Chapter 3

Nondimensionalisation

The first and arguably the most important step in the analysis of a system of differential equations.

It involves scaling each variable (dependent and independent) by a typical or reference value, leaving
a nondimensional variable whose typical scale is O(1).

Nondimensionalisation or problem normalisation has several important uses:

1. It identifies the dimensionless groups (ratios of dimensional parameters) which control the
solution behaviour.

2. Terms in the equations are now dimensionless and so allows comparison of their sizes.

This allows the identification of the important (i.e. dominant) terms in the equations and their
interaction in different regimes, giving insight into the structure of solutions and the dominant
physical mechanisms at work.

In particular, negligible terms can be identified leading to simplification in many circumstances.

3. It allows estimates of the effects of additional features to the original model through the new
dimensional group(s) associated with the additional term(s). This allows measurement of the
effect of the physical feature(s) in the model.

4. Finally, it can reduce the number of parameters ocurring in the problem by forming the nondi-
mensional parameters or dimensionless groups.

1
CHAPTER 3. NONDIMENSIONALISATION 2

3.1 Scaling

If an equation has a variable u, say, then we nondimensionalise that variable by writing, for example,

u = [u]ū

where [u] is the chosen scale (with the same dimensions as u) and ū is the corresponding dimensionless
variable. If a system of equations describes a real process then it is dimensionally homogeneous i.e.
consistent. The process of nondimensionalisation will necessarily give a set of equations, each of whose
terms is dimensionless, after division through by the dimensions of the equations. It is then possible
to compare terms in a meaningful way.

The art of nondimensionalisation lies in the choice of scales. There is no right or wrong way to do
it (other than to only partially nondimensionalise the equations) and in more complicated problems,
the choice of scales can be the difficult part of the analysis. The basic principle is that the scales must
ultimately be chosen self-consistently by balancing terms in the equations. Because the purpose is to
obtain ‘properly scaled’ equations in which the largest dimensionless terms are numerically of O(1),
the simplest choices arise when the scales can be chosen so that all the dimensionless parameters are
O(1).

This provides our rationale. Given no other information, one assumes a priori that dimensionless
variables and their derivatives are O(1), until we are forced to assume otherwise. It is only when
inconsistencies arise that the process of rescaling becomes necessary. The generic situation in which
this happens is where singular perturbation theory is appropriate. In general, not all dimensionless
parameters can be choosen to be O(1), in which case they are first assumed to be O(1) and then the
limit in which they become large is taken.

3.2 Examples

3.2.1 Example 1

The number of atoms N (t) at time t of a radiactive substance is governed by the differential equation

dN
= −λN
dt

with an initial condition N = N0 at t = 0. Here λ is a decay constant with units of [time]−1 .

CHAPTER 3. NONDIMENSIONALISATION 3

We nondimensionalise as follows
t̄
N = N0 N̄ , t= ,
λ
where N0 is taken as the reference value for N and 1/λ for the time scale. This gives the dimensionless
problem
dN̄
= −N̄ with N̄ = 1 at t̄ = 0,
dt̄
for N̄ (t̄).

3.2.2 Example 2

The motion of a linearly damped pendulum is governed by the equation

d2 θ dθ
ℓ + k + g sin θ = 0, (3.1)
dt2 dt

with the initial conditions

dθ
at t=0 θ = θ0 and = ω0 . (3.2)
dt
Here θ(t) represents the angle that the pendulum makes to the vertical at time t, the initial angle
being θ0 and initial angular speed ω0 . The dimensional parameters are the length ℓ of the pendulum,
the coefficient of resistance k and acceleration due to gravity g.

We nondimensionalise as follows
θ0
θ = θ0 θ̄, t= t̄,
ω0
using the initial values to give characteristic scales for the dependent variable θ and independent
variable t. Thus we obtain the dimensionless problem

d2 θ̄ dθ̄ ( )
2
+K + G sin θ0 θ̄ = 0, (3.3)
dt̄ dt̄

subject to
dθ̄
at t̄ = 0 θ̄ = 1 and = 1, (3.4)
dt̄
where we have introduced the dimensionless parameters

kθ0 gθ0
K= , G= ,
ω0 ℓ ω02 ℓ

in addition to θ0 .

The dimensional problem has 5 parameters, the four dimensional parameters ℓ, k, g, ω0 and the dimen-
sionless parameter θ0 . The dimensionless problem has three dimensionless parameters K, G, θ0 .
CHAPTER 3. NONDIMENSIONALISATION 4

If observational values are available for the parameters ℓ, k, g, ω0 , θ0 then these may be used to infer
the sizes of the dimensionless groups K, G, θ0 . The sizes of these dimensionless parameters determine
whether they are to be taken as O(1) or a suitable limit needs to be considered in which their values
are small or large. The latter cases lead to either a regular or singular perturbation problem.

As an example, suppose we have the situation in which θ0 = π/4 radians, ω0 = 1 radian per second,
ℓ = 1 m, k = 0.1 m/s and g = 10 m/s2 . Then K ≈ 0.079, G ≈ 7.9, θ0 ≈ 0.79. Thus we are interested
in analyzing the dimensionless problem in the limit K → 0 with G = O(1), θ0 = O(1) which should
be regular.

If in contrast k = 100 m/s then we would have K ≈ 79, G ≈ 7.9, θ0 ≈ 0.79. Thus we are now
interested in analyzing the dimensionless problem in the limit K → ∞ with G = O(1), θ0 = O(1)
which is singular.

3.2.3 Example 3

Consider the following boundary-value problem (BVP) for one-dimensional heat flow in a bar,

∂u ∂2u
in 0 < x < ℓ, t > 0 ρc = k 2 + q, (3.5)
∂t ∂x

at x = 0 u = u0 , (3.6)
∂u
at x = ℓ −k = h(u − ui ), (3.7)
∂x
at t = 0 u = ui , (3.8)

where ui is the initial temperature (which is also the external surrounding temperature), u0 is the
temperature at the end of the bar which is raised above ui , h denotes the heat transfer coefficient,
ρ, c, k denote the density, specific heat and conductivity of the bar respectively. The length of the bar
is ℓ and q represents a constant heat source term. All variables are assumed dimensional, this being
the statement of the dimensional problem.

We nondimensionalise as follows

ℓ2
x = ℓx̄, t= t̄, u = ui + (u0 − ui )ū,
κ

with κ = k/ρc and introduce the two dimensionless parameters

qℓ2 hℓ
Q= , H= ,
k(u0 − ui ) k
CHAPTER 3. NONDIMENSIONALISATION 5

to obtain

∂ ū ∂ 2 ū
in 0 < x̄ < 1, t̄ > 0 = + Q, (3.9)
∂ t̄ ∂ x̄2

at x̄ = 0 ū = 1, (3.10)
∂ ū
at x̄ = 1 − = H ū, (3.11)
∂ x̄
at t̄ = 0 ū = 0. (3.12)

It is assumed that H, Q are O(1) quantities. If they are not, then the appropriate limit needs to be
taken which will lead to either a regular or singular perturbation problem.

3.2.4 Example 4

Consider the Navier-Stokes equations of an incompressible Newtonian viscous fluid

∇.v = 0,
( )
∂v
ρ + (v.∇)v = −∇p + µ∇2 v + ρF,
∂t
( )
∂T
ρc + (v.∇)T = k∇2 T + Φ,
∂t
where v is the velocity, p is the pressure, T is the temperature and are functions of spatial coordinates
x and time t. Also, ( )2
1 ∂vi ∂vj
Φ= µ +
2 ∂xj ∂xi
is the viscous dissipation term. The dimensional parameters of density ρ, specific heat c, conductivity
k and viscosity µ are assumed constant.

The body force per unit mass F is assumed to be gravity so that F = g = gḡ with g = |g|.

These equations arise from the application of the physical laws of conversation of mass, momentum
and energy.

In the momentum equation, the physical effects modelled are: inertia, viscous forces, gravity respec-
tively.

In the heat equation, the physical effects modelled are: heat convection (or advection), heat conduc-
tion, viscous dissipation respectively.

The problem is completed by specification of suitable boundary conditions to a give a well-posed BVP.

Consequently typical scales or reference values for the variables will be given.
CHAPTER 3. NONDIMENSIONALISATION 6

Let U be a typical velocity scale

L be a typical length
T1 − T0 be a typical temperature deviation (T0 is an ambient or equilibrium temperature)
then we may nondimensionalise as follows

L
x = Lx̄, v = U v̄, t = t̄, p = P p̄, T = T0 + (T1 − T0 )T̄ ,
U

where P is to be determined, which gives

∇.v̄
¯ = 0,
( )
∂ v̄ P ¯ µ ¯2 Lg
+ (v̄.∇)v̄
¯ =− 2
∇p̄ + ∇ v̄ + 2 ḡ,
∂ t̄ ρU ρU L U
( )
∂ T̄ k ¯2 µU
+ (v̄.∇)
¯ T̄ = ∇ T̄ + Φ̄,
∂ t̄ ρcU L ρcL(T1 − T0 )
where ( )2 ( )
1 ∂v̄i ∂v̄j ∂ ∂ ∂
Φ̄ = + , ∇
¯ = , , .
2 ∂ x̄j ∂ x̄i ∂ x̄1 ∂ x̄2 ∂ x̄3
Introduce the dimensionless parameters
ρU L
Reynolds number Re = compares effects of inertia and viscous forces
µ
U2
Froude number F r = compares inertia and gravity
Lg
ρcU L UL
Peclet number P e = = compares convection to conduction
k κ
µU 2
Brinkman number Br = compares viscous dissipation with heat conduction
k(T1 − T0 )
µc Pe
Prandtl number P r = = compares viscous terms to those of heat conduction
k Re
Hence we have
∇.v̄
¯ = 0,
( )
∂ v̄ ¯ + 1 ∇ ¯ 2 v̄ + 1 ḡ,
+ (v̄.∇)v̄ = −∇p̄
¯
∂ t̄ Re Fr
( )
∂ T̄ ¯ T̄ = 1 ∇¯ 2 T̄ + Br Φ̄,
+ (v̄.∇)
∂ t̄ Pe Pe
where we have chosen P = ρU 2 i.e. the pressure scale is based on inertial forces.

The choice of pressure scale is not unique and could have been based on viscous forces so that

µU ρU 2
P = = .
L Re

The appropriate scalings are dictated by the problem and the numerical values of the dimensionless
parameters.
CHAPTER 3. NONDIMENSIONALISATION 7

The identification of dimensionless parameters allows model similitude i.e. a model will represent
(dynamically similar to) a practical situation if the values of the dimensionless parameters are the
same for both.

The limits of large and small Reynolds number Re are of particular interest.

Simplification of the equations by neglecting a term multiplied by a small dimensionless parameter is

the first step in a systematic procedure for obtaining an asymptotic expansion for the full solution in
terms of the small parameter.

If we consider the above model with 1/F r = Br = 0 and no temperature variation then

v̄ = v̄(x̄, t̄, Re) and p̄ = p̄(x̄, t̄, Re).

If Re ≪ 1 then we may expand in regular powers of Re to obtain the asymptotic expansions

p̄ = p̄0 (x̄, t̄) + Re p̄1 (x̄, t̄) + . . . , v̄ = v̄0 (x̄, t̄) + Re v̄1 (x̄, t̄, Re) + . . . ,

where p̄0 and v̄0 are the solutions to the simplified model with Re = 0.

This is a regular perturbation procedure and may be used if p̄(x̄, t̄, Re) and v̄(x̄, t̄, Re) are regular
functions of Re near Re = 0.

In some circumstances the expansion is invalid and a singular perturbation procedure must be used.
In these cases the solution of the problem with Re = 0 and the solution for Re ̸= 0 but Re ≪ 1 are
very different.

An example is the high Reynolds number flow Re ≫ 1. A regular expansion in powers of 1/Re gives
the inviscid flow equations at leading order. In this case the highest derivtive term ∇
¯ 2 v̄ is neglected

and consequently not all boundary conditions can be satisfied (usuall the no slip conditon). Thus thin
regions develop where this term must be brought back, which are termed boundary layers.
CHAPTER 3. NONDIMENSIONALISATION 8

3.2.5 Example 5

The one-phase Stefan problem for the temperature T (x, t) in the water phase for a melting ice problem
can be written as
∂T ∂2T
in 0 < x < s(t), t > 0 ρc =k 2, (3.13)
∂t ∂x

at x = 0 T = T0 , (3.14)
∂T ds
at x = s(t) T = 0, −k = Lρ , (3.15)
∂x dt
at t = 0 s = 0, (3.16)

where s(t) denotes the moving ice/water interface, L is the latent heat per unit mass, ρ, c, k are the
density, specific heat and conductivity of the water. The water (liquid phase) occupies the region
0 < x < s(t) and the ice (solid phase) x > s(t). Initially there is no water present i.e. s(0)=0. T0
(> 0) is the temperature of the water at the fixed boundary x = 0.

We nondimensionalise as follows

x = ℓx̄, s = ℓs̄, t = β t̄, T = T0 T̄ ,

where the scales ℓ and β have to be found. The system (3.13)–(3.16) becomes
∂ T̄ k β ∂ 2 T̄
in 0 < x̄ < s̄(t̄), t̄ > 0 = , (3.17)
∂ t̄ ρc ℓ2 ∂ x̄2

at x̄ = 0 T̄ = 1, (3.18)
∂ T̄ ℓ2
Lρ ds̄
at x̄ = s̄(t̄) T̄ = 0, − = , (3.19)
∂ x̄ kT0 β dt̄
at t̄ = 0 s̄ = 0, (3.20)

The governing equation (3.17) suggests

ρc 2
β= ℓ ,
k
and the Stefan condition on the moving interface then suggests introducing the dimensionless param-
eter λ where
Lρ ℓ2 L
λ= = ,
kT0 β cT0
which is commonly termed the Stefan number. The dimensionless problem is thus
∂ T̄ ∂ 2 T̄
in 0 < x̄ < s̄(t̄), t̄ > 0 = , (3.21)
∂ t̄ ∂ x̄2

at x̄ = 0 T̄ = 1, (3.22)
∂ T̄ ds̄
at x̄ = s̄(t̄) T̄ = 0, − =λ , (3.23)
∂ x̄ dt̄
at t̄ = 0 s̄ = 0, (3.24)
CHAPTER 3. NONDIMENSIONALISATION 9

where the length scaling ℓ remains arbitrary; this is not unexpected given that the original problem
had no inherent spatial length scale. The fact that ℓ is not fixed leads to the existence of a similarity
solution of this problem termed the Neumann solution (one of the few explicit solutions that exist for
moving boundary problems).

3.3 Dimensional analysis

The topic of dimensional analysis formalises the procedure of nondimensionalisation. An important

theorem in which is the Buckingham Pi theorem:

If n quantities Q1 , Q2 , . . . , Qn (dependent, independent variables and parameters) involving r separate

fundamental dimensional components (usually r = 3, these being mass, length, time, i.e. M,L,T, but
see the table below) are related by a unique dimensionally consistent function f (Q1 , . . . , Qn ) = 0,
then we can find n − r dimensionless combinations of Qi , say Πj (Q1 , . . . , Qn ), j = 1, . . . , n − r, such
that the solution can be expressed as F (Π1 , . . . , Πn−r ) = 0.

As a first illustration of this theorem, consider Example 2. Let f (θ, t, ℓ, k, g, θ0 , ω0 ) = 0 be the solution
of the IVP (3.1)–(3.2). There are two dimensionless quantities (θ, θ0 ) and n = 5 dimensional quantities
involving r = 2 fundamental dimensions (L,T). Thus the solution can be expressed in terms of θ, θ0
and n − r = 3 dimensionless quantities namely F (θ, θ0 , t̄, K, G) = 0 as shown by (3.3)–(3.4).

As a second illustration of this theorem, consider Example 3. Let f (u−ui , x, t, ρc, k, q, h, ℓ, u0 −ui ) = 0
be the solution of the BVP (3.5)–(3.8). There are n = 9 dimensional quantities involving r = 4
fundamental dimensions (M,L,T,Θ). Thus the solution can be expressed in terms of n − r = 5
dimensionless quantities namely F (ū, x̄, t̄, Q, H) = 0 as shown by (3.9)–(3.12).

As a third illustration, we consider Example 4. Let f (v, p, T − T0 , x, t, ρ, µ, c, k, T1 − T0 , g, L, U ) = 0

be the solution of the dimensionless governing equations. There are n = 17 dimensional quantities
involving r = 4 fundamental dimensions (M,L,T,Θ). Thus the solution can be expressed in terms
of n − r = 13 dimensionless quantities namely F (v̄, p̄, T̄ , x̄, t̄, Re, F r, P e, Br) = 0 as shown by the
dimensionless equations.

As a fourth illustration, we consider Example 5. Let f (T, s, x, t, ρ, c, k, L, T0 ) = 0 be the solution

of the moving boundary problem (3.13)–(3.16). There are n = 9 dimensional quantities involving
r = 4 fundamental dimensions (M,L,T,Θ). Thus the solution can be expressed in terms of n − r = 5
dimensionless quantities namely F (T̄ , s̄, x̄, t̄, λ) = 0 as shown by (3.21)–(3.24).
CHAPTER 3. NONDIMENSIONALISATION 10

Note 1. The function f may be a solution of a PDE, a BVP or IVP.

Note 2. If a PDE involves fewer fundamental dimensions than dimensional quantities, it must admit
a simplified solution in accordance with the Buckingham Pi Theorem.

The International System (SI) of fundamental units are:

Fundamental Dimension Base Unit

Length (L) metre, m
Mass (M) kilogram, kg
Time (T) second, s
Electric current (A) ampere, A
Thermodynamic temperature (Θ) kelvin, K
Amount of substance (X) mole, mol
Luminous intensity (I) candela, cd