OJ,s6. / x where. T and P., are linear operators which will be discussed in more. O), S .... (2.31 ) {B cos 6~ A sin Q)ue + [(B sin 6 + A cos d)rÑr+ {Cico - s)m] = 0.
TRANSACTIONS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 176, February
1973
INITIAL-BOUNDARYVALUE PROBLEMSFOR HYPERBOLIC SYSTEMSIN REGIONSWITHCORNERS. 1(1) BY
STANLEY OSHER ABSTRACT. In recent papers Kreiss and others have shown that initialboundary value problems for strictly hyperbolic systems in regions with smooth boundaries are well-posed under uniform Lopatinskii conditions. In the present paper the author obtains new conditions which are necessary for existence and sufficient for uniqueness and for certain energy estimates to be valid for such equations in tegions with corners. The key tool is the construction of a symmetrizer which satisfies an operator valued differential equation.
I. Introduction.
estimates
for mixed
differential
symmetrizer
problem
with variable improved
[ill,
in regions
with corners.
seems
[6] has
to have
recently
in regions
real
with
[lOl,
to permit
estimate. higher
in Kreiss'
classical
and uniqueness
in the nature
energy
of the symmetrizer
was
and by Rauch
obtained
work was done by Hersh
results
estimates.
the
operator
His result
independently
Earlier
around
problems.
initial-boundary
coefficients,
[12]
partial
hyperbolic
L2 norm.
valued
Sakomoto
value
for a strictly
order equation.
obtaining
hyperbolic
type of mixed
in the
complex
existence
without
work came
boundary
is well-posed
for a single
coefficients
that a general
a smooth
energy
The work will revolve
of its own.
a semigroup
constant
for certain
a significance
results
l3J, who obtained
problems
initial-boundary
to obtain
similar
will be to obtain
for general
shown
coefficients
by Ralston
of papers
value
of a new symmetrizer
Kreiss value
series
initial-boundary
equations
introduction This
The aim of this
for equations
An important
he used
with
new idea
in order to obtain
his estimate. The symmetrizer homogeneous appear
in the Kreiss
analogous
we shall
differential
condition
Lopatinskii
use the symmetrizer
result
in a simpler
below
These
involves
for elliptic
introduced
certain
are the same
for well-posedness,
condition
paper,
Received
introduce
equation.
and indeed
systems
here to obtain
[8l.
solutions
exponential which
We shall,
special
cases
to a
solutions appear
that in the
in a later
of Kreiss'
fashion.
by the editors
December
27, 1971.
AMS(MOS)subject classifications (1970).Primary 35L50, 35L30; Secondary 78A45. trizer,
Key words and phrases. Hyperbolic energy estimate, well-posedness.
(') Research
partially
supported
equations,
under N.S.F.
141
initial
boundary
conditions,
symme-
Grant GP29-273. CopyrightC 1973, Americin Mjthemiticil Society
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
142
STANLEY OSHER In the regions
space
problem
give
a simple
natural
with corners
need
counterexample
and necessary
does
exist,
down the answer
has certain
a weakening
elliptic
boundary here.
The situation of Toeplitz
invertibility
invertibility Special
value
to the condition
in a natural
We then construct
for constant we shall
It should
remains
encountered operators
paper
the functional
be noted
that
that
this
sym-
calculus
in the elliptic
[4] has shown
near a corner,
an
In order to do
in a future
Kondrat'ev
for case,
that
in distinction
an
to what
section.
here appears
to be analogous
in two dimensions,
of the corner hyperbolic
such
way, and we can
coefficients.
show
is both
a new norm, with which
and hence
in the last
of the two half space
in connection
which
applies.
this
half
We shall
coefficients
properties,
problem
for each
problem.
for constant
of the norm at the corner,
We discuss
condition
a new condition
coefficients,
smoothness
Kreiss
of the corner
if a solution
at least
operators
modulo
happens
case.
is obtained,
pseudo-differential
study
in this
with variable
metrizer
below,
We introduce that
then we are led directly
estimate
problems
below.
in the sense
write
a priori
discussed
not imply well-posedness
[February
problems
e.g.
to that found in the
Douglas
is necessary,
and Howe [2l, where
but not sufficient,
for
problem.
differential
with water
waves
equations
in such regions
over sloping
wedge
problems
in optics,
e.g.
Kraus
author
[7], used
a new Weiner-Hopf
beaches,
and Levine
technique
e.g.
[5].
have
Peters
Recently,
to obtain
a closed
been
studied
[9], and in Kupka
and the
form solution
to
such a problem. II. Statement system
with
of the problem;
constant
results.
We begin
by considering
the hyperbolic
coefficients, n
(2.1)
Au + Bu + 22 C.u y
where
z = (zol3 z,,---, 4''«'
tem is hyperbolic
y=3
z ), u and
in the sense
(2.2)
'
Zi
+ Du - u = F(x, y, z, t),
F are complex r
that
for the
valued
m x m matrices
m vectors
A, B, C
and the sys-
the matrix
det [Ait; + Bit] + Cico - A] = 0
has only purely
imaginary
roots.
Here Cioj = 2"=3 Cico-, co = (d>3, • • • , ù>n), and 0. (We shall square
to solve
this
x, y, ¿>0;
(2.3)
use the convention
root of the sum of the squares
We wish
with
always
initial
problem
that
| | of a vector
of its components.) in the region:
-°° < z. < °°,
/' = 3, 4,• • •, n,
conditions
u(x, y, z, 0)= 0.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
is the
+
1973]
INITIAL-BOUNDARY VALUE PROBLEMS. I We further
assume
that
0
(2.4)
143
■••
a. 0, k= p+ I,-.-,
m.
b '
conditions
u\0,
\
' / '••V
0
boundary
that
of the form
y, z, t) - SuU{0, y, z, t) = g(y, z, t),
(2.6) (b)
{Tu)m(x, n, z, t) - R{TtAlV(x, 0, z, t) = h{x, z, t),
where
u1 = («j,.. III
« 5 and rows,
/ =(«j,...,
R are constant
(m ~ p) columns
It is clear the corner
the two half
space
. with
an=
\T
,
«
IV
having
(ai+1>- • •, «m)T, / = ("p+1,---,
/ rows
«„)
and
\T
■
(m — I) columns,
and
p
respectively.
speed
of propagation
to be well-posed
considerations,
in any reasonable
that in order for
sense,
it is necessary
that
problems:
Equation
(a)
af) matrices
from finite
problem
• , tvpT,
(2.1)
boundary
in region
0 < x,
condition
(2.6(a))
t < oo, —oo < y, z. < oo,
?
and Equation
(b)
(2.1)
.
in region
0 < y, t < oo, - oo < x, z. < oo,
t
with boundary condition
^
(2.6(b))
be well-posed. We first transform
Laplace
transform
in y and all the
(2-7) [it is easily
z.
(2.1)
in time and use
variables.
We have,
ux + A~1{Bíxú2 + ¡Coi - s)u= shown
may be set equal
that the matrix to zero
D plays
(2,3).
In problem
(a), Fourier
if F = 0,
0,
only a trivial
throughout.]
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
x > 0. role in this problem
and
144
STANLEY OSHER It is easily
seen
that
Kreiss'
co ?, to are real, We can show linearly
(e.g.
independent
condition
[3J), that
them at x = 0, and obtain
(2.8)
this which
condition
with
ordinary decay
for both half spaces
for all
77> 0, £ real.
differential
equation
exponentially
as
has
exactly
/
x —>+ 00. Normalize
/ functions
$y(x, ú>2,ú>, s), Kreiss'
is valid
s = r¡ + iç
solutions
[February
7= 1, 2,— , /.
is, at x = 0,
(2.9)
determinant
[/, - S] [ {, ■. . , 4^] / 0
for all oj2, co, s, with Real s > 0. This formed
condition
boundary
is natural
condition
in that
if one wishes
of (2.6(a)),
to solve
the inverse
of this
(2.7) with the trans-
matrix
must
appear
in
the solution. Kreiss'
condition
for problem
(b) is completely
analogous.
We make an assumption. (2.1)
Assumption.
Kreiss'
We now proceed
by first
condition
assuming
for problems
that a solution
F = 0, h = 0, and then
seeing
(2.1')
Aux + Bu y + (Cico - s)u = 0,
(2.6')
(a)
(2.6')
(b) Let
what this
is valid
assumption
ul{0, y, co, s) - Sull(0,
We have
x, y>0. J —
y, cd, s) = g{y, to, s),
v be the
function
defined
for
y>0,
— «> 0, which
if 0 < y < oo,
if -oo < y < o.
if obeys
Aw + Bw + (Ci(ú-s)w x
= ß(3(y - 0)íXx, 0, oj, s),
y
x > 0,
(2.15)
with boundary
J
conditions
wKo, y, oj, s) - SwU(Q, y, oj, s) = g(y, oj, s)
if y > 0,
wKo, y, oj, s) - S^'HO, y, co, s) = 0
if y < 0.
We can Fourier
transform
in the analogous
solution
if we require
for x > 0, — oo < y < oo with
w\x, y, co, s) = 0
plane
0,0;, s) = 0,
0, tu, s) for x > 0.
let
w(x,
Then
145
does
indeed
in y and proceed
fashion exist
as above,
to this
to solve
finally
problem,
this
obtaining
in the right
half
w{0, y, co, s).
If a
then it must be true that
- (2.27)(b)
real,
the estimate
'dy{l+
the equation
this
method
Moreover,
we assume
in the manner we shall
discussed
derive
our first
above
main theorem
in (2.10)
to (2.18)
below.
will not be applicable
the existence
similar
((/ - TM )~ )*.
it is not too hard to obtain
of the estimates
Moreover,
oj, £.
operator
assumptions,
dy(i + \co\i+\s\2)K>U)dy,
in the case
of a symmetrizer
of variable
has an independent
coeffi-
abstract
interest.
The operators Kondrat
ev's
(y (d/dy)r
paper
[4] on elliptic
them here for technical of the matrices
K2 = 0 does
right-hand
reasons
in (2.24)
certain
However,
e.g.
Kupka
and Osher
tion
is often
cuss
this
disguised
in conical
concern
(K2
fashion
regions.
certain
if there
of eigenvalues
exist
is the constant
in
We introduced
multiplicities
We are not sure
K~ > 0 does.
matter
the solvability integral
sometimes
itself
[71. The
quite
equations
(2.6)
but
singular
to handle.
integral
equations
which
and (2.25).
not suffice,
in a somewhat
cases
in which
appearing
on the
side of (2.27)(a).)
In order to check analyze
appear
here
in a forthcoming
Definition. u(x,
Let
vector
function
which
are true for each
exact
solutions which
Thus,
with estimate
which,
algebra
nontrivial.
arising
of (2.26),
equations
of these must
(2.27),
in general,
equations
be checked
it is not surprising
often require
special
we must
are rather
can be obtained,
in the Kreiss
that
analysis.
difficult
condi-
the singular
We expect
to dis-
paper.
S(x, y, co, s) be the Fourier-Laplace
y, x, t). We shall
obtain
estimates
on/
transform Jz/(x,
of any
y, co, s)\ dxdy
co, s.
Define for fixed ru, s
(2-28)
W\2p_Q- f^1
de ¡~(l+\co\2+
\s\2)Vrdr
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
£
\(r £Ju(r,
6, co,
148
STANLEY OSHER
Q is some real analogous
number,
P is a nonnegative
integer,
defined
on the boundary,
e.g.
53(ex/2?, e>/2f0
(2.39)
-K,
£
(e-X/2{d/d\Yb,
e~k/2(d/dXYh)(l + |oj|2 + |s|2)
o
7=0
III. Proof of Main Theorem
the auxiliary
(3.1) with the boundary
(3.2)
I using
the symmetrizer.
We begin
problem
Adx + Bvy + {Cico-s)v=
F,
x, y>0,
conditions
?'(0, y, oj, s) = 0,
(TD) ln(y, 0, oj, s) = 0.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
by considering
2.
150
STANLEY OSHER
If we inner
product
the equation
part of both sides,
(3.3)
with
v, integrate
[February
over
x, y > 0, and take
the real
we have
(i-«)B^5,9+xg|îbii2i0+ic9iîb2i2>o r¡0 > 0, with
only consider
[6] and Ralston
r¡0 any positive
U in the neighborhood
[10]
have
(4.7)
shown
that
of every
there
point
exists
constant.
Thus
we
C = Cn> T] = 0. Kreiss
a smooth
W, with
W~lM\W
K where Re /Vj- 23^;
(4.8)
for 7/ = 0, £' = £q, / > 2,
M.(¿¿, 0)
k. is purely imaginary,
(4.9)
k. /= k. if / ¡¿ /,
M.(C, r,') = M.(¿¿,0) + JN.iO + 0((r,')2)+ Ey(£').
where
£ .(£ ) has
lower
left corner
For each
purely
imaginary
entries,
of N 7{¿, ), which
block
AL, consider
Det(Alf -*')=
(4.10)
we call
£.(£»)
= 0, and the element
n s .L , has
nonzero
the characteristic
real
in the
part.
equation
Det(AI;' - k!. - (*' - k'.))
= (*'. _ k')s + i»"1,',, 7
, + '*(£' - C>
'si
0
,
s 1
+ odT,'i2+ K'-g2), Re nsl
¿0,lmmsl
Thus
the roots
= 0. are
(4.11) k! = k'.- l-P-W'st-
W
- *,+
ocivi2 + K* -¿¿I2)]1*.
Th is means that, if |£' - £'0\ < k(¿,'0)[t]']
, the roots of M, split up with l/2s,
l/2(s
part if 77 > 0. The remainders
+ 1), or l/2(s
have negative
- 1), having positive
real
real part.
Let
U. with
transformation
in the region
branch
0 < 77 < k. AC'r),
at Ç = ¿jQ, r¡ = 0,suchthat
UP
(Y-lM.ÏÏ.
(4.14)
are taken.
;
Re m['¡ > 25^, Re M^ < - 25^. We may do this
the compactness
for each
block
and for each
of the unit ball \ç \ = 1.
We may now use this
lemma
in order to obtain
transform
(4.4) with respect
to y, multiply
variables
g = Uh. Then (4.4) becomes
h +
(4.15)
A.
The result
then
follows
by
Q.E.D. by A"
'II
'12
0
'2 2
the solutions
to (4.4).
Fourier
, then make the change
of
h= 0,
hence
exp(-Lnx)ij
b=
(4.16) where
h.
is some
(m - I) vector
valued
= F"
(4.17) as a function
which
obeys
We may proceed
function
of A,,
'exp{-Ll{x)hl
U
0
(4.4).
in the same
way and obtain
"exp(-f\
V
(4.18) where
h? is some
valued
function
(m - p) vector
obeying
(4.4).
valued
(4.19)
^J_^exp(^2 ' expl-L
lly)h2
0 function
We now switch
1
■si
s, co). We define
of (,°°) independent
/>/2)/l£=F^_
that we may premultiply
of 0, then the result
7-1
/.:[?
V
(4.26) 72*(n/2)Av= FM-2\X£ We make the change
,
0)).
= /, + ;2 •
still
satisfies
At 0 = 77/2, we consider
(4.27)
rcos
now
(4.25) In this
0'
dco
N
#*,)=
of variables
5 (-In x.)
-L,
t
(oj.) exp(/oj-(A-
"exp(-0,
e"'))Av(r)¿r,
0
0
V*(co2)Av{r)dr.
We then
/S(x) = 0,
1
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
have,
letting
x. S2|p| 2_0,
and, if p satisfies
(4.34), then
(4-38)
(j*{0)Bv,NJ*(0)Bv) 52(7^27-22 + l).
Let
(4.52)
^jv.-^ug^^^-,
and
(4.53)
Nn = -S2.
Thus
N21 = 52(/+T*1Tn)+Re52(-T22
- 7*, 7^(7^
- l)~ 1Tn
(4 54) = 52[l-Re(T22+r*)(Tu>s-l)-An], and
^2=4l[T22-7íT^(T«..-l)"1 _12(It V. Verification
of the estimates.
(3) of Main Theorem scalar
(5.1)
operators
II. We begin
_ t*1(t We shall by noticing
_ i)-1 _ 7*). now verify
conditions
that the operator
of the form
q(co , s, co) I
exp(-A(w2,
s, co)x)x!f(x)dx,
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
7*
(1), (2) and involves
1973]
159
INITIAL-BOUNDARY VALUE PROBLEMS. I
where
A and
q are
except
for certain
C°° functions algebraic
of all their
branch
points
variables
where
if s = r) + iç,
multiple
r¡ > t/„ > 0,
eigenvalues
of
B~ {\ico? - s + Cico) occur.
However,
(5.2)
because
of (2.25),
\x' exp(-A(oj2,
Let
x = 1/Re
we have
s, co)x)q(co2,
s, oj)| < Kj expi-S^rj),
x > 0.
A(oj2, s, oj); we then have
(5.3)
\q((o2, s, oj)| < Kl2[Re A(oj2, s, co)]'.
If we integrate
(5.1)
by parts
j times,
we obtain
^(oj2, s, oj)
(5.4)
/ t9V j
exp(-A(oj2,
s, co)x)[—j
• (x'f{x))dx
A?(oj?, s, co We need
now to estimate
(5.5) i.e.
J~
we view this
exp(-A(oj2, s, co)x)f{x)= PxFf,
as an operator
(5.6)
acting
on the Fourier
transform
of /:
f™ exp(-ico2x)f(x)dx= Ff= f(co2)
which takes the
the operator
lower
oj2 into half
plane
- z'A(oj2, s, co). In order into
the unit
circle
to estimate
and then
use
this,
we shall
Theorem
transform
1 of Carleson
[l].
Notice
(5.7)
/.U+l)\
[//=/
is a unitary
formation
map of L2[0,
which
takes
oo] one-to-one
f(i(z
2K
VU-1)7(1-«) onto
H
of the circle.
We seek
the trans-
+ \)/{z - 1))(1/(1 - z)) into f{- ik{i{z + l)/{z-l)))(!/(!
-z)\
Let (z + 1)
M =~U-i)
(5.8) Then
the transformation
The function
A is an eigenvalue
We can show therefore,
as
for some positive
_,.
.
z + i
\z) = -r.
is
Tg = g(p-H-iA(p(z))))
(5.9)
expansion
in question
p
of ß~
p'(z)
p'(p-H-¿A(pU)))) (A/oj2 + Czoj - s) with positive
(x>2I 0 that the function
integer
A(1/oj2)
p.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
real part.
has a Puiseux
160
STANLEY OSHER
LFebruary
À-,
(5-10)
A(1A) = — + T kxüL'*-1 W2
7=1
with A_j / 0. We can differentiate
this
and obtain CO
(5-11)
X'(co2)= A_, + X A.(l - ;/>X«u2)-**. 7= 1
Thus,
the function
- z'A(cu2) is one-to-one
if co2 > M(s, co), also
if &>. ) < M0(l + \s\2 + |W|2)Ql/2
for some
öi > 0.
We shall
only estimate
the operator
in (5.5) for co2 > M. This
corresponds
to
bounding
(5.13)
j\(/A(-zA(M^))))|2
-^-f-^
lAp-H-ittpiJ0))))} for e> 0j > 0. By the one-to-one
property
of A, we may make the change
of
variables
(5.14)
z = /A(AA)~ VA)Ï = ei6,
0.15)
|pU)A| -
so we must
, '^
Q0,
We may replace
f(a>2)= Af+{l-A)f, A(co2.
co, s)
The quantity wish
We next
of the form
(5.21) where
161
I
is in C°°,
A= 1
if |oj j| > M(s, co) + 1,
A= 0
if |oj,| < Al(s, oj),
involving
(1 - A) f can be easily
0 < A < 1.
estimated
appropriately.
We thus
to estimate
(5.22)
xl
We may integrate
(5.23)
I
by parts
j times,
s, co)x)qf(co2)dco2.
and obtain
f°° exp(-A(oj„ s, co)x)U—) q(co , s, co)f(co )dco . JM r 2 \cho2 A (oj2, s, co)/
We can use (5.3), modulo
different
(5.10) K,
and (5.11)
and
In order to estimate
Q22-
of {x(d/dx)Y
on them.
(5.1) with respect
of the form
Because
K,(l
+ |oj|2
to obtain
Of course,
products
differentiate bound
exp(-A(oJ2,
the same
a similar
of these
operators
of Assumption
(2.3)
3'
. Thus
as for 7'22, is valid
we must consider or Lemma
to oj2, and the effect + |s|2)
estimate
argument
(4.1),
for T...
the effect we may
will be only an increase
we are considering
in
expressions
of the form (5.24)
j"1 dx
y > 0 and
I
q is uniformly
we have
(modulo
(5.25)
Í
This (5.26)
dco exp(;'«J2y
becomes
dx I
- A(oj2, s, to)x){co2y)'q{co2,
bounded.
Integrate
by parts
/ times
s, co)y—{xTf(x))j with respect
tooj2,
constants)
exp(/oJ2y) (-t--)
a sum of terms
[w72exp(-A(oj2, s, új)x)í7(oj2,s, oj)](—]U7('
2> of the form
J "° dx J °^ exp{ico2y-)ico2,s,co)x)[cot'2xp
q\.co2, s, oj)]i^-j
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
{xrf(x))dx,
dx,
162
STANLEY OSHER
p = 0, 1,.. by parts
have
., /, where p times
terms
(5-27)
in x, recall
have
2
0 /)_ Next integrate
i Q(p) + |s\z) ! > co2/X(co2,
s, co). We then
* **/(*)«&.
shown
IVIp,Q