Chapter 8

Chapter 8: Hyperbolic partial differential
equations
Similarity and Transport Phenomena in Fluid Dynamics

Christophe Ancey
Chapter 8: Hyperbolic partial differential equations
• Hyperbolic problems
• One-dimensional problems
• Characteristic equation
• Shock formation
• The Riemann problem
• Generalization to multidimensional problems
• Linear systems
• Nonlinear systems
• Shallow-water equations
Similarity and Transport Phenomena in Fluid Dynamics 2

o
Hyperbolic problems
Hyperbolic problems arise frequently in fluid mechanics (and continuum mechanics).

For instance, in hydraulic engineering:
Dimension 1: nonlinear convection equation, for example the kinematic wave
equation, which describes flood propagation in rivers
∂h √ ∂h 5/3
+K i = 0,
∂t ∂x
with h flow depth, K Manning-Strickler coefficient, et i bed gradient;
Dimension 2: Saint-Venant equations (also called the shallow water equations)
∂h ∂hū
+ = 0,
∂t ∂x
∂ ū ∂ ū ∂h τp
+ ū = g sin θ − g cos θ − ,
∂t ∂x ∂x %h
with ū flow-depth averaged velocity, h flow depth, θ bed slope, τp bottom shear
o
Hyperbolic problems
Dimension 3: Saint-Venant equations with advection of pollutant

∂h ∂hū
+ = 0,
∂t ∂x
∂ ū ∂ ū ∂h τp
+ ū = g sin θ − g cos θ − ,
∂t ∂x ∂x %h
∂ϕ ∂ϕ
+ ū = 0,
∂t ∂x
with ϕ pollutant concentration.
All these equations are evolution problems of the form
∂f
+ A(f ) · ∇f = S(f )
∂t
with f the dependant function, S the source term (possibly a differential operator,
e.g. diffusion), A a matrix.
o
Hyperbolic problems
Hyperbolic problems share a number of properties

• they describe systems in which information spreads at finite velocity
• this information can be conserved (when the source term is zero) or altered
(nonzero source term)
• solutions can be discontinuous
• smooth boundary and initial conditions can give rise to discontinuous solutions
after a finite time

o
Characteristic equation for one-dimensional problems
Let us first consider the following advection equation with n = 1 space variable and
without source term:
∂tu(x, t) + a(u)∂xu(x, t) = 0,
subject to one boundary condition of the form:
u(x, 0) = u0(x) at t = 0.
Note the this PDE is equivalent to
∂tu(x, t) + ∂xf [u(x, t)] = 0,
0 1
with a = f (u) when f is C continuous.
A characteristic curve is a curve x = xc(t) along which the partial differential
equation ∂f U + a∂xU = 0 is equivalent to an ordinary differential equation.
o
Characteristic equation
Consider a solution u(x, t) of the differential

t
system. Along the curve C of equation
x = xc(t) we have: u(x, t) = u(xc(t), t) and
the rate change is:
du(xc(t), t) ∂u(x, t) dxc ∂u(x, t)
x = + .
dt ∂t dt ∂x
u0 Suppose now that the curve C satisfies the
equation dxc/dt = a(u) :
du(x, t) ∂u(x, t) ∂u(x, t)
= +a = 0.
dt ∂t ∂x
Characteristic curves

o
Any convection equation can be cast in a characteristic form:

∂ ∂ du(x, t) dx
u(x, t) + a(u) u(x, t) = 0 ⇔ = 0 along straight lines C: = a(u).
∂t ∂x dt dt
Since du(x, t)/dt = 0 along xc(t), this means that u(x, t) is conserved along this
curve. Since u is constant a(u) is also constant, so the curves C are straight lines.
This holds true for linear and nonlinear systems.
If the source term is non zero, this does not change the final equation (except for
the right-hand term), but u is no longer conserved.

o
When this equation is subject to an initial condition, the characteristic equation can
be easily solved. As u is constant along the characteristic line, we get
dx
= a(u) ⇒ x − x0 = a(u)(t − t0),
dt
with the initial condition t0 = 0, u(x, t) = u0(x). We then infer
x − x0 = a(u0(x0))t
is the equation for the (straight) characteristic line emanating from point x0.
Furthermore, t ≥ 0 u(x, t) = u0(x0) since u is conserved. Since we have:
x0 = x − a(u0(x0))t, we then deduce:
u(x, t) = u0(x − a(u0(x0))t).

o
Shock formation
Consider the convective nonlinear equation:

∂ ∂
u(x, t) + f [u(x, t)] = 0,
∂t ∂x
with initial condition u(x, 0) = u0(x) and f a given function of u. This equation
can be solved simply by the method of characteristics.
du dx
= 0 along curves = λ(u),
dt dt
0
where λ(u) = f (u). We deduce that u is constant along the characteristic curves.
So dx/dt = λ(u) = c, with c a constant that can be determined using the initial
condition: the characteristics are straight lines with slopes λ(u0(x0)) depending on
the initial condition:
x = x0 + λ(u0(x0))t.
o
Shock formation
t
Since u is constant along a characteristic curve, we find:
u(x, t) = u0(x0) = u0 (x − λ(u0(x0))t)
The characteristic lines can intersect in some cases,
tB especially when the characteristic velocity decreases:
0
λ (u) < 0. What happens then? When two
x characteristic curves intersect, this means that
potentially, u takes two different values, which is not
Characteristic curves and shock
possible for a continuous solution. The solution
formation
becomes discontinuous: a shock is formed.

o
Shock formation
When two characteristic curves interest, the differential ux becomes infinite (since u
takes two values at the same time). We can write ux as follows
0
0 ∂x0 0 1 u0(x0)
ux = u0(x0) = u0(x0) 0 0
= ,
∂x 1 + λ (u0(x0))u (x0)t 1 + ∂xλ(x0)t
0 0
where we used the relation: λ (u0(x0))u (x0) = ∂uλ∂xu = ∂xλ. The differential ux
0
tends to infinity when the denominator tends to 0, i.e. at time: tb = −1/λ (x0). At
the crossing point, u changes its value very fast: a shock is formed. The s = s(t)
line in the x − t plane is the shock locus. A necessary condition for shock
occurrence is then tb > 0:
0
λ (x0) < 0.
Therefore there is a slower speed characteristic.
o
Shock formation
u
The characteristic curves that are causing the shock
form an envelope curve whose implicit equation is given
by:
0
x = x0 + λ(u0(x0))t et λ (u0(x0)) + 1 = 0.
x
After the shock, the solution is multivalued, which is
x=s
impossible from a physical standpoint. The multivalued
Shock position part of the curve is then replaced with a discontinuity
positioned so that the lobes of both sides are of equal
area.

Shock formation: Rankine-Hugoniot equation
Generally, we do not attempt to calculate the envelope of characteristic curves,

because there is a much simpler method to calculate the trajectory of the shock.
Indeed, the original PDE
Z xRcan be cast in the integral form:
d
u(x, t)dx = f (u(xL, t)) − (u(xR, t)),
dt xL
where xL and xR are abscissa of fixed point of a control volume. If the solution
admits a discontinuity in x = s(t) on the interval [xL, xR], then
Z xR Z s Z xR
d d
u(x, t)dx = u(x, t)dx + u(x, t)dx ,
dt xL dt xL s
That is:
Z xR Z s Z xR
d ∂ ∂
u(x, t)dx = u(x, t)dx + u(x, t)dx + ṡu(xL, t) − ṡu(xR, t).
dt xL xL ∂t s ∂t
o
Shock formation: Rankine-Hugoniot equation
Taking the limit xR → s and xL → s, we deduce:

ṡJuK = Jf (u)K,
where
+ −
JuK = u − u = lim u − lim u,
x→s,x>s x→s,x<s
The + and − signs are used to describe what is happening on the right and left,
respectively, of the discontinuity at x = s(t).
In conclusion, we must have on both sides of x = s(t):
ṡJuK = Jf (u)K
This is the Rankine-Hugoniot equation.

o
Riemann problem
u0
uL We call Riemann problem an initial-value
problem of the following form:
uR
∂tu + ∂x[f (u)] = 0,
x (
uL if x < 0,
u(x, 0) = u0(x) =
x − at = 0 uR if x > 0,
t
uL with uL et uR two constants.
uR
This problem describes how an initially
x
piecewise constant function u, with a
Riemann problem for the linear case discontinuity in x = 0 changes over time. This
problem is fundamental to solving theoretical
and numerical problems.
Riemann problem: linear case
u0
uL Let us consider the linear case f (u) = au,
with a a constant. The solution is
uR
straightforward:
(
x
uL if x − at < 0,
u(x, t) = u0(x − at) =
x − at = 0 uR if x − at > 0.
t
uL The discontinuity propagates with a speed a.
uR
Riemann problem for the linear case

Riemann problem: nonlinear case
00
In the general case (where f 6= 0), the Riemann problem is an initial-value problem
of the following form:
∂tu + ∂x[f (u)]
( = 0,
uL if x < 0,
u(x, 0) = u0(x) =
uR if x > 0.
00
with uL and uR two constants. Assume that f > 0 (the case of a non-convex flow
will not be treated here). We will show that there are two possible solutions:
• a solution called rarefaction wave (or simple wave), which is continuous;
• a discontinuous solution which represents the spread of the initial discontinuity
(shock).

o
Rarefaction wave. The PDE is invariant under the transformation x → λx and

t → λt. A general solution can be sought in the form U (ξ) with ξ = x/t.
Substituting this general form into the partial differential equation, we obtain an
ordinary differential equation of the form:
0 0
(f (U (ξ)) − ξ) U = 0.
There are two types of solution to this equation:
0 00 0 0
• rarefaction wave: (f (U (ξ)) − ξ) = 0. If f > 0, then f (uR) > f (uL) ; equation
0 0 0
f (U ) = ξ admits a single solution when f (uR) > ξ > f (uL). In this case, uL is
0 0
connected to uR through a rarefaction wave: ξ = f (U (ξ)). Inverting f , we find
out the desired solution
0(−1)
u(x, t) = f (ξ)
o
0
• constant state: U (ξ) = 0. This is the trivial solution u(x, t) = cst. This solution
does not satisfy the initial problem.
The solution is thus a rarefaction
 wave. It reads x 0
 u L if ≤ f (uL ),
t



0(−1) 0 x 0
u(x, t) = f (ξ) si f (uL) ≤ ≤ f (uR)
 x t
0

 uR if ≥ f (uR).

t

o
Shock wave
Weak solutions (discontinuous) to the hyperbolic differential equation may exist.
Assuming a discontinuity along a line x = s(t) = ṡt, we get: Jf (u)K = ṡJuK. The
solution is then: (
uL if x < ṡt,
u(x, t) =
uR if x > ṡt.
Then a shock wave forms, with its velocity ṡ given by:
f (uL) − f (uR)
ṡ = .
uL − uR

o
Selection of the physical solution

00 0
Two cases are to be considered (remember that f > 0). We call λ(u) = f (u) the
characteristic velocity (see section below), which is the slope of the characteristic
curve (straight line) of the problem.
00
• 1st case: uR > uL. Since f > 0, then λ(uR) > λ(uL). At initial time t = 0, the
0
characteristic lines form a fan. Equation ξ = f (U (ξ)) admits a solution over the
interval λ(uR) > ξ > λ(uL);
• 2nd case: uR < uL. Characteristic lines intersect as of t = 0. The shock
propagates at rate λ(uR) < ṡ < λ(uL). This last condition is called Lax condition;
it allows to determining whether the shock velocity is physically admissible.

o
u0
uR
uL
t
L)
x − mt = 0
u
t
λ(
uL
=
x
uR )t
λ(
x=
uR
x

Non-convex flux
For some applications, the flux is not convex. An example is given by the equation
of Buckley-Leverett, reflecting changes in water concentration φ in a pressure-driven
flow of oil in a porous medium:
φt + f (φ)x = 0,
2 2 2 −1
with f (φ) = φ (φ + a(1 − φ) ) and a a parameter (0 < a < 1). This fonction
has an inflexion point. Contrary to the convex case, for which the solution involves
shock and rarefaction waves, the solution is here made up of shocks and compound
wave resulting from the superimposition of one shock wave and one rarefaction
wave.

o
Exercise 1
x
h0
x=
Solve Huppert’s equation, which describes fluid motion over an inclined plane in the low
Reynolds-number limit:
2
∂h ρgh sin θ ∂h
+ = 0.
∂t µ ∂x
The solution must also satisfy the mass
Z conservation equation
h(x, t)dx = V0
where V is the initial volume V0 = `h0
o
Generalization to higher dimensions: terminology
Terminology
We study evolution equations in the form:
U t + A(U )U x + B = 0,
with A an n × n matrix. B is a vector of dimension n called the source. The
system is homogeneous if B = 0. It is a conservative form when
∂
U t + F (U ) = 0,
∂x
with A(U ) = ∂F /∂U .
The eigenvalues λi of A represent the speed(s) at which information propagates.
They are the zeros of the polynomial det(A − λ1) = 0. The system is hyperbolic if
A has n real eigenvalues.
o
Generalization to higher dimensions: terminology
If a function satisfies an evolution equation:

ut + [f (u)]x = 0,
then we can create an infinity of equivalent PDEs: [g(u)]t + [h(u)]x = 0 provided
0 0 0
that g and h are such that h = g f . As long as the function u(x, t) is
continuously differentiable, there is no problem, but for weak solutions (exhibiting a
discontinuity), then the equations are no longer equivalent. We must use the original
physical equation (usually expressing conservation of mass, momentum or energy).

o
Left and right eigenvectors
Take the particular case n = 2 for illustration. The matrix A has two real
eigenvalues λ1 and λ2 together with left eigenvectors v 1 and v 2:
v i · A = λiv i.
It also has two right eigenvectors w1 et w2:
A · wi = λiwi.
Let us assume that A has the following entries
" #
ab
A= ,
cd

o
Left and right eigenvectors
Then we get
   √ 
1 a−d+ ∆ √
√  a+d+ ∆
v 1 =  d − a + ∆  , w1 =   , associated with λ1 = ,
  
2c
1 2
2c
   √ 
1 a−d− ∆ √
√  a+d− ∆
v 2 =  d − a − ∆  , w2 =   , associated with λ2 = ,
  
2c
1 2
2c
2
avec ∆ = (a − d) + 4bc. Note that
v 1 · w2 = 0, and v 2 · w1 = 0.

o
Diagonalization
Linear system: When the eigenvectors are constant

v i · U t + v i · A(U )U x + v i · B = 0.
thus:
v i · U t + λiv i · U x + v i · B = 0.
We pose ri = v i · U and obtain
rt + Λ · rx + r · B = 0
where Λ = diag{λ1, λ2}. The system is now made of independent PDEs
dr 1 dxc, 1(t)
+ r 1 · B = 0 along x = xc, 1(t), = λ1,
dt dt
dr 2 dxc, 2(t)
+ r 2 · B = 0 along x = xc, 2(t), = λ2,
dt dt
o
Diagonalization: nonlinear system
Nonlinear system: We seek new variables r = {r1, r2} such that:

v 1 · dU = µ1dr1,
v 2 · dU = µ2dr2,
where µi are integrating factors such that dri are exact differential. We have:

∂r1 ∂r1
µ1dr1 = µ1 dU1 + dU2 = v11dU1 + v12dU2.
∂U1 ∂U2
Identifying the various terms leads to:
∂r1 v11
= ,
∂U1 µ1
and
∂r1 v12
= .
∂U2 µ1
o
By taking the ratio of the two equations above, we get:

∂r1 v11 ∂r1
= ,
∂U1 v12 ∂U2
The Schwartz theorem states that ∂xy f = ∂yxf and so from du(x, y) = adx + bdy,
we deduce that ∂y a = ∂xb. Here this gives us the relation
∂ v12 ∂ v11
= .
∂U1 µ1 ∂U2 µ1
The integrating factor can also be deduced from ∂r1/∂U2 = 1/µ1 when the entries
of v 1 are properly selected such that v11 = 1. Note that
∂r1 v11 ∂r1
= ⇒ w21∂r1/∂U1 + w22∂r1/∂U2 = 0 ⇒ w2 · ∇r1 = 0
∂U1 v12 ∂U2
Definition: r1 is said to be a 2-invariant of the system.
o
The characteristic equation associated with the equation above is

dU1 dU2 dr1
= = ,
v12 v11 0
which leads to an integral. The first equation of the differential system is equivalent
to:
dU
v1 · + v 1 · B = 0,
dt x=X1(t)

where x = X1(t) satisfies dX1/dt = λ1. This is the 1-characteristic curve:

dr1
µ1 + v 1 · B = 0.
dt x=X1(t)

Similarly for r2:
dr2
µ2 + v 2 · B = 0.
dt x=X2(t)

o
In a matrix form:
dr
+ S(r, B) = 0,
dt r=X(t)

along two characteristic curves r = X(t) such that dX(t)/dt = (λ1, λ2); S is the
source term whose entries are µiSi = v i · B. The new variables r are called the
Riemann variables. For B = 0, they are constant along the characteristic curves
and thus they are called Riemann invariants.

o
Exercise 2
Consider the Saint-Venant equations:

∂th + ∂x(uh) = 0, (1)
∂tu + u∂xu + ∂xh = 0, (2)
Determine the Riemann invariants and plot the characteristic curve for the
dam-break problem
– initial velocity −∞ < x < ∞ u(x, 0) = 0
– initial depth x < 0 h(x, 0) = h0
x>0 h(x, 0) = 0

o
The Riemann problem: linear systems
Consider the following linear hyperbolic problem:

∂U ∂U
+A· = 0,
∂t ∂x
where A is an n × n matrix with n distinct real eigenvalues. We thus have
−1
A = R · Λ · R , with R the matrix associated with the change of coordinates
(the columns are the right eigenvectors of A) and Λ a diagonal matrix whose
−1
entries are λi. Making use of the change of variables W = R · U leads to
∂W ∂W
+Λ· = 0.
∂t ∂x
This is a system of independent linear hyperbolic PDEs: ∂twi + λi∂xwi = 0, whose
solution takes the form wi = ωi(x − λit).

o
The inverse change of variables leads to e U = R · W

Xn
U= wi(x, t)r i,
i=1
where r i is a right eigenvector A associate to λi and wi is ith entry of W . The solution
results from the superimposition of n waves travelling at speed λi; these waves are
independent, do not change form (this form is given by the initial condition ωi(x, 0)r i).
When all but one elementary waves are constant (∂xωi(x, 0) = 0), then the resulting
wave is called a j-simple wave
X n
U = ωj (x − λj t)r j + wi(x, t)r i,
i=1, i6=j
Information propagates along the j-characteristic curve (all others wi are constant).
o
The Riemann problem takes the form

∂U ∂U
+A· = 0,
∂t ∂x
with (
U ` if x < 0,
U (x, 0) = U 0(x) =
U r if x > 0.
We now expand U ` et U r in the eigenvector basis r i
n n
(`) (r)
X X
U` = wi r i et U r = wi r i ,
i=1 i=1
(`) (r)
with w` = wi et wr = wi vectors with constant entries.

o
The Riemann problem involves n scalar problems

(
(`)
w if x < 0,
wi(x, 0) = (r)
w if x > 0.
The solution to these advection equations is
(
(`)
wi if x − λit < 0,
wi(x, t) = (r)
wi if x − λit > 0.
We call I(x, t) the largest index i such that x − λit. The solution reads
I n
(r) (`)
X X
U (x, t) = wi r i + wi r i .
i=1 i=I+1

o
x = λ2 t
U * Consider the case n = 3. The solution in the

A
M x − t space breaks down into ”wedges” where
U*r
x = λ1t x = λ3t U is constant and separated by characteristic
UA Ur curves x = λit. At any point M, we can
determine the value taken by U by plotting
x
X − λ3T X − λ2T O X − λ1T the characteristic curves issuing from M
toward the x-axis.

o
Exercise 3
Consider the wave equation

2 2
∂u 2∂ u
2
= c 2
.
∂t ∂x
with initial data
u(x, 0) = u0(x) and ut(x, 0) = u1(x)
Solve the equation.
o
The Riemann problem: nonlinear systems

∂U ∂U
+A· = 0,
∂t ∂x
where A = ∇F . This equation is invariant to the stretching group
(x, t) → (λx, λt). We seek solutions in the form U (x, t) = W (ξ, uL, uR), with
ξ = x/t
dW dW
−ξ + ∇F · =0
dξ dξ
0
• W (ξ) = 0, this is the constant state;
0
• W (ξ) is a right eigenvector of ∇F associate to ξ for all values taken by ξ. The
curve W (ξ) is tangent to the right eigenvector w.

o
t
uR Generalizing the concept seen for 1D
hyperbolic equations, we define a rarefaction
uL
wave as a simple wave function of ξ = x/t

x 

 uL si x/t ≤ ξ1,
u(ξ) = W (ξ, uL, uR) si ξ1 ≤ x/t ≤ ξ2,

uR si x/t ≥ ξ2.


x
where uR and uL must satisfy
λk (uL) < λk (uR)

o
From the original PDE

dW dW
−ξ + ∇F · =0
dξ dξ
0
we deduce that W is a right eigenvector and that
ξ = λk (W ),
and on differentiating with respect to ξ, we get
0
1 = ∇uλk (W ) · W (ξ),
0 0 −1
Since W is a right eigenvector, W (ξ) = αwk , thus α = [∇uλk (W ) · wk ] . The
function W is solution to the ODE
0 wk
W (ξ) = ,
∇uλk (W ) · wk
o

∂U ∂U
+A· = 0,
∂t ∂x
where A = ∇F . This equation is invariant to the stretching group
(x, t) → (λx, λt). We seek solutions in the form U (x, t) = W (ξ, uL, uR), with
ξ = x/t
dW dW
−ξ + ∇F · =0
dξ dξ
0
• W (ξ) = 0, this is the constant state;
0
• W (ξ) is a right eigenvector of ∇F associate to ξ for all values taken by ξ. The
curve W (ξ) is tangent to the right eigenvector w.

o
t
uL A shock wave is a non-material surface
x = s(t) across which the solution is
discontinuous ẋ = s. The Rankine-Hugoniot
x relation must hold
x = s ′t ṡ(uL − uR) = f (uL) − f (uR),
to which we add the Lax entropy condition
x λk (uL) > s > λk (uL),

(jump in the kth field: we speak of a k-shock
wave)

o
λi
λ1
Summary
t λn
The solution to the Riemann problem:
∂U ∂F (U )
+ = 0,
∂t ∂x
subject to
(
U L si x < 0,
UL UR U (x, 0) = U 0(x) =
U R si x > 0.
involves n + 1 states separated by n waves
x
related to each eigenvalue.

o
For linear systems, the eigenvalues define shock waves. For nonlinear systems,
different types of waves are possible:
• shock wave: in this case, the Rankine-Hugoniot holds
0
s [U ]x=s(t) = F (U (xL)) − F (U (xR)) along with the entropy condition
0
λi(U L) > si > λi(U R)
• contact discontinuity (when an eigenvalue is constant or such that ∇uλk · wk = 0):
the Rankine-Hugoniot relation holds, with the condition λi(U L) = λi(U R)
• rarefaction wave: the characteristics fan out λi(U L) < λi(U R), self-similar
solutions.

o
R2 R1
Hugoniot locus
UL
UR The solution to the Riemann problem:
∂U ∂F (U )
+ = 0,
U* ∂t ∂x
S2 subject to
S1 (
U L si x < 0,
U (x, 0) = U 0(x) =
U R si x > 0.
involves n + 1 states separated by n waves
related to each eigenvalue.

o
Example: The Saint-Venant equations
Let us consider the Saint-Venant equations:

∂th + ∂x(uh) = 0,
2
∂thu + ∂xhu + gh∂xh = 0.
We introduce the unknowns U = (h, hu), the flux function
2 2
F = (hu, hu + gh /2) and the matrix A:
!
∂F 0 1
A= = 2
.
∂U gh − u 2u
The conservative form is:
∂u ∂u
+A· = 0.
∂t ∂x

o
√
Eigenvalues and eigenvectors for the conservative formulation (with c = gh)
i=1 i=2
λi u−c u+c
1 1
wi { , 1} { , 1}
u−c u+c
3c 3c
wi · ∇λi
2(c − u) 2(c + u)

o
If we take (h, u) as variables, then the system is put in a nonconservative, but some
2 2
solutions are easier to work out. With U = (h, u), F = (hu, hu + gh /2) and
matrix A: !
∂F uh
A= = ,
∂U gu
∂u ∂u
+A· = 0.
∂t ∂x

o
√
Eigenvalues and eigenvectors for the nonconservative formulation (with c = gh)
i=1 i=2
eigenvalues λi u − c u + c
c c
right eigenvectors wi − , 1 , 1
n gc o n gc o
left eigenvectors v i − , 1 , 1
h h
Riemann invariants ri u − 2c u + 2c

o
Shock conditions
σ JhK = JhuK,
2 2
σ Jhu] = Jhu + gh /2K,
with σ the shock velocity. In a frame related to the shock wave, then v = u − σ and
h1v1 = h2v2,
2 2 2 2
h1v1 + gh1/2 = h2v2 + gh2/2.
There are two families
• 1-shock: σ < uL − cL et uR − cR < σ < uR + cR. vL > vR: the flux goes from left to
right when vL > 0;
• 2-shock: σ > uR + cR et uL − cL < σ < uL + cL. vR > vL: the flux goes from right
to left when vL > 0. Similarity and Transport Phenomena in Fluid Dynamics 54
o
Let us determine the Hugoniot locus, i.e., the points (h2 v2) connected to (h1 v1) by a
1- or 2-shock wave
h2v2 − h1v1
σ= ,
h2 − h1
2 2 2
(h2u2 − h1u1) 2 gh2 2 gh1
= h2u2 + − h1u1 − ,
h2 − h1 2 2
This gives us the shock speed and u2(h2|h1 v1) :
r
g h1 + h2
u2 = u1 ∓ (h2 − h1) ,
r 2 h 1 h2
g h2
σ = u1 ∓ (h1 + h2) .
2 h1

o
Rarefaction waves
We seek Riemann invariants rk , defined as ∇urk · wk = 0. We work with the variables
(h, u). The first invariant is:
∂r ∂r
−c + λ1 = 0,
∂h ∂u
whose characteristic equations are
du dh
=− .
g c
An integral is u + 2c. For the second invariant, we find u − 2c.
√ √
Along a 1-rarefaction wave, we have: u2 + 2 gh2 = u1 + 2 gh1 and the invariant
r1 = u + 2c is constant along any characteristic curve associated with the eigenvalue
λ1 = u − c (when these fan out, r1 is in the cone formed by the characteristics).

o
3
2
Returning to the variables (h, q = hu), we
1
deduce
S1
0 • Along a 1-rarefaction wave, we get:
u
√ √
-1 q2/h2 + 2 gh2 = q1/h1 + 2 gh1;
S2 • Along a 2-rarefaction wave, we get:
-2
√ √
-3 R1 q2/h2 − 2 gh2 = q1/h1 − 2 gh1.
0 1 2 3 4 5
h
Show and rarefaction waves in the (h, u)

space. Arbitrarily the curves are issuing from
(h, u) = (1, 0)
o
Working out the solution to the Riemann problem

The construction method consists of introducing an intermediate state u∗. The state
(h∗, u∗) can be connected to a left state (hL, uL) through a 1-wave
 √ √
 S1(h∗| hL, uL) = uL + 2 ghL − 2r gh∗
 if h∗ < hL 1-rarefaction wave
u∗ = h ∗ + hL
 R1(h∗| hL, uL) = uL − (h∗ − hL) g
 if h∗ > hL 1-shock wave
2h∗hL
It can be connected to a right state (hR, uR) through a 2-wave
 √ √
 S2(h∗| hR, uR) = uR − 2 ghR + r
 2 gh∗ if h∗ < hR 2-rarefaction wave
u∗ = h ∗ + h R
 R2(h∗| hR, uR) = uR + (h∗ − hR) g
 if h∗ > hR 2-shock wave
2h∗hR

o
3
We begin with 1-waves, then 2-waves as
2 R2
information on the left gauche is primarily
1 B
S1 conveyed by the smallest eigenvalue, then the
u
0 L R
others.
-1 A
Note that tangents to the curves R1 et S1 are
S2
-2 R1 the same. Note also that an intermediate
0 1 2 3 4 5
state is possible only if:
h p p
uR − uL < 2( ghR + ghL).
Solution to the Riemann problem for
(hL, uL) = (1, 0) et (hR, uR) = (2, 0) For hL = 0 (hR = 0, resp.), then the 1-shock
wave (the 2-shock wave, resp.) is undefined.

o
Homework
Let us consider the (dimensionless) governing equations for a

visco-elastoplastic material in a simple shear
∂u ∂τ
=1+ ,
∂t ∂z
∂τ ∂u
= − F (τ ),
∂t ∂z
1/n
with F (τ ) = max (0, |τ | − 1) τ /|τ |. The boundary and initial
Lacaze, L., A. Filella, and O. Thual, Steady and unsteady shear flows of a viscoplastic fluid in a
cylindrical Couette cell, Journal of Non-Newtonian Fluid Mechanics, 220, 126-136, 2015.
conditions are u = 0 at z = 0, τ = 0 at z = 1, and τ = u = 0 at
t = 0. Cast the system into its characteristic form. Write a
numerical code to solve the resulting system.

o

Chapter 8

Uploaded by

Copyright:

Available Formats

Chapter 8

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 8

Uploaded by

Copyright:

Available Formats

Chapter 8: Hyperbolic partial differential

Similarity and Transport Phenomena in Fluid Dynamics

Similarity and Transport Phenomena in Fluid Dynamics 2

Hyperbolic problems arise frequently in fluid mechanics (and continuum mechanics).

Dimension 3: Saint-Venant equations with advection of pollutant

Hyperbolic problems share a number of properties

Similarity and Transport Phenomena in Fluid Dynamics 5

Consider a solution u(x, t) of the differential

Similarity and Transport Phenomena in Fluid Dynamics 7

Any convection equation can be cast in a characteristic form:

Similarity and Transport Phenomena in Fluid Dynamics 8

Similarity and Transport Phenomena in Fluid Dynamics 9

Consider the convective nonlinear equation:

Similarity and Transport Phenomena in Fluid Dynamics 11

Similarity and Transport Phenomena in Fluid Dynamics 13

Generally, we do not attempt to calculate the envelope of characteristic curves,

Taking the limit xR → s and xL → s, we deduce:

Similarity and Transport Phenomena in Fluid Dynamics 15

Riemann problem for the linear case

Similarity and Transport Phenomena in Fluid Dynamics 17

Similarity and Transport Phenomena in Fluid Dynamics 18

Rarefaction wave. The PDE is invariant under the transformation x → λx and

Similarity and Transport Phenomena in Fluid Dynamics 20

Similarity and Transport Phenomena in Fluid Dynamics 21

Selection of the physical solution

Similarity and Transport Phenomena in Fluid Dynamics 22

Similarity and Transport Phenomena in Fluid Dynamics 23

Similarity and Transport Phenomena in Fluid Dynamics 24

If a function satisfies an evolution equation:

Similarity and Transport Phenomena in Fluid Dynamics 27

Similarity and Transport Phenomena in Fluid Dynamics 28

Similarity and Transport Phenomena in Fluid Dynamics 29

Linear system: When the eigenvectors are constant

Nonlinear system: We seek new variables r = {r1, r2} such that:

By taking the ratio of the two equations above, we get:

The characteristic equation associated with the equation above is

Similarity and Transport Phenomena in Fluid Dynamics 34

Consider the Saint-Venant equations:

∂tu + u∂xu + ∂xh = 0, (2)

Similarity and Transport Phenomena in Fluid Dynamics 35

Consider the following linear hyperbolic problem:

Similarity and Transport Phenomena in Fluid Dynamics 36

The inverse change of variables leads to e U = R · W

The Riemann problem takes the form

Similarity and Transport Phenomena in Fluid Dynamics 38

The Riemann problem involves n scalar problems

Similarity and Transport Phenomena in Fluid Dynamics 39

U * Consider the case n = 3. The solution in the

Similarity and Transport Phenomena in Fluid Dynamics 40

Consider the wave equation

Consider the following linear hyperbolic problem:

Similarity and Transport Phenomena in Fluid Dynamics 42

Similarity and Transport Phenomena in Fluid Dynamics 43

From the original PDE

Consider the following linear hyperbolic problem:

Similarity and Transport Phenomena in Fluid Dynamics 45

x λk (uL) > s > λk (uL),

Similarity and Transport Phenomena in Fluid Dynamics 46

Similarity and Transport Phenomena in Fluid Dynamics 47

Similarity and Transport Phenomena in Fluid Dynamics 48