Week 9 lectures

1. Isentropic 1-D Gas Dynamics Equations

In one space dimension, the equations for isentropic fluid dynamics
(5.34)-(5.35) of week 8 notes become
(1.1) ρt + uρx + ρux = 0
1
(1.2) (ut + uux ) + px = 0
ρ
(1.3) St + uSx = 0,
where from equilibrium
thermodyanmics, p = p(ρ, S) is considered
∂p
known. Since dp = ∂S ρ
dS + c2 (ρ, S)dρ, it follows from (1.3) and
(1.1) that
(1.4) pt + upx = c2 (ρt + uρx ) = −c2 ρux
Between (1.2) and (1.4), we seek to take linear combination of the two
equations to find suitable characteristic variables. If we multiply (1.2)
by ρc and add/subtract from (1.4), then we obtain
(1.5) (pt + (u + c)px ) + ρc (ut + (u + c)ux ) = 0

(1.6) (pt + (u − c)px ) − ρc (ut + (u − c)ux ) = 0

Equations (1.5)-(1.6) in addition to (1.3) are the complete set of hyper-
bolic equations for determination of p, u and S is characteristic form
since we may write them as following set of ODEs on characterics
dp du dx
(1.7) + ρc = 0 , on =u+c
dt dt dt
dp du dx
(1.8) − ρc = 0 , on = u−c
dt dt dt
dS dx
(1.9) = 0 , on =u
dt dt
Note the characteristic speeds are u, u±c. The latter two are associated
with sound waves when fluid motion is present. If we linearize the
nonlinear equations (1.7)-(1.9) about a quiescent state u = 0 with
uniform density ρ0 , we obtain
dp du dx
(1.10) + ρ0 c0 on = c0
dt dt dt
dp du dx
(1.11) − ρ0 c0 on = −c0
dt dt dt
1
2

dS dx
(1.12) = 0 on =0
dt dt
Integrating (1.12),
(1.13) S − S0 = H(x)
In the case of a homentropic flow, H(x) = 0. In that case, integration
of (1.10) and (1.11) results in
(1.14) p − p0 + ρ0 c0 u = F (x − c0 t) ,

(1.15) p − p0 − ρ0 c0 u = G(x + c0 t) ,
implying
1
(1.16) p − p0 = (F (x − c0 t) + G(x + c0 t))
2

1
(1.17) u= (F (x − c0 t) + G(x + c0 t))
2ρ0 u0
Returning to nonlinear equations (1.7)-(1.9), if we have a homen-
tropic flow, then p = p(ρ), c2 = p′ (ρ) and it follows that
Z ρ
c(ρ′ ) ′

d dx
(1.18) dρ + u = 0 , on =u+c
dt ρ′ dt

ρ
c(ρ′ ) ′
Z 
d dx
(1.19) dρ − u = 0 , on =u−c
dt ρ′ dt
Recall for that for a gas p = Cργ and so, c2 = γCργ−1 ; using this (1.18)
and (1.19) implies
 
d 2c dx
(1.20) + u = 0 , on =u+c
dt γ − 1 dt

 
d 2c dx
(1.21) −u = 0 , on = u−c
dt γ−1 dt
and we have on integration the Riemann invariants
2 dx
(1.22) c ± u = constant , on =u±c
γ−1 dt
 3

2. Case of a moving piston

The equations (1.22) simplify further if initial conditions are such
that one of the Riemann invariants is trivial, as it can be used in the
other to turn it into a first order nonlinear PDE for one scalar variable.
This is the case for the piston problem as described below. When such
simplifications are not possible, the Riemann invariants (1.22) can still
be used to compute or analyze solutions.
Consider an initially quiescent gas u = 0 in an semi-infinite cylindri-
cal container where the piston boundary is at x = X(t), with X(0) = 0.
We assume initial density and entropy to be ρ = ρ0 , S = S0 , each of
which are constants. The domain in t − x plane is shown in Figure 1.

t
dx /dt =u+c

x=X(t)
dx/dt =u−c x=t c
0

u=0,S=S0 , ρ=ρ x
0

Figure 1. Two sets of characteristics shown in the t − x domain

to the right of a moving piston x = X(t)

On the set of characteristics dx

dt
= u − c, which intersects the piston
x = X(t) as well as the x-axis, where u = 0, c = c0 ,
2c 2c0
(2.23) −u=
γ−1 γ−1
This can be used to eliminate c completely from the equations. There-
fore, on the characteristic set
dx γ+1
(2.24) = u + c = c0 + u,
dt 2
we obtain from the Riemann invariant
2c 2c0
(2.25) + u = 2u + = constant
γ −1 γ−1
or equivalently
 
γ+1
(2.26) u t + c0 + u ux = 0
2
4

In the region in the t − x plane region, where characteristic set (2.24)

intersects x = X(t), (2.25) implies
(2.27) u=u = Ẋ(τ )
piston

where t = τ is the initial value of t where the set of characteristics

(2.24) intersects intersects x = X(t). Using (2.27), integration of (2.24)
results in
 
γ+1
(2.28) x = X(τ ) + c0 + Ẋ(τ ) (t − τ )
2
If and when it is possible to invert (2.28) to solve τ = τ (x, t) it produces
a classical solution to the PDE (2.26) in the form
(2.29) u = Ẋ (τ (x, t)) ,
By taking partial derivative of (2.28) with respect to τ , the implicit
function theorem condition of inversion is:
γ−1 (γ + 1)
(2.30) − Ẋ(τ ) − c0 + Ẍ(τ ) (t − τ ) 6= 0
2 2
For a piston moving to the left, i.e. Ẋ < 0, we still have − γ−1
2
Ẋ(τ ) −
c0 < 0 when the piston speed is not comparable to sound speed. Under
these conditions, if Ẍ ≤ 0, then (2.30) is satisfied since each term is
< 0. In the special case when Ẋ = −V = constant, then solution in
γ+1


the region x < c0 − 2 V t is simply
(γ − 1)
(2.31) u(x, t) = −V0 , c = c0 − V
2
and for x > c0 t,
(2.32) u(x, t) = 0 , c(x, t) = c0
In the intermediate region c0 t ≥ x ≥ c0 − γ+1


2
V t, we have a simple
wave solution
2 x 
(2.33) u(x, t) = − c0
γ+1 t
which may be verified to be a solution of the PDE (2.26) for t > 0
by direct substitution. These are continuous solutions. However, if
Ẍ(τ ) > 0 for any τ , then it is easily seen inversion condition (2.30)
will not always be valid. This gives rise to characteristic curves (2.28)
intersecting, which corresponds to classical solution developing singu-
larities. Beyond such time, physically reasonable solutions are ones
where one allows solutions to undergo jumps, i.e. we allow for weak
solutions.
 5

2.1. Shock Waves. When waves break, i.e. singularities form in clas-
sical solution, then the inversion condition (2.30) becoming invalid. In
that case, we must return to the basic derivation of conservation of
mass momentum and energy for x ∈ (x1 , x2 ) to determine additional
conditions that determine where to place one or more shocks i.e. dis-
continuities across which ρ, u and S may jump. We study these re-
lations in one space dimension for simplicity, though the idea is much
more general. Conservation of mass from week 1 notes implies
d x2
Z
(2.34) ρdx + [ρu]xx21 = 0 ,
dt x1
Conservation of momentum in the absence of body forces implies
  x2
d x2

2 ∂u
Z
 2 x2
(2.35) ρudx + ρu + p x1 = µ+λ ,
dt x1 3 ∂x x1
Conservation of energy implies
(2.36) x2    x2
d x2 1 2
Z     
1 2 dT 2 du
ρu + ρE dx+ ρu + ρE u + pu = k + µ+λ
dt x1 2 2 x1 dx 3 dx x1
where [.]xx21 is the evaluation of the quantity at x1 subtracted from eval-
uation at at x2 . The terms on the right in (2.35) and (2.36) are due to
molecular diffusion effects and can be ignored outside of a shock region.
If x = Xs (t) denotes the location of a shock where there is discontinuity
of flow quantities, then by taking x2 = Xs (t) + ǫ and x1 = Xs (t) − ǫ
and taking the limit of ǫ → 0+ , with U = Ẋs , while ignoring molecular
effects, it follows that
(2.37) − U [ρ] + [ρu] = 0 ,

− U [ρu] + ρu2 + p = 0 ,
 
(2.38)
    
1 2 1 2
(2.39) − U ρu + ρc + ρu + ρE u + pu = 0 ,
2 2
One avoids the problem of overlapping characteristics for the case Ẍ >
0 by enforcing conditions (2.37)-(2.39) across the two sides of a shock
x = X(τ ); away from the shock one uses the classical solution such as
the one obtained in (2.29). In this context, it is useful to note that
p
for a polytropic gas under conditions of isentropic flow, E = (γ−1)ρ . We
avoid going through any more details here, though there are excellent
texts on the subject of shocks (see for instance Whitham, Linear and
Nonlinear Waves, Wiley).
6

3. Incompressible free boundary problems: Motion of

Bubbles
Thus far, we have discussed incompressible flow in a fixed domain Ω.
There are many instances, when this is not appropriate. For instance
in the motion of a bubble, the boundary itself evolves in time. The
same is true for water waves. On a free boundary, such as between two
fluids, or between fluid and vacuum, the no-slip condition u = v, used
for a solid boundary, is no longer appropriate. Consider first the case
of a fluid boundary ∂Ω with another fluid of negligible viscosity and
pressure variation, i.e. vacuum conditions. Since an infinitesimal free
boundary has infinitesimal mass, the forces on the two sides must be
in balance. This corresponds to(1)


(3.40) − pni + 2µTij nj = −pni + µ ∂xj ui + ∂xi uj nj = −σκni
where σ is the surface tension and κ the curvature for in 2-D, and mean
curvature in 3-D. Note T is the viscous stress tensor and (3.40) is a
vector relation. It is called the Stress condition. By taking the dot
product of relation (3.40) with n, we obtain
(3.41) − p + ni Ti,j nj = −σκ , on∂Ω
By taking dot product with respect to a tangent vector τ , tangent to
the interface for which τ · n = 0, we obtain from (3.40)
(3.42) τi Ti,j nj = 0 or τ · T · n = 0
Equations (3.41) and (3.42) are mathematically equivalent to (3.40)
since (3.42) is true for any tangent vector τ . Physically, (3.41) implies
that that the normal component of stress across a free surface is bal-
anced by surface tension effects, while (3.42) is a statement that there
is no tangential stress.
It is to be noted that (3.40) or its equivalent form (3.41)-(3.42) is
only valid when there is fluid motion only on one side of the domain
∂Ω. This is not valid when there is fluid motion on both sides of the
interface, as it is for strong winds blowing on top of an ocean surface.
Generalization of (3.41) and (3.42) in those cases is that the jump in
normal stress across two sides of ∂Ω is
[−p + n · T · n] = σκ
where as the jump in tangential stress across two sides of ∂Ω is
[τ · T · n] = 0

(1)Here we are returning to dimensional quantities

 7

Further, unlike the case of a fixed boundary, we need an additional

equation to determine the location of the free boundary as part of the
problem. This is determined by setting
(3.43) u · n = Vn
where Vn is the normal velocity of the interface. For a free boundary
described implicitly by a scalar relation F (x, t) = 0. This relation is
found by noting that a point on the free surface can be characterized
by x = X(t). Then F (X(t), t) = 0; therefore, taking time derivative
Ft + Xt · (∇F ) = 0 = Ft + Vn |∇F |, since by definition Vn is the normal
component of surface motion, and therefore Xt · n = Vn . This implies
Ft
that (3.43) may be replaced by u · n = − |∇F |
, or
(3.44) Ft + u · (∇F ) = 0 , on F (x, t) = 0
This is called the kinematic condition.
3.1. Inviscid Irrotational Free boundary. Consider the simplest
case, when the flow is inviscid, i.e. viscosity effects are neglected.
Further, we assume that the flow is irrotational. Then the equations
simplify. In Ω, we have
(3.45) ∆Φ = 0
Then, since Bernoulli equation is valid everywhere in Ω, we have
p 1
(3.46) ∂t Φ + + V (x) + |∇Φ|2 = 0,
ρ 2
where we assumed body force b = −∇V . In the case of gravity V =
gx3 , gravity being alligned in the negative x3 -axis Using this in the
pressure equation (3.41), we obtain after noting that viscous stress
tension T = 0 here, we obtain on ∂Ω
1
(3.47) Φt + V (x) + |∇Φ|2 = σκ
2
Equation (3.45) in Ω, together with pressure condition (3.47) and kine-
matic boundary condition (3.44) completely specifies the free boundary
problem if Ω is finite. However, if ∞ ∈ Ω, then we have to add an ad-
ditional condition at ∞. For instance, a bubble in 3-D with changing
volume will introduce a source at ∞ will introduce a source flow at ∞:
m(t)x
(3.48) u = ∇Φ ∼ as x → ∞,
4π|x|3
where m(t) = dtd (Bubble Volume) is source strength. On the other
hand, if there is a uniform flow U0 at ∞, then we need to specify as
8

flow past a solid body,

(3.49) u ∼ U0 as x → ∞
3.2. Spherical Bubble. First consider the simplest case, the motion
of spherical oscillating bubble in a fluid with no body force b = 0. This
is physically realistic for small bubbles, where gravity does not play an
important role. In this case, Φ = Φ(r, t), r = |x|. There is no θ, φ
dependence and the boundary of the sphere is r = R(t) and the domain
Ω is given by r > R(t). Therefore, we have from Laplace’s equation in
spherical coordinates:
2
(3.50) ∂r2 Φ + ∂r Φ = 0
r
This implies
B(t) B(t)
(3.51) Φ(r, t) = A(t) + , implying ur = − 2
r r
Therefore, since F (x, t) = r − R(t), the kinematic condition (3.44)
implies that
dR dR B
(3.52) 0= − Φr (R(t), t) = + 2
dt dt R
Since increase/decrease in volume of the bubble 4π 3
R3 (t), is effectively
a source/sink, it follows that from (3.52),
d 4π 3 dR
(3.53) m(t) = R = 4πR2 = −4πB
dt 3 dt
Now, consider the pressure condition: on ∂Ω, i.e. on r = R(t):
1 2σ
(3.54) ∂t Φ + (∂r Φ)2 =
2 ρR
From representation (3.51), this becomes
dB
dA B2 2σ
(3.55) + dt + 4
=
dt R 2R ρR
Further, using Bernoulli equation as x → ∞, we find from (3.46), (3.55)
dB
p∞ d B2 2σ
(3.56) = −∂t Φ = − A = + dt + 4
−
ρ dt R 2R ρR
Thus, in terms of specified p∞ , by using (3.53) and (3.56), we obtain
after some algebra
 2
d2 R 3 dR 2σ p∞
(3.57) R 2 + + =−
dt 2 dt ρR ρ
 9

You can analyze this ODE if you like to study oscillation of bubble for
given p∞ /ρ. If p∞ /ρ is a constant (3.57) is a automomous system which
can be further reduced to a first order system and further analyzed; or
your could study the phase plane the usual way about the equilibrium,
which happens to be
p∞
(3.58) r = R0 ≡ −
2σ
assuming p∞ < 0. If p∞ > 0, there is no equilibrium, since the bubble
will eventually contract to zero size since the acceleration is clearly
negative for all time.
You can also study the response of the bubble to sound by considering
p∞
ρ
= C0 + C1 cos ωt for constant C0 and C1 .
3.3. Nonspherical Perturbation to a bubble. Suppose the bubble
is now perturbed a bit about the equibrium position r = R0 , deter-
mined from (3.58) with p∞ < 0 and independent of time and with no
body force, i.e. V = 0. We assume that the perturbation is not nec-
essarily spherically symmetric. In that case, in spherical coordinates,
the boundary ∂Ω of the bubble is given by
(3.59) r = R0 + ǫf (θ, φ, t)
and we seek to study the evolution of f with time, with given initial
perturbation
(3.60) f (θ, φ, 0) = f0 (θ, φ)
We will choose ǫ << 1, and seek solution for the linearized problem.
In order to simplify the problem, we have to take boundary conditions
at r = R0 + ǫf and apply them at the spherical unperturbed boundary
r = R0 . This is done by Taylor expanding the boundary condition
in powers of ǫ, assuming that this expansion is possible (this assume a
priori that the boundary shape is analytic. We note that in equilibrium
Φ = 0, since there is no flow in equilibrium. So, we may assume
(3.61) Φ(r, θ, φ, t) = ǫΨ(r, θ, φ, t)
Further, mean curvature is given by κ = ∇ · n. In our case, note that
in polar coordinates
(3.62)
−1/2
ǫ2 ǫ2 fφ2
 
ǫfθ 1 2
n = 1, − − fφ 1+ f +
R0 + ǫf sin θ(R0 + ǫf ) (R0 + ǫf )2 θ (R0 + ǫf )2 sin2 θ
Recalling that that for any vector F, expressed in spherical coordinates,
1 1 1
(3.63) ∇ · F = 2 ∂r r 2 Fr +


∂θ (sin thetaFθ ) + ∂φ Fφ
r r sin θ r sin theta
10

We obtain
2 2ǫ ǫ2
(3.64) κ= − 2f − Lf + O(ǫ2 )
R0 R0 R0
where the differential operator L in θ and φ is defined by
1 ∂2
 
1 ∂ ∂
(3.65) L≡+ sin θ +
sin θ ∂θ ∂θ sin2 θ ∂φ2
The pressure boundary condition (3.47) becomes
ǫσ
(3.66) ǫΨt (R0 , θ, φ) = − 2 (2f + Lf ) + O(ǫ2 )
R0
The kinematic condition (1.4) becomes
∂Ψ
(3.67) ǫft (θ, φ, t) − ǫ (R0 , θ, φ, t) = O(ǫ2 )
∂r
Equation (3.66) and (3.67) are now applied on the unperturbed bound-
ary r = R0 . We have to solve for Ψ(r, θ, φ, t) outside this perturbed
boundary with condition
∂ 2 Ψ 2 ∂Ψ ∂2Ψ
 
1 ∂ ∂Ψ 1
(3.68) 0 = ∆Ψ = + + sin θ +
∂r 2 r ∂r r 2 sin θ ∂θ ∂θ r 2 sin2 θ ∂φ2
Now, it is known that the differential operator L defined in (3.65) has
eigenfunctions
(3.69) LYl,m(θ, φ) = −l(l + 1)Yl,m(θ, φ)
where l ≥ 0 is an integer, and Yl,m are called spherical harmonics, given
by
(3.70) Yl,m (θ, φ) = eimφ Pl,m (cos θ) , where − l ≤ m ≤ l
and Pl,m (z) are called associated Legendre functions that satisfy the
differential equation
′ m2
(1 − z 2 )Pl,m
′

(3.71) − Pl,m = −l(l + 1)Pl,m
1 − z2
It is known that the set
(3.72) {Yl,m (θ, φ)}l=0,..∞,m=−l..l
forms a complete orthogonal set on the unit sphere. An arbitrary func-
tion f (θ, φ) can be written in terms of a linear combination of Yl,m (θ, φ)
in the L2 sense. In particular, the we may express perturbation about
a sphere is given by
∞ X
X l
(3.73) f (θ, φ, t) = al,m (t)Yl,m(θ, φ)
l=0 m=−l
 11

for some set of quantities al,m . Similarly, if we express

∞ X
X l
(3.74) Ψ(r, θ, φ, t) = Bl,m (r, t)Yl,m(θ, φ)
l=0 m=−l

Plugging (3.74) into (3.68) after using LYl,m (θ, φ) = −l(l + 1)Yl,m, we
obtain
∞ X l  2 
X ∂ Bl,m 2 ∂Bl,m l(l + 1)Bl,m
(3.75) 0 = Yl,m (θ, φ) + − ,
l=0 m=−l
∂r 2 r ∂r r2

implying
∂ 2 Bl,m 2 ∂Bl,m l(l + 1)Bl,m
(3.76) + − =0
∂r 2 r ∂r r2
So, solving we get a linear combination of r l and r −l−1 . The only
acceptable solution is a multiple of r −l−1 since r l does not vanish as
r → ∞. Therefore, it follows that
bl,m (t)R0l+1
(3.77) Bl,m (r, t) =
r l+1
Using (3.73), (3.74) and (3.77) in the linearized boundary condition
(3.66), (3.67), we obtain for each (l, m),
d σ
(3.78) bl,m = − 2 (−l2 − l + 2)al,m
dt R0

d (l + 1)
(3.79) al,m + bl,m = 0
dt R0
Or, eliminating bl,m between the two relations, we obtain
d2 al,m (l + 1)(l2 + l − 2)σ
(3.80) = − al,m
dt2 R03
The solution is obviously sinusoidal for l ≥ 1, with frequency
s
(l + 1)(l + 2)(l − 1)σ
(3.81) ωl =
R03
This describes the linearized motion of an oscillating bubble that os-
cillates due to surface tension effects when disturbed from equilibrium
The general shape will of course be given by (3.73), with al,m deter-
mined from initial values of shape distortion, that determines al,m (0)
and velocity that determines bl,m (0).
12

Remark 3.1. Note that the above calculation involved a linearization

and throwing away the nonlinear term. Generally, keeping the nonlin-
ear term makes it a much more difficult mathematical problem. Next
class, I will show how you can formulate such free boundary problems
in terms of fixed boundary problem in 2-D through the use of conformal
map. There is also another way of handling such problems in both 2-D
and 3-D. This is through the use of dipole or vortex sheet method.