JOURNAL OF DIFFERENTIAL EQUATIONS 65, 250-268 (1986)

Analysis of a Singular Hyperbolic System

of Conservation Laws
ELI L. ISAACSON*
Department qf Mathematics, University of Wyoming,
Laramie, Wyoming 82071

J. BLAKETEMPLE*,+
Mathematics Research Center, Madison, Wisconsin

Received March 8. 1985

We solve the Riemann and Cauchy problems globally for a singular system of n
hyperbolic conservation laws. The system, which arises in the study of oil reservoir
simulation, has only two wave speeds, and these coincide on a surface of codimen-
sion one in state space. The analysis uses the random choice method of Glimm
(Comm. Pure Appl. Math. 18 (1965) 697-715). % 1986 Academic press, IIIC.

1. INTRODUCTION

In the study of enhanced oil recovery, the scalar Buckley-Leverett

equation

s, + (%).x = 03 (1.1)
models the water flooding of an oil reservoir [6]. Water flooding involves
the injection of water, which is immiscible with oil, into certain wells of the
reservoir to force oil out at others. In this case, s = s(x, t) is the saturation
of water (i.e., the volume fraction of water in the total fluid, 0 < s d 1), and
g= g(s) is the particle velocity of the water.
Consequently, since the total volumetric flow rate is fixed [2, 6, 71, the
fraction of the flow rate associated with the water is sg(s) =f(s). In this
problem, the fractional flow curve f(.) is nonconvex: f increases from 0 to

*Work supported by FINEP Grant 4.3.82.0719.00and by CNPq while the authors were
visiting the Pontificia Universidade Catolica do Rio de Janeiro, RJ, Brazil.
+Sponsored by the United States Army under Contract DAAG29-80-C-0041.
250
0022-0396186$3.00
Copyright 0 1986 by Academic Press, Inc
All rights of reproduction in any form reserved
 SINGULAR HYPERBOLIC SYSTEM 251

I s
FIG. 1. A typical S-shaped curve corresponding to f(.) in system (1.1) and f(., c) in
systems (1.2), (1.3).

1, f’(0) = f’( 1) = 0, and f has one inflection point. We say f is S-shaped.

(See Fig. 1.)
The 2 x 2 system of nonlinear conservation laws

(1.2)

models the polymer flooding of an oil reservoir [2, 71. (See also [3].) In a
polymer flood, water thickened with polymer is injected into the reservoir.
A polymer is a solute of water which inhibits its flow, and thus improves
the oil displacement. In this case, s = s(x, t) is the saturation of the aqueous
phase (the solution of polymer and water), c = c(x, t) is the concentration
of polymer in the water, g = g(s, SC)is the particle velocity of the aqueous
phase, and f(s, c) = sg(s, SC)is the fraction of the flow rate associated with
the equeous phase. For fixed c, the fractional flow curvef(., c) has the same
qualitative properties as in the Buckley-Leverett problem.
In the present paper, we study a natural generalization of Eqs. (1.1) and
(1.2) given by the (n + 1) x (n + 1) system

u,+ (gUL=O, (1.3)

where U= (s, SC) is the vector of conserved quantities, c = (c,,..., c,),

g = g(s, SC),and {(s, c): 0 <s < 1, 0 < ci < 1) is the state space. Again, the
function f(s, c) = sg(s, SC) is important for our analysis. Our only
assumption for system (1.3) is that f is smooth and the curves f(*, c) have
the S-shape of the Buckley-Leverett problem for each fixed c. Our main
results are:
252 ISAACSON AND TEMPLE

THEOREM A. The Riemann problem for ( 1.3) has a unique solution ,ftir
arbitrary initial data (sL, c,) and (sR, cR). Moreover, for every positive t, the
solution (s(x, t), c(x, t)) is in L&, with respect to x and dependscontinuously
on t.

THEOREM B. There is a singular transformation

Y-c(s, c) + WE (2, c)
such that, if the initial data (s,,(x), Q(X)) satisfies TV(z,(.), co(.)) < co, then
there exists a global weak solution 17(x, t) of the Cauchy problem for (1.3).
Moreover, U(x, t) is in L;O, with respect to x and dependscontinuously on t.
To prove Theorem A we construct the solutions explicitly, and to prove
Theorem B we use the random choice method (RCM) of Glimm [ 11.
System (1.3) can be viewed as a model for the flooding of a reservoir by
water containing n additives. Our present assumptions relax the
requirement that the additives inhibit the water flow as in [2, 71. This
system is of interest in the study of conservation laws because of the new
phenomena which arise. The system is nowhere strictly hyperbolic and, in
fact, has at most two distinct eigenvalues which coincide on a surface of
codimension one.
There are two new features in the present analysis. First, we analyze the
Riemann problem entirely in the 2-dimensional sf-plane, and not the
(n + 1)-dimensional state space. This enables us to handle the higher
dimensional problem. It also makes unnecessary the assumption that f(s, .)
be monotone for each fixed s. Second, unbounded variation occurs in
approximate solutions of the RCM for systems (1.2), (1.3). As in [7], we
prove convergence by bounding the variation in s as measured under a
singular transformation. Here we must incorporate a supplemental bound
on the total variation in c because extra degrees of freedom occur in (1.3)
that do not occur in (1.2). Also, as in [7], it is necessary to randomize in
space as well as time to handle the singularity.

2. THE RIEMANN PROBLEM

2.1. Preliminaries
In this section we consider the Riemann problem; i.e., the initial value
problem

u,+ (gU).x
=0,
(2.1)
U(x,0)= U,(x),
 SINGULAR HYPERBOLIC SYSTEM 253

lope = g = f = constant

FIG. 2. In system (1.2) and (1.3) the surface g = g, e const is given by all points (s, c) such
that (s, f(s, c)) lies on the line {(s, f): f= g,s}.

where

U,(x)= uuL’ :;;

i RY
Here, U = (s, b), b = SC,0 <s < 1, 0 < bi< s, and f= sg satisfies the con-
stitutive assumptions of system (1.3). (See Figs. 1, 2.) Here, U is the vector
of conserved quantities. Let
U=@(V), v= (s, c).

We let “state space” refer either to the space of allowed U’s or the
corresponding values of V. The map CDis contractive, one-to-one, and
regular except at s = 0. We solve the Riemann problem for arbitrary values
of V, and VR. We note that when s = 0, the solution is not unique in the
U-variables, but specifying a value of c determines a unique solution in the
V-variables.
2.2. Notation
First, because our analysis to follow uses V-space instead of U-space, we
take g to be a function of V = (s, c); i.e., g(V) z g 0@(V). The set of points
where f, = f/s is crucial in our analysis. At these transition points the
system (1.3) fails to be strictly hyperbolic. Let s=(c) denote the unique
positive value of s for which f(s, c)/s = fs(s, c). The existence and uni-
queness of sT for each c follow from the fact that f(., c) is S-shaped (see
Fig. 3). The transition surface Y is the subset of state space given by
5 = {(s, c): s = ST(C)}.
254 ISAACSON AND TEMPLE

FIG. 3. Construction of sT(c) and gT(c) from the graph off(., c),

We also define the region

Y= {(s,c), O<s<ss,(c)}
to the left of f and the region
a= {(s, c): s,(c)<.36 l}

to the right of Y. For each c, the unique state on the transition surface is

VT(C)= MC), cl.

For each c this gives a unique g-value given by (see Fig. 3)

gdc) = g( VT(C)).
2.3. S-waves and C-waves
By differentiating (1.3), we obtain the system
U,+A(U) U,=O,
where

A(U)=A(s,b)=gZ+
gb = ( gb, >...) gb.1.

The following lemma is a consequence of the fact that A is a rank one per-
turbation of a scalar matrix.
 SINGULAR HYPERBOLIC SYSTEM 255

LEMMA 2.1. The eigenvalues of A are g = f/s and g + sg, + b. g, = fs.

Off the transition surface Jr, the eigenvalue g has multiplicity n, and f, is
simple. The eigenvectors corresponding to the eigenvalue g span the tangent
space to the surface g = const. The eigenvector corresponding to the eigen-
value fs is u = (s, b). On F there are only n independent eigenvectors.
From this lemma we obtain

THEOREM 2.2. The integral manifolds of the eigenvectors corresponding

to the eigenvalue g are the connected components of the surfaces g = const.
The integral curves of the eigenvector corresponding to the eigenvalue f, are
the lines c = const.
The Hugoniot locus [4] of a state U, consists of those states UR for
which the discontinuous function

UL x < at,
U(x, t) = u
i R x > at,

is a weak solution of (2.1). An equivalent statement is that U = UR satisfy

the Rankine-Hugoniot jump condition:

g(U)U-g(Ud uL=du- UL)

for some number 0. A straightforward calculation yields

THEOREM 2.3. The Hugoniot locus of a state UL is the set

{U: g(u)=g(u,)}u{u:c=c,}.

From Theorems (2.2) and (2.3), it is clear that there are two families of
elementary waves: C-waves (concentration waves) consist of contact dis-
continuities joining two states with equal g-values; S-waves (saturation
waves) solve the scalar Riemann problem that is obtained from (2.1) when
c is constant. To insure the uniqueness of solutions lo the Riemann
problem, we assume that the elementary waves of each family satisfy an
entropy condition. We require that the S-waves satisfy the standard
entropy condition for a scalar conservation law. Our entropy condition for
the C-waves is that C-waves can join two states only if both states are in 9
or both are in W. This is equivalent to the generalization of the Lax
entropy conditions given by Keytitz and Kranzer [3].
we let vL +’ VR (resp. VI -bc VR) indicate that VL on the left can be
connected to VR on the right by an S-wave (resp. C-wave). Two con-
secutive arrows (e.g., VL +S VI +’ VR) indicate that the waves can be
256 ISAACSON AND TEMPLE

slope = gT(CL)
I
slope = gT(cL) f
/

a I s b I s

C I s
FIG. 4. (a) For each c, (s, c) E WI iff (s, f(s, c)) lies on the shaded segment off(., c)). (That
is, (s, c) E Yeand (s, f(s, c)) lies below the line with slope gT(cL).) (b) For each c, (s, c) E S2
iff g,(c) < g,(c,) and (s,f(s, c)) lies on the shade segment of f(., c). (c) For each c,
(s, c) E B3 iB g*(c) 2 g,(c,) and (s, f(s, c)) lies on the shaded segment of f(., c).

composed; i.e., the wave speedsincrease from left to right, and we say that
the speedsare compatible.

2.4. Solution of the Riemann Problem

We present the solution in six cases:

(I) VL E 9 and VR in one of (see Fig. 4)
(i) W1=.%n{Vg(V)~gT(cL)}
(ii) % = 9 n { V: g=(c) 6 &cd},
(iii) 4e,= (9, u Se,1’ = (if: tzT(C)2 gT(CL)l\%
 SINGULAR HYPERBOLIC SYSTEM 257

(II) V, E 9 and VR in one of

(i) %=Wn{V:g(V)<g(V,)),
(ii) z~9f-1 {V: gT(c)<g(VL)},
(iii) LZ3E (SC;u L&)‘= { V: g=(c) > g( V,)}\Yi.
These cover all possibilities. As the regions have common boundaries, one
must verify that the solutions presented for initial data in the overlap
actually agree. In fact, two such solutions V(x, t) agree in L’ for each t.
Finally, the observations referred to below are listed in Section 2.5 and
follow from the assumption that f(., c) is S-shaped. That solutions depend
continuously on the data in L,‘,, is a consequence of the continuity in each
region, together with the agreement of solutions at the boundaries of
regions:
(1) vL= (SL, C,)EB.
Case (i). V, E ~8~. Solution: V, +’ Vr --)’ V,. The intermediate state
VI is given by
VI = h, CL)
where si is chosen so that g( Vi) = g( VR), and I/, E W. State I/, exists by
observation (01). The wave speeds are compatible by observation (03)
(given below).
Case (ii). V, E B*. Solution: L’, +’ Vi --+= Vu --ts VVR. The inter-
mediate states I/, and V,, are given by

where si is chosen so that g( V,) = g,(c,) and su = sT(cR). State Vi exists by

observation (01). The wave speeds are compatible by observations (03)
and (04).
Case (iii). VRE $I$. Solution: V, +’ L’, dc T/i, +’ V,. The inter-
mediate states V, and Vii are given by

where si = sT(cL) and su is chosen so that g( I’,,) = gT(cL). State Vi, exists
by observation (02). The wave speeds are compatible by observations
(03) and (05).
(II) V, = (sr, c,) E 9.
258 ISAACSON AND TEMPLE

Case (i). VRE 3,. Solution: V, -+’ Vi -+’ V,. The intermediate state
VI is given by
VI = (SIYcd
where s, is chosen so that g( Vi) = g( V,) and Vi E 9% State Vi exists by
observation (01). The wave speeds are compatible by observation (06).
Case (ii). V, E &. Solution: V, +’ V, + c Vu -+’ V,. The inter-
mediate states V, and Vu are given by

VII = (%I> CR 12

where sI is chosen so that g( Vi) = g,(c,) and su = sr(cR). State Vi exists by

observation (01). The wave speeds are compatible by observations (06)
and (04).
Case (iii). V, E Z3. Solution: V, --t c Vi -+’ I/, The intermediate state
VI is given by
VI = @I, cd
where sI is chosen so that g( Vi) = g( Vr). State Vi exists by observation
(02). The wave speedsare compatible by observation (05).
2.5. Observations in the sf-plane
In the observations below, we let Z(g) be the line in the sf-plane with
slope g passing through the origin. We say I( go) intersects the curve
f=f(., cO)at the point V,, = (so, cO) if g( V,,) = g,. (In the sf-plane, the two
curves actually intersect at (so, f( V,)). E.g., see Fig. 2.) Also, the wave
speeds in observations (03) through (06) refer to those in the scalar
Buckley-Leverett equation:
(0 1) If I( go) intersects f = f(., cO) at a point in 9, and if I( g, ) inter-
sects f=f(., ci) with g, 2 go, then I( go) also intersects f =f(., cl) at a
point in 9?.
(02) If 0 < go 6 gPT(c),then I( go) intersects f =f(., c) in 9.
(03) The wave speedsin V, +s V2 are at most g( V,) if V,, Vz E @.
(04) The wave speeds in V, +s V, are at least g( V,) if V, E Y and
V,EZ.
(05) The wave speedsin V, +’ V, are at least g( VI) if V, E 9 and
either (a) V, E 9 or (b) VzE a with g( Vz) 3 g( VI).
(06) The wave speeds in VI +’ V, are at most g( V2) if V, E 9,
V,E~, and g(V,)2g(V2).
 SINGULAR HYPERBOLIC SYSTEM 259

3. SOLUTION OF THE CAUCHY PROBLEM

3.1. Preliminaries
In this section we use the random choice method to solve the Cauchy
problem (2.1) with general initial data. First define the variable z by

z=z(V)= gdc)- &A

v if VE8,
(3.1)
id v - g-r(c) if VET.

Let E V + W be defined by

w= (z(V),c). (3.2)
The transformation ‘Y is one-to-one and regular except at the transition
surface Y. We define the strength of S- and C-waves, as well as the
strengths of waves in the variable z, as follows. Let S, C denote arbitrary S-
and C-waves, respectively. Let V, and V, denote the left and right states of
the wave, respectively. Then define
lSI=~v,-VR~~~SL-SRI+~CL-CR~
= ISL-SRI + lclL-clRl + ..’ + ICnL-CnRI,
ICI = IV,- VRI, (3.3)
ICI,= IcL-%I,
ISI,= I%-ZRI,
if (3.4) holds,
if (3.4) does not hold,

where the condition is

VL, V,EZ and gr( VL) 2 gl-( VRL

or (3.4)
VL, v,Ea and gT( vd G gT( VR 1.

The general Riemann problem has a solution of the form SLCSR; i.e., two
S-waves separated by a C-wave with the convention that SR= 0 if C= 0.
We let y = SLCSR and set

IYI= lSLl + ICI + ISRI,

Irl,= ISLIz+ ICI,+ ISRI,.

The following lemma is crucial in the convergence proof of Section 4.

260 ISAACSON AND TEMPLE

LEMMA 3.1. For ever-y E > 0 there exists a constant M(E) depending only
on E such that
if’ IsI >E, then Is/ d M(E) IsI;, (3.5)
if ICI>&, then ICI GM(E){lCI;+ ICI,}. (3.6)

Proof: Suppose ISI > E. Then Isy- sT(cy)l > s/2 for q = L or R. The map
(z, c) + (s, c) is a one-to-one regular map off the transition surface Y.
Therefore the Jacobian is bounded off any neighborhood of Y. Conse-
quently, there is a constant M,(E) so that

b-G dMI(&)lZL-ZR/. (3.7)

To prove (3.6), we use the fact that sT is a differentiable function of c.
Thus,
Is-dcd - sl-(CR)1G Y 1%- %I
for some number y. (We may assume y > 1.)
Now suppose ICl > E.If ICl, > &/2y, then we can take M2(s) = 2y(n + l)/~
since ICI d n + 1. On the other hand, if ICI, d .5/2y,then

(1 -WY)&< ICI -ICI,= b-SRI

G 1%- S-r(CL)I+ IS-JCL)- ST(%)I + I.%(%) -SRI
d ISL-ST(CLh +YlCIc+ IsR-sT(%)I
d ISL- sr(cdI + 42 + I.%- +(%)I.
Rearranging gives Isy- sT.(cy)l> (y - 1)e/4y for q = L or R. Therefore, we
can use (3.7) again to obtain

ICI = b-SRI + ICI,

GM,((Y- 1)@Y)lzL-z,I + ICI,
GM,((Y- W2Y){l~Iz+ ICI,>
provided M, 3 1.
3.2. The Random Choice Method
We first define the approximate solutions generated by the RCM. Let h,
k be mesh lengths in x, t, respectively. Let k= k(h) be chosen so that the
Courant-Friedrichs-Lewy condition holds:

(3.8)
 SINGULARHYPERBOLIC
SYSTEM 261

For every integer i and nonnegative integer j, define the mesh points
xi = ih, tj = jk in the xt-plane. Let a denote a sequence
a = {a,>, (3.9)
where 0 < aV< 1; i.e., an element of the measure space

A= n F4nJ (3.10)
iCZZ
jeZ+

with Lebesgue measure denoted by m. For each h, a, and initial data V,(x),
we define the approximate solution
Vh(x, t) = VJx, t; a)

generated by the RCM, by induction on j as follows: Let

vh(x, 0) = vo(xj + a,h) for xi<x<xi+r, (3.11)

and for 0 6 t < t,, let Vh(x, t) be the solution of (3.1) obtained by solving
the Riemann problems posed in (3.11). Here I/h(x, t) is well defined for
t < t, by (3.8).
Now assume VJx, t) has been defined for times t < tj. Then define V,, for
t = tj by setting

Vh(X, tj) = Vh(Xi + a;jh, tj) for xi<x6xi+r, (3.12)

and define V, for tj < t < tj+ r by solving the corresponding Riemann
problems with “initial data” VJx, tj). Again, Vh(x, t) is defined for all time
by (3.8). Let

Wh(x, t) = Wh(x, t; a) = Y( Vh(x, t; a)), (3.13)

U,(x, t) - Uh(x, t; a) = @(VJx, t; a)), (3.14)
W,(x) = w Vcdx)), (3.15)
U,(x) = @(V,(x))- (3.16)

Let yii = SkC,Sf be the waves appearing in the solution of the Riemann
problem centered at (xi, t,). Note that all waves have nonnegative speeds.
Define

W)=C {Iriilz+ IC,l,). (3.17)

262 ISAACSON AND TEMPLE

Notation. In the following lemma and throughout the remainder of the

paper, we let A4 denote a generic constant depending only on the functionf
and any explicit arguments.

LEMMA 3.2. Zf TV{ W,(.)} <MO< co, then

w+ l)GF(j)<F(O)<MM,, (3.18)
TV{ w/i., t,} Q MM,. (3.19)

Proof: Inequality (3.18) holds because both F=(j) E ‘& lC,I, and
r;Z(j) = C 1~~1,are nonincreasing functions of j. This can be proven by
induction on j. That F, is nonincreasing follows easily from the fact that the
change in c in any Riemann problem is due to a single jump; i.e., the jump
in c across the C-wave. That Fz is nonincreasing follows from the fact that
the sum of the wave strengths in z for any sequence of elementary waves
connecting a given left and right state is minimized by the waves in the
Riemann problem solution (cf. [7]). Inequality (3.19) follows immediately
from (3.18) together with the definition of W = YJ(V), since for tj < t < t,, ],
TV{ W,(., t,} <F(j) d F(0) < MM,.

Let the index set X(E, a) be defined by

X=X(&,a)= {(i,j): IS;1 <sand ISi1 <sand lC,I GE}. (3.20)

LEMMA 3.3. Zf (i, j)$ X(E, a), then

IYijlam){lY~lz+ K,lJ. (3.21)

Proof If (i, j)$X(s, a), then either ISkI > E or IS;1 >E or lC,l >E. If
lC,I > E, then by (3.6)

Since IS&l < 1 for q = L or R, we obtain

lyiil= Is;1 + Is;1 + IC,I

d2+J4(4{IGlz+ K,lc)
d(2/&+1)M(&){IClil;+ICiilc)
-w{IYijlz+ lCylc)
 SINGULAR HYPERBOLIC SYSTEM 263

where M(E) is generic. The case IS;1 > E, q = L or R, follows from (3.5) in
the same manner.
3.3. Convergence of the Approximate Solution
We prove the following result.

THEOREM 3.4. For every a E A and sequence of mesh lengths h + 0, a sub-

sequence U, of approximate solutions converges uniform/y in L[,, to a
function U(x, t) = U(x, t; a); i.e., U,(*, t) converges to U(., t) in
L’[ -M, M], un$ormly on 0 < t < T for any positive M, T. Moreover,
ll4.7 f)llL:,, is a continuous function of t.
We use the following result of Oleinik [ 1] :

LEMMA 3.5. Let h denote a positive sequence tending to zero, and let
W,,: (x, t) + R” be a sequence of functions satisfying

TV{ W,(., t)} < M, < 00 for all h, t, (3.23)

II W,(., t) - W,(,, t)ll LI d M( It - zl + h}. (3.24)

Then a subsequence of the functions W,, converges uniformly in L:,, to a

function W(., t). Moreover, W(., t) satisfies

IIW(., t)- Wt., z)lIp GMlt-4 (3.25)

Proof of Theorem 3.4. The approximate solutions WJx, t) = W,,(x, t; a)

satisfy the conditions of Lemma 3.5 for all a E A. The first condition follows
from Lemma 3.2 with M, = MM,. To see that the second condition is
satisfied, let tjl and t,2 be the times with
z - k d tj, d t < t < tjz < t + k.

Then for x E (xi, xi+ 1],

I w/k t) - w/Ax, 7116 5 { lYijlz+ IKjlJ

j=jl

Therefore,

s m I W,,(x,t) - W,Sx,z)l dx
-cc
264 ISAACSON AND TEMPLE

Thus (3.24) holds for the approximate solutions.

We now show that V, also converges to V uniformly in L,‘,,,. Note that
Yy-’ is continuous with compact domain; so it is uniformly continuous.
Suppose T> 0. For 0 < t d T, let
E= E(M, h, t, p)
= {XE C-M, Ml: IWJX, t)- W(x, t)l <p}
Then

But SE< 2M6(~), which is uniformly small in [0, T]. (Here, 6 is the
modulus of continuity of Y.) Also

sC- M,Ml\E
Ivh- VI ~2(~+1)~{C-w~I\q

IW,- Wldx. (3.26)

The last member of (3.26) is uniformly small for h small because W, -+ W

uniformly in L:,,. Consequently the convergence of V, to V is uniform in
L:,,. Since @ is also uniformly continuous, Uh also converges to U
uniformly in L:,,. The uniform continuity of Y and @ together with (3.25)
imply that IIV., t)ll L;ocis a continuous function of time. This completes the
proof of Theorem (3.4).
3.4. The Weak Solution
We say that U(x, t) is a weak solution of (1.3) if it satisfies

‘Q(U,f#)=jj -mJ<x< +m
lJ&+gU&dxdt+jl‘
-cc
U(x,O)$(x,O)dx=O
I>0
(3.27)
 SINGULAR HYPERBOLIC SYSTEM 265

for every smooth C$with compact support. Let

(3.28)

THEOREM 3.6. Suppose U,(x) is initial data satisfying

U,(x) = @(V,(x)) (3.29)

and

TV{W,(.)}<M,<oO. (3.30)

Then there exists a set N c A, m(N) = 0, such that if a E A\N, then the
function U(x, t; a) of Theorem (3.5) satisfies (3.27) for all 4 of compact sup-
port.
Proof of Theorem 3.6. Let h +O be a sequence of mesh lengths such
that U,(x, t; a) + U(x, t; a) in the senseof Theorem 3.4. We will show that
there exists a set N of measure zero in A such that if a E A\N, then

D(h, a, 4) -, 0

for all smooth 4 of compact support. This completes the proof because U,,
converges to U uniformly in L:,,.
Since Uh is an exact solution on [ tj, tj+ 1), we can integrate by parts in
(3.27) to write

(3.31)

where

D&h, a, 4) = [zth [u(X, tj +) - u(x, tj -)] 4(x, tj) dx. (3.32)

Thus

D,(h, a, 4) G h 11~11
m 1~~1 (3.33)

Let dij be the constant function 4 s &xi, tj). Then

W, a,4)= c DJh,a,4)
= T D&h, a, dij) + c D& a, 4 - 4~). (3.34)
ij

505!h5.2-9
266 ISAACSON AND TEMPLE

But

(3.35)

where
.F = {i: 3j with [xi - h, xi + h] x [It,, t, + k] n supp(4) not empty},
f = {j: 3i with [xi- h, x,+h]x[t,,f,+k]nsupp(fj)notempty}.
Also, set
9=9x$@.
By Lemma 3.3, with
4= (i: (i,j)eX}),

we obtain
1 IYJ = c + c 63&Mh-‘+M(&)F(j)~3&Mh-‘+M(E)
ie./ ic./n.%, ic.Pnx;

where we have applied Lemma 3.2. (We now allow the generic constant M
to depend on the test function 4.) Thus by (3.35),

I~Dh a, 4-4.)~ 6 M IlV~ll,{3~ + ww

17
This implies that

(3.36)

as h -+ 0 as follows: for any 6 > 0, choose E< 6/(6M IlV& m}. Then for all
h-c 6/{6M(&)M /VdII m} we have

Consider now the first sum in the right-hand side of (3.34). We have

(3.37)
 SINGULAR HYPERBOLIC SYSTEM 267

since

DiiDk,da =O for (i, A Z (k 0,

sA

by the independence of sampling points. But for any a E A,

1 D&h, a, 4ii)2

= c + c
(i,j)czYnX (i,j)sYnX[

G 1 232~2+ 1
114112,h h2W4*(lyijlz+ ICgld*
11~112, (3.38)
(WE 9 (i&E 9

where we have applied Lemma 3.3. But with E< 1 we obtain

and

1 lhll’, h*WG*(ly,l,+ IC,l.)*

G C WE) h*C (Irijz+ IC,Id2

icd I
2

ii
G C W4h2 1 (luijl,+ ICijlJ
is%
< C M(E) h2F(j)* < M(c)h.
ieb

Thus from (3.38),

Using this estimate in (3.37) yields

c D&h, a, dij) 2 6 ME + M(E)h. (3.39)

0 >
268 ISAACSON AND TEMPLE

Now as in (3.36), (3.39) implies that

C D,,@,a, 4,) ’ da + 0 (3.40)

I, >
as h --t 0. Therefore there exists a set N of measure zero in A and a sub-
sequenceh + 0 such that, if a E A\N, then

C D&h, a, &I -, 0 (3.41)

iJ

as h + 0. Therefore

D(h, a, d) + 0
as h -+ 0 by (3.36), (3.41), and (3.34). This achieves the desired result for a
particular 4, since N depends on 4. To extend this to arbitrary 4 E CA, we
can apply the diagonal process in a direct manner as in [ 1, 71. This com-
pletes the proof.

ACKNOWLEDGMENTS

We would like to thank Professors Dan Marchesin and P. J. Paes-Leme for their generous
hospitality during our stay at PUC/RJ.

REFERENCES

1. J. GLIMM, Solutions in the large for nonlinear hyperbolic systems of equations, Comm. Pure
Appl. Math. 18 (1965), 697-715.
2. E. ISAACSON, Global solution of a Riemann problem for a non-strictly hyperbolic system of
conservation laws arising in enhanced oil recovery, Rockefeller University preprint.
3. B. KEYFITZ AND H. KRANZER, A system of non-strictly hyperbolic conservation laws arising
in elasticity theory, Arch. Rat. Mech. Anal. 72 (1980), 219-241.
4. P. D. LAX, Hyperbolic systems of conservation laws, II, Comm. Pure Appl. Math. 10
(1957), 537-566.
5. P. D. LAX, Shock waves and entropy, in “Contributions to Nonlinear Functional Analysis”
(E. H. Zarantonello, Ed.), pp. 603-634, Academic Press, New York, 1971.
6. D. W. PEACEMAN, “Fundamentals of Numerical Reservoir Simulation,” Elsevier, New
York, 1977.
7. B. TEMPLE, Global solution of the Cauchy problem for a class of 2 x 2 nonstrictly hyper-
bolic conservation laws, Adu. in Appl. Math. 3 (1982) 335-375.