Nano Mechanics and Materials:

Theory, Multiscale Methods and Applications

by
Wing Kam Liu, Eduard G. Karpov, Harold S. Park

5. Introduction to Multiple Scale Modeling
Motivation for multiple scale methods
Coupling of length scales
Bridging scale concurrent method
Molecular dynamics (MD) boundary condition
Numerical examples
1D wave propagation
2D dynamic crack propagation
3D dynamic crack propagation
Discussion/Areas of improvement for bridging scale
Conclusions and future research
Role of Computational Methods
Nano- and micro-structure
Electronic structure
Molecular mechanics
Continuum
mechanics
Potentials
Const.
laws
Plasticity

Multiscale methods
Manufacturing
platform
Function
Performance
Reliability
Computations
and design
Prediction
Validation
Structural and material design
Optimization
Prediction and validation
Examples of Multi-Scale Phenomena in Solids
17:1
250:1
250:1
200 m
Shear bands
Mechanics of carbon nanotubes
Figures: D. Qian, E. Karpov, NU
Shaofan Li, UC-Berkeley
Movie: Michael Griebel, Universitt Bonn
Why Multiscale Methods?
Limitations of industrial simulations today:

a) Continuum models are good, but not always adequate
Problems in fracture and failure of solids require improved constitutive models
to describe material behavior
Macroscopic material properties of new materials and composites are not readily
available, while they are needed in simulation-based design
Detailed atomistic information is required in regions of high deformation or
discontinuity

b) Molecular dynamics simulations
Limited to small domains (~10
6
-10
8
atoms) and small time frames (~nanoseconds)
Experiments, even on nano-systems, involve much larger systems over longer times

Opportunities: 1) Obtain material properties by subscale (multiscale) simulation
2) Enrich information about material/structural performance across scales
via concurrent multiscale methodologies
Hierarchical vs. Concurrent

Hierarchical approach
Use known information at one scale to generate model for larger scale
Information passing typically through some sort of averaging process
Example: bonding models/potentials, constitutive laws


Concurrent approach
Perform simulations at different length scales simultaneously
Relationships between length scales are dynamic
Classic example: heat bath techniques


Macroscopic, Atomistic, Ab Initio Dynamics (MAAD)
Finite elements (FE), molecular
dynamics (MD), and tight binding (TB)
all used in a single calculation
(MAAD)
MAAD = macroscopic, atomistic, ab
initio dynamics
Atomistics used to resolve features of
interest (crack)
Continuum used to extend size of
domain
Developed by Abraham (IBM),
Broughton (NRL), and co-workers
From Nakano et al, Comput. In Sci. and Eng., 3(4) (2001).
MAAD: Concurrent Coupling of Length Scales
( ) ( )
( ) ( )
( )
/
/
, , , ,
, ,
,
Tot FE FE MD
MD MD TB
TB
H H H
H H
H
= +
+ +
+
u u u u r r
r r r r
r r

Scales are coupled in
handshake regions

Finite element mesh
graded down to atomic
lattice in the overlap region

Total Hamiltonian is
energy in each domain, plus
overlap regions
Nakano et al,
Comput. In Sci. and
Eng., 3(4) (2001).
Broughton, et al, PRB 60(4) (1999).
Handshake
at MD/FE
interface
Quasicontinuum Method
Tadmor and Phillips, Langmuir 12, 1996
Developed by Ortiz, Phillips and coworkers in 1996.
Deformation is represented on a triangulation of a subset
of lattice points; points in between are interpolated using
shape functions and summation rules

Adaptivity criteria used to reselect representative lattice
points in regions of high deformation

Applications to dislocations, grain boundary interactions,
nanoindentation, and fracture (quasistatic modeling)

Cauchy-Born rule assumes 1) continuum energy density
can be derived from the atomic potential; 2) deformation
gradient F describes deformation at both continuum and
atomic scales, and therefore serves as the link. Thus, atomic
deformation has to be homogeneous

Issues: non-local interaction, long dislocations/ill
conditioning, separation of scales, finite temperatures,
universal scenarios

Later improved by Arroyo, Belytschko, 2004, in application
to CNT: PRB 69, article 115415
T
W c
=
c
P
F
Challenges
Large number of degrees of freedom at the atomic scale
Interfaces: mismatch of dynamic properties, and other issues
Consistent and accurate representation of meso-, micro- & nano-
level behavior within continuum models
Multiple time scales
Potentials
Interdisciplinary nature of multiscale methods
- continuum mechanics
- classical particle dynamics (MD), and lattice mechanics
- quantum mechanics and quantum chemistry
- thermodynamics and statistical physics
Atomic scale plasticity: lattice dislocations
Finite temperatures
Entropic elasticity, soft materials
Dynamics of infrequent events: diffusion, protein dynamics
Algorithmic issues in large scale coupled simulations
Typical Issues
1. True coarse scale discretization and coupling between the scales

2. Handling interfaces where small and large scales intersect; handshake is
expensive and non-physical; spurious wave reflection

3. Double counting of the strain energy

4. Implementation: usage of existing MD and continuum
codes is hard; parallel computing

5. Dynamic mesh refinement/enrichment

6. Finite temperatures

7. Multiple time scales and dynamics of infrequent events


- BSM has resolved issues 1-2, and partially 3-6.

- The alternative: MSBC method, where issues 1-4 DO NOT ARISE
Typical interface model
The Bridging Scale Method
( )
( ) ( ) ( )
( ) ( )
( ) ( ) ( ) ( ) ( )
x x x
x x
x x x x x
' = +
=
' = = =
u u u
u Pu
u u Pu I P u Qu
= +
Two most important components:
- bridging scale projection
- impedance boundary conditions applied
MD/FE interface in the form of a time-history integral

Assumes a single solution u(x) for the entire domain.
This solution is decomposed into the fine and coarse
scale fields:
( ) x u
( ) x u
( ) x ' u
BS projection
T
1 T
A
A

=
=
M N M N
Q I NM N M
Within the bridging scale method, the MD and FE formulation exist
simultaneously over the entire computational domain:
+ =
The total displacement is a combination of
the FE and MD solutions:

Multiscale Lagrangian

Lagrangian formulation gives coupled,
coarse and fine scale, equations of motion

' = + = + u u u Nd Qq
T
T
( )
( )
A
U = c
=
c =
Md N f u
f
u M q Q f u
Bridging-Scale Equations of Motion
MD, q
FEM, d MD + FE, (q, d)
( )
T T T
1 1
( , , , ) ( , )
2 2
A
L U = = + d d q q d Md q Q M q d q
Impedance Boundary Conditions / MD Domain Reduction
MD degrees of freedom outside the
localized domain are solved implicitly
+
Due to atomistic nature of the model, the structural impedance is evaluated computed at
the atomic scale.
The MD domain is too large to solve, so that we eliminate the MD degrees of freedom
outside the localized domain of interest.

Collective atomic behavior of in the bulk material is represented by an impedance force
applied at the formal MD/continuum interface:
FE + Reduced MD +
Impedance BC
MD
FE

( )
T
0
( )
( ) ( ) ( ) ( )
t
A
t d t t t t
=
= +
}
Md N f u
M q f u q u
1D Illustration: Non-Reflecting MD/FE Interface
Impedance boundary conditions allows non-reflecting coupling of the fine and coarse grain solutions
within the bridging scale method.

Example: Bridging scale simulation of a wave propagation process; ratio of the characteristic lengths
at fine and coarse scales is 1:10

Direct coupling with continuum Impedance BC are involved
Over 90% of the kinetic wave energy
is reflected back to the fine grain.
Less than 1% of the energy is reflected.
Why is Multiscale Modeling Difficult?
Wave reflection at MD/FE interface
Larger length scales (FE) cannot represent wave lengths typically found at
smaller length scales (MD)
Also due to energy conserving formulations for both MD and FEM


MD MD FEM FEM
The phase velocity of progressive
waves is given by





Dependence on the wave number:




Value v
0
is the phase velocity of the
longest waves (at p 0).
-1 -0.5 0.5 1
0.5
1
1.5
2
2.5
3
/ p t
0
/ e e
continuum

v
p
e
=
-6 -4 -2 2 4 6
0.2
0.4
0.6
0.8
1
/ p t
continuum

0
/ v v
0 0
1
2sin sin
2 2
p v p
v p
e
e
= =

Incompatible Dispersion Properties of Lattices and Continua

lattice structure
lattice structure
Issues in Multiscale Modeling

Preventing high frequency wave reflection
- Need reduced MD system to behave like full MD system
- Reflection of high frequency waves can lead to melting of the
atomistic system

Need for a dynamic, finite temperature multiple scale method

True coarse scale representation
- No meshing FEM down to MD lattice spacing
- Different time steps for MD and FEM simulations

Mathematically sound and physically motivated treatment of high
frequency waves emitted from MD region at MD/FE interface

Selected References
Quasicontinuum Method
E. Tadmor, M. Ortiz and R. Phillips, Philosophical Magazine A 1996;
73:1529-1563
Coupled Atomistic/Discrete Dislocation method (CADD)
L. Shilkrot, R.E. Miller and W.A. Curtin, Journal of the Mechanics
and Physics of Solids 2004; 52:755-787
Bridging Domain method
S.P. Xiao and T. Belytschko, Computer Methods in Applied
Mechanics and Engineering 2004; 193:1645-1669
Review articles:
W.A. Curtin and R.E. Miller, Modelling and Simulation in Materials
Science and Engineering 2003; 11:R33-R68
W.K. Liu, E.G. Karpov. S. Zhang and H.S. Park, Computer Methods
in Applied Mechanics and Engineering 2004; 193:1529-1578