Applied Mathematics Lecture Notes

Fatihcan M. Atay

Spring 2023
Introduction

Applied math deals with: Modeling, transformation, approximation.

Transformation: Laplace, Fourier, and many others.
Approximation does not mean vague or uncertain solutions. On the contrary, one aims
to obtain detailed knowledge about the nature of the problem by approximations.
e = limnæŒ (1 + n1 )n = 2.71828 . . .
Applied math is not synonymous with: numerical solutions, data analysis, etc., al-
though there is an overlap in the topics.
Textbook: Mark Holmes, Introduction to the Foundations of Applied Mathematics,
2nd ed., Springer, 2019. We will cover the first four chapters and then use some additional
sources.
Most applied problems will involve differential equations, so a basic knowledge of dif-
ferential equations as well as linear algebra is essential.

1
Chapter 1

Dimensional analysis

1.1 Introduction
Numbers in math.
fi = 3.141592653589793 . . . is defined as the ratio of a circle’s circumference to its diam-
eter, or half the circumference of the unit circle, or the area of the unit circle. Babylonians
(around 2000 B.C.): fi ¥ 3 or later fi ¥ 25/8 = 3.125. Archimedes (around 250 B.C.)
usually credited for fi ¥ 22/7, Chinese Zu Chongzi (5th century A.D.): fi ¥ 355/113 =
3, 1415929204 . . .
e = 2.718281828459045 . . .
We can do calculations such as fi + e = 5.859874482048838...
Some other constants in physical applications: Gravitational acceleration on the surface
of the earth g ¥ 9.81m/s2 , the speed of light in vacuum c ¥ 299, 792, 458 m/s. It is
meaningless to add g + c; the dimensions are different.

1.1.1 Projectile problem

Example 1.1. An object thrown vertically upwards. x = distance measured from surface
of the earth. Ignore air resistance.
Newton’s second law (F = ma) and the law of gravitation give (ignoring air resistance)
d2 x mM
m = ≠G
dt2 (R + x)2
m is the mass of object, M is the mass of earth, R is the radius of earth.
We know that the gravitational force equals ≠mg on the surface (x = 0):
⇢M
m
⇢
⇢g = G
m
⇢
R2
=∆ GM = gR2

so that
d2 x R2 g
=≠ . (1.1)
dt 2 (R + x)2
Initial conditions x(0) = 0, dt (0)
dx
= v0 . This is a nonlinear second-order ODE; an exact
solution is not apparent.

2
CHAPTER 1. DIMENSIONAL ANALYSIS 3

Fact: R = 6371 km. Historically, Eratosthenes (ca. 240 BC) first calculated the
circumference of earth to be about 40,000 km. Shadow’s angle ◊/2fi = s/circumference of
earth

Approximate solution. Approximating assumption: x π R. Then R + x ¥ R. Eq.

(1.1) becomes
d2 x
= ≠g
dt2
which can be solved to yield
1
x(t) = ≠ gt2 + v0 t.
2
Are we justified in the assumption x π R? Let’s check the maximum value of x:

xÕ (t) = ≠gt + v0 .

Hence, the max value occurs at tM = v0

g and

v02
xM = x(tM ) = . (1.2)
2g

So the approximation is valid under the condition v02 /(2g) π R. We write this in dimen-
sionless form as v02 /(2gR) π 1.
(Note that the approximate solution could also be found by energy considerations, i.e.,
by equating kinetic energy at bottom to potential energy at top: 12 mv02 = mgxM ; however,
the potential energy term also uses the same approximation that x is sufficiently small so
that the gravitational force is constant during the motion.)
We have so far ignored air resistance; we’ll come to that later.

Exact solution for maximum height. With v = dt ,

dx
equation (1.1) becomes

dv R2 g
=≠ .
dt (R + x)2
CHAPTER 1. DIMENSIONAL ANALYSIS 4

Express v as a function of x instead of t, that is, v = v(x) = v(x(t)). By the chain rule,
dv dv dx dv
= = v.
dt dx dt dx
Substituting into the differential equation,
dv R2 g
v=≠
dx (R + x)2
which can be solved by separation of variables:
R2 g
v dv = ≠ dx
(R + x)2
v2 R2 g
=∆ = +c
2 (R + x)
The integration constant c is found from the initial condition v(x = 0) = v0 :
v02 R2 g
= +c
2 R
v 2
=∆ c = 0 ≠ Rg
2
Hence,
v2 R2 g v2
= + 0 ≠ Rg
2 (R + x) 2
Û
2R2 g
=∆ v(x) = ± + v02 ≠ 2Rg
(R + x)
as a function of x. (The plus sign corresponds to upward motion and minus sign to
downward motion.) At the highest point, x = xM and v = 0; therefore,
2R2 g
+ v02 ≠ 2Rg = 0.
(R + xM )
Solving for xM gives
Rv02
xM = .
2Rg ≠ v02
Dividing through by gR,
Á Á 1 v02
xM = R =R , where Á = . (1.3)
2≠Á 2 1 ≠ 2Á gR
For Á small (specifically for |Á| < 2), the fraction can be expanded in a power series
A B
Á Á Á2 Á3
xM =R 1+ + + + ... .
2 2 4 8
v2
The first term 12 RÁ gives our previous approximate solution xM = 2g0 , which was
obtained under the assumption Á π 1. The remainder terms give corrections to this.
Note that Á is a dimensionless quantity.
CHAPTER 1. DIMENSIONAL ANALYSIS 5

1.2 Physical dimensions

Fundamental dimensions: L: length, M : mass, T : time. Independent set, i.e., one of
them cannot be written in terms of the other two. One can also add temperature ◊ (for
thermodynamic problems) and current I (for electrical problems).
Quantity Dimension
Mass M Quantity Dimension
Length L Angle 1
Time T Angular velocity T ≠1
Area L2 Angular acceleration T ≠2
Volume L3 Force M LT ≠2
Density ML ≠3 Energy M L2 T ≠2
Frequency T ≠1 Power M L2 T ≠3
Period T Viscosity (dynamic) M L≠1 T ≠1
Wavelength L Temperature ◊
Velocity LT ≠1 Electric current I
Acceleration LT ≠2
Notation: Given a physical quantity q, the fundamental dimensions of q will be de-
noted as [q]. In the case of when q is dimensionless, [q] = 1.
In the projectile example, [v0 ] = L/T , [x] = L, [g] = M/L2 , and [xM /R] = 1.
Dimensional analysis is independent of the system of units used.
An equation is said to be dimensionally homogeneous if it does not depend on the
specific unit system being used. For example, Newton’s Law F = ma and the projectile
equation (1.1) are dimensionally homogeneous. If we specify a system of units (say, SI)
and set R = 6371 km and g = 9.81 m/s2 , then the resulting equation

d2 x 398184
=≠
dt 2 (6371 + x)2

is not dimensionally homogeneous.

1.2.1 Maximum height of projectile

Example 1.2. Referring to the projectile problem, we want to determine the maximum
height of the projectile, xM . We assume it depends on the parameters v0 , m, g:

xM = f (m, v0 , g)

for some unknown function f . These parameters should combine in a certain way to
produce the correct dimension for xM . Hypothesis: there are numbers a, b, c such that

[xM ] = [ma v0b g c ].

Using the fundamental dimensions,

3 4b 3 4c
L L
L = Ma .
T T2
CHAPTER 1. DIMENSIONAL ANALYSIS 6

Equating the exponents,

L: b+c=1
T : ≠ b ≠ 2c = 0
M : a=0

Therefore, a = 0, b = 2, and c = ≠1, so that

[xM ] = [v02 g ≠1 ].

Hence,
v02
xM = –
g
for some arbitrary (and dimensionless) number –.
Observations:

1. Although we don’t know – from this calculation, we can still get useful information
on how the projectile height xM depends on the parameters. For example, the height
is quadrupled if the initial velocity is doubled.

2. The value of – can be determined by (in principle a single) experiment: We measure

the initial velocity v0 and the height xM and determine – from the formula above.

The formula for xM obtained by dimensional analysis above is the same as the formula
(1.2) obtained by solving the ODE (1.1) under the assumption v02 /(2gR) π 1 (with – = 12 ).
However, it looks a bit different than the exact solution (1.3). Specifically, we do not have
R in our approximate formula. We can try the following alternative hypothesis for the
form of the solution
xM = f (v0 , g, R) (1.4)
having the product form
xM = v0b g c Rd
for some numbers b, c, d. (We already know that the solution does not depend on mass.)
In terms of dimensions,
3 4b 3 4
L L c d
L= L .
T T2
Therefore,

L: b+c+d=1
T : ≠ b ≠ 2c = 0.

With less equations than unknowns, the solution will not be unique. If we keep c arbitrary,
then b = ≠2c and d = 1 ≠ b ≠ c = 1 + c. Thus, for some dimensionless constant — and
arbitrary exponent c,
3 4c A B≠c
gR v02
xM = —v0≠2c g c R1+c = —R = —R
v02 gR
CHAPTER 1. DIMENSIONAL ANALYSIS 7

Other choices of the free parameter can yield different expressions. For example, if we
keep d arbitrary, then b = ≠2d + 2 and c = d ≠ 1. Thus, for some dimensionless constant
— and arbitrary exponent d,
3 4d A B≠d
v2 gR v2 v02
xM = —v0≠2d+2 g d≠1 Rd =— 0 =— 0
g v02 g gR
In fact, since — and c are arbitrary,
A B≠c1 A B≠c2
v02 v02
xM = —1 R and xM = —2 R
gR gR

for any values of —1 and c1 , and therefore

S A B≠c1 A B≠c2 T
v02 v02
xM = R U—1 + —2 V
gR gR

and similarly, S T
A B≠c1 A B≠c2
v02 v02
xM = R U—1 + —2 + ···V
gR gR

for some sequence of —i , ci . We conclude that xM has the general form

A B
v02
xM = R · f
gR

for some function f . (We have subsumed the constants —i into the function f .) This
is a reduction from (1.4) where the unknown function depends on three variables to an
unknown function depending on the single dimensionless product = v02 /(gR).
Comparing now with the exact solution (1.3), we see that the unknown function is
Á 1
f (Á) = ,
2 1 ≠ 2Á

where Á = v02 /(gR). This function is not of the product form Ác for any c; however,
expanding in a power series, we have
A B
Á Á Á2 Á3
f (Á) = 1+ + + + ···
2 2 4 8

which converges for |Á/2| < 1.

1.2.2 Drag on a sphere

Fluid dynamics is generally a difficult topic. The Navier-Stokes equations are nonlinear
PDEs describing the motion of viscous fluids. One of the seven Millennium Problems of
the Clay Mathematics Institute concerns the Navier-Stokes equations, with a prize tag
of $1 million. The Navier-Stokes equations are useful for describing the physics of many
phenomena of scientific and engineering interest. They may be used to model the weather,
CHAPTER 1. DIMENSIONAL ANALYSIS 8

ocean currents, water flow in a pipe and air flow around a wing. They help with the design
of aircraft and cars, the study of blood flow, the design of power stations, the analysis of
pollution, etc.
The resistance experienced by an object moving in a fluid is not easy to characterize.
We will consider a very simple geometry as a start.
A spherical object of radius R moving in a fluid at speed v experiences a resistive drag
force. Hypothesis: The drag force depends on

DF = f (R, v, fl, µ)

where fl is the density of the fluid and µ is the (dynamic) viscosity. The viscosity of a fluid
is a measure of its resistance to deformation at a given rate. Viscosity is a material property
that relates the viscous stresses in a material to the rate of change of a deformation (the
strain rate). For liquids, it corresponds to the informal concept of "thickness": for example,
syrup has a higher viscosity than water.
Comparing dimensions,
[DF ] = [Ra v b flc µd ].
Expressing in terms of fundamental dimensions

M LT ≠2 = La (L/T )b (M/L3 )c (M/LT )d = La+b≠3c≠d T ≠b≠d M c+d .

gives

L : a + b ≠ 3c ≠ d = 1
T : ≠b ≠ d = ≠2 (1.5)
M :c+d=1

If we keep d arbitrary, then c = 1 ≠ d, b = 2 ≠ d, and a = 2 ≠ d. Hence,

3 4d
µ
DF = –R2≠d v 2≠d fl1≠d µd = –flR2 v 2
Rvfl
for some arbitrary numbers – and d. It follows that the drag force has the form

DF = flR2 v 2 F ( )

where the quantity = Rvfl µ

is a dimensionless product. In fluid mechanics often its inverse
is used, called the Reynolds number:
Rvfl
Re = .
µ

(In a general problem, R is some characteristic length of flow geometry.)

Hence, the drag force has the form

DF = flR2 v 2 G(Re)

for some function G, G(Re) = F (1/Re), which has been experimentally measured.
CHAPTER 1. DIMENSIONAL ANALYSIS 9

We see from the figure that there is a range of Re values (around 103 < Re < 105 ) for
which the drag force is constant.
For smaller values of Re (Re < 102 , occurs for very small flow speeds, known as
Stokes flow or creeping flow, which occurs in paint, MEMS devices, viscous polymers and
microorganisms, and is a linear simplification for the Navier-Stokes equations valid in
the limit Re æ 0), the dependence is linear in the log-log plot, which means log(G) =
c1 ≠ c2 log(Re), or G = Rec3c2 for some constants c1 , c2 , and c3 = 10c1 , which can be found
by curve-fitting to data. Actually, Stokes flow can be exactly solved, and the values are
found to be c3 = 6fi and c2 = 1. Therefore, at low speeds
6fi 6fiµ
DF ¥ flR2 v 2 = flR2 v 2 = (6fiµR)v
Re Rvfl
i.e., the drag force is proportional to speed of the sphere at low Reynolds numbers. This
is the basis for models that take resistive forces proportional to speed of the object (called
Stokes drag).
Larger values of the Reynolds number is associated with turbulent flow and complex
dynamics.
Scale models. The dimensionless quantities play an important role in scale models,
such as tests conducted in wind tunnels or towing tanks. For example, if the (spherical)
object is too large (say, R = 20 m) to make experiments in a wind tunnel, one can make a
smaller model (say, R̄ = 0.5 m). That’s a reduction by a factor of 40. Assuming the same
fluid is used in the experiment (i.e., same fl and µ) then the velocity must be 40 times
higher for the model to have the same Reynolds number and to use the empirical values
shown in the figure above.

1.2.3 Toppling dominoes

We want to model the velocity v of the wave generated by toppling dominoes.
Relevant parameters: d = distance between dominoes, h = height, w = width of domi-
noes, and the gravitational constant g.

[v] = [da hb wc g e ]

(L/T ) = La Lb Lc (L/T 2 )e = La+b+c+e T ≠2e

CHAPTER 1. DIMENSIONAL ANALYSIS 10

Therefore,

a+b+c+e=1
≠2e = ≠1
1 1
We have e = 2 and b = 2 ≠ a ≠ c, so that
3 4a 3 4c
a 1/2≠a≠c c 1/2
 d w
v = –d h w g = – hg
h h
where – is an arbitrary constant. Therefore, the general solution must be of the form

v= hgF ( 1, 2)

where F is an arbitrary function of the dimensionless products 1 = hd and 2 = wh . In

this way, the original problem involving an unknown function of four parameters is reduced
to an unknown function of two dimensionless parameters.
If it is assumed that the dominoes are very thin, so that the width is small compared
to height, w π h, then one can take 2 = wh = 0 and the solution form simplifies to

v= hgG( 1)

for some function G. Experiments for a particular type of dominoes have yielded the data
shown in the figure below.

It can be seen that for 1 < 0.5, G is approximately constant at a value 1.5. Thus,

v ¥ 1.5 hg.

For a typical domino with h = 5 cm, we find that the wave velocity is about v ¥ 1 m/s.

1.3 Buckingham Pi Theorem

Suppose a physical quantity q depends on n physical parameters or variables p1 , p2 , . . . , pn .
By physical we mean the quantity is measurable. Each can be expressed in fundamental
dimensions

[q] = L¸0 T t0 M m0
[pi ] = L¸i T ti M mi
CHAPTER 1. DIMENSIONAL ANALYSIS 11

where for definiteness we have assumed the fundamental dimensions M, L, T , but other
dimensions like temperature and current can be added as necessary.
Modeling assumption: q = f (p1 , p2 , . . . , pn ) and this relation is dimensionally homoge-
neous, that is, it is independent of the system of units used in measurements.
The question is whether there are numbers a1 , a2 , . . . , an so that

[q] = [pa11 pa22 · · · pann ].

Substituting in the dimensions gives the system of equations.

¸1 a1 + ¸2 a2 + · · · + ¸n an = ¸0
t1 a1 + t2 a2 + · · · + tn an = t0
m1 a1 + m2 a2 + · · · + mn an = m0

In matrix form
Aa = b (1.6)
where Q R
Q R a1 Q R
¸1 ¸2 . . . ¸n c d ¸0
c d c a2 d c d
A = a t 1 t2 . . . tn b , a=c
c .. d ,
d b = a t0 b .
m1 m2 . . . mn a . b m0
an
The dimension matrix A has as many rows as the number of fundamental dimensions and
as many columns as the number of parameters q is assumed to depend on. The matrix
equation (1.6) may have no solution. In this case the assumption that q depends on
p1 , . . . , pn is incomplete and additional parameters are needed.

Definition 1.1. (Dimensionally complete.) The set p1 , p2 , . . . , pn is dimensionally com-

plete for q if it is possible to combine the pi ’s to produce a quantity with the same dimension
as q. If it is not possible, the set is dimensionally incomplete for q.
From now on we assume that the pÕi s are dimensionally complete. To find the general
solution of (1.6), first consider the associated homogeneous equation Aa = 0. Let a1 , . . . , ak
be a basis for the null space N (A). Then the general solution of the homogeneous equation
is c1 a1 + · · · + ck ak . If ap is any particular solution of (1.6), then the general solution of
(1.6) is
a = ap + c1 a1 + · · · + ck ak
where ap is any vector that satisfies (1.6) and c1 , . . . , ck are arbitrary numbers.
Each of the basis vectors ai for the null space N (A) gives rise to a dimensionless
product. Indeed, if we write in component form

ai = (–, —, . . . , “)€ ,

then Q R
Q R–
¸1 ¸2 . . . ¸n c d
c d c— d
c
Aai = a t1 t2 . . . tn b c . d = 0. (1.7)
.d
m1 m2 . . . mn a . b
“
CHAPTER 1. DIMENSIONAL ANALYSIS 12

Then the corresponding dimensionless product is

i = p–1 p—2 · · · p“n .

This follows because [pi ] = L¸i T ti M mi , so that i has dimensions
L¸1 –+¸2 —+···+¸n “ T t1 –+t2 —+···+tn “ M m1 –+m2 —+···+mn “
but the exponent of L, namely ¸1 – + ¸2 — + · · · + ¸n “, is zero by (1.7), and similarly the
exponents of T and M are zero. Hence i is dimensionless.
Since the ai ’s are independent, the dimensionless products 1 , . . . , k are independent.
Finally, if we write the particular solution ap in component form
ap = (a, b, . . . , c)€ ,
then Q R
Q aR Q R
¸1 ¸2 . . . ¸n c c b d
d ¸0
c d c d
Aap = a t1 t2 . . . c .. d
tn b c d = a t0 b . (1.8)
m1 m2 . . . mn a . b m0
c
Then the quantity
Q = pa1 pb2 · · · pcn
has the same dimension as q. This is proved in a similar way: The dimension of Q is
[Q] = L¸1 a+¸2 b+···+¸n c T t1 a+t2 b+···+tn c M m1 a+m2 b+···+mn c = Lq0 T t0 M m0 = [q]
where we have used (1.8).
Combining the above arguments, we have proved the following theorem.
Theorem 1.1 (Buckingham Pi Theorem (1915)). . Assuming the formula q = f (p1 , p2 , . . . , pn )
is dimensionally homogeneous and dimensionally complete, then it is possible to reduce it
to one of the form q = Q · F ( 1 , 2 , . . . , k ), where 1 , 2 , . . . , k are independent dimen-
sionless products of p1 , p2 , . . . , pn . The quantity Q is a dimensional product of p1 , p2 , . . . , pn
with the same dimensions as q.
In this way, the original formula for q which contains an unknown function of n variables
can be reduced to a formula that contains an unknown function of k dimensionless variables,
where k = dim N (A).
Example 1.3. In the example of drag on a sphere, we had the equations (1.5). So, in the
matrix equation (1.6) we have
Q R
Q R a Q R
1 1 ≠3 ≠1 cbd 1
c d c d c d
A = a0 ≠1 0 ≠1b , a = c d, b = a≠2b .
acb
0 0 1 1 1
d
To solve Aa = b, we form the augmented matrix
Q R
1 1 ≠3 ≠1 1
c d
(A|b) = a 0 ≠1 0 ≠1 ≠2 b
0 0 1 1 1
CHAPTER 1. DIMENSIONAL ANALYSIS 13

By a sequence of elementary row operations (add the second row to first, add 3 times the
third row to first, multiply the second row by ≠1) , we obtain
Q R
1 0 0 1 2
c d
a 0 1 0 1 2 b.
0 0 1 1 1

Recall that the elementary row operations do not change the null space. The matrix on
the left has rank 3, so the dimension of the null space is k = 1. To find a basis for the null
space, we solve Q R
Q R a Q R
1 0 0 1 c d 0
c dc b d c d
a 0 1 0 1 bc d=a 0 b
a c b
0 0 1 1 0
d
which implies a = b = c = ≠d; hence ( ≠1 ≠1 ≠1 1 )€ is a basis for the null space.
For a particular solution, we solve
Q R
Q R a Q R
1 0 0 1 c d 2
c dc b d c d
a 0 1 0 1 bc d=a 2 b
a c b
0 0 1 1 1
d

If we keep d as a free parameter, then c = 1 ≠ d, b = 2 ≠ d, and a = 2 ≠ d. For the particular

choice d = 0, ap = ( 2 2 1 0 )€ is then one particular solution. Therefore, the general
solution is Q R Q R Q R
a 2 ≠1
c b d c 2 d c ≠1 d
c d c d c d
c d=c d + dc d
a c b a 1 b a ≠1 b
d 0 1
for some arbitrary constant d. With these exponents, we obtain as before the drag force
3 4d
µ
DF = flR2 v 2
Rvfl
for arbitrary d, so DF must have the form

DF = flR2 v 2 F ( ).

In the notation of the Buckingham Pi theorem, the dimensionless quantity is = Rvfl

µ
and
2 2
the dimensional quantity is Q = flR v , which has the same dimension as the drag force
DF .

1.4 Nondimensionalization and scaling

In the projectile problem
d2 x R2 g
= ≠
dt2 (R + x)2
CHAPTER 1. DIMENSIONAL ANALYSIS 14

we will nondimensionalize the variables as

t = tc ·, x = xc u
where tc is a characteristic time and xc is a characteristic length to be determined later. The
important observation is that the new time and space variables, · and u, are dimensionless.
To transform the equations of motion to new variables, note that by the chain rule,
d d· d 1 d
= =
dt dt d· tc d·
3 4
d2 d d 1 d2
2
= = .
dt dt dt t2c d· 2
Then the projectile equation (1.1) becomes
1 d2 gR2
(x u) = ≠
t2c d· 2 (R + xc u)2
c

The next step is to collect parameters into dimensionless groups. There is no unique way
to do this. For example, we can factor the denominator as R(1 + xc u/R) or xc (R/xc + u).
The first choice allows us to cancel the R in the numerator, so following this choice, we
have
xc d2 1
(u) = ≠ .
2
gtc d· 2 (1 + xc u/R)2
There are several different ways to define the characteristic scales. We consider the following
cases.
Ò
1. xc = R and tc = g.
R
Then

d2 1
(u) = ≠
d· 2 (1 + u)2

where u = x/R and · = t/ R/g. This would be the cleanest form if we were to
solve the equation analytically. All the parameters now appear in the nondimensional
variables and the equation does not contain any other symbols than u and · .
2. Another possibility is suggested by our solutions of the reduced problem: we scale
time and length by the initial velocity and the characteristic acceleration g:
v02 v0
xc = , tc =
g g
so that
xc v2
Á= = 0
R gR
is a dimensionless (and small) parameter and the equation becomes
d2 1
(u) = ≠ ,
d· 2 (1 + Áu)2
which would be a good choice for approximate solutions involving powers of Á (regular
perturbation expansion).
CHAPTER 1. DIMENSIONAL ANALYSIS 15

3. A third choice could be xc = R as in the first option above but tc = R/v0 instead, so
that the equation becomes
xc d2 v02 d2 1
(u) = (u) = ≠
2
gtc d· 2 gR d· 2 (1 + u)2
that is,
d2 1
Á (u) = ≠
d· 2 (1 + u)2
where Á = v0 /(gR) as before. This also looks promising for small values of Á; however,
it has a very different character as the previous equation, as we will see in detail when
doing perturbation methods (singular perturbation expansion). One can already see
a big difference from a regular perturbation case: here, when Á = 0 we don’t have a
differential equation any more, as the left hand side vanishes!

1.5 Similarity Variables and the Diffusion Equation

Diffusion equation: u(x, t) is the density of a substance at location x and time t,

ˆu ˆ2u
=D 2 (1.9)
ˆt ˆx
where x œ (0, Œ), t > 0 and D is a positive constant called the diffusion coefficient. We
assume boundary conditions

u|x=0 = u0 , u|xæŒ = 0

and initial condition

u|t=0 = 0.
It is assumed that u0 > 0.
Dimensional identities regarding derivatives:
5 6 C D
df [f ] d2 f [f ]
= , =
dt [t] dt2 [t]2
5 6 5 6 C D
ˆu [u] ˆu [u] ˆ2u [u]
= , = , = .
ˆt [t] ˆx [x] ˆt ˆx [t][x]
These follow from the definition of the derivative.
Dimensions: [u] = M/L3 , [uxx ] = M/L5 and [ut ] = M/(T L3 ) . To find the dimensions
of D, consider the dimensions in (1.9),

[Duxx ] = [ut ]
M M
[D] 5 =
L T L3
L2
[D] =
T
Also, from the boundary condition, [u0 ] = [u] = M/L3 .
CHAPTER 1. DIMENSIONAL ANALYSIS 16

1.5.1 Dimensional analysis

The assumption that
u = f (x, t, D, u0 )
is reasonable since these are the only parameters in the equation. The question is whether
we can find numbers a, b, c, d such that
[u] = [xa tb Dc udo ].
Comparing dimensions,
A Bc 3 4d
M L2 M
= La T b = La+2c≠3d T b≠c M d .
L3 T L3
Thus,
a + 2c ≠ 3d = ≠3
b≠c=0
d=1
If we take a as a free parameter, we have d = 1 and b = c = ≠a/2; therefore,
3 4a
x
u = –u0 Ô
Dt
for some arbitrary dimensionless constant –. We conclude that the general solution has
the form 3 4
x
u = u0 F Ô = u0 F (÷) (1.10)
Dt
Ô
for some function F . The variable ÷ = x/ Dt is called a similarity variable; it is a
dimensionless product that involves the independent variables in the problem.
If in the above derivation we keep c as a free parameter, then we obtain d = 1, b = c and
a = ≠2c, which yields a form 3 4
Dt c
u = –u0
x2
which is equivalent to the above form since the exponent c is arbitrary. This derivation
implies that u = u0 G(›) is a function of the similarity variable › = Dt/x2 . Although the
two representations are equivalent up to this point, for the subsequent analysis it is easier
to work with ÷ than ›, because the diffusion equation (1.9) involves second derivative
with respect to x; see the derivation below.

1.5.2 Similarity Solution

Ô
From u = u0 F (÷) and ÷ = x/ Dt, by the chain rule,
ˆu ˆ÷
= u0 F Õ (÷)
ˆt ˆt
≠x
= u0 F Õ (÷)
2D1/2 t3/2
÷
= ≠u0 F Õ (÷) .
2t
CHAPTER 1. DIMENSIONAL ANALYSIS 17

Similarly,
ˆu ˆ÷ 1
= u0 F Õ (÷) = u0 F Õ (÷) Ô
ˆx ˆx Dt
ˆ2u 1
= u0 F ÕÕ (÷) .
ˆx2 Dt
Substituting into the diffusion equation (1.9) gives
÷ 1
≠u0 F Õ (÷) =⇢
⇢u0 F ÕÕ (÷)
D
2t⇤ ⇢t⇤
⇢
D
1 Õ
F (÷) = ≠ ÷F (÷).
ÕÕ
2
Note that ÷ œ (0, Œ) since x œ (0, Œ) and t > 0.
We must also transform the boundaryÔ and initial conditions. Knowing the form of the
solution (1.10) u = u0 F (÷) = u0 F (x/ Dt)
u|x=0 = u0 : When x = 0 we have ÷ = 0. This yields F (0) = 1.
u|xæŒ = 0: From (1.10) we obtain F (Œ) = 0.
u|t=0 = 0: Since ÷ is not defined for t = 0, we deal by taking limits, namely
3 4
x
lim u0 F Ô =0
tæ0+ Dt
for any positive value of x. Since u0 is nonzero, we require that F (Œ) = 0, which is the
same condition obtained from the boundary condition above.
We have thus transformed the original problem involving a PDE to a problem involving
the ODE
1
F ÕÕ (÷) = ≠ ÷F Õ (÷), ÷ œ (0, Œ)
2
with the conditions F (0) = 1 and F (Œ) = 0. This linear ODE is easy to solve: Let G = F Õ .
Then
1
GÕ (÷) = ≠ ÷G(÷).
2
The general solution is s 1 2
G(÷) = –e≠ 2 ÷ d÷ = –e≠÷ /4 .
for some arbitrary constant –. Since, G = F Õ , we integrate from 0 to ÷ to obtain
⁄ ÷ ⁄ ÷ ⁄ ÷
2 /4
G(s) ds = F (s) ds = F (÷) ≠ F (0) = –
Õ
e≠s ds.
0 0 0

Using the condition F (0) = 1, we have

⁄ ÷
2 /4
F (÷) = 1 + – e≠s ds.
0
s Œ ≠s2 /4 s 2 Ô
The condition F (Œ) = 0 implies 1 + – 0 e ds = 0. Note that 0Œ e≠s /4 ds = fi (see
Ô
below). Thus – = ≠1/ fi.
CHAPTER 1. DIMENSIONAL ANALYSIS 18

Ô
The value fi follows from the evaluation of the so-called Gaussian integral (also called
the Euler-Poisson integral) ⁄ Œ
2 Ô
I= e≠x dx = fi.
≠Œ

One way to derive is to use the trick (due to Poisson) of calculating I 2 instead and using
polar coordinates:
3⁄ Œ 42 3⁄ Œ 4 3⁄ Œ 4
2 2 2
I2 = e≠x ds = e≠x dx e≠y dy
≠Œ ≠Œ ≠Œ
⁄ Œ ⁄ Œ
≠x2 ≠y 2
= e e dx dy
≠Œ ≠Œ
⁄ Œ ⁄ Œ
2 +y 2 )
= e≠(x dx dy
≠Œ ≠Œ
⁄ 2fi⁄ Œ
2
= e≠r r dr d◊
0 0
2
-Œ
-
= 2fi · (≠ 12 )e≠r -
r=0
=fi
Ô
which proves that I = fi. Now, with the change of integration variable x = s/2 and
dx = 12 ds, we have
⁄ Œ ⁄ Œ ⁄ Œ
≠x2 ≠x2 2 /4 Ô
I= e dx = 2 e dx = e≠s ds = fi.
≠Œ 0 0
We have thus calculated
⁄ ÷
1 2 /4
F (÷) = 1 ≠ Ô e≠s ds
fi 0

or with a change of integration variable r = s/2 and dr = 12 ds,

⁄ ÷/2
2 2
F (÷) = 1 ≠ Ô e≠r dr.
fi 0

The integral is not among the elementary functions; however, the expression above arises
so often that it leads to the definition of a special function.
Definition 1.2. The error function erf is defined as
⁄ z
2 2
erf(z) := Ô e≠r dz
fi 0

and the complementary error function erfc is defined as

⁄ z
2 2
erfc(z) := 1 ≠ erf(z) = 1 ≠ Ô e≠r dz.
fi 0

These special functions appear, among other places, in probability theory: The error
function corresponds to the area under the Gaussian function (bell curve). For a Ô random
variable Z that is normally distributed with mean 0 and standard deviation 1/ 2, the
probability that Z falls in the range [≠z, z] equals erf(z) (for z Ø 0). The complementary
error function erfc(z) gives the probability that Z does not lie in [≠z, z].
CHAPTER 1. DIMENSIONAL ANALYSIS 19

Hence, we have found the solution of the diffusion problem as

3 4
x
u(x, t) = u0 erfc(÷/2) = u0 erfc Ô .
2 Dt

2
erfc(x)
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
≠3 ≠2 ≠1 0 1 2 3
x
1
t=0.1
t=1.0
t=5.0
t=15
0.8 t=50

0.6
u(x,t)

0.4

0.2

0
0 2 4 6 8 10
x
In conclusion, similarity variables and dimensional analysis provide a powerful tool for
solving PDEs. In fact, it is one of the few methods that can be used to solve nonlinear
PDEs. However, it is limited to problems of certain specific forms. For example, if x in the
above example is confined to a finite interval (0, ¸) then dimensional analysis will give us
two independent similarity variables, which is not a reduction over the original variables
x, t.