One Dimensional Hyperbolic Conservation Laws:

Past and Future

Alberto Bressan
arXiv:2310.16707v1 [math.AP] 25 Oct 2023

Department of Mathematics, Penn State University,

University Park, Pa. 16802, USA.
[email protected]

October 26, 2023

Abstract

Aim of these notes is to provide a brief review of the current well-posedness theory for
hyperbolic systems of conservation laws in one space dimension, also pointing out open
problems and possible research directions. They supplement the slides of the short course
given by the author in Erice, May 2023, available at: sites.google.com/view/erice23/speakers-
and-slides.

1 Introduction

Aim of these notes is provide a brief review of the current well-posedness theory for hyperbolic
systems of conservation laws in one space dimension, also pointing out open problems and
possible research directions. They supplement the slides of the short course given by the
author in Erice, May 2023, available at: sites.google.com/view/erice23/speakers-and-slides.
Section 2 introduces basic definitions, including the concept of weak solution, and various
admissibility conditions. Section 3 describes several approximation methods. The main results
on global existence of weak solutions and their continuous dependence on initial data are
recalled in Sections 4 and 5. The recent advances, on the uniqueness of weak solutions that
satisfy the Liu admissibility condition, are covered in greater detail in Section 6. The relevance
of these results toward error bounds for all kinds of approximate solutions is discussed in
Section 7, together with two specific open problems. Finally, Section 8 is devoted to solutions
with possibly unbounded total variation, recalling the main known results and pointing out
some research directions.

1
2 Basic concepts

2.1 Hyperbolic systems.

A system of conservation laws in one space dimension has the form

ut + f (u)x = 0. (2.1)

In components, this can be written as

     
u1 f1 (u) 0
∂  . 
 . + ∂  .  .
 .  =  ..  . (2.2)
∂t  .  ∂x  .   
un fn (u) 0

Here u = (u1 , . . . , un )T is the vector of conserved quantities, while f = (f1 , . . . , fn )T

is the vector of fluxes. Conservation laws provide the fundamental mathematical models
in continuum physics [41]. A primary example is provided by the Euler equations of gas
dynamics, accounting for the conservation of mass, momentum and energy [43].
Smooth solutions to the system of PDEs (2.1) can be obtained by solving the equivalent
quasilinear system
.
ut + A(u)ux = 0, where A(u) = Df (u). (2.3)

We say that the system (2.1) is strictly hyperbolic if at every point u the n × n Jacobian
.
matrix A(u) = Df (u) has n real distinct eigenvalues

λ1 (u) < λ2 (u) < · · · < λn (u). (2.4)

When this holds, one can find bases of right and left eigenvectors, say {r1 , . . . , rn }, {l1 , . . . , ln },
with

A(u)ri (u) = λi (u)ri (u), li (u)A(u) = λi (u)li (u), i = 1, . . . , n. (2.5)

These vectors can be normalized so that

(
1 if i = j,
ri (u) = 1, li (u)rj (u) =
0 ̸ j.
if i =

The behavior of eigenvalues of Df (u) strongly affect the nature of solutions to (2.1). Following
classical literature [51], we say that the i-th characteristic field is genuinely nonlinear if the
directional derivative of the eigenvalue λi in the direction of the corresponding eigenvector
ri (u) satisfies
∇λi (u) · ri (u) > 0 for all u. (2.6)
On the other hand, we say that the i-th characteristic field is linearly degenerate if

∇λi (u) · ri (u) = 0 for all u. (2.7)

Throughout the following we assume that the flux function f is at least twice continuously
differentiable, so that the above derivatives are well defined.

2
Example 2.1. In Lagrangian coordinates, the Euler equations of isentropic gas dynamics
take the form (
vt − ux = 0,
(2.8)
ut + p(v)x = 0.
Here ρ is the density of the gas, v = ρ−1 is specific volume, u is the velocity and p = p(v) is the
pressure. A natural choice for the pressure is p(v) = kv −γ , with 1 ≤ γ ≤ 3. The eigenvalues
of the Jacobian matrix !
. 0 −1
A = Df =
p′ (v) 0
are p p
λ1 = − −p′ (v) , λ2 = −p′ (v)
Since p′ (v) < 0, the system is strictly hyperbolic. A further computation reveals that both
characteristic fields are genuinely nonlinear.

2.2 Weak solutions.

A key feature of hyperbolic conservation laws is that, even for initial data ū ∈ C 1 (i.e.,
continuously differentiable), the gradient ux of the solution may blow up at a finite time
T (see Fig. 1). In order to prolong the solution also for t > T , one must work within a
space of discontinuous functions, interpreting the equation (2.1) in distributional sense. In
the following, L1loc (Ω) denotes the space of locally integrable functions defined on an open
subset Ω ⊂ R × R, with values in Rn . Moreover, Cc1 (Ω) denotes the space of continuously
differentiable functions with compact support.

u
’
f(u) u(T)
u(0)

x
Figure 1: An example where the gradient ux of the solution becomes unbounded at a finite time T .
For t > T , the solution contains a shock and must be interpreted in distributional sense.

Definition 2.1. Let u = u(t, x) be a function defined on an open set Ω ⊆ R × R. We say that
u is a weak solution to the system of conservation laws (2.1) if u, f (u) ∈ L1loc (Ω) and
ZZ
for all ϕ ∈ Cc1 (Ω).

uϕt + f (u)ϕx dxdt = 0 (2.9)

Definition 2.2. A function u = u(t, x) is a weak solution to the Cauchy problem

ut + f (u)x = 0, u(0, x) = ū(x), t ∈ [0, T ], (2.10)

3
if the map t 7→ u(t, ·) is continuous with values in L1 (R; Rn ), satisfies the initial condition in
(2.10), and moreover
Z T Z +∞ 
ϕ ∈ Cc1 ]0, T [×R


uϕt + f (u)ϕx dxdt = 0 for all
0 −∞

u(0) u(t)
u+
_
u

0 λt x

Figure 2: A shock with left and right states u− , u+ , moving with speed λ.

Notice that the identity (2.9) is obtained multiplying (2.1) by the test function ϕ and inte-
grating by parts.
The simplest example of a discontinuous solution to (2.1) is a single shock, shown in Fig. 2.
(
u− if x < λt ,
u(t, x) = +
(2.11)
u if x > λt .

It is well known that the above function is a weak solution if and only if the shock speed λ
and the left and right states u− , u+ satisfy the Rankine-Hugoniot equations

λ · (u+ − u− ) = f (u+ ) − f (u− ). (2.12)

In other words, the vector equation (2.12) states that

[speed] × [jump in the state] = [jump in the flux].

. −
t x =λ u

τ +
u

x
ξ

Figure 3: A point of approximate jump.

To state a version of the Rankine-Hugoniot conditions which applies to more general solutions,
we introduce

4
Definition 2.3. The function u = u(t, x) has an approximate jump at the point (τ, ξ) ∈ R2
if there exists vectors u+ ̸= u− and a speed λ such that, setting
(
. u− if x < λ t,
U (t, x) = +
(2.13)
u if x > λ t,

one has Z r Z r
1
lim 2 u(τ + t, ξ + x) − U (t, x) dxdt = 0. (2.14)
r→0+ r −r −r

We say that u is approximately continuous at the point (τ, ξ) if (2.14) holds with u+ = u−
(and λ arbitrary).

Theorem 2.1. (Rankine-Hugoniot equations) Let u be a bounded weak solution of (2.1),

having an approximate jump at a point (τ, ξ). Then the left and right states u− , u+ and the
speed λ satisfy the Rankine-Hugoniot equations (2.12).

For a proof, see for example [16]. Writing the Rankine-Hugoniot equations in the form
Z 1
+ − + − Df θu+ + (1 − θ)u− · (u+ − u− ) dθ


λ (u − u ) = f (u ) − f (u ) =
0

= A(u+ , u− ) · (u+ − u− ),

we see that

(i) The jump u+ − u− is an eigenvector of the averaged matrix A(u+ , u− ).

(ii) The speed λ coincides with the corresponding eigenvalue.

2.3 Admissibility conditions.

In general, solutions to the Cauchy problem (2.10) may not be unique, as soon as discontinu-
ities are present. To single out a unique weak solution one needs to impose further admissibility
conditions on the shocks. These can be derived by three different approaches:

• Stability w.r.t. small perturbations.

• Vanishing viscosity approximations.

• Entropy dissipation.

1. A stability condition. Consider first a scalar conservation law. In this case, the Rankine-
Hugoniot condition (2.12) simply states that the speed of a shock with left and right states
u− , u+ must be
f (u+ ) − f (u− )
λ = . (2.15)
u+ − u−
Looking at the graph of the function f (u) (see Fig. 4), this means

speed of the shock = slope of the secant line through u− , u+ .

5
 f
f

_ _
u u* u+ u+ u* u

Figure 4: For a scalar conservation law, according to (2.15) the speed of a shock is the slope of a
secant line to the graph of f .

u(0) u(t)
u+

_
u

u(0) u(t)
u+

u* stable
_
u

u(0) u(t)
u+

u* unstable
_
u

Figure 5: A solution containing a single shock (top figure) can be perturbed into a solution which
initially contains two nearby shocks. If the shock behind travels faster than the shock ahead (center
figure), then the original shock is stable. If the shock behind is slower than the shock ahead (lower
figure), the original shock is unstable.

6
To check the stability of the solution (2.11), we can perturb the shock by inserting an inter-
mediate state u∗ ∈ [u− , u+ ]
The original shock will be stable w.r.t. this perturbation iff

[speed of jump behind] ≥ [speed of jump ahead].

By (2.15), this means

f (u∗ ) − f (u− ) f (u+ ) − f (u∗ )
≥ . (2.16)
u∗ − u− u+ − u∗
Interpreting the two sides of (2.16) as slopes of secant lines to the graph of f , as shown in
Fig. 4 one obtains the following stability conditions.

(i) when u− < u+ the graph of f should remain above the secant line through u− , u+ .

(ii) when u− > u+ , the graph of f should remain below the secant line through u− , u+ .

The two above cases are equivalent to one single inequality, namely

speed of the shock [u− , u+ ] ≤ speed of any intermediate shock [u− , u∗ ]. (2.17)

For every intermediate state u∗ between u− and u+ , the stability of the shock thus requires

f (u+ ) − f (u− ) f (u∗ ) − f (u− )

≤ . (2.18)
u+ − u− u∗ − u−

The formulation (2.18) is particularly important, because it can be extended to any n × n

strictly hyperbolic system of conservation laws. To state this general admissibility condition,
we recall that, for any given left state u− ∈ Rn and i ∈ {1, . . . , n}, one can find a curve
s 7→ Si (s) of right states which can be connected to u− by an i-shock. More precisely (see
Fig. 6),
d
Si (0) = u− , Si (s) = ri (u− ),
ds s=0
and for every s the Rankine-Hugoniot equations hold:

f Si (s) − f (u− ) = λi (s) Si (s) − u− ,

(2.19)

for some speed λi (s).

u2 u+ = Si (σ)

_ u*= S i(s)
u
_
ri(u )
u1

Figure 6: The i-shock curve s 7→ Si (s) through the left state u− .

The Liu admissibility condition for general shocks can now be stated as follows.

7
 u+

u*

_
u

Figure 7: According to the Liu condition, the shock with left and right states u− , u+ is admissible if
its speed is slower that the speed of every intermediate shock, joining the states u− , u∗ .

Definition 2.4. A shock of the i-th family, connecting the states u− and u+ = Si (σ) is
Liu-admissible if the speeds of all intermediate shocks satisfy
λi (s) ≥ λi (σ) for all s ∈ [0, σ]. (2.20)

2. Vanishing viscosity limits. From physical considerations, it is often natural to assume

that the “good” solutions to the system of conservation laws (2.1) are those obtained as limit
of solutions to vanishing viscosity approximations
uεt + f (uε )x = ε uεxx . (2.21)
In particular, we say that the shock solution (2.11) is vanishing viscosity admissible if it
can be obtained as a limit of solutions to (2.21).
In the case of a single shock, the equivalence between the vanishing viscosity and the Liu
admissibility condition for a single shock has been proved in [56] and in [7]. More generally,
the analysis in [10] shows that every limit of vanishing viscosity approximations satisfies the
Liu condition at every point of approximate jump.

3. Entropy admissibility condition. Given the hyperbolic system of conservation laws

(2.1), a scalar function η(u) is called an entropy with entropy flux q(u) if
Dη(u) · Df (u) = Dq(u). (2.22)
We observe that (2.22) is a system of n first order PDEs for the 2 functions η, q of the variables
(u1 , . . . , un ). In general, this is overdetermined and has no solution if n > 2. However, there
are relevant physical systems where a nontrivial entropy can still be found.
By (2.22), every smooth solution to (2.1) satisfies the additional conservation law
η(u)t + q(u)x = Dη(u) ut + Dη(u)Df (u)ux = 0.
On the other hand, the entropy may not be conserved in the presence of shocks.

Definition 2.5. Assume that the hyperbolic system of conservation laws (2.1) admits a con-
vex entropy η(u) with entropy flux q(u). We say that a weak solution u = u(t, x) is
entropy admissible if it satisfies the inequality
η(u)t + q(u)x ≤ 0 (2.23)
in distributional sense. That means
ZZ
for all ϕ ∈ Cc1 ,

η(u)ϕt + q(u)ϕx dxdt ≥ 0 φ ≥ 0. (2.24)

8
As a special case, the shock solution at (2.11) is entropy-admissible iff

λ η(u+ ) − η(u− ) ≥ q(u+ ) − q(u− ).

 
(2.25)

It is well known that, if a convex entropy exists, every limit of vanishing viscosity approxima-
tions satisfies the entropy admissibility conditions [16, 41].

Remark 2.1. In the classical theory of gas dynamics, the second law of thermodynamics
implies that the physical entropy should increase in time. To reconcile this fact with the
decrease in the entropy stated at (2.23), it suffices to observe that the physical entropy is
concave down, while the entropies considered by the mathematical theory are always convex
functions. Hence the change in the sign.

3 Approximation methods

Several techniques for constructing approximate solutions to the Cauchy problem

ut + f (u)x = 0, u(0, ·) = ū, (3.1)

have been considered in the literature. Generally speaking, these methods are known to
converge to the exact solution in two main cases:

(i) For a scalar conservation law, based on Kuznetsov’s estimates. See for example [48].

(ii) For general n × n systems, as long as the exact solution remains C 1 , i.e., continuously
differentiable.

On the other hand, for weak solutions to n × n hyperbolic systems, the convergence of approx-
imations requires a careful analysis. In various cases, the convergence still remains an open
problem.
In many algorithms, a basic building block is provided by the Riemann problem, where the
initial datum is piecewise constant with one single jump at the origin:
(
u− if x < 0,
u(0, x) = +
(3.2)
u if x > 0.

For an n × n hyperbolic system, assuming that every characteristic field is either genuinely
nonlinear or linearly degenerate, the general solution to the Riemann problem was first con-
structed by Lax [51]. As shown in Fig. 8, it consists of n + 1 constant states

u− = ω0 , ω1 , ··· , ωn = u+ ,

where each couple of states (ωk−1 , ωk ) are separated by an admissible shock or by a centered
rarefaction wave of the k-th family. Solutions u = u(t, x) to a Riemann problem are self-
similar, in the sense that they are invariant w.r.t. a rescaling symmetry:

u(t, x) = u(θt, θx) for all θ > 0.

9
 t

ω2
ω1

ω = u−
0
ω3 = u+

0 x

Figure 8: A typical solution to a Riemann problem for a 3 × 3 hyperbolic system.

A brief survey of different approximation methods for the Cauchy problem (3.1) is given below.

1. The upwind Godunov scheme. To simplify the presentation, we shall assume that all
characteristic speeds satisfy
λi (u) ∈ [0, 1]. (3.3)
This is not restrictive, because if λi (u) ∈ [−M, M ] one can achieve (3.3) by the simple
coordinate change
x̃ = x + M t, t̃ = 2M t. (3.4)

The Godunov (upwind) scheme starts by constructing a grid in the t-x plane with step size
∆t = ∆x = ε, see Fig. 9.

• The grid points are (tj , xk ) = (j · ∆t , k · ∆x).

• At each time tj , j ≥ 0, the approximate solution u(tj , ·) is piecewise constant with jumps
at the points xk :

u(tj , x) = uj,k for xk ≤ x < xj,k+1 .

• For tj ≤ t < tj+1 the solution is computed by solving the corresponding Riemann
problems at each point of jump (tj , xk ), for every integer k.

• At time tj+1 the solution is again approximated by a piecewise constant function, and
the procedure can repeat.

In the Godunov scheme, at the time tj+1 the function

.
u(tj+1 −, ·) = lim u(t, ·)
t→tj+1 −

is replaced by a piecewise constant function, equal to its average on each interval [xk , xk+1 ].
Namely Z xk+1
. 1
uj+1,k = u(tj+1 −, x) dx . (3.5)
xk+1 − xk xk

10
 t

2∆t

∆t

0 ∆x 2∆ x x

Figure 9: An approximate solution obtained by solving a Riemann problem at each node of the grid.

A remarkable property of this scheme is that, in order to compute uj+1,k there is no need
to actually construct the solution to a Riemann problem. Indeed, applying the divergence
theorem on each square of the grid (see Fig. 10), by the conservation law (2.1) one immediately
obtains
uj+1,k = uj,k + f (uj,k−1 ) − f (uj,k ). (3.6)
The finite difference scheme (3.6) is called the (upwind) Godunov scheme. While this is easy
to implement numerically, a rigorous convergence analysis of this scheme is still lacking, for
general solutions containing shocks.
u j+1,k

f(u j,k−1) f(u j,k )

u j,k−1 u j,k

Figure 10: By the conservation law (2.1), the average value of the solution u on the top side of the
square is equal to the value uj+1,k computed by the formula (3.6).

2. The Glimm scheme. This scheme is similar to the Godunov scheme, with one major
difference. At each time tj , we have to replace the function u(tj −, ·) with a piecewise constant
function having jumps at the points xk . In the Godunov scheme this is achieved by computing
the average value on each interval [xk , xk+1 ]. In the Glimm scheme [44], the restarting is
achieved by random sampling. Namely, inside each interval [xk , xk+1 ] we choose a random
point x∗j,k . The value u(tj −, x∗j,k ) of the solution to the Riemann problem at this particular
point is taken to be the value of the function u(tj , ·) on the entire interval.
More precisely:

• We choose a random sequence of numbers θ1 , θ2 , θ3 , . . . uniformly distributed on [0, 1].

• At each time tj , for every k we consider the random point x∗j,k = (k + θj ) · ∆x and define

u(tj , x) = u(tj −, x∗j,k ) for all x ∈ [xk , xk+1 [ . (3.7)

11
 t

2∆t * * * * * θ = 1/3
2

∆t * * * * * θ1=1/2

0 ∆x 2∆ x x

Figure 11: An approximate solution obtained by solving a Riemann problem at each node of the grid.
In the Glimm scheme, at each time tj = j ∆t the solution is sampled at the points marked by an
asterisk, depending om the random sequence θ1 , θ2 , . . .

As later proved by T.P.Liu [55], instead of a random sequence one can use a deterministic
sequence of numbers θ1 , θ2 , θ3 , . . . ∈ [0, 1] which is uniformly distributed, so that

# j ; 1 ≤ j ≤ N, θj ∈ [0, λ]
lim = λ for each λ ∈ [0, 1]. (3.8)
N →∞ N

θ3 θ1 θ2

0 λ 1

Figure 12: A sequence of numbers in the interval [0, 1]. For every λ, the percentage of points θi ,
1 ≤ i ≤ N which fall inside the subinterval [0, λ] should approach λ as N → ∞.

A simple way of generating such a sequence is to write decimal digits in inverse order:

θ1 = 0.1 , ... , θ759 = 0.957 , ... , θ39022 = 0.22093 , ... (3.9)

The relevance of the assumption (3.8) is illustrated in Fig. 13. As in (2.11), consider a solution
containing a single shock, traveling with speed λ ∈ ]0, 1[ .
Fix a time interval [0, T ] and take ∆x = ∆t = T /N . Call x(tj ) the location of the shock at
time tj , in the approximate solution. By construction (see Fig. 13, left), we have

• If θj ≤ λ, then x(tj ) = x(tj−1 ) + ∆x.

• If θj > λ, then x(tj ) = x(tj−1 ).

Hence, at time T , the position of the shock in the approximate solution is


x(T ) = # j ; 1 ≤ j ≤ N, θj ∈ [0, λ] · ∆t

# j ; 1 ≤ j ≤ N, θj ∈ [0, λ]
= · T → λT as N → ∞,
N

12
 θj < λ * *
t
*
*
_
u=u *
* u = u+
*
*
θj > λ * *
*
∆t *

∆x x

Figure 13: Left: at each time step, the position of the shock remains unchanged if θj > λ, while it
jumps forward if θj < λ. Right: a single shock solution with speed λ ∈ [0, 1] and an approximate
solution constructed by the Glimm scheme. As the grid size approaches zero, convergence to the exact
solution is achieved if and only if the limit in (3.8) holds.

provided that the uniform distribution assumption (3.8) holds.

3. Front tracking approximations. In the Glimm scheme, the Riemann problems are
solved on a fixed grid in the t-x plane. In a front tracking algorithm, the points where new
Riemann problems are solved depend on the solution itself.

• The construction starts by approximating the initial data ū ∈ L1 with a piecewise

constant function v̄.
• At each point where v̄ has a jump, an approximate solution to the Riemann problem is
constructed, within the class of piecewise constant functions. As shown in Fig. 14, this
solution can be prolonged up to the first time t1 where two fronts interact.
• At time t1 , we construct a piecewise constant approximate solution to the new Riemann
problem generated by the interaction, and prolong the solution until a further interaction
occurs.
• By inductively solving the new Riemann problems at the times t2 , t3 , . . . where two fronts
interact, a piecewise constant approximate solution is constructed for all times t ≥ 0.

For a single conservation law, the front tracking method was first introduced by Dafermos [38].
To apply this method to n × n systems, one needs to make sure that the number of wave fronts
does not become infinite in finite time. This requires some technical provision, such as the
introduction of “non-physical fronts” [5, 13]. For a comprehensive presentation we refer to
[16, 41, 48].

4. Vanishing viscosity approximations. Starting with the hyperbolic system (2.1) and
adding a small diffusion term, one obtains the quasilinear parabolic system:

uεt + A(uε )uεx = εuεxx , u(0, x) = ū(x). (3.10)

13
 t
t4

t3

t2
t1

0 x

Figure 14: An approximate solution constructed by front tracking.

Here A(u) = Df (u) is the n × n Jacobian matrix of the flux function. Letting ε → 0+, it is
expected that the solutions to (3.10) will converge to the unique solution to the hyperbolic
Cauchy preblem (3.1). For initial data with small total variation, a rigorous proof of this
convergence was given in [10].

5. Jin-Xin relaxation approximations. These are obtained by solving the second order
wave equation
ut + f (u)x = ε(uxx − utt ) , u(0, x) = ū(x)

As ε → 0, uniform BV bounds and convergence to a unique limit have been proved by

S. Bianchini in [9]. The convergence rate has not been studied in detail.

6. The method of lines. In this case, approximate solutions are obtained by discretizing
space while keeping time continuous.
Fix a mesh size ∆x = ε > 0. Approximating the partial derivative f (u)x by a finite difference,
the system of conservation laws (2.1) is replaced by a countable family of ODEs

d f (Uk−1 (t)) − f (Uk (t))

Uk (t) = , k ∈ Z,
dt ε
for the variables Uk (t) ≈ u(t, kε), see Fig. 15.
As ε → 0, uniform BV bounds and convergence to a unique limit have been proved by
S. Bianchini in [8].
Uk−1 Uk
t

0 ∆x k ∆x x

Figure 15: The method of lines.

7. Backward Euler approximations. In this case we discretize time while keeping space

14
continuous. Choosing ∆t = ε as time step, the Backward Euler approximation takes the form

u(t + ε) = u(t) − εf (u(t + ε))x . (3.11)

Setting
v(x) = u(t, x), w(x) = u(t + ε, x), A(w) = Df (w),
at every time step one needs to solve

w(x) = v(x) − εDf (w(x)) wx (x),

which leads to the ODE

 
v(x) − w(x)
w′ (x) = A−1 (w(x)) . (3.12)
ε

By performing a change of coordinates similar to (3.4), one can assume that all characteristic
speeds (i.e., all eigenvalues of the Jacobian matrix A(u) = Df (u)) are contained inside the
interval [1, 2]. This guarantees that in (3.12) all matrices A(w) have a uniformly bounded
inverse. For a fixed ε > 0, existence and uniqueness of L1 solutions to (3.12) have been
studied in [36], together with traveling wave profiles. However, as ε → 0, uniform BV bounds
and convergence to a unique limit remain an open question.

8. Periodic mollifications. These approximations are again constructed by discretizing

time. Fix ε > 0 and set tk = εk, k = 0, 1, 2, . . . On each subinterval [tk , tk+1 [ , the function u
is defined to be a classical solution:

ut + Df (u)ux = 0, t ∈ [tk , tk+1 [ .

At each time tk , before any shock is formed, the solution is restarted by performing a convo-
lution with a mollifying kernel:
u(tk ) = Jε ∗ u(tk −).

Letting ε → 0, we expect that the approximations should converge to an admissible weak

solution to the original system (2.1). This is well known in the scalar case [48]. However, for
general n × n systems, uniform BV bounds and convergence to a unique limit have not been
proved.

9. Nonlinear diffusion approximations. These take the form

ut + f (u)x = ε B(u)ux x , (3.13)

where B(u) is a (possibly degenerate) n × n diffusion matrix. Since in many physical systems
the viscosity depends on the macroscopic variables, it would be of great interest to prove
rigorous convergence results for solutions of (3.13). However, apart from the case where B is a
constant, invertible matrix [10], establishing uniform BV bounds and convergence to a unique
limit as ε → 0 remains a challenging open problem.

15
4 Global existence of weak solutions

If the initial datum ū : R 7→ Rn is smooth, a unique local in time C 1 solution to the Cauchy
problem (3.1) can be constructed by the method of characteristic, as the fixed point of a
contractive transformation [16, 63]. However, for large times, as shown in Fig. 1 the gradient
of the solution can become unbounded. Global in time solutions can only be obtained in a
space of discontinuous functions.
For scalar conservation laws, also in several space dimensions, a general existence-uniqueness
theorem was proved in the famous paper by Kruzhkov [50]. Shortly afterwards, an alternative
proof based on the theory of nonlinear contractive semigroups was given by Crandall [37].
For n × n hyperbolic systems, the first global existence theorem for weak solutions to the
Cauchy problem (3.1) was proved in a celebrated paper by Glimm [44].

Theorem 4.1. (Global existence of weak solutions). Consider the one-dimensional

Cauchy problem (3.1) for a strictly hyperbolic system of conservation laws, where each char-
acteristic field is either linearly degenerate or genuinely nonlinear.
Then there exists a constant δ > 0 such that, if the initial data satisfies

ū ∈ L1 (R; Rn ), Tot.Var.{ū} ≤ δ,

then (3.1) has a weak solution u = u(t, x) defined for all t ≥ 0.

If the system admits a convex entropy, a global solution exists which is entropy admissible.

The proof is achieved by constructing a sequence of approximate solutions (um )m≥1 according
to the Glimm scheme (see Fig. 11). Here we let the grid size ∆t, ∆x → 0 as m → ∞.

• By carefully estimating the strength of new waves produced by nonlinear wave interac-
tions, one obtains a uniform bound on the total variation of all approximate solutions.
Namely if Tot.Var.{ū} is sufficiently small, then Tot.Var.{um (t, ·)} remains small for all
times t ≥ 0 and every m ≥ 1.

• By Helly’s compactness theorem, one obtains a convergent subsequence um → u in

L1loc (R+ × R ; Rn ).

• Relying on the assumption that the sequence (θk )k≥1 used for random sampling (3.7) is
uniformly distributed on [0, 1], one proves that with probability one the limit function
u = u(t, x) is a weak solution to the Cauchy problem.

For many years, the Glimm scheme provided the only tool for a rigorous analysis of weak
solutions to hyperbolic conservation laws. Among the first such studies, the asymptotic be-
havior of solutions as t → +∞ was analyzed by T.P.Liu [57]. The assumption of genuine
nonlinearity or linear degeneracy of each characteristic field was removed in [58]. Alternative
proofs, relying on front tracking approximations, were later given in [3, 5, 13].
We remark that, in all these results, the smallness of the initial data is a key assumption which
has never been removed. This leads to the following question:

16
If the total variation of the initial data u(0, ·) = ū(·) is bounded but possibly large, does the
total variation of the solution u(t, ·) remain bounded for all times t > 0, or can it blow up in
finite time?
A counterexample constructed by Jenssen [49] shows that, in some cases, the total variation of
a weak solution can indeed blow up in finite time. However, the hyperbolic system considered
in this example does not admit any strictly convex entropy. In particular, the construction
does not apply to any of the systems of conservation laws which are relevant for continuum
physics.
Open Problem #1. Consider a strictly hyperbolic system of conservation laws (2.1), ad-
mitting a strictly convex entropy.

• Construct an example of an entropy admissible weak solution u = u(t, x) whose total

variation blows up in finite time.

• Or else, prove that every solution, whose total variation is initially bounded, remains
with bounded variation for all times t > 0.

τ
T

λT λT+h y

Figure 16: Left: if their speeds are slightly changed, the three wave fronts will interact in different
order, producing new waves of different strengths. Right: a sketch of the interaction pattern considered
in [17], leading to blow up of the total variation in finite time.

The question of global BV bounds versus finite time blow up is not resolved even for the
2 × 2 system of isentropic gas dynamics (2.8). The recent analysis in [17] only points out
how difficult the problem really is. By slightly changing the wave speeds, one can arrange
so that wave fronts cross each other in a different order (see Fig. 16). For the same initial
data, one can construct approximate solutions whose total variation remains bounded, and
other approximate solutions whose total variation blows up in finite time, depending on the
interaction pattern. It is hard to say what happens for the exact solution.

5 Continuous dependence on initial data

For a wide class of evolution equations, continuous dependence on initial data is achieved by
showing that the distance between any two solutions satisfies the differential inequality
d
u(t) − v(t) ≤ C u(t) − v(t) , (5.1)
dt

17
for some constant C. In turn, by Gronwall’s lemma this implies

u(t) − v(t) ≤ eCt u(0) − v(0) . (5.2)

This approach can be applied to scalar conservation laws, even in a multidimensional space
Rd . As proved in [37, 50], a scalar conservation law generates a contractive semigroup on
L1 (Rd ). Namely, for every couple of entropy admissible solutions u, v one has

u(t) − v(t) L1
≤ u(s) − v(s) L1
for all 0 ≤ s < t. (5.3)

On the other hand, the inequality (5.1) does not hold for weak solutions to hyperbolic systems.
As shown in Fig. 17, the L1 distance between two nearby solutions can increase rapidly during
short time intervals.

jumps in u
jumps in v
t

t2 || u(t) − v(t) || 1
L
t1

0 t t t
x 1 2

Figure 17: Left: a solution u which initially contains two approaching shocks, and a second solution
v which differs from u only in the location of the shocks. During the short time interval [t1 , t2 ] the
solution u contains three shocks, while v still has two. Right: the L1 distance between the two solutions
remains constant for t ∈ [0, t1 ] and for t ≥ t2 , but increases rapidly during the interval [t1 , t2 ].

We observe that the Glimm scheme does not provide insight on the continuous dependence of
solutions. Indeed, the approximate solutions constructed by the Glimm scheme do not depend
continuously on initial data. With reference to Fig. 18, left, an arbitrarily small change in the
speed of a shock may place it to the right or to the left of the sampling point x∗j,k . In this
case, the piecewise constant approximation changes value on the entire interval [xk , xk+1 ].
Understanding how weak solutions depend on initial data was the primary motivation for
developing an alternative approximation scheme based on wave front tracking [13]. As shown
in Fig. 18, let u be a piecewise constant approximate solution, and let v be a perturbed solution
obtained by slightly shifting the location of the jumps in u. At any time t ≥ 0, the L1 distance
between the two solutions is measured by
X X
u(t) − v(t) L1 ≈ |σα | · |ξα | = [jump strength] × [shift]. (5.4)
α α

By carefully estimating how the right hand side of (5.4) changes at interaction times, one
obtains a bound on the distance between the two solutions, at every time t ≥ 0.
Theorem 5.1. Let (2.1) be a strictly hyperbolic n×n system of conservation laws, and assume
that each characteristic field is either genuinely nonlinear or linearly degenerate. Then there
exists a domain D ⊂ L1 (R; Rn ) and a semigroup S : D×R+ 7→ D with the following properties.

18
 Glimm scheme front tracking
t

x*j,k
*

x
u
σα ξα

xα x

Figure 18: For the Glimm scheme, approximate solutions do not depend continuously on the initial
data. On the other hand, by shifting the location of the wave fronts, one can understand how front
tracking approximations are affected by changes in the initial data.

(i) The domain D contains all functions ū ∈ L1 with sufficiently small total variation.
.
(ii) The map (ū, t) 7→ u(t, ·) = St ū is a uniformly Lipschitz continuous semigroup:

S0 ū = ū, Ss (St ū) = Ss+t ū,

St ū − Ss v̄ L1
≤ L · ∥ū − v̄∥L1 + |t − s| for all ū, v̄ ∈ D, s, t ≥ 0.

(iii) For every initial data ū ∈ D, the trajectory t 7→ u(t, ·) = St ū provides an admissible
solution to the Cauchy problem (3.1).

Trajectories of the semigroup can be obtained as limits of front tracking approximations. This
theorem was first proved in [19] in the case of 2 × 2 systems. A more elaborate proof, valid
for n × n systems, was later worked out in [21]. These earlier proofs relied on a homotopy
method: the distance between two solutions u, v was estimated by constructing a 1-parameter
family of solutions uθ , θ ∈ [0, 1], connecting u with v. At every time t ≥ 0, the length of the
path θ 7→ uθ (t, ·) provides a bound on the distance u(t) − v(t) L1 .
Using ideas introduced by T.P.Liu and T.Yang [60], an alternative proof was given in [30].
This approach relies on the construction of a Lyapunov functional Φ : D × D 7→ R+ with the
following properties.

• Φ is equivalent to the L1 distance:

1
· v−u L1
≤ Φ(u, v) ≤ C · v − u L1
,
C
for every couple of piecewise constant functions u, v ∈ L1 (R; Rn ) with small total vari-
ation.

19
 • Φ is non-increasing in time, along couples of (front tracking approximate) solutions:

Φ(u(t), v(t)) ≤ Φ(u(s), v(s)) for all t ≥ s ≥ 0. (5.5)

Given two piecewise constant functions u, v, the functional Φ(u, v) is defined as follows. For
each x ∈ R, we uniquely determine intermediate states

u(x) = ω0 (x), ω1 (x), . . . , ωn (x) = v(x),

such that every pair ωi−1 (x), ωi (x) is joined by a (possibly non-admissible) shock of the i-th
family (see Fig. 19). Calling qi (x) the strength of this shock, the L1 distance between u and
v can now be estimated as
Z +∞ X n
∥u − v∥L1 ≈ |qi (x)| dx .
−∞ i=1

To achieve the decreasing property (5.5), suitable weights Wi must be inserted. Roughly
speaking, Wi (x) measures the total strength of waves in u and in v that approach an i-shock
located at x. The functional Φ thus takes the form
Xn Z ∞
Φ(u, v) = |qi (x)| Wi (x) dx .
i=1 −∞

For all details we refer to [30]. This approach greatly simplified the earlier proofs, and is now
adopted in most textbooks on the subject [16, 41, 48].

ω3 = v(x)
λ3 σα xα
q v
3
ω2

λ1 ω1 u
q
1
ω0= u(x)
x xα

Figure 19: Constructing the functional Φ(u, v). At each point x, the strengths q1 (x), . . . , qn (x) of the


shocks connecting u(x) with v(x) can be regarded as the scalar components of the jump u(x), v(x) .

All previous results were proved in the same setting as in Glimm’s theorem, where each char-
acteristic field is either linearly degenerate or genuinely nonlinear. Eventually, this assumption
was entirely removed by the approach based on vanishing viscosity approximations [10]. In
this case, one does not even need that the hyperbolic system be in conservation form.
Theorem 5.2. Consider the Cauchy problem for a strictly hyperbolic system with small vis-
cosity
ut + A(u)ux = ε uxx , u(0, x) = ū(x) . (5.6)
If Tot.Var.{ū} is sufficiently small, then (5.6) admits a unique solution uε (t, ·) = Stε ū, defined
for all t ≥ 0. Moreover, for some constants C, L independent of ε, one has

Tot.Var. Stε ū ≤ C Tot.Var.{ū} ,


(BV bounds)

20
 Stε ū − Stε v̄ L1
≤ L ∥ū − v̄∥L1 . (L1 stability)
As ε → 0, the solutions uε converge to the trajectories of a semigroup S such that

St ū − St v̄ L1
≤ L ∥ū − v̄∥L1 for all t ≥ 0 .

If the system is in conservation form: A(u) = Df (u) for some flux function f , then every
trajectory t 7→ u(t, ·) = St ū provides a weak solution to the Cauchy problem

ut + f (u)x = 0, u(0, x) = ū(x).

Moreover, the Liu admissibility conditions (2.20) are satisfied at every point of approximate
jump.

For a proof, see [10] or the lecture notes [14]. These vanishing viscosity limits can be regarded
as the unique viscosity solutions of the hyperbolic Cauchy problem

ut + A(u)ux = 0 u(0, x) = ū(x) .

We remark that all results stated in Theorem 5.2 hold for the system (5.6) with “artificial
viscosity”, where the diffusion acts uniformly on all components (u1 , u2 , . . . , un ) of the solution.
In several physical models, the diffusion acts differently on different components, and depends
on the state u as well. This leads to
Open Problem #2. Extend the results of Theorem 5.2 to hyperbolic systems with nonlinear
diffusion:

ut + A(u)ux = ε B(u)ux x , u(0, ·) = ū , (5.7)
where B(u) is a positive semidefinite viscosity matrix. Assuming that the initial data ū has
small total variation, prove uniform bounds on Tot.Var.{u(t, ·) and study the limit of these
solutions as ε → 0+.
In the conservative case where A(u) = Df (u), we expect that the vanishing viscosity limit
should be unique, and provide a weak solution to the hyperbolic system (2.1). On the other
hand, in the non-conservative case, as ε → 0+ different limits may well be obtained, depending
on the choice of the viscosity matrices B(u) in (5.7).

6 Uniqueness of weak solutions

Having constructed a Lipschitz semigroup of admissible weak solutions, which are limits of
vanishing viscosity approximations (and of front tracking approximations as well), it becomes
entirely clear which is the unique “good” solution to the Cauchy problem (3.1). Namely,
the semigroup trajectory t 7→ u(t, ·) = St ū. From this point of view, uniqueness becomes a
marginal issue in the overall theory. Some authors barely mention the problem [59], focusing
instead all the attention on the continuous dependence on initial data.
On the other hand, a general uniqueness theorem can be quite useful if we want to study the
convergence of different approximation methods. Without a uniqueness result, one may even
suspect that these algorithms converge to different limit solutions.

21
As soon as a semigroup of solutions has been constructed, to establish a uniqueness theorem
it suffices to come up with a set of conditions that uniquely characterize the semigroup tra-
jectories. Various ways to do this have been worked out in [23, 28, 29]. The proofs rely on the
following elementary error estimate (see Fig. 20).

Lemma 6.1. Let S : D × [0, ∞[ 7→ D be a Lipschitz semigroup satisfying

∥St u − Ss v∥ ≤ L · ∥u − v∥ + L′ · |t − s| .

Then, for every Lipschitz continuous map w : [0, T ] 7→ D one has

Z T( )
w(t + h) − Sh w(t)
w(T ) − S w(0) ≤ L · lim inf dt . (6.1)
T
0 h→0+ h

For a proof, see [15, 16]. The integrand on the right hand side of (6.1) can be interpreted as
the instantaneous error rate of the approximate solution w(·) at time t.

w(T)

S w(t+h)
w(t+h) T−t−h
S w(t)
w(t) Shw(t) T−t

S w(0)
T
w(0)

Figure 20: The distance between w(T ) and ST w(0) is bounded by the length of the path t 7→ ST −t w(t),
t ∈ [0, T ]. Since the semigroup amplifies distances at most by a factor L, this leads to the formula
(6.1).

To prove that a solution u = u(t, x) of the Cauchy problem (3.1) coincides with the semigroup
trajectory, it now suffices to show that

u(τ + h) − Sh u(τ ) L1
lim inf = 0 (6.2)
h→0+ h

for a.e. time τ ∈ [0, T ].

To fix ideas, w.l.o.g. we assume that all wave speeds (i.e., all eigenvalues of the matrices
A(u) = Df (u)) are contained in the interval [−1, 1]. Given the function u(τ, ·), following the
approach introduced in [15] we split the real line inserting points
. .
−∞ = x0 < x1 < x2 < · · · < xN = +∞,

so that (see Fig. 21)


Tot.Var. u(τ, ·) ; ]xi−1 , xi [ < ε, i = 1, 2, . . . , N. (6.3)

22
 u(τ)

τ +h

τ
x xk x
k−1 k+1

Figure 21: Given the function u(τ, ·), we insert points xk so that the total variation on each open
interval ]xk−1 , xk [ is < ε.

We now estimate
1 ∞
Z
u(τ + h, x) − Sh u(τ )(x) dx
h −∞
X 1 Z xk +h X 1 Z xk −h
= u(τ + h, x) − Sh u(τ )(x) dx + u(τ + h, x) − Sh u(τ )(x) dx
h xk −h h xk−1 +h
k
X X k
= Ak (h) + Bk (h) .
k k
(6.4)
We claim that, by choosing ε > 0 small, the limit as h → 0 of the right hand side of (6.4) can
be made arbitrarily small. Indeed, assume that at (τ, xk ) the solution u is either approximately
continuous, or has an approximate jump, as in (2.14). Then

• On each interval [xk − h, xk + h], the function u(τ + h, ·) and the semigroup solution
Sh u(τ ) are both compared with the piecewise constant function
( .
. u+ = u(τ, xk +) if x > xk + λ(t − τ ),
Uk (t, x) = − .
u = u(τ, xk −) if x < xk + λ(t − τ ).

• On each of the remaining intervals [xk−1 + h, xk − h] the function u(τ + h, ·) and the
semigroup solution Sh u(τ ) are both compared with the solution Wk of the linear Cauchy
problem with constant coefficients

wt + Ak wx = 0 w(τ, x) = u(τ, x), (6.5)

.

where Ak = Df u(τ, yk ) for some yk ∈ ]xk−1 , xk [ .

These comparisons are based on two lemmas. Equivalent results were proved in Theorem 2.6
of [16] and in [25], respectively. Since they play a crucial role in the uniqueness results, we
include here the complete proofs.

Lemma 6.2. Assume that the map t 7→ u(t, ·) is Lipschitz continuous with values in L1 (R).
Moreover, let (τ, ξ) be a point of approximate jump for u, so that (2.14) holds. Then
Z h
1
lim u(τ + h, ξ + x) − U (h, x) dx = 0. (6.6)
h→0+ h −h

23
Proof. Assume that, on the contrary, there exists a decreasing sequence hm → 0 such that
Z hm
1
u(τ + hm , ξ + x) − U (hm , x) dx ≥ δ > 0 for all m ≥ 1.
hm −hm

By Lipschitz continuity, for some constant L this implies

Z hm
u(τ + hm + s, ξ + x) − U (hm + s, x) dx ≥ δhm − Ls for all m ≥ 1, s ≥ 0.
−hm

Setting
δ
s̄ = , rm = hm (1 + s̄),
L
we now obtain
Z rm Z rm
1
2
u(τ + t, ξ + x) − U (t, x) dxdt
rm −rm −rm
Z (1+s̄)hm Z hm
1
≥ u(τ + t, ξ + x) − U (t, x) dxdt
(1 + s̄)2 h2m hm −hm
Z s̄hm
1 1 δ2
≥ (δhm − Ls) ds = · > 0.
(1 + s̄)2 h2m 0 (1 + s̄)2 2L

This contradicts the assumption (2.14).

Estimating the difference u(τ + h, ·) − Wk (τ + h, ·) requires more work. Indeed, here we are
approximating u with the solution to a linearized problem. This is accurate as long as u
remains close to a constant, not only at time τ , but on the entire trapezoidal domain
n o
Γk = (t, x) ; t ∈ [τ, τ + h], x ∈ Jk (t) , (6.7)
. 
Jk (t) = ]xk−1 + (t − τ ), xk − (t − τ ) , (6.8)
as shown in Fig. 21. For a solution constructed by front tracking, or by the Glimm scheme,
the assumption that u(τ, ·) has small total variation on ]xk−1 , xk [ implies that u has few waves
(and hence is nearly constant) also on the domain of dependence Γk in (6.7). However, in
principle this may not be true for more general weak solutions. For example, the analysis
in [6] shows that the Godunov scheme can ampify the total variation by an arbitrarily large
factor.
For this reason, additional assumptions were required in earlier papers. Namely “Tame Vari-
ation” in [28], “Tame Oscillation” in [23], and “Bounded variation along space-like curves”
in [29]. For a class of 2 × 2 systems, the recent analysis in [34] has shown that any entropy
admissible weak solution taking values in the domain of the semigroup always satisfies the
bounded variation condition in [29]. Therefore, one does not need this additional assumption
to achieve uniqueness.
Following [22, 25], we show here that uniqueness holds also for fully general n × n systems,
without any of the previous regularity assumptions.
To appreciate the underlying idea, consider the scalar function

V (t) = Tot.Var. u(t, ·) ; Jk (t) ,

24
where Jk (t) is the interval introduced in (6.8). Notice that V is lower semicontinuous, hence
measurable. Assume that τ is a Lebesgue point for V . Since V (τ ) ≤ ε, this implies that the
set of times where V (t) > 2ε is very small. Indeed,

meas t ∈ [τ, τ + h] ; V (t) > 2ε
lim = 0. (6.9)
h→0+ h

At time τ + h the difference between u and the solution Wk to the linearized equation (6.5)
can be estimated as
Z xk −h
.
Ek (h) = u(τ + h, x) − Wk (τ + h, x) dx
xk−1 +h
Z τ +h



= O(1) · Tot.Var. u(t, ·); Jk (t) · Df u(t, ·) − Df u(τ, yk ) dt (6.10)
τ L∞
Z τ +h
= O(1) · V (t) · u(t, ·) − u(τ, yk ) L∞
dt.
τ

If V (t) ≤ 2ε for all t ∈ [τ, τ + h], we would be in the Tame Variation case. Both factors in the
integrand on the right hand side of (6.10) have size O(1) · ε. Hence Ek = O(1) · hε2 , as proved
in [16, 15, 28]. In the general case, there is an additional error, measured by how much the
solution u can change during the intervals of time where V (t) > 2ε. If τ is a Lebesgue point
for V , since the map t 7→ u(t, ·) is Lipschitz continuous, by (6.9) we obtain the slightly weaker
bound
Ek (h) = O(1) · hε2 + o(h),
where the Landau symbol o(h) denotes a higher order infinitesimal as h → 0. We now state
a more precise result in this direction. As before, we assume that all characteristic speeds are
contained in the interval [−1, 1].

Lemma 6.3. Let t 7→ u(t, ·) be a weak solution to the strictly hyperbolic system (2.1), Lipschitz
continuous with values in L1 (R; Rn ). Assume that

(i) Tot.Var. u(t, ·) ≤ M for all t ≥ 0.

(ii) Tot.Var. u(τ, ·) ; ]a, b[ ≤ ε.

(iii) Setting n  o
V (t) = Tot.Var. u(t, ·) ; a + (t − τ ) , b − (t − τ ) , (6.11)

the limit (6.9) holds.

Call W = W (t, x) the solution to the linear Cauchy problem with constant coefficients

wt + Aw
e x = 0, w(τ, x) = u(τ, x), x ∈ R, (6.12)


e = Df u(τ, ξ) for some ξ ∈ ]a, b[ . Then there holds
with A
Z b−h
1
lim sup u(τ + h, x) − W (τ + h, x) dx = O(1) · ε2 . (6.13)
h→0+ h a+h

25
Proof. By Theorem 4.3.1 in [41], the bound (i) on the total variation implies that the map
t 7→ u(t, ·) is Lipschitz continuous with values in L1 (R; Rn ).
Let λi (u), li (u), ri (u), i = 1, . . . , n, be respectively the eigenvalues and the left and right
eigenvectors of the matrix A(u) = Df (u). For notational convenience, call ũ = u(τ, ξ), and
let
λ̃i = λi (ũ), ˜li = li (ũ), r̃i = ri (ũ),
be the corresponding eigenvalues and left and right eigenvectors of the matrix A
e = Df (ũ).

Since W solves the linear problem (6.12), one has

˜li · W (t, x) = ˜li · W τ, x − (t − τ )λ̃i = ˜li · u τ, x − (t − τ )λ̃i .

Following the proof of Theorem 9.4 in [16], fix any two points
. 
ζ ′ , ζ ′′ ∈ J(t) = a + (t − τ ), b − (t − τ ) .

(6.14)

Assuming ζ ′ < ζ ′′ , consider the quantity

Z ζ ′′
.
Ei (ζ ′ , ζ ′′ ) = ˜li ·


u(t, x) − W (t, x) dx
ζ′
Z ζ ′′  (6.15)


= ˜li · u(t, x) − u τ, x − (t − τ )λ̃i dx.
ζ′


We apply the divergence theorem to the vector u, f (u) on the domain
.
n o
Di = (s, x); s ∈ [τ, t], ζ ′ − (t − s)λ̃i ≤ x ≤ ζ ′′ − (t − s)λ̃i , (6.16)

shown in Fig. 22.

ζ’ ζ"
t
Di
s J(s)

τ a x
b

Figure 22: The domain Di considered at (6.16).

Since u satisfies the conservation equation (2.1), the difference between the integral of u at the
top and at the bottom of the domain Di is measured by the inflow from the left side minus
the outflow from the right side of Di . By (6.15) it thus follows
Z t  
′ ′′ ˜li · f (u) − λ̃i u (s, ζ ′ − (t − s)λ̃i ) ds


Ei (ζ , ζ ) =
τ Z t  
˜li · f (u) − λi u (s, ζ ′′ − (t − s)λ̃i ) ds


−
Z t τ (6.17)


li (ũ) · f (u′ (s)) − λi (ũ) u′ (s) − f (u′′ (s)) − λi (ũ)u′′ (s) ds


=
Zτ t
F ũ, u′ (s), u′′ (s) ds,


=
τ

26
where we set
. .
u′ (s) = u s, ζ ′ − (t − s)λ̃i , u′′ (s) = u s, ζ ′′ − (t − s)λ̃i ,

.



F (u, u1 , u2 ) = li (u) · f (u1 ) − λi (u)u1 − f (u2 ) − λi (u)u2 .

Observing that

• F (u, u2 , u2 ) = 0,

• Du1 F (u, u1 , u2 ) = li (u) · (Df (u1 ) − λi (u)I),

• Du1 F (u, u, u2 ) = li (u) · (Df (u) − λi (u)I) = 0,

one obtains

F (u, u1 , u2 ) = F (u, u1 , u2 ) − F (u, u2 , u2 )

Z 1
= Du1 F (u, u2 + σ (u1 − u2 ) , u2 ) dσ · (u1 − u2 )
0
Z 1
= [Du1 F (u, u2 + σ (u1 − u2 ) , u2 ) − Du1 F (u, u, u2 )] dσ · (u1 − u2 )
0


= O(1) · |u1 − u| + |u2 − u| · |u1 − u2 | .

In turn, this yields

Z t  
′ ′′
Ei (ζ , ζ ) = O(1) · u′ (s) − u′′ (s) · u′ (s) − ũ + u′′ (s) − ũ ds. (6.18)
τ

Recalling the definitions at (6.11) and (6.14), for any x ∈ J(s) we now compute

u′ (s) − ũ ≤ V (s) + u s, x − u (τ, x) + |u (τ, x) − ũ| ≤ V (s) + u s, x − u (τ, x) + V (τ ).

Integrating w.r.t. x over the interval J(s), dividing by its length and using the Lipschitz
continuity of the map t 7→ u(t, ·), we obtain
Z
1
|u′ (s) − ũ| ≤ V (s) + V (τ ) +
u(s, x) − u(τ, x) dx
meas J(s) J(s) (6.19)
= V (s) + ε + O(1) · (s − τ ) =
˙ g(s).

An entirely similar estimate holds for u′′ (s) − ũ . Therefore

Z t
Ei (ζ ′ , ζ ′′ ) = O(1) · u′ (s) − u′′ (s) · g(s) ds
τ
Z t n o
Tot.Var. u(s) ; ζ ′ − (t − s)λ̃i , ζ ′′ − (t − s)λ̃i · g(s) ds

= O(1) ·
τ

= O(1) · µi ]ζ ′ , ζ ′′ ] .

27
Here µi is the Borel measure defined by
Z t n

 o
µi ]c, d[ = Tot.Var. u(s) ; c − (t − s)λ̃i , d − (t − s)λ̃i · g(s) ds,
τ

for any open interval ]c, d[ ⊂ J(t).

As proved in Lemma 9.3 of [16], one has
Z n Z
X
˜li · u(t, x) − W (t, x)


u(t, x) − W (t, x) dx = O(1) · dx
J(t) i=1 J(t)
Xn Z t
= O(1) · µi (J(t)) = O(1) · V (s) · g(s) ds.
i=1 τ

In turn, this implies

∥g∥∞ t V (τ ) t
Z  Z Z 
1
u(t, x) − W (t, x) dx = O(1) · V (s) − V (τ ) ds + g(s) ds .
t − τ J(t) t−τ τ t−τ τ
(6.20)
We now observe that, for all s > τ sufficiently close to τ , the function g introduced at (6.19)
satisfies
g(s) ≤ V (s) + ε + O(1) · (t − τ ) . (6.21)
Thanks to (6.9), taking the limit of (6.20) as t → τ + we thus obtain
Z
1
lim sup u(t, x) − W (t, x) dx = O(1) · V (τ ) (V (τ ) + ε) = O(1) · ε2 , (6.22)
t→τ + t − τ J(t)

proving (6.13).

Going back to instantaneous error estimate (6.4), by Lemma 6.2 it follows

lim Ak (h) = 0 for all k = 1, . . . , N − 1. (6.23)
h→0+

Moreover, using Lemma 6.3, for some constant C and every k = 1, . . . , N we obtain
lim sup Bk (h) ≤ C ε2 . (6.24)
h→0+

Observing that the number of intervals in the partition is N = O(1) · ε−1 , we conclude
N −1 N
!
X X
lim sup Ak (h) + Bk (h) ≤ 0 + N · Cε2 = O(1) · ε. (6.25)
h→0+ k=1 k=1

Since ε > 0 can be chosen arbitrarily small, this implies (6.2).

We can now state the main uniqueness theorem in [22].

Theorem 6.1. Let (2.1) be a strictly hyperbolic n×n system of conservation laws and consider
the semigroup of vanishing viscosity solutions S : D × R+ 7→ D, constructed in Theorem 5.2.
Then, any weak solution t 7→ u(t, ·) ∈ D, which takes values in the domain of the semigroup and
satisfies the Liu admissibility conditions (2.20) at every point of approximate jump, coincides
with a semigroup trajectory:
u(t, ·) = St u(0) for all t ≥ 0. (6.26)

28
Notice that, by Theorem 5.2, every limit of the vanishing viscosity approximations (5.6) is
a weak solution to (2.10) and satisfies the Liu admissibility conditions. The above result
provides a converse: every weak solution to (2.10) which is Liu-admissible (and has suitably
small total variation, so it lies within the domain D) actually coincides with a semigroup
trajectory. Therefore it is obtained as the unique limit of the viscous approximations (5.6).
Sketch of the proof. 1. The assumption that u = u(t, x) is a weak solution and its
total variation Tot.Var. u(t, ·) remains uniformly bounded implies that u : R+ × R 7→ Rn
is a BV function of the two variables t, x. By a general structure theorem, the set of its
approximate jumps is countably rectifiable [1], i.e., it can be covered by countably many
Lipschitz curves (see Fig. 23, left). However, since u is a solution to a hyperbolic system,
at each point of approximate jump the Rankine-Hugoniot equations hold. In particular, the
speed of these jumps must be uniformly bounded. As proved in [22], the set of approximate
jumps is contained in the graphs of countably many Lipschitz functions (see Fig. 23, left).

x = ϕℓ (t), ℓ ≥ 1. (6.27)

To simplify our notation, w.l.o.g. we shall assume that all characteristic speeds λi (u) are
contained in the interval [−1, 1] and all functions ϕℓ have Lipschitz constant 1.

2. In addition to the functions ϕℓ we consider the countably many functions

ϕξ+ (t) = ξ + t, ϕζ− (t) = ξ − t, (6.28)

where ξ ∈ Q is rational. We relabel the set of all these functions as

{ϕℓ ; ℓ ≥ 1} ∪ {ϕξ+ ; ξ ∈ Q} ∪ {ϕζ− ; ζ ∈ Q} = {ψj ; j ≥ 1}, (6.29)

and consider the countably many functions

 n o
 Tot.Var. u(t, ·) ; ψ (t), ψ  if ψi (t) < ψj (t),
. i j
Wij (t) = (6.30)
 0 otherwise.

We observe that each function Wij is measurable. Therefore there exists a null set N such
that every τ ∈ R+ \ N is a Lebesgue point for all the countably many functions Wij .

t t

x x
Figure 23: Left: for a general BV function u : R+ × R 7→ Rn the set of approximate jumps can be
covered by countably many Lipschitz curves. Right: if u is a solution to (2.1), all of its jumps travel
with bounded speed. Hence the set of jumps can be covered by graphs of uniformly Lipschitz functions.

3. As shown in Fig. 24, we now insert points y1 < y2 < · · · < yN so that the total variation
of u(τ, ·) on each open interval ]yk−1 , yk [ is < ε.

29
 • If u(τ, ·) has a jump at yk , then by construction (τ, yk ) lies on the graph of one of the
functions ϕℓ .

• If u(τ, ·) is continuous at yk , then we can find two nearby points yk′ < yk < yk′′ , lying on
one of the rational lines in (6.28), namely

yk′ = ϕξ+ (τ ), yk′′ = ϕζ− (τ ), for some ξ, ζ ∈ Q,

and furthermore
Tot.Var. u(τ, ·) ; [yk′ , yk′′ ]

< ε.

In the end, we can cover the real line with finitely many points yk and open intervals Ik =
]ak , bk [ with the following properties.

(i) At each point yk , the function u(τ, ·) has an approximate jump, satisfying the Liu ad-
missibility condition. By Lemma 6.2 this implies
Z yk +h
1
lim u(t, x) − U (t − τ, x − yk ) dx = 0. (6.31)
h→0+ h yk −h


(ii) On each open interval Ik the total variation satisfies Tot.Var. u(τ, ·) ; Ik < 2ε. More-
over, both endpoints of Ik lie on the graph of one of the functions ψj at (6.29), say

ak = ψi (τ ), bk = ψj (τ ),

for some i, j ≥ 1. Since all functions ϕℓ have Lipschitz constant ≤ 1, for t ≥ τ we have
n  o n  o
Tot.Var. u(t, ·) ; ak + (t − τ ), bk − (t − τ ) ≤ Tot.Var. u(t, ·) ; ψi (t), ψj (t) .

Therefore, calling Wk the solution to the linearized equation

  
. ak + bk 
wt + A
ek wx = 0, w(τ, x) = u(τ, x), A
ek = Df u τ, ,
2

by Lemma 6.3 it follows

Z bk −h
1
lim sup u(τ + h, x) − Wk (τ + h, x) dx = O(1) · ε2 . (6.32)
h→0+ h ak +h

Since the semigroup trajectory v(t, ·) = St−τ u(τ ) satisfies the same estimates (6.31)-(6.32) as
u(t, ·), we conclude that, at any Lebesgue time τ ∈ R+ \ N ,
N
u(τ + h) − Sh u(τ ) L1
X
lim sup ≤ C1 ε2 ≤ C2 ε, (6.33)
h→0+ h
k=1

for some constants C1 , C2 . Indeed, the number of intervals Ik in the partition is N = O(1)·ε−1 .
Since ε > 0 can be chosen arbitrarily small, this achieves the proof.

30
 u (τ)

φ
t

τ y
y’ y" x
k−1 k−1 k

ak Ik bk

Figure 24: Covering the real line with points yk where u(τ, ·) has a jump, and open intervals Ik =
]ak , bk [ where the total variation is < 2ε.

7 Error estimates

Having constructed a Lipschitz semigroup of admissible solutions to (2.1), it is of interest to

estimate the L1 distance between an approximate solution constructed by one of the algorithms
described in Section 3 and the exact solution.
Results in this direction were proved in [15] for the front tracking method, in [4, 31, 12] for
the Glimm scheme, and in [27, 33] for vanishing viscosity approximations.
More precisely, consider the n × n hyperbolic system (2.1), assuming that all characteristic
fields are genuinely nonlinear. The estimate in [33] shows that the distance between the
solution uε of the viscous approximation (3.10) and the exact solution u(t) = St ū to the
Cauchy problem (3.1) can be estimated as
√
uε (t, ·) − u(t, ·) L1 = O(1) · (1 + t) Tot.Var.{ū} ε| ln ε|. (7.1)

Next, consider an approximate solution uGlimm constructed by the Glimm scheme, with a
grid of step size ∆t = ∆x = ε. Choosing sampling points as in (3.9), the analysis in [31] has
established a similar convergence rate:
uGlimm (t, ·) − u(t, ·) L1
lim √ = 0 for all t > 0. (7.2)
ε→0 ε| ln ε|

For other approximation methods, such as periodic mollifications, the backward Euler scheme,
or fully discrete numerical schemes, no a priori BV bounds are currently available. In partic-
ular, it is known that the Godunov scheme can amplify the total variation by an arbitrarily
large factor [6]. For this reason, it seems more promising to look for a posteriori error bounds.
Namely, assume that an approximate solution to the Cauchy problem (3.1) has been con-
structed, whose total variation remains small for all times t ∈ [0, T ]. Using this additional
information, we seek a bound on the error
uapprox (t, ·) − uexact (t, ·) L1
. (7.3)
We outline here an approach which is in a sense “universal”, i.e., it does not make reference
to any particular approximation method.
Given ε > 0, consider an approximate solution u = u(t, x) with the following properties.

31
 Definition 7.1. Let (2.1) be an n×n strictly hyperbolic systems of conservation laws, endowed
with a strictly convex entropy η, with entropy flux q. We say that u = u(t, x) is an ε-
approximate solution to the Cauchy problem (2.10) if u(0, ·) − ū L1 ≤ ε and moreover the
following holds.

(ALε ) Approximate Lipschitz continuity:

∥u(τ, ·) − u(τ ′ , ·)∥L1 ≤ M |τ − τ ′ | + ε for all τ, τ ′ ≥ 0

(Pε ) Approximate conservation law and approximate entropy inequality:

For every strip [τ, τ ′ ] × R and every test function φ ∈ Cc1 (R2 ), one has
Z Z Z τ ′Z
′

u(τ, x)φ(τ, x) dx − u(τ , x)φ(τ, x) dx + uφt + f (u)φx dx dt
τ (7.4)
≤ ε(τ ′ − τ + ε) ∥φ∥W 1,∞ .

Moreover, assuming φ ≥ 0, one has the entropy inequality

Z Z Z τ ′Z
′ ′

η(u(τ, x))φ(τ, x) dx − η(u(τ , x))φ(τ , x) dx + η(u)φt + q(u)φx dxdt
τ
′
≥ − ε(τ − τ + ε) ∥φ∥W 1,∞ .
(7.5)

In the above setting, the paper [18] has established a posteriori error estimates, assuming that
the total variation of the ε-approximate solution remains small, so that u(t, ·) remains within
the domain of the semigroup. However, the estimates in [18] also required a “post processing
algorithm”, tracing the location of the large shocks in the approximate solution. This is related
to the assumptions of “tame variation”, “tame oscillation” or “bounded variation along space-
like curves” which were used respectively in [28], [24] and in [29] to prove uniqueness of
solutions. In essence, these additional assumptions rule out configurations such as the one
shown in Fig. 25.

u
t
u(t 3)
u(t2 )
u(t1 )
x

Figure 25: An approximate solution u where the total variation remains small at all times. However,
oscillations appear and disappear at different regions on the x-t plane.

The recent paper [25] has shown that, for a system endowed with a strictly convex entropy,
these additional regularity conditions are not needed to achieve uniqueness:
Theorem 7.1. Let (2.1) be a strictly hyperbolic n × n system, where each characteristic field
is either genuinely nonlinear or linearly degenerate, and which admits a strictly convex entropy
η(·). Then every entropy admissible weak solution u : [0, T ] 7→ D, coincides with a semigroup
trajectory.

32
We observe that, in the setting of the above theorem, the dissipation of a single entropy suffices
to single out the Liu-admissible shocks.
As proved in [25], the compactness of the family of approximate solutions, together with the
uniqueness of the limit, yields a uniform convergence rate:
Corollary 7.1. In the above setting, given T, R > 0, there exists a function ε 7→ ϱ(ε) with
the following properties.

(i) ϱ is continuous, nondecreasing, with ϱ(0) = 0.

(ii) Let t 7→ u(t, ·) ∈ D be an ε-approximate solution to (3.1), with u(t, ·) supported inside
the interval [−R, R] for all t ∈ [0, T ]. Then one has
u(t) − St ū L1
≤ ϱ(ε) for all t ∈ [0, T ]. (7.6)

This corollary shows that such a “universal rate of convergence” must exist. However, it does
not offer clues on how the function ρ(·) looks like.

Open Problem #3. Let (2.1) be a strictly hyperbolic n×n system, where each characteristic
field is either genuinely nonlinear or linearly degenerate, and which admits a strictly convex
entropy η(·). Provide an asymptotic estimate on the universal convergence rate ϱ(·) in (7.6),
as ε → 0.
Based on the earlier estimates (7.1)-(7.2), in the genuinely nonlinear case one may conjecture
that ϱ(ε) ≈ ε1/2 | ln ε|.
The key feature of the bound (7.6) is that it holds for any ε-approximate solution satisfying
(ALε )-(Pε ), regardless of the method used to construct the approximation. All the algorithms
considered in Section 3 generate ε-approximate solutions, in the sense of Definition 7.1. See
Section 6 in [18] for details.

One can speculate whether a similar universal convergence rate can be valid for general n × n
systems, not necessarily endowed with a strictly convex entropy. For these systems, semigroup
trajectories are characterized by the Liu admissibility condition, as in Definition 2.4. To reach
our goal, we should replace the ε-approximate entropy condition (7.5) with some sort of ε-
approximate Liu condition. This leads to

Open Problem #4. Introduce a definition of “ε-approximate Liu admissible solution”, valid
for general n × n hyperbolic systems, possibly not endowed with a strictly convex entropy.
A bit more precisely, what is needed here is a suitable definition such that the following
properties will be satisfied.

• Approximate solutions with small total variation constructed by the various methods
described in Section 3 should all satisfy the ε-approximate Liu condition, with ε → 0 as
the step size in the approximation (or the viscosity coefficient) approaches zero.
• Given a convergent sequence of approximations un → u, if each un is an εn -approximate
Liu admissible solution with εn → 0, then the limit solution u should be Liu-admissible
in the original sense.

33
8 Solutions with unbounded variation

As remarked in Section 4, for solutions with large initial data it is a hard open question to
decide whether the total variation remains bounded for all times. It is thus natural to consider
solutions in the larger space L∞ (R; Rn ), possibly with unbounded variation.
For 2×2 systems, existence of weak solutions with L∞ data was proved in a fundamental paper
by DiPerna [42], based on compensated compactness. See also [41, 61, 64] for a comprehensive
account of this approach. Existence of L∞ solutions remains a largely open problem for general
n × n systems.
Unfortunately, compensated compactness works as a “black box”. It provides an abstract
result on the existence of solutions, but it does not yield information about uniqueness, con-
tinuous dependence, or the qualitative structure of these solutions. Some of the few results
on the regularity of L∞ solutions can be found in [35, 46].
In this direction, it would be of interest to construct a continuous semigroup of admissible
solutions, defined on a domain larger than BV .

Open Problem #5. Given an n × n hyperbolic system of conservation laws, extend the
e ⊆ L∞ (R; Rn ), also containing
semigroup of vanishing viscosity solutions to a larger domain D
functions with unbounded variation.

A continuous semigroup of solutions defined on the entire space L∞ (R; Rn ) was constructed
in [24] for some Temple class systems, and more recently in [26] for 2 × 2 systems in triangular
form. But apart from a few special cases the problem is wide open.
As suggested in [2], in general it may not be possible to construct a continuous semigroup
defined on the entire space L∞ . Instead, one could consider some intermediate domain D e⊂
∞
L , borrowing ideas from the theory of intermediate spaces used in the analysis of parabolic
equations [47, 62]. Of course, we do not expect that the extended semigroup will be Lipschitz
continuous. Its modulus of continuity will strongly depend on the regularity properties of
functions u ∈ D.
e

In addition to compensated compactness, another approach is worth mentioning here. In their

classical memoir [45], Glimm and Lax consider the Cauchy problem for a genuinely nonlinear
2 × 2 system. Assuming that the initial data ū has sufficiently small L∞ norm, they prove
that a global weak solution exists, globally in time. Indeed, the total variation (which initially
may well be infinite) becomes locally finite at every time t > 0. See also [11] for a shorter
proof, based on front tracking approximations. The uniqueness and continuous dependence of
these solutions still remains an open problem.

In the opposite direction, it would also be of interest to find examples of Cauchy problems
admitting multiple solutions. In [32] a 3 × 3 strictly hyperbolic system has been constructed,
together with bounded, measurable initial data, leading to an infinite number of solutions.
However, this example does not have physical relevance because the system does not admit
convex entropies. We thus conclude with

34
Open Problem #6. Construct an example of an n × n strictly hyperbolic system, endowed
with a strictly convex entropy, together with initial data ū ∈ L∞ (R; Rn ), such that the Cauchy
problem admits two distinct entropy admissible solutions.

Acknowledgement. This research was partially supported by NSF with grant DMS-2306926,
“Regularity and approximation of solutions to conservation laws”.

References

[1] L. Ambrosio, N. Fusco, and D. Pallara, Functions of Bounded Variation and Free Dis-
continuity Problems. Clarendon Press, Oxford, 2000.

[2] F. Ancona, S. Bianchini, A. Bressan, R. M. Colombo, and K. T. Nguyen, Examples

and conjectures on the regularity of solutions to balance laws. Quarterly Appl. Math. 81
(2023), 433–454.

[3] F. Ancona and A. Marson, A wave front tracking algorithm for N × N non genuinely
nonlinear conservation laws. J. Differential Equations 177 (2001), 454–493.

[4] F. Ancona and A. Marson, Sharp convergence rate of the Glimm scheme for general
nonlinear hyperbolic systems. Comm. Math. Phys. 302 (2011), 581–630.

[5] P. Baiti and H. K. Jenssen, On the front-tracking algorithm J. Math. Anal. Appl. 217
(1998), 395–404.

[6] P. Baiti, A. Bressan, and H. K. Jenssen, BV instability of the Godunov scheme, Comm.
Pure Appl. Math. 59 (2006), 1604–1638.

[7] S. Bianchini, On the Riemann problem for non-conservative hyperbolic systems, Arch.
Rational Mech. Anal. 166 (2003), 1-26.

[8] S. Bianchini, BV solutions of the semidiscrete upwind scheme. Arch. Rational Mech.
Anal. 167 (2003), 1–81.

[9] S. Bianchini, Hyperbolic limit of the Jin-Xin relaxation model. Comm. Pure Appl. Math.
59 (2006), 688–753.

[10] S. Bianchini and A. Bressan, Vanishing viscosity solutions of nonlinear hyperbolic sys-
tems, Annals of Mathematics 161 (2005), 223–342.

[11] S. Bianchini, R. M. Colombo, and F. Monti, 2 × 2 systems of conservation laws with L∞

data. J. Differential Equations 249 (2010), 3466–3488.

[12] S. Bianchini and S. Modena, Quadratic interaction functional for general systems of
conservation laws. Comm. Math. Phys. 338 (2015), 1075–1152.

[13] A. Bressan, Global solutions to systems of conservation laws by wave-front tracking, J.

Math. Anal. Appl. 170 (1992), 414–432.

35
[14] A. Bressan, BV solutions to systems of conservation laws by vanishing viscosity. In:
Hyperbolic systems of balance laws, P. Marcati Ed., Lecture Notes in Math. 1911 Springer,
Berlin, (2007), pp. 1–78.
[15] A. Bressan, The unique limit of the Glimm scheme, Arch. Rational Mech. Anal. 130
(1995), 205–230.
[16] A. Bressan, Hyperbolic systems of conservation laws. The one dimensional Cauchy prob-
lem. Oxford University Press, 2000.
[17] A. Bressan, G. Chen, and Q. Zhang, On finite time BV blow-up for the p-system, Comm.
Partial Diff. Equat. 43 (2018), 1242–1280.
[18] A. Bressan, M. T. Chiri and W. Shen, A posteriori error estimates for numerical solutions
to hyperbolic conservation laws. Arch. Rational Mech. Anal. 241 (2021), 357–402.
[19] A. Bressan and R. M. Colombo, The semigroup generated by 2 × 2 conservation laws,
Arch. Rat. Mech. Anal. 113 (1995), 1–75.
[20] A. Bressan and R. M. Colombo, Unique solutions of 2 × 2 conservation laws with large
data, Indiana Univ. Math. J. 44 (1995), 677–725.
[21] A. Bressan, G. Crasta, and B. Piccoli, Well posedness of the Cauchy problem for n × n
systems of conservation laws, Amer. Math. Soc. Memoir 694 (2000).
[22] A. Bressan and C. De Lellis, A remark on the uniqueness of solutions to hyperbolic
conservation laws. Arch. Rational Mech. Anal., to appear.
[23] A. Bressan and P. Goatin, Oleinik type estimates and uniqueness for n × n conservation
laws, J. Differential Equations 156 (1999), 26–49.
[24] A. Bressan and P. Goatin, Stability of L∞ solutions of Temple class systems, Differential
& Integral Equat. 13 (2000), 1503–1528.
[25] A. Bressan and G. Guerra, Unique solutions to hyperbolic conservation laws with a strictly
convex entropy. Preprint 2023, available on arXiv:2305.10737.
[26] A. Bressan, G. Guerra, and W. Shen, Vanishing viscosity solutions for conservation laws
with regulated flux, J. Differential Equations 299 (2019), 312–351.
[27] A. Bressan, F. Huang, Y. Wang, and T. Yang, On the convergence rate of vanishing
viscosity approximations for nonlinear hyperbolic systems, SIAM J. Math. Analysis 44
(2012), 3537–3563.
[28] A. Bressan and P. LeFloch, Uniqueness of weak solutions to systems of conservation laws,
Arch. Rational Mech. Anal. 140 (1997), 301–317.
[29] A. Bressan and M. Lewicka, A uniqueness condition for hyperbolic systems of conservation
laws, Discr. Cont. Dyn. Syst. 6 (2000), 673–682.
[30] A. Bressan, T. P. Liu and T. Yang, L1 stability estimates for n × n conservation laws,
Arch. Rational Mech. Anal. 149 (1999), 1–22.
[31] A. Bressan and A. Marson, Error bounds for a deterministic version of the Glimm scheme.
Arch. Rational Mech. Anal. 142 (1998), 155–176.

36
[32] A. Bressan and W. Shen, Uniqueness for discontinuous O.D.E. and conservation laws,
Nonlinear Analysis, T.M.A. 34 (1998), 637–652.

[33] A. Bressan and T. Yang, On the rate of convergence of vanishing viscosity approximations,
Comm. Pure Appl. Math 57 (2004), 1075–1109.

[34] G. Chen, S. Krupa, and A. Vasseur, Uniqueness and weak-BV stability for 2x2 conserva-
tion laws, Arch. Rational Mech. Anal. 246 (2022), 299–332.

[35] G. Q. Chen and M. Torres, On the structure of solutions of nonlinear hyperbolic systems
of conservation laws. Comm. Pure Appl. Anal. 10 (2011), 1011–1036.

[36] M. T. Chiri and M. Zhang, On backward Euler approximations for systems of conservation
laws. Nonlin. Diff. Equat. Appl., submitted.

[37] M. G. Crandall, The semigroup approach to first order quasilinear equations in several
space variables. Israel J. Math. 12 (1972), 108–132.

[38] C. Dafermos, Polygonal approximations of solutions of the initial value problem for a
conservation law. J. Math. Anal. Appl. 38 (1972), 33–41.

[39] C. Dafermos, Generalized characteristics and the structure of solutions of hyperbolic

conservation laws. Indiana Univ. Math. J. 26 (1977), no. 6, 1097–1119.

[40] C. Dafermos, The second law of thermodynamics and stability. Arch. Rational Mech.
Anal. 70 (1979), 167–179.

[41] C. Dafermos Hyperbolic conservation laws in continuum physics. 4-th Edition, Springer,
2016.

[42] R. J. DiPerna, Convergence of approximate solutions to conservation laws. Arch. Rational

Mech. Anal. 82 (1983), 27–70.

[43] L. Euler, Principes généraux du mouvement des fluides. Mém. Acad. Sci. Berlin 11 (1755),
274–315.

[44] J. Glimm, Solutions in the large for nonlinear hyperbolic systems of equations. Comm.
Pure Appl. Math. 18 (1965), 697–715.

[45] J. Glimm and P. Lax, Decay of solutions of systems of nonlinear hyperbolic conservation
laws. Mem. American Math. Soc. 101, Providence, R.I. 1970.

[46] W. Golding, Unconditional regularity and trace results for the isentropic Euler equations
with γ = 3. arXiv:2207.05821.

[47] D. Henry, Geometric theory of semilinear parabolic equations. Lecture Notes in Mathe-
matics 840, Springer, Berlin, 1981.

[48] H. Holden and N.H. Risebro, Front tracking for hyperbolic conservation laws. Springer-
Verlag, New York, 2002.

[49] H. K. Jenssen, Blowup for systems of conservation laws, SIAM J. Math. Anal. 31 (2000),
894–908.

37
[50] S. Kruzhkov, First-order quasilinear equations with several space variables, Mat. Sb. 123
(1970), 228–255. English transl. in Math. USSR Sb. 10 (1970), 217–273.

[51] P. Lax, Hyperbolic systems of conservation laws II. Comm. Pure Appl. Math. 10 (1957),
537–566.

[52] R. J. LeVeque, Numerical Methods for Conservation Laws. Birkhäuser, Basel, 1990.

[53] M. Lewicka, Stability conditions for patterns of non-interacting large shock waves, SIAM
J. Math. Anal. 32 (2001), 1094–1116.

[54] M. Lewicka, The well posedness for hyperbolic systems of conservation laws with large
BV data, Arch. Rational Mech. Anal. 173 (2004), 415–445.

[55] T. P. Liu, The deterministic version of the Glimm scheme, Comm. Math. Phys. 57 (1975),
135-148.

[56] T. P. Liu, The entropy condition and the admissibility of shocks. J. Math. Anal. Appl.
53 (1976), 78–88.

[57] T. P. Liu, Linear and nonlinear large-time behavior of solutions of general systems of
hyperbolic conservation laws. Comm. Pure Appl. Math. 30 (1977), 767–796.

[58] T. P. Liu, Admissible solutions of hyperbolic conservation laws Mem. Amer. Math. Soc.
30 (1981), no. 240.

[59] T. P. Liu, Shock Waves. American Mathematical Society, Providence, RI, 2021.

[60] T. P. Liu and T. Yang, L1 stability for 2 × 2 systems of hyperbolic conservation laws. J.
Amer. Math. Soc. 12 (1999), 729–774.

[61] Y. Lu, Hyperbolic conservation laws and the compensated compactness method. Chapman
& Hall/CRC, Boca Raton, FL, 2003.

[62] A Lunardi, Analytic esmigroups and optimal regularity in parabolic problems, Birkhäuser,
Basel, 1995.

[63] B. L. Rozdestvenskii and N. N. Yanenko, Systems of quasilinear equations and their

applications to gas dynamics. Transl. Math. Monogr. 55, American Mathematical Society,
Providence, RI, 1983.

[64] D. Serre, Systems of Conservation Laws 2. Cambridge University Press, Cambridge, 2000.

[65] J. Smoller, Shock Waves and Reaction-Diffusion Equations. Springer-Verlag, New York-
Berlin, 1983.

38