5
0.4

0.3 2

0.2 potential 1.5

outer solution potential
0.1 velocity
outer solution velocity 1
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x
(a) (b)

Results for the dimensionless velocity and potential along with mole fractions for the asymmetric case of a
Figure 9.7
1:1 electrolyte mixture. Here the electric field corresponds to 6 v over a channel of length L = 3.5 µm; the
channel height h = 25 nm. (a) Velocity and potential. The outer solutions of the singular perturbation analysis
are also shown. (b) Mole fractions. Here the zeta potential is −20 mV at y = 0 and −10 mV at y = 1. The
Boltzman distribution is used for the mole fractions, and the concentrations of each electrolyte are 0.1 M in
the upstream reservoir. Here the potential is plotted as φ − ζ .

Two examples of walls having different ζ potentials are depicted in Figures 9.7
and 9.8, both for a 1:1 electrolyte. In Figure 9.7 ζ ∗ = −20 mV at y = 0 and ζ ∗ =
−10 mV at y = 1 so that at y = 0, ζ = −20/26 = 0.77, and at y = 1, ζ = 0.38.
Numerical solutions have been calculated for both the inner and outer regions,
and the outer solution for the velocity and potential obtained asymptotically is

−3
2.5 x 10
8 Cations
Anions
2
7
Potential and Velocity

1.5
6
Mole Fractions

1 5

0.5 4

3
0
potential
2
−0.5 outer solution potential
velocity
outer solution velocity 1
−1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
y x
(a) (b)

Results for the dimensionless velocity and potential (a) along with mole fractions (b) for the asymmetric case
Figure 9.8
of a 1:1 electrolyte mixture. Here the electric field corresponds to 0.06 V over a channel of length L = 3.5 µm;
the channel height h = 25 nm and the NaCl solution concentration is 0.1 M in the reservoir. The ζ potential
is −40 mV at y = 0 and 20 mV at y = 1, and the potential is plotted as φ − ζ .
 318 Essentials of Micro- and Nanofluidics

Ports/Vents

Electrode
Nano-channel
Array

Receiver/Donor chambers with

Electrolyte solutions

An electro-osmotic pumping system with a nanopore membrane; flow is from left to right, with the donor region
Figure 9.9
on the left and the receiver region on the right. Typically, each channel is a slit pore, and typical dimensions
in the direction of flow L ∼ 1 µm, spanwise W ∼ 40 µm, with the height h ∼ 10–50 nm.

indicated by the dashed lines in Figures 9.7a and 9.8a. The potential curve is
shifted up by ζ to better compare the potential and velocity. Note the linear
character of the velocity in the outer region, while the potential remains constant
in the core; as mentioned earlier, the variation of velocity will lead to increased
dispersion.
In Figure 9.9, the walls are actually of opposite-sign ζ potential leading to
reversed flow near y = 1. The dimensional flow rate using only the outer solution
is Q ∗ = 1.37 × 10−18 m3 /sec for the parameters of Figure 9.7 and Q ∗ = 2.76 ×
10−18 m3 /sec for the parameters of Figure 9.8.

9.2.6 Species velocities in electro-osmotic flow: Electromigration

Notation alert: For the most part, in this section, the equations will be in dimen-
sional form.

As has already been noted in Chapter 3, the flux of a given species in an

electrolyte solution is due in general to Fickian diffusion, electrical migration,
and bulk fluid motion. Thus most biomolecular transport studies begin with the
expression for the dimensional flux of species A in the form

N A = −cD A ∇ X A + cm A z A X A E + cX A V (9.38)

where D A is the diffusion coefficient, m A z A = z A (F D A /RT ) is often termed

the ionic mobility, F is Faraday’s constant, c is the total concentration, X A is the
mole fraction, u is the dimensional velocity, and E is the dimensional electric
field. Also, R is the universal gas constant, and T is temperature.
In particular, equation (9.38) in the streamwise, x direction of Figure 9.1 is
given by
∂XA
N Ax = −cD A + cm A z A X A E x + cX A u (9.39)
∂x
319 Electrokinetic Phenomena

The quantity N Ax is a flux and has units of mole/m2 sec, and the quantity
N Ax
= uA (9.40)
c
where c is the total concentration, including solvent, is thus the velocity of the
transported species A in the x direction, and so the speed of the molecule at any
point is given by
N Ax ∂XA
uA = = −D A + m A z A X A Ex + X Au (9.41)
c ∂x
Then the position of any ion or biomolecule assumed to be residing at a point in
the channel is obtained by solving
dxA
= uA (9.42)
dt
subject to the appropriate initial conditions.
To estimate the order of magnitude of each term in this equation, consider a
typical channel in the nanopore membrane in Figure 9.9 (Chen & Conlisk, 2008).
First, the diffusion term for a length L = 10−6 m = 1 µm and D A = 10−9 m2 /sec
is
∂XA m2 m
DA ∼ 10−9 X A /(10−6 m) ∼ 10−3 X A (9.43)
∂x sec sec
for a typical ionic solute having a change in concentration X A over the length
L. The term proportional to the electric field, for ionic species, is termed electro-
migration, and its magnitude is
mole m2 C
m A F z A X A E x ∼ 4 × 10−14 × 96,500 × 0.03V /10−6 m
J sec mole
(9.44)
∼ 10−3 z A X A (9.45)

for E x = 3 × 104 V/m. The species convective term is of the order

e E x RT
X AU ∼ X A ∼ 5 × 10−4 X A (9.46)
Fμ
where the viscosity of water is used for μ and the velocity scale U = e E x φ0 /μ.
Note that the convective term nominally is of the same order of magnitude as
the migration term and that the diffusion term is much smaller if X A  X A .
Also, the electromigration term for a cation is greater than zero, whereas that for
an anion is less than zero. Thus, for a channel having negatively charged walls,
a cation or other biomolecule of positive charge (assuming that the biomolecule
can be represented as a point charge) will be transported faster than an anion in an
electro-osmotic flow. This is the basis for electrokinetic separation of molecules,
particularly ions and biomolecules.
Equation (9.39) shows that a cation will move in a direction opposite to the
anion in an electrolyte solution in the absence of an electro-osmotic flow (EOF).
The cation will move to the cathode and the anion to the anode. This is depicted in
 320 Essentials of Micro- and Nanofluidics

(a) Without EOF (b) With EOF

Sketch of the motion of ions in an electrolyte solution in a channel having negatively charged walls.
Figure 9.10

Figure 9.10a. In the presence of electro-osmotic flow between negatively charged

walls, the cation and anion will move in the same direction for large enough
electro-osmotic flow velocity. The cation will travel faster than the anion; this is
depicted in Figure 9.10b.
The simplest model for the motion of a biomolecule in a nanopore mem-
brane is to consider it to be a single species within the multicomponent mixture
(say, Phosphate Buffered Saline (PBS)). Consider the transport of an uncharged
biomolecule such as glucose. Being neutral, glucose is transported solely by the
bulk fluid motion if the concentration is uniform because its valence is zero. Thus
an uncharged molecule will always move to the receiver region in Figure 9.10.
The motion of an anion or a negatively charged biomolecule will be controled
by the magnitude of the electric field. As noted, in the fully developed region of
the channel, the electromigration term is less than zero and the bulk convection
term is greater than zero for an anion, and so depending on the magnitude
of each term, the anion could move to the receiver or stay in the donor. It is
this electrodynamic interaction that determines the transport properties of the
membrane. However, because such a biomolecule may adsorb to the surface,
not all of the analyte that originated in the donor may reach the receiver. If the
polarity is reversed and the negative electrode is on the donor side, the flow will be
reversed.
The discussion here assumes steady flow, which means that there is an infinite
amount of solute in the donor region. In practice, this is often not the case, and
solute may be injected periodically as a plug rather than in a continuous manner.
This situation will require an analysis similar to Wilson (1948) for charged
solutes; this transient problem for uncharged solutes is discussed in Chapter 5.

9.2.7 Current and current density in electro-osmotic flow

In many flow-through biochemical sensing devices, changes in electrical current
are often used to identify an analyte (Braha et al., 2000). The definition of the
electrical current density in a flowing electrolyte solution having concentration
gradients is

J=F z i Ni (9.47)
i
321 Electrokinetic Phenomena

where
Di
Ni = −Di ∇ci + zi Fci E + ci u (9.48)
RT
The units of current density are amp/m2 . In Section 7.18, we analyzed a permse-
lective membrane having an imposed current, and in this section, we extend the
calculation of the current in a stagnant medium having no gradients presented in
Section 6.8 to include both electro-osmotic flow and thus velocity in addition to
potential and concentration gradients. In this section, the current is generated by
flow and the presence of the EDLs and is not imposed.
Notation alert: In this section, J is the current density and not the molar flux of
species A relative to the bulk velocity field J A , as in Chapter 3. Also, here there
should be no confusing the current I for ionic strength.
It is convenient to take the origin of the coordinate system at the centerline of
the channel. Integrating the streamwise component of the current density across
the channel, the total current per unit width of the channel is given by
h
I = Jx dy (9.49)
−h

Substituting for the streamwise flux, and neglecting streamwise gradients of

concentration, as is common (see the previous section), the current per unit width
is given by
F2  2 h h 
I = Ex z Di ci dy + F z i ci udy (9.50)
RT i i −h −h i

Taking the dimensional velocity field to be

e
u(y) = E x (φ − ζ ) (9.51)
μ
then
e2 h ∂ 2φ
I = 2hσe E x − E x (φ − ζ ) dy (9.52)
μ −h ∂ y2
where σe  is the average electrical conductivity of the solution defined by
F2 1  2 h
σe  = z Di ci dy (9.53)
RT 2h i i −h

with
F2  2
σe = z Di ci (9.54)
RT i i

being the electrical conductivity. Note that we have replaced the volume charge
density with the second derivative of the potential from Poisson’s equation. Since
φ = φ(y), the partial derivative ∂ 2 φ/∂ y 2 = d 2 φ/dy 2 is really a total derivative.
322 Essentials of Micro- and Nanofluidics

Integrating by parts, the total current is given by

 
 2 h dφ 2
I = 2hσe E x + E x e dy (9.55)
μ −h dy
It then remains to specify a solution for the potential. If the EDLs are overlapped,
and the potential is symmetric about the centerline, recall that the potential is
given by
ζ cosh κ y
φ= (9.56)
cosh κh
where κ is the inverse Debye length. Performing the integration,
  
e2 ζ 2 2 sinh (2/) − 1
I = E x 2hσe  + (9.57)
μh  2 cosh2 (1/)
Note that the current–voltage relationship is linear because E x = −/L, where
 is the voltage drop across the channel. Equation (9.57) is the analogue of
the equation V = I R for lumped parameter static systems; here the V is really a
voltage drop, which we have written as .
The equation for the current can be written in dimensionless form. From
equation (9.57), the quantity
E x e2 φ02
I0 = (9.58)
μh
where φ0 = RT /F has units of C/m sec = amp/m and so can be used for the
current scale. Thus, in dimensionless form,

I sinh(2/) − 1
I∗ = =  + (ζ F/RT )2 2 (9.59)
I0 cosh2 (1/)
where  = 1/κh = O(1) and  = 2h E x σe /I0 .
Thus, in general, the current in the channel consists of two parts: the molar flux
due to the migration of ions induced by the electric field, called the conduction
current (the first term in equation (9.59)), and the flux due to the bulk electro-
osmotic flow of the electrolyte, the convective current. The convective current is
displayed in Figure 9.11. Note that if the EDLs are thin ( → 0) (recall we can
do that because the cosh solution approaches the correct solution as  → 0), the
current due to the convective term is nonzero only within the EDLs on the walls
and thus contributes little to the current density. That is why water, while a better
conductor than air, is still a dielectric.

9.2.8 Electro-osmotic flow in an annulus

As in classical viscous flow, there are a number of canonical geometries for which
the electro-osmotic flow field may be obtained, analytically and numerically.
Electro-osmotic flow in a cylinder is covered in the exercises at the end of this
chapter, and in this section, electro-osmotic flow through an annulus having fixed
 323 Electrokinetic Phenomena

1.8
ζF/RT =0.5
1.6 ζF/RT =1.0
ζF/RT =1.5
1.4 ζF/RT =2.0

Dimensionless current
1.2

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
ε

The convective current as a function of  for several values of the dimensionless ζ potential.
Figure 9.11

walls is considered. If the inner cylinder is moving, this problem is a simplified

model for the electrophoretic transport of a DNA molecule through a nanotube,
discussed in detail in Chapter 12.
Notation alert: As you might have guessed by now, all the electro-osmotic
flow problems discussed so far vary in a single dimension. Thus all the partial
derivatives could be written as total derivatives, and as such, I have made no
attempt to be consistent in which notation to use. Either set at notation works
for me.

The dimensionless governing equations for the velocity and potential in fully
developed electro-osmotic flow of a binary monovalent electrolyte, with cation g
and anion f in a long annulus, shown in Figure 9.12, are
 
∂ 2φ 1 ∂φ
 2
+ = −β (g − f ) (9.60)
∂r 2 r ∂r
 
∂ 2u 1 ∂u
 2
+ = −β (g − f ) (9.61)
∂r 2 r ∂r

along with the boundary conditions that at r = α, 1, φ = ζ, u = 0. Here g and

f are mole fractions, and as before, β = c/I . These equations can be solved
numerically in the nonlinear, high-potential case, and an analytical solution may

Geometry for flow through an annulus.

Figure 9.12
 324 Essentials of Micro- and Nanofluidics

1.5

1
u and φ−ζ

0.5
α=0.9
α=0.8
α=0.6
α=0.4
0
0 0.2 0.4 0.6 0.8 1
(r− α)/(1−α)

Results for the electro-osmotic flow in an annulus; here the potential and the velocity are equivalent. The
Figure 9.13
different curves correspond to different values of α = a/b, the ratio of the radii, and the constant surface
potential boundary condition is used. Here b = 50 nm and V /L = 0.6 V/3.5 µm, and the ionic strength of
each species is 0.1 M in the upstream reservoir.

be obtained in the DH approximation. Note that again, in dimensionless form,

u = φ − ζ.
In DH limit, the governing equation for the potential is
 
1 ∂ ∂φ φ
r = 2 (9.62)
r ∂r ∂r 
This is a modified Bessel equation, and applying the boundary conditions at
r = α and r = 1, the solution for the potential is
         
ζ I0 r K 0 1 − K 0 α + ζ K 0 r −I0 1 + I0 α
φ (r) =     (9.63)
I0 α K 0 1 − I0 1 K 0 α
where I0 and K 0 are modified Bessel functions of order 0 and  = λ/b.
The results for the potential and velocity are shown in Figure 9.13. The curves
are similar to those seen in a channel; note that the EDLs appear to thin as α → 0.
This is really not the case and is a result of normalizing the horizontal axis to
be on the domain [0, 1]. If the inner cylinder is moving and the outer cylinder
is fixed (the Couette flow problem), the velocity induced by the inner cylinder
moving can be added linearly to the preceding velocity.

9.2.9 Electro-osmotic flow in nozzles and diffusers

Electro-osmotic flow in micro- and nanonozzles is important in many applica-
tions, for example, in electrical measurements on living cells (Lehnert et al.,
2002) and for the injection and manipulation of DNA fragments in a mass spec-
trometer (Luginbuhl et al., 2000). Nozzles are also sometimes used for single
molecule detection, analysis, and separation of biomolecules.
 325 Electrokinetic Phenomena

y*

2hi h* (x*)
x*
+ 2ho
−

The geometry of a typical micro-nano nozzle; the cross section is assumed rectangular.
Figure 9.14

A sketch of a typical nozzle geometry is depicted in Figure 9.14. The nozzle

(or diffuser) is assumed to have a rectangular cross section for simplicity, and the
spanwise width of the channel is assumed to be much larger than the height of
the channel, so the problem can be considered two-dimensional. The length of
the channel is L, and the channel height is assumed to be linear; also,
ho − hi ∗
h ∗ (x) = h i + x (9.64)
L
where h i , h o are the half heights of inlet and outlet of the channel, and in dimen-
sionless form,
ho − hi
h(x) = 1 + x (9.65)
hi
where h = h ∗ /h i and x = x ∗ /L. The walls of the channel are defined
by y = ±h(x). The imposed electric field E ∗ satisfies Maxwell’s equations
in the form ∇ • E ∗ = 0 so that if i denotes the imposed electric field,
∗ ∗
E xi (x)A(x) = E xi (0)A(0), where A is the cross-sectional area of the channel.
The total dimensionless electric field may be written

hi E0 E x ∂φ
ET x = − 1 (9.66)
φ0 ∂x
hi E0 E y ∂φ
ET y = − (9.67)
φ0 ∂y

where 1 = h i /L. If the imposed electric field at the inlet E 0 = E i∗ (0) is taken as
the scale of the imposed electric field, the dimensionless imposed electric field in
the x direction is E xi = 1/h(x).
Solutions for the potential and velocity are discussed under the DH approxi-
mation and the lubrication approximation for which (h i − h o /L)Re  1 in the
case of a nozzle, with a corresponding restriction for a diffuser. The Reynolds
number is based on the inlet height h i ; see Section 4.9 for the derivation of this
restriction.
326 Essentials of Micro- and Nanofluidics

THIN EDLS:   1
For thin EDLs, the electro-osmotic velocity is similar to that in a straight channel,
and
ζ  −( h−y ) h+y
u eof = e  − e−(  ) − 1 (9.68)
h
where ζ is the dimensionless potential on both walls; here, for negatively charged
walls, ζ < 0, and so u eof > 0. Superimposing a pressure driven velocity, if nec-
essary, the two-dimensional velocity field is given by

1 dp  2 ζ  −( h−y ) h+y
u= y − h2 + e  − e−(  ) − 1 (9.69)
2 dx h
Because the cross-sectional area of the channel varies with x, there is a trans-
verse velocity that is obtained by integrating the continuity equation
 
1 d 2 p y3 2 dp  1 φw h 
v = −1 − h 2
y − h 3
+ hh (y + h) − (y + h)
2 dx2 3 3 dx h2
  
φw h  φw h   − h−y − h+y − 2h
+ + e  − e  − e  + 1 (9.70)
h2 h
The pressure is an unknown, as noted in Section 4.9, and is obtained by requiring
that the continuity equation is satisfied; the result is a differential equation for the
pressure, which is given by

1 3 d2 p  2 dp ζ h  − 2h ζ h 
1 − e− 
2h
h 2
+ h h − e  +
2
=0 (9.71)
3 dx dx h h
Assuming that the nozzle is connected to reservoirs, the boundary conditions for
the pressure are taken to be

p = pi , x =0 and p = po , x =1 (9.72)

It should be noted that there is a rather extensive literature on the appropriate

boundary conditions to be used in lubrication problems, and these are the simplest
that can be considered (Pinkus & Sternlicht, 1961).
Integrating equation (9.71) once between the inlet of the channel and a given
position x,
 2h  
3 dp e−   2h
h = 3ζ − − Ēi 1, +C (9.73)
dx h h 

where C is an integration constant and Ēi is the exponential integral given by

∞ e−t
Ēi(x) = − dt (9.74)
−x t

Equation (9.73) can be integrated for   1, neglecting the first term in equa-
tion (9.73) because it is exponentially small and using the asymptotic expansion
 327 Electrokinetic Phenomena

1 1
x=0
x=0.2
x=0.4
0.5 x=0 0.5 x=0.6
x=0.2 x=0.8
x=0.4 x=1.0
x=0.6
0 x=0.8 0
y

y
x=1.0

−0.5 −0.5

−1 −1
0 0.02 0.04 −1.5 −1 −0.5 0 0.5 1
U V −3
x 10
(a) streamwise velocity (b) transverse velocity

The streamwise and transverse velocity of EOF in a microdiffuser. The height of the diffuser is 20 µm at the
Figure 9.15
inlet and 130 µm at the outlet, and the length of the diffuser is 650 µm;  = 5 × 10−5 , and the EDLs are thin
compared to the diffuser. The imposed electric field is 8000 V/m, and the ζ potential is −15 mV. The pressure
is zero both at the inlet and the outlet. The concentration of each electrolyte component in the upstream
reservoir is 0.1 M.

for the exponential integral; applying the boundary conditions,

po h 2o (h 2 − h i2 ) + pi h i2 (h 2o − h 2 )
p(x) =
h 2 (h 2o − h i2 )
  
ζ ζ h i2 + h i h o + h 2o ζ
+ −  2 +  (9.75)
h h3 h h h i h o (h i + h o ) h h i h o (h i + h o )

It is seen that the pressure distribution comprises two parts – that induced by the
inlet and outlet pressure difference as well as the presence of the EDLs – and for
thin EDLs, this correction is small. If there is no pressure difference ( pi = po ),
then the pressure field is due entirely to the presence of the electro-osmotic flow;
the pressure is small, and p = O().
Results for the electro-osmotic flow in a microdiffuser are shown in Figure 9.15.
In this example, the height of the diffuser is 20 µm at the inlet and 130 µm at the
outlet, and the length of the diffuser is 650 µm; the parameter  2 = 0.04  1.
Each electrolyte concentration is 0.1 M in the reservoir, and the ζ potential of
the PMMA walls is −15 mV (Kirby and Hasselbrink, 2004c). The Debye length
is 0.7 nm ( = 5 × 10−5 ), and the EDLs are very thin compared to the height of
the diffuser. As required by continuity, the velocity decreases in the streamwise
direction.

OVERLAPPED EDLS:  ∼ 1
The solution for the potential in this case is formally the same as for the straight
channel to leading order in 1 , and
y
cosh 
φ=ζ h
(9.76)
cosh 
328 Essentials of Micro- and Nanofluidics

where y is measured from the centerline of the channel; the velocity is

 y

φ ζ cosh 
u eo f = = −1 (9.77)
h h cosh h

The v velocity is obtained from the continuity equation, as discussed earlier,

and
   
1 d 2 p y3 2 3 dp  ζ h  sinh h sinh y
v = 1 − −h y− h +
2
hh (y +h) + 
2 dx2 3 3 dx h cosh2 h
  
ζ h  sinh y ζ yh  ζ h  1 tanh2 h ζ 1 h  tanh h ζ h
+ 2 h − 2 + + −
h cosh  h h h2 h
(9.78)

Again, using the continuity equation, the equation for the pressure is given by
   
1 3 d2 p  2 dp ζ h 2 h ζ h h ζ h
h + h h = − tanh − tanh + (9.79)
3 dx2 dx h  h2  h
At this point, further integration of this equation is difficult, and the solution for
pressure has to be found numerically.
Because of the complexity of the pressure equation, a numerical calculation
is used to obtain the pressure distribution within the nozzle. Second-order finite
difference methods are used to discretize the pressure equation (9.79), as dis-
cussed in Chapter 10, and the Thomas algorithm is used to solve the resulting set
of difference equations.
Figure 9.16 shows the electro-osmotic flow in a nanonozzle. The height of
the nanonozzle at the inlet is 13 nm, and the length of the nozzle is 65 nm; the
height at the outlet is 8 nm, and so the parameter 12 = 0.04  1. The Debye
length is 2.5 nm ( = 0.19), and the EDLs are overlapped, as shown in the
parabolic potential and velocity profiles in Figure 9.16. The imposed electric field
is 8000 V/m. The streamwise velocity is parabolic, and the transverse velocity is
O(1 ) and small compared to the streamwise velocity.
In Chapter 12, we will return to this problem because this device has applica-
tions in delivering DNA and other biomolecules to cells. An opening is created
in the cell membrane, and material is often delivered to the cell through a nozzle
of the type described here. The process of creating an opening in a living cell is
called electroporation.

9.2.10 Dispersion in electro-osmotic flow

As was seen in the case of Poiseuille flow in Section 5.7, dispersion of a solute
is characterized by a dispersion coefficient, which is indicative of how much a
solute can spread, over and above molecular diffusion, as it travels down a channel.
Dispersion of a solute is caused by the local variation of the fluid velocity.
 329 Electrokinetic Phenomena

1 1
x=0
x=0.2
x=0.4
0.5 0.5 x=0.6
x=0 x=0.8
x=0.2 x=1.0
x=0.4
x=0.6
y

0 0

y
x=0.8
x=1.0

−0.5 −0.5

−1 −1
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 −1 −0.5 0 0.5 1
U V −3
x 10
(a) streamwise velocity
(b) spanwise velocity

Results for electro-osmotic flow in the converging nanonozzle. The height of the nozzle is 13 nm at the inlet
Figure 9.16
and 8 nm at the outlet, and the length of the nozzle is 65 nm;  = 0.19, and the EDLs are overlapped. The
imposed electric field is assumed to be 8000 V/m, and the ζ potential of the walls is −5 mV. The pressure is
zero both at inlet and outlet. The concentration of each electrolyte solution in the upstream reservoir is 0.1 M.

In the exercises of Chapter 5, you showed that for a very wide channel, the
dispersion coefficient for an uncharged solute in a Poiseuille flow is
 
2h ∗2 u2
K = DA 1 + (9.80)
105D 2A
Dispersion also occurs in electro-osmotic flow when the EDLs are overlapped.
Notation alert: In this section, it is convenient to place the origin of the y
coordinate at the centerline of the channel. In this case, the electro-osmotic
velocity is slightly different from that when the coordinate origin is at the bottom
wall, as in the Poiseuille flow in Chapter 4.
As with the current calculation, it is convenient to place the origin of the y
coordinate at the centerline of the channel. Assuming that the DH approximation
holds, the expression for the dimensional EOF velocity in a channel with the origin
at the centerline of the channel having overlapped double layers is given by
 
e E x ζ cosh y
u= −1 (9.81)
μ cosh h
where y = y ∗ /λ and h = h ∗ /λ = κh ∗ . The average velocity for this distribution
is given by
 
e E x ζ tanh h
u = −1 (9.82)
μ h
Griffiths and Nilson (1999) obtained the expression for the dispersion coefficient
in the form
   
1 uλ 2
K = DA 1 + f (h) (9.83)
3 DA
330 Essentials of Micro- and Nanofluidics

where f is the function

(6 + h 2 ) sinh2 h − 92 h sinh h cosh h − 32 h 2

f (h) = (9.84)
(h cosh h − sinh h)2
Substituting for the average velocity, the dispersion coefficient becomes
    2 
1 U λ 2 tanh h
K = DA 1 + − 1 f (h) (9.85)
3 DA h

where U = e E x ζ /μ is a velocity scale based on the ζ potential. Note that the

quantity Peλ = U λ/D A is a Peclet number based on the Debye length λ. In the
limit h/λ → ∞, that is, for very thin EDLs, the dispersion coefficient approaches
the value
 
1 2
K ∼ D A 1 + Peλ (9.86)
3
and as the Peclet number Peλ → 0, there is no dispersion at all: K = D A .
While we have calculated the dispersion coefficient for the case of overlapped
double layers, dispersion can occur due to other effects. We have already men-
tioned Joule heating, but dispersion may also occur due to geometrical irregular-
ities, such as curvature and roughness, and due to variation of channel properties,
such as changes in ζ potential.
Field flow fractionation (FFF) (Giddings et al., 1976; Griffiths & Nilson, 2006)
is the term given to a variety of methods to separate and identify analytes based on
size and charge. FFF generally uses a pressure-driven flow in the main streamwise
direction and an applied field, such as electric and magnetic fields in the transverse
direction, which is perpendicular to the primary flow direction. The result is that
species will travel at different speeds, depending on their position relative to the
walls of the channel, their size, and their charge. This leads to the formation of
bands of particle motion similar to what occurs with ions in electro-osmotic flow
(see Figure 9.11).
A field that requires a fundamental understanding of dispersion is chromato-
graphy, which is the term given in the chemistry community to the separation
of mixture components flowing through a tube or channel. The carrier fluid can
either be a liquid or a gas, and the solid medium is usually a porous material.
In chromatography, adsorption of an analyte on the walls of a device is a central
feature of the separation process.
The solutes separate into bands of sample constituents due to dispersion,
each migrating with a speed characteristic of its relative interaction with the
stationary and mobile phases. Thus an analyte with a strong interaction with
the stationary phase moves more slowly than one with a weak interaction with the
stationary phase.
Chromatographic techniques can be classified according to the state of matter
of the mobile phase. In gas chromatography (GC) (Lee et al., 1984), the mobile
phase is a gas (the sample constituents have to be vaporizable), and in liquid
chromatography (LC) (Snyder & Kirkland, 1979), the mobile phase is a liquid. In
 331 Electrokinetic Phenomena

supercritical fluid chromatography (SFC) (Anton et al., 1998), the mobile phase
is a supercritical fluid.

9.3 Electrophoresis: Single particles

9.3.1 Introduction
The subject of electrophoresis, also known in the earlier literature as cataphore-
sis, is a vast one that has exploded in recent years with the development of
lab-on-a-chip devices for separations and rapid molecular analysis. As discussed
in Abramson (1931), electrophoresis seems to have first been described by F.
Reuss in a paper he published in 1809. Reuss observed the motion of clay par-
ticles relative to the electro-osmosis of water through a quartz-sand medium.
Smoluchowski (1918) and Henry (1931) made significant theoretical contribu-
tions to the study of electrophoresis. Today there are a number of books on the
subject, Shaw (1969) for one, and there are a number of journals, including one
appropriately named Electrophoresis.
Applications of electrophoresis permeate colloid science. Even in the absence
of dispersion, charged molecules such as proteins, amino acids, DNA, RNA,
polymers, and other molecules can be separated based on charge, shape, and
size using electrophoresis. These separations can be performed in channels or
cylindrical pores having uncharged walls (no electro-osmosis) or charged walls.
Clearly uncharged molecules can also be separated from charged molecules.
Electrophoretic separation of colloidal particles is achieved by passing a sample
in a solvent through that conduit that can be a channel or tube. Electrophoretic
separations can occur in the absence or presence of flow. Of course, if flow
is added, say, the walls are charged, separation will still occur in the manner
described in Section 9.2.6.
There are different experimental variants of electrophoresis; capillary elec-
trophoresis or capillary zone electrophoresis, using a tube to separate species
based on different size and charge, is perhaps the most common. The solvent is
often an aqueous solution, but higher-viscosity solvents, such as gels, may also be
used, a process known as gel electrophoresis. Recall that a gel is a liquid made up
of high-molecular-weight material, long-chain polymers; these polymers could
consist of proteins, carbohydrates, and polysaccharides.
It should be noted that electrophoresis and electromigration are similar pro-
cesses. The difference between the two is that in electrophoresis, the size of the
molecule or particle is explicitly accounted for, usually based on the Stokes drag
law, and the influence of the particle EDL is often included, as will be seen.
Conversely, though the size of the particle in electromigration is accounted for in
the Stokes–Einstein expression for the diffusion coefficient, the influence of the
particle EDL is not included. Thus the term electromigration is usually reserved
for ions and very small molecules dissolved in solution.
In this section, the fundamental theory of electrophoresis is presented, and the
work of Smolukowski and Henry is covered in detail.
332 Essentials of Micro- and Nanofluidics

Electrokinetic Soil Remediation

The electrokinetic remediation process. Image courtesy of Sandia National Laboratory.

Electro-osmotic flow can be used in soil remediation. Electrodes are placed in

the treatment area as shown in the figure above and the flow through the porous
soil media drives the cations toward the cathode and the anion toward the anode.
Using this procedure, harmful negatively charged pollutants, such as heavy metals
(e.g. chromium ions, Cr O4−2 ) and polar organics in mud sludge, can safely be
removed. In the figure the lead Pb+2 moves to the negatively charged cathode and
the chromium moves to the positively charged anode.
Note that the electrokinetic remediation process in soil takes place in a porous
media so that the EOF is through channels of arbitrary shape and tortuosity.
Nevertheless, the basic physical principles involved in electrokinetic remediation
are the same as that in a channel of rather uniform dimension.

9.3.2 Electrophoretic mobility

Consider the motion of a single charged particle in a neutral solvent in a stagnant
fluid. The insertion of the particle will create an EDL of opposite charge to the
surface of the particle to keep the system electrically neutral. Because of the
presence of the EDL, there will be locally a net body force of (dimensional)
magnitude ρe E x , where E x is the dimensionally applied electric field. If the
particle is negatively charged, it will move in a direction opposite to the direction
of the electric field, and the solvent will only be affected by the motion of the
particle; that is, far from the particle, the solvent velocity vanishes.
Suppose that the Debye length is much larger than the particle radius, as
depicted in Figure 9.17a. Then, equating the electrical force with the viscous
drag using Stokes’s drag law results in

q E x = 6πμUe a (9.87)
 333 Electrokinetic Phenomena

(a) λ >> a (b) λ << a

A particle moving in an electric field. (a) Debye length much greater than the particle radius. (b) Debye length
Figure 9.17
much smaller than the particle radius, a.

where q is the net charge on the particle. Thus the electrophoretic velocity is
given by
q Ex
Ue = (9.88)
6πμa
As with electro-osmotic flow, the electrophoretic mobility is defined as
Ue
μe = (9.89)
Ex
Recall that the potential due to a charged sphere in the DH limit is given by
a
φ(r ) = ζ e(a−r)/λ (9.90)
r
The total charge on the sphere was calculated in Chapter 7, and the result is
 
2 dφ ∞ 1 1
q = 4πe a | = 4πe a ζ 2
+ (9.91)
dr a λ a
Substituting for q into equation (9.88),
2 ζ e E x a
Ue = for 1 (9.92)
3 μ λ
This is the so-called Debye-Hückel (DH) expression for the electrophoretic veloc-
ity (Figure 9.22a).
In the limit of very large a/λ, larger particles, and a thin double layer, curvature
effects may be neglected. In this limit, the electrophoretic velocity is equivalent
to the electro-osmotic velocity far from a flat surface, and we have already shown
this velocity to be given by
ζ e E x a
Ue = for 1 (9.93)
μ λ
This is the Helmholtz–Smolukowsky expression (Figure 9.22b).
Now consider the case in which the particle is immersed in an electro-osmotic
flow having a velocity U . Then the mixture will experience a net electrical force
and will move in the direction of the electric field. Thus the balance of electrical
and drag forces becomes
q E x = 6πμ(Ue − U ) (9.94)
334 Essentials of Micro- and Nanofluidics

and the particle can go either in the same direction as the electric field or in the
opposite direction. The velocity U corresponds to the electro-osmotic velocity far
from the particle. For thin double layers on the bounding surfaces of the channel,
the electro-osmotic velocity U will be constant.

9.3.3 Henry’s solution

In the previous section, several assumptions were made in the calculation of the
electrophoretic velocity of a single colloidal particle. First, it was assumed that the
local electric field were unaffected by the presence of the particle. This is unlikely
to be the case in practice. Second, it was assumed that the dielectric constant of
the particle and the fluid were the same. And finally, the electrophoretic velocity
of the particle was calculated only for  = λ/a  1 and   1. Henry (1931)
removed these assumptions; his work is the subject of this section.
Notation alert: In this section, we will use the notation in the chemistry com-
munity for λ = 1/κ. As has been mentioned before, chemists like to refer to the
ratio of the particle radius to the Debye length as κa.
Henry was concerned with the dependence of the electrophoretic velocity on
the shape of the particle and considered both spheres and cylinders in his analysis.
It is assumed that the Reynolds number is small and thus that the Stokes’s flow
assumption holds. The solution for the applied electric field is simply added to
the solution for the EDL, as we have done previously. The potentials due to the
applied field satisfy

∇ 2ψ = ∇ 2ψ = 0 (9.95)

where
   
1 ∂ 2 ∂ 1 ∂ ∂
∇ = 2
2
r + 2 sin θ (9.96)
r ∂r ∂r r sin θ ∂θ ∂θ
and the potential has been assumed to be symmetric about the transverse angle,
where the prime indicates the potential inside the particle. The boundary condi-
tions are continuity of surface charge at the particle surface:

∂ψ  ∂ψ
e = e at r = a (9.97)
∂r ∂r
Continuity of the potential requires

ψ =ψ at r = a (9.98)
ψ = −E x r cos θ as r → ∞ (9.99)

The solutions for the potentials are given by

 
ξ a3
ψ = −E x r + 2 cos θ (9.100)
r
335 Electrokinetic Phenomena

outside the particle and

ψ  = −E x (1 + ξ )r cos θ (9.101)
inside the particle, where ξ = (e − e )/(2e − e ).
Turning to the momentum equation, the velocity field satisfies
μ∇ 2 u − ∇ p = ρe ∇(ψ + φ) (9.102)
∇ ·u =0 (9.103)
where φ is the potential due to the EDL on the sphere. The sphere is assumed to
be held fixed in the flow, and thus the boundary conditions in a coordinate system
fixed to the sphere are
u r = −U cos θ as r → ∞ (9.104)
u θ = U sin θ as r → ∞ (9.105)
ur = u θ = 0 as r = a (9.106)
Henry obtained the solution for the velocity field in a rather elegant but some-
what lengthy manner. The most important result is the form of the force balance
on the particle; this is given by
a ∂φ
−6πμUe a + e E x dr + e E x a 2 |r =a + q E x = 0 (9.107)
∞ ∂r
By Gauss’s law, the last two terms cancel, and thus
e E x a
Ue = dr (9.108)
6πμ ∞
where  is defined by
∂φ r ∇ 2φ
= + ξ a 3r dr (9.109)
∂r ∞ r
4

In general, a value for  can be computed from a numerical solution for the
potential. Avoiding such a situation (remember that Blasius (1908) had computed
a numerical solution to the boundary layer equations in 1908), Henry suggested
that for a first approximation, the DH approximation could be used (equation
(9.90)). In this case, performing the integrations and expanding for large κa > 20,
  2  3 
e ζ E x 1 1 1
Ue = 1−3 + 25 − 220 + ··· (9.110)
μ κa κa κa
and for κa < 5,

2e ζ E x κa 5(κa)3 (κa)4 (κa)5
Ue = 1+ − − +
3μ 16 48 96 96
  
(κa)4 (κa)6 κa e −t
+ − eκa dt (9.111)
8 96 ∞ t
It is seen that there is a gap in the validity of the two formulas between κa = 5
and κa = 20. Of course, these formulas are only valid for potentials less than
26 mV, although we saw previously in Chapter 7 that the DH approximation for
 336 Essentials of Micro- and Nanofluidics

1.16 1.5

1.14
1.4
1.12

1.1 1.3
f(κ a)

f(κ a)
1.08

1.06 1.2

1.04
1.1
1.02 Curve Fit Curve Fit
Henry Henry
1 1
0 1 2 3 4 5 0 100 200 300 400 500
κa κa
(a) (b)

Plot of Henry’s solution and the curve fit for different ranges of κ a.
Figure 9.18

the potential is accurate at much higher potentials. Note that there is an error in
Henry’s original paper that is corrected by Dukhin and Derjaguin (1974). It is
important to note that this solution neglects the induced asymmetry of the ionic
cloud surrounding the particle as it moves through the fluid.
There is a useful curve-fit to Henry’s formula, and this formula was developed
by Masliyah and Bhattacharjee (2006). If Henry’s formula is written in the form
Ue 2e ζ
μe = = f (κa) (9.112)
Ex 3μ
then a simple but very accurate curve-fit is given by
3 1
f (κa) = −  (9.113)
2 2 1 + A −B
where A = 0.072 and B = 1.13; this function is plotted in Figure 9.18, along
with Henry’s solution for the entire range of κa.
It is seen that Henry’s model for the electrophoretic velocity does not appear
to depend on size for κa  1. This fact was actually demonstrated by Abramson
(1931). He showed in a series of papers that shape does not affect the elec-
trophoretic mobility of quartz, glass, and clay particles of irregular forms and
of size varying from 3 to 15 µm. Moreover, Abramson (1931) also found that
the electrophoretic mobility of leucocytes, blood platelets and red blood cells are
the same as their irregularly shaped aggregates. This, he suggested, is because the
particles get covered with an adsorbed layer of protein and so behave as if they
have the same surface characteristics.

9.3.4 The full nonlinear problem

While Henry’s solution is elegant and gives good results at low potentials, at
high surface charge and potential of the particle, Henry’s solution neglects some
important effects. His theory does not account for the distortion of the coun-
terion cloud surrounding the particle, as depicted in Figure 9.19, and cannot
describe the high potentials that occur in the range 5 < κa < 20. Overbeek (1943)
 337 Electrokinetic Phenomena

and Booth (1950) both sought to remove

this restriction, but their results are
expressed as a power series in the ζ poten-
tial and thus can only describe the result for
moderate values of the potential.
Wiersma et al. (1966) produced numer-
ical solutions, but because of convergence
problems, they could not produce results for
Electrophoretic motion of a colloidal particle at a ζ potential greater than ζ = 125 mV or
Figure 9.19
high potential, showing asymmetry of the coun- a value of 1/ = κa = 2.765. O’Brien and
terion cloud. White (1978) calculated the solution to the
problem, avoiding the convergence issues
encountered by Wiersema et al. For higher values of the potential, where the
Henry and Wiersema et al. solutions are not valid, they found that a maximum
occurs in the mobility for κa > 6 and near a value of Fζ /RT ∼ 5. O’Brien and
White’s results are depicted in Figure 9.20. Recall that the results for κa → ∞
correspond to the DH expression for the electrophoretic velocity, while the case
κa → 0 corresponds to the Helmholtz–Smolukowsky expression. The works of
Wiersema et al. and O’Brien and White are for a positively charged particle;
however, the results apply by simple sign changes to the case of a negatively
charged particle.

Electrophoretic mobility of a colloid particle from O’Brien and White (1978): (a) κ a ≤ 2.75; (b) κ a ≥ 3. Here
Figure 9.20
y = F ζ /R T , and E is the dimensional electric field.
 338 Essentials of Micro- and Nanofluidics

The effect of multivalent ionic species was also considered by O’Brien and
White. They found that the effect of counterion valence is substantial, with
the mobility significantly decreasing as the counterion valence increases. The
increase in counterion valence from 1 to 3 decreases the electrophoretic mobility
by a little over a factor of 3; that is, the decrease is linear in valence.

9.4 Streaming potential

We now proceed to the third electrokinetic phenomenon, the streaming potential.

When an electrically conducting fluid is passed through a channel or tube under an
imposed pressure drop, there is an electrical potential difference set up as a result
because the net current in the channel must vanish (i.e., there are no upstream and
downstream electrodes). From the governing equations in the direction of flow in
the fully developed region, the streamwise momentum equation is given by
d 2u dp
μ = − ρe E x (9.114)
dy 2 dx
where ρe is the volume charge density and E x is the streamwise component of the
induced electric field. Note that as we have discussed previously, the velocity is
composed of a pressure-driven component and
an electrically driven component. The stream-
ing potential corresponds to the potential asso-
ciated with E x in equation (9.114). In this sec-
tion, the manipulations will be performed in
dimensional form, and it is left to the exercises
to perform the analysis in dimensionless form.
Geometry for the calculation of the stream-
Figure 9.21 The geometry of interest is a channel that
ing potential.
is wide and long compared to its height. The
geometry is depicted in Figure 9.21. Note that the origin is taken at the center of
the channel so that the height is 2h and the flow is assumed to be fully developed.
This coordinate system is chosen so that a comparison with the result for the
cylindrical tube given by Newman (1972) can be made easily.
Recall that the current density is defined by

J=F z i Ni (9.115)
i

where Ni is the dimensional flux, given by

Di
Ni = −Di ∇ci + zi Fci E + ci u (9.116)
RT
Integating the streamwise component of the current density across the channel to
obtain the total current, the result is
h
I = Jx dy (9.117)
−h
339 Electrokinetic Phenomena

Substituting for Jx , at zero current, the result for the induced electric field is
given by
F2  2 h h 
Ex z Di ci dy + F z i ci udy = 0 (9.118)
RT i i −h −h i

where we have neglected the streamwise gradients of the concentration.

Further progress may be made by specifying the form of the velocity field.
We know that the electrically driven component is proportional to the electrical
potential, and so the velocity may be written as
e 1 dp 2
u(y) = E x (φ − ζ ) + (y − h 2 ) (9.119)
μ 2μ d x
Substituting into equation (9.118) and noting from the equation for electrical

potential that i F z i ci = −e d 2 φ/dy 2 , it follows that2
e2 h d 2φ e d p d 2φ h
2hσe E x − E x (φ − ζ ) dy − dy = 0 (y 2 − h 2 )
μ−h dy 2 2μ d x
−h dy 2
(9.120)
where σe  is the average electrical conductivity of the solution defined in Sec-
tion 9.2.7.
Finally, integrating the two integrals by parts, we obtain
 2
e2 h d 2 φ e d p h dφ
2hσe E x + E x dy + y dy = 0 (9.121)
μ −h dy 2 μ d x −h dy
so that the induced streaming potential electric field is given by

e d p h
∂φsp y dφ dy
μ d x −h dy
Ex = − =− 2 h
 (9.122)
∂x 2hσe  + μe −h dφ
2
dy
dy

Equations (9.121) and (9.122) are the analogues of the equation for a cylindrical
tube given by Newman (1972), p. 194, equations (63-10).
As a final step in the analysis, note that
h dφ h
y dy = (yφ) |hh − φdy (9.123)
−h dy −h

= 2hζ − 2hφ (9.124)

so that the electric field per width of the channel is given by

e d p
∂φsp μ dx
(ζ − φ)
Ex = − = −2h  (9.125)
∂x 2  h 2
2hσe  + μe −h dφ dy
dy

For thin EDLs, the convective portion of the total current is very small and of
O(), where  = λ/ h in dimensionless terms because over much of the channel,

2
Because fully developed flow is assumed, ∂φ/∂ y = dφ/dy, though we need not make this sharp
distinction, as discussed previously.
 340 Essentials of Micro- and Nanofluidics

the concentrations will be constant and i z i ci = 0 to preserve electroneutrality.
Thus, for thin double layers, as with the electrical current discussed in Section
9.2.7, the streaming potential difference is very small.
The ζ potential can be measured by using the streaming potential. In a
microchannel having thin EDLs, the potential will be a constant at a given cross
section we can choose to be zero, and in this case, equation (9.125) becomes

ζ e d p
Ex = (9.126)
σe μ d x

With the definition d p/d x = p/L, solving for the ζ potential,

σe μφsp
ζ = (9.127)
e p
Measurement of φsp = φupstream − φdownstream and p = pupstream − pdownstream
then yields the average ζ potential in the channel.
As an example, the transmembrane pressure drop across a nanopore membrane
simulating the function of the kidney (Conlisk et al., 2009) is about p = 2 psi.
In a single slit pore making up the membrane of 2h = 8 nm, in 0.1 M NaCl in
plasma, a typical average flow speed for the pressure-driven component is U =
1.83 × 10−5 m/sec. For a surface charge density of −3 × 10−3 C/m2 , the ratio

|Usp | e RT |E sp |
= ∼ 0.0017 (9.128)
U μFU
so that the streaming potential is small. Note that the streaming potential velocity
opposes the pressure-driven velocity.
Analogous to the streaming potential is the streaming current, for which the
electric field vanishes rather than the current; a sketch of a streaming current
setup appears in Figure 9.22. Consider the situation in which two electrodes
inserted near the channel inlet and exit
are connected outside the channel through
resistanceless electrical connectors. In this
situation, both electrodes are at the same
potential, and no electric field can exist
within the channel (E x = 0 in equation
(9.116)). The current I , originating from
the net pressure-driven motion of charges
within the channel, can be measured in the
electrical circuit outside the channel. This
is known as the streaming current method
of calculating the ζ potential on the channel
wall.
Through simplification of the expression
An experimental setup for a streaming current
for I obtained earlier in a manner similar
Figure 9.22
measurement of the ζ potential. Contributed by to that in the streaming potential method
Professor Susan Olesik. for λ  h, the current per unit width of the
 341 Electrokinetic Phenomena

channel is given by
2he ζ p
I = (9.129)
μL
Note that, unlike equation (9.127), the current predicted by the preceding equa-
tion has no dependence on the conductivity σ of the electrolyte. Thus no esti-
mate of electrolyte conductivity is necessary for the streaming current method,
unlike for the streaming potential method. However, more sensitive instru-
mentation may be necessary to measure the typically 1 nA/kPa of current
generated at room temperature in the streaming current method than the typi-
cally 10–100 mV/kPa of voltage generated in the streaming potential method
(Erickson & Li, 2001).

9.5 Sedimentation potential

The sedimentation potential, also called the Dorn effect (Masliyah and Bhat-
tacharjee, 2006), is the potential induced by the fall of a charged particle under an
external force field at zero current (Fig-
ure 9.23). It is analogous to the streaming
potential in the sense that a local electric
field is induced as a result of particle motion
under the action of gravity or centrifugal
fields. Here we discuss the case of a parti-
cle falling under the action of gravity. Just
as in electrophoresis, the fall of a particle
induces a streaming current caused by the
A charged particle falling under gravity, illus- distortion of the EDL around the particle.
Figure 9.23
trating the sedimentation potential. The governing equations for the sedi-
mentation potential are the same as for elec-
trophoresis of a particle, with the addition of the driving gravitational force term.
Thus the methods of determining the sedimentation velocity of a charged particle
are similar to those already described for electrophoresis. Moreover, the local
velocity field is entirely analogous to the Stokes velocity profile discussed in
Chapter 4 for an uncharged particle.
Recall that the sedimentation velocity of a single uncharged particle is given
by
 
2 a2 g ρp
US = −1 (9.130)
9 ν ρ

If ρ p /ρ > 1, then the particle motion is downward in the direction of gravity.

Analogous to electrophoresis, a force balance on a charged particle leads to the
sedimentation potential, given by

e ζ (ρ p − ρ)g
ES = (9.131)
μσ∞
 342 Essentials of Micro- and Nanofluidics

where σ∞ is the electrical conductivity of the fluid far from the particle. In a
dilute suspension of particle volume fraction α, the sedimentation potential is
αe ζ (ρ p − ρ)g
ES = (9.132)
μσ∞
Both formulas for the sedimentation potential are valid for κa  1 and for α  1.
As with electrophoresis, the effect of κa is not incorporated in the preceding
simple formulas. The sedimentation of a charged particle has been analyzed in
detail by Stigter (1980) numerically and by Ohshima et al. (1984). Ohshima et al.
show, in particular, that the sedimentation velocity for a charged sphere is the
same as that for an uncharged sphere in the limit of a thin EDL, as noted in
Newman (1972) (i.e., λ/a  1). In particular, the electric field is found to be
given by
αe ζ (ρ p − ρ)g
ES = − f (κa) + · · · (9.133)
μσ∞
where α is the volume density of the particles and ζ is the ζ potential of the
particles. Here f (κa) is a function similar to that derived for electrophoresis, and
Ohshima et al. find that for Fζ /RT < 2,

f (κa) = 1 + 2eκa E 5 (κa) − 5eκa E 7 (κa) (9.134)

where E n are the exponential integrals of order n, defined by

∞ dt
E n (x) = x n−1 e−t (9.135)
x tn
Note that equation (9.133) indicates that the electric field acts upward for posi-
tively charged particles (ζ > 0) and downward for negatively charged particles
(ζ < 0), assuming that gravity acts downward in the positive coordinate direction.

9.6 Joule heating

As already mentioned, at small channel heights, electric fields are more efficient
than pressure gradients in driving fluid flow. However, a by-product of applying
an external electric field is that it acts as a source of energy, thereby causing
unwanted temperature gradients within the channel; that is, Joule heating is the
process by which electrical energy is converted to thermal energy because of a
resistance to electrical current flow.
Variable temperature affects a number of properties, including viscosity, elec-
trical conductivity, electrical permittivity, and thermal conductivity (Xuan et al.,
2004). Viscosity decreases with rise in temperature, and this change in viscosity
may lead to increased dispersion.
The current flowing through a microfluidic device generates heat. The electrical
power input per unit volume when current passes through a straight microfluidic
channel of constant cross section carrying a flowing electrolyte is given by Pin =
Jx E x , where Jx is the current density from equation (9.48). The axial gradient of
343 Electrokinetic Phenomena

the concentration of each ion is significant only a few channel widths from the
ends of the channel, and so the diffusional component of current can be neglected.
Thus, the electrical power, is,

Pin = σe E x2 + ρe E x u (9.136)

where σe = (F 2 /RT ) i zi2 ci Di is the electrical conductivity. The last term in
equation (9.136) is mechanical work that makes no contribution to the change in
thermal energy of the fluid.
Neglecting spanwise variations, the governing equation for the temperature is
thus
 
d dT
k(T ) = σe (T )E x2 (9.137)
dy dy
Note that the thermal conductivity and the electrical conductivity are functions
of the local temperature. The temperature dependence of the thermal conductiviy
has already been discussed, and it is seen that

k(T ) = A + BT + C T 2 (9.138)

where A, B, and C are constants. For water, A = −0.383, B = 5.254 × 10−3 , and
C = −6.369 × 10−6 . The variation of the electrical conductivity with tempera-
ture is linear and increases with increasing temperature. Usually, the dependence
is of the form (Masliyah and Bhattacharjee, 2006)

σe (T ) = σe1 (1 + K (T − T1 )) (9.139)

where K is a constant and K = 0.02◦ C−1 for freshwater. Equation (9.137) can
be solved numerically for the temperature given boundary conditions at the two
walls.
Joule heating is a major factor in decreasing the efficiency of microdevices
but does not seem to be a major factor in nanodevices. To see this, if the walls
of channel dissipate the volumetric generation, then a heat balance on a control
volume in the channel results in

2q y A = σe E x Ah (9.140)

from which it is seen that

1
qy = σe E x h (9.141)
2
That is, the heat flux required to be dissipated is much less for a nanochannel than
for a microchanel. Indeed, Xuan et al. (2004) have shown that in microchannels,
Joule heating affects the velocity distribution significantly, changing the classical
concave EOF velocity distribution to a convex one. Steep temperature drops
were also observed near the ends of the channel. Erickson et al. (2003) have
investigated the Joule heating effect in PDMS systems and found significant
increases in temperature because of the relatively low value of the PDMS thermal
conductivity.
 344 Essentials of Micro- and Nanofluidics

9.7 Chapter summary

This book is all about flows of electrically conducting fluids in small channels. As
the smallest dimension approaches the nanoscale, it has been shown that pressure
driven flow cannot be employed to move fluids.
Thus moving fluids around electrokinetically becomes much more effective.
In this chapter, the four main electrokinetic phenomena have been described:

1. Electro-osmosis (electro-osmotic flow): the bulk motion of a fluid caused

by an electric field
2. Electrophoresis: the motion of a charged particle in an otherwise motionless
fluid or the motion of a particle relative to a bulk motion
3. Streaming potential: the potential induced by a pressure gradient at zero
current flow; streaming current: the current induced by a pressure gradient
at zero electric field
4. Sedimentation potential: the electric field induced when charged particles
move relative to a liquid under a gravitational, centrifugal, or other force
field.

Electrophoresis and sedimentation potential refer to particles, and electro-

osmosis and streaming potential refer to electrolyte solutions.
By far the greatest emphasis in this chapter is on electro-osmosis and elec-
trophoresis. We have derived the solution for the velocity and potential in a wide
rectangular channel (most common in micro- and nanofluidics) for both thin and
thick EDLs. The method of matched asymptotic expansions permits the derivation
of the solutions for arbitrary values of the ionic valences for different values of
the ζ potential on the walls, a situation made tractable by the development of
modern fabrication procedures such as injection molding and laser ablation. The
electrical current passed by such nanochannels in electro-osmotic flow has also
been calculated. For thin EDLs, the current is dominated by the conduction cur-
rent. We have discussed in particular how ions of different valence can move at
different speeds in both micro- and nanochannels, a process called electromigra-
tion. Ions also move at different speeds due to dispersion, a process we discussed
for uncharged molecules in Chapter 5.
Electrophoresis is discussed in this chapter from the point of view of a simple
force balance for thin and thick EDLs to full numerical solutions of the governing
equations. Of particular value is a curve-fit to the numerical results of Henry
(1931), an analysis that spans all but a small range of the parameter κa developed
by Masliyah and Bhattacharjee (2006).
We ended this chapter with short discussions of the sedimentation potential
and Joule heating, the latter of which can lead to significant dispersion.
The electrical properties of small ions and small to large biomolecules can be
analyzed, identified, mixed, separated, and manipulated in nanopore membranes
in a variety of applications. Molecules can be separated based on their size and
345 Electrokinetic Phenomena

charge due to dispersion and due to their different electrophoretic mobilities;

small ions can be separated based on their different electro-osmotic mobilities.
These separation properties can be used to estimate a given unknown molecule’s
size and charge. The field of chromatography, the separation of molecules, par-
ticularly biomolecules, relies on basic knowledge of electro-osmotic flow and
electrophoresis. It is the electrically conducting nature of the underlying fluid and
the imbedded particles that gives rise to the applications described in Chapter 12.

EXERCISES
9.1 What is the best mechanism for achieving a flow rate of 10−6 µL/min in a
20 nm channel? Pressure driven or voltage driven? Assume a 1:1 electrolyte
in an aqueous solution.
9.2 Calculate the velocity profile under the DH approximation for electro-
osmotic flow through a cylinder of radius a. Assume a z:z electrolyte in an
aqueous solution and that φ = ζ at r = a. What is the primary effect of a
larger valence?
9.3 An electro-osmotic flow is present between two plates. At y = 0, the poten-
tial is φ = ζ0 , and at y = 1, it is φ = ζ1 . If the pressure gradient is zero
and the DH approximation is valid, find the relationship between ζ0 and ζ1
and other parameters such that the dimensionless volume flow rate Q = 0.
Assume that the EDLs are thin, and calculate the flow rate through O(),
where  = λ/ h. Write this relationship in dimensional form as well as using
the potential scale φ0 = RT /F.
9.4 A flow field induced by a combined pressure-driven and electro-osmotic
flow is present between two parallel plates. The ζ potential is the same on
the two plates. Under the DH approximation, find the relationship between
the pressure gradient and the ζ potential such that the volume flow rate van-
ishes. Reproduce Figure 9.3 by solving for the electro-osmotic component
numerically for h =20 nm for a reservoir concentration of 0.1 M = 0.05 +
0.05 M for electrolytes with valence z = ±1 and compare with the result
in the DH limit.
9.5 Suppose an electrolyte mixture consists of three species (g, f, r) of valence
zr = 2, z g = 1, z f = −1. Show that in this case, the electrical neutrality
condition in the core of the channel leads to

g0 r0
x3 − x − 2 =0
f0 f0

where x = eψo = u o , the outer solution for the potential and velocity, and the
superscript zero denotes the mole fraction at the wall. Compare this result
with the solution without species r. It has been observed in experiments that
the electro-osmotic velocity can be significantly reduced with the addition
of a multivalent cation in a negatively charged channel. If the ionic strength
346 Essentials of Micro- and Nanofluidics

is kept constant, under what conditions will the velocity be significantly

reduced.
9.6 A biomolecule of valence z = −15 and diffusion coefficient D = 5 ×
10−11 m2 /sec are being transported in an aqueous solution in a rectangular
channel in a sodium–chloride solution. The concentrations in the reservoir
upstream are [Na] = [Cl] = 0.15 M and [B] = 6.7 × 10−4 M. Estimate the
convection velocity of the biomolecule with respect to the elecro-osmotic
velocity if the pressure gradient is negligible.
9.7 In dimensionless form, show that the electrical current density is conserved:
∇ • J = 0.
9.8 Using the singular perturbation techniques discussed in Appendix A,
develop an expression for the outer solution for the velocity in a cylin-
drical tube.
9.9 Determine the electro-osmotic velocity profile in a fluid film on a plate that
is being dragged out of an electrolyte solution at a velocity U0 . Assume
that the shear stress vanishes at the free surface. Write the equation for the
velocity in terms of the electrical potential. (Can this be done?) Assume
that the surface charge density is fixed at the plate. What is the boundary
condition on the potential at the free surface if the fluid outside the film is air?
9.10 An analytical solution for the flow in an annulus has been obtained
analytically under the DH approximation. Solve the governing equations
numerically for a pair of ions of valence z = ±1 and ζ = 100 mV on each
wall. Compare the result for the velocity with the DH approximation result.
9.11 Polystyrene beads are often used as simulants for biomolecules. Assume
that the beads have a radius of a = 40 nm, the surrounding mixture is a
NaCl solution having a total electrolyte concentration of 1 mM far from
the particle, and a surface charge density of σ = −0.01 C/m2 . Assuming
that the electric potential outside the particle is

φ(r) = a/r ζ e(a−r)/λ

estimate the electrophoretic velocity of the particle. Compare this value

with the value obtained from an electromigration calculation of 1 mm/sec.
9.12 A mixture contains two species of colloidal particles that are negatively
charged and spherical. The concentration of the 1:1 aqueous electrolyte
is 0.03 M = 0.015 + 0.015 M, and the one species of colloidal is 100 nm
in diameter and has a ζ potential of ζ = −40 mV. The other species has
a diameter of 20 nm and has ζ = −20 mV. Estimate the time it takes to
travel a distance L = 100 µm in a channel of height h = 5 µm.
9.13 Show that the equation for the streaming potential in a cylindrical capillary
of radius a is given by

∂φsp e a 2 d p ζ − φav
Ex = − =− 
∂x μ d x a 2 σ  + e2  a dφ 2
2r dr
e μ 0 dr
347 Electrokinetic Phenomena

Write the equation in dimensionless form as

Usp ζ − φ
= a  2
Up μσe 
+ 0 dφ r dr
e2 φ02 dr

where U p is the velocity scale for the pressure-driven flow and Usp is the
streaming potential velocity based on E x .
9.14 Derive the equation for the streaming potential in a wide channel, including
slip at the wall. Where does the slip length appear in the equation? Does
it increase or decrease the streaming potential?
9.15 Consider a pressure difference P = 2 psi across the length L = 4 µm
of a nanochannel with slit-shaped pores of height 2h = 10 nm and width
W = 45 µm connected to reservoirs containing 0.143 M sodium–chloride
solutions. The surface charge density on each nanochannel wall is
σ = 2.8 × 10−3 C/m2 . Find the streaming potential drop φ, assuming
that the EDLs do not overlap and the flow rate through the nanochannel in
picoliters per second (pico ≡ 10−12 ). What is the the percentage reduction
in the flow rate due to the streaming potential effect?
9.16 As mentioned in the text, another way to measure the ζ potential is by the
streaming current method. Show explicitly that the current is given by
2he ζ p
I =
μL
and thus the ζ is determined by the properties of the electrolyte and the
measured current.
9.17 Compare the dispersion coefficient for pressure-driven flow and for
electro-osmotic flow. Calculate the ratio

K
DA
−1
γ = PR
K
DA
−1
EOF

and show that its value is

105 f (κh)
γ =
24(κh)2
where f is the function defined by equation (9.84). Find the numerical
value of both the EOF and the dispersion coefficients for κh = 1 and
κh = 100. Which type of flow is better for achieving minimum dispersion?
9.18 This is an open-ended exercise. Write a research report on nonlinear
electrokinetic phenomena. Pay special attention to electro-osmosis of the
second kind and induced charge electro-osmosis. How are they similar
and/or different?
 10 Essential Numerical Methods

10.1 Introduction

Mechanical engineers design new products for consumer use: engines for auto-
mobiles, airplanes, and other devices; cars; air-conditioners; heat pumps; com-
pressors; fans; hair dryers; and all sorts of other products. Increasingly, mechan-
ical and chemical engineers are involved in the design of biomedical devices for
drug delivery systems, biochemical sensing, and rapid molecular analysis. In this
chapter, the basic numerical techniques used to provide design and performance
criteria of these devices are described. A simplified view of a general design
process is depicted in Figure 10.1.
In this chapter, we shift gears a bit and discuss some basic concepts associated
with numerical methods.1 These methods are required when no simple analytical
solution is possible. What is meant by the term analytical is that no solution can
be found in terms of simple functional forms such as the polynomial, trigono-
metric, exponential, logarithmic, or hyperbolic functions. In the following, much
attention is focused on the basic methods required to solve a nonlinear ordinary
differential equation. This requires several different capabilities:
r Numerical differentiation
r Solving sets of linear(ized) equations
r Numerical integration

There are many situations for which numerical methods are required in micro-
and nanofluidics. For example, determining the dependence of the ζ potential
on pH in Chapter 7 requires a numerical zero-finding technique. The non-linear
Poisson equation for the electrical potential discussed in the preceding chap-
ter requires a numerical solution for the potential. In general, the solution of
the potential equation for multicomponent and multivalent mistures requires
a numerical solution. Indeed, the same is true of the momentum equation in
these cases if the analogy between the velocity and potential cannot be invoked.

1
This chapter is dedicated to Professor James David Allan Walker, who taught me all I know about
numerical analysis.

348
 349 Essential Numerical Methods

Feasibility Study PT

Conceptual Design

Analytical Design CAE

Detail Design CAD

Testing

Manufacture

A simplified view of a typical device design process. The abbreviation PT means “programming tool,” which
Figure 10.1
could be a self-contained toolbox such as Matlab or a self-written computer program such as FORTRAN or
C++.

I am sure that the astute reader will note that experimental data acquired in
micro- and nanofluidics will require processing using curve-fitting techniques,
just like at the macroscale.
The foundation of numerical analysis is the concept of function approximation
embodied in the Taylor series. Any continuous and differentiable function has
a Taylor series valid locally around a given point, and this concept is discussed
first. This is followed by the discussion of zero finding and interpolation and
the most common means of comparing with sets of experimental data: curve
fitting.
Numerical differentiation and integration formulas are then derived from the
Taylor series approximation to a given function. Next, the various methods of
solution of linear systems of equations, which arise in the discretization of ordi-
nary differential equations of the boundary value type, are presented. Next is
a discussion of the solution of initial value problems so common in problems
involving chemical and biochemical kinetics and in the solution of problems
governed by a constant Hamiltonian (i.e., constant energy). An example is the
numerical solution of the many-body problem that is the core of molecular dynam-
ics simulations; these problems often require the energy to remain constant, a
constraint that is satisfied by what are called symplectic integrators. The con-
cept of symplectic integrators is seldom, if ever, discussed in numerical methods
courses taught in engineering.
 350 Essentials of Micro- and Nanofluidics

Following this section, the numerical solution of the Poisson–Nernst–Planck

system is described as an example. Finally, the chapter ends with a short discussion
of partial differential equations and a section on verification and validation of
numerical solutions, essential tools in determining the accuracy of a model for
describing complex physical systems.
In presenting the material in this chapter, some judgment had to be made as
to which methods to include. The basic methods in each area described here
work for the vast majority of problems that the student is likely to encounter, and
these are small enough to fit nicely in the Matlab environment. The only area of
numerical analysis that is not covered here is eigenvalue methods, primarily for
lack of space.
There are many books on this subject, and good introductions to these topics,
including eigenvalue problems, are given in the books by Fausett (2008) and
Gilat and Subramaniam (2008), written for undergraduate engineers. There are
many more advanced-level books that include an introduction to computational
fluid dynamics, and these texts are far beyond the scope of this book. A Google
search for “numerical analysis” will yield thousands of hits, including archival
papers discussing special numerical methods for more complicated problems.
These types of methods are not discussed here.
For many students, this chapter may be a review and can be skipped. Conversely,
because numerical analysis permeates much of the material in this book, this
chapter is still useful as a reference for those students whose memory has faded
a bit.
Throughout this chapter, example Matlab files will be introduced, which the
students can use to begin writing their own “m-files.” In addition, for most of the
subjects addressed in this chapter, Matlab has its own built-in modules. These
are discussed where appropriate as well.

10.2 Types of errors

Before getting into Taylor series and numerical differentiation, we note that there
are several types of errors associated with numerical computation. The two types
of fundamental errors are the following:

1. Round-off error: this is the error incurred when a computer formulates

a floating point number (2.36540) with a fixed word length. On most
computers, each number is represented by a 64-bit word length.
2. Truncation error: this error is associated with the particular mathematical
formula chosen, a formula usually based on truncation of a Taylor series.
This type of error can and should be minimized, within reason.

The first type of error is associated with how a computer represents a number.
Numbers on a computer are represented in base 2, and thus all of the numbers
 351 Essential Numerical Methods

in this representation are zeros and ones. For example, the number 6.25 is repre-
sented in base 2 as

2ˆ4 2ˆ3 2ˆ2 2ˆ1 . 2ˆ{-1} 2ˆ{-2} 2ˆ{-3}

0 0 1 1 . 0 1 0
0 + 0 + 4 + 2 + 0 + .25 + 0

Actually, there are many more spaces for digits, and in modern computers, 64
bits are actually available for number representation. Of these 64 bits, 52 spaces
are used for the number, 11 for the exponent, and 1 for the sign of the number
(Gilat and Subramaniam, 2008; Fausett, 2008). Very large numbers and very
small numbers are difficult to represent because of the finite number of digits
(52) that can represent the number on a 64-bit machine. A large number may need
to be chopped or rounded; for example, 2/3 can be written as 0.6666 or 0.6667.
In the first case, the number is chopped, and in the second case, the number is
rounded up. The error associated with these procedures is termed round-off error.
Round-off errors do not often arise in the type of computations described in this
book.
Conversely, truncation errors are associated with the order of numerical approx-
imation to a given function or functional operation such as differentiation. For
example, truncation error occurs in representing a function by a three-term Taylor
series because a Taylor series contains an infinite number of terms. Truncation
errors will be discussed extensively in this chapter.
In addition, there are other, more subtle types of errors that involve some
decisions on the part of the practitioner. These errors include the following:
r Experimental error that shows up in the original data set: this is most often
associated with the uncertainty in experimental data (for which there should
be error bars) that may need to be curve fit
r Blunders or programming errors
r Propagated error or error that builds up as the calculation progresses: this is the
case in solving first-order ordinary differential equations (ODEs) and parabolic
partial differential equations (PDEs).

In this chapter, we will be most concerned with truncation errors, blunders, and
propagated errors. Round off error is usually not a factor, but we need to know
that it exists.

10.3 Taylor series

So much of numerical analysis is based on Taylor series that it is useful to review it

now. Taylor series play a crucial role in zero finding, interpolation, differentiation,
and integration.2

2
For many students, this section may be skipped.
352 Essentials of Micro- and Nanofluidics

Table 10.1. Taylor series of f (x ) = e x

x ex 1 term 2 term 3 term
0.3 1.3499 1.0000 1.3000 1.3450
0.2 1.2214 1.0000 1.2000 1.2200
0.1 1.1052 1.0000 1.1000 1.1050

The Taylor series of a function f (x) is a polynomial representation such that

the coefficients are specified in terms of the derivatives of the function f (x)
in a given range of the independent variable x. Suppose that the behavior of a
function, the Taylor series of a function about the point x = c, is required. Then
the Taylor series of f is defined as
(x − c)2 (x − c)3
f (x) = f (c) + f  (c)(x − c) + f  (c) + f  (c) + · · · (10.1)
2! 3!
This is an infinite series; obviously, an infinite number of terms cannot be cal-
culated, and the series will be truncated at the particular number of terms that
will give the accuracy required. This truncation process leads to the so-called
truncation error. Thus, for a series of N + 1 terms,
N
f (k) (c)
f (x) ∼
= (x − c)k (10.2)
k=0
k!

where f (0) (x) = f (x) and N is an integer. Very often, if c is close to x, then N
can be as small as 2 or 3.
For example, consider the exponential function f (x) = e x . Here take c = 0,
and the approximating series is known as the MacLaurin series. Here f (0) = 1;
f  (0) = 1 and f  (0) = 1, and so
x2
ex = 1 + x + + ··· (10.3)
2
Now suppose it is necessary to find the value of e.3 using the Taylor series. This
is in anticipation that the values of much more complicated functions may be
required. Table 10.1 shows the result of a calculation. Note that the closer the
evaluation point is to zero – the expansion point – the better the approximation.
The Taylor series of a function is really a way to approximate the function
in a particular region. To be accurate, the Taylor series must converge, and so
the point of evaluation, that is, x, must be somewhat near the expansion point c.
We should not calculate e1 using the preceding MacLaurin series around x = 0
unless we are preppared to take a large number of terms or take c closer to x = 1.
Indeed, a Taylor series converges rapidly near the point of expansion, but it may
converge slowly or not at all far from that point.
When f (x) is given as a numerical function, values of f (x) between the given
points may be calculated using Taylor series. Table 10.2 shows a set of data that
is the output of a set of experiments to measure temperature at several points in a
 353 Essential Numerical Methods

Table 10.2. Temperature as a function of position in a solid

x (m) 0 0.5 1.0 1.5 2.0
T (o C) 15.00 12.70 10.00 7.32 5.00

solid. The heat flux q or the heat transfer rate per unit area is proportional to the
derivative of the temperature, with the proportionality constant being the thermal
conductivity. To calculate the heat flux at x = 1 m, the Taylor series is calculated
about x = 1 m; that is, c = 1 m. Then

T (x) = T (1) + T  (1)(x − 1) + · · · (10.4)

What is x in this equation? Well, we can take x to be any one of the values in Table
10.2, but it is better that it be close to x = c = 1 m. So let us take x = 1.5 m, and
then the only unknown in the equation is T  (1), which is proportional to the heat
flux there. Thus, after truncating the Taylor series after two terms,
T (1.5) − T (1) 7.32 − 10 C
T  (1) ∼ = = −5.36 (10.5)
1.5 − 1 0.5 m
This is only an approximation, but it will get better as the x = 0.5 gets smaller.
As will be seen later, this is the forward difference approximation to the derivative.

10.4 Zeros of functions

10.4.1 Numerical methods
Consider the following question: for a given function f (x), for what values is
the equation f (x) = 0? There are a wide range of applications for which this
is necessary. In most applications, by preliminary analysis of the problem, the
approximate location of the zero(s) will be known, and this feature will help
in finding the solution. An example of a function with two zeros is depicted
in Figure 10.2. In this section, methods to find the real zeros of functions are
discussed.
As an example of the need for finding a zero of a given function, consider the
Redlich–Kwong equation of state as an extension of the ideal gas law (Moran &
Shapiro, 2007), which is
RT a
p= − (10.6)
V̄ − b V̄ T 1/2 (V̄ + b)

where a and b are constants that depend on the gas considered. Here V̄ is the
molar volume in m3 /kmole. The constant b is intended to account for the finite
volume of the molecules neglected in the perfect gas model, and the constant
a is meant to account for intermolecular forces. This equation is substantially
empirical in nature, with a and b determined by curve fits to experimental data,
 354 Essentials of Micro- and Nanofluidics

f(x)

zero

a b x

Sketch of a function with two zeros.

Figure 10.2

and so, empirically, it can also be used for liquids. For example, in metric units
for water,
 
m3 m3
a = 142.59 bar K1/2 and b = .021 (10.7)
kmole kmole

Note that given the pressure and temperature, the molar volume V̄ cannot be
determined analytically. After multiplication by the two denominators, equation
(10.6) becomes a polynomial of order 3. For this case, there are explicit formulas
for the roots of a third-order polynomial of the form

x 3 + Ax 2 + Bx + C = 0 (10.8)

and these are given in a table in Appendix B.3 of the book by Murray (2001).
The form of the roots can get complicated as the order of the equation increases,
and in many cases, it is easier to calculate the roots numerically.
Three methods of numerically finding zeros are described here:
r Bisection
r Secant method
r Newton’s method

Before implementing any of these methods, it is useful to graph the func-

tion because in any of these methods, a good initial guess is crucial to rapid
convergence. Only the case of simple zeros is discussed here, in which

f (x) = (x − x 0 )F(x), F(x 0 ) = 0 (10.9)

BISECTION
The idea in this method is to search for an interval in which f (a) × f (c) < 0, in
which case, for any continuous function f , there must be a zero between a and c.
This situation is depicted in Figure 10.3. Assuming that a zero has been bounded
 355 Essential Numerical Methods

f(x)

x =a
1

α x2 =c
x

Sketch of a function with a single zero. Here α

Figure 10.3
denotes the location of the zero.

by graphing, for example, a pseudo-Matlab script might go something like

this:

Set x_{1}=a and x_{2}=c

Set x_{3}=\frac{(x_{1}+x_{2})}{2}
Calculate f(x_{3})
Is f(x_{3}) < \epsilon where \epsilon is a pre-assigned
tolerance?
If yes, stop and the zero, f(\alpha)=0, \alpha=x_{3}
If no, is f(x_{1}) \times f(x_{3}) <0?
If yes, x_{2}=x_{3} go back to line 2
If no, x_{1}=x_{3} go back to line 2

The advantage of bisection is that it is simple; the disadvantage is that convergence

is slow.

Notation alert: In this and other sections in this chapter, the symbol  is used
for the numerical preassigned tolerance. A good starting value is  = 10−4 .

SECANT METHOD
In this method, the calculation of x3 (Figure 10.4) is different. Passing a straight
line through x 1 = a and x 2 = c,
f (x) − f (x2 ) f (x 2 ) − f (x1 )
= (10.10)
x − x2 x2 − x1
and evaluating at x = x3 , where f (x 3 ) = 0,
(x 2 − x1 ) f (x 2 )
x3 − x2 = − (10.11)
f (x 2 ) − f (x1 )
Note that the zero need not be bounded for the secant method to work. The
advantage of this method is that it is faster than bisection, and the only calculation
that is different from the bisection procedure is the way x3 is calculated.
 356 Essentials of Micro- and Nanofluidics

f(x)

x =a
1

α x2 =c
x
x3

The secant method.

Figure 10.4

NEWTON’S METHOD
Here again assume that a zero exists at x 3 and that the guess, x, is close to x 3 . In
Newton’s method, the zero need not be bounded and we take the equation of the
straight line tangent to the curve at x to generate an approximation for x 3 (Figure
10.5). Then, expanding the function f in a Taylor series and chopping after one
term, the approximation to the root is

f (x 3 ) = f (x) + f  (x)(x3 − x) (10.12)

Because f (x 3 ) = 0, the estimate of the zero is

x 3 = x − f (x)/ f  (x) (10.13)

Putting this equation in iterative form, the n + 1st iterate is given by

xn+1 = x n − f (x n )/ f  (xn ) (10.14)

f(x)

α x
x

x3

Newton’s method for finding zeros.

Figure 10.5
 357 Essential Numerical Methods

0.5

f(x)
−0.5

−1

−1.5

−2
0 0.5 1 1.5 2 2.5 3
x

Sketch of f (x ) = log10 (1 + x 2 ) − 2e−x .

Figure 10.6

The advantage of this method is that it is faster than the secant method; a dis-
advantage is that f  (x) must be known, and the scheme may not converge if the
guess is not close enough. Newton’s method is ideal if f  (x) can be calculated
analytically.

MATLAB FUNCTION: x = fzero(‘function’, guess, tol)

This Matlab function uses a combination of bisection and the secant method
along with inverse interpolation. Consider the function

f (x) = log10 (1 + x 2 ) − 2e−x (10.15)

A Matlab script to plot the function may look like the following:

% Zero finding example

dx=0.2
xend=3
x0=0
x=x0:dx:xend
y0=log10(1+x.*x) -2.*exp(-x)
plot(x,y0)

Figure 10.6 indicates a zero near x = 1.5. An m-file defining the function will
look as follows:

function y=zeroex1(x)
y=log10(1+x.*x)-2.*exp(-x)

and this file is called zeroex1.m, created in the Matlab edit window. Now use the
Matlab command window to find the zero:

>> fzero(’zeroex1’,1.5)

Here the tolerance has not been specified, and Matlab provides the value intrin-
sically. The program iterates, and the output is
358 Essentials of Micro- and Nanofluidics

Zero found in the interval: [1.4151, 1.56]

ans=

1.4244

>>

Working in the command window in Matlab is the quickest way to get an

answer to this simple problem. However, once Matlab is shut down, the result is
lost. Writing a Matlab program called an m-file, which can be saved, is the most
efficient way to deal with this issue.
As an example, returning to the Redlich–Kwong equation of state, for
carbon monoxide gas at T = 215 K, a = 17.26 bar m3 K1/2 /mole2 , b =
0.02743 m3 /mole, and p = 69.2 bar. The gas constant is 0.08314 KJ/Kgmole◦ K.
Using the secant method with (a, c) = (0, 0.5) with a tolerance of  = 0.001,
V̄ = 0.2268 m3 /kg. There are no other roots in the interval (a, c) = (0, 10). If
other real roots occur, then one has to choose the correct root on physical grounds.
This example could also have been solved using the methods for polynomials
discussed next.
These methods are not confined to analytically defined functions. For example,
the bisection method may easily be applied to a numerical function. Newton’s
method can be as well, except f  would be computed numerically.

10.4.2 Polynomials
A polynomial has the generalized form

N +1
PN (x) = an x N +1−n = a1 x N + · · · + a N x + a N +1 (10.16)
n=1

The following statements may be made about polynomials:

1. PN (x) has N zeros; they may be real and distinct, real and repeating, or
complex. If the coefficients an are real, the complex roots occur in conjugate
pairs.
2. One can estimate the location of all real roots of PN (x) = 0 by plotting
PN (x) for −1 ≤ x ≤ 1 and PN (z) for −1 ≤ z ≤ 1 with z = 1x .
3. The largest real root of PN (x) is approximated by the largest root of

a1 x + a2 = 0 and a1 x 2 + a2 x + a3 = 0 (10.17)

4. The smallest root is approximated by the smallest root of

a N x + a N +1 = 0 and a N −1 x 2 + a N x + a N +1 = 0 (10.18)

5. The sum of all the roots is −a2 /a1

 359 Essential Numerical Methods

Any of the methods described here may be used to find the zeros of a poly-
nomial. In the case of multiple real roots, the value of the zero must be isolated
based on the initial guess of the local zero. This information may be obtained by
plotting the function.

MATLAB FUNCTION: r = roots( p)

Suppose the root of f (x) = x 3 + x 2 − 3x − 3 is desired. Then an m-file in the
edit window to find the roots may look as follows:

>> p=[1 1 -3 -3]

>> r=roots(p)

The vector or array p comprises the coefficients of the polynomial. In the com-
mand window, execute the m-file by invoking its name “> name1,” and in the
command window, it is found that

p =

1 1 -3 -3

r=

-1.7321
1.7321
-1.0000

As mentioned previously, the Redlich–Kwong equation of state can be put in

polynomial form as
 
a a
p V̄ − RT V̄ − b + RT b + 1/2 V̄ +
3 2 2
=0 (10.19)
T bT 1/2
This equation, given the constants a, b and the temperature T , can be solved using
the roots program. In this case, the roots routine can be called from a program
that calculates the constants in the polynomial, unlike the preceding example, for
which the constants are given. In these types of problems, care should be taken
that all the terms in the equation have the same units.

10.5 Interpolation

Interpolation is the process of evaluating a function at a point where the function

is not specifically known. The need arises when measurements are made at
discrete points or when solutions to problems are obtained numerically. Clearly
measurements of a specific quantity or a computation of a numerical solution at an
infinite number of points cannot be obtained. The function may be approximated
by a polynomial or by some other set of functions such as cubic splines. In this
360 Essentials of Micro- and Nanofluidics

Table 10.3. A function f

tabulated at a discrete
number of points
x0 f0
x1 f1
x2 f2
x3 f3
x4 f4

section, polynomial interpolation is described and then used to develop the basic
numerical integration formulas.
Notation alert: The symbol h in this chapter is generally used for the spacing
of the mesh points at which a function is tabulated, as is usually the convention
in numerical analysis. Recall that the symbol h has been heretofore used for the
height or half-height of a micro- or nanochannel.

10.5.1 Linear interpolation

Suppose a function f is tabulated at a discrete number of points, as in Table 10.3,
and it is necessary to find the value of the function at some point x, x0 < x < x1 .
Such a value of x can be expressed as

x = x0 + s(x1 − x0 ) = x 0 + sh for 0 < s < 1 (10.20)

where it is assumed that the data are tabulated at equal intervals, defined as
h = xi − xi−1 . The simplest function that can be used to evaluate the function f
between any two points is a straight line, and for x0 < x < x1 , the straight line is
defined by
x1 − x x − x0
p(x) = f (x0 ) + f (x1 ) (10.21)
x1 − x0 x1 − x0
and this case is depicted in Figure 10.7a. This equation can be generalized for
any of the intervals [xi−1 , xi ], and the result is
xi − x x − xi−1
p(x) = f (xi−1 ) + f (xi ) (10.22)
xi − xi−1 xi − xi−1
Note that by design, the function p agrees with the function f at xi−1 and xi .
If the points are tabulated at equal intervals, equation (10.21) is simplified to
 
f1 − f0
p(x) = p(x0 + sh) = f 0 + sh (10.23)
h
and now this equation looks like the first two terms of a Taylor series of f about
x0 . Here s is determined by the value x and is a known quantity. To make this
equation accurate, sh must be small; that is, the interval h should be as small as
possible.
 361 Essential Numerical Methods

N=1
N=2

f p(x) y

x0 x1 x
x
(a) linear interpolation (b) quadratic interpolation

Sketch of interpolating polynomials: (a) linear interpolation; (b) quadratic interpolation. The open symbol in (b)
Figure 10.7
denotes a functional value at a point in between the tabulated points.

Let us find f (1) and f (2.6) for the data set in Table 10.4 using linear interpo-
lation. Note that the data are not tabulated at equal intervals. Furthermore, it is
clear that f (1) = 9 from the linear interpolation formula; that is, f (1) is just the
average of f (0) and f (2). Now, for x = 2.6, x0 = 3, and s = −0.4,
 
f1 − f0
p(2.6) = p(x0 + sh) = f 0 + sh
h
= 28 − 0.4(1)(28 − 11)/1 = 28 − 6.8 = 21.2 (10.24)

10.5.2 The difference table

Suppose a function is tabulated at equal intervals in the form shown in Table 10.5.
Then a difference table can be formed by calculating successive differences.
Each column is composed of a difference between two previous differences.
Let us calculate the difference table for f (x) = x 2 . The results are depicted in
Table 10.6.
Note that the third differences are zero; in general, for a polynomial of degree
N , the N + 1st differences are all zero. The difference table is especially useful
for assessing the order of polynomial by which a set of data may be approximated.
As in Table 10.6, the successive differences must decrease for a well-behaved
function. Clearly the difference table can also be used to assess the accuracy and
consistency of an experimental data set. The difference table may also be used
for extrapolation or trying to predict the value of a function outside the range of
the tabulated data. This is risky, however, and should be avoided. The difference
table can even be used to assess the accuracy of a discretization of an ordinary or
partial differential equation, as will be seen.

Table 10.4. A sample set of tabular data

x 0 2 3 4
f 7 11 28 63
362 Essentials of Micro- and Nanofluidics

Table 10.5. The difference table

x0 f0  2 3 4
x1 f1 f1 − f0
x2 f2 f2 − f1 f 2 − f y1 + f 0
x3 f3 f3 − f2 f 2 − 2f 1 + f 0 .
x4 f4 f4 − f3 f 2 − 2f 1 + f 0 . .

10.5.3 Lagrangian polynomial interpolation

In general, if a function is defined by N + 1 points, the function can be approxi-
mated by a polynomial of degree N whose values agree with the numerical func-
tion at the nodal points. If the data set in Table 10.3 is considered, a quadratic func-
tion is passed through any three points. Then three equations in three unknowns
can be written to find each of the coefficients ai in the polynomial:
p(x) = a3 + a2 x + +a1 x 2 (10.25)
The coefficients ai can be calculated directly, and the result is in analogy with
equation (10.22):
(x − x1 )(x − x2 ) (x − x 0 )(x − x 2 ) (x − x0 )(x − x 1 )
p(x) = f0 + f1 + f2
(x0 − x1 )(x0 − x2 ) (x1 − x0 )(x1 − x2 ) (x2 − x0 )(x2 − x1 )
(10.26)
Note that the points need not be equally spaced.
The general form of an interpolating polynomial of order N is

N
p N (x) = li (x) f (xi ) (10.27)
i=0

where
   
x − x0 x − xn
li (x) = ... (10.28)
xi − x0 xi − xn
Note that this expression is a product of N factors, and so li is a polynomial of
degree N . The denominators of the li are just numbers, and p N (x j ) = f (x j ), that
is, the interpolating polynomial matches the function at each of the grid points.
Suppose there are 10 points running from x = 1 to x = 10. Then, theoretically,

Table 10.6. The difference table for f (x ) = x 2

x f (x )  2 3 4
0 0
1 1 1
2 4 3 2
3 9 5 2 0
4 16 7 2 0 0
 363 Essential Numerical Methods

N=4 N=2

y y

x x
(a) a bad interpolating polynomial (b) a good interpolating polynomial

Sketch of interpolating polynomials. As the order of the polynomial increases, the deviation between the
Figure 10.8
functional points and the interpolating polynomial may be large; N is the number of intervals. (a) In this case,
the interpolating polynomial is not accurate. (b) A quadratic polynomial locally fitted between the first three
points is much better.

a polynomial of degree 11 could be fit through all of the points. However, this is
dangerous, and if the value of f at x = 2.5 of Table 10.4 is desired, for example,
the best solution is to locally put a quadratic polynomial through three nearby
points. If the interpolating polynomial is of too large a degree, it may oscillate,
as depicted qualitatively in Figure 10.8.
The general form of the polynomial in equation (10.27) includes the case of
unequal intervals. In this case, a divided difference table may be formed. Given a
function tabulated at two points b > a, the divided difference is given by d =
f b − f a /b − a. Each column proceeds as in the difference table calculation; one
computes 2d , 3d , and so on.

10.5.4 Newton interpolation formulas

The Lagrangian formulas given earlier can be simplified, and consider the
quadratic interpolation formula. In this section, assume that the numerical func-
tion is tabulated with a uniform spacing. Given three points, say, x 0 , x 1 , and x2 ,
the task is to find the value of the function at a point x = x0 + sh, where h is
constant. Then, substituting into equation (10.26) and after some algebra, the
interpolating polynomial p(x) is given solely in terms of s by

s(s − 1)
p(x) = f 0 + s( f 1 − f 0 ) + ( f 0 − 2 f 1 + f2 ) (10.29)
2

where 0 < s < 2. This equation can be written in terms of the entries in the
difference table as

s(s − 1) 2
p(x) = f 0 + s f 0 +  f0 (10.30)
2

where, for example,  f 0 = f 1 − f 0 .

364 Essentials of Micro- and Nanofluidics

Table 10.7. Error associated with the Newton

interpolation formulas
Formula Error
h3
Forward 3!
s (s − 1)(s − 2)y (3) (ξ )
h3
Backward 3!
s (s + 1)(s + 2)y (3) (ξ )
h3
Central 3!
s (s 2 − 1)y (3) (ξ )

The variable ξ is a point between x 0 and x 2 whose value

is unknown.

Note that one term has been added to the linear interpolation formula. Equation
(10.30) is called a forward formula because the formula uses points ahead of x0 .
Recall that sh should be as small as possible. If, for example, the required
interpolation point is near x2 , this formula would not be appropriate because s is
too large. Near x 2 , the interpolating point is given by x = x2 + sh and now s < 0
so that
s(s + 1)
p(x) = f 2 + s( f 2 − f 1 ) + ( f0 − 2 f1 + f2 ) (10.31)
2
where −2 < s < 0. This formula is called a backward formula because the points
behind x2 are used as a way of keeping s small.
There is also a central formula, this is

s s2
p(x) = f 1 + ( f 2 − f0 ) + ( f 0 − 2 f 1 + f 2 ) (10.32)
2 2
where now −1 < s < 1 and x = x 1 + sh.
Which of the forward, backward, or central formulas is used depends on where
the function is to be evaluated in a given table of values. Near the beginning of
the difference table, use the forward formula; near the end, use the backward
formula; and away from either end, use the central formula. The central formula
should always be used, if possible. There are truncation errors associated with
both the Newton and Lagrangian formulas; the error associated with the Newton
formulas is depicted in Table 10.7.
As an example, Yoda and coworkers (Sadr et al., 2006) have measured the fluid
velocity near the wall for electro-osmotic flow in a micron-sized channel with
the results depicted in Table 10.8. The ionic strength in the upstream reservoir is
37 µM, and the surface charge density of the channel is σ = 8.63 × 10−4 C/m2 .
The Debye length is estimated at λ = 18 nm. There are nine points here, and
theoretically, a 10th-order polynomial could be fit to these nine points.

Table 10.8. Experimental data for the velocity near the wall in electro-osmotic flow in a
microchannel (Sadr et al., 2006)
y(nm) 0 30 60 90 120 150 180 210 240
m
u( sec ) × 10 5
0 0.33 2.01 2.33 2.51 2.61 2.65 2.69 2.69
365 Essential Numerical Methods

To illustrate the use of Newton’s formula, suppose that the value of the velocity
at y = 100 nm for Yoda’s data in Table 10.8 is required. First consider the forward
formula with (x0 , x1 , x2 ) = (90, 120, 150). Then s = (100 − 90)/30 = 0.33, and
the linear formula gives
m
p(100) = 2.33 + 0.33(2.51 − 2.33) = 2.39 × 10−5 (10.33)
sec
For the quadratic forward formula, the result is
0.33(0.33 − 1)
p(100) = 2.33 + 0.33(2.51 − 2.33) + (2.33 − 2 × 2.51 + 2.61)
2
m
= 2.40 × 10−5 (10.34)
sec
Finally, for the central formula x 0 = 60, x 1 = 90, and x 2 = 120, and again for
s = 0.33, we have
.33 .332
p(100) = 2.33 + (2.51 − 2.01) + (2.01 − 2 × 2.33 + 2.51)
2 2
m
= 2.40 × 10−5 (10.35)
sec
Note that the differences between the linear and the quadratic formulas are small
and that the forward and central formulas give the same result. Thus this data set
is well behaved. Note that only two digits have been kept in the solution because
the data are only accurate to two significant digits. This error in the data thus
produces error in the estimate of the velocity at y = 100 nm.

10.5.5 Matlab interpolation functions

Matlab has two functions that can be used for interpolation. One is polyfit, called
in the form p = polyfit(x, f, m), where p is the output list of coefficients for the
best fit to the function and m is the input order of the interpolating polynomial.
The function polyval( p, xrange) then evaluates the polynomial at any point x. As
with the zero-finding routines, it can be used in the command line of Matlab or
in a Matlab m-file.
The function interp1 calculates the interpolated value of a function at a single
point. The format is f i = interp1(x, f, xi, method ), where x and f have the
same meaning as in this section, xi is the value of x corresponding to the value
of f i, and  method is the method of interpolation, which can be one of several
methods, such as quadratic interpolation. If no method is specified, then linear
interpolation is used.
A data set that contains oscillations is not a well-behaved data set. To illustrate
the problem associated with oscillations, consider the data set in Table 10.9. The
Matlab program looks as follows:
n=5;
x=[0 2 4 5 6 7];
 366 Essentials of Micro- and Nanofluidics

Table 10.9. Table of numerical data to illustrate the problem with oscillations in a data set
x 0 2 4 5 6 7
y 0 0.58 0.34 0.33 −0.28 −0.53

y=[0 .58 .34 .33 -.28 -.53];

x0=0;
dx=0.001;
xend=7.0;
xrange=x0:dx:xend;
p=polyfit(x,y,n5);
f=polyval(p,xrange);
plot(x,y,’o’,xrange,f);
xlabel(’x’);ylabel(’y’);
title(’Plot of data points and functions’);
legend(’Actual Data’,’Polyfit’);grid;

Note that there are five intervals, and thus a fifth order polynomial (n = 5) can
theoretically be fit through the five points. The result of this is shown in Figure
10.9. Note the significant oscillation in the first, second, and last intervals. Such a
curve is clearly not accurate for 0 < x < 4. This is clearly not a very good result,
but the situation is improved by the use of cubic splines, which is considered
next.

10.5.6 Cubic spline interpolation

Straightforward polynomial interpolation of two types has been discussed. First
was polynomial interpolation over N + 1 points. For N sufficiently large, it has
been shown that polynomial interpolation of the function over the entire range can

1.5
Actual Data
Polyfit
1

0.5
y

−0.5

−1
0 1 2 3 4 5 6 7
x

Using a single fifth-order polynomial to interpolate a data set is not always best.
Figure 10.9
 367 Essential Numerical Methods

Si
y(x) y(x)

S
i−1

x
x xi x i+1
x i−1
(a) (b) cubic spline interpolation

(a) Sketch of a series of polynomial approximations. (b) Two successive intervals for the cubic spline
Figure 10.10
approximation.

lead to oscillatory errors. Second, piecewise polynomial interpolation has been

discussed and is endorsed. The trouble with piecewise polynomial interpolation
is that the interpolating polynomials are not differentiable at the end points of
each interval, as shown in the schematic of Figure 10.10a.
There is an alternative to this type of interpolation: cubic splines. A spline
was a draftman’s device used to draw smooth curves. A smooth cubic polyno-
mial is used in each interval. For technical reasons, odd powers of polynomials
seem to work better than even powers. For example, in quadratic spline interpo-
lation, the second derivative is still discontinuous at the grid points or “knots.”
There are many other kinds of splines, but we confine ourselves to the cubic
spline.
The cubic spline interpolation procedure begins with two successive intervals
[xi−1 , xi ], and [xi , xi+1 ], i = 1, N where N is the number of intervals (Figure
10.10b); then the cubic spline interpolating function is

Si (x) = ai (x − xi )3 + bi (x − x i )2 + ci (x − xi ) + di (10.36)

valid for i = 1, . . ., N . The coefficients are determined by the boundary condi-

tions at the end of each interval.
Now, in each interval, [xi , xi+1 ], let z i = S  (xi ), the second derivative of the
spline function evaluated at x i . Note that S  is a linear function that takes the
values z i at x = xi and z i+1 at x = xi+1 Clearly Si can be written
z i+1 zi
Si (x) = (x − x i ) + (xi+1 − x) (10.37)
hi hi
and integrating twice,
z i+1 zi
Si (x) = (x − xi )3 + (xi+1 − x)3 + C(x − xi ) + D(xi+1 − x) (10.38)
6h i 6h i
368 Essentials of Micro- and Nanofluidics

To find the constants C and D, note that at the grid points, the spline must agree
with the functional values of y and

Si (xi ) = yi (10.39)

Si (xi+1 ) = yi+1 (10.40)

which gives two equations in two unknowns for C and D. The result is
 
z i+1 zi yi+1 h i z i+1
Si (x) = (x − xi ) +
3
(xi+1 − x) +
3
− (x − xi ) (10.41)
6h i 6h i hi 6
 
yi+1 h i zi
+ − (xi+1 − x) (10.42)
hi 6
This is the result for the cubic spline; however, the z i , z i+1 are unknown, and we
need to be able to calculate them. To do this, differentiating equation (10.42),
z i+1 zi yi+1 h i z i+1 yi h i zi
Si (x) = (x − x i )2 + (xi+1 − x)2 + − − +
2h i 2h i hi 6 hi 6
(10.43)
and evaluating at x = xi ,
zi h i z i+1 h i yi+1 − yi
Si (xi ) = − − + (10.44)
3 6 hi
Similarly,

 zi−1 h i z i h i−1 yi − yi−1

Si−1 (xi ) = + + (10.45)
6 3 h i−1
The advantage of splines over direct polynomial interpolation is that the inter-
polating function derivative can be made continuous at the mesh points or knots
xi by requiring

Si (xi ) = Si−1


(xi ) (10.46)

and so

h i−1 z i−1 + 2(h i−1 + h i )z i + h i z i+1 = 6(bi − bi−1 ) (10.47)

where
yi+1 − yi
bi = (10.48)
hi
Equations (10.47) are a system of tridiagonal linear equations and are valid for
i = 2, N − 1; the points i = 1 and i = N are the end points. We will be solving
systems of this type later in this chapter. At the end points, because there are
no continuity conditions, we must specify something about the z i . The most
common cubic spline conditions are that z 1 = 0, z N = 0. Other conditions can
also be used: z 1 = z 2 , z N = z N −1 , which is a first order approximation to a zero
derivative at the end points. The results are usually not affected greatly by use of
either of these conditions.
369 Essential Numerical Methods

Let us interpolate the data set in the following Matlab program to find
y(.7), y(1.7), y(4.7) using cubic splines using the Matlab spline function. The
Matlab program looks as follows:

x = [0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0];
y = [0.0 0.03060 0.11490 0.23209 0.35283 0.44606 0.48609
0.45863 0.36413 0.21785 0.04657];
y_spline1 = interp1(x,y,1.7,’spline’);
y_spline2 = interp1(x,y,0.7,’spline’);
y_spline3 = interp1(x,y,4.7,’spline’);
disp(’Interpolation Estimates’);
disp(’ x ’);
disp(’ 1.7 ’);
disp(’ y ’);
disp(y_spline1);
disp(’ x ’);
disp(’ 0.7 ’);
disp(’ y ’);
disp(y_spline2);
disp(’ x ’);
disp(’ 4.7 ’);
disp(’ y ’);
disp(y_spline3);

Running this program yields the following:

>> splines1
Interpolation Estimates
x
1.7
y
0.2817

x
0.7
y
0.0589

x
4.7
y
0.1507

>>
 370 Essentials of Micro- and Nanofluidics

As another example, consider the data set in the following Matlab program.
Let us find and plot the cubic spline interpolation curve. The Matlab file is similar
to that above, but now the Matlab spline command is used:

xdata=[0.0 0.6 1.5 1.7 1.9 2.1 2.3 2.6 2.8 3.0]
ydata=[-0.8 -0.34 0.59 0.59 0.23 0.1 0.28 1.03 1.5 1.44]
x1=0;
dx=0.1;
x2=3.0;
xspline=x1:dx:x2;
yspline=spline(xdata,ydata,xspline)
plot(xdata,ydata,’o’,xspline,yspline,’-’)
title(’Cubic Spline Curve’)
xlabel(’x-axis’),ylabel(’y-axis’)

The results are shown in Figure 10.11. Note that the curve is smooth and approx-
imates the numerical function reasonably well between the mesh points.

10.6 Curve fitting

Curve fitting is merely another form of approximation of a function. In interpo-

lation, a polynomial was passed through a few (good) or all of the points (bad for
too many points). However, as with all experimental methods, these data will have
some error owing to the experimental method used. Thus it may be unrealistic to
require an approximating polynomial to pass through each data point because the
experiment is only repeatable within the experimental error. Thus a lower-order
polynomial may not pass through any of the points may be appropriate. The
question is, what function gives the best fit? Usually, the number of data points

1.5

0.5
y

−0.5

−1
0 0.5 1 1.5 2 2.5 3
x

Cubic spline interpolation of the data set just discussed. Note that the curve is smooth and differentiable
Figure 10.11
between the mesh points.
 371 Essential Numerical Methods

k Least−Squares
Linear Curve−Fit

Qualitative sketch of the thermal conductivity with temperature and a postulated curve fit.
Figure 10.12

greatly exceeds the order of the polynomial, so the question of how to determine
the coefficients of the polynomial must be addressed.
The results of this section are useful because they will yield one approximating
function for an entire set of data, unlike the interlacing polynomials concept. Fur-
thermore, the function should differ only a little from the data, even if additional
data points are added.
The thermal conductivity (Figure 10.12) of a material is often approximated
by a linear function, as done in Chapter 2:
k = k0 (1 + bT ) (10.49)
with T in ◦ C and k0 and b constant. Let us write this equation in the form
k(T ) = α + βT (10.50)
where α and β are constants fit to experimental data. The question is, how should
α and β be chosen so that the difference between the data and the approximating
function is, in some sense, a minimum?
Let
f i = α + βT (10.51)
and define the error as
ei = ki − f i (10.52)
Define the error E as
E = e12 + · · · + e2N (10.53)
The least squares criterion requires that E be a minimum, and at a minimum,

∂E N
=− 2(ki − (α + βT )) = 0 (10.54)
∂α i=1

∂E N
=− 2Ti (ki − (α + βT )) = 0 (10.55)
∂β i=1
372 Essentials of Micro- and Nanofluidics

These are two equations in two unknowns for α and β, which can be solved by
the methods to be discussed in Section 10.9.
Curve fitting is the most common way of generating an approximate function
to represent a given data set. This is because the experimental data have errors
in them, and also, a curve fit usually leads to a single functional form valid over
all data points. That is not to say that there may be parameter ranges over which
a given set of data requires two or even more functional forms; nevertheless, the
curve-fitting procedure is usually optimal for analyzing experimental data.
A set of data may not be well approximated by a linear curve fit. In this case,
a similar procedure could be used to approximate a data set with a polynomial
curve fit. Moreover, if a given function f (x) ∼ Ax β , then a linear curve fit of the
logarithm of the function is possible:

ln f ∼ ln A + β ln x (10.56)

Matlab’s interpolation functions polyfit and polyval may also be used for
curve-fitting purposes. Matlab recognizes whether the order of the approximating
polynomial is less than N − 1, where N is the number of points in the data set,
and automatically uses the curve-fit procedure.
As an example, Sadr et al. (2006) have measured the electro-osmotic mobility
μe = EUx inside the electrical double layer as a function of reservoir concentration
(Figure 10.13). The velocity U is the average velocity over a distance of about
200 nm from the wall. The data appear in the Matlab file:

conc=[ 0.19 1.9 3.6 18.4 36]; % mM

mob=[ 5.14 2.41 2.02 1.40 1.14]; %mobility x 10ˆ4 cmˆ2/V/s
conc2=log10(conc)
mob2=log10(mob)
p1=polyfit(conc2,mob2,4)
fit1=polyval(p1,conc2);
p2=polyfit(conc2,mob2,2)
fit2=polyval(p2,conc2);
p3=polyfit(conc2,mob2,1)
fit3=polyval(p3,conc2);
plot(conc2,fit1,’-’,conc2,mob2,’o’,conc2,fit2,’*’,
conc2,fit3,’-.’)
xlabel(’log conc’)
ylabel(’log mobility’)

Note that the log of both the concentration and the mobility are being curve
fit with first a fourth-order polynomial through each point (interpolation), then a
quadratic, then a linear fit. Note that all the functions seem to work well as shown
by the results in Figure 10.13. The linear curve fit is defined by

log10 U = −0.2816 log10 c + 0.4856 (10.57)

 373 Essential Numerical Methods

0.8

0.7

0.6

log Mobility 0.5

0.4

0.3

0.2

0.1

0
−1 −0.5 0 0.5 1 1.5 2
log Concentration

Curve fit to the mobility – concentration data taken by Sadr et al. (2006). The solid line is a fourth-order
Figure 10.13
polynomial fit, and the dashed line is the least squares linear curve fit.

and note that the fourth order polynomial is very accurate even away from the
grid points, indicating that this is a well-behaved data set.

10.7 Numerical differentiation

As has been seen many times, there are many instances where numerical differ-
entiation is required. Often a given physical quantity that has been measured is
defined only at a discrete number of points, as was seen in the previous section.
Numerical differentiation is also required in the discretization of ordinary and
partial differential equations. Moreover, certain physical quantities are defined by
derivatives, and thus numerical differentiation is required as part of the numeri-
cal solution of ordinary and partial differential equations. For example, the shear
stress is defined in terms of the derivative of the velocity, and the surface charge
density is defined in terms of the derivative of the electrical potential.
The formulas for the derivative of a given function, analytical or numerical,
come from the Taylor series, as with the various methods of function approxima-
tion such as interpolation.

10.7.1 Derivatives from Taylor series

Consider the function f (x), and recall that the Taylor series approximation to
f (x) near x is given by

h2 h3
f (x + h) = f (x) + f  (x)h + f  (x) + f  (x) + · · · (10.58)
2! 3!
where h is small. Note that this formula is written in a slightly different form
from equation (10.1) for reasons that will become obvious.
374 Essentials of Micro- and Nanofluidics

The forward difference approximation to the derivative is obtained by neglect-

ing the second derivative term and
f (x + h) − f (x)
f  (x) ∼
= (10.59)
h
Evaluating at a discrete value of x, say, xi ,
f i+1 − f i
fi ∼
= (10.60)
h
In a similar way, the backward difference approximation is obtained by replacing
h by −h, and the result is
f i − f i−1
fi ∼
= (10.61)
h
Because the finite difference formulas are approximations, there is error; for
the forward and backward formulas, the error is
h
E∼
= f  (ξ ) (10.62)
2
where ξ is some point satisfying x ≤ ξ ≤ x + h. Note that the error gets smaller
as the grid size h gets smaller.
A more accurate formula, may be obtained by adding the backward formula
and the forward formula, resulting in the central difference approximation:
f i+1 − f i−1
f i ∼
= (10.63)
2h
The error in the central difference approximation is smaller than the error in the
forward or backward difference, being
2
h
E∼
= f  (ξ ) (10.64)
3
This central difference approximation should be used whenever possible.
The second derivative may be obtained from Taylor series in a similar manner.
Adding

h2 h3
f (x + h) = f (x) + f  (x)h + f  (x) + f  (x) + · · · (10.65)
2! 3!
to the corresponding formula for f (x − h) yields the central difference approxi-
mation to the second derivative:
f i+1 − 2 f i f i−1
f i ∼
= + O(h 2 ) (10.66)
h2
It will turn out that derivatives higher than the second will not generally be
required.
The expressions for the first and second derivatives in the forward, backward,
and central differences appear in the interpolation formulas discussed previously.
375 Essential Numerical Methods

10.7.2 A more accurate forward formula for the first derivative

There are times when the more accurate central formula cannot be used. An
example is the calculation of the shear stress at a surface or the surface charge
density in an electrolyte solution. In these cases, the last point in the fluid is on the
boundary, and using a simple forward formula may not be accurate enough, espe-
cially when the function is varying rapidly near the surface. However, the Taylor
series approximations can be manipulated to derive a more accurate formula.
Consider the forward formula with h replaced by 2h:
4h 2 8h 3
f (x + 2h) = f (x) + f  (x)2h + f  (x) + f  (x) + ··· (10.67)
2! 3!
and the standard forward formula is
h2 h3
f (x + h) = f (x) + f  (x)h + f  (x)
+ f  (x) + · · · (10.68)
2! 3!
Multiply the last equation by –4 and add to the first, we find
4h 3
f (x + 2h) − 4 f (x + h) = −3 f (x) − 2 f  (x)h + f  (x) + ··· (10.69)
3!
Then
− f (x + 2h) + 4 f (x + h) − 3 f (x))
f  (x) ∼
= (10.70)
2h
or, in discrete form,
− f i+2 + 4 f i+1 − 3 f i
fi ∼
= (10.71)
2h
The error in this formula is still O(h 2 ), yet all of the points are on one side of the
point i. Such formulas are called sloping difference formulas.
The temperature distribution through a copper slab with a mean conductivity
k = 83 W/m/C generates the data in Table 10.10. We might be asked to calculate
the heat flux at y = 0. Note that the temperature distribution within the slab is
nonlinear, and so the heat transfer cannot be from conduction alone. The definition
of the heat flux is
Q dT
q= = −k (10.72)
A dy
If the two point forward formula is used for dT /dy, then
Q dT W (28.6 − 30) C W
= −k = −83 × = 290.5 2 (10.73)
A dy mC .4 m m
This formula is not very accurate and would generally not be used unless abso-
lutely necessary. The more accurate three-point sloping difference formula results

Table 10.10. Results from experimental measurements of the temperature in a slab

y(m) 0.0 0.4 0.8 1.2 1.6 2.0
T (C) 30.0 28.6 26.2 22.1 20.0 18.3
 376 Essentials of Micro- and Nanofluidics

in
Q dT W (−26.2 + 4(28.6) − 3(30)) C W
= −k = −83 × = 186.75 2
A dy mC .4 m m
(10.74)
Note that this answer is considerably different from the two-point forward for-
mula.
At an interior point such as y = 1.2, the two-point central difference formula,
should be used, and
Q dT W (20.0 − 26.2) C W
= −k = −83 × = 643.25 2 (10.75)
A dy mC .8 m m
For comparison, the result using the backward formula is
Q dT W (22.1 − 26.2) C W
= −k = −83 × = 850.75 2 (10.76)
A dy mC .4 m m
and the forward result is
Q dT W (20.0 − 22.1) C W
= −k = −83 × = 435.75 2 (10.77)
A dy mC .4 m m
Note the considerable difference in the results and that the result using the central
difference is just the average of the forward and backward formulas.
As will be seen later in this chapter, using the central difference approximations
to the first and second derivatives is the key step in the numerical solution of
second-order differential equations of the boundary value type. This will lead to
a linear system of equations of tridiagonal form that is quickly and easily solved
by several methods. More accurate formulas can be derived for the first and
second derivatives; formulas for higher derivatives, such as the third and fourth
derivatives, may also be developed. However, these formulas are often difficult
to use and lead to considerably more complicated sets of linear equations when
solving an ordinary differential equation of the boundary value type. Thus only
the first and second derivatives will usually be required. These other formulas are
compiled in several places, in particular, Gilat and Subramaniam (2008).

10.8 Numerical integration

Integration and differentiation are complementary in the sense that the integral
of the first derivative of a function is the function itself. There are many cases
when a function must be integrated numerically; that is,
b
I = f dx (10.78)
a
or
x
Ii = f dy (10.79)
a
 377 Essential Numerical Methods

b
Area= f(x) dx
a

f(b)
f(x)
f(a)
Area

a b x

The trapezoidal rule is obtained by assuming

Figure 10.14
that the function to be integrated is linear in the
integration variable. The integral is simply the
area under the curve.

Ii is an indefinite integral because x, a variable, appears on one of the integration

limits, while I is a definite integral. As has been seen, the volume flow rate of a
fluid flowing in a duct or tube is defined by

Q= ud A (10.80)
A

where u is the fluid velocity and A is the cross-sectional area through which the
fluid flows.
As with differentiation, there are just a few formulas that are accurate enough
for most problems, although typically, textbooks discuss many more methods
than what are described here. Different integration formulas are derived based
on how the function is approximated over a given interval; typically, a linear or
quadratic approximation of the function is adequate, and only these formulas will
be discussed in detail.

10.8.1 The trapezoidal rule

Suppose that the definite integral of a function must be calculated and that the
integrand is given by a numerically generated function. In that interval, [a, b] f
may be approximated by a linear function, as in Figure 10.14, and

f (x) ∼
= p(x) = A + Bx (10.81)

Note that
a( f (a) − f (b))
A = f (a) − (10.82)
b−a
and
f (b) − f (a)
B= (10.83)
b−a
378 Essentials of Micro- and Nanofluidics

so that p(x) agrees with f (x) at the points a and b. Then, substituting into the
integral,
b b
f (x)d x ∼
= p(x)d x (10.84)
a a

b b B 2
p(x)d x = (A + Bx)d x = A(b − a) + (b − a 2 ) (10.85)
a a 2
Substituting for A and B,
b b h
f (x)d x ∼
= g(x)d x = ( f (b) + f (a)) (10.86)
a a 2
where h = b − a. This is the trapezoidal rule. Note that 1/2( f (a) + f (b)) is just
the average of the function value over the interval [a, b]. Thus
b h
f (x)d x = ( f (b) + f (a)) + error (10.87)
a 2
and the error is just the error in approximating the function by Taylor series:
1
error ∼
=− (b − a)3 (10.88)
12
Equation (10.87) can be extended to more than one interval; consider the case
of two intervals. Then, because the integral is additive,
c b c
f (x)d x = f (x)d x + f (x)d x (10.89)
a a b

and using the formula for the trapezoidal rule,

c b−a c−b
f (x)d x = ( f (b) + f (a)) + ( f (b) + f (c)) + E (10.90)
a 2 2
If the grid points are equally spaced, then
c h h
f (x)d x = f (a) + +h f (b) + f (c) + E (10.91)
a 2 2
In general, for N points,
c h  N
f (x)d x = ( f 0 + f N +1 ) + h fj + E (10.92)
a 2 j=1

A typical mesh is depicted in Figure 10.15. The error is now

1
E∼
= − (c − a)h 2 f  (ξ ) (10.93)
12
where a < ξ < c, but ξ is unknown. The trapezoidal rule is the simplest of the
integration formulas to derive, and it is also accurate enough for most purposes
(Figure 10.15).
 379 Essential Numerical Methods

A typical mesh for an integration scheme. The first numbering system ( j = 0 − N ) is often used in textbooks.
Figure 10.15
The second numbering system is used in programming. Note that the mesh size h = (b − a)/N in the first
case and h = (b − a)/(N − 1) in the second case. Note that the numbering system includes the end points.

Consider the data set in Table 10.11; the Matlab file for the trapezoidal rule is
straightforward:

%A Program for the Trapezoidal Rule,

x=[0 .5 1.0 1.5 2.0 2.5 3.0 3.5 4.0];
f=[1.90 2.39 2.71 2.98 3.20 3.20 2.98 2.74 2.63];
plot(x,f)
xlabel(’x’)
ylabel(’f’)
N=length(x);
h=(x(N)-x(1))/(N-1);
sum=f(1);
for i=1:N-1;
sum=sum+2.*f(i);
end
sum=sum+f(N);
Int=sum.*h./2

Note that the counter begins with i = 1. If the name of this integration m-file
is ‘trapint,’ then running the m-file produces (without the plot shown)

>>trapint

Int =

12.1825

>>

This value is compared with that for Simpson’s rule, discussed next. Note that
the trapezoidal rule may be used if a function is tabulated at unequal intervals.

Table 10.11. Numerical data set for calculating the definite integral
x 0 .5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
f (x ) 1.90 2.39 2.71 2.98 3.20 3.20 2.98 2.74 2.63
 380 Essentials of Micro- and Nanofluidics

b
Area= f(x) dx
a

f(b)
f(x)
f(a)
Area

x 0= a x 1 x =b x
2

The Simpson’s 1/3 rule is obtained by assuming

Figure 10.16
that the function to be integrated is quadratic
in the integration variable.

10.8.2 Simpson’s rules

There are many Simpson’s rules, depending on how the function is approximated.
The most common rule is the Simpson’s 1/3 rule, in which a quadratic function
is used to approximate the numerical function. To derive the formula, the Newton
forward interpolation formula is used (Figure 10.16):

s(s − 1) 2
p(x) = f 0 + s f 0 +  f0 (10.94)
2
where  f 0 = f 1 − f 0 is called the forward difference operator. Then
x2 x2
f (x)d x = p(x)d x + E (10.95)
x0 x0

Now recall that x = x0 + sh, and so

 
x2 2 s(s − 1) 2
p(x)d x = h f 0 + s f 0 +  f 0 ds (10.96)
x0 0 2

x2 h
p(x)d x = ( f0 + 4 f1 + f2 ) (10.97)
x0 3

and thus
x2 h
f (x)d x = ( f0 + 4 f1 + f2 ) + E (10.98)
x0 3

and the error E turns out to be

1 5 iv
E =− h f (ξ ) (10.99)
90
381 Essential Numerical Methods

If the function is approximated by a cubic polynomial over four points, or three

intervals, another Simpson’s rule, the 3/8 rule, is obtained:
x3 3h
f (x)d x = ( f0 + 3 f1 + 3 f2 + f3) + E (10.100)
x0 8
where now
3 5 iv
E =− h f (ξ ) (10.101)
80
Note that the error has not been reduced by adding another point.
If the function is tabulated at N + 1 points, then the Simpson’s 1/3 rule
becomes
⎛ ⎞
x N +1 
N −1 
N −2
h⎝
f (x)d x = f0 + 4 fj + 2 f j + f N +1 ⎠ + E (10.102)
x0 3 jodd jeven

where N is the total number of intervals and to use Simpson’s rule, N must be
even. The error is now
(x N − x0 ) 4 iv
E =− h f (ξ ) (10.103)
180
a reduction in error of O(h 2 ) compared to the trapezoidal rule.
Let us compare the value of the definite integral using the trapezoidal rule and
Simpson’s 1/3 rule. The matlab m-file for Simpson’s rule is similar to that for the
trapezoidal rule:

% function I=simp(f,a,b,n)
%A Program for Simpson’s Rule
x=[0 .5 1.0 1.5 2.0 2.5 3.0 3.5 4.0];
f=[1.90 2.39 2.71 2.98 3.20 3.20 2.98 2.74 2.63];
N=length(x);
plot(x,f)
xlabel(’x’)
ylabel(’f’)
N=length(x);
h=(x(N)-x(1))/(N-1);
sum=f(1);
for i=2:2:N-1
sum=sum+4.*f(i);
end
for i=3:2:N-2
sum=sum+2.*f(i);
end
sum=sum+f(N);
Int=sum.*h./3
382 Essentials of Micro- and Nanofluidics

Running the Matlab file produces

>>simpint

Int =

11.2583

>>

which is a bit different than the trapezoidal rule value. The Simpson value is
considered more accurate, based on its smaller truncation error.
The question arises of what happens if the function is tabulated over an odd
number of intervals. Then Simpson’s rule cannot be used near the beginning of
the data. One option is to use the trapezoidal rule; however, the trapezoidal rule
is much less accurate than Simpson’s rule. The solution is to use a high-accuracy
forward interpolation formula that is of the same order as Simpson’s rule, O(h 5 ).
Such a formula leads to
x+h h
f (x)d x = (9 f 0 + 19 f 1 − 5 f 2 + f 3 ) (10.104)
x 24
for the integration over the first interval, and Simpson’s 1/3 rule may be used to
complete the calculation.

10.8.3 Matlab integration functions

Matlab has several built-in functions that can be used for integration. The first
is quad, which is called as I = quad( function , a, b, tol), where a, b are the
integration limits and tol is a number related to the error; the smaller the value
of tol, the smaller the error. This Matlab function uses a form of Simpson’s rule
that adapts the grid size locally based on the behavior of the function.
The other function of interest here is trapz, called by I = trapz(x, f ). This
Matlab function is designed for numerically given functions like that discussed
in the preceding example. It uses the trapezoidal rule, as you might guess from
its name.

10.8.4 The indefinite integral

Suppose a function f (x) is perhaps the result of some other numerical computa-
tion. Then the indefinite integral is given by
x
Ii (x) = f (x)d x + J (a) (10.105)
a

Note that the indefinite integral Ii = Ii (x). By the trapezoidal rule, then,
1
Ii (x j+1 ) = J j+1 = J j + h( f j + f j+1 ) for j = 1, . . . , N (10.106)
2
383 Essential Numerical Methods

With Simpson’s rule, we need three points, and so we have to start at x2 , for
which
h
Ii2 = ( f (a) + 4 f 1 + f 2 ) (10.107)
3
and in the general case,
h
Ii (x j+1 ) = J j+1 = J j−1 + ( f j−1 + 4 f j + f j+1 ) for j = 2, . . . , N
3
(10.108)
To obtain Ii (1), a sloping integration formula that is of the same order of accuracy
as the Simpson’s rule could be used, as discussed earlier, for the definite integral.
For example, the electric field is defined as the gradient of the electrical
potential. In one dimension, the electric field is defined to be
dφ
E =− (10.109)
dy
so that the potential is given as the indefinite integral of the electric field or
y
φ(y) = − E(y)dy (10.110)
y0

10.8.5 Other formulas

It should be pointed out that there are many other formulas such as Newton–Cotes
formulas and Romberg integration, often called Richardson extrapolation. The
Newton–Cotes formulas result from different approximations to the function over
the interval of integration. Romberg integrations utilize solutions for different
mesh sizes to enhance the accuracy of the final result. I have found that the
trapezoidal rule and the Simpson’s 1/3 rule are sufficient for most applications.
In fact, I have never used any formulas other than the trapezoidal rule and
Simpson’s 1/3 rule for a standard integration.
The methods described here should not be used for Fourier integrals of the
form
π
I = f (x) cos nxd x (10.111)
0

because for large values of n, the integrand becomes highly oscillatory. Two
approaches that can be used are the Filon approach (Abramowitz & Stegun,
1972) and the fast Fourier transform (Fausett, 2008).

10.8.6 Grid (mesh) size

The question is, how do we know that the answer for the integral is correct?
Well, the answer lies in the number of points taken in the integration: N in
the two preceding Matlab files for the trapezoidal and Simpson’s rule. For a
numerical function such as earlier, N cannot be changed, so the accuracy of the
integral is fixed. However, when an analytical function is to be integrated, such as
384 Essentials of Micro- and Nanofluidics

f (x) = e−x , the value of N should be increased systematically until the answer
for successive values of N is within a specified tolerance such as 10−4 . Obviously,
a larger value of N may be required if the function varies rapidly; an example
of a function that varies rapidly near x = 0 is f (x) = e−Ax with A  1. Try
numerically integrating this function over the interval [0, 1] with A = 100. Of
course, the answer may be obtained analytically and is I = 1/A(1 − e −A ).
This issue is easily fixed by changing variables. Noting that e−Ax is zero for
large A, except near x = 0, the solution is to change variables to X = Ax, and
then, for x = 1, X = A  1, and so now the interval [0, 1] has become [0, ∞]
in X in the limit A → ∞. In this case, the numerical solution may easily be
computed.
There is a method in choosing mesh size, equivalent to choosing N . If the
integration is over [0, 1], a good starting point is choosing N = 11, which gives
10 intervals, and h 1 = 0.1. Note that the value of N includes the end points. Once
the solution for N = 11 is calculated, the procedure is to repeat the calculation
for N = 21 , giving h 2 = h 1 /2 = 0.05. This process is repeated until the answer
for two successive grid sizes or mesh differs by a small specified tolerance. If
three-digit accuracy is required (e.g., 0.1235 and 0.1237 are three-digit accurate),
then the tolerance is tol = 10−3 , that is, I1 − I2 < 10−3 .
The Romberg integration process is defined by combining the solution for h 1
and h 2 to achieve O(h 4 ) accuracy, and the Romberg result is defined by

4I (h/2) − I (h)
IR = + O(h 4 ) (10.112)
3

This procedure is equivalent to Richardson extrapolation in the solution of ordi-

nary and partial differential equations.
In general, the process of determining the numerical accuracy of a given
integral is part of a general process of verification, which is discussed later in this
chapter.

10.8.7 Singularities
It is important to mention that if there is singular behavior in the region of
integration, all these integration methods will fail. For example, the integral

1
I = ln xd x (10.113)
0

cannot be performed numerically since ln 0 = ∞ and no numerical procedure

will succeed. Conversely, the function f (x) = x −α for 0 < α < 1 does exist:

1 1
x −α d x = (10.114)
0 1−α
385 Essential Numerical Methods

Despite that the integral does exist, none of the integration formulas discussed
here will work. Moreover, the function x α with α > 0 has a singularity in the first
derivative at x = 0, and any Taylor series approximation to the function will thus
fail.
There is no general rule on how to deal with integrals having singularities.
There are several possible approaches, however, and we discuss these briefly. The
first method is the removal of the singular part by performing the singular portion
of the integral analytically, and then the remaining regular part can be performed
numerically. For example,
π π
(ln x + e−x sin x)d x = π(ln π − 1) + e−x sin xd x
2 2
(10.115)
0 0

and the second integral can be performed numerically. Second, a local series
expansion can be used to remove the singularity. Consider the evaluation of the
integral
1
I = J0 (x)x −1/3 d x (10.116)
0

where J0 is the Bessel function of order 0. Near x = 0, the Bessel function

behaves like
x2
J0 = 1 −
+ O(x 4 ) (10.117)
22
Thus, near x = 0, where the integrand is singular, the integral can be performed
analytically with the rest of the integral being performed numerically. The integral
can be evaluated by picking a small number, say, x = 0.1, and splitting the integral
into two parts:
0.1 1
I = J0 (x)x −1/3 d x + J0 (x)x −1/3 d x (10.118)
0 0.1

For the first integral, the behavior of the Bessel function near x = 0 is used, say,
the first two terms in the series, and the integration is performed analytically. The
rest of the integral can be performed numerically.
Finally, sometimes a change in variable will regularize an integral. For example,
b f (x)d x
I = (10.119)
0 x 1/2
can be regularized by letting η = x 1/2 so that
√
b f (x)d x b
I = =2 η f (η2 )dη (10.120)
0 x 1/2 0

It should be mentioned that singularities are often the sign of a flaw in a model of
a given system. Models of physical systems should not contain singularities unless
some physical aspect of the system is not considered. An example is a boundary
layer. From an inviscid point of view, the velocity profile is discontinuous near
 386 Essentials of Micro- and Nanofluidics

a solid surface. However, if viscous forces are included, the velocity profile is
regularized.

10.9 Solution of linear systems

Sets of linear equations occur very often in practice. Consider, for example, the
ordinary differential equation

y  + my = e−x (10.121)

where m is a constant subject to the boundary conditions y(a) = 1, y(b) = 0.

To solve this equation numerically, the interval (a, b) is broken up into N − 1
smaller intervals each of width (h = b − a)/(N − 1); recall that this was also
done in the section on integration, as depicted in Figure 10.15. Let j denote a
mesh point; approximating the second derivative by a central difference, then
at j,
y j+1 − 2y j + y j−1
+ my j = e−x j for j = 2, . . . , N − 1 (10.122)
h2
or

y j−1 − (2 − mh 2 )y j + y j+1 = h 2 e−x j for j = 2, . . . , N − 1 (10.123)

These are N − 2 equations in N − 2 unknowns, and the system is said to be

tridiagonal. The solution to sets of linear equations is required in curve-fitting
problems, as has been seen.
For a small number of equations, say, N ≤ 3 or 4, you were probably taught to
solve them by Cramer’s rule. Let us consider the 2 by 2 system

a11 x 1 + a12 x2 = b1
a21 x 1 + a22 x2 = b2
with b1 = 0. Then Cramer’s rule provides the solution as
   
b a  a 
 1 12   11 b1 
   
 b2 a22   a21 b2 
x1 =    
 , x2 =  a  (10.124)
 a11 a12   11 a12 
   
 a21 a22   a21 a22 

as long as the determinant a11 a22 − a12 a21 = 0. Cramer’s rule is easy for N = 2
but really gets complicated for N larger. It can be shown, moreover, that the
number of multiplications for N equations is Mc = (N − 1)(N + 1). For N = 20,
Mc = 97 × 1019 multiplications. Clearly a method is required that requires far
fewer operations because 50 or 100 or 10,000 or more equations will often be
required to solve. Gauss elimination is considered first, which is useful for N up to
several hundred, and then iterative methods are considered, which should be used
for the really large systems that occur in solving partial differential equations.
387 Essential Numerical Methods

Gauss elimination is the main method of solving sets of linear equations,

particularly when the matrix of coefficients is not sparse, meaning the matrix is
“full.” The idea is to reduce one of the equations in the set to one that can be
solved immediately for one of the variables and then solve for the rest.
Consider the set of equations

2x1 + x2 − 3x3 = −1
−x1 + 3x 2 + 2x3 = 12
3x1 + x2 − 3x3 = 0

The matrix of coefficients is given by

⎡ ⎤
2 1 −3
⎢ ⎥
⎣ −1 3 2 ⎦ (10.125)
3 1 −3

The procedure is to systematically eliminate elements in the matrix of coefficients

to get this system in the form of x3 = b, where b is a known constant. To do this,
multiply the first equation by +1/2 and add to the second equation to obtain

2x 1 + x2 − 3x3 = −1
7 1 1
0 + x 2 + x 3 = 11
2 2 2
3x1 + x2 − 3x3 = 0

Eliminate x 1 in the third equation by multiplying the first equation by −3/2 and
adding to third, which, after some simplification, yields

2x 1 + x2 − 3x3 = −1
7x2 + x3 = 23
−x2 + 3x3 = 3

Now x 2 is eliminated in the third equation by multiplying the second equation by

1/7 and adding to the third equation:

2x 1 + x 2 − 3x 3 = −1
7x 2 + x3 = 23
1 23
3 x3 = +3
2 7
Now we can solve for x3 = 2 and then back substitute to get x 2 and x 1 :
23 − x 3
x2 = =3
7
−1 − x 2 + 3x3
x1 = =1
2
388 Essentials of Micro- and Nanofluidics

The final form of the set of equations is called upper triangular form. For the
Gauss elimination procedure, the number of operations can be shown to be
MG = (1/3)N 3 + N 2 − (1/3)N , which, for N = 20, MG = 3060, much fewer
than for Cramer’s rule.
This procedure is particularly amenable to programming because the procedure
systematically eliminates elements in each column; let us write the system of
equations as

a11 x 1 + a12 x2 + · · · + a1N x N = b1

a21 x 1 + a22 x2 + · · · + a2N x N = b2
..
.
a N 1 x1 + a N 2 x2 + · · · + a N N x N = b N

and the converted system as

s11 x 1 + s12 x 2 + · · · + s1N x N = d1

s22 x2 + · · · = d2
..
.

s N −1N −1 x N −1 + s N −1N x N = d N −1
sN N x N = dN

Then the back substitution procedure is

dN
xN =
sN N
s N −1N −1 x N −1 + s N −1N x N = d N −1
..
.

and so on down to

N
s1 j x j = d1
j=1

To generate the si j s, which is the elimination process,

s1 j = a1 j j = 1, N
d1 = b1

That is, the first line is left alone. For the second row,
a21
s 2 j = a2 j − a1 j j = 2, 3, 4, . . . , N
a11
a21
d2 = b2 − b1
a11
To successively remove the coefficients in the first row, multiply the first equation
by a21 /a11 and subtract from the second equation. This eliminates the x1 term in
389 Essential Numerical Methods

the second equation. The general term is

ai1
si j = ai j − a1 j i ≥ 2, j = 2, 3, . . . , N
a11
ai1
d i = bi − b1
a11
In this way, the coefficients multiplying x 1 in every equation are eliminated. Next
eliminate the coefficients of x 2 in the third through N th equations, performing
the same operations on the second column, and so on.
In the scheme, it was assumed that aii = 0, and so all of the divisions of
coefficients can be performed accurately. However, if one of the diagonal elements
aii is small, then significant error may be incurred on division by that element.
Pivoting prevents the error that would be incurred in a division by a small diagonal
coefficient (i.e., if one aii ∼ 0); in addition, pivoting can be used to improve the
accuracy of the scheme as well. Basically, in pivoting, the order of the equations
is altered. For example, consider the following system:
.0001x 1 + 1.0000x2 = 1.0000
1.0000x1 + 1.0000x2 = 2.0000
If the computer has no round-off error, then the exact solution is x1 = 1.0001
and x2 = 0.9999. But every computer has round-off error, so let’s suppose that
the computer rounds to four digits. Then, without pivoting;
0.0001x1 + 1.0000x2 = 1.0000
−1.000 × 104 x2 = −1.000 × 104
and to four significant digits,
x2 = 1.0000 x1 = 0.0000 (10.126)
This answer is clearly wrong! To pivot, merely change the order of the equations:
1.0000x1 + 1.0000x2 = 2.0000
0.0001x1 + 1.0000x2 = 1.0000
Now, to four digits, x1 = 1.0001 and x 2 = 0.9999.

10.9.1 Solving sets of linear equations in Matlab

There are two ways to solve a linear system in Matlab. If the linear system is
defined by Ax = b, where A is the matrix of coefficients of the linear system,
then left division yields the solution x = A b. Conversely, the inverse of the
matrix A can be calculated; the inverse is defined as that matrix A−1 such that
A−1 A = I , where I is the identity matrix. Then the solution x is given by the
statement x = A−1 ∗ b or by x = inv(A) ∗ b, where inv(A) denotes the inverse;
check the Matlab help for cautions on the use of the inv command. For example,
in the problem just discussed, the answer in Matlab is produced by the script
390 Essentials of Micro- and Nanofluidics

>> A = [0.0001 1
1 1]

A=

0.0001 1
1 1

>>inv(A)

ans=

-1.0001 1.0001
1.0001 -0.0001

>> b=[2
1]

b=

1
2

>> x=A\b

x=

1.0001
0.9999

10.9.2 Iterative solution to linear systems

Gauss elimination is fine for linear systems of less than roughly several hundred
equations; for larger numbers of equations, iterative methods are often faster and
more efficient. Moreover, these methods can be faster than Gauss elimination
when the coefficient matrix is sparse, for example, tridiagonal systems that have
elements on the diagonal of the matrix and the two diagonals adjacent.
As an example, consider the following problem:

8x 1 + x2 − x3 = 8
2x1 + x2 + 9x3 = 12
x1 − 7x 2 + 2x3 = −4
391 Essential Numerical Methods

In each equation, we solve for the variable with the largest coefficient; rounding
to three digits,
x1 = 1 − .125x 2 + .125x3
x2 = .571 + .143x1 + .286x 3
x3 = 1.333 − .222x1 + .111x 2
We begin with some initial estimate of the solution, say, x1 = x 2 = x3 = 0, and
successively compute values using the previously calculated values. If n denotes
iteration, then, say,
x1n+1 = 1 − .125x2n + .125x3n (10.127)
This procedure is called Jacobi iteration. The following iteration scheme is
generated for a zero initial guess:
Iteration x1 x2 x3
1 1 .571 1.333
2 1.095 1.095 1.048
3 .995 1.026 .969
4 .993 .990 1.000
5 1.002 .998 1.004
6 1.001 1.001 1.001
7 1.000 1.000 1.000
The Jacobi iteration scheme is, in general,

bi  ai j (n)
N
xin+1 = − x i = j n = 1, . . . (10.128)
aii a j
j=1 ii

Note that all the values of x j on the right-hand side of the equation are at the
previous or (nth) iteration.
We can do much better by using updated x j as soon as they are generated. For
example, in the Jacobi solution of the previous equation,
(n+1) (n) (n)
x2 = .571 + .143x1 + .286x3 (10.129)
Because x1 has already been updated, it is more efficient to use the latest value:
(n+1) (n+1) (n)
x2 = .571 + .143x 1 + .286x3 (10.130)
This is called the Gauss–Seidel iteration. The Gauss–Seidel iteration produces
the following:
Iteration x1 x2 x3
1 1 .714 1.032
2 1.041 1.014 .990
3 .997 .996 1.002
4 1.001 1.000 1.000
5 1.000 1.000 1.000
392 Essentials of Micro- and Nanofluidics

Note that this iteration scheme converges faster than the Jacobi scheme and would
always be employed in practice. The Gauss–Seidel iteration is, in general,

bi 
i−1
ai j (n+1) N
ai j n
xin+1 = − xj − x n = 1, 2, 3, . . . (10.131)
aii a
j=1 ii
a i
j=i+1 ii

The successive overrelaxation method or (SOR) speeds up Gauss–Seidel further

by adding a relaxation factor to the solution:

xin+1 = αxin+1 + (1 − α)xin (10.132)

where 1 < α < 2. Typically α ∼ 1.5 works very well.

A sufficient condition for convergence of these iteration schemes is that


N
 
|aii | > ai j  i = j i = 1, . . . , N (10.133)
j=1

That is, the set of equations is diagonally dominant. Discretization of ordinary and
partial differential equations, if done properly, will result in diagonally dominant
systems.
Last, because these are iterative schemes, a test for convergence is required. In
general, a relative test should be used; if xold(i) denotes the solution at the nth
iteration and xnew(i) is the solution at the n + 1st iteration, then the convergence
test is
 
 xnew(i) − xold(i) 
  < eps (10.134)
 xnew(i) 

where eps is a small number, say, 10−4 . A Matlab program using the Gauss–Seidel
iteration looks like the following:
% Gauss-Seidel solution of a set of linear equations.
function x=gseidel(A,b,max,epsi,n,guess)
% Solution of Ax=b
% xnew is the solution vector
% A is the coefficient matrix
% b is the right side of the linear system
% define new matrix C to be used in solving for xnew
C=-A ;
xold=guess;
xnew=xold;
xnew=xnew’;
xold=xold’;
for i=1:n
C(i,i)=0;
end
393 Essential Numerical Methods

for i=1:n
C(i,1:n)=C(i,1:n)/A(i,i);
end
for i=1:n
d(i,1)=b(i)/A(i,i);
end
iter=1;
while(iter <= max)
xold=xnew;
for j=1:n
xnew(j)=C(j,:) * xnew + d(j,1);
end
if abs((xnew-xold)./xnew)<epsi
x=xnew
disp(’Gauss-Seidel iteration scheme converged.’)
disp(iter);
return;
end
% disp([i xnew’]);
iter=iter+1;
end
disp(’Result after max iterations:’)
x=xnew;

The colon in the column entry, C( j, :), means to perform the operation for all
columns.
The preceding Matlab m-file is self contained. If a separate file is used to
define the matrix coefficients, and for the actual Gauss–Seidel iteration, after
definition of matrix A and the right-hand side in the calling program, the solution
is obtained by writing x = gseidel(A, b, max, epsi, n, guess).

10.9.3 Tridiagonal systems

As has been mentioned, tridiagonal systems arise in the solution of ordinary and
partial differential equations. One example is the nonlinear ordinary differential
equation for the potential in an electrolyte solution. Consider a system of equa-
tions, say, N = 5:

a1 x 1 + c1 x2 = d1
b2 x1 + a2 x 2 + c2 x3 = d2
b3 x2 + a3 x 3 + c3 x4 = d3
b4 x3 + a4 x 4 + c4 x5 = d4
b5 x4 + a5 x5 = d5
394 Essentials of Micro- and Nanofluidics

Let us work with the one-dimensional arrays ai , bi , ci , di , i = 1, . . . , N . Here

it is necessary to eliminate only one element from each equation in each col-
umn; that is, x1 from the second equation, x2 from the third equation, and so
on. To do this, multiply the first equation by −b2 /a1 and add to the second
equation:

a1 x 1 + c1 x 2 = d1
 
b2 b2
a2 − c1 x 2 + c2 x 3 = d2 − d1
a1 a1
b3 x2 + a3 x 3 + c3 x 4 = d3
..
.

We have eliminated x 1 from equation 2. Now eliminate x2 from equation 3 by

multiplying equation 2 by

b3
− (10.135)
a2 − ca1 b1 2

and adding; the last three equations are as follows:

 
b3 c 2 b3 (d2 − ba21 d1 )
a3 − x 3 + c 3 x 4 = d 3 −
a2 − ca1 b1 2 a2 − ca1 b1 2
b 4 x 3 + a4 x 4 + c 4 x 5 = d 4
b 5 x 4 + a 5 x 5 = d5

We can now see a pattern developing; if a superscript denotes the number of the
elimination operation, then

(1) c1 b2
a2 = a2 −
a1
(2) b3 c2
a3 = a3 − (1)
a2
(1) b2 d1
d2 = d2 −
a1
(1)
(2) b3 d2
d3 = d3 − (1)
a2

The general form of these newly formed arrays, say, α and γ , is

bi+1 ci
αi+1 = ai+1 − f or i = 1, . . . , N − 1 and α1 = a1 (10.136)
αi

bi+1 γi
γi+1 = di+1 − f or i = 1, . . . , N − 1 and γ1 = d1 (10.137)
αi
395 Essential Numerical Methods

The system eventually ends up with all terms below the diagonal eliminated, and
this form is
a1 x 1 + c1 x 2 = d1
α2 x 2 + c2 x 3 = γ2
α3 x 3 + c3 x 4 = γ3
α4 x 4 + c4 x 5 = γ4
α5 x 5 = γ5
We now back substitute
γ5
x5 =
α5
γ4 − c4 x5
x4 =
α4
..
.
γi − ci xi+1
xi =
αi

for i = N − 1, . . . , 1.
As an example, consider the system
x1 + 2x2 = 2
x1 + 3x 2 + 2x3 = 7
x2 + 4x3 = 15
Computing the arrays α, γ :
α1 = a1 = 1
b2 c1 1×2
α2 = a2 − =3− =1
α1 1
b3 c2 1×2
α3 = a3 − =4− =2
α2 1
γ1 = 2
b2 γ1 1×2
γ2 = d2 − =7− =5
α1 1
γ3 = 15 − (1)(5) = 10
γ3 10
x3 = = =5
α3 2
5 − (2)(5)
x2 = = −5
1
2 − (2)(−5)
x1 = = 12
1
A Matlab program for this elimination algorithm is
396 Essentials of Micro- and Nanofluidics

% Tridiagonal system solver.

function x=tridiag(a,b,c,d,n)
alpha(1)=a(1);
gamma(1)=d(1);
for i=1:n-1
alpha(i+1)=a(i+1)-b(i+1)*c(i)/alpha(i);
gamma(i+1)=d(i+1)-b(i+1)*gamma(i)/alpha(i);
end
x(n)=gamma(n)/alpha(n)
for i=n-1:-1:1
x(i)=(gamma(i)-c(i)*x(i+1))/alpha(i)
end

Different schemes use slightly different elimination procedures. The algorithm

that goes by the name of the Thomas algorithm defines the two arrays αi and γi
as
c1 d1
α1 = , γ1 =
a1 a1
ci di − bi di−1
αi = , γi = for i = 2, . . . , N − 1
ai − bi ci−1 ai − bi ci−1

with the back substitution step as

γn
xn =
αn
(γi − ci xi+1 )
xi = for i = N − 1, . . . , 2
αi
The Thomas algorithm is extremely popular, but both these elimination schemes
work well.

10.9.4 Ill-conditioning and stability

A matrix is said to be ill-conditioned if there is a large difference between the
largest and the smallest eigenvalues in a system defined as the one-dimensional
array of real numbers λ that satisfies Ax = λx. The condition number is defined
as the difference between the largest and smallest eigenvalues. Consider the
following system:

1 3
x1 + x2 =
2 2
1 1 5
x1 + x2 =
2 3 6
for which the condition number is 19.281 (Fausett, 2008). The Matlab command
to compute the condition number is cond, and the script here may look as follows:
397 Essential Numerical Methods

> A=[1 1/2

1/2 1/3]
> A= 1.0000 0.5000
0.5000 0.3333
> cond(A)
> ans =

19.281
>

A condition number that is much greater than 1 indicates a high possibility of ill
conditioning.
We have already seen what kinds of errors can occur without pivoting. Diffi-
culties also occur if, say, the coefficients on the right-hand side are not known for
certain. For example, the function on the right-hand side of an ordinary differ-
ential equation may be the result of experimental data; this often happens. The
uncertainty can drive a system into an ill-conditioned state, rendering the calcu-
lation of the solution unstable in the sense that a small change in the coefficients
can lead to large differences in the solution.
Consider, for example, the following system, which is chopped at two signifi-
cant digits:

3.02x1 − 1.05x 2 + 2.53x 3 = −1.61

4.33x 1 + .56x2 − 1.78x 3 = 7.23
−0.83x1 − 0.54x 2 + 1.47x 3 = −3.38

Solving this system using Gauss elimination with pivoting yields the reduced
system

4.33x1 + .56x 2 − 1.78x 3 = 7.23

0x1 − 1.44x 2 + 3.77x 3 = −6.65
0x 1 + 0x2 − .00362x3 = .00962

keeping three digits, except in the last equation, where we keep five. The solution
to this reduced system is

x = (0.88, −2.35, −2.66) (10.138)

whereas the true solution is

x = (1, 2, −1) (10.139)

Thus more precision is required here, even though the right-hand side data are only
known to, say, two digits. Moreover, if a small change in one of the coefficients
is made, the solution to the linear syatem will be much different. Try changing
 398 Essentials of Micro- and Nanofluidics

0.56; in the second equation to 0.50; then

x = (0.94, 0.12, −1.72) (10.140)

Very unstable indeed!

In the next two sections, we apply the methods of solving linear systems of
equations to the solution of ordinary differential equations.

10.10 Solution of boundary value problems

10.10.1 Introduction
Numerical differentiation and the methods of solution of linear systems provide
the means to solve both linear and nonlinear ordinary differential equations. In
this section, the focus is on the solution of second-order boundary value problems.
Higher-order equations, for example,

z (iv) + z  = f (x) (10.141)

which often arise in the field of solid mechanics, can be solved as a system of
second-order equations. Recall that the Navier–Stokes equations and the Pois-
son equation for the potential are second-order equations. For this fourth-order
equation, define

z = s
w = s  = z  (10.142)
w + s = f (x)

Each of these equations can be solved for the variables (z, s, w).
In general, second-order linear equations are of the form

y  + p(x)y  + r (x)y = f (x) (10.143)

where p, r , and f are known functions of x . A second-order nonlinear equation

is one, for example, like

y  + yy  = f (x) (10.144)

or, in a more general form,

q(x, y, y  )y  + p(x, y, y  )y  + r (x)y = f (x) (10.145)

If f (x) = 0, the equation is said to be homogeneous.

Second-order equations of the boundary value type must satisfy boundary
conditions at the end points of the interval. In general, as has been discussed
before for specific equations, they are usually of four types:

1. End values specified: y(a) = A, y(b) = B

2. End derivatives specified: y  (a) = dy/d x(a) = C, y  (b) = dy/d x(b) = D
399 Essential Numerical Methods

3. Linear combinations of y, y  specified: E y  + F y = C, x = a, H y  +

Ry = S, x = b, where E, F, C, H, R, S are constants
4. Some combination of type 1, 2, or 3 or some other nonlinear conditions:
y  (a) + y 2 (b) = 0, y(b) = B

The boundary conditions are determined by the physical conditions imposed on

the boundary such as the no-slip and slip condition in fluid mechanics, the surface
charge density condition for the electrical potential, or the no-flux condition for
a chemical species in an electrolyte solution at a solid wall.
At this point, consider the simplest type of equation, the linear equation, subject
to specified end values. It should be pointed out that in many cases, closed-form
solutions are available such as when p, r in equation (10.143) are constants and
f = 0.

10.10.2 Linear equations

Often linear equations have an analytical solution in the form of simple functions
such as the exponential function or polynomials. However, sometimes the solution
to a given equation is so complicated that it is just as easy to solve the problem
numerically. For example, evaluating a Bessel function of order zero (Abramowitz
& Stegun, 1972) that arises in cylindrical domains and is the solution of the linear
ordinary differential equation

x 2 y  + x y  + x 2 y = 0 (10.146)

involves evaluating a complicated integral numerically.

Thus consider the equation

y  + p(x)y  + r (x)y = f (10.147)

subject to y(a) = A, y(b) = B. The interval [a, b] is divided into N − 1 inter-

vals of equal length (N points), using second-order central differences to approx-
imate the derivatives and evaluating the functions p, r, f at x j , results in

y j+1 − 2y j + y j−1 y j+1 − y j−1

2
+ pj + r j y j = f j + E j for j = 2, . . . , N − 1
h 2h
(10.148)
Note that the end points are defined by j = 1 and j = N and that E is the
error incurred in the differencing and E = O(h 2 ) for the central difference
approximation. Multiplying by h 2 and collecting terms multiplying y j+1 , y j
and y j−1 , the tridiagonal system is obtained according to
  , -  
h h
y j−1 1 − p j + y j −2 + h 2 r j + y j+1 1 + p j
2 2

= h2 f j + h2 E j for j = 2, . . . , N − 1 (10.149)
400 Essentials of Micro- and Nanofluidics

Table 10.12. Numerical solution for the

linear ODE
Numerical Exact
y0 = 0 0
y 1 = 0.18023 0.17990
y 2 = 0.38735 0.38682
y 3 = 0.64993 0.64945
y 4 = 1.0 1.0

Provided that the solution y j is well behaved, the error term, which, on multi-
plication by h 2 , is E = O(h 4 ), is neglected, and a solution accurate to three or
four digits may be obtained by successively increasing the number of grid points,
thereby reducing the grid size, that is, by increasing N . The value of N includes
the end points, and so there are actually N − 1 intervals.
Let us solve the equation y  − y = x with y(0) = 0, y(1) = 1 using h = 1/4
or N = 5. Here  is a constant and p = 0, r = −1, f = x, and our equation
becomes

y j−1 − 2 + h 2 y j + y j+1 = h 2 x j with x j = j h f or j = 1, . . . , N − 1
(10.150)
Writing the tridiagonal system explicitly for  = 1 yields

−2.0625y2 + y3 = .015625
y2 − 2.0625y3 + y4 = .031250
y3 − 2.0625y4 = −.953125

This is a tridiagonal system of equations, which may be solved by either of our

elimination methods. This equation does have an exact solution; for  = 1, it is
2  −x 
y= e − ex − x (10.151)
e−1 −e 1

The system of equations yields the solution found in Table 10.12.

Considering that only four intervals have been used (N = 5 in this numbering
system), the agreement is very good. Comparing the numerical result with an
exact solution is one way to verify that the numerical result is correct. However,
in most cases, an exact solution will not be available, and as with integration, the
value of N is increased systematically until the successive solutions agree to a
specified tolerance. To do this, the solution would be calculated for N = 11, then
N = 21, and so on until the desired accuracy is obtained. The reason for these
values of N is that it is easy to compare the numbers because half of the points
are at the same locations for each increase in N . When solutions for two different
mesh sizes agree to within a specified tolerance, the solution is said to be verified.
Most of the time, programming errors are caught when the grid is refined and the
solutions do not agree. A typical Matlab file for this problem is
401 Essential Numerical Methods

% First example in the linear ODE section.

epsilon=1;
n=11;
al=1.;
dx=al/(n-1);
n1=n-1;
y(1)=0;
y0=y(1);
y(n)=1;
yn=y(n);
x=[0:dx:1];
for j=1:m
for i=2:n1
a(i)=-(2.*epsilon+dx*dx);
b(i)=epsilon;
c(i)=epsilon;
d(i)=x(i)*dxˆ2;
end
d(n-1)=-yn*c(n-1)+d(n-1);
c(n-1)=0.;
y=tridiag(n,a,b,c,d,y0,yn);
% yan is the analytical solution for epsilon=O(1) and yan2
% is the analytical solution for epsilon<< 1 near x=1
for i=1:n
yan(i)=2./(exp(-1)-exp(1))*(exp(-x(i)/sqrt(epsilon))-
exp(x(i)/sqrt(epsilon)))-x(i);
yan2(i)=2*exp(-(1-x(i))/sqrt(epsilon))-x(i);
end

figure(1);
plot(x,y,’*’,x,yan2,’o’);
xlabel(’x’);
ylabel(’y’);

end

Numerical solutions of this equation will be difficult as the value of  decreases.

As  → 0, the second derivative becomes increasingly large near x = 1, and the
solution approaches the result
√
y = 2e−(1−x)/ 
−x (10.152)

The result for various values of  is depicted in Figure 10.17.

As  becomes smaller and approaches zero, we say that this is a singular
perturbation problem because over much of the domain, y = −x, and thus we
 402 Essentials of Micro- and Nanofluidics

1.5

ε= 0.50000 Numerical
ε= 0.50000 Analytical
1 ε= 0.00500 Numerical
ε= 0.00500 Analytical
ε= 0.00005 Numerical
y
0.5 ε= 0.00005 Analytical

−0.5

−1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Results for several values of  for the linear ordinary differential example. Note that finite difference algorithm
Figure 10.17
is extremely robust and able to resolve the boundary layer, or region of rapid variation, down to  = 5 × 10−4 .

cannot satisfy the boundary condition at x = 1. These types of problems were

encountered in Chapter 7 and will be seen in subsequent chapters.
Notation alert: Here is where we run into another notation problem. In this
chapter, h has denoted mesh size, but in Chapter 7, h denotes a channel height.
To avoid confusion in the discretization of the potential equation between h the
mesh size and h the channel height, subsequently, we will use H to denote the
channel height and  = λ/H .
Specifically, recall that the potential in the electrical double layers between two
plates satisfies, in the Debye–Hückel limit,
d 2φ
2 = z2φ (10.153)
dx2
for a solution containing a pair of ionic species of valence z. The resulting
difference equations are

 2 φ j−1 − (2 2 + z 2 h 2 )φ j +  2 φ j+1 = 0 (10.154)

If the boundary conditions are taken to be φ = ζ at x = 0, 1, where x is scaled

on the channel height, then the first and last equations have a nonzero right-
hand side: φ0 = φ N = ζ . Here  = λ/H , where H is the channel height and is
equivalent to  as used in Chapter 7.
For   1, that is, for extremely thin electrical double layers (EDLs), we will
need to solve the problem only within the EDL. Near Y = 0, we blow up the EDL
by defining a boundary layer variable Y = x/ so that the governing equation
is
d 2φ
= z 2φ (10.155)
dY 2
subject to φ = ζ at Y = 0. The other boundary condition must be applied at a
large value of Y and in the limit  → 0, Y → ∞ so that if the core is electrically
neutral, the boundary condition is φ = 0 as Y → ∞. But numerically, what is
403 Essential Numerical Methods

∞? Well, as discussed in Chapter 7, for this problem, ∞ ∼ 5. In the general

case, to simulate ∞, the boundary is systematically increased until the solution
does not change away from the outer boundary. Of course, there is an analytical
solution to this problem, which is

φ(Y ) = ζ e −zY (10.156)

and it is not necessary to solve the problem numerically. Note that Y = x ∗ /λ,
where x ∗ is dimensional, and so the boundary layer analysis is equivalent to
taking the length scale to be the Debye length and not the channel height.
Normally for ordinary differential equations, the direct elimination methods
are preferable; it is worth mentioning that the difference equations may also be
solved using the iterative methods of Jacobi or Gauss–Seidel or, better, successive
overrelaxation. However, the advantage of iterative methods is greatest when the
number of equations is very large, say, 1000 or more, and this usually does not
happen with ordinary differential equations. To use the iterative methods, the
equations must be diagonally dominant, that is,
   h  
 
h 

 
−2 + h 2r j  ≥ 1 + p j  + 1 − p j  j = 2, . . . N − 1 (10.157)
2 2
It turns out that this condition will hold if the mesh size
2
h <   j = 2, . . . N − 1 (10.158)
pj
and r j < 0. If r j > 0, iterative methods can fail, although even if diagonal
dominance is not satisfied, iterative methods may still converge. Difference
equations derived from ordinary differential equations are usually diagonally
dominant.
Numerical methods are usually required for nonlinear equations, and this
subject is considered next.

10.10.3 Nonlinear equations

Notation alert: Just a reminder that the symbol h is used for mesh length in this
chapter and not for channel height.
Nonlinear ordinary differential equations cannot normally be solved directly
using the linear system solvers discussed earlier. However, the equation can first
be linearized by assuming that some portion of the nonlinear term is known. Then
the linear system solver can be used iteratively until the process converges. The
specific steps are as follows:

1. Linearize the nonlinear terms in the difference equations by guessing their

values.
2. Solve the difference equations by a direct method: the Thomas algorithm.
3. Return to step 1, unless two successive iterations differ by less than a
prescribed tolerance.
404 Essentials of Micro- and Nanofluidics

Consider the following equation as an example:

y  + p(x)y 2 y  + r (x)y 3 = f (x) (10.159)

y(a) = A, y(b) = B (10.160)

At a typical mesh point, the finite difference approximation to equation (10.159)

is
h  
y j+1 − 2y j + y j−1 + p j y 2j y j+1 − y j−1 + h 2r j y 3j
2
= h 2 f j + O(h 4 ) for j = 2, . . . , N − 1 (10.161)

This is a tridiagonal system of equations for the N − 2 values of y j at the interior

grid points. The problem is that these equations are nonlinear; here the difference
equation should be linearized. To do this, if n denotes the iteration number, for
this equation,
h  2, -  2
y nj+1 − 2y nj + y nj−1 + p j y n−1
j y nj+1 − y nj−1 + h 2r j y n−1 j y nj = h 2 f j
2
(10.162)
and rearranging,
   
h  2  2
y nj−1 1 − p j y n−1j + y n
j −2 + h 2
r j y n−1
j
2
 
h  2
+ y nj+1 1 + p j y n−1
j = h2 f j (10.163)
2
Thus the coefficients of the linear system are
 2
a j = −2 + h 2r j y n−1
j j = 2, . . . , N − 1
h  n−1 2
bj = 1 − pj yj j = 2, . . . , N − 1
2
h  2
c j = 1 + p j y n−1
j j = 2, . . . , N − 2
2
d j = h2 f j j = 2, N − 1

Note that the boundary conditions in the Matlab program are used for j = 2 and
j = N − 1; for example, if the function is specified at the end point, then
 
h  n−1 2
d2 = h f 2 − y1 1 − p1 y1
2
(10.164)
2
and depends on the boundary value y1 . The same is true for d N −1 .
In general, the solution is guessed initially, and then equations (10.163) are
solved successively until
 
 y n−1 
 j 
1 − n  < eps all j = 2, . . . N − 1 (10.165)
 yj 

where eps is a pre-determined tolerance; usually, eps = 10−4 or smaller.

405 Essential Numerical Methods

The success or failure of the iterative process can depend on the way the
equation is linearized. The best way is to linearize the lowest-order terms in the
group of nonlinear terms. For example, suppose the nonlinear term is
yy  (10.166)
We would not linearize y  :

j+1 − y j−1
y n−1 n−1

yy  ∼ yy  p = y nj (10.167)
2h
Recall that the nonlinear form of the Poisson equation for the electric potential
for binary and symmetric electrolytes is, in the notation of the present section,
d 2φ
= z sinh zφ
2 (10.168)
dx2
In this case, there is no choice but to linearize the right-hand side, and the
discretized version is
 2 φ nj−1 − 2 2 φ nj +  2 φ nj+1 = zh 2 sinh zφ n−1
j (10.169)
It should be mentioned here that these systems of nonlinear discretized equa-
tions can also be solved by using zero-finding techniques. However, I have never
found that to be an efficient way to proceed, and those methods are prone to
divergence (Gilat and Subramaniam, 2008). Of course, the method described
here is also prone to divergence, but in my experience, linearizing the equations
and using the direct or iterative solvers is the preferred method.
Often second-order equations are solved as two first-order equations; this
approach is called the shooting method (Gilat and Subramaniam, 2008). In my
opinion, the method is cumbersome to program and involves several estimates
of initial values at the point x = a and iteration until the boundary condition at
x = b is satisfied. Yet the method is very popular, and the Matlab function bvp4c
uses this method to solve second-order boundary value problems. Nevertheless,
I do not recommend this method.
An example of the solution of a nonlinear equation in the context of a system
of two equations is discussed next.

10.10.4 Systems of ordinary differential equations

Consider the case where the system of equations is generated from a higher-order
ordinary differential equation; consider the Blasius boundary layer problem
f  + f f  = 0
f  (0) = f (0) = 0 (10.170)
f  (∞) = 1
where, as noted earlier, the function f  is the velocity. This equation is most
easily solved as a system of equations because the third derivative leads to a
406 Essentials of Micro- and Nanofluidics

nontridiagonal system, which is more difficult and time consuming to solve. To

convert this equation to a system of equations, write
x
f = g f = gd x + f 0 ; f 0 = f (0) = 0 (10.171)
0

g  + f g  = 0 g(0) = 0, g(∞) = 1 (10.172)

To determine ∞, we follow the steps mentioned previously and assume that x∞ is

some “large” number; usually 5–8 is enough, but the precise number depends on
the problem. Several values of x∞ should be checked to make sure the solution
does not change much for the smaller values of x. The quantity x ∞ should be as
small as possible, obviously. The discretization of equation (10.172) results in

g j+1 − 2g j + g j−1 g j+1 − g j−1
+ f j =0 (10.173)
h2 2h
and multiplying by h 2 ,
   
h h
g j−1 1 − f j − 2g j + g j+1 1 + f j = 0 j = 2, . . . , N − 1 (10.174)
2 2
Using the trapezoidal rule with the first internal mesh point labeled as j = 2,
h
fj = g j−1 + g j+1 + f j−1 j = 2, . . . , N (10.175)
2
The problem is that f j in equation (10.174) is unknown. The final linearized
equation is thus
   
h n−1 h n−1
g j−1 1 − f j
n
− 2g j + g j+1 1 + f j
n n
= 0 j = 2, . . . , N − 1
2 2
(10.176)
where n denotes the latest iteration level. Note that only d N −1 = 0 because
y(N ) = 1 and thus
   
h h n−1
g nN −2 1 − f Nn−1 − 2g n
= −g n
1 + f (10.177)
2 −1 N −1 N
2 N −1
The procedure here would be as follows:
r Specify x∞ , N .
r Guess f  or g, f  = x, say, g = x 2 /2.
r Put the guessed value of f into equation (10.176) .
r Solve using the tridiagonal system solver.
r Integrate to get f .
r Check for convergence.

A Matlab program similar to the linear ODE solver can be written by adding a
convergence step, which may look as follows:
jc=0
for i=2:n-1;
small(i)=abs((f(i)-fold(i))/f(i));
407 Essential Numerical Methods

if small(i)>eps
jc=1
end

Here “jc” is a parameter that signifies whether convergence has been achieved;
here “jc = 1” means that the scheme is not converged. This relative test requires
all entries of the small array to be less than the convergence criteria eps
“(jc = 0); if only one entry is greater than eps, the tridiagonal system must
be solved again.

10.10.5 Derivative boundary conditions

As noted in Chapter 3, many heat, mass, and electrochemical problems involve
derivative boundary conditions. This entails some adjustments to the tridiagonal
solver, and there are two ways to deal with the derivative boundary condition.
Consider now the equation

y  + p(x)y  + r (x)y = f (x) (10.178)

but now with

y0 = y(a) = A, y  (b) = B = constant (10.179)

The discretization of equation (10.178) leads to the set of difference equations of

the form

b j y j−1 + a j y j + c j y j+1 = h 2 f j j = 1, . . . , N − 1 (10.180)

However, note that y(b) (y N ) is unknown, and so the standard back substitution
procedure is not possible. There are two ways to deal with this issue.

CENTRAL DIFFERENCES
Here assume that the equation holds at x N = b. Thus, for j = N ,

b N y N −1 + a N y N + c N y N +1 = d N = h 2 f j (10.181)

where y N +1 is a fictitious point outside the boundary. To eliminate y N +1 , approx-

imate the derivative at x N = b by central differences:
y N +1 − y N −1
=B
2h
y N +1 = 2h B + y N −1 (10.182)

Substituting into equation (10.181), the last equation becomes

b N y N −1 + a N y N + c N (2h B + y N −1 ) = d N = h 2 f j (10.183)

and on rearranging,

(b N + c N ) y N −1 + a N y N = d N − 2h Bc N (10.184)
408 Essentials of Micro- and Nanofluidics

The tridiagonal system now has another equation (10.184) added to it, and so the
system of equations to be solved is

b j y j−1 + a j y j + c j y j+1 = d j j = 2, . . . , N − 1

(b N + c N ) y N −1 + a N y N = d N − 2h Bc N (10.185)

and there are N − 1 equations in N − 1 unknowns instead of N − 2 equations in

N − 2 unknowns.
As a simple example, consider the equation y  − y = x with y(0) = 0, y  (1) =
1, and using h = 1/4. Discretization leads to

b j = c j = 1, a j = − 2 + h 2 = −2.0625 (10.186)

So the equation at x = b = x N would read, with B = 1,

(1 + 1)y N −1 − 2.0625y N = .0625 − 2 × .25 × 1 × 1

2y N −1 − 2.0625y N = −.4375

A weakness of this method is that the error in our difference equations is O(h 4 )
(i.e., h 2 × h 2 ) when multiplying through by h 2 ; however, the N th equation,
because a central difference equation is used, becomes only h 2 × h = h 3 , accu-
rate after multiplication by h. This criticism is somewhat overshadowed by the
ease of incorporation into the numerical scheme.

SLOPING DIFFERENCES
Instead of using a central difference formula, a sloping difference formula could
be used for the derivative:
3y N − 4y N −1 + y N −2 
y  (b) = + O h2 (10.187)
2h

Multiplying through by h, it is seen that the equation is only O h 3 again. But
adding another term to equation (10.187) by suitable manipulation of Taylor
series expressions for the first derivative results in
11y N − 18y N −1 + 9y N −2 − 2y N −3 
y  (b) = + O h3 (10.188)
6h

which, when multiplying through by 6h, is O h 4 accurate. To incorporate
equation (10.187) into the numerical scheme, suppose we have used the Thomas
algorithm and computed the arrays [αi ], [γi ]. Then recall that
γi − ci yi+1
yi = i = N − 1, . . . , 1 (10.189)
αi
assuming y0 is the left end point. If y  (b) = B, using equation (10.187),

3y N − 4y N −1 + y N −2 = 2h B (10.190)
409 Essential Numerical Methods

and using equation (10.189),

γ N −1 + c N −1 y N
y N −1 =
α N −1
γ N −2 + c N −2 y N −1
y N −2 =
α N −2
γ N −2 + c N −2 (γ N −1 + c N −1 y N ) /α N −1
y N −2 =
α N −2
Substituting these two values into equation (10.190),

4 (γ N −1 + c N −1 y N ) γ N −2 + (c N −2 γ N −1 + c N −2 c N −1 y N ) /α N −1
3y N − + = 2h B
α N −1 α N −2
(10.191)
Solving for y N ,

[3α N −1 α N −2 − 4c N −1 α N −2 + c N −2 c N −1 ] y N
= 2h Bα N −1 α N −2 + 4α N −2 γ N −1 − α N −1 γ N −2 − c N −2 γ N −1 (10.192)

Equation (10.192) determines y N because everything else is known. The same

can be done with equation (10.188); however, the algebra is rather complicated.
When equation (10.188) is used, the accuracy problem is solved.

10.10.6 Convergence tests and Richardson extrapolation

As mentioned earlier, the solutions to the nonlinear boundary value problems
and other nonlinear problems require the iterative calculations of the solutions to
linearized systems of equations. Thus programs must include a test for solution,
and if y denotes the solution to the system of equations, this test at the nth iteration
is written
 
 y n − y n−1 
 j j 
  < eps all j (10.193)
 y nj 

or in the present notation, all the internal mesh points 2 ≤ j ≤ N − 1. It is

important to note that this equation must hold for each point in the mesh j. This
is a relative test that indicates how many significant digits have been retained in the
iterative process; for example, if eps = 10−4 , this formula means that successive
iterates must agree to within four significant digits at each internal mesh point.
This test is independent of the magnitude of the dependent variable y, in contrast
to the absolute test, which omits the denominator in equation (10.193). In that
case, if y  O(1), the algorithm may converge, even if there are no significant
digits that agree in successive iterations.
In any case, common sense should govern the choice of a convergence criterion,
the value of eps. For very small values of y, the relative test may fail, and in
410 Essentials of Micro- and Nanofluidics

that case, the absolute test can be used locally. In this case, we would usually be
satisifed with testing down to a minimum value of y, say, y ≥ 10−7 .
Another test for the accuracy of the algorithm is to use Richardson extrapo-
lation. The idea is that if the formulas used to approximate the derivatives of a
boundary value problem are O(h 2 ) accurate, then at any internal mesh point, the
solution is

y 1j = yT j + Ah 2 + Bh 4 + · · · (10.194)

where A and B are constants associated with the error in the Taylor series
approximation to the derivative and yT j is the “true” solution. As noted earlier,
to determine accuracy, we would reduce the step size h to h/2 and repeat the
calculation. Thus we have a second solution of the form
h2 h4
y 2j = yT j + A +B + ··· (10.195)
4 16
if it is assumed that A and B are approximately constant. The term Ah 2 may be
eliminated, and solving for the “true” solution,
4y 2j − y 1j Bh 4
yT j = − + ··· (10.196)
3 4
at each mesh point. The key assumption is that A and B are approximately constant
in the interval (x j − 2hn , x j + 2hn ). This will not normally be exact; however, the
approximation will get better as the mesh size decreases. Extrapolation by itself
is risky, and we would use this Richardson deferred approach to the limit after
halving the interval twice and comparing the extrapolated result with the finest
mesh calculation.

10.10.7 Solving boundary value problems with Matlab functions

The Matlab function bvp4c will solve boundary value problems of order 2 or
greater by converting the ODE to a system of first-order equations. This function
solves the resulting first-order system using a finite difference approach. It is
called in the form

solution = bvp4c( ode , odebc , initsol)

where ode is a file that contains the system of equations, odebc is the statement
of the boundary conditions, and initsol is the initial guess to the solution.
Consider the simple second-order linear equation considered previously: y  −
y = x with y(0) = 0, y(1) = 1 for  = 1. To use bvp4c, this equation is written
in the form

y = z (10.197)
z = y + x (10.198)

The ode file containing the system of first-order equations is

 411 Essential Numerical Methods

function dydx=ode(x,y)
dydx=[y(2) y(1)+x];

and the boundary condition file is

function bc=odebc(ya,yb)
y0=0; y1=1;
bc=[ya(1)-y0 yb(1) -y1];

The main Matlab program looks as follows:

initsol=odeinit(linspace(0,1,11),[1,1])
sol=bvp4c(’ode’,’odebc’,initsol)
plot(sol.x,sol.y(1,:))
xlabel(’x’);xlabel(’y’)

The “linspace” command generates a vector x = [0, 1] using N = 11 points.

The Matlab function “bvp4c” is certainly not limited to linear equations, and,
for example, the Blasius problem can be solved in this manner.

10.11 Solution of initial value problems

10.11.1 Introduction
By definition, first-order differential equations are of the initial-value type. There
are many ways to solve these types of equations using both single-step and
multistep methods. To begin to understand these methods, consider a first-order
equation of the form

dy
= f (x, y) with y(0) = y0 (10.199)
dx

where f (x, y) is a given function. In some cases, an analytical solution can be

found; for f (x, y) = 1 and y(0) = 1, the solution is clearly

x2
y(x) = 1 + (10.200)
2

As you might guess, such a simple equation often does not occur in applications.
There are several methods for solving such problems, and these methods are
discussed next.
There are a wide number of applications for which first-order equations must
be solved. As has been seen, the streamlines, pathlines, and streaklines of a fluid
flow are governed by first-order equations. First-order equations are particularly
common in biochemical reaction kinetics; for example, the action of enzymes is
governed by a set of first-order differential equations.
412 Essentials of Micro- and Nanofluidics

Michaelis and Menten (Murray, 2001) suggested that enzymes react on sub-
strates according to the two-step reaction
k1 k2
S+E SE → P + E (10.201)
k−1

with each reaction governed by rate constants k1 and k2 and the reverse rate
constant k−1 . This theory of enzyme action is meant to describe the case of
a dilute solution; in concentrated solutions, the rate constants may depend on
concentration. This equation states that one molecule of S combines with one
molecule of E to form the complex S E, which then produces one molecule of
P. The law of mass action states that the reaction rate is proportional to the
concentration, and letting, for example, s = [S], the concentration of S with
corresponding expressions for E, P, S E, there are four equations:
ds
= −k1 es + k−1 c
dt
de
= −k1 es + (k−1 + k2 )c
dt
dc
= k1 es − (k−1 + k2 )c
dt
dp
= k2 c (10.202)
dt
where c = [S E]. The units of the rate constants are sec−1 M−1 , where M is
molar. The initial conditions are

s(0) = s0 , e(0) = e0 , c(0) = p(0) = 0 (10.203)

The enzyme E only facilitates the reaction, and so adding the second and third
equations and integrating,

e(t) + c(t) = e0 (10.204)

Substituting into equations (10.202) (actually, just the first two equations), there
are two distinct equations:
ds
= −k1 e0 s + (k1 s + k−1 )c (10.205)
dt
dc
= k1 e0 s − (k1 s + k−1 + k2 )c (10.206)
dt
with the appropriate initial conditions on s, c as earlier. Note that p may be calcu-
lated once c is known. The rate constants are usually obtained from experiment,
and in the general case, these are two nonlinear first-order differential equations
for the concentration s and the complex se.
First-order equations are also common in drug delivery models, called com-
partment models (Saltzman, 2001). Pharmacokinetics is the study of the effect
of a drug on the body. In its simplest form, pharmacokinetics is the study of the
injection of a drug into one or more compartments in the body that is subse-
quently eliminated in at least one compartment. The major organs of the body are
413 Essential Numerical Methods

often modeled as two compartment regions in which the drug may be injected,
absorbed, or eliminated. For example, if the concentration of a drug is assumed to
be constant within a compartment, a well-stirred model, then the concentration
within compartment 1 is determined by the equation (Saltzman, 2001)

dc1 
= Q c0 e− τ − c1
t
(V1 + V2 k(c2 )) (10.207)
dt
where k is an equilibrium constant but is a function of the concentration in the
second compartment c2 , Q is the volume flow rate of blood through the organ, V1
and V2 are the volumes of the two compartments, and τ is a time constant. This
equation is a simple statement of conservation of mass of the drug. Of course,
the concentration of drug in the compartment may not be constant, leading to a
partial differential equation.
In this section, several different methods of solving such first-order differential
equations are described, beginning first with an analytical method using the Taylor
series.

10.11.2 Taylor series method

This method allows the determination of the initial behavior of the solution of
a differential equation but is not suited for the development of a longer-term
numerical solution. Let us consider the equation

dy
= e−x (y 3 + 1) with y(0) = 2 (10.208)
dx
Note that the function f (x, y) is nonlinear in y and that an analytical solution, a
solution expressible in simple functional form, is not evident. The problem is to
compute the solution y for x > 0.
To do this, a Taylor series approximation about t = 0 is given by

x2
y(x) = y(0) + y  (0)x + y  (0) + ··· (10.209)
2
Because y(0) = 2, the derivative dy/d x(0) is also known; moreover, the second
derivative is also known simply by differentiating equation (10.208). In this way,
the first three terms of the Taylor series are

x2
y(x) = 2 + 9x + 99 + ··· (10.210)
2
Of course, this is not the full solution because there are theoretically an infinite
number of terms in the Taylor series. However, equation (10.210) gives a good
approximation to the function y for very small values of x. The disadvantage is
that the Taylor series will not converge for larger values of x. There is value in
this approach because the Taylor series method can be used as a starting formula
for a multistep method, discussed later.
 414 Essentials of Micro- and Nanofluidics

A graphical depiction of Euler’s method.

Figure 10.18

10.11.3 Euler methods

Euler methods approximate the derivative dy/d x by a first- or second-order finite
difference; the simplest formula is the Euler forward formula:

y j+1 = y j + h f (x j , y j ) for j = 0, 1, 2, · · · (10.211)

where n denotes the number of the time step and h = x is length of the step in
x. This formula approximates the derivative with a forward difference formula.
This Euler forward formula is explicit in that the solution yn+1 follows from the
values before. This method is thus easy to program and is fast computationally,
and the step h = x can be changed easily. However, the local error is relatively
large, and

E L = h 2 y  (0) + · · · (10.212)

and over a number of time steps of O(1/ h), the global or propagated error
E G ∼ O(h). In practice, very small steps are required for this method that in
general is not used. A graphical or geometric interpretation of the Euler method
is depicted in Figure 10.18.
The accuracy can be increased by using a central difference approximation to
the derivative, which is equivalent to the equation
h
y j+1 = y j + f (x j , y j ) + f (x j+1 , y j+1 j = 0, 1, 2, . . . (10.213)
2
The only difference is that the function f is evaluated at the midpoint of the
interval. The error is now E ∼ O(h 3 ); however, because y j+1 is unknown, this
equation is implicit. The value y j+1 can be predicted using the standard Euler
formula, equation (10.210). However, error is incurred in just guessing y j+1 , and
so iteration is necessary. The procedure could go as follows:
Step 1. Predict y j+1 using the Euler formula
P
y j+1 = y j + h f (x j , y j ) (10.214)

where P indicates the prediction step.

Step 2. Correct this value with the modified Euler formula:
h
y Cj+1 = y j + f (x j , y j ) + f (x j+1 , y j+1
P
(10.215)
2
where C indicates the correction step.
415 Essential Numerical Methods

P
Step 3. At this point, we have values for y j+1 and y Cj+1 , and we need to check
to see if they are close. We would use a relative test to check to see if
 
 yC − y P 
 j+1 j+1 
  < eps (10.216)
 y Cj+1 

where  is a specified small number that may vary with the problem. A value that
often gives good results is  = 10−4 .

Step 4. If the test is not satisfied, then repeat step 2 using the formula

h
j+1 = y j +
y n+1 f (x j , y j ) + f (x j+1 , y nj+1 (10.217)
2
where n denotes the number of the iteration.
Generally, only a few iterations are necessary, and it is possible that choosing a
small enough h = x will remove the need to correct or iterate. It can be shown
that for this process to converge,
 
 ∂f 
h  < 2 (10.218)
 ∂y 

which imposes an upper limit on the mesh length h.

Consider the example problem

dy
= 1 + y 2 , y(0) = 0 (10.219)
dx
which has the exact solution y = tan x. Suppose h = 0.1 to start, and use two
correction steps. Then the solution is displayed in Table 10.13. Note that the
modified Euler solution is much better than the Euler solution, although the
error even for the modified Euler is growing. In practice, the step size for this
problem should be smaller rather than attempting to use more corrector steps. In
practice, even the modified Euler method is not used much, and there are better
single-step methods, such as Runge–Kutta methods, and these are discussed
next.

Table 10.13. Solution to the ordinary differential equation

given by equation (10.219)
x Euler Mod. Euler Exact
0 0 0 0
0.1 0.1000 0.1005 0.1003
0.2 0.2010 0.2031 0.2027
0.3 0.3050 0.3100 0.3093
0.4 0.4143 0.4238 0.4227
0.5 0.5315 0.5478 0.5464
416 Essentials of Micro- and Nanofluidics

10.11.4 Runge-Kutta methods

Runge–Kutta methods are single-step methods that can be made as accurate
as necessary by increasing the number of function evaluations. Runge–Kutta
methods are of the form
y j+1 = y j + a1 k1 + a2 k2 + a3 k3 + · · · + a M k M (10.220)
where the ai are coefficients, depending on the order of the method, and the ki
are values of the function f evaluated at specific points:
k1 = h f (x j , y j ) (10.221)

k2 = h f (x j + α1 h, y j + β1 k1 ) (10.222)

k3 = h f (x j + α2 h, y j + β2 k2 ) (10.223)
and so on. The number of terms taken in the expression is the order of the
method; in equation (10.220), the order of the method is M. The arrays αi and βi
are chosen from the expression for the Taylor series of the function y about y j ,
or
h 2  h 3 
y j+1 = y j + hy j + y + yj + · · · (10.224)
2 j 6
where, for example, y j = y  (x j ). The equation of interest is
y  = f (x, y) (10.225)
and consider the second-order Runge–Kutta formula. Note that
df
y  (x) = (10.226)
dx
but f = f (x, y) so that
df ∂f d f ∂y
= + (10.227)
dx ∂x dx ∂x
Substituting into equation (10.224),
 
h2 ∂ f d f ∂y
y j+1 = y j + h f j + + (10.228)
2 ∂x dx ∂x j
∂y
where f j = f (x j , y j ) and f j = ∂x j
. Now compare equation (10.228) with the
formula
y j+1 = y j + a1 k1 + a2 k2 (10.229)
Here it is clear that k1 = h f j , and a1 = 1 will make the second term in each
of the equations the same. From the definition of k2 = h f (x j + α1 h, y j + β1 k1 ),
expanding in a Taylor series,
∂f ∂f
k2 = h f (x j + α1 h, y j + β1 k1 ) = h f j + α1 h 2 + β1 h 2 k 1 f j + ···
∂x j ∂y j
417 Essential Numerical Methods

Equations (10.228) and (10.229) match to O(h 2 ) if the following conditions

hold: a1 + a2 = 1, a2 α1 = 1/2, and a2 β1 = 1/2. Note that these conditions have
multiple solutions but that the standard symmetrical forms used most often are
a1 = a2 = 1/2 and α1 = β1 = 1. Thus the second-order Runge–Kutta formula is
given by
1
y j+1 = y j + (k1 + k2 ) + O(h 3 ) (10.230)
2
where k1 = h f j and k2 = h f (x j + h, y j + k1 ).
Higher-order formulas are obtained by taking more terms in the Taylor series,
and again, there are multiple solutions for the coefficients ai , αi , and βi . The most
common fourth-order formula is
1
y j+1 = y j + (k1 + 2k2 + 2k3 + k4 ) + O(h 5 ) (10.231)
6
where

k1 = h f j (10.232)

k2 = h f (x j + h/2, y j + k1 /2) (10.233)

k3 = h f (x j + h/2, y j + k2 /2) (10.234)

k4 = h f (x j + h, y j + k3 ) (10.235)

This is the Runge–Kutta method in most use. The fourth-order Runge–Kutta has
several advantages:
r It is very easy to program.
r It is cumulative fourth order accurate after a number of steps of O(1/ h).
r Since it is a one-step method, it is self-starting.
r While there are some Runge–Kutta methods that are unstable in some sense,
this method is highly stable.

The only real disadvantage of the fourth order Runge–Kutta method is that
there are at least four function evaluations at each time step, adding to the expence
of the calculation, and all of the information is thrown away as it is calculated.
In addition, there is no good estimate of the error incurred. As with boundary
value problems, to show that a given solution is accurate, the solution should be
calculated using two different grid sizes, say, h and h/2. If the solution agrees to
a sufficient number of digits, then the solution is said to be verified.
To understand a Runge–Kutta calculation, consider the linear ODE
dy 1
= 3x + y (10.236)
dx 2
418 Essentials of Micro- and Nanofluidics

with y(0) = 1. Assuming h = 0.1, the values of the ki for the fourth-order scheme
at the first nonzero value of x can be calculated as
 
1
k1 = h f (0, 1) = 0.1 0 + = 0.05 (10.237)
2

k2 = h f (0.05, 1.015) = 0.1(0.15 + 0.5125) = 0.06625 (10.238)

k3 = h f (0.05, 1.033125) = 0.06666 (10.239)

k 4 = 0.08333 (10.240)

With these values,

1
y(0.1) = (k1 + 2k2 + 2k3 + k4 ) = 1.06652 (10.241)
6
The value at x = 0.2 is calculated similarly, beginning with the value y(0.1).
The exact solution to this problem may be found as

y(x) = 13e x/2 − 6x − 12 (10.242)

and the value y(0.1) = 1.06652, which is identical to the numerical result; actu-
ally, the two results agree to the eighth decimal place. For a larger time step,
for example, h = 0.2, the numerical solution gives y(0.4) = 1.47820, while the
exact result is y(0.4) = 1.47823. Note that now, the two results deviate in the
fifth place, and as x increases, the error will get much bigger.
The procedure is easily extendable to systems of equations. Suppose the system
of equations is defined by

dy
= f (x, y, z), y(x0 ) = y0 (10.243)
dx

dz
= g(x, y, z), z(x0 ) = z 0 (10.244)
dx
Then the ki can be defined for the first equation in terms of the values of f , and
the other coefficients, say, li , are defined in terms of the functional values of g;
for example, at a given time step n, k1 = h f (x j , y j , z j ) and l1 = hg(x j , y j , z j ).
The solutions for (y, z) at the j + 1 time step are given by equation (10.224).
The fourth Runge–Kutta method is the most popular method of all the single-
step methods; it is easy to program and is self-starting and extremely stable.
Conversely, it has four function evaluations, and it is wasteful because it throws
away all the information it has calculated. For these reasons, Runge–Kutta meth-
ods are not often used in computationally intensive problems such as for molecular
dynamics simulations, which are Hamiltonian systems. Nevertheless, it is a good
method and can be used in most problems with good success.
419 Essential Numerical Methods

10.11.5 Adams–Moulton methods

Multistep methods are those methods derived by Taylor series that span two or
more time steps. The most common Adams-Moulton explicit method is given by
 
h 251 5 (5)
yn+1 = yn = (55 f n − 59 f n−1 + 37 f n−2 − 9 f n−3 ) + O h yn
24 720
(10.245)
Another formula can be derived that includes the value of the right-hand side
f n+1 , resulting in the formula
 
h 19 5 (5)
yn+1 = yn = (9 f n+1 + 19 f n + 5 f n−1 + f n−2 ) + O h yn (10.246)
24 720
Note that the error associated with equation (10.246) is on the order of 10 times
less than that associated with equation (10.245). Also it is seen that equation
(10.246) is implicit. Thus an iterative scheme similar to a modified Euler scheme
can be performed with a predictor, corrector step corresponding to equations
(10.245) and (10.246), respectively. It is left as an exercise for the reader to
apply the Adams–Moulton method to the simple ODE addressed in the previous
section.
This method is the best of the ones discussed so far. Note that it is fast and
requires only two function evaluations at each time step, compared to Runge–
Kutta, which requires four evaluations. However, because it is a multistep method,
a starting formula must be used, and Runge–Kutta could be used for this purpose
because the error is of the same magnitude. This method sometimes appears under
the name Adams–Bashforth. There are many other such methods corresponding
to their order and the different departure points for the Taylor series.

10.11.6 Symplectic integrators

It has been mentioned that a second-order initial value problem should usually be
solved as a system of first-order equations. For certain problems associated with
computing particle trajectories, the highly accurate Runge–Kutta and Adams–
Moulton methods may not yield optimal results. This is especially true for sys-
tems defined by a Hamiltonian, an example being the computation of particle
trajectories in classical molecular dynamics simulations discussed in Chapter 11.
The methods described in this section are not normally included in presentations
of numerical methods in engineering textbooks.
A Hamiltonian system is defined as a system of equations that can be written
as
∂H
ξ̇ = (10.247)
∂η
∂H
η̇ = (10.248)
∂ξ
420 Essentials of Micro- and Nanofluidics

where H is the Hamiltonian and ξ̇ = dξ /dt. In classical mechanics, for example,

ξ = y is often a particle position and η is its momentum η = mv = m ẏ; m is a
particle mass.
For example, the harmonic oscillator is defined by the equation
d2 y
+ ky = 0
m (10.249)
dt 2
which may be split into two first order equations for the positions and momentum
as
∂H
ẏ = (10.250)
∂p
1 ∂H
ṗ = − (10.251)
m ∂y
where the Hamiltonian is given by
1 2 p2
ky +
H= (10.252)
2 2m
The key feature of systems of equations defined by Hamiltonians is that the total
energy in the system, the Hamiltonian, H = constant = H0 , is a constant fixed
by its value at the initial time. To show this, note that the work done by a force is
given by
2
W12 = F(y)dy (10.253)
1

where the force F may be defined by a potential

d
F(y) = − (10.254)
dy
and so
W12 = 2 − 1 (10.255)
In the same way, integrating the force part of the equation
1
W12 =
m(v22 − v12 ) (10.256)
2
and equating the two expressions for the work done, it is seen that H1 = H2 ,
meaning that d H /dt = 0 and
H = constant = H0 (10.257)
Numerical schemes that preserve this feature of the problem are called symplectic.

Suppose the harmonic oscillator problem is solved by using the explicit Euler
method. Then
yn+1 = yn + hpn (10.258)

pn+1 = pn − hyn (10.259)

421 Essential Numerical Methods

where it has been assumed that the spring constant and the mass have the
value of 1, k = m = 1. Then, calculating the Hamiltonian directly, it is evident
that
1 2 p2
Hn+1 = yn+1 + n+1 = Hn (1 + h 2 ) (10.260)
2 2
so that the Hamiltonian grows at each time step, violating the condition that the
Hamiltonian be constant; that is, the Euler method is not symplectic. In the same
way, it can be shown that the classical fourth-order Runge–Kutta method and the
Adams–Moulton method are not symplectic, although Runge–Kutta methods can
be made symplectic by adjustment of the coefficients (Hairer et al., 2006).
By adjusting the explicit Euler scheme so that one of the equations is implicit,
the scheme will be symplectic. Suppose the Euler scheme for the harmonic
oscillator is given by

yn+1 = yn + hpn+1 (10.261)

pn+1 = pn − hyn (10.262)

Then direct substitution for the Hamiltonian shows that

1 2 p2
Hn+1 = yn+1 + n+1 = Hn + h 2 ( pn+1
2
− yn2 ) (10.263)
2 2
Because both p and y are harmonic, the error in the Hamiltonian is bounded and
does not grow. This algorithm is called the symplectic Euler method.
Consider now the general second-order initial value problem defined by
d2y
= f (y, t) (10.264)
dt 2
subject to
dy
y= = 0 at t = t0 (10.265)
dt
Let v = ẏ; then the second order equation can be written as two first order
equations according to

v̇ = f (10.266)

ẏ = v (10.267)

The second-order equation (10.264) may be solved directly by approximating

the second derivative at time t = tn by the central difference approximation, and

yn+1 − 2yn + yn−1 = h 2 f n for n = 1, · · · (10.268)

Note that this method, called a Verlet algorithm, is reversible in the sense that
exchanging n + 1 and n − 1 and changing h to −h does not alter the equation.
This is an especially important property for systems of equations that describe
particle paths (Hairer et al., 2006).
 422 Essentials of Micro- and Nanofluidics

d
q′

q
εe

Two charged particles in an electrolyte solution.

Figure 10.19

Equation (10.268) is equivalent to the set of two approximations to the first-

order systems
h
vn+1/2 = vn + fn for n = 1, · · · (10.269)
2
yn+1 = yn + hvn+1/2 for n = 1, · · · (10.270)
and the use of the central difference approximation
yn+1 − yn−1
vn = for n = 1, · · · (10.271)
2h
To complete the scheme, one method of updating the velocity at the point n + 1/2
is
vn+1 = vn + h f n+1/2 for n = 1, · · · (10.272)
In the molecular dynamics literature, these symplectic methods are called Verlet
methods (Verlet, 1967).

The Verlet Algorithm

The term Verlet algorithm refers to Loup Verlet who “invented” the algorithm
and published the paper in 1967 (Verlet, 1967). According to Hairer et al. (2006),
this method actually appeared in Newton’s Principia. The method appears to
have been used in astronomy to predict the trajectories of the planets in the solar
system by Störmer (1907). Thus, the method is often called the Störmer-Verlet
algorithm. In the numerical solution of partial differential equations, such as the
heat equation, it is often called the leap-frog method.

For example, consider the case of two colloidal particles of mass m and total
charge q and q  in a fluid of permittivity e . For simplicity, the origin can be taken
to be the position of one of the particles, and the problem is reduced to finding
the trajectory of the second particle relative to the first. The situation is depicted
in Figure 10.19. Assume that the only force on the particles is the electrostatic
force (omitting the 4π)
e qq 
F= (10.273)
r2
and that the fluid is at rest. The equations of the two particles can be made
dimensionless on the initial distance between the two particles, and then the time
scale is given by
md 3 e
τ= (10.274)
| qq  |
 423 Essential Numerical Methods

1.5
0.01

1
0.008

0.5

H−Hi
0.006
y

0
0.004

−0.5 0.002

−1 0
−2 −1.5 −1 −0.5 0 0.5 0 50 100 150 200 250
x t
(a) (b)

The Verlet algorithm integration of the elliptical orbit of two charged particles around each other: (a) the
Figure 10.20
trajectory; (b) time evolution of the deviation of the Hamiltonian from its initial value, H − H i . Here x 0 = 1 − β,

y0 = 0, v x (0) = 0, and v y (0) = v0 , with v0 = (1 + β)/(1 − β) and β = 0.6.

where m is the mass of the particle and d is the distance between the two
particles. This time scale can be used to define the dimensionless time variable.
The dimensionless equations of motion of the particle are thus given by
d2x x
= −α  3/2
(10.275)
dt 2 x + y2
2

d2 y y
2
= −α  3/2
(10.276)
dt x 2 + y2
where
qq 
α= (10.277)
| qq  |
If α = 1, the particles are of the same charge, and the dimensionless equations
are the same as the classical two-body problem of Kepler in celestial mechanics,
for which the force is given by
Gm 1 m 2
F= (10.278)
r2

and the time scale for a particle of mass m is τ = h 3 /mG. For α = 1, the
particles will revolve around each other in an elliptic orbit as in the two-body
problem of celestial mechanics. This is a Hamiltonian system with
1 2 1
H= (vx + v 2y ) +  (10.279)
2 x + y2
2

where, in dimensionless form, the momentum is equivalent to the velocity px =

vx . Thus the Verlet algorithm can be used to integrate the system forward in time.
As a specific example, consider the case where x(0) = x 0 and y(0) = y0 , with
ẋ(0) = 0 and ẏ(0) = v0 . The solution for the trajectory using the Verlet algorithm
can be compared with that using the simple explicit Euler method. The results are
depicted in Figures 10.20 and 10.21. The results for the Verlet algorithm in Figure
10.20 indicate that the trajectory is elliptical and that the Hamiltonian remains
 424 Essentials of Micro- and Nanofluidics

1.5 0.25

1
0.2

0.5
0.15

H−Hi
0
y

0.1
−0.5

0.05
−1

−1.5 0
−3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5 0 50 100 150 200 250
x t
(a) (b)

Euler integration of the elliptical orbit of two charged particles around each other: (a) the trajectory; (b) time
Figure 10.21
evolution of the deviation of the Hamiltonian from its initial value, H − H i . Here x 0 = 1 − β, y0 = 0, v x 0 = 0,

and v y0 = v0 , with v0 = (1 + β)/(1 − β) and β = 0.6.

approximately constant, with a periodic deviation from the initial value of small
amplitude. For the explicit Euler method, the trajectory expands dramatically and
the Hamiltonian (the total energy) grows with time, clearly violating the constant
energy condition very early in the computation.
It is left as an exercise to solve these equations using a fourth-order Runge–
Kutta method or the Adams–Moulton method, neither of which, in its standard
form, is symplectic. The interested reader should consult the comprehensive
treatment of symplectic algorithms by Hairer et al. (2006).

10.11.7 Stiff equations and stability

Stiff differential equations arise when a set of initial value problems contains
widely varying time scales. This will occur when the coefficients of a system
of initial value problems are large. Consider, for example, the set of equations
(Fausett, 2008)
ẋ = 98x + 198y (10.280)

ẏ = −99x − 199y (10.281)

with x(0) = 1 and y(0) = 0. The solution is given by
x(t) = 2e−t − e−100t (10.282)

y(t) = −e−t + e−100t (10.283)

Clearly there are two time scales τL = 1 and τ S = 0.01, a short time scale τ S ,
and a long time scale τ L . If the short time scale must be resolved, then very small
time steps are required. Conversely, because the short time scale is short, we may
choose not to resolve it.
Stiff equations may be solved by several methods. One way is to use an implicit
solver. To see this, consider the equation
ẏ + βy = 0 (10.284)
425 Essential Numerical Methods

where β is a constant. Euler’s method gives

yn+1 = (1 − βh)yn (10.285)
If 1 − βh > 1, the solution grows without bound, and the Euler method is said to
be unstable. Thus the time step must satisfy h < 2/β for stability, and if β  1,
very small time steps are required. Conversely, suppose that the backward Euler
method is used, which is defined by
yn+1 = yn − βhyn+1 (10.286)
leading to
yn
yn+1 = (10.287)
1 + βh
which will be stable, though not necessarily accurate for all h. As can be seen,
implicit methods are usually more stable than explicit methods for the same time
step.
Another method that can be used is to approximate the right-hand side of the
equation at the mid-point of an interval and this equation is
yn+1 = yn + 0.5h( f (tn+1 , yn+1 ) + f (tn , yn )) (10.288)
Applying this equation to the preceding simple equation, we obtain
2 − βh
yn+1 = yn (10.289)
2 + βh
and so the method is stable. More detail on the nature of the stability of the other
numerical methods is given in a variety of texts.
Another way to solve stiff equations is to use a variable time step, the magnitude
of which can be based on an estimate of the magnitude of the right hand side of
the equation. Other algorithms are described by Fausett (2008) and the references
therein.
The stiffness problem may also be addressed using the method of matched
asymptotic expansions, as presented in Appendix A for second-order boundary
value problems. Stiff equations are common in chemical kinetics problems for
which reactions occur on different time scales and lead to singular perturbation
problems that often may be solved analytically, without recourse to the computer
– a powerful tool indeed. Consider the action of enzymes discussed in Section
10.11.1. There the two equations governing the concentration of the substrate
material and the complex ES (= c) were derived according to (Murray, 2001)
ds
= −k1 e0 s + (k1 s + k−1 )c (10.290)
dt
dc
= k1 e0 s − (k1 s + k−1 + k2 )c (10.291)
dt
subject to
s(0) = s0 and c(0) = 0 (10.292)
Here e0 is the concentration of the enzyme; most biochemical reactions require
very little of the enzyme. Thus, in most cases of practical interest,  = e0 /s0  1.
426 Essentials of Micro- and Nanofluidics

If the variables are made dimensionless as t ∗ = k1 e0 t, s ∗ = s/s0 , and c∗ = c/e0 ,

then (after dropping the star)
ds
= −s + (s + K − γ )c (10.293)
dt
dc
 = s − (s + K )c (10.294)
dt
with

s(0) = 1 c(0) = 0 (10.295)

Here K = (k−1 + k2 )/(k1 s0 ) = K m /s0 and γ = k2 /k1 s0 ; K m is called the

Michaelis constant. Also   1, and thus there are two time scales, t, the long
time scale, and t/ = O(1), the short time scale. This system of equations is
thus stiff. As an example of the order of magnitude of some of these parameters,
standard values may be s0 = 1 µM, e0 = 0.012 µM, k1 = 0.1 (µM)−1 sec−1 ,
k−1 = 0.1 sec−1 , and k2 = 0.3 sec−1 (M stands for molar, as usual). Then
K m = 4 µM, and  = 0.012.
The leading-order solutions to equations (10.293) and (10.294) are obtained by
expanding the variables as s = S0 +  S1 + · · · and c = C 0 + C1 + · · · . Then
S0 and C 0 satisfy
d S0
= −S0 + (S0 + K − γ )C0 (10.296)
dt

S0 − (S0 + K )C0 = 0 (10.297)

subject to S0 (0) = 1 and C 0 (0) = 0. Immediately, we see that

S0
C0 = (10.298)
S0 + K
and note that this solution cannot satisfy the initial condition that C 0 = 0. This
problem will be addressed in the analysis to follow.
Substituting for C 0 into equation (10.296), we find that
d S0 γ S0
=− (10.299)
dt S0 + K
which can be integrated to yield

S0 + K ln S0 = A − γ t (10.300)

where A is a constant. The solutions C0 and S0 are termed the outer solutions, in
the language of matched asymptotic expansions.
Setting  = 0 in equation (10.294) means that we have lost the ability to satisfy
the initial condition that c(0) = 0. The problem defined by equations (10.293)
and (10.294) with the initial conditions is thus termed a singular perturbation
problem. The preceding solutions for S0 and C0 are on the time scale t = O(1):
the long time scale, as discussed earlier. To complete analysis, the solution on the
427 Essential Numerical Methods

short time scale t = O() must be calculated. To do this, define a new variable
tˆ = t : the short time scale. Then, on this time scale,
ds
=0 (10.301)
d tˆ
dc
= s − (s + K )c (10.302)
d tˆ
In the same way as earlier, we expand the variables as s = ŝ0 +  ŝ1 + · · · and
c = ĉ0 +  ĉ1 + · · · , and thus
d ŝ0
=0 (10.303)
d tˆ
so that ŝ0 = 1 to satisfy the initial condition. The solution for ĉ0 is found to be
1 
ĉ0 = 1 − e(1+K )tˆ (10.304)
1+K
after satisfying the initial condition ĉ0 (0) = 0. Note that as tˆ → ∞, the complex
reaches the asymptotic value ĉ0 = 1/(1 + K ), which is equal to the limit of the
outer solution as t → 0:

lim ĉ0 = lim C0 (10.305)

tˆ→∞ t→ 0

The solutions ĉ0 and ŝ0 are termed the inner solutions in the language of matched
asymptotic expansions.
It remains to find the constant A. This is done by matching the inner and outer
solutions according to (see Appendix A)

lim ŝ0 = lim S0 (10.306)

tˆ→∞ t→ 0

resulting in A = 1, a result that could have been anticipated from the form of the
solution.
The inner and outer solutions for s and c can be combined in what is called
the uniformly valid solution, and for any variable, this is defined by

fUV = f inner + f outer − CP (10.307)

where the common part (CP) is the term that appears in both the inner and outer
solutions. For s, the common part is sCP = 1, and for c, the common part is
cCP = 1+K
1
. Thus the one term, uniformly valid solution for c is
1 S0 1
c0,UV = C0 + ĉ0 − = − e−(1+K )tˆ (10.308)
1+K S0 + K 1+K
S0,UV + K ln S0,UV = 1 − γ t (10.309)

The uniformly valid solutions for the substrate, the complex, and the enzyme
are shown in dimensionless form in Figure 10.22. Note the boundary layer behav-
ior near t = 0, which is very difficult to resolve in a fully numerical method. Here
the parameters given earlier are used, and note that E + C = 1 for  = 0.1 for
 428 Essentials of Micro- and Nanofluidics

E
0.8

0.6
S, C, E

0.4
S
0.2
C

0
0 1 2 3 4
t

Uniformly valid solution for the Michaelis–Menten substrate, enzyme, complex temporal development. Here
Figure 10.22
E + C = 1 for  = 0.1 for these values of the parameters.

these values of the parameters. Note also that the substrate (S) variation near
t = 0 is much slower than that of the complex C; in the kinetics literature, this
fact is used to justify solving for the substrate concentration only on the long time
scale.

10.11.8 Solving initial value problems using Matlab functions

The Matlab functions ode23, ode45, and ode113 may be used to solve initial
value problems. These functions are called with a statement of the form

[t y] = ode23(’f’, tspan, y0)

where y is the solution, tspan is the interval over which the time integration is to be
performed, say, tspan = [t0 tf], and y0 is the initial condition. For a system
of equations, f and y0 are vectors. The function ode23 uses two second- and
third-order Runge–Kutta solvers, whereas ode45 uses a combination of fourth-
and fifth-order methods. The function ode113 is a variable step size routine that
uses a variety of Adams–Moulton methods. There can be additional entries in
the calling statements for each of these functions, and the Matlab help resources
should be consulted for details. The Matlab functions ode15s and ode23s are
designed specifically to solve stiff initial value problems.

10.12 Numerical solution of the PNP system

As was discussed in Chapter 7, the Poisson–Nernst–Planck (PNP) system is a set

of second-order equations for the solutes in the mixture and the electrical poten-
tial, and these equations can be solved numerically in one dimension using the
429 Essential Numerical Methods

methods discussed for boundary value problems. The equations are discretized
using the second-order accurate central difference approximations and for all the
derivative terms. Recall that the equation for the potential is given in dimension-
less form by

d 2φ
2 = −ρe (10.310)
dy 2
where ρe is the dimensionless volume charge density and the concentrations are
scaled on ionic strength. The finite difference equation for the potential using
central differencing is given by

h2
φi−1 − 2φi + φi+1 = − (z a X a,i + z b X b,i ) (10.311)
2
for i = 2, . . . , N − 1 interior grid points, in a mixture having two solute species
of valences z a and z b .
If the ions are assumed to be point charges, then the equations for the mole
fraction have the classical Boltzman distribution. In the solution for the potential,
that distribution can be used explicitly. Alternatively, the equation for the mole
fraction can be solved in the form
d2 Xa z A X a (z a X a + z b X b ) d Xa d Xa
− + zA =0 (10.312)
dy 2  2 dy dy
Using central differencing, the difference equations are given by
   
1 h2 2
X a,i−1 1 − z a (φi+1 − φi−1 ) − X a,i 2 + 2 z a X a,i
4 
 
1 h2
+ X a,i+1 1 + z a (φi+1 − φi−1 ) = 2 z a z b X a,i X b,i (10.313)
4 
Note that the quantities inside the parentheses are the linearized coefficients of
the tridiagonal system.
The difference equations for the mole fractions are nonlinear, and so iteration
is required. Either equation can be solved first, with the result substituted into
the potential equation. The set of linearized tridiagonal difference equations can
be solved using a direct tridiagonal solver such as the Thomas algorithm or
iteratively using the Gauss–Seidel scheme with successive overrelaxation (SOR).
The iteration scheme is assumed to converge when
 
 X aNew − X aOld 
  < eps (10.314)
 X aNew 

at all grid points, where X a is the mole fraction of species a and eps is the
convergence criterion, usually chosen at first to be eps = 10−4 .
These equations are subject to boundary conditions that can involve specified
values of the potential and mole fractions or derivative boundary conditions. In
 430 Essentials of Micro- and Nanofluidics

dimensionless form, the boundary condition for the potential using the surface
charge density is

dφ −σ 0 h
= at y = 0 (10.315)
dy e φ0
for example. In this case, σ 0 is assumed known, and the sign on the right side of the
equation depends on the direction of the outward unit normal to the surface. Note
that the right side of equation (10.315) is dimensionless. As has been discussed, a
central difference approximation for the derivative at the boundary can be written
and, after elimination of the fictitious point, yields an additional equation, which
is added to the tridiagonal system.
The no-flux condition on the mole fractions corresponds to
d Xa dφ
+ za X a = 0 at y = 0 (10.316)
dy dy
A similar procedure is used here, with the dφ/dy assumed known in the calcu-
lation of the solution for X a .
Of course, because the equations are second-order, another boundary condition
is required. This depends on the value of ; for   1, we would use a specified
condition as y → ∞ or, for  not so small, a boundary condition at the other wall
could be used.
All the numerical methods used to solve this problem have been discussed in
detail, and putting all these modules together to solve the PNP system numerically
is left as an exercise.

10.13 Partial differential equations

In Chapter 3, at least in the linear case, second-order partial differential equa-

tions are classified into three groups: parabolic, elliptic, and hyperbolic equa-
tions. Though the classification procedure is not easily formulated for nonlinear
equations; this classification system is usually assumed to apply for nonlinear
equations. This classification is based on how information travels in each case;
for example, information travels in one direction, forward, in parabolic equa-
tions; in two directions for hyperbolic equations; and information is instaneously
felt everywhere for elliptic equations. Each of these types of equations must be
solved by different methods based on the type of equation. While partial differ-
ential equations do not play a large role in this text, it is useful to discuss how to
solve them.
As we indicate in Section 3.15, parabolic equations are characterized by the
unsteady mass (or heat) transport equation
∂c A
= D A ∇ 2c A (10.317)
∂t
 431 Essential Numerical Methods

where ∇ 2 is the Laplacian operator. Parabolic equations have a single characteris-

tic direction. Parabolic equations are solved by marching the solution forward in
time: the characteristic direction. Hyperbolic equations are solved by the method
of characteristics along the two directions that information travels: the two char-
acteristic curves. Elliptic equations must be solved at all points simultaneously.
The two-dimensional Poisson equation for the electrical potential discussed in
the previous section is an example of an elliptic equation.
In this section, the focus is exclusively on elliptic and parabolic equations,
which are common in micro- and nanofluidics. Hyperbolic equations are less
common in microscale and nanoscale flows.

10.13.1 Elliptic equations

As indicated in Section 3.15, elliptic equations have no characteristic directions.
Thus information travels instantaneously in all directions. To illustrate the numer-
ical discretization and solution procedure, let us return to the two-dimensional
form of the potential equation in dimensionless form, which is
 
∂ 2φ ∂ 2φ
 2
+ = −ρe (10.318)
∂ y2 ∂z 2

where ρe is the dimensionless volume charge density. This equation could repre-
sent the potential due to the EDLs in a square channel (Bhattacharyya & Conlisk,
2005). Each of the second derivatives is represented by its second-order central
derivative expression, and the resulting difference equation is given by

y 2
φ j−1,k − 2φ j,k + φ j+1,k + γ (φ j,k−1 − 2φ j,k + φ j,k+1 ) = − ρej,k (10.319)
2

where γ = y 2 /z 2 . The labeling of the

grid points is depicted in Figure 10.23, and
(j+1,k)
the three-dimensional channel geometry is
depicted in Figure 10.24. This is a set of
linear or, in the case of nonlinear equations,
linearized equations in each of the y, z
(j,k−1) (j,k) (j,k+1) directions that may be solved directly by
sweeping each direction. However, unlike
ordinary differential equations, the differ-
ence equations are not tridiagonal; thus
(j−1,k) a Gauss elimination procedure would be
used. However, because the matrix of coef-
Notation for the labeling of the grids surround-
Figure 10.23
ing the central point ( j , k). ficients is relatively sparse, there is much
wasted computation.
 432 Essentials of Micro- and Nanofluidics

y x

Geometry of the channel; u, v, and w are the fluid velocities in the x , y, and z directions, respectively.
Figure 10.24

Conversely, the equation can be solved iteratively using SOR, and this is the
method that is preferred. To do this, solve for φk, j and
1 y 2
φ j,k = (φ j−1,k + φ j+1,k + φ j,k−1 + φ j,k+1 ) + 2 ρej,k (10.320)
2 + 2γ 
Note that for the volume charge density equal to zero and for γ = 1, the value
of φk, j is just the average of the points around it. Both the direct method and the
iterative method just discussed will be successful if there is diagonal dominance
and the mesh is swept in an orderly fashion. Normally the mesh would be swept
horizontally (through the columns) – and then vertically (through the rows) –
or the order could be reversed. In the latter case, in Matlab, two “for” loops are
required:

for i=2:m-1
for j=2:n-1

where m is the number of rows and n is the number of columns. Generally, for
most problems, the iterative methods converge in far fewer than 100 iterations.
Iterative methods also lend themselves to nonlinear problems because iteration
is already being performed; that is, a separate iterative loop to account for the
nonlinearity is not required.
As with the one-dimensional examples, the SOR method is preferred based on
the Gauss–Seidel iteration. It can be shown based on the analysis of the spectral
radius of the coefficient matrix that the Gauss–Seidel iterative method converges
twice as fast as the Jacobi iteration after a large number of iterations (Smith,
1985).
Specific examples of the solution of elliptic equations, including the Poisson
equation for the potential, are left for the exercises.

10.13.2 Parabolic equations

As noted previously, parabolic PDEs have only one set of characteristics, and
so information travels in a single direction. This fact has a considerable effect
on the numerical method used. Parabolic PDEs can be integrated forward in the
timelike variable; the timelike variable need not physically refer to time but can
 433 Essential Numerical Methods

θ=1 θ=0

θ=0 y
y=0 y=1

The explicit scheme to solve the diffusion equation with specified boundaty conditions; θ is the dimensionless
Figure 10.25
temperature.

be a spatial variable as well. In what follows, however, the timelike variable will
be referred to as being time.
In solving parabolic equations, the solution may easily be found explicitly.
However, because the error grows in time, very small time steps are required.
To remedy this, the equation can be solved implicitly, resulting in a tridiagonal
system of equations that can be solved by the direct and iterative methods we
have discussed previously. This situation is similar to the case for initial value
ordinary differential equations discussed earlier.

EXPLICIT METHODS
Suppose θ represents a dimensionless temperature in the unsteady and one-
dimensional heat conduction equation. In an explicit method, the equation is
approximated at the previous time step, as shown in Figure 10.25. The result for
the temperature θ is

θn+1, j − θn, j θn, j+1 − 2θn, j + θn, j−1

= (10.321)
t y 2
 
θn+1, j = θn, j + r θn, j+1 − 2θn, j + r θn, j−1 (10.322)

θn+1, j = r θn, j+1 + (1 − 2r )θn, j + r θn, j−1 (10.323)

which is an explicit expression for the dimensionless temperature at the new time
step. Note that the scheme is dependent on the ratio of the time step to the spatial
step r = t/y 2 , and for r = 1,

θn+1, j = θn, j+1 − θn, j + θn, j−1 (10.324)

The advantage of explicit methods is that it is easy to program; however, a

disadvantage is that the time step must be very small. In fact, the explicit method
is only valid for r ≤ 0.5; that is, the scheme is convergent and stable only for
r ≤ 0.5 (Smith, 1985).
 434 Essentials of Micro- and Nanofluidics

t+1/2 Δt

θ=1 θ=0

θ=0 y
y=0 y=1

The Crank–Nicolson scheme to solve the diffusion equation.

Figure 10.26

IMPLICIT METHODS: CRANK-NICOLSON

Referring to Figure 10.26, in this case, the equation is approximated at t j+1/2 =
1
(t
2 j+1
+ t j ) = t + 12 t, and the difference equation is
θn+1, j − θn, j 1 θn+1, j+1 − 2θn+1, j + θn+1, j−1
=
k 2 h2
1 θn, j+1 − 2θn, j + θn, j−1
+ (10.325)
2 h2
or
 
2 θn+1, j − θn, j = r θn+1, j+1 − 2θn+1, j + θn+1, j−1

+ r θn, j+1 − 2θn, j + θn, j−1 (10.326)

Collecting terms,

−r θn+1, j−1 + (2 + 2r ) θn+1, j − r θn+1, j+1

 
= 2θn, j + r θn, j+1 − 2θn, j + θn, j−1
= r θn, j−1 + (2 − 2r ) θn, j + r θn, j+1 (10.327)

and this is a triadiagonal system of equations to solve for the temperature at each
time. Thus the tridiagonal solver or the Thomas algorithm discussed previously
can be used to solve the system. Note that r appears again, but here the value of r
need not be so small. Indeed, the Crank–Nicolson method is stable for all values
of r (Smith, 1985).
There are many other schemes to solve these problems, such as the Keller–
Box scheme, in which the second-order equation is converted to two first-order
equations. This scheme is very good for nonuniform grids. The Crank–Nicolson
scheme is easy to program and is the method of choice in most cases.

NONLINEAR EQUATIONS
Nonlinear partial differential equations are handled in the same way as differen-
tial equations. The lowest-order portion of the nonlinear term is linearized. For
 435 Essential Numerical Methods

example, for the equation

 
∂T ∂ ∂T
ρc = K (T ) (10.328)
∂t ∂y ∂y
with the thermal conductivity given by

K (T ) = K 0 (1 + αT )

the nonlinearity is in the conductivity term:

 
∂ ∂T ∂2T ∂ K ∂T
K (T ) = K (T ) 2 + (10.329)
∂y ∂y ∂y ∂y ∂y
In the first term at each iteration, K (T ) and ∂ K /∂ y are known, and the tridiagonal
system is solved in the Crank–Nicolson scheme as discussed previously, in a way
similar to the Blasius equation.

10.13.3 The Matlab PDE solver

Matlab has a PDE solver for both parabolic and elliptic equations. The solver
pdepe solves initial boundary value problems in one space variable and one time
variable. However, there must be at least one parabolic equation in the given
system of equations. As with Matlab’s ODE solver, pdepe converts the PDE into
an ODE with the ime integration performed with the Matlab function ode15s.
The calling syntax is

sol=pdepe(m,pdefun,icfun,bcfun,xmesh,tspan)

where m = 0, 1, 2 denotes rectangular, cylindrical, and spherical coordinates,

respectively. The various functions in the calling statement have the same inter-
pretation as for the ODE solver.

10.14 Verification and validation of numerical solutions

There are two main classes of activity for assessing the accuracy and validity
of numerical calculations of scientific and engineering problems. Verification
is the term used to determine whether the discretized problem converges to
the continuum problem as the number of mesh point and time steps increases,
i.e., whether the given equations have been solved correctly. This procedure is
used for both ordinary and partial differential equations (Figure 10.27). One
method of assessing the accuracy is to identify the number of digits to which
two successive solutions agree. This methodology applies to the finite difference
methods described in this chapter and also to finite volume and finite element
procedures that are beyond the scope of this book.
Validation is the term given to the process of determining whether the correct
physical problem has been solved (Figure 10.27). Validation answers the question,
 436 Essentials of Micro- and Nanofluidics

Illustration of the process of verification and validation of a numerical solution of an ordinary or partial
Figure 10.27
differential equation. Adapted from Oberkampf et al. (1998).

is the computational model an accurate physical representation of the actual real-

world problem? Both terms were developed to apply to computational fluid
dynamics simulations, but these concepts also apply to the much simpler cases
in this book (Oberkampf et al., 1998).
These are serious questions, especially in situations where even the physical
properties of a material are unknown. This is one element in the definition of
the term uncertainty. Moreover, in industry, there are often fiscal constraints that
limit the extent to which a given code can be verified and validated. These issues
and more are described by Oberkampf et al. (1998) in great detail; in this section,
we present a short summary of these concepts.
As we have already discussed, the most common way of verifying a computer
code is to systematically reduce the grid and compare successive results. Indeed,
the main objective of verification is to quantify the effect of the grid on the results.
The objective of grid and time step refinement is to estimate the discretization
error of the numerical solution and to expose programming errors and blunders.
The discretization error should monotonically approach zero as the grid size and
time step approach zero. The defining characteristic in this monotonic region is
that the order of accuracy of the difference equations being solved is constant as
the grid and time step are reduced. When this is true, Richardson’s method can
be used to extrapolate to an additional accuracy level. At this point, one can say
that the numerical scheme is both grid and time step convergent. When a code
is accurate to second order, the truncation error is reduced by 4 when the grid
size is reduced by 2. This is because the error associated with the discretization
is O(h 2 ), where h is the grid size in any given direction.
An additional complementary means of verifying a code is to compare with
an available analytical solution – a solution that can be expressed in “simple”
functions such as the circular functions (sine and cosine), exponentials, or com-
binations of exponentials; the hyperbolic functions; or a Fourier series. A similar
procedure could be used to compare with an already verified numerical solution.
Both types of solutions are called benchmark solutions. One example of a bench-
mark solution is the Blasius flat plate boundary layer.
437 Essential Numerical Methods

Note that both these methods of verification may require simplification of

the code to an extent that many of the features of the code used to describe
a given physical problem may not be exercised. Thus a systematic reduction
of the grid size is the best means of verifying a code. Blunders or coding
errors can be identified in this way; if the code is second-order accurate and
the results do not exhibit second-order reduction in error, something is wrong.
Similar comments apply to time step reduction. Of course, another source of
error is lack of convergence of an iterative process, and this aspect of the code
must be tested as well. A summary of the verification process is depicted on
Figure 10.27.
A relatively recent means of verification of numerical solutions is the method
of manufactured solutions (Roache, 1998; Roache & Steinberg, 1994; Oberkampf
& Blottner, 1998; Roy et al., 2004). In this method, given a partial differential
equation, a specific analytical form is chosen and substituted into the equation,
generating source terms on the right-hand side. The modified equation, including
source terms, is then discretized and solved numerically. A symbolic manipulator
such as Mathematica can be used to generate the source terms (Roy et al., 2004).
Because the manufactured solution generated the right-hand side, the numerical
solution of the modified equation should be the manufactured solution itself. A
measure of the accuracy is thus the difference between the numerical result and
the assumed manufactured solution, or

E = f Num − f Man (10.330)

at each grid point. While manufactured solutions are of greatest value when the
grid cannot be decreased, as in coupled problems involving large sets of equations,
the method can also be used to determine the accuracy of numerical solutions of
ODEs and even integrals.
The most common means of validation is to compare the computational results
with experimental data. As is well known, the experimental data contain error,
and the owner of the experimental data should always include error bars for a
proper comparison. The computational results need not exactly agree with the
experimental data. Often in difficult and large systems, the computational model
may only predict trends. In the simpler unit problems described here, comparison
with the experimental data is likely to be much better. As the complexity of the
physical system and its model increases, so, too, does the validation process. A
summary of the validation process is also depicted on Figure 10.27.
Calibration, or the process of adjusting certain physical parameters to better
agree with experimental data, is not validation. However, in very difficult prob-
lems, where crucial information is not available, calibration may be a necessary
substitute.
Validation at the nanoscale, especially of dynamic phenomena, requires addi-
tional scrutiny. Although techniques such as transmission electron microscopy
(TEM) and near-field scanning optical microscopy (NSOM) are capable of sub-
micron spatial resolution, most techniques that are nonintrusive and capable of
interrogating flow in sealed microchannels are limited to length scales (i.e., spatial
 438 Essentials of Micro- and Nanofluidics

resolutions) of at least 1 µm, and hence validation of spatially complex nanoscale

computations with experiment is simply not possible. Thus, at length scales below
about 1 µm (see Section 3.17), computations in most cases can only be validated
using integrated properties such as flow rate or overall force measurements.
Verification and validation activities are much more complicated than is
described here (Roache, 1998; Oberkampf et al., 1998). However, for the prob-
lems discussed in this book, this section is a good start. Moreover, the concepts of
verification and validation can refer to any of the individual numerical procedures
such as integration, differentiation, or interpolation.

10.15 Chapter summary

In this chapter, the basic principles of numerical methods are described, from zero
finding to the numerical solution of partial differential equations. The chapter is
written to be self-contained and is heavily weighted with example Matlab scripts
that can be used for all the problems described in this book. The methods presented
in this chapter work; indeed, it is not the intent to present all the numerical
methods that can be used in a given problem. Thus, for example, in the section
on integration, the focus is on the basic trapezoidal rule and Simpson’s 1/3 rule.
The Matlab intrinsic functions that can be used are also identified. Many, though
not all, of the example problems and the problems at the end of this chapter are
drawn from topics within the purview of micro- and nanofluidics.
The chapter begins with a discussion of the concept of the Taylor series, which
is the most fundamental means of function approximation. After the introduction
of the Taylor series, several methods of finding zeros of nonlinear functions are
described. Next, the related concepts of interpolation and curve fitting are pre-
sented, and the limitations of using higher-order polynomials in the interpolations
are identified. In fact, it is shown that sometimes curve fitting is preferred when
the data set is noisy.
Numerical differentiation and integration are discussed next; the concept of
a central difference that is much more accurate than the forward and backward
differences is introduced. Dealing with derivative boundary conditions is also
described.
Leading into the discussion of the solution of ordinary and partial differential
boundary value problems, the solution of linear systems of equations is presented.
Both direct solution methods and iterative methods are presented, and it is noted
that iterative solution methods are preferable for large numbers of equations,
whereas direct methods are preferable for smaller sets of equations, although the
cutoff for the use of each is not distinct.
Solutions of initial value problems are discussed as well. Particularly relevant
for micro- and nanofluidics is the concept of symplectic integrators, methods
that preserve the condition that the Hamiltonian remain a constant. This property
was demonstrated for the classical two-body problem. Symplectic methods are
439 Essential Numerical Methods

used exclusively in molecular systems governed by a Hamiltonian in molecular

dynamics simulations discussed in Chapter 11.
The chapter concludes with the presentation of the methods of solving linear
and nonlinear ordinary and partial differential equations. In particular, it is shown
how the Poisson–Boltzman equation can be solved numerically for both constant
potential and constant surface charge density boundary conditions.
After completion of this chapter, the student should be able to write a Matlab
script to solve problems in the areas of numerical analysis described earlier. The
tools that Matlab provides are powerful and, in most cases, reasonably straight-
forward to implement. As can be seen, many of the line drawings presented in
this book are drawn using Matlab scripts.
With the knowledge of the material of this chapter, the student should be able to
r Find a zero of a numerical or analytical function such as finding the ζ potential
given the surface charge density
r Curve fit or interpolate a given set of data
r Find the volume flow rate through a channel or tube given a numerical function
r Numerically calculate the solution for the electrical potential or velocity for
an arbitrary mixture of ionic components of arbitrary valence
r Numerically calculate the solution for the mole fractions to verify the accuracy
of the Poisson–Boltzman distribution
r Numerically integrate first-order differential equations in the time domain to
find the end products of a biochemical reaction
r Numerically integrate stiff equations using both numerical and asymptotic
perturbation methods
r Begin to solve parabolic and elliptic partial differential equations numerically

This is but a small list of the possible applications; there are many more.

EXERCISES
10.1 Find the solution of
x 2 + 2 − ex
x=
3
accurate to 10−5 by reformulating the equation as a zero-finding problem.
Use the following methods:
a. From a sketch or several calculated values of the function in the
zero-finding formulation, determine the approximate location of the
solution.
b. Find the solution by hand calculations using Newton’s method. Show
x1 , x2 , f (x1 ), f  (x1 ) and the error at each step.
c. Find the solution with a computer using the secant method.
d. Write an m-file using Newton’s method, starting at the same point
used in the hand calculation.
440 Essentials of Micro- and Nanofluidics

10.2 The natural frequencies of vibration of a uniform beam clamped at one

end and free at the other are solutions of

cos(βb)cosh(βb) = −1

where β = ρω2 /E I , b = 1.5m, ω = frequency, E I = flexural rigidity,

and ρ is the density of the beam. Determine the three smallest values of
β that satisfy this equation to four significant figures. Note that β > 0
10.3 The solution of partial differential equations required to model phenomena
such as surface adsorption and heat transfer arising in microfluidic appli-
cations often requires, as an intermediate step, the solution of problems
such as

x tan x = p

where x needs to be solved for given a numerical value of p, where p > 0.

Multiple values of x will satisfy this equation.
a. Sketch the graphs of f (x) = cot(x) and g(x) = x and find the approx-
imate location of the roots of equation (3) when p = 1.
b. Calculate the three smallest nonzero roots of equation (3) using
an iterative root-finding procedure such as Newton’s method for
p = 1.
c. Use a built-in zero-finding function in Matlab, fzero, to find
the three smallest positive roots of xn of Equation (3) for p =
0.01, 0.1, 1, 10, 100, 1000. Plot and tabulate your results.
10.4 For a charged surface immersed in a multispecies electrolyte solution, the
surface ζ potential is related to the surface charge density σ by
   
RT σ 2  0 z i Fζ
= ci exp − 1
2e F 2 i
RT

where F is Faraday’s constant, R is the gas constant, and T is the tem-

perature, which can be taken to be 300 K in the calculation. The surface
charge density is −0.01 C/m2 , and the concentration of each species in
the solution (ci0 ) is listed in the table. Find the surface potential ζ .
Species ci0
Na+ 0.14214 M
Cl− 0.15834 M
Ca2+ 0.00810 M

10.5 Consider the data set of Table 10.9:

a. Draw by hand an interpolating curve by intuition.
b. Use Matlab to verify the results depicted in Figure 10.9.
c. Calculate y(1) using this polynomial.
d. Find an interpolating function using cubic splines.
e. Calculate y(1) using the cubic splines.
All the curves should appear on the same plot. What do you conclude?
441 Essential Numerical Methods

10.6 The set of data points in the table is the result of an experiment that
yields the velocity distribution in a fluid flow above a plane wall; the
data are presented in dimensionless form; u(y) is the velocity in the
boundary layer in the x direction. Using a Newtonian interpolation for-
mula, find the velocity at y = 1.7, y = 0.25, and y = 2.8. Use only
second-order accurate formulas, then curve ft the data with an appropriate
polynomial.

y 0.0 0.5 1.0 1.5 2.0 2.5 3.0

u(y) 0.000 0.402 0.682 0.852 0.942 0.981 0.995

10.7 For the data set of the previous problem, do the following:
a. Calculate du/dy at y = 2 using the forward formula, the back-
ward formula, and the central formula. Which formula is most accu-
rate?
b. Calculate d 2 u/dy 2 at y = 2.
c. Calculate the wall shear stress using μ = 1.0 × 10−3 Nsec/m2 .
Assume that y is in millimeters.
10.8 The following problem involves the integration schemes discussed in the
text.
a. Write a Matlab function routine to calculate a definite integral using
the trapezoidal rule.
b. Write a Matlab function routine to calculate a definite integral using
the Simpson 1/3 rule.
c. Test your routine by calculating the integral of the function f (x) =
e−x over the range (0, 6) by taking 11, 21, and 41 points. What do you
conclude? Compare your answers with the “exact” result.
d. Write an m-file routine to calculate the indefinite integral of the func-
tion in b over the range from (0,8). What do you notice about the
values of the indefinite integral past 6?
10.9 This problem involves the Simpson rules.
a. Write a Matlab function routine to calculate a definite integral using
the Simpson 1/3 rule.
b. Test your routine by calculating the integral of the function f (x) =
e−x over the range (0, 2) by taking 11, 21, and 41 points. What do you
2

conclude? Compare your answers with the “exact” result of 0.99532.

c. Write an m-file to calculate the integral using QUAD.
d. Write a Matlab function routine to calculate the indefinite integral of
the function in b over the range from (0,3). What do you notice about
the values of the indefinite integral past 2?
10.10 The total volume flow rate through the boundary layer is defined by

3
Q= u(y)dy
0
442 Essentials of Micro- and Nanofluidics

where u(y) is in units of m/sec and Q is in units of m2 /sec. Using the

data in the table of problem 10.6, do the following:
a. Write a Matlab script to calculate the volume flow rate Q using
Simpson’s rule. Compare with the value using the trapezoidal rule.
y
b. Calculate the indefinite integral Q(y) = 0 U (σ )dσ from the given
data points using the Trapezoidal rule. Assume Q(0) = 0. In the
solution, include a table of y, U (y) and Q(y), and give the grid
spacing. Compare with the Simpson’s rule result.
10.11 The data in the table of exercise 10.6, above are from Sadr et al., 2006
in a channel of height h = 10 µm and width W = 25 µm. Noting that
the velocity is constant beyond y = 210 nm across the entire channel,
assuming symmetry (i.e., that there is another layer with the same velocity
as in the table, calculate the volume flow rate through the entire channel.
What percentage of the flow rate is confined to the EDL, that is, y ≤
210 nm.
10.12 Measurements of some parameters of a physical system give a model of
the system represented by the following set of linear equations:

2.98x 1 − 1.05x 2 + 2.53x 3 = −1.61

−1.66x 1 − 1.08x 2 + 2.94x 3 = −6.76
4.35x 1 + 0.56x 2 − 1.78x 3 = 7.23

a. Find the solution to this system of equations by hand calculations

using Gaussian elimination.
b. Solve the system using Matlab.
c. Repeat b after changing the 0.56 coefficient to 0.50.
d. Calculate the determinant of the two coefficient matrices.
By noting what happens to the solution as we change the coefficient 0.56
to 0.50, what conclusions can be made about the model’s susceptibility
(or, in engineering terms, robustness) to model coefficient errors (and thus
measurement errors) in terms of finding the solution?
10.13 Solve the system of equations

x 1 + 2x2 = 2
x1 + 3x2 + 2x3 = 7
x2 + 4x3 = 15

by hand using the Thomas algorithm. Compare the result with the answer
in the text for the alternative elimination algorithm.
10.14 Consider the tridiagonal system of equations defined by aii = 4 + h for
i = 1 to N , ai+1,i = 1 for i = 1 to N − 1 and ai,i+1 = 1 for i = 1 to
N − 1. Take the right-hand side ai,N +1 = k 2 . All other entries in the
matrix are zero.
443 Essential Numerical Methods

a.Solve this system using both the Jacobi and Gauss–Seidel iterative
schemes. For simplicity, take N = 5, h = .05, and k = .1. For the
iterative techniques, start with a zero initial guess; the number of
iterations required for convergence should be printed out along with
the solution.
b. Write a program to solve this system of equations using the two tridi-
agonal elimination algorithms discussed in class. No two dimensional
arrays are to be used in the programs.
10.15 Solve the one-dimensional linearized Poisson equation for the potential
d 2φ
= zφ
dY 2
subject to

φ = ζ at Y = 0

where φs is a dimensionless ζ potential and at large distances,

φ → 0 as Y → ∞

numerically for φs = 1 and z = 1, 3. Compare with the analytical result

φ = ζ e−zY

10.16 Solve the one-dimensional Poisson equation

d 2φ
= z sinh zφ
dY 2
subject to

φ = ζ at Y = 0

where φs is a dimensionless ζ potential and at large distances,

φ → 0 as Y → ∞

numerically using the SOR method for φs = 1, 4. Compare with a solution

using the direct solver and with the analytical solution given by
 
zζ
zφ = 4 tanh−1 e−zY tanh
4
Compare the nonlinear result with the linear result. At what value of the
ζ potential does the Debye–Hückel approximation not hold? What is your
criterion?
10.17 Repeat the previous problem for the Neumann boundary condition
dφ
= −σ at Y = 0
dY
with σ = 1, 4. What is the dimensional ζ potential at the wall?
444 Essentials of Micro- and Nanofluidics

10.18 Solve the one-dimensional Poisson equation in cylindrical-polar coordi-

nates:
d 2φ 1 dφ
2
+ = z sinh zφ
dr r dr
subject to

φ = ζ at r = 1

dφ
= 0 at r = 0
dr
Note that the value of φ at r = 0 is unknown, and so an extra equation
needs to be added to the linearized system of equations.
10.19 Consider the nonlinear ordinary differential equation

d2 y dy
δ 2
− − y2 = 0
dx dx
subject to boundary conditions
dy
δ (0) = y(0) − 1
dx
y(1) = 1

Solve this equation for δ = 1. Determine the number of points

a.
required for three-digit accuracy to the right of the decimal point.
Use the Thomas algorithm.
b. Repeat for δ = 0.1, 0.01. What do you notice about the solution near
x = 0?
c. Plot y(x) for the solutions in (a), (b).
10.20 Using the preceding solutions to the Poisson equation for the potential
as a guide, solve the Poisson–Nernst–Planck system for the case of two
species of possibly different valences:
 
d dg dφ
+ zg g =0
dy dy dy
 
d df dφ
+zf f =0
dy dy dy

d 2φ
2 = −(z g g + z f f )
dy 2
with z f and z g taking arbitrary values. Assume that g is the cation and f
is the anion. The boundary condition on the potential is

φ = ζ at y = 0
φ → 0 far from the wall
445 Essential Numerical Methods

and the no-flux conditions at the wall y = 0 for the mole fractions g, f
are
dg dφ
+ zg g =0
dy dy
df dφ
+ zg f =0
dy dy
and far from the wall the mixture is electrically neutral and the concen-
tration of each species is 0.1M. Calculate the solution for z g = 2 and
z f = −1 and the dimensionless potential at the wall is ζ = 0.1. Is Gra-
hame’s solution recovered?
10.21 Extend the previous problem to the channel for the same boundary con-
ditions at y = 1 as at y = 0.
10.22 Using a method of your choice, solve Airy’s equation,
d2 y
− ty = 0
dt 2
subject to y(0) = 1 and ẏ(0) = 1 for 0 ≤ t ≤ 5.
10.23 The spring in a spring-mass damper system often behaves in a nonlinear
way. Suppose that a mass is subjected to force that is periodic in time;
then Newton’s law applied to the vertical deflection of the mass due to the
force may be reduced to an equation of the form
dy 2 1
2
+ y − y 3 = 3 sin(3ωt)
dt 6
This is called Duffing’s equation. In this equation, take ω = 0.893, and
the two initial conditions are y = y  = 0 at t = 0, where y  = dy/dt is
the vertical velocity of the mass.
a. Using the fourth-order Runge–Kutta or the Adams–Moulton method,
calculate the solution to t = 20 sec using a step size of h = t =
2 sec.
b. Repeat for h = 1, 0.5 sec to check the accuracy. Is this sufficient for
three-digit accuracy?
c. The motion is approximately periodic. Determine the period from a
plot of yvst for the appropriate time step. Plot y  = 0 on the vertical
axis and y on the horizontal axis for a single period. This is the phase
plane.
d. Compare your results with the Matlab solver ode45.
10.24 For Duffing’s equation, find the Hamiltonian and formulate the problem
as a set of two first-order equations in Hamiltonian form. Solve the equa-
tions using the explicit Euler equation, and compare the result with the
symplectic Euler result and your results from the previous problem.
10.25 For a square channel, the Poisson equation for the potential is given by
 
∂ 2φ 2∂ φ
2
 2
+A = −(z g g + z f f )
∂ y2 ∂z 2
446 Essentials of Micro- and Nanofluidics

where A is the aspect ratio of the channel. Solve this equation for a
symmetric electrolyte with

φ = ζ at y = 0, 1 and z = 0, A
for A = 1, 2. Assume that the concentration of each ionic species is
0.016M in an upstream reservoir, z g = −z f = 1, h = 20 nm and φs =
−120 mV. Use the solid wall boundary condition for the mole fractions.
10.26 Write a program to solve the linear diffusion equation
∂θ ∂ 2θ
= 2
∂t ∂y
with θ(0, t) = 1, θ(1, t) = 0 and at time t = 0 θ(y, 0) = 0.
a. Solve using the explicit scheme for r = 1/6 and for r = 1 out to a
time when the steady state is reached. What do you conclude? Be
sure to assess accuracy, and provide a list of values indicating that the
solution is three-figure accurate.
b. Plot the solution as a function of time at y = 0.25, 0.5, 0.75.
c. Repeat the first two parts using the Crank–Nicolson scheme.
 11 Molecular Simulations

11.1 Introduction

In this chapter, the focus is on presenting the fundamentals of molecular sim-

ulations, particularly molecular dynamics (MD) and its derivatives, particularly
nonequilibrium molecular dynamics (NEMD), or MD in a system that is exposed
to an external field. The number of molecular simulation tools has grown sub-
stantially over the last few years, and it would be impossible to mention all the
applications that have appeared in the literature. MD simulations have been used
in a variety of ways; four of the most important ways in the context of this text
that such calculations have been used are as follows:
r Calculating transport properties of new fluids and mixtures for which such
properties have not been measured or as a test of the accuracy of such experi-
mental measurements
r Predicting the equilibrium conformation of complex biomolecules and poly-
mers
r Verifying fluid dynamics boundary conditions (Koplik et al., 1989)
r Predicting the flow of pure fluids and complex mixtures in very small channels
and circular pores, where the continuum approximation may break down and
where direct experimental measurements are not available

Indeed, it is the last class of problems that is of particular interest in this chap-
ter. It is certainly to be noted that while restricted to very short length and
time scales, molecular simulations provide information that cannot be found any
other way.
As has been seen throughout this book, ionic and biomolecular transport
devices are now being used for drug development and delivery, single molecule
manipulation, detection and transport, and rapid molecular analysis. Many of
these processes are illustrated by natural ion channels a few angstroms in diameter
that serve as ion-selective nanoscale conduits in the body, which allow nutrients
in and waste products out. Electrokinetic effects, such as electro-osmosis and
electrophoresis, can be efficiently utilized to accomplish the desired transport
of fluid. As the dimensions of the channels in these new devices shrink from

447
448 Essentials of Micro- and Nanofluidics

the micro- to the nanoscale, molecular effects may be important for an accurate
description of the flow field. However, in this context, continuum theory must
still provide a quantitative framework for understanding microfluidics and its
interface on nanofluidics.
The complexity of MD simulations is significant. There are literally thousands
of potentials from which to choose to describe the molecular interactions between
individual atoms and molecules (Sadus, 1997). In this respect, the objective of
this chapter is limited. The purpose here is to introduce the reader to the basic
concepts of molecular simulations, focusing specifically on MD and how they
differ from continuum simulations in both philosophy and scope. The focus is on
presenting the equations for both pressure-driven and electro-osmotic flow. Appli-
cations to problems involving biomolecular transport are also discussed. There
are many assumptions that go into a molecular simulation, and the computational
requirements are substantially greater than for continuum calculations.
This book has considered liquid flows exclusively, and this chapter will focus
on a particular type of molecular simulation, MD (and NEMD). We discuss
only deterministic methods; thus we will not discuss those methods in which
some part of the system is not treated explicitly, such as direct simulation Monte
Carlo (DSMC) techniques (Bird, 1994), which is a statistical method used pri-
marily for gas flows, although it has recently also been used for liquid flows.
The same is true for those methods that treat the solvent implicitly such as
Brownian dynamics. The primary objective of this chapter is to introduce the
student and reader to the principles and processes used to develop a MD code
and to point the reader to the fundamental literature on the subject, including
existing codes, some of which are available free to the public. Indeed, in today’s
world, the researcher using molecular simulation techniques can access a num-
ber of rigorously validated software packages that will be discussed later in the
Chapter.
There are several monographs on this subject written at various levels. For
good reason, the majority of the books are at the more advanced level (Allen &
Tildesley, 1994; Sadus, 1997; Rapaport, 2004; Hinchcliffe, 2003; Frenkel and
Smit, 2002; Leach, 1996; Tuckerman, 2010). These books require some knowl-
edge of chemistry, the structure of molecules, intermolecular forces, bonding,
and other detailed aspects of computational chemistry. We hope that the essence
of MD and NEMD simulations can be presented clearly and effectively with a
minimum of such background material although a look back at Chapters 6, 7,
and 8 may help set the stage.
There are some lucid review articles relevant to the presentation in this chapter;
these articles present the concepts of MD in a way that makes them accessible to
the engineering community. These are the reviews by Sagui and Darden (1999), in
which the focus is on electrostatic effects, and the review by Allen (2004), which
focuses on presenting the fundamentals of MD simulations without a lot of the
detail. My belief is that this chapter will more closely follow the philosophies of
these reviews.
 449 Molecular Simulations

Generally, the MD process comprises several different steps. These steps

include the following:
r Choosing the potential characterizing the nature of the forces between atoms
r Specifying the initial and boundary conditions of the system of atoms
r Specifying how the force field is to be evaluated
r Choosing the numerical method to solve Newton’s laws of motion
r If there is an external field imposed on the system, introducing a mechanism
to dissipate the energy that is being introduced into the simulated system

The last step is called thermostating. We will discuss all these topics. I should
mention that in this chapter, it is not necessary to make the distinction between
atoms and molecules, and the terms atom, molecule, particle, and body will be
used interchangeably.
The chapter begins with an overview
of the nature of molecular simulations,
describing the various levels of molecu-
lar simulation from computing the elec-
tronic wave function of the electron cloud
that results in the potential energy between
atoms to the approximations inherent in
MD calculations. This section is followed
by a discussion of the potentials used,
Isaac Newton followed by a presentation of the details
involved in MD and NEMD calculations
that require the solution to Newton’s laws of motion for many molecules. The
chapter ends with a discussion of the types of MD and NEMD computational
packages that are available.
Caution: This chapter is not for the experienced molecular simulator! It is geared
to the late undergraduate or beginning engineering student who is a complete
novice when it comes to molecular simulations.

11.2 The molecular world

As we know, molecules, and indeed, all matter, is made up of atoms, which

are arranged in a specific way to form molecules. The individual atoms are
held together by chemical bonds and consists of a positively charged nucleus
consisting of protons and neutrons surrounded by a cloud of negatively charged
electrons. An atom as a whole must be electrically neutral.
The behavior of a collection of atoms and molecules is completely determined
by solving the time-dependent Schröedinger equation for the wave function  for
all the electrons and nuclei in the sample. The wave function contains probabilistic
information about the location of each particle as the system evolves in time. For
450 Essentials of Micro- and Nanofluidics

example, let us focus on a single particle (without specifying what the particle
actually is) and suppose that particle is constrained to move in the x direction
only. Then the wave function  = (x, t) and quantum mechanics tells us that
the probability that the particle will be found between x and x + d x at time t is
| (x, t) |2 d x. The unit of the wave function is (length)−d N /2 , where N is the
number of particles in the system and d is the dimensionality of the space (Styer,
1996). A good and lucid introduction to the principles of quantum mechanics is
the old and short monograph by Gillespie (1970).
At the most basic level, then, a molecular simulation would result in the deter-
mination of the wave function for all of the electrons and the nuclei in the system;
such a wave function is a function of the electron and nuclei coordinates from
which the probabilities that a range of velocities that will occur can be determined,
within the limitations imposed by quantum mechanics (Gillespie, 1970).
Of course, this is a daunting task, even for a few particles. Thus some simpli-
fications must be made. One approximation results from the observation that the
mass of an electron is much smaller than that of a proton, by over 1800 times.
Thus the dynamics of the protons in a system can be decoupled from the dynam-
ics of the electrons and this is called the Born–Oppenheimer approximation.
In the Born–Oppenheimer aproximation, the electronic wave function is first
obtained assuming the nucleus is infinitely massive and stationary. Thus, in this
approximation, the wave function provides the electronic energy as a function
of nucleus position. This electronic energy becomes the potential energy surface
on which nuclear dynamics takes place. Quantum mechanics could also be used
to determine nuclear dynamics, but in the overwhelming majority of molecular
simulations, the nuclei are assumed sufficiently massive to be described by the
classical equations of motion.
The Born–Oppenheimer approximation requires solving for the wave function
for many electrons. For example, suppose the wave function is desired for a
system of 4100 water molecules, a system that contains 1000 electrons. Then, for
the three-dimensional system, the wave function dimension is 3000 – a daunting
task indeed!
The process of solving the Schröedinger equation for the wave function is
called an ab initio calculation, Latin for “from first principles.” Molecular dyn-
mics calculations where the electrons are treated explicitly are called ab initio
molecular dynamics. Even though approximations are made to reduce the compu-
tational recourses required, these are still the most expensive MD methods. Thus,
often the Born-Oppenheimer electronic energy is assumed to be of a relatively
simple form based on physically based information. These are called empirical
potentials and often come from experimental data or by fitting ab initio results.
The next level of approximation is to perform all of the calculations using
some potential energy function that uses empirically determined coefficients;
this is what is most commonly termed the classical MD method. This method
is best illustrated by the Lennard–Jones potential, which will be discussed in
the next sections, when we present the details of MD calculations. It is this
 451 Molecular Simulations

method that is used most often today for molecular calculations. Despite that this
empirical method is the most computationally efficient, it still can take days to
weeks for a reasonably long computation. The empirical potential method has
often been termed molecular mechanics. A class of hybrid methods has also been
developed in which the electrons are treated explicitly, while empirical potentials
are used for the balance of the problem. These methods are known as quantum
mechanics–molecular mechanics (QM–MM) methods.

11.3 Ensembles

The state of a thermodynamic system is characterized by a given set of state vari-

ables or properties. These properties may include pressure, temperature, com-
position, density, surface tension, and free energy. Statistical thermodynamics
deals with the structure of matter on the molecular level, that is, on the micro-
scopic level. A microscopic state is a specific realization of the system where all
velocities and positions of the particles in the system are known. Conversely, a
macroscopic state corresponds to averages over all the microscopic states, result-
ing in macroscopic quantities such as density, pressure, and temperature. All
microscopic states that are a realization of a specific macroscopic state belong to
it and form what is known as an ensemble.
MD simulations are usually performed under well-specified thermodynamic
conditions, and thus specific ensembles can be identified. Although any set of
thermodynamic variables that specifies a state could be used to determine the
ensemble, only a few specific combinations are of practical relevance in MD and
NEMD simulations. The most common ensembles are the following:
r Microcanonical ensemble (NVE): the thermodynamic state of a system is
defined by its composition N (the number of particles and their type), its
volume V , and the total energy E; in this ensemble, the Hamiltonian is constant
r Canonical ensemble (NVT): the thermodynamic state of a system is defined
by its composition N , its volume V , and temperature T
r Isothermic-isothermal ensemble (NPT): the thermodynamic state of a system
is defined by its composition N , its pressure P, and temperature T

The NEMD calculation of the Poiseuille and electro-osmotic flows in a nanochan-

nel and discussed in Section 11.11 are examples of NVT ensembles. Armed with
this background, we can begin to discuss the details of a typical MD calculation,
beginning with the potentials.

11.4 The potentials

Although it seems to apply to any molecular level theory, the term molecular
dynamics is often used to designate the simulation of many-particle systems by
452 Essentials of Micro- and Nanofluidics

solving classical equations of motion. Typical MD simulations involve trajectories

of 102 –106 particles evolving for times of 102 –106 ps in simulations cells of 101 –
103 nm. Recall that 1 ps = 10−12 s.
To perform MD simulations, the important intermolecular forces must be
determined; thus the nature of the molecules, whether they are polar or nonpolar,
charged or uncharged, and their size must be known. An MD simulation begins
with the determination of an intermolecular potential  having units of energy
from which the force is determined by
F = −∇ (11.1)
In the nineteenth century, it was believed that the interaction potential between
any two bodies of mass m 1 and m 2 was given by (Israelachvili, 1992)
Cm 1 m 2
(r ) = − (11.2)
rn
where C is a constant. Thus the total potential energy associated with the effect
of all of the particles on the other is given by
L
tot = N 4πr 2 dr (11.3)
a

where N is the number of particles in the region between r and r + dr and L is

some length scale that satisfies L  a. Substituting for the potential, we find that

  n−3 
4πC N a
tot =− 1− (11.4)
(n − 3)a n−3 L
Thus we see that for n > 3, the total energy is independent of the distance L,
the size of the box. These are the short range forces. For n < 3, the total energy
depends on L, and these potentials are associated with a long-range force. Recall
in Coulomb’s law for the electric field that n = 1.
Typical empirical potentials used in a molecular simulation can be classified
into two groups: those for bonded interactions and those for nonbonded interac-
tions. A typical potential may be of the form
  
= kbi (bi − b0 ) + kθi (θi − θ0 ) + kϕ cos(nϕ + δ) + 1)
bonds angles torsions

   Ai j 12  Bi j 6   qi q j
+ − + (11.5)
i j<i
ri j ri j i j<i
ri j

where ri j = ri − r j is the distance between two particles. In this expression, the

first three terms are bonded potentials and the last two terms are nonbonded
potentials. The first term deals with the bond length, the second with the change
in bond angles, and the third with changes in the bond rotation. The fourth term in
the equation is the Lennard–Jones potential, describing short-range attractive and
repulsive forces, and the fourth is the long-range Coulomb interaction. In many
empirical MD simulations, the bonded interactions are neglected. This equation
 453 Molecular Simulations

could describe the potential for a set of atoms or for small molecules, or even for
small sugars or proteins.
An alternative to the repulsive force in the Lennard–Jones potential, the ri−12
j
term, is the Buckingham potential, defined by
 6
−βab ri j γab
 B = αab e − (11.6)
ri j

and describes the interaction between atoms of type a and b, and αab , βab and γab
are constants. Note that the Buckingham potential behaves very differently from
the repulsive Lennard–Jones term as ri j → 0.
In equation (11.5), it is the long-range electrostatic interactions that are the
most troublesome from a computational perspective because they decay only as
1/ri j . Indeed, the efficiency of MD calculations and the eventual results depend on
how these sums are calculated. This and other computational issues are discussed
later in the chapter. The constants in equation (11.5), the km , Ai j , and Bi j , are
often obtained from an ab initio calculation.

11.5 Using the Lennard–Jones potential

In this section, we will describe the basic computational methodology for an MD

simulation using the Lennard–Jones potential, while recognizing that there are
many other potentials describing intermolecular potentials in the literature.
The Lennard–Jones potential is perhaps the most popular description of the
nonbonded interaction between two particles. The cutoff form of the Lennard–
Jones potential is most commonly used in the form
 12  6
σ σ
 L J (ri j ) = 4W − for ri j < rc (11.7)
ri j ri j
= 0 for ri j ≥ rc (11.8)

where W is the well depth at dφ L J /dr = 0, and σ in the Lennard–Jones system

is considered to be a particle diameter. Here rc ∼ 2.5σ is a cutoff radius and
ri j = | ri − r j |, the distance between two particles. A sketch of this potential is
given in Figure 11.1. The first term in the potential is the repulsive force term,
and the second term represents the attractive part of the force field, the van der
Waals energy. A Lennard–Jones fluid is a collection of smooth, colliding balls.
Note here that the “balls” are not polar, and so strictly speaking, the Lennard–
Jones potential cannot describe polar substances like water; nevertheless, a
Lennard–Jones calculation can describe some generic properties of liquids, some
of which apply to water.
The Lennard–Jones potential has a discontinuity at r = rc that could adversely
affect the numerical results. To avoid this, a shifted Lennard–Jones potential is
 454 Essentials of Micro- and Nanofluidics

φ LJ

σ r
εW

Sketch of the Lennard–Jones potential.

Figure 11.1

defined as
 12  6
σ σ
 L J,shift (ri j ) = 4W − −  L J (rc ) for ri j < rc (11.9)
ri j ri j
= 0 for ri j ≥ rc (11.10)

In this case, there is not a discontinuity at r = rc , which means that the forces
will always be finite.
The force corresponding to the potential is given by the gradient of the poten-
tial, and for the shifted potential,

Fi j = −∇ L J (11.11)

and performing the differentiation, we find that

   14  8
48W σ 1 σ
Fi j = − ri j (11.12)
σ2 ri j 2 ri j

for ri j < rc , and zero otherwise. Note that the potential is continuous at ri j = rc
but the force is discontinuous there; however, experience has shown that the
discontinuity does not have a serious adverse affect on the computation if r c is
chosen to be sufficiently large. Then Newton’s law is given by

d 2 ri  N
mi = Fi = Fi j (11.13)
dt 2 j=1

where N is the number of particles (i.e., atoms or molecules) and m i is the mass
of each molecule. The sum in equation (11.13 ) excludes the value i = j.
455 Molecular Simulations

The problem can also be formulated in terms of the Hamiltonian, which is

defined as
N
pi2 
N
H= +  L J (ri j ) (11.14)
i=1
2m i< j

where pi is the momentum of particle i and the force associated with the Hamil-
tonian is given by

F = −∇ H = m a (11.15)

Note that we have described the interaction potential between the atoms or
molecules in the fluid; however, if the fluid is confined by walls, the interaction
potential between the walls and fluid also must be specified. If the walls are taken
to be smooth round spheres, then the interaction potential can also be of the
Lennard–Jones type, but with possibly different different values of the Lennard–
Jones parameters, σ and W , although they are taken to be the same in some cases
(Zhu et al., 2005).
Equation (11.13) is usually put in dimensionless form to reduce the number of
parameters required to vary to describe the system; chemists often refer to these
dimensionless variables as reduced units. We define dimensionless variables by
chosing σ , m, and W to be the units of length, mass, and energy, respectively.
Thus, for example, the dimensionless length is
r
r∗ = (11.16)
σ

The unit of time is mσ 2 /W , and substituting into equation (11.13), we have
(leaving out the asterisk, of course)
 N  
d 2ri −14 1 −8
= 48 ri j − ri j ri j + Fexti (11.17)
dt 2 j=1
2

where, again, the sum excludes the value i = j. Here Fexti is an external force
field, perhaps the electric field in electro-osmotic flow.
A popular option to obtain the interaction parameters for a Lennard–Jones
potential of different species is to use the Lorentz–Berthelot combining or mixing
rules (Allen & Tildesley, 1994). The combining rules are empirical expressions
that provide cross-interaction parameters (σ, W ) for molecules from the param-
eters of the pure atoms. For example, it is assumed that the collision diameter (σ )
for interaction between the species a and b is the arithmetic mean of the values
for the two pure species:
1
σab = (σaa + σbb ) (11.18)
2
The well depth (W ) is given as the geometric mean:

W ab = (W aa × W bb ) (11.19)
 456 Essentials of Micro- and Nanofluidics

11.6 Molecular models for water

In Chapter 7, we discussed in qualitative terms the structure of water. In this

section, we discuss the determination of the Lennard–Jones parameters of water
and the specific interactions associated with the water molecule.
The requirements for a water model are that it must be simple, be computa-
tionally tractable, and reproduce the basic properties of water. These objectives
are not easy to meet and sometimes conflict; thus there is no single water model
that is commonly used in MD or NEMD simulations. The computational cost is
a crucial factor, and most useful water models include only pairwise potentials
with no explicit three-body terms. One of the first MD simulations of water was
reported by Rahman and Stillinger (1971), and since then, many other water
models have been proposed.
The water monomer (a single water molecule) can be described as a rigid or as
a flexible entity, allowing all degrees of freedom for the OH bonds and HOH bond
angle, the latter being computationally demanding. In rigid models, the SHAKE
algorithm developed by Ryckaert et al. (1977) is generally used to constrain the
bond lengths.
Molecular simulations of liquids have been confined almost exclusively to the
implementation of pairwise rigid water models; three of the currently most used
pairwise water models are the following:
r TIP3P water model developed by Jorgensen et al. (1983)
r Simple Point Charge (SPC) model of Berendsen et al. (1981)
r Simple Point Charge Extended (SPC/E) model of Berendsen et al. (1987)

These models use three interaction sites and rigid bond lengths and a rigid
angle. The three models include two positive charges corresponding to the hydro-
gen partial charges and one negative charge corresponding to the oxygen partial
charge; moreover, the van der Waals interaction between two water molecules is
computed by using a Lennard–Jones potential acting only on the oxygen site. The
TIP3P and SPC models are similar, differing only in the values of the angles and
bond lengths of the water molecules and in the hydrogen charge; the Lennard–
Jones parameters are shown in Table 11.1.
SPC/E is an updated version of SPC including a correction in the potential
parameters to include polarization. Detailed studies have been performed to ana-
lyze and compare the accuracy and behavior of these classic water models (van der
Spoel et al., 1998; Mark & Nilsson, 2001). It should be noted that the assumption
of a rigid water molecule is an approximation, and thus results for some proper-
ties will be substantially different from the corresponding experimental values.
Interactions between water and another species, say, ions, must also undergo a
calibration process to obtain the Lennard–Jones parameters from experimental
results.
 457 Molecular Simulations

Table 11.1. Original interaction parameters for the SPC, SPC/E, and TIP3P water models
Parameters SPC SPC/E TIP3P
distance (O-H) 0.1 nm 0.1 nm 0.096 nm
◦ ◦
angle (H-O-H) 109.47 109.47 104.52◦
W 0.6502 kJ mol−1 0.6502 kJ mol−1 0.63639 kJ mol−1
σO 0.3166 nm 0.3166 nm 0.31506 nm
qO –0.82e –0.8472e –0.834e
qH 0.41e 0.4238e 0.417e

Because of the computational intensity of MD simulations, boundary con-

ditions are important. In simulations in nanochannels, the boundary conditions
can be applied directly to the simulation domain; typically, in a 6 nm channel,
there will be about 20 water molecules across the channel (Zhu et al., 2005), and
the boundary conditions can be applied directly. Conversely, periodic boundary
conditions must be used in the other two directions if the dimensions are larger
to reduce the computation time to a reasonable amount – these conditions are
discussed next.

11.7 Periodic boundary conditions

Because MD simulations are so time consuming, the region of computation must

necessarily be limited. Consider a system of, say, N = 1000 atoms or particles
in a three-dimensional box. Using walls to specify the boundaries of the system
inevitably introduces significant and unwanted perturbations into the calculations.
Typically, the influence of the walls in an MD simulation of Poiseuille flow will
extend approximately five molecular diameters from the wall. If the walls of a box
of particles are of length nσ , where σ is the molecular diameter (Figure 11.2), the

A simulation cell showing the need for periodic boundary conditions in molecular dynamics simulations. Figure
Figure 11.2
contributed by Professor Sherwin Singer.
 458 Essentials of Micro- and Nanofluidics

Sketch of the primary simulation simulation and its periodic images in two dimensions. Figure contributed by
Figure 11.3
Professor Sherwin Singer.

fraction of the number of particles affected by the walls is given approximately

by

6(5n 2 σ 3 ) 30
fraction = = + O(1/n 2 ) (11.20)
n σ
3 3 n
Thus, to make sure the walls do not adversely affect the numerical results, we
should take n  30. For n = 300, there will be a total of 27 × 106 particles
required, an impossible computation, and still 10 percent of the system would
be perturbed by walls. Using periodic boundary conditions offers a more prac-
tical way of reducing the computational load while maintaining computational
relevance.
The bulk behavior of the fluid using MD calculations can be accomplished by
assuming periodic boundary conditions in much the same way that Fourier series
are used in continuum periodic problems. The simulation cell and its periodic
images are shown in Figure 11.3. Using periodicity allows the simulation of a
system that is infinite and without explicit walls. Periodicity does introduce some
unphysical features, but these are much less severe than including walls explicitly.
The interpretation of the vector n is best understood by comparing this expression
with a typical Fourier series expansion in three dimensions.
So how many particles are enough? It is generally thought that for most
problems, a region bounded by two walls in a given direction should be about 20
particle diameters at a minimum. Consider a molecular dynamics simulation of
water. A single water molecule has a diameter of D = 0.3 nm and so 20 diameters
is h = 6 nm. In a three-dimensional box, length L = 6 nm, there are about 7200
water molecules so that 30/n = 30/19 > 1 since 72001/3 ∼ 19. Thus periodic
boundary conditions must be used. Using periodic boundary conditions reduces
the severity of unphysical artifacts caused by the 30/n fraction of particles being
affected by the wall. Remember, a Lennard–Jones fluid is not water, and typically,
the number density of Lennard–Jones particles is taken to be n = 0.8/σ 3 . A
simulation using Lennard–Jones balls for the solvent described in the work of
Zhu et al. (2005) contained 7757 solvent molecules, 31 cations, and 12 anions in
their model of electro-osmotic flow. Periodic boundary conditions were used in
two of the dimensions.
459 Molecular Simulations

In using periodic boundary conditions, we write the vector ri as

ri,n = ri + n L (11.21)

where n = (n x , n y , n z ) is a vector of integers with n = (0, 0, 0) being the primary

cell. This is shown in Figure 11.3. Suppose the primary cell is a cube of length L.
What is required for the calculation is the potential per cell; we divide the calcu-
lation into describing interactions within the same periodic cell and interactions
between particles in different cells.
The potential, or energy per cell due to all the interactions between the particles
within the same cell, the intracell interactions, is given by

1 
Ncell 
N
intra = ([ri + n L] − [r j + n L]) (11.22)
Ncell n i< j

where Ncell is the number of particles within the cell. Strictly speaking, this
equation is accurate only in the limit as Ncell → ∞, which we will omit in this
discussion. Canceling the n terms and then noting that the sum over n gives just
Ncell , we have


N
intra = (ri − r j ) (11.23)
i< j

That is, we need only use the primary cell (n = 0) to calculate the intracell
interactions.
The intercell component of the energy per cell, the interactions between parti-
cles in different cells, is given by

1 
Ncell 
Ncell 
N
inter = ([ri + n L] − [r j + m L]) (11.24)
2Ncell m n=m i, j

and we note that

([ri + n L] − [r j + m L] = ([ri − r j + (n − m)L) (11.25)

so that all cells with the same relative displacement give the same result. The 2
in the denominator of the intercell potential expression is to avoid counting each
i, j pair twice and is an alternative to writing the sum over i < j. Here m denotes
the vector over all of the image cells. The inner sum includes i = j, and the i < j
is noted to prevent double counting. Now let us look at the sums over n and m;
both these sums give an identical contribution, as can be shown by simply taking
Ncell = 2 and in one dimension, say, n = 0, 1, 2. Thus the intercell potential, can
be written as

1 cell 
N N
inter = (ri − r j + n L) (11.26)
2 n=0 i, j
 460 Essentials of Micro- and Nanofluidics

Combining the two potentials, the total potential is given by


N
1
N cell 
N
total = (ri − r j ) + (ri − r j + n L) (11.27)
i< j
2 n=0 i, j

and this is the potential that is used to describe the state of the system when
periodic boundary conditions are used. We can combine these two terms, the first
having n = 0 and
 N
Ncell 
total = (ri − r j + n L) (11.28)
n=0 i< j

where the prime indicates that i = j is omitted for n = 0.

One more approximation is made. If a cutoff method is used, the cutoff should
not be so large that the motion of a given particle is influenced by its own image.
Thus all terms in n, except the one for which | ri − r j + n L | is the smallest. This
approximation is called the minimum image convention (Leach, 1996), and the
potential is written as
N
1
total = Minn (ri − r j + n L) (11.29)
2 i, j

The minumum image convention is the primary means of truncating the series
in n.
Now that we have the potential, we need to know how to sum the series. The
potentials of interest contain both short- and long-range forces; the short-range
forces can be handled by the minimum image convention. The long-range elec-
trostatic force fields require a different approach, the Ewald summation, and this
is discussed next.

11.8 The Ewald sum

In a brilliant paper, Ewald (1921) showed how to sum the long-range electrostatic
force contribution to the potential in MD simulations. These interactions decay
slower than r −3 and thus cannot be truncated like the short-range interactions;
moreover, the series is only conditionally convergent, meaning that the end result
depends on the order of the summation.
There are several ways to derive the Ewald method; for example, we know that
1 erf (αr ) erfc(αr )
= + (11.30)
r r r
where α is some parameter to be defined and erf and erfc are the error and
complementary error functions, respectively. The first term is identified as the
long-range component of the electrostatic potential since erf(∞) = 1, and the
second term is recognized as the short-range component of the electrostatic
potential. The use of the error function in this identity can be motivated by the
following arguments, using the potential due to a set of N charges of charge qi .
461 Molecular Simulations

The long-range potential at a location r due to a point charge placed at ri

satisfies Poisson’s equation:
1
∇ 2 φi L = − qi δ(r − ri ) (11.31)
e
where δ is Dirac’s δ function defined by δ(0) = ∞, and δ = 0 otherwise. The δ
function is defined by its action on a function f as

f (x)δ(x)d x = f (0) (11.32)

If there are many charges, then the sum of all the charges over a volume leads to
a volume charge density ρe that replaces the individual chages qi . In this case,
the solution to the Poisson equation (11.31) in three dimensions is given by
1 ρe (r0 )
φi L (r ) = dV0 (11.33)
4πe V | r − r0 |
Note that this integral is singular at r = r0 , and so its numerical evaluation
cannot be done effectively. Knowing that ions and biomolecules are really not
point charges, a well-known approximation to the δ function is the Gaussian
distribution

G α = (2α 2 )3/2 e−(αr)

2
(11.34)
√
where the standard deviation of this distribution is 1/ 2α. The student can plot
the Gaussian distribution for various values of α to show that

lim G α (r ) = δ(r ) (11.35)

α→∞

Let us now write the Poisson equation in a three-dimensional, spherically

symmetric domain as
1 ∂2 Gα
(r α ) = − (11.36)
r ∂r 2 e
Integrating twice, it is evident that
1 r
r α = G α (r )dr (11.37)
2α 2 e 0

Using the definition of G α and noting that

2 r
e−t dt
2
erf (r ) = (11.38)
π 0

we see immediately that

1
αL = erf (αr ) (11.39)
e r
Thus the long-range contribution to the potential due to a charge qi having a
Gaussian distribution centered at r = ri is
1 qi
i L = erf(α | r − ri |) (11.40)
e | r − ri |
462 Essentials of Micro- and Nanofluidics

A similar argument for the short-range component due to a field of charges

given by
ρi S = qi (δ(r − ri ) − G α (r − ri )) (11.41)
can be used to show that the short-range potential induced by particle i is
1 qi
i S = erfc(α | r − ri + n L |) (11.42)
e | r − r i |
What Ewald really did was to write the overall charge distribution of particle i
due to a charge q j that has short- and long-range components as
ρei,T = ρei,L + ρei,S = qi G α ((| r − ri |) + qi (δ(r − ri ) − G α (r − ri )) (11.43)
Using these two results for the long-range and short-range potentials due to all
the charges, the total energy can be written, in terms of the simulation cell and
all of its periodic neighbors, as

1 
Ncell 
qi q j
L = erf (α | ri − r j + n L |) (11.44)
2e n i, j | ri − r j + n L |

and, for the short-range forces, as

1 
Ncell 
qi q j
S = erfc(α | ri − r j |) (11.45)
2e n i, j | ri − r j + n L |

where the term i = j for n = 0 is omitted from the sum. Often the self-energy
√
term is extracted from the long-range interaction; since lim erfc(r ) = 2/ π, this
r →0
quantity is given by

α  2
N
self =√ q (11.46)
π i=1 i
and the total potential can be written as
Tot =  L +  S − self (11.47)
and now we add in the n = 0 term for i = j in the long-range sum.
The short-range contribution can immediately be summed in real space; how-
ever, the long-range sum still needs work because erf (∞) = 1. So what do we do?
We will focus on the one-dimensional situation and then extend the result to three
dimensions. Because the system is periodic, we use the finite Fourier transform
√
of the long-range component of the charge density defined by (i = −1 in the
exponential of the Fourier transform!)
π
fˆ(k) = f (x)e−ikx d x (11.48)
−π

which results in
√ 
N
k2
ρ̂ L (k) = 2π α 2 q j e−ikx j − 4α2 (11.49)
j=1
 463 Molecular Simulations

Taking the Fourier transform of the Poisson equation, the long-range potential is
given by
√ 2 N
2
−ikx j − k 2
2π α e 4α

ˆ L (k) = qj (11.50)
e j=1
k2

Now write the inverse Fourier transform to obtain

√
2πα 2  
N N
1 −ikx j − k 22 ikx
 L (x) = qj e 4α e (11.51)
2πe j=1 k=1 k 2

If the system is 2π/L periodic, then writing

2πm
k= (11.52)
L
α2 N  m −2 2πim (x−x )− π 2 m 2

φ L (x) = qj e L j
Lα 2 (11.53)
(2π)5/2 e L j=1 m=0 L

Now, because of the exponential decay factor, the last term, the series is now
rapidly convergent, and the electrostatic potential that is long range in physical
space becomes short range in transform space, often called reciprocal space.
It is now a matter of writing the equation in three dimensions, including all
the sums, and the final result for the total long-range component of the energy
reads
1    m −2 2πi m (x −x )− π 2 m 2
Tot,L = qi q j e L i j Lα2 (11.54)
2π L 3 i< j m=0
L

with again, for i = j, m = 0 is omitted.

We can now move on to a discussion of the numerical issues faced in MD
simulations.

11.9 Numerical issues

11.9.1 Time integration
As discussed in Section 10.11.6, a particularly useful method for MD simulations
is to replace the second derivative by a discrete central difference approximation;
this approximation may be derived simply from Taylor series approximations,
and the approximation at time level k for any position x is defined by
d 2 xi xik+1 − 2xik + xik−1
= + O(t) (11.55)
dt 2 t 2
Such an approximation in the context of MD simulations is called the Verlet
(1967) algorithm, and it is symplectic, as noted in Chapter 10, preserving the
initial value of the Hamiltonian reasonably well. It is also noted that Newton’s
equations of motion are reversible in time, as is this Verlet algorithm.
464 Essentials of Micro- and Nanofluidics

It is well known that particle trajectories computed from satisfying Newton’s

law are extremely sensitive to the initial conditions. Thus chaos is present in
any system whose particle trajectories are calculated from Newton’s law. In a
potential flow calculation, the author and his colleagues found chaos in the
motion of a system of three potential vortices above a plane wall (Conlisk et al.,
1989).
As an experienced numerical analyst, I am reading this and saying; so what
good are the MD results? Well, I went searching for an explanation and found
it in Frenkel and Smit (2002). The objective of an MD simulation is not to
precisely predict the trajectories of every particle in the system. The aim of an
MD simulation is to predict the state of a system in a mean sense, the same
way that a continuum approach is used to predict the state of a system, perhaps
electro-osmotic flow in a channel. This fact is best stated by Frenkel and Smit
(2002) themselves, who point out that in a MD calculation,

We wish to predict the average behavior of a system that was prepared in an

initial state about which we know something (e.g. total energy) but by no means
everything. In this respect, MD simulations differ fundamentally from numerical
schemes for predicting the trajectory of satellites through space: in the latter case we
really wish to predict the true trajectory. We cannot afford to to launch an ensemble
of satellites and make statistical predictions about their destination. However in
MD simulations, statistical predictions are good enough.

This explanation does not rigorously prove that statistical results would be no
different if we could compute trajectories with infinite precision, but it can be
shown that the results for the particle trajectories are not in too much error; see
the discussion at the top of page 73 of Frenkel and Smit (2002).
The most time consuming step in an MD simulation is the calculation of the
forces, especially the long-range forces. To do this, at a given particle location,
we need to calculate the contribution of all of the other particles on the particle of
interest. Thus, with the minimum image convention, there will be N (N − 1)/2
pairs of interactions that must be evaluated at each time step so that the time
required for the force calculation scales as N 2 , where N is the number of particles.
For a relatively small value of N = 100, this means that there are 104 evaluations
of the force at each time step!

11.9.2 Truncation of interactions

Short-range interactions are truncated by assuming that the potential vanishes
beyond a cutoff radius ri j > rc . A typical value for the cutoff radius is rc =
2 − 3σ . Alternatively, often a tail contribution of the form is added:

∞
tail = ρ N φ(r )4πr 2 dr (11.56)
rc
 465 Molecular Simulations

Schematic depiction of the simulation cell for electro-osmotic flow in a channel showing the cations and
Figure 11.4
anions. The solvent molecules are not shown. The no-slip condition is applied at y = ± y1 = ywall − σ/2.

where ρ N is the average number density. For the Lennard–Jones potential, the
result for the tail contribution is
    3 
8 1 σ 9
σ
tail = πρ N σ 3 − (11.57)
3 3 rc rc

where often ρ N σ 3 = 1. Note that the tail contribution will diverge unless the
potential decays faster than r −3 , as discussed earlier. The Coulomb interaction,
the long-range electrostatic force, decays as r −1 and is evaluated using the Ewald
summation procedure discussed in the previous section.

11.9.3 Boundary conditions

Often the wall particle interactions with other wall particles and the fluid particles
are assumed to have a Lennard–Jones potential, with perhaps different values of
the well depth and σ . Because the particles do not have a unique measure of their
spatial extent (atoms are “fuzzy”), the question of the effective channel height is
an issue. Consider the situation in Figure 11.4. The stationary wall particles are
assumed to be located at y = ±ywall . An effective height of the channel is thus
defined by y1 = ywall − 1/2σ , while in the continuum regime, ywall = h. Thus,
in Figure 11.4, y1 = h − σ/2. As will be seen, this is an important consideration
in the NEMD examples discussed in Section 11.11.

11.10 Postprocessing

So now we have completed the MD (or NEMD) simulation and have thousands
of values for the positions and velocities of all the particles. We have run the
simulation long enough that the system has reached equilibrium in the sense that
the positions and velocities remain relatively constant with time. As we have seen,
the purpose of molecular simulations, and MD in particular, is not to precisely
 466 Essentials of Micro- and Nanofluidics

Example of “binning” for an MD simulation of electro-osmotic flow. The number of bins is dependent on the
Figure 11.5
problem and N b = 5 is an illustration only.

calculate the trajectories of all the particles but to predict the average behavior of
the system.
In a typical MD simulation, positions and velocities for the different species
in the system are saved as functions of time. These sets of stored values are used
to extract the static and dynamic properties of the system.
To compute properties from the stored particle trajectories, such as density,
concentration, and velocity profiles, a method for averaging the noisy data is
required. The many-body problem is chaotic in the sense that each particle
trajectory is strongly dependent on its original condition. Thus the trajectories
can only be calculated statistically or as averages of the trajectories of many
partiles. Thus the computational box is divided into bins; the binning resolution
depends on the property to be measured and the configuration of the simulated
system. Consider the calculation of the electro-osmotic flow velocity in a channel,
as depicted in Figure 11.5. We have a number of particles in each bin, and the
average velocity is defined by

1
u = ud A (11.58)
A
Thus the average in each bin is reduced to a sum (equivalent to using the trape-
zoidal rule) as

1 
Nbi
ubi = uj (11.59)
Nbi j=1

where Nbi is the number of particles in each bin. The local velocity takes the value
ū bi at the center of each bin and becomes a “velocity at a point” in continuum
language. In fact, the velocity ū bi is actually the average of the center of mass
velocities of the molecules in each bin.
Of course, the spatial averaging process takes place only after a suitable number
of time steps, and in fact, in a similar way, the results are averaged in temporal
bins as well. For example, the time average of the velocity of a particle i after a
 467 Molecular Simulations

time t may be

1 t
û i = u i (τ )dτ (11.60)
t 0

which can also be turned into a sum. Of course, writing this average after a long
time t assumes that the results are independent of the initial condition. Since the
many-body problem is chaotic, this will certainly not be the case, and really we
should make a number of runs, averaging over a set of initial conditions. Doing
this results in an ensemble average. It is seen that the postprocessing procedure
is extremely complicated, and certainly this presentation is not sufficient to just
go out and postprocess numerical data. In fact, an important part of the analysis
of the results of MD simulations is to use the methods of statistical mechanics
(Frenkel and Smit, 2002), the discussion of which is far beyond the scope of this
presentation.
So why does statistical mechanics get involved? An MD simulation is based
on the numerical solution of a many-body problem. As studied by Conlisk et al.
(1989), chaos is present in interacting systems that follow Newton’s laws. There-
fore, in typical MD systems, chaos is expected to be present, although when an
equilibrium state is established, the mean properties averaged over a sufficient
number of trajectories seem independent of the initial configuration of the system.
See the discussion on the objectives of an MD simulation and the quote from
Frenkel and Smit (2002) in Section 11.9. Further analysis of this issue can be
found in the literature related to chaos in equlibrium and nonequilibrium statisti-
cal mechanics (Castiglione et al., 2008; Rice, 2000), and there is more in Frenkel
and Smit (2002).

11.11 Nonequilibrium molecular dynamics

11.11.1 Introduction
NEMD refers to the situation where molecules or particles are subjected to an
external force field. The use of NEMD simulations to describe the trajectories of
molecules in multicomponent mixtures characteristic of problems in biology and
chemistry in a bulk fluid flow is in its relative infancy because of the relatively
large computation time required. Yet most molecular dynamics simulations are
of the NEMD type.
It is important to note that when an external force is applied to a system, a
simulation cell will rapidly heat up if heat is not removed. The term thermostat is
given to the procedure used to remove the excess energy to keep the temperature
of the cell constant. Various means have been used to do this such as coupling
with a heat bath (Berendsen et al., 1981) and the Nosé–Hoover and Isokinetic
thermostats (Heyes, 1998; Rapaport, 2004). In the next section, we discuss the
MD simulation of Poiseuille flow.
 468 Essentials of Micro- and Nanofluidics

2.5
u 0.1

ρ 1.5 u

0.05
1

0.5
ρ

0 0
-10 -5 0 5 10
y

Steady state dimensionless density ρ and velocity distributions u established under conditions of planar
Figure 11.6
Poiseuille flow for a pure Lennard–Jones fluid. Flow was induced by a constant force, f = 0.005, applied to
all particles.

11.11.2 Poiseuille flow

Building on the work of Travis and Gubbins (2000), Zhu et al. (2005) has used
NEMD to simulate Poiseuille flow in a small channel; the flow is assumed to be
in the x direction, as in Figure 11.6. Periodic boundary conditions are used in the
x direction, the direction of flow, and the z direction, with dimensionless lengths
L x = 21.77 and L z = 21.83, respectively.
It is well known that the Navier–Stokes equations are based on the units of
force per unit volume. Indeed, the governing equation for fully developed flow in
a channel, repeated from Chapter 4, is given by

d 2u 1 dp
= (11.61)
dy 2 μ dx

for which the solution is given by

1 dp 2
u(y) = (y − h 2 ) (11.62)
2μ d x

where the zero y location is at the center of the channel of height h.

However, in molecular dynamics simulations, it is the force that is specified,
and so to compare with the NEMD result, the continuum solution must be mod-
ified. The pressure gradient has units of N /m 3 or force per volume. To put the
continuum result in a molecular setting, we write the governing equation in the
form

d 2u f x ρN
=− (11.63)
dy 2 μ

where ρ N is the number density of the particles and f x is the external imposed
force on the system. The continuum result written in terms of these molecular
469 Molecular Simulations

variables is given by
fx ρN  2
u(y) = y1 − y 2 (11.64)
2μ
where μ is the viscosity, and y1 = h − σ/2 (as discussed previously). The pres-
sure gradient is now recognized as
dp
= fx ρN (11.65)
dx
The dimensional distribution of velocity can be made dimensionless by defin-
ing a dimensionless number density, viscosity, and force by ρ N∗ = ρ N N A σ 3 ,
√
μ∗ = μσ 2 / mW , and f x∗ = f x σ/W . The dimensionless velocity should then
√
be scaled by W /m and given in dimensionless variables by (of course, after
dropping the asterisk)
fx ρN  2
u(y) = y1 − y 2 (11.66)
2μ
and the lengths are scaled on σ and not h.
The flow is induced by a constant force, f x = 0.005, applied to all particles.
The dimensionless viscosity is given the value of μ = 2.13 (Zhu et al., 2005), and
the dimensionless value for the wall height is h = ywall = 10.76, as in Table 11.2.
The velocity profile obtained from simulations is compared with the standard
result for an incompressible fluid, with overall number density ρ = 0.8 and no-
slip boundary conditions enforced one particle radius from each wall in the
y direction: y1 = 10.26, and y1 is now dimensionless.
The results of the simulations are shown in Figure 11.6. Note that the continuum
expression for the Poiseuille flow in molecular variables slightly underpredicts
the NEMD result, but the difference is not large. Also note the molecular layering
near the walls, where the dimensionless number density fluctuates, and thus, from
the point of view of NEMD, the flow is not incompressible.
We now address a more difficult problem: electro-osmotic flow.

11.11.3 Electro-osmotic flow

In this section, we will assume that the external field is due to an electric field in
the form of the Coulombic interaction (Zhu et al., 2005)
zi z j
C = ζ (11.67)
ri − r j
where the variables are in dimensionless form and
e2
ζ = (11.68)
e σ
is the dimensionless interaction energy, e is the electron charge, and e is the
permittivity of the medium.
Here, as in the Poiseuille flow discussed earlier, the ions and wall particles have
the same diameter as the solvent, and all particles have the same mass. Strong
 470 Essentials of Micro- and Nanofluidics

Table 11.2. Parameters used in the NEMD simulation of electro-osmotic flow by Zhu et al. (2005)
Run Total Cations Anions Solvent Wall ywall ζ IS M (cation) M (anion)
A 8638 31 12 7757 836 10.76 5 7 0.2206 0.08538
B 8638 31 12 7757 836 10.76 1 1.4 0.2206 0.08538

ion–solvent attractions are needed to dissolve ions in solution. This means that the
dimensionless ion–solvent interaction parameter must be large to keep the ions in
solution because the Lennard–Jones spheres modeling the solvent are nonpolar
(in contrast to water). The value of the dimensionless ion–solvent well depth, IS ,
depends on the Coulomb interaction strength. Explicit values of IS and other
parameters are given in Table 11.2. Similar to the Poiseuille flow, the continuum
expression for electro-osmotic flow must be put in molecular variables; the details
are given by Zhu et al. (2005).
The molarities in Table 11.2 are based on a particle diameter of σ = 2.8818 Å,
which would make the total solution molarity equal to that of water, 55.5 M. The
concentrations of ions were chosen to be those estimated for physiological pH salt
solutions. Assuming the same value of σ , the wall charge density corresponds to
2.4 × 105 electron charges per µm2 .
The Gaussian isokinetic thermostat (Evans & Morriss, 1990) is used to main-
tain the system temperature constant. A dimensionless time step of t = 0.01 is
√
used, and the dimensional time is scaled on τ = m L J /σ . As with the Poiseuille
flow, only the y and z components of the velocity are thermostatted because there
is a nonzero streaming velocity in the x direction.
Velocity profiles for the solvent are exhibited in Figure 11.7a for ζ = 5 and
Figure 11.7b for ζ = 1. Note that for the smaller value of the interaction param-
eter, there is little ion exclusion from the wall, and there is no need to correct
for excluded volume effects. For ζ = 5, however, a correction of one ion radius
gives excellent agreement with the continuum Poisson-Boltzmann (PB) result.

0.05
0.06 0.045
0.05 0.04
0.035 Δy = 2
0.04 0.03 Δy = 1
μe μe 0.025
0.03
0.02
0.02 0.015 Δy = 0
0.01
0.01 (unmodified PB)
0.005
0 0
–10 –5 0 5 10 –10 –5 0 5 10
y y
(a) (b)

Dimensionless electro-osmotic mobility μe = u/E x across the channel. The dimensionless electric field
Figure 11.7
E x = 2. (a) Interaction parameter ζ = 5 and (b) interaction parameter ζ = 1. For ζ = 5, ion exclusion from

the wall is significant. The characteristic velocity is U 0 = W /m L J where  is the energy scale in the
LJ-potential and the electric field is scaled on E 0 = W /σ e.
 471 Molecular Simulations

Table 11.3. MD/NEMD packages and their availability on the Web

Package Web site About
GAUSSIAN http://www.gaussian.com ab initio quantum chenistry
LAAMPS http://laamps.sandia.gov MD/NEMD (free)
GROMACS http://www.gromacs.org MD/NEMD (free)
AMBER http://ambermd.org MD force fields (free)
CHARMM http://www.charmm.org molecular mechanics
NAMD http://www.ks.uiuc.edu/Research/namd/ MD/NEMD (free)

GAUSSIAN, an ab initio package, is included because the users of other packages may use force fields derived from
GAUSSIAN.

The anion and cation concentrations for ζ = 5 show significant layering near
the surface, with significant density fluctuations; however, ions are completely
absent from the wall layer. This is due to the strong ion–solvent interaction that
prevents the ion from shedding its solvated solvent molecules to any great extent.
Conversely, for ζ = 1, the ions can populate the inner wall layer readily. For a
discussion of other simulations that suggest that the Newtonian fluid assump-
tion may break down near the interface (Qiao & Aluru, 2003b; Freund, 2002),
see Zhu et al. (2005). Such behavior is not seen in the present work if the indicated
correction of one ion radius is used in specifying the actual width of the channel.

11.12 Molecular dynamics packages

11.12.1 Introduction
The focus in this discussion is on empirical MD simulators as opposed to ab initio
packages that calculate atomic interactions based on quantum chemistry. Some
of the most popular classical MD software packages are Groningen Machine for
Chemical Simulations (GROMACS). Large-Scale Atomic/Molecular Massively
Parallel Simulator (LAMMPS), Chemistry at Harvard Molecular Mechanics
(CHARMM), and Assisted Model Building with Energy Refinement (AMBER).
All these packages are open source and readily available on the Web. The Web
site locations are given in Table 11.3.
GROMACS or LAMMPS appear to be the tools of choice among researchers
working on nanoscale problems in fluid dynamics. For example, Qiao and Aluru
(2003a) have explored electrokinetic problems using GROMACS, and Priezjev
et al. (2005) have demonstrated the use of LAMMPS to study slip behavior in
wetted films.

11.12.2 What MD/NEMD simulators do

What kinds of information do these packages contain?
 472 Essentials of Micro- and Nanofluidics

1. Force fields: These packages contain a number of interaction potentials.

Usually in nanofluidics, we are interested in modeling of the electrical dou-
ble layer, solvation of ions, surface tension, slip, contact lines, interaction of
liquids, and polyelectrolyte with carbon nanotubes. For these applications,
the Lennard–Jones potential and electrostatic potentials are often sufficient,
but more sophisticated models of the interaction between water molecules
may be required (e.g., SPC/E; Berendsen et al., 1981). Both GROMACS
and LAMMPS have a large number and variety of force fields from which
to choose.
2. These packages also allow the user to choose the integrator, include con-
straints (e.g., constant temperature in NEMD), and provide boundary con-
ditions. To perform MD simulations in ensembles where energy is not
fixed and/or in a domain of finite extent in a certain direction, a variety
of constraints, such as thermostats, barostats, and periodic boundaries, are
needed. Usually, they are built into the MD packages.
3. These codes have been optimized so that the force fields are calculated effi-
ciently. Portability across hardware and platforms is also a consideration.
4. Ease of use and legacy: A close connection between the problem of inter-
est and the sample problems that come with the installation of a pack-
age ensures shorter lead times in obtaining meaningful results. LAMMPS
comes with examples for atomistic simulation of Couette and Poiseuille
flow.

A factor that enhances the versatility of newer packages like NAMD and
LAMMPS (both released first in 2002) is that they can use the force fields and
input configurations of CHARMM and AMBER. The ease of preprocessing,
postprocessing, and visualization should also be considered in choosing a pack-
age, for example, the trajectories and initial coordinates of atoms in GROMACS
can be visualized without a need for extra software; whereas LAMMPS needs
software like VMD or RasMol for visualization.

11.13 Summary

In this chapter, the reader has been introduced to molecular simulations, specif-
ically what is termed molecular dynamics. There are a number of applications,
and relative to the subject of this text, these applications include the following:
r Calculating transport properties of new fluids and mixtures for which such
properties have not been measured or as test of the accuracy of such measure-
ments
r Predicting the equilibrium conformation of complex biomolecules and
polymers
r Verifying the no-slip condition (Koplik et al., 1989)
473 Molecular Simulations

r Predicting the flow of pure fluids and complex mixtures in very small channels
and circular pores, where the continuum theory may break down

A number of steps are involved in the building of a MD code, and these steps
have been discussed in this chapter. These steps include the following:

r Choosing the potential characterizing the nature of the forces between atoms
r Specifying the initial and boundary conditions of the system of atoms
r Specifying how the force field is to be evaluated
r Choosing the numerical method to solve Newton’s laws of motion
r If there is an external field imposed on the system, rescaling the temperature
to insure that it is constant

The last step is called thermostating.

In today’s world, it is typically not necessary for a researcher to build MD code
from scratch. We have discussed several packages that are available to the public,
three of them free of charge. That is not to say that these packages are easy to
use, however!

EXERCISES
One note: The goal of the exercises following is to help the student understand
the fundamentals of why the MD results are processed the way they are. The
reason is that all MD simulations exhibit chaotic behavior at some level. This
is why the methods of statistical mechanics are often used in postprocessing the
MD data.

11.1 Return to the two-body problem discussed in Section 10.11.6 and add a
third body. Write the governing equations in dimensionless form for the
motion of the three particles. Calculate and analyze the trajectories. What
do you find? Be sure to reduce the time step to make sure the trajectories
are accurate. Arrange the particles in a triangle initially. Perturb the initial
condition of one of the particles by 1 percent. Are the trajectories any
different?
11.2 Explicitly write out the governing equations for the motion of the N -
body problem for N = 3 from equation (11.17). Assume a Lennard–Jones
potential.
11.3 Solve this system in dimensionless form numerically using the Verlet algo-
rithm and compare with the same result using Runge–Kutta. Be sure to
verify the results by reducing the time step. Calculate the Hamiltonian.
What do you find? Assume the particles are in a triangle initially. Plot the
trajectories as a function of time. Perturb the initial condition of one of
the particles by 1 percent of the distance from one other particle. Do the
trajectories change?
474 Essentials of Micro- and Nanofluidics

11.4 This is an open-ended exercise. Write a report on chaos in fluid mechanics.

A good initial source is the paper by Conlisk et al. (1989). Chaos theory
has also been used as a model for turbulent flow.
11.5 Write out the governing equations as in exercise 11.1, but include the
electrostatic term.
11.6 This is an open-ended exercise. Think about the simulations described in
Section 11.11. Using the results of your chaos report from exercise 11.4, is
the system in exercise 11.1 chaotic? If so, how must you view the results?
11.7 This exercise requires some of the research from your chaos report. Now
return to exercise 11.1 and perform a time series analysis of the trajectories.
What measures are used to analyze the trajectories?
11.8 Write a report on statistical mechanics and its relationship to molecular
dynamics simulations. What is a Green–Kubo relation, and how is it used?
 12 Applications

12.1 Introduction

As has been mentioned in Chapter 2, applications of micro- and nanofluidic

analyses include drug delivery and its control, DNA manipulation and transport,
protein separations, rapid molecular analysis, renal assist devices, biochemical
sensing, and cancer treatment. In the present chapter, several of these applications
are described, and modeling tools are developed using the ideas developed in the
preceding chapters.
Generally, many of the biomedical devices being tested for the preceding
aplications are essentially synthetic micro- or nanopore membranes consisting
of an array of channels, much as in Figure 1.1. For example, these devices can
be used to deliver insulin to a diabetes patient, as needed, in a highly controlled
manner. They have also been considered for use as a renal assist device (RAD)
whose purpose is to replace some of the functions of the kidney; this device is
discussed later. DNA has been shown to be able to navigate through approximately
cylindrical nanopores, and this problem is also discussed in this chapter; the in
vivo transport of DNA through a nanopore has been demonstrated by Bayley and
Cremer (2001).
More general applications of nanotechnology and additional information on
some of the applications described may be found in the short monograph edited
by Malsch (2005) and in the monograph by Ciofalo et al. (1999), respectively. The
latter book lists centers and groups that are active in either nanoscale physiological
systems or other biological flow problems as of its publication in 1999. Since
then, that list has expanded considerably (see Bohn (2009) and the references
therein).
Many biomedical applications are described in the four-volume set titled
BioMEMS and Biomedical Nanotechnology, edited by Ferrari (2006). The titles
of each of the four volumes include “Biological and Biomedical Nanotechnol-
ogy” (Lee & Lee, 2006), “Micro/Nano Technology for Genomics and Proteomics”
(Ozkan & Heller, 2006), “Therapeutic Micro/Nano Technology” (Desai & Bhatia,
2006), and “Biomolecular Sensing, Processing and Analysis” (Bashir & Wereley,
2006).

475
 476 Essentials of Micro- and Nanofluidics

The purpose of this chapter is to investigate several application areas and

illustrate the use of the modeling techniques described in the previous chapters.
The main application areas discussed here are
r DNA transport
r Transport processes in a RAD
r Biochemical sensing

These three areas were chosen because of the crucial role that micro- and nano-
fluidics plays in the operation of the devices in each area. These areas are certainly
not all inclusive of the myriad applications, and the reader is encouraged to consult
the references listed.
In particular, there are four areas that are not covered in this chapter, non-
biomedical problems that are of intense interest:
r Electrokinetic remediation
r Batteries
r Fuel cells
r Water desalination and purification

Each of these application areas includes electrochemistry as the primary process

driver, and the reader of this text will clearly understand the processes that occur
in these subject areas. The reader will be introduced to these areas as exercises
at the end of this chapter.

12.2 DNA transport

DNA transport through tubes is used for DNA sequencing, the determination
of the order of the individual nucleotide bases that make up the DNA, DNA
repair, injection of the repaired DNA into cells (electroporation), and sensing.
In the sensing area, Chang et al. (2004) and Fan et al. (2005) have performed
experiments in an effort to understand the dependence of the electrical current
through a nanochannel filled with a DNA strand on various flow parameters such
as electrolyte concentration, flow geometry, and applied electric fields. These
experimental studies are closely related to (and inspired by) efforts to develop a
novel DNA sequencing mechanism (Bayley & Cremer, 2001; Kasianowicz et al.,
1996; Deamer & Akeson, 2000; Meller et al., 2001). It should be noted that the
model described here can easily be generalized to other long-chain, cylindrical
polymers.
Both natural and synthetic nanopores have been used to study the transport
process, called translocation experimentally. Kasianowicz et al. (1996), Bayley
and Cremer (2001), and Meller et al. (2001) experimentally investigated the
translocation of single-strand DNA (ssDNA) through natural α hemolysin pores
to characterize the DNA translocation properties by analyzing the ionic current
signals.
 477 Applications

b−a
Lp
i

Up

Side view of the geometry of the circular tube and circular DNA strand. Here the DNA strand of radius a is
Figure 12.1
moving to the right; the outer radius is denoted by b. The notation L pi denotes the length of the particle inside
the tube and is equal to the tube length. The direction of motion of the particle depends on the surface charge
densities of the particle and the wall.

Researchers have also investigated DNA transport through synthetic nanopores

that are claimed to be more stable, offer more flexible choices in pore shape and
surface properties, and are easier to integrate into micro- and nanofluidic devices.
Fan et al. (2005), Li et al. (2003), and Storm et al. (2005) investigated the translo-
cation of double-stranded DNA (dsDNA) through a synthetic nanopore exper-
imentally as a function of applied votlage, electrolyte concentration, and other
physical parameters. In particular, Li et al. (2003) demonstrated the capability
of observing individual molecules of dsDNA translocation, and the experimen-
tal translocation rate for dsDNA through synthetic nanopores is ∼0.01 m/s for
3 kbps (kbps = 1000 base pairs) and 10 kbps dsDNAs. Chen et al. (2004) showed
that the DNA translocation velocity varies linearly with the applied voltage drop,
increasing as the voltage drop is increased. The DNA velocity was demonstrated
to be independent of the DNA length based on the experimental data for 3 kbps,
10 kbps, and 48.5 kbps dsDNA. Conversely, the experimental data of Storm et al.
(2005) on 11.5 kbps and 48.5 kbps dsDNA show that longer DNA strands move
more slowly through the nanopore than shorter ones, in conflict with Li et al.
(2003). This point is thus a matter for further research.
In the next few sections, we describe a continuum model for translocation of
DNA through a tube idealized as the moving inner cylinder in the electro-osmotic
flow through an annulus discussed in Section 9.2.8.

12.2.1 How does DNA move?

Let us describe in general terms how DNA may move through a tube. The simplest
model of a dsDNA molecule is a solid cylinder; it is assumed for simplicity that
the dsDNA lies within a cylindrical pore. The radius of the dsDNA is about
a = 2 − 3 nm, and the synthetic nanopore is usually about b ∼ 10 nm in radius.
DNA, whether single or double stranded, is negatively charged, and so a cathode
placed upstream and an anode placed downstream will drive the dsDNA molecule
through the tube. Thus the dsDNA is electrophoretically driven through the tube
in concert with an electro-osmotic flow. The geometry of the problem is depicted
in Figure 12.1.
 478 Essentials of Micro- and Nanofluidics

b−a It is assumed that the molecule and the

tube are charged with possibly differing
Lp
charge densities and that the gap between
the tube and the molecule is big enough so
that the fluid flow between the tubes can
(a) be described by a continuum analysis (Zhu
b−a et al., 2005). In addition, it is assumed that
Lp
i electrostatic and other local conditions do
not prevent the biomolecule from entering
the tube; this is a separate problem and
involves possible deformation mechanics,
Up
a feature that can be modeled by Brownian
(b) dynamics simulations (Hur et al., 2000).
b−a
Lp
The length of the particle1 that is inside
i
the tube is denoted by L pi (t), and the var-
ious parameter ranges are shown in Fig-
Up
ure 12.2.2 As the particle enters the chan-
(c) nel, only a portion of the tube is occu-
b−a pied. During this time, a transient pressure
Lp
i
gradient is induced because the channel is
only partially filled. As the leading edge
of the DNA approaches and reaches the
exit of the tube, there is another short tran-
Up
sient period in which the pressure gradient
(d)
relaxes to zero (provided that the trailing
Schematic of a cylindrical particle in a cylin- edge of the biomolecule has not reached
Figure 12.2
drical tube. In all cases, the gap between the the entrance), resulting in a sudden but
particle and the wall of the channel is assumed
not large change of electrophoretic veloc-
much smaller than the length of the particle. (a)
Length of the particle is less than the length of ity. During the time when the biomolecule
the tube (Zhao & Bau, 2007). (b) Length of the fully occupies the channel, the flow is fully
particle is much longer than the length of the developed and steady. As the trailing edge
tube; partial entry. (c) Length of the particle is
of the biomolecule approaches and enters
much longer than the length of the tube; par-
ticle fills the tube. (d) Length of the particle is the tube, another transient period results
much longer than the length of the tube; partial in the development of a transient-induced
exit. Not all of the particle is shown in (b)-(d). pressure gradient, and the electrophoretic
velocity suddenly changes again. The cur-
rent density changes when the DNA occupies the pore, and thus transient current
density changes can be expected. These transient periods – their duration, in
particular – can be used to determine the length of the DNA strand of interest.

1
The term particle will be used extensively to denote the DNA strand. Indeed, the model described
here is generic and will apply to any long-chain charged polymer, natural or synthetic.
2
The parameter ranges described here have been deduced based on unpublished work by the author
and former graduate student Prashanth Ramesh.
479 Applications

The direction of motion of the DNA is determined by the relative magnitude

of the surface charge densities of the particle and the tube. The direction of
the EOF is always from left to right for a negatively charged tube. For the tube
surface charge density σ 1 > σ 0 , the particle will move to the right and in the
same direction as the EOF. Conversely, for σ 1 < σ 0 , the particle will move to
the left and against the EOF. This result will be demonstrated later and is an
important consideration in designing rapid molecular analysis tool kits.
Though the transient periods and their length are important in DNA analy-
sis systems, these transient periods are short in duration, and the predominant
translocation velocity may be determined using a steady flow analysis. This is
discussed in the next section.

12.2.2 Mathematical model

The governing equations for the motion of DNA are the same as in the case of
electro-osmotic flow in an annulus, which is considered in Section 9.2.8, except
for the partial entry and partial exit cases and the fact that the particle (i.e., DNA)
velocity is unknown. Assuming that the transient periods are very short on the
time scale of the electro-osmotic flow and that the flow is fully developed and
steady, the governing equation for the axial electro-osmotic velocity is given in
dimensionless form by
 
∂ w ∂w 
2 ∂p
2
1
2 + = − − β zi X i (12.1)
∂r 2 r ∂r ∂z i

where the pressure is scaled on the viscosity. Similarly, the equations for the mole
fractions and the potential are
 
∂2 Xi 1 ∂ Xi ∂ 2φ 1 ∂φ ∂ X i ∂φ
+ + zi X i + + zi =0 (12.2)
∂r 2 r ∂r ∂r 2 r ∂r ∂r ∂r

 
∂ 2φ 1 ∂φ 
 2
+ = −β zi X i (12.3)
∂r 2 r ∂r i

Of course, as mentioned earlier, the concentrations can also be scaled on the ionic
strength, for which β = 1.
It is the boundary conditions that differ from those considered in Section 9.2.8.
The speed of the particle must be determined by a force balance. The boundary
conditions on the walls are obtained by assuming the no-slip condition at the two
walls and using the force balance on the particle surface at r = α, with α = a/b
to determine the particle velocity. The dimensional force balance at the particle
surface is given by
∂w∗
λe E x + 2παbμ =0 (12.4)
∂r ∗
480 Essentials of Micro- and Nanofluidics

where λe is the line charge density on the particle and μ is the dynamic viscosity;
E x is the constant externally applied electric field. Rearranging and nondimen-
sionalizing this expression using the length scale b and the standard electro-
osmotic velocity scale (see Chapter 3), the dimensionless boundary condition for
velocity at the inner wall is given by

∂w −λe
= (12.5)
∂r 2παe φ0

The boundary conditions for the potential are obtained using the surface charge
density boundary condition at the particle surface:

∂φ ∗ −σ 0
= (12.6)
∂r ∗ e

where σ 0 , the surface charge density of the particle, is approximated as a uni-

formly distributed charge corresponding to the charge per unit length, λe . Thus

λe
σ0 = (12.7)
2παb
In dimensionless form, equation (12.6) becomes

∂φ −λe
= (12.8)
∂r 2παe φ0
−σ 0 b
= (12.9)
e φ0

The no-flux boundary condition is used as the boundary condition for mole
fractions. Equations (12.2), (12.3), and (12.1) are thus subject to the boundary
conditions

∂w −σ 0 b ∂φ −σ 0 b ∂ Xi ∂φ
= , = , + zi X i = 0 at r = α (12.10)
∂r e φ0 ∂r e φ0 ∂r ∂r

∂φ σ 1b ∂ Xi ∂φ
w = 0, = , + zi X i =0 at r = 1 (12.11)
∂r e φ0 ∂r ∂r

As long as the particle fills the tube, there is no pressure gradient, and in
this case, the governing equations for the velocity and potential are seen to be
identical, and the boundary conditions differ only at the particle surface. Thus
the velocity and the electrical potential differ by a constant

w =φ+C (12.12)

where C is a constant. The governing equations can be solved numerically using

either a direct tridiagonal solver or an iterative technique, as discussed in Chap-
ter 10. In the next section, we discuss the results of the numerical simulation.
 481 Applications

σ0 = −0.01 C/m2 σ0 = −0.01 C/m2

0.4
0.024
σ1 = −0.02 C/m2
Dimensionless Potential

0.3 σ1 = −0.02 C/m2 0.022

σ1 = −0.01 C/m2
1 2
σ = −0.01 C/m

Mole Fraction
and Velocity

0.02 1 2
0.2 σ = −0.005 C/m

0.018
0.1
σ1 = −0.005 C/m2 0.016 σ1 = −0.005 C/m2
Na
0 σ1 = −0.01 C/m2
Potential 0.014 Cl
Velocity σ1 = −0.02 C/m2
−0.1
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
r r
(a) Potential and Velocity (b) Mole Fractions

The numerical results for (a) potential difference, velocity and (b) mole fractions for the case of a two-component
Figure 12.3
aqueous electrolyte mixture. The surface charge density of the inner cylinder is −0.01 C/m2 , while that of the
outer cylinder varies from −0.005 C/m2 to −0.01 C/m2 up to −0.02 C/m2 . The electrical current is calculated
to be 450.1 pA. The dimension of the annulus is b = 10 nm, α = 0.2.

12.2.3 Results
Results in this section are for dsDNA, for which the surface charge density is a
matter of some debate and quoted values range from −0.01 C/m2 to −0.8 C/m2 .
To illustrate the behavior, computations were carried out for three cases, with
the surface charge density on the inner cylinder, the DNA, held constant at
−0.01 C/m2 and the outer wall surface charge density varying from −0.005 C/m2
up to −0.02 C/m2 , with an applied electric field corresponding to 1.0 V over a
channel of length 10 µm and with a fully dissociated electrolyte concentration of
1 M in the upstream reservoir. The outer radius is 10 nm and the inner cylinder
radius is 2 nm in the calculations, which is about the radius of dsDNA. Figure 12.3
shows the dimensionless velocity and potential and notes that the double layers
are thin.
The variation of the DNA translocation velocity with electrolyte concentra-
tion is shown in Figure 12.4. It is evident from this plot that the concentration
of the electrolyte taken to be its value in the upstream reservoir plays a strong
role in the translocation velocity of the DNA. Although the dsDNA has zero
velocity when both surfaces have equal surface charge density (irrespective of
the electrolyte concentration), the velocity varies nonlinearly with concentration
when the difference of surface charge densities is nonzero. Figures 12.4a and
12.4b show the variation of inner cylinder velocity as a function of the concen-
tration for two representative cases, one with σ 1 = −0.005 C/m2 and the other
with σ 1 = −0.02 C/m2 , with the inner cylinder surface charge density held at
σ 0 = −0.01 C/m2 for both cases. Note that the velocity shown in Figure 12.4a
is in the direction opposite to the EOF, whereas in Figure 12.4b, the DNA veloc-
ity is in the same direction as the EOF. We are now in a position to calculate
 482 Essentials of Micro- and Nanofluidics

0 2 1
σ = −0.01 C/m & σ = −0.005 C/m
2 σ0 = −0.01 C/m2 & σ1 = −0.02 C/m2
−0.05
0.28

Dimensionless DNA Velocity

Domensionless DNA Velocity

−0.06
0.24

−0.07
0.2
−0.08

0.16
−0.09

0.1
0.5 1 1.5 2 0.12
0.5 1 1.5 2
Concentration
Concentration
(a) Δσ < 0
(b) Δσ > 0

Nondimensional velocity as a function of reservoir electrolyte concentration (in M) for two different surface
Figure 12.4
charge densites on the outer wall. Here σ = σ 0 − σ 1 .

the total current and its variation with time – the main reason to perform these
calculations.

12.2.4 DNA current

The current density and total current in electro-osmotic flow are discussed in Sec-
tion 9.2.7. In this section, we apply those concepts to the annular model of DNA
translocation. The dimensionless current density in the streamwise z direction is
given by
  ∂ Xi ∂φ

Jz = zi + zi X i − zi + PeX i w (12.13)
i
∂z ∂z

where = E 0 L/φ0 , Pe = ReSc is the Peclet number, and the dimensionless

current is defined by Jz = Jz∗ /cU0 . The total current in the channel is seen to
comprise four parts: the streamwise concentration gradient, molar flux due to the
migration of ions under the applied electric field, the flux due to the perturbation
potential gradient, and the flux due to the bulk velocity of the electrolyte. The
current through the channel is the integral of the current density across the entire
cross section of the channel:
1
I = 2π Jz r dr (12.14)
α

We can now summarize the theoretical picture of the change in current due
to the transport of a biomolecule through the tube. Note that the current density
contains two terms with axial gradients and two terms that are proportional to
the mole fraction of species i. As the biomolecule enters the channel, the current
density should be dominated by the two gradient terms. The streamwise current
density will jump from its value when the pore is empty to its value when the pore
483 Applications

is full. This time period is so short, however, that it will appear as an instantaneous
rise or drop. During the time period when the biomolecule is within the pore,
the current density will be independent of time and dominated by the terms
proportional to the mole fraction, or
 1
I = 2π zi ( z i X i + PeX i w) r dr (12.15)
i α

and it is seen that it is crucially dependent on the two dimensionless parameters

and Pe.
For thin electric double layers in which the core is electrically neutral and the
mole fractions are constant, the streamwise current density reduces to

I = π (1 − α 2 )X io z i2 (12.16)
i

where X io is the (constant) mole fraction of the electrolyte in the core region and
also in the upstream reservoir. Thus the current density is solely dependent on
the concentration of the electrolytes in the core of the channel. Of course, in the
overlapped double-layer case, the integral must be computed numerically, and
the bulk flow term will play, perhaps, a significant role.
So how do the values of the particle velocity and the current drops compare
with experiment? This is discussed next.

12.2.5 Comparison with experiment

Chen and Conlisk (2010) have incorporated the effect of upstream and down-
stream reservoirs on the DNA translocation velocity and have compared the
computational with experimental data in Table 12.1. The detailed experimental
parameters are also listed in the table. The geometry of the nanopore used in the
Storm et al. (2005, 2003) experiments is similar to the pore shown in Figure 12.5.
In this case, the radius of the nanopore at the inlet region is very large compared
to the Debye length, and an asymptotic solution is used to calculate the electro-
osmotic flow velocity (Chen & Conlisk, 2010). In Li et al.’s (2003) work, the
inlet radius is 50 nm, and the total length is 500 nm; the radius of the cylindrical
pore is 5 nm, and the length of the cylindrical pore is 10 nm. The results for the
DNA velocity calculated numerically compare well with the experimental data
in both cases.

Table 12.1. Comparison on DNA velocity between the numerical results and the experimental data
Source σ 1 (C/m2 ) L DNA (µm) V DNA (m/s) Exp V DNA (m/s) Num
Storm et al. (2005) −0.20 3.91 0.013 0.015
Li et al. (2003) −0.14 0.40 0.010 0.012

Here σw is the surface charge density of the nanopore; LDNA is the length of the DNA used in the experiments;
CKCl = 1 M is the concentration of the KCl solution; VDNA Exp is the experimental data, and VDNA Num is the numerical
result.
 484 Essentials of Micro- and Nanofluidics

SiO2
50nm
R0=5nm

50nm

350nm

The geometry of the synthetic nanopore used for calculation. A 300 nm long conical silica nanopore is connected
Figure 12.5
to a 50 nm long cylindrical pore. The radii of the large and small ends of the conical pore are 50 nm and 5 nm,
respectively (Storm et al., 2003).

As shown in Figure 12.6, the current results compare very well with the
experimental data from Storm et al. (2005) for 11.5 kbps DNA. The base-
line current defined as the current through the nanopore without DNA in it
is Ibaseline = 7085 pA, and the amplitude is I = 160 pA. The corresponding
experimental data is Ibaseline = 7,100 pA and I = 140 pA. The relative current
change (I /I ) is 0.020 for experimental data and 0.022 from the calculation.

12.3 Development of an artificial kidney

12.3.1 Background
End-stage renal disease (ESRD), characterized by a loss of the essential protein
albumin, among other factors, affects over 375,000 Americans and is increasing

7200
Calculated
Experimental(Storm2005)
7150

7100
I(pA)

7050

7000

6950

6900
0 0.5 1 1.5 2
t(ms)

The results for current through the nanopore as a function of time (Chen, 2010). For the numerical results, the
Figure 12.6
baseline current is I baseline = 7085 pA, and the amplitude is I = 160 pA. The corresponding experimental
data on I = 7100 pA and I = 140 pA (Storm et al., 2005).
 485 Applications

Sketch of an implantable renal assist device (RAD). From Conlisk et al. (2009).
Figure 12.7

in prevalence at an annual rate of 8 percent (Fissell & Humes, 2006; Humes

et al., 2006).3 Treatment of ESRD patients by renal transplant is hindered by
shortage of donor organs, and dialysis is expensive, inconvenient, and often
unsuccessful. Tissue engineering of an artificial kidney is an alternative strategy
to dialysis in the treatment of ESRD. The extracorporeal RAD is a bioartificial
kidney that combines hemofiltration, the filtration of small ions, and other small
molecules from the blood with cell therapy to provide the metabolic, endocrine,
and immunological functions of a healthy kidney. A sketch of such a device
appears as Figure 12.7.
Dialysis is the primary renal replacement therapy for most ESRD patients. The
most common implementation of dialysis is hemodialysis, whereby toxins in a
patient’s blood are filtered through a dialyzer. The key component of the dialyzer
is a semipermeable membrane. Blood is pumped over one side of the membrane
at a rate of 200–500 mL/min, while a clean electrolyte solution (dialysate) is
pumped at a similar or higher flow rate on the other side. The membrane is
engineered to allow for diffusive transport of small molecules, including ions
such as sodium and chloride, from the blood, while retaining higher-molecular-
weight compounds such as albumin, coagulation proteins, and immunoglobulins
in the bloodstream. This configuration results in diffusion of urea, creatinine, and
other toxins from blood into the dialysate.
The nanopore membranes described previously in this text can also be used
to filter the small ions, such as sodium and chloride, through the membrane,
while retaining the (relatively) large proteins such as albumin. The pores in
the membrane are on the order of <10 nm in the smallest dimension and are
very wide, W ∼ 40–100 µm of rectangular cross section, and about 4 µm long.
These dimensions are mentioned elsewhere in this book. These filtration methods

3
Shuvo Roy and William Fissell wrote significant portions of this section, and the author is grateful
for their permission to use this material.
 486 Essentials of Micro- and Nanofluidics

Side View
u Nanopore Membrane

Feed Permeate

UF UP y
x a
2h

nanochannels
L=4 μm

L=4 μ m

(a) (b)

(a) A synthetic nanopore membrane for the proposed RAD (Fissell et al., 2007); a side view of a nanopore
Figure 12.8
membrane. (b) Sketch of a spherical particle in a rectangular channel; side view. The dimension into the plane
of the paper is about W = 10 µm. A scanning electron microscope image of the top view of the membrane
appears in Figure 4.1.

mimic filtration mechanisms in nature such as the cell membrane, the respiratory
membrane, the vascular endothelium, and the glomerular basement membrane, a
key structure in the kidney (Scherrer & Gerhardt, 1971; Deen et al., 2001; Martini,
2001), as well as technology for water treatment, desalination, pharmaceuticals,
and food and beverage processing (Zeman & Zydney, 1996). The mathematical
modeling of such a membrane is described next.

12.3.2 The nanopore membrane for filtration

A semipermeable or permselective membrane allows the passage of certain
molecules while restricting the passage of others and is an important component
of filtration mechanisms in nature; the cell membrane, respiratory membrane,
vascular endothelium, and glomerular basement membrane in the kidney are all
semipermeable membranes.
Figure 12.8a is a sketch of a typical membrane used for hemofiltration in the
RAD (Fissell et al., 2007). Figure 12.8b shows an individual nanopore of width
2h confining a molecule of radius a. The membrane consists of a large number
of nanopores (N ∼ 104 ) and connects an upstream feed microchannel where the
average flow speed is U F to a downstream permeate microchannel where the
average flow speed is U P (Figure 12.8a).
The characteristic dimensions of the feed and permeate channels that serve
as the donor and receiver reservoirs in the other applications discussed here
are on the scale of millimeters, and the flow rate in the feed channel is of the
order of milliliters per second; in typical experiments discussed here, an external
pump is used to drive this flow past the membrane (Fissell et al., 2007). The
487 Applications

flow speed U P in the permeate channel is solely due to permeation of the feed
solution through the membrane pores and is not imposed externally, as in dialysis
or hemodiafiltration. The flow across the nanopore membrane is driven by a
transmembrane pressure drop of p ∼ 1 − 2 psi, established between the feed
and permeate solution (Figure 12.8). This is about the same pressure drop that
occurs across the native kidney. Figure 12.8b shows the geometry of an individual
nanopore and the coordinate system chosen for the theoretical model.
To simplify the theoretical problem, the pores are assumed to be straight,
nonintersecting, and of uniform width. The width 2h of each nanopore is ∼10 nm;
for example, 2h = 8 nm for the membranes described in Fissell et al. (2007). As
is evident from Figures 12.8a and 12.8b, the other dimensions of the nanopore
are large enough for the pore to be treated as a slit. A hemofiltration chip in
the first generation of the RAD consists of an array of several (five to nine)
such membranes. The chip is fabricated using a state-of-the-art nanofabrication
technology that involves deposition of a sacrificial layer of SiO2 for the definition
of nanopore size and can be controlled precisely enough to result in a nearly
monodisperse pore size distribution (<1% variation across the chip) for 5 nm
pores.
The sieving coefficient is a measure of the ability of a membrane to retain
solutes, in this case, primarily albumin. The sieving coefficient of a solute with
concentration c P in the permeate channel and c F in the feed channel is defined
by S = c P /c F . A small sieving coefficient S ∼ 10−4 is desirable, and the inverse
of the sieving coefficient 1/S can be considered a quantitative measure of its
selectivity (Zeman & Zydney, 1996).

12.3.3 Hindered transport

A sketch of the geometry of a nanopore membrane is depicted in Figure 12.8.
In this case, the channels are very wide, W ∼ 100 µm, and so the flow may be
assumed to be one-dimensional. Because the pores of the membrane are so small
(h < 10 nm), the diameter of an albumin molecule (d = 7 nm) is comparable to
the size of the pore, and transport within the pore will be hindered by the close
proximity of the walls. The flux of a biomolecule of valence z in the streamwise
direction in the nanopores of interest is given by

∂c A
N A = −K d D A + z K d m A E x∗ c A + K c u ∗ c A (12.17)
∂x∗
where the quantities in this equation are all averaged over the cross section of the
pore. For example, u ∗ is the average velocity, and c A is the average concentration
in the pore. If the ratio of the size of the particle to the pore height is large
enough, the particle or biomolecule will tend to stay in the center of the channel.
The hindrance factors K c and K d can be calculated based on this so-called
centerline approximation. Conversely, Dechadilok and Deen (2006) have shown
488 Essentials of Micro- and Nanofluidics

that this approximation is not necessary, and they obtain

1−3.02χ 2 +5.776χ 3 −12.3675χ 4 +18.9775χ 5 −15.2185χ 6 +4.8525χ 7

Kc =
1−χ
(12.18)

1+ 9
χlog(χ) − 1.19358χ + 0.4285χ 3 − 0.3192χ 4 + 0.08428χ 5
Kd = 16
1−χ
(12.19)
where χ = a/ h is the ratio of the solute radius (i.e., albumin) to the channel
half-height.
Equation (12.17) can be written in dimensionless form as in Conlisk et al.
(2009), and the flux N A takes the form
   
1 dX
N A = Kc − − z Ex X + u X (12.20)
Pe H d x
assuming only streamwise mass transport. Here the dimensionless quantities are
defined by x = x ∗ /L, u = u ∗ /U0 , where U0 is the centerline velocity in Poiseuille
flow and E x = E x∗ L/φ0 and N = N ∗ /(U0 cX 0 ), where X 0 is the mole fraction of
species A, albumin at the entrance to the pore, and c is the total concentration
of the mixture. Here, as usual, φ0 = RT /F = 26 mV. The quantity Pe H =
K c U0 L/K d D A is the hindered transport Peclet number and can be interpreted as
the characteristic ratio of axial convection speed and axial diffusion speed of a
solute constrained by the pore.
The condition of a constant flux of the species requires d N A /d x = 0, leading
to the governing equation for the species mole fraction X as

d2 X dX
2
= Pe H u(1 + s) (12.21)
dx dx
This equation is subject to boundary conditions at the entrance and exit of the
pore. These boundary conditions require some thought.
As a solute approaches the entrance of a single pore in a nanopore membrane,
the concentration or mole fraction of the solute at the pore entrance, say, x = 0,
changes sharply from its value in the bulk feed solution. This effect can be
quantified using a partition coefficient F, defined by

X (0) = X F F (12.22)

where X (0) is the solute mole fraction at the slit pore entrance and X F is the the
same quantity in the bulk feed stream. Similarly, near the exit of the pore, we
define X (1) = X P P, where the subscript P denotes permeate. The value of the
mole fraction X P depends on the permeate flux and so is unknown.
What about the boundary condition at x = 1? The total flux of the solute at
the exit of the slit pore should balance the net flux in the permeate stream. The
489 Applications

net flow efflux through this region must equal the permeate flux,
X (1)
NA = u X P = (12.23)
P
Inserting N A from equation (12.20) (evaluated at x = 1) into equation (12.23) at
 
dX 1
|x=1 = Pe H u 1 + s − X (1) (12.24)
dx P Kc
where the dimensionless parameter
z D E x∗ F K d z E x∗
s= = (12.25)
RT u ∗ K c Pe H u
incorporates the effect of any applied or induced electric field (E x ) and
−1 < s < ∞.
Equation (12.21) can be solved with the boundary conditions (12.22) and
(12.24) to give (Conlisk et al., 2009)
1 + [(1 + s)P K c − 1] exp [−Pe H u(1 + s)(1 − x)]
X (x) = X F F (12.26)
1 + [(1 + s)P K c − 1] exp [−Pe H u(1 + s)]
The sieving coefficient S defined as X P /X F = X (1)/X F P calculated from this
distribution is
(1 + s)F K c
S= (12.27)
1 + [(1 + s)P K c − 1] exp [−Pe H u(1 + s)]
The asymptotic forms of equation (12.27) for large and small Peclet numbers are

S∞ = F K c (1 + s) for Pe H  1 (12.28)
F
S0 = for Pe H  1 (12.29)
P
Note that for large Peclet number, S∞ → 0 as s → −1. Recall that it is desired
that the sieving coefficient for albumin be small, specifically, S ∼ 10−4 . This can
be achieved by adjusting the electric field E x .
In the case of a negatively charged species like albumin, a negative value of
s implies an applied electric field directed from feed to the permeate side. The
case s = −1 means that convection effects are balanced by electromigration. In
this case, from equation (12.21), the mole fraction is linear, and
 
Pe H u
X = X0F 1 − x (12.30)
P K c + Pe H u
for which the sieving coefficient is
FP K c
S= (12.31)
P K c + Pe H u
and is independent of K c since Pe H = K c u L/K d D, but the sieving coefficient
is dependent on K d .
Conlisk et al. (2009) discuss explicitly the determination of the feed and
permeate partition coefficients F and P as well as presenting results for the
 490 Essentials of Micro- and Nanofluidics

1.20

1.00

Seiving coefficient (S)

0.80

0.60

0.40

0.20 BSA
2
−σ (C/m )=0.04,0.02,0.01 (z=−20)
p
0.00
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
a
h

Experimental sieving coefficients of bovine serum albumin (BSA), thyroglobulin, and carbonic anhydrase (CA)
Figure 12.9
in 1 × PBS, 10 × PBS (circles colored black), and bovine blood with clotting factors removed as a function
of the ratio a/h of protein hydrodynamic radius (a) and pore half-width (h) and comparison with theoretical
curves obtained under the assumption of steric partitioning (solid curve) and electrostatic partitioning as
applicable to bovine serum albumin (dashed curves). The shapes of the symbols indicate the type of protein:
circles, triangles, and squares stand for BSA, thyroglobulin, and CA, respectively. The circle colored black near
a/ h = 0.65 is the only data point in blood. The experiments were conducted in membranes of pore widths
2h = 7, 9.69, 10.9, 12.78, 42 nm.

mole fraction and sieving coefficient. For example, under the assumption that the
solution is dilute and there are no significant electrostatic or other long-range
interactions between the solute molecules and the pore wall,

F =P =1−χ (12.32)

where χ = a/ h is the ratio of the solute radius (a) to the pore half-width (h)
(Deen, 1987). The expression describes what is termed steric partitioning, the
partitioning due to the finite size of the particle or biomolecule. Thus the boundary
condition at the entrance to the pore is X (0) = X F (1 − χ).
We have just seen that an imposed electric field can significantly reduce the
sieving coefficient. This is also true if the semipermeable membrane itself is
charged. Conlisk et al. (2009) compare the results of their analysis for charged
pores with experiments conducted at the Cleveland Clinic. The theory for a
charged solute in a charged pore is compared with these experiments in Fig-
ure 12.9. All data in 1 × PBS feed solution are indicated with hollow symbols,
the data in 10 × PBS are indicated with solid symbols colored black, and the only
datum in blood (also identifiable uniquely by a horizontal coordinate value of
a/ h = 0.65) is indicated with a solid symbol colored red. The dashed curves are
obtained using a charge number of −20 for bovine serum albumin (BSA) and for
three different values of surface charge (σ = −0.01, 0.02, 0.04 C/m2 ) density
on the pore walls. The experiments were conducted in membranes with the pore
widths, 2h = 7, 9.69, 10.9, 12.78, and 42 nm, and so different experiments could
be performed at the same value of χ. The dashed curves indicating electrostatic
partitioning give significantly lower sieving coefficients, beginning at χ ∼ 0.7,
than for steric (size exclusion) effects alone. This value of χ indicates that the
nanopore membrane should have 2h ∼ 10 nm or smaller. In addition, it would
 491 Applications

seem that the use of a charged membrane can reduce the sieving coefficient
substantially.

12.4 Biochemical sensing

12.4.1 Introduction
In Section 12.2, we spoke of the fact that the DNA translocation through a
nanopore can significantly change the current through the nanopore. How that
current change is determined is part of the function of biochemical sensing. In this
section, we qualitatively describe the processes involved in biochemical sensing.
In this section, we focus primarily on electrochemical sensors and optical sensors.
The Sandia device depicted in Figure 1.2a is an example of an electrochemical
sensor.
Chemical and biochemical sensors are used to detect chemical and biochemical
species of interest: the analyte. The essential parts of a biochemical sensor are the
recognition element or a receptor and a transducer that transforms the event into
a signal that represents the effect of the event. Chemical or biochemical sensors
that use a biological recognition element, such as antibodies, enzymes, nucleic
acids, cells, or a synthetic receptor mimicking that found in nature, are called
biosensors (McNaught & Wilkinson, 1997; Spichiger-Keller, 1998).
There are many areas of application, including, for example, implantable
nanopore arrays for glucose monitoring; because of the size of the market, glu-
cose monitoring has been one of the major applications in this area (Ravindra
et al., 2007). The ultimate goal is the continuous monitoring of glucose levels
in the blood or urine using an implantable device that will automatically adjust
glucose concentration to the proper level. Such continuous monitoring eliminates
the need for unpleaseant and painful injections.
Other applications include detection of airborne pathogens, such as nerve and
mustard gas, and explosives such as TNT and biological warfare agents such
as anthrax and ricin; detection and removal of environmental pollutants (e.g.,
electrokinetic remediation); and clinical applications such as treatment of cancer,
measurement of key biological entities sustaining human health such as folic acid,
vitamin B12 , biotin, and so on, determining pH level, and detecting pollutants
such as botulinum and anthrax in water.
An objective that is now possible in biochemical sensing is the ability to detect
and analyze single molecules. Indeed, biochemical sensing and, in particular,
single molecule analysis permeates the four-volume work of Ferrari (2006); each
of the four volumes contains at least several papers whose basic themes are
biochemical sensing, and volume IV is titled Biomolecular Sensing, Processing
and Analysis.
There are a number of reviews on the subject, including Vo-Dinh (2006),
Spichiger-Keller (1998), Rosi and Mirkin (2005), Jianrong et al. (2004), D’Orazio
(2003); and Scheller et al. (2001). The books on biosensors by Eggins (1996) and
 492 Essentials of Micro- and Nanofluidics

Flow

Solution
+
Analyte

Transducer
Recognition
Analyte Element

Measureable
Signal

Sketch of a biosensor showing the recognition element and the transducer.

Figure 12.10

Kress-Rogers (1997) are a good source for background on biosensors in general.

A partial list of companies in the biosensing business is given by Ravindra et al.
(2007).
A sensor utilizes a specific means of detecting a given analyte such as the
action of an antibody for a specific analyte; this biochemical reaction will then
produce a signal (the transducer) such as a color change or a change in electrical
potential. Nature has created a particularly large variety of membrane sensors,
such as ion channels, that can discriminate between a large number of stimuli
(Braha et al., 2000b; Bayley and Cremer, 2001).
The two most important properties of a sensor are its sensitivity and its selec-
tivity. Sensitivity of a sensor is related to its limit of detection, that is, the smallest
concentration of an analyte in a sample that will yield a measureable output signal
based on a given input signal. If the sensitivity of a sensor is low, false negatives
will occur. Selectivity refers to the ability to detect a given target. If a sensor is
not selective enough, false positives will occur. A generic sketch of a biosensor
is depicted in Figure 12.10.
An extensive discussion of all of the possible applications and the types of
sensors available is clearly beyond the scope of this section. The purpose here is
to introduce the reader to the components of a sensor, specifically, a biosensor,
and how micro- and nanofluidics can be used in their operation. The small size of
these systems means that in a biomedical application, the device can be brought
to the patient; such devices are called point-of-care (POC) systems.

12.4.2 What is a biosensor?

The recommended definition of a biosensor (Thévenot et al., 2001) from the
International Union of Physical and Applied Chemistry (IUPAC) is as follows:

A biosensor is a self-contained integrated device which is capable of providing

specific quantitative or semi-quantitative analytical information using a biological
 493 Applications

Conceptual diagram of the biosensing principle and schematic classification of biosensors according to
Figure 12.11
receptors and transduction mechanism. (Reproduced from Vo-Dinh (2006)).

recognition element (biochemical receptor) which is in direct spatial contact with

a transducer element. A biosensor should be clearly distinguished from a bioana-
lytical system which requires additional processing steps, such as reagent addition.
Furthermore, a biosensor should be distinguished from a bioprobe which is either
disposable after one measurement, i.e. single use, or unable to continuously monitor
the analyte concentration.

Biosensors can be classified according to their receptors as well as their trans-

ducers (Vo-Dinh, 2006). Figure 12.11 adapted fromVo-Dinh (2006) provides
a conceptual diagram of a biosensor and summarizes these two classification
schemes. The first 10 pages of Vo-Dinh (2006) are an excellent overview of
biosensors for nonspecialist engineers and scientists. When several biosensors
are integrated on a microchip to perform detection of several molecules, the
integrated system is called a biochip (Vo-Dinh, 2006).

12.4.3 Receptor-based classification of biosensors

The immune system of biological organisms produces antibodies that can target
specific foreign proteins (antigens) introduced into the body fluids. This type of
interaction is the basis for antibody biosensors that perform immunoassays.4 One
example of this antigen–antibody pairing is streptavidin, which is the antigen
to the biotin antibody. Biotin is a B-complex vitamin that is soluble in water
and promotes healthy cell growth; it is present in eggs and some vegetables.
Streptavidin is a protein that is present in the bacterium streptomyces. Nucleic
acid biosensors are typically based on DNA hybridization, or the study of the

4
An immunoassay is a test that uses the binding of antibodies to antigens to identify and measure
certain substances.
494 Essentials of Micro- and Nanofluidics

sequence of ATGC pairings. Enzymes can be used as catalysts in various pro-

cesses, including the processes for detection of an enzyme-tagged molecule;
Enzyme-Linked ImmunoSorbent Assay (ELISA) is a well-known immunoassay
to detect the presence of antibodies or antigens in a given sample.

12.4.4 Transducer-based classification of biosensors

As shown on Figure 12.11 an alternative to a receptor-based classification system
is a transducer-based classification scheme. The most commonly used detection
methods are optical, electrochemical, and mass-based systems.

OPTICAL DETECTION METHODS

Optical methods typically use microscopes equipped with sources of radiation
such as plane polarized light, laser radiation, X-rays, or UV rays that illuminate the
zone to be interrogated in the system of interest. Other common optical system
accessories are cameras to record images, prisms, optical fibers, wave guides,
and the necessary electronics. Fluoroscopy uses fluorescent-tagged molecules
for optical detection. Conventional fluoroscopy uses volumetric illumination. In
total internal reflection fluorescence (TIRF) microscopy, selective illumination
of a surface is possible using evanescent waves; this allows adsorption of analyte
to a solid substrate to be detected (Axelrod et al., 1984; Sadr et al., 2004). The
intensity of fluorescence (Staubli et al., 2005) as well as a decrease in fluorescence
intensity (quenching) may be measured (Humbert et al., 2005).
Another precise optical technique is surface plasmon resonance, which uses
surface illumination and measures electron charge oscillations instead of fluo-
rescence (Homola et al., 1999). Optical detection systems are very precise and
are often capable of giving real-time information (Brecht & Gauglitz, 1997),
although they are very expensive.

OTHER DETECTION METHODS

Colorimetric methods use changes in color of a species used as a tag (such
as a 4-hydroxyazobenzene-2 carboxylic acid (HABA) tag with streptavidin)
to detect a binding event (Green, 1970). In many cases, electrochemical sig-
nals, such as electrical conductivity (conductimetry), current (amperometry), or
voltage (voltametry), are used for detection (D’Orazio, 2003; Scheller et al.,
2001). Atomic force microscopy offers a technique based on changes in the
force field associated with a binding event (Allen et al., 1996; Rosi & Mirkin,
2005). Some detection techniques depend on the measurement of the change in
mass when the analyte attaches to a surface containing a receptor, for example,
through measurement of acoustic vibration frequency (Thévenot et al., 2001;
Vo-Dinh, 2006). Novel nanofabricated sensors, such as nanowires, colloidal gold
particles, and carbon nanotubes, are also being used for biosensing (Rosi &
Mirkin, 2005).
495 Applications

12.4.5 Evaluation of biosensor performance

The key parameter affecting the performance of a biosensor is the minimum
concentration of analyte the biosensor can detect; this can range from as low
as attomolar (Kemery et al., 1998; Gong et al., 2008) to micromolar (Leca-
Bouvier & Blum, 2005). The efficiency of a biosensor can be assessed either by
determining the ratio of the number of moles of analyte captured on the wall
in the solid phase (Figure 12.10) to the maximum capacity of the solid phase
or by the ratio of the number of moles captured in the solid phase to the total
number of moles supplied from the moving solution. This can be quantified using
binding curves (a general term common in separations literature) or dose response
curves (more common in the drug delivery literature) with solution phase analyte
moles/concentration as the independent variable and the solid phase captured
moles as the dependent variable.
Mixing a solution is often used to enhance binding of an analyte, either to
a surface or to another molecule in the bulk. Mixing a sample with a solution
can be used to attach the analyte (target) to a complementary tagging molecule.
Surface delivery is often required to ensure that the target molecules interact with
a detection surface and is often used in conjunction with mixing to enhance the
concentration of the analyte at the detection site.
Mott et al. (2006), Golden et al. (2007), and Mott et al. (2009) have devel-
oped a toolbox for optimized mixing in a microfluidic channel using complex
herringbone-like structures (i.e., grooves) to capture the analyte. The applica-
tion is the detection of TNT in seawater; the Reynolds number is on the order
of Re ≤ 10. Several groove configurations are depicted in Figure 12.12a. The
channel has a width of W = 1016 µm and a height h = 325 µm. The effect
of the grooves is to steer fluid toward the sidewalls to the center of the chan-
nel and toward the floor and the ceiling – the interrogation zones. Using what
they call advection maps, the authors were able to sample over 500,000 groove
configurations for optimizing mixing.
In Figure 12.12b are results for grooves pointing both upstream and down-
stream. The value of N denotes the total number of grooves on the floor and the
ceiling – the sample has passed. The value N = 20, for example, indicates that the
sample has passed 10 grooves on the floor and 10 grooves on the ceiling. The solid
symbols indicate that particles have not yet reached within 0.1h of the ceiling or
the floor. The hollow cylinders indicate particles that have come within 0.1h of the
ceiling or the floor. Note that by N = 50, almost all of the particles tracked have
reached the target threshold. Mott et al. (2009) found that the results for the fea-
tures pointing downstream increased the number of particles reaching the target
threshold by almost 3 times compared to the baseline channel with no grooves.
Low Reynolds number vortices may also be generated as a result of an engi-
neered change in surface charge density or ζ potential and used to improve surface
delivery of an analyte; the situation is depicted in Figure 12.13. Anderson and
Idol (1985) appear to have been the first to investigate this phenomenon (see also
 496 Essentials of Micro- and Nanofluidics

(a) Schematic of several features (grooves) placed on the ceiling and floor of a channel to enhance surface
Figure 12.12
delivery. (b) Results for both upstream and downstream facing features; the lighter shade indicates that a
sample has come within 10% of the channel height to the surface. Note that nearly all of the sample has
reached this threshold. From Mott et al. (2009).

Erickson & Li, 2002) analytically, and Stroock et al. (2000) have observed this
effect experimentally (see also Chen et al., 1997, and the references therein). As
has been seen in electro-osmotic flow, the direction of the velocity field depends
on the sign of the wall surface charge. An abrupt change in the wall charge or
potential of the same sign will cause a “vortex” to form. The results are slightly
different from a modeling perspective for the case of a changing wall charge com-
pared with a changing ζ potential, but both result in a vortex that can enhance
surface delivery. It should be noted that because of the discontinuous nature of the
potential and the finite Peclet number, the solution for the concentrations deviates
from the Poisson–Boltzmann distribution. Note that in Figure 12.13, the vortex
covers about 25 percent of the channel; it turns out that the size of the vortex
decreases with decreasing Peclet number and that the measure of the deviation of
the numerical solution from the Boltzmann distribution increases with increasing
Peclet number.

12.4.6 Nanopores and nanopore membranes for biochemical sensing

While uncommon just a few years ago, many research groups around the world
are exploring the use of nanopores and nanopore membranes for sensing many
 497 Applications

(a) (b)

Schematic of a channel having a potential discontinuity.

Figure 12.13

different molecules. Much of the work is still in the research stage, and some of
these concepts have been applied to defense-related problems.5
It is well known that miniaturization of a given device for detection and rapid
molecular analysis has several advantages. First, because the device is small, the
amount of analyte and sample required will be small. Also, it is clear that the
smaller the flow path, the shorter will be the required analysis times.
Braha et al. (2000) showed experimentally that the occupation of a pore by
ions could be detected by a modulation of the current passing through the pore.
This has been demonstrated theoretically in the preceding analysis of the DNA
transport problem. They used the natural α hemolysin (αHL) channel as the
recognition element, a pore that is 10 nm long and consists of a water-soluble
293-amino acid polypeptide chain that naturally occurs in biological membranes.
As was mentioned before, Bayley and his co-workers (Bayley & Cremer, 2001)
could identify various types of DNA down to single base-pair resolution as they
pass through a α hemolysin nanopore. They also showed that electro-osmotic
flow could significantly increase the effectiveness of the binding of a neutral
molecule (in this case, β−cyclodextrin) to an αHL pore.
While early micro- and nanofluidic devices used some form of the silica family
of surfaces for the channels, many researchers have used PDMS and PMMA for
their ease of use and inexpensive fabrication procedures. Saleh and Sohn (2006)
have demonstrated that a single nanopore can be used for a wide variety of appli-
cations, including immunoassays and DNA sizing. Using the fact that the ratio of
the change in current to the total current is proportional to the ratio of the volume
of a particle to the volume of the pore (as shown theoretically in Section 12.2.4),

5
The Defense Advanced Research Projects Agency (DARPA) program Engineered Biomolecular
Nanodevices and Systems (MOLDICE) was created to investigate the use of biological sensing
elements to detect pathogens.
 498 Essentials of Micro- and Nanofluidics

they were able to measure the radius of a DNA strand, about a = 2 nm, a value
that was used in the section describing the transport of DNA through a tube.

12.5 Chapter summary

The purpose of this chapter is to introduce the reader to several application

areas and illustrate the use of the modeling techniques described in the previous
chapters. The main application areas discussed here are as follows:
r DNA transport and electroporation
r Transport processes in a RAD
r Biochemical sensing

These three areas were chosen because of the crucial role that micro- and nanoflu-
idics plays in the operation of the devices in each area. These areas are certainly
not all-inclusive of the myriad applications, and the reader is encouraged to con-
sult the references listed in this chapter, especially the four-volume set edited by
Ferrari (2006) and the MEMS handbook of Gad-el Hak (2001).

EXERCISES
The exercises here are meant to be open ended, and each exercise contains
a writing assignment. Some of these exercises may be appropriately assigned
after Chapters 9 and 7 are covered. The writing assignments can be of a length
determined by the instructor, depending on the purpose of the exercise. These
exercises are designed for the student to make extensive use of the Web.

12.1 Electro-osmotic flow can be used to remove metal contaminants from satu-
rated clay and other soil materials. Using the governing equations for EOF
in a porous media in the paper by Shapiro and Probstein (1993), and the
parameters defined therein, determine the total amount of the contaminant
that can be removed as a function of energy input. Write a paper describing
the basic principles of an electrokinetic remediation (ER) process and your
results. Define the efficiency of an electrokinetic remediation process.
12.2 Carbon easily binds to carbon in a unique way. Carbon nanotubes or
buckytubes can be synthesized into a hollow cylinder made up of a
single array of carbon atoms; these tubes are called single-wall carbon
nanotubes (SWCNT). These tubes have a variety of applications. Write a
paper describing the the applications of SWCNTs to biochemical sensing.
12.3 Carbon nanotubes are also used for cancer diagnostics. Write a paper
discussing the specific way in which a carbon nanotube is used in cancer
diagnostics.
12.4 Another application of nanopore membranes is the purification of water.
Write a report on the basics of water purification using reverse osmosis,
499 Applications

and identify the difference between reverse osmosis membranes and

nanofiltration membranes. Also explain the use of nanoparticles, such as
carbon nanotubes, for the purification of water.
12.5 The world depends on a steady supply of fresh water for a variety of
activities. Unfortunately, only 1 percent of the total supply of water is
fresh water. Nanopore membranes can also be used to desalinate water.
Explain how such membranes can be used to desalinate water, and write
a paper on your results. Using a given supply of salt water, determine
how efficiently the membrane removes the salt from the water.
12.6 Write a report on the electrochemical aspects of the operation of a battery.
12.7 Repeat the previous exercise for a fuel cell.
12.8 Chromatography is the process of separating out chemical species in a
porous medium by selective adsorption on a solid surface. Neglecting
diffusion, using a simple rectangular control volume in a porous medium
of void fraction , derive first order partial differential equations
governing the concentration of the solute in the liquid and that adsorbed
on the substrate surface. Consult Rhee et al. (1986, page 33) if necessary.
12.9 The artificial kidney depends on a pressure-driven flow for its sieving
mechanism. For the parameters of the artificial kidney, write a report
discussing the influence of the streaming potential.
12.10 Equation (12.26) determines the influence of an externally imposed
electric field on the mole fraction of an analyte. Compare the sieving
coefficient for the imposed field with that for a charged membrane.
12.11 Design an implantable drug delivery system for the delivery of insulin to
a type I diabetes mellitus patient. Write a paper on your design. Be sure
to begin the paper with a definition of diabetes and why the patient needs
additional insulin.
12.12 Ion channels supply cells with needed nutrients and are tailored to their
specific function. Ion channel walls are made up of charged proteins
and are essentially natural nanopores. Write a paper to characterize the
specific ion channel of your choice (there are many), and identify the role
of ion channels in the disease cystic fibrosis.
12.13 Design a nanoparticle for the treatment of a lung cancer tumor.
12.14 What is a quantum dot?
Appendix A
Matched Asymptotic Expansions

A.1 Introduction

Why study perturbation techniques? The answer lies in the fact that we can
simplify the problem significantly if we study the structure of the equation. In
some cases, analytical solutions may be obtained, which eliminates any need
for computational work. More commonly, the application of these techniques
results in a problem that may be solved numerically much easier than the original
problem.
Perturbation methods had been known since the days of the ancient astronomers
to predict the trajectories of planets subject to small disturbances. The modern
concept of regular and singular perturbation problems was developed in the
1960s and refined over the 1970s in fluid mechanics to investigate the high
Reynolds number flow past a flat plate, the boundary layer. The function of
a boundary layer is to bring the velocity to zero on a solid body. Because
the boundary layer is very thin, of O(Re−1/2 ), the velocity will vary rapidly
across the layer, making numerical solutions very difficult, if not impossible. In
the intervening years, boundary layer–type behavior, the rapid variation of any
quantity over a short distance or period of time, has been used in a wide variety
of disciplines to describe such behavior.
There are a number of books on this subject, and these appear in the bibliog-
raphy; they include Kevorkian and Cole (1981, 1996), VanDyke (1975), Nayfeh
(1973), Bellman (1964), Bender and Orszag (1999), Holmes (1995), Hinchcliffe
(2003), and Howison (2005). The book by VanDyke (1975) is specifically oriented
toward fluid mechanics, whereas the others are more general treatments.

A.2 Terminology

Before moving on to some examples, we need to define some terms commonly

used in perturbation theory. Ordering of numbers and functions with respect to
the small parameter is important. We say, for example, 2 = O(3), which is stated
in words as “2 is of the order of 3”; similarly, when a given number is “about”
10 times another number, we say 1  10, or “1 is much less than 10.” Similarly,
501
 502 Appendix A. Matched Asymptotic Expansions

10  1. There are no hard boundaries for this terminology, and these ordering
symbols should be treated as estimates; for example, we might say 2  10, but
this estimate is at the boundary of validity. In the case of functions, f (x)  g(x)
means that “f is much less than g, for all x.” The technical definitions of the two
symbols x  y and x = O(y) may be expressed in terms of the small parameter
 as f () = O(g()) as  → 0 if
f ()
lim <∞ (A.1)
→0 g()
and f ()  g() as  → 0 if
f ()
lim =0 (A.2)
→0 g()
It should be noted that f  g is sometimes written f = o(g), or “f is little o of
g.”
As specific examples, we write

sin 2 = O() as  → 0

cos() = O(1) as  → 0

We could also write

sin 2 = o(1) as  → 0

Also, we should note that

e−    m
1
for all m.

A.3 Asymptotic sequences and expansions

Perturbation methods most always involve a small parameter that often and per-
haps usually arises as a result of a nondimensionalization process. Consider the
series

f = f0 +  f1 +  2 f2 + · · · (A.3)

where   1. The functions gn =  n are called gauge functions and satisfy the
condition that gn+1  gn or gn+1 = o(gn ) since
gn+1
=
gn
The sequence of functions gn for n = 0, 1, 2, . . . is called an asymptotic sequence,
and the expansion of f is called an asymptotic expansion.
Equation (A.3) looks like a Taylor series expansion, and so the question of
convergence in the mathematical sense may be posed. However, there are two
limiting processes at work here: the limit  → 0 and the limit n → ∞. For a
Taylor series, we take the limit n → ∞ before even thinking about the value
 503 Appendix A. Matched Asymptotic Expansions

of . In asymptotic and perturbation analysis, the limit  → 0 is of paramount

importance. In some sense, then, in an asymptotic analysis, the error should get
smaller as  → 0. Asymptotic analysis is a highly pragmatic subject, and it is
customary to proceed until experience suggests otherwise.
To illustrate the concept, consider the general problem of finding a solution to
the equation

L f = a +  P( f ) (A.4)

where a and P do not depend on ; for simplicity, we assume that a is a constant

and that P is some operation on f . Then, expanding f in an asymtotic series and
equating coefficients of , we find

L( f 0 +  f 1 +  2 f 2 + · · · ) = a +  F( f 0 +  f 1 +  2 f 2 + · · · ) (A.5)
L( f 0 ) = a
L( f 1 ) = F( f 0 )

and so on, where we have assumed that a = O(1). For example, suppose a = 2
and L( f ) = f and P( f ) = f 2 . Then the solution is obviously

f = 2 + 4 + 16 2 + E (A.6)

where E = O( 3 ) is the error. Note the rising coefficients of the increasing
powers of , and we might ask the question, does the series converge? For fixed 
as n → ∞, the answer is no because the ratio of the coefficients gets larger and
not smaller, as in a convergent Taylor series. But in the limit  → 0, for fixed n,
the answer is yes!

A.4 Regular perturbations

A typical example of this type of perturbation is the following problem:

y  − y = 0,   1 (A.7)

subject to

y(0) = 1, y(1) = 0 (A.8)

The exact solution to this problem is

sin h  1/2 (1 − x)
y= (A.9)
sin h  1/2
Note that the small parameter  does not multiply the highest derivative of the
equation. This problem is easy to solve, but let us anticipate solving problems of
a more difficult nature and try an asymptotic expansion of the form

y = y0 + y1 (x) +  2 y2 (x) + · · · (A.10)

 504 Appendix A. Matched Asymptotic Expansions

Each term in the asymptotic expansion must combine to satisfy the equation
and the boundary conditions that are obtained by substituting directly into the
equation; thus

n = 0 y0 = 0 y0 (0) = 1, y0 (1) = 0

n = 1 y1 = y0 y1 (0) = 0, y1 (1) = 0
n=2 y2 = y1 y2 (0) = 0, y2 (1) = 0

for the first three terms of the asymptotic expansion. It should be noted that
an asymptotic series need not converge in the mathematical sense, although the
solution should approach the true solution as more terms are included.
These equations are easy to solve, and the result is

y0 = 1 − x (A.11)
x
y1 = − (x − 1)(x − 2) (A.12)
6
and so on. Typical of these methods is that the solution for higher-order terms in
the expansion gets more complicated, and for this reason, only one or two terms
in the expansion are usually calculated. This is often sufficient for an accurate
solution to a physical problem. To show that the perturbation solution agrees with
the exact solution for small , we can expand the exact solution in a Taylor series
for small , and the result is
x
y = 1 − x −  (x − 1)(x − 2) + · · · (A.13)
6
which agrees with the first two terms in the perturbation expansion. For a reg-
ular perturbation problem, one expansion is sufficient to describe the solution
everywhere in the domain. This is not the case for a singular perturbation.

A.5 Singular perturbations

We should caution here that there are a wide variety of singular perturbation
problems, and the singular nature of the problem may appear in a number of
ways. The most common way of telling whether the problem is of the singular
perturbation type is that the small parameter multiplies the highest-order dervative
in the equation. As noted earlier, this is the case when we set Re = ∞ in the
Navier–Stokes equations.
Let us consider a specific example; we first consider a simple ordinary differ-
ential equation. Consider the equation

y  + y  + y = 0,   1 (A.14)

subject to

y(0) = 1, y(1) = 0 (A.15)

505 Appendix A. Matched Asymptotic Expansions

The solution is easily calculated as

y = Aeα1 x + Beα2 x (A.16)

where α1 and α2 are given by

α1 ∼ −1 −  as  → 0
1
α2 ∼ − + 1 +  as  → 0

Applying the boundary conditions, we find that

y(x) = e−x(  −1−) − e−(  −1−) e(+1)(1−x)

1 1
(A.17)

Let us analyze the solution and further simplify the exact solution. Note that
for  > 0 and  small, y  1, except when x = O(). Let us define η = x/;
then, to leading order in the small parameter , the solution may be expressed as

y = e−η (A.18)

Apparently, for η = O(1), the solution is given by the preceding equation. This
is called the inner solution. Note that it satisfies the boundary condition at η =
x = 0. Away from the boundary η  1, y = 0 is the solution. This is called the
outer solution.
Let us apply the concepts of singular perturbation theory to this problem. To
solve this problem, we must determine which terms in the equation are important
and where. We note that if  = 0, then

y + y = 0 (A.19)

and the solution is

y = Ae−x (A.20)

This is our outer solution. However, A is an arbitrary constant, and at this point,
we do not know how to determine it. We cannot satisfy both boundary conditions
because we have only one constant. If we choose to satisfy the boundary condition
y(0) = 1, then A = 1, whereas if we choose to satisfy y(1) = 0, then A = 0.
Which do we choose? Clearly a regular perturbation expansion procedure will
not work. (Why?) We must decide how to bring in the first and second derivative
terms.
Suppose we decide to balance the two derivative terms and take y  = O(y  );
then
d
= O( −1 ) (A.21)
dx
and the boundary layer variable is
x
η= (A.22)

506 Appendix A. Matched Asymptotic Expansions

With this scaling, the governing equation becomes

d 2Y dY
2
+ + Y = 0 (A.23)
dη dη
where Y is our boundary layer–dependent variable. Thus we expand Y as

Y = Y0 + Y1 + · · · (A.24)

and Y0 satisfies

Y0 + Y0 = 0 (A.25)

This equation is easy to solve, and the result is

Y0 = C + De−η (A.26)

We have assumed that the boundary layer is at x = 0, and so we must satisfy the
boundary condition at η = 0. Thus

C+D=1 (A.27)

The solution is thus

Y0 = 1 + D(e−η − 1) (A.28)

This is our inner solution.

The question is, how do we connect or match the inner and outer solutions? In
the inner solution, we have A as the arbitrary constant, and in the inner solution,
D is the arbitrary constant. A reasonable procedure is to take the limit of the outer
solution as the boundary layer is approached and the limit of the inner solution
away from the boundary layer. This is called a matching principle by Van Dyke
(1975), and mathematically, the condition is written as

lim y0 = lim Y0 (A.29)

x→0 η→∞

Clearly, in our problem, this requires

A =1− D (A.30)

Because the outer solution is valid away from x = 0, we require it to satisfy the
boundary condition at x = 1. In this case, A = 0, D = 1, and we have

Y0 = e−η (A.31)

y0 = 0 (A.32)

The question you may be asking is, why can there not be a boundary layer at
x = 1? You can show very easily that this assumption leads to an exponentially
growing solution in the boundary layer variable:
1−x
ξ= (A.33)

507 Appendix A. Matched Asymptotic Expansions

This is unacceptable, and thus no boundary layer can be present at x = 1. You

can also show that there cannot be a boundary layer of thickness  1/2 ; that is, we
cannot balance d 2 y/d x 2 with y.
As a final note, the solution can be written

y = e−η + 0 − 0
y = inner + outer − common part

and the solution is now valid everywhere on the interval 0 ≤ x ≤ 1. We say that
the solution is uniformly valid. The common part is that part of the solution that
is common to both the inner and the outer solution. Another way of matching
the two solutions is to write the outer solution in inner variables and the inner
solution in outer variables, equate the two, and take the limit as  → 0. In this
case, this procedure yields
Ae −η = 1 + D(e−  − 1)
x
(A.34)

which yields, in the limit  → 0,

A =1− D (A.35)

as we have already seen.

Appendix B
Vector Operations in Curvilinear
Coordinates

If  is a scalar function and A = A 1 e1 + A2 e2 + A3 e3 is a vector function of

orthogonal curvilinear coordinates u 1 , u 2 , u 3 (e1 , e2 , e3 are unit vectors in the
direction of increasing u 1 , u 2 , u 3 , respectively), then the following results are
valid:
1 ∂ 1 ∂ 1 ∂
1. ∇ = grad = e
h 1 ∂u 1 1
+ e
h 2 ∂u 2 2
+ e
h 3 ∂u 3 3
, -
∂ ∂
2. ∇ • A = div A = 1
h1h2h3 ∂u 1
(h 3 h 1 A2 ) + ∂u∂ 3 (h 1 h 2 A3 )
(h 2 h 3 A1 ) + ∂u 2
 
 h e h 3 e3 
 1 1 h 2 e2
 
3. ∇ × A = curl A = h 1 h12 h 3  ∂/∂u 1 ∂/∂u 2 ∂/∂u 3 
 
 h 1 A1 h 2 A2 h 3 A3 
,  
4. ∇ 2  = Laplacian of  = h 1 h12 h 3 ∂u∂ 1 hh2 h1 3 ∂u
∂
+ ∂
∂u
h 3 h 1 ∂
h 2 ∂u 2
 - 1 2
∂ h 1 h 2 ∂
+ ∂u 3 h 3 ∂u 3

where h 1 , h 2 , h 3 are the scale factors:

     
 ∂r   ∂r   ∂r 
h1 =    
, h2 =  , h 3 =  
∂u 1  ∂u 2  ∂u
3


B.1 Cylindrical coordinates

u1 = r h 1 = 1 e1 = er
u2 = θ h 2 = r e2 = eθ
u3 = z h 3 = 1 e3 = ez

B.2 Spherical coordinates

u1 = r h1 = 1 e1 = er
u2 = θ h2 = r e2 = eθ
u3 = φ h 3 = r sin θ e3 = eφ

508
 509 Appendix B. Vector Operations in Curvilinear Coordinates

B.3 Rectangular coordinates

u1 = x h1 = 1 e1 = i
u2 = y h2 = 1 e2 = j
u3 = z h3 = 1 e3 = k
Reference: M. R. Spiegel, Theory and Problem of Vector Analysis: Introduction
to Tensor Analysis, New York, McGraw-Hill.
Appendix C
Web Sites

Here are some Web sites that contain much useful information on micro- and
nanofluidics. This list is not meant to be complete, and I would welcome additional
information on cool Web sites that you may have. In addition, there are individual
researchers’ Web sites at various universities that I have left out. I have also left
out individual company and publisher Web sites. Also, I have grouped these sites
into categories; however, understand that in micro- and nanofluidics, this may be
dangerous.

C.1 Fluid dynamics and micro- and nanofluidics

1. http://www.efluids.com/
Very informative Web site that includes many applets and demos covering
fluid dynamics in general.
2. http://www.emicronano.com/
Very informative Web site that includes images, movies, demos, pictures,
and solver tools for micro- and nanofluids. Run by the same people as
efluids, this Web site concentrates on smaller-scale fluids.
3. http://www.lab-on-a-chip.com/
This site is an RSS feed site that has many current news headlines and
articles from university labs covering microfluidics, microarrays, and labs-
on-a-chip. It also has a list of upcoming events regarding microfluidics and
other information in the field.
4. http://www.rsc.org/Publishing/Journals/lc/Chips and Tips/index.asp
This is a Q&A Web site dedicated to answering questions and providing
information for those interested in labs-on-a-chip.
5. http://www.aip.org/tip/INPHFA/vol-9/iss-4/p14.html
This is a feature article from The Industrial Physicist that explains the
explosion of micro- and nanofluidics technologies in recent years. The
site describes several technologies benefitting from new discoveries in
the microfluidics area and discusses the differences between micro- and
nanofluidics.

510
 511 Appendix C. Web Sites

C.2 General nanotechnology

1. http://nanoHUB.org/
This site provides online nanocomputation tools in a variety of disciplines
such as quantum transport, nanomedicine, and others. Researchers con-
tribute computation tools for use by registered clients. There are currently
over 800 contributors. The site also provides a series of courses in mechan-
ics, biology, and chemistry.
2. http://www.nnin.gov/
This is the site for the National Nanotechnology Infrastructure Network
(NNIN), funded through the National Science Foundation. It is an inte-
grated network of facilities for sharing of nanoscale experimental and
fabrication tools. The url http://www.nnin.org/nnin compsim.html is the
computational portal of NNIN, which operates in a manner similar to
nanoHUB.

C.3 Wikipedia

1. http://en.wikipedia.org/wiki/Microfluidics
Standard definitions and basic information about microfluidics from
Wikipedia. Discusses continuous-flow microfluidics, digital microfluidics,
DNA chips, molecular biology, optics, acoustic droplet ejection, and fuel
cells.
2. http://en.wikipedia.org/wiki/Nanofluidics
Standard definitions and basic information about nanofluidics from
Wikipedia. Discusses theory and includes several equations, specifically
the Poisson–Boltzmann equation. It also has a few paragraphs on nanos-
tructure fabrication and general applications of nanofluidics technologies.
3. http://en.wikipedia.org/wiki/Lab-on-a-chip
Contains lots of information on labs-on-a-chip, and includes a history of
the technology, advantages, disadvantages, fabrications, and relations to
global health and examples.
4. http://en.wikipedia.org/wiki/List of microfluidics research groups
Contains a large list of microfluidics research groups throughout the world
and has links to more information about each group and its research.
5. http://en.wikipedia.org/wiki/Capillary electrophoresis
Wikipedia article that gives in-depth information on capillary electrophore-
sis – a good overview.
Appendix D
A Semester Course Syllabus

Syllabus: Introduction to Micro- and Nanofluidics for Engineers

and Physical Scientists (Three hour lecture, 1 hour lab)

Syllabus for a 4 hour (3 hour lecture, 1 hour lab) semester course 15 weeks long,
assuming 30 lectures of length 1 hour 12 minutes, two lectures per week, and
does not include exam periods. The lab meets once per week, with two lectures
on fabrication and experimental methods – the content of lecture 6a. If the class
meets three times per week, prorate lectures to 45.
Objective: The objective of this course is to introduce students to the basic
physical foundations of incompressible fluid mechanics appropriate to micro-
and nanosize conduits. The course will emphasize the fundamental principles
involved in the formulation and solution of problems in fluid mechanics and
mass transfer for pressure-driven and electrically driven motions of biofluids and
individual components such as ions. On completion of the course, the student
should be able to extract from a raw physical situation the essential principles
from which a useful fluid mechanical model may be developed.
The lab will provide students with an introduction to the design, fabrication,
and testing of microfluidic and nanofluidic devices and how practical devices can
be analyzed theoretically.

Meeting Topic Reading

1 Overview and Historical Perspective 1.1–1.17
Convergence of Molecular Biology and Engineering
2 Transport Properties 2.1–2.3
Viscosity and Diffusion Coefficient
Electrical Permittivity
Surface Tension and Wettability
3 Thermodynamics 2.5, 2.6
4 Kinematics/Forces 3.3, 3.4
Conservation of Mass 3.5
Navier–Stokes Equations 3.6
5 Mass Transport 3.7

512
513 Appendix D. A Semester Course Syllabus

Meeting Topic Reading

Electrostatics: Poisson Equation 3.8
Energy Transport 3.9
6 Boundary Conditions 3.11
Dimensional Analysis 3.12
Fluid and Heat and Mass Transfer Analogies 3.13
6a Fabrication/Experiments/Theory 3.17
7 Fully Developed Flow in Pipes and Channels 4.1–4.3
Slip Flow 4.4
8 Fully Developed Suction Flows 4.7
Developing Suction Flows 4.8
9 Slow Flow Past a Sphere 4.11
Sedimentation 4.12
Blood Flow 4.13
10 The Temperature Distribution in Pipe and Channel Flow 5.1–5.4
11 Graetz Problems 5.5
12 Taylor–Aris Dispersion 5.7
The Stochastic Nature of Diffusion: Brownian Motion 5.9
13 Electric Field 6.1–6.4
14 Electrical Potential 6.5–6.7
Current and Current Density 6.8
Maxwell’s Equations 6.9
15 Electrochemistry 7.1–7.4
16 Chemical Potential, Electrochemical Potential 7.5–7.7
17 Charged Surfaces 7.9, 7.10
18 Electrical Double Layer 7.11–7.13
19 ζ Potential 7.14
Surface Charge Density
20 EDL/Cylinders/Sphere 7.15, 7.16
Semipermeable Membranes 7.18
21 Molecular Biology Chapter 8
DNA and Proteins
22 Electrosmosis 9.1–9.2.3
23 Continued 9.2.4–9.2.6
24 Continued 9.2.8–9.2.10
25 Electrophoresis 9.3.1, 9.3.2
26 Continued 9.3.3, 9.3.4
27 Streaming Potential 9.4
Sedimentation Potential 9.5
Joule Heating 9.6
28 Molecular Simulations Chapter 11
29 Applications Chapter 12
30 Continued Chapter 12
 Bibliography

Abramowitz, M., & Stegun, I. A., eds. 1972 Handbook of Mathematical Func-
tions with Formulas, Graphs and Mathematical Tables. Washington, DC: National
Bureau of Standards.
Abramson, H. A. 1931 The influence of size, shape, and conductivity on cataphoretic
mobility, and its biological significance. J. Phys. Chem. 35, 289–308.
Acheson, D. J. 1990 Elementary Fluid Mechanics. Oxford: Clarendon Press.
Adrian, R. J. 1991 Particle-imaging techniques for experimental fluid mechanics.
Ann. Rev. Fluid Mech. 23, 261–304.
Adrian, R. J., & Yao, C. S. 1985 Pulsed laser technique application to liquid and
gaseous flows and the scattering power of seed materials. Appl. Opt. 24, 44–52.
Alberts, B., Bray, D., Lewis, J., Raff, M., Roberts, K., & Watson, J. D. 1994 Molecular
Biology of the Cell. New York: Garland.
Alberts, B., Bray, D., Hopkin, K., Johnson, A., Lewis, J., Raff, M., Roberts, K., &
Walter, P. 1998 Essential Cell Biology. New York: Garland.
Allen, M. P. 2004 Introduction to molecular dynamics simulation. In Computational
Soft Matter: From Synthetic Polymers to Proteins, NIC Series, vol. 23 (ed. Nor-
bert Attig, Kurt Binder, Helmut Grubmuller & Kurt Kremer), pp. 1–28. Julich,
Germany.
Allen, M. P., & Tildesley, D. 1994 Computer Simulation of Liquids. Oxford: Clarendon
Press.
Allen, S., Davies, J., Dawkes, A. C., Davies, M. C., Edwards, J. C., Parker, M.
C., Roberts, C. J., Sefton, J., Tendler, S. J. B., & Williams, P. M. 1996 In situ
observation of streptavidin-biotin binding on an immunoassay well surface using
an atomic force microscope. FEBS Lett. 390(2), 161–164.
Anderson, J. D. 1982 Modern Compressible Flow: With Historical Perspective. New
York: McGraw-Hill.
Anderson, J. L., & Idol, W. K. 1985 Electroosmosis through pores with nonuniformly
charged walls. Chem. Eng. Commun. 38, 93–106.
Anton, K., & Berger, C. 1998 Supercritical Fluid Chromatography with Packed
Columns: Techniques and Applications. New York: Marcel Dekker.
Archer, D. G., & Wang, P. 1990 The dielectric constant of water and the Debye-Hückel
limiting law slopes. J. Phys. Chem. Ref. Data 19, 371–411.

515
516 Bibliography

Aris, R. 1956 On the dispersion of a solute in a fluid flowing through a tube. Proc. R.
Soc. A 235, 67–77.
Aris, R. 1959 On the dispersion of a solute by diffusion, convection and exchange
between phases. Proc. R. Soc. A 252, 538–550.
Atkinson, B., Brocklebank, M. P., Card, C. C. H., & Smith, J. M. 1969 Low Reynolds
number developing flows. AIChE J. 15, 548–553.
Axelrod, D., Burghardt, T. P., & Thompson, N. L. 1984 Total internal reflection
fluorescence. Ann. Rev. Biophys. Bioeng. 13, 247–268.
Barcilon, V., Chen, D.-P., & Eisenberg, R. S. 1992 Ion flow through narrow membrane
channels: Part II. SIAM J. Appl. Math. 52, 1405–1425.
Barcilon, V., Chen, D. P., Eisenberg, R. S., & Jerome, J. W. 1997 Qualitative properties
of steady-state Poisson–Nernst–Planck systems: Perturbation and simulation study.
SIAM J. Appl. Math. 57, 631–648.
Barrat, J. L., & Bocquet, L. 1999 Large slip effect at a nonwetting fluid-solid interface.
Phys. Rev. Lett. 82, 4671–4674.
Bashir, R., & Wereley, S., eds. 2006 BioMEMS and Biomedical Nanotechnology:
Volume IV Biomolecular Sensing, Processing and Analysis. New York: Springer.
Batchelor, G. K. 1967 Introduction to Fluid Dynamics. Cambridge: Cambridge Uni-
versity Press.
Bavier, R., & Ayela, F. 2004 Micromachined strain gauges for the determination
of liquid flow friction coefficients in microchannels. Measure. Sci. Technol. 15,
377–383.
Bayley, H., & Cremer, P. S. 2001 Stochastic sensors inspired by biology. Nature 413,
226–230.
Bazant, M. Z., Chu, Kevin T., & Bayly, B. J. 2005 Current-voltage relations for
electrochemical thin films. SIAM J. Appl. Math. 65, 1463–1484.
Becker, O. M., & Karplus, M. 2006 A Guide to Biomolecular Simulations. Dordrecht,
Netherlands: Springer.
Bellman, R. E. 1964 Perturbation Techniques in Mathematics, Physics, and Engi-
neering. New York: Holt, Rinehart and Winston.
Bender, Carl M., & Orszag, Steven A. 1999 Advanced Mathematical Methods for
Scientists and Engineers: Asymptotic Methods and Perturbation Theory. New
York: Springer.
Berendsen, H. J. C., Grigera, J. R., & Straatsma, T. P. 1987 The missing term in
effective pair potentials. J. Phys. Chem. 91, 6269–6271.
Berendsen, H. J. C., Postma, J. P. M., van Gunsteren, W. F., & Hermans, J. 1981
Interaction models for water in relation to protein hydration. Intermolecular
Forces, vol. 3 (ed. B. Pullman Reidel, Dordrecht. The Netherlands), pp. 331–
342.
Berman, A. S. 1953 Laminar flow in channels with porous walls. J. Appl. Phys. 24,
1232–1235.
Bernoulli, D. 1738 Hydrodynamics. Strasbourg: Bernouli. English Translation Dover,
New York.
517 Bibliography

Bhattacharjee, S., & Elimelech, M. 1997 Surface element integration: A novel tech-
nique for evaluation of DLVO interaction between a particle and a flat plate. J.
Colloid Interface Sci. 193, 273–285.
Bhattacharjee, S., Elimelechi, M., & Borkovec, M. 1998 DLVO interaction between
coloidal particles: Beyond Derjaguin’s approximation. Croatica Chem. Acta 71,
883–903.
Bhattacharyya, S., & Conlisk, A. T. 2005 Electroosmotic flow in two-dimensional
charged micro- and nanochannels. J. Fluid Mech. 540, 247–267.
Bhushan, B., ed. 2007 Springer Handbook of Nanotechnology, 2nd ed. New York:
Springer.
Bianchi, F., Wagner, F., Hoffmann, P., & Girault, H. H. 1993 Electroosmotic flow
in composite microchannels and implications in microcapillary electrophoresis
systems. Science 261, 895–897.
Bird, G. A. 1994 Molecular Gas Dynamics. Oxford, UK: Clarendon Press.
Bird, R. B., Stewart, W. E., & Lightfoot, E. N. 2002 Transport Phenomena, 2nd ed.
New York: John Wiley.
Blasius, H. 1908 Grenzschichten in flussigkeiten mit kleiner reibung. Z. Math. Phys.
56, 1–37.
Bockris, J. O’M., & Reddy, A. K. N. 1998 Modern Electrochemistry, Volume 1 Ionics,
2nd ed. New York: Plenum Press.
Bohn, P. 2009 Nanoscale control and manipulation of molecular transportin chemical
analysis. Ann. Rev. Anal. Chem. 2, 279–296.
Booth, F. 1950 The cataphoresis of spherical, solid non-conducting particles in a
symmetrical electrolyte. Proc. R. Soc. London A 203, 514–533.
Braha, O., Gu, Li-Qun, Zhou, Li, Lu, Xiaofeng, Cheley, S., & Bayley, H. 2000a
Simultaneous stochastic sensing of divalent metal ions. Nat. Biotechnol. 18, 1005–
1007.
Brecht, A., & Gauglitz, G. 1997 Recent developments in optical transducers for
chemical or biochemical applications. Sensors Actuators B Chem. 38, 1–7.
Bretherton, F. P. 1961 The motion of long bubbles in tubes. J. Fluid Mech. 10,
166–188.
Breuer, K., ed. 2005 Microscale Diagnostic Techniques. Berlin: Springer.
Brown, G. M. 1960 Heat or mass transfer in a fluid in laminar flow in a circular or
flat conduit. A I Che J. 6, 179–183.
Bruus, H. 2008 Theoretical Microfluidics. New York: Oxford University Press.
Burgeen, D., & Nakache, F. R. 1964 Electrokinetic flow in ultrafine capillary slits. J.
Phys. Chem. 68, 1084–1091.
Butler, J. N. 1998 Ionic Equilibrium: Solubility and pH Calculations. New York: John
Wiley.
Castellan, G. W. 1983 Physical Chemistry, 3rd ed. Menlo Park, CA: Benjamin Cum-
mings.
Castiglione, P., Falcioni, M., Lesne, A., & Vulpiani, A. 2008 Chaos and Coarse
Grainning in Statistical Mechanics. Cambridge, UK: Cambridge University Press.
518 Bibliography

Chang, H., Kosari, F., Andreadakis, G., Alam, M. A., Vasmatzis, G., & Bashir, R.
2004 DNA-mediated fluctuations in ionic current through silicon oxide nanopore
channels. Nanoletters 4, 1551–1556.
Chang, H.-C., & Yeo, L. Y. 2010 Electrokinetically-Driven Microfluidics and Nanoflu-
idics. Cambridge: Cambridge University Press.
Chang, R. 2000 Physical Chemistry for the Chemical and Biological Sciences. Sausal-
ito, CA: University Science Books.
Chapman, D. L. 1913 A contribution to the theory of electrocapillarity. Philos. mag.
25, 475–481.
Chen, D. P., Lear, J., & Eisenberg, R. 1997 Permeation through an open channel:
Poisson–Nernst–Planck theory of a synthetic ion channel. Biophys. J. 72, 97–116.
Chen, L., & Conlisk, A. T. 2008 Electroosmotic flow and particle transport in
micro/nano nozzles and diffusers. Biomed. Microdevices 10, 289–298.
Chen, L., & Conlisk, A. T. 2009 Effect of nonuniform surface potential on electroos-
motic flow at large applied electric field strength. Biomedical Microdevices 11,
251–258.
Chen, L. & Conlisk, A. T. 2010 DNA translocation phenomena in nanopores, Biomed-
ical Microdevices, 12, 235–245.
Chen, P., Gu, J., Brandin, E., Kim, Y.-R., Wang, Q., & Branton, D. 2004 Probing
single DNA molecule transport using fabricated nanopores. Nano Letters 4, 2293–
2298.
Chen, R.-Y. 1973 Flow in the entrance region at low Reynolds numbers. J. Fluids
Eng. 95, 153–158.
Chu, K. T., & Bazant, M. Z. 2005 Electrochemical thin films at and above the limiting
current. SIAM J. Appl. Math. 65, 1485–1505.
Churchill, R. V. 1969 Fourier series and boundary value problems McGraw-Hill, 2.
Ciofalo, M., Collins, M. W., & Hennessy, T. R. 1999 Nanoscale Fluid Dynamics in
Physiological Process: A Review Study. Southampton, UK: WIT Press.
Condon, E. U., & Morse, P. M. 1929 Quantum Mechanics. New York: McGraw-Hill.
Conlisk, A. T., Guezennec, Y. G., & Elliott, G. S. 1989 Chaotic motion of an array
of vortices above a flat wall. Phys. Fluids A 1, 704–717.
Conlisk, A. T., Datta, S., Fissell, W. H., & Roy, S. 2009 Biomolecular transport
through hemofiltration membranes. Ann. Biomed. Eng. 37(4), 732–746.
Constant, F. W. 1958 Theoretical Physics. Reading, MA: Addison-Wesley.
Conway, B. E. 1981 Ionic Hydration in Chemistry and Biophysics. New York: Elsevier.
Crick, F. H. C., & Watson, J. D. 1953 Molecular structure of nucleic acids. Nature
171, 737–738.
Cui, S. T. 2004 Molecular dynamics study of single-stranded DNA in aqueous solution
confined in a nanopore. Molecular Phys. 102, 139–146.
Currie, I. G. 2003 Fundamental Mechanics of Fluids, 3rd ed. New York: Marcel-
Dekker.
Cussler, E. L. 1997 Diffusion: Mass Transfer in Fluid Systems, 2nd ed. Cambridge:
Cambridge University Press.
519 Bibliography

Czarske, J., Buttner, L., Razik, T., & Muller, H. 2002 Boundary layer velocity mea-
surements by a laser Doppler profile sensor with micrometre spatial resolution.
Measure Sci. Technol. 13, 1979–1989.
Datta, S. & Ghosal, S. 2008 Dispersion due to wall interactions in microfluidic
separation systems. Phys. Fluids 20, 012103–1–012103–14.
Daune, M. 1993 Molecular Biophysics: Structures in Motion. Oxford: Oxford Uni-
versity Press.
Dawson, T. H. 1976 Theory and Practice of Solid Mechanics. New York: Plenum
Press.
Day, M. A. 1990 The no-slip condition of fluid mechanics. Erkenntis 33, 285–
296.
Deamer, D. W., & Akeson, M. 2000 Nanopores and nucleic acids: Prospects for
ultrarapid sequencing. Tibtech 18, 147–151.
Debye, P., & Hückel, E. 1923 The interionic attraction theory of deviations from ideal
behavior in solution. Z. Phys. 24, 185–206.
Dechadilok, P., & Deen, W. M. 2006 Hindrance factors for diffusion and convection
in pores. Ind. Eng. Chem. Res. 45, 6953–6959.
Deen, W. M. 1987 Hindered transport of large molecules in liquid-filled pores. AIChE
J. 33, 1409–1425.
Deen, W. M., Lazzara, M. J., & Myers, B. D. 2001 Structural determinants of glomeru-
lar permeability. Am. J. Physiol. Renal Physiol. 281, 579–596.
Denbigh, K. 1971 Principles of Chemical Equilibrium, 3rd ed. Cambridge: Cam-
bridge University Press.
Derjaguin, B. V. 1934 Friction and adhesion IV: Theory of adhesion of small particles.
Kolloid Z. 69, 155–164.
Derjaguin, B. V., & Landau, L. D. 1941 Theory of the stability of strongly charged
lyophobic colloids and the adhesion of strongly charged particles in solutions of
electrolytes. Acta Physicochim. 14, 633–662.
Desai, T., & Bhatia, S., ed. 2006 BioMEMS and Biomedical Nanotechnology: Volume
III Therapeutic Micro/Nano Technology. New York: Springer.
Devasenathipathy, S., & Santiago, J. G. 2005 Electrokinetic flow diagnostics. In
Microscale Diagnostic Techniques (ed. Kenny Breuer), pp. 113–154. Berlin:
Springer.
Devasenathipathy, S., Santiago, J. G., & Takehara, K. 1998 Particle tracking tech-
niques for electrokinetic microchannel flows. Exp. Fluids 25, 316–319.
D’Orazio, P. 2003 Biosensors in clinical chemistry. Clin. Chim. Acta 334, 41–69.
Drazin, P. G., & Riley, N. 2006 The Navier–Stokes Equations: A Classification of
Flows and Exact Solutions. Cambridge: Cambridge University Press.
Dukhin, S. S., & Derjaguin, B. V. 1974 Electrokinetic phenomena. In Surface and
Colloid Science vol. 7 (ed. E. Matijevic), pp. 1–351. John Wiley.
Eggins, B. R. 1996 Biosensors: An Introduction. New York: John Wiley.
Einstein, A. 1905a A new determination of molecular dimensions. PhD thesis, Uni-
versity of Zurich.
520 Bibliography

Einstein, A. 1905b On the motion of small particles suspended in liquids at rest

required by the molecular-kinetic theory of heat. Ann. Phys. 17, 549–560.
Einstein, A. 1956 Investigation on the Theory of the Brownian Movement, 4th ed.
New York: Dover.
Elimelech, M. 1998 Particle Deposition and Aggregation: Measurement, Modelling
and Simulation. Burlington, MA: Butterworth-Heinemann.
Erickson, D., & Li, D. 2001 Streaming potential and streaming current methods for
characterizing heterogeneous solid surfaces. J. Colloid Interface Sci. 237, 283–
289.
Erickson, D., & Li, D. 2002 Influence of surface heterogeneity on electrokinetically
driven microfluidic mixing. Langmuir 18, 1883–1892.
Erickson, D., Sinton, D., & Li, D. 2003 Joule heating and heat transfer in
poly(dimethylsiloxane) microfluidic systems. Lab on a Chip 3, 141–149.
Ethier, C. R., & Simmons, C. A. 2007 Introductory Biomechanics: From Cells to
Organisms. Cambridge: Cambridge University Press.
Evans, D. J., & Morriss, G. P. 1990 Statistical Mechanics of Nonequilibrium Liquids.
London: Academic Press.
Ewald, P. P. 1921 The calculation of optical and electrostatic grid potential. Ann. Phys.
(Leipzig) 64, 253–287.
Fan, R., Karnik, R., Yue, M., Li, D., Majumdar, A., & Yang, P. 2005 DNA translocation
in inorganic nanotubes. Nanoletters 5, 1633–1637.
Fausett, L. V. 2008 Applied Numerical Analysis Using Matlab, 2nd ed. Upper Saddle
River, NJ: Prentice Hall.
Fawcett, W. R. 2004 Liquids, Solutions, and Interfaces: From Classical Macroscopic
Descriptions to Modern Microscopic Details. Oxford: Oxford University Press.
Ferrari, M., ed. 2006 BioMEMS and Biomedical Nanotechnology. New York:
Springer.
Ferrer, M. L., Duchowicz, R., Carrasco, B., de la Torre, Jose G., & Acuna, A. U.
2001 The conformation of serum albumin in solution: A combined phospho-
resceence depolarization-hydrodynamic modeling study. Biophys. J. 80, 2422–
2430.
Feynman, R. P. 1961 There’s Plenty of Room at the Bottom. New York: Reinhold.
Fissell, W. H. 2006 Developments towards an artificial kidney. Expert Rev. Med.
Devices 3, 155–165.
Fissell, W. H., & Humes, H. D. 2006 Tissue engineering renal replacement therapy. In
Tissue Engineering and Artificial Organs, Section 5, chap. 60 (ed. J. D. Bronzino),
60, pp. 1–14. Boca Raton, FL: CRC Press.
Fissell, W. H., Manley, S., Dubnisheva, A., Glass, J., Magistrelli, J., Eldridge, A.,
Fleischman, A., Zydney, A., & Roy, S. 2007 Ficoll is not a rigid sphere. Am. J.
Physiol. Renal Physiol. 293, F1209–F1213.
Franks, F. 1972 Water: A Comprehensive Treatise, 7 vols. New York: Plenum Press.
Freifelder, D. 1987 Molecular Biology, 2nd edn. Boston: Jones and Bartlett.
Frenkel, D., & Smit, B. 2002 Understanding Molecular Simulations from Algorithms
to Applications, 2nd ed. San Diego, CA: Academic Press.
521 Bibliography

Freund, J. B. 2002 Electroosmosis in a nanometer scale channel studied by atomistic

simulation. J. Chem. Phys. 116, 2194–2200.
Friedman, M. H. 2008 Principles and Models of Biological Transport, 2nd ed. New
York: Springer.
Fung, Y. C. 1981 Biomechanics: Mechanical Properties of Living Tissues. New York:
Springer.
Fuoss, R. M., & Onsager, L. 1955 Conductance of strong electrolytes at finite dilu-
tions. Proc. Nat. Acad. Sci. U.S.A. 41, 274–283.
Gad-el Hak, M. 2001 The MEMS Handbook. Boca Raton, FL: CRC Press.
de Gennes, P. G., 2002 On fluid/wall slippage. Langmuir 18, 3413–3414.
Gibbs, J. W. 1961 The Scientific Papers of J. W. Gibbs. New York: Dover.
Giddings, J. C., Yang, F. J., & Myers, M. N. 1976 Flow-field-flow fractionation: A
versatile new separation method. Science 193, 1244–1245.
Gilat, A., & Subramaniam, V. 2008 Numerical Methods for Scientists and Engineers.
New York: John Wiley.
Gillespie, D. T. 1970 A Quantum Mechanics Primer. Scranton, PA: International
Textbook.
Gillespie, D. 1999 A singular perturbation analysis of the Poisson–Nernst–Planck
system: Applications to ionic channels. PhD thesis, Rush Medical School,
Chicago.
Gillespie, D., & Eisenberg, R. S. 2001 Modified Donnan potentials for ion transport
through biological ion channels. Phys. Rev. E. 63, 061902-1–06192-8.
Glazer, A. N., & Nikaido, H. 2007 Microbial Biotechnology: Fundamentals of Applied
Microbiology, 2nd ed. Cambridge: Cambridge University Press.
Golden, J. P., Floyd-Smith, T. M., Mott, D. R., & Ligler, F. S. 2007 Target delivery in
a microfluidic immunosensor. Biosensors Bioelectr. 22, 2763–2767.
Goldstein, S. 1965a Modern Developments in Fluid Dynamics Volume I. New York:
Dover.
Goldstein, S. 1965b Modern Developments in Fluid Dynamics Volume II. New York:
Dover.
Gong, M., Kim, B. Y., Flachsbart, B. R., Shannon, M. A., Bohn, P. W., & Sweedler,
J. V. 2008 An on-chip fluorogenic enzyme assay using a multilayer microchip
interconnected with a nanocapillary array membrane. IEEE Sensors J. 8, 601–607.
Gouy, G. 1910 About the electric charge on the surface of an electrolyte. J. Phys. A
9, 457–468.
Grahame, D. C. 1953 Diffuse double layer theory for electrolytes of unsymmetrical
valence types. J. Chem. Phys. 21, 1054–1060.
Granicka, L. H., Kawiak, J., Snochowski, M., Wojcicki, J. M., Sabalinska, S., &
Werynski, A. 2003 Polypropylene hollow fiber for cells isolation: Methods for
evaluation of diffusive transport and quality of cells encapsulation. Artificial Cells
Blood Substitutes Biotechnol. 31, 249–262.
Green, N. M. 1970 Spectrophotometric determination of avidin and biotin. Methods
Enzymol. 18, 418–424.
Gribbin, J. 1997 Richard Feynman: A Life in Science. New York: Dutton.
522 Bibliography

Griffiths, S. K., & Nilson, R. H. 1999 Hydrodynamic dispersion of a neutral nonre-

acting solute in electroosmotic flow. Anal. Chem. 71, 5522–5529.
Griffiths, S. K., & Nilson, R. H. 2006 Charged species transport, separation, and
dispersion in nanoscale channels: Autogenous electric field-flow fractionation.
Anal. Chem. 78, 8134–8141.
Guo, L. J. 2004 Recent progress in nanoimprint technology and its applications. J.
Appl. Phys. D: 37, R123–R141.
Hairer, E., Lubich, C., & Wanner, G. 2006 Geometric Numerical Integration:
Structure Preserving Algorithms for Ordinary Differential Equations, 2nd ed.
Heidelberg, Germany: Springer.
Hardy, R. C., & Cottingham, R. L. 1949 Viscosity of deuterium oxide and water in
the range 5◦ C to 1250◦ C. J. Res. Na. Bur. Standards 42, 573–578.
Haynes, W. M., ed. 2011–2012 Handbook of Chemistry and Physics, 92nd ed. Cleve-
land, Ohio: CRC Press.
Helmholtz, H. L. F. 1897 Uber den einflu der elektrischengrenzschichten bei gal-
vanischer spannung und der durch wasserstromung erzeugten potentialdiffernz.
Ann. Physik. 7, 337–387.
Henry, D. C. 1931 The cataphoresis of suspended particles, Part I. The equation of
cataphoresis. Proc. R. Soc. London A 133, 106–129.
Heyes, D. M. 1998 The Liquid State: Applications of Molecular Simulations. Chich-
ester, UK: John Wiley.
Hille, B. 2001 Ion Channels of Excitable Membranes, 3rd ed. Sunderland, MA:
Sinauer Associates.
Hill, T. L. 1963 Thermodynamics of Small Systems, Part I. New York: W.A. Benjamin.
Hill, T. L. 1964 Thermodynamics of Small Systems, Part II. New York: W.A. Ben-
jamin.
Hinchcliffe, A. 2003 Molecular Modeling for Beginners. John Wiley.
Hollerbach, U., Chen, D. P., & Eisenberg, R. 2001 Two- and three-dimensional
Poisson–Nernst–Planck simulations of current flow through gramicidin a. J. Sci.
Comput. 16, 373–409.
Holmes, M. H. 1995 Introduction to Perturbation Methods, 2nd ed. New York:
Springer.
Homola, J., Yee, S. S., & Gauglitz, G. 1999 Surface plasmon resonance sensors:
Review. Sensors Actuators BC 54, 3–15.
Honig, C. D. F., & Ducker, W. A. 2007 No-slip hydrodynamic boundary condition
for hydrophilic particles. Phys. Rev. Lett. 98, 053101.
Howison, S. 2005 Practical Applied Mathematics. Cambridge: Cambridge University
Press.
Hughes, W. F., & Gaylord, E. W. 1964 Basic Equations of Engineering Science. New
York: Schaum.
Humbert, N., Zocchi, A., & Ward, T. R. 2005 Electrophoretic behavior of streptavidin
complexed to a biotinylated probe: A functional screening assay for biotin-binding
proteins. Electrophoresis 26, 47–52.
523 Bibliography

Humes, H. D., Fissell, W. H., & Tiranathanagul, K. 2006 The future of hemodialysis
membranes. Kidney Int. 69, 1115–1119.
Hunter, R. J. 1981 Zeta Potential in Colloid Science. London: Academic Press.
Hur, J. S., Shaqfeh, E. S. G., & Larson, R. G. 2000 Brownian dynamics simulations
of single DNA molecules in shear flow. J. Rheol. 44, 713–742.
Icenhower, J. P., & Dove, P. M. 2000 Water behavior at silica surfaces. In Adsorption
on Silica Surfaces (ed. Eugene Papirer), pp. 277–295. New York: Marcel-Dekker.
Iler, R. K. 1979 The Chemistry of Silica. New York: John Wiley.
Incropera, F. P., & Dewitt, D. P. 1990 Fundamentals of Heat and Mass Transfer, 3rd
ed. New York: John Wiley.
Ishido, T., & Mizutani, H. 1981 Experimental and theoretical basis of electrokinetic
phenomena in rock-water systems and its application to geophysics. J. Geophys.
Res. 86(83), 1763–1775.
Ishijima, A., & Yanagida, T. 2001 Single molecule nanoscience. Trends Biochem.
Sci. 26, 438–444.
Israelachvili, J. 1992 Intermolecular and Surface Forces, 2nd ed. London: Academic
Press.
James, R. O. 1981 Surface ionization and complexation at the colloidl/aqueous elec-
trolyte interface. In Adsorption of Inorganics at Solid–Liquid Interfaces (ed. M. A.
Anderson & A. J. Rubins), pp. 219–261, chap. 6. Ann Arbor, MI: Ann Arbor
Science.
Jianrong, C., Yuqing, M., Nongyue, H., Xiaohua, W., & Sijiao, L. 2004 Nanotech-
nology and biosensors. Biotechnol. Adv. 22, 505–518.
Jorgensen, P. L. 1990 Structure and molecular mechanism of Na, k-pump. In Monova-
lent Cations in Biological Systems (ed. Charles Alexander Pasternak), pp. 117–154.
Boca Raton, FL: CRC Press.
Jorgensen, W. L., Chandrasekhar, J., Madura, J. D., Impey, R. W., & Klein, M. L.
1983 Comparison of simple potential functions for simulating liquid water. J.
Chem. Phys. 79, 926–935.
Judy, J., Maynes, D., & Webb, B. W. 2002 Characterization of frictional pressure
drop for liquid flows through microchannels. Int. J. Heat Mass Transfer 45, 3477–
3489.
Karniadakis, G., Beskok, A., & Aluru, N. 2005 Microflows and Nanoflows. New
York: Springer.
Kasianowicz, J. J., Brandin, E., Branton, D., & Deamer, D. W. 1996 Characterization
of individual polynucleotide molecules using a membrane channel. Proc. Natl.
Acad. Sci. U.S.A. 93, 13770–13773.
Kays, W. M., & Crawford, M. E. 1980 Convective Heat and Mass Transfer, 2nd ed.
New York: Mcgraw-Hill.
Keilland, J. 1937 Individual activity coefficients of ions in aqueous solutions. J. Am.
Chem. Soc. 59, 1675–1678.
Kemery, P. J., Steehler, J. K., & Bohn, P. W. 1998 Electric field mediated transport in
nanometer diameter channels. Langmuir 14, 2884–2889.
524 Bibliography

Kestin, J. 1978 Thermal conductivity of water and steam. Mech. Eng. Mag. August,
47.
Kevorkian, J., & Cole, J. D. 1981 Perturbation Methods in Applied Mathematics. New
York: Springer.
Kevorkian, J., & Cole, Julian D. 1996 Multiple Scale and Singular Perturbation
Methods. New York: Springer.
Kirby, B. J. 2010 Micro- and Nanoscale Fluid Mechanics: Transport in Microfluidic
Devices. Cambridge, UK: Cambridge University Press.
Kirby, B. J., & Hasselbrink, E. F. 2004a Zeta potential of microfluidic substrates: 1.
Theory, experimental techniques, and effects on separations. Electrophoresis 25,
187–202.
Kirby, B. J., & Hasselbrink, E. F. 2004b Zeta potential of microfluidic substrates: 2.
Data for polymers. Electrophoresis 25, 203–213.
Kirby, B. J., & Hasselbrink, E. F. 2004c Zeta potential of microfluidic substrates: 2.
Data for polymers. Electrophoresis 25, 203–213.
Kjelstrup, S., & Bedeaux, D. 2008 Non-Equilibrium Thermodynamics of Heteroge-
nous systems. New Jersey: World Scientific.
Knox, J. H., & McCormack, K. A. 1994 Temperature effects in capillary elec-
trophoresis. 1: Internal capillary temperature and effect upon performance. Chro-
matographia 38, 215–221.
Koltun, W. L. 1965 Precision space-filling atomic models. Biopolymers 3, 665–
679.
Koplik, J., Banavar, J. R., & Willemson, J. F. 1989 Molecular dynamics of fluid flow
at solid surfaces. Phys. Fluids A 1, 781–794.
Kress-Rogers, E., ed. 1997 Handbook of Biosensors and Electronic Noses: Medicine,
Food and the Environment. Boca Raton, FL: CRC Press.
Kuo, T.-C., Jr, Cannon, D. M., Jr, Shannon, M. A., Bohn, P. W. & Sweedler, J. V. 2003
Hybrid three-dimensional nanofluidic/microfluidic devices using molecular gates.
Sensors and Actuators A 102, 223–233.
Lamb, S. H. 1945 Hydrodynamics, 6th ed. New York: Dover.
Landau, L. D., & Levich, B. V. G. 1942 Dragging of a liquid by a moving plate. Acta
Physicochim. 17, 42–54.
Landers, J. P., ed. 1994 Handbook of Capillary Electrophoresis. Boca Raton, FL:
CRC Press.
Langhaar, H. L. 1942 Steady flow in the transition length of a straight tube. J. Appl.
Mech. 9, 55–58.
Latini, G., Grifoni, R. C., & Passerini, G. 2006 Transport Properties of Organic
Liquids. Southhampton, UK: WIT Press.
Lauga, E., Brenner, M. P., & Stone, H. A. 2005 Microfluidics: The no-slip condition.
In Handbook of Experimental Fluid Mechanics (ed. J. Foss & A. Yarin), pp. 1219–
1240. New York: Springer.
Leach, A. R. 1996 Molecular Modeling: Principles and Applications. Essex, UK:
Longman.
525 Bibliography

Leal, L. G. 2007 Advanced Transport Phenomena. New York: Cambridge University

Press.
Leca-Bouvier, B., & Blum, L. J. 2005 Biosensors for protein detection: A review.
Anal. Lett. 38, 1491–1517.
Lee, A. P., & Lee, L. James, eds. 2006 BioMEMS and Biomedical Nanotechnology:
Volume I Biological and Biomedical Nanotechnology. New York: Springer.
Lee, L. J. 2006 Nanoscale polymer fabrication for biomedical applications. In
BioMEMS and Biomedical Nanotechnology: Volume I Biological and Biomedi-
cal Nanotechnology (ed. Abraham P. Lee & L. James Lee), pp. 51–96. New York:
Springer.
Lee, M. L., Yang, F. J., & Bartle, K. D. 1984 Open Tubular Column Gas Chromatog-
raphy: Theory and Practice. New York: John Wiley.
Lee, P.-S., & Garimella, S. V. 2006 Thermally developing flow and heat transfer in
rectangular microchannels of different aspect ratio. Int. J. Heat Mass Transfer 49,
3060–3067.
Lehnert, T., Gijs, M., Netzer, R., & Bischoff, U. 2002 Realization of hollow SiO2
micronozzles for electrical measurements on living cells. Appl. Phys. Lett. 81,
5063–5065.
Lempert, W. R., Magee, K., Ronney, P., Gee, K. R., & Haugland, R. P. 1995 Flow
tagging velocimetry in incompressible flow using photo-activated nonintrusive
tracking of molecular motion. Exp. Fluids 18, 249–257.
Levich, V. G., & Krylov, V. S. 1969 Surface-tension driven phenomena. Ann. Rev.
Fluid Mech. 1, 293–316.
Levin, Y., & Flores-Mena, J. E. 2001 Surface tension of strong electrolytes. Europhys.
Lett. 56, 187–192.
Li, D. 2004 Electrokinetics in Microfluidics. Amsterdam: Elsevier.
Li, J., Gershow, M., Stein, D., Brandin, E., & Golovchenko, J. A. 2003 DNA molecules
and configurations in a solidstate nanopore microscope. Nat. Mater. 2, 611–
615.
Li, Z., & Liu, B. C.-Y. 2001 A molecular model for representing surface tension for
polar liquids. Chem. Eng. Sci. 56, 6977–6987.
Liou, W. K., & Fang, Y. 2006 Microfluid Mechanics: Principles and Modeling. New
York: McGraw-Hill.
Luginbuhl, P., Indermuhle, P.-F., Gretillat, M.-A., Willemin, F., de Rooij, N. F., Gerber,
D., Gervasio, G., Vuilleumier, J. -L., Twerenbold, D., Dugelin, M., Mathys, D., &
Guggenheim, R. 2000 Femtoliter injector for DNA mass spectrometry. Sensors
Actuators B 63, 167–177.
Malsch, N. H., ed. 2005 Biomedical Nanotechnology. Boca Raton, FL: Taylor and
Francis.
March, H. W., & Weaver, W. 1928 The diffusion problem for a solid in contact with
a stirred liquid. Phys. Rev. 31, 1072–1082.
Mark, P., & Nilsson, L. 2001 Structure and dynamics of the TIP3P, SPC, and SPC/E
water models at 298 K. J. Phys. Chem. A 105, 9954–9960.
526 Bibliography

Martin, F., Walczak, R., Boiarski, A., Cohen, M., West, T., Cosentino, C., & Ferrari,
M. 2005 Tailoring width of microfabricated nanochannels to solute size can be
used to control diffusion kinetics. J. Controlled Release 102, 123–133.
Martini, F. 2001 Fundamentals of Anatomy and Physiology, 5th ed. Prentice Hall.
Masliyah, J. H., & Bhattacharjee, S. 2006 Electrokinetic and Colloid Transport Phe-
nomena. Hoboken, NJ: John Wiley.
Maxwell, J. C. 1847 On Faraday’s lines of force. Trans. Cambridge Philos. Soc. 10,
27–83.
McCammon, J. A., & Harvey, S. C. 1987 Dynamics of Proteins ansd Nucleic Acids.
Cambridge: Cambridge University Press.
McNaught, A. D., & Wilkinson, A. 1997 Compendium of Chemical Terminology
(Gold Book), Malden: Blackwell.
Meagher, R. J., Light, Y. K., & Singh, A. K. 2008 Rapid, continuous purification
of proteins in a microfluidic device using genetically-engineered partition tags.
Lab-on-a-Chip 8, 527–532.
Meller, A., Nivon, L., & Branton, D. 2001 Voltage-driven DNA translocations through
a nanopore. Phys. Rev. Lett. 86, 3435–3438.
Moran, M. J., & Shapiro, H. N. 2007 Fundamentals of Engineering Thermodynamics,
6th ed. New York: John Wiley.
Moran, M. J., Shapiro, H. N., Munson, B. R., & Dewitt, D. P. 2003 Introduction to
Thermal Systems Engineering. New York: John Wiley.
Mott, D. R., Howell, P. B., Golden, J. P., Kaplan, C. R., Ligler, F. S., & Oran, E. S. 2006
Toolbox for the design of optimized microfluidic components. Lab-on-a-Chip 6,
540–549.
Mott, D. R., Howell, P. B., Obenschain, K. S., & Oran, E. S. 2009 The numerical tool-
box: An approach for modeling and optimizing microfluidic components. Mech.
Res. Commun. 36, 104–109.
Munson, B. R., Young, D. F., & Okiishi, T. H. 2005 Fundamentals of Fluid Mechanics,
2006th ed. New York: John Wiley.
Murray, J. D. 2001 Mathematical Biology I: An Introduction, 3rd ed. New York:
Springer.
Murray, J. D. 2003 Mathematical Biology II: Spatial Models and Biological Appli-
cations, 3rd ed. New York: Springer.
Murrell, J. N., & Jenkins, A. D. 1982 Properties of Liquids and Solutions, 2nd ed.
Chichester, UK: John Wiley.
Nayfeh, A. H. 1973 Perturbation Methods. New York: John Wiley.
Newman, J. S. 1972 Electrochemical Systems. Englewood Cliffs, NJ: Prentice Hall.
Nguyen, N. T., & Wereley, S. T. 2002 Fundamentals and Applications of Microfluidics.
Norwood, MA: Artech House.
Oberkampf, W. L., & Blottner, F. G. 1998 Issues in computational fluid dynamics
code verification and validation. AIAA J. 36, 687–695.
Oberkampf, W. L., Sindir, M. M., & Conlisk, A. T. 1998 G-077-1998 guide for the
verification and validation of computational fluid dynamics simulations. Tech. Rep.
American Institute of Aeronautics and Astronautics.
527 Bibliography

O’Brien, R. W., & White, L. R. 1978 Electrophoretic mobility of a spherical colloidal

particle. J. Chem. Soc. Faraday Trans. II 74, 1607–1626.
Ohshima, H., Healy, T. W., White, L. R., & O’Brien, R. 1984 Sedimentation veloc-
ity and potential in a dilute suspension of charged spherical colloidal particles.
J. Chem. Soc. Faraday Trans. II 80, 1299–1317.
Onsager, L., & Samaras, N. N. T. 1934 The surface tension of debye-huckel elec-
trolytes. J. Chem. Phys. 2, 528–536.
Oseen, C. W. 1910 Uber die sstokes’sche formel und uber eine verwandte aufgabein
der hydrodynamik. Ark. Math. Astron. Fys. 6.
Overbeek, J. TH. G. 1943 Theory of the relaxation effect in electrophoresis. Kolloide
Beihefte 54, 287–364.
Oyanader, M., & Arce, P. 2005 A new and simpler approach for the solution of the
electrostatic potential differential equation: Enhanced solution for planar, cylindri-
cal and annular geometries. J. Colloid Interface Sci. 284, 315–322.
Ozkan, M., & Heller, M. J., ed. 2006 BioMEMS and Biomedical Nanotechnology: Vol-
ume II Micro/Nano Technology for Genomics and Proteomics. New York: Springer.
Papirer, E., ed. 2000 Adsorption on Silica Surfaces. New York: Marcel Dekker.
Persello, J. 2000 Surface and interface structure of silica. In Adsorption on Silica
Surfaces (ed. Eugene Papirer), pp. 297–342. New York: Marcel-Dekker.
Peters, T. 1996 All About Albumin: Biochemistry, Genetics and Medical Applications,
3rd ed. San Diego, CA: Academic Press.
Pinkus, O., & Sternlicht, B. 1961 Theory of Hydrodynamic Lubrication. New York:
McGraw-Hill.
Plawski, J. L. 2001 Transport Phenomena Fundamentals. New York: Marcel-Dekker.
Priestley, J. 1767 The History and Present State of Electricity. London: Printed for
J. Dodsley, J. Johnson and T. Cadell.
Priezjev, N. V., & Troian, S. M. 2006 Influence of periodic wall roughness on slip
behavior at liquid/solid interfaces: Molecular scale simulations versus continuum
predictions. J. Fluid Mech. 554, 25–48.
Priezjev, N. V., Darhuber, A. A., & Troian, S. M. 2005 Slip behavior in liquid films on
surfaces of patterned wettability: Comparison between continuum and molecular
dynamics simulations. Phys. Rev. E 71, 41608.
Probstein, R. F. 1989 Physicochemical Hydrodynamics. Boston: Butterworths.
Proudman, L., & Pearson, J. R. A. 1957 Expansions at small Reynolds numbers for
the flow past a sphere and a circular cylinder. J. Fluid Mech. 2, 237–262.
Qiao, R., & Aluru, N. R. 2003a Atypical dependence of electroosmotic transport on
surface charge in a single-wall carbon nanotube. Nano Lett. 3, 1013–1017.
Qiao, R., & Aluru, N. R. 2003b Ion concentrations and velocity profiles in nanochan-
nel electroosmotic flows. J. Chem. Phys. 118, 4692–4701.
Quinke, G. 1859 Ueber eine neue Art elekrischer Ströme. Prog. Ann. 107, 1–47.
Rahman, A., & Stillinger, F. H. 1971 Molecular dynamics study of liquid water. J.
Chem. Phys. 55, 3336–3359.
Rapaport, D. C. 2004 The Art of Molecular Simulation, 2nd ed. Cambridge: Cam-
bridge University Press.
528 Bibliography

Ravindra, N. M., Prodan, C., Fnu, S., Padroni, I., & Sikha, S. K. 2007 Advances in
the manufacturing, types and applications of biosensors. JOM 59, 37–43.
Raymond, K. W. 2007 General, Organic and Biological Chemistry: An Integrated
Approach, 2nd ed. New York: John Wiley.
Reid, R. C., Prausnitz, J. M., & Poling, B. E. 1987 The Properties of Gases and
Liquids, 4th ed. New York: McGraw-Hill.
Revil, A., Pezard, P. A., & Glover, P. W. J. 1999 Streaming potential in porous media
1. Theory of the zeta potential. J. Geophys. Res. 104, 20021–20032.
Rhee, H.-K., Aris, R., & Amundson, N. R. 1986 First-Order Partial Differential
Equations: Volume 1 Theory and Applications of Single Equations. Englewood
Cliffs, NJ: Prentice Hall.
Rhee, H.-K., Aris, R., & Amundson, N. R. 1989 First-Order Partial Differential
Equations: Volume 2 Theory and Applications of Hyperbolic Systems of Quasilin-
ear Equations. Mineola, NY: Dover.
Rice, S. A. 2000 Active control of molecular dynamics: Coherence versus chaos. J.
Stat. Phys. 101, 187–212.
Richardson, S. 1973 On the no-slip boundary condition. J. Fluid Mech. 59, 707–719.
Roache, P. J. 1998 Verification and Validation in Computational Science and Engi-
neering. Socorro, NM: Hermosa.
Roache, P. J., & Steinberg, S. 1994 Symbolic manipulation and computational fluid
dynamics. AIAA J. 22, 1390–1394.
Robinson, R. A., & Stokes, R. H. 1959 Electrolyte Solutions. New York: Academic
Press.
Roco, M. 2005 Converging technologies: Nanotechnology and medicine. In Biomed-
ical Nanotechnology (ed. Neelina H. Malsch). Boca Raton, FL: Taylor and Francis.
Rosi, N. L., & Mirkin, C. A. 2005 Nanostructures in biodiagnostics. Chem. Rev. 105,
1547–1562.
Roy, C. J., Nelson, C. C., Smith, T. M., & Ober, C. C. 2004 Verification of
Euler/Navier–Stokes codes using the method of manufactured solutions. Int. J.
Numer. Methods Fluids 44, 599–620.
Russel, W. B., Saville, D. A., & Schowalter, W. R. 1991 Colloidal Dispersions.
Cambridge: Cambridge University Press.
Ryckaert, J.-P., Ciccotti, G., & Berendsen, H. J. C. 1977 Numerical integration of the
Cartesian equations of motion of a system with constraints: Molecular dynamics
of n-alkanes. J. Comput. Phys. 23, 327–341.
Sadr, R., Yoda, M., Zheng, Z., & Conlisk, A. T. 2004 An experimental study of
electro-osmotic flow in rectangular microchannels. J. Fluid Mech. 506, 357–
367.
Sadr, R., Yoda, M., Gnanaprakasam, P., & Conlisk, A. T. 2006 Velocity measurements
inside the diffuse electric double layer in electroosmotic flow. Appl. Phys. Lett. 89,
044103-1–044103-3.
Sadr, R., Hohenegger, C., Li, H., Mucha, P. J., & Yoda, M. 2007 Diffusion-induced
bias in near-wall velocimetry. J. Fluid Mech. 577, 443–456.
529 Bibliography

Sadus, R. J. 1997 Molecular Simulation of Liquids: Theory, Algorithms and Object-

Orientation. Amsterdam: Elsevier.
Sagui, C., & Darden, T. A. 1999 Molecular dynamics simulations of biomolecules:
Long range electrostatic effects. Annu. Rev. Biomolecular Structure, 28, 155–179.
Saleh, O. A., & Sohn, L. L. 2006 An On-Chip Artificial Pore for Molecular Sensing.
New York: Springer.
Saltzman, W. M. 2001 Drug Delivery: Engineering Principles for Drug Therapy.
Oxford: Oxford University Press.
Saltzman, W. M. 2009 Biomedical Engineering. Cambridge, UK: Cambridge Uni-
versity Press.
Scheller, F. W., Wollenberger, U., Warsinke, A., & Lisdat, F. 2001 Research and
development in biosensors. Curr. Opi. Biotechnol. 12, 35–40.
Scherrer, R., & Gerhardt, P. 1971 Molecular sieving by the Bacillus megaterium cell
wall and protoplast. J. Bacteriol. 107, 718–735.
Schnell, E. 1956 Slippage of water over nonwettable surfaces. J. Appl. Phys. 27,
1149–1152.
Sears, F. W., & Zemansky, M. W. 1964 University Physics, 3rd ed. Reading, MA:
Addison-Wesley.
Shapiro, A. P., & Probstein, R. F. 1993 Removal of contaminants from saturated clay
by electroosmosis. Environ. Sci. Technol. 27, 283–291.
Shaw, D. 1969 Electrophoresis. London: Academic Press.
Shereshefsky, J. L. 1967 A theory of surface tension of binary solutions I. Binary
liquid mixtures of organic compounds. J. Colloid Interface Sci. 24, 317–322.
Sinha, M. K., Roy, D., Gaze, D. C., Collinson, P. O., & Kaski, J.-C. 2004 Role of
ischemia modified albumin, a new biochemical marker of myocardial ischemia, in
the early diagnosis of acute coronary syndromes. Emerg. Med. J. 21, 29.
Smith, G. D. 1985 Numerical Solutions of Partial Differential Equations, 3rd ed.
Oxford: Oxford University Press.
Smoluchowski, M. 1918 Versuch einer mathematischen theorie der koagulation
kinetic kolloider losungen. Z. Phys. Chem 92, 129–135.
Snyder, L. R., & Kirkland, J. J. 1979 Introduction to Modern Liquid Chromatography.
New York: John Wiley.
Spichiger-Keller, U. E. 1998 Chemical Sensors and Biosensors for Medical and
Biological Applications. Wiley-VCH.
Spoel, D., van der, van Maaren, P. J., & Berendsen, H. J. C. 1998 A systematic study
of water models for molecular simulation: Derivation of water models optimized
for use with a reaction field. J. Chem. Phys. 108, 10220–10230.
Stachel, J., ed. 1998 Einstein’s Miraculous Year. Princeton, NJ: Princeton University
Press.
Staubli, T., Stæckli, T., Knapp, H. F., Alpnach, S., Lausanne, S., de Neuchtel, U.,
& Neuchtel, S. 2005 Fast immobilization of probe beads by dielectrophoresis-
controlled adhesion in a versatile microfluidic platform for affinity assay. Elec-
trophoresis 26, 3697–3705.
530 Bibliography

Stern, O. 1924 The theory of the electrolytic double layer. Z. Elektrochem. 30, 508–
516.
Stigter, D. 1980 Sedimentation of highly charged colloidal spheres. J. Phys. Chem.
84, 2758–2762.
Storm, A. J., Chen, J. H., Zandbergen, H. W., & Dekker, C. 2003 Fabrication
of solid-state nanopores with single-nanometre precision. Nat. Mater. 2, 537–
540.
Storm, A. J., Chen, J. H., Zandbergen, H. W., & Dekker, C. 2005 Translocation of
double-strand DNA through a silicon oxide nanopore. Phys. Rev. E. 71, 051903.
Störmer, C. 1907 Sur les trajectoires des corpuscules electrises. Arch. Sci. Phys. Nat.
Geneve 24, 5–18, 113–158, 221–247.
Stroock, A. D., Weck, D. M., Chiu, D. T., Huck, W. T. S., Kenis, P. J. A., Ismagilov,
R. F., & Whitesides, G. M. 2000 Patterning electro-osmotic flow with patterned
surface charge. Phys. Rev. Lett. 84, 3314–3317.
Styer, D. F. 1996 Common misconceptions regarding quantum mechanics. Am. J.
Phys. 64, 31–34.
Tabeling, P. 2005 Introduction to Microfluidics. Oxford: Oxford University Press.
Tahery, R., Modarress, H., & Satherly, J. 2005 Surface tension prediction and ther-
modynamic analysis of the surface for binary solutions. Chem. Eng. Sci. 60, 4935–
4952.
Taylor, G. I. 1953 Dispersion of soluble matter in solvent flowing slowly through a
tube. Proc. R. Soc. London A 219, 186–203.
Taylor, G. I. 1954 Conditions under which dispersion of a solute in a stream of solvent
can be used to measure molecular diffusion. Proc. R. Soc. London A 225, 473–
477.
Taylor, R., & Krishna, R. 1993 Multicomponent Mass Transfer. New York: John
Wiley.
Terrill, R. M. 1964 Laminar flow in a uniformly porous channel. Aeronaut. Q. XV,
297–299.
Terrill, R. M., & Shrestha, G. M. 1965 Laminar flow through parallel and uniformly
porous walls of different permeability. Z. Angewan. Math. Phys. 16, 470–482.
Terrill, R. M., & Thomas, P. W. 1969 On laminar flow through a uniformly porous
pipe. Appl. Sci. Res. 21, 37–67.
Thévenot, D. R., Toth, K., Durst, R. A., & Wilson, G. S. 2001 Electrochemical biosen-
sors: Recommended definitions and classification. Biosensors Bioelectronics 16,
121–131.
Thompson, P. A., & Troian, S. M. 1997 A general boundary condition for liquid flow
at solid surfaces. Nature 389, 360–362.
Thust, M., Schoning, M. J., Frohnhoff, S., & Arens-Fischer, R. 1996 Porous silicon
as a substrate material for potentiometric biosensors. Measure. Sci. Technol. 7,
26–29.
Tokaty, G. A. 1971 A History and Philosophy of Fluid Mechanics. New York: Dover.
Travis, K. P., & Gubbins, K. E. 2000 Poiseuille flow of Lennard–Jones fluids in
narrow slit pores. J. Chem. Phys. 112, 1984–1994.
531 Bibliography

Tuckerman, M. 2010 Statistical Mechanics: Theory and Simulation. Oxford: Oxford

University Press.
Turns, S. R., Thermal-Fluid Sciences: An Integrated Approach, Cambridge, UK:
Cambridge University Press, 2006.
Tyrrell, H. J. V., & Harris, K. R. 1984 Diffusion in Liquids: A Theoretical and
Experimental Study. London: Butterworth.
Van Dyke, M. 1975 Perturbation Methods in Fluid Mechanics, 2nd ed. Stanford, CA:
Parabolic Press.
Venema, P., Hiemstra, T., & van R., Willem H. 1996 Comparison of different site bind-
ing models for cation sorption: Description of pH dependency, salt dependency,
and cation-proton exchange. J. Colloid Interface Sci. 181, 45–49.
Venturoli, D., & Rippe, B. 2005 Ficoll and dextran vs. globular proteins as probes
for testing glomular permselectivity: Effects of molecular size, shape, charge and
deformability. A. J. Physiol. Renal Physiol. 288, 605–613.
Verlet, L. 1967 Computer experiments on classical fluids I. thermodynamical prop-
erties of Lennard–Jones molecules. Phys. Rev. 159, 98–103.
Verwey, E. J. W., & Overbeek, J. T. G. 1948 Theory of Stability of Lyophobic Colloids.
Amsterdam: Elsevier.
Vo-Dinh, T. 2006 Biosensors and biochips. In Biomolecular Sensing, Processing and
Analysis (ed. R. Bashir, Steve Wereley, & Mauro Ferrari), pp. 4–33. New York:
Springer.
Volkov, A. G., Paula, S., & Deamer, D. W. 1997 Two mechanisms of permeation of
small neutral molecules and hydrated ions across phospholipid bilayers. Bioelec-
trochem. Bioenergetics 42, 153–160.
Wang, S., Hu, Xin, & Lee, L. J. 2008 Electrokinetics induced asymmetric transport
in polymeric nanonozzles. Lab-on-a-Chip 8, 573–581.
Wang, S., Zeng, C., Lai, S., Juang, Y.-J., Yang, Y., & Lee, L. J. 2005 Polymer
nanonozzle array fabricated by sacrificial template imprinting. Adv. Mater. 17,
1182–1186.
Wang, Y., Bhushan, B., & Maali, A. 2009 Atomic force microscopy measurement of
boundary slip on hydrophilic, hydrophobic and superhydrophobic surfaces. J. Vac.
Sci. Technol. A 27, 1–7.
Wehausen, J. V., & Laitone, E. V. 1960 Surface waves. In Handbuch der Physik (ed.
E. Flugge), vol. IX, pp. 446–758. Berlin: Springer.
Wereley, S. T., & Meinhart, C. D. 2010 Recent advances in micro-particle image
velocimetry. In Ann. Rev. Fluid Mechanics, vol. 42, pp. 557–576. Palo Alto: Annual
Reviews.
White, F. M. 2003 Fluid Mechanics, 5th ed. New York: McGraw-Hill.
White, F. M. 2006 Viscous Fluid Flow, 3rd ed. New York: McGraw-Hill.
White, F. M., Barfield, B. F., & Goglia, M. J. 1958 Laminar flow in a uniformly porous
channel. J. Appl. Mech. 25, 613–617.
Wiersma, P. H., Loeb, A. L., & Overbeek, J. T. G. 1966 Calculation of the elec-
trophoretic mobility of a spherical colloid particle. J. Colloid Interface Sci. 22,
78–99.
532 Bibliography

Wilson, A. H. 1948 A diffusion problem in which the amount of diffusing substance

is finite. Philos. Maga. 54, 48–58.
Wilson, S. D. R. 1982 The drag-out problem in film coating theory. J. Eng. Math. 16,
209–221.
Xuan, Xiangchun, X., Bo, S., David, & Li, D. 2004 Electroosmotic flow with Joule
heating effects. Lab-on-a-Chip 4, 230–236.
Yoda, M. 2006 Nano-particle image velocimetry. In Biomolecular Sensing, Process-
ing and Analysis (ed. Rashid bashir & Steve Wereley), pp. 331–348. New York:
Springer.
Zaltzman, B., & Rubinstein, I. 2007 Electroosmotic slip and electroconvective insta-
bility. J. Fluid Mech. 579, 173–226.
Zeman, L. J., & Zydney, A. L. 1996 Microfiltration and Ultrafiltration: Principles
and Applications. New York: Marcel-Dekker.
Zhao, H., & Bau, H. H. 2007 On the effect of induced electroosmosis on a cylindrical
particle next to a wall. Langmuir 23, 4053–4063.
Zhu, W., Singer, S. J., Zheng, Z., & Conlisk, A. T. 2005 Electro-osmotic flow of a
model electrolyte. Phys. Rev. E 71, 041501.
 Index

anion, 24 eukaryotic cell, 298

aqueous solutions, 14 fatty acid, 300
artificial kidney, 484f Golgi apparatus, 299
centerline approximation, 487 hormones, 300
end-stage renal disease, 484 insulin, 300
hindered transport, 487 ion channels, 300
influence of electric field, 489 lipid bilayer, 298
nanopore membranes as, 485 membrane transport, 301f
sieving coefficient, 487 mitochondria, 299
atomic force microscope, 284 nucleus, 299
organic molecules, 299
Batchelor, George, 194 plasma membrane, 300
batteries, 270 prokaryotic, 298
Blasius boundary layer ribosomes, 299
similarity analysis, 153 role of testosterone, 300
singular perturbation problem, 153 steroid hormones, 300
Blasius boundary layer equations, 150 chemical and biochemical sensing, 491f, 492, 493
blood flow, 173f electrochemical sensors, 491
boundary conditions, 106f atomic force microscopy, 494
apparent slip, 108 biological recognition elements, 493
electrostatics, 114 biosensor performance, 494
at a free surface, 112 ELISA, 494
heat transfer, 116 immunoassay, 493
heat transfer convection coefficient, 116 low Reynolds number vortices, 495
mass transfer, 113 nanopores and nanopore membranes used as,
mass transfer convection coefficient, 113 496
no-slip, 109, 111 nucleic acid biosensors, 493
slip condition, 110, 147 optical detection methods, 494
slip length, 108, 110, 147 optical sensors, 491
surface tension effects, 113 surface plasmon resonance, 494
velocity, 107 chemical equilibrium, 236
boundary layer, 17 chemical potential, 236f
Brownian motion, 133, 196f chromatography
diffusion coefficient, 200 gas, 330
bubble in a tube, 166f liquid, 330
classification of partial differential equations,
carbon nanotubes, 494, 498 128f
Casson fluid, 173 characteristic curves, 129
cation, 24 elliptic equations, 128
cells, 298f hyperbolic equations, 128
cell death, 298 parabolic equations, 128
cell doctrine, 298 colloid science, 29
cytoskeleton, 298 colloidal systems
cytosol, 299 van der Waals forces in, 31
endoplasmic reticulum, 299 DLVO theory, 31

533
534 Index

continuity equation, 88 electrical permittivity, 54, 214

continuum approximation, 75 relative permittivity, 54
Couette flow, 45 electrochemical potential, 243f
Crick, Francis, 285 Donnan equilibrium, 244
Nernst equation, 244
Debye–Hückel approximation, 256 electrochemistry, 230f
Derjaguin approximation, 275 ζ -potential, 251
dielectric constant, 25 ζ -potential liquid side view, 261
dielectrics, 24 ζ -potential solid side view, 262
diffusion coefficient, 43, 48f acids and bases, 244
human serum albimun in serum, 293 buffer solution, 244
Stokes–Einstein equation, 48 charge polarization, 251
dimensional analysis, 117f covalent bond, 231
dimensionless parameters, 118 Debye length, 254
drag coefficient, 118 Derjaguin approximation, 276
dynamic similarity, 117 electrical double layer, 251
fully developed flow, 119 equilibrium constant, 245
DNA hydration of ions, 25, 31, 232
base pair, 286 hydrophilic molecules, 232
hydrogen bond in, 286 hydrophobic molecules, 232
role in synthesizing proteins, 286 neutral pH, 246
transcription, 286 polar molecule, 232
DNA transport, 476f Stern layer, 27, 251
and biochemical sensing, 478 surface element integration, 277
comparison with experiment, 483 electrokinetic phenomena, 306f
current calculation, 482 capillary electrophoresis, 331
direction and surface charge density, 477 discovery of electro-osmosis, 19
mathematical model, 479 electro-osmosis, 306
through α-hemolysin pores, 476 electro-osmotic flow, 306
translocation, 476, 481 electro-osmotic mobility, 311
drug delivery electromigration, 319
compartment models, 412 electrophoresis, 306, 331
pharmacokinetics, 412 electrophoretic mobility, 333
dynamic viscosity, 16, 45f electrophoretic velocity, 333
gel electrophoresis, 331
electric field, 214, 214f Joule heating, 342
Coulomb’s law, 214 relationship between electromigration and
due to a charged plate, 217 electrophoresis, 331
due to a charged wire, 216 sedimentation potential, 306, 341
Gauss’s law, 218 streaming current, 340f, 347
electrical conductivity, role in Joule heating, 343 streaming potential, 306, 338f
electrical conductors, 24 electrokinetic remediation, 332
electrical double layer, 26f, 231, 251f electrolyte solution, 14, 25f
anode, 28 current and current density, 320f
asymmetric multivalent mixtures, 259f electrical conductivity, 267
Boltzmann distribution, 254 as electrical conductors, 24
cathode, 28 Lewis acid defined, 26
on a cylinder, 265 Lewis base defined, 26
Debye length defined, 27 electro-osmotic flow, 307f
Debye–Hückel picture, 251 in an annulus, 318f
electroosmotic pump, 28 chromatography, 330
excluded volume effect, 253 current and current density, 316f
Gouy–Chapman picture, 252 dispersion in, 328f
Helmholtz picture, 251 electro-osmotic pump, 3
history, 20 field flow fractionation, 330
inner Helmholtz plane, 253 in nozzles and diffusers, 320f
ionic strength, 27, 72, 239, 256, 264, 268, 308, volume flow rate in, 9, 310
324, 345 electrophoresis, 331f
Stern layer, 27, 251 full numerical solution for the electrophoretic
triple layer model, 252 mobility, 336
electrical insulators, 24 Henry’s solution, 334
535 Index

electrostatics, 213f heat, mass, momentum, and electrical analogies,

Coulomb’s law, 24 123
current, 225 hydration of ions, 25
current density, 225 Albert Einstein’s contribution, 234
differential form of Gauss’s law, 223 hydrophilic surface, 51, 58, 108, 112
dipole moment, 221 hydrophobic surface, 58, 108, 112, 148
electric dipole, 221
electric field definition, 24 ideal solution, 239
electric potential, 220 ill-conditioned matrix, 396
electrical conductance, 225 initial value problems
electrical conductivity, 225 Euler forward formula, 414
electrical permittivity of a vacuum, 24 Euler methods, 413
electrical resistance, 225 modified Euler method, 414
Gauss’s law, 218 Taylor series method, 413
image charge, 228 interpolation
induced dipole, 222 cubic splines, 367
Maxwell’s equations, 226 difference table, 361
nonpolar molecules, 221 divided difference table, 363
Ohm’s law, 225 Lagrangian polynomial interpolation,
Poisson equation, 100 361
Poisson’s equation, 223 linear interpolation, 360
polar molecules, 221 Newton interpolation formulas, 363
surface charge density, 216 truncation error, 364
volume charge density, 216 ion channels, 7, 34, 301
energy equation gating of, 302
conduction heat transfer, 104 transmembrane potential, 302
derivation, 102f ions, 14
equation of charge conservation, 101 valence, 26
Euler’s equations, 16
experimental methods and Brownian motion, kinematics, 77
133 Eulerian description of fluid motion, 78
Lagrangian description of fluid motion,
Fick’s law, 43 78
history, 18 material derivative, 79
Fourier, Joseph, 17, 188 pathlines, 80
Fourier’s law, 42, 103 streaklines, 81
history, 17 stream function, 80
Franklin, Rosalind, 285 streamlines, 79

Gauss’s law, volume charge density in, 219 Lab-on-a-chip, 4

gels, 30 law of mass action, 412
Gibbs function, 230, 236, 236f lubrication, 160f
equilibrium constant and, 241 lubrication approximation, 162
role in equilibrium, 240 Reynolds equation, 164
Grahame’s equation, 261 lyophilic colloids, 30
lyophobic colloids, 30
Hamiltonian, 349, 419, 455
heat transfer, 180f MacLaurin series, 352
forced convection, 22 mass flow rate, 60
free convection, 22 mass spectrometry, 34
Graetz problem, 190f mass transfer
heat transfer coefficient, 186 chemical reaction, 22, 26
heat transfer resistance, 182 convection, 22
in a channel, 182 diffusion, 22
Joule heating, 183 electrical migration, 99
mean temperature, 186 entrance length, 188
Nusselt number, 182, 189, 191, mass averaged velocity, 98
206 mass density, 94
Peclet number, 189 mass fraction, 95
thermal entrance length, 184, 190 molar concentration, 94
thermally fully developed, 186 molar mass, 94
536 Index

mass transfer (cont.) Jacobi iteration, 391

molarity, 94 Michaelis-Menten problem, 425
mole fraction, 95 numerical differentiation, 373
Nernst–Planck equation, 99 Romberg integration, 384
species velocity, 98 round-off errors, 350
in a thin film, 192f Runge-Kutta methods, 415
matched asymptotic expansions, 313, 344 sets of linear equations, 386
micro total analysis systems, 4 Simpson’s rules for integration, 379
molecular bonding, 233f singular perturbation problem, 401
atoms, 233, 449 solution of linear boundary value problems,
covalent bond, 233 399
hydrogen bond, 232 solution of nonlinear ordinary differential
ionic bond, 233 equations, 403
valence, 233 stability of initial value solvers, 424
van der Waals force, 233 stiff initial value problems, 424
molecular dynamics, 447f successive overrelaxation, 392
and electro-osmotic flow, 465 symplectic integration methods, 420
Ewald sum, 460 Thomas algorithm, 396
Lennard Jones potential, 453 trapezoidal rule integration, 377
minimum image convention, 460 tri-diagonal linear equations, 386
periodic boundary conditions, 457 tri-diagonal systems of equations, 393
Poiseuille flow, 468 truncation error, 350
postprocessing, 465, 473 validation, 435
for protein structure and conformation, 283 verification, 435
molecular simulations, 447f numerical differentiation
ab initio molecular dynamics, 450, 453, 471 backward difference approximation,
Born–Oppenheimer approximation, 450 374
empirical potentials, 450, 452 forward difference approximation, 374
ensemble, 451 central difference approximation, 374
Schröedinger equation, 449 sloping difference formula, 375
numerical solution of partial differential
nanopore membranes, 4, 33, 140 equations, 430f
for diabetes treatment, 499 elliptic equations, 431
for nanofiltration, 499 parabolic equations, 430
for reverse osmosis, 499 numerical solution of the boundary layer
Navier–Stokes equations, 88f equations, 405
history, 16 numerical solution of the Poisson-Nernst-Planck
indicial notation, 90 equations, 428f
stress tensor, 90
surface and body stresses, 83f Ohm’s law, 101
vorticity, 86 osmotic pressure, 280
Newtonian fluid, 16, 87
no-slip condition, history, 17, 107, 148 parallel plate viscometer, 45
Non-Newtonian fluids, 126f pH, 26
nucleic acids, 285 pipe and channel flow, 141f
chromosomes, 285 entrance length, 142
DNA, 285 fully developed regime, 143
genes, 285 point-of-care systems, 492
RNA, 285 Poiseuille flow, 143f
numerical analysis, 348f history, 17
Adams–Moulton methods, 418 solution derived, 144
central difference approximation, 374 volume flow rate in, 10, 144
curve-fitting, 370 volume flowrate in, 144
definite integral, 377 Poiseuille flow analogy, 224
derivative boundary conditions, 407 Poisson equation, 100
Gauss Elimination, 386 Poisson–Boltzmann distribution, 255
Gauss-Seidel Iteration, 391 polymers, 249
ill conditioned matrix, 396 biopolymers, 30, 250
ill-conditioning and stability of linear systems dimer, 250
of equations, 396 monomer, 249
indefinite integral, 377 PDMS, 250
initial value problems, 411 PMMA, 250
537 Index

tetramers, 250 silicon, 246

trimers, 250 singular perturbation problem, 314
polysacchrides, 286 ssDNA, 285
glucose, 287 Stokes flow past a sphere, 169f
porous media, 162 drag coefficient, 172
protein synthesis, 288 history, 16
gene expression, 288 pressure, 171
gene role in, 288 Stokes–Einstein equation, 50, 171
proteins, 287f Stokes’ drag law, sedimentation, 172
amino acids role in structure, 290 streaming potential, history, 19
antibodies, 288, 295 Sturm–Liouville problem, 191
antigen, 288, 295 suction flows, 155f
apolipoproteinE, 294 developing, 158
binding site, 295 fully developed, 155
bioinformatics, 292 surface and body forces, 83f
C-terminus, 290 surface charge density
conformation, 290 liquid side view, 260f
denatured, 290 related to potential, 43f
enzymes, 288 solid side view, 248f, 262
function, 288 surface tension, 55, 55f
human insulin, 294 capillary number, 58
human serum albumin, 290, 292, 484 hydrophilic surface, 58
igG, 294 hydrophobic surface, 58
insulin sequencing, 291 surfactants, 59
interferon, 294 Weber number, 58
ion channels, 292 Young’s law, 58
ligand, 295
lysozyme, 294 Taylor, G. I., 194
movement proteins, 289 Taylor–Aris dispersion, 195f
N-terminus, 290 dispersion coefficient, 197, 198
nutrient proteins, 289 Taylor series, 351f, 413f
peptide bond, 289 temperature and concentration boundary layers,
polypeptide backbone, 289 205f
primary structure, 289 thermal conductivity, 42, 52, 182f
protein binding, 295 thermal sciences, 22
quartenary structure, 292 thermodynamics, 40f, 62f
regulatory proteins, 289 closed system, 63
secondary structure, 291 equilibrium, 63
structural proteins, 289 first law of thermodynamics, 65
structure, 289 and head loss, 69
tertiary structure, 291 and losses in pipe and channel flows, 68
transport proteins, 289 open system, 63
use of molecular dynamics simulations, perfect gas, 62
292 second law of thermodynamics, 66
thermodynamic property, 64
Redlich-Kwong equation of state, 353 transport properties, 18, 41f
Reuss, F. F., 19
RNA, 286 unsteady mass transport in a membrane, 201f

semipermeable membranes, 271f van der Waals forces, 31, 453, 456
concentration polarization, 275 Verlet, Loup, 422
current–voltage relationship, 273
induced charge electrical douible layers, Water, 231f, 452f
275 extended simple point charge model, 232
limiting current, 274 simple point charge model, 232
silica surface, 246f structure of, 231
equilibrium constants, 247 well-posed problems, 130f
isoelectric point, 247
Nernstian–Stern layer model, 249 zerofinding, 353f
polyethelene glycol treatment, 247 bisection method, 354
potential determining ions, 247 Newton’s method, 354
surface charge density, 248 secant method, 354