Euler equation and Navier-Stokes equation

WeiHan Hsiaoa
a
Department of Physics, The University of Chicago

E-mail: [email protected]

A BSTRACT: This is the note prepared for the Kadanoff center journal club. We review the basics
of fluid mechanics, Euler equation, and the Navier-Stokes equation. The stability of the solution is
discussed by adapting Landau’s original argument.
Contents

1 Introduction 1

2 Ideal Fluid 2
2.1 Continuity Equation 2
2.2 Momentum Equation 2
2.3 Momentum Equations as Conservation Laws 3

3 Viscous Fluid 3
3.1 Equation of Motion 3
3.2 Incompressible Fluid 4
3.3 Energy Dissipation 5
3.4 Law of Similarity 5

4 Stability Condition 6

5 Open Questions 9

6 Summary 10

1 Introduction

In this quarter, we wish to cover the some basics of turbulence in the first 3 talks. As a warm up, the
first talk aims to cover some building blocks in the field of fluid mechanics. Fluid dynamics describes
collective motion of enormous particles macroscopically. Therefore, we no longer use the coordinates
of each individual. Instead, the velocity field v as a function of spacetime is the desired variable.
On top of the velocity, we need 2 thermodynamic variables and the equation of state to complete the
problem. In many cases, we choose the fluid density ρ and pressure p.
In d space dimensions1 , we have (v, ρ, p) up to d + 2 variables. The equation of state provides 1
constraint. To fully solve for them, we need extra d + 1 equations. We will see shortly they are
provided by conservation laws and force balancing equations. In the following sections we are going
to introduce them for different types of fluids mainly following the book by Landau [1]. Also useful
is the classic textbook by Batchelor [2].
The talk and the note is organized as follows. We start will the ideal fluid, in which case dissipation is
absent, to set up the general idea of fluid dynamics. Following the idea, we introduce the viscous term
to the stress tensor. By arguing the general form it can has, we derive the Navier-Stokes equation.
1
In this note, we consider the case d = 2 or 3.

–1–
Basic properties, energy dissipation and law of similarity, are discussed. Following this section, we
pave the path toward turbulence by studying the stability of a viscous, incompressible and steady flow,
giving an estimate of the critical Reynolds number. Then the we will ended with some open problems.

2 Ideal Fluid

2.1 Continuity Equation

The first equation is given by the conservation of particle number. The idea is essentially the same
as what we studied in electromagnetism and we will merely give an one-minute review. Looking at a
given region V , the rate of change for the particle number in V is
Z
∂
dd x ρ. (2.1)
∂t V

Accordingly, the particles flowing out through ∂V is the surface integral

Z Z
da · ρv = dd x ∇ · (ρv). (2.2)
∂V V

Equating (2.1) to −(2.2), we have the continuity equation

∂
ρ + ∇ · (ρv) = 0. (2.3)
∂t
2.2 Momentum Equation
The second hydrodynamic equation is Newton’s second law. In a fluid, in addition to the total applied
force, there is internal momentum transfer due to pressure p, which is
Z Z
− da p = − dd x ∇p. (2.4)

Therefore −∇p is recognized as the force density acting on fluid. According to Newton’s law, it
should be the source of the rate of change of the momentum

dv ∂ 
ρ =ρ + v · ∇ v. (2.5)
dt ∂t
This time derivative accounts for the fact that the fluid particles also move in space. It can actually
be regarded as a kind of derivative, using which one measures the force along the velocity streamline.
The second term v · ∇v, the inertial force, is the connection term owing to moving with the flow.
Identifying the rate of velocity change and pressure, we have

∂v 1 1
+ (v · ∇)v = − ∇p + (· · · ), (2.6)
∂t ρ ρ

where (· · · ) refers to other plausible external forces acting on the fluid with no dissipation.
We note that the continuity equation and Newton’s second law give d + 1 equations. Since there is no

–2–
heat production owing to dissipation. The fluid motion can be considered adiabatic. The equation of
state then is given by isentropic condition
ds ∂s
= + v · ∇s = 0. (2.7)
dt ∂t
An alternative form of conservation can be written in terms of the entropy density ρs.
∂ ∂ρ ∂s
(ρs) = s +ρ = −(s∇ · (ρv) + ρv · ∇s) = −∇ · (ρsv)
∂t ∂t ∂t
∂
⇒ (ρs) + ∇ · (ρsv) = 0. (2.8)
∂t
In this expression, ρsv is interpreted as the entropy flux density.

2.3 Momentum Equations as Conservation Laws

In the previous paragraphs we have derived a set of equations that determine (v, ρ, p) completely. In
this paragraph, we are about to spend some lines on writing the Newton’s second law as a conservation
law in the spirit of ∂t (· · · )+∇·(· · · ) = 0 with the help of equation of continuity. Now the momentum
density ρvi and its temporal partial derivative are given by
∂(ρvi ) ∂vi ∂ρ
=ρ + vi
∂t ∂t ∂t
= − ρ(vj ∂j )vi − ∂i p − vi ∂j (ρvj ) = −∂j (δij p + ρvi vj ) = −∂j Πij . (2.9)

Recall the material in Griffiths chapter 8, Πij is the momentum flux density. In the following, instead
of Euler equation, we will implement this one as we generalize the discussion to viscous fluids.

3 Viscous Fluid

3.1 Equation of Motion

In this section consider the existence of viscosity, or the internal friction between particles constitute
the fluid. Including this effective does not change the conservation of particle number and conse-
quently does not change the continuity equation
∂ρ
+ ∇ · (ρv) = 0. (3.1)
∂t
What is changes is the momentum equation and the proper equation of state. Let us recall the mo-
mentum equation derived in the last section
∂
ideal : ρvi = −∂j Πij , Πij = pδij + ρvi vj . (3.2)
∂t
It represents the reversible momentum transfer owing to the action of pressure and mass. In the
presence of viscosity, an additional term is needed to represent the irreversible momentum transfer
from large |v| to small |v|. We write
0
viscous : Πij → pδij − σij + ρvi vj := −σij + ρvi vj . (3.3)

–3–
The stress tensor σ is responsible for the momentum transfer not directly due to mass, incorporating
conservative part and dissipative part.
A immediate question is that what is the general form of σ 0 in terms vi or its derivative. We want to
argue the general form from 2 physical observations.
The first one is that the internal friction does not manifest in uniform flow. It must be proportional
to ∂i vj , to the first order in derivative expansion. Given such a tensor, we can decompose it into the
trace part, symmetric and traceless part, and the skew symmetric part.
2
∂i vj → δij ∇ · v, ∂i vj + ∂j vi − ∇ · v, and ∂i vj − ∂j vi . (3.4)
d
The second observation is that this effect also disappears under rigid rotation v = Ω × r. It implies
only symmetric terms are retained. Consequently, we have the general decomposition

0
 2 
σij = η ∂i vj + ∂j vi − δij ∇ · v + ζδij ∇ · v, (3.5)
d
where η and ζ are coefficients of viscosity. It can be shown that both of them are positive. Using the
equation of motion ∂t ρ + ∇ · (ρv) = 0, the above equation can be written as
∂ 
0
ρ + v · ∇ vi = −∂i p + ∂k σik . (3.6)
∂t
There is no reason a priori to forbidden η and ζ from being fields. Nonetheless, in the following, we
will be treating them as constants. Therefore, the momentum equation becomes
∂   1 
ρ + v · ∇ v = −∇p + η∇2 v + ζ + η ∇(∇ · v). (3.7)
∂t d
Equations (3.6) and (3.7) are the Navier-Stokes equation.

3.2 Incompressible Fluid

We have modified the momentum equations in the presence of viscosity. Due to dissipation and
the heat produced. There is no reason to assume adiabatic process ds/dt = 0. A model dependent
equation of state has to be proposed to provide with sufficient constraints. For the case we will
investigate in the rest of the talk, we narrow our attention to a certain kind of fluid, the incompressible
flow defined by ρ = const. As such, ρ is no longer regarded as an unknown. The continuity equation
reduces to ∇ · v = 0. The momentum equation becomes
∂  ∇p
+v·∇ v =− + ν∇2 v (3.8)
∂t ρ
with ν = η/ρ being the kinematic viscosity.
To make the equation simpler, we take the curl of the full equation. Note that
1
(v · ∇)v = ∇v 2 − v × (∇ × v) (3.9)
2
⇒∇ × (v · ∇)v = −∇ × [v × (∇ × v)]. (3.10)

–4–
Thus, we arrive at
∂
(∇ × v) = ∇ × [v × (∇ × v)] + ν∇2 (∇ × v). (3.11)
∂t
On the other hand, taking the divergence of the equation gives us a Poisson type equation

−∇2 p = −ρ∇ · (v · ∇)v = ρ∂i ∂k (vk vi ). (3.12)

Using (3.11), the velocity profile can be solved. Plugging the profile into (3.12), we can solve for the
variable p.
To completely determine the solution, boundary conditions are required. Contrast to the ideal fluid,
here we demand v = 0 over the boundary owing to internal friction.

3.3 Energy Dissipation

Next we discuss the rate at which the energy is dissipated. The kinetic energy of the fluid in a region
V is given by
Z
1
E= dd x ρv 2 . (3.13)
2 V
∂ 2
For incompressible fluid, the rate of change Ė is proportional to ∂t (v ). Dotting (3.6) by v,
1∂ 2 0
ρ v = −ρv · (v · ∇)v − v · ∇p + vi ∂j σij . (3.14)
2 ∂t
0 = ∂ (v σ 0 ) − σ 0 ∂ v . Using ∇ · v = 0, v v ∂ v = 1 v ∂ v 2 = 1 ∂ (v v 2 ) and v ∂ p =
vi ∂j σij j i ij ij j i i i i j 2 i i 2 i i i i
∂i (vi p). Thus,
Z Z i Z
ρ ∂ h 1 p
Ė = dd x v 2 = − da · ρv + − v · σ0 − 0
dd x σik ∂k vi . (3.15)
2 ∂t ∂V 2 ρ V
0 = η(∂ v + ∂ v ),
As we push ∂V to infinity and write σik i k k i
Z
η
Ė = − dd x (∂i vj + ∂j vi )2 . (3.16)
2 V

Since Ė < 0 and the integrand is positive definite, physically we see η > 0.

3.4 Law of Similarity

Let us think about what we can say using the scaling property of the Navier-Stokes equation. In this
section we give the argument for steady and incompressible flows and look at the problems in which
the fluids flow past solid bodies of the same shape with different length scales.
Under this circumstance, the only parameter appearing in the equation of motion is the kinematic
viscosity ν. The main stream has a characteristic velocity u and the boundary conditions give a
characteristic linear dimension `. Their dimensions are
L2 L
[ν] = , [u] = , [`] = L. (3.17)
T T

–5–
From these informations we can define a dimensionless parameter, the Reynolds number, as

u`
R= . (3.18)
ν
Other dimensionless quantities are essentially functions of R. R can be given a direct physical inter-
pretation as we write

u` u2 /`  v · ∇v 
R= = ∼ , (3.19)
 
ν νu/`2 ν∇2 v

which can interpreted as the competition between the inertial force, the non-linear term, and the
viscous force.
It turns out that we can measure v and coordinates x in units of u and ` respectively. Formally
v
= v(x/`, R). (3.20)
u
The conclusion is if 2 incompressible and steady flows with the same Reynolds number flowing past
similar solid bodies, they flows are then given by the same function of x/`. The same implication
applies to the ratio p/(ρu2 ) and other physical quantities. This is the law of similarity.
Generalizing this law as more dimensionful physical quantities come into play in possible. For exam-
ple, as gravity is significant in the problem of interest, another dimensionless parameter, the Froude
number,

u2
F= (3.21)
`g
would be defined beside the Reynolds number. Flows are similar as they have the same Reynolds
number and Froude number.
It can even be generalized to non-steady flows, in which a time scale is in the game. For instance, as
we immerse an oscillating solid body into the fluid, its inverse oscillation frequency τ introduces a
time scale. As such, the so-called Strouhal number is defined as
uτ
S= . (3.22)
`
Flows with the same S and R exhibit similar motion.
To end this section we note that the story is quite different for ideal fluids, whose η = 0. In those
cases, the Euler equation is essentially scale invariant.

4 Stability Condition

Either Euler equation or Navier-Stokes equation is nonlinear. In particular, there is a parameter ν, or

equivalently, R, in the Navier-Stokes equation, which we can tune and drive the equation from linear
regime to nonlinear one. In the extremely viscous case ν → ∞, or R → 0, the equation is dominated
by dissipation and linear. Perturbative study can be carried out. As we look at a range of R, we may

–6–
Figure 1. This figure depicts the stream lines of fluids of different R flowing past a cylinder. Going from
small R to large ones, we see the stream lines go from Laminar to turbulent. This figure is adopted from Ref.
[5]. To be more elaborated, as R ∼ 10, vortices start forming. As R → 100, those vortices form vortex streets.
Turbulent phenomenon occurs as R reaches 104 .

first look at their phenomenologies illustrated in figure 1.

From figure 1, we understand as R increases beyond a threshold, the fluid becomes turbulent, which
roughly speaking is the chaotic behavior of fluid. Here we try to establish this picture by examining the
fate of a little deviation from a given steady solution v0 (x). In our experience, owing to nonlinearity,
a little deviation added to a solution may lead to dramatic change in its nature, and leads to the
investigation of stability.
We write

v(t, x) = v0 (x) + δv(t, x) (4.1)

p = p0 + δp. (4.2)

To first order in δ, we have

∂
δv + (v0 · ∇)δv + (δv · ∇)v0 = −ρ−1 ∇δp + ν∇2 δv, ∇ · δv = 0. (4.3)
∂t

–7–
At this order δv satisfies a linear equation. It turns out we can look at the normal modes δv ∼ e−iωt
and the flow is said to be stable if Im[ω] < 0.
More precisely, we say as R < Rc , the critical Reynolds number, all ω’s have negative imaginary part.
At the critical point, one ω becomes real, and as it slightly goes larger than Rc , its imaginary part
becomes positive. We write this frequency as

ω = ω 0 + iω 00 , ω 00  ω 0 (4.4)

and the deviation assumes the form

00 0
δvω = A(t)f (x), A(t) ∝ eω t e−iω t . (4.5)

We comment that this expression holds true at small time. If we are to consider the evolution of the
magnitude averaged over a period (ω 00 )−1  τ  (ω 0 )−1 , to the leading order, we would naively
d
write dt |A|2 = 2ω 00 |A|2 . Landau conjectured a subsequent term

d
|A|2 = 2ω 00 |A|2 − α|A|4 , (4.6)
dt
where α is the Landau constant which can be positive or negative. Let us argue for the case that
ω 00 > 0 and α > 0, |A|2 assumes a finite limit value and therefore the solution is not stable. Solving
the above equation, we obtain
1 α 00
2
= 00
+ const × e−2ω t (4.7)
|A| 2ω

and |A|2 has an asymptotic value 2ω 00 /α. It may concerns one that as we plug the asymptotic value
back into the differential equation, each term on the right-hand of (4.6) is of the same order. Conse-
quently, it seems not reasonable to drop β|A|6 or any other higher order terms. A more proper way
we should imagine this process is that equation is valid at small time. As |A|2 increases such that the
existing 2 terms are of the same order, the next order should be included self-consistently. Owing to
the assumption that ω 00 is slightly greater than 0 as R exceeds Rc , if we are slightly above the critical
value, we see that
p
lim |A| ∝ (R − Rc ). (4.8)

It is a reminiscent of Landau’s theory of phase transition. To end this section, we provide an estimate
for the critical value Rc . Suppose we do not assume the smallness of δv. The fully expanded equation
reads
∂ 1 1
δv + (v0 · ∇)δv + (δv · ∇)v0 + (δv · ∇)δv = − ∇p + ∇2 δv. (4.9)
∂t ρ R
Let us dot this equation by δv.
∂ i
δv i δv + δv i v0j ∂j δv i + δv i δv j ∂j v0i + δv i δv j ∂j δv i = −ρ−1 δv i ∂i p + R−1 δv i ∂j ∂j δv i . (4.10)
∂t

–8–
 1 ∂ 2
The first term becomes 2 ∂t δv . Moving other terms to the right-hand side, and using, when assuming
∂i δv i = ∂i v0i = 0,
1
δv i v0j ∂j δv i = ∂j (v0j δv 2 ) (4.11)
2
1
δv δv ∂j δv = ∂j (δv j δv 2 ),
i j j
(4.12)
2
we have
1∂ 2
δv = −δv i δv j ∂j v0i − R−1 ∂j δv i ∂j δv i
2 ∂t
 1 
+ ∂j R−1 δv i ∂j δv i − ρ−1 δv j p − δv 2 (v0j + δv j ) . (4.13)
2
Integrating the whole equation over the domain, the total derivative vanishes owing to boundary con-
ditions. The left-hand side is the rate of change of the kinetic energy of perturbation.
Z Z
∂ d i j i 1 D
δE = − d x δv δv ∂j v0 − dd x tr[∇v(∇v)T ] := T − . (4.14)
∂t R R
It provides with us a lower bound for R. If
D D
R < min ≤ , (4.15)
T T
we see ∂t δE < 0 strictly and thus it flow is stable.

5 Open Questions

In the previous section, we have seen the stability of the solution fails as the Reynolds number ex-
ceeds a critical value. In particle, the limit of R → ∞ is identified as fully developed turbulence. It
corresponds to, formally, ν → 0 at fixed u` and it will be the theme in the next talk.
Over the previous sections, we discuss the physical essence of the Euler and Navier-Stokes equations.
In this section we would like to list some mathematical problems that remain open. To motivate some
of them, we note that the fully developed turbulence features irregular behaviors of the solutions to
Navier-Stokes equation. The following are some long standing problems, existence and smoothness
of Navier-Stokes solutions on R3 or T3 , and the breakdown of these solutions on R3 or T3 , given a
specific class of initial data and constraints such as smoothness and energy bound [4]. Partial progress
has been made by Leray for weak solutions. Quantitatively, a function v is a weak solution to the
incompressible Navier-Stokes equation if
Z
d
 ∂ 2

d x v· + v · (v · ∇) + νv · ∇ + p∇ · φ = 0, (5.1)
∂t
Z
dd x v · ∇ψ = 0, (5.2)

where φ (subject to ∇ · φ = 0) and ψ are compactly supported test functions. Leray has shown in
existence of weak solutions (v, p), whereas the uniqueness is yet to be known.

–9–
6 Summary

To summarize, in this talk we have covered the basic hydrodynamic equations, Euler equations and
Navier-Stokes equation. In particular, we discuss the stability of solutions as we tune the Reynolds
number. We saw from qualitative study that chaotic behavior appears as R exceed a critical value, and
analyzed the story using Landau’s argument. Finally we point out mathematically the Navier-Stokes
equation has a lot of problems to be solved, which is a reason that this equation draws interest across
different fields in science.
The author also thank for the comments and questions received during the talk, which help the author
to refine the existing note.

References

[1] L. D. Landau and E. M. Lifshitz, Fluid Mechanics, Course of Theoretical Physics Volume 6. 2nd Ed.
(1987).
[2] G. K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press (1967).
[3] Denis Bernard, Turbulence for (and by) Amateurs, arXiv:cond-mat/0007106.
[4] C. L. Fefferman, Existence and Smoothness of the Navier-Stokes Equation.
[5] F. van den Bosch, Physical Processes in Astronomy, Summary 12.

– 10 –