A Derivation of the Navier-Stokes Equations

Neal Coleman

Neal Coleman graduated from Ball State in 2010

with degrees in Mathematics, Physics, and Economics.
He is pursuing a PhD in Mathematics at Indiana
University, Bloomington. This paper is based on a
project he did in a PDE class with Dr. Mohammed.

Introduction
The Navier-Stokes equations are a set of second-order partial differential equa-
tions relating first and second derivatives of fluid velocity, which is represented
as a smooth vector field. While simple in principle, they are enormously dif-
ficult to solve; in fact, no proof has yet been found guaranteeing even the
existence of a smooth solution in just three dimensions. Annoying as this is
to mathematicians, the consequences of a lack of solution are far-reaching: the
Navier-Stokes equations govern continuum phenomena in all areas of science,
from basic hydrodynamical applications to even cosmology.

Preliminaries
To understand and appreciate the Navier-Stokes equations, one must first be
familiar with some of the basic concepts of fluid dynamics. We begin with the
distinction between intensive and extensive properties.
An intensive property of a fluid is a property whose value does not depend
upon the volume of measurement. For instance, pressure, density, momentum,
and velocity are intensive properties. Mass, volume, and surface area are not.
An extensive property of a fluid is a property whose value does depend upon
the volume of measurement. (That is, it is a property that is not intensive.)
Intensive properties are useful to us because they can be evaluated meaningfully
and generally on differential elements.
The second concept we require is the continuum hypothesis (not related to
set theory). Because we are dealing with a macroscopic fluid, we ignore that
the fluid is composed of zillions of discrete particles and thus assume that the

20 B.S. Undergraduate Mathematics Exchange, Vol. 7, No. 1 (Fall 2010)

properties of the fluid may be described by continuous (actually, often C ∞ )
functions.
As a matter of fact, if we “zoom in” far enough, we find that intensive prop-
erties become ill-defined or discontinuous; since temperature, for instance, is
the sum of the kinetic energies of the particles of a material, at the atomic scale
temperature becomes a discrete function of position. But in the macroscopic
limit, because of the relatively high particle density, we may, and do, regard
temperature (and other intensive properties) as continuous functions.
The continuum hypothesis is relevant not just to continuum mechanics,
but to all macroscopic physics that deal with continuous systems, including
electrodynamics, thermodynamics, materials science, and hydrodynamics.
The Navier-Stokes equations deal chiefly with the relationship between the
intensive properties pressure and velocity. Pressure p may be considered a C ∞
function p : Rn × [0, ∞) → (0, ∞), and velocity u a C ∞ vector field over Rn
that varies with time. Generally, for physically applicable situations, we take
n = 2 or 3.
We require a formulation of several conservation laws. Conservation laws are
essential to physics; they permit us to use the constancy of various quantities to
relate systems at different points in time. We shall need the law of conservation
of mass and the law of conservation of momentum. In particular, the law of
conservation of mass states that, given an isolated system, the amount of matter
present in the system remains constant over time. Alternatively, mass may be
neither created nor destroyed.
Let us apply the law of conservation of mass to a fluid. Consider an arbitrary
volume V . Let dA be a surface element on the surface ∂V of V . On dA, the
density ρ of the fluid is constant, as is the velocity u. If ηA is the outward unit
normal to dA, then in a time dt, the volume of fluid that flows through dA is
u · ηA dAdt, so the mass that leaves through dA is
dM
dM = ρu · ηA dAdt, or = ρu · ηA dA.
dt
Integrating over the surface, we have the entire mass flux across ∂V equal to
the rate of change of the mass within V :
dM
ZZ ZZZ
=− ρu · ηA dA = − (∇ · (ρu))dτ,
dt ∂V V

using the divergence

RRR theorem. RRR
But M = V
ρdτ , so dM/dt = V
dρ/dtdτ . This implies
ZZZ  
∂ρ
0= + ∇ · (ρu) dτ.
V ∂t
This is true for every V ⊆ Rn , so we have that everywhere
∂ρ
+ ∇ · (ρu) = 0.
∂t
This is known as the mass continuity equation; it describes the movement of
mass through a continuous fluid.

B.S. Undergraduate Mathematics Exchange, Vol. 7, No. 1 (Fall 2010) 21

 When dealing with fluids, it is often useful to assume incompressibility:
namely, that the density of a fluid is constant. In other words, given a container,
it is impossible to ever pack more fluid into it or take fluid out without changing
the volume. This is equivalent to saying that ∇·u = 0; using the mass continuity
equation, if density is constant then we have ∇ · (ρu) = ρ∇ · u = 0; ρ > 0, so
∇ · u = 0.
Stress and body forces are the two other important concepts we shall need.
Body forces are, generally, forces per unit volume. They may be characterized
by long-range bulk forces, such as gravity or the electromagnetic forces, and
internal forces, which are caused by internal stresses induced by viscosity. A
stress is merely an internal force acting on an imaginary surface within a body;
it is described in units of force per unit area. The simplest example is the stress
within a body subject to an axial uniform external force; for example, a pillar.
The stress then is merely the total load divided by the cross-sectional area of
the pillar. More generally, stresses vary from point to point within a body, and
at a point are different for different planar slices through that point.
For a body, the stress at a point acting on a planar surface can be decom-
posed twice: the stress vector, into n components, and the surface into n planes
each perpendicular to an axis. The ith component of the stress T on a surface
with unit normal n = (nj ) can therefore be described by Ti = σij nj , where σij
is a second-degree tensor known as the stress tensor.
Finally, there are several conditions that must be imposed on solutions
of the Navier-Stokes equations over Rn in order to ensure physical realism.
They must not blow up as one moves to infinity, so the initial condition – a
divergence-free C ∞ vector field u0 (x) – must be either spatially periodic or be
bounded on all of Rn . Moreover, the kinetic energy R of the fluid, which depends
on the square of the speed, must be bounded: |u(x, t)|2 dτ < C for all t > 0
and some positive constant C.

Derivation
The derivation of the Navier-Stokes can be broken down into two steps: the
derivation of the Cauchy momentum equation, an equation governing momen-
tum transport analogous to the mass transport equation derived above; and
the linking of the stress tensor to the rate-of-strain tensor in order to simplify
the Cauchy momentum equation.

Cauchy Momentum Equation

We consider an incompressible, viscous fluid filling Rn subject to an external
body force f described as a time-variant vector field f : Rn × [0, ∞) → Rn . Let
fi denote the components of force. We begin the derivation of the Navier-Stokes
equations by first deriving the Cauchy momentum equation.
Consider a volume element dτ in Rn . The total body force acting in the xi
direction on dτ is due to fi and to forces caused by the stress tensor σij . Let
us consider these stress forces. At x, the component σij of the stress tensor
represents the force per unit area in the xi direction acting at a point on a

22 B.S. Undergraduate Mathematics Exchange, Vol. 7, No. 1 (Fall 2010)

plane cut through Rn normal to the xj direction. There are n components of
the stress tensor pointing in the xi direction: σij , j ∈ {1, . . . , n}. Each of these
stresses varies over the element dτ in the xj direction by some small amount;
call it ∂σij . Since dτ has side lengths of dxi , the rate of stress variation is thus
∂σij /∂xj . This represents a body force on dτ .
Therefore, in the xi direction, the total force (body force multiplied by
volume) on dτ is:
X ∂σij
Fi = fi dτ + dτ.
j
∂xj

The momentum of the above differential volume element in the xi direction

is Pi = ρui dτ ; we use Newton’s second law, F = dP/dt, and differentiate
momentum, using the chain rule and noting that each xi is a function of time
and that uj = dxj /dt:
 
dPi d ∂ui
X ∂u i
= (ρui dτ ) = ρ  + uj dτ.
dt dt ∂t j
∂xj

Finally, equating the two expressions for total force on dτ , we have:

 
X ∂σij ∂u i
X ∂u i
fi dτ + dτ = ρ  + uj dτ ;
j
∂xj ∂t j
∂xj

letting Ω be an arbitrary volume in Rn and integrating,

   
Z X ∂σij Z
∂u i
X ∂ui
fi +  dτ = ρ + uj  dτ,
Ω j
∂xj Ω ∂t j
∂xj

or    
Z X ∂σij ∂u ∂u
 − ρ i + i 
X
fi + uj dτ = 0.
Ω j
∂xj ∂t j
∂xj

Since the integrand is continuous and Ω is arbitrary, we have the Cauchy mo-
mentum equation:
 
X ∂σij ∂u i
X ∂ui
fi + = ρ + uj .
j
∂xj ∂t j
∂x j

Stress and Rate of Strain

As it stands, the problem is intractable: we have no idea what σij is! But,
turning to physical intuition, we have motivation for further assumptions: the
stresses on the volume element should be related to the velocity gradient. For
example, consider a simple stream of water where the flow is unidirectional
and the speed of flow is related to distance from the bottom. If we consider a

B.S. Undergraduate Mathematics Exchange, Vol. 7, No. 1 (Fall 2010) 23

volume element at some height y from the bottom and follow it as it flows, we
will notice it begin to shear as the top flows more quickly than the bottom.
This gives some physical basis to the definition of a Newtonian fluid : one
where the stress on a volume element is linearly related to the velocity gradient.
In the simple example above, we express this relationship as

du
σ=µ ,
dy

where µ is a proportionality constant known as viscosity.

More generally, consider a fluid element dτ in Rn . The stresses acting on it
can be broken down into two components: a normal uniform stress, known as
pressure, which is the average of all of the normal stresses
P
σii
p=−
n

(it is negative by convention); and a deviatoric stress τij = σij −pδij , composed
of the non-normal stress components and the deviation of the normal stresses
from the pressure. The deviatoric stress is responsible for the deformation of
the volume element, and is therefore related to the velocity gradient.
Now we concern ourselves with figuring out what the velocity gradient ac-
tually looks like. Consider a point q = (qi ) ∈ Rn and the associated velocity
(ui (q)). A small distance dq away from q, the velocity will have changed by
some small quantity du given by the rule
X ∂ui
dui = dq · ∇ui = dxj .
j
∂xj

We can rewrite this as

X 1  ∂ui ∂uj
 X 1  ∂ui ∂uj

dui = + dxj + − dxj ,
j
2 ∂xj ∂xi j
2 ∂xj ∂xi

or
1X
dui = (ij + ωij )dxj .
2 j

Here we set ij = (∂ui /∂xj + ∂uj /∂xi )/2 and ωij = (∂ui /∂xj − ∂uj /∂xi )/2;
the former is known as the rate of strain tensor and the latter as the vorticity
tensor. The rate of strain tensor is responsible for the deformation of the
volume element dτ ; the vorticity tensor is responsible for the rotation of the
element. This can be seen by an examination of the geometry of the element
at the endpoints of a time interval (t, t + dt).
Here, we apply the assumption that our fluid is Newtonian: the deviatoric
stress tensor is proportional to the rate of strain tensor, or τij = 2µij . This
permits us to simplify the Cauchy momentum equation, and in particular the

24 B.S. Undergraduate Mathematics Exchange, Vol. 7, No. 1 (Fall 2010)

second term on the left side:
X ∂σij X ∂
= (τij + pδij )
j
∂xj j
∂xj
X ∂ ∂p
= (2µij ) +
j
∂xj ∂xi
 
∂p X ∂ ∂ui ∂uj
= +µ +
∂xi j
∂xj ∂xj ∂xi
!
∂p X ∂ 2 ui ∂ 2 uj
= +µ +
∂xi j
∂x2j ∂xj ∂xi
∂p ∂ X ∂uj
= + µ∇2 ui + µ
∂xi ∂xi j ∂xj
∂p ∂
= + µ∇2 ui + µ ∇ · u = ∇p + µ∇2 ui ,
∂xi ∂xi
using, variously, the continuity of second derivatives and the fact that our fluid
is incompressible.
Setting density to 1 for convenience and substituting back into the momen-
tum equation, we finally have the Navier-Stokes equations for an incompressible
Newtonian fluid :
∂ui X ∂ui ∂p X ∂uj
+ uj = fi + µ∇2 ui − ; = 0.
∂t j
∂xj ∂xi j
∂xj

One can also write the above Navier-Stokes equations more succinctly and
in more familiar form as follows.

∇·u=0 Mass Continuity Equation,

∂u
+ u · ∇u = f + µ∇2 u − ∇p Cauchy Momentum Equations.
∂t

Explanation and Significance

These equations are thought to govern the motion of all fluids; they thus are
essential to the physics of continuum mechanics. Practically every area of sci-
ence that makes macroscopic approximations at least touches on fluid dynamics
– from cosmology and the motion of galaxies to nuclear fusion and the con-
finement of plasma, from electrodynamics and the flow of charges to aerospace
engineering and the design of airplanes, the Navier-Stokes equations are vital
to the understanding of the universe.
Yet, it is not known whether physically realistic smooth solutions are guar-
anteed to exist on all of R3 ! The Clay Mathematics Institute has identified
this question as one of the top challenges facing mathematicians and physicists
in the first several decades of the twenty-first century, and has thus issued a

B.S. Undergraduate Mathematics Exchange, Vol. 7, No. 1 (Fall 2010) 25

$1,000,000 prize to any individual who can provide proof of, or counterexample
to, the existence of smooth solutions to the form of the Navier-Stokes equations
here derived.
More specifically, the challenge requires a mathematician to prove that any
acceptable (periodic or bounded) initial condition to the Navier-Stokes equa-
tions has a solution, or that there is an acceptable (periodic or bounded) initial
condition for which the Navier-Stokes equations do not admit a solution.
The solution of the Navier-Stokes equations is, as one might expect, very
difficult, and will require new insight into the very foundations of partial dif-
ferential equations. It is crucial, then, that we continue to advance in this area
of mathematics and physics. For more discussion on the problem we refer the
reader to the reference [1].

References
[1] The Millennium Prize Problems
http://www.claymath.org/millennium/Navier-Stokes_Equations/

26 B.S. Undergraduate Mathematics Exchange, Vol. 7, No. 1 (Fall 2010)