Lecture Notes in

Mathematics
Edited by A. Dold and B. Eckmann

1192

Equadiff 6
Proceedings of the International Conference
on Differential Equations and their Applications
held in Brno, Czechoslovakia, Aug. 26-30, 1985

Edited by J. Vosmansk} and M. Zl~,mal

I I

Springer-Verlag
Berlin Heidelberg New York Tokyo
Editors

Jaromir Vosmansk~
J.E. Purkyn~ University, Department of Mathematics
Jan&~,kovo n~.m 2a, 662 95 Brno, Czechoslovakia

Milo~ Zl&mal
Technical University, Computing Centre
Obr~ncQ miru 21,602 00 Brno, Czechoslovakia

Published in co-edition with Equadiff6, J. E. Purkyn~ University, Department of

Mathematics, Brno, Czechoslovakia.

Sole .distribution rights outside the East European Socialist Countries, China, Cuba,
Mongolia, Northern Korea, USSR, and Vietnam:
Sprirger-Verlag Berlin, Heidelberg, NewYork, Tokyo

Mathematics Subject Classification (1980): 34-02, 35-02, 65-02, 73-02, ?6-02,

80-02

ISBN 3-540-16469-3 Springer-VeAag Berlin Heidelberg New York Tokyo

ISBN 0-387-16469-3 Springer-Verlag New York Heidelberg Berlin Tokyo

All rightsreserved.No partof this publicationmaybe reproduced,storedin a retrievalsystem,

transmittedin anyform byanymeans,mechanical,electronic,photocopying,recordingor
otherwise, without the previousconsent in writing from the Publisher.
© Equadiff6 and Springer-VerlagBerlin Heidelberg 1986
Printed in Czechoslovakia
Printing: Tisk, Brno
Binding: BeltzOffsetdruck, Hemsbach/Bergstr.
214613140-543210
 III.

PREFACE

Following the tradition of the previous Conference EQUADIFF 1-5,

held periodically in Prague (1962, 1977), Bratislava (1966, 1981) and Brno
(1972), The 6th Czechoslovak Conference on Differential Equations and
Their Applications EQUADIFF 6 was held in Brno from August 26 to Au-
gust 30, 1985. The Conference was organized by the University of J. E. Pur-
kyn~ in Brno with support of the International Mathematical Union in
cooperation with the Technical University in Brno, the Mathematical Insti-
tut of the Czechoslovak Academy of Sciences, Society of Czechoslovak Ma-
thematicians and Physicists, sponsored by the Faculty of Mathematics and
Physics of the Charles University in Prague, the Faculty of Mathematics
and Physics of the Comenius University in Bratislava, the Czech Technical
University- in Prague, the Faculty of Science of the PalackS~ University in
Olomouc, the Faculty of Science of the University of P. J. Safarik in Kogice,
the School of Mechanical and Electrotechnical Engineering in Plzefi and
the School of Transport and Communications in Zilina.
EQUADIFF 6 was prepared by the Organizing Committee president by
M. Zlgmml, chairman, and J. Vosmansk:~, executive secretary, with the help
of the local organizing staff.
The topic of this meeting were differential equations in the broad
sense including numerical methods of their solutions and applications. The
main goal was to stimulate cooperation among various branches in differ-
ential equations.
The Conference was attended by 473 participants (207 from Czechoslo-
vakia, 266 from abroad) and 62 accompanying persons from 31 countries.
36 participants from abroad were granted the financial support.
92 invited mathematicians from abroad took part in the Conference
and together with Czechoslovak scientists delivered plenary lectures and
other invited lectures and communications in sections. The participants
had the opportunity to deliver their papers as communications, at the pos-
ter session or in the form of the enlarged abstracts (without oral presenta-
tion).
The scientific program comprised 10 plenary lectures and 64 main lec-
tures in the following sections:

1. Ordinary Differential Equations (20)

2. Partial Differential Equations (16)
3. Numerical Methods (14)
4. Applications (14)
 Iv.

In addition 251 p a p e r s w e r e p r e s e n t e d

a) as c o m m u n i c a t i o n s in 9 s i m u l t a n e o u s s u b s e c t i o n s (136)
b) at the p o s t e r session (461)
e) in the form of enlarged a b s t r a c t s (70)

Besides the scientific p r o g r a m the participants and the a c c o m p a n y i n g

p e r s o n s could e n j o y a rich social program.
Two slightly different parallel editions of this v o l u m e are published.
The Springer-Verlag edition contains 9 plenary lectures and 48 main lec-
t u r e s in sections representing the substantial p a r t of lectures p r e s e n t e d at
the Conference. The EQUADIFF 6 edition for the participants of the Confer-
ence and for the socialist c o u n t r i e s contains also S u p p l e m e n t consisting of
7 additional contributions. These c o n t r i b u t i o n s are not fully compatible
with the conditions for the Lecture Notes publication and their revised ver-
sion could n o t be arranged.

Editors
CONTENTS

PREFACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ~i i
LIST OF PAPERS PRESENTED AT THE CONFERENCE . . . . . . . . . . . . . . . . . . . ix
LIST OF PARTICIPANTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv i i

PLENARY LECTURES:

F E I S T A U E R M.: Critical p o i n t t h e o r y a n d n o n l i n e a r d i f f e r e n t i a l
equations ................................................ 3
FRIEDMAN A.: F r e e b o u n d a r y p r o b l e m s in fluid d y n a m i c s . . . . . . . . . . . . . . . 17
KA~LrR J.: M e t h o d of R o t h e in e v o l u t i o n e q u a t i o n s . . . . . . . . . . . . . . . . . . . . . 23
KUFNER A.: B o u n d a r y v a l u e p r o b l e m s in w e i g h t e d s p a c e s . . . . . . . . . . . . . . . 35
MAWHIN J.: Critical p o i n t t h e o r y a n d n o n l i n e a r d i f f e r e n t i a l e q u a t i o n s . . . . . . 49
N~UWIAN F.: O r d i n a r y l i n e a r d i f f e r e n t i a l e q u a t i o n s - a s u r v e y
o f t h e global t h e o r y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
REKTORYS K.: N u m e r i c a l a n d t h e o r e t i c a l t r e a t i n g of e v o l u t i o n
p r o b l e m s by t h e m e t h o d of d i s c r e t i z a t i o n in t i m e . . . . . . . . . . . . . . . . . . . 71
STETTER H. J.: A l g o r i t h m s for t h e i n c l u s i o n o f s o l u t i o n s of o r d i n a r y
initial v a l u e p r o b l e m s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
TRIEBEL H.: R e c e n t d e v e l o p m e n t s in t h e t h e o r y of f u n c t i o n s p a c e s . . . . . . . . . 95

LECTURES PRESENTED IN SECTIONS:

A. Ordinary differential equations:

BARTUSEK M.: On p r o p e r t i e s o f o s c i l l a t o r y s o l u t i o n s
of nonlinear differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
BURTON T. A., DWIGGINS D. P.: U n i q u e n e s s w i t h o u t c o n t i n u o u s
dependence ...... ; ........................................ 115
FIEDLER B., BRUNOVSKY P.: C o n n e c t i o n s in s c a l a r r e a c t i o n d i f f u s i o n
equations with Neumann boundary conditions .................... 123
GREGUS M.: On a c e r t a i n b o u n d a r y v a l u e p r o b l e m o f t h e t h i r d o r d e r . . . . . . . . 129
KRBEC P.: On n o n p a r a s i t e s o l u t i o n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
KREITH K.: U n i f o r m z e r o s Ibr b e a d e d s t r i n g s . . . . . . . . . . . . . . . . . . . . . . . . . . 141
KURZWEIL J., JARNIK J.: P e r r o n i n t e g r a l , P e r r o n p r o d u c t i n t e g r a l
and ordinary linear differential equations . . . . . . . . . . . . . . . . . . . . . . . . 149
MULDOON M. E.: On t h e z e r o s of s o m e s p e c i a l f u n c t i o n s :
differential equations and Nicholson-type fornmlas . . . . . . . . . . . . . . . . . 155
SEDA V.: S u r j e c t i v i t y a n d b o u n d a r y v a l u e p r o b l e m s . . . . . . . . . . . . . . . . . . . . 161
 vt

SVEC M.: Some p r o b l e m s c o n c e r n i n g the e q u i v a l e n c e s of two

s y s t e m s of differential e q u a t i o n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
TRENCH W. F.: Linear p e r t u r b a t i o n s of g e n e r a l d i s c o n j u g a t e
equations ................................................ 181
TVRDY M.: On o p t i m a l c o n t r o l of s y s t e m s with i n t e r f a c e side
conditions ................................................ 187

B. Partial differential equations

BEBERNES J. W.: A d e s c r i p t i o n of blow-.up for the solid fuel

ignition model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
BRILLA J.: Spectral a n a l y s i s of n o n - s e l f - a d j o i n t elliptic o p e r a t o r s . . . . . . . . . . 197
CHANG K, Ch.: On the m o u n t a i n p a s s l e m m a . . . . . . . . . . . . . ............. 203
GAJEWSKI H.: On u n i q u e n e s s a n d s t a b i l i t y of s t e a d y - s t a t e c a r r i e r
d i s t r i b u t i o n s in s e m i c o n d u c t o r s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
GIAQUINTA M.: Partial r e g u l a r i t y of m i n i m i z e r s . . . . . . . . . . . . . . . . . . . . . . . 215
KREJCi P.: Periodic s o l u t i o n s of p a r t i a l differential e q u a t i o n s
with hysteresis ............................................ 221
KUCERA M.: S t a b i l i t y a n d b i f u r c a t i o n p r o b l e m s
for r e a c t i o n - d i f f u s i o n s y s t e m s w i t h u n i l a t e r a l c o n d i t i o n s . . . . . . . . . . . . . 227
MAZ'YA V. G.: B o u n d a r y i n t e g r a l e q u a t i o n s of e l a s t i c i t y in d o m a i n s
with piecewise smooth boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
SIMADER C. G.: Higher r e g u l a r i t y of w e a k s o l u t i o n s of s t r o n g l y
n o n l i n e a r elliptic e q u a t i o n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
STARA J., JOHN O.: Some r e g u l a r i t y r e s u l t s for q u a s i l i n e a r
parabolic systems .......................................... 247
TRUDINGER N. S.: Classical b o u n d a r y v a l u e p r o b l e m s
for M o n g e - A m p e r e type e q u a t i o n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
VALLI A.: Q u a l i t a t i v e p r o p e r t i e s of the s o l u t i o n s to the
Navier-Stokes e q u a t i o n s for c o m p r e s s i b l e fluides . . . . . . . . . . . . . . . . . . . 259

C. Numerical methods

AGARWAL R.: On G e l ' f a n d ' s m e t h o d of c h a s i n g for s o l v i n g

multipoint boundary value problems ............................ 267
AXELSSON O.: Stability a n d e r r o r e s t i m a t e s v a l i d for i n f i n i t e
time, for s t r o n g l y m o n o t o n e a n d i n f i n i t e l y stiff e v o l u t i o n
equations ................................................ 275
BREZZI F.:Recent r e s u l t s i n the a p p r o x i m a t i o n of free b o u n d a r i e s . . . . . . . . . . 285
CERM_AK L., ZL.~MAL M.: F i n i t e e l e m e n t s o l u t i o n of a n o n l i n e a r
diffusion p r o b l e m with a m o v i n g b o u n d a r y . . . . . . . . . . . . . . . . . . . . . . . 291
DESCLOUX J., FERRO R.: A n a l y s i s of T h a c k e r ' s m e t h o d for s o l v i n g
the l i n e a r i z e d s h a l l o w w a t e r e q u a t i o n s . . . . . . . . . . . . . . . . . . . . . . . . . . 295
GOERISCH F., ALBRECHT J.: The c o n v e r g e n c e of a n e w m e t h o d
for c a l c u l a t i n g l o w e r b o u n d s to e i g e n v a l u e s . . . . . . . . . . . . . . . . . . . . . . . 303
JANOVSK~z V., MAREK I., NEUBERG J.: B i f u r c a t i o n a n a l y s i s of s t i m u l a t e d
Brillouin scaterring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
KRIZEK M.: S u p e r c o n v e r g e n c e r e s u l t s for l i n e a r t r i a n g u l a r e l e m e n t s . . . . . . . . 315
NEDELEC J. C.: Mixed finite e l e m e n t in 3D in H ( d i v ) a n d H ( c u r l ) . . . . . . . . . . . 321
NITSCHE J. A.: F r e e b o u n d a r y p r o b l e m s for Stokes' flows . . . . . . . . . . . . . . . . 327
SCHMIDT J. W.: E n c l o s i n g m e t h o d s for p e r t u r b e d b o u n d a r y v a l u e
p r o b l e m s in n o n l i n e a r difference e q u a t i o n s . . . . . . . . . . . . . . . . . . . . . . . 333
 VII

THOMI~E V.: E r r o r e s t i m a t e s for finite e l e m e n t m e t h o d s

for s e m i l i n e a r p a r a b o l i c p r o b l e m s with n o n s m o o t h d a t a . . . . . . . . . . . . . 339
WHITEMAN J. R.: S i n g u l a r i t i e s in two- a n d t h r e e - d i m e n s i o n a l
elliptic p r o b l e m s a n d finite e l e m e n t m e t h o d s for t h e i r t r e a t m e n t . . . . . . . 345
ZENISEK A.: Some n e w c o n v e r g e n c e r e s u l t s in finite e l e m e n t
t h e o r i e s for elliptic p r o b l e m s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

D. A p p l i c a t i o n s

BOLEK P., FOIST J., KOZEL K., POLASEK J.: M a t h e m a t i c a l s o l u t i o n

of d i r e c t a n d i n v e r s e p r o b l e m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
COLLATZ L.: Einige A n w e n d u n g e n der M e h r d i m e n s i o n a l e n Apprixi-
m a t i o n s t h e o r i e zur L 6 s u n g s e i n s c h l i e i ~ u n g bei R a n d w e r t a u f g a b e n . . . . . . 367
KODNAR R.: A p o s t e r i o r i e s t i m a t i o n s of a p p r o x i m a t e s o l u t i o n s
for s o m e t y p e s of b o u n d a r y v a l u e p r o b l e m . . . . . . . . . . . . . . . . . . . . . . . . 373
KUB|~EK M., HOLODNIOK M.: N o n l i n e a r d y n a m i c s y s t e m s --
bifurcations, continuation methods, periodic solutions . . . . . . . . . . . . . . 379
MARTENSEN E.: The r o t h e m e t h o d for n o n l i n e a r h y p e r b o l i c
problems ................................................. 387
MEISTER E.: Some s o l v e d a n d u n s o l v e d c a n o n i c a l p r o b l e m s
of diffraction t h e o r y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ~. 393
NECAS J.: E n t r o p y c o m p a c t i f i c a t i o n of the t r a n s o n i c flow . . . . . . . . . . . . . . . . 399
NEUSTUPA J.: The global e x i s t e n c e of w e a k s o l u t i o n s of t h e
mollified s y s t e m of e q u a t i o n s of m o t i o n of v i s c o u s
c o m p r e s s i b l e fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
SADOVSK~ Z.: B i f u r c a t i o n s n e a r a d o u b l e e i g e n v a l u e of tile
r e c t a n g u l a r plate p r o b l e m with a d o m a i n p a r a m e t e r . . . . . . . . . . . . . . . . 415
SMITALOV~ K.: Dealy m a k e s p r o b l e m s i n p o p u l a t i o n m o d e l l i n g . . . . . . . . . . . 421

SUPPLEMENT (in the EQUADIFF 6 edition

only)

ARSCOTT F.: AnalyticM a n d c o m p u t a t i o n a l p r o b l e m s

in h i g h e r special f u n c t i o n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
CHANTURIA T. A.: O n o s c i l l a t i o n of s o l u t i o n s of l i n e a r o r d i n a r y
differential e q u a t i o n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
FAZEKAS F.: D e t e r m i n i s t i c a n d s t o c h a s t i c v e c t o r differential
e q u a t i o n s a p p l i e d in t e c h n i c a l s y s t e m s t h e o r y . . . . . . . . . . . . . . . . . . . . . 435
HASLINGER J.: Shape o p t i m i z a t i o n in c o n t a c t p r o b l e m s . . . . . . . . . . . . . . . . . 445
KLOTZLER R.: Zur a n a l y t i s c h e n L 6 s u n g a l t e r u n d n e u e r
geometrischer Optimieringsprobleme ........................... 451
KOSHELEV A. I.: On the s m o o t h n e s s of the s o l u t i o n s to the elliptic
systems .................................................. 459
KUSANO T.: On the a s y m p t o t i c b e h a v i o r of s o l u t i o n s of n o n l i n e a r
ordinary differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
 VIII

LIST OF FURTHER MA1N LECTURES

PRESENTED AT THE CONFERENCE

PLISS V.: Stable and unstable manifolds of hyperbolic s y s t e m s (plenary lecture)

A T K I N S O N F. V.: Critical cases of certain ground-state problems for nonlinear
wave equations
B O B R O W S K I D.: Boundary-value problems for r a n d o m differential equations
EVERITT W. N.: On linear ordinary quasi-differential equations
H E D B E R G L. I.: Sobolev spaces and nonlinear potential theory
L A Z A R O V R.: Superconvergence of the gradient for triangular finite elements
M A R K O W I C H P.: The semiconductor device equations
M A S L E N N I K O V A V. N.: Boundary value problems for second order elliptic
o _ equations in domains having non-compact and non-smooth boundaries
PUZA B.: Ob odnom metode analiza razreshimosti kraevykh zadach dlja
o b y k n o v e n n y k h differentsialjnykh uravnenii
SELL G.: Lyapunov exponents and oscillatory behavior equations with negative
feedback
SUSSMANN H. J.: A theory of envelopes and high order optimality condition for
bang-bang controls
 IX

LIST OF PAPERS PRESENTED

AT THE CONFERENCE

I. P A P E R S P R E S E N T E D AS COMMUNICATIONS IN SECTIONS

A. Ordinary differential equations

ANGELOV V.: A coincidence theorem in uniform spaces and applications
ANGELOVA D.: Asymptotic and oscillation properties on functional - differential
equations
ANDRES J.: Higher kind periodic orbits
AUGUSTYNOWICZ A.: On the existence of continuous solutions of operator
equations in Banach spaces
BERKOVIC L. M.: A constructive approach in the theory of differential equations:
Factorization and transformations
BIHARI I.: A second order nonlinear differential inequality
BRESQUAR A. M.: Asymptotic solutions for the oscillatory differential equation
BIANCHINI R. M., CONTI R.: Local and global controllability
CADEK M.: Pointwise transformations of linear differential equations
DLOTKO T.: Initial functions as controls
DOSLA Z.: Differential equations and higher monotonicity
DOSLY O.: Transformations of linear differential systems
ELBERTA.: Eigenvalue estimations for the halflinear second order differential
equations
FENYO I.: On the interrodifferential equation
x ( t ) + kf~Jn(2~/tz) (t/z) n/2 X(k~ (z)dz = F(t)
FISHER A.: Almost periodic solutions of systems of linear and quasilinear
differential equations with almost periodic coefficients and with time lag
FOFANA M. S.: The stability of a special differential equation
FOLTYI~'SKA I.: An oscillation of solutions of nonlinear integro-differential
equations system
GARAY B. M.: Parallelizability in Banach spaces: Examples and counterexamples
GRAEF J. R., SPIKES P. W., ZHANG B. G.: Sufficient conditions for the oscillatory
solutions of a delay differential equation to converge to zero
GREGU~ M.: Nontrivial solutions of a nonlinear b o u n d a r y value problem
HABETS P.: On periodic solutions of nonlinear second order differential
equations
HADDOCK J.: Phase spaces for functional differential equations
HALICKX M.: Existence of regular synthesis for two classes of optimal control
problems
HATVANI L.: A generalization of the invariance principle to n o n a u t o n o m o u s
differential systems
JAROS J.: Oscillation criteria for forced functional differential inequalities
KARTAK K.: Generalized absolutely continuous solutions of ODE
KHEKIMOVA M.: Periodicheskie i kraevye zadachy dlya singulyarno
v o z m u s h c h e n n y k h sistem s impulsnym vozdeystvem
KISIELEWICZ M.: Compactness and upper semicontinuity of solutions set
of neutral functional - differential inclusions
KRISZTIN T.: On the rate of decay of solutions of functional differential equations
with unbounded delay
KRUPKOVfi. O.: The inverse problem of the calculus of variations
KULEV G., BAJNOV D.: ,,Prakticheskaya ustoichivost" sistem s impuljsnym
vozdeistviem pri postoyano deictvuyushchik vozmushcheniyakh
KUPPER T.: Identification through forced bifurcation
LAFORGIA A.: Turan - type inequalities for the zeros of the ultraspherical
and laguerre polynomials
LAITOCHOVA J.: Global transformations of linear second order differential
equations of a general form
LALLI B.: Oscillatory behavior of nonlinear differential equations with deviating
arguments
MARUSIAK P.: Oscillation theorems for nonlinear differential systems with
general deviating arguments
MAKSIMOV V. P.: O nekotorykh novykh napravleniyakh rozvitiya teorii
nelinejnykh uravnenii s posledeistviem
MEHRI B.: A note on existence of a periodic solution for certain non-linear second
order differential equation
MOLLER M.: Boundary-eigenvalue problems depending nonlinearly on the
parameter
MIHALIKOVA B.: O koleblemosti reshenii sistem differentsialjnykh uravnenii
MIKOLAJSKI J.: On nonoscillatory solutions of some systems of differential
equations
MIRONENKO V.: Reflective function of a system
MOSON P.: Quasi-periodic solutions of 4-dimensional systems
MULDOWNEY J. S.: The converse of Polya's mean value theorem
OMARI P.: Periodic solutions of lineard equations (a joint work with
F. ZANOLIN)
PUDEI V.: Zum Problematik der Extremall6sungen yon linearen Differentialchlei-
chungen n-ter Ordnung
RONKOV A.: Linear inequalities for functions defined in partialy ordered spaces
SHKIL N. I.: About periodical solutions of systems of second order differential
equations
SCHAAF R.: Time maps and global solution branches
SCHNEIDER K. R.: Integralmanitblds of periodic solutions of a u t o n o m o u s
differential equations
SIMSA J.: Asymptotic integration of linear differential equations of order N under
mild integral smallness conditions
TERJt~KI J.: On the stability of solutions of functional differential equations
with infinite delay
VANDERBAUWHEDE A.: Bifurcation of subharmonic solutions in time reversible
systems
VOLKMANN P.: Un theoreme d'existence pour les equations integrales de Volterra
dans les espaces de Banach
VRDOLJAK B.: On solutions of the lagerstrom equation
WYRWINSKA A.: Integrability of certain nonlinear differential equation with
deviating arguments
ZANOLIN F.: On a dynamical system in the Lienard plane
 tI

B. Partial differential equations

BIROLI M.: Wiener obstacles for A2

BOJARSKI B.: Microlocal analysis of linear transmission p r o b l e m s
DLOTKO T.: Geometric description of quasilinear parabolic equations
DRY,BEt( P.: Destabilizing effect of certain unilateral conditions for the s y s t e m of
reaction-diffusion type
DZIUK G.: A simple climate modell
FILO J.: On a nonlinear diffusion equation with nonlinear b o u n d a r y conditions:
Method of lines
FILA M.i. Connecting orbits in certain reaction diffusion equations
HEGEDUS J.: Zadaehi s o p r y a z h e n i y a dlya nekotorykh ellipticheskikh
i giperbolicheskikh uravnenii
HUEBER H.: Dirichlets problem for some hypoelliptic differential operators
KAMONT Z.: Weak solutions of first order partial differential equations with
a retarded a r g u m e n t
KAWOHL B.: Starshaped r e a r r a n g e m e n t and applications
KOLOMY J.: On accretive operators
LEWIS R. T.: The eigenvalues of elliptic differential o p e r a t o r s
LORENZI A.: An inverse problem for a quasilinear parabolic equation
in divergence form
MUSTONEN V.: Topological degree of mappings of m o n o t o n e type
and applications
NARAZAKI T.: Global classical solutions of semilinear evolution equation
NAUMANN J.: Liouvilte p r o p e r t y and regularity lor parabolic s y s t e m s
NETUKA I.: The best harmonic approximation
()TANI M.: Existence and non-existence of non-trivial solutions of some
nonlinear degenerate elliptic equations
PULTAR M.: Numerical methods of solution of hyperbolic equations
ROTHER W.: Generalized Thomas-Fermi-von Weizs/icker equations
SAL\~ R.: The equations of viscous incompressible n o n - h o m o g e n o u s fluids:
On the existence and regularity
SHOPOLOV N.: The first b o u n d a r y problem of a parabolic equation with
arguments reversing their roles
SOKOLOWSKI J.: Differential stability of solutions to constrained optimization
problems for p.d.e.
SPECK F.-O.: B o u n d a r y value problems for elliptic convolution type equations
SZULKIN A.: Minimax principles tbr lower semieontinuous functions and
applications to elliptic b o u n d a r y value problems
gVEC A.: Spectrum of spheres
TERSIAN S.: Characterizations of the range of Neumann problem for semilinear
elliptic equations
TIBA D.: Control of nonlinear hyperbolic equations
~IVRO J.: A b o u n d a r y value problem for quasilinear hyperbolic systems
of differential-functional equations
VERHULST F.: The Galerkin-averaging method lbr a nonlinear Klein-Gordon
equation

C. Numerical methods

AMIRALIEV G.: Towards the numerical solution of the system of Boussinesq

equation
 XII

BALLA K.: On error estimation of the approximative solution for certain singular
differential equations of Riccati type
BURDA P.: Finite element solution of a problem of potential flow
DECHEWSKI L. T.: A method for error estimation of numerical solutions
of differential equations
ELSCHNER J.: On suboptimal convergence of finite element methods
FRIVALDSZKY S.: Lineare und nichtlineare Mehrschrittverfahren mit variablen
Koeffizienten
FROHNER M.: Galerkin techniques and the method of lines applied to Burger's
equation
GUDOVICH N. N.: Ustoichivye raznostnye metody proizvoljnogo porjadka
approsimatsii dlja differentsialjnykh uravnenij
HAN H.: Nonconforming finite element approximation of Navier-Stokes equations
HEINRICH B.: On finite difference methods with fern-character for elliptic
problems
HLAVACEK I.: Shape optimization by the dual finite element method
CHOW Y.-M.: Initial-value methods for computing eigenvalues of two point
b o u n d a r y value problem
JOVANOVICH B.:
KRETZSCHMAR H.: Stabile zweischichte Differenzenverfahren
PIRC V.: On the possibility of calculation of zero points of solution of second
order differential equations
PR_~GER M.: Numerical illustration of the dimension reduction method
PROESSDORF S.: Spline approximation methods for singular integral equations
REGIIqSK~. T.: Superconvergence of external approximation for two-point
boundary value problems
ROOS H.-G.: Feedback grid generation via monotone discretization
S~_NDIG A.-M.: Fem error estimates for elliptic b o u n d a r y value problems
in domains with conical points
SEGETH K.: On the numerical evaluation of integrals involving Bessel functions
STANKIEWtCZ R.: Approximate methods for temporally inhomogeneous parabolic
equation
STREHMEL K.: Stability of linear implicit methods lbr retarded differential
equations
TAUFER J., VITASEK E.: Transfer of b o u n d a r y conditions for two-dimensional
problems
VULCHANOV N. L.: Numerical integration of asymptotic two-point b o u n d a r y
value problems for ODE
WEINER R.: Partitioned adaptive Runge-Kutta methods for the solution of stiff
and nonstiff differential equations
D. Applications
ANTES H.: Dual complementary variational principles in Reissner's plate theory
BECKERT H.: The bending of plates and their stability region
BOCK I.: Optimal control problems for yon
BRILLA I.: Bifurcation theory of the time dependent Karman equations
FARKAS M.: Competitive exclusion by zip bifurcation
JARUSEK J.: Optimalheating of bodies with constrains on stresses
JEDRYGA T. M.: An estimation of moment of the solution of a random operator
integral equation of Volterra's type
LAMZYUK V.: Ob odnom metode svedeniya granichykh zadach k nachaljnym
i ego ispoljzovanii pri reshenii zadach matematicheskoi fiziki
 XIII

LOVISEK J.: Optimal control of a variational inequality

MARKO L.: Buckled states of circular plates
MOSZNER Z.: On pseudo-processes and their extensions
NEDOMA J.: Contact problem in thermoelasticity. Coercive case.
POLCAR P., KOTOUL M.: On the numerical solution of two-dimensional stress
wave propagation problem
PETROV K.: Automodel of motion partial gaseus mixture in electric field
RUMPEL H.: Mathematische Modelle der Fluidmechanik
RI)ZICKOVA H.: On the transport-diffusion algorithm
SKIERCZYNSKI B.: Application of the methods of the sensitivity analysis
in obtaining the solution of nonlinear differential equations
SOBOTKA Z.: Solutions of ordinary non-homogeneous linear differential
. ecjuations following from rheological models
STEPAN G.: Delay of reflexes in balancing
VRKOC I.: Integral equations attached to skin effect

H. P A P E R S P R E S E N T E D AT THE POSTER S E S S I O N

BARTUZEL S.: Variational approach to certain diffusion problem

BARViNEK E.: The spectral theorem for normal diagonable operators on a real
Hilbert space
Bl~DA P.: On some global properties of a predator-prey model
CURGUS B.: Eigenfunction expansions associated with ordinary differential
operators with an indefinite weight function
CHERKAS L. A.: Periodicheskie resheniya avtonomnoi sistemy s fazovym
prostranstvom
DESPERAT T.: Difference methods for the solutions of differential-algebraic
systems
FARAGO I.: Dvykhshagovyi a-ustoichivyi metod dlya resheniya zadachi
khemosorbtsii
FARZAN R.: Zadacha rasprostraneniya elektromagnitnykh voln v sredakh
o neodnorodnostyami
GERGO L.: Adaptive finite element methods
GOPFERT A.: Approximation by solutions of elliptic equations
GRYSA K.: On use of a certain ordinary differential equation to finding the sums
of Dini series
HOROVA I.: On the variational principles for Dirichlet boundary-value problem
INVERNIZZI S.: Nonuniform nonresonance for jumping nonlinearities
JANKOWSKI J.: Green function application so numerical solving b o u n d a r y
problems
JANOVSKA D., MAREK I.: About the monotonicity of temple quotiens
KAFKA J.: One aspect of the discretization of Maxwell's equations
KAROLYI K.: Parameter estimation in problents of chemical reaction kinetics
KHUSAINOV D.: Ispoljzovanie vtorogo metoda Lyapunova optimizatsii kriteriev
kachestva funktsionirovaniya dinamicheskikh sistem
KOSTOVA T.: Qualitative behavior of the solution of a class of equations
generalizing Michaelis-Menten kinetics
KRBEC M.: Maximal operators and imbedding theorems
KUNCHEV O. I.: Some extremal problems for high order elliptic equations
KUTEV N.: Fully nonlinear, nonuniformly elliptic equations
LAPTINSKIi V. N.: Ob odnom metode konstruktivnogo analiza periodicheskikh
reshenii differentsialjnykh uravnenii
 XIV

LIPPOLD G.: Error estimation and adaptive refinement in finite element methods
LITEWSKA K.: Some applications of the finite elements methods to the system
of differential equations
MEGAHED F., HAMAD G. D., SALEM Sh.: On some integral inequalities
in n-independent variables
MYJAK J.: On the set of solutions of a diff. inclusion
NAZAROV V.: Gladkost" reshenii obyknovennogo differentsialjnogo uravneniya
s otklonyayushchimsya argumentom v prostranstvakh Rum"e
NGUEN DONG A.: Issledovanie vliyaniya razlichnykh periodicheskikh
i sluchainykh vozbuzhdenii na sistemu Van-Der-Polya
PAVLIKOVA E.: Higher monotonicity properties of zeros of a third order
differential equation
PAVLOV V. A., NEVIDOMSKII A. I.: Reshenie matrichnogo uravneniya Rikkati
ob odnoi zadache teorii optimaljnogo upravleniya
PAIVARINTA L.: The uniqueness of the one dimensional inverse problem
PEIZ~R J.: An algorithm for solving the multi-point boundary value problems
for ODE
PERINOVA V.: Fokker-Planck equation for free-electron laser
PETROV I.: An inverse problem for Maxwell equations
POPENDA J.: On the discrete generalizations of Gronwall's inequality
RETI P.: Geometrical methods in chemical kinetics
SCHIMMING R.: Laplace-Lie differential operators with a logarithm-free
elementary solution
TABISZ K.: Asymptotic behavior on solutions free b o u n d a r y problem
T~tBOAS P.: Periodic solutions of a forced Lotka-Volterra equation
WAKULICZ A.: Convergence of a class of differential inclusion approximations
W~ERBOWSKI J.: Asymptotic and oscillatory behavior of solutions of differential
inequalities generated by retarded and advanced arguments
ZITNAN P.: Lower bounds for the eigenvalues of the equation Au = Bu
by residual defect method
The papers of the following authors were also presented at the Poster Session:
BOGDANOV R., BOUZNASKI E., VASSILEVSKI P.

HI. P A P E R S P R E S E N T E D IN THE FORM OF E N L A R G E D A B S T R A C T S

ANIKb~AESEI G.: Optimal synthesis for a class of nonlinear control problems

ANTONCHIK V.: Odno obobshechenie priznaka ustoichivosti Kh. Massera dlya
nepravi~nykh sistem
ASTROVSKII A. I.: Differentsialjnaya upravlyaemost" lineinykh nestatsionarnykh
sistem v klasse funktsii Chebyshcheva
BOEV T.: Uniqueness and singularities of solutions of linear operators
and applications
BORZYMOWSKI A.: A Goursat problem for a polyvibrating equation
of Di Mangeron
CHAUVEHEID P.: Green functions for some over-determined b o u n d a r y value
problems
CHOCHOLATY P.: Finite element simulation of an axisymetric acoustic
transmission system
KHUSAINOV D., YUNJKOVA E., IVOKHIN E., ZHUiKOVA A.: Ispoljzovanie vtorogo
metoda Lyapunova v optimizatsii kriteriev kachestva funktsirovaniya
dinamicheskikh sistem
TSEREMENSKIi A.: Stabilizatsiya v chastotnoi oblasti
 ×Y

DIBLIK J.: On conditional stability of solutions of linear systems

DOKTOR P.: On uniqueness periodic solution of a certain parabolic equation
DOLEZAL J.: New aspects of computer-aided design of dynamical s y s t e m s
FEDORENKO L.: Ob ustoichiw)sti resheni[ stokhasticheskikh differentsialjnykh
uravnenii parabolicheskogo tipa
FREILING G.: Irregular b o u n d a r y value problems
GAISHUN I. V.: Spektraljnye kriterii eksponentsialjn0i dikhotomii dlya uravnenii
v polnykh proizvodnykh
GONCERZEWICZ J.: On a b o u n d a r y value problem with radial s y m m e t r y
for the porous medium equation
GOROKHOVIK S. YA.: Dostatochnye usloviya lokaljno~ upravlyaemosti
nelineinykh sistem
GOROWSKI J.: On the oscillatory properties of solutions of certain elliptic
equation
GROGER K.: Equations modeling semiconductor devices with high carrier
densities
HACIA L.:Approximate solutions of integral equations of the mixed type
HACIK M.: A note to a certain p r o p e r t y of Bessel functions
HAVARNEANU T.: On an operatorial equation of hereditary type
HYB W.: On the s p e c t r u m of flow on the two dimensional torus
IGNATYEV V. N., ZADORIN A. I.: A finite difference method on nonuniform mesh
for a singular perturbation problem
INVERNIZZI G. C.: Periodic solutions of forced oscillators at resonance
IONESCU I. R., SOFONEA M.: Existence stability and large time behaviour
of the solution for a nonlinear viscoelastic problem
KAFKA J.: One aspect of the discretization of Maxwell's equations
KALENYUK P. I., BARANETSKIi YA. E.: Predstavtenie resheni[ nekotorykh
klassov k r a e v y k h i nachaljnokrae~Tkh zadach dlya lineinykh uravnenH
s chastnymi proizvodnymi
KALININ A. I., ROMANYUK G. A.: Optimizaksiya lineinykh v o z m u s h c h e n n y k h
sistem na baze o p o r n y k h i asimptoticheskikh metodov
KAPANADZE D.: O plotnosti elektricheskogo zaryada na poverkhnosti
p r o v o d y a s h c h e g o parallelepipeda
KARLSSON T.: Wiener's criterion and obstacle problems for vector valued
functions
KLIC A.: Bifurcations in symmetric systems
KOVRIGIN A. B.: Filjtr Kahnana s vyrozhdennymi s h y m a m i v nablyudeniyakh
KUBEN J.: Time-optimal control of two-dimensional s y s t e m s
KUBIACZYK I., RZEPECKt B.: Existence theorem for ordinary differential
equations
KVEDARAS B.: Application of Laplace transformation method to the solution
of a strongly degenerate elliptic equation
KWAPISZ M.: An extension of Bielecki's method of proving of global existence
and uniqueness results for functional equations
LASKIN M. B.: Obosnovanie skhodimosti m e t o d a Brauna dlya vypuklo-vognutykh
funktsii s p o m o s h c h y u funktsii L y a p u n o v a
LIZANA M.: Bounded, almost-periodic and periodic solutions of certain singularly
perturbed s y s t e m s with delay
LAITOCH M.: On central dispersions of the first kind and the theory of linear
difference quations
LUNGU N., MURESAN M.: On the n u m b e r of small-amplitude limit cycles
of certain systems of differential equations
 XV~

MALEC M.: Estimations of the measure of noncompactness and an existence

theorem
MERENKOV YU. N.: Kriterii ustoichivosti ura dlya funktsionaljno-differentsialj-
nykh uravnenii
MIRICA S.: Marginal characteristics solutions for Hamilton-Jacobi equations
MORAV(~iK J.: Globaljnaya ekvivalentnost" i lineinye differentsialjnye
uravneniya tret"ego poryadka vse resheniya kotorykh stremyatsya k nulyu
MORCHALO J.: Asymptotic behaviour of the solutions of differential-difference
equations
NADZIEJA T.: Shadowing temma for family ot s-trajectories
NAZMUTDINOV A. T., MD~ARISOV I. KH.: Otsenka verkhnego chisla osovykh
tochek vtoroi gruppy
NKASHAMA M. N., IANNACCI R.: Periodic solutions of second order
delay-differential systems
OKRASINSKI W.: On asymptotic solutions of some nonlinear problems
POTRA T.: Finite element of spline type for elliptic partial differential systems
PTASHNIK B. I., BERNIK V. I.: Zadacha tipa Dirikhle dlya differentsialjnykh
uravnenH v chastnykh pro izvodnykh sostavnogo tipa
RAGAB A. A., OWAIDY H. EL, ZAGHROUT A. A. S.: On oscillations of nonlinear
differential equations
RASVAN V.: Stability of a integro-differential system occurring in nuclear
reactor dynamics
RIZUN V. I.: Metod vspomogateljnykh funktsii i ego primereniya
RUDYKH G. A.: Svoistva integraljno[ krivoi neavtonomnoi sistemy
differentsialjnykh uravnenii
RZEPECKI B.: On bounded solutions of a linear differential equation with
a nonlinear perturbation in the case of Banach spaces
SAMOILENKO A. M., BORISENKO A. D., BORISENKO S. D.: Limit behaviour of the
solution of the Cauchy problem for parabolic equations with coefficients
depex}ding on parameter
SIMERSKA C.: Generalized L-splines as a solution of n-point b o u n d a r y value
problem
SIUDUT S.: Some remarks on the singular integrals on the line group
SKOROBOGAT"KO V. YA.: Svyaz" obratnoi zadachi elektrorazvedki
s mnogotochechnoi zadachei dlya obyknovennogo differentsialjnogo
uravneniya
SOSULSKI W.: Generic properties of generalized differential equations
of hyperbolic type
SPIGLER R.: Numerical treatment of certain parabolic partial differential
equations
SZMANDA B.: Oscillation of solutions of higher order difference equations
TRYHUK V.: The contribution to a linear differential delay equation of the first
order
VASSILEVSKI P. S.: Numerical solution of Poisson's equation on regions
partitioned into substructures
VERNESCU B.: Homogenization of a transmission problem in porous media flow
VORNICESCU N.: Existence of optimal control without convexity
WIEGNER M.: On the asymptotic behaviour of solutions of nonlinear parabolic
equations
ZACHARIAS K., GAJEWSKI H.: On a mathematical model of polymerization
by particle growth and coalescence
 XVII

LIST OF
PARTICIPANTS

page
AUSTRALIA Vulchanov N. Naumann J.
Trudinger S. 251 Proessdorf S.
CANADA Riedrich P.
AUSTRIA Arscott F. (427) Roos H. G.
Markowich W. Atkinson F.V. Rumpel H.
Stetter H. J. 85 Cree G.C. S~indig A. M.
Lalli B.S. Schimming R.
Muldoon M.E. 155 Schmidt J. W. 333
BELGIUM Muldowney J. Schneider R.
Chauveheid P. Sinclair R. Schneider K. R.
Habets P, Schultze B. W.
Mawhin J. 49 CHINA Strehmel K.
Vanderbauwhede A. Chang Kung-Ching 203 Triebel H. 95
Unger S.
BRAZIL Hart H. Weiner R.
Taboas P. Zacharias K.
EGYPT
BULGARIA Hamad G.D. GREAT BRITAIN
Angelov V. Everitt W.
FINLAND Whiteman J, R. 345
Angelova D.
Mustonen V.
Angelova N.
Bainov D. HUNGARY
Bouznaski E. FRANCE Balla K.
Khekimova M. Nedelec J. C. 321 Baranyi J.
Kostova-Vassilevska T. B~da P.
Kulev G. GDR Bihari I.
Kunchev O. Beckert H. Bodocs L.
Kutev N. Elschner J. Elbert A.
Lazarov R. FrShner M. Farago I.
Manolov S. Gajewski H. 209 Farkas M.
Petkov K. GrSger K. Farzan R.
Petrov I. Grund F. Fazekas (435)
Petrov K. M. Heinrich B. Feny5 I.
Popov V. K16tzler R. (451) Fofana M.
Ronkov A. Kretschmar H. Frivaldszky S.
Shopolov N. Lippold K. Garay B. M.
Tersian S. Malonek H. Gergo L.
Vassilevski P. Mfiller E. Gruber T.
 kVITI

Hatvani L. POLAND Potra T.

Hegedfis J. Augustynowicz A. Rasvan V.
Karolyi K. Bartuzel G. Sofonea M.
Krisztin T. Bobrowski D. Stavre R.
Moson P. Bojarski B. Tiba D.
Reti P. Borzymovski A. Varsan C.
Stephan G. Choczewski B. Vernescu B.
Terjeki J. Desperat T. Vornicescu N.
Dlotko Tad,
IRAN Dlotko Tom. SINGAPORE
Mamourian A. Folt~nsk~t I. Agarwal R. P. 267
Mehri B. Goncerzewicz J. Chow Y. M.
Gorowski J.
Grysa K. SPAIN
IRAQ Hacia L,
A1-Faiz M. Vega C.
Hyb W.
Jankowski J.
ITALY Jedryka T. M, SWEDEN
Biroli M. Kamont Z. Daho K.
Bresquar A. M. Kisielewicz M. Hedberg L. I.
Brezzi F. 285 Kubiaczik I. Karlsson T.
Conti R. Litewska M. Szulkin A.
Gatteschi A. Matkowski J. Thome~ V. 339
Gatteschi L. Mikolajski J.
Giaquinta C.G. 215 Morchalo J. SWITZERLAND
Invernizzi C. G. Moszner Z. Desloux J. 295
Invernizzi C. S. Muszynski B. Schwarz H. R.
Laforgia A. Myjak Z.
Magnaghi-Delfino P. Nadzieja T. USA
Marchi V. Okrasinski W. Bebernes J. 193
Nkhasama M. Olech Cz. Burt~)n T. 115
Omari P. Popenda J. Friedman A. 17
Paparoni F. Reginska T. Graef J. R.
Salvi S. Rzepecki B. Haddock M.
Torelli A. Skierczynski B. Kreith K. 141
Valli A. 259 Sokolowski J. Lewis R.
Zanolin F. Sosulski W. Sussmann H.
Stankiewicz R. Sell G. R.
Szmanda B. Trench W. F. 181
JAPAN Tabisz K.
Kusano T. (465) Tabor J.
Narazaki T. Turo J. USSR
Otani M. Wakulicz A. Amiraltyev G.
Werbowski J. Berkovic L.
JUGOSLAVIA Wyrwinska A. Bogdanov R.
Curgus B. Chanturia T. A. (431)
Jovanovic B. Cherkas L.
RUMANIA Chusainov D. Ja.
Vrdoljak B.
Aniculaesei G. Gorochowik S.
Ionescu I. Gudovic N. N.
NETHERLANDS Lungu N. Koshelev A. I. (459)
Axelsson O. 275 Morozan T. Kurzhanski A. B.
Verhulst F. Muresan M. Lamzyuk V. D.
 XIX

Mamedov Ja. D. Schmidt D. Fut~k J.

Maslennikova V. Simader Ch. 24:3 Galajda P.
Mazya V. G. 235 Speck F. O. Gregug M. 129
Mironenko V. Volkmann P. Gregu~ M.
Nazarov V. Wiegner P. Groschaftov~ Z.
Pavlov V. A. Zacharias W. Gruskovfi V.
Pliss V. Hfieik M.
Shkil J. Halickgt M.
Vqjtenko S. CZECHOSLOVAKIA Haslinger J. (445)
A n d r e s J. Hage~k A.
VENEZUELA Barnovsk~ M. Hejda P.
Lizana P. M. Bartgk J. H e r r m a n n L.
Bartofiovgt J. Hlavfi~ek I.
Bartu~ek M. 109 Holafiov~ V.
VIETNAM
Barvinek E. Holodniok M.
Nguen Dong An Blaheta R. Horgtk P.
Vinh D. Belohorec S. Horov~ I.
Blagko R. H r i c i ~ k o v f i D.
WEST GERMANY Bock I. Hfebi~ek J.
A l b r e c h t J. 303 Boh~e Z. Huga A.
A l b r e c h t S. Borfivka O. Hu~a A.
Antes H. Brilla I. Chocholat:~ P.
Bticker Ch. Brilla J. 197 Janovsk~ D.
Byrne C. M. B r u n o v s k 9 P. Janovsk:~ V. 309
Collatz L. 367 B u r d a P. J a r n i k J.
Dziuk G. Cadek M. Jarog J.
E b e r h a r d W. Cermfik L. 291 Jarugek J.
F i e d l e r B. 123 Cerfianovfi V. J o h n O.
Freiling G. DaHk J. Ka~ur J. 23
Garus R. Dan~eek J. Kafka J.
G6pfert A. Diblik J. Kalas J.
Gorenflo N. Dolansk:~ P. Kafiovsk:~ V.
Gorenflo R. Dole~al J. Kart~k K.
Hahn H. G. Dole2al V. Klie A.
Hilsmann J. Doktor P. Kmef T.
Hofmann J. Dogl~ Z. Kodn/tr R. 373
Heuber H. Dogl~ O. Kohout V.
Kawohl B. Drgtbek P. Kojeckgt J.
Kettler M. Dr~palik V. Kolfifov~i J.
K o n o p k a M. Dula J. Kolom:~~J.
Kuban A. Dutko M. Komornik J.
Ktipper M. Eliag J. Kop~iekov~ M.
Ktipper T. F~zikov~ V. Kottas J.
Kupzik E. Feireisl E. Kozet K.
Martensen E. 387 F e i s t a u e r M. 3 Krbec M.
Meister E. 393 F i d l e r J. Krbec P.
M611er M. Fila M. Krej~i P. 221
Miiller-Retkowski Fialka M. Krupkov/t O.
Niedack-Nad M. Filo J. Kfi~ek M. 315
Ostermann-Emden F i s c h e r A. Kuben J.
Ptakties H. F o r t J. Kubieek M. 379
Rother W. Francfl J. Ku~era M. 227
Schaaf R. Fuller J. Kufner A. 35
Kulcsar ~. Rachfinkovgt I. Zl~mal M.
Kurzweil J. 149 R~tkosnik J. 2gt~ik T.
Laitoch M. Raudensk~ M. 2eni~ek A. 353
Laitochovfi J. Rektorys K. 71 Zidek J.
Lovicar V. Rosa V. Zitfian P.
Lovi~ek J. Roubi~ek T.
Luke~ J. Rovder J.
Malec M. Rusngtk M.
Mal:~ J, Rfi~V2kovfi H.
Marek I. Sadovsk:~ Z. 415
Marek M. Schneider Z.
Marko L. Schwabik ~.
Maru~iak P. Segeth K.
Maslowski R. Seiler J.
Medved' M. Seman J.
MihalikovgL B. Simersk~i C.
Milota J. Smitalov~ K. 421
Mikunda J. Sobotka Z.
Mika S. Stan6k S.
Moll I. Star~ J. 247
Morav6ik J. Stragkraba I.
Ne6as J. 399 Santavfi S.
Nedoma J. Seda V. 161
Nedoma Jos. ~imko E.
Netuka I. Simga J.
Neuman F. 59 Sitka J.
Neustupa J. 409 Si~ol~k F.
Ni~fiansk~ J. ~i~ol~kov~ D.
Nosfirov~ A. Solt6s P.
Novfikovfi E. Solt6s V.
Novotn~ A. ~t6dr~ M.
Ohriska J. Svarc R.
Ol~th R. Svec A.
Orgulik V. ~vec M. 171
Osi~ka J. ~ver~k V.
Pato~ka Z. Taufer J.
Pavlikov~ E. Tryhuk V.
Pavlu~ M. Tur~ok S.
Pelant J. Tuzar A.
Pekfir J. Tvrd~ M. 187
Pe~inovfi V. U~nikovfi D.
Petru~ka I. Ullrich M.
Petzeltov~ H. Valko V.
Pir~ V. Vencko J.
Pokornfi O. Ve~i~ik M.
Polfi~ek J. Viszus E.
Polcar P. VitAsek E.
Prfiger M. V o r ~ e k J.
Pudei V. Vosmansk~ J.
Pultar M. Vrko~ I.
Prima B. V ~ e ~ l E.
Quittner P. Zezula R.
R~b M. Zimka R.
MATHEMATICAL A N D NUMERICAl, STUDY
OF NONLINEAR PROBLEMS FLUID
MECHANICS
M. FEISTAUER
Faculty of Mathematics and Physics, Charles University
Malostranskg ndm. 25, 118 O0 Prague 1, Czechoslovakia

INTRODUCTION

The study of flew problems in their generality is very diffi-

cult since real flows are three-dimensional, nonstationary, viscous
with large Reynolds numbers, rotational, turbulent, sometimes also
more-fase and in regions with a complicated geometry. Therefore, we
use simplified, usually two-dimensional and non-viscous models. (The
effects of viscosity are taken into account additionally on the basis
of the boundary layer theory.)
Here we give a s u ~ a y of results obtained in the study of boun-
dary value problems describing twP,dimensiQnal~ non-viscous~ statio-
nary or quasistationa~f incompressible or subsonic compressible flows
with the use of a stream function r

I. STREAM FUNCTION FOP~v[ULAT!ON OF THE PROBLEB,I

On the basis of a detailed theoretical and numerical analysis

of various types of flow fields (plane or axially symmetric 'channel
flow, flow past an isolated profile, cascade flow etc.) a unified
conception for the stream function-finite element solution of flow
problems was worked out.
We start from the following assumptigns:
I) The domain ~ C R 2 filled by the fluid is bounded with a piece-
wise smooth, Lipschitz-continuous boundary 8~. (Usually ~ has
the form of a curved channel with inserted profiles.)

2) 5~ = rD N j=1 j=1 P P, , r~
 +
are arcs or simple closed curves, rp, rp are piecewise linear ares,
rp is obtained by translating Fp in a given direction by a given
distance. This translation is represented by a one-to-one mapping
Zp: -pr-°nt°~*p.
-+ F D and FN are formed by finite numbers of arcs. Of
course, all these arcs and simple closed curves are mutually disjoint,
except neighbouring arcs that have common initial or terminal points.
We assume that rD # ~.
3) The differential equation has the form
z (b(x,u,(Vu)2)Ux)x. : f(x,u,(Vu) 2) in ~. (1.1
i:~ l 1
4) We admit the following boundary/ conditions:
u|F D = u D (Dirichlet condition), (1.2
~u
b(.,u,(Vu)2)~-~}r N : -¢N (Neumann condition), (I.3

u(Zp(X)) : u(x) + Q, (1.4,a)

8u
(b(. ,u,(Vu) 2)%-~)(ZP(x)) 8u) (x)
= -(b(. ,u,(Vu)2)-~--n (1.4 b)

X erp (periodicity conditions),

Ulrl3 = + = eonst,

I b(. ,u,(Vu)2 )6-~ndS

~u = -v~ (velocity circulation (1.5,b)
conditions)
j : I,...,KI,
Ulr~ = uS + : const, (1.6,a)
~u
~yn(Zj : 0, zj ~ r~, (Kutta-Joukowski
conditions)
trailing (1.6,b)

j = I,...,K T.
• . j

UD, UN, u~, u~ are given f u n c t i o n s , Q, v~ - given c o n s t a n t s , zj E r T

are given trailing stagnation points, u is an unknown function and
q~, q~ are unknown constants.
The contact of some boundary conditions is prohibited e.g. (1.2)
and (].5,a-b). It is also necessary to consider the consistency of
some types of these conditions as e.g. (1.2) and (1.4). For concrete
examples see [ 7 - 10 ] .

2. THE PROBLB{ WITHOU~ T R A I L I N G CONDITIONS

Since the problem (1.1) - (1.5) without trailing conditions has

better properties from the mathematical point of view than the general
problem (1.1) - (1.6), we shall treat these problems separately.
 2.1. Variational f o r m u l a t i o n of the problem (1.1) - (1.5). We
shall seek a w e a k s o l u t i o n in the w e l l - k n o w n S o b o l e v space H I ( o ) :
= W~( ~ ). We define the set
V = {ve C (~); vlr D O, v(Zp(X)) = v(x), x e rp,

v Ir~ = const} (2.1)

and the space
V = Iv E HI( ~)~ vlr D = O, v(Zp(X)) = v(x), X e r p ,

v I~ = const (in the sense of traces on a~)]. (2.2)

The validity of the following a s s e r t i o n is important:

The set V is dense in V, i.e.

HI
(~) = v. (2.3)
It is not easy to prove this. For a cascade flow p r o b l e m see [10]
Further, let u *C HI(~ ) satisfy
a) u*IrD : UD, b) u~Ir~ : u~, (2.4)
C) u*(Zp(X)) = ~'~(x) + Q, x e rp.

Very• often the existence of this u* follows from the fact that u D and
u~ are indefinite integrals of functions from L2(F D) and L2(r~) ,
r e s p e c t i v e l y (of. [20]).
Under the above notation the problem (1.1) - (1.5) is (formally)
equivalent to the following variational formulation: Find u such that
a) u e HI(~ ), b) u - u* e V, (2.5)
c) a(u,v) = m(v) ¥ v e V,
where
a(u,v) = f(b(.,u,(vu)2)vu.vv + f(.,u,(Vu)2)v)dx (2.6,a

Yu,vE H I ( ~ ),

N
m(v) : - Z v~vlrJ - /¢NVdS, ve V. (2.6,b
j:1 rN
2.2. Finite element discretization. Let [~ be a p p r o x i m a t e d by
a p o l y g o n a l domain ~h and let T h be a t r i a n g u l a t i o n of ~h with
usual properties. We denote by oh = {PI,...,PN] the set of all ver-

tices of Th" Let the common points of' rZ, r N etc. and also the
points of a~, where the c o n d i t i o n of smoothness of ah is not satis-
fied, belong to ~. Moreover, let ~h N a~ h C a~ ~ and ~ e r p m ~ ~l

Zp (pj ) E r +p N ~l" Hence, the sets r D' F N etc " are approxima-
ted by arcs or curves rDh , FNh etc. c a~ h in a natural way.
An approximate solution is sought in the space of linear confor-
ming t r i a n g u l a r elements
 Wh = { Vh e C ( K h ) ; VhiT is affine WT e Th} (2.7)
The discrete problem is written down quite analogously as the conti-
nuous problem (2.5,a-c): Find u h such that
a) u h e Why b) u h - ~ j ~ e Vh (2.8)

c) ah(Uh,V h) : mh(v h) VVhE Vh,

whe r e
V h = { v h c Wh~ VhlFDh : 0, Vh(ZP(Pj)) : Vh(Pj) , Pj E Sh 0 Fp,
Vhlfldh : eonst } (2.9)

The function u~ E W h has the properties

a) Uh(P i) = UD(Pi) , Pi E ~h a FD,

b) Uh(P i) : u J ( P i ) ' :Pi C ~h n F J, (2.~0)
c) Uh(ZP(Pi)) : Uh(Pi) + Q, Pi E ~ h n r p and
plays the same role as u* in the continuous problem. Further,

ah(Uh,Vh! = f (b(.,Uh~(VUh)2)VUh.VVh+ f(.,Uh,(VUh)2)Vh)dX

~h (2.11 ,a)
KI
mh(Vh) = - ~ ¥~hlFJlh / C~hVhdS. (2.11 ~b)
j:1 r
Nh

Usually, the integrals in (2.11,a-b) are evaluated by c o n v e n i e n t

int
numerical quadratures. Then, instead o f Uh, a h a n d mh we h a v e Uh,
int int
ah and m h in (2.8,a-c).
The problem (2.8,a-c) leads to a system of al6ebraic equations
A(~)[ : F(~). (2.~2)
Here ~ = (ul,...,um) T is a vector with components defining the appro-
ximate solution, A(Z-) is an nxn ( n < N ) s y ~ e t r i c positive definite
matrix for all ~ c R and F:R -~-R .
~i TI n
Now let us introduce the properties of the functions b and f:
I ) b and f depend on x E [ , u c R 1 , ~ _> O ( n := (vu)2i '.
2) b, f and their derivatives ab/$xi, ab/~u, ab/an , 9f/[~xi etc. are
continuous and bounded.
3) b % b 1 > O, b] = const, ~b/~n > O.

4) ab (x, u,s2)s 2,
~-~n I ~ (x,u s 2 [)s _ onst ~ E ~ ~ ~ i

5) b(Zp(x),u+Q, rl ) = b(x,u,n ) Vx E Fp, u ~ RI , n ~ 0.

f satisfies the secortd inequality in 4) and the assumption 5 ).
2.3. The solvability of the problem (2.5,a-c) is a consequence
of the monotone and p s e u d o m o n o t o n e operator theory ([ 19, 221 ). If
the flow is irro%ational (b = b(x, q ), f = 0), then the form a(u,v)
satisfies the c o n d i t i o n of strong monotony and the s o l u t i o n is unique.
These results for various types of flows are contained in [i, 4, 5,
10, 15].
2.4. The study of the discrete probl...e..m. Its solvability easily
follows from the Brower's fixed point theorem and the properties of
b and f (cf. [19, 13] ). M u c h more complex is the q u e s t i o n on the con-
v e r g e n c e of the finite element method, since by Strang ([23]) we have
c o m m i t e d three variational crimes (approx:Jmation of 0 by a polygonal
domain; W h ~ HI( ~ ), V h ~ V; numerical integration), the p r o b l e m is
nonlinear and boundary conditions are n o n h o m o g e n e o u s and nonstandard.
We shall consider numerical quadratures of p r e c i s i o n d=1 with
nonnegative coefficients. Let ~ N and ~ be piecewise of the class C 2.
Let us c o n s i d e r a regular system of t r i a n g u l a t i o n s {Th}h~(O,ho)

of ~ h (ho > 0 is suf~zclently

~ " small) and study the b e h a v i o u r of
Uh, if h - 0+.
By I!.II1, ~ h we denote the usual norm in HI( O h) and put

IVll,a h : (/ (v v)2dx) I/2 (2.13)

h
It is important that
llvhlll, O h <
- c IVhl i, ~ h ~ VhE Vh v h e (0,ho) (2.14)

with a constant c > 0 independent of v h and h (see [ 13] or [ 24] ).

By [21], the s o l u t i o n u of the continuous p r o b l e m and the func-
tion u '~ posses the C a l d e r o n extensions from ~ to a domain ~ such
that ~ ' ~h C ~ vh 6 (O,ho) and u, ~ E H I(~ ).
Further, let us assume that
llu"~ u"ll * 0 if h - 0+. (2.15)
n 1,0 h
In some cases (of. e.g. [ 10]) u "~{ e v1+aw2 ( ~)' a > O, and we can put
tf~ : rh~{ (= the Lagrange interpolation of u ~ ). Then, since
II u* - rhff~ tl 1, £ h _< ch e tl ~11,vv21+ @ ( ~ ) (2.16)

(with c independent of u ~ and h) we have (2.15).

First let us c o n s i d e r an irrotational flow. The study of the
convergence u h ~ u, if h - 0+, is based on the f o l l o w i n g results.
2.4.1. Theorem. There exist ~ , K > 0 such that

ah(u I ,Ul-U 2) - ah(u 2 Ul-U 2) > ~ I u 1-u~ 2 (2.17)

' I,[2 h ,

lah(Ul,V) - ah(u2,v)l -< K llu1-u2111, ~ h II vii1, ~ h (2.18)

v Ul,U2~V E HI( ~ h ), ~h 6 (O,ho)
and
 int, 2 (2.17,~)
a hint,[u I ,u I -u2) - ah ~u2'ul-u2) > ~lUl-U211' ~h

int.ku I , v )
I ah - a hint(u2~v) I -< .,K IIUl-U2Ul , ~h~Vlll , ~ h (2"18~'~)

vu 1 , u 2 , v e Wh , v h e (O,ho)
Proof follows easily from the properties of the functions b and
f, the Mean Value Theorem and [3, Theorem 4.1.5]
Now let us introduce abstract error estimates.
2.4.2. Theorem. There exist constants At, A2, A 3 independent of
h such that
Ilu-uhlll ~h -< A I winf
h~ IIU-Whll + (2.19)
, u~+V h I, ~ h

+ A 2 sup (lah(U,Vh)-mh(vh)i/ llvh[l )

Vh~ Vh 1,~h
and
II int ,t _< ( 2.20 )
Uh-Uh "I , ~ h

A3 sup (I a h ( U h , V h ) - a ~ n t ( u h , v h) 1 + l m h ( V h ) - m l n t ( v h ) l )/llvhll 1 2 .
v h ~ Vh ' h
Proof is a consequence o f Theorem 2 . 4 . 1 .
2.4.3 . Theorem. Let u , u ~ E H 2 ( ~ ). Then ~U-Uhnl, ~ h : O(h).

Proof. We apply the technique common in linear problems (cf.

[ 3] ) based on estimates (2.19) and (2.16) with ~ : I and a similar
estimate for u. This, the use of" Green's theorem and the fact that
meas(( f~- ~h ) u ( ~ h - ~ )) -< ch2 give the result. (Another approach
avoiding the use of Green's theorem is used in [ 18] .)
2.4.4. Theorem. Let u E H I ( ~ ) and let (2.15) be satisfied. Then
lira 11U_Uhll = O.
h~O+ 1, ~ h
Proof. From (2.3) and (2.15) we get
lira inf '~+V ~u-whll I ~ : O.
h - O+ w h~ u h h ' h
The convergence of the second term in (2.19) to zero is proved by
introducing convenient modifications ~hE V of v h ~ Vh" Hence,
a(u,~ h) : m(~h).
This, the estimates of Vh-~ h derived in [24) and the estimates of
a(u,~h)-ah(u,v h) and mh(Vh)-m(~h) imply the desired result.
2.4.5. Theorem. If we use numerical integration of precision
d=1 with nonnegative coefficients and 8~ ' ~ N are pieeewise of
class C 2, then RUh-u~ntlll, ~ h = O(h).
 Proof follows from the estimate (2.20), [ 3, T h e o r e m 4.1.5]
and the boundedness of {Uh}hE(O,ho).
2.4.6. Remark. The convergence of the finite element sglution
w i t h numerical i n t e g r a t i o n applied to a n o n l i n e a r elliptic problem
was proved in [13]. A more complex analysis will be given in [18].
For general rotational flows, instead of strong monotony, we
have p s e u d o m o n o t o n y only. T h e n by the a p p l i c a t i o n of methods from
the p s e u d o m o n o t o n e operator theory ([ 19, 22]) we get the f o l l o w i n g
result:
2.4.7. Theorem. Let ~ be a polygonal domain and let the con-
ditions (2.3) and (2.15) be satisfied. Let the forms a and m are
evaluated by means of numerical quadratures of p r e c i s i o n d:1 w i t h
normegative coefficients. Hence, a and m are a p p r o x i m a t e d by ah:=
shint and mh:= m hint, respectively. T h e n it holds:
I) To each h e (O,h o) at least one s o l u t i o n u!h of (2.8,a-c) exists.
2) There exists~ c > O such that IlUhII1, ~ s
c for all h 6(O,ho).
3) If [Uhn]n:] is a subsequenoe of the system {Uh}he(O,ho) ,
hn ~ 0 and u h
- u weakly in H I ( Q ) for d
n - ~' , then u is a s o l u -
n
tion of the continuous p r o b l e m (2.5,a-c) and u h - u strongly in
HI(e). n
Proof. Let A : H I ( ~ ) - (HI( ~))* be the operator defined by
the r e l a t i o n
(A(u),~ = a(u,v), u, v 6 H I ( c ) . (2.21)
F r o m the properties of b and f it follows:
a) A is L i p s c h i t z - c o n t i n u o u s and bounded.
b) A satisfies the g e n e r a l i z e d property (S), i.e. it holds:
Zn, z E V , zn ~ z weakly, ~ - ~ strongly, < A ( ~ + Z n ) - A ( ~ + z ) ,
zn-z) ~ 0 ~ u*=Unn+Zn ~ u : ~ +z strongly.
The proof of the a s s e r t i o n a) follows from the properties of
the functions b and f. Let us show that also b) is valid. We assume
that Zn, z E V , zn z weakly, un u strongly, u=u~+z, Un= +z n
and
J (A(u n) - A(u ), z n - z ) ~ O.
n
If we put
I = a(Un,Un-U) - a(U,Un-U) ,
n
then
J : I + a ( u , u ~ - u , ) - a (u~, Un_d ~. )
An n
Since u" - u'*we find out that
n
a( ~'¢ ~ ~ ~
U,Un-U~ - a(Un,Un-U ) - O.
Hence, In - O.
 I0

From the d e f i n i t i o n of the form a it follows that

2 au n au) a ( U n - U )
In : I{ z ( b ( . , U n , ( V U n ) 2 ) .........b ( . , u , ( v u ) 2) ,. +
i:l ~}Xi ax i ax i

+ (f(.,Un,(VUn)2) - f(.,u,(V u)2))(Un-U)]dx .

As u n u weakly in HI( 9 ), u n u strongly in L 2 ( ~ ). The p r o p e r -

ties of f and b imply the r e l a t i o n s

In >
- -
~ I Un-UI ~ ,~ + Cn~ ~ > 03

Cn : 9]{(b( . ,u n , ( V ~ r l ) 2 ) - b(.,u,(Vu)2))Vu . V(Un - u) +

+ (f(.,Un,(~TUn)2) - f(.,u,(Vu)2))(Un - u)]dx,

c n : (h(.,(VUn)2) -b(.,u,(Vu)2))Vu . V(u n - u ) +

+ (f(.,Un,(Vun)oo2) - f ( . , u , ( V u ) 2 ) ) ( u n - u)dx,
(The s e q u e n c e {llUn-UIll,9 }n:1 is b o u n d e d . ) From this and e q u i v a l e n c e
of the norms II.111, ~ and i .I i ,n in the space V we a l r e a d y conclu-
de that u n u strongly.
Since {u } - is a b o u n d e d set and A is a b o u n d e d operator,
we c a n a s s u m e that w e U h a v e a sequence Un:=U h such that
n
hn 0, u n u weakly in H I (~) , (2.22)

A ( u n) ~ X weakly in (HI( 9 ))~ .

In v i e w of (2.15), it is e v i d e n t that u=u~'~+z, z C V and z =z ~ z

n nh
weakly.
Similarly as in [ 13] or [18]+we derive the e s t i m a t e s
la(Uh,V h) - ah(Uh,Vh)l -< chllVh}lI ,~ v v h C Vh v h e (0,ho) (2.23)

Im(v h) - m h ( v h) I -< chllvhlll,9 VVh e Vh vhe (0,ho) (2.24)

with c independent of v h and h.
Let v E V . By (2.16) Vn::r h v - v in H I ( 9 ), V n E V h . We
have n n

< A ( U n ) , V n} = m(v n) + (a(Un~V n) - a h (Un~Vn)) +

n
+ (m h (v n) - m(v n)).
n
F r o m this, (2.22~-(2.24) and (2.3) we d e r i v e the r e l a t i o n
(X,v) : re(v) vv E V. (2.25)
Further, by (2.21)-(2.25),
( A(u n) - A(u), zn - z > -~ 0.
Now, if w e use the g e n e r a l i z e d property (S) of the operator- A, we
find out that z n ~ z and thus, u n - u (strongly). As the operator
A is continuous, A(u) = lim A(u n) = X. By (2.25),

< A(u),v ) : m(v) vv e V,

which we wanted to prove.

2.4.8. Remark. Instead of Lipschitz-continuity of the operator

A it is sufficient to use its demicontinuity: "u - u strongly ~
n
A(u n) ~ A(u) weakly. The proof of the conw.~rgence of the appro-
ximate solution obtained without numerical integration is similar
(and of course more simple). The case of the problem in a nonpoly-
gonal domain ~ remains open.

3. ON THE GENERAL PROBLEM (T.I) - (1.6)

In practice the complete problem (1.1) - (1.6) is very important,

but its mathematical study is unfortunately much more difficult becau-
se of the discrete trailing conditions (1.6,b). Therefore, the results
are not so complete as in the case of the problem (1.1) - (1.5) and
we present here only a brief surway.
3.1. The solvability of the continuous problem has to be studied
in classes of classical solutions. The main too] for proving the sol-
vability are appriori estimates of solutions to linear and nonlinear
elliptic equations and the strong maximum principle. The study was
successful for incompressible irrotational and rotational flows ([6,
8]) and for irrotational compressible flows ([9]). The solvability of
the general rotational compressible flow problem re~ains open.
3.2. Finite element discretization. Let us consider a triangula-
tion Yh of the domain ~h with the properties• from 2.2. Moreover, we
assUmeyh that to each trailingzj point~ zj E~hr~ there exists a triangle ~ ~
E with vertices~ Pj = and P~ E such that the side S. :
J $J
is normal to r~. Then, if we discretize the condition (1.6,b) by its
finite-difference analogue and consider (].6,a), we derive the condi-
tions (for simplicity we assume that u~ = O)

"~ e ~h n r (3.1)
Now the discrete problem to (1.1) - (].6) is written down in
the following way: Find u h such that
a) u h E Wh, b) u h - u~ E ~ h ' (3.2)

e) ah(Uh,V h) : mh(v h) v v h E V h-
Here,
 12

V h = { Vhe Wh; vhlrDh : O, Vh(ZP(Pi)) = Vh(Pi), (3.3)

, Vh, = const, = o

~h IV h eWh; VhiFDh = O, Vh(ZP(Pi)) = Vh(Pi), (3.4)

Pi e ~h n r ~ , vh,r~h = const, vh'rShUS j = const } ,

a) u~eWh, u~(P i) = uD(Pi) , P i e ~h nrD ' (3.5)

e) u~(Zp(Pi)) = d~(P i) + Q, Pi e ~hn r~,

d) ~ I r J Thus j = O.

a h and m h are again defined by (2.11,a-b).

The problem (3.2,a-c) is equivalent to a system (2.]2). Since
Vh # ~h' the m a t r i x A(~) is not more symmetric. However, if all an$1es
of all T e Th are less then or equal to 90 °, then A(~) is an irreducib-
ly diagonally dominant m a t r i x and the system (2.12) has a solution.
Under the same assumption, w i t h the use of the discrete m a x i m u m prin-
ciple, we can prove the conversenqe of the method: if u e C 2 ( ~ ) and
the p r o b l e m is linear, then IIU-UhlJL~( e h ) ~ ch for all h e (O,ho).
For details see [ 14 ].

4. ITERATIVE S O L U T I O N OF THE D I S C ~ T E PROBLEM

It is convenient to distinguish several cases:

4.1. Irrotational incompressible flow (b = b(x), f = 0): The
system (2.12) is linear and we use the S 0 R method.
4.2. Irrotational compres@$ble flow (b = b(x,n ), f = 0): A m o n g
the methods we have tested the f o l l o w i n g iterative process occurs as
an effective one:

a) 9 ° E Rn (a convenient initial approximation) (4.1)

b) B~ k+] = Bu-k - ~ ( A ( ~ k ) ~ k - F(~k)), k k O, a > O.
The speed of the convergence depends on the choice of a (its estimate
can be obtained on the basis of the b e h a v i o u r of the f u n c t i o n b) and
of a p r e c o n d i t i o n i n g positive definite m a t r i x B.
~.3. Rotational incompressible flow (b = b(x), f = f(x,u)):
Similarly as in [ 7] we can apply a Newton relaxation method. If the
vorticity is too strong, it is better to proceed as in the following
 13

case.
4.4. Rotational compressible flow: As a s u f f i c i e n t l y robust the
method of least squares and c o n j u g a t e gradients by G l o w i n s k i et al.
appears (see[ 2]). The details w i l l be the subject m a t t e r of an in-
t e n d e d paper.

5. E X A M P L E S

As a simple test p r o b l e m we introduce a flow t h r o u g h a plane

channel. On the inlet (left side of the boundary - see Fig. I) and
outlet (right side of the boudary) we c o n s i d e r the N e u m a n n condition
~uf~n = O. On the l o w e r wall we put u = 0 and on the u p p e r wall u =
= 25. We c o n s i d e r a r o t a t i o n a l flow d e s c r i b e d by the e q u a t i o n
Au = -200 a r c t g u.
The u n i q u e n e s s of this boundary value p r o b l e m is not sure.
This p r o b l e m was successfully solved by the m e t h o d of least squ-
ares and conjugate gradients s t a r t i n g f r e m the s o l u t i o n of the corres-
ponding linear irrotational flow ~ u = 0). In Fig. I we see the
triangulation used. The i t e r a t i v e p r o c e s s was stopped after 6 conju-
gate g r a d i e n t i t e r a t i o n s , w h e n the r e s u l t i n g value of the cost fun-
ctional was ]0 -5 . F o r o n e - d i m e n s i o n a l minimization the g o l d e n - s e c t i o n
m e t h o d was applied. In Fig. 2 the c a l c u l a t e d velocity field is p l o t -
ted. It is i n t e r e s t i n g w i t h b a c k w a r d f l o w s caused by a s t r o n g v o r t i -
city.
The second example represents an i n d u s t r i a l application of the
presented theory and n u m e r i c a l methods - a result of a c a s c a d e flow
calculation (cf. [10 - 14, 16~ 17]). I n Fig. 3 w e show velocity
vectors plotted in the d o m a i n r e p r e s e n t ~ g o n e period of a cascade of
profiles.

F o r other examples see [11, 12, 17 I.

Fig. I
 14

° • • • •

Fig. 2

/- ,1 '

; /
/
," l { "
.//
I

Fig. 3
 15

REFERENCES

[1] J.Benda, M.Feistauer: Rotational subsonic flow of an ideal comp-

ressible fluid in axially symmetric channels. Acta Po!ytechnica,
7(IV,3), 1978, 95-105 (in Czech).
[2] M.O.Bristeau, R.Glowinski, J.Periaux, P.Perrier, O.Pironneau, G.
Poirier: Application of optimal control and finite element
methods to the calculation of transonic flows and incompressible
viscous flows. Rapport de Recherche no. 294 (avril ]978), LAB0-
RIA IRIA.
[3] Ph.G.Ciarlet: The Finite Element Method for Elliptic Problems.
North-Holland, Amsterdam-New York-Oxford, 1978.
[4] M.Feistauer: On two-dimensional and three-dimensional axially
symmetric flows of an ideal incompressible fluid. Apl.mat. 22
(1977), ]99-214.
[51 M.Feistauer: Mathematical study of three-dimensional axially sym-
metric stream fields of an ideal fluid. Habilitation Thesis,
Faculty of Math.and Physics, Prague, 1979 (in Czech).
[61M.Feistauer: Solution of elliptic problem with not fully speci-
fied Dirichlet boundary value conditions and its application in
~ d r o d y n a m ~ c s . Apl.mat. 24(1979), 67-74.
[7] M. Feistauer: Numerical solution of non-viscous axially symmetric
channel flows. In: Methoden und Verfahren der mathematischen
Physik, Band 24, 65-78, P.Lang-Verlag, Frankfurt am Main-Bern,
1982.
[8] M.Feistauer: Mathematical study of rotational incompressible non-
viscous flows through multiply connected domains. Apl.mat. 26
(198]) 345-364.
[91 M.Feistauer: Subsonic irrotational flows in multiply connected
domains. Math.Meth. in the Appl. Sci. 4(1982), 230-242.
[10] M.Feistauer: On irrotational flows tlhrough cascades of profiles
in a layer of variable thickness. Apl.mat. 29(1984), 423-458.
[11 ] M.Feistauer: Finite element solution of non-viscous flows in cas-
cades of blades. ZAMM 65(1985),4, TI91-T194.
[12J M.Feistauer: Mathematical and numerical study of flows through
cascades of profiles. In: Proc. of "International Conference on
Numerical Methods and Applications" held in Sophia, August 27-
September 2, 1984 (to appear).
[13] M.Feistauer: On the finite element approximation of a cascade flow
problem. Numer.Math. (to appear).
[14] M.Feistauer: Finite element s o l ~ o n of flow problems with trai-
ling conditions (to appear).
[15] M.Feistauer, J.~im~nek: Solution of subsonic axially symmetric
stream fields. Apl.mat. 20(1975), 266-279.
[]6] M.Feistauer, J.Felcman, Z . V l ~ e k : Calculation of irrotational
flows through cascades of blades in a layer of variable thickness.
Research report, ~KODA Plzen, 1983 (in Czech).
[]71M.Feistauer, J.Felcman, Z.Vla~ek: Finite element solution of flows
through cascades of profiles in a layer of variable thickness.
Apl.mat. (to appear).
 Io

[181M.Feistauer, A.~eni~ek: Finite element solution of nonlinear

elliptic problems (submitted to Numer. Math.)
[19] J.L.Lions: Quelques Methodes de R@solution des Probl@mes aux
Limites non Lin@aires. Dunod, Paris, 1969.
[20] J.Ne~as: Uber Grenzwerte von Funktionen, welche ein endliches
Dirichletsches Integral haben. Apl.mat. 5(1960)~ 202-209.
[2]] J.Ne~as: Lea M@thodes Directes en Th4orie des ~quations Ellipti-
ques. Academia, Prague, 1967.
[22] J.Ne~as: Introduction to the Theory of Nonlinear Elliptic Equati-
ons.T~ubner-Texte zur Mathematik, Band 52, Leipzig, 1983.
[23] G.Strang: Variational crimes in the finite element method. In:
The Mathematical Foundations of the Finite Element Method with
Applications to Partial Differential Equations (A.K. Aziz, Ed.),
Academic Press, New York, ]972, 689-710.
[24] A.Zeni~ek: How to avoid the use of Green's theorem in the
Ciarlets's and Raviart's theory of variational crimes (to appear).
FREE BOUNDARY PROBLEMS IN FLUID
DYNAMICS
A. FRIEDMAN
Nortkwestern University
Evanston, Illinois 60201, U.S.A.

The velocity potential of a 2-dimensional ideal incompressible

and i r r o t a t i o n a l f l u i d s a t i s f i e s A% = 0; f u r t h e r , B e r n o u l l i ' s l a w
2
IV~I + 2p : c o n s t , y i e l d s IV%t : c o n s t , o n the (free) b o u n d a r y o f t h e
fluid in c o n t a c t w i t h air. S i n c e v% is t a n g e n t i a l to the free bound-
ary, the stream function u (i.e., the h a r m o n i c conjugate o f %)
satisfies:

Au = 0 in t h e fluid
8u (I)
u = c, 7~g : ~ o n t h e f r e e b o u n d a r y

where c,X are constants. If w e take gravity into account, then i is

replaced by ~ (a > 0, g > 0) w h e r e the gravitational force is in
the upward direction.
In addition to (i) w e m u s t impose boundary conditions
3u
u : u0 or 5-~ : u I (2)

on the fixed boundary

as w e l l as a condition at infinity. For exa[~le
(i) f o r a s y m m e t r i c jet flow from a nozzle Z we have:

~ u : Q
Au: 0 u = ly a t oo,
Xh -- Q
u = 0 Figure 1

where h is t h e a s y m p r o t i c height of the free boundary as x - ~

(ii) fora symmetric cavitational flow with nose I w e have:

U = 0 ~ ~

u> 0
C
Figure 2
 18

u : y(1 + O(i)) Vu : e(l + o(i)) where e : (0,1) and o(i) ~ 0 if

2 y2

Problems such as (i), (ii) have been solved by several methods

over the last i00 years. The general procedure has been to use con-
formal mappings or the h o d o g r a p h transformation in o r d e r to r e d u c e
problems such as (i), (ii) to n o n l i n e a r integral equations (of a r a t h -
er c o m p l i c a t e d type) and then apply the Leary-Schauder fixed point
theorem; for d e t a i l s see [141123][24] and the r e f e r e n c e s in [ 121,[22] .
Another approach based on a v a r i a t i o n a l principle was developed in
[ 19] ,[ 20] .
In the last few years Aft, Caffarelli and Friedman have developed
a new variational approach to e s t a b l i s h existence of solutions for
free boundary problems of g e n e r a l ideal fluids [2-4,8,9],This w o r k has
a l s o been extended totwoflnlds (flowing ~ e - b y - s i d e ) [ ~ - 4 , ~ 9 ] ° W e shall ex-
plain theessenceofthemethodinthesimplestcase (i)(Figure I, above).
Consider the functional

J(v) : SIVv - i~ x 12dxdy, v E K

Iv<Q] X E

where
I : y = g(x), _oo < X < 0, g monotone <b : g(0>, A --(0,b)),
E : {(x,y); -~ < x < ~, 0 < y < b}, ~.: {(x,y); 0 < y < g(x),
-~< x_< 0},
R +2 = {(x,y)} x > 0, y > 0],

= ~. UR 2
+, ~ : ~ N {x > -U},

K = [v 6 H I ( R 2 ) , V = Q if y -> g(x), x < 0 or y >

_ b, x > 0;
v(x,0) = 0 if - ~ < x < ~,
v(-u,y) = h u ( y ) , a n d 0 < v S Q a.e.} ;

here h (y) is a g i v e n function monotone in y, h (0) = 0, h (g(-u)) = Q.

Consider the p r o b l e m : Find v = v in K such that
min J(v) = J(ux, ) .
veK u
U
It is e a s i l y seen that this problem has a solution. The solution is
harmonic in ~ . \ E and is a l o c a l minimizer in E of

J(v) ~ I(JVvI2 + 12 )dxdy .

X {v<Q}

Alt and Caffarelli [ i] s t u d i e d the local minimizer v of ~ a n d

proved that ~ is L i p s c h i t z continuous and that the free boundary
8 {3 < Q} N E u is l o c a l l y analytic.
 ~9

LEMMA i. The minimizer is u n i q u e .

Indeed, suppose Ul, u 2 are two m i n i m i z e r s and introduce
u~(x,y) = Ul(X - e,y) and

E
v I : u I A u2, v 2 : u~ v u2 .

Denote by Je the functional J = JX~p corresponding to the translation

x ~ x + e of ~p, K~. One verifies that

Je(u[) + J ( u 2) = J c ( v I) + J ( v 2) ,

which implies that J ( u 2) : J ( v 2 ) , i.e., u~ v u 2 is a m i n i m i z e r . Conse-

quently u[ v u 2 is s m o o t h , which implies that either u~ ~ u 2 or
u le _< u 2 e v e r y w h e r e . Since u el < u2 n e a r the boundary, we d e d u c e that
u[ < u 2 t h r o u g h o u t the d o m a i n . Taking e - 0 we g e t u I ~ u2, and
similarly u 2 S u I.
Taking u I = u 2 in the above argument we get:

~--d ul,p a 0 .

Thus the a n a l y t i c free boundary r = Fl,~ has the form

FX, p : x = fx, (y) .

N e x t we have:

LEMMA 2. fl, (y) is c o n t i n u o u s and finite if a n d o n l y if

h < y < b, w h e r e h = Q/X.

LEMMA 3. X -- fx,p(b) is c o n t i n u o u s (fx,p(b) = fx, (b + 0)).

LEMMA 4. If X is s u f f i c i e n t l y small then fx (b) < 0; if

X < Q/b and I~ - Q/bl is s m a l l enough, then fx,p(b) > 0.

From Lemmas 3, 4 we d e d u c e that there is a v a l u e X = X(u) such

that f (b) = 0, i.e., there is a " c o n t i n u o u s fit" at A. Fro this
X~p
value of X, (ux,u, FX~ u) "almost" solves the jet p r o b l e m . In o r d e r to
complete the construction of a s o l u t i o n we let p - ~ l(~) ~ X a n d
denote the limiting ux~, FX~U by u,F.

LEMMA 5. Continuous fit implies smooth fit.

More precisely the c u r v e I U F is n o t o n l y continuous at the
 20

point of d e t a c h m e n t A, b u t it is a l s o C 1 at A, a n d V u is u n i f o r m l y
C 1 in {u < Q } - n e i g h b o r h o o d of A.

F
THEOREM. There exists a unlque classical solution of the
symmetric jet p r o b l e m (i).
Existence was already outlined above~ uniqueness is p r o v e d by a
comparison argument [21].

The above procedure has been extended to t h r e e - d i m e n s i o n a l axi-

ally symmetric jets [2] , 2 - d i m e n s i o n a l asymmetric flows [ 3], to f l o w s
in a g r a v i t y field [4], to r o t a t i o n a l flows [16] a n d to c o m p r e s s i b l e
fluids [8][9] } some cavity problems are treated in [ 1 3 ] [ 1 8 ] .
Two-fluid problems are treated in [ 5 - 7 ] . Here u + a n d u- are h a r -
monic and

~u + ~u-
- ~ = ~ on the free b o u n d a r y . (3)

In a t w o - f l u i d flow in a p o r o u s media of a r e c t a n g u l a r dam, the free

boundary condition c a n be r e d u c e d to

au + ~u-
- ~ = cos(x,~) (4)

which is s i m i l a r to ( ~; in (3) I is n o t a priori given, whereas in

(4) a d e g e n e r a c y occurs at p o i n t s where cos(x,~) = 0. P r o b l e m (q) is
studied in [ i0] where existence of a s o l u t i o n is p r o v e d having a C1
free boundary.

Lemma 5 has been extended in [ ii] to q u a s i l i n e a r elliptic opera-

tors and to m o r e general boundary conditions 8u/3~ = f. T h e assertion
is t h a t e i t h e r F U £ is C 2 at A or it is o n l y in C 3/2 a n d the
curvature of r goes to ± ~ as x i 0.
Other physical flow problems lead to f r e e b o u n d a r y conditions as
above. We m e n t i o n the p r o b l e m of freezing in a c h a n n e l because of h e a t
s i n k at the o r i g i n [25]. Thus

&u : -MS in {u > 03

where 8 = Dirac measure, -u is the t e m p e r a t u r e , and

u > 0 in the ice,

~u
~v - ~ on the free boundary;

a n d M are given positive constants. Assuming that the c h a n n e l ~ is

symmetric with respect to teh y - a x i s it w a s recently proved by F r i e d -
man and Stojanovic [ 17] that the p r o b l e m has a unique solution with
free b o u n d a r y concave to the ice. This implies that if B~ c o n s i s t s of
p curves I i convex to ~ t h e n the free b o u n d a r y consists of at m o s t p
arcs ("fingers") concave to ~, e a c h connecting an adjacent pair li,

li+ 1 •

R e f e r e n c e s

|i] H.W. A l t a n d A . C a f f a r e l l i , e x i s t e n c e and r e g u l a r i t y f o r a minimum

problem w i t h f r e e boundary. J . R e i n e Angew. Math. 105 (1981>,
105-144.
[2] H.W. Aft, L.A. Caffarelli a n d A. Friedman, Axially symmetric jet
flows, Arch. Rat. Mech. Anal. 81 <~983), 97-149.
[3] H.W.AIt, L.A.Caffarelli and A. Friedman , Asymmetric j e t flows,
Comm. P u r e Appl. Math. 35 <1982), 29 - 68.
~4] H.W.AIt, L.A.Caffarelli a n d A. F r i e d m a n ~ Jet flows w i t h g r a v i t y ,
J. Reine Angew. Math. 331 (1982), 58-]03,
[5] H.W.AIt, L.A. Caffarelli a n d A. Friedman, Variational problems
w i t h two phases and t h e i r f r e e boundaries, Trans. Amer. Math.
Soc., 282 <1984), 431-461.
[6] H.W.AIt, L.A.Caffarelli and A. F r i e d m a n , J e t s w i t h two f l u i d s I:
one free boundary, I n d i a n a Univ. Math. J., 33 (1984), 213-247.
[7] H.W.AIt, L.A.Caffarelli a n d A. Friedman , Jets w i t h two f l u i d s II:
two f r e e boundaries, Indiana Univ. Math. J., 33 (1984), 367-391.
[8] H.W.AIt, L.A.Caffarelli and A. F r i e d m a n , A free boundary problem
for quasi-linear elliptic e q u a t i o n s , Ann. Sc. Norm. Sup. Pisa,
ii <1984), 1-44.
[9] H.W.AIt, L.A.Caffarelli and A. Friedman , Compressible flows of
jets and c a v i t i e s , J. Diff. Eqs., 56 ( 1 9 8 5 ) , 82-141.
[10] H.W.AIt, L.A.Caffarelli and A. Friedman , The dam problem w i t h two
fluids, Comm. P u r e Appl. M a t h . , 37 (1984), 601-646.
[Ii] H.W.Alt, L.A.Caffarelli and A. Friedman ~ Abrupt and smooth
s e p a r a t i o n of f r e e boundaries i n flow problems, Scu. Norm. Sup.
Pisa, to a p p e a r .
I12] G.D. Birkhoff a n d E.H. Zarantonello, Jets, wakes and cavities,
Academic Press, N e w York, 1957.
[13] L.A. Caffarelli a n d A. Friedman, Axially symmetric in finite
cavities, I n d i a n a Univ. Math. J. 30 (1982), 135-160.
 22

[14] R. Finn, Some t h e o r e m s on d i s c o n t i n u i t y of p l a n e f l u i d

motion, J.D'Analyse Math. 4 (1954/6), 246-291.
[15] A. Friedman, Variational principles and f r e e boundary p r o b l e m s ,
John Wiley, New York, 1982.
[16] A. Friedman, Axially symmetric cavities in rotational flows,
Comm. in P.D.E., 8 (1983), 949-997.
[17] A. Friedman and S. Stojanovic, A free boundary problem
associated with icing in a c h a n e l , to appear.
[18] A. Friedman and T.I. Vogel, Cavitational flow in a chanel with
oscillatory wall, Nonlinear Analysis, 7 (1983), 1157-1192.
[ 19] P.R. Garabedian, H. Lewy and M. Schiffer, Axially symmetric
cavitational flow, Ann. of Math. 56 (1952), 560-602.
[20] P.R.Garabedian and D.C. Spencer, Extremal methods in cavita-
tional flow, J. Rat. Mech. Anal. 1 (1952), 350-409.
[21] D. Gilbarg, Uniqueness of a x i a l l y s y m m e t r i c f l o w s w i t h f r e e
boundaries, J. Ra. Mech. Anal. 1 (1952), 309-320.
[22] D. Gilbarg, Jets and cavities, in Handbuch der Physik, vol. 9,
Springer-Verla~ New York, 1960.
[23] J. Leray, Les Probl~mes de r e p r e s e n t a t i o n conforme de H e l m h o l t z
I, II, Comm. Math. Helv. 8 (1935), 149-180; 250-263.
[24] J. Serrin, E x i s t e n c e t h e o r e m s f o r some h y d r o d y n a m i c a l f r e e
boundary p o r b l e m s , J . R a t . M e c h . A n a l . 1 ( 1 9 5 2 ) , 1 - 4 8 .
[25] L. Tiny, D.S. Ahluwalia and M.J. Miksis, S o l u t i o n s of a c l a s s
of mixed f r e e boundary p r o b l e m s , SIAM J. Appl. Math., 43 (1983),
759-775.
METHOD OF ROTHE IN EVOLUTION
EQUATIONS
J. KA(~UR
Institute of Applied Mathematics, Comenius Un,iveYsity
Mlynskd dolina, 842 15 Bratislava, Czechoslovakia

The a i m of this paper is to p r e s e n t Rothe's method (also c a l l e d method

of lines , or the m e t h o d of s e m i d i s c r e t i z a t i o n ) as an e f f i c i e n t theore-
tical tool for solving a broad scale of e v o l u t i o n problems . Using time
discretization , evolution nroblems are a p p r o x i m a t e d by c o r r e s p o n d i n g
elliptic problems by m e a n s of w h i c h an approximate solution for the o r i -
ginal evolution Droblem is c o n s t r u c t e d . By a relatively simple tech-
nique convergence of the approximate solution to the solution of the
original evolution problem is p r o v e d . Thus , unlike some abstract me-
thods for the analysis of e x i s t e n c e and u n i q u e n e s s problems for e v o l u -
tion equations , Rothe's method has a strong numerical aspect . At the
same time it g i v e s a first good insight into the structure of the solu-
tion of the investigated evolution problems .
Rothe's method introduced bv E . R o t h e in 1930 has been used and d e v e -
loped by m a n y authors , e.g.O.A. Lady{enskaja ; T.D. Ventce~ ; A.M. Ii-
jin , A.S. Kala~nikov , O.A. O l e j n i k ; ~.J. I b r a g i m o v ; P.P. Mosolov ;
K. Rektorys[10] in l i n e a r and q u a s i l i n e a r parabolic problems . Nonline-
ar a n d abstract parabolic problems has b e e n studied bv J. Ka~ur[2]-[6];
J. N e ~ a s [ 9 ] ; A.G. Kartsatos , E.M. P a r r o t t [ 7 ] , [8] etc. Linear and q u a s i -
linear hyperbolic equations has been considered by J. J e r o m e ; E. M a r -
tensen ; M. P u l t a r ; J. Ka~ur , etc. A modification of Rothe's method
(discretization in x- v a r i a b l e ) has b e e n used by V.N. Faddeeva ;
W. W a l t e r ; C. Corduneanu , etc. Time and space discretization for sol-
ving evolution problems has b e e n employed by m a n y authors ,e.g. R. G l o -
winski , J.L. Lions , R. T r ~ m o l i 4 r e s [ l ] ; M. Zl~mal[ll] ; A. Zeni~ek[12];
H.W. Aft , S. L u c k h a u s etc. , using similar technique to t h a t of R o t h e ° s
method . For the m o r e complete references we r e f e r the reader to [6]
Efficiency of R o t h e ' s method we demonstrate in s o l v i n g :
I. A n o n l i n e a r parabolic problems ;
II, V a r i a t i o n a l inequalities
III. H i g h e r order equations

I. A n o n l i n e a r parabolic problems.
 24

Let V be a reflexive B-space with its dual V* and let H be a Hil-

bert space . Let I I. 11, I. I be the norms in V , H , respectively .
We assume that V~ H is a dense set in V and H . By <f,v> we de-
note the dualitv for f E V* and v~ V . Scalar product in H we denote
by (.,.) . Let St be the interval [-q,t] for t ~ [O,T] ~ I , q ~ 0 .

An operator F : L ( S T , H ) --> L (ST,H) is a Volterra operator iff

u(s) = v(s) a.e. in St implies F(u)(t) = F(v)(t) for any te ST .

We assume A : V -- V ~ to be a coercive maximal monotone operator .

Consider the equation

(i • i) d ud (t t ) + Au(t) : f(t,F(u)(t)) a.e. t E I , u : % in So

where % : So - H is a g i v e n Lipschitz continuous function

( % ~ Lip(S o ~H ) and f E Lip(IxH - H) • Coerciveness of A we assume

in t h e form

(1.2) <Au,u> ~ I l u l l p ( l lul I)- C l I U l 2 C2 ¥ u E V

where p : R+ ~ R+ and p(s) - ~ for s ~ ~ . Lipschitz continuity

of f is e x p r e s s e d in the form

(1.3) Jf(t,v) - f(t',v')[ ~ C(It-t'I + Jt-t'I Ivl + Iv-v~I ) V t,t'e I,

V v,v'( H . We assume that F maps L i p C S T --H) into Lip(ST-~H) and

(1.4) I iF(u) - F ( v ) t f C ( S T , H ) ~ Cl I u - V l I C ( S T , H )

- du
(1.5) IF(u)(t) - F(u)(t')1 < It-t IL(I lul I C ( S T , H ) ) ( 1 + I [~J IL ~ ( S t , H ) )

V t,t'£S T , t" < t and u &Lip(ST-H ) where L:R+~R+ is cont.f•

Solving (I.I) we apply Rothe's method in the following way :

Let n be a positive integer , h = T n -I , ti = ih . Successively for
i=l,...,n we look for the solution u i - Ui,nE V~H of the elliptic
equation
u-ui_ 1
(1.6) (-~-- ,v) + <Au,v> : (f(ti,F(~i_l)(ti)),v) V v~ V{3H

where uo =~(0) and ~i_l E LiD(ST~H) is d e f i n e d by

I ~ on S0
~i-i --~ _ ~(0) on [O,h]
Ui-l'n" uj_ 1 + (t-tj_l)h-l(uj-uj_l), tj _< t < t j ÷ 1
 I for j:l,...,i-i
Ui_ 1 on [ti, T]

The existence of ui4 V~ H is a s s u r e d by the following argument. The

element Ui_l h-I + f(ti,F(~i_l)(ti)) is in H and the operator

Ah : V~ H -- V * + H defined by [AhU,V ] : ~(u,v) + <Au,v> is a c o e r c i -

ve maximal monotone . Hence theory of m o n o t o n e operators guarantee the

existence of ui . Uniqueness of ui is a consequence of strict mono-
tonicity of Ah . By means of u i ~ ui, n we construct Rothe's function
Un(t) and the corresponding step function ~n(t)

(1.7) Un(t) = ui_ 1 + ( t - t i _ l ) h - l ( u i - u i _l ), t i _ l ~ t ~ t i , i=l,...,n

(1.8) ~n(t) = ui for ti_l<t~t i , i=l,...,n , Un(0) = uo .

Then (1.6) can be rewritten in t h e form

d u (t)
n
(1.9) (~--~--uL ,v) + <A~n(t),v> = (fn(t,F(~n_l)(t)),v) V v ( V~ H

where fn(t,v) = f(ti,v) for ti_]<t~t i , i=l,...,n . First , we prove

a priori estimates for {Un] (see Lemmas 1,2,3 ) and then we take the
limit as n ~ ~ in (1.9) . We obtain

Theorem i. Let A : V -- V e be maximal monotone and let A%(0)~ H .

If (1.2) - (1.5) are satisfied then there exists the unique solution
u of (i.I) in t h e following sense : u e L (I,V) , u ~ Lip(I ~H) ,
du
d--[E L (I,H) and Au E L (I,H). Moreover , the estimate

2 < C
flu n - u~ I C ( I , H ) -

takes place where {Un] is from (1.7).

A priori estimates we obtain in t h e following way.

Lemma i. The estimate luil ~ C takes place for all n , i=l,...,n.

Proof. We put u:u i , v = hu i into (1.6) . We sum it u p for i=l,...,j.

Using (1.2) -(1.4) we estimate

J 2h
lujl 2 ~ C 1 + C 2 ~ m a x lUkl
i 1 l~k~i
and hence
 !6

J 2h
m a x lUk 12 ~ C 1 + C 2 ~ m a x lUkl
l~k~j i:l l~k~i

Thus Gronwall's Lemma implies the required result.

Lemma 2. The estimates

(~ i0) ! ~ I ~c , ItuiI[ ~ c

hold for all n , i:l,...,n .

Proof. We s u b t r a c t (1.6) for u=uj ' V=6hUj ~ u j -hu j - i from (1.6) for

u=uj_] , V=~hU j . Owing to the m o n o t o n i c i t y of A and (1.3) -(1.5) we

obtain

(l.ll) I~hUjl ~ 16hUj_iI + C(h + max I{hUkih )

isk~j

because of L e m m a 1 . From (1.6) for {:i, u:u I , V : 6 h U 1 we c o n c l u d e

l+hUll ~ c
since Uo:%(0) and AuoE H . Thus successively from (I.Ii) we o b t a i n

J
16hUjl ~ C1 + C2 i~l m a x 16hUkl h
l~k~i

which similarly as a b o v e implies 16hUi [ ~ C . Using this estimate and

(i.i0) in (1.6) for u:u i , v:u i we o b t a i n I lui[ [ ~ C because
of (1.2)
As a consequence of (i.i0) and (1.6) for u=u i we h a v e

I<Aui,v>l ~ CIvl V n , i:l,...,n .

The nrevious a priori estimates can be r e w r i t t e n in the form

dun(t)
(i.12) I ~ I ~ c , I lUn(t)IIvA H ~ c , 1<Aun(t),v>l ~ CIv I

C
(1.13) fUn(t) - Un(t')L -<CIt - t'l, fUn(t) - Un(t)l -<

du
Lemma 3. T h e r e exists an u E L (I,V), u~ Lip(I +H) with ~-~E L (I,H)
such that
2 C dUn du
I lUn-Ul IC[I,H) -< --n ' dt -~--dt in L2CI,H) and AUn(t)-~ Au(t) in V n

<also in H) ~ t ~ I .

Proof. Subtract (1.9) for n=r from (1.9) for n=s where

v = Ur(t) - Us(t) . Using <1.12) and (i.13) we estimate

 27

(I.14) ~d } U r - U s l 2 ~ C( ~1 + ~1 + Ilur_l-url
~ 2
iC(St,H) + i lus_l_usl iS(St,H)

2
+ I IUr-Us[IC(St,H)).

Integrating (1.14) over (0,t) and taking into account the e s t i m a t e

dUr(~) 2
2 c sup Id--7---I < c
]l~r-i - UrlIC(St,H) ~ ~2 T e [O,t] - ~2

?
we C o n c l u d e un ~ u in C(I,H) a n d the e s t i m a t e )lUn-U116(i,H)~
C

Hence and from (1.12) , (1.13) we obtain

~n(t)"~ u(t) , IIAUn(t)ll 5 N C , (IA~n(t) I N C ) and

<A~n(t), Un(t) - u(t)> -- 0 V t6 I

Hence maximal monotonicity of A implies AUn(t)--~ Au(t) in V ~ (more-

over in H ).
Proof of T h e o r e m i. We integrate (1.9) over (Yl,~2) and take the
limit as n - ~ . Owing to L e m m a 3 we c o n c l u d e that u is a s o l u t i o n
of (i.i) since TI,~ 2 e I are arbitrary . Uniqueness follows from
(I.I) by s t a n d a r d arguments .

Remark i. Theorem 1 holds true also when A : V ~ Ve is n o n s t a t i o n a r y

under the following assumptions :
A(t) : V ~ Ve is m a x i m a l monotone ¥ t~ I ;

A(t)u = V¢(t,u) , i.e. A(t) are p o t e n t i a l (¢: IxV ~ R )

<A(t)u,u> a
f luJ Ip(Iluil) , p(t) ~ ~ for t ~
d2
II~--~ A ( t ) u l l ~ + I l~t2 A ( t ) u [ l ~ S C 1 + C2P(I lul I)

For the p r o o f it s u f f i c e s to c o m b i n e the techniques used in t h e p r o o f

of T h e o r e m 1 with those used in [3]

Remark 2. A m o d i f i c a t i o n of T h e o r e m 1 with m-accretive operators

A(t) : D ~ V ~ V ~ (t& I) satisfying

llA(t)v - A(t')vl I S It-t~IL(l lul I)( 1 +I IA(t)vl I )

and w i t h the right hand side f = G(t,ut) (at f i x e d t the operator

G transforms the v a l u e s of ut(s) = u(t + s) , s [-q,0] into H )
satisfying LiDschitz like condition has been obtained by A.G. Kartsa-
tos and M.E. Parrott in [7]

I I. V a r i a t i o n a l inequalities .
 2S

Let ¢ be a proper (%: V --> (-~,~] , % ~ ~ ), c o n v e x and lower semi-

continuous (l.s.c.) function on V . We assume A : V ~ V e to be a
bounded maximal monotone operator . Consider the variational inequality

~du(t)
(2.1) L ~ ,v-u(t)) + <Au(t),v-u(t)> + ~(v) - ~(u(t))

(f(t,F(u)(t)),v-u(t)) V vEV~H , a.e. t& I , u : ~ on S

o

where ~ and F are the same as in S e c t i o n I. W e use the approxima-

tion scheme

(2.2) ( 6 h U i , V - U i) + <Aui,v-ui> + ~(v) - ~(u i) ~ (f(ti,F(~i_l)(ti)),v-

ui ) ~V V E V ~ H
where ~i-i is t h e same as ~n S e c t i o n I . It is e l l i p t i c variational
inequality with respect to ui provided Ul,...,ui_ 1 are known .
Coerciveness of A is assumed in t h e form : There exists v E V with
o
¢(v ) < ~ such that
o

<2.3) (<hu,U-Vo> + ¢ ( u ) ) . I lul I-I ~ ~ for I lul I - ~ .

Then , existence of ui£ V~H satisfying (2.2) is g u a r a n t e e d by the

well-known results from elliptic variational inequalities . Here ,
instead of Au E H we assume : There exists z E H such that
o o

(2.4) ( Z o , V - U o) + <AUo,V-Uo> + $(v) - $ ( u o) ~ ( f ( 0 , F ( u % ) ( 0 ) ) , v - u o)

V v& V6H where u : ~(0)

o
Since we have less possibilities in t h e test function v than in S e c -
tion I , we assume

(2.5)
~ either $(0) < ~ , or
~du
L IF < u ) ( t ) - F<u)(t')I ~ cit-t I I l~-~I IL ( S t , H )

for u 6 L i p ( S T --H). Let us put i=j , v=uj_ 1 into (2.2) and then i=j-l,
v=uj . Adding these inequalities the values with } are eliminated and
we are in t h e same situation as in the case of equations . Thus , we
obtain the same a priori estimates (except of I < A u i , v > I ~ C l v I ) as in
Section I . Since Un(t) ~ u(t) we have $(u(t)) ~ lim inf %(Un(t))
( # is a l s o weakly l.s.c, on V ). F r o m this information and from

dun(t)
(2.6) <AUn(t),Un(t)-v> ~ ~(v) - ~(~n(t)) + ( dt 'V-Un(t)) -

(f(t,F(~n_l)(t)),V-Un(t))
 29

for v:u(t) we conclude lim sup <AUn(t),Un(t)-u(t)> ~ 0 . Since A

is a pseudomonotone ooerator we obtain

<Au(t),u(t)- v > ~ lim inf <A~n(t),~n(t) - v> V v £ V .

Then integrating (2.6) and taking lim inf i n (2.6) we obtain the solu-
n -
tion of (2.1) by the same arguments as in S e c t i o n I .

Theorem 2. Let A : V - Ve be a bounded maximal monotone operator .

If (1.3) , (1.4), (2.3)-(2.5) are satisfied then there exists the
unique solution of (2.1) with the same properties as in T h e o r e m 1 .

Remark 3. T h e o r e m 2 holds true also in t h e case of the operator A be-

ing nonstationary under the assumptions of Remark ! •

Similarly , the following types of evolution variational inequali -

ties can be solved

a) (d2u(t) ,v- ~d u ( t ) ) + b ( t ,.-d-u~(-t-)- , v-_ ~d u ( t ) ) + a(t;u(t),v- ~d u (. t ) ~ +

dt 2

~(v) - ~ <.du(t).
--~) A (f(t),v- -d-u~( t ) ) u(0) = Uo , d ud---{---
( 0 ) = U1 ;

.du(t) du(t)~ du(t)) .du(t)

b) <d-d-t---'v- d - - ~ " + a(t;u(t),v- ~ + ~(v) - ~<--~ ) _>

(f(t),v- ~d u (, t ) ] u(0) : Uo ;

c) a(t;u(t),v- ~d u ( t ) ) + #(v) - % .du(t),

< ~ ) ~ (f(t),v- ~d u (, t ) h ,u(0):Uo

v ~ V~H , a.e. t ( I .

Here V , H are Hilbert spaces and b(t;u,v) , a(t;u,v) are continu-

ous bilinear forms in u,v ~ V ( t ~ I) . We use the approximation sche-
me
1
a I) ~(6hUi-~hUi_l,V-~hUi ) + b(ti;6hUi,V-6hU i) + a(ti;ui,v-~hU i) +

~(v) - }(~hUi ) ~ (f(ti),V-6hUi)

b I) (6hUi,V-6hUi) + a(ti;ui,v-6hU i) + ~(v) - }(6hUi ) ~ (f(ti),V-6hU i)

c I) a(ti;ui,v-~hU i) + ~(v) - % ( 6 h U i ) _> ( f ( t i ) , V - G h U i)

i-i
V v ~ V~H , i = l , . . . ,n . If w e express u i = u o + 6hUi h + ~ 6hUih
i-I j:l
and a(ti;ui,v ) = ha(ti;6hUi,V ) + ~ ha(ti;6hUj ~v ) + a ( t i ; U o ~ V )
j=l
in al), bl), c I) then we obtain the ellintic variational inequalities
 30

with respect to 6hU i . We shall assume

2.7) a(t;u,v) = a(t;v,u)

2.8) a(t;u,u) + ~ l u l 2 ~ Cilull ( ~_>0 ) ;

2.9) b(t;u,u) ~ -Clul 2

2.10) dp
1 a(t;u,v) I ~ C1 lul I I Ivl I (p:l,2 in the cases a),b)
dtP p=l in the case c) ) .
/
2.11) l~b(t;u,v)l~cllull ILvlL
We a s s u m e that there exist So~ H , Zoe H such that

2.12) a (So,V-U 1 ) + b(0;Ul,V-Ul) + a(0;Uo,V-Ul) + %(v) - ¢(U I)

( f ( 0 ) , v - U I) ;

2.12) b (Zo,V-Z o) + a ( 0 ; v - z o) + ¢(v) - }(z o) ~ (f(0),v-z O) ;

2.12) c a ( 0 ; U o , V - Z o) + ¢(v) - ~(z o) ~ ( f ( 0 ) , v - U O)

V v~ V~ .By the same w a y as above (see a l s o [5])we obtain

Theorem 3. If (2.7) - (2.12) are satisfied and if f, df d2f ( L 2 ( I , V , )

dt' dt 2
df
(or f, ~ - ~ L2(I,H) in a), b) ) then there exists the u n i q u e solution

of a), b) , c) , respectively , with the following properties :

du C
u~ C(I,V) , ~( L (I,V) , [lUn - u I l~(i,V ) ~

where {Un} is the corresponding sequence of R o t h e ' s function . More-

over , in the c a s e s a) , b) we h a v e
du d2u dun(t) du(t) I2 < C
d--[E C ( I , H ) , ~ L (I,H) , i-- dt _ ~ V t ~ I .
dt 2 dt
Remark 4. In the fact in the cases a) , b) a perturbed symmetry of
a(t;u,v) can be a s s u m e d . Let ao(t;u,v) be c o n t i n u o u s bilinear form
in u,v~ V (t ~ I) s a t i s f y i n g lao(t;u,v) I CI lul I Ivl It s u f f i c e s
to a s s u m e a (t;u,v) + ao(t;u,v) is s y m m e t r i c .

Remark 5. In the cases of v a r i a t i o n a l inequalities a) , b) a more ge-

neral problem (corresponding to p r o b l e m (2.1)) with a right hand side
f(t,F(u)(t)) can be c o n s i d e r e d . If (1.3),(1.4),(2.5) are satisfied
then Theorem 3 holds true .

Remark 5. U s i n g time and space discretization the variational inequali-

ties a) , b) have been solved in [i]. A special case of (2.1) (A is
asymptotically linear , } ~ #K - i n d i c a t r i x of the closed convex set K
in V ) have been solved in[12] .
 31

III. Higher oder evolution equations.

In t h i s section we a p p l y Rothe's method to the e q u a t i o n s of the form

m-i dkw(t) dm-lw

(3.1) G(t) dmw(t--) + ~ Ak(t) ~ = g(t,w, dtm_l )
dt m k=0 dt k "'''

dkw(0) = Wk k=0, ,m-i where Ak(t)E~(V,V*) G(t)E ~ ( H , H )

dt k ' . . . . ,
(&~(V,V*)), g ~ L i p ( I × [ V ] m ~ H ) and V , H being Hilbert spaces with
V~ H dense in V and H . The equations of type (3.1) include the
governing equations of q u a s i s t a t i c and dynamic problems of v i s c o e l a s t i c
plates and s h a l l o w shells (see [13]). We assume that either i) Am_l(t)
is V-elliptic , or ii) Am_2(t) is V-elliptic . Operator G(t) is
supposed to be symmetric and H-elliptic . Using transformation

dm-lw dm-2w
u = - - in the c a s e i) , or u = - - (m ~ 2 ) in the case ii)
dt m-I dtm-2

the e q u a t i o n (3.1) can be reduced to the form

du(t)
E) i G(t) ~ + A(t) u(t) = f(t,u,F(u)(t)) or

E)ii G(t) d2u(t)

_ _
+ B(t) ~du(t) + A(t)u(t) du
= f(t,u,~-~,F(u)(t))
dt 2
where
t t
(3.2) F(u)(t) = (/ u ds ..... / (t-s)P~u(s) ds) (p=m-2 in i) ,
0 0
p:m-3 in ii) )
The problem E) i has b e e n considered in S e c t i o n I . Now , we formulate
Problem 3.1 w h i c h includes the p r o b l e m E)..
±l
Let V , V1 ; H , HI be H i l b e r t spaces and let <u,v> V , <x,y> H be
the continuous pairings between u ~ Vl, veV and x~H 1 , y 6 H , respec-
tively . Let a(t;u,v) , b(t;u,v) be the same as in S e c t i o n I and let
G(t;u,v) be a c o n t i n u o u s bilinear form for u,vE H . Consider the o p e -
rators fv E L i p ( I x V --V 1 ) , fH & L i p ( I x V x H - H I ) and V o l t e r r a type ope-
rators FV: L i p ( S T --V) -- L i p ( S T - V ) , FH: Lip(S T-H) ~ Lip(S T ~H) .
du du
Problem 3.1. To find u ~ C(I,V~H) with ~-~ L (I,V~H), ~-~E C ( I , H )
d2u
~ L (I,H) such that
dt 2

G(t;d2u(t) . du(t)
(3.3) dt 2 ,v) + b[t;--~,vj + a(t;u(t),v) =

du
<fv(t,F(u)(t)),V>v + <fH(t,Fv(u)(t),FH(~-{)(t)),V>H
du
holds for all v~ VDH and u = ~ , ~-~ = ~ on SO where ~Lip(S o ~V),
~ELip(S O-H) are g i v e n functions .
 32

TO solve Problem 3.1 we use the a p p r o x i m a t i o n scheme

1
(3.4) ~ G(ti;~hUi-~hUi_l,V) + b(ti;~hUi,V) + a(ti;ui,v) :

<fv(ti,F(~i_l)(ti)),v> V + <fH(ti,Fv(~i_l)(ti),FH(6h~i_l)(ti)),v> H

v v & VNH where ~i-i is the same as in Section I and @h~i_l is

c o n s t r u c t e d by means of @ , 6hUl,...,ghUi_l bv the same way as ui_ I.
S i m i l a r l y as above (3.4) can be t r a n s f o r m e d to the e l l i p t i c e q u a t i o n
w i t h respect to @hUi provided 6 h U l , . . . , 6 h U i _ 1 are k n o w n .
The s o l u t i o n of P r o b l e m 3.1 and the c o n v e r g e n c e of our a p p r o x i m a -
tion scheme we obtain under the following assumptions

3.5) G(t;u,v) = G(t;v,u)

2
3.6) CllUl 2 ~ G(t;u,u) ~ C2~u
s
d
3.7) I~-~ G(t;u,v) IN c fuR Ivl

3.8) IIFR(U) - F R ( V ) I I C ( S T , R ) ~ CI lu - vl IC(ST,R ) for R =V , H ;

3.9) IIFR(U)(t)-FR(U)(t')II R ~ t-t" I L(I IUIIC(ST,P0)(I+I Idd--~lIL (St,R))

where L : R+ ~ R + is c o n t i n u o u s , t,t'&I , t'<t and
F(u) is from (3.2). Analogously to (2.12) we assume :
There exists So£ H such that

3.10) G(0;So,V) + b(0;$(0),v) + a(0;~(0),v) = <fv(0,0),v> v +

<fH(0,Fv(~)(0),FH(~)(0)),v> H V v ~ VnH i

Theorem 4. Suppose fv ~ L i p ( I x V ~ Vl), fH(IXVxH~HI) (see (1.3)) and

$(0)(V . If (3.5) -(3.10) are s a t i s f i e d then there exists the unique
solution of P r o b l e m 3.1 . Moreover, the e s t i m a t e s
du
(3 1]) llUn - uli2 ~ C n dul 12 < C
• C(I,V) n ' I ]dt dt L (I,H) -

hold where {un] is the c o r r e s p o n d i n g sequence of Rothe°s functions .

By a similar technique used in Sections I and II, successively we

obtain a priori estimates

lUll ~ C r 16hU i ~ C
and then
I~uil ~ c ,I!6hUil I s C , IIuil I S C .

Similarly as in Lemma 3 a priori estimates (3.11) can be p r o v e d .Then

taking the limit as n-~ in a p p r o x i m a t i o n scheme (3.4) we c o n c l u d e
Theorem 4 .
 33

Examole . Problem 3.1 can be interpreted in t h e following way .

We p u t V : ~(~) , V 1 : W22 , H = h2(~) : H1 where ~c R N . Consider

c2
a(t;u,v) = ~ (x,t) Diu DJv dx for u,v(W2(~)

b(t;u,v) = Jil ~i][jI-<2 ~ bij (x,t) DIu D3V dx (or b(t;u,v) = [~] uv dx).,

t
<fv(t,F(u)(t) ,v> v = / Av ] (t-s) p A u ( s ) ds d x (p _>i) ;
0

6/ v Au(~(t)) dx ~Lip(ST--ST)
If ~ (t)-< t ;
, (du ~(t)
<fH(t Fv(u)(t ,FH ~ ) ( t ) ) , v > H : v / K(s,t) Au(s) ds dx ]
-q
m(t)
f (s,t) dsdx ,
- q

Bilinear form G(t;u,v) can be interpreted in the following way .

i) H : L2(a) : H 1 , G(t;u,v) : / uv dx .
dmw
Then the first term in (3. I) is of the form
dt m ;
2) H : L2, (~) : {u ;/ ~u 2 d x <~] , H] : L i(~) , <u,v> H : ] uv dx
2, -- 9
where ~(x) > 0 , e & LI(~) . We consider Cla(X) _< g ( x , t ) ~ C2~(x)

Then G(t.u,v) : / g(x,t) uv dx (u,vE L2 (~)) g e n e r a t e s a degenerate

dmw
first term in (3.1) in the form g(x,t) . dtn-
~ ;

3) H = V = [I2 , H I = V 1 = W 2 2 . Then G(t}u,v) generates the first

dmw
term in (3.1) in t h e form G(t) where G(t)E ~ ( V , V f=) is a s y m -
dt m
metric ~ V-elliptic operator .

References

[i ] R. G l o w i n s k i , J.L. L i o n s , R. T r e m o l i @ r e s : Analyse num4rique des

des in4quations variationnelles . Dunod, P a r i s 1976

[2 ] J. Ka~ur : The Rothe method and nonlinear parabolic equations of

arbitrar V order . Theory of nonlinear operators - Summer school-
Neuendorf 1972 . Akademie-Verlag . Berlin 1974, 125 - 131 .

[3 ] -- : Application of Rothe's method to n o n l i n e a r evolution equa-

tions . Mat. ~asopis Sloven. Akad. Vied 25, 1975 , 63 - 81

[4 ] -- : Method of R o t h e and nonlinear parabolic boundary value

problems of a r b i t r a r y order . Czech. Math. J., 28 103 , 1978
 3~

[s] -- : On an approximate solution of variational inequalities .

Math. Nachr. , 123 , 1985 , 63 - 82 .

[6~ - - : Method of Rethe in Evolution Equations . TEUBNER-TEXTE zur

Mathematik , Leipzig , to appear .

[7 ] A.G. Kartsatos , 2~.E. Parrott : A method of lines for a nonlinear

abstract functional differential e~uations . Trans . Am. Mth. Soc.
V-286 , N-I , 1984 , 73 - 91 .

[8 ] -- : Functional evolution equations involving time dependent

maximal monotone operators in Banach spaces . Nonlinear analysis
Theory , Methods and Applications . Vol.8 , 1984 , 817-833 .

[9 ] J. Ne~as : Applications of Rothe's method to abstract parabolic

equations . Czech. Math. J. 24 , 1974 , 496-500 .

[i0] K. Rektorys : On application of direct variational methods to the

solution of parabolic boundary value problems of arbitrary order
in the space variables . Czech. Math. J. ,21, 1971 , 318-339 .

[II] M. Zl~mal : Finite element solution of quasistationary nonlinear

magnetic fields . RAIRO , Anal. Num., V-16, 1982, 161-191 .

[12] A. Zen~ek : Approximation of parabolic variational inequalities .

Aplikace matematiky, to appear .

[13] J. Brilla : New functional spaces and linear nonstationary prob-

lems of mathematical physics . EQUADIFF 5 - Proceedings of the con-
ference held in Bra%islava, 1981 . TEUBNER-TEXTE zur Mathematik ,
Band 47 , Leipzig , 1982 .
BOUNDARY VALUE PROBLEMS
IN WEIGHTED SPACES
A. KUFNER
Mathematical Institute, Czechoslovak Academy of Sciences
115 67 Prague 1, Czechoslovakia

1. Introduction

Let us c o n s i d e r a linear differential operator of o r d e r 2k of the

form

(Lu)(x) = [ ~ [ , I B t £ k (-1) [ e [ D a ( a B(x)Dgu(x) ) , x 6 ~ , (1.1)

together with the a s s o c i a t e d bilinear form

a (u,v) = ~ J
{aa~(x)Dgu(x)Dav(x) dx . (1.2)

Here a 8 are given (real) functions defined on the d o m a i n ~ 6 ~N

The usual procedure for solving a boundary value problem for the
operator i proceeds in the following fundamental steps:

(i) Choose an a p p r o p r i a t e Banach space V such that the form

a(u,v) is defined a n d continuous on V x V and elliptic on V , i.e.,
that there exist constants cI > 0 and co > 0 such that the following
conditions are fulfilled:

la(u,v) I ~ cii lul IV11vl IV for e v e r y u, v 6 V (1.3)

(continuity of a(u,v) ), and

a(u,u) ~ c0[ lul I~ for e v e r y u V (1.4)

(ellipticity of a(u,v) ).

(ii) Use the Lax-Milgram Lemma in o r d e r to o b t a i n a s s e r t i o n s a b o u t

the e x i s t e n c e and uniqueness of a (weak) s o l u t i o n in the space V .
If t h e c o e f f i c i e n t s ass of the o p e r a t o r i are bounded, i.e., if

a 8 6 L~(Q) for I~/, [BI ~ k , (1.5)

and if the o p e r a t o r L is (for s i m p l i c i t y ) ~niformly elliptic, i.e.,
if t h e r e exists a constant cO > 0 such that
 36

aaS(x)~ {B ~ c01~I 2 (1.6)

I~I, isI~ k
for every ~ c ~M ( ~ = {~, I~I s k} ), then the first step m e n t i o n e d
above c a n be r e a l i z e d if we choose for the space V the Sobolev space
~,2(~) or o n e of its subspaces selected according to the t y p e of the
boundary conditions.

If o n e or b o t h of the c o n d i t i o n s (1.5) and (1.6) are v i o l a t e d , i.e.,

if o p e r a t o r s with singular coefficients appear condition (1.5) is
not fulfilled - or if the o p e r a t o r becomes degenerate - the q u a d r a -
tic f o r m on t h e left-hand side in (1.6) is o n l y p o s i t i v e semidefinite
- then the Sobolev spaces wk'2(~) cannot be u s e d in g e n e r a l . In t h e s e
cases, appropriate weighted S~bolev spaces can be c o n s t r u c t e d w h i c h re-
p l a c e the c l a s s i c a l s p a c e s w k ' 2 ( ~ ) . T h e weight functions a p p e a r i n g in
t h e s e n e w s p a c e s are determined by the coefficients of the operator, a n d
the m e t h o d of the p r o o f of the corresponding existence and u n i q u e n e s s
theorem for w e a k solutions is the same as in t h e case of c l a s s i c a l So-
bolev spaces, the main tool being the L a x - M i l g r a m Lemma.

On the o t h e r hand, there appear boundary value problems in w h i c h the

operator i satisfies conditions (1.5) and (1.6) but the r i g h t - h a n d
side in t h e e q u a t i o n lu = f or the r i g h t - h a n d sides in t h e b o u n d a r y
conditions Biu = gi , i = 1,...,k ( Bi being boundary operators) be-
have in s u c h a way that the classical Sobolev spaces cannot be used:
the function f is n o t an e l e m e n t of the d u a l space (wk'2(~)) ~ or
s o m e of the functions gi are not traces of functions from wk'2(~)
on the b o u n d a r y 8Q of ~ . Also in s u c h cases, weighted spaces can be
sometimes used for o b t a i n i n g assertions about existence and uniqueness
of w e a k solutions. The bilinear form a(u,v) is c o n s i d e r e d to be d e f i -
n e d on a weighted space V or on a product of two weighted spaces
V 1 x V 2 , and it is n e c e s s a r y to s h o w for which such spaces conditions
(1.3), (1.4) or t h e i r certain modifications are fulfilled.

In w h a t follows, we give a survey of results obtained in t h e s e two

directions of a p p l i c a t i o n of w e i g h t e d Sobolev spaces to the solution of
boundary value problems.

1.1. D e f i n i t i o n.........
of
....... the w e i g h t e d space. Let I < p < ~ and let ~ be
a domain in RN . For k 6 ~ and for N - d i m e n s i o n a l multiindices a
such that I~I =< k let a = ~(x) be weight functions, i.e. measur-
able a n d a.e. in d positive functions, and let us d e n o t e S =
{~, lal ~ k} . The w e i g h t e d Sobolev space

wk'P(d~S) (1.7)
 37

is d e f i n e d as the set of all functions u = u(x) , x ~ ~ , such that

l lull ,p,s = i~t<~

~ I ID~u(x) Ip ~ (x) dx < ~ , (1.8)

the derivatives D~u being considered in t h e sense of distributions.

Further, let

w~'P(a~S) (1.9)

be the closure (if it is m e a n i n g f u l ) of the set C~(~) with respect to

the norm (1.8).

1.2. Theorem. Let us suppose that

g
e L~ O C (~) for lal ~< k "
(I IO)

Then the linear set wk'P(~;S) is a B a n a c h space wlth,respect

to the norm I l'I ' defined by (I 8 ). - if, moreover,
Ik,p, S

~a £ L11oc(e) for I~[ =< k , (1.11)

then the linear set W~0'P(~S) is a B a n a c h space with respect

to the same norm.

1.3. Remark. Conditions (1.10), (1.11) are rather restrictive. In t h e

paper A. KUFNER, B. OPIC E2] it is s h o w n how to modify the definition
of the weighted spaces if (1.10) and/or (1.11) are not fulfilled.
The most frequent type of weight functions ~ are the so called power
type weights

a s (X) = [dist(x,M)~ E

with M C [ and e = E(a) real numbers. If M is a s u b s e t of the

boundary 8~ of Q , then conditions (1.10) and (1.11) are obviously
satisfied.

2. Operators with singular or degenerating coefficients

Let us consider the operator i from (Ioi) and let us suppose that
its coefficients fulfil the following conditions:

a
~a > 0 a.e. in ~ ;
a , a -I
~ E L lIo c ( ~ ) , lal < k ; (2.1)

there exist constants cI > 0 , cO > 0 such that

 38

la~B (x) I < c 1 4 a a ~ ( x ) a ~ (x) a.e. in Q (2.2)

for

2
~. aa~(x)~a~ > cO ~ a (x)~ a.e. in ~ (2.3)

for all ~ 6 ~M .

Conditions (2.1) indicate that the w e i g h t e d spaces W k'2(~;s) and

~ '2(Q;S) with ~ = a~a , ,
~I~I =< k , i.e. with

S = {a = a (x) Is[ < k} (2.4)

are B a n a c h spaces. From the following theorem we see t h a t these weighted

spaces are just the right tool for solving boundary value problems.

. 2.1. Theorem. Let the operator L fro~ (1.1) fulfil conditions (2.1)
- (2.3). Let S be given by (2.4); ~et f e (W~'2(~;S)) and u0
wk'2(~;S) . Then there exists one and only one weak solution u
wk~2(~;S) of the Dirichlet problem for the equation Lu = f , i.e.,
such a function u" that

u - u0 £ ' (~) (2.5)

and

a(u,v) = < f,v > for every v E W O) , 2u( ~ . (2.6)

Moreover, there is a constant c > 0 such that

I Lullk,2,s~C6I{fl{~ + I I%llk,2,S) (2.7)

I d e a of the p r o o f : Condition (2.2) implies the continuity of the

bilinear form a(u,v) from (1.2) on V x V with V = wk'2(~;S) and
conditions (2.3) imply its ellipticity while (2.1) guarantees that the
k,2
space V and its subspace W0 (~;S) are w e l l defined. A standard
application of the L a x - M i l g r a m Lemma then yields the existence and uni-
queness of a w e a k solution u 6 wk'2(~;S) as w e l l as the e s t i m a t e
(2.7) which e x p r e s s e s the c o n t i n u o u s dependence of the solution on
the d a t a of the b o u n d a r y value problem.

2.2. Remarks. (i) In the sequel, we shall give two examples of b o u n -

dary value problems which go beyond the frame of conditions (2.1) -
(2.3), but for w h i c h again existence and uniqueness of a w e a k solution
c a n be p r o v e d . These examples indicate that conditions (2.1) - (2.3) can
be s u b s t a n t i a l l y weakened and that the a d e q u a t e weighted space c a n be
constructed in a m u c h more spphisticated way. A detailed description of
 39

the (rather complicated) construction of these spaces can be found in

A. KUFNER B. OPIC [ I " ] , [3].

(ii) Although Theorem 2.1 is a s i m p l i f i e d version of an applica-

tion of weighted Sobolev spaces to the solution of boundary value prob-
lems, some of its conditions can be weakened: E. g. condition (2.3) fol-
lows from (2.2) if the constant cI in (2.2) is sufficiently small, i.
e. if c I < I/(M - 1) where M is the number of multiindices a such
that lel ~ k .

(iii) The restriction to the Dirichlet problem in T h e o r e m 2.1 is

not substantial, either; other boundary value problems can be handled
in t h e same manner.

2.3. Exampl e . Let us c o n s i d e r the differential operator of order two,

i.e. k = I :
N
(Lu) (x) = - Z ~ 78 ~(ai(x) ~.Su
~V-~ ) + a0(x)u
i=1 1 1

where ai > 0 for i = I .... ,N but_ a 0~< 0 , a0 = - Ib 0 with b0 >

0 , I >__ 0. W e suppose that ai, ai I E L l o c ( Q ) for i = 0,1 ..... N .
Here, one of the conditions in (2.1) is n o t fulfilled, b u t if w e t a k e
.1,2
S = { b 0 , a I ..... aN} , then W]'2([~;S) and w0 (~;S) are the adequate
spaces which can be used for deriving existence and uniqueness theorems
provided the following inequality ~olds' for all U ~ CO(Q) :

f lu(x) ]2b0(x) dx < c ~

N ! ~8u
i
2 ai(x) dx (2.8)
i=I

with a constant c independent of u , and provided the constant 1 in

a 0 =- Ib 0 is sufficiently small, namely I < 1/c .

2.4. Example. Let us consider the plane domain ~ = (0,~) x (0,~) (i.e.
N = 2 ) and the fourth order operator

~2 61 62 82u
(Lu) (x) ax13x2(X I x 2 axlax-----~-
) -

- ~--9--I sl x 2~2 ! ~ 2 )
I¢IYI x2~2 ma u7) 2 xI
Here we have two possibilities:

(i) We can prove existence and uniqueness of a weak solution of the

E2
Dirichlet problem in the anisotropic space W ' (~;S) normed by
 40

lUl 12 = IUl 2 X 1
($1-2 x 262-2 dx +
i 8~1 2 X Y1 X2
I X2 dx

(2.9)
BI B2 O2u 2 61 62
8u 2 2 Xl
+ f 8x x2 dx + f ~x18x2 x1 x2 dx

provided 81 ~ I , 62 ~ I (these last conditions are caused by the

fact thatthe e l l i p t i c i t y constant cO equals
-2 2
16(6 1 - 1) (6 2 - I) + l ).

(ii)~ We c a n p r o v e existence and uniqueness in a n o t h e r anisotropic

space wE'2(~;S) normed by the expression obtained by o m i t t i n g the se-
cond and third integrals in (2.9) provided Y1 = 61 ' Y2 = 62 - 2
#I = 61 2 , B2 = 62

2.5. Remarks. (i) Example 2.4 shows that the structure of the opera-
tors as w e l l as of the w e i g h t e d spaces can be m o r e general than that
mentioned in f o r m u l a (1.1) and in D e f i n i t i o n 1.1; in p a r t i c u l a r , ani-
sotrop~c o p e r a t o r s and spaces c a n be t r e a t e d by o u r m e t h o d .

(ii) In E x a m p l e 2.3, the estimate (2.8) played an i m p o r t a n t role.

Estimates of s u c h a type, which c a n be v i e w e d as c o n t i n u o u s imbeddings
of WI'2(~;S) into the w e i g h t e d L2-space L 2 ( Q ; b 0 ) , are v e r y useful
tools both in the theory and in a p p l i c a t i o n s of w e i g h t e d Sebolev spaces.

2.6. Nonlinear o~rators. Let us c o n s i d e r the nonlinear o p e r a t o r

(Lu) (x) = ~ (-I) I~] D a a (X;6kU(X)) , x £ C , (2.10)

I~I <__k
where 6kU = {DSu, 18 I ~ k} . Using the theory of monotone operators,
results concerning existence of w e a k solutions of b o u n d a r y value pro-
blems for the e q u a t i o n [u = f in the w e i g h t e d space ~'P(~;S) with
I < p < ~ and S : {~ , lal ~ k} can be d e r i v e d provided the "coeffi-
cients" a (x;~) of the o p e r a t o r k satisfy the following three con-
ditions:
(i) the weighted growth condition

la (x;6) [ ~ ~l/P(x) [g (x) + c ~ I~61 p-I ~/P(x)] , (2.ii)

IBIA k

I~I < k for a.e. x ~ ~ and all < £ RM with given constants c > 0
and functions g e Lq(~) , q = p/(p - I) ;

(ii) the u s n a l monotonicity condigson

 4!

(2.i2)
Z
lal <_k

f o r a. e. x E ~ and all ~, n E RM ;

(iii) a weighted coercivity condition

Z a (x;~)£~ ~ c o $ l~al p o (x) (2.13)

!a!!k ~ lal~ k a

for a.e. x ~ ~ and all ~ E ~M with a given constant cO > 0 .

The following assertion is a n o n l i n e a r analogue of Theorem 2.1.

2.7. Theorem. Let S = {o a, lal ~ k} be a given family of weight

functions and wk'P(~;S) the corresponding weighted space with p > I.
Let the operator i from (2.10) fulfil conditions (2.]I) - (2.]3). Let
f 6 (w~'P(9;S)) ~ and u0 C wk~P(~;S) be given. Then there exists at
least one weak solution u C wk'P(~;S) of the Dirichlet problem for
the equation ku : f , i.e., such a function u that

u - u0 E W~o'P(~;S)
and
b

If the inequality in (2.12) is strict, then the solution u is

uniquely determined.

Idea of the p r o o f : Let us consider the form

l

The operator } , defined by the formula a(u,v) = <}u,v > , is, in v i e w

of condition (2.11), a bounded operator from wk'P(~;S) into its dual.
To find a solution u of the Dirichlet problem means to find a func-
tion ~ E X = ~v ' P ( ~ ; S ) such that a(u + Uo, v) = <f,v> for every
v 6 X , i.e., that 9(~ + u 0) = f . If w e denote by T the operator
from X to X~ defined by T~ = 9([ + u 0) , then our problem reduces
to the solution of the equation Tu = f in X with a given f ~ X*
Conditions (2.]I) - (2.13) guarantee that the operator T is b o u n d e d ,
demicontinuous, monotone and coercive, and so, the existence of at
least one solution of the Dirichlet problem follows by applying the
method of monotone operators. Uniqueness follows by contradiction if w e
assume that the inequality in (2.]2) is s t r i c t .
 42

2.8 Example. As a t y p i c a l example of a n o n l i n e a r operator closely con-

nected with the w e i g h t e d Sobolev space wk'P(~;S) , S = {~ = Oa(X);
lal ~ k} , we can c o n s i d e r the o p e r a t o r

(iu) (x) = E {-I) laID~{IDau(x) IP-I sgn D~u(x) c ix) }

3. Ellipti c operators with "bad" right-hand sides

Let us n o w s u p p o s e that the o p e r a t o r i from (1.1) satisfies con-

dition (1.4) w i t h a space V C wk'2(~) ( V is c h o s e n in a c c o r d a n c e
w i t h the type of the b o u n d a r y conditions: for instance, V = W~'2(~)
for the D i r i c h l e t problem and V = wk'2(~) for the N e u m a n n problem).

Further, let S = {~ , la{ ~ k} be a f a m i l y of w e i g h t functions

and let us d e n o t e by I/S the f a m i l y {I/~, l~I ~ k} . The functions
I/~ e are w e i g h t functions as w e l l and c o n s e q u e n t l y , we can c o n s i d e r
the p a i r of w e i g h t e d Sobolev spaces

H I = ~'2(~;S) and H 2 = ~'2(~;I/S) (3.1)

Rewriting the b i l i n e a r form a(u,v) from (1.2) in the form

a(u,v) = ~ fa ~ ( x ) D ~ u ( x ) $ o a ( x ) D ~ v ( x ) / ~ I dx (3.2)

we c o n c l u d e immediately from H~Ider's inequality that a(u,v) is a

continuous bilinear form on H I x H2 .

This last p r o p e r t y replaces the c o n t i n u i t y condition (1.3). If we

now replace the e l l i p t i c i t y condition (1.4) by the pair of conditions

sup la(ul,u2) I K cjl lujl IH ' , i, j = 1,2 , i # j , (3.3)

lluilIH ~I 3
1

- we say in this case that a(u,v) is (Hi,H2)-elliptic - we can

again derive assertions about existence and uniqueness of w e a k solutions
of the e q u a t i o n [u = f in HI (or its s u b s p a c e s selected according
@
to the type of the b o u n d a r y conditions), f 6 H 2 . The m a i n tool is a
m o d i f i e d version of the La~-Milgram Lemma due to J. N E C A S [I~,[2], who
also proposed the m e t h o d roughly described above,

3.1, P r o b l e m . For what weight functions ~ ~ lal _~ k, the c o n d i t i o n s

(3.3) are s a t i s f i e d ? More p r e c i s e l y : For w h a t w e i g h t functions ~ does
 43

the e l l i p t i c i t y condition (1.4) with V a subspace of the classical

Sobolev space ~,2(~) imply the weighted (H1,H2)-ellipticity?

3.2. Power type weights. L e t us c o n s i d e r weight functions of the form

as(x) = [dist(x,M)~ e for all lal ~ k (3.4)

where M is an m - d i m e n s i o n a l manifold, M C ~ , and c is a real

number. The corresponding weighted space wk'2(~;S) will be d e n o t e d by
~'2(~; (dist)C) , so t h a t w e h a v e w k ' 2 ( ~ ; (dist)-~) for wk'2(~;I/S)

In the case of the Dirichlet problem the solution of P r o b l e m 3.1

is g i v e n by the following statement.

3.3. Theorem. There exists an interval J containing 0 such that

for c ~ J the (Hi,H2)-ellipticity conditions (3.31 are satisfied with

H I = W k,2
0 (~; (dist) e) , H 2 = ~ 0 , 2 ( ~ ; (dist)-e)

The p r o o f is b a s e d on the imbedding

W I0, 2 ( ~ ; (dist) e) ~ L 2 ( ~ ; (dist) ~-2 1 (3.5)

which holds for e # 2 + m - N (m = d i m M) . Using the e l l i p t i c i t y con-

dition (1.4) and repeatedly the imbedding (3.5) (also f o r h i g h e r deriva-
tives and for b o t h ~ and -e ) we obtain the lower estimates (3.3)
with constants c. w h i c h d e p e n d on the c o e f f i c i e n t s of the o p e r a t o r
3
(i.e. on the L ~ - n o r m of aa6 , on the e l l i p t i c i t y c o n s t a n t c o ), on
geometrical properties of ~ (especially on m = dim M a n d on the
smoothness of M ) a n d on the n o r m of t h e imbedding operator from (3.5
(i.e. mainly on the v a l u e of the p a r a m e t e r £ ). The interval J is
determined by the requirement of the p o s i t i v i t y of t h e c o n s t a n t s cj .
A detailed derivation c a n be found in A. KUFNER [I 2 w h o extended to ar-
bitrary M C 3~ the ideas developed by J. N E ~ A S [I~ for the c a s e M =
~ .

3.4. The size of J . Theorem 3.3 states the existence of an interval

J ; consequently, the e x i s t e n c e and uniqueness of a w e a k solution of
the D i r i c h l e t problem for the equation iu = f in ~'2(~; (dist) e) is
guaranteed provided e 6 J . For applications it is n e c e s s a r y to k n o w
the e x a c t size of the interval J of a d m i s s i b l e powers e in the
weight function (3.4). It d e p e n d s on k , ~ and M , but the e s t i m a t e s
derived in the p r o o f of T h e o r e m 3.3 are v e r y rough, and therefore, it
is n e c e s s a r y to e v a l u a t e the interval J in e v e r y particular case se-
 44

parately. For example, for the o p e r a t o r i = - A it c a n be shown that

Theorem 3.3 h o l d s for Isl < I so t h a t w e h a v e J = (-1,1) but in
the case of M = {Xo} with x0 8~ (i.e., the case of the w e i g h t
function Ix - x0 Is ) w h e r e ~ is a p l a n e domain with the outer cone
property at the p o i n t x0 (the cone b e i n g characterized by the angle
) we have a better estimate Igl < 2~/(2~ - ~) . In t h i s connection,
let us m e n t i o n the recent result of J. V O L D ~ I C H [I~ w h o h a s shown that
for any given s ~ 0 , I~I arbitrary small, a second order elliptic
differential operator i can be c o n s t r u c t e d (depending on ~ ) such
that the Dirichlet problem for i has no solution in the space
W I ' 2 ( ~ ; (dist) S)

In the case of o t h e r than power type weights, certain results con-

cerning weight functions of the type

o (x) = s ( d i s t ( x , M ) ) ,

s = s(t) a positive function on (0, ~) , have been derived by B. OPIC.

These results as w e l l as o t h e r examples concerning the Dirichlet problem
can be found in the b o o k A. KUFNER [I].

3.5. Other boundar[ yalue problems. For non-Dirichlet problems, Problem

3.1 has been investigated only for p o w e r type weights, and results si-
milar to T h e o r e m 3.3 h a v e been established. The fundamental difference
as c o m p a r e d to the D i r i c h l e t problem consists in t h e fact that serious
restrictive conditions appear. E. g., in the c a s e of the N e u m a n n problem,
where one has to w o r k with the space W k , 2(~;S) instead of W k0 , 2 ( ~ ; S )
the following analogue of T h e o r e m 3.3 holds.

3.6. Theorem. Let for ~ C ~N and M C 8~ with m = dim M the follo-

wing condition hold:

N-m>2k+1
-- (3.6)

Then there exists an interva~ J containing 0 such that for ~ ~ J

the (Hi,H2)-ellipticity conditions (3.3) are satisfied with HI =
• ~ = wk, 2
w k ' 2 ( ~ ; (dlst) ) , H2 (~; (dist) -s) (and c o n s e q u e n t l y , existence
and u n i q u e n e s s of a w e a k solution u 6 HI of the N e u m a n n problem for
an e l l i p t i c equation of o r d e r 2k is g u a r a n t e e d ) .

The p r o o f uses the s a m e ideas as the p r o o f of T h e o r e m 3.3, but it

is b a s e d on the imbedding

W1'2(~;(dist) g) % L2(~; (dist) ~-2)

which, in c o n t r a r y to the imbedding (3.5), holds only for e > 2 + m -
 45

N . This difference leads to the u n p l e a s a n t restriction (3.6).

3.7. Remark. F o r second order e q u a t i o n s , i.e. for k = 1, condition

(3.6) has the form
N - m > 3
and e x c l u d e s many important and interesting special cases of M as
points (vertices - m = 0 ) or lines (edges - m = I ) on the b o u n -
daries of d o m a i n s ~ of d i m e n s i o n N = 2 or N = 3, respectively.
Nonetheless, for s o m e special domains (cubes) and special operators
( - A ), J. V O L D ~ I C H derived results analogous to Theorem 3.3, even
in the case if (3.6) is v i o l a t e d . For details see A. KUFNER, J. V O L D ~ I C H
D]-

3.@. Another approach. The m e t h o d described above is a l i t t l e more

complicated than the u s u a l method mentioned in I n t r o d u c t i o n : It n e e d s
a pair of B a n a c h spaces and two " e l l i p t i c i t y " conditions (3.3) instead
of o n e simpler condition (1.4) and involves the Lax-Milgram-Ne~as Lemma
mentioned in the b e g i n n i n g of this section. In the p a p e r of A. KUFNER,
J. RI[KOSNfK [I], another method is p r o p o s e d which uses only one (weigh-
ted} space and requires the c l a s s i c a l version of the L a x - M i l g r a m Lemma.

Let us d e s c r i b e the m e t h o d for the Diriohlet problem. We i n t r o d u c e

a new b i l i n e a r form b by the formula
b(u,v) = a(u,ov) (3.7)

where ~ is a (sufficiently smooth) weight function, and consider the

weighted space wk'2(~;S) with the family S = {a (x) = ~(x) for all
lal ~ k} as w e l l as the corresponding space W~'2(~;S) . For a given
functional f 6 (W~'2(~;S)) and a given function u 0 6 wk'2(~;S) , we
say that the function u 6 wk'2(~;S) is a a-weak solution of the Di-
riohlet problem for the operator i if

u - u0 6 ~'2(~;s)
and
b(u,v) = < f,v > for e v e r y v 6 W k,2
0 (~;S)

(provided b(u,v) is m e a n i n g f u l for u, v 6 w k ' 2 ( ~ ; S ) ).

Let us f u r t h e r consider a weight function o which satisfies the

following conditions: There exist a weight g0 and c o n s t a n t s cI , c2
such that

f 12 o0(x) dx Cli~N ~ !
lu(x) 81/ 2
8x i
a (X) dx (3.8)
 46

for every u C C~(~) and

Ivo(x) 12/o(x) ~ c2o0(x) a.e. in ~ . (3.9)

Then it c a n be shown that the form b(u,v) is b o u n d e d on

WI'2(~;S) x WI'2(~;S) . Further, it c a n be shown that if t h e constant
elo 2 is sufficiently small then the form b(u,v) satisfies the ellip-
ticity condition b(u,u) => c0q lu I I 21 , 2 ~ S ' so t h a t the existence and
uniqueness of a ~-weak solution follows by a standard application of
the Lax-Milgram Lemma.

3.9. Remarks. (i) The last result was derived for k : I , i.e. for
the second order o p e r a t o r s only. For k > I , we have to consider
weights o which fulfil conditions (3.8), (3.9) repeatedly (i.e. for
o there must exist the corresponding o0 , for J0 the corresponding
(o0) 0 etc. k-times),

(ii) The pair of conditions (3.8), (3.9) on J can be replaced

by the single condition

IVo (X) I ~ c2J (X) a.e. in ~ (3. 10)

(in t h e case k = I . For such weights we again deduce that b(u,v)

is c o n t i n u o u s and, moreover, for c2 > 0 sufficiently small also ellip-
tic, so that existence and uniqueness of a J-weak solution in W I ' 2 ( ~ ; S )
follows in a s t a n d a r d way.

3.10. E x a m p l e s . (i) For 0(x) = ~dist(x,M)] e , condition (3.9) is sa-

tisfied with o0(x) = ~ d i s t ( x , M ) ] ~-2 and c2 = 2 and condition (3.8)
is s a t i s f i e d with cI = le - II - ½ if ~ # I and with cI =
Is + N - m - 2 1 - ½ if e # m + 2 - N ( m = dim M , N = dim ~ ). C o n -
sequently, we obtain an assertion about the existence and uniqueness of
a (dist)£-weak solution u of the Dirichlet problem in t h e space
WI'2(~; (dist) C) provided I~l is sufficiently small, i.e. E £ J~
where J is a n interval containing the origin. Thus, we have obtained
a result similar to %Z%eorem 3.3, and a comparison of the interval J
from Theorem 3.3 with the interval J shows that (at l e a s t in some
special cases) J ~ J so that our second approach improves the set of
admissible powers.

(ii) The weight o(x) = exp(~ dist(x,M)) satisfies condition (3.10)

with the constant c2 = I~I . Weights of such a type are suitable for
unbounded d o m a i n s ~ and the existence and uniqueness of a o-weak so-
lution is g u a r a n t e e d for IEI small.
 47

3~ii. O t h e r boundary value problems can be dealt with in t h e same manner,

and similar difficulties arise as in t h e first approach. E. g., if we
consider the Neumann problem in t h e space WI'2(~;S) with o(x) =
[dist(x,M)] ~ , we obtain a result about the existence and uniqueness
for E 6 J which is the same interval as in the case of the Dirichlet
problem, but under the restrictive condition N - m ~ 3 . Further, one
can show that for N - m = I our method cannot be used while for
N - m = 2 we find that admissible values are positive ~'s from J
On the other hand, the mixed boundary value problem admits existence
for ~ @ J without restriction on the dimension of M .

3.12. Remark. Since the form a(u,v) was derived from the operator k
by using Green's formula for the integral flu v dx , v E C~(~) , the

form b(u,v) = a(u,ov) can be derived in the same way from the integ-

ral fLu(or) dx = foiu v dx . Consequently, we can treat our a-weak so-

lution as the solution of a boundary value problem for the operator

oku . Since o(x) > 0 a.e. in ~ , the difference between a weak and
a ~-weak solution is m o r e or less formal.

References

KUFN~I j n. :
Weighted Sobolev spaces. J. Wiley & Sons, Chichester-New York
-Brisbane-Toronto-Singapore 1985

KUFNE~_~ A.; OPIC, B.:

The Dirichlet problem and weighted spaces I. ~ a s o p i s P~st. Mat.
108(1983) , 381-408
[2] How tO define reasonably weighted Sobolev spaces. Comment.Math.
Univ. Carolinae 2--5(3) (1984) , 5 3 7 - 5 5 4
[3] The Dirichlet problem and weighted spaces II. To appear in Ca-
sopis P~st. Mat.

KUFNER, A. ; V O L D ~ I C H , J. :
[11 The Neumann problem in weighted Sobolev spaces,
Math. Rep. Roy. Soc. Canada

KUFN~ Linear A. ; R ~ K O S N f K , J. :
elliptic boundary value problems and weighted Sobolev
spaces: A modified approach. M a t h . S l o v a c a 3--4(1984), N o . 2
185-197

N E ~ A S , J.:
[I] Sur une m~thode pour r~soudre les dquations aux d~riv~es par-
tielles du type elliptique, voisine de la variationnelle. A n n .
Scuola Norm. Sup. Pisa 16(1962), 305-326
 48

[2~ Les m~thodes directes en thdorie des dquations eiliptiques.

Academia, Prague & Masson et C le, Paris 1967

V O L D ~ I C H , J.:
[I] A remark on the solvability of boundary value problems in
weighted spaces. To a p p e a r in C o m m e n t Math. Univ. C a r o l i n a e
CRITICAL POINT THEORY
AND NONLINEAR DIFFERENTIAL
EQUATIONS
J. MAWHIN
Institut MathAmatique, U~iversitd de Louvain
B-1348 Louvain-la-Neuve, Belgium

1. I N T R O D U C T I O N
The variational approach to b o u n d a r y value problems for d i f f e r e n -
tial equations consists in w r i t i n g the problem, whenever it is p o s s i -
ble, as an a b s t r a c t equation of the form

(i) ~(u) : 0

where ¢ : E ~ E* is of the form ¢ = ~', w i t h ~' the G ~ t e a u x derivative

of a r e a l function 9 defined on a B a n a c h space E. In this w a y the
search of s o l u t i o n s for (i) is e q u i v a l e n t to the determination of the
critical points of ¢, i.e. of the zeros of ~'. Such a viewpoint can
be traced at l e a s t to F e r m a t , with his m i n i m a l type principle to f i n d
the law of r e f r a c t i o n for the light.

Since Fermat also, we k n o w that the p o i n t s at w h i c h ~ achieves

its e x t r e m u m s are critical points of 9. Thus, any way which succeeds
in p r o v i n g , directly, that ~ has a maximum or a m i n i m u m provides a
way of p r o v i n g the existence of a s o l u t i o n of (i). This is the
so-called direct method o f the c a l c u l u s of v a r i a t i o n s which goes
back to G a u s s , Kelvin, Dirichlet, Hilbert, Tonelli and others. More
recent work deals with Droving the e x i s t e n c e of critical points at
which ~ does not achieve an extremum (saddle points). This paper
surveys some of the r e c e n t work in this direction. A systematic
exposition of m a n y aspects of the v a r i a t i o n a l approach to b o u n d a r y -
-value problems for o r d i n a r y differential equations will be given in
[11].

For definiteness, we shall consider a system of o r d i n a r y differ-

ential equations of the form

(2 ) u" + ~u = V F ( x , u ) (V : D u )
 50

on a compact interval I = [a.b] , s u b m i t t e d to h o m o g e n e o u s boundary

conditions, say, of Dirichlet, Neumann or periodic type. For sim-
plicity, we assume here that F and VF are continuous on I × R N. We
could as w e l l consider elliptic partial differential equations. It
is w e l l known that the spectrum of - d2/dt 2 submitted on I to the
above boundary conditions has the form

(0S) ~i < 12 < ....

Moreover, (2) is the Euler-Lagrange equation associated to the

functional

: H ~ R, u ~ Q (u) + II F(.~u(.))

where

Q (u) = / i ( i / 2 ) ( lu'l 2 - ~ l u 1 2 ) ,

I(I;RN), H I ( I ; R N ) o r H~(I,R N) = {u E HI(I,R N) : u(a) =

and H = H0
= u(b)} with their usual norm denoted by ll.ll. S o l v i n g (2) with one
of the above boundary condition is thus equivalent to finding a
critical point of ~ o n H, i.e. a point u @ H such that

(3) ~'(u) = O.

If c = ~(u) with u a critical point, c is c a l l e d a critical value

for ~.
The simplest situation for (3) to h o l d is w h e n ~ has a global
minimum (which requires of course ~ to be b o u n d e d from below).
Since Hammerstein [6] in 1930 (in the Dirichlet case) we know
that ~ will have a global minimum whenever

(4) ~ < XI

and

(5) F(x,u) ~ - ( B / 2 ) l u l 2 - y(x)

for some 8 < 11 - e, y E Ll(I) and all (x,u) E I × R N. In fact, ~ is

coercive (~(u) - +~ for ILuH ~ ~) because, by (4) and (5), ~ is
bounded from below by a coercive quadratic form. Moreover $ is w e a k l y
lower semi-continuous so that ~ has a global minimum by a classical
result. We shall discuss now situations where (4) and (5) do n o t
hold.
 5]

2. THE CASE OF ~ : k 1 A N D ~I F C O E R C I V E ON THE K E R N E L

The situation is a l r e a d y m o r e complicated when ~ : iI ( r £ s o n a n c e

at the l o w e s t eigenvalue) and c o n d i t i o n (5) is no m o r e sufficient
for the e x i s t e n c e of a c r i t i c a l points as s h o w n by a l i n e a r e q u a t i o n
violating the F r e d h o l m alternative condition. To m o t i v a t e the
introduction of a n e w - s u f f i c i e n t condition, let us f i r s t c o n s i d e r
the case w h e r e VF is b o u n d e d .

a) The
. . . . .case
. . . . .w.h.e.r.e. .VF
. . . .is. . .b .o .u .n .d .e d
Writing u(x) : ~(x) + u(x) w i t h u • H 1 the e i g e n s p a c e of 1 1 and
E HI = H~, w e h a v e ~(u) = Q1 (~) + ]I [ F ( ' ' ~ ( ' ) ) + F(.,u(.)) -
1
F(.,~(.))] ~ QI (u) + fI F ( . , ~ ( . ) ) -
1
Mil~tlL2 Z ClJt~Jt2 - c2tI~IL + IIF(.,~(.))~

where M is an u p p e r b o u n d for IVFI on I × R N, and we s h a l l recover

coercivity for ~ if we a s s u m e that

(6) I I F(.,v(.)) - +~ as llvli ~ ~ in HI

(coercivity of t h e averaged F on l h e k e r n e l ) . Such a c o n d i t i o n was

first introduced by Ahmad, Lazer and Paul [ i] a n d it generalizes
the c l a s s i c a l Landesman-Lazer conditions. As ~ is a g a i n w . l . s . c . ,
the e x i s t e n c e of a m i n i m u m is i n s u r e d .

The b o u n d e d n e s s of VF c a n be r e p l a c e d bv the c o n v e x i t y of
F(x,.) for each x 6 I. In this case, if (6) also holds, there
e x i s t s u0 e HI such t h a t

(7) fI V F ( ' ' U 0 ( ' ) ) ~ : 0 for all ~ • ~i"

Moreover, by c o n v e x i t y and u s i n g (7) we h a v e

~(U) ~ QI (~) + II [ F ( ' ' ~ 0 ( ' ) ) + (VF(''~0('))'u - ~0

1

(8) = Qkl (~) + /I F ( . , u0- (.)) + ] I ( V F ( . , ~ 0 (.)),u)

cifI~i12 - c21111i - c 3
 5_
~

Thus each m i n i m i z i n g sequence (ub) for ~ has (ub) b o u n d e d in the

norms II.ll and lJ.lJ ~. On the o t h e r hand, by c o n v e x i t y again
L

F(X,~k/2) ~ ( i / 2 ) F ( x , u k) + ( i / 2 ) F ( x , - u k )

and hence,

~(u k) ~ 211 F ( . , U k / 2 ) - I I F ( - , - u k)

2] I F ( . , ~ k / 2 ) - c4,

which, by (67, implies that (Uk) is b o u n d e d and ~ has a m i n i m u m .

Let us r e m a r k that w h e n F(x,.) is s t r i c t l y convex for each x @ I

a n d a = ~i' it can be s h o w n that 6) is n e c e s s a r y and s u f f i c i e n t for
the e x i s t e n c e of a s o l u t i o n [ iO] .

L e t us a s s u m e now t h a t

(9) II F ( . , N ( . ) ) ~ -~ as LiviD ~ ~ in HI"

As this s i t u a t i o n only h o l d s in t r i v i a l situations w h e n F(x,.) is

convex, let us a s s u m e a g a i n that VF is b o u n d e d . By (9), we h a v e

~(v) : II F(.,v(.)) - -~ as lJvJJ ~

in HI' so that ~ is no m o r e b o u n d e d f r o m b e l o w a n d has no g l o b a l

minimum. O n the o t h e r hand, on HI'

~(w) : QI (w) + ]I IF(.,0) + (F(.,w(.)) - F(.,0))]

1

a c IIiwll2 - c211wJl - c 3

and hence ~I~I is b o u n d e d from below (even c o e r c i v e ) . Consequently,

there exists R > 0 such t h a t

sup ~ < inf

HIA3B(R) H1
 53

This suggests the u s e of the following saddle type or m i n i m a x t h e o r e m

of R a b i n o w i t z [15], introduced to g i v e a variati~onal proof of the
Ahmad-Lazer-Paul results [i].

LEMMA i. L e t E be a Banach s p a c e and ~ ~ C I ( B , R ) . Assume t h a t there

exists a decomposition E = Ei • E2 with dim El < = and R > 0 s u c h
that

sup 9 < inf

EINSB(R) E2

~t

E : {o e C(E,E) T o(u) : u on 8B(R)}

and

(i0) c = inf max ~(o(s)) (Z inf ~)

oeE seB(R)AE 1 E2

Assume t h a ~ i f there is a (u k) ~ c h that ~ ( u k) ~ c and ~ ' ( u k) ~ O,

then c is a critical value. IPalais-Smale type condition PS* at c).
Then ~ ha~ a c r i t i c a l point with critical value c.

This theorem c a n be p r o v e d by d e f o r m a t i o n techniques [12] or

Ekeland's variational lemma [ 4] .
In the above case w i t h E = H, E 1 = HI' E2 = HI' the P S * - c o n d i t i o n
holds for each c and ~ has a critical point.
The above results are summarized in the following.

THEOREM i. Assume t h a t

fiF(.,v(.)) - + ~ aS IIviL ~ ~ in HI

(the eigenspace of ~i) and t h a t either VF i s bounded or F i s convex in

u. Then ( 2 ~ ) with the suitable boundary conditions has at least a
i
solution which minimizes ~. A s s u m e t h a t

/I F ( . , v ( . ) ) ~ -~ as llvlt -~ oo i n T{1

and t h a t VF i s bounded. Then (2~i ) w i t h the suitable boundary condi-

tions has at least a solution u with ~(u) : c given by (10) with
E1 = HI t h e eigenspaee of ~i"
 54

3. THE CASE OF ~ : 11 A N D F PERIODIC

An interesting situation in w h i c h (6) does not h o l d occurs

when

F(x,u + T e ) = F(x,u) (i ~ i ~ N)
l l

for all x E I, u E R N a n d some T > 0. (I ~ i ~ N).

1
This implies t h a t F a n d VF a r e b o u n d e d on I × R N. T h e r e f o r e

$(u) = QI (~) + /I F ( . , u ( . ) )
(il) i
ClllUJi2 - c 2 •

so ~ is b o u n d e d from below a n d any m i n i m i z i n g sequence (u k) is s u c h

that (Uk) is b o u n d e d in the norms I].Ii and JJ.li .
L

Then, 11 = 0 and HI ~ RN is the space of c o n s t a n t mappings from

[a,b] into R N. M o r e o v e r ,

(12) ~(u + Tie i) : e(u) (i ~ i ~ N)

for all u E H, so t h a t any m i n i m i z i n g sequence can be s u p p o s e d ,

without loss of g e n e r a l i t y , such that

N
IUkl ~ ( E T~) 1/2
i:l
Thus ~ has a bounded minimizing sequence and h e n c e a minimum. This
result is due to W i l l e m [18] and (independently and in s p e c i a l cases)
Hamel [5] and Dancer [3]. The existence of a s e c o n d solution was
proved by M a w h i n - W i l l e m [8,9] using the m o u n t a i n pass lemma, a variant
of Lemma I. T h e i r approach was extended to s y s t e m s of the form

d 3L ~L
dt 84 (u,~) - ~ (u,~) = 0

by C a p o z z i , Fortunato and Salvatore [2]. See also Pucci-Serrin [13,

14] for a b s t r a c t critical point theorems motivated by this situation.

b) 2h~-~a~_Q£_Dixl~h!~_hgu~arX_~Qn~i~ie~a

The Dirichlet case strongly differs f r o m the other ones because

 55

2
~x
1 1 = (b_a)7 > 0 and HI = span(sin ~-c~) w h i c h imply t h a t we loose the

periodicity property (12) of ~. The problem has been studied by W a r d

[17] for N = 1 and

(13) F(x,u) : G(u + E(x))

where G is c o n t i n u o u s and T-periodic and E : I ~ R is c o n t i n u o u s .

Indeed, Ward considered explicitely the p r o b l e m

V" + 1IV = g(V) + e(t)

v(a) : v(b) = 0

when g ( v + T) = g(v), ]~ g = 0 and e E HI' which reduces to the a b o v e

case by a t r i v i a l change of v a r i a b l e s .
A possible w a y of a p p r o a c h , slightly different from Ward's one,
makes u s e o f the following lemma which c a n be p r o v e d by a d e f o r m a t i o n
technfque or E k e l a n d variational lemma.

LEMMA 2. Let E be a B a n a c h space and ~ c CI(E,R} be bounded from below

and s a t i s f y PS ~ at c = inf ~. Then ~ has a minimum.

Using an e x t e n s i o n of the Riemann-Lebesgue lenm~a, one can p r o v e

that ~ associated to F g i v e n in (13)satisfies the PS*-condition at
each b % 0 and that ~I~i satisfies PS * at e a c h b E R. T h u s the
existence of a c r i t i c a l point is i n s u r e d by Lemr~a 2 e x c e p t when

0 : inf 9 < inf

H
1

The above mentioned Riemann-Lebesgue type lemma also implies that, on

HI~ ~(v) ~ 0 as llviI ~ ~. Thus, there exists some R > 0 such that

max ~ < !nf

HI N ~ B ( R ) H1

and then c given by the Rabinowitz lemma is g r e a t e r or e q u a l to inf

and h e n c e nonzero. Consequently, this c is a c r i t i c a l value for ~.

The above results can be summarized in the following.

T H E O R E M 2. A s s u m e that
 56

F(x,u + T.e,) : F(x,u) (i < i < N)

ll

with N = i and F of the form (13) in the Dirichlet case. Then (21)
l
with the suitable boundary condition has at least one solution.

4. T H E CASE OF li-I < ~ -< ~' (i > 2)

In this case, ~ is n e i t h e r bounded from below nor from above, as

Q (v) - - ~ on Hi_ 1 : span of eigenfunctions of kl,...,li_l and
O~(v) - + ~ on Hi+l : span of eigenfunctions of
li+l'

a) ~ h £ _ S ~ _ ~ h ~ S _ ~ < _ i ~ _ ~ S ~

Then one can use the Rabinowitz Lemma in a way similar to the
case where ~ = l I and ~! F ( . , v ( . ) ) ~ - ~ as ~v~ ~ ~ if the extra
condition

(14) fI F ( . , v ( . ) ) d x ~ + ~ or - ~ as Irvll ~ ~ in the

eigenspace of 1
1

holds when ~ = ki" One choose in this case E 1 = H i' E 2 = H i o r E l =

= ~i+l' E2 = Hi+l according to the sign of ~ in (14).Under these
conditions (2) has at least one solution. This is essentially a
result of Ahmad-Lazer-Paul [ i] and Rabinowitz [15].

Then, sharper results can be obtained without boundedness

assumption of VF through the use of the Clarke-Ekeland dual least
action principle which reduce the study of the critical points of
to that of an associate
dual function ~ involving the (possibly
d2
generalized) inverse of -- + I I and the Legendre-Fenchel transform
dt 2 i
of F(x,.). Under reasonable conditions o n F, % is b o u n d e d f r o m b e l o w
and, in this way the existence of a solution is in particular insured
when

lim sup F(x,u) ~ 6 < i+l i (unif. i n x E I)

2
lUl ~

and (if ~. = e),

1

fl F ( x , f ( x ) ~ I x ~ + oo a s IIvll ~ oo i n the eigenspace of h i.

 57

See [ IO] for. g e n e r a l results in this direction.

c) The case w h e r e F is p e r i o d i c a n d ~ = I
....................................... 1

Results are known only when N = 1 a n d F has the form (13). The
proof, due to L u p o and Solimini [16,7] is such more delicate because
the PS e is n o t satisfied at c = 0. T h i s requires, in a d d i t i o n to the
classical Rabinowitz saddle point theorem, ether saddle point
theorems of the same type and some topological arguments (together
with the Riemann-Lebesgue-type lemma mentioned above).

The above results can be summarized in the following

THEOREM 3. A s s u m e that li_ 1 < a ~ I i (i ~ 2) and t h a t one of the

following conditions holds:

i) VF is bounded an~, whenever ~ = ~ ,

i
IiF(.,v(.)) - + ~ or - ~ as llvn ~ ~ i n the eigenspace of li

Ai+ 1 - I .
ii) F(X,.) is convex, l i m sup F ( x ,Zu ) < 8 < .
- 2 l (unif " in x @ I)~
lul -- ~ lul

and, whenever ~ : li,

(14) /IF(.,v(.)) - + ~ as llvll ~ ~ i n the eigenspace of i

1

iii) ~ : Ai" N : i a n d F has the form (13) wi~h G T-periodic.

Then the problem (2) with any of the boundary conditions has at
least one s o l u t i o n .

One can show that (1%) is n e c e s s a r y and sufficient when F(x,.) is

strictly convex [ IO] .

References

[ i] S . A H M A D , A . C . L A Z E R a n d J . L . P A U L , E l e m e n t a r y c r i t i c a l p o i n t t h e o r y
a n d p e r t u r b a t i o n s of e l l i p t i c b o u n d a r y v a l u e p r o b l e m s at r e s o n a n c e ,
I n d i a n a Univ. Math. J. 25 (L976) 9 3 3 - 9 4 4 .
[2] A.CAPOZZI, D.FORTUNATO and A.SALVATORE, Periodic solutions of
Laoranqian s y s t e m s w i t h b o u n d e d p o t e n t i a l , to a p p e a r .
[3] E . N . D A N C E R , On the use of a s y m p t o t i c in n o n l i n e a r b o u n d a r y value
p r o b l e m s , Ann. Mat. P u r a ADD1. (4) 131 (1982) 167-185.
[4] I.EKELAND, Nonconvex minimization problems, Bull. Amer. Math. Soc.
(NS) 1 (1979) L~43-474.
[5] G.HAMEL, Ober erz~mgene Schwinqunaen bei endlichen Amplituden, Math.
Ann. 86 (1999) 1-13.
 5~

[6] A.HAMMERSTEIN, N i c h t l i n e a r e I n t e ~ r a l a l e i e h u n g e n nebst Anwendun-

gen, Acta Math., 54 (1930) 117-176.
[7] D.LUPO and S.SOLIMINI, A note on a resonance problem, Proc. Royal
Soc° Edinburgh, Ser. A, to appear.
[8] J . M A W H I N and M.WILLEM, Multiple solutions of the periodic b o u n d a r y
value p r o b l e m for some forced p e n d u l u m - t y p e equations, J. Diff.
Equations 52 (1984) 964-287.
[9] J . M A W H I N and M.WILLEM, Variational methods and boundary value
problems for vector second order d i f f e r e n t i a l equations and
a p p l i c a t i o n s to the p e n d u l u m equation, in " N o n l i n e a r Analysis and
Optimization", Leot. Notes in Math. No 1107, Springer, Berlin,
1984, 181-192.
[IO] J . M A W H I N and M.WILLEM, Critical points of convex p e r t u r b a t i o n s of
some indefinite q u a d r a t i c forms and semi-linear b o u n d a r y value
p r o b l e m s at resonance, Ann. Inst. H. Poincar6, A n a l y s e non-
lin6aire, to appear.
[ ii] J . M A W H I N and M.WILLEM, "Critical P o i n t T h e o r y and H a m i l t o n i a n
Systems", in preparation.
[12] P.S.PALAIS, Critical p o i n t theory and the m i n i m a x principle, in
Proc. Symp. Pure Math. vol. 15, Amer. Math. Soc., Providence,
1970, 185-212.
[13] P.PUCCI and J.SERRIN, E x t e n s i o n s of the m o u n t a i n pass theorem,
J. Funct. Anal. 59 (1984) 185-210.
[14] P.PUCCI and J.SERRIN, A m o u n t a i n pass theorem, J. D i f f e r e n t i a l
Equations, 57 (1985).
[15] P.RABINOWITZ, Some m i n i m a x theorems and a p p l i c a t i o n s to nonlinear
partial d i f f e r e n t i a l equations, in "Nonlinear Analysis, a volume
d e d i c a t e d to E°H.Rothe", A c a d e m i c Press, New York, 1978, 161-178.
[16] S.SOLIMINI, On the solvability of some elliptic p a r t i a l differen-
tial equations with the linear part at resonance, to appear.
[17] J.R.WARD, A b o u n d a r y v a l u e p r o b l e m w i t h a p e r i o d i c nonlinearity,
J. N o n l i n e a r Analysis, to appear.
[18] M.WILLEM, O s c i l l a t i o n s forc4es de svst~mes hamiltoniens, Publ.
S~min. Analyse non lin~aire Univ. BesanGon, 1981.
ORDINARY LINEAR DIFFERENTIAL
EQUATIONS - A SURVEY OF THE GLOBAL
THEORY
F. NEUMAN
Mathematical t n s t i t u ~ of the Czechoslovak Academy of Sciences, branch Brno
Mendlovo ndm. 1, 603 O0 Brno, Czechoslovakia

I. H i s t o r y
Investigations of l i n e a r differential equations from the p o i n t
of t h e i r transformations, canonical forms and invariants started in
the last century. In 1834 E.E. K u m m e r [6] studied transformations of
the second order equations in the form involving a change of the
independent variable and m u l t i p l i c a t i o n of t h e d e p e n d e n t Variable.
Till the end of the last century several mathematicians dealt also
with higher order equations. Let us m e n t i o n at least E. Laguerre,
A.R. Forsyth, F.Brioschi, G.H.HalDhen from many others. Perhaps the
most known result from this period is the so c a l l e d Laguerre-F0rsyth
canonical f o r m of linear differential equations characterized by the
vanishing of the c o e f f i c i e n t s of the (n - 1)st and (n - 2)nd
derivatives.
However as late as in 1892 p. S t a c k e l (and one year later
independently S.Lie) proved that the f o r m of t r a n s f o r m a t i o n conside-
red b y K u m m e r (as w e l l as all his successors) is the m o s t general
pointwise transformation that converts solutions of any linear homo-
geneous differential equation of the o r d e r greater th~n one into
solutions of an e ~ u a t i o n of the same kind. In fact, only this result
justified backwards the w h o l e previous investigations.
Already in 19.10 G.D. Birkhoff [ i] p o i n t e d out that the investig-
ations, considered in the real domain, were of local character. He
presented an e x a m p l e of the t h i r d order linear differential equation
that cannot be transformed into any e q u a t i o n of the L a g u e r r e - F o r s y t h
canonical f o r m on its w h o l e interval of d e f i n i t i o n .
The local nature of m e t h o d s and results is not suitable for
dealing with problems of g l o b a l character, as b o u n d e d n e s s , periodicity,
asymptotic or o s c i l l a t o r y behavior and other properties of s o l u t i o n s
that necessarily involve investigations on the w h o l e intervals of
definition.
Only to d e m o n s t r a t e that even in the m i d d l e of this century there
 60

were just isolated results of a global character and no systematic

theory, let me m e n t i o n G. Sansone's example of the third order linear
differential equation with all o s c i l l a t o r y solutions. This result
occured as late as in 1948 in spite of the fact that the question
about the existence or n o n e x i s t e n c e of such an equation is as old as
the p r o b l e m of f a c t o r i z a t i o n of linear differential operators.
It is now some 35 years ago that O. B o r ~ v k a started the systema-
tic study of global properties of the second order linear differential
equations. He deeply developed ~is theory and summarized his original
methods and results in his m o n o g r a p h [3] that appeared in 1967 in Ber-
lin and in an extension version in 1971 in London.
For linear differential equations of the second and higher orders
there have occurred results of a global character in papers of several
mathematicians. Let me m e n t i o n at least N.V.Azbelev, J.H.Barrett,
E. Barvfnek, L.M.Berkovi~, T.A.Burton, Z.B.Caljuk, T.A.Chanturija, W.A.
Coppel, W.N.Everitt, M.Gregu~, H.Guggenheimer, G.B.Gustafson, M.Hanan,
Z.Hust~, I.P.Kiguradze, V.A. Kondratjev, M.K. Kwong, M.Laitoch, A.C.
Lazer, A.Ju. Levin, W.T.Patula, M.R~b, G.Sansone, S.Stan~k, J.Suchomel,
C.A.Swanson, V.~eda, M.~vec, M.Zllmal from several others. However,
there was still no unified and systematic theory of global properties
of linear differential equations of an a r b i t r a r y order enabling us to
fortell what can and what cannot happen in global behavior of solu-
tions.
In the last 15 years we d i s c o v e r e d enough general approach and
methods, we introduced new useful notions and d e r i v e d results giving
answers to substantial questions and solving basic problems in the
area of global properties of linear differential equations of an
arbitrary order. O.Bor~vka's methods and results for the second order
equations were at the b e g i n n i n g of our approach to equations of
arbitrary orders and they still play an important role in the w h o l e
theory. We cannot see the p o s s i b i l i t y how to handle the general situ-
ation w i t h o u t having had his results at our disposal.
Algebraic, topological, analytical and geometrical tools together
with methods of the theory of d y n a m i c a l systems and functional equ-
ations make it possible to deal with problems concerning global pro-
perties of solutions by contrast to the previous local investigations
or isolated results. Theory of categories, Brandt and E h r e s m a n n grou-
poids, Cartan's moving-frame-of-reference method among other differen-
tial geometry methods, and functional equations are some of the means
used in our approach.
 61

The t h e o r y in q u e s t i o n includes also effective methods for sol-

ving several special problems, e.g. concerning the g l o b a l equivalence
of two g i v e n e q u a t i o n s , or f r o m the a r e a of q u e s t i o n s on distribution
of z e r o s of s o l u t i o n s , disconjugacy, oscillatory behavion, etc.

II. G l o b a l T r a n s f o r m a t i o n s
For n ~ 2, let P n ( Y , X ~ I } denote a linear homogeneous ordinary
linear differential equation

(n-l)
y(n) + Pn_l(x)y + . . . + Po(X) : 0 ,

where Pi 6 C°(I), i = 0,1,...,n - i, are real c o n t i n u o u s functions

defined on an o p e n i n t e r v a l I of reals. Similarly, Qn(z,t;J) denotes

z(n) + qn.-i (t)z(n-l) + ' ' ' + qo (t) = 0 , qi e C°(j) ,

i = 0,1,...,n - i, J C R b e i n g an o p e n interval.
We say t h a t P (y,x;I)
n
is globally trans~formable into Q (z,t;J)
n
if
there exist
a function f E cn(j), f(t) # 0 on J, and
a Cn-diffeomorphism h of J into I,
such that
z(t) : f(t) . y(h(t)), t e j
is a s o l u t i o n of Q n ( Z , t ; J ) whenever y is a s o l u t i o n of P n ( Y , X ; I ) .
This d e f i n i t i o n complies with the m o s t g e n e r a l f o r m of a p o i n t w i s e
transformation derived by Stackel. The b i j e c t i v i t y of h g u a r a n t e e s the
transformation of s o l u t i o n s on their w h o l e intervals of d e f i n i t i o n ,
i.e. the g l o b a l i t y of the t r a n s f o r m a t i o n . Let me r e m a r k also, that
recently M.~adek derived St~ckel's result without any d i f f e r e n t i a b i l i t y
assumption, [4].
It a p p e a r s to be c o n v e n i e n t to w r i t e the g l o b a l transformation in
the f o l l o w i n g form. Let y = (yl,...,yn)T be the v e c t o r column function
whose c o o r d i n a t e s j y i are l i n e a r l y independent solutions of the e q u a t i o n
Pn(Y,X;I) for i = l,...,n. L e t us call the y a fundamental solution of
Pn(Y,x;I). Similarly, let z d e n o t e a fundamental solution of the
equation Qn(Z,t)J). Then there exists a nonsingular n by n c o n s t a n t
m a t r i x C such that
(~) z(t) : C.f(t).y(h(t)), t e J .
The g l o b a l transformation expressed expiicitely by this f o r m u l a w i l l be
d e n o t e d by ~ = <Cf,h) , a n d we shall w r i t e
Y
Pn(Y,X;I)e : Qn(Z,t~J) ,
or s h o r t l y
 62

P ~ Q o
The relation of global transformability is an equivalence
relation. Hence the set A of all linear homogeneous differential
equations of all orders greater than and equal to two, is d e c o m p o s e d
into the classes of globally equivalent equations.
Let B be one of the classes of the equivalence. For each three
equations P, Q and T of the class B there exist global tranformations
and 8 such that
P~ = Q and Q8 = T .
If we define a composition ~B of the t r a n f o r m a t i o n s ~ and 8 by
(P~)8 = P(eS) = T ,
we introduce a certain algebraic structure into each class B of
globally equivalent equations. This algebraic structure considered on
the w h o l e set Ais a special category, called the Ehresmann groupoid.
Linear differential equations are o b j e c t s and global transformations
are m o r p h i s m s of the category. The same a l g e b r a i c structure restricted
to any class B of g l o b a l l y equivalent equations is a special Ehresmann
groupoid, called the Brandt groupoid.
The basic (and in fact, the only) structural notion of a B r a n d t
groupoid is the so called stationary group of any of its objects. In
our case of d i f f e r e n t i a l equations, the stationary group G(P) of an
equation P is formed by all global transformations that t r a n s f o r m the
equation P into itself, i.e.
G(P) = {~; P~ = P}
It can be shown that the stationary groups of any two e q u a t i o n s
P a n d Q from the same e q u i v a l e n t class B are conjugate:

if P~ = Q then G(P) = ~G(Q)~ -I

Having a special (canonical) object (equation) S B in the class B

of e q u i v a l e n t equations, all global transformations transforming P into
Q are d e s c r i b e d by the formula

y-IG(SB)6, where P = SBY and Q = SB6.

W e could observe that in each area of m a t h e m a t i c s where a struc-

ture of an E h r e s m a n n groupoid occurs as it is also in our case, the
following basic problems have to be solved in order to d e s c r i b e the
structure of sets of objects and t r a n s f o r m a t i o n s in this area, and in
this manner, to f o r m a foundation of the c o r r e s p o n d i n g theory:
I. F i n d s u f f i c i e n t and/or necessary conditions (if even effective,
the better) under which two g i v e n objects, two g i v e n e q u a t i o n s are
 63

equivalent, i.e. c r i t e r i o n of g l o b a l e q u i v a l e n c e .
2. C h a r a c t e r i z e all possible s t a t i o n a r y groups a c c o r d i n g to the
classes of equivalence.
3. Find, construct canonical objects, e q u a t i o n s in each class of
equivalent equations.
In w h a t follows we shall answer the m e n t i o n e d questions for
linear differential equations of a r b i t r a r y orders.
First, let us introduce also a g e o m e t r i c a l r e p r e s e n t a t i o n of our
global transformations very useful in the sequel w h e n d i f f e r e n t
geometrical approaches are applied.
Again, let an e q u a t i o n P be r e p r e s e n t e d by its (arbitrary, but
fixed) fundamental solution y, c o n s i d e r e d n o w as a curve in n - d i m e n -
sional vector space Vn, the i n d e p e n d e n t variable x ranging through the
interval I and b e i n g the p a r a m e t e r of the curve. Due to the f o r m (e)
of a global transformation,
the change x = h(t) is only a reparametrization,
the fa'ctor f(t) selects only another curve b u t on the same
cone K formed by straight lines g o i n g through the origin 0 6 V and
n
all points of the o r i g i n a l curve y,
the m a t r i x C performs a centroaffine mapping.
We m a y conclude that each fundamental solution, or curve z of any
equation Q globally equivalent to the e q u a t i o n P is a section of a
cone in n - d i m e n s i o n a l vector space o b t a i n e d as a c e n t r o a f f i n e image of
a fixed c o n e d e t e r m i n e d by a fixed curve y.
Now, let us come to answer the a b o v e m e n t i o n e d basic questions.

III. Global Equivalence

A sufficient and n e c e s s a r y condition for global equivalence of
the second order linear d i f f e r e n t i a l e q u a t i o n s was found by 0. Bor~v-
ka [3] in the sixties. First, some definitions:
The maximal number of zeros of n o n t r i v i a l solutions of an equa-
tion of the second order P2 gives the type of the equation: either
finite, a n integer m, or infinite. Moreover, the e q u a t i o n P2 b e i n g of
finite type m is c a l l e d of general kind, if it admits two linearly in-
dependent solutions w i t h m - i zeros, everything considered on the w h o -
le interval of definition. Otherwise, P2 b e i n g of finite type m is
c a l l e d of S p e c i a l kind. If the e q u a t i o n P2 is of i n f i n i t e type then
its k i n d is either oae-side oscillatory or both-side oscillatory.
Now Bor6vka's criterion reads as follows:
Two second order l i n e a r d i f f e r e n t i a l equations are g l o b a l l y
e q u i v a l e n t i f and only i f t h e y are of t h e sa~¢ type and a t the same
 64

time o f t h ~ same k i n d .
Our criterion of global equivalence of equations of higher
orders needs the following notion. Let
(p) u'' + p(x)u : 0
be an equation of the second order whose coefficient p belongs to the
class cn-2(I), and let u I and u 2 denote two of its independent solu-
tions. Define n functions

Yl :: un-1 n-2 n-1

1 ' Y2 :: ul "u2''''' Yn :: u2

These functions are of the class cn(I) and they are linearly indepen-
dent. Hence they can be considered as solutions of the uniquely de-
termined n-th order linear differential equation, called the i t e r a t i v e
equation iterated from the equation (p). We denote the iterative
equation by p[n] (y,x;I), or simply by p [ n ] The differential expres-
sion of the iterative equation normalized by the unit leading coeffi-
cient will be denoted as {pin] f. It can be shown (e.g. [5]) that

ip[n] r : y(n) + (n+l

3 )p(x)y (n-2) + z<
^,n+l
4 )P'(x)y(n-3 )+''"

In order to find whether two given linear differential equations

of the n-th order, Pn(Y,X;I) and Qn(Z,t;J) with sufficiently smooth
coefficients are globally equivalent, we rewrite them in the form

Pn(Y,X; I) = IP[n] i + rn_3(x)y (n-3) + rn_4(x)y(n-4)+... : 0

and
Qn(Z,t;J) : Iq[ n] I + Sn_ 3 (t)z (n-3) + Sn_4(t)z(n-4)+... : 0 ,

where the first three coefficients of Pn and Q coincide with the

coefficients of the iterative expressions Ip[n~l and fq]n] i, respec~
tively. I f t h e e q u a t i o n Pn i s g l o b a l l y transformable into the equation
Qn b y means o f a g l o b a l transformation with the change x = h(t), then
A. t h e second order equation u'' + p(x)u = 0 on I i s globally
transformable into v'' + q(t)v = 0 on J w i t h the same c h a n g e x ~ h(t)
of the independent variable,
B. t h e following relations are satisfied
rn_3(h(t))h'3(t) = Sn_3(t) on J

rn_4(h(t))h'4(t) = Sn_~(t) on J where Sn_3(t) = 0,

rn_5Ch(t))h'5(t) = Sn_5(t) on J w h e r e Sn_3(t) : Sn_4(t) = 0,

etc.
Due to condition B the criterion is i n general effective, that
 65

means, that it is e x p r e s s i b l e in terms of q u a d r a t u r e s of c o e f f i c i e n t s

of g i v e n equations. L e t us r e c a l l that for the second order equations
the c r i t e r i o n is n o t e f f e c t i v e in this sense, since it r e q u i r e s the
number of zeros of solutions.

IV. Stationary Groups

Stationary groups for the second order equations, called groups
of dispersions, were studied and completely described by O. B o r ~ v k a
[3] in the s i x t i e s . Some results on s t a t i o n a r y groups of linear
differential equations of an a r b i t r a r y order were obtained in 1977
mainly by u s i n g the theory of functional equations [iiI .
In 1979 J. P o s l u s z n y a n d L.A. Rubel [15] characterized (up to c o n -
jugacy) those transformations, called motions, of a l i n e a r differential
equation into itself that consist in a c h a n g e of the independent
variable only.
Finally, in 1984 o n the b a s i s of o u r criterion of g l o b a l equi-
valence a complete c h a r a c t e r i z a t i o n of all possible stationary groups
was derived [14]. Here is the list of the groups up to c o n j u g a c y of
linear differential equations of all o r d e r s considered with respect to
global transformations in the m o s t general form, i.e., involving
changes both the independent and the dependent variables:

a tan x + b
i. The functions h : R ~ R, h{x> = Arctan c t a n x + d' lad - bcl = 1

a tan x
2. h : R + ~ R+, h(x) = Arctan ~ t a n x ~ i/~ , a # 0

3m. F o r each positive integer m, h : (0,m~) - (0,m~),

h(x) = Arctan .......... a tan x

c tan x ~ i/a ' a % 0

4m. For each positive integer m, h : (0,m~ - ~/2) ~ (0,m~ - ~/2),

h(x) = Arctan(k tan x) a n d h(x) = Ar~tan(k cot x), k > 0
5. T h e functions h : R ~ R, h<x) = x + c a n d h(x) = -x + c, c e R
6. T h e increasing functions from 5
7. T h e functions h : R ~ R, h(x) = x + k a n d h(x) : -x + k, k 6 Z
8. T h e increasing functions from 7
9. id R a n d -id R
10. O n l y id R.
These groups range from the maximal one, a three-parameter group
in c a s e i, t h r o u g h an i n f i n i t e cyclic group in c a s e 8, to the t r i v i a l
group in c a s e 10 c o n s i s t i n g from the identity only. L e t me p o i n t out
that the m a x i m a l group has already occured as the fundamental group in
 66

Bor6vka's investigations of the second order equations.

For each case of the stationary groups we can characterize the
corresponding equations and each of the cases listed here actually
occurs. E.g., the case i takes place exactly when the e q u a t i o n is an
iterative equation of an a r b i t r a r y order iterated from a both-side
oscillatory second order equation.
L e t us n o t e that if we c o n s i d e r global transformations with only
increasing c h a n g e s of the i n d e p e n d e n t v a r i a b l e then, up to c o n j u g a c y ,
t h e r e are 5 possible cases of s t a t i o n a r y g r o u p s with respect to lhe
number of parameters as a n n o u n c e d in 1982 [12].

V. C a n o n i c a l Forms
The next important notion is the n o t i o n of c a n o n i c a l forms of
linear differential equations. Such forms were studied from the e a r l y
beginning af i n v e s t i g a t i o n of the e q u a t i o n s in the m i d d l e of the last
century.
W~ h a v e mentioned that already in 1910 G.D. Birkhoff pointed out
that the so c a l l e d L a g u e r r e - F o r s y t h c a n o n i c a l form i s not g l o b a l . It
c a n be s h o w n [13] that also the o t h e r canonical form that has occurred
in the literature, the so c a l l e d Halphen c a n o n i c a l form i s not g l o b a l
either.
For constructions of g l o b a l c a n o n i c a l forms we m a y p r o c e e d in two
ways, either we u s e a c e r t a i n geometrical approach, or w e m a y apply the
criterion of g l o b a l equivalence.
First let us e x p l a i n shortly our geometrical approach. We h a v e
seen that fundamental solutions z, c o n s i d e r e d as c u r v e s in an n - d i m e n -
sional vector space, corresponding to a l l equations globally equivalent
to o n e equation with a fundamental solution Y, a c u r v e y, a r e o b t a i n e d
as s e c t i o n s of a c o n e determined by the c u r v e y. To find a canonical,
that means, a special equation in the c l a s s of e q u i v a l e n t equations,
we n e e d a special section of the cone. By a p p l y i n g Cartan's moving-
frame-of-reference method we come unfortunately again to the H a l p h e n
forms that are not global. However, if w e c o n s i d e r the e u c l i d e a n
n-dimensional space and take the c e n t r a l projection of o u r curves and
then their length parametrization, we o b t a i n special sections of the
cone, special curves. Fortunately, this c a n be d o n e without any restric-
tions on the w h o l e intervals of d e f i n i t i o n . Then by u s i n g differential
geometrical methods the explicit forms of the special, canonical equa-
tions corresponding to the special curves are obtained.
These g l o b a l c a n o n i c a l forms are
n = 2: y'' + y = 0 on (different) I C R,
 67

n = 3: y'''- (p'(x)/p(x))y'' + (i + p 2 ( x ) ) y ' - ( p ' ( X > / p ( x } ) y = O

on I C R,
(one) a r b i t r a r y function p e CI(1), p(x) % 0 on I,
atc.
For n : 2 the canonical equations coincide with the canonical
forms studied by O.Borevka.
There is also another procedure producing global canonical forms.
This p r o c e d u r e is a n a l y t i c a l and the c o n s t r u c t i o n is based on our
criterion of global equivalence. A m o n g many d i f f e r e n t global canonical
forms o b t a i n e d by this a p p r o a c h [13] the following equations

y ( n ) + 0.y(n-l) + l.y(n-2) + Pn_3(x)y(n-3)+...+ Po(X)y : o,ICR,

are global canonical forms.for equations with s u f f i c i e n t l y smooth

coefficients. They are c h a r a c t e r i z e d by their first three c o e f f i c i e n t s
i, 0, 1 .
Comparing with the local L a g u e r r e - F o r s y t h canonical forms having the
corresponding sequence
i, 0, 0 ,
we may conclude that if Laguerre and F o r s y t h had taken 1 as the coef-
ficient of the (n-2)nd d e r i v a t i v e instead of their zero they w o u l d
have got global forms instead of their local.

VI. Invariants
Invariants of linear d i f f e r e n t i a l equations with respect to trans-
formations have been d e r i v e d from the middle of the last century either
directly, or m a i n l y on the basis of the H a l p h e n canonical forms. These
invariants are local.
A global invariant of the second order linear d i f f e r e n t i a l
equations is in fact their t~pe:finite (a positive integer) or infinite,
and their kind, as introduced and derived bv O.B0r~vka in the sixties.
Due to the criterion of global equivalence we have now also
global invariants for equations of an arbitrarq order. Indeed, the
t y p e and k i n d o f t h e equation (p): U'' + p(X)U : 0 on I i s a global
invariant of the n-th order equations P rewritten in the form
n
Pn(Y,X~I) : Ip[n]'(y,x~I)l + rn_3(x)y(n-3)+... = 0 .

Another interesting invariants have o c c u r z e d recently. It is a bit

misleading fact that each second order e q u a t i o n with only continuous
c o e f f i c i e n t s can be globally transformed into an e q u a t i o n with even
analytic coefficients, e.g., into y', + l.y = 0 on some I C R. For
higher order equations the degree of the smoothness of their coeffi-
 68

cients is in some r e s p e c t an i n v a r i a n t property. F r o m m a n y results of

this kind let me introduce at least the following simplest one:
If the c o e f f i c i e n t s of the e q u a t i o n Pn(Y,X~I) satisfy

Pn-i 6 cn-2(I), Pn-2 E cn-3(I),..., pj 6 cJ-I(I) for some

j S n - 1 ,
then the c o e f f i c i e n t s of any g l o b a l l y equivalent equation to the
Pn(Y,X;I) have t h e same o r d e r o f d i f f e r e n t i a b i l i t y .
VII. Equations with Solutions of P r e s c r i b e d Properties
The m a i n idea h o w to c o n s t r u c t linear differential equations with
solutions of some p r e s c r i b e d properties is based on the following
"coordinate approach".
Having global canonical forms (the g l o b a l i t y is essential), each
linear differential equation P of an a r b i t r a r y order can be "coordi-
nated" b y a couple {S,~} consisting of its global canonical f o r m S and
of the global transformation ~ converting S into P, i.e., P = Se.
If we succeed to r e f o r m u l a t e a given property of solutions of P
equivalently into p r o p e r t i e s of S and ~, we may c o n s t r u c t all required
equations. Also problems concerning relations among certain properties
are then c o n v e r t e d into (sometimes simple, or e v e n a l r e a d y solved)
problems from the t h e o r y of functions.
By u s i n g this a p p r o a c h there w e r e constructed linear differential
equations that have important applications in d i f f e r e n t i a l and integral
geometries. E.g., it was p o s s i b l e to g e n e r a l i z e Blasehke's and S a n t a l J ' s
isoperimetric theorems, [8].
Connections between boundedness of solutions and their L2-pro -
perties were easily explained by the a b o v e m e t h c d [7].
Relations between distributions o f z e r o s and a s y m p t o t i c behavior
of the solutions were also d e e p l y studied by means of the c o o r d i n a t e
approach.
There is also a n o t h e r way, a geometrical one, h o w to see what
happens with zeros of solutions and how to c o n s t r u c t equations with
prescribed distribution of zeros of their solutions.

VIII. Zeros of S o l u t i o n s
This qeometrical approach is b a s e d on the r e p r e s e n t a t i o n of a
fundamental solution y of an e ~ l a t i o n Pn(Y,x;I) as ~ curve in n-dimen-
sional vector or even e u c l i d e a n space V mentioned in the p r e v i o u s
n
sections.
Let the curve v be the central nro~ection of the curve y onto the
 69

unit sphere S n _ 1 in the space V n without a change of p a r a m e t e r x. E a c h

solution y of P n ( Y , X ; I ) c a n be w r i t t e n as a s c a l a r product c . y where
c is a n o n z e r o constant vector in V . L e t H(y) denote the hyperplane
n
H(y) := {d E Vn~ c . d : 0}
going through the o r i g i n and corresponding to the v e c t o r c. E v i d e n t l y
0 = y ( x O) : c . y ( x O) = c . V(Xo)lY(Xo)I ~ c . v ( x O) : 0
since lY(Xo)i * O. T h u s we h a v e shown that
to each s o l u t i o n y of t h e e q u a t i o n p t h e r e c o r r e s p o n d s a h y p e r -
n
plane H(y) i n V going t h r o u g h t h e o r i g i n such t h a t
z e r o s o f t h e s o l u t i o n y o c c u r as p a r a m e t e r s of i n t e r s e c t i o n s of
t h e p a r t i c u l a r h y p e r p l a n e H(y) w i t h t h e curve Y~ and v i c e v e r s a .
Multiplicities of z e r o s o c c u r as orders of c o n t a c t s , [9].
Let us r e c a l l that all this happens on the u n i t sphere, a compact
space, where strong topological tools are at our d i s p o s a l .
Several open problems were solved and m a n y complicated constuc-
tions were easily explained by u s i n g this approach, [ i0] . As a s i m p l e
demonstration of the m e t h o d let us p r e s e n t Sansone's result by con-
structing a third order l i n e a r differential equation with all oscil-
latory solutions.
For this purpose it is s u f f i c i e n t to h a v e an e n o u g h smooth (of the
class C 3) c u r v e u on the u n i t aphere S 2 in 3-dimensional space w i t h o u t
points of inflexion (that m e a n s , that Wronskian of u is n o n v a n i s h i n g )
such that each plane going through the o r i g i n intersects u for infini-
telv many values of p a r a m e t e r . The picture of a closed "Drolonqed
cycloid" infinitely manv times surrounding the equator as its parameter
ranges f r o m - ~ to +°° m a y serve as an e x a m p l e of a c u r v e with the
required property.

IX. A p p l i c a t i o n s
To the e n d of m y survey let me m e n t i o n some fruitful applications
of the p r e s e n t e d theory.
The above methods were succesfully applied to s y s t e m s of l i n e a r
differential equations. E.g., construction of c e r t a i n second order sys-
tems w i t h only periodic solutions, [ i0] , p l a y s an important role in geo-
metry of m a n i f o l d s whose all geodesics are closed [21.
By u s i n g the a b o v e approach there were solved some problems con-
 70

cerninq l i n e a r and n o n l i n e a r d i f f e r e n t i a l e q u a t i o n s and s y s t e m s w i t h

one or s e v e r a l d e l a y s . There are useful applications in generalized
differential equations and l i n e a r d i f f e r e n t i a l expressions with quasi-
derivatives as w e l l . Last but not least, there are many fruitful
connections with the theory of functional equations.

References
[ i] Birkhoff, G.D.: On the s o l ~ t i o ~ of ordinary linear homogeneous d i f f e r e n t i a l
equations of the ~nird ord~, Annals of M a t h . 12 (1910/11) , 103-124o
[2] Besse, A.L.: ~a~ifoldS All of Whose Geodesi~-~e Closed, Ergenisse, vol. 93,
Springer, Berlin ~ New York, 1978.
[3] Bor~vka, 0.: ~near differentiat~ansformationen ~. Ordnu~g, VEB Berlin 1967;
Linear Diff~e~tic~ TRansformations of the Second Order, ~ e English Univ.
Press, London 1971.
[4] Cadek, M.:A form of general poi~/d~i~e t/u~nsforma/oLo~ of line~ different~
eqb~ons, Czechoslovak Math. J. (in print)o
[5] n u s t ~ , z . : Die Iteration homogen~ linear D i f f e r e n t i a l g l e i ~ u n g e n , Publ. Fac.
Sci. Univ. J.E. Purkyn~ (Brno) 4__4--9(1964), 23-56.
[6] Kummer, Eo : ~e gener~i q u ~ ~ q u ~ o n e differen~o~i tert~ ordi~, progr.
Evang. Konigl. & Stadtgymnasiums Liegnitz 1834~
[7] Neuman, F.: Relation b ~ e e n the d ~ t ~ b ~ i o n o~ the zeros 0~ the solutions of
a 2nd order linear d i f f e r e n t i a l equation and the boundedness of these s o l u t i o n ,
Acta Math. Acado Sci. Bungar. 19 (1968), i-6.
[8] Neuman, F.: Linear differen2i~-equat~o~ of the secJgnd ord@/% a ~ their
applications, l~end. Math~ 4 (1971), 559-617o
[9] Neuman, F.: Geometrical approach to linear diff~e~tial e q ~ o ~ of the n-th
order, Rend. Mat. 5 (1972), 579-602.
[I0] Neu~nan, F.: On ]J~o-proble~ abou~ oscill~on of linear diffe~%e~ial equiz~ions
of the th/rd order, J. Diff. Equations 15 (1974), 589-596~
[ Ii] Neuman, F.: On solutions of the vector ~ctional equation y(~(x)) =
= f(x).A.y(x), aequationes Math. 16 (1977), 245-257.
{ 12] Neuman, F.: A survey of global p r ~ t i e s of linear differential equations of
the n-th ord~/%, in: Lecture Notes in Math. 964,543-553.
[!3] Neuman, Fo: G10boLf_canonical for~ of ~ne~--~iff~re~ti~ equations, Math.
Slovaca 33, (1983), 389-394.
[ 14] Neuman, F-~: StationaAy groups of linear differential eq~l~ons, Czechoslovak
Math. J. 34 (109) (1984), 645-663.
[ ~5] Posluszny,---~. an----dRubel, L.A.: The ~ot~on of an o r d i ~ y d i f f e r e ~ eq~a~on,
J. Diff. Equations 34 (1979), 291-302.

Details will aDDear in

Neuman,F.:0rd~nar~ Linear Differe~tialEQuations, Academia Publishing House,

Prague & North Oxford Academic Publishers Ltd., Oxford.
NUMERICAL AND THEORETICAL
TREATING OF EVOLUTION PROBLEMS
BY THE METHOD OF DISCRETIZA ON
IN TIME
K. REKTORYS
Technical University Prague
Thdkurova i. 166 29 Prague 6, Czechoslovakia

More than fifty years ago, E. Rothe suggested an a p p r o x i m a t e method of

solution of p a r a b o l i c problems. He d i v i d e d the interval I = [0,T 3 for
the v a r i a b l e t into p subintervals I. of the l e n g t h h = T/p and
J
at e a c h point t = jh , j = I .... ,p , he a p p r o x i m a t e d the f u n c t i o n
J
u(x,tj) by the function zj(x) and the d e r i v a t i v e 8u/St by the d i f -
ference quotient Ezj(x) - zj_1(x~/h . Starting with z0 given by
z0(x) = u(x,0) = u0(x) , he found, successively for j = I ..... p , the
approximations z (x) as s o l u t i o n s of the so a r i s e n ordinary boundary
3
value problems. The p r o b l e m , solved originally by E. Rothe, was a very
simple one. However, his m e t h o d turned out to be a v e r y useful tool for
solution of substantially raore c o m p l f c a t e d evolution p r o b l e m s (at first
linear and q u a s i l i n e a r parabolic problems of the second order in n di-
mensions later parabolic problems of a r b i t r a r y order, nonlinear prob-
lems hyperbolic problems, the Stephan problem, integrodifferential prob-
lems, mixed parabolic-hyperbolic problems, etc.). The development of the
Rothe method, called also the m e t h o d of d i s c r e t i z a t i o n in time, or the
horizontal method of lines, is c o n n e c t e d with such names as O. A. L a d y -
~enskaja, T. D. V e n t c e l , A. M. Iljin, A. S. K a l a ~ n i k o v , O. A. O l e j n i k ~
J. I. I b r a g i m o v , P. S. M o s o l o v , O. A. Liskovec, R. D. R i c h t m a y e r ~ N. N.
Jan~nko, M. Zl~mal, J. Ne~as, J. Ka~ur, A. G. Kartsatos, M. E. Parrot:
W. Ziegler, J. W. J e r o m e , E. M a r t e n s e n a n d his school~ U. v. Welck, J.
Naumann, C. C o r d u n e a n u , etc. Theoretical as w e l l as n u m e r i c a l questions
have been examined (existence and convergence theorems~ regularity ques-
tions, numerical aspects, etc.). Many of the o b t a i n e d results were obtai-
ned as w e l l by o t h e r methods - method of c o m p a c t n e s s , theory of s e m i -
groups, method of m o n o t o n e operators, Fourier transform, etc. (A. F r i e d -
man, M. Krasnoselskij, P. E. Sobolevskij, F. E. B r o w d e r , J. L. Lions,
E. M a g e n e s , H. Br4zis, V. Barbu, D. P a s c a l i ~ M. G. C r a n d a l , W. v. Wahl,
etc.). As c o n c e r n s numerical methods, related to the Rothe method, the
methods of s p a c e - or t i m e - s p a c e discretization were applied (V. N.
Fad6jeva, J. D o u g l a s , T. D u p o n t , M. Zllmal, R. G l o w i n s k i , J. L. Lions,
 72

R. T r e m o l i ~ r e , P. A. Raviart, W. W a l t e r , K. GrSger, etc.). Each of the

mentioned methods, including the R o t h e method, has its p r e f e r e n c e s and
its d r a w b a c k s . However, the R o t h e method has its significance both as
a numerical method and theoretical tool. Existence theorems are p r o v e d
in a c o n s t r u c t i v e way. Thus no o t h e r methods are n e e d e d to g i v e preli-
minary information on e x i s t e n c e , or r e g u l a r i t y of the solution as re-
quired in m a n y other numerical methods when questions on c o n v e r g e n c e , or
order of c o n v e r g e n c e , etc. are to be a n s w e r e d . The Rothe method is a
stable method. To the solution of e l l i p t i c problems generated by t h i s
method~ current methods, especially the v a r i a t i o n a l ones, c a n be a p p l i e d .
As concerns theoretical results, they are obtained in a r e l a t i v e l y sim-
ple way, as u s u a l . Moreover, the Rothe method, being a very natural one,
makes it p o s s i b l e to get a particularly good insight into the structure
of the solutions. Often a brief inspection of the corresponding elliptic
problems gives an information what can be e x p e c t e d as c o n c e r n s proper-
ties of the solution. This is w h y I prefer it.

In 1971, a slightly different technics than that applied currently in

this m e t h o d appeared in my w o r k [2j, m a k i n g it p o s s i b l e to t r e a t corres-
ponding elliptic problems in a p a r t i c u l a r l y simple way. This technics
was followed by o t h e r authors (in o u r country J. Ne~as, J. Ka~ur) and
became a base for w o r k of my seminar at the T e c h n i c a l University in
Prague. Results obtained in this seminar were summarized in m y m o n o g r a p h
[lJ in 1982. I would like to p r e s e n t some of t h e m here, pointing out
the v e r y simple way in w h i c h they have been obtained.

.i!i. E x i s t e n c e and convergence theorem. Let us s t a r t w i t h a relatively

simple parabolic problem
8u
8-~ + Au = f in G x (0,T) , (I)

u(x,o) = o , (2)

B.u = 0 on F x (0,T) , i = I,...,~ , (3)

l
C.u = 0 on F X (0,T) , i = 1,...,k-~ . (4)
1

Here, G is a b o u n d e d region in EN with a Lipschitz boundary F ,

A = ~ (-I) lil D i ( a i j ( x ) D J 1 (5)

Lit,tJt<k
with aij bounded and measurable in G , f 6 L2(G) ; (3), or (4) are
(linear) boundary conditions, stable (thus c o n t a i n i n g derivatives of
orders ~ k - I ), or u n s t a b l e with respect to t h e o p e r a t o r A , respec-
tively. Denote
 73

V ri v; V e w ~(~k ) ( G ) , Biv = 0 on F in the sense of traces,

(6)
i = 1,...,p}

let ((.,.)) be the bilinear form, corresponding to t h e operator A and

to the boundary conditions (3), (4), familiar from the theory of varia-
tiinal methods. (Roughly speaking, ((v,u)) is o b t a i n e d of (v,Au) , ap-
plying to every integral / v D i ( a i j D3u)- d x i-times the Green theorem
G
in the usual way, see e.g. [I~, o r [3 3 . F o r e x a m p l e , if A = - A and
u = 0 on F is p r e s c r i b e d , then

V = W 1)(G) and ((v,u)) = i=1~ 3xi 1 dx .)

Let the form ((.,.)) be hounded in V x V and V-elliptic, i.e. let two
positive constants K and ~ (independent of v and u ) exist such
that the inequalities

l((v,u))~ ~JlVllvllUIIv (7)

((v,v)) m ~llvllv2 (8)

hold for all v, u 6 V . Let to the ,solution of the problem (I) - (4)
the Rothe method be applied. Denote

zi(x ) - Zi_1(X)
Zi(x) = h , i : I ..... p . (9)

(Thus Zi(x) "corresponds" to the derivative 8u/~t for t = ti .) I n

the weak formulation, we have to find successively for j = I .... ,p ,
the functions z. e V , s a t i s f y i n g ,
3
((v,zj)) + (v,Zj) = (v,f) V v 6 V , (10)

with z0(x) = u(x,0) = 0 . Under the assumptions (7), (8), each of these
problems has exactly one solution z. ~ V . Apriori estimates: Put v =
3
Z1 = (z I - z 0 ) h = Zl/h into (10) written for j = I . We obtain

h((ZI,Z1)) + (ZI,ZI) = (Z1,f) (11)

Because of (8) and I (Zl,f) I & I IZll I If[l , (11) yields

I iZll I2 =< I IZl[ [ I If[ [ => I IZII =< IIfl I . (12)

Subtracting (I0), written for j - I , from (10) gives

h ( ( v , Zj)) + (v, Zj - Zj_1) = 0

Putting v = Z. , w e obtain, in a s i m i l a r way as b e f o r e ,

3
llzjlZ m IfZj_llr , j = 2,...,p
 74

what gives, together with (12)

Itzjll < IIfll : cI (13)

Let us refine our division, considering the divisions dn with the

steps h n = hl/2n-1 , n = I , 2,. .- , hI = h . Denote the corresponding
functions
n n
z - z.
Z
n. ,
Zn = J 3-1
3 ] h
n
The estimate (13) having been obtained independently of the lengh of the
step h , it r e m a i n s valid as w e l l for the division dn ,

I Iz~.ll < cI (14)

3 =
Because z n3 = (zjn _ z n3- i) + ... + (z nI - z0)
n , it follows

II=nllu <= Jhn(ilzJ I1 + "'" + IlZl It) --< Tel = c2 . <15)

Putting then v = z n. into (10) written for the functions zn and Zn

3 3 ]
and using (8), we get

l lznl
3
Iv =< c3 (16)
(14), (15) and (16) are the basic needed a priori estimates. They have
actually been obtained in a very, very simple way. What follows, is a
standard procedure, now. Let

t - tn
n n n j-1
Un(t) = zj_ I + (zj - zj-1 ) - h (17)
for tnj-I < t <= tn3 ' J = I, .... p . 2 n-1

(n = 1,2,... ) (the so-called Rothe functions), or

n
/Z I for t = 0 ,
Un(t)
z9 for t~ = t~ j = I .... p - 2 n-1 (18)
3 3 -1<t< 3' '
(n = 1,2,... ) be abstract funcions, considered as functions from I =
~,~ into V , Or L2(G ) , respectively. In consequence of their form
and of (16) and (14), they are uniformly bounded (with respect to n ).
in L2(I,V ) , or L2(I,L2(G)) , respectively (even in C(I,V) , or
L (I,L2(G))). The space L2(I,V ) , and L2(I,L2(G)) being Hilbert spa-
ces, a subsequence {unk } , or {Unk } can be found such that

u n k --~ u in L2(I,V) , Unk -~ U in L2(I,L2(G)) (19)

Now, (17), (i80 imply

 75

t
U (T)dT : U (t) vn
n n
0

yielding easily
t
/ U(T)d~ : u(t) ,
0

and thus U : u' in L I ( I , L 2 ( G ) ) . Consequently, u 6 AC(I,L2(G)) and u(0):

: 0 in C ( I , L 2 ( G ) ) . So the function u satisfies

u E LiI,V) A AC(I,L2(G)), (20)

u' = U • L2(I,L2(G)) , (21)

u(0) : 0 in C ( I , L 2 ( G ) ) . (22)

L4oreover, on m a s e of i n t e g r a l identities (iO) and of (19), one comes

to the integral identities
T T T
j ((v,u)) dt + / (v,u') dt : / (v,f) dt. (23)
0 0

A function with the p r o p e r t i e s (20) - (23) is c a l l e d the w e a k solution

of the p r o b l e m (i) - (4).
Uniqueness: L e t Ul, ul b e two functions satisfying (20) - (23). Then
their difference u = u 2 - u I has the p r o p e r t i e s (20) - (22) and satis-
fies
T T
J ((v,u)) dt + I (v,u') dt : 0 v v E L2(I,V).
0
Let a E I be arbitrary. Choose
/u(t) for 0 < t < a,
v(t)
=~0 for a < t < T.
We h a v e
T a
][ (v,u') dt = / (u,u') dt = ~li lu(a) i 12 _ ~iI lu(0) i 12 = ~llu(a
) I [12.
0 0

In c o n s e q u e n c e of (8) we thus obtain flu(a) 112 = 0; the p o i n t a having

been chosen arbitrarily, u = 0 in I.
Uniqueness implies in the familiar way that u ~ u (not o n l y u ~ u)
n nk
in L2(I,V). Moreover, for e v e r y t 6 I the sequence {Un(X)} is b o u n d e d
in V and thus compact in L2(G). The functions Un(t) being uniformly
 76

bounded in i, in t h e m e t r i c of the space L2(G) , and equicontinuous on

base of (13), the Ascoli theorem can be applied, implying (strong)
uniform convergence, in I, o f {Un] to u.
Sum/narising, we thus have:

Theorem i. Let (7), (8) be satisfied, let f E L2(G). Then there exists
exactly one weak solution of the problem (I) - (~) and

un u in L 2 ( I , V ) , u n ~ u in C ( I , L 2 ( G ) ) (24)

Remark 1. By a more detailed treatment it can be proved that even un

u in C ( I , V ) , u' E L (I,L2(G)). We shall not go into details here.
See [ i] .

2. Error Estimates. As can be excepted, to get an e f f i c i e n t error es-

timate, some supplementary assumptions are needed: Let the assumptions
of Theorem 1 be fulfilled. Let, moreover,

f E V, Af E L2(G), ((v,f)) : (v,Af) v v E V. (25)

k~en 2
Mjh
lu(x,tj) - z j ( x ) ll £ 2 ~, , j = 1 ..... p, (26)

where M : IIAf I !. If, moreover, the coefficient C of positive defini-

teness can be easily found, for which thus

((v,v)) h c211vll 2 v v ~v

holds, then the following (slightly better) estimate can be used:

_C 2
I lu(x,tj) - zj(x)I I < Mh (i - e jh), j = 1 ..... p. (27)
= 2C 2

Proof of (26) (the proof of (27) is similar): Let us investigate the

division di (for the notation see the text following (13)) and denote

2 i
z2i zi = qi' i = 0,1,...,p

(with z}l = z ).i The functions 2

z2i satisfy the integral identities

2 1 2 2
((v,z2i)) + ~-/~ ( V , Z 2 i - Z2i_l) : (v,f) ~ V E V.

Subtracting the integral identity (i0), with i written for j, w e obtain

((V,Z2i - Zi)) + (V, (Z i - Z.)l - (z22i-2-zi-i)) :

~(v, 2 _ ~ 2 2
z2i 2z2i_i + z2i_2) v v E V,

or, denoting
 77

2 ~ 2 n n n
s~ = z i - 2z -I + zi-2 (s~ : zi - 2 Z i - l + zi-2 in g e n e r a l )
(h/2)~ 1 h2 --' '
n

l(v 1 1 ) : ~ 2
((v,q~)) + ~ 'qi - q i - i - (V,S2i) v V E V,

with q~ : z~ - z 0 : 0. P u t t i n g v : q~ for i : 1 and t a k i n g (8) into

account, we g e t
h2 2 h2M
ILq~ll~T [~s211~ 4 ,
because under the a~sumptions (28)
n
I Isi I I ~ M for all n and i.

Similarly, for i : 2 we obtain

h2M 2h2M
qlq~lJ~ llq~II+~ 4 ,
and, finally,
ih2M

n n+l n
In g e n e r a l , we h a v e , for qi : z2i - zi '
~n i(h/2n-l)2M ,... 2n-i
i lqill ~ . . . . 4 , i : o,i , p.

NOW,

IlUn(X,tj) - z . ( x ) I I : IlZ~n_lj z~ II <

3 3 =

I n-I + n-2 +'''+ 2

q2j + qji I I <
lq2n-2 j q2n-3j =
• 2n-l.
~-r- (J + 22-~+- -+

Coming to the limit for n ~ ~ (what is a l l o w e d because of T h e o r e m i),

we o b t a i n (26).
The estimate (26) is v e r y sharp (and the m e r e is the e s t i m a t e (27)), as
can me seen from the following simple example:
o u _ ~2u : sin x in (0,~) x (0,7), (28)
~t 8x 2 i

u(x,O) : 0, (29)
u(O,t) : O, U(~,t) : O. (30)

The assumptions (25) are e a s i l y established. Further, we h a v e

 78

M = IIAfll = II- (sin x)''II = llsin xll = J(~/2)

Choosing, for, example, h = 0.01 and j = 20, or j = 40, the Rothe

method yields
z20 = 0.1805 sin x ,
z40 = 0.3284 sin x ,

respectively. The exact solution is k n o w n (this was the reason why such
a simple example has been chosen): u = I - e -t) sin x . Thus

u(x, 0.20) = 0.1813 sin x ,

u(x, 0.40 = 0.3297 sin x .

The actual errors then are

I u(x, 0 2 0 ) - z20 I = 0.000811sin x I = 0.00101 ,

(31)
I u(x, 0 4 0 ) - z401 = 0.001311sin x I = 0.00163 ,

while (26) gives

20.0.012
i u(x, 0.20) - ' z 2 0 I 2 $~ = 0.00125 ,
(32)
40.0.012
flu(x, 0.40) - z4011 £ 2 4~ = 0.00251 .

Finding C = 1 (see [lj, p. 90) and using (27), we get the estimates

Ilu(x, 020) - z2011 < °i°----!

(1 - e-°'2)#~ = 0.00113
(33)
IluCx, 0.40) - z4011 a g~QJ (1 - e-0"4)$~ = 0.00206

which are still better then the estimates (32). The example demonstrates
very well the sharpness of the estimates (26), (27) and the fact that
they cannot be substantially improved.

3. Nonhomo~eneous initial and bounda[y conditions. Let us first in-

vestigate the problem (I) - (4) w i t h homogeneous equation and nonhomo-
geneous initial condition u(x,0) = u0 ~ L2(G ) , i.e. the problem
8u
8-~ + A u = 0 in G x (0,T) , (34)

u(x,0) = u0(x) (35)

Biu = 0 on £ x (0,T) , i = I,...,~ , (36)

Giu = 0 on F x (0,T) , i = 1,...,k-~ . (37)

The corresponding integral identities are

((v,zj)) + (v,Zj) = 0 V v C V , j = I ..... p , (38)

 79

with Zj = (zj - z j _ 1 ) / h , z0 = u 0 . (39)

Similarly as in the case of the p r o b l e m (I) - (4), w e c o m e to t h e inequa-

lity
llzjll ~ IIzj iii
However, if o n l y u 0 6 L2(G) is a s s u m e d , it is n o t p o s s i b l e to p u t v =
= ZI into the first of the integral identities (38) to o b t a i n a simple
estimate for I [ZII I as in (11), (12), because we h a v e not ZI C V here,
in g e n e r a l . Thus an e x i s t e n c e theorem is d e r i v e d , first for "suffici-
ently smooth" u 0 = s E V, m o r e precisely for u 0 from the set M of such
functions s @ V for w h i c h a unique g E L2(G ) e x i s t s satisfying

((v,s)) : (v,g) v v e V.

Putting then

z. : r + s
3 j
into the integral identities

((v,zj)) + h(V,Zj - z9_ !) : 0 v v E V,

with z 0 -- s, we come to the identities

((v,rj)) + h(v,rj - rj_l) = -(v,g) v v e V,

with r 0 : 0, c o r r e s p o n d i n g to the p r o b l e m (I) - (4) in w h i c h u is

replaced by r and f b y -g. Having obtaiMed its w e a k solution r(t), the
weak solution of (34) - 437) w i t h u 0 : s 6 M is d e f i n e d b y u(t) : r(t)+
+ s. M o r e o v e r , if we p u t z. for v into the original integral identities,
3
we obtain, subsequently,

I IZl I I < ! Is I I, I Iz2 I I < Ilzl I I < Ils ] I, etc.

The function u(t) being the limit, in C ( I , L 2 ( G ) ) , of the corresponding

Rothe sequence, it f o l l o w s

liU(t) I I ~ Ils II f o r all t @ I.

Now, the form ((v,u)), being V-elliptic, the set M is d e n s e in V, thus

as w e l l in L2(G). Let u 0 E L2(G ) and let s i 6 M, i = 1,2,..., be Can
arbitrary) sequence converging to u 0 in L 2 ( G ) . Then the sequence of
corresponding weak solutions u(i)(t) is a C a u c h y sequence in
C(I,L2(G)) , because

llu(J)(t) - u(k)(t) II ~ I Isi - Skl I ~ t E I.

 ~0

Its limit is then called the very weak solution of the problem (34) -
(37). Omviously, this very weak solution is u n i q u e l y determined by
the initial function u 0 E L2(G).
Amour convergence, in C(I,Li(G)) , of the corresponding Rothe sequence
to this very weak solution as w e l l about nonhomogeneous boundary con-
ditions see [i].

4. The Ritz-Rothe method. Let us investigate the problem (34) - (37)

with f 6 L2(G) on the right-hand side of (34) instead of zero. The so-
lution u(t) of this problem is the sum of solutions of the problems
(I) - (4) a n d (34) - (37). (The problems are linear.) The corresponding
integral identities when applying the Rothe method are:

((v,zj)) + ~I (v, zj - z j _ 1) = (v,f) ~ v 6 V , j = I , . . . ,p , (40)

with z0 = u 0 . Let us solve each of these problems approximately - to

be concrete, by the Ritz method (or b y a method with similar properties).
So let v 1,...,v n be the first n terms of a base in V and let zI
be the Ritz approximation of the function z I . Put zI instead of zI
into the second of the identities (40),
m I m *
((v,z2)) + ~ (v, Z 2 - Zl) = (v,f) (41)

and let z2 be the Ritz approximation of the function z2 , etc. We

thus can construct the function
z. - z.
• * .3 - 1
u 1(t) = zj_ I + h - (t - t j _ I) for t j _ I <= t < tj , (42)
j = I ..... p

which is a n a n a l o g u e of the Rothe function u1(t ) (41) announces that

using the Ritz method, the errors become cumulated with increasing j •
Fortunately, according to a very simple law: Subtract (41) from the se-
cond of the identities (40). We obtain

((v, z 2 - Z2)) + (v, (z 2 - z2) - (z I - Z l ) ) = 0 V V ~ V .

%
Putting v = z 2 - z2 , we get

iJz2- ~211 a 11zi - ~iII

Etc. Using this result, convergence of this "Ritz-Rothe" method is
easily proved: Let ~ > 0 be given. According to Theorem i, such a
(fine) division of the interval I into p subintervals can be found -
let us preserve the notation h for the length of these subintervals -

that
 S1

I lU(t) -- Ul(t) II < ~ v t @ I

(where u l ( t ) is the c o r r e s p o n d i n g Rothe function), Denote q = e/2p.

Let the n u m b e r of terms in the Ritz approximation be sufficiently
large so that

~Iz I - z~ll <

oe fulfilled. Then - as just shown -

llz2- z21t < n.

Let the R i t z approximation z~ of ~2 be such that

again. Thus

IIz2- z%11 < 2~.

In a s i m i l a r way we come to the e s t i m a t e s

II~j- ~ll<J~ :~, j : 3 . . . . . p.

Because of the 'form of the Rothe functions Ul(t) and u~(t) (they are
piecewise linear in t), we h a v e

] lU(t) - U~(t)I I < ~ V t E I.

5. Regularity of the solution, a) Regularity with respect to t ,

smmoothing effect. In [II , Chap. 12 a n d 13, regularity properties of the
weak, or v e r y weak solutions with respect to t are examined. We shall
not go into details and show the v e r y simple idea of t h e s e investiga-
tions on t h e e x a m p l e of the p r o b l e m (34) - (37). Let the form ((.,.))
satisfy (7) and (8) (boundedness and V - e l l i p t i c i t y ) and let it be, more-
over, syn~netric in V , i.e. let

((v,u)) = ((u,v)) ~ v, u 6 V (43)

be fulfilled. Thus ((.,.)) has the p r o p e r t i e s of a s c a l a r product. Let

h be s u f f i c i e n t l y small (in o r d e r that the p o i n t s t o , 2t 0 , etc., in-
vestigated below, lie in the interval [0,T] ) and choose an a r b i t r a r y
tO 6 (0,T) such that t o = jh ( j being a positive integer). Take
the first of the integral identities (38) and p u t v = z I . We o b t a i n

h((Zl,Zl)) + (z I, Z I - u 0) = 0 .

I 2
Writing (z I, z I - u0) in the form ~ (I IZll I + I IZl - u 0 I2 -

- Jlu0J I2) , we g e t

h((Zl'Zl)) + ~1 IlZl I i2 a ~
1
ItUotl 2
 82

Similarly,
1 12 1 2
h((z2'z2)) + 2 llz2 I <~ 2 llZl I I '

I 12 I 2
h((zj,zj)) + 2 I Izj I ~ 2 I Izj-111
Making the sum, we o b t a i n

J
h ~ ((zi,zi)) <__~i flu01 12 (44)
i=1
Putting, in the s e c o n d of the i d e n t i t i e s (38), V = z 2 - z I , w e get,
similarly,
I
((z 2 - z I, z2)) + [ (z 2 - z I' Z 2 - z I ) = 0 ,

I [( (z2,z 2 ) ) + ( (z 2 - z I , z 2 zl)) - ((z1'zl))] ~ 0 ,

((z2,z2)) < ((Zl,Zl)) .

G o i n g on in this way, w e o b t a i n

((zj,zj)) ~ ((zj_1,zj_1))~ ... ~ ((z2,z2)) ~ ((Zl,Zl)) . 445)

Thus r e p l a c i n g in 444) all the s u m m a n d s by ((zj,zj)) and t a k i n g into

account that jh = t O , we h a v e

((zj,zj)) < ~ flu0112 (46)

= 2t 0
and, because of the V - e l l i p t i c i t y of the f o r m 44.,.)) (see 48)),

II~jLIv ~ #(2~t o) Iluol I 447)

In c o n s e q u e n c e of 445), this r e s u l t h o l d s for all l a r g e r indices, too,
and a l s o for all d i v i s i o n s d with n > I ,
n =

< ~
I Iz~IIV = ~ ( 2 e t 0)
I lUol I ~ t~ ~ t O
1
(48)
U s i n g this result, interchanging the role of zi and Zi and a s s u m i n g
2t 0 6 (0,T) , we get, similarly,

< --I-- I lu011 V tn 2t0 449)

IIz~ll = 2t 0 i

A simple consideration leads then to the c o n c l u s i o n that for the re-

strictions ~(t) a n d ~'(t) of the f u n c t i o n s u(t) and u'(t) on the i n t e r -
val [2t0,T] , (48)~ (49) imply

6 L2<~t0,T],V ) , u' 6 L 2 ( ~ t 0 , T ] , L 2 ( G ) ]
G o i n g on in the same way, we p r o v e similarly (assuming 4t 0 ~ (0,T) )

~' 6 L 2 ( ~ t 0 , T ] , V ) , ~'' 6 L2[~t0,T],L2(G)] ,

 83

etc. Let n 6 (0,T) be a r b i t r a r y , q > 0 an a r b i t r a r y integer. Chosing

t O ~ n/(2q + 4) , we come, in this way, to the result that

~(i+I) ~ L2(~,~,V 1 ~(i+2)

what implies, among others,

~<i) c Ac([n,T3,v] , ~(i+I) G AC[[~,T~,L2<G)] , i = 0 ..... q

The numbers n and q having been chosen arbitrarily, we have come,
in this v e r y simple way, to the f o l l o w i n g

Theorem 2. L e t (7), (8), (43) be f u l f i l l e d , u0 { L2(G) . Then the v e r y

weak solution u(t) of the p r o b l e m (34) - (37), considered as an a b s -
tract function ~,T] + V , has on the interval (0,T] continuous de-
rivatives of all orders.

L e t us r e m a r k that applying the just shown idea in a p r o p e r l y modified

way, J. Kadur obtained rather strong regularity results for the equation
8u/St + A(t)u = f(t) . See [43 .

b) Re@ularity with resp@ct to x . While the b a s i c method how to e x a -

mine regularity with respect to t has b e e n shown in [I], the idea how
to o b t a i n regularity results with respect to x belongs to J. Ka~ur.
Regularity results known for e l l i p t i c problems are utilized. For details
see ~3-

6. O t h e r parabolic problems. Using the same technics as above, linear

parabolic equations of the form ~u/~t + A(t)u = f(t) , nonlinear equa-
tions, integrodifferential equations as w e l l as some nontraditional prob-
lems (problem with an integral condition, for e x a m p l e ) c a n be e x a m i n e d .
For details see [I~.

7. H y p e r b o l i c problems. Also hyperbolic problems c a n be t r e a t e d in the

same way. Under the a s s u m p t i o n of b o u n d e d n e s s in V x V , V-ellipticity
and V - s y m m e t r y of the form ((.,.)) , an e x i s t e n c e and convergence theo-
r e m has been derived, in [I~, a n d convergence of the "Ritz-Rothe method"
proved. Moreover, for f [ V , the following error estimate has been
found, in a s i m i l a r way as in the case (28):
I lu(x,tj) - zj(x) I I ~ Mh 3 j(j + I) with M = $(~ ((f,f))]. (50)

For some regularity results see [53 . F o r generalization to the c a s e of

quasilinear hyperbolic equations see [~ .

R e f e r e n c e s

D~ R E K T O R Y S , K.: The Met)~od of Discretization in Time and Partial Dif-

ferential Equations. Dordrecht-Boston-London, D. R e i d e l 1982.
 ~4

[23 REKTORYS, K.: On Application of Direct Variational Methods to the

Solution of Parabolic Boundary Value Problems of Arbitrary Order.
Czech. Math. J, 21 (1971), 318-339.
~3] REKTORYS, K.: Variational Methods in Mathematics, Science and Engi-
neering, 2nd Ed. Dordrecht-Boston-London, D. Reidel 1979.
[~ KA~UR, J.: Method of Rothe in Evolution Equations. Leipzig, Teub-
ner. To appear.
[~ PULTAR, M.: Solution of Abstract Hyperbolic Equations by Rothe Me-
thod. Aplikace matematiky 29, (1984), 23-39.
ALGORITHMS FOR THE INCLUSION
OF SOLUTIONS OF ORDINARY INITIAL
VALUE PROBLEMS
H. J. STETTER
Technical University Vienna
A-1040 Wien, Austria

Introduction

Customary numerical algorithms do not produce bounds ~or the true so-
lution of the specified p r o b l e m but an approximate solution. Information
about the remaining error is obtained from a secondary problem:
Given the original problem and an approximatc solution, find an ap-
proximation to its error.

It is obvious that this does not eliminate the u n c e r t a i n t y about the

quality of the approximate solution. This is tolerable because most prob-
lems are only approximations of real-life situations. Nevertheless, there
arise situations where rather concise information about the error of an
approximate solution must be obtained. In the following, we will analyze
the structure of this task for initial value problems for systems of
first order ordinary differential equations.

The Problem

We formulate our task in analogy to the secondary problem above. The

original problem is (y(t) £ ~s)

y'(t) = f(t,y(t)), y(O) = Yo ' t E [O,T] , (1.1)

with sufficient regularity in a sufficiently large n e i g h b o r h o o d of the

unique solution trajectory (t,y(t)).

Inclusion problem: Given an approximate solution ~ : [O,T] ~ s of (1.1)

Find ~ : [O,T] ~ ~ s such that

e(t) := ~(t) - y(t) C ~(t) , t E [O,T] . (1.2)

Here ~ denotes the power set; normally we have to restrict the range of
 86

to an ~asily representable subset of ~ R s like the set H R s of all in-

tervals in ~ s . While the computation of norm bounds for e is a spe-
cial case of (1.2), we will primarily be interested in componentwise
lower and upper bounds w~Nch may well be Of equal sign. Often we will
be satisfied with producing values of ~ at a sequence of arguments
to,tl,..,t n C [O,T].

It is clear that the inclusion (1.2) of e implies an inclusion

y(t) £ ~(t) - ~(t) (1.3)

for the true solution y of (1.1). The algorithms which we will consider
are also immediately applicable to the case of strips of true solutions
(y(t) £ ~ R s ) as they appear for a set-valued initial condition y(O) £ Y o
£ ~s in (1.1).

Naturally, the inclusion problem becomes the more delicate the tight-
er an inclusion we request. It is clear, however, that we cannot gene-
rate an inclusion of a prespecified maximal width in a one-pass step-by-
step procedure for a general initial value problem.

The generation of numerical solutions for the inclusion problem (1.2)

has been studied by many scientists and a good number of algorithms have
been proposed. One of the early investigations is by N.J. Lehmann [4];
it is remarkable that he has already suggested the use of symbol manipu-
lation systems in this connection.

For lack of space, we cannot systematically list and comment the va-
rious contributions. A very extensive bibliography on the subject is to
be found, e.g., in Nickel [7]. Our own bibiliography contains only some
typical examples of specific approaches.

Rather than sketching the historic development, we will attempt to

display a common conceptual framework for most of the algorithms which
have been proposed. This should help in their understanding and evalua-
tion and stimulate the further analysis and development of the area.

Local Analysis

Except in trivial situations, a numerical algorithm for (1.2) cannot

cover the interval [O,T] at once. Hence we consider at first one step
in a forward stepping algorithm: We have arrived at t _ I and obtained a
set ~ - I such that e(t _i) E ~ _ I . In the construction of a corresponding
 87

set ~ at t = t _1+h , we have to regard all solutions y(t;tv_l,yv_ l)

of (1.1) which pass through an admissible value Yv-1 at tv_ I.

Local problem: Find ~ such that (see Fig.])

V

~(tv) - y(t ;tv_1,y~_]) E ~ for all Yv-1 ~ ~(tv-l) - ~-I " (2.1)

Obviously, the use of the l o c a l exac£ inclusion

Ev :: { ~ ( t v ) - Y ( t ~ ; t u - l ' Y u - 1 ) : Yv-1 £ ~ ( t v - 1 ) - E v - 1 } (2.2)

for ~ would keep the inclusion optimally tight. By (2.1), we have

N
E c E and we can use the interior difference

D := ~ ~ E E IP~ s (2.3)
v x)

to represent the excess of ~ over E .

(For two sets in a linear space, with A c B , the interior difference B~A
is the unique set C which satisfies A + C = B. Obviously, O£BTA. The
norm of B~A is the Hausdorff distance of A and !3.)

ms ~._~(t)
Fig. I

tv_ 1 t

It appears that the local excess D of (2.3) is the natural analogon

to the local error in a stepwise algorithm for (1.1). Its size, expres-
sed e.g. by

lID II := max [Id]l

dED

may be used as a quantitative measure Df the (local) accuracy of an in-

clusion algorithm. For s >I and ~ - ] ' ~ v £ ~ ks; D will generally not be
an interval because E v ¢ H ~ s.

Typically, the computation of ~ will be based,, at best, on

 88

correct representation of a few derivatives w.r.t, t

- correct representation of l i n e a r terms in the deviation e between

and y (first order perturbation analysis)

strict bounding of remainder terms, nonlinearities, etc.

Round-off error effects will be caught by the use of directed rounding.

We assume that their influence is negligible compared to the leading lo-
cal excess terms.

A local excess analysis for an algorithm of this kind leads to

D v = o(hP +I) + O(h ~ _ i ) + higher order terms (2.5)

where ~v-1 := diam ~v-1" The appearance of the second term seems unavoid-
able, even if quadratic terms in e are evaluated:

Take y,=y2 so that y'(Yo+e) = y~ + 2Yoe + e 2. Assume e E [-~,~] =: E

and compute bounds for y'(Yo+e) by interval evaluation of y~ + 2YoE + E 2.
With E 2 = [0,~2], one obtains

hy)(Yo+e) £ h . [Yo2 . 2Yo~ y~ +2Yo~ + 2 ]

whose lower bound creates an excess of -h8 2. The reason is the d e p e n ~

d e n c e between the multiple occurences of E in a quadratic expression.

Stability

If our algorithm accounts for the linear terms (linearized about

~(t)) in the deviations correctly, the excess D generated in the step
towards t should propagate like a local perturbation at t during the
further integration. In other words, our local excesses should accumu-
late like the local errors in a one-step algorithm for (1.1), at least
for sufficiently small steps h and a sufficiently narrow inclusion
strip.

However, computationally there arises the necessity to r e p r e s e n t in-

clusions in terms of a s e m i o r d e r of the A s based on c o m p o n e n t s , e.g.
by componentwise intervals. Not always such a semiorder is preserved by
the differential system (1.1): The initial value problem (1.1) is cal-
led q u a s i m o n o t o n e w.r.t, a semiorder ~ in A S if
 89

w'(t) ~> f(t,w(t)) , t £ [O,T] ,

implies w(t) ~> y ( t ) , t ¢ [O,t]
w(O) ~> Yo

(Criteria for quasimonotony and related theorems may be found in Walter

[9] . )

If (1.1) is not quasimonotone w.r.t, componentwise semiorder, the

following happens (see Fig.2):

Consider an inclusion interval ~v-1 = ~v-1 + [-1,+1] • dr_ I, with

dr_ I ~ O. The variational equation of (1.1) (near the solution trajec-
tory) maps ~v-I into E v = G ~v-] + "''' but E~ is not represented or in-
cluded by e v + [-I,+I] .Gdv_ 1.

/ G

variational
e equation

Fig. 2

The s m a l l e s t interval including E is rather specified by

~ = IGi-d_ 1 • (3.1)

Since G ~ I +hvJv_1, (3.1) simulates the use of the modified Jacobian

v-] := D (5.z)

i.e. only diagonal elements are not replaced by their modulus. For a
non-quasimonotone problem we have J+ * J which implies

p(IGI) > p(G) ; (3.3)

hence the excess of an inclusion is amplified more v i o l e n t l y than small

 90

perturbations.

This "wrapping" effect (see e.g. Jackson [3]) equally appears in a

direct use of the variational equation: To include a solution of

e' (t) = J(t) e(t) + ....

lower bound of e must be used with negative elements of J in the

the upper
computation of the upper
lower bound of e. This corresponds to the use of the
2s×2s-matrix

[°::+ o::l
which has the combined spectrum of J = D + R + +R- ) and J+ =
( D+R+-R - ) = ( D + IRI ).

A well-known remedy (cf. e.g. Moore [6]) is the following: Represent

the inclusion at each t~ by E~ = A ~_ , A~ E ~ s x s . If

E = G ,v_ I g v - t + Cv

is the exact local inclusion (see (2.2)), compute

A := Gv,v_ 1 A _I , (3.4)

A A IC]
E := E _ I + W[A; , (3.5)

where W is ~he wrapping operation which maps a bounded set in ~ s into

the smallest enclosing interval. Now the correct propagation is main-
tained by (3.4), and each inclusion increment C is only wrapped twice
(instead o f n - v , t i m e s ) , by (3.5) and in the "output" formation

~ = W[A ~ ] (3.6)

(3.6) is needed for the evaluation of various terms in the step pro-
ceeding from t . Hereby, C may be distorted by a factor lIA lltJA~IH
= cond(A ), see (3.5) and (3.6).

Hence, in the use of (3.4)/(3.5) the growth of cond(A~] has to be

monitored and the representation must be restarted if necessary:

^ ^
= A E =: I E
 91

Other tricks to counteract the effect of (3.2) - see e.g. Conradt

[1, section 6] - also suffer if the condition of A of (3.4) grows
v
with v.

None of the p r e s e n t l y suggested algorithms is suitable for stiff

problems (1.1) because polynomial approximations in t are used.

The A c c u m u l a t e d Excess

At some tv, we compare the computed inclusion ~ to the true error

e(tv) (cf. (1.2)) or the true lnCluslon set E(t ) in case of a set ini-
tial condition Yo for (1.1) and define

global (=accumulated) excess X := ~ ~ E(t v) . (4.1)

The computational c o n t i n u a t i o n of the inclusions from t _ I to t

- propagates the excess X _ I present in ~v-1'

generates the additional excess D =: h D , cf. (2.3),

- may introduce further excess by wrapping.

Let us assume that (1.1) is q u a s i m o n o t o n e w.r.t, componentwise semi-

order so that we may disregard the wrapping effects and that all devia-
tions are s u f f i c i e n t l y small so that a p e r t u r b a t i o n approach is justi-
fief. Then we have (cf. "Local A n a l y s i s " and "Stability")

X = X _I + h J X _I ÷ h D (4.2)

For s u f f i c i e n t l y small h , X v ~ X(t ) where

X'(t) = J(t) X(t) + D(t) , X(O) = {0} , (4.3)

if we assume that we start the inclusion correctly. From (2.5) we have,

with appropriate A and F,

D(t) = A(t) h p + r(t) (diam ~(t)) 2 + small terms . (4.4)

Case I: Point initial condition for (1.1)

Here ~(t) = e(t) + X(t) so that diam ~(t) = diam X(t). This leads,
with (4.4) and (4.3), to
 92

X(t;h) = X(t) h p + O(h p + I ) (4.5)

as long as the "small terms" do n o t become dominant.

This m e a n s that the tZghtness of our i n c l u s i o n s (of the e r r o r as well

as of the true solution) depends on the o r d e r p of the a l g o r i t h m and on
the s t e p s i z e used, as in usual in o.d.e, algorithms.

Case 2: I n t e r v a l initial condition Y for (1.1)

o

d i a m ~(0) = d i a m E(O) = d i a m Yo =: So > 0 ,

(4.6)
diam ~(t) ~ d/am E(t) = diam Y(t) = r ( t ) ~ o + O(e~)

Substitution of (4.4)/(4.6) into (4.3) now y i e l d s

X(t;h,eo) ~ X1(t)hP + X2(t)~ + higher order terms . (4.7)

This m e a n s that a r e d u c t i o n of h c a n n o t improve the i n c l u s i o n beyond the

second term. However, this unavoZdable excess from the q u a d r a t i c terms
in the d e v i a t i o n is 0 ( ~ ) while the e r r o r (and solution) tube d i a m e t e r
is 0(~o) , see (4.6).

The b e h a v i o r (4.5) and (4.7) is well displayed in n u m e r i c a l computa-

tion; results of some e x p e r i m e n t s are s h o w n in T a b l e I.

Methods

We c a n o n l y sketch the two fundamental approaches and, must refer to

the l i t e r a t u r e for more detail:

I) D e f e c t Correction: Let dCt) := ~'(t) - f(t,7(t)) denote the de-

feet of 7- T h e n we have (cf. (1.2) and (2.2))
t t
M
e(tv) = e(t _ I) + ~ [f(s,~(~)) - f(T,~(~)-e(T))]d~ + .[ d ( ~ ) d'c ,

tv. 1 tv- 1
t t

Ev -~ gv-I + f J+(T)E(T) dT+ S incl. {d(~)} dT

tv- 1 tv- 1 (s.1)
t
v
+ f incl. {2nd d e r i v , terms w.r,t, e} dT
tv- I
 93

The solution of the integral inequality (5.1) may be bounded by ap-

proximating t-he resolvent kernel and bounding the remainder. In the last
term of (5.1), an a priori estimate for e in [t _1,t ~] must be used.
This approach was initiated by Schr6der (e.g. [8]); an elaborate al-
gorithm has been described by Marcowitz [5] and Conradt [I].

2) Local Expansion: Let y be a truncated Taylor-expansion about

tv_ 1; denote y(i) (~;~,~) =: fi(~,~ ). Then
p-1 h i
e(t ) = e(t i) + ~ ~ [fi(t 1,Yv_1) - fi(t _ 1 , ~ _ i - e _i)]
i=I
hP
- p-T f p ( X , y ( T ) )

-11 h i
~ f: ,7~_ 1 )

p-1 h i hp (5.2)
+ i=I~ ~.~ [ f ~ ( t v - l ' ~ - 1 - ~ - 1 ) - fi(t~-l'~-1)] ~v-1 ~ fp([tv-1'tv]'Y~ )

where Y is an a priori estimate for y in I t - 1,tv]. The approach was

initiated by Moore (e.g.[6]); a detailed analysis of an algorithm based
upon (5.2) has been presented by Eijgenraam [2].
Obviously, an efficient imp£ementation of an inclusion algorithm for
(1.2) must rely on a powerful Computer AZgchra system for the automatic
generation of procedures for derivatives and bounds of various kinds,
and it must also use an Int¢rva£ Arithmetic system which automatically
handles interva')is properly (with correct rounding). As both kinds of
programming tools are becoming more widely available in standardized
forms, the design of transportable and easily usable software for the
inclusion problem (1.2) should now become feasible.

References
[11 J. C~NRADT~ Ein ~nterval~verfahren ~ur Einschlie~ung des Fehlers e~ner N~zherungsl~mng...~
Freiburger Intervall-Berichte 80/1, 1980.
[2] Tracts
P. EIJGENRAAM,
144, 1981. The solution o f initial value problems using interval arithmetic, Math. Centre
[3] L.W.JACKSON, Interval arithmetic error-bounding algorithms, SINUM 12(1975) 223-238.
[4] N.J. LEHMANN, Fehlerschranken ~ r NaherungslOsungen bei Differentialgleichungen, Numer. Math.
10(1967) 261-288.
[5] U.(1975)MARCOWITZ249-275.
'
Fehlersch?itzung bei Anfangswertaufgaben yon gew. Diffgln_., Numer. Math. 24
[6] R.E.MOORE, IntervalAnalysis, Prentice Hall lnc., 1966.
[7] K.#2590,NICKELt982.Using
interval methods for the numerical solution o f ODEs, MRC Tech. Summary Rep.
[8] J. SCHRODER, Fehlerabsch?~tzung T i t Rechenanlagen bei gew. Diffgln. 1. Ordn., Numer. Math. 3
(1961) 39-61.
[9 ] W.WALTER, Differential - undlntegralungleichungen, Springer-Tracts in Nat. Phil. vol. 2, 1964.
 £ -6 c = 2 -4
= 0 = 2
0 O O ~o = z-z
2-1 .25 ( - 2 ) .27 ( - 2 ) 17.3 .38 ( - 2 ) 6.11
27.7 26.9 18.2
2-2 .9l (-4) .10 ( - 3 ) .64 .21 ( - 3 ) .34 .29 ( - 2 ) 1.16
18.7 11.8 3.23 2.27
2-3 .48 ( - 5 ) .85 ( - 5 ) .054 .65 ( - 4 ) .10 .13 (-2) .51
17.4 2.79 1.~9 1.39
2-4 .28 (-6) • 31 (-5) .020 .075
.47 ( - 4 ) .93 (-3) .37
16.7 1.23 1.13 1.16
2-5 .16 ( - 7 ) .25 ( - 5 ) .016 .41 ( - 4 ) .066 .80 ( - 3 ) .32
16.4 1.06 1.06 I .07
2-6 .to (-8) ,23 ( - 5 ) .015 .39 ( - 4 ) .062 .74 (-3) .29
4~
16.2 1.03 1.03 1.04
2-7 • 63 (-9) .23 ( - 5 ) .015 .38 ( - 4 ) .060 .72 ( - 3 ) .28

diam E(~o) = .157 (-3) 4.0 .626 (-3) 4.0 .253 (-2)

Table 1. Excess X as a f u n c t i o n of h and c o.

80 8O
The p r o b l e m was y, = _y2 Yo : [1 --2-, ] +-2-] t £ [0,9]

The a l g o r i t h m u s e d was an i m p l e m e n t a t i o n of (5.2), w i t h p = 4.

The m a i n figures display diam X at t = 9, cf. (4.1). The i t a l i c figures are q u o t i e n t s of their two
neighbors. The right-hand figures in the c o > 0 c o l u m n s are the q u o t i e n t s d i a m X(h,~o) / d i a m E(Co) .
RECENT DEVELOPMENTS IN THE THEORY
OF FUNCTION SPACES
H. TRIEBEL
Sektion Mathematik, Universitiit Jena
DDR-6900 Jena, Universit~its Hochhaus

i. I n t r o d u c t i o n
The w o r d "function spaces" covers nowadays rather different bran-
ches and techniques. In o u r c o n t e x t function spaces means spaces of
functions and distributions defined on the r e a l euclidean n-space R
n
which are isotropic, non-homogeneous and unweighted. More precisely,
this survey deals with the spaces B ps, q a n d F ps , q on R n w h i c h cover
Holder-Zygmund spaces, Sobolev-Slobodeckij spaces, Besov-Lipschitz
spaces, Bessel-potential spaces and spaces of H a r d y type. F i r s t we try
to d e s c r i b e how the d i f f e r e n t approaches are interrelated, inclusively
few historical remarks. Secondly, we outline some very recent develop-
ments which, by the o p i n i o n of the author, not only unify and simplify
the theory of f u n c t i o n spaces under consideration considerably, but
which also may serve a starting point for further studies.

2. H o w to M e a s u r e Smoothness?
Let R be the r e a l euclidean n-space. The classical devises to mea-
n
sure smoothness are derivatives and differences. If o n e w i s h e s to ex-
press smoothness not only locally but globally, in o u r case on Rn, then
function spaces, e.g. of L p - t y p e , s e e m to be a n a p p r o p r i a t e tool. We
m
use standard notations for the derivatives D e and the d i f f e r e n c e s A h,

D = al an if x : ( X l , . . . , x n) 6 Rn, ~ = (~l,...,en),l~i =
8x I ...Sx n
n
= Ze.
j=l 3
and
m m-l.l
A f(x) = f ( x + h) - f(x)~ Ah = Ah Ah

if x 6 R n, h E Rn, and m = 2,3,... Furthermore,

lif[Lp1[ = ( f{f(x) IPdx) I/p , 0 < p S ~ ,

Rn

with the ssual modification if p = ~. Recall that S and S" s t a n d for

the S c h w a r t z space of a l l c o m p l e x - v a l u e d infinitely differentiable
rapidly decreasing functions on R n a n d the space of all c o m p l e x -
 96

-valued tempered d i s t r i b u t i o n s on R , r e s p e c t i v e l y . Of course, the

n
spaces L with 0 < p S ~ h a v e the u s u a l m e a n i n g ( c o m p l e x - v a l u e d
P
functions).

Definition i. (i) (H~lder-Zygmund spaces). Let s be a p o s i t i v e

number and let m be a n i n t e g e r w i t h 0 < s < m. Then

C s = {flf 6 L~'YflcSllm = nflL U + sup l h l - S l ~ f ( x ) l < ~}. (I)

xe R n
0 #he R n
(ii) (Sobolev spaces). Let 1 < p < ~ and let m be a n a t u r a l
number. Then
W ~ = {flf e Lp, nf~wmn~ = ~ lIDafIL II < ~}. (2)

Remark I. L e t 0 < s < 1. T h e n

lJflCS~l= sup If(x) l + sup If(x) - f(y)l (3)

2
xeR x%y Ix-yt
n
are the f a m i l i a r norms in the H o l d e r spaces C s. If s is a p o s i t i v e
fractional number, i.e. 0 < s : [ s] + {s} w i t h Is] integer and 0 <
< {s} < 1 t h e n (~ c a n be e x t e n d e d by

liD~fiL~n + ~ ILD~fIc{S} il
OSI~ISI s) I ~ l = l s] i "

The c o r r e s p o n d i n g spaces are the w e l l - k n o w n Holder spaces (on R n) as

they had been used since the twenties. It h a d b e e n d i s c o v e r e d b y A.
Zygmund [29] in 1945 that it is m u c h m o r e e f f e c t i v e to use h i g h e r dif-
ferences than derivatives combined with first differences. Definition
l(i) m u s t be u n d e r s t o o d in this sense. In p a r t i c u l a r if s is g i v e n
t h e n all the a d m i s s i b l e norms IIflcSH are equivalent to e a c h other. The
m
spaces ~D have been introduced by S.L. Sobolev [16] in 1936. T h e d e r i -
vatives involved must be understood in the s e n s e of d i s t r i b u t i o n s .
In the f i f t i e s several attemps hade been made to e x t e n d the s p a c e s
from Definitio I, to fill the gaps b e t w e e n 1 W2,...
Lp, Wp, and to re-
place the s u p - n o r m in (i) by o t h e r norms. On the b a s i s of q u i t e d i f f e -
rent motivations S.M. N i k o l ' s k i j introduced in the e a r l y f i f t i e s the
spaces A ps , ~ w i t h s > 0, 1 < p < ~ (we a l w a y s prefer the n o t a t i o n s
used below which are d i f f e r e n t f r o m the o r i g i n a l ones) and L.N.Slobo-
deckij, N.Aronszajn and E.Gagliardd defined the s p a c e s A~,p~ w i t h s > 0,
1 < p < ~. T h e n e x t m a j o r step came around 1960. L e t F a n d F -I b e the
Fourier transform and its i n v e r s e on S', r e s p e c t i v e l y . Let
s
I sf = F - I [ ( 1 + i~12)2Ff], f 6 S', - ~ < s < ~ (4)

Definition 2. (i) (Besov-Lipschitz spaces). L e t s > 0, i < p <

 97

and 1S q S ~. L e t m he an i n t e g e r with m > s. T h e n

As = {f{f 6 Lp, lfflA s" ~ +

P,q p,qHm-~ HflLp
1
÷ ( Ilhl-squA~f(.)[L iSq ~h )q < ~}
Rn P lht n
(usual modification if q = ~).
(ii) (Bessel-Dotential sDaces). Let -~ < s < ~ and 1 < p <
Then
Hs = {flf e S',HfIHSll = HI fIL II < ~] . (6)
p p s p

Remark 2. T h e B e s o v - L i p s c h i t z spaces As have been introduced by

P,q
O.V. Besov [2,3] (following the way paved by S . M . N i k o l ' s k i j ) . They
proved to be o n e of the m o s t successful scales of f u n c t i o n spaces. The
two sup-norms in (i) (with respect to x E R n a n d h E R n) are splitted
in (5) in an L p - n O r m a n d an L q - n O r m . In some sense these spaces are the
appropriate extensions of the spaces C s in the w a y described above
and they fill the gaps between the Sobolev spaces in a r e a s o n a b l e way,
although the Sobolev spaces are not special cases of the spaces As
P,q
if p ~ 2. As in the c a s e of the spaces C s all She a d m i s s i b l e norms
nfIA~,qn
m = (with different m's) are pairwise equivalent. The spaces
H s h a v e b e e n i n t r o d u c e d bv A . P . C a l d e r o n [5] and N.Aronszajn, K.T.
P
S m i t h [ i]. F i r s t we r e m a r k t h a t
Hs = Ws if s = 0 , i , 2 , . . . a n d 1 < p <
P P
In o t h e r w o r d s , a l s o t h e s p a c e s Hps fill the g a p s between the S o b o l e v
spaces and extend these spaces to n e g a t i v e values of s. B u t m o r e impor-
tant, successful method, the F o u r i e r - a n a l y t i c approach, or t h e s p e c -
tral approach, which we discuss in the n e x t section.

3. T h e Fourier-Analytical Approach
We return to (4) a n d (6). Let ~ be the Laplacian on R n and let E
b e the identity. Recall that
(E - A ) f : F-I[(1 + i~J2)Ff], f 6 S' .
More general, the fractoonal powers of E - A a r e g i v e n by
Is s
(E - A ) 2 f = F-I[(1 + J~t2)~Ff] , feS', -~<s< ~
In o t h e r words, f 6 H s if a n d o n l y if (E - ~ ) s / 2 f 6 L . This gives a
P P
better feeling w h a t is g o i n g o n in (6). In p a r t i c u l a r , s m o o t h n e s s is
measured in the F o u r i e r image b y the w e i g h t - f u n c t i o n g(~) = (i , i ~ 1 2 ) s/2,
and the g r o w t h of t h i s w e i g h t - f u n c t i o n at i n f i n i t y represents the d e g r e e
of s m o o t h n e s s . Let h(~) be a n o t h e r positive smooth weight-function,
not necessarily of t h e a b o v e polynomial type. In o r d e r to p r o v i d e a
better understanding of the F o u r i e r - a n a l y t i c a l method we dare a b o l d
speculation: If h l ( ~ ) and h 2 ( { ) are two w e i g h t - f u n c t i o n s with the
same b e h a v i o u r at i n f i n i t y then they g e n e r a t e the s a m e s m o o t h n e s s
class in the a b o v e sense. It c o m e s out that s o m e t h i n g of this type is
correct (via F o u r i e r multiplier theorems), but we s h a l l not try to
m a k e this v a g u e assertion more precise. But on the b a s i s of this
speculation we try to r e p l a c e the a b o v e w e i g h t - f u n c t i o n g(~) =
= (i + I~12) s/2 by m o r e h a n d s o m e weight-functions which offer a great-
er f l e x i b i l i t y . If I~I ~ 2 j w i t h j = 0,i,2,... t h e n g(~) ~ 2 js. H e n c e
one can try to r e p l a c e g(~) by a step f u n c t i o n g(~) w i t h g(~) ~ 23s if
[~i ~ 2 j. This r e p l a c e m e n t is a little bit too crude, but a smooth
version of this idea is just w h a t we want. We g i v e a p r e c i s e formula-
tion. Let ~({) E S w i t h
1
supp 9 C {~I~ S l~i ~ 2]
and
~(2-J{) = 1 if ~ # 0 .

Functions with t h e s e properties~ exist. Let ~j(~) = ~(2-3~) if j = 1,2,

3,... and ~0(~) = 1 - ~l~j(~). Then ~0(~) has also a c o m p a c t support.
j s
The d e s i r e d substitute of (I + I~i2) 2 is n o w g i v e n by ~ 2Jsgj({). We
introduce the p s e u d o d i f f e r e n t i a l operators j=0

~9(D)f(x) = F-l[~j(~)Ff](x), x e Rn, j = 0,1,2, .... f 6 S'. (7)

This makes sense because by the P a l e y - W i e n e r - S c h w a r t z theorem ~j(D)f(x)

is an a n a l y t i c function in R n for a n y f E S'. F u r t h e r m o r e , by a t h e o r e m
of P a l e y - L i t t l e w o o d type we h a v e
IIfIHSll II( 12Js~j(D)f(.)12)i/21Lpll, - ~ < s < ~, 1 < p<~ (8)
P j=0
(in the s e n s e of e q u i v a l e n t norms). This is the s u b s t i t u t e we are l o o k -
ing for. N o w we c a n a s k q u e s t i o n s . D o e s it m a k e s e n s e to r e p l a c e the
12-norm in (8) b y a n l q - n o r m (or q u a s i - n o r m ) ~ 0 < q ~ ~ ? Is it r e a s o -
nable to i n t e r c h a n g e the r o l e s of Lp and 12 (or m o r e general lq) in
(8)?

Definition 3. (i) L e t - ~ < s < ~, 0 < p S ~ a n d 0 < q S ~. T h e n

B sp,q= [fif 6 S',llfJB~,qll = (3~02Jsqu~j(D)f(.)ILpllq)i/q.= < ~] ~ (9)

(usual m o d i f i c a t i o n if q = ~) .
(ii) Let - ~ < s < ~, 0 < p < ~ a n d 0 < q ~ ~. T h e n

Fs = {flf e S',IIflF s n = ~( ~ 2 J s q l ~ j ( D ) f ( . ) l q ) I / q l L U<~] (i0)

P'q P'q ~ j=0 P
 99

(usual m o d i f i c a t i o n if q = ~ ).

Remark 3. F o r a l l admissiSle values s,p,q the s p a c e s B ps, q and

F ps , q are quasi-Banach spaces (Banach spaces if p -
> 1 and q > _ i)~ a n d
t h e y are independent of the c h o s e n function ~ (in the s e n s e of e q u i v -
alent quasi-norms). Maybe this fact is n o t so a s t o n i s h i n g if p a n d q
are r e s t r i c t e d by 1 < p < ~ and I < q < ~, because in t h o s e cases the
Fourier multiplier theory for Lp w i t h 1 < p < ~ and its v e c t o r - v a l u e d
counterparts c a n be taken as h i n t s that something of t h i s type m a y be
valid. But it was a big surprise, also for the creators of t h i s theory,
that these definitions make sense even if 0 < p s 1 (and 0 < q ~ i).
The only exception is p = ~ in the c a s e of the s p a c e s Fs (but e v e n
P,q
in t h i s case o n e c a n do s o m e t h i n g after appropriate modifications).
The above definition of the s p a c e s B s is d u e to J . P e e t r e [ 1 1 , 1 2 ] .
P,q
The spaces Fs h a v e b e e n i n t r o d u c e d by the a u t h o r [19], P . I . L i z o r k i n
P,q
[I0] a n d J . P e e t r e [13]. F r o the g r e a t e r p a r t of the t h e o r y of t h e s e
spaces a restriction to p Z i, q ~ i w o u l d be a r t i f i c a l . But from a
technical point of v i e w such a restriction often simplifies the p r o o f s
because one has the e l a b o r a t e d technique of B a n a c h space theory at
hand (and o n e avoids a lot o f p i t f a l l s which are so a b u n d a n t if p<l).
Systematic treatments of the theory of the spaces B s and F s have
P,q P,q
been given in [14] (mostly restricted to B s with 1 ~ p S ~ ) and
P,
[23] (with [21,22] as forerunners, cf. a l s o ~ 2 0 ] ) . A g a i n o n e c a n a s k
questions. What is the use of these spaces? What is the c o n n e c t i o n of
these spaces and those ones introduced in S e c t i o n 2? As far as the
latter question is c o n c e r n e d one has the following answer.

T h e o r e m i. (i) Let s > 0. T h e n

Cs = B s (11)
(ii) Let i < p < ~ and - ~ < s < ~. T h e n
Hs = Fs (12)
P P,q '
(in p a r t i c u l a r , ~ = Fs if m = 0,1,2, and 1 < p < ~ )
p,2 . . . .
(iii) L e t s > 0, i < p < ~ and i ~ q ~ ~. Then
A ps , q = B ps , q
(1 3)
(iv) L e t 0 < p < ~. Then F 0 is a ( n o n - h o m o g e n e o u s ) space of
p,2
Hardy type.

Remark 4. Proofs may be found in [23], cf. also Sections 6 and 7.

4. P o i n t s Left Open
The Fourier~analytical approach proved to be v e r y useful in c o n -
 lO0

nection with applications to linear and n o n - l i n e a r partial differen-

tial equations, cf. [20,23] as far as linear equations are concerned.
In the r e c e n t l y developed method of p a r a - m u l t i p l i c a t i o n s by J . M . B o n y
and Y . M e y e r (in order to o b t a i n local and m i c r o l o c a l smoothness asser-
tions for n o n - l i n e a r partial differential equations) characterizations
of type (II) play a c r u c i a l role. An e x t e n s i o n of these m e t h o d s to the
full scales Bs and F s has b e e n given by T . R u n s t [15] (there one
P,q P,q
can also find the n e c e s s a r y r e f e r e n c e s to the papers by Bony, Meyer).
There is no c l a i m that this p a p e r gives a s y s t e m a t i c description
of the h i s t o r y of those function spaces which are t r e a t e d here. We o-
mitted few i m p o r t a n t developments. But we w i s h to m e n t i o n at least few
key-words and some m i l e s t o n e - p a p e r s . Interpolation theory plays a cru-
cial role in the theory of function spaces since the sixties. The out-
standing papers a r e those ones of J.-L.Lions, J.Peetre [9] a n d A.P.
Calderon [6]. A s y s t e m a t i c approach to the theory of function spaces
from the s t a n d p o i n t of i n t e r p o l a t i o n theory has b e e n g i v e n in [20].
Another important approach to the theory of function spaces is the
real v a r i a b l e method in the theory of H a r d y spaces and the e l a b o r a t i o n
of the t e c h n i q u e of m a x i m a l functions. The m i l e s t o n e - p a p e r in this
field is C.Fefferman, E.M.Stein [7].

5. H a r m o n i c and T h e r m i c Extensions
The interest in H a r d y spaces has its o r i g i n in c o m p l e x function
theory: traces of h o l o m o r p h i c functions in the unit disc or the upper
half-plane on the r e s p e c t i v e boundaries. A generalization of this idea
yields a characterization of functions and d i s t r i b u t i o n s of the spaces
Bs and F s on R as traces of h a r m o n i c functions or temperaturs in
p,q p,q n
+
Rn+ 1 = [(x,t)Ix • Rn, t > 0} cn the h y p e r p l a n e t = 0, w h i c h is identi-
n ~2
fied w i t h R n. We r e f o r m u l a t e this problem as follows. Let A =
j=l ~x~
3
be the L a p l a c i a n in R
and let f 6 B s or f E F s . W h a t can be said
n p,q p,q
(in the sense of c h a r a c t e r i z i n g properties) about the solutions u(x,t)
and v(x,t) of the p r o b l e m s

( ~zu + AU)(x,t) = 0 if (x,t) • + u(x,0) f(x) if x E R (14)

~t2 Rn+l} = n

(harmonic extension) and

(Sv - Av)(x,t) = 0 if (x,t) • Rn+ll

+ v(x,0) = f(x) if x • R n (15)

(thermic extension)? At least in a formal w a y the solutions u(x,t) and

 101

V(x,t) are known,

t
u(x,t) = P(t)f(x) = c ~ n+l f(y)dy, x e Rn, t>0 (16)
R
n (|x_yl2÷ t 2) 2
(Cauchy-Poisson semigroup) and
2
- m
n _Ix-~l
v(x,t) = W(t)f(x) = ct 2 fe 4t f(y)dy, x e Rn, t > 0 (i 7)
R
n
(Gauss-Weierstrass semigroup). If f E S' is given, then (17) makes sen-
se. F u r t h e r m o r e , (16) m u s t be u n d e r s t o o d in the f o l l o w i n g theorem via
limiting procedures. If a is a real number we put a+ -- max (0,a).

Theorem 2. Let 9 0 E S with 90(0) # 0.

(i) Let - ~ < s < ~, 0 < p < ~, and 0 < q < ~. Let k and m be non-
n e g a t i v e integers with k > n( 1 - i)+ + max (s,n( 1 - i) ) and 2m > s.
Then P P1 +
1
U~0(D)flLp[I + (.0f t(k-s)q" ~0kp(t)f
~t Lpliq ~-dt )q (18)

and s 1
i
ll~0(D)flL II + ( f t (m-~)qll0mw(t)flL iiq at )q (19)
P 0 3t m P ~-

(modification if q = ~ ) are e q u i v a l e n t quasi-norms in B s . If s >

n( 1 - I)+ then II~0(D)flL p [i in (i 8) ~ (i 9) can be r e p l a c e d P'qil
by fIL P II.

(ii) Let -~ < s < ~, 0 < p < ~ and 0 < q < ~. Let k and m be non-
-negative integers w i t h k > m i n (np , q ) + max (s,n(~i~ - i)+) and 2m > s.
Then
]
H~o(D)flLp~ + ll(0~it(k-s)q ~t~.)lq~kp(t)f~ ~)qILpll (20)

1
and II~0(D)flLpII + I{( ~I t ( m - ~ ) q i ~~m w ( t ) f ( . ) l q ~dt
- )qILpll
-- (21)
0 @ tm
(modification if q = ~ ) are e q u i v a l e n t quasi-norms in F s . If s >
P,q
n(p - i)+ then ll~0(D)flLpll in (20), (21) can be r e p l a c e d by ]IfILplI.

Remark 5. C h a r a c t e r i z a t i o n s of the above type have a long history.

As far as the c l a s s i c a l B e s o v - L i p s c h i t z spaces A s and the B e s s e l -
P,q
- p o t e n t i a l spaces H s are c o n c e r n e d the first c o m p r e h e n s i v e t r e a t m e n t
P
in the sense of the above t h e o r e m has b e e n g i v e n by M o H . T a i b l e s o n [ 18],
cf. also T . M . F l e t t [ 8]. In this c o n t e x t we m e n t i o n also the books by
P.L.Butzler, H.Berens [ 4] and E.M. Stein [ 17] w h e r e one can find
many informations about characterizations of the above type (for the
classical space) and the s e m i g r o u p s from (16) and (17), cf. also [20,
 I02

2.5.2, 2.5.3]. More recent results (characterizations of the spaces

B ps, q and Fp,q
s in the sense of the above theorem) have been obtained
b y G.A.Kaljabin, B . - H . Q u i and the author. The above formulation has
been taken over from [25] (cf. also [23, 2.12.2] w h e r e we also gave re-
ferences to the papers by B.-H.Qui and G.A.Kaljabin).

6. Unified A p p r o a c h
Up to this moment we said nothing how to u n d e r s t a n d that the
a p p a r e n t l y rather different approaches via derivatives, differences,
Fourier-analytical decompositions, harmonic and thermic extensions,
always yield the same spaces B s and F s In [23] we proved equiva-
p,q p,q"
lence assertions of the above type m o s t l y by rather specific arguments,
cf. also [14,22]. But recently it became clear that there exists a
unified approach which covers all these methods, at least in principle,
and which sheds some light on the j u s t - m e n t i o n e d problem. We follow
[25] where [24] may be considered as a first step in this direction.
The basic idea is to extend the admissible functions ~ and ~j in (7)
and (9), (i0), Such that c o r r e s p o n d i n g (quasi-)norms in the sense of
(9), (i0) cover a u t o m a t i c a l l y c h a r a c t e r i z a t i o n s of type (18), (19) and
(5). We recall that

~(tD)f(x) = F-l[~(t.)Ff](x) = ct k 8kp(t)f(x) if ~(~) =

= j~ike_l~ I ~t k (22)

and
~(~t D)f(x) : ct m 8mW(t)f(x> if ~(~) ~ J~12me - I ~ 2 (2 3)
8t m

Furthermore we r e m a r k that the discrete quasi-norms in (9) and (i0)

have always continuous counterparts, i.e.
! i
II~0(D)flLp II + ( 0 I t-sqll~(td)f(.)l Spllq ~---
d t )q (24)

is the c o n t i n u o u s substitute of the q u a s i - n o r m in (9) and

I i
[I~0(D)flLpll + If(0/ t-sqI~(tD)f(" )lq ~--dt)qILpll (25)

is the continuous substitute of the q u a s i - n o r m in (I0). This replace-

ment of "discrete" quasi-norms by "continuous" ones is a technical mat-
ter and has nothing to do with the extension of the class of admissible
~'s which we have in mind. If one puts (22),(2 3) in (24)~(25) then
one obtains (18)-(21)° Of course one has to clarify under what condi-
tions for the parameters involved this procedure is correct. However
before giving some details we ask how to incorporate derivatives and
 103

differences in this Fourier-analytical concept. We h a v e

~(D)f(x) = cD~A~f(x) if ~(~) = ~ ( e i~h- i) m, (26)

n
with ~ = (el,...,en#, m natural number, and Sh = ~ {jhj, ~e= ~ii...
...$~n. T h e t h r e e functions ~ in (29),(2 3)~(26) h~ve in c o m m o n that
n
t h e y t e n d to tero if ISt + 0 (even if e = 0 in (28)). In a d d i t i o n the
functions ~ from (22),(2 3) h a v e the same property if I~I + =. If o n e
compares these functions ~ with the function ~ from Section 3 used
in D e f i n i t i o n 3 then it s e e m s to be at least plausible that one can
substitute ~ in (9)~(10) by t h e functions ~ from (22)~(2 3) if k a n d
m are chosen sufficiently large. As for t h e function ~ form (26)
this question is m o r e delicate. First one has no d e c a y if ~ t e n d s to
infinity and secondly one has not only to h a n d l e an i s o l a t e d function
but a family of functions parametrized by h E R (and, maybe, by e).
n
We r e t u r n to these questions later on a n d formulate a result which co-
vers in p r i n c i p l e all cases of interest.
L e t h(x) C S a n d H(x) E S with supp h C {yl lyl ~ 2}, s u p p H C {yI
1 1
K lyl ~ 4}, h(x) = i if ]xl ~ i, and H(x) = i if ~ K Ixl S 2.

Theorem 3. L e t 0 < p ~ ~, 0 < q ~ ~ and - ~ < s < ~. L e t s 0 and s1

be two real numbers with
1
So+ n ( ~ - i)+ < s < s I and s I > n(~ - i)~ . (27)

Let ~0(~) and ~(~) be two infinitely differentiable complex-valued

functions on R n a n d R n- {0}, respectively, which satisfy the T a u b e r i ~
conditions
1
I~0(~)I > 0 if I~t ~ 2 a n d I~(~)I > 0 if ~ ~ l~I ~ 2. (28)
let p = rain (1,p) a n d
~I(F-I ~(z)h(z) )(y)IPdy < ~ ~ (29)
R sl
n Izl

-ms0P
sup 2 fl(F-l~(2m.)H(.))(y)IPdy < ~ , (~)
m= i, 2, .. Rn

and (~) with ~0 i n s t e a d of ~. T h e n

ll~0(D)flLpll + ( I t-sqll~(tD)f(.)ILpllq ~ )q (3ql

0

(modification if q = ~ ) is an e q u i v a l e n t quasi-norm in B s .
P,q
Remark 6. T h i s formulation coincides essentially with Theorem 3 in
[25] • Of c o u r s e , ~(tD)f = F-l[ ~ ( t . ) F f ] ( x ) and (31) coincides with (24)
This theorem has a direct counterpart for the spaces Fs .Furthermore
s s P'q .
there are some modifications (both for B p , q a n d Fp,q) where not onzy a
 104

single function ~ b u t families of these functions are involved, cf.

the c o n s i d e r a t i o n s in front of the above theorem. Maybe the crucial
conditions (29) and (30) look somewhat complicated and seem to be hard
to check. But this is not the case, in p a r t i c u l a r for functions of type
(26) the f o r m u l a t i o n s (29)~(30) are well adapted. Furthermore, if one
uses
IIF-I~ILvn S cUllH ~ ~, 0 < v ~ i, 1
6 > n ( ~ - ~), (~)

then one can replace (29)~(30) by more h a n d s o m e - l o o k i n g conditions,

6
where only B e s s e l - p o t e n t i a l spaces H 2 (or even S o b o l e v spaces W~) are
involved.

Remark 7. T h e o r e m 2 follows from T h e o r e m 3 and its F s -counte[-

P,q
part. One has to use the functions ~ from (22)~(2 3).

7. C h a r a c t e r i z a t i o n s via D i f f e r e n c e s
In principle one can put ~ from (26) in T h e o r e m 3 and its F p,q-
s
counterpart. One can calculate under what conditions for the pamaterers
(29)~(30) are satisfied. However as we p o i n t e d out in front of T h e o r e m
3 one has to modify T h e o r e m 3, because one needs now theorems with fa-
m i l i e s of functions ~ instead of a single function ~. T h i s can be
done, details m a y be found in [25]. We formulate a result what can be
o b t a i n e d on this way.

1
T h e o r e m 4. (i) Let 0 < p S ~, 0 < q S ~ and n ( ~ - i)+< s < m,
where m is a natural number. Then
1
llflLpll + ( lhl~f lhl-sqnAmflLpnqh dhlhl~ )q (33)

(modification if q = ~ ) is an e q u i v a l e n t q u a s i - n o r m in B s
n P'q
(ii) Let 0 < p < =, 0 < q S ~ and min(p,q) < s < m, where m is a
natural number. Then
1
llflLpll + 0( f lhl-sqj(~f)(.)lq dh )qlLpU (~)
lhlSl lhl n

(modification if q = ~ ) is an e q u i v a l e n t q u a s i - n o r m in F s
P,q
Remark 8. We refer for details to [25] w h e r e we proved m a n y other
theorems of this type via F o u r i e r - a n a l y t i c a l a p p r o a c h from Section 6
and few a d d i t i o n a l considerations. H o w e v e r the t h e o r e m itself is not
new, it may be found in [23, 2.5.10, 2.5.12]. But the proof in [23] is
more c o m p l i c a t e d and not so clearly based on F o u r i e r - a n a l y t i c a l results
in the sense of T h e o r e m 3. On the basis of T h e o r e m 4 one has n o w also a
 105

better understanding of (ii) and (i 3). W e p r e f e r e d in the a b o v e t h e o -

rem a formulation via d i f f e r e n c e s only. B u t one can r e p l a c e some
differences by d e r i v a t i v e s , as it is also s u g g e s t e d by (26).

8. T h e L o c a l A p p r o c a h
The original Fourier-analytical approach as d e s c r i b e d in S e c t i o n 3
does not r e f l e c t the l o c a l n a t u r e of the s p a c e s B s and F s o If
P,q P,q
x E R n is g i v e n t h e n one n e e d s a k n o w l e d g e of f on the w h o l e R n in
o r d e r to c a l c u l a t e ~j(D)f(x) in (7). T h i s stands in s h a r p c o n t r a s t to
the derivatives D~f(x) and the d i f f e r e n c e s A~f(x) with lh[ ~ 1 as t h e y
h a v e b e e n u s e d above. However the e x t e n d e d Fourier-analytical method
as d e s c r i b e d in S e c t i o n 6 gives the p o s s i b i l i t y to c o m b i n e the advan-
t a g e s of the o r i g i n a l Fourier-analytical approach a n d of a s t r i c t l y
local p r o c e d u r e . We give a description. L e t k 0 6 S, and k E S w i t h
supp k 0 C [y[ [y[ ~ i], supp k C [y[ [yl ~ I],
(Fk0)(0) • 0 and (Fk)(0) # 0.
n ~2
Let k N = ( ~ 2 )Nk, w h e r e N is a n a t u r a l number. We introduce the
j:l 8xj
means

K(kN,t)f(x) : fkN(Y)f(x + ty)dy, x 6 Rn, t > 0, (~

R
n
where now N = 0,1,2,... This m a k e s sense for any f E S'.

Theorem 5. (i) L e t - ~ < s < ~, 0 < p ~ ~ and 0 < q S ~. L e t

0 < c < ~, 0 < r < ~ and 2 N > m a x (s,n(~ - i)+). Then
1
r
lIK(k0,e)flLpl[ + ( f t-sqllK(kN,t)flLpll q ~ )q (
0
(modification if q = ~ ) is an e q u i v a l e n t q u a s i - n o r m in B s
p,q"
(ii) L e t - ~ < s < ~, 0 < p < ~ and 0 < q s ~. L e t 0 < e < ~,
0 < r < ~ and 2N > m a x ( s , n ( ~ - i)+). T h e n
P i
nK(k0,E)fILpa + ]l(0frt-sqJK(kN,t)f(.)lq t--dt )qlLpt[ (37

(modification if q : ~ ) is an e q u i v a l e n t quasi-norm in F s .
P,q
Remark 9. It c o m e s o u t t h a t the a b o v e t h e o r e m c a n be o b t a i n e d
from Theorem 3 a n d its F s - c o u n t e r p a r t . O n the o t h e r h a n d it is c l e a r
P,q
that (3) describes a local procedure.

Remark I0. W i t h the h e l p of T h e o r e m 5 one can s i m p l i f y and unify

several proofs in [23], cf. e.g. [26]. B u t it is a l s o a n a p p r o p r i a t e
 106

tool to handle psudodifferential operators, cf. [28], and to introduoe

spaces of B s and F s type on c o m p l e t e Riemannian manifolds (which
P,q P,q
are not n e c e s s a r i l y compact), cf. [27].

References

[i] Aronsazaj,N., Smith,K.T., Theom] of B~sel potentials, I. Ann. Inst. Four~ier

(Grenoble) ii (1981), 385-476.
[2] Besov,O.V., 0n a fam//y of funct/on spaces. Embeddings and exte~ions, (Russian)
Dokl. Akad. Pall< SSSR 126 (1959), i163-i165.
[ 3] Sesov,0.V., On a family of function spaces in connectio~ with embeddings and
e x t e r i o r s , (Russian) Trudy Mat. I n s t . $teklov 6@ (1961), 42-'81.
[ 41 B u t z e r , P . P . , Berens,H., Semi-Groups of Op~ators and Approximation, Springer;
Berlin, Heidelberg, New York, 1967.
[5] Calder~n,A.P., ieb~gue spac~ of functio~ and d i s t ~ i b ~ , "Part. Diff.
[q.", Proc. Syrup. Math. 4, AMS (1981), 33-49.
[6] Calderen,A.P., Intermediate spaces and i}~terpolation, the complex method,
Studia Math. 24 (1964), 113-190.
[ 7] Fefferman,C., Stein,E.M., H p spaces of 6euera/ u~L/ables, Acta Math. 129
(1972), 137-193.
[ 8] Flett,T.M., T e m p £ ~ u ] ~ , Bessel p o t e n ~ and L e p s c ~ z spaces, Proc. London
Math. Soc. 32 (1971), 385-451.
[9] Lions,J.-L., Peetre,J., SuA une claSS d" espaces d' interpolation, Inst.
Hautes Etudes Sci. Publ. Math° 19 (1964), 5-68.
[ iO] Lizorkin,P.I., Properties of f u ~ o ~ of th£ spaces Ar (Russian) Trudy
Mat. Inst. 8teklov 131 (1974), 158-181. P'@'
[ii] Peetre,J., Su~ l ~ ~spaces de Besou, C.R. Acad. Sci. Paris, S~r. A-B 264 (1967),
281-283.
[ 12] Peetre,J., Remarq~es SuA les espaeeS de Besov, Le ca6 0 < p < I, C.R° Acad. Sci.
Paris, SSr. A-B 277 (1973), 947-950.
[13] Peetre,J., On spae~ of Triebel-Lizorkin type, Ark. Mat. 13 (1975),123-130.
|14] Peetre,J., New Thought6 on Besou Spaees, Duke Univ. Math. Series~ Durham, 1976.
[15] Runst,T., Para-differential op~u~tors in spaces of Triebel-Lizorkin and Besov
type, Z. Analysis Anwendungen.
[ 16] Sobolev,S.L., M~thode nouvelle ~ re6ou~e le probl~me de Cauehy pour les
[qumtio~ lin[aJ~es hyp~teolique~ no~afes, Mat. Sb. i (1936),39-72.
[17] Stein,E.M., Singular Integrafz and Niff~entiability Fropert/es of Fuact/ons,
Princeton Univ. Press~ Princeton, 1970.
[ 18] Taibleson,M.H., On the theory of iipsch/£z spaces of d/etn/but/ons on euc]~/dean
n-space, [,/[, J. Math. Mechanics 13 (1964), 407-479; (1965), 821-839.
[19] Triebel~H., SpaceS of distributions of Besov type on euclidean n-space, D~Lity,
Interpolation, Ark. Mat. i i (1973). 13-64.
[20] T r i e b e l , H . , Interpolation Theory, Function Spaces, Differential Operators,
North-Holland,.Amsterdam, New York, Oxford, 1978.
[21] Triebel,H., Fou]ue~]~A~fysis and F a n , o n Spaces, Te~mbner, Leipzig, 1977.
[22] Triebel,H., Spaces of Besov-Ha~-Soboleu Type, Teubner, Leipzig, 1978.
[23] Triebel~H., The0ry 0f F~no~0n Spaces, Birkh~user, Boston 1983, and Geest &
Porting, Leipzig, 1983~
[ 24] Triehel,H., Ch~uzct@Y~zations of Besov-Hardy-Sobolev spaces via harmonic
function~, temperatures, and related mea~, J. Approximation Theory 35 (1982)~
275-297.
[ 25] Triebel,H., C h ~ z a t i o n 6 of Besov-Hardy-Soboleu spaces, a unified approach.
[26] Triebel,H., Diffeomorph~m properties and poin~wise multi~eas for spaces of
Bes~v-Hardy-Sobolev type.
[27] T r i e b e l , H . , Spac~ of Besov-Hardy-Sobolev type.on complete Riemannian manifolds.
[28] T r i e b e l , H . , Pseudo-diff~entia£ operators in --F~q'Spaces"
[29] Zygmund,A., Smooth functions, Duke Math. J. 12 (1945), 47-7@.
ON PROPERTIES OF OSCILLATORY
SOLUTIONS OF NON-LINEAR
DIFFERENTIAL EQUATIONS
OF THE n--TH ORDER
M. BARTU~EK
Department of Applied Mathematics, Fac. of Science, University of J. E. Purkyn6
Jand~kovo ndm. 2a, 662 95 Brno, Czechoslovakia

Consider the d i f f e r e n t i a l equation

(I) y(n) = f(t,y,...,y(n~l)), n a 2

where f : D ~ R is c o n t i n u o u s , ~ = R+ X R n, R+ = [0,~), R = (-~),

there exists a number a 6 {0,1} such that

(2) (~l)af(t,xl,...,Xn)Xl a 0 in D.

Definition. T h e s o l u t i o n of (I) d e f i n e d on R+ is c a l l e d p r o p e r if
y is not t r i v i a l in a n y n e i g h b o u r h o o d of ~. T h e s o l u t i o n of (i)
d e f i n e d on [0,b) is c a l l e d non-continuable if e i t h e r b = ~ or b <
n~lly(i)
and (t)l =
i=0
The solution y of (i) d e f i n e d on [0,b), b S ~ is c a l l e d oscilla-
t d r y if t h e r e e x i s t s a s e q u e n c e of its zeros tending to b and y is n o t
trivial in a n y left n e i g h b o u r h o o d of b.
Denote the set of all o s c i l l a t o r y solutions of (i), d e f i n e d on
|0,b) b y 0 [ 0 , b ). Let 0 [ 0 , ~ ) = 0 and N = {1,2,...}.

Many papers (see e.g. [6]) are d e v o t e d to the s t u d y of c o n d i t i o n s

under which oscillatory solutions exist• But the p r o b l e m of b e h a v i o u r
of such s o l u t i o n s for n > 2 is n o t s o l v e d in a s u i t a b l e way. W e t o u c h
some p r o b l e m s concerning the b e h a v i o u r of o s c i l l a t o r y solutions.

I. Definition. T h e p o i n t c 6 [0,b) is c a l l e d H - p o i n t of y. if t h e r e
exist sequences {t k} i' {~k } of n u m b e r s of [0,b) such that
(t k - c ) ( ~ k - c) > 0, y(t k) : 0, y(~k ) # 0, k E N.

In [4] some p r o p e r t i e s of zeros of y C 0 [ 0 , ~ ) w e r e studied for the

l i n e a r c a s e of (I) Especially, it w a s shown, that e v e r y zero of y(i)
i = 0,1,...,n-i is s i m p l e in some n e i g h b o u r h o o d of +~. T h i s r e s u l t is
generalized in [ I] for the e q u a t i o n (i) if the i n t e r v a l (0,b) d o e s not
contain H-points. Moreover, the f o l l o w i n g statement w a s proved:

Theorem I . Let either n = 2n0, nO + ~ be odd, or n be odd.and let

y E 0[0,b). Then there exist at most two H-points in the interval [0,b).
 tJO

If there 6x~st two q nes cI < c2, then 9(t) ~ 0 qn [ 6 1 , 6 2 ] .

If n = 2n0, n 0 + a is e v e n the statement of the theorem 1 is n o t

valid as it is s h o w n by the following

Theorem 2. Let n = 2, ~ = i. T h e r e exist continuous functions

f : D ~ R with the property (2), y E 0[~,~) and a sequence {Tk} 1 o f
numbers such that ~k c R+, l i m Tk = ~ a n d T k i s the H-point of y.
k~
Proof. In [ 5] it is s h o w n that there exist continuous function
a : R+ (-~,0) and numbers b 6 R+, k E (0,I) such that the d i f f e r e n -
tial equation y'' : a(t) l y ( t ) I k s g n y(t) has an o s c i l l a t o r y solution on
[0,b) a n d y(t) - 0 on [b,~).
Let T E [0,b) be an a r b i t r a r y z e r o of y' a n d d e n o t e h = b - T.
Define ~ : R+ -- (-~,0) and ~ : R+ ~ R+ in the following way: a,~ are
periodic on [T, ~) with the p e r i o d 2h,

~(t) = a(t), ~(t) -- y(t) for t 6 [0,b)

~(t) = a(2b - t), ~(t) = y(2b - t) for t 6 [b,b + h].

From this, according to y'(T) = 0 we g e t that ~ E C0(R+), ~ E CI(R+).

B y u s e of substitutions t ~ x, x = 2(b + ih) - t, t E [b + (i - l)h,
b + ih], i = 0,i,2,... c a n be p r o v e d t h a t ~ is a s o l u t i o n of y'' =

= ~(t)ly(t)iksgn y(t). As b is H - p o i n t of y a n d ~ , too, we c a n p u t

T k = b + 2kh, k E N. T h e theorem is p r o v e d .
n n
II. Let n_ be t h e e n t i r e p a r t of ~. P u t for y E C 0(R.), m E N
Jm(t;y) - rtU~Tm T2 ~
- 0 ~ 0 .... 0 ~ Y(Zl)dZl'''d~m' J0 (t:,y) = y(t), t E R+
n-n0-1
a+i n-i n
(3) Z(t;y) = Z (-i) ( n ) n--~-i)
2 J2i (t;~y(i)]2)"
i=0
The following Lemma was proved in [ i] :

Lemma. Let y be a s o l u t i o n of (I) d e f i n e d on R

+ and let either
n = 2n0, n O + ~ be o d d o r n be o d d . Then
no-X
z(n-Z)(t;y) = Z (-1)a+iy(n'i-l)y(i)(t) ,
i=0
_ )no+a (n o )
+ i (-i (n - 2 n 0 ) [ y (t)] 2
2

z(n)(t;Y) = (-l)~y(n)(t) y(t) ÷

no+~[y(n 0 )
+ (-i) (t)]2(n - 2n 0 - i) > 0, tER+.
 1ll

In the p r e s e n t part we skull study the a s y m p t Q t i c behaviour of

proper oscillatory solutions of (1) under the a s s u m p t i o n s

(4) n -- 2n 0 + i, n o E N .

Definition. Let y E 0 and lim z(n-l)(t;y) = c. T h e n y E 01 (y E

t-~
e 0 2) if c = ~ (c < ~).
It is shown in [ii that
for y e 01 lim suply(t) l = ~ holds. The
t~o~
behaviour of y 6 0 2 is different.

T h e o r e m 3. Let (4) be v a l i d and l e t continuous functions g,

gl: R+ ~ R+ e x i s t s u c h that g(x) > 0 i n some neighbourhood of x = 0,
l l m inf g(x) > 0 and

(5) g(IXll) -< I f ( t , xl, .... Xn)l <_ g l ( I X 1 1 ) in D

holds. L e t y E 0 2 . T h e n c = O, l i m y ( i ) ( t ) = 0, i -- 0 , 1 , . . . , n - 2 and
t~
y(n-l) is bounded.

Proof° Let M 6 (0,~) be a n u m b e r such that M l =

rain g(x) > 0.
MS x< ~
Let D 1 = {t : t 6 R+, ly(t) I S M}, D 2 = R+ - DI, Yi(t) = y(t) for
t 6 Di , yi(~) = 0 for t 6 R+ Di, i z 1,2. It is clear (by use of (5))
that Yl 6 L (R+) '
. (n) 6 L~(R+)
Yl
According to Lemma and (5)

(6) ~ > z(n-l)(~;y) - z(n~l)(0}y) = 0/~(-l)~y(n)(t)y(t)dt Z

a 0 ~ g(ly(t~)ly(t)Idt a O f g(JY2(t)l)lY2(t)Idt Z

oo
> M 1 0 f lY2(t)Idt ;

0f°°[Y~n)(t)Id t S ~ 0f~lY2 n) (t)Y2(t),dt _< ~ 0 f ~ l y { n ) ( t ) Y (t)Idt<

Thus Y2 E LI(R+), y~n) E LI(R+) and a c c o r d i n g to [3, V, §4] and (5)

(7) [y(i)(t)l S K < ~, t E R+, i = 0,1,2,...,n-l.

Let [tk}l, [~k}l be sequenoes, such that 0 S t k < ~k < t k + l ' lim t k
k..,.~ ,

y ( t k) = 0, y ' ( ~ k) = 0t y(t) * 0 on (tk,~k)t k E N. T h e n t by u s e o f (g)

and (7)
1 ~ rk
> 01 g ( l y ( t ) l ) l y ( t ) I d t a ~ k~ 1 t~ g ( l y ( t ) I ) l y ( t ) l ly'(t)Idt S
 112

1 ~ tY(~k)
<- K k~Z 0I g(s)sds .

Thus l i m y(t) = 0 a n d a c c o r d i n g to K o l m o g o r o v - H o r n y Theorem ([ 4],

t~
p. 167) a n d (7) we c a n c o n c l u d e that

(8) lim y(i)(t) = 0, i = 0,i,2,...,n-2 .

t~
Let c % 0. By i n t e g r a t i o n and by u s e of L e m m a we g e t the existence of
6 R+ s u c h that

(9) ~Z(t;y) l >_ ~ tel tn-l, t 6 [~,~).

(n o )
As a c c o r d i n g to (8) lim y (t) = 0, it f o l l o w s from (3) t h a t

]Z(t;y)1 _< A ( t ) t n-l, l i m A(t) -- 0

t~
which contradicts to (9). T h e theorem is p r o v e d .

III. This paragraph contains some remark concerning the e x i s t e n c e of

proper oscillatory solutions of (i). The case ~ = 0 was investigated in
[7] .

Definition. The equation (i) has Property A 0 if e v e r y proper

solution of (i) is o s c i l l a t o r y for n e v e n and is e i t h e r oscillatory or

(i0) lim y(i)(t) = 0 ,

t~
i = 0,1,...,n-i for n odd. The equation (i) h a s Property A 1 if e v e r y
proper solution is e i t h e r oscillatory or (i0) holds for i -- 1,2,...
...,n-l.

The following theorem gives us s u f f i c i e n t conditions for the

existence of p r o p e r oscillatory solutions if e = 1.

Theorem 4. L e t ~ = i and b o t h n , n 0 be e v e n (n be o d d ) . Let (I)

have Property A0 ( P r o p e r t y AI). Let continuous functions h : R+ - R+,
: R+ ~ ( 0 , ~ ) exist such that ~ is non-decreasing, Of ~
dt ( t ) - ~ and

(ii) If(t, Xl, .... Xn) I ~ h(t)~ ( ~ Ixil ) in D

i=l
hold. Then every non-eontinuable solution y of (7), s a t i s f y i n g
z(n-l)(o;y) > 0 is oscillatory and p r o p e r .

Proof. Let y be a n o n - c o n t i n u a b l e solution of (i) for w h i c h

(n-l)
(12) Z (0;y) > 0.
 113

According to the assumptions of Theorem and [ 6, Th. 12.1] y is

either proper or lim y(i)(t) = 0, i = 0,1,2,...,n-i. As by virtue of
t~
Lemma the function z(n-l)(0;y) is non-decreasing, we can conclude
that y is proper.
Further, in both cases, it follows from L e m m a o f Kiguradze
([5], Lemma 14.1) that in case of y be non-oscillatory lim z(n-l)(t~y) --
t~
= 0 holds. Thus we get the contradiction to (12) and Lemma. The
theorem is proved.

Remark i. The conditions, under which (i) has Property A 0 or A 1

were studied by many authors, see e.g. [6].
2. For the linear case of (i) the existence of oscillatory
solutions from the set 02 was proved in [5].

References

[i] BARTU~EK,M., 0, properties of o s c i l l a t o r y s o l u t i o n s of ordinar~

differential inequalities and e q u a t i o n s , D i f . U r a v . (to appear,
in Russian).

[2] BARTU~EK,M., On o s c i l l a t o r y s o l u t i o n of t h e d i f f e r e n t i a l equation

of t h e n-th order, A r c h . M a t h . ( t o a p p e a r ) .
[3] BECKENBACH E.F., BELLMAN,R., Inequalities, Springer-Verlag, Berlin,
1961.

[4] ELIAS,U., O s c i l l a t o r y s o l u t i o n s an Extremal points for a l i n e a r

d i f f e r e n t i a l equation, Arch. Ration Mech. and Anal., 71, No 2,
177-198, 1979.

[5] HEIDEL,J.W., Uniqueness, c o n t i n u a t i o n , and n o n o s c i l l a t i o n f o r a

second order n o n l i n e a r d i f f e r e n t i a l equation, P a c i f . J . M a t h . , 1 9 7 0 ,
32, NO 3, 715-721.

[6] KIGURADZE,I.T., Some s i n g u l a r boundary value problems f o r o r d i n a r y

differential equations (in Russian), Tbilisi Univ. Pres, Tbilisi
1975.

[7] KIGURADZE,I.T., On asymptotic behaviour of solutions of nonlinear

non-autonomou~ ordinar 9 differential equations, Colloq. math. soc.
I. Bolgai, 30. Qualitative theory of diff. eq., Szeged, 1979, pp.
507-554.

[8] KIGURADZE I.T., On vanishing at i n f i n i t y of s o l u t i o n ~ of ordinary

differential equation~, Czech. Math. J. 33 (i08), 1983, 613-646.
UNIQUENESS WITHOUT CONTINUOUS
DEPENDENCE
T. A . B U R T O N a n d D. P. D W I G G I N S
Department of Mathematics, Southern Illinois University
Carbondale, Illinois 62901, U.S.A.

i. Introduction. In the classical theory of o r d i n a r y differential

equations if s o l u t i o n s of a s y s t e m

(1) x' = h(t,x)

are uniquely determined by initial conditions, then the solutions are

continuous in the initial conditions. But the situation is much differ-
ent for differential equations in infinite dimensional spaces. Suffi-
cient conditions for this to hold have been discussed in [8-16] and [19-
20]. Recently, Schiller [18] constructed a fairly abstract example of
a differential equation in the Banach space ~ of bounded sequences with
the supremum norm in which solutions are unique but are not continuous
in initial conditions.
We p r e s e n t a simple example of the same behavior and point out that
the real difficulty is that there are many topologies for the initial
condition space.

2. Continuity in initial conditions. Consider the system

t
(2) x' = h(t,x) + f q(t,s,x(s))ds

in which h: (-~,~) × Rn + Rn , q: (-~,~) x (-~,~) × Rn ÷ Rn , with h and

q continuous pointwise. To fix the function space we suppose all solu-
tions start at t O = 0. Then, to s p e c i f y a solution of (2) we require a
continuous initial function ¢: (-~,0] -÷ R n such that
def 0
~(t) = f q(t,s,*(s))ds

is continuous for t > 0. We m a y t h e n use the Schauder fixed point the-

orem to show that the system
t
(3) x' = h(t,x) + ~ qCt,s,x(s))ds + ~(t), x(0) = *(0)

has a solution x(t,0,~) satisfying (3) on an interval [0,~), for some

> 0, withix(t,0,~) = ~(t) on (-%0].
System (2) is well defined using pointwise continuity in Rn a n d
there is an initial function set X consisting of continuous functions
for w h i c h ~ is c o n t i n u o u s for t > 0 (X m a y be empty). Without putting
any topology at all on X the p r o b l e m s of e x i s t e n c e , uniqueness, and
continuation of s o l u t i o n s are w e l l - d e f i n e d . But to c o m p l e t e a classical
 116

fundamental theory for (2) we want to say that for each # E X if there
is a u n i q u e solution x(t,0,#) on [0,8] and if {~n } is a s e q u e n c e in X
converging to ~ then s o l u t i o n s x(t,0,~n) converge to x(t,0,#) on [0,B].
W h i l e we are q u i t e w i l l i n g to accept any type of c o n v e r g e n c e of
x(t,0,~n) to x(t,0,~) on [0,8], the m e a n i n g of ~n c o n v e r g i n g to ~ must
be specified.
In a g i v e n p r o b l e m we f r e q u e n t l y have a wide d e g r e e of f r e e d o m in
our c h o i c e of t o p o l o g y for the initial condition space. Recent prob-
lems call for u n b o u n d e d initial functions, plentiful compact subsets
of these initial functions, and'continuity of the t r a n s l a t i o n maP.
These requirements lead us to a l o c a l l y convex topological vector
space (Y,p) w i t h ~ 6 Y if ~: (-~,0] + R n is c o n t i n u o u s and for #,~ 6 Y
then
eo

(4) P(¢,~) = ~ 2 - k [ p k ( ~ , ~ ) / ( l + Pk(¢,~))]

k=l

where pk(~,~) = max I¢(s) ,(s) l and I'I is any n o r m on R n. For

-k<s<0
motivations see [1- 5] and [7]. Problems are also effectively treated
using a Banach space with weighted norm as the same references show.

EXAMPLE 1 . Consider the linear scalar equation

t 3
(s) x' ~ x + $ [x(s)/(t s+l) ]as
_co

which has the zero solution and it is unique. In fact, if ~: (-oo,0] ÷R n

0
is any continuous function for which #(t) = ~ [,(s)/(t-s+l)5]ds is con-

tinuous for t > 0, then there is one and only one solution x(t,0,~)
defined on [0,oO. Note that the set X is not empty. We now s h o w t h a t
solutions x(t,0,~) are not continuous in (Y,P).

PROOF. Define a sequence {~n } c X by

= I 0 if -n <_ s < 0

L -n(s+n) if s < -n.

Notice that

P(¢n,0) = E 2-kok(¢n,0)/[1 + pk(~n,0)]

k=l
oo

< ~ 2 -k + 0 as n +
-k=n

and so {~n ) c o n v e r g e s to the zero f u n c t i o n in (Y,p). Now, for 0 < t < 1

 117

and n > 2 we have

0
~n(t) = ~ [¢n(S)/(t - s + l ) 3 ] d s

-n
>_ -nf [(s+n)/(-s+Z)5]ds >_1/16.
-oo

Hence, we a r e considering the equation

t
x' = x + ~0 [x(s)/(t -s + l)3]ds + ~n(t)

> x + (1/16)

so that continuity of x(t,0,%) in ~ fails.

Sch~ffer suggests that the absence of continuity in his example
may be the result of his space, ~, being neither separable nor reflex-
ive. But our sequence {%n} is contained in a compact subset of (Y,p)
so the subset is separable and it may be embedded in a Banach space.
One can show that (Y,p) is not reflexive. However, since (Y,p) is
Frechet it is barreled (cf. []7; p. 60]).

PROPOSITION I. Let {~n } be the sequence of Example 1 in (Y,p).

Then {~n } is contained in a compact subset of (Y,9).

PROOF. Define a continuous function g: (-~,0] + [0,~) by

g(s) = sup ~n(S). Then g is a continuous piecewise linear function.
n
Moreover, if s > -n, then g(s) is Lipschitz with constant n. Let
~: (-~,0] + [0,~) be the piecewise continuous linear function defined
by a(-n) = n. Then the set

s = {4~ ~ Y [ ] ~ ( s ) l <__gCs] on <-~, o],

l¢(u) ~(v) I !~(lul +Lvl +1) tu-vL}

is compact in (Y,~) (cf. [7; p. 2]) and ~ontains {~n ). This completes
the proof.

To see that S can be embedded in a compact subset of a Banach

space, for the function g defined in the proof of Proposition i, define
~(s) = [g(s) + 1] 2. Then define the Banach space (Z,I-i~) by ~ E Z if
E Y and if

I~[~ = sup I~<s)[/~Cs)

-~<s<O

exists. This is a Banach space and S is compact in it.

 t18

PROPOSITION 2. The set S in the proof of Proposition 1 is con-

tained in a.reflexive subspace of Y.

PROOF. Let Q = L(S) be the linear hull of S (i.e., Q is the space

formed by taking linear combinations from S.). Now Q is a subspace of
the locally convex metric space Y and so Q is a locally convex metric
space. Horeover, as S is closed, Q is closed and complete. Hence, Q
is a Frechet space and is barreled. Q is not compact since Q is un-
bounded in the sense of Treves [21; pp. 136-7]. However, closed and
(Treves) bounded subsets of Q satisfy boundedness and Lipschitz condi-
tions similar to those of S, and so must be compact. Therefore, Q is
a reflexive space (of. Treves [21; p. 373]). This completes the proof.

Hence, continuity in initial conditions is not guaranteed by the

separability and reflexivity of the space.

3. Fading memory. In general, (2) makes sense only when there

is a fading memory. Consider the scalar equation
t
x' : A(t)x + ~ C(t-s)x(s)ds.

At the very least we w i s h to consider all bounded continuous % E Y.

Since we w a n t %(t) to be continuous for t > 0 we n e e d to ask that

~tC(u)ldu < ~. T h e n by [6] there is a continuous increasing function

r: [0,~o) + [1,~) such that r(t) ~ ~ as t -~ ~ a n d f IC(u) lr(u)du < ~.

We t a k e g(s) = r(-s) so that if l*(s) l < yg(s) fo~ s o m e y > 0, then

0 IC(t -s),(s)lds 2 ~ /i C(u) lr(u)du

-m t

and this tends to 0 as t ÷ ~. tn summary, if we a d m i t bounded % then

we c a n admit unbounded % and the memory of ~ fades in %(t) as t + ~.
A similar result holds for nonlinear systems as may b e seen in [5].
The function g is central to the study of delay equations and its
role may b e seen in [1 - 5] and [8]. For (2) to have meaning we e x p e c t
to be able to require } to be continuous in t for bounded continuous
~. But i n many p r o b l e m s one quickly learns that unbounded ~ are needed;
however, we s h o w i n [5] that if bounded } make ~ c o n t i n u o u s then s o do
certain classes of unbounded }. And t h i s gives rise to the function g
which is the weight for a Banach space <x,l.lg). In order to have a
unified theory of existence, continuity, boundedness, stability, and
periodicity we w o r k entirely :in t h i s Banach space. The importance of
that unity is illustrated in [4].
In a private communication Kaminogo informs us that he improved
our Example 1 by using bounded initial functions and has obtained
 119

continuou~ dependence results for b o u n d e d initial functions.

As a step toward completion of a u n i f i e d theory we n o w p r e s e n t a
result on c o n t i n u o u s dependence on initial conditions using unbounded
initial functions. In p r e p a r a t i o n for that result we n o w suppose that
for the equation (2) there is a c o n t i n u o u s function g: (-~,0] + [i,~)
which is d e c r e a s i n g , g(0) = i, and g(r) ÷ ~ as r ~ -~. Form the Banach
space (X,].ig) with 9 ~ X if 9: (-~,0] ÷ R n is c o n t i n u o u s ,

191 def sup Ig(t) I/g(t )

g -~<t<O

exists, and there is a n o n e m p t y subset U c X for w h i c h the following

definition holds.

DEF. A set U c X is an e x i s t e n c e set for (2) if ~ E U i m p l i e s

~(t) is c o n t i n u o u s for t > 0.

DEF. Let U be an e x i s t e n c e set for (2). Then (2) has a fading

memory with respect to U if for each ~ 6 U, each J > 0, and each £ > 0
there is a 6 > 0, a D > 0, and an H > 0 such that if ~ 6 U, I~-~Ig < 6,
and 0 < t < J then
D
(a) ~ lq(t,s,?(s)) - q(t,s,~(s))ids < s and
0
(b) ]~ q ( t , s , ? ( s ) ) d s l i M.

THEOREM. Let U be an existence set for (2) and let (2) have a
fading memory with respect to U. Suppose there is a 9 E U such that
x(t,0,~) is unique on some interval [0,tl]. Then x(t,0,9) is contin-
uous in ~ in the following sense: If {~n } c U and 19 - ~ n l g ÷ 0 as
n + ~, then IQ9 - Q ~ n l g ÷ 0 as n ÷ ~ where (Q~)(t) = x(t+tl,0,, ) for
-~ < t 2 0 and x(t,0,?n) is any solution of (21) w i t h initial function

~n"
PROOF. Let x(t,0,9 ) be defined on [0,tl] and suppose it is not
continuous in 9- Then for some s > 0 and for each 6k > 0 there exists
?k E U a n d t k E [ 0 , t 1] with IX(tk,0,,) - x(tk,0,~k) I h e. We may,
assume t k ÷ S E [ 0 , t 1] by picking a subsequence if necessary. Moreover
we m a y a s s u m e the t k chosen so that {x(t,0,?k)} is bounded on [0,S].
Thus, {x'(t,0,*k)) is bounded on [0,S] and so {x(t,0,?k)} is an equi-
continuous sequence with a convergent subsequence, say {x(t,0,~k)}
again, with limit ~(t). We m a y w r i t e
 120

t
Xk(t ) = x(t,0,~bk) = ~k(0) + ; h(S'Xk(S))ds

t u tO

t -D
+ £ ~_~ [q(u,s,,k(S)) - q(u,s,,(s))]ds du
t 0

+ fD[ q ( u , s , ~ k ( S ) ) - q(u,s,¢(s))]ds du

f o r a n y D > 0. Let ~1 > 0 b e g i v e n and let 0 < t < S. Then there is

a D > 0 such that
t -D
I~ f_~ [ q ( u , s , ~ k ( S ) ) - q(u,s,*(s))]ds dul < e 1.

For this D > 0, then {~k(S)) converges uniformly to ,(s) on [-D,0].

Hence, we may t a k e the limit as k + ~ and find that xk(t) -~ n ( t ) and
t
n(t) = ,(0) + £ h(s,n(s))ds

t u tO
+ ~ 't q ( u , s , n ( s ) ) d s du + ~ ~_~q(u,s,*(s))ds d u .

Thus, n and x(t,0,~) satisfy the same equation. Since that equation
has a unique solution, ~(t) = x(t,0,~). This contradicts
iX(tk,0,~) - X(tk,0,~k) I > E and completes the proof.

REFERENCES

i. Arino, O., Burton, T., and Haddock, J., Periodic solutions of

functional differential equations, Royal Soc. Edinburgh, to
appear.
2. Burton, T. A., Volterra Integral and Differential Equations,
Academic Press, New York, 1983.
3. , Periodic solutions of nonlinear Volterra equations,
Funkcial. Ekvac., to appear.
4. , Toward unification of periodic theory, in Differential
Equations; Qualitative Theory (Szeged, 1984), Colloq. Math. Soc.
J~nos Bolyai, 47, North Holland, Amsterdam.
5. , Phase spaces and boundedness in Volterra equations, J.
Integral Equations, to appear.
6. Burton, T. and Grimmer, R., Oscillation, continuation, and
uniqueness of solutions of retarded differential equations, Trans.
Amer. Math. Soc. 179(1973), 193-209.
 121

7. Corduneanu, C., Integral Equations and Stability of Feedback

Systems, Academic Press, New York, 1973.
8. Haddock, J., A friendly space for functional differential equa-
tions with infinite delay, to appear.
9. Hale, J. K., Dynamical systems and stability, J. Math. Anal.
Appl. 26(1969), 39-59.
I0. Hale, J. K. and Kato, J., Phase spaces for retarded equations,
Funkcial. Ekvac., 21(1978), 11-41.
II. Hino, Y., Asymptotic behavior of solutions of some functional
differential equations, Tohoku Math. J. 22(1970), 98-108.
12. , Continuous dependence for some functional differential
equations, ibid., 23(1971), 565-571.
13. , On stability of the solutions of some functional diff-
erential equations, Funkcial. Ekvac., 14(1970), 47-60.
14. Kappel, F. and Schappacher, W., Some considerations to the funda-
mental theory of infinite delay equations, J. Differential Equa-
tions, 37(1980), 141-183.
15. Naito, T., On autonomous linear functional differential equations
with infinite retardations, J. Differential Equations, 21(1976),
297~315.
16. , Adjoint equations of autonomous linear functional dif-
erential equations with infinite retardation, Tohoku Math. J.,
28(1976), 135-143.
17. Schaefer, H. H., Topological Vector Spaces, Macmillan, New York,
1966.
18. Sch~ffer, J. J., Uniqueness without continuous dependence in
infinite dimension, J. Differential Equations, 56(1985), 426-428.
19. Schumacher, K0, Existence and continuous dependence for functional
differential equations with unbounded delay, Arch. Rat. Mech.
Anal., 67(1978), 315-335.
20. Seifert~ G., On Caratheodory conditions for functional differen-
tial equations with infinite delays, Rocky Mt. J. Math., 12(1982),
615-619.
21. Treves, T., Topological Vector Spaces, Distributions and Kernels,
Academic Press, New York, 1967.
CONNEC ONS IN SCALAR REACTION
DIFFUSION EQUATIONS
WITH NEUMANN BOUNDARY CONDITIONS
B. FIEDLER and P. BRUNOVSKY
Universit~it Iteidelberg, Inst. of Applied Mathematics
Inst. e~Applied Mathematics Comenius University,
Im Neuenheimer Feld 294, Mlynskd dolina, 842 15 Bratislava
tteidelberg, West Germany Czechoslovakia

We c o n s i d e r the flow of a o n e - d i m e n s i o n a l reaction diffusion

equation

u t = Uxx + "f(u) (i)

on the interval x C (0,i) with Neumann boundary conditions

Ux(t,0) : Ux(t,l) : 0. (2)

Given two stationary solutions v, w of (1),(2) (i. e. solutions of

v" + f(v) : 0, v'(0) : v'(1) : 0 ) (3)

we say that v connects to w if t h e r e exists a solution u(t,x) of (i),

(2) for t ~ (- ~,~) such that

lim u(t,.) = v, lim u(t,.) = w. (4)

t--~ t -~
For ordinary differential equations trajectories connecting sta-
tionary points have been studied in the context of s h o c k waves [3,10]
and t r a v e l l i n g waves [i0]. F o r ( 1 ) , (2) the p r i n c i p a l motivation for
studying connections is s o m e w h a t different. As a r g u e d by Haie [4] the
f l o w on the m a x i m a l compact invariant set A d i s p l a y s the e s s e n t i a l qua-
litative features of t h e f l o w of (i), (2). Since (1),(2) is a g r a d i e n t
system, under mild growth conditions on f at i n f i n i t y A consists of
stationary solutions and connecting trajectories. Therefore, determi-
ning all stationary solutions and their connecting trajectories, we
know the e s s e n t i a l Dart of the flow.
For special classes of n o n l i n e a r i t i e s the p r o b l e m of identification
of p a i r s of stationary solutions admitting connections has b e e n studied
by C o n l e y and Smoller [2, i0] and Henry [5, 6] w h o solved the p r o b l e m
completely for f satisfying f(0) : 0 and b e i n g qualitatively cubic-like.
In [i] we have given an a l m o s t complete answer to the following question
concerning equation (i) w i t h Dirichlet boundarv conditions for g e n e r a l
f:
 124

(Q)Given a stationary solution v, w h i c h stationary solutions does it

connect to?
Similarly as in [i], to d i s t i n g u i s h the w ' s to w h i c h v connects
we introduce a scalar characteristics of the complexity of stationary
solutions. However, while in [i] this is the m a x i m a l number of sign
changes (called zero number, z), in o u r case its r o l e w i l l be p l a y e d
bv the lap n u m b e r 1 introduce by M a t a n o [7]. F o r a given function v on
[9, i] l(v) is, by d e f i n i t i o n , the m i n i m a l number of i n t e r v a l s Ij i n t o
which [0, i] can be p a r t i t i o n e d so t h a t v is s t r i c t l y monotone on e a c h
I. and l(v) : 0 for v c o n s t a n t .
3
F o r v s t a t i o n a r y we d e f i n e the instability (Morse) index i(v) as
the number of n e g a t i v e eigenvalues of the p r o b l e m

y" + fC(v(x))+~) V : 0 (5)

y'(0) : y'(1) : 0. (6)

Bv a S t u r m - L i o u v i l l e separation of zeros argument one obtains for v ~ c n s t

l(v) S i(v) S l(v) + i. (7)

The stationary solution v is c a l l e d hyperbolic if I = 0 is not an e i g e n -

value of the p r o b l e m (5),(6).
Given v hyperbolic, for 0 ~ k ~ l(v) we d e n o t e by ~ k ( Z k ) the sta-
tionary solution ~ (~) satisfying i(~) : k with smallest ~(0)> max
Range v ( i(~) = k with largest ~(0) c R a n g e v, respectively). By Q(v)
we d e n o t e the set of stationary solutions which v connects to. The
following theorem is an almost complete answer to (Q):
Theorem. Let f be C 2 and let

lim f ( s ) / s < 0 (8)

Isl--=
Let v be a h y p e r b o l i c solution of (3).
(i) If v is c o n s t a n t or i(v) = l(v) then

~(V) : {Vk, Vk: 0 -< k < i(v)}

(ii) If v(0) = m a x v / m i n v and i(v) = l(v) + I then

e(v) : el u f~2 u ~3'

where

~i : {Vk: 0 _<k < i ( v ) } ,

0 <_k < i ( v ) - l}
~2 : {~k:
and e i t h e r
g3 : [~k: k : i(v) - i}

or ~3 c o n s i s t s of one or s e v e r a l stationarv solutions w with Range w c

Range v and i(w) < i(v).
 125

Note that from [7] it follows that there are no other cases
possible except of (i) - (ii).
The proof of this theorem proceeds along the lines of the proof of
the analogous theorem of [i] for the Dirichlet case. Therefore, the de-
tails of its outline given below can easily be completed from [I].
To establish connections we focus on the case f(0) =0, v ~ 0 here
for simplicitv. Let the zero number z ( u ( t , . )) d e n o t e the number of sign
changes of x J~ u ( t , x ) , 0 <x <i - cf. [i]. Then z ( u ( t , . )) is d e c r e a s i n g
with t [7, 8] and we may define the dropping times

tk = inf {t ~ 0: z(u(ti)) S k} ~
and
Tk = tanh tk 6 [0, ]].

Note that t k S tk_ 1 ~ ... ~ t O . If t k < tk_l, the sign

~k = sign u(t,0], t k < t < tk_ 1

is independent of t. We collect all this information in the map

Y = (Y0 . . . . . Yk .... ) where
: )i/2
Y0 ~0 (I - ~0

Yk = ~k(Tk-i Tk )I/2

Taking n = i(v) - I and a small sphere En around v in t h e d n s t a b l e mani-

fold wU(v) ,

v: En ÷ S n

turns out to be a continuous and essential mapping into the standard n-

snhere S n. In p a r t i c u l a r , y is surjective. Therefore, for any 0 S k <
i(v) , 0 6 {-i, I} there exists a u0 6 wU(v) such that Y(U0) = oe k w h e r e
e k denotes the k-th unit vector. Hence the trajectory u(t) through u0
connects v to a stationary w with

z( w - v) : k and sign (w(0) - v(0) = ~ .

The fact that w = v k for a = I and w = vk for ~ = -i (the latter in

case (i) follows from the following two lemmas:
Lemms I. T h e stationary solution v does not connect to w if t h e r e is a
stationary with w(0) between v(0) and w(0) such that

z(v - ~) sz(w - ~).

Lemma 2. For stationary solutions v, w
l ( v ) k i if R a n g e w < Range v
z(v - w) : { (9)
0 if R a n g e v O Range w = @

Note that up to interchange of v a n d w all possible cases are taken

 126

care of in L e m m a 2.
The p r o o f s of t h e s e lemmas can e a s i l y be o b t a i n e d by a d a p t i n g
those of the corresponding lemmas from [i]. The first one is b a s e d on
the m a x i m u m Qrinciple, the second employs the p h a s e plane portraits of
(v, v ' ) a n d ( w , w'~ one notes that b e t w e e n two successive local extrema
of v t h e r e is p r e c i s e l y one intersection point of v and w in c a s e
Range w C Range v.
Concluding we n o t e that ~k and ~ k can e a s i l y be identified from
the global bifurcation diagram of the p a r a m e t r i c equation

u t = Uxx + a2f(u)

as g i v e n e. g. in [9]. Also, we n o t e that by f u r t h e r analysis we can

identify the m e m b e r s of ~3 m o r e precisely.
Figure 1 illustrates the T h e o r e m for a p a r t i c u l a r f. P o i n t s on one
curve re~resent stationary solutions with the same lap n u m b e r which
increases from curve to c u r v e by o n e from left to r i g h t starting with
1 = i. Case (i) a p p l i e s to v in the left p a r t of F i g u r e 1 with l(v)=0,
i(v) = 3. C a s e ( i i ) applies to v in the right part of F i g u r e i. In this
case l(v) = 8, i(v) = 9 , v8 ~ ~3 and all candidates for ~3 not exclu-
ded by (il) are marked by "?". By further analysis we are able to s h o w
that connections do e x i s t to those solutions marked by "!" a n d do n o t
exist to those solutions marked by "x".

REFERENCES

i. P. B r u n o v s k ~ , B. F i e d l e r : C o n n e c t i n g orbits in scholar r e a c t i o n diffu -

sion e q u a t i o n s , to a p p e a r
2. C. Conley, J. Smoller: T o p o l o g i c a l t e c h n i q u e s in r e a c t i o n d i f f u s i o n
e q u a t i o n s . In " B i o l o g i c a l G r o w t h and S p r e a d " , Proc. H e i d e l b e r g
a.d. 1979, J ~ g e r , Rost, T a u t u e d i t o r s , S p r i n g e r L e c t u r e N o t e s in
Biomathematics 38, 473 - 483
3. I. M. G e ~ f a n d : Some p r o b l e m s in the t h e o r y of q u a s i l i n e a r equations.
U s p e c h i Matem. N a u k 14 (1959), E n g l i s h t r a n s l a t i o n ~ I S Trans-
l a t i o n S e r i e s 2, 29 (1963), 295 - 381
4. J. Hale, L. M a g a l h a e s , W. Oliva: A n I n t r o d u c t i o n to I n f i n i t e Dimen-
s i o n a l D y n a m i c a l S y s t e m s - G e o m e t r i c T h e o r y . Aopl. Math. Sci.
47, S p r i n g e r 1984
5. D. Henry: G e o m e t r i c T h e o r y of S e m i l i n e a r P a r a b o l i c Equations.
L e c t u r e N o t e s in H a t h e m a t i c s 840, S p r i n g e r 1981
6. D. Henry: Some i n f i n i t e dimensional Morse-Smale systems defined by
parabolic equations, to a p p e a r in J o u r n a l of D i f f e r e n t i a l
Equations
7. H. M a t a n o : N o n i n c r e a s e of the lap n u m b e r of a s o l u t i o n for a o n e -
dimensional s e m i l i n e a r p a r a b o l i c e q u a t i o n . Publ. Fac. Sci. Univ.
K y o t o Sec. IA, 29 (1982), 401 - 441
 127

8. K. Nickel: G e s t a l t a u s s a g e n ~ber L ~ s u n g e n p a r a b o l i s c h e r D i f f e r e n t i a l -
gleichungen. Crelle's J. fdr Reine und Angew. M a t h e m a t i k 211
(1962), 78 - 94
9. P.Poli~ik: G e n e r i c b i f u r c a t i o n s of s t a t i o n a r y solutions of the Neu-
mann p r o b l e m for reaction d i f f u s i o n equations. Thesis, Komen-
sky University, B r a t i s l a v a 1984
iO. J. Smoller: Shock Waves and R e c t i o n - D i f f u s i o n Equations. Grundlehren
der Math. Wiss. 258, Springer 1982
 128

X ....

!
1 o
I I
I
I
I ~q
I
i

! !
°t c~
+

>l 7

c~

~F

c~O ~4

l
l
!
W3 A 1
ON A CERTAIN BOUNDARY VALUE
PROBLEM OF THE THIRD ORDER
M, GREGUS
F a c u l t y q / M a t h e m a t i c s a n d Physics, C o m e n i u s University
M l y n s k d dolina, 8 4 2 15 Bratislava, Czechoslo~'akia

i. T h e boundary value problem of the form

(i) y''' + If(x) + ~g(x)]y' + ~h(x)y = 0

(2) y(-ap~) = y(a,l) = 0, a > 0

a t
(3) X f [r(t)-k] { g ( t ) y ( t , X ) + f [h(T) - g'(T)]y(~,X)d~}dt =
-a -a
a t
= f [r(t)-k] { f ( t ) y ( t , X ) + / f(T)y' ( T , X ) d ~ } d t ,
-a -a

where f'(x), g'(x), h(x), r''(x) are continuous functions on the in-
terval (-a,a> and k is a c o n s t a n t , will be s t u d i e d .
The boundary condition (3) is in the integral form. For the first
time, such a condition was formulated in [ i] a n d the p r o b l e m (1), (2),
(3) is a n a t u r a l generalization of the problem discussed in [i] .
It w i l l be shown that under certain conditions on the function
r(x), the problem (i), (2), (3), is e q u i v a l e n t to the b o u n d a r y
problem (I), (4), where

(4) y(-a,~) = y''(-a,X) = 0, y(a,X) = 0 a > 0 .

In o r d e r to solve the problem (1), (4), the theory of the third

order linear differential equation [2] c a n be a p p l i e d . Moreover some
special results will be formulated.

2. C o n s i d e r the problem (I), (2), (3). Let the functions f, g,

h, r fulfil the c o n d i t i o n s formulated in S e c t i o n I. T h e n the following
theorem is true.

THEOREM i . The problem (11, (21, (51 is equivalent to the

problem (I), (4) if the function r = r(tl solves t h e problem

(5) r'' + f(tlr = kf(t)

(6) r(-a) = k, r(a) = k .

 130

P~00~. Integrating the d i f f e r e n t i a l equation (I), written in

the form

y''' + {['f(x) + lg(x)]y}' + {-f'(x) + l[h(x) - g'(x)]}y = 0 ,

t e r m by t e r m from -a to x, x ~ a, and considering (2), we get

x
y'' + f ~)y + lg(x)y + / {-f' (~) + l[h(T) - g ' ( r ) ] } y ( T , l ) d r =
-a
= y''(-a,~)

Now suppose that y''(-a,l) = 0, multiply the last equality by

r(x)-k, where k is a c o n s t a n t , and integrate it f r o m -a to a°

We c o m e to t h e e q u a l i t y

a
(7) f [r(t)-k][y''(t,l) + f(t)y(t,l)]dt =
-a
a
= f [r(t)-k] { g ( t ) y ( t , ~ ) +
-a
t
+ f [h(~)-g'(~)]y(~,X)dT}dt-
-a
a t
- f [r(t)-k] { - f ( t ) y ( t , x ) ÷ f f(~)y' ( ~ , X ) d T } d t .
-a -a

The right-hand side of (7) contains the expression which stands in

condition (3). Therefore it is n e c e s s a r y to p r o v e that the integral
o n the left-hand side of (7) is e q u a l to zero.
Calculate this integral. Under the c o n d i t i o n s (2), it f o l l o w s
that
a
f [r(t)-k][y''(t,x) + f(t)y(t,X)]dt = y'(a,X)[r(a)-k] -
-a
a
-y'(-a,~)[r(-a)-k] + f [r''(t) + f(t)r(t)-k f(t)]y(t,X)dt .
-a

This implies that the b o u n d a r y condition (3) will be f u l f i l l e d if

y''(-a,l) = 0 and if t h e function r(t) solves the b o u n d a r y problem
(5), (6). Thus the theorem is p r o v e d .

3. In [3] the p r o b l e m

(8) y''' + {[ 1 + l g ( x ) ] y } ' = 0

a
(9) y(-a) = y(a) = 0, f (cos t - cos a)g(t)y(t)dt = 0
-a
 131

and its g e n e r a l i z a t i o n

(i0) y''' + [i + Ig(x)]y' + lh(x)y = 0

a
(Ii) y(-a) = y(a) = 0, f (cos t - cos a){g(t)y(t) +
-a
X
+ I [h(T)-g'(~)]y(T)d~}dt = 0
-a

where a > 0, g'(x), h(x) are c o n t i n u o u s functions on <-a,a> , w e r e

discussed.

REMARK I. The problems (8), (9), and (I0), (ii) are special
cases of the p r o b l e m (I), (2), (3).

Clearly, if w e suppose f(x) = I, h(x) = g'(x), k = c o s a, we get

that (8)~ (9) is a s p e c i a l case of (i)~ (2)~ (3) and from Theorem 1
it f o l l o w s h a t r(x) = cos x. Similarly, if f(x) = 1 and k = cos a
we get that (i0)~ (ii) is a s p e c i a l case of (i), (2)~ (3) if
r(x) = cos x and k = cos a. But r(x) = cos x solves the p r o b l e m
(5), (6), where k = cos a and f(x) = I.
In [3] it h a s been proved that under the c o n d i t i o n a = ~/2 the
problems (8)~ (9) and (10)~ (ii), respectively are equivalent to the
problems (8), (4) and (10), (4) respectively.
Now we prove the following theorem (the f o r m u l a t i o n will be o n l y
for the equation (8), in the case of the e q u a t i o n (I0) the e q u a t i o n
is s i m i l a r ) .

THEOPJZIM 2 . Let g(x) be c o n t i n u o u s on <-a,a> and let

0 < a < ~f2. Then the problem (8), (13), where
a
(I 3) y(-a) = y(al = o, f [r(t)-1] g(t)y(tldt = 0
-a

is equivalent to the problem I8), (4J if

a
(14) r(x) = f G(x,~)d~ ÷ ~l(x) + ~2(x) ,

where G(x,~) is the Green function of the problem

r'" + r = O, r(-a) = r(a) = O, 0 < a < ~ ,

~l(X), and ~2(x}, respectively, are the solutions of the problem

r ' " + r = 0 9 r ( - a ) = i . r ( a ) = o, and of the problem r" + r = o,

r ( - a ) = o, r ( a } = I r e s p e c t i v e l y .
 t32

The proof of T h e o r e m 2 is s i m i l a r to t h a t of T h e o r e m i. I n t e g r a t i n g
(8) term by term from -a to x ~ a and considering (2), we g e t
y'' + y + Ig(x)y : y''(-a,l).
Let y''(-a,X) = 0, multiply this equation by r(x)~ and integrate
it f r o m -a to a. We o b t a i n

a a
[~ i t •
(15) - f y (t) + y ( t ) ] [ r ( t ) - l ] d t = I f [r(t)-l]g(t)y(t)dt
-a -a

It is n e c e s s a r y to f i n d such an r(x) that the inte9ral on the left-

hand side of (15) be e q u a l to zero.
Calculating it w e g e t

a
f [y''(t) + y(t)][r(t)-l]dt = y'(a)[r(a)-l] - y'(-a)[r(-a)~] +
-a
a
+ f y(t)[r''(t) + r(t)-l]dt .
-a

From this equality it f o l l o w s that r(x) mqst solve the problem

r'' + r = 1
r(-a) = i ~ r(a) = I

Thus the theorem is p r o v e d .

REFERENCES

[i] Lockschin,A., ~ber die Knick~ng eines gekr~mmten S t a b l e s , ZAMM,

16, 1936, 49-55.
[2] Gergu~,M., Linear D i f f e r e n t i a l Equation of t h e Third Order
(in s l o v a k ) , Veda, Bratislava 1981•
[ 3] G r e g u ~ , M . , On Some A p p l i c a t i o n of Ordinary D i f f e r e n t i a l Equations
i n Physics, Proc. T h i r d . Conf. Diff. Equ. a n d A p p l . R o u s s e 1985
(to a p p e a r ) .
ON NONPARASITE SOLUTIONS
P. KRBEC
Aeronautical Research and Test Institute
199 05 Prague 9, Czechoslovakia

i. I n t r o d u c t i o n
We s h a l l investigate the differential relation
(i) ~ E F(t,x), x(0) = x0

where F : U ~ K, U = <0,1) X BI, K is the set c o m p r i s i n g nonempty,

compact subsets of some ball in R n, B 1 is the unit ball in R n. J a r n f k
and Kurzweil [2] proved that if F ( t , x ) is c o n v e x t h e n we c a n suppose
F to be S c o r z a - D r a g o n i a n . These authors and many others (see e.g. [ i] ,
[2], [3], [ I0] , [12]) have studied the c o n v e x case very thoroughly.
The nonconvex r.h.s, has been attacked too, certain very strong
results being obtained e.g. by O l e c h [7] , T o l s t o n o g o v [ i0] , [ ii] , Vr-
ko~ [12]. It is e a s y to see that to o b t a i n some reasonable existence
theorem in n o n c o n v e x case it is n e c e s s a r y to s u p p o s e F to be c o n t i n u -
ous. It is a w e l l known fact that the solutions of x 6 F are then
dense in the set of all solutions of x E c o n v F, see e.g. Tolstonogov
[9] .
It is t e m p t i n g then to u s e the F i l i p o v respectively Krasovskij
operation to d e f i n e generalized solutions of x 6 F(t,x), F being
possibly nonconvex° To be more specific, we can define the solution
of ~ E F ( t , x ) through the r e l a t i o n ~ 6 G(t,x) where

G(t,x) = N N co'nv F ( t , B 6 ( x ) - N) or
6>0 ~(N)=0

G(t,x) = A c--~ F(t,B6(x))

6>0
The main problem is t h a t introducing even the solution of x = f(x), f
discontinuous real valued function, through Filippov or e v e n K r a s o v s k i j
operation we c a n o b t a i n certain meaningless solutions.

2" E x a m p l e I. (Sentis [8])

Let f : R -- R, f(x) = -i for x a 0, f(x) = +i for x < 0.
Then x(t) = 0 is a ( u n i q u e ) Filippov solution of the C a u c h y problem
= f(x), x(0) = 0, t E <0,i) . T h i s t y p e of s o l u t i o n is c a l l e d sliding
motion and there are good reasons to c o n s i d e r it to be the solution.
 134

O n the o t h e r hand let f(x) = I for x ~ 0, f(x) ~ -i for x < 0.

Then the C a u c h y problem ~ = f(x)~ x(0) = 0 has the Filippov solution
x+(t) = t, x (t) = -t a n d x (t) = 0 for t • <0,1al) , x (t) =
- a a
= sgn a . ( t - la[) for t Z laJ. All the Xa(.) solutions are physically
meaningless, they are called parasite solution. For the exact defini-
tion of s l i d i n g and parasite solution see [4] or S e n t i s [8].

3. G e n e r a l i z e d solutions
Our aim is to d e f i n e the solution of ~ • F ( t , x ) in s u c h a manner
that all the sliding solutions are retained a n d all parasite are
expelled. The first definition of t h i s type was given by S e n t i s [8] in
1976 and it w a s as f o l l o w s :

Definition I. F u n c t i o n y(.) : <0,11 -- R n is a g - s o l u t i o n of the

differential relation ~ 6 F(t,x)~ x(0) -- x 0 on (0,i} iff there exists a
sequence {yn } n --1 of p i e c e w i s e linear functions and a sequence {hn}n__ 1
of d i v i s i o n s such that (denote Y n ( h k) by xkn a n d ~(h n) by u n)

i) lira lh I -- 0 ,
n
n~
0
i i) x n = x 0

iii) for e v e r y p o s i t i v e i n t e g e r n a n d k -- 0 , i , . . . , ~ there are

k 6 k k k n k+l k n k k+l k
an F ( h n , X n) a n d e • R such that x -- x + a (h - h ) +
n n n n n n
+ k
n
a n d yn ( • ) is l i n e a r on e v e r y (hk,"n k+l.
n ), k = 0 , i , . . . , ~ n

vn
iv) lim Z II£kll = 0
n~ k= 1 n
v) l i m Yn = y uniformly on <0,i) .
n
Sentis introduced this definition to c o v e r the case (cl s t a n d s for
closure)
F(t,x) = N N cl f ( B 6 ( t , x ) - N) a n d h i s d e f i n i t i o n works
6>0 N C R n + I
~(N)=0
well for such right-hand sides. He p r o v e d that any classic solution of
6 F(t,x) (i.e. any absolutely continuous function x(.) such that
x(t) e F(t,x(t)) a.e.) is a g - s o l u t i o n , any g-solution of ~ • F ( t , x ) is
a classic solution of x • c o n v F(t,x) and there a r e no p a r a s i t e solu-
tions.

4~ Example 2.

For R n= R s e t F l ( t , x ) = [-i} for x < 0 and e v e r y t, F ~ ( t , x ) =

 135

= {-i~i} for x = O a n d e v e r y t and F l ( t , x ) = {l}for x > 0 and e v e r y t,

F2(t,x) = Fl(t,x) for t d y a d i c a l l y irrational and e v e r y x. F o r t =
= (k/2m), k odd, set F 2 ( t , X ) = Fl(t,x) for x ~ < - i [ 2 m, i/2~ and
F2(t,x) = [-I,i} for x 6 (-I/2m, i / 2 ~ o T h e n b o t h F 1 a n d F 2 are u.s.c.
mappings a n d B{t 6 (0,i) l~(Fl(t,x ) ~ F2(t,x))} = 0.

T h e f u n c t i o n y(.), identically equal to zero on <0,i) is not a

g-solution of ~ 6 F l ( t , x ) , x(0) = 0 b u t it is a g - s o l u t i o n of the
relation x 6 F2(t,x), x(0) = 0 o n <0,i} .
This example shows t h a t e v e n for F u.s.c, the s o l u t i o n does
• %
d e p e n d on v a l u e s w h i c h F obtalnes on a set w h o s e projection on t - a x i s
is of m e a s u r e zero. In the s e q u e l we s h a l l m o d i f y the d e f i n i t i o n of
the g - s o l u t i o n to a v o i d this d i s c r e p a n c y .

5. R e @ u l a r Generalized Solutions
Let F be Scorza-Dragonian. Denote GMF = {(t,x,y)ly 6 F(t,x),
t ~ M} i.e. G M F is the g r a p h of the p a r t i a l m a p p i n g FI((0,1)_M)XB .
W e set G~F = N cl GMF and d e f i n e a m u l t i v a l u e d mapping F ~ through
~(M)=0
Me<0,1)
its g r a p h i.e. we set g r a p h F ~= G~F. It is p o s s i b l e to p r o v e that there
exists a set M 0 C (0~i} , u ( M 0) = 0 a n d G~F = cl G. F, so our d e f i n i t -
ion is m e a n i n g f u l l . The set G e F is c l o s e d hence F~0is u.s.c. If the
mapping F is u.s.c, too t h e n F ~ C F b e c a u s e
g r a p h F ~ = cl G M o F C cl GF = GF and {t 6 (0,i) l ~ ( F ~ ( t , x ) # F ( t , x ) ) } C M 0
x
Joe. its m e a s u r e is zero. W e define the s o l u t i o n of x 6 F ( t , x ) t h r o u g h
I

the S e n t i s g-solution of ~ 6 F ~ ( t , x ) ; resulting type of s o l u t i o n being

i

called rg s o l u t i o n It r e ~ a i n s all the nice p r o p e r t i e s of S e n t i s g-so-

l u t i o n a n d is i n d e p e n d e n t ~ o n behaviour of F o n a set of m e a s u r e zero
(in t). If the m a p p i n g F is s u p p o s e d to be o n l y S c o r z a - D r a g o n i a n we
have o n l y g r a p h F ~ C cl GF and F~(t,x) D F(t,x) for t ~ Mq, nonetheless
the r g - s o l u t i o n can be d e f i n e d too. There is f o l l o w i n g characterisation
of r g - s o l u t i o n :

Theorem i. L e t F be a S c o r z a - D r a g o n i a n mapping. Then a function

y(.) is an r g - s o l u t i o n of ~ 6 F ( t , x ) iff for e v e r y M C (0~I) , ~(M)= 0
t h e r e are s e q u e n c e s {yn}~l and {hn }~n=l such that all c o n d i t i o n s of
Definition 1 are f u l f i l l e d and U h N M = #.
n=l n
To p r o v e the
I
t h e o r e m we w i l l u s e the f o l l o w i n g trivial lemma.

Lemma. L e t us s u p p o s e a 6 F ~ ( t , x ) , M C [0,1], p(M) -- 0. T h e n

there are sequences {(tn,Xn)}n= 1 and {an}n= 1 such t h a t a n 6 F ~ (tn,Xn) ,
 136

t n ~ M, n~lim (tn, Xn,a n ) ~ (t,xt~).

Proof. From a 6 ~ ( t , x ) we o b t a i n as a c o n s e q u e n c e of the

identity GF e = GeF and of Lemma 1 that 6 GF ~ : cl G M oM F,(t,x,a)
0
z(M 0 U M) : 0. Hence there exists a sequence {tn,Xn,a n] - (t,x,a) such
that t ~ M 0 O M and a e F(t ,x ). Since F ~ ( T ~ ) : F(~) for • ~ %
n n n n
the proof is complete.

Proof of the theorem: {t e [0,1]IB Since F~(t,X) : F(ttx)} C M 0 ,

xe R n
~(M 0) : 0, the "only if" part of the t h e o rem follows immediately. To
prove the "if" part let y(.) be an r g - s o l u t i o n and M C [0,1] , u(M) = 0.
Then there is a sequence {yn ] ~ y and the sequence {h n] such that the
c o n d i t i o n s (i),...,(v) from D e f i n i t i o n 1 are f u l f i l l e d with F ~ instead
of F. C o n d i t i o n (iii) w r i t t e n e x p l i c i t l y has the f o l l o w i n g form:
,,k+l. k a k- k+l h~) + k a k @ ~ (h~ k
Yn~nn ) = Yn(hn ) + n~hn en' n 'Yn(hn ))"
As a c o n s e q u e n c e of Lemma we o b t a i n that Yn,hn,a
n k k and ekn can be
-- --k --k --k
r e p l a c e d by Y n , h n , a n , S n such that

(2) ~ : {0 : ~0 < ~i <...< ~ n +: : i} n M :

n n n n
u
n k
for every n : 1 , 2 , 3 , . . . , ~ < h k+l, ([k _ h k) < i/(n . Vn ), 2 Ib~ II ~ 0
n n n k:l n
as n ~ ~ and

(3) Yn (~k+l)n : Y n ( ~ ) + ~ k ( ~ kn+ i n ~) + ~k,n ~kn E F ~ ( ~ , y n ( ~ ) )

for n = 1,2,... and k = 0 , 1 , 2 , . . . , v n.

We can p r o c e e d for e x a m p l e v a ~ i f o l l o w s . For every n = 1,2,... we
set ~0n : h0n : 0, y n ( ~ ) : Xn, ~nn : i, ~n(1) : yn(!) ' ~0n : a0"n Let
us d e n o t e i/(n~ ) by p. As a c o n s e q u e n c e of Lemma we can choose [k,
k n ~
~ and %n' such that ( 2 ) i s f u l f i l l e d and lhk- n hkl <n P' % C B p ( ~ h ~ ) )
~ kn E ~ (~k k), ~k E Bp(a k) holds for k : lt2 ..... 9n" We set ~n([~) =
= %k and choose such ~k that (3) is fulfilled. Then
n n
-k - --k+l ~ n ( ~ k ) _ ~ k ( ~ k + l _ [k)
en : Yn(hn ) - n n n
and

U e k H < RYn(hn--
--k+l) _ Yn(hk+l)u + ilYn(hk ) _ ~n(~k)~ + ~kn- akin

]i~k+l _ ~kl, + ,lakil( ~k+In - hk+l'n + ,~k _ h knl) +

+ llYn(hk+l)n - Yn (hk) - ak(hk+in

n - hnk)U < 3p + 2p + [lekll

Hence lim z ~ k ~ = 0. S i m i l a r l y we o b t a i n lim ~n : y u n i f o r m l y on [0,i]

n-~
 137

and the p r o o f is c o m p l e t e .
It m e a n s that using division to c o n s t r u c t a solution we can avoid any
set of m e a s u r e zero.

6. G a u g e approach
To d e f i n e rg-solution we n e e d F to be S c o r z a - D r a g o n i a n (due to the
definition of F*) b u t by m e a n s of a v o i d i n g the sets of m e a s u r e z e r o we

can define the rg-solution for q u i t e a general system. In the sequel,

using gauge approach, we introduce another procedure to d e f i n e solut-
ions. Let us r e m i n d that a gauge is an a r b i t r a r y real valued positive
function and a division A = {ty] is s u b o r d i n a t e d to a g a u g e 6 (or A is
6-fine, A < 6) iff t i + i- t 1 < 6(t 1 ) " We shall say t h a t a set Q is a
gauge set iff for e v e r y positive constant c there exists a 6 @ ~ such
that sup 6(t) < c and for e v e r y 61~...,6nE Q there exists a 6 E Q such
that 6 N min(61,...,6n).
There is a w e l l ' k n o w n theorem about 6-fine divisions saying that
for e v e r y 8 there is a 6-fine division which is finite, see K u r z w e i l
[6]. In o u r case this theorem doesn't hold because we o p e r a t e with so
called left divisions. But a similar theorem holds with a countable di-
visions. Let us n o t e that using general division instead of left o n e we
don't succeed in r e j e c t i n g parasite solutions.
Let Q be a g a u g e set. We shall say that y is an 9 - s o l u t i o n of
E F(t,x)~ x(0) : x 0 iff all items of D e f i n i t i o n 1 are fullfiled with
6-fine division, 6 E ~ i.e.

¥ ¥ 9 3 3 ~ (I~AI < ~, ly - XAI < ~) .

~>0 8~0 A<6 ~A [A XA
The following theorem c a n be p r o v e d .

Theorem 2. L e t F be b o u n d e d and let Q be a g a u g e set. Then there

exists an Q - s o l u t i o n .

Proof: Let p > 0 be s u c h that IIyll _< p for all y @ F(t, x),
(t,x) C [0,i] X R n and let K be t h e set of all x(.) c C(<0,1)) such
that
a) Ix(t) l -< p for every t @ [0,i]
and
b) ]X(t I) - x(t2)t < plt I t2[ for every t l , t 2 @ [0,i] .
The K with the norm max is the compact metric space. Let 6 E ~. W e
shall construct a set S 6 C K. Let SJ6 be the set of all functions
fulfilling all the conditions of D e f i n i t i o n 1 and such that (see
 138

condition iv) 2 ll~.ll ~ sup 6(t) It can be proved, by the m e t h o d of

1
transfinite i
sequences (see [13]), that S J6 is non-empty. Every
J
function x(.) E S 6 can be modified, by subtracting jumps E i in points
t. of d i v i s i o n 4, to obtain a function y(.) E K. This procedure results
l
in a set S 6 C K. The set K is compact, hence ~ ~6 # @' It is easy to
6e~
see that every function x('.), X E N $6 is an ~ - s o l u t i o n , w h i c h
Be
completes the proof.
Let us denote ~0 = {6(.)16~ a(6) > 0}, ~r = {6(.)IS(t) ~ a(6) a.e.,
a(6) > 0}. Then it is possible to prove that Q 0 - s o l u t i o n s are e x a c t l y
the Sentis g-solutions and Qr-solutions are precisely the rg-solutions.
Using the results m e n t i o n e d above we can say that for F u.s.c, the
gauge set ~ r is the good one to define a solution. But this is not true
for F S c o r z a - D r a g o n i a n because ~r-SOlutions are the solutions of ~ ~ F e,
F ~ b e i n g u.s.c., F ~ D F a.e. Hence we cannot expect Q r - S O l u t i o n s to be
solutions of x E conv F. So a natural p r o b l e m arises:
W h a t is the smallest but sufficient gauge set for S c o r z a - D r a g o n i a n
r i g h t - h a n d side?

References

[i] JARNfK,J. Constructing t h e Minimal D i f f e r e n t i a l R e l a t i o n w i t h

P r e s c r i b e d S o l u t i o n s , ~asop. pro p~st. matem. 105 (1980),
311-315.

[2] JARNfK,J., KURZWEIL,J., On Conditions on Right Hand Sides of

Differential Relations, ~asop. p~st. matem. 102 (1977), 334-339.
[3] JARNfK,J., KURZWEIL,J., Sets of Solutions of Differential
Relations, Czech. Math. Journ. 106 (1981).
[4] KRBEC,P., On Nonparasite G e n e r a l i z e d S o l u t i o n s of D i f f e r e n t i a l
R e l a t i o n s , ~asop. p~st. mat. 106 (1981).
[5] KRBEC,P., On Nonparasite S o l u t i o n s of D i f f e r e n t i a l R e l a t i o n s ,
to appear.

[6] KURZWEIL,J., Nichtabsolut konvergente Integrale, T e u b n e r - T e x t e

zur M a t h e m a t i k 26, B.G. Teubner, Leipzig 1980.

[7] OLECH,C., E x i s t e n c e of S o l u t i o n s of non-convex o r i e n t o r f i e l d s ,

D i f f o r d 1974, Summer School on O r d i n a r y D i f f e r e n t i a l Eauations,
Brno 1979.

[8] SENTIS,R., Equations differentielles a second membre measurable,

B o l e t i o n U.M.I. (5) 15-B (1978), 724-742°
 139

[9] T rSTOGONOV,A.A., 0 p l o t n o s t i i 9 r a n i ~ n o s t l mno~estva r e ~ e n i j

d i f f e r e n c i a l n o g o v k l j u ~ e n i j a v bana~hovom p r o s t r a n s t v e , DAN 1981,
No. 2, tom. 261.

"[I0] TOLSTOGONOV,A.A., K teoremam s r o v n e n i j a d l j a d i f f e r e n c i a l n y e h

v k l j u ~ z n i j v lokalno uypuklom ~ a o s t r a n s t v e . I. S u ~ e s t v o v a n i e
r e ~ e n i j , Differencialnye uravnenija 1981, tom. XVII, No. 4,
If. Svo]stva re~eni]. Differencialnye uravnenija 1981, tom. XVII,
No. 6.

[ii] TOLSTOGONOV,A.A., 0 d i f f e r e n c i a l n y c h v k l j u d e n i j a e h v banaehouych

pros t r a n s t v a c h i nepreryvnyc~ s e l e k t o r a c h , S i b i r s k i j matem.
~urnal 1981, tom. XXII No.4.

[12] VP~KO~,I., A ~ew D e f i n i t i o n and some M o d i f i c a t i o n s

of Filippov
cone, Lecture Notes in Math. 703, Equadiff IV Proceedings, Prague
1977, Springer Verlag, Berlin-Heidelberg-New York 1979.
[13] KRBEC,P., Weak s t a b i l i t y of Multiualued D i f f e r e n t i a l Equations,
Czechoslovak Math. Journal 26 (i01), 1976.
UNIFORM ZEROS FOR B E A D E D STRINGS
K. KREITH
University of Califo~'nia
Davis CA 95616, U.S.A.

I. Introduction. Early a t t e m p t s to model the f u n d a m e n t a l v i b r a t i o n of a

musical s t r i n g f o c u s s e d on three p h y s i c a l p r o p e r t i e s which were believed to

underlie this p h e n o m e n o n : isochronism, the pendulum principle and a

"simultaneous c r o s s i n g of the axis" [3]. W i t h the discovery that the small

vibrations of such strings are d e s c r i b e d by h y p e r b o l i c partial differential

equations, interest in these p h y s i c a l concepts declined. Isochronism became

embodied in the "small amplitude assumption" which u n d e r l i e s the l l n e a r i t y of

the r e s u l t i n g equation, while the pendulum principle (asserting that restoring

forces are proportional to d i s p l a c e m e n t s from e q u i l i b r i u m ) t u r n e d out to be

incorrect for the wave equation. The notion of a simultaneous crossing of the

axis has become identified with separation of v a r i a b l e s and does not seem to

have been pursued in its own right.

However given a linear hyperbolic PDE of the form

(1.1) utt - Uxx + p(x,t)u = O,

with p(x,t) continuous and p o s i t i v e for 0 < x < L and t > O, and g i v e n

boundary conditions such as

(~ .2) Ux(O,t) - Ux(L,t) = O,

the question of a s i m u l t a n e o u s c r o s s i n g of the axis is an important one.

Specifically, the question arises whether it is possible to a s s i g n C a u c h y data

of the form
 142

(I .3) u(x,0) = 0; ut(x,0) = g(x)

for 0 < x < L such that the solution of (1.1)-(1.3) will satisfy

u(x,T)~- 0 for 0 < x < b

for some T > 0. We shall refer to such solutions as having a uniform zero at

t = T.

Except in t h e ease of separation ~ variables there seems to be no simple

answer to this question. As such it is n a t u r a l to c o n s i d e r a semi-discrete

approximation to (1.1)-(1.3) corresponding to a beaded string. In this context

one obtains [7] a system of ordinary differential equations of the form

(1.4) --- + G(t)u = 0

dt 2

subject to initial conditions of the form

(1.5) u(0) ~ 0; u ' ( O ) = g.

The problem of choosing g in (1.5) so as to satisfy ~(T) = O for some T > 0

is n o w the classical problem of establishing the existence of conjugate points

of zero relative to (1.4). Such problems have been studied by M. Morse [9] and

W. T. Reid [10] in the more general context of H a m i l t o n l a n systems. More

recently Ahmad and Lazer [I] have also studied conjugate points under the

assumption the entries of G(t) satisfy appropriate posltivity conditions.

In the case at hand, the matrix function G(t) is a Jacobi matrix given by

gii(t) = 2 ÷ Pi(t) for I < i < n,

(I . 6 ) gi,i-1 ~ gi-l,i = -I for 2 < i < n,

gij ~ 0 for li-jl ~ 2.

 143

While this m a t r i x f u n c t i o n has the s y m m e t r y r e q u i r e d in [9] and [I0], the

variational criteria for c o n j u g a t e points established t h e r e i n are based on

positive definiteness and p r o v i d e no information regarding the sign of the

solution which realizes a particular conjugate point. Also, the essential

indefiniteness of G(t) prevents the techniques of [I] from being b r o u g h t to

bear in e s t a b l i s h i n g uniform zeros for s o l u t i o n s of (1.4). Accordingly,

criteria for the existence of uniform zeros of (1.1) would seem to r e q u i r e the

development of novel t e c h n i q u e s for e s t a b l i s h i n g the existence of conjugate

points for (1.4).

2. The Oppositional Mode of Vibration. A special case of interest in

connection with (1.4) and (1.6) is that where the initial data

~ col(g I .... ,gn ) in (1.5) satisfies

(-1)3gj < O; I < j < n.

In this case the solution of (1.4) and (1.5) also satisfies (-1)Juj(t) < 0 for

sufficiently small values of t and is s a i d to be (initially) in an

oppositional mode of vibration.

The s p e c i a l Jacobi form of (I .6) makes tractable the problem of

establishing the e x i s t e n c e of the c o n j u g a t e point T whose corresponding

solution u(t) = col(u1(t),...,Un(t) is in an oppositional mode for 0 < t < T.

Indeed, if we define v(t) = col(v1(t) ..... Vn(t)) by

vj(t) = (-1)Juj(t)

then v(t) is a solution of

v" + F(t)v ~ 0

(2.2)
 144

where fij = Igijl and fi ~ Igil for I ~ i, j ~ n. Because of the positivity

properties of F(t) and f one can apply the techniques of Ahmad and Lazer [I]

te e s t a b l i s h the existence of a conjugate point T for (2.2) which is realized

by solution ~(t) whose components are positive for 0 < t < T.

A nonlinear version of this problem has been considered by Duffin [4] in

connection with the "plucked string" (corresponding to a r i g h t focal point).

Indeed given appropriate pos~tivity conditions on h(t,v) one can also use the

techniques of Krasnoselskii [6; Ch. 7.4] to establish the existence of p o s i t i v e

solutions of boundary value problems of the form

v" + h(t,v) = 0

(2.3)

v(0) ffi v ( T ) ffi O,

leading to more general equations which allow for solutions in this oppositional

mode.

While of interest, these results are of little help in establishing uniform

zeros for (1.1). For as we seek to approximate (1.1) by s y s t e m s such as (1.4)

and let n ÷ ~, solutions in the oppositional mode do not converge to solutions

of (I .I ).

For this reason one is led to the more difficult problem of establishing

the existence of c o n j u g a t e points for (1.4) which are r e a l i z e d by p o s i t i v e

solutions.

3. Positive Solutions. In case the matrix G(t) given by (1.6) is a c o n s t a n t

Jacobi matrix, the e x i s t e n c e of a c o n j u g a t e point realized by a p o s i t i v e

solution can be e s t a b l i s h e d by a l g e b r a i c means. In this case (1.4) can be

written as

(3.1) G-Iu '' + u = 0

 t45

where G -I is totally posit lv e in the sense of Gantmacher and Krein [5]. As

shown in [5], it now follows that G -I has n simple positive eigenvalues

11 > 12 > ....> An > 0, where ~I corresponds to an eigenfunction !I which may

be taken to be positive. Accordingly the choice ~ = ~I in (1.5) leads to

~I " ii sin t/@~- I and T = ~W~I. (It also f o l l o w s from [5] that the

solution --nU = ~n sin t/V'-~n corresponds to the o p p o s i t i o n a l mode of

vibration considered in §2).

In order to deal with non-constant G(t) in (1.4) it will be n e c e s s a r y to

give a n o n - a l g e b r a i c argument for the existence of the above solution ~1(t).

To that end we consider the case where

(3.2) O(t) = r 0 + E(t),

r0 being a constant matrix with entries

Yii = p > O; Yi,i-1 = Yi-l,i = -I; 7ij = 0 otherwise

and E(t) - dlag(~1(t), .... ~n(t)) playing the role of a p e r t u r b a t i o n of F O.

By [5] FO has positive eigenvalues ~I < u2 < "'" < ~n' for which we establish

the following property.

3.1 Lemma. ]For sufficiently large values of p the eigenvalues of r0 satisfy

(3.3) I< E < ; 2 < i < n.

I
Proof. The eigenvalues of Pi FO satisfy ~ ~I < "'" < p! ~n and tend to

I as p ~ ®. Therefore (3.3) follows for the eigenvalues of p F0 and for the

eigenvalues of F0 as well.

In order to establish topological c r i t e r i a for the e x i s t e n c e of uniform

zeros it will be useful to regard solutions of

(3.4) ~" + [r0 + ~(t)]~ = 0; ~(0) ~ 0, u,(0) =

 ~6

as trajectories in ~n which emanate from the origin with initial velocity ~.

We seek to show the existence of g > 0 such that the corresponding t r a j e c t o r y

+
e x i t s the p o s i t i v e n-tant ~ through the origin. To that end we denote the

normalized eigenvectors of F0 (corresponding to the eigenvalues ui ) by ~i'

+
requiring that ~I lie in ~ n and, more generally, that the sum of the

components of each ~i be n o n n e g a t i v e . This sign convention has the

c o n s e q u e n c e that when we express any ~ 0 in the form g = c i ¢ I ÷ ... + c A ,

then ci ~ 0 for ] < i < n.

As in [4] we define a c o n t a c t point of a trajectory u(t) as its first

point of intersection with a coordinate plane. An exit point is a contact point

at which the trajectory also crosses that coordinate plane. In the oppositional

mode one can r e a d i l y show [4] that such first c o n t a c t points are also exit
+
points, but this need not be the case for trajectories in IR . However, the
n
following t h e o r e m shows that under the c o n d i t i o n of Lemma 3.1 such an
+
equivalence also exists for trajectories in ~n"

3.2 Theorem. If v(t) is a trajectory of

_v,, + ro~ : o; v_(o) = o, v_,(o) : g > o,

and if c o n d i t i o n (3.3) is s a t i s f i e d , then the point at w h i c h v(t) first

intersects a coordinate plane bounding ~+ is also an exit point.

n

Proof. Suppose the contact point occurs at t = to and lies in the c o o r d i n a t e

plane (~,~j) = O, w h e r e e. is a unit vector along the positive v -axis.

--J 3
Because of (3.3) and the fact that the ¢i c o m p o n e n t has m a x i m a l a m p l i t u d e

among the characteristic directions, it follows that we must have

3~
(3-5) ~- < ~ I
2 to < ~ and ~ <4~i to < 2-

for 2 < i < n. Writing _v(tO) in terms of the eigenvectors of r0 leads to

the equation
 147

(3.6) (c1~1-%A2-...-c~,~j) = 0

for appropriate choice of positive constants c l , . . . , c n. If now ~ ( t O) were

no_!an exit point we would also have (~'(to),ej) = 0 and, because of (3.5),

(dl¢l+d2¢2 + .... d ¢ ,~j) = 0

for positive constants d 1 , . . . , d n. This contradicts (3.6) and completes the

proof.

Remarks

I. Given specific eigenvalues satisfying (3.3) the above argument remains

valid under small perturbations of the trajectories. Accordingly Theorem

3.2 remains valid for (3.4) when ~(t) is sufficiently small.

2. The fact that contact points are also exit points assures that contact

points will vary continuously with initial data, This observation is

crucial to the proof of Theorem 3.4 below.

3. In the case of opposltional vibrations the fact that the initial velocity

vector g has gj = 0 assures that the resulting trajectory exits the

oppositlonal quadrant at t = 0 across v. = O. Lemma 3.3 shows that for

3
(3.4) a very different situation exists.

3-3 Lemma (Crossover property). If in Theorem 3.2 the vector g has gj = O,

then the trajectory v(t) does not exit ~ + across the coordinate plane
-- n

(Z,e_j) = o.

Proof. The proof is similar to that of Theorem 3.2 by writing g = c i ¢ I + 02~

where ~ lies in ¢~. The fact that all components of $ will satisfy
37
H <~'~i to < -5- at the time of contact precludes an exit across the plane

(z,ej) = O.
 148

The above properties of trajectories of ~'' + F0~ : 0 lead to the

existence of a conjugate point as follows. Among all initial velocity vectors
g satisfying g > 0 all llgll = i we define

T : {g : v(t) exits R+ across (~,ej) ~ 0}.

3 -- -- n

A well known corollary to Sperner's lemma then leads to the fact that
n
A T # @ and the following result.
j:l 3
3.4 Theorem, Under the hypotheses of Theorem 3.2, and for sufficiently @mal~
perturbations ~(t), th__~es[stem (3.4) has a conjugate point of zero which is
realized by a trajectory in R +.
n

References

]. S. Ahmad and A. Lazer, On the components of extremal solutions of second

order systems, SIAM J. Math. Anal. 8(1977), 16-23.

2. P. Alexandroff and H. Hopf, Topolo~ie, Berlin, Springer Verlag, 1935.

3. J. Cannon and S. Dostrovsky, The Evolution of DynamiQs, Vibration Theory

from 1687 to 1742, New York, Springer Verlag, 1981.

4. R . J . Duffin, Vibration of a beaded string analyzed topologically, Arch.

Rat. Mech. and Anal. 56(1974), 287-293.

5. F. Gantmacher and M. Krein, Oscillation Matrices and Kernels and Small

Vibrations of Mechanical Systems, Moscow, State Publishing House, 1950.

6. M . A . Krasnoselskii, Positive Solutions of Operator Equatlons, Gronlngen,

Noordhoff, 1964.

7. K. Krelth, Picone-type theorems for seml-discrete hyperbolic equations,

Proc. Amer. Math. Soc. 88(]983), 436-438.

8. K. Kreith, Stability criteria for conjugate points of indefinite second

order differential systems, J, Math. Anal. and Applic., to appear.

9. M. Morse, A Generalization of the Sturm separation and comparison theorems

in n-space, Math. Annalen I03(1930), 72-91.

10. W. T. Reid, Sturmian Theory for Ordinary Differential Equations. New York,
Springer Verlag, 1980.
PERRON INTEGRAL, PERRON PRODUCT
INTEGRAL AND ORDINARY LINEAR
DIFFERENTIAL EQUATIONS
J. KURZWEIL and J. JARN[K
Mathematical Institute, Czechoslovak Academy of Sciences
115 67 Prague 1, Czechoslovakia

I. Perron integral and Perron product integral

A finite set 6 = { X o , t l , x I ..... tk,Xk} is c a l l e d a partition of an

interval [a,b] if

a = x 0 < x I < ... < xk = b , xj_ I ~ tj ~ xj

for j = 1,2 ..... k . Let ~ : [a,b] + (O,~) (no c o n t i n u i t y or measura-

bility properties required). A partition A is s a i d to be ~ - f i n e if
Exj_1,xj] C (tj - 6(tj), tj + ~(tj) I .

k
Let f : [a,b] + ~ , put S(f,A) = 3)~ 1.f (=t 9 (xj - xj_ 1) . It is w e l l

known (cf. [13 , [2]) that the following two c o n d i t i o n s are equivalent:

f is P e r r o n integrable (P-integrable) over [a,b~ ,

b (~'J)
q = (P) I f(t) dt
a

for e v e r y e > 0 there exists such a 6 : [a,b] + (0,~) that

(I .2)
lq - S(f,A) I ~ e for e v e r y ~-fine partition A of Ea,b3

Condition (1.2) makes good sense since

for e v e r y ~ : [a,b~ * (0,~) there exists a 6-fine

(I .3)
partition A of Ea,b] .

1.1. REMARK. The p r o o f of (1.3) is easy: If (1.3) were false for a

on [a,bJ , it w o u l d be false either for 6 on Ea, (a + b ) / ~ or for
on [(a + b)/2, b~ a n d this procedure, if c o n t i n u e d , leads to a c o n -
tradiction.

Denote by M the ring of r e a l or c o m p l e x n x n matrices. For

 150

A : [a,b] ÷ M and a partition A of [a,b] put

P(A,A) = (I + A ( t k ) ( X k - X k _ l ) ] . . . (I + A ( t l ) ( X I - X0) 1 ,

P(A,A) = exp[A(tk)(X k - Xk_1))... e x p [ A ( t l ) ( X 1 - x0) ] .

The following result is w e l l known (cf. [4], [5]): If A is c o n t i n u o u s

and if U is the m a t r i x solution of

= A(t)x , (1.4)

U(a) = I , then both P(A,A) , ~(A,A) converge to U(b) in the follo-

wing sense:

For every e > 0 there exists such an n > 0 that

I Iu(b) - P(A,A)I I ~ a , I IU<b) - ~(A,A)I I ~ e for e v e r y
(I .5)
partition A of [a,b~ satisfying xj - xj_ I < n ,
j = 1 ..... k .

In [5~ the L e b e s g u e product integral was introduced in a w a y analo-

gous to the u s u a l introduction of t h e B o c h n e r integral and it w a s proved
that U(b) is e q u a l to t h e L e b e s g u e product integral of exp(A(t) dt)
over [a,b~ provided A is L e b e s g u e integrable in t h e u s u a l sense. In
the n e x t definition, the l i m i t i n g process from (1.2) is a p p l i e d to the
product P(A,A) - of c o u r s e without any continuity or m e a s u r a b i l i t y con-
dition on A .

1.2. DEFINITION. Let Q E M be r e g u l a r . A is said to be Perron pro-

duct-integrable over [a,b~ (P-integrable), Q is c a l l e d the Perron
b
product integral (P-integral) of A and denoted by PI(I + A(t) dt) ,
a
if for e v e r y ~ > 0 there exists such a 6 : [a,b~ ÷ (0, ~) that
IIQ - P(A,A)I I ~ ~ for e v e r y 6-fine partition A of [a,b3

1.3. REMARK. The same concept of the P-integral is o b t a i n e d if p(A,A)

is r e p l a c e d by ~(A,A) in D e f i n i t i o n 1.2.
b
The integral PI(I + A(t) dt) has properties analogous to t h o s e
a
b
of the integral (P)~f(t) dt . T h e p r o p e r t i e s of the latter are listed
a
in S e c t i o n 2, the a n a l o g o u s p r o p e r t i e s of t h e f o r m e r in S e c t i o n 3. In
Section 4 some relations to A C G , - f u n c t i o n s and to the equation (1.4)
are m e n t i o n e d .
 151

2. Properties of the P e r r 0 n integral

t
If f is P - i n t e g r a b l e over [a,b] then F(t) = (P)If(s) ds
J (2.1)
a
exists for t 6 (a,b~ . We put F(a) = 0 .

If f is P - i n t e g r a b l e , then F is c o n t i n u o u s and
(2.2)
F(t) = f(t) a.e. Moreover, f is m e a s u r a b l e .

Let f be P - i n t e g r a b l e over [a,b] . Then the following

assertion holds: if C C ~a,b] is of m e a s u r e zero and
> 0 , then there exists such a 6 : C ÷ (0,~) that
r
IF(qj) - F(~j) I < e provided ~j e C , ~4J < Tj < qj (2.3)
j=l
--<--<j+1 and [%j,qj] C [~j - ~(~j), Yj + 6(Tj)) for j =

Let F : [a,b 3 ~ ~{ have derivative a.e. and satisfy the

assertion from (2.3). Put f(t) = F(t) if F(t) exists, (2.4)
f(t) arbitrary otherwise.
b
Then (P)If(t) dt exists and equals F(b) - F(a)
a
t t
?

If (P)If(s) ds exists for t < b and if lim (P)If(s) ds

aJ b t+b- aj (2.5)

q £ ~ , then (P)If(s) ds exists and equals q .

a

3. Properties of the P e r r o n product integral

If A is P - i n t e g r a b l e over [a,b~ then U(t) =

t (3.1)
PI(I + A(s) ds] exists for t ~ (a,b~ . We put U(a) = I .
a

If A is P - i n t e g r a b l e , then U(t) is r e g u l a r at e v e r y t ,
U is c o n t i n u o u s and U(t)u-l(t) = A(t) a.e. Moreover, (3.2)
A i is m e a s u r a b l e .

Let A be P-integrable. Then the assertion (2.3) holds

(3.3)
with F replaced by U .
 152

Let U : [a,b3 ÷ M be c o n t i n u o u s , regular at e v e r y t and

differentiable a.e., a n d let it s a t i s f y the m o d i f i e d asser-
t i o n of (2.3) (cf. (3.3)). Put A(t) = U(t)U-1(t) if U(t)
b (3.4)
exists, A(t) arbitrary otherwise. Then PI(I + A(t) dt)
a
exists and equals U(b)U-1(a) .

t
If F I(I + A(S) ds) exists for t < b and if
a
t
lim Pf(I + A(s) ds) = Q e M is regular, then
(3.5)
t+b-
b a
PI(I + A(s) ds) exists and e q u a l s Q .
a

4. ACG,-solutions of l i n e a r o r d i n a r y differential equations

The c o n c e p t of an A C G ~ - f u n c t i o n (cf. [~ ) extends without complica-

tions to f u n c t i o n s with values in f i n i t e d i m e n s i o n a l l i n e a r spaces. A
function u : [a,b~ ~ Rn (e n) is c a l l e d an A C G , - s o l u t i o n of (1.4) if
u is an A C G , - f u n c t i o n and s a t i s f i e s (1.4) a.e. In an a n a l o g o u s manner
the c o n c e p t of a m a t r i x A C G ~ - s o l u t i o n of (1.4) is to be u n d e r s t o o d .

It is w e l l k n o w n that F from (2.1) is an A C G , - f u n c t i o n and that

every ACG~-function is the p r i m i t i v e of its d e r i v a t i v e (in the s e n s e of
(2.1), (2.2)). It f o l l o w s from (2.3) and (2.4) t h a t F is an A C G , - f u n c -
tion iff it s a t i s f i e s (2.4). It f o l l o w s from (3.1) - (3.4) that U from
(3.1) is an A C G , - f u n c t i o n a n d that e v e r y A C G ~ - f u n c t i o n U : [a,b3 ÷ M
can be w r i t t e n in the f o r m
t
U(t)U-I(a) =PIII + A(s) ds)
a
provided U(t) is r e g u l a r for e v e r y t . At the same time w e h a v e
t t
U(t) - U(a) : (P)[ 6(s) ds : (P)I A(s) U(s) ds
w

a a
for t E [a,b] , that is, U is an A C G , - m a t r i x solution of (1.4).

Thus w e h a v e o b t a i n e d a c l a s s of L D E ' s the s o l u t i o n s of w h i c h are

ACG~-functions; these LDE's have the u s u a l existence and u n i q u e n e s s pro-
perties.
 153

Denote by H([a,b~) the set of such A : [a,b3 ÷ M that

b
P~(I + A(t) dt I exists. Let us find some effective conditions for
a

A ~ H([a,b 3 )

Assume that c > 0 , B : ~c,c 3 + M is c o n t i n u o u s , I + B(t) is

regular for t £ [-c,c 3 and B is l o c a l l y absolutely continuous on
~c,c 3 \{0} . We have

if A(t) = ~(t)~ + B(t)3 -1 a.e. then A 6 H(~c,c~) , (4.1)

if k 6 {0,1 .... } , A(t) = B(t)~ - B(t) + ... + (-1)kBk(t)~

c (4.2)
t

a.e., ~I I B ( t ) B k + 1 ( t ) I I dt < ~ , then A E H([-c,~) .

d

--C
In t h e case (4.1), I + B(t) is a f u n d a m e n t a l matrix of (1.4), in the
case (4.2) the substitution x = ~ + B(t)~y leads to the result.

Let ~ > 0 , ~ > 0 , T, S E M . If a < 1 + ~ , then there exists

such a continuous B : ~ ÷ M that B(0) = 0 , B(t) =

Iti-S[T cos ]t1-8 + S sin [tt-sl for t ~ 0 . Let c > 0 be so small

that I + B(t) is r e g u l a r for t 6 [-c,c3 . Then (4.1) may be applied.
If a < I + 8/2 then (4.2) may be applied with k = 0 . If I + 8/2
< I + 2~/3 then (4.2) may be applied with k = I ; moreover, if
t
TS - ST ~ 0 then IA(s) ds is u n b o u n d e d for t + 0- so that
I -I
(P) [A(t) dt does not exist.
J
-I

5. The Saks-Henstock Lermaa

In t h e proof of the properties (2.2) and (2.3) of the Perron integ-

ral the key part is p l a y e d by the following

5.1. LEMMA (Saks, Henstock). Assume that f is P - i n t e g r a b l e over

t
I

Ea,b~ , F(t) = (P)I f(s) ds . Let e > 0 and let the gauge corres-
a

pond to e according to (1.2). Let

~j, ~j, nj 6 [a,b~ , ~j ~ ~j ~ qj ~ ~j+1

(5.1)
[~j,~j~ C (xj - 6(Tj), ~j + ~(~j)) , j = 1,2 ..... r .
 154

Then
r
If(T'j) (~j - 6j) -'F(qj) + F(~j) I < 2e .
9=I

For the P e r r o n product ~integral, an a n a l o g o u s role in the p r o o f of

the p r o p e r t i e s (3.2) and (3.3) is p l a y e d by

5.2. LEM~LA. There exist e0 > 0 and K > 0 depending on n o n l y so

that the f o l l o w i n g holds:

Assume that A is P - i n t e g r a b l e over ~a,b3 ,

t
U(t) PI(I + A ( s ) ] d s , U(b) = Q .
a
Let 0 < s < E0/I IQ-III and let the g a u g e 6 correspond to e accor~
d i n g to D e f i n i t i o n 1.2. Let (5.1) hold. Then
r
I II + A(~j) (n'j - {j) - U < n j ) U -1(~:j)II 5_ K~ .
j=1

R e f e r e n c e s

Eli K U R Z W E I L , J.: Nichtabsolut konvergente Integrale. Teubner, Leipzig

1980, T e u b n e r - T e x t e zur M a t h e m a t i k , 26.
[2] K U R Z W E I L , J.: The integral as a Limit of integral Sums. J a h r b u c h
~ b e r b l i c k e M a t h e m a t i k 1984, 105-136, B i b l i o g r a p h i s c h e s I n s t i t u t
A G 1984.
[3] S C H L E S I N G E R , L.:Einfu'hrung in die Theorie der gew~hnlichen Diffe-
rentialgleichungen auf funktionentheoretischer Grundlage. B e r l i n
1922.
[4~ GANTMACHER, F. R.: Theory of Matrices. Moskva 1966 (Russian).
E5] DOLLARD, J. D. and F R I E D M A N , CH. N.: Product Integration with Appli-
cations to Differential Equations. Univ. Press, C a m b r i d g e 1979.
E6] SAKS, S.: theory of the Integral. M o n o g r a f i e m a t e m a t y c z n e VII. G.E.
Stechert, N e w York - W a r s z a w a 1937.
ON THE ZEROS OF SOME SPECIAL
FUNCTIONS: DIFFERENTIAL EQUATIONS
AND NICHOLSON-TYPE FORMULAS
M. E. MULDOON
Department of Mathematics, York University
North York, Ontario M3d 1t'3, Canada

1. Introduction. There a r e many r e s u l t s in the literature on s p e c i a l

f u n c t i o n s c o n c e r n i n g t h e way i n which a z e r o o f a f u n c t i o n c h a n g e s w i t h r e s p e c t t o
one o f t h e p a r m l e t e r s on which t h e f u n c t i o n depends. Methods b a s e d on d i f f e r e n t i a l
equations, in particular S t u r m i a n methods, a r e o f t e n u s e f u l i n t h e s e d i s c u s s i o n s .
O t h e r methods a r e r e l a t e d t o i n t e g r a l r e p r e s e n t a t i o n s f o r t h e f u n c t i o n s add seem t o
be p r o v a b l e , though n o t e a s i l y d i s c o v e r a b l e , by d i f f e r e n t i a l e q u a t i o n s methods.
Among t h e s e a r e methods b a s e d on N i c h o l s o n ' s f o r m u l a [13, p.444]

(1) J2(z) + ~(z)= ~ I Ko(2Z simh t ) c o s h 2vt d t , Re z > 0 ,

w 0
and a companion formula

4
(2) Ju(z)a ~(z)l~ - Yu(z)o J u ( z ) / a ~ = - W o K0(2z sinh t)e -2~t dt , Re z > 0 ,

from which it follows [13, p.508] that

m

(3) dc/dv = 2c ~ 10(2c sinh t)e -2vt dt .

0
Here J and Y are the usual Bessel functions, K0 is the modified Bessel

function and, in (3), c = c(v,k,a) is an x-zero of the linear combination

Cu(x) = COS a J u ( x ) - s i n a Y (x) .

Formula (1) was used by L. Lorch and P. Szego [9] t o show some r e m a r k a b l e
sign-re~larity properties of the higher k - d i f f e r e n c e s of the sequence {c(~,k,a)}
in the case J~J -> ~ • Beyond i t s o b v i o u s u s e t o show t h a t c increases with v ,

(3) has been u s e d t o g e t f u r t h e r i n f o r m a t i o n a b o u t t h e s e z e r o s ; s e e [10,11] f o r

references. A. E l b e r t h a s u s e d (3) t o show t h a t j~k(=C(~,k,0)) i s a concave

f u n c t i o n on ~ on (-k,®) . E l b e r t and A. L a f o r g i a have u s e d (9) in s e v e r a l r e c e n t

.2
papers. They p r o v e d , f o r example, t h a t Jvk i s a convex f u n c t i o n o f ~ on (0,m)

[6J and t h e y have shown r e c e n t l y ( p e r s o n a l c o m u n i c a t i o n ) that d 3 j~k/d~3 > 0 ,

O<v<~

2. Other Nicholson-type formulas. The usefulness of (I), (2) and (3) suggests
the desirability of having similar formulas for other special functions. L. Durand
[3,4] has given results analogous to (i) for some of the classical orthogonal
polynomials. The simplest of these, for Hermite functions, is [3, p.371]
 156

4,) e X2CH]Cx) + ]4x)l = 2A+IF(A+I) [-0 e - ( 2 A + l ) t + x 2 t a n h t ( c o s h t s i n h t ) - 1 / 2 d t .

Durand does n o t u s e d i f f e r e n t i a l e q u a t i o n s b u t p o i n t s o u t [3, p . 3 5 5 ] t h a t , once t h e

results a r e known, t h e y can be c h e c k e d b y d i f f e r e n t i a l equations methods. In fact
J.E. Wilkins, Jr. [14] 4see a l s o [12, p p . 3 4 0 - 3 4 1 ] ) p r o v e d (1) by s h o w i n g t h a t b o t h
sides satisfy t h e same t h i r d o r d e r d i f f e r e n t i a l e q u a t i o n and h a v e t h e same
asymptotic behaviour as z - +~ . I {10] d i d t h e same f o r e q u a t i o n (2) u s i n g a
third-order nonhomogeneous e q u a t i o n .
More r e c e n t l y , I have tried t o d i s c o v e r w h e t h e r t h e r e i s a n a t u r a l way i n w h i c h
these formulas arise in a differential equations setting. I present such a setting
here for Bessel functions but it is not clear to me yet whether the method applies
to a general situation of which the Bessel function case would be a particular
example. It turns out to be convenient to consider the more general formulas
[2,7,13]

(5) J/~(z)Ju(z) + Y (z)Yu(z) = ~ ~0 K _~(2z sinh t)[e(/~+~)t+ e-(/~+U)tcos(/A-u)w]dt ,

oo

aO

(8) J (z)Y (Z)-Ju(z)Y 4z) = ~

z ~0 Ku+"(2z sinh t)[e(Ze-/a)tsin /~ - e(/~'-u)tsin us]dr.

These are all valid for Re z > 0 with JHe(t~ + u)[ < 1 in (6) and (8) and
jRe(u - /~)J < I in (5) and (7).
Clearly 41) is got from (5) by setting y = u while, as pointed out in [2],
(2) (and hence (3)) is got from (7) by dividing by y -~ and letting # -~ v .
Dixon and Ferrar [2, p.142] find an analogue of (2) based on a similar treatment of
(8).
The corresponding analogue of (3) is

dc/d~ = -(2c/w) K2~(2c sinh t)[2t sin w~ - w cos us]dr , c > 0 , luj < ~ ,

but this is both more complicated and has a smaller range of validity then (3).

3. A differential e q u a t i o n s p r o o f o f 46). The p r o o f s o f 45), 47) and (8) a r e

quite similar t o t h a t w h i c h we w i l l g i v e f o r 46). We may c l e a r l y suppose that /a
and u are real and that z i s r e a l and p o s i t i v e a n d we w r i t e z = e0 so that (6)
becomes
(9) J/~(eO)Ju(eS ) + Y/~(eS)Y(e 0)
m

= ~ ~ K~(2eOsinh t)[e(~-U)tcos us + e-(~-U)tces /~]dt .

w 0
 157

The functions Ju(e e) , Y ( e s) satisfy [13, p.99]

d2y/dO 2 + (e 20 - u 2 ) y = 0
s o t h a t t h e l e f t - h a n d s i d e o f (9) i s a s o l u t i o n o f [13, p . 1 4 6 ]
(10) Leu m (D2 - b2)(D2 - a2)u + 4e2e(De + 1)(D e + 2)u = 0

where e = /a + u , b = p - u . There is a standard method [I; 8, Ch. 8] for finding

an integral representation

(11) u(S) = F k(S,t) v(t) dt

Q

for a solution of (10). We t r y t o f i n d a l i n e a r differential operator

m
Mt = k~ 0 ink(t) Dk

and a function x(e,t) such that

(12) L8 k ( 8 , t ) = Mt x ( S , t ) .

We t h e n d e t e r m i n e v(t) as a solution of Mt v = 0 where Mt is the adjoint of

Mt , i . e .
R
Rt v = Z ( - 1 ) k Dk[mk(t)v] .
k=O
Then ( l l ) is a solution of L8 u = 0 provided a end ~ are chosen so that

z (_1)~ ( ~ v)(¢) . ( k - ~ - l ) ~ _- 0 .
k=l ~=0 a

(The differentiations in (13) are with respect to t .) Most of the standard

applications of the method are to second order equations and with s = k and its
success depends on being able to solve the equation ~t v = 0 . In the present

case, i f we c h o o s e
(14) k(8,t) = Ka(2e# s i n h t )

we h a v e t h e c o n v e n i e n t " f a c t o r i z a t i o n "
(15) LO k ( 8 , t ) = (D~ - D2)(D 2 - a 2) k ( O , t )

w h i c h i s o f t h e form (12) w i t h
x(@,t) = (D2 - 2 ) Ka(2e@ sinh t) = 4e 2e sinh2t Ka(2e 8 sinh t)

2 _ b2
and Mt = ~t = Dt Thus we get v(t) : c I e ("-u)t + c 2 e -(~t-u)t and it is

easily shown t h a t (13) h o l d s i f we c h o o s e ot = 0 , ~ = m To d e t e r m i n e c1 and

c2 we u s e [13, Ch.7]

Jr(x) au(x) + ~(x) YvCx) = ~2 co ( ~ + _~( 2_ 2 ) s i ~ ( _ ~ + O(x-3) x " , a,

WX

a n d , u s i n g [12, Ch.9] ,
 158

Ka(2X sinh t)e At dt = x x -I + aAx x -2 ÷ O.x

-3.() , x ~ ®
0
4 COS ('R'~/2) 8 sin(wa/2)

Then, by c o m p a r i n g c o e f f i c i e n t s , we g e t c 1 = cos u~ , c 2 = c o s #x so (9), and

hence ( 6 ) , is proved.
The key to the success of the method in the present case is the factorization
(15) arising from the choice (14) for the kernel k(#,t) . The choice of a function
of the form f(2e 8 sinh t) may be motivated by the fact that for a polynomial P
we have P(D#) f(2e # sinh t) = P(tanh t D t) f(2e 8 sinh t) . We see from (i0) that

L@ f(2e # sinh t) can be expected to take on a relatively simple form if we choose

f to satisfy
(D~- - a 2) f(2e 8 sinh t) = 4e 28 sinh2t f(2e 8 sinh t)

But this is the modified Bessel equation satisfied by f = Ka .

4. Another Nicholson-type formula. Here we give a differential equations

proof of
m

4 ~ r(u) (~)
= •'-w x u ;j Ku(xu)(u 2 + 4) u-1/2 u u du , x > 0 , u > -
1
0
the special case n = 0 of [3, p.368, (42)]. It is convenient to write this
formula in the form
m

.2
(17) j~(x) 2
+ y~(x) = ka(xt) (t~/2 + 4) -(a+4)/(28+4) ta/2-I dt , x > 0 , a > 0
0

where we h a v e a d o p t e d an a d h o c n o t a t i o n for the generalized Airy functions:

J a ' Ya are appropriately normalized solutions of

(18) y" + xay = 0

where a = -2 -' 1/u , while ka is a suitable solution vanishing at +~ of

(19) y,, - x % = 0 .
In the special c~e a = 1 (u = - 1 / 3 ) , (17) becomes
U
Ai2(-x) + Bi2(-x) - 24(2/3)1/6 ; t - 1 / 2 ( t 3 + 4) - 5 / 6 A i ( x t ) dt , x • 0 .
J~ r(1/6) 0
I n o r d e r t o p r o v e (17) we n o t e t h a t , u s i n g (18) and [13, p . 1 4 5 ] its left-hand side
satisfies
L x u • (D 3 + 4x a Dx + ~ x a - 1 ) u = 0

end, truing ( 1 9 ) , we find that

Lx ka(xt) : Mt x¢-1 ka(xt)
where
 159

Mt = ( t a+3 + 4t)D t + a ( t a+2 + 2)

and
Mt : - ( t a + 3 + 4 t ) D t + ( 2 a - 4 - 3 t a+2) •

NOW Mt v = 0 has the general solution

v(t) = ct a/2-1 ( t a+2 + 4) - ( a + 4 ) / ( 2 a + 4 )

We n o t e t h a t the condition (13) i s a l s o s a t i s f i e d with a = 0 , ~ = ® leading to
(17) apart from a constant factor. To evaluate the constant we return to the form
(16) and use
J u2 (x) + ~ (x) -- ~+2 O(x-2) , x~®

and, using [12, Ch.7],

[ Z_(xu)(u 2 + 4) L~'-1/2 u ~ au = 1 2u-i ~r(Zu)/r(~) ÷ 0(x -2) , x ,

J 0 ~ x

5. Zeros of generalized Airy functions. M.S.P. Eastham (private

communication) raised the question of showing that the smallest positive zero x(X

of a solution o f (18), satisfying y(O) = 0 , decreases as a increases,

0 < a < ~ T h i s , and more, h a s b e e n p r o v e d by A. L a f o r g i a and t h e a u t h o r ( t o b e
published) using results (due t o E l b e r t and L a f o r g i a ) b a s e d on (3) and t h e
w e l l - k n o w n c o n n e c t i o n b e t w e e n (18) and t h e B e s s e l e q u a t i o n . I t would be n i c e t o
show t h i s u s i n g (18) d i r e c t l y . The Sturm c o m p a r i s o n t h e o r e m i s n o t a p p l i c a b l e in
a n y o b v i o u s way b e c a u s e x~ is not monotonically increasing in a for each x in
an i n t e r v a l (0,b) , b > 1 . This raises t h e q u e s t i o n o f w h e t h e r one can f i n d an
a n a l o g u e o f (3) ( o t h e r t h a n t h e awkward f o r m u l a g o t by t r a n s f o r m i n g (3) i t s e l f ) for
dxa/d~ . What we n e e d i n e f f e c t is a result t h a t b e a r s t h e same r e l a t i o n t o (17) a s

(3) d o e s t o ( 1 ) . One way t o a p p r o a c h t h i s p r o b l e m w oul d b e t o f i n d an i n t e g r a l

representation for J a Y~ - J~ Ya which satisfies a known f o u r t h o r d e r d i f f e r e n t i a l

equation.
A p e r h a p s more t r a c t a b l e problem would be t o f i n d t h e a p p r o p r i a t e
generalization o f (4) f o r
_x 2
e [.^(x) ~(x) -G^(x) Uu(x)l .
This would give, in particular, a formula for the derivative with respect to A of
a zero of a Hermite function.
 160

References

I. H. Bateman, The solution of linear differential equations by means of definite

integrals, Trans. Cambridge Philos. Soc. 21 (1909), pp. 171-196.
2. A.L. Dixon and W.L. Ferrar, Infinite integrals in the theory of Beasel
functions, Quart. J. Math. Oxford I (1930), pp. 122-145.
3. I,. Durand, Nicholson-type integrals for products of Gegenbauer functions and
related topics, i n T h e o r y and A p p l i c a t i o n o f S p e c i a l F u n c t i o n s (H. Askey, ed. ),
Academic P r e s s , New York, 1975, pp. 3 5 3 - 3 7 4 .
4. L. Durand, Product Pormu]as and Nicho]son-type integrals for Jscobi functions.
I: S~m~.ry of results, SIAM J. Math. Anal. 9 (1978), pp. 76-86.
5. A. E l b e r t , Concavity of the zeros of Besse] functions, Studia Sci. H a t h .
Hungar. 12 ( 1 9 7 7 ) , pp. 8 1 - 8 8 .
8. A. E l b e r t a n d A. L a f o r g i a , On t h e s q u a r e o f t h e z e r o s o f B e s s e l f u n c t i o n s , sIAM
J . H a t h . A n a l . 15 ( 1 9 8 4 ) , pp. 206-212.
7. G.H. Hardy, Some f o r m u l a e i n t h e t h e o r y o f 8 e a s e l f u n c t i o n s , P r o c . London Math.
Soc. 23 (1925), pp. Ixi-lxiii.
8. E.L. I n c e , Ordinary D i f f e r e n t i a l Equations, Longmans, London, 1927; reprinted
Dover, New York, 1956.
9. L. Lorch and P. Szego, Higher monotonicity properties of certain
Storm-Liouville functions, Acta Math. 109 (1963), pp. 55-73.
i0. M.R. H~ldoon, A d i f f e t ~ e n t i a l e q u a t i o n s p r o o f o f a N i c h o l s o n - t y p e f o r m u l a , Z.
Angew. H a t h . Mech. 61 (198~_), pp. 598-$99.
11. M.E. Muldoon, ]'he v a r i a t i o n w i t h r e s p e c t t o o r d e r o f z e r o s o f Bewsel f u n c t i o n s ,
Rend. Sere. H a t . Univ. P o l i t e c . T o r i n o 39 ( 1 9 8 1 ) , pp. 15-25.
12. F.W.J. O l v e r , A s y m p t o t i c s and S p e c i a l F u n c t i o n s , Academic P r e s s , New York a n d
London, 1974.
13. G.N. Watson, A t r e a t i s e on t h e Theory o f B e a s e l F u n c t i o n s , 2nd e d . , c a m b r i d g e
University Press, 1944.
14. J . E . Wilkins, J r . , NichoJson's i n t e g r a l f o r JZn(Z) + ~n(z) , Bull. Amer. Math.

Soc. 54 (1948), pp. 232-234.

SURJECTIVITY A N D BOUNDARY VALUE
PROBLEMS
V. SEDA
Faculty of Mathematics and Physics, Cornenius University
Mlynskd dolina, 842 15 Bratislava, Czechoslovakia

In t h e p a p e r we shall deal with an i n i t i a l and a boundary problem

for t h e f u n c t i o n a l differential equation with deviating argument x'(t) =
= f[t,x (t)l in a B a n a c h space whereby the functions of the state space
are d e f i n e d in the interval (-~,0] as w e l l as w i t h the g e n e r a l i z e d
boundary value problem for a s y s t e m of d i f f e r e n t i a l equations in R n . T h e
main tool for proving the e x i s t e n c e of a s o l u t i o n to t h e s e problems
will be s o m e theorems on surjectivity of an o p e r a t o r .

i. S u r j e c t i v i t y of an o p e r a t o r .
Let (E, I.[) b e a r e a l Banach space, ~ * X C E and S : X - E. W e
recall t h a t S is compact if S is c o n t i n u o u s and maps bounded sets into
relatively compact sets. Similarly T : X ~ E is s a i d to be a condensing
map if T is c o n t i n u o u s , bounded (i.e. maps bounded sets into bounded
sets) and for every bounded set A C X which is n o t relatively compact
we h a v e e(T(A)) < ~(A) where ~ is the K u r a t o w s k i measure of n o n c o m p a c t -
ness. A simple example of a c o n d e n s i n g map is o n e of the form U + V
where U : X - E is a s t r i c t contraction and V : X - E is a c o m p a c t map.
L e t G * ~ be an o p e n subset of E and d e n o t e b y G the closure of G.
Let T : ~ ~ E be a c o n d e n s i n g map, a @ E. If the set A = {x 6 G : x -
- T(x) = a} is c o m p a c t (possibly empty), then the d e g r e e deg(I - T,G,a)
is d e f i n e d in the sense of N u s s b a u m [6] whereby I is the identity.
Notice that A will certainly be c o m p a c t if G is b o u n d e d a n d T is s u c h
that x - T(x) * a for all x E %G (boundary of G) ([6], p. 744). If T is
compact, then the d e g r e e above agrees with the classical Leray-Schauder
degree.
Denote B the real Banach space of all continuous functions
x : [0, ~ ) ~ E s u c h that there exists l i m x(t) = x(~) ( E E) for t ~ ~.
The norm in B is d e f i n e d by 111xi]2= s u p { I x ( t ) 1 : 0 ~ t < ~} for e a c h
x E B. Let, further, U(r) = {x C E : [xM < r}. Using the d e g r e e theory
for c o n d e n s i n g perturbations of identity, the topological principle in
[8], p. 241, c a n be g e n e r a l i z e d as follows (for p r o o f , see [9],[ 10]).

Theorem I. Let g : E ~ B be a c o n t i n u o u s mapping. Denote by g ( x , t )

 162

the value o ~ g(x) 6 B at the point t 6 [0,~ (g(x, ~) : lim g ( x , t ) for

t ~ ~).~Assume that
(i) v(x) : inf{Ig(x,t)1 : 0 ~ t ~ ~] - ~ for Ix[ -- ~ ;
(ii) the m a p p i n g I - g(.,t) is c o n d e n s i n g for e a c h t 6 [0,7
(iii) for e a c h y 6 E there is an r0> 0 such that
deg(g(.,0) - y, U(r0),0) ~ 0 ;
(iv) g(x,.) is c o n t i n u o u s in t, u n i f o r m l y in x 6 U(r) for e a c h
r > 0. T h e n for each t 6 [0,~]
g(E,t) = E.

Proof. Let y E E, tO 6 [0,7 . By (i), there is an r 0 > 0, ryl <

< r 0 ~ such that y ~ g(SU(r0),t ) for each t e [0,~] . H e n c e the m a p p i n g
G : U(r0) X {0,~] ~ E defined by G(x,t) : x - g(x,t) + y is c o n t i n u -
ous and G ( x , t ) ~ x f o r x e ~U(r0) , t E [0,~] , By (ii), G(.,t) is a
condensing map for b E [0,~] and (iv) i m p l i e s that G(x,.) is c o n t i n u -
ous in t, u n i f o r m l y in x 6 U ( r 0 ~ . Hence, by Corollary 2 in [ 6] , p . 7 4 5 ,
and (iii), for e a c h to, 0 S t O < ~,
deg(I - G(.,~), U(r0),0 ) : deg(I - G(.,0), U(r0),0 ) :
: deg(g(.,0) - Y, U ( r 0 ) , 0 ) ~ 0.
As to the set S : {x 6 U ( r 0 ) : g ( x , t 0) - y : 0] , e i t h e r it is n o t
compact o r in case it is c o m p a c t we can use P r o p o s i t i o n 5 f r o m [ 6],
p. 744, and h e n c e , in b o t h cases it is n o n e m p t y .
Corollary 2 as w e l l as P r o p o s i t i o n 5 from [6] can be applied to
the case tO ~, too, since then t = tg ~ s m a p s [0,i] continuously on
[0, ~] and ~nstead of the f u n c t i o n G(x,t) we consider Gl(X,S ) =
= G(x,tg ~ s), x 6 U ( r 0 ) , s e [0,i] .

Remark. Clearly the a s s u m p t i o n (iii) is s a t i s f i e d if g ( x , 0 ) : x

for e a c h x E E.
O n the b a s i s of the Schauder theorem on d o m a i n invariance ([2],
p. 72) the following result c a n be proved. ([ i0]~.

Theorem 2. L e t T : E -- E be s u c h that
(a) l i m IT(x)I : ~ ;
Ixl-~
(b) I - T is c o m p a c t ;
(c) T is l o c a l l y one-to-one, i.e. for e a c h point x0E E there is a
neighbourhood N of this point such that TIN is o n e - t o - o n e . Then T(E)=E.
 163

Proof. The assumptions (b), (c) i m p l y t h a t T is an o p e n m a p p i n g ,

ioe. it m a p s open sets onto open sets. Hence T(E) is an o p e n subset of
E. Let {yn} C T(E) be a convergent sequence and Y O = l i m Yn" T h e n w e
n~
can find a sequence {xn} such that T ( x n) = Yn" A s s u m p t i o n (a) is
equivalent to the statement that the inverse image of a bounded set at
the m a p p i n g T is a b o u n d e d set. Hence the sequence {xn} is b o u n d e d
together with the sequence {yn]. By (D), there is a s u b s e q u e n c e {xm]
of {xn} and a point x 0 6 E such that xm - Ym = Xm - T(Xm) ~ Xo as
m ~ ~. Then
lim X m = YO + Xo' and b y c o n t i n u i t y of T, T ( x 0 + yo ) = YO"
m~
Thus YO 6 T(E) and T(E) is c l o s e d . As E is c o n n e c t e d , T(E) = E.

Corollary i. L e t T : E - E be such that

(a) lim IT(x)[ = ~ ;
Lxi~
(b) I - T is c o m p a c t ;
(c) T is o n e - t o - o n e .
T h e n T is a h o m e o m o r p h i s m of E o n t o E and there is a c o m p a c t mapping
TI: E -- E such that T-I: I - T 1 where T -I is the inverse mapping to T.

Proof. By Tl~eorem 2 and its p r o o f w e h a v e that T(E) = E and the

mapping T "I is c o n t i n u o u s . Hence T is a h o m e o m o r p h i s m . For T -I we
have the identity I - T -I = (T - I) o T -I, By (a), T -I is a b o u n d e d
mapping and thus, by (b), I - T -I = T 1 is c o m p a c t .

If E = R n, then Theorem 1 is t r u e w i t h o u t assuming assumptions

(ii),(iv) and in T h e o r e m 2 instead of the a s s u m p t i o n (b) it s u f f i c e s to
assume the c o n t i n u i t y of T. C h o o s i n g properly the m a p p i n g g : Rn- B (B
now means the B a n a c h space of all continuous functions x : [0,i] ~ Rn
with the supnorm, 1.1 is the euclidean norm in R n a n d (.,.) the s c a l a r
product in this space) we get the following

Corollary 2. L e t T : R n~ R n be a c o n t i n u o u s mapping such that

(i) lim iT(x)i = ~ ;

(ii) either there is an x 0 6 R n s u c h that

T(x)-Xo= k ( x - x O) implies k Z 0 for each x E R n, x # x0 ,
or
there is an rl> 0 such that (x,T(x)) > 0 for all
--
x 6 Rn t

IX] ~ r I
 164

or
T is locally one-to-one.
Then
T ( R n) = R n.

Proof. a. Consider the first case that there is a n x 0 E R n s u c h

that

(~) T(x) - x 0 = k(x - x 8) implies k Z 0 for each x @ R n, x # x 0.

without loss of generality we may assume that x0 = 0. Let the mapping

g : R n -- B b e d e f i n e d b y

g(x,t) = tT(0) for x = 0, 0 ~ t ~ i,

g(x,t) = [ (i - t) Ixl + t i T ( x ) i ] .[ i(l - t ) x + t T ( x ) i] -I
[ (I - t)x + tT(x)] for x % 0, 0 ~ t < i,
g(x,t) = T(x) for x ~ 0, t = i.

By (a) the mapping g is w e l l defined• Further g(x,.) is c o n t i n u o u s in

[0,i] for each x 6 R n and thus, g maps R n into B. Clearly

(~) g(x,0) = x, g(x,l) = T(x) for each x 6 R n.

Now we prove that g is continuous. Let x % 0 be an arbitrary but fixed

point from Rn and y be a point sufficiently close t o x. Then

(i - t ) x + t T ( x ) (i - t ) ~ + t T ( ~ )
Ig(x,t) - g(y,t)l ~ T<T~'-t)x~-£T(x)i - i(i ~ t ) y + t T ( y ) l~ "

• [ (i - t) Ixl + t I T ( x ) I] +

+ I(1 - t ) ( I x l - lYl) + t(IT(x)l - IT(y) l)l, 0 S t ~ i.

Clearly the second term on the right-hand side is less or equal to

(y) (I - t) Ix - Yi + tiT(x) - T(y)1, 0 ~ t S 1.

As to the first term, there is a constant k > 0 such that this term is
less or equal to

ki(l - t)y + t T ( y ) i-I• J[ (i - t ) x + tT(x)] • I(i - t ) y + tT(y) l -

- [ (i - t)y + tT(y)] .i(l - t)x + tT(x) l l

kl(l - t)y + t T ( y ) } -I l[ (I - t ) ( x - y) + t(T(x) - T(y))]

I(1 - t ) y + tT(y)l + [ (i - t ) y + t T ( y ) ] . [ i(l - t ) y + tT(y)I-

- i (i - t ) x + t T ( x ) I] i .

Hence the first term is less or equal to

(6) 2k[(l - t) Ix - Yl + tiT(x) - T ( y ) l], 0 ~ t ~ i.

The inequalities (7) and (6) give

 165

Ig(x,t) - g(y,t) l S (2k + i)[ (i - t) Ix - Yl + tiT(x) - T(y)I],

0 S t S i,

which proves the continuity of g at x % 0. In a s i m i l a r way it can b e

shown t h a t g is c o n t i n u o u s at 0.
Now we derive properties (i), (iii) of g f r o m T h e o r e m 1 and this
will complete the p r o o f of t h i s p a r t of C o r o l l a r y 2. As Ig(x,t)1 =

: (i - t)]xl + tIT(x)l Z m i n ( I x l , I T ( x ) I), c l e a r l y (i) is s a t i s f i e d .

(iii) follows from (8).
b. Suppose that there is an r > 0 such that
1
(~) (x,T(x)) a 0 for all x E R n, ixl ~ r .
1
Consider the m a p p i n g g which is d e f i n e d for each x E R n, 0 S t ~ i, b y

g(x,t) = (i - t ) x + tT(x).

Clearly g : R n -- B and g is c o n t i n u o u s . F u r t h e r g s a t i s f i e s (~). By (x),

ig(x,t) l2 (I - t)21xt 2 + t 2 IT(x) 12 ~ ~i [ (i - t)IxI + t I T ( x ) i] 2
Hence g satisfies assumption (i) as w e l l as (iii) of T h e o r e m i. By
this theorem the result follows.
c. The statement of C o r o l l a r y 2 in c a s e that T is l o c a l l y one-to-
one follows directly from Theorem 2.

2. F u n c t i o n a l Differential Equations With Deviatin ~ Argument

F i r s t we formulate the initial-value problem for t h e s e equations
which includes the p r o b l e m from [ 12] ,[ 4] and is r e l a t e d to one in [ i],
[ 3]. F o r d e t a i l s and proofs, see [9]. We shall employ the notations:
(E,i. I) is a r e a l Banach space.
The state space C is the B a n a c h space of all c o n t i n u o u s and
bounded mappings x : (-~, 0] ~ E w i t h the sup-norm 11. II.
: [0, ~) ~ (0, ~) is a n o n d e c r e a s i n g continuous function.
The deviation ~ : [ 0, ~) ~ R is a c o n t i n u o u s mapping such that
~o(0) = 0.
f : [0, ~) X C ~ E is a c o n t i n u o u s m a p p i n g .
+
a = m a x (a,0) for e a c h a 6 R, s g n 0 -- 0, s g n a = 1 for e a c h a > 0.
Finally, if x : (-~,~) ~ E is a c o n t i n u o u s mapping which is b o u n -
ded in (-~,0] a n d u 6 R, then x u is the function defined by
x u (s) -- x ( u + s) for all s, - ~ < s < 0.
Clearly X u E C.
Th~ initiaZ-uaZ~£ prob2.cm in the c a s e t h a t h E C is u n i f o r m l y
continuous in (-~,0]
(i) x'(t) = f[t,x (t)]
(2) x0= h
 166

means the p r o b l e m to find a function x which is c o n t i n u o u s in (-~,~),

x(t) : h(t) for all t e (~,0] , x is d i f f e r e n t i a b l e in [0, ~) and it
satisfies (i) at e a c h point f r o m [0,~). Since ~, f are continuous and
h is u n i f o r m l y continuous, the problem (i)~(2) is e q u i v a l e n t to the p r o -
blem: To find a continuous solution of the integral equation
t
(3) x(t) = h(0) + f f[s,X~(s)]ds (0 ~ t < ~)
0
which satisfies (2).
Consider the following assumptions:
(AI) The function ftlf(s,0)Ids is ~ - b o u n d e d in [0,~), i.e. I ftlf(s,0)t
0 0
dsl/ %(t) (0 ~ t < ~) is b o u n d e d .
(A2) There exists a nonnegative, locally integrable in [0, ~) real func-
tion n such that
If(t,z I) - f(t,z2)l ~ n(t) 11zI- z211
for e v e r y Zl,Z 2 6 C and t q [0,~).
t
(A3) T h e f u n c t i o n f n ( s ) d s is % - b o u n d e d in [0,~).
0
(A4) T h e r e e x i s t s a qt 0 S q < i, such t h a t
f t n(s) sgn ~+(s)~[~+(s)]ds ~ q~(t) (0 S t < ~).
0
t
(A5) T h e r e is a K > 0 s u c h that f If(s,0)Ids ~ K for all t, 0 ~ t < ~.
0
t
(A6) There is a q, 0 ~ q < i, s u c h that f n(s)ds ~ q, 0 ~ t <
0
The existence of a unique ~-bounded solution to (1)~(2) is
guaranteed by

Lemma I. If the assumptions (AI)-(A4) are satisfied, then there e-

xists a unique ~-bounded in [0, ~) solution x(t) of (i)~(2), i.e. Ix(t)1
/~(t) is b o u n d e d in [0,~).

Proof. Let D be the v e c t o r space of all continuous mappings

x : (-~,~) ~ E wh±ch are b o u n d e d in (-~,0] and ~ - b o u n d e d in [0,~),
Dh = {x q D : x(t) = h(t), - ~ < t -< 0 } . Let F be the Banach space of
all continuous and % - b o u n d e d mappings x : [ 0, ~) - E with the n o r m
ilKil1 -- sup Ix(t) i/~(t). Then in v i e w of the assumptions of the lemma
the m a p ~ t ~ T defined by
T(x)(t) = h(t), - ~ < t _< 0,
T(x)(t) = h(0) + ~ f[s,x (s)]ds, 0 -< t < OO I

maps D h i n t o D h or c o n s i d e r i n g only the restriction of f u n c t i o n s from

D h to [0,~), T : G -- G w h e r e G = Ix 6 F : x(0) = h(0)] is a closed
 167

s u b s e t of F. By (A2) and (A4) JT(x)(t) - T ( y ) ( t ) { / ~ ( t ) <

t t
S ~ n(s)llx (s) - y~(s)llds/~(t) < llx - yll I f n(s) sgn ~+(s).
0
~ [ ~ + ( s ) ] d s / ¢ ( t ) ~ q ~ x - y{Ii. The B a n a c h f i x e d p o i n t t h e o r e m g i v e s the
result.

By c o n s i d e r i n g the bounded solutions of ~

problem (I)~ (2) we can p r o v e

Lemma 2. If the a s s u m p t i o n s (AI)-(A4) are s a t i s f i e d and ~ is b o u n -

ded, then for the u n i q u e b o u n d e d solution x(t) of (i)~(2) there exists
lim x(t) : c ( 6 E).
t~

Proof. By (3) and (A2),

t t for 0 ~ t I < t 2 < ~ w e h a v e ix(t2) -
X(tl) I ~2[f(s,0)ids + ~2 n(s)lix (s)llds. In v i e w of (AI), (A3) and

the b o u n d e d n e s s of %, b y the C a u c h y - B o l z a n o criterion the r e s u l t

follows.

Denote this u n i q u e b o u n d e d solution of (I)~(2) as x(t,h). Then the

continuity of the b o u n d e d solution of (i)~(2) in h is p r o v e d in

Lemma 3. S u p p o s e that (A2)~(A5) and (A6) are s a t i s f i e d . Then for

a n y h l , h 2 • C, h I, h 2 are u n i f o r m l y continuous in (-~,0] and hl(0) =
= h2(0) -- 0
llxt(.,h 2) - xt(.,hl)ll -< llh2- hlllv(t) ~ 0 -< t < ~ ,
w h e r e v(t) is the u n i q u e real b o u n d e d continuous solution of
<4) v(t) -- i ~ I t n(s) v [ ~ + ( s ) ] d s , 0 _< t <
0

Proof. D e n o t e u(t) = llxt(.,h2) - xt(.,hl)ll, 0 _< t < ~. By (3) and

(A2) it f o l l o w s t h a t

t
Ix(t,h2) -x(t,hl) I _< lh2(0) - hl(0)i + / n(s)u[~+(s)]ds,
0 0 _< t < ~,
t
and h e n c e u(t) _< "llh2 - nln + -i n ( s ) u [ ~ + ( s ) ] d s , 0 _< t < ~. S i n c e u is
bounded and c o n t i n u o u s , b y the g e n e r a l i z e d G r o n w a l l l e m m a the r e s u l t
follows.

Lemma 4. A s s u m e that (A2)~(A5) and (A6) are s a t i s f i e d . L e t h E C,

h(0) -- 0 and let h be u n i f o r m l y continuous in (-~,0]. Let {z k} , ZkE E,
k = 1,2,..., be a s e q u e n c e w i t h lim IZkl = ~. D e n o t e mk= inf{Ix(t,h +
k~
+ Zk)l : 0 < t < ~}. T h e n
 168

lim m k = co.
k~

Proof, It is si'milar to t h a t of L e m m a 3 in ~ 8], p. 240.

By u s i n g Theorem 1 where g(x0,t) = x ( t , h + x 0) the f o l l o w i n g

boundary value pro6lem for (I) c o n be solved. A n a r b i t r a r y p o i n t Xl6E

and an i n i t i a l function h 6 C, h is u n i f o r m l y continuous in (-°°,0] ,
h(0) = 0, are given. To f i n d a p o i n t x 0 E E s u c h t h a t
(5) lim x ( t , h + x 0) = x I .
t~
Theorem 3. A s s u m e that (A2 ~) ~(A5),(A6) as w e l l as the a s s u m p t i o n :
(A7) T h e r e e x i s t s a ql' 0 < ql < I, such that for the b o u n d e d continu-
ous s o l u t i o n v(t) of the e q u a t i o n (4) the i n e q u a l i t y
v(t) _< 1 + ql' 0 < t < oo ,
is true,
are s a t i s f i e d . Let x 1 6 E and let h 6 C, h be u n i f o r m l y continuous in
(-~,0] , h(0) = 0. T h e n t h e r e e x i s t s exactly one x 0 6 E s u c h t h a t (5) is
true.

Proof. Define a mapping g : E ~ B in this way. G i v e n an x 0 6 E,

let g ( x 0 , t ) = x ( t , h + x 0) for 0 <_ t < ~ and let g ( x 0 , ~ ) --
-- l i m x ( t , h + x0). L e m m a 1 and 2 g u a r a n t e e t h a t g is w e l l d e f i n e d .
t~
By L e m m a 3 g is c o n t i n u o u s and L e m m a 4 i m p l i e s t h a t the c o n d i t i o n (i)
in T h e o r e m 1 is s a t i s f i e d . Clearly (iii) in that t h e o r e m h o l d s . Let
r > 0, t I < t, Ix2i < r. Then Jg(x2,t) - g(x2,tl)l < ~ t f ( s , 0 ) I d s +
t t tl
+ tlf n ( s ) U x +(s )(.,h + x 2 ) U d s _< tlf f f ( s , 0 ) I d s + tl~ n(s) [M 1 +

+ (UhR + r ) K l ] d S , w h e r e M 1 -- sup ~x +(s)(.,O)~ , K 1 = sup v ( ~ + ( s ) ) .

0<s<~ 0<s< ~
If t < tl, w e g e t a s i m i l a r i n e q u a l i t y . This i m p l i e s t h a t (iv) is
satisfied.
Consider the m a p p i n g U = I - g(. ,t) for a f i x e d t 6 [0,~]. Then
b y L e m m a 3 and (A7) IU(x0) - U(Y0)t < ~
0 n(s)Ux +(s ) - y + ( s ) N d s _<

_< Ix 0 - y01(v(t) - i) < q l I x 0 - y0 I. H e n c e U is a s t r i c t c o n t r a c t i o n

and thus a condensing mapping. By T h e o r e m i, g ( E , t ) = E for e a c h
t 6 [ 0,~]. S i n c e U is a s t r i c t contraction, ig(x0,t) - g(y0,t)l >
>_ (i - q l ) I X 0 - y0 i w h i c h implies that g(.,t) is a h o m e o m o r p h i c mapping
of E o n t o itself.

Remarks. i. In case E = R n, T h e o r e m 3 is v a l i d w i t h o u t assuming

(A7). Of c o u r s e u n i q u e n e s s of x 0 n e e d not be true.
2. T h e o r e m 3 extends the m a i n r e s u l t f r o m [8], p. 239,
 169

and in the c a s e ~(t) -- t, 0 < t < ~, is s t r o n g e r than Theorem I in

[ii], p.3.

3. G e n e r a l i z e d Boundary Value Problem for D i f f e r e n t i a l Systems

The generalized boundary value problem for a d i f f e r e n t i a l system
(~) x' -- f ( t , x ) , t E i, x 6 Rn ,
and a given continuous mapping T (not n e c e s s a r i l y linear) of t h e space
C ( i , R n) of a l l c o n t i n u o u s n-dimensional vector functions defined in i
into R n can be defined as a p r o b l e m of f i n d i n g a solution x(t) of t h e
system (@) on the interval i for w h i c h T(x) is a g i v e n vector r in Rn,
i.e.
(7) T(x) = r .
The topology in C ( i , R n) is g i v e n in two d i f f e r e n t cases. If i = [a,b]
is a c o m p a c t interval, then we consider the topology of u n i f o r m conver-
gence, while in c a s e i is a n o n c o m p a c t interval, e.g. i -- (a, °°) , t h e n
we use the topology of l o c a l l y uniform convergence.

Theorem 4. L e t f -- f(t,x) be a c o n t i n u o u s function on i X R n and

let t h e e q u a t i o n (S) h a v e the following properties:
(a) T h e r e is a p o i n t t06 i such that for e a c h vector x06 R n there
exists a unique solution x(t) on i to the initial-value problem (6)~
(8) x ( t 0) -- x 0
and either:
(b) F o r e a c h solution x of (6)~(8) the following implication is
true:
If T(x~ = k x ( t 0 ) ~ x(t 0) ~ 0, t h e n k _> 0 ,
OZ'I

(c) The problem (6),(7) has at m o s t one solution for e a c h r 6 R n.

Then in the case (a),(b) a sufficient condition and in the case (a),(c)
a necessary and sufficient condition that there exist at l e a s t one
solution of the p r o b l e m (6)~(7) for e a c h r 6 R n is t h a t the following
compactness condition be s a t i s f i e d :
(d) If {Xk} is a s e q u e n c e of s o l u t i o n s of (6) on the interval i
such t h a t {T(Xk) } is b o u n d e d , t h e n t h e r e is a s u b s e q u e n c e {Xk(1)} such
that {Xk(1) } is c o n v e r g i n g in C ( i , R n ) .
The proof is b a s e d on Corollary 2 and the K a m k e convergence lemma.

References

[ I] ANGELOV,V.G., BAJNOV,D.D., On the Existence and Uniqueness of a

Bounded S o l u t i o n to F u n c t i o n a l D i f f e r e n t i a l Equations of N e u t r a l
Type i n a Banaeh Space (In R u s s i a n ) . Arch. Math. 2, S c r i p t a Fac.
Sci. Nat. UJEP Brunensis XVII: 65-72 (1981).
 170

[2] DEIMLING,K., N i e h t l i n e a r e Gleichungen und Abbildungsgrade, Sprin-

ger-Verlag, Berlin 1974.
[3] HALE,J.K., Theory of Functional D i f f e r e n t i a l Equations, AppI.
Math. Sci., Vol. 3, Springer-Verlag, New York 1977.
[4] HALE,J.K., Retarded EquatZons With I n f i n i t e Delays, Lecture Notes
in Mathematics, Vol. 730, Springer-Verlag, Berlin, 157-193 (1979).
[5] HARTMAN,Ph., Ordinary D i f f e r e n t i a l Equations, John Wiley, New York
1964.
[6] NUSSBAUM, R.D., Degree Theory for Local Condensing Maps, J. Math.
Anal. Appl. 37, 741-766 (1972).
[7] OPIAL,Z., Linear Problems for Systems of Nonlinear D i f f e r e n t i a l
Equations, J. Differential Equations 3, 580-594 (1967).
[8] SMfTALOV~,K., On a Problem Concerning a Functional D i f f e r e n t i a l
Equation, Math. Slovaca 30, 239-242 (1980).
[9] ~EDA,V., Functional D i f f e r e n t i a l Equations With Deviating Argument
(Preprint).
[10] ~EDA,V., On S u r j e c t i v i t y of an Operator (Preprint).
[ i i ] ~VEC,M., Some P r o p e r t i e s of Functional D i f f e r e n t i a l Equations,
Bolletino U.M.I. (4) ii, Suppl. Fasc. 3, 467-477 (1975).
[12] WEBB,G.F., A c c r e t i v e Operators and E x i s t e n c e for Nonlinear Func-
t i o n a l D i f f e r e n t i a l Equations, J . D i f f e r e n t i a l E q u a t i o n s 14,
57-69 (1973).
 SOME PROBLEMS CONCERNING
THE EQUIVALENCES OF TWO SYSTEMS
OF DIFFERENTIAL EQUATIONS
M. SVEC
Faculty of Mathematics and Physics, Comenius University
Mlynskd dolina, 842 15 Bratislava, Czechoslovakia

Consider two systems

(i) x(t) = f ( t , x t)

(2) 9(t) 6 f ( t , y t) + g ( t , y t)

where t 6 J = (0,~), x : R -- R n, y : R ~ R n, x t = x ( t + s), yt = y ( t + s),

s E (-~,0) . D e n o t e by C = C ( - ~ , 0 ; R n) the space of all functions
: (-~,0> ~ Rn which are b o u n d e d and continuous with sup normi1.11.
Then f : J x C ~ R n, g : J x C -- {the set of all nonempty subsets of Rn}.
Further properties of f and g will be g i v e n later. However, we w i l l still
assume that f a n d g are such that the e x i s t e n c e of the solutions of (i)
and (2) is g u a r a n t e e d on J. i.i is the v e c t o r norm in R n. If A C R n,
then IAl ~ sup{lal : a 6 A}.
Our a i m is to e s t a b i l i s h the c o n d i t i o n s which give the p o s s i b i l i t y
of p a i r i n g of the solutions x(t) of (i) a n d y(t) of (2) in s u c h a way
that we will be able to say something about the a s y m p t o t i c behaviour
of the d i f f e r e n c e y(t) - x(t) = z(t). Assume that x(t) is given. Then,
proceeding formally, substituting y(t) by z(t) + x(t) in (2), w e get

(3) ~(t) e - f ( t , x t) + f(t,zt+ xt) + g(t,zt+ xt )

We h a v e to p r o v e the e x i s t e n c e of such solution z(t) to the functio-

nal differential inclusion (3) that lim z(t) = 0 as t - ~ ( t h e case
of a s y m p t o t i c equivalence) or t h a t z(t) £ L (J), p ~ 1 (the c a s e of
P
p-integral equivalence). There are m a n y m e t h o d s h o w to do it, e.g. use
the v i a b i l i t y theory, m e t h o d of f i x e d p o i n t , method of L i a p u n o v function.
F i r s t we w i l l use the v i a b i l i t y theory.

Theorem i. a) L e t be f : JxC - R n continuous and let it s a t i s f y the

Lipschitz condition
(4) If(t,~ I) - f(t,~2)] ~ L(t)~ I- ~2~, L(t) E LI(J)

for e a c h (t,~l) ~ (t,~ 2) E JxC.

b) L e t g be an u p p e r semicontinuous map f r o m J x C to t h e n o n e m p t y
compact convex subsets of R n a n d let
Ig(t,~)1S G0(t,H~II) a.e. on J
 172

where G0(t,u) : JxJ -- J is m o n o t o n e nondecreasing in u for e a c h fixed

t E j and is i n t e g r a b l e on J for e a c h f i x e d u C j.
n
c) L e t x : R ~ R be a b o u n d e d s o l u t i o n of (i).
d) L e t there exist a solution u : J ~ J to the differential
equation
(5) ~(t) : -L(t)u - G0(t,u + ilxtDI) ~ - G ( t , u ) , u(0) > 0
e) L e t be
K(t) ~ {x @ Rn: fxl ~ u(t)}, t @ J
f) Let be
~t @ J, K(t) ~ {~ • C : ~(0) • K(t)}
g) Let be the image of the graph (K) by the m a p
F(t,~) ~ -f(t,xt) + f(t,~ + x t) + g(t,~ + x t)
relatively compact.
h) L e t for
vt • J , v ~ such that ~(0) • K(t)~x • K(t)
F<t,~) N DK(t,~(0))(i ) ~
where DK(t,~(0)) is the c o n t i n g e n t derivative of K at (t,~(0)). Then
for e a c h ~ • K(0) there exists a solution z(t) to the functional
inclusion such that
(6) for a l m o s t all t 6 j, ~(t) @ F ( t , z t)
(z) 0 =

which is v i a b l e in the sense that

(7) ~t • J, z(t) e K(t) (iz(t)i S u(t))

Remark i. E v i d e n t l y , if l i m u(t) = 0 as t ~ ~, then also l i m z(t)=

= 0 as t - ~ a n d if u(t) • L (J), p ~ i, t h e n also the restriction
p
z ( t ) I j • Lp(J) h o l d s .

Remark 2. It f o l l o w s from the p r o p e r t i e s of f a n d g that F is an

upper semicontinuous map from JxC to the nonempty compact convex sub-
sets of R n a n d
(8) IF(t,~)J ~ L(t) ll~lJ + G0(t,Ji~ll + ilxtll) ~ G(t, li~)
Evidently, G : J x J -- J is n o n d e c r e a s i n g in u for e a c h fixed t • J and
integrable in t for e a c h fixed u • J.

Remark 3. In o u r c a s e the basic space is R n. T h e set valued map K

defined b y e) is u p p e r semicontinuous and therefore its g r a p h is clo-
sed.
The proof of the T h e o r e m I. f o l l o w s immediatly from the time depen-
dent Viability Theorem. (See e.g. [ I] ~)

Remark 4. The m o s t important condition is the condition h) w h i c h

 173

is n e c e s s a r y in our case b e c a u s e R n has a finite d i m e n s i o n . This fol-

lows f r o m the v i a b i l i t y theory. If t > 0 and [~(0)[ < u(t) t h e n (t,~(0))
6 int (graph (K)) and t h e r e f o r e the c o n t i n g e n t cone T g r a p h ( K ) ( t , ~ ( 0 ) ) =

= R n~ . Evidently, in this case the c o n d i t i o n h) is s a t i s f i e d .

As to w h a t c o n c e r n s the e x i s t e n c e of the s o l u t i o n u(t) f r o m the
condition d) we have the f o l l o w i n g lemma.

Lemma i. Let be s a t i s f i e d a) f r o m the T h e o r e m I. Let i) G0(t,c) E

-1
e LI(J ) for e a c h c ~ 0 7 ii) l ~ m i~f(c 0 / G0(s,c)ds) = 0. T h e n t h e r e ex-
ists a s o l u t i o n u : J ~ J of the e q u a t i o n (5) such that lim u(t) = 0 as
t - ~. If, m o r e o v e r , iii) tG0(t,c) E LI(J) for e a c h c ~ 0, t h e n this
solution u(t) E Lp(J), p Z i.

The p r o o f of this L e m m a 1 can be m a d e v i a S c h a u d e r fixed point

theorem.

Theorem 2. Let be s a t i s f i e d a) and b) f r o m the T h e o r e m i. c') Let

y : R ~ R n be a b o u n d e d solution to the f u n c t i o n a l differential inclus-
ion (2).
d r ) Let u : J - J be a s o l u t i o n of the e q u a t i o n
(9) ~(t) = - L ( t ) u - G0(t,JiYtlI) ~ - G l ( t , u ) , u(0) > 0
e' ) Let be
K 1 ~ {x e Rn: IxI ~ u(t)}, t 6 J
f' ) L e t be
~t 6 J, K 1 ~ [~ 6 C : ~(0) 6 Kl(t)}
g') Let be the i m a g e of the g r a p h (K I) by the m a p
F l ( t , ~ ) £ f(t,~ + yt ) - f ( t , y t) - g ( t , y t)
relatively compact.
h') L e t for
vt 6 J,v~ s u c h that ~(0) 6 K l ( t ) , v x e Kl(t )
Fl(t,~ ) N DKI(t,~(0))(I ) %
where D K I ( t , ~ ( 0 ) ) is the c o n t i n g e n t derivative of K 1 at (t,~(0)). Then
for e a c h ~ E KI(0 ) t h e r e e x i s t s a s o l u t i o n z(t) to the f u n c t i o n a l
d i f f e r e n t i a l i n c l u s i o n such t h a t
(1o) for a l m o s t all t e j, ~(t) 6 Fl(t,zt)
(z) 0 =
which is v i a b l e in the s e n s e t h a t
t E J, z(t) E Kl(t ) ([z(t)[ ~ u(t))
The s i m i l a r remarks as R e m a r k I - 4 held also in this case. The p r o o f
o f the T h e o r e m 2 f o l l o w s also immediatly f r o m the time d e p e n d e n t Via-
bility Theorem.
 174

Lemma 2. L e t be s a t i s f i e d a) from the Theorem 1 and i) f r o m

Lemma I. T h e n
t s
u(t) = exp(- O
~ L(s)ds) ~ exp( f L(v)dv)G0(s,lly It)ds
t 0 s
is a s o l u t i o n of (9) such that l i m u(t) = 0 as t - ~. If, moreover,
iii) from Lem/na 1 holds, then also u(t) E Lp(J), p ~ I holds true.
The proof c a n be m a d e immediately.
From Theorem i, T h e o r e m 2, L e m m a 1 and Lemma 2 we get

Theorem 3. L e t be s a t i s f i e d all conditions of T h e o r e m 1 and 2.

Then the c o n d i t i o n s i) a n d ii) from Lemma 1 guarantee the a s y m p t o t i c
equivalence between the set of all bounded solutions of (i) a n d the set
of a l l b o u n d e d solutions of (2). If, moreover, the c o n d i t i o n iii) from
Lemma i is s a t i s f i e d , then there exists also the D- integral equiva-
lence, D ~ i, b e t w e e n the a b o v e mentioned sets of s o l u t i o n s .

Now, we w i l l consider the same problem of a s y m p t o t i c and inteqral

equivalences for the systems (i) and (2) bv use of f i x e d Doint method.
Hencefort~ we w i l l assume that the followinq assumptions are satisfied:
(F) If(t,~ I) - f(f,~2)l ~ L(t)w(ll~l - ~21l)
w h e r e L(t) e LI(J), w : J ~ J is a c o n t i n u o u s function, f~L(t)dt = S,
sup w(r) ~ s-l~u, ~ < i; 0
<0,u>
(Hl) gft,~) is n o n e m p t y compact convex subset of R n for e a c h (t,~)6
EJxC;
(H 2) for e v e r y fixed t E J gft,~) is u p p e r semicontinuous in ~;
(H 3) for e a c h m e a s u r a b l e function z : R ~ R n such that z{(_~ 0> 6 C
there exists a measurable selector v : J - R n such that
vft) e g ( t , z t) a.e. on J
We set M ( z f t ) ) ~ {all m e a s u r a b l e selectors belonging to zft)}.
(H 4) there exists a function G0: J x J -- J s u c h that e) G 0 ( t , u ) is m o -
notone nondecreasing in u for e a c h fixed t E J and G0(t,u) E ~ (J) for
any fixed u 6 J; 8)Ig(t,~){ S G0ft,{l~{{) a.e. on J~
y) lira inf ( u -I f~G {t,u)dt) = 0 uniformly for t 6 J.
u ~ 0 0

Lemma 3. L e t z : R -- R n be a m e a s u r a b l e and bounded function. Then

for e a c h v(t) 6 M(z(t)) we have v(t) e LI (J).

Proof. It f o l l o w s from (H4).

Lemma 4. L e t be s a t i s f i e d (F), (~) - (H4). Let be B = {z:R~Rn:

continuous a n d bounded} and B u = {z 6 B : nz~ S u}. L e t x : R -- R n be a
bounded solution of (I) a n d let be ~ ~ C given. Then the o p e r a t o r T
 175

defined on B by the relation : for z @ B it is

oo
(Tz)(t) : {- f ~ [ f ( S , Z s + x s) - f(S,Xs)]ds - f v(s)ds : v(t) @
t t

@ M(z(t) + x(t))} for t E J

(Tz) 0 = {~(t) - ~(0) - f~[ f ( S , Z s + x - f(S,Xs)]ds -

0 s

- f v(s)ds} , for t ~ 0
0
maps B - 2 B, is c o m p a c t and upper semicontinuous in B a n d there exists
such u 6 J that T maps B u into cf(Bu). ( c f ( B u) is the set of a l l
closed and convex subsets of B .)
u

Proof. Let be z(t) 6 B. Then Uzll = S < ~ and n y (F) w e have

oo oo
0~ If(s'z s + Xs ) - f ( S ' X s ) I d s -< of L(s)w(llzsIl)ds _<

max w(~) f~ L(s)ds <

O~T~8 0

By L e m m a 3 v(t) ~ M(z(t) + x(t)) is from LI(J). Thus the operator T is

well defined. Evidently, for z(t) E B (Tz)(t) is a subset of B.
Let be llx(t)ll = p. C o n s i d e r the set B u" Let be z(t) E B u and let
be ~(t) e (Tz)(t). Then there exists such v(t) 6 M(z(t) + x(t)) t h a t

~(t) = - f~[ f ( s , z s + x ) - f(S,Xs)] ds - f v(s)ds, t 6 J

t s t

~(t) = ~(t) - ~(0) - f~[f(S,Zs+ Xs ) _ f ( S , X s ) ] d s -

0

- f~v(s)ds, t e (-~,0)
0
and

(~) l~(t)l ~ max w(r) /~ L(s)ds + /~ G (s,p + u)ds = K < ~,

O~rsu 0 0 0
tEJ

Thus the functions ~(t) 6 (Tz)(t) are uniformly bounded by the constant
K and b e c a u s e for each z(t) 6 B u we get the same constant K, w e may
conclude that TB u is the set of c o n t i n u o u s and uniformly bounded
functions.

Let be 0 ~ t I < t2. Then for ~('t) 6 (Tz)(t), z(t) 6 B u we have

t t
l~(t 2) - ~(tl)l S f 2 1 f ( S , Z s + Xs) - f(S,Xs)Ids + ~21v(s)Ids S
tI tI
 176

t t
S max w{r~ f2 L ( s ) d s + ~2 G 0 ( s , ~ e ~ds
0srsu t t
1 1

From this w e conclude that all functions f r o m TB u are e q u i c o n t i n u o u s

on J. M o r e o v e r , to e a c h e > 0 there e x i s t s t0(e) > 0 such that f o r
t0(E) S t I < t2 w e h a v e

[~(t 2) - ~(tl)I < m a x w(r) f ~(s)ds + G0(S,p + u)ds <

0~r~u to to

Then from this, from the uniform boundedness and f r o m the equiconti-
nuity of all f u n c t i o n s of TB u it f o l l o w s t h a t TB u is c o m p a c t in the
topology of u n i f o r m convergence.
Evidently, to each m o u n d e d set A C B there exists such u E J that
A C B u and TA C TB u . F r o m t h i s it f o l l o w s t h a t T is c o m p a c t in B.
L e t be Zn(t), z(t) 6 B and let {Zn(t)} c o n v e r g e to z(t) in B, i.e.
uniformly on R. T h e r e f o r e , the set {Zn(t), z(t), n = 1,2,..} is
bounded in B. Thus there exists u ~ 0 such that Zn(t) 6 Bu, z(t) C Bu
and TB u is a c o m p a c t set. L e t hn(t) 6 (Tz)(t), n = 1,2,... Evidently
h n ( t ) E TBu, n = 1,2... The set TBu being compact there exists a
subsequsnce {hni(t) } of {hn(t) } , w h i c h converges uniformly to a
function h(t) 6 TBu. Then to e a c h hn(t) there exists Vn(t) E M(zn(t) +
+ x(~)), n = 1,2,... such that

hn(t) = - S~ [ f ( s , ( Z n ) s + x s) - f ( S , X s ) ] d s - f~ v (s)ds,
t t n
t 6 J, n = 1,2,...

hn(t) : ~(t) - ~(0) - f ~ [ f ( s , ( Z n ) s + x s) - f(S,Xs)]ds -

0

- Vn(S)ds , t 6 (-~,0~
0

By L e m m a 3 we have

Iivn(t) II1 < f~ G 0 ( s , u + p)ds <

0
It m e a n s that the sequence {Vn(t)} is b o u n d e d in LI(J). Furthermore, if
{Ek}, E k C J, is a n o n i n c r e a s i n g sequence of s e t s such that ~ E k -- ~,
k=l
then

lim If V n ( S ) d S l -< l i m f IVn(S)Ids _< l i m f G 0 ( s , u + p)ds = 0

k~ Ek k~ Ek k~ Ek

Then (see [2], Th. IV. 8.9) it is p o s s i b l e to c h o o s e from {Vn(t)} a

 177

subsequence {vnk(t)} which weakly converges to some V(t) 6 Li(J).

Now, b e c a u s e {znk(t)} converges to z(t) in B a n d vnk(t) E
g(t,znk(t)), k = 1,2,..°, using (H2) , to g i v e n e > 0 and t 6 J t h e r e
exists M = M(t,e) such that for any n k ~ M we h a v e

g(t,Znk(t)) C 0 (g(t,z(t)))

where 0£(g(t,z(t))) is e - n e i g h b o u r h o o d of the set g ( t , z ( t ) ) . It m e a n s

that for all n k ~ N v n k ( t ) C 0 (g(t,z(t))).
Consider the s e q u e n c e {vnk(t)}, n k ~ N. Then (see [2] , C o r o l l a r y
V.3.14) it is p o s s i b l e to c o n s t r u c t such c o n v e x combinations f r o m Vnk,
n k ~ N, d e n o t e t h e m gm(t), m = 1,2,... t h a t the s e q u e n c e {gm(t)}
converges to v(t) in LI(J). Then by Riesz theorem there exists a
subsequence {gmi(t) } of {gm(t)} which converges to v(t) a.e. on J.
From the c o n v e x i t y of 0 e ( g ( t , z ( t ) ) ) a n d f r o m the fact that vnk(t) @
6 0 (g~t,z(t))) it f o l l o w s t h a t g m i ( t ) e 0 ( g ( t , z ( t ) ) ) , i = 1,2,... and,
t h e r e f o r e , v(t) e ~ ( g ( t , z ( t ) ) ) . F o r c~ 0 w e get that v(t) e g(t,z<t)).
£
R e c a l l t h a t t w a s a f i x e d p o i n t and that g ( t , z ( t ) ) w a s a c o m p a c t
convex s u b s e t of Rn°
Thus

h(t) = - / ~ [ f ( s , z s + x s) - f ( S , X s ) ] d s - /~ v(s)ds
t t
is w e l l d e f i n e d and h(t) 6 (Tz)(t) for t 6 J. It f o l l o w s f r o m the w e a k
convergence of {vnk(t) } to v(t) in LI(J) t h a t the s u b s e q u e n c e {hnk(t)}
of the s e q u e n c e {hn(t)} , i.e. for t @ J

hnk(t) = - tf [f(s, (znk )s + Xs) - f(S'Xs)]ds - t S~ v n k ( S ) d s

converges to h(t) a.e. on J. H o w e v e r , the f u n c t i o n s hnk(t) belong to

the c o m p a c t set TB . T h e r e f o r e , t h e r e e x i s t s a s u b s e q u e n c e of the
u
sequence {hnk(t)} w h i c h c o n v e r g e s to a f u n c t i o n ~(t) u n i f o r m l y on J.
It m e a n s t h a t ~(t) = h(t) 6 (Tz)(t) a.e. on J. W i t h this w e end the
proof of the u p p e r s e m i c o n t i n u i t y of the o p e r a t o r T.
Consider n o w Bu. L e t be z(t) 6 Bu, ~(t) e (Tz)(t). Then f r o m (n),
(F) and y) f r o m (H 4) w e g e t for 0 < c < 1 - ~ the e x i s t e n c e of such
2
u > 0 that

I~(t)I S ~ u + (D + u)c < (e + 2 c ) u < u

Thus ~(t) 6 B u and TB u C Bu. We h a v e already proved that TB u and a l s o

(Tz)(t), z(t) 6 Bu, are c o m p a c t and, therefore, a l s o closed. From the
hypotheses (HI) , (H 3) it f o l l o w s that M(z(t)) is n o n e m p t y and convex,
 178

therefore, (Tz)(t) is also nonempty and convex. Thus T maps B u in

cf <B).

Lemma 5. L e t be satisfied (F), (~) - (H4). L e t B, B u be as in

Lemma 4. L e t y : R -- R n be a b o u n d e d solution of (2) on J. Let ~ E C
be given. Then the o p e r a t o r T 1 defined o n B b y the r e l a t i o n s : for
z(t) 6 B it is

(TlZ)(t) = {- f~[ f ( S , Z s + ys ) - f ( S , Y s ) ] d s +
t

+ f v(s)ds, v(t) 6 M(y(t))}, t e J

t

(TlZ) 0 = [~(t) - ~(0) - 0f~[f(S'Zs+ Ys) - f(S'Ys)]ds +

+ f v(s)ds}, t ~ 0
0
maps B - 2 B, is c o m p a c t and upper semicontinuous in B and there exists
such u 6 J that T maps B u in c f ( B u ) .

The proof of this Lemma can be m a d e in ~he same way as the p r o o f

of L e m m a 4.

From Lemma 4 and Lemma 5 follows

Theorem 4. L e t b e satisfied (F),(HI)-(H4). Then between the set of

all b o u n d e d solutions of (1) a n d the set of all bounded solutions of
(2) t h e r e is the a s y m p t o t i c equivalence. Moreover, if
(Ii) tL(t) E ~ (J), tG0(t,c ) @ ~ (J) for e a c h c ~ 0
then there is p - i n t e g r a l equivalence, p Z i, b e t w e e n the a b o v e mentio-
ned sets of b o u n d e d solutions of (I) a n d of (2).

Proof. L e t be x(t) a bounded solution of (i) on J a n d let ~ @ C be

given. T h e n by L e m m a 4 there exists a ball B u C B such that T maps Bu
into cf(Bu),T is u p p e r semicontinuous a n d TB u c o m p a c t . Thus by F a n
fixed point theorem T has a fixed point z(t) e Bu, i.e. there exists
v(t) E M(z(t) + x(t)) such that

z(t) = - f~[ f ( S , Z s + x s) - f(S,Xs)]ds - f~v(s)ds, t @ J

t t
(z)0(t) = ~(t) - ~(0) - f~[ f ( S , Z s + X s ) - f ( S , X s ) ] ds- f ~ v ( s ) d s , t ~ 0
0 0
Evidently, lim z(t) = l i m ( y ( t ) - x(t)) = 0 as t ~ ~ a n d y(t) = x(t) +
+ z(t) is a b o u n d e d solution of (2). Moreover, if (ii) is satisfied, we
get
 179

Iz(t)l <- <0,uzu~Sup(r)tf~L(s)ds + tf%0(s'Ux]i + ILzil)ds

Thus by Lemma 2 from [3] z(t) 6 Lp(J), p > i.

Let now y(t) be a bounded solution of (2) and let ~ 6 C be given~
Then by Lemma 5 there exists a ball BuC B such that the operator T1
maps B u into cf(Bu) , T 1 is upper semicontinuous and TIB u is compact.
Thus Fan fixed point theorem gives the existence of a fixed point of
T 1 in Bu, i.e. there exists v(t) 6 M(y(t)) such that

z(t) -- - f~[f(S,Zs+ ys ) - f(S,Ys)]ds + f~v(s)ds, t 6 j

t t
(z)0(t) = ~(t) - ~(0)- f~[ f(S,Zs+ ys )- f(S,Ys)]ds + f~v(s)ds,
0 0
t < 0
Evidently, lim z(t) = lim(x(t) - y(t)) = 0 as t - ~ and x(t) = y(t) +
+ z(t) is a bounded solution of (I). Moreover, if (ll) is satisfied,
the n
Iz(t)] _< sup w(r) f~L(s)ds + f~G0(s,UyH)ds
( 0,11 zI~} t t
which by Lemma 2 from [3] means that z(t) 6 L (J), p >_ I.
P

References

[ i] AUBIN,J.P., CELLINA,A., D i f f e r e n t i a l I n c l u s i o n s , Springer Verlag,

1984, A Series of Comprehensive Studies in Mathematics.

[2] DUNFORD,N., SCHWARTZ,J.T., Linear Operators, General Theory, 1958,

Interscience Publishers, New York, London.

[3] H A ~ K , A . , ~VEC,M., Integral equivalence of two systems of

d i f f e r e n t i a l equations, Czech. Math. J., 32 (i07), 19B2, 423-436.
LINEAR PERTURBATIONS
OF GENERAL DISCONJUGATE EQUATIONS
W. F. TRENCH
Drexel University
Philadelphia, Pennsylvania, U.S.A.

Suppose that pl,...,Pn_l, q E C[a,~] , Pi > 0, and

( 1) f~ Pidt = ~, 1 -< i < n - I,

and define the quasi-derivatives

(2) LoX = x ; LrX = ~r(Lr_iX) ' , 1 < r _< n

(with Pn -- i). We will give conditions which imply that the

equation
(3) LnU + q(t)u = 0

has solutions which behave as t ~ ~ like solutions of the equation

L x = 0.
n
Let 10 -- 1 and
t
Ij(t,s~ qj ..... qi ) = f qj(w)Ij_l(W,slqj_ 1 .... ,qi)dw, J -> i.
s
Then a principal system [ 2] for L x -- 0 is g i v e n by
n
xi(t) -- I i _ l ( t , a ; P l , . . . , P i _ l ) , 1 _< i < n ;
in fact,

L r X i ft~"~ = ~ I i _ r _ l ( t , a ; P r + l , . . . , P i _ l ) , 0 < r < i - 1 ,

(4)
[ 0, i S r < n - 1 .

We also define

Yi(t) = In_i(t,a;Pn_l,...,pi) , 1 < i <_ n,

and
= i LrXi(t)' 0 S r ~ i - i,
(5) d i r (t)
[ i/Ir_i+l(t,a;Pr,...,pi), i S r ~ n .
We give sufficient conditions for (3) to h a v e a solution /ul.
such that
(6) LrU i = LrX i + o(dir) (t ~ ~), 0 S r S n - I,

for some given i in {l,o..,n}. This formulation of the question is

 182

due to Fink and Kusano, and the best previous result on this question
is the following special case of a theorem obtained by them in [ I].

THEOREM i. If

(7) f~xiYilqlds < ~,

then (3) ha~ a ~ o l u t i o n ui which satisfies (6).

Our results require less stringent integrability conditions. We

need the following lemma from [4].

LEYLMA I . Suppose that Q e CI t o , ~ ) f o r some t O >_ a, that

f~YiQdt converges (perhaps conditionally), and t h a t

sup I / y i Qd sl <
_ ~(t), t > to,
Tat
where ~ is nonincreasing and c o n t i n u o u s on [t0~). Define

K(t;Q) -- f~In_i(t,s;p i .... ,Pn.1)~(s)ds,

t
and, for t >_ t o , let
J(t;Q) = K(t;Q) if i = l;
or
t
J(t;Q) = f pl(s)K(s;Q)ds = l!(t, to;PiK( ;Q))ds
to
if i -- 2; or
J(t;Q) = Ii_ 1 t, to;Pl ..... Pi_iK(;Q)
if 3<i<_n.
Then
(8) L n J ( t ; ~) = - Q ( t ) , t a to,
and
{ ~ ( t o ) d i r ( t ) , 0 <_ r < ~ - 2,
ILrJ( ;Q) I -< t a to;
2'9(t)dir(t) , i - i < r < n - 1,
moreover, if lim ~(t) = 0, then also
t~
Lr(J(t;Q)) = o(dir(t)) , 0 <_ r <_ i - 2.

The following assumption applies throughout.

ASSUMPTION A. Let f Yixiqds converge (perhaps conditionally),

and suppose that

(9) E(t) = I Yixiqds = O(~(t))

t
 |B3

with ~ nonincreasing on [a, ~) , and

(i0) l i m ~(t) = 0
t-~o
If t O > a, let B(t 0) be the set of f u n c t i o n s h such t h a t
L0h,...,Ln_ 1 h 6 C[t0,~ ) and
0(dir), 0 _< r -< i - 2,
Lrh = t -> t o ,
0(~dir), i - 1 _< r <_ n - 1 ,

with norm H II d e f i n e d by

(il) llhn = sup m a x ~ ILrh(t)l ILrh(t)I )}

t->t0 L~(t0)dir(t) (0<r<_i-2), 2 ~ ( t ) d i r ( t ) (i-l_<r_<n-i

Then Lemma 1 with Q -- q v and ~ -- K~ implies the f o l l o w i n g lemma.

LEM~4A 2. If v e C[t0,~ ) and

i f~yiqvdsl ~ K~(t), t -> t o f

t
then
J(;qv) e B(t O)
and
llJ( ; q v ) , ~ K .

Now define the t r a n s f o r m a t i o n T by

(12) (Th)(t) = J ( t } q x i) + J ( t ; q h ) .

Lemma 2 and Assumption A imply that J( ;qx i) E B(t0) for all t0> a.
We n e e d o n l y impose further conditions which will i m p l y that /~yiqhds
converges (perhaps conditionally) if h 6 B(t0) , and that

I I~yiqhdsl S llhll~(t;t0)~(t) , t a tO ,
t
where ~ d o e s not d e p e n d o n h, and
(131 sup ~ ( t ~ t 0) = 8 < 1
t->t0
if t o is s u f f i c i e n t l y l a r g e , L e m m a 2 will then i m p l y that T is a
contraction mapping of B ( t 0) into itself, and t h e r e f o r e that t h e r e
is an hi in B ( t 0) such that Th i = h i, It w i l l then follow from
(8) and (12) that ui = xi + h i is a s o l u t i o n of (3). M o r e o v e r , Lemma
3 with Q = qu i will i m p l y that

(14) LrUi~ - L r X i = ~ o(dir), 0 S r ~ i - 2

[ 0(~dir), i - 1 S r S n - 1 .

T h e n e x t l e m m a can be o b t a i n e d from (9) a n d i n t e g r a t i o n by parts.

 184

See [31 for the p r o o f of the special case w h e r e PI: "'" : Pn : i.

LEMMA 3. Let

(15) H 0 = yiq ; Hi(t) = t / Pj_iHj.ldS, 1 ~ j ~ i (Po = i).

Then (9) i m p l i e s that

(16) Hj = 0 ( ~ / L j _ i x i ), 1 ~ j ~ i ,

and that the integrals

(17) J~ p j ( L j x i ) H j d s , 0 ~ j ~ i - i,

all converge. Moreover, if the c o n v e r g e n c e is a b s o l u t e for some j = k

with 0 ~ k ~ i - 2, then it is a b s o l u t e for k S j ~ i - Io

THEOREM 2. If

(18) IEN (~(t)) -i f ~ p i _ l I H i _ l ; ~ d s = A < 7i

t~ t
then (3) has a solution u, which satisfies (14).

Proof. Integration by p a r t s yields

T i-i T
(19) t/ Y i q h d s = - j=iZH (L] j_l h)ItT + tf P i - i H i - l ( L i - l h ) d s

if h E B(t0) and 2 < i ~ n; if i = i, then the s u m on the r i g h t

is v a c u o u s and (19) is t r i v i a l . (Recall (2) a n d (15).) Now (5),(9),
(11)2(18), and Lemma 3 imply t h a t we c a n let T ~ ~ in (19) and infer
(13) w i t h
i-I
(20) o ( t ; t 0 ) : 9 ( t 0 ) ( ~ ( t ) ) -I Z IH ( t ) I L j _ i x i ( t ) +
j=l j

+ 2 ( ~ ( t ) ) -I f~Pi_llHi_ll~ds -
t
From (16), the s u m on the r i g h t side of (20) is b o u n d e d on [a,~) ;
hence, (i0) a n d (18) imply (13) for t o s u f f i c i e n t l y large. This
completes the p r o o f .

With i = i, (18) r e d u c e s to

(~(t)) -I f yllql~ds < ~ •

t~ t
which is w e a k e r t h a n (7), s i n c e x I = I. T h e next two c o r o l l a r i e s show
that(18) is a l s o w e a k e r than (7) if 2 < i < n.

COROLLARY i. If 2 S i ~ n and

(21) I pk(LkXi)(Lk_ i x i ) - i d t <

for some k in {i ..... i - 1}, then (3) has a solution ui which
 185

satisfies (14).

Proof. F r o m (16),

(22) Pk(LkXi)IHkL S MPk(LkXi)(Lk_IXi)-i ~

for some c o n s t a n t M, so (21) i m p l i e s that (17) w i t h j = k

converges absolutely. F r o m the c l o s i n g s e n ~ n c e of L e m m a 3, this
means that

Pi_llHi_lldS < ~ ,
which obviously implies (18) w i t h A =0.

COROLLARY 2. If 2 ~ i ~ n and
s
(23) t J~pi-l(s)(af Pi_l(W)dw)-l~2(s)ds = o(~(t)),

then (3) has a s o l u t i o n ui which satisfies (13) .

Proof. From (22) w i t h k = i - I and (4), (23) i m p l i e s (18) w i t h

A = 0.

THEOREM 3. If i < i <- n - i and

_ ~ S -i 1
(24) ~ (~(t)) i f ~(s)Pi(S)( f Pi(w)dw) IHi(s)Ids = B < ~ ,
t- ~ t a
then (3) has a s o l u t i o n which satisfies (14).

Proof. Lemma 3 a n d our p r e s e n t assumption enable us to c o n t i n u e

the i n t e g r a t i o n by p a r t s in (19) by one m o r e step, to o b t a i n
oo i oo
f yiqhds = ~iHj(t)Lj_lh(t) + f PiHi(Lih)ds.
t j: t
Because of (5) (with r = i) and (11), this y i e l d s
i-i
~ ( t ; t 0) = ~ ( t 0 ) ( ~ ( t ) ) -I ~ IHj ( t ) I L j - l X i ( t ) + 2Hi(t) +
j=l
s -I
+ 2(~(t)) -I f ~ ( s ) P i ( S ) ( I Pi(w)dw) IH.(s)Ids.l
t a
Now (10) and (16) i m p l y (20) for t O s u f f i c i e n t l y large. This c o m p l e t e s
the proof.

COROLLARY 3. If 1 S i ~ n - i and

(25) f Pi(S)( J Pi(W)dw) le2(s)ds = o(~(t)),

t a
then (5) has a solution ui which satisfies (14).

Proof. From (16) w i t h j = i, it f o l l o w s that (25) i m p l i e s (24)

 186

with B : O.

R e f e r e n c e s .

[i] A.M.Fink and T . K u s a n o , N o n o s c i l l a t i o n theorems for a class of

p e r t u r b e d d i s c o n j u g a t e d i f f e r e n t i a l equations, J a p a n J . M a t h .
9 (1983), 277 - 291.

[2] W.F.Trench, Canonical forms and p r i n c i p a l systems for g e n e r a l

d i s c o n j u g a t e e q u a t i o n s , Trans. Amer. Math. Soc. 189 (1974),
319 - 327.

[3] W.F.Trench, Evetual iiseq~ugacy of a l i n e a r d i f f e r e n t i a l equation,

Proc. Amer. Math. Soc. 89 (1983), 461 - 466.

[4] W.F.Trench, Asymptotic theory of p e r t u r b e d g e n e r a l d i s c o n j u g a t e

equations II, Hiroshima Math. J. 14 (1984), 169 - 187.

[5] W.F.Trench, Eventual diseonjugacy of a l i n e a r d i f f e r e n t i a l

equation, Proc. Amer. Math. Soc. 89 (1983), 461 - 466.
ON OPTIMAL CONTROL OF SYSTEMS
WITH INTERFACE SIDE CONDITIONS
M. TVRD~"
Mathematical Institute, Czechoslovak Academy of Sciences
115 67 Prague 1, Czechoslovakia

Let 0 < T < 1 . Denote by Dn the space of functions x : E0,1~ ÷

Rn which are absolutely continuous on D,~J and on (T,]] and such
that their derivatives x are square integrable on E0,1] ( x 6 L 2 ).
n
We want to establish necessary conditions for a local extremum of the
functional of the type

F : (x,u) ~ D n x L2m ~ go [x(0)] + g~(x(~+)] + g1(x(1))

I (0.1)
+ /h(s,x(s),u(s)) ds e R
J
0
subject to the constraints

x(t) - i(t)x(t) - B(t)u(t) : 0 a.e. on [0,1] (0.2)

and
I
Mx(0) + Nx(~+) + K(s) x(S) ds = 0 . (0.3)
f .
0

I. Preliminaries

Throughout the paper the elements in R are considered to be co-

n
lumn n-vectors. Given a c 6 Rn , c* denotes its transposition. Given
a Banach space X , I I-I IX and X denote the norm on X and the
dual of X , respectively. For any x 6 X and # C X , the value of
the functional # on x is d e n o t e d by <x,~> X . If Y is a l s o a Ba-
nach space, then L(X,Y) denotes the space of linear continuous map-
pings of X into Y . For A ~ L(X,Y) , N(A) , R(A) and A denote
its null space, range and adjoint, respectively.

Furthermore, L2 denotes the space of functions x : ~,I] ÷ R

n
square integrable on ~,I~ , equipped with its usual norm denoted by
II-I !L . The norm on Dn is d e f i n e d by x E D n -> Ilxl 1D = Ix(0) I +

Ix(~ + )! + IlXIIL . Obviously Dn is isometrically isomorphic with

 188

L n~ x R 2 n " Its dual will be identified with L n2 X R 2 n ' while

< x,~ >D = a ' x ( 0 ) + b*x(~+) + < x,w >L =

I

a'x(0) + bex(T+) + I W*{S) ~(S) ds

0

for any x ~ Dn and 9 = (w,a,b) G L2 x R x R

n n n
We shall keep the following assumptions.

ASSUMPTIONS. A(t), B(t) and K(t) are square integrable on ~,I~

matrix valued functions of the types n x n , n x m and k x n , re-
spectively, M and N are k x n-matrices. The functions g0(x) ,
gT(x), g1(x) and h(t,x,u) are continuous and continuously differen-
tiable with respect to x and u .

2. Lagran@e Multiplier Theorem

Let us d e f i n e

x(t) - A(t)x(t) 1
A : x ~ Dn ÷ I ,
Mx(0) + Nx(~+) + [K(s) x(s) ds
0

L 2 + iB(t)~(t)l
B : U 6 m

and

T : (x,u) e D n x L2m ~ A x - Bu .

2 L2
Then A E i(Dn, L2n x R k) , B 6 [(L m, n x R k) and T C

i ( D n x L m'
2 L n2 x R k) and the constraints (0.2), (0.3) may be replaced
by the operator equation for (x,u) ~ Dn x L2
m

T(x,u) = 0 . (2.1)

The operator A is r e l a t e d to interface boundary value problems.

It is k n o w n (cf. [I]) that under our assumptions A is n o r m a l l y sol-
vable i e. (f,r) 6 L 2 x Ru belongs to its range iff < Y'f >L +
" n
yr = 0 for all (Y,Y) 6 N ( A~* ) (N(A*) C L 2n x R k ). It w a s also shown
in [1] that N(A*) consists of all (Y,Y) C L~ x R k for which there
exists a z E D such that z*(t) = y* (t) + y~K(t) a.e. on ~,I]
n
and
 189

- ~*(t) - z*(t)A(t) + y*K(t)A(t) = 0 a.e. on [_0,1] , (2.2)

- z*(0) + x*M = 0 , z*(~-) = 0 , (2.3)

- z*(~+) + y*N = 0 , z*(1) = 0 . (2.4)

It is easy to see t h a t 0 ~ d i m N(A) + d i m N(A*) < ~ . H e n c e we m a y

apply Proposition 1.2 of [6~ to o b t a i n n e c e s s a r y and sufficient condi-
tions for the c o m p l e t e controllability of the s y s t e m (O.2), (0.3).

PROPOSITION. R(T) = L2n x R k iff the o n l y couple (z,y) ~ Dn x R k ful-

filling (2.2) - (2.4) together ~ith

- z*(t)B(t) + x*K(t)B(t) = 0 a.e. on E0,1-] (2.5)

is the trivial one: z(t) = 0 on ~,1~ and X = 0 .

L e t us s u p p o s e that R(T) = L2n x R k a n d let (x0,u 0) C D n x L 2m be

such that T ( x 0 , u 0) = 0 . F r o m the a b s t r a c t Lagrange Multiplier Theorem
(cf. [~ 9.3, Theorem I) we o b t a i n that if (x0,u 0) is a local e x t r a -
m u m on N(T) of the f u n c t i o n a l F defined by (0.1) then there exists
a couple (Y,Y) 6 L2 x R k such t h a t each (x,u) ~ D x L2 satisfies
n n m

EF' (x 0,u0) ~ (x,u) = < T(x,u), (y,y) > , (2.6)

L~ x Rk

where F' (x0,u0) stands for the F r e e h e t derivative of F at the p o i n t

(x 0,u 0) with respect to (x,u) ( F' (x0,u 0) E i(D n x L 2m, R) ). Inser-
ting the e x p l i c i t form (0.1) of F into (2.6), applying the i n t e g r a t i o n
by p a r t s formula and t a k i n g into a c c o u n t that
I 1
(x,u) X + a'x(0) + b*x(~+) + I w*(s) x(s) as + I v*(s) u(s) as e R
0 0

is the zero functional on D x L2 iff a = b = 0 , w(s) = 0 and

n m
V(S) = 0 a.e. on ~,I~ we o b t a i n the f o l l o w i n g result.

THEOREM (La~ange Multipliers). Let R(T) = L n2 x R k • T h e n (x0,u 0 )

D n X L2m is a local extremum o~ F on N(T) only if

x0(t) - A(t)x0(t) - B(t)u0(t) = 0 a.e. on E0,13 , (2.7)

Mx0(0) + NX0(T+) + f K(S) x0(s) ds = 0 (2.8)

0
and there exist z C Dn and x ~ Rk such that
 190

ah
z* (t) z*(t)A(t) + y*K(t)A(t) = (~(t,x0(t),u0(t)))
(2.9)
a.~. on EO,1] ,

z ~(0) + y~"M = (OgO(xo(O)))

~x * ' z* (T-) = 0 ' (2.10)

Bg~ , rSgl , )*
- z*(T+) + y~N = B---~--~x0(~+))) * , z*(1) = [~-~--%x0(1)) , (2.11)

Bh
z~(t)B(t) + y~K(t)B(t) = (~(t,x0 (t) ,u0 (t) ))* ,
(2.12)
a.e. on [0,1_]

REMARK. Related topics were treated e.g. in [2], [3], [5].

R e f e r e n c e s

[I] BROWN, R. C., TVRDY, M. and VEJVODA, O.: Duality theory for21inear
n-th order integro-differential operators with domain in L de-
termined by interface side conditions. Czech. Math. J. 32, m (107)
(1982), 183-196.
~] HALANAY, A.: Optimal control of periodic solutions. Rev. Roum.Math.
Pures et Appl., 19 (1974), 3-16.
~3] CHAN, W. L . , S. K. NG : Variational control problems for linear
differential systems with Stieltjes boundary conditions. J. Austral,
Math. Soc. 20 (1978), 434-445.
[4j LUENBERGER, D. G.: Optimization by vector space methods,J. W i l e y &
Sons, New Y o r k - L o n d o n - S y d n e y - T o r o n t o , 1969.
~] MARCHI0, C.: (M,N,P)-con~rollabilit~ completa, Questioni di cont-
rollabilitY. I s t i t u t o U. Dini, Firenze, 1973/2, 14-26.
[63 TVRD~, M.: On the controllability of linear Fredholm-Stieltjes
integral operator, Functional-Differential Systems and Related
Topics. (Proc. Int. Conference, ed. M. Kisielewicz) (1983), 247-
252.
A DESCRHrFION OF BLOW-UP
FOR THE SOLID FUEL IGNITION MODEL
J. W. B E B E R N E S
Department of Mathematics, University Colorado
Boulder, CO 80309, U.S.A.

The nondimensional ignition model for a supercritical high acti-

vation energy thermal explosion of a solid fuel in a bounded container
can be described by

(i) ut Au = e u

(2) u(x,0) = }(x) > 0, x 6 R, u(x,t) = 0, x 6 ~, t > 0

where ~ = {x E ~n: ix I < R} and ~ is radially decreasing, i.e.,

~(x) > ~(y) > 0 whenever ixl < IYl < R and A~ + e ~ > 0 on R.

Assume R > 0 is such that the radially symmetric solution u(x,t)

blows up in finite time T > 0. Then by the maximum principle u(-,t)
is radially decreasing for t E [0,T) and ut(x,t) ~ 0 for all
(x,t) 6 ~ = ~ x [0,T).

Friedman and McLeod [4] recently proved that blow-up occurs only
at the origin x = 0 and in addition that u(x,t) satisfies the fol-
lowing estimates: I) u(x,t) ! - 2 £nlx I + c for all ~ < 1 and
(x,t) E F; II) there exists t < T such that I V u ( x , t ) I < 2e u ( 0 ' t ) / 2 ,
t 6 [~,T), Ixl < R; III) there exists ~ > 0 such that ut(x,t) h
R R
@e u(x't) , t £ [~ u T) r x E [ - ~ , ~ ] ; a n d iv) - £ n ( T - t ) --
< u ( 0 , t ) _<
-~n(T-t) - In~, t E [} r T) , 6 > 0

Since u(x,t) is radially symmetric, the initial boundary value

problem (1)-(2) can be reduced to a problem in one spatial dimension.
Let D = { (r,t): 0 j t ! T, 0 < r < R}. Then if r = Ixl, v(r,t) :
u(x,t) satisfies:

(3) vt = Vrr + __n-i v + ev

r r
(4) v(r,0) = #(r), Vr(0) = 0, v ( R , t ) = 0.

To study the asymptotic behavior of v as t ÷ T, consider the fol-

lowing change of variables: T = -£n(T-t), ~ = r(T-t) -I/2, e = v +
In(T-t) = v - T whose inverse is t = T - e -T , r = ~ e - T / 2 , v = 8 -
£n(T-T). The domain D transforms to D' = { (~,T): 0 < N < Re T/2,
T > -£nT} and 8(N,T) = v - T solves
 194

(6) 8(n,-£nT) = % ( q T I/2) + £nT

@ (0,T) = 0, @(Re T/2 ,T) = -T

n
The following theorem is s i m i l a r to a r e s u l t proven by G i g a - K o h n [5].

Theorem 1. As T ÷ + ~, the s o l u t i o n 8(q,T) tends uniformly to a

function y(~) on c o m p a c t subsets of ~+ where y(q) is a s o l u t i o n
of the p r o b l e m :

(8) y'(0) = 0, y(0) = ~ > 0

which is g l o b a l l y Lipschitz continuous and nonincreasing in ~.

Thus, to d e s c r i b e how the b l o w - u p occurs at (T,0) for (1)-(2),

we n e e d to a n a l y z e the solutions of the steady-state equation (7)-(8)
which are globally Lipschitz and are nonincreasing on [0,~).

Theorem 2. For n = 1 or 2, the o n l y solution of (7)-(8) which is

globally Lipschitz continuous and nonincreasing in ~ is y(q) - 0.

Proof. F o r n -- i, this result was first proven by B e b e r n e s - T r o y [ 2].

The following proof is e s s e n t i a l l y due to D.Eberly. F o r n > 2, t h e
proof fails. Let
g(n) = ~ Y'(n) + 1
and h(n) = y''(~)+ --~- y, (n)
n-i

where y(n) is a s o l u t i o n of (7) - (8).

T h e n g(n) s a t i s f i e s
g,, + (n - 1 n)g,
(9) ~ ~ - ~ + (ey l)g = 0

[g(0) = i, g'(0) = 0
a n d h(n) satisfies

h'' + (~ - 1 _ ~)h" + (e y - l)h _< 0

l
z
(i0)
h(0) = 1 - e ~, h'(O) = 0.
It is c l e a r that g(n) > 0 on I = [ 0 , x 0) w h e r e x0 6 (0,~] .
Set W(q) = gh' - g'h, then W(n) satisfies

(zz)
~
•
W'

~W(O)
+ (n-

: 0
D
1 2)W = -eY(y')2g(n) < 0
 t95

on I. This implies W(n) S 0 on I and hence h(n)/g(n) s h(0)/g(0) =

= 1 - e ~ on I. T h u s , we have
(12) h(n) S (1 - e ~ ) g ( D ) on I.
We now must consider two cases. We assume now that n = 1 or 2.
a) If x 0 < ~, then g ( x 0) = 0 and g'- ~ = - ~Y < 0 implies
g(B) < 0 for all n > n 0, T h u s (ny')' = ng(n) - ney < 0 and y(~) is
not globally Lipschitz on [0,~).
b) If x 0 = + ~ a n d g(q) > e > 0 for all n ~ 0, then (ny')' < 0
by (12) and again y'' (B) < 0. If lim inf g(n) = 0 as n ~ ~ with
g(n) > 0, w e observe that (ii) can be solved for h(n) to give

(13) h(n) = (1 - ea)g(n) -

q 1 e s2/4 Sue-U2/4
- g(~) f ( f eY(y')2g(u)du)ds
0 g2(s) s 0

By analyzing (13), we can show that h(~) - - ~ as n ~ + ~. Once again

we have that y'' (B) < 0 f o r n large and y(n) cannot be globally
Lipschitz on [0,~). This completes the proof in dimensions 1 and 2.
As an immediate consequence of theorems 1 and 2, w e have

Theorem 3. L e t n = 1 or 2. A S t -- T - , v(r,t) - in(T - t) -I -- 0

uniformly on 0 S r S c(T - t) I/2.
These results will appear i n [3].
Several open questions remain. What can be said for n ~ 3? W h a t
happens outside the parabolic domain r S c ( T - t ) I/2 as t - T-?

References

[i] J.Bebernes and D.Kassoy, A m a t h e m a t i c a l a n a l y s i s of blowup f o r

thermal reactions - t h e s p a t i a l l y monhogeneous c a s e , SIAM J .
Appl. Math. 40 (1981), 476-484.

[2] J.Bebernes and W.Troy, Nonexistence for the Kassoy problem,

SIAM J. Math. Analysis, submitted.

[3] J.Bebernes, A.Bressan and D.Eberly, A description of blow-~p

for the solid fuel ignition model, submitted.

[4] A.Friedman and B.McLeod, Blow-up of positive solutions of

semilinear heat equations, Indiana Univ. Math. J. 34 (1985),
425-447.
 196

[5] Y.Giga and R.Kohn, A s y m p t o t i c a l l y s e l f - s i m i l a r blow-up of

s e m i l i n e a r heat e q u a t i o n s , Comm. Pure Appl. Math. 38 (1985),
297-320.

[6] D. Kassoy and J.Poland, The thermal e x p l o s i o n confined by a

c o n s t a n t temperature boundary:I. The i n d u c t i o n period s o l u t i o n ,
SIAM J. Appl. Math. 39 (1980), 412-430.
SPECTRAL ANALYSIS OF
NON-SELF-ADJOINT
ELLIVI C OPERATORS
J, BRILL~
Institute of Applied Mathematics and Computing Technique, Comenius University
842 !5 Bratislava, Czechoslovakia

i. I n t r o d u c t i o n
Many important problems of m a t h e m a t i c a l physics lead to analysis
of the d i f f e r e n t i a l e q u a t i o n
n 8k
E Ak u = f, in ~, (i)
k:0 8~t

where A k are symmetric positive d e f i n i t e elliptic operators of order

2m. W h e n d e a l i n g with analysis of these equations we assume that ~ -
the d o m a i n of d e f i n i t i o n is b o u n d e d and 89 - the b o u n d a r y is suffi-
ciently smooth. We consider homogeneous boundary conditions and non-
homogeneous initial conditions.
When applying Laplace t r a n s f o r m we arrive at

A(p)~ = nE pkAk u = fe, (2)

k=0

where a tilde denotes the Laplace transform and f* includes initial

conditions. The o p e r a t o r A(p) is a c o m p l e x symmetric n o n - s e l f - a d j o -
int elliptic operator.
For analysis of equations (2) we have introduced [ 1 - 2] spaces
of analytic functions v a l u e d in S o b o l e v spaces, w h i c h are isomorphic
to w e i g h t e d a n i s o t r o p i c S o b o l e v spaces c o n v e n i e n t for analysis of e-
quations (I).
Now we shall deal with spectral analysis of c o m p l e x symmetric
operators and show that it is possible to obtain similar results on
existence of eigenvalues and c o m p l e t e n e s s of sets of e i g e n v e c t o r s as in
the case of symmetric compact operators.

2. Spectral analysis
Operators A(p) are complex symmetric operators. Thus it holds
Ae(p) = A(p) and

(Ax,x) = (X,%£) . (S)

When AkA 1 # AIA k i.e. when operatore A k are n o n c o m m u t a t i v e

 198

AA* # A e A (4)

and A(p) is a nonnormal operator. Thus for their analysis it is not

possible to apply the spectral theory of symmetric c o m p a c t operators.
However it is possible to generalize some of its results.
I.C. G o k h b e r g and M.G. Krein [3] delt with the spectral analysis
of (i) from the point of view of a nonlinear e i g e n v a l u e p r o b l e m

n
Z IkAk e = 0 . (5)
k=0

When applying this approach we cannot use valuable results of the li-
near spectral theory.
T h e r e f o r e for the D ~ o b l e m under c o n s i d e r a t i o n we define a linear
eigenvalue p r o b l e m considering the equation

n
A(p)e(p) = kZ0PkAke(P)= = l(p)e(p) , (6)

where l(p) for w h i c h the solutions of (6) exist are eigenvalues and
the c o r r e s p o n d i n g solutions e(p) are e i g e n v e c t o r s of (6). Both
eigenvalues and e i g e n v e c t o r s are in general functions of the parameter
p. Eigenvalues in the sense of (5) are values of p for which

l(p) = 0 (7)

and the c o r r e s p o n d i n g values of e(p) are e i g e n v e c t o r s of (5).

For nonnegative real values of p A(p) is a symmetric positive
definite elliptic operator. Thus it has discrete spectrum and a
complete pairwise o r t h o g o n a l set of eigenvectors. Then there exists a
n e i g h b o u r h o o d ~Pl of the positive real semiaxis Pl' where A(p) has

the compact inverse B(p) = A-I(p) and BI(p) = Re B(p) and B2(p) =
= Im B(p) are positive symmetric c o m p a c t operators.
The we can prove:

-l(p)
T h e o r e m i. The operator B(p) = A has at least one nonzero ei-
genvalue and its eigenvalues and eigenvectors are solutions of the
v a r i a t i o n a l problem

min m a x [I(B e,~)1 - lull(e,~)l], ~ = i/l (8)

 199

Proof: As BI(p) a n d B2(p) are p o s i t i v e operators the t r a c e of

B(p) is n o t e q u a l to zero. Therefore B(p) is n o t a q u a s i - n i l p o t e n t
operator a n d has at l e a s t o~e n o n z e r o eigenvalue. Further the G a t e a u x
derivative of (8) y i e l d s the c o n d i t i o n

1 [(B e , ~ ) ( B ~,e) + (B e , ~ ) ( B ~,h)] -

I(B e,~)l
(9)
1
-Ipll(e,-~) I [(e,,'~)(e',e)+ (e,e')(~,h)] = 0

What is f u l f i l l e d by

B e = He . (l 0)

Analysis of the s e c o n d G a t e a u x derivative shows that (i0) is a

s a d d l e p o i n t of (8).

Theorem 2. E i g e n v e c t o r s of a c o m p l e x symmetric operator B(p) and

eigenvectors of its a d j o i n t Be(p) = B(p) form biorthogonal systems
which can be b i o r t h o n o r m a l i z e d .

Proof: For ~k # Ul it h o l d s H k ( e k , ~ I) = ( A e k , e I) and Hl(ek,~l) =

= ~ l ( e l , ~ k) = ( A e l , e k) = (Aek,~l). Then

(u k - ~ l ) ( e k , ~ I) = 0 . (ii)

Hence for H k % Pl (ek'el) = 0 and e i g e n v a l u e s ek,e I form biorthogonal

systems.
Points p, w h e r e it h o l d s (e(p), e(p)) = 0, w i l l be c a l l e d excep-
tional points of the o p e r a t o r B(p). W e c a n prove:

Theorem 3. Symmetric complex compact operators B(p) = A-l(p) are

semisimple with exception of e x c e p t i o n a l points.

Proof: We s h a l l m a k e the p r o o f for an e i g e n v e l u e of the m u l t i p l i c -

ity two. In this case the J o r d a n canonical f o r m w i l l be

Be I = ~e I + e 2 ,

Be 2 = ue 2 •

After biorthogonalization x 2 = e2, x I = kle I + k2e 2 we a r r i v e at

 200

BX I = ~ixl + ax 2
(1 3)
Bx 2 = #x 2 •

Multiplying the f i r s t e q u a t i o n (I 3) by x 2 and the s e c o n d one by x I we

arrive at

(AxI,~ 2) = a (x2,~ 2) ,
(:[ 4)
( A x 2 , ~ I) = 0 ,

w h a t can be f u l f i l l e d only when (x2,x 2) = (e2,~ 2) = 0. In a s i m i l a r

w a y we can p r o v e our a s s e r t i o n also for e i g e n v a l u e s of h i g h e r m u l t i -
plicity.
This t h e o r e m h o l d s also for c o m p l e x symmetric matrices. W h e n the
eigenvector e n belonging to the e i g e n v a l u e ~ of the m u l t i p l i c i t y n ful-
fil the c o n d i t i o n (en,e n) # 0 the c o r r e s p o n d i n g canonical f o r m is
diagonal and the m a t r i x is simple. J. H. W i l k i n s o n has s h o w n an exa-
m p l e of a c o m p l e x symmetric matrix, w h i c h c a n n o t be d i a g o n a l i z e d . It
is [ 4]

This m a t r i x has a t w o - f o l d eigenvalue ~ = i and the e i g e n v e c t o r

e 2 = [i, -i] , thus (e2,~ 2) = 0 a n d a c c o r d i n g to the a b o v e r e s u l t s
the matrix cannot be d i a g o n a l i z e d .
Then similarly as in the case of s y m m e t r i c compact operators we
can construct a complete s y s t e m of e i g e n v e c t o r s . It holds:

Theorem 4. Operators B(p) = A-l(p) a n d A(p) h a v e w i t h e x c e p t i o n

of e x c e p t i o n a l points a countable complete set of e i g e n v e c t o r s el,e2,
e3,.., biorthogonal or b i o r t h o n o r m a l to the c o m p l e x conjugate set of
eigenvectors of the a d j o i n t o p e r a t o r s ~ and ~ corresponding to
e i g e n v a l u e s ~I' ~2' ~3 .... ( resp. ~k = I/~k) w i t h t~iI ~ t~21 ~ I~31~...
s u c h t h a t for f = B h w e h a v e

f = l(f,eL)e k = E(f,ek)~ k , (16)

k k

what corresponds to c o v a r i a n t and contravariant expansions of v e c t o r s ,

respectively. Then it h o l d s
 20f

UfH 2 = Z(f, e k } ( f k , ~ k ) (17)

k

The proof is s i m i l a r to that for s y m m e t r i c compact operators.

At e x c e p t i o n a l points it is n e c e s s a r y to r e p l a c e a basis with the
eigenvector by an other biorthonormal basis of the s u b s p a c e corresponr
ding to the m u l t i p l e eigenvalue.
Finally we can prove the basic theorem on a n a l y c i t y of e i g e n -
values and e i g e n v e c t o r s of A(p).

Theorem 5. Suppose that A(p) = A0+ PAl+ p2A 2 +...+ pnAn , w h e r e

A k are p o s i t i v e definite elliptic operators. Suppose that ~ is an
eigenvalue of m u l t i p l i c i t y m of the o p e r a t o r A(p) at P0' w h e r e P0
assumes real nonnegative vaues. Then there exist ordinary power series
~l(p - p 0 ) , . . . , ~ m ( p - p0 ) and power series in H i l b e r t space el( p - p0 ),
...,em( p - p0 ) all c o n v e r g e n t in a n e i g h b o u r h o o d of P0' w h i c h satisfy
the f o l l o w i n g conditions:

I. ei( p - p0 ) is an e i g e n v e c t o r of A(p) belonging to the ei-

genvalue li( p - p0 ), i.e.

A(p)ei( p - p0 ) = ~i(p - P 0 ) e i ( P - p0 ), i = l,...,m , (18)

li(0) = l, i = l,...,m and the e i g e n v e c t o r s ei( P - p0 ) form w i t h

eigenvectors ~j(p - p0 ) of ~ biorthonormal sets, i.e.

(ei(P - p0 ), ~ j ( p - p0 )) = 8ij , i,j = l,...,m, (19)

2. T h e r e exists such a n e i g h b o u r h o o d of I and a p o s i t i v e number

p such that the s p e c t r u m of C(p - p0 ) = ~(p) for p w i t h Ip - p0 I < P
c o n s i s t s e x a c t l y of the p o i n t s kl( p - p 0 ) , . . . , ~ m ( p - p0 ).
Proof can be done by a g e n e r a l i z a t i o n of results of E . R e l l i c h [5].
F. R e l l i c h proved such t h e o r e m for an o p e r a t o r A(e) for small real
v a l u e s of e. He r e s t r i c t e d h i m s e l f to o r t h o n o r m a l systems of eigen-
vectors. T h e n scalar p r o d u c t of a n a l y t i c f u n c t i o n s are a n a l y t i c o n l y
at real v a l u e s of the p a r a m e t e r ~ and the W e i e r s t r a s s preparation
theorem Can be a p p l i e d only to real v a l u e s of e. I n t r o d u c i n g of
biorthonormal sets of e i g e n f u n c t i o n s and scalar products (f,~) enables
to apply the W e i e r s t r a s s preparations theorem also to c o m p l e x values
of p.
Moreover after introducing biorthonormal sets of e i g e n v e c t o r s it
 202

is possible to generalize the proof also for complex values of P0"

Similarly it is possible to generalize other theorems of F.Rellich.

References

[i] Brilla,J., New f u n c t i o n a l spaces and l i n e a r n o n s t a t i o n a r y problems

of mathematical p h y s i c s , Proceedings of Equadiff 5, Bratislava
1981, Teubner, Leipzig 1982, 64-71.

[2] Brilla,J., Novye f u n k c i o n a l ' ~ e p r o s t r a n s t v a i linejnye nestacionar-

nye problemy matemati@eskoj f i z i k i , Proceedings of the 7th Soviet-
Czechoslovak Conference, Yerevan State University 1982, 49-58.

[3] Gokhberg,I.C., Krein, M.G., Vvedenie v teoriju linejnych nesa-

mosoprjazhennyeh operatorov, Nauka, Moskva, 1965.
[4| Wilkinson,J.H., The algebraic Eigenualue Problem, Clarendon Press,
Oxford, 1965.

[5] Rellich,F., Perturbation Theory of Eigenvalue Problems, Gordon and

Breach, New York - London - Paris, 1969.
ON THE MOUNTAIN PASS LEMMA
KUNG-CHING CHANG
Department of Mathematics, Peking University
Beijing, China

In this paper, I propose to d e s c r i b e a generalized Mountain Pass

Lemma (MPL, in short), which extends the o r i g i n a l MPL due to
Ambrosetti and Rabinowitz [ I] in two a s p e c t s :
(a) f r o m a B a n a c h space to a c l o s e d convex subset,
(b) from the strong separation condition of v a l u e s of f u n c t i o n s
to a w e a k e r one.
Three applications on m u l t i p l e solutions of v a r i a t i o n a l inequality,
semilinear elliptic BVP, and minimal surface are p r e s e n t e d .
i. L e t ~ be a B a n a c h space. L e t C be a c l o s e d convex subset of ~.
L e t Q a n d S be two c l o s e d subsets of C.
We say that the b o u n d a r y BQ a n d S link w . r . t . C, if
(I) BQ ~ S = ~,
(2) for each ~ : Q ~ C continuous, satisfying
~IBQ = idlaQ ,
we h a v e
#(Q) NS # #
Suppose that f : C ~ R 1 is a r e s t r i c t i o n of a C 1 f u n c t i o n defined
on a n e i g h b o r h o o d of C. A c c o r d i n g to the v a r i a t i o n a l inequality theory,
we say x 0 E C a c r i t i c a l point of f w . r . t . C, if
<f'(Xo), x - x O) ~ 0 ~ x 6 C ,

where (,) is the d u a l i t y between ~ a n d I.

For x ~ 6 ~, and x I 6 ~, let us d e f i n e

@X~OXl = S u p { < x ~, x - Xl) I x e C w i t h llx - XlIl ~ i}

We extend the Palais Smale (P.S° in short) Condition w.r°t. C as

following:
For any sequence {Xn} C C, such that f(x n) is b o u n d e d , and
|-f'(Xn)~ x - 0 has a convergent subsequence.
n
THEOREM i. S u p p o s e that f satisfies the P.S. Condition w.r.t. C,
and that B s 6 R 1 such that
Sup[f(x)l x e aQ} S ~ ,
S u p { f ( x ) l x e Q} < +~ ,
and
 204

f(x) > e~ ~ x 6 S.
Then one of the three possibilities occurs:
(i) ~ is an a c c u m u l a t e point of c r i t i c a l values.
(2) ~ is a c r i t i c a l value with uncountable K .
(3) c = inf S u p f(x) > e is a c r i t i c a l value,
A E F x 6A
where F = {A -- ~(Q) I ~ E C ( Q , C ) , with ~I~Q-- idlsQ] .

The proof depends on [61 and the following deformation lemma.

L e t K be the critical set of f. ~ a 6 R I, d e n o t e K a = f-l(a) n K
and f = {x E CI f(x) -< a} .
a

DEFORMATION LEMMA. Suppose that c is the u n i q u e critical value

of f in the interval [c,b) and that K is c o u n t a b l e , then f is a
c c
strong deformation retract of fb\Kb .

Proof. It is a c o m b i n a t i o n of the proofs given in K.C. Chang [ 5],

Chang, Eells [ 7] a n d Z.C. Wang [ 19]. A pseudo gradient vector field and
an a s s o c i a t e flow were constructed in [ 7] for f 6 C 2-0 and finite K c,
it w a s proved in [ 5]. A n improvement which enables to c o v e r our
conditions, was given in [ 19].

Proof of T h e o r e m i. If n o n of these cases occurred, then there

would exist ~ > 0 a n d #0 E C ( Q , C ) such that:
= c, f-l(c, c+~] n K = ~, K is c o u n t a b l e and
c
#0(Q) c fc+e
A c c o r d i n g to t h e d e f o r m a t i o n lemma, there is a c o n t i n u o u s ~:
fc+e - fc" Since ~ o ~0 6 C ( Q , C ) with ~ o ~0{~Q= idI@Q, we h a v e
(# 0 # 0 ) ( Q ) n s~ % ~. It i m p l i e s
Sup[f(x)l x 6 # o #0(Q)} > e = c .
This is a c o n t r a d i c t i o n .
As c o r o l l a r i e s , we have

COROLLARY i. Suppose that x 0 6 C is a l o c a l minimum, and that

x I 6 C such that f ( x 0) ~ f ( x I ), t h e n f has a critical point other
than x0 .
In c a s e C = ~, this w a s obtained in K . C . C h a n g [2,4] in 1982.
Obviously~ it i m p l i e s some results in D.G. de F i g u e i r e d o [8], D.G. de
Fiqueiredo S.Solmini [9], and Pucci-Serrin [12].

COROLLARY 2. Suppose that f has two local minima, then there

exists a third critical point.

2. W e p r e s e n t three applications of T h e o r e m 1 (or its c o r o l l a r i e s ) .

 205

(1)Variat:ione~l Inequality
Let ~ b e an o p e n subset in R 3, a n d let g be a n o n n e g a t i v e
measurable function defined o n ~.

THEOREM 2. The functional

f(u) = /[ ~(Vu) 2 - ~1u 3 + gu] (l)

n
i(a)
has at l e a s t two c r i t i c a l points w.r.t, the p o s i t i v e cone P in H 0 .

}
THEOREM 3. Let % 6 H~ ~), and let C = {u 6 H 0l(~)l 0 s u(x) s~(x)
a.e.}. Assume that
i n f { f ( u ) l u e C} < 0 . (2)
Then f(u) has at l e a s t three critical points w.r.t.C.

Outline 0f the p r o o f . It is e a s y to see that u I = 0 is a l o c a l

minimum, and that the global minimum u 2 of f is a t t a i n a b l e . The condi-
tion (2) implies u I = u 2. C o r o l l a r y 2 implies the c o n c l u s i o n of
Theorem 2.Similarly, Corollary 1 implies the c o n c l u s i o n of T h e o r e m 2.

REMARK i. T h e condition (2) is s a t i s f i e d , if ~(X] is large enough.

REMARK 2. F o r similar considerations, see C.Q. Zhung [20] and

A. Szulkin [18].

(2) A c o m b i n a t i o n of the v a r i a t i o n a l method and the sub - a n d

super-solutions.
Let ~ be an o p e n bounded domain with smooth boundary 8~ in R n, and
let g E C ¥ ( ~ X RI,RI), for some o < y < i, be a f u n c t i o n satisfying

Ig(x,t)[ S c(l + [tl ~)

for some constants C > 0 and ~ < ~n+2 if n ~ 3.
t
THEOREM 4. Let G ( x , t ) = f g(x,~)d~. Assume that the functional
0
f(u) = f[~(Vu)
2 - - G(x,u(x))]dx

satisfies the P.S. condition in the s p a c e H~(~), and that f is u n b o u n d e d

below. Moreover if t h e r e exists a pair of s t r i c t sub- and super-solutions
of the e q u a t i o n
-Au = g(x,u) in ~ ,

uIo~ = 0 .
Then the e q u a t i o n h a s at least two distinct solutions.
For a proof, cf. K.C.Chang [2]. A considerable simplification can
be found in K . C . C h a n g [5].
 206

Many a p p l i c a t i o n s d e r i v e d from this theorem, w h i c h includes the

superlinear A m b r o s e t t i Prodi type problem, a nonlinear eigenvalue
problem, A m a n n three solution theorem, and a resonance problem. See
K . C . C h a n g [ 3]. The superlinear A m b r o s e t t i Prodi type p r o b l e m was
rediscussed in de F i g u e i r e d o [8] and de F i g u e i r e d o Solimini [9].

(3) Minimal s u r f a c e s
Let M be a compact oriented surface of type (p,k), and let (N,h)
be a compact R i e m a n n i a n m a n i f o l d with nonpositive sectional curvature.
~f ~ is a conformal structure on M compactible with its orientation,
then we write (M,u) for the a s s o c i a t e d Riemann surface.
For a map ~ : (M,p) ~ (N,h), the e n e r g y is

E(~) = ~ fld~]2dxdy .
M

Let r = {Fi} ~ be a set of d i s j o i n t oriented Jordan curves in N

s a t i s f y i n g an i r r e d u c i b i l i t y condition, which prevents the d e g e n e r a c y
of topological type.

T H E O R E M 5. If #l• : (M,~ i) ~ (N,h), i = 0,1 are homotopic

admissible conformal isolated E-minima, then there is a conformal
structure u on M and an admissible conformal harmonic map
# : (M,u) - (N,h) h o m o t o p i c to both, w h i c h is not an E-minimum.
A special case, in which M is a borded planar domain and N is
E u c l i d e a n space R n, is due to M o r s e - T o m p k i n s and S h i f f m a n [ 13,14,15].
If M is a disc or an annulus and N = R n, that special case has been
reproved by struwe [16,17].
In p r o v i n g this theorem, c o r o l l a r y 2 is applied. The c l o s e d convex
set is the following

C =~k X ~(p.k)~

where ~={u & C ° A HI/2([0,2~ ] ,Rl)lu is weakly monotone, and

,2k~ 2k~
ul--~) = -~-- , for k = 0,1,2,3},

and ~(p,k) denotes the T e i c h m ~ l l e r space of compact oriented surface M

of type (p,k). The M u n f o r d c o m p a c t n e s s t h e o r e m is aDnlied to v e r i f y the
P.S. Condition.
For d e t a i l s see Chang Eells [6,7].
 207

R e f e r e n c e s

[i] A. Ambrosetti, P.H. Rabinowitz, Dual u a % i a t i o n a l methods

in critical p o i n t t h e o r y a~d a p p l i c a t i o n s , $, F u n c t A n a l .
14 (1973), 349-381.

[2] K.C. Chang, A v a r i a n t mouJ~t~i~ pa~s ~emm,~. Scientia Sinica 26,

no. 12, (1983), 1241-1255.

[3] , V a r i a t i o n a l method anl t h e s u b - and s u p e r - s o l u t i o n s ,

ibid, 1256-1265.

[41 , An e x t e n s i o n of minimax p r i n c i p l e . Symp. DD 3 (1982)

Changchuan Jilin.

[5] , Infinite d i m e n s i o n a l Morse t h e o r y and i t s applica-

tions, Lecture Notes of the 22nd Session of the Seminaire de
mathematiques superieures at Montreal in 1983.

[6] K.C. Chang, J. Eells, Harmonic maps and minimal s u r f a c e

c o b o u n d a r i e s , Proc. Lefschetz Centenary. Mexico (1984).
[7] , U n s t a b l e minimal s u r f a c e c o b o u n d a r i e s , Preprint,
April 1985 Univ. of Warwick.

[8] D.G. de Figueiredo, On t h e s u p e r l i n e a r A m b r o s e t t i - P r o d i problem,

MRC Tech. Rcp. #2522, 1983.

[9] D.G. de Figueiredo, S. Solmini, A v a r i a t i o n a l Approach to

superlinear elliptic p r o b l e m s , Comm. in PDE, 9 (7), (1984),
699-717.

[I0] M. Morse, C.B. Tompkins, The e x i s t e n c e of Minimal s u r f a c e s of

general critical types, Ann. Math. 40 (1939), 443-472.

[Ii] ..... , U n s t a b l e minimal s u r f a c e s of h i g h e r t o p o l o g i c a l

structure, Duke Math. J. 8 (1941), 350-375.

[12] P, Pucci, J. Serrin, A mountain pass theorem, to appear.

[13] M. Shiffman, The P l a t e a u problem f o r m i n i m a l s u r f a c e s of

arbitrary t o p o l o g i c a l s t r u c t u r e , K m e r . J . M a t h . 61 ( 1 9 3 9 ) ,
853-882.

[14] , The P l a t e a u p r o b l e ~ f o r non-relative minima,

Ann. Math. 40 (1939), 834-854.

[15] , U n s t a b l e minimal s u r ~ c e s with several

b o u n d a r i e s , Ann. Math. 43 (1942), 197~22.
 208

[16] M. Struwe, 0~ a eritical p o i n t th.eorq f o r minimal s u r f a c e s

s~a~nin~ a w~¢ in R n, ~° reine anq. Math. 349 (1984), 1-23.

[17] , ~ norse theor~ for annulus-t~De minimal

~ur~aces , ~re~rimt.
[ 18] A. Szulkin, M ~ m ~ x ~rinc~ [~les for lower semitontinuous
functions a~,d ~ : ~ c a t i ~ n s to n o n l i n e a r boundarq v a l a e p r o b l e m s ,
PreDrint.

[19] Z.O. Wanq, R em~l~s a,n t~£ d £formation lemma Ito appear).

[ 20I C.Q. Zhung, ~as2er Thesi~ ait tanzhou Univ. 1985.

ON UNIQUENESS AND STABILITY
OF STEADY-STATE CARRIER
DISTRIBUTIONS IN SEMICONDUCTORS
H. GAJEWSKI
Karl- Weierstrafl-Institut fiir l[4athematik der Akademie der Wissenschqften der DDR
1086 Berlin, Mohrenstrafie 39, .ODR

In this paper we e s t a b l i s h a si1~le smallness condition guaran-

teeing the b a s i c equatioHs for c a r r i e r distributions in s e m i c o n d u c t o r s
to p o s s e s s a unique steady-state solution. Uri~er this condition arbi-
t~=~y ~Derturbations of the steady state decay expo~entially in time.

i. I n t r o d u c t i o n

L e t G be a b o u n d e d Lipschitzian domain in R d r
d < 3
- •
Let the
boundary S of G be the u n i o n of two d i s j o i n t parts S1 and $2,
S1 closed in S, m e s S 1 > 0. A f a m i l i a r model of c a r r i e r transport in
a semiconductor device occupying G is g i v e n bv the system [10,13]

-Au = (a/~J(f + p - n), (1.1)

qn t = V . J n- qR, JD = q ~ n (kvn - nVu), (1.2)

~D t = -v. Jn- qR. Jn = -q~p(kVn + pVu~ , (1. 3)

u = U s , n = N s, p = P on R + X S ~.Vu=~.Vn=~.Vp=0 on
s I' (1.4)
R+X S 2

n(0,x) = n0(x), p(0,x) = P0(X). x @ G . (1.5)

Here
u is the e l e c t r o s t a t i s t i c potential,
n and D are the m o b i l e electron and h o l e densities,
J and J are the c u r r e n t densities,
n p
f is the net density of i o n i z e d impurities,
q is the electron charge,
e is the d i e l e c t r i c Dermitivitv of the semiconductor material,
R = (np- n~)/(T(n+p+2ni)) is the r e c o m b i n a t i o n rate,
n. is the intrinsic semiconductor carrier density,
l
T is the e l e c t r o n a n d hole lifetime,
~n and ~p are the (constant) electron and hole mobilities,
Us, N s and Ps are given boundary values,
is t h e o u t w a r d unit normal at a n y p o i n t of S 2.
 210

In the expressions for the current densities the E i n s t e i n relation

Dn. p = kDn,p between diffusion coefficients and mobilities is used.
(k = k B T / g , kb= Boltzmann constant. T = absolute temperature.)
The carrier transport equations (1.1)-(1.3) were derived by V a n
Roosbroeck [ ii] in 1950 a n d are now generally accepted. The first sig-
nificant report on u s i n g numerical techniques to s o l v e these equations
for c a r r i e r s in an o p e r a t i n g semiconductor device structure has been
published by G u m m e l [6] in 1964. Since then, the n u m e r i c a l modelling
of semiconductor devices proved to be a p o w e r f u l tool for d e v i c e de-
signers (see [ 13]) .
In s p i t e of their physical, a n d technological relevance, the d e v i -
ce e q u a t i o n s received relatively little attention from the s i d e of
mathematical analysis. To o u r k n o w l e d g e , the first matematical paper
devoted to t h e s e equations appeared in 1972. In t h i s paper Mock [7]
proved the solvability of the steady-state equations associated to
(I.I)-(1.5) supposing that ~n = ~ a n d R = 0. M o r e recently, Seidman
[ 12], the author [3] and Gr~ger ~5] have published more general exi-
stence theorems for steady states. All these results are b a s e d on m a -
ximum principle and compactness arguments.
As to the instationary problem (1.1)-(1.5), again Mock [8] was
the first to p r o v e a global existence and uniqueness result in a spe-
cial situation. Recently, the author [2] and Gajewski&Gr~ger [4] could
show the existence and u n i q u e n e s s of q l o b a l solutions under rather
qeneral assumDtions. Of c o u r s e , the crucial step in t h e s e papers con-
sists in f i n d i n q appropriate a-priori estimates. Such estimates are
obtained by means of a p h v s i c a l l v motivated LiaDunov function a n d an
iteration technique due to M o s e r and Alikakos.
One of the essential open questions arising from the V a n Roosbroeck
equations is t h a t of the u n i q u e n e s s and stability of s t e a d v states.
General answers to t h i s question are n o t to be e x p e c t bv physical
reasons [I,I0]. A sDecial result in t h i s direction [7] concerns the
case of s m a l l perturbations of the thermal equilibrium which results
from the assumption

U s - k l o g ( N s / n i) = U s + k l o g ( P s / n i) = c = const, on S I
and is q i v e n by

N = n i exp((U - c)/k)~ P = n i exD((c - u)/k) ,

where U is the (uniaue) solution of the n o n l i n e a r boundary value

problem

-~U = (q/e)(f - 2n i s i n h ( ( U - c)/k))) in G,

 211

BU : U s , where BV ~ {v on SI, ~ , V v on S 2] & n d Us: 0 on S 2.

The therm~l equilibrium has been shown to be q l o b a l l v ~svmDtoticallv

stable (comb. [9]for the special case S = S 2 and [2,4] for m o r e ge-
neral situations). In fact, it w a s proved in [4] that for r e a s o n a b l e
initial values the solution (u(t),n(t).p(t)) of (1.I)-(i.5) conver-
aes to the co~resnondinq thermal equilibrium (U,N,P) exponentially
in time. The proof of t h i s result heavily upon the observation
that the function

L(t): ](kq(n(log(n/N)-l)*N+p(log(p/P)-l)+P)+(~/2)IV(U-u)i2)dx
G
is m o n o t o n o u s l y d e c r e a s i n g .
The main DurDose of the present DaDer is to s t a t e another kind of
smallness condition implying uniqueness as w e l l as g l o b a l asymptotic
stability of stationary solutions. Our smallness condition involves
the e s s e n t i a l physical parameters and c a n be e a s i l y checked.

2. R e s u l t s
L e t L 2, L~. i be
H2 the u s u a l space of f n n c t l o n s defined on G.
We use the followina notations

Ivl 2= fv2dx, Ivl = vrai m a x v, llv[l2 = flVvl 2 = dx ,

G G
v = {v 6 H~/ v = 0 o n SI}, W = {v • (H~ N L ) 3 / v 2 , v 3 > 0 in G].

We assume that f • L~ and that the b o u n d a r y values c a n be r e p r e s e n -

ted by functions ( U s , N s , P s) • W. Let i be the smallest eigenvalue
of t h e p r o b l e m
-Av = kv in G, Bv : 0 on S ,
such that we have
llvl 2 _< llvll2 , v • V . (2.i)

Now we can S~ate our results.

Theorem i. L e t (U,N.P) • W be a s t a t i o n a r y solution of (i.i)-(1.4)

such that
1
r(Q) = ~]'f (~(F + Q) + ~ (i + - - Q )) < i
2n i
where
F = IfI~, Q = 4( ~Nl~ + IPl~, ~ = m i n ( ~ n , ~ p) •

Then (U,N,P) is u n i q u e in W.

Remark. A s to e x i s t e n c e results for steady states (U,N,P) • W we re-

fer to [3]. In this paper also explicit bounds for INI~ a n d IPI~ can
be f o u n d which involve only f a n d the b o u n d a r y values.
 212

Theorem 2. Suppose 0 ~ no, P0 @ L~. Let {u,n,p) be the solution

of (l.i)-(1.5) and let (U,N,P) he a s t a t i o n a r y solution satisfying
the h y p o t h e s e s of T h e o r e m i. Then for t ~ 0 the f o l l o w i n g estimates
are valid with a : 2kl~(l - r(Q))

~pln(t)-Ni2+ ~nlP(t)-p[2~ e-at(~pln0-Ni2+ ~nlD0-Pl 2)

~llu(t)-UI ~ IIu(t)-UIi ~ ( q / ( ~ l ) ) ( I n ( t ) - N { + Ip(t)-Pl) .

Remark. The e x i s t e n c e and u n i q u e n e s s of the t i m e - d e p e n d e n t solu~

tion (u,n,p) is g u a r a n t e e d by [4] , T h e o r e m I.

3. Proofs
We d e n o t e by (.,.) the L 2 - s c a l a r product as well as the pai-
ring b e t w e e n the H i l b e r t space V and its dual V ~ C L2. We intro-
duce the set
M = {[N,P} 6 (H i2 N L ~ 2 , N,P ~ 0 on G, N=Ns, P=Ps on SI} .
Finally, we define an o p e r a t o r A @ (M ~ (Ve) 2) by
(A{N,p],[hl,h2]) = ~ p ( ( ~ n ( k V N - N V U ) ~ V h l ) + (R,hl)) +

+ Un((Up(kVP+PVU),Vh2) + (R,h2)) ~hl,h 2 E V ,

w h e r e R = R(N,P) and U = U(N,P) is the solution of the b o u n d a r y va-
lue p r o b l e m
-AU = (q/~)(f + P - N), BU = U on S .
s
The m a i n tool for p r o v i n g our results is the f o l l o w i n g m o n o t o n i c i t y
p r o p e r t y of the operator A.

Lemma. Let [Nj,Pj] 6 M, j=l,2, N2S 9, P2 ~ ~ in G, N,P=cons. Set

Q=4(N+~). Then it holds with m=~ n Upk(1-r(Q)), N=NI-N2, P=PI-P2 ,

(A[NI,P I] - A [ N 2 , P 2] ,[N,P]) ~ m( [IN112+ IIptl2) .

Proof. S e t t i n g UI= U(NI,PI) , U2= U ( N 2 , P 2 ) , U:UI-U 2 and using (2.1)

we get
IIull2= (q/~)(P - N,U) S (q/~)IP - NIIUI S (q/(~l))lIP - NOIIIuI]
and c o n s e q u e n t l y
IIU~ = ( q / ( c ~ ) ) E P - NI ~ (q/(el))llp - Nit,IU1 ~ (q/(El))IP-Nl (3.1)
Thus we find
(kVN-N!VUI+ N2VU2,VN) + (kVP + PIVUI - P2VU2,VP) =
= k(UNII2+ UpH 2) _ (NVUI+ N2VU,VN) + (PVUI+ P2VU,VP) =

= k(~N~2+ RpR 2) + ( q / ( 2 e ) ) ( p 2 - N 2 , f + PI-NI)÷(P2VP-N2VN,?U) =

= k(UNU2+ Hpll2) + ( q / ( 2 e ) ) ( ( ( N - p ) 2 , N I + P I ) - (N2,f + 2P 2) +

+ (p2,f _ 2N2) + 2(NP,N2+ p2)) + (P2V P - N2VN,VU) _>

 2i3

k(iIN1[2+ ilPll2) - (q/(2eX))((F + N+ 3P]IBNII2+ (F + 3N+P) I!Pil2 +

+ 2(NIINII + pllmll)lip - Nil)

k(1 - (q/(2el))(F + Q))(lINi[ 2 + ~mJl2)

On the other hand, setting a = ~ ( N + P + 2n ), we get

J J 3 ±
(R I- R2,N) = ( ( 1 / a l ) ( N P l + N 2 P - ( ( N 2 P 2 - n ~ ) / a 2 ) ~ ( N + P)),N)
a -(I/X)((N/(2al))(IINII2+ llPll2) + (Q/(16al))(IINII + I[pI[)IIN~ +
+ (i/(8~))(liN~ 2 + llml12))

-(I/(41T)((N/ni)(IINTI2 + ~pll2) + (Q/(igni)(311N[12 + IIpll2) =

+ (I/2)(IIN~ 2 + I[P~2)).

Evidently, an analogous estimate holds for (R I- R2,P). Now the lem-

ma is an immediate consequence from these estimates.

Proof of T h e o r e m I. Using the o p e r a t o r A we can rewrite the

s t a t i o n a r y p r o b l e m as follows.
A[N,P] = 0, [N,P] 6 M . (3.2)
From this it becomes clear that the t h e o r e m follows easily from the
lemma.

Proof of T h e o r e m 2. We can write (1.1)-(1.5) in the compact form

[ ~ p n t , ~ n p t] + h[n,p] = 0, [n(t),p(t)] E M, n(0) = no, p(0) = P0"

Hence, using (3.2) and the lemma, we get

0 ~ ~ ( ~ p l n - NI2+ ~nlp - P [ 2 ) t + kX~(l - r)(~pln - N l 2 + ~ n l p - p[2).

Applying a well-known differential inequality and (3.1) we obtain the

theorem.

Remark. Our lemma can also be used in order to find r e l a x a t i o n

p a r a m e t e r s ~ such that the i t e r a t i o n sequence ([Nj,Pj]) d e f i n e d by

-A[hl,h 2] = b A[Nj,Pj], hl,h 2 6 V, j = 0,i ....

Nj#I= Nj + hl, Pj+l = Pj + h2, [N0,P 0] e M

converges to a s t a t i o n a r y solution.

References

[ i] B O N ~ - B R U E V I C H , V . L . , ZVJAGIN,I.P., M I R O N O V , A . G . , Spatial electrical

instability i n semiconductors (russian), M o s c o w 1972.
[2] GAJEWSKI,H., On e x i s t e n c e , uniqueness and a s y m p t o t i c behavior of
t h e basic equations for c a r r i e r t r a n s p o r t in semiconductors,
ZAMM 65, (1985), 101-108.
 214

[3] GAJEWSKI,H., On t h e e x i s t e n c e of s t e a d y - s t a t e carrier

distributions i n s e m i c o n d u c t o r s , In: Prohleme und Methoden der
Mathematlschen Physik, Teubner-Texte zur M a t h e m a t i k 63. (Ed. V.
Friedrich u. a.).
[4] GAJEWSKI,H., GROGER,K., On the basic equations for carrier trans-
port in semiconductors, j. Math. Anal. AppI., to appear.
[5] GROGER,K., On s t e a d y - s t a t e carrier distributions in semiconductor
d e v i c e s , to appear.
[6] GUMMEL,H.K., A s e l f c o n s i s t e n t iterative scheme f o r o n e - d i m e n s i o -
nal steady state transistor calculations, IEEE T r a n s . Electron
Devices ED-II (1964), 455-465.
[7] MOCK,M.S., On e q u a t i o n s d e s c r i b i n g s t e a d y - s t a t e carrier distribu-
tions i n a s e m i c o n d u c t o r d e v i c e , Comm. Pure Appl. Math. 25 (1972),
781-792.
[8] MOCK,M.S., An i n i t i a l v a l u e problem from s e m i c o n d u c t o r device
theory, SIAM J. Math. Anal. 5 (1974), 597-612.
[9] MOCK,M.S., A s y m p t o t i c b e h a v i o r of s o l u t i o n s of t r a n s p o r t equati-
ons f o r s e m i c o n d u c t o r devices, J. Math. Anal. Appl. 49 ( 1 9 7 5 ) ~
215-225.
[i0] MOCK,M.S., A n a l y s i s o f m a t h e m a t i c a l models of s e m i c o n d u c t o r devi-
c e s , Dublin 1983.
[Ii] VAN ROOSBROECK,W., Theo%y of t h e f l o w of e l e c t r o n s and h o l e s i n
Germanium and o t h e r s e m i c o n d u c t o r s , Bell Syst~ ~ Tech. J 29 ( 1 9 5 0 ) ,
560-623.
[12] SEIDMAN,T.I., S t e a d y s t a t e s o l u t i o n s of d i f f u s i o n - r e a c t i o n sy-
stems with electrostatic convection, Nonlinear Analysis 4 (1980),
623-637.
[13] SELBERHERR,S., A n a l y s i s and s i m u l a t i o n of s e m i c o n d u c t o r devices,
Wien-New York 1984.
PARTIAL REGULARITY OF MINIMIZERS
M. GIAQUINTA
Instituto di Matematica Applicata, Universita di Firenze
Via S. Marta, Firenze, Italy

After the e x a m p l e s shown by E. De Giorgi, E. Giusti-M. Miranda,

V.G. Mazja, J. Nedas, J. Sou~ek, it is well known that the m i n i m i z e r s
of v a r i a t i o n a l integrals

(i) ~[u~] = /F(x,u(x), Du(x))dx

n
in the vector valued case, even in simple situations, are in general
non continuous. There is only hope to show partial regularity of mini-
mizers, i.e. regularity except on a c l o s e d set h o p e f u l l y small.
The study of the partial regularity of m i n i m i z e r s and of solutions
of non linear elliptic systems starts w i t h the w o r k s by M o r r e y and
Giusti-Miranda in 1968, and it is the aim of this lecture to refer
about some of the results obtained. I shall restrict myself to some
results concerning the p a r t i a l regularity of m i n i m i z e r s referring to
[7] for a general account.
L e t me start by stating the most general and recent result.

THEOREM i. L e t ~ be a b o u n d e d o p e n s e t i n R n and l e t
F(x,u,p) : ~ X R N X R nN ~ R be a f u n c t i o n such that
i) Ipl m ~ F ( x , u , p ) ~ e O l P l m, m ~ 2
ii) E i s o f c l a s s C2 w i t h r e s p e ~ t 2 t o p and

IFpp(X,u,p)[ ~ Cl(l + Ipl 2 ) 2

m

iii) (i + Ipl 2) 2 F ( x , u , p ) is H~lder~continuous in (x,u) uniformly

with respect to p
iv) F is strictly quasi-convex i.e.for all Xo,Uo,Po and all
E CO(~ , RN)

Hl,m N
Let u E loc[~,R ) be a m i n i m i z e r for
~[ u ; ~ ] = fF(x,u,Du)dx

i.e. ~[ u ; s u p p ~] ~ ~[ u + ~ supp ~]. Then there exists an o p e n s e t

o
such that u E CI'U(eo,RN), moreover meas (~ - C o ) = 0 .

Theorem I, p r o v e d in [ 12] , is the result of a series of steps due

 216

tO d i f f e r e n t authors.
Under the s t r o n g e r ~ c o n d i t i o n of ellipticity
m-2
F . ~ ~ v(l + [pl 2) 2 i~12 ~ @ RuN; v > 0
i 3 I 3
P~PB
theorem 1 was p r o v e d for m ~ 2 by C . M o r r e y and E . G i u s t i , for i < m < 2
bv L.PeDe in 1968 in the case F : F(p); in th~ c a s e m : 2, F : F ( x , u , p )
by Giaquinta - Giusti and Ivert in 1983, in the case m ~ 2, F = F ( x , u , p )
by Giaquinta - I v e r t in 1984. F r o t h e s e r e s u l t s I refer to [7] [9]
[ ii] . U n d e r the w e a k e r assumption of ~ u a s i - c o n v e x i t y in (2) it w a s
p r o v e d by L. E v a n s [5] in the case F : ~(p), m ~ 2.
T h e case 1 < m < 2 is open, and e s s e n t i a l l y o p e n are all the
questions concerning the s i n g u l a r set; for i n s t a n c e
i. w h a t a b o u t the s t r u c t u r e of the s i n g u l a r set? w h a t a b o u t the
Hausdorff dimension of the s i n g u l a r set?
2. are t h e r e r e s o n a b l e structures under w h i c h m i n i m i z e r s are r e g u l a r ?
(see the i n t e r e s t i n g paper [22])
3. w h a t a b o u t the s t a b i l i t y or i n s t a b i l i t y properties of the s i n g u l a r
set? or w h a t a b o u t topological properties of the set of s m o o t h
minimizers?
We have r e s u l t s improving theorem 1 roughly o n l y in case of
quadratic functionals if we e x c l u d e the case in w h i c h F does not
depend explicitly on u. So let us c o n s i d e r a quadratic functional

(3) A(u) : IA~(X'U)DculDBu3dxzj

where the c o e f f i c i e n t s A~ are s m o o t h (for e x a m p l e Holder-continuous)

13
and s a t i s f y the e l l i p t i c i t y c o n d i t i o n

~B ~ B 2 RnN
(4) Aij(x,u)$i~ j ~ [${ v~ e

Notice that the f u n c t i o n a l A is not d i f f e r e n t i a b l e . Concerning the

strong condition of e l l i p t i c i t y (%), we r e m a r k that t h e r e is not m u c h
hope to w e a k e n it. In fact in [14] it is s h o w n that for w e a k solutions
of the s i m p l e q u a s i l i n e a r system

/A~(x,u)D ulDB~3dx = 0 v~ E H ; ( c , R N)

with coefficients satisfying the s t r i c t Legendre-Hadamard condition

A j~ ( x , ul ) g a g3 S n i 4 >- t { 1 2 t n J 2 v$ e R n v~ e R N

Caccioppoli's inequality m a y not be true; and C a c c i o p p o l i ' s inequality

is i n d e e d the s t a r t i n g point for the r e g u l a r i t y theory.
 217

THEOREM 2 . IGiaquinta ~ Giusti [8]) - L e t u b~ ~ m i n i m i z e r for

A(~l.Then the H~usdorff dimension af t h e singular ~et ~- ~0 i s
strictly less than n-2. In p a r t i c u l a r minimizers are smooth in
dimension n = 2. 1)

N o W the f i r s t n a t u r a l question is w h e t h e r the s i n g u l a r i t i e s are at

most isolated in d i m e n s i o n n : 3, w h e r e f i r s t we c a n h a v e singularities.
The q u e s t i o n is o p e n in that g e n e r a l i t y , but it h a s a p o s i t i v e answer
under the extra assumption t h a t the c o e f f i c i e n t s s p l i t as

(5) A~(x,u) = G~8(x)gij(u)

THEOREM 3 . (Giaqainta - Gi~sti [I@]) - L e t u be a b o u n d e d m i n i -

mizer of
fG:~g[x)gij(~}:D uiO~uJdx

w h e r e G and g a r e s m o o t h s y m m e t r i c definite positive matrices. Then i n

dimension n = 3 the singularities of u are at most isolated and i n
general the singular set of ~ ~as H a u s d o r f f dimension no l a r g e r than
n - 3.~

THEOREM 4 . ( J Q s t - Meier |1811 - Under t h e assumption of t h e o r e m 3

if u is a bounded mi n i m i z e ~ ~ith smooth boundary datum, then singulari-
ties may o c c u r only far from~ t h e b~undary.

We r e c a l l t h a t solut~o.ns of q u a s i l i n e a r elliptic systems may

instead have singularities at the b o u n d a r y [6].
T h e functio~nal (3) (~) (5) that c a n be rewritte~n as

(8) g(u) = ]GaB(x)g~j(u)D uiDsuJ~G dx

where
G(x) : det(G ~Cx)) CGe6(x)) : (G~6(x)) -I

represents in l o c a l cc~ordinate~ the e n e r g y of a m a p b e t w e e n two

Riemannian manifolds ~ : M n ~ M N with metric tensors respectively
GeB" gij" Smooth stationary points are c a l l e d karmonie mapS. We r e f e r
to [2][ 3]~17] for m o r e information.
From~ the g e n e r a l p o i n t of v i e w of d i f f e r e n t i a l geometry, theorems

i) A c t u ~ l l y , under some m o r e r e s t r i c t i v e assumptions, in the g e n e r a l

situation of t h e o r e m 1 minimizers are also s m o o t h in d i m e n s i o n 2,
see [ 7] .
 218

2 and 3 are limited.

n
In fact, while we c a n a l w a y s localize in M , this is in g e n e r a l
not possible in the target manifold M N, e x c e p t that we a s s u m e that it
is c o v e r e d by one chart <or, w o r s e still, that u is c o n t i n u o u s ) . In the
general setting of a m a p f r o m M n into M N, theorems 2 and 3 have been
proved independently by S c h o e n - Uhlenbeck [ 19] [ 201.
At this p o i n t we may resonably ask w h e t h e r the (bounded) mini-
mizers of (6) m a y be r e a l l y singular. In that respect the classical
result by E e l l s - Sampson [4] can be read: if the sectional curvature
of MN i s non-positive then the minimizers (as w e l l as t h e s t a t i o n a r y
points) of (6) are smooth. H i l d e b r a n d t - Kaul - Widman [ 15] in the
case of t a r g e t manifold with positive sectional curvature proved:
i f u(M n) i s c o n t a i n e d i n a g e o d e t i c b a l l Br(q) which i s d i s j o i n t from
t h e c u t l o c u s of i t s c e n t e r and has r a d i u s
(7) R <
2v~
where k i s an upper bound f o r t h e s e c t i o n a l c u r v a t u r e , then the mini-
m i z e r s {and even t h e s t a t i o n a r y p o i n t s ) are smooth.
In c a s e of a m a p from the unit ball BI(0) of R n i n t o the standard
sphere S n of R n + l c o n d i t i o n (7) m e a n s that U(Bl(0)) is s t r i c t l y contain-
ed in a h e m i s p h e r e . Hildebrandt - Kaul - Widman showed that the equator
map u ~': d e f i n e d by u~'~(x) = (T~l,0) is a s t a t i o n a r y point for ~(u).
Then J~ger-Kaul [16] proved that u ¢~ is a minimizer for n > 6,
while it is even unstable for n < 7; m o r e recently Baldes [ I] s h o w e d
that u ~ is stable e v e n for n = 3 if c o n s i d e r e d as a m a p p i n g from
BI(0) into a suitable ellipsoid.
In g e n e r a l we h a v e

THEOREM 5 . ( S c h o e n - U h l e n b e c k [ 2 1 ] , G i a q u i n t a - S o u d e k [ 1 3 ] ) Every
energy m i n i m i z i n g map u from a domain i n some n - d i m e n s i o n a l Riemannian
N
m a n i f o l d i n t o t h e h e m i s p h e r e S+ i s r e g u l a r p r o v i d e d n ~ 6, and i n
g e n e r a l i t s s i n g u l a r s e t has H a u s d o r f f d i m e n s i o n no l a r g e r t h a n n - 7.

References

[I] Baldes,A., Stability and Uniqueness P r o p e r t i e s o f t h e Equator Map

from a Ball into an Ellipsoid, Math. Z. 185 (1984), 505-516.
[2] EELLS,J., Lemaire,L., A report on harmonic maps, Bull. London Math.
Soc. i0 (1878), 1-68.
[3] E E L L S , J . , L e m a i r e , L . , Selected topics in harmonic maps, C B M S
R e g i o n a l C o n f e r e n c e series.
[4] EELLS,J., Sampson,J.H., Harmonic mappings of Riemannian m a n i f o l d s ,
Amer. J. Math. 86 (1964), 109-160.
 2t9

[5] Evans,C.L., Quasico~vexity and p a r t i a l r e g u l a r i t y i n t h e c a l c ~ s

of v a r i a t i o n s , p r e p r i n t 1984.
[61 G i a q u i n t a , M . , A counterexample to the boundary r e g u l a r i t y of
s o l u t i o n s to e l l i p t i c q u a s i l i n e a r s y s t e m s , m a n u s c r i p t a math. 14
(1978), 217-220.
[7] G i a q u i n t a , M . , M u l t i p l e i n t e g r a l s i n the Calculus of Variations
and non l i n e a r e l l i p t i c s y s t e m s , Annals. Math. Studies n 105,
Princeton University Press, 1983.
[8] Giaquinta,M., Giusti,E., On the r e g u l a r i t y of t h e minima of
v a r i a t i o n a l i n t e g r a l s , Acta Math.148 (1982), 31-46.
[9] Giaquinta,M., Giusti,E., Differentiability of minima of non
differentiable functionals, Inventiones Math. 72 (1983), 285-298.
[i0] Giaquinta,M., Giusti,E., The singular set of the minima of
c e r t a i n q u a d r a t i c f u n c t i o n a l s , Ann.Sc.Norm.Sup. Pisa ii (1984),
45-55.
[ ii] Giaquinta,M., Ivert,P.A., Partial regularity for minima of
v a r i a t i o n a l i n t e g r a l s , Arkiv fof Math.
[12] Giaquinta,M., Modica,G., Partial regularity of minimizers of
quasiconuex integrals, Ann. Inst. H. Poincar4, Analyse non
lineaire.
[13] Giaquinta,M., Sou~ek,J., Harmonic maps into a hemisphere, Ann.
Sc. Norm. Sup. Pisa.
[14] Giaquinta,M., Sou~ek,J., Cacciopoli's inequality and Legendre-
Hadamard condition,Math. Ann. 270 (1985), 105-107.
[15] Hildebrandt,S., Kaul,H, Widman,K.O., An existence theorem for
harmonic mappings of Riemannian m a n i f o l d s , Acta Math. 138 (1977),
1-16.
[16] J ~ g e r , W . , K a u l , H . , R o t a t i o n a l l y symmetric harmonic maps from a
b a l l i n t o a s p h e r e and t h e r e g u l a r i t y problem f o r weak s o l u t i o n s
of e l l i p t i c s y s t e m s , J. reine u. angew. Math. 343 (1983), 146-161.
[17] J o s t , J . , Harmonic mappings between Riemannian manifolds, C e n t r e
for Math. Anal., Australian National Univ., vol. 4, 1983.
[18] Jost,J., M e i e r , M . , Boundary r e g u l a r i t y for minima of c e r t a i n
quadratic f u n c t i o n a l s , Math. Ann. 262 (1983), 549-561.
[19] Schoen,R., Uhlenbeck,K., A regularity theory for harmonic maps,
J. Diff. Geom. 17 (1982), 307-335.
[20] Schoen,R., Uhlenbeck,K., Boundary r e g u l a r i t y and miscellaneous
r e s u l t s on harmonic maps, J. Diff. Geom. 18 (1983), 253-268.
[21] Schoen,R., Uhlenbeck,K., Regularity of minimizing harmonic maps
into the sphere, Inventiones math.
[22] Uhlenbeck,K, R e g u l a r i t y for a cla~s of n o n l i n e a r e l l i p t i c
s y s t e m s , Acta Math. 138 (1977), 219-240.
PERIODIC SOLUTIONS OF PARTIAL
DIFFERENTIAL EQUATIONS
WITH HYSTERESIS
P. KREJCi
Mathematical Institute, Czechoslovak Academy of Sciences
115 67 Prague 1, Czechoslovakia

Introduction.

In m e c h a n i c s of p l a s t i c - e l a s t i c bodies or in the t h e o r y of e l e c t r o -
magnetic field in f e r r o m a g n e t i c m e d i a w e are led to t h e c o n s i d e r a t i o n
of h y s t e r e s i s phenomena. T h e r e are v a r i o u s approaches to the m a t h e m a t i -
cal d e s c r i p t i o n of h y s t e r e s i s (cf. [I]). E x i s t e n c e results for P D E ' s
with hysteresis nonlinearities are d u e to V i s i n t i n (see e.g. [5]). We
g i v e h e r e a s u r v e y of r e s u l t s of [23, [3~, [~, w h e r e w e p r o v e the e x i s -
t e n c e of p e r i o d i c solutions to the p r o b l e m s

utt U x x ± F(u) = H(t,x) , u(t,0) = u(t,~) = 0 ([23), (P1)

ut - (F(Ux))x = H(t,x) , u(t,0) = u(t,1) = 0 ( [33 ), (P2)
utt - [F(Ux))x = H(t,x) , Ux(t,0) = Ux(t,z) = 0 ( [43 ), (P3)

where F is the I s h l i n s k i [ hysteresis operator and H is a g i v e n time-

-periodic function with an a r b i t r a r y period ~ > 0 .

I. Ishlinski[ operator (cf. [1], [23).

We f i r s t d e f i n e simple hysteresis operators v ÷ lh(V) , fh (v)

for h > 0 and for p i e c e w i s e monotone continuous inputs v : [o,T~ -~ R 1

as f o l l o w s :

max {lh(V)(to) , v(t) - h}, t E [t0,tl] ,

lh(v)(t) =< if v is n o n d e c r e a s i n g in [t0,t1~
(1.1)
sin {Zh(V ) (to) , v(t) + h}, t E [t0,tl] ,
if v is n o n i n c r e a s i n g in [t0,t1~ ,
 222

0 , if ]v<0)l ~ h
/ h ( V ) (0) = v(O) - h , v(O) > h
v(O) + h , v(O) < - h .

f h ( v ) (t) = v ( t ) - / h ( V ) (t) . (I .2)

For v , w continuous and piecewise monotone we have (see [I],

p. 16)

l/h(V)(t) - /h(w)(t){ ~ max {Iv(s) - w ( s ) I; s £ [0,~ } (1.3)

This property enables us to d e f i n e /h(V) , fh(v) for arbitrary

continuous inputs. Moreover, it follows from (].3) that lh ' fh map
continuously the space C([0,T~) of all continuous functions into it-
self.

Let us introduce the space C , w > 0 , of all continuous w-pe-

w
riodic functions v : RI ~ RI with sup-norm I I" I I . F o r v E C the
W

functions /h(V) , fh(v) are w-periodic for t .~ ~ , hence lh ' fn

can be considered as continuous operators C + C
W

Let further g : [0,-) + E0,~) be a function of class C 2 ( 0 , ~)

satisfying

(i) g is increasing, g(O) = 0 , 0 < g'(0+) < ~

(ii) g(h) ~ a-h ~ for some a > 0 , ~ ~ (0,1)

and for every h ~ 0 .
(1.4)
(iii) Put ~(r) = inf {-g''(h); 0 < h ~ r} . We r e q u i r e the
existence of constants b > 0 , B 6 (0,~] such that
lim inf y(r) r 2-B = b .

(iv) B > I/3 , 3~ < 4B .

For v ~ C we define

(i) F ( v ) (t) = - i fh (v) (t) g'' (h) dh

0
(I .5)

(ii)i L(v)(t') = i £h(v)(t) (g-1)''(h) dh + (g-l)'(0+) v(t) ,

0

where g(g-1(h)) = g-1(g(h)) = h .

Roughly speaking ,the dependence of F on v (L on v) can be

represented by a system of hysteresis loops constituted by parts of the
 223

graph of g (g-1 , respectively) for v nondecreasing and -g (_g-1 •

respectively) for v nonincreasing. The operators F , L have the fol-
lowing properties:

(i) F, L : C ÷ C are continuous

(ii) L = F -I

(iii) IIF(v)11 = g(llvlI)

(iv) g'[max {I IvII, I IwII}l II v-wll ~ llF(v) - F(w)ll
< g' (0+) llv - wll
(v) For v absolutely continuous F(v) , L(v) are absolutely
continuous and the inequalities

I(F(v))'(t)I <__ g ' ( o + ) Iv'(t) l ,

(I .6)
l(L(v))'(t)l <= (g-~)'(llvll) Iv'(t) l
are satisfied almost everywhere.

(vi) For v' absolutely continuous we have

2~
I o , Iv1 ft 1
~ 0
2u~ a~

(L(v))'V'' <= - ~
0
where ~(r) = inf {(g-1)''(h); 0 < h < r} °

For functions of two variables u(t,x), t 6 RI , x 6 I , such that

u(.,x) E C for each x ~ I we define

F(u)(t,x) = F(u(.,x)) (t) , L(u)(t,x) = L(u(.,x)) (t)

2. Existence results.

We introduce the Banach spaces or time-periodic functions u(t,x)

u : R1 x [0,£] ÷ RI , £ > 0 , u(t + ~, x) = u(t,x) , ~ > 0 :
c~(E0,z3): the space of all continuous functions with norm
IHulII = max { l u ( t , x ) I; t RI, x ~ [0,£]} ;

L P (0,£), I <= p <= ~: the space of all measurable functions such that

,Ulp = (] I ' u ( t ' x ) lp d x dt] I/p < ~ for p < ~ ,

0 0
lul~ = s u p ess { l u ( t , x ) I; t RI x ~ E0,1] } <
 224

with norm l'Ip ;

2 I/2
L2(O,1; Llto) = {U 6 Llo(o,1); lu[2,1 = (I(I[U(t,x) l dt) dx) < ~}
0 0
w i t h norm I " 1 2,1

Theorem. (2.1)
(i) Let H & L~(0,~) be given such that Htt ~ L~/2(0,~) and
let (1.4)(i) - (iii) hold. Then there exists at least one
solution U ~ C ([0,~]) , Utt - Uxx E L2(0,~)
u t ~ L3(0,~) to the problem (PI) such that (PI) is satis-
fied almost everywhere.
9
(ii) Let H ~ C -~-([0'17) be given such that H t e L~(0,1) and
let (1.4)(i) - (iii) hold. Then there exists at least one
classical solution u E C (E0,13) , u t, u x, (F(Ux))x
, L2 3
C ([0,13) utt e ( 0 , I ) , Utx ~ L (0,I) , Uxx ~ L~(0,1)
to the p r o b l e m (P2)

(iii) Let H E L 1(0 ~) be given such that H E L3/2(0,~)N

~ X OJ

L2(O,rf; L1) , [ IH(t,x) dx dt = 0 , and let (1.4)(i) - (iv)

0 0
hold. Then there exists at least one solution u e C ([0,~3) ,
Ux & C ([O,rt~) , Utt , (F(Ux))x ~ Ll(O,lt) , Utx ~ L3(O,rr) to
the p r o b l e m (P3) such that (P3) is s a t i s f i e d almost every-
where.

Sketch of the proofs.

(i) We use the classical Galerkin method. We denote

Wjk(t,x) = ej(t) sin kx , j integer, k natural, (2.2)

where ej(t) = sin 2 ~ t for j > 0 and cos 2~j~ t for j =< 0 .

For m > I put

m m
u(t,x) = ~ ~lUjk Wjk(t,x ) , (2.3)
m j=-mk

where the coefficients Ujk are solutions of the algebraic system

2~ ~ 2~
I I(~tt- ~xx ± F(u))wjk dx dt = I IH Wjk dx dt , (2.4)
~0 m ~0
 225

j = -m,...,m , k = 1,...,m .

3
Multiplying (2.4) by [2~j] u kJ
. , summing over j and k and
using (1.6)(vi) we g e t

Y (llluIII)lUt 123 ~ c o n s t .
m m
Similarly we h a v e [utt U x x I 2 ~ c o m e t . (I + IIIF(u)III) . Classical em-
m m m
bedding theorems and (1.4), ( 1 . 6 ) ( i i i ) y i e l d IIIulll
m
c o n s t . (lutt - Uxxl2 + lutl3) ~ c o n s t . (lllulll I - ~ / 2 + llIulll~ + I), hence
m m m m m
ll]ulll, lutl 3 , lutt - U x x l 2 ~ const.
m m m m

These estimates imply the solvability of (2.4). On the o t h e r hand

there exists a subsequence {u} of {u} such that utt - U x x +
n m n n
utt - Uxx in ~2(0,~) weak, ut ÷ ut in L3(0~ ,~) w e a k and u ÷ u
n n
in C ([0,~) strong. Thus, we c a n p a s s to the l i m i t in (2.4) and the
proof is c o m p l e t e .

(ii) We r e p l a c e the p r o b l e m (P2) by the following system of o r d i n a r y

differential equations (space discretization):

v[]- n ( F ( n ( v j + I - vj)) - F(n(vj - v j _ 1 ) ) ) = hj , (2.5)

v0=Vn=0,
(j+1)/n
t

where hi(t) = n ~ H(t,x) dx , j = I ..... n-1 .

J
j/n
Using (1.6)(iv), (vi) w e d e r i v e apriori estimates for v. independent
]
of n which ensure the e x i s t e n c e of p e r i o d i c s o l u t i o n s of (2.5). We
further put
I
u(t,x) = vj(t) + (nx - j)(v~+1(t)~ - vj(t)) + ~ L(Vs+1(t) - vs(t))
J J
n
÷ (nx - j ) 2 ( v j + 2 ( t ) - 2 v j + 1 ( t ) + v j ( t ) ) ~--

t E RI , x ~ [j/n, (j+1)/n~ , j = 0,1 ..... n-1 , V n + I = - Vn_ I

A straigtforward computation and the c o m p a c t n e s s argument show

that there exists a subsequence of u which converges in s u i t a b l e to-
n
pologies to a s o l u t i o n u of (P2).

(iii) We first solve the auxiliary problem

L(v t - 9)t - Vxx = ~x ' v(t~0) = v(t,~) = 0

 226

where ~(x) = !~ H(t,<) dt d~: , x 6 [0,~] ,

0 0
t
~(t,x) = IH(T,X) dT - t~ IH(T,X) dT , t ~ R1 , x E [0,~
0 0
We use again the G a l e r k i n approximation scheme of the type (2.2), (2.3),
(2.4). In an a n a l o g o u s way we d e r i v e the f o l l o w i n g estimates:
2
~(l]IVt - ~lll)IVttl3 < const.
m m

IVxtl~ =
< (g-l)' (lllvt - ~l]l)IVtt 122 + const, lvtl 3 "
m m m m
Following (1.4) there exists ~ > 0 such that I/~(r) <

c o n s t . ( r 3-I/~-6 + I) for every r > 0 . The space {u 6 LI(0,~); ut 6

L3(0~ ,~) , Ux E L2(0,~)}~ is c o m p a c t l y embedded into C ([0,~]) , hence

IIIVtm - @Ill < eonst.(Ivttl

3 m + IVtxl2 + I) _<o const.(IIlvtm - 91111-6/2 + 1).
Similarly as above we get
lllVtlll, IVttI3 , IVtxl2 , IVxxI2 <= eonst.,
m m m m
so that we can repeat the a r g u m e n t of (i).

The s o l u t i o n u of (P3) is then given by the f o r m u l a

t 2~
u(t,x) = ](Vx + ~)(~,x) dT - t I (V x + ~)(~,X) dT+
co X o3

+ ~L(v t - ~)(~,~) d~ + const., t ~ RI , x E [0,~3

0

R e f e r e n c e s

] K R A S N O S E L S K I ~ , M. A. and POKROVSKI~, A. V.: Systems with hystere-

sis. Nauka, Moscow, 1983 (Russian).
~] KREJ~f, P.: Hysteresis and periodic solutions of nonlinear wave
equations. To appear.
[3] KREJ~I, P.: Periodic solutions to a parabolic equation with hyste-
resis. To appear.
~3 KREJ~, P.: Periodic solutions to a quasilinear wave equation with
hysteresis. To appear.
[~ VISINTIN, A.: On the Preisach model for hysteresis. Nonlin. Anal.,
Theory, Meth. & Appl. 8 (1984), pp. 977-996.
STABILITY AND BIFURCATION PROBLEMS
FOR REACTION-DIFFUSION SYSTEMS
WITH UNILATERAL CONDI ONS
M. KUCERA
Mathematical Institute, Czechoslovak Academy of Sciences
115 67 Prague 1, Czechoslovakia

Let us consider a reaction-diffusion system of the type

u t = dsu + f(u,v)
on <0,~) x ~ (RD)
vt = Av + g(u ¥)

where f , g are real smooth functions on R2 , d is a nonnegative

parameter (diffusion coefficient), [~ is a bounded domain in Rn . Sup-
pose that u, v > 0 is an isolated solution of f(u,v) = g(u,v) = 0 .
Thus, u , ~ is a stationary spatially homogeneous solution of (RD)
with Neumann boundary conditions and also with the boundary conditions

u = U on (0,~) × FD , 8 nu _ 0 on (0,~) x "~ N ' (BCl)

v : v on (0,~) x FD , 8n
v = 0 on (0,~) × v• N ' (BC 2)

8f
where rD U FN : 6~ (the boundary of ~ ). Set b11 = 7 ~ ( u , v ) , b12 =

~-~(u,v) , b21 , b22 , b21,b22

bll > 0, b21 > 0, b12 < 0, b22 < 0, bll + b22 < 0, det B > 0. (B)
Then there exists the greatest bifurcation point dO of (RD), (BC) at
which spatially nonhomogeneous stationary solutions bifurcate from the
branch of trivial solutions { [d,u,v]; d 6 R} ; the solution u , ~ is
stable for any d > dO and unstable for any 0 < d < dO . (All eigen-
values of the corresponding linearized problem have negative real parts
for d > dO and there exists a positive eigenvalue for d < dO .) F o r
the case of Neumann conditions and n = I (i.e. ~ = (0,1) ) see e.g.
[8], for the general case see [2]. Notice that (B) is fulfilled in mo-
dels connected with population dynamics, chemistry, morphogenesis etc.
In these cases u represents a prey or an activator, v represents a
predator or an inhibitor. The existence of stationary spatiall~ nonhomo-
geneous solutions explains the occurence of the so called striking pat-
terns.
 228

The aim of this lecture is to present some results obtained by the

author together with P. Drlbek, M. Mikov~ a n d J. N e u s t u p a , showing how
this situation can change by introducing unilateral conditions given by
a cone in a suitable Hilbert space. One of the simplest examples are uni-
lateral boundary conditions

v = ~ on (0,~) x PD , v > ~ ~v
(1)
(v - 8v
-v)~-~ = 0 on (0,~) x r0 , 8a nv = 0 on (0,~) x (PN~P 0)

where P0 is a g i v e n subset of PN " Roughly speaking, the spatially

homogeneous solution u , v becomes unstable even for some d > do and
the greatest bifurcation point shifts to the right of do if c l a s s i c a l
conditions are replaced by unilateral ones for v ; the greatest bifurca-
tion point shifts to the left if u n i l a t e r a l conditions are introduced
for u .
In what follows, we shall suppose ~ = ~ = 0 without loss of gene-
rality.

I. Abstract formulation.

Let K be an arbitrary closed convex cone in the space V =

I
{u Q W 2 (~) ; u = 0 on PD} with its vertex at the origin.• Consider the
problem

I[ut(t,x)~(X) + d v u (t,x) v~ (x) - f I u ( t , x ) , v ( t , x ) l ~ ( x ) ] = 0

~2
v(t,.) £ K ,

I{vt(t,x) [~(x) - v(t,x)] + vv(t,x) v[~(x) - v(t,x)] - (2)

~2
- g(u(t,x),v(t,x)) [~(x) - v(t,x)]} dx a 0
for all ¢ e V , ~ 6 K , a.a. t >= 0 ,

n
where Vu-v~ = i=}Jluxi~xi . By a solution we can understand a couple

u, v ~ L2(0,T;V) such that ut, vt ~ L2(0,T;V) o

Notice that the choice K = {v e V; v _> 0 on P0} corresponds to

the problem (RD), (BC 1), (I). ((2) is o b t a i n e d by multiplying the equa-
tions in (RD) by test functions, integrating by parts and using (BC I) ,
(1) .)
In general, we can define a solution of (RD) with the boundary con-
ditions (BC 1 ) a n d unilateral conditions for v given by K as a couple
u , v satisfying (2). The corresponding linearization reads
 229

i [ut(t,x)#(x) + dVu(t,x)V#(x) - IbllU(t,x) + b12v(t,x))#(x) ] dx

= 0 , v(t,.) 6 K ,

I { v t ( t , x ) (~(x) - v(t,x)) + Vv(t,x).V[~(x) - v(t,x)] (3)

- (b21u(t,x) + b22v(t,x)) {*(x) - v(t,x))} d x => 0

for all ¢ ~ V , ~ 6 K , a.a. t > 0 .

Analogously, we can consider (RD) with (BC2) and unilateral conditions

for u , i.e.

u ~ K ,
r
J [ u t ( ~ - u) + dVu'V(~ - u) - f ( u , v ) (9 - u)] dx > 0
(4")
for all ~ 6_ K , a.a. t > 0 ,

I[vt~ + Vv-V~ - g(u,v)~] dx = 0 for all ~ ~ V , a.a. t > 0 .

2. Destabilization.

I
EXAMPLE 1. Consider FD = ~ (i.e. V = W2 ) , K = {v C V; v ~ 0
on ~I . Then (2) corresponds to a free-boundary problem

u t = dAu + f(u,v) on <0,~) x ~ ,

v t = Av + g(u,v) on Q+ ,

v = 0, g(u,0) ~ 0 on <0, ~) x ~ - Q+ , (5)

8u Ov
Ux. , Vx. are continuous, ~n 3n 0 on (0,~) × ~ ,
1 l

where the domain Q+ = {[t,x] ~ <0,~) × ~; v(t,x) > 0} is u n k n o w n . The

couple u(t,x) = exp(bllt)- ~ , v(t,x) = 0 satisfies the linearization
of (5) (i.e. also (3)) classically for any ~ < 0 . It follows that the
trivial solution of (3) is u n s t a b l e for any d , even for d > dO , and
even with respect to spatially homogeneous perturbations. Notice that
spatially homogeneous solutions of (3) (in o u r special case) are solu-
tions of the inequality
U(t) ~ K
c (6)
< Ut(t) - BU(t), ~ - U(t) > ~ 0 for all ? 6 Kc , a.a. t ~ 0 ,

where U = [u,v] , Kc = {[~,~] ~ R2; ~ ~ O~ , <.,.> is t h e scalar

product i n R 2. It is n o t difficult to describe all trajectories of (6)
and characterize also some spatially homogeneous solution of (2) under
the assumption (B). A s a consequence it is p o s s i b l e to prove also the
 230

instability for (2) for any d > 0 (see [7]).

Notice that the eigenvalue problem determining the stability of the

trivial solution of (RD), (BC) can be written in the vector form

D(d)AU + BU = IU (7)

(with the boundary conditions (BC)), where U = [u,v] , D(d) = (~i~)

AU = [Au,Av] . The eigenvalue problem with unilateral conditions corres-
ponding to (3) reads

I [ d v u . V~ - (bllu + b12v - /u)~] dx = 0 for all V

v ~ ~ , (8)

I[vv-v(~ - v) - (b21u + b22v lv) (~ - v)] dx o for all ¢ e K.

Denote by E(d,h) the set of all solution of (7), (BC) (for g i v e n d ,

~ R ). N o t i c e that E(do,O) # {0} because do is a bifurcation point
of (RD), (BC). Further, we shall suppose that

there exists a completely continuous operator ~ : V ÷ V

(a p e n a l t y operator) satisfying < 8 u - ~v, u - v > ~ 0 ,
~(tu) = t~u for all u, v e V , t ~ 0 , ~u = 0 for all (P)
u e K , < 3v,v > > 0 for all v ~ K , < Bv,u > < 0 for
all v ~ K , u 6 K ° ,

where <.,.> is t h e inner product in V , K° and 3K is the interior

and the boundary of K . This assumption is fulfilled in examples. For
the cone K mentioned in Section 1 we can consider the penalty opera-
tor defined by

<~v,~> = - [ v- ~ dx for all v, ~ ~ V ,

F0

where v- denotes the negative part of v .

THEOREM I. Let (P) , (B) hold and E ( d 0 , 0 ) (~ V x K ° # ~ . Then for

any d 6 ( d o , d 1) (with some d l > do ) there exists a solution o]" (3)
of the type U(t) = exp (l t ) . u a with Id > 0 , Ud C ~K .

THEOREM 2. Let (B) hold and E(do,0) ~ V x K° # ~ , dim E(d0,0) = I,

meas F~ > 0 . Then there exists a bifurcation point dI > do of (2) at
which spatially nonhomogeneous stationary solutions bifurcate from
{[d,O,O]; d ~ R} .

"P r o o f of Theorem I is based on a modification of the method deve-

loped in [5]. We shall explain main ideas only (more precisely see [3]
 231

cf. also [2]). It is s u f f i c i e n t to s h o w t h a t for a n y d 6 (do,d I) the-

I
re e x i s t s a positive eigenvalue Id of (8) w i t h the c o r r e s p o n d i n g ei-
I I I-
genvector U d = [Ud,Vd] 6 V x 8K . S u p p o s e t h a t d i m E ( d 0 , 0 ) = l . (The ge-
neral case c a n be r e d u c e d to t h i s situation - see [3].) Choose a fixed
d > dO and consider the system with the p e n a l t y

I[dVu'V~ - (bllU + b 1 2 v - lu)¢] dx = 0 for all ¢ ~ V ,

(9)
f[~v'V~ - (b21u + b 2 2 v - Iv)~] dx + s < B v ~> = 0 for all ~ ~ V .

It is e q u i v a l e n t to (7), (BC) for e = 0 and its eigenvalues a n d ei-

genvectors approximate those of (8) for ~ + + ~ . (The last a s s e r t i o n
can be p r o v e d by standard penalty method technique.) We shall consider
only solutions of (9) satisfying the n o r m condition

IluII2 ( = IIull2 + Ilvt!2 ) - 1 +~ (lO)

Set Ca = ~ ' ~ R x''V x V x R; Ilull # 0 ,(9), (10) is fulfilledT

(the c l o s u r e in R x V x V x R ). T h e m a i n idea is to s h o w t h a t the
greatest eigenvalue Id of (7), (BC) can be joined with an e i g e n v a l u e
~dI of (8) by a c o n n e c t e d (in R × V x V x R ) subset C+
d of Cd and
to p r o v e ldI > 0 on the b a s i s of the properties of t h i s b r a n c h C d+
(for any d e (d0,d I) with some d I > d O ). The e x i s t e n c e of a g l o b a l
+
continuum Cd C Cd of s o l u t i o n of (9), (10) s t a r t i n g at lid,0,0 ] in
the d i r e c t i o n Ud = [Ud,Vd] E E(d,l d) ~ V x (-K°) follows from a slight
generalization of a D a n c e r ' s bifurcation result [I]. (Setting x =
[U,~] , (9), (10) can be w r i t t e n as the u s u a l bifurcation equation
x - L(~)x + N(~x) = 0 in the s p a c e X = V x V x R with compact linear
operators L(~) depending continuously on ~ 6 R and a small compact
perturbation N ; "starting in the d i r e c t i o n Ud m e a n s t h a t for any
+
6 > 0 there is [~ U £] ~ C d with { IU/I IUI I - Udl I < $ in a n y n e i g h -
bourhood of [ld,0,0] .) An e l e m e n t a r y investigation of s o l u t i o n s of
(9), (10) yields that in a s m a l l neighbourhood of [Id,0,0] can be o n l y
solutions [I U s] of (9), (10) satisfying I > Id and that for all
solutions of (9), (10) different from lid,O,0] the following implica-
tions hold: i > ld ~ v ¢ 8K ; v ~ K =~--->~ # Id " T h i s together with
+
the fact that Cd s t a r t s in the d i r e c t i o n Ud ~ V x K and w i t h the
+
connectedness of C+
d implies ~ > id , v ~ K for any [I,U,~] ~ Cd ,
+
U = [u,v] . It f o l l o w s that Cd cannot intersect an a n a l o g o u s branch
Cd of s o l u t i o n of (9) (10) starting in the d i r e c t i o n - Ud G V x K ° .

Dancer's result (see []], Theorem 2) states that in t h i s case C+

d is
 232

unbounded. It f o l l o w s that there exists a sequence {[In,Un,en ] } C C+

d I
with s ÷ + ~ . The penalty method technique gives ~'n --~ l I ' U n + Ud'
n I I
where ld and Ud is an e i g e n v a l u e and the corresponding eigenvector
of (8), U I ~ V x ~K . If ~ dI > 0 was not true for all d ~ ( d 0 , d I)
with some d I > dO then we would obtain I dI ÷ 0 for some d n + do+
n
because id ÷ 0- , I dI => id . We could suppose U dI ÷ U ~ 8K and this
n n
would contradict the assumptions dim E(d0,O) = I , E(d0,0) N V x K°
. (Any solution U of (8) w i t h d = do , I = 0 lies in E(do,0)
under the last assumption.)

P r o o f of Theorem 2 is b a s e d on the same method as that of Theorem

1. T h e greatest bifurcation point dO of the stationary problem corres-
ponding to (RD), (BC) can be joined (raughly speaking) with a bifurca-
tion point dI > do of (2) by a branch of solutions of the correspon-
ding penalty equation (with the variable d instead of k ). S e e [4].

REMARK I. If meas FD = 0 in Theorem 2 (the c a s e of Neumann conditi-

ons) then the bifurcation point dI can coincide with infinity in a
certain sense (see [4]).

REMARK 2. If K = {v C- V; v > 0 on FO} then E(do,0)C] V x K° ~

holds if t h e r e exists U = [u,v] C E(d0,0) with v > 6 on F0 ( 6 >
0 ).

3. Stabilization.

I.
EXAMPLE 2. Consider the cone K = {u £ V; u ~ 0 on 9} with V = W2
again. Then spatially constant solutions of the linearization of (4)
are solutions of (6) w i t h Kc = {@ = [%,~] C R2; % ~ 0} . If B has
a pair of complex eigenvalues and b12 , b22 < 0 then any solution of
(6) (coinciding with the solution of U t = BU as long as it is in
K° × V ) touches the line { [0,v] ; v ~ 0} after some time to and
then coincides with the solution of the type u(t) = 0 , v(t) =
exp (b22(t - to))" ~ . It follows that the trivial solution of the line-
arization of (4) is s t a b l e with respect to s p a t i a l l y homogeneous per-
turbations even if t h e trivial solution of U t = BU is u n s t a b l e (more
precisely see [7]). Of course, in this way we cannot obtain any infor-
mation about the stability with respect to nonhonogeneous perturbations.

THEOREM 3. Let (B) hold and let E(d 0) A K x V = {0}, meas FD > O.
O'
 233

Then there is no bifurcation point of (4) at which stationary spatially

nonhomogeneous solutions bifurcate from {[d,O,Ol; d e R~ in
(d O - ~, +~) (with sbme 6 > 0 ).

P r o o f . Introduce the inner product <., .> and the operator A

in V by

<u,~> = ~VuV~ dx <Au,~ > = u ~ dx for all u, ~ C V

J

The linearization of the stationary problem corresponding to (4) can be

written as

u 6 K ,

<du - b11Au - b12Av, ~ - u> ~ 0 for all ~ ~ K ,

(11)
v - b21Au - b22Av = 0 .

Calculating v from the second equation in (11) and substituting to

the first inequality we obtain the inequality of the type

U ~ K ,

<du - Tu, ~ - u> ~ 0 for all ~ & K

(12)

with a compact linear symmetric operator T in V . It follows that

any bifurcation point dI of our unilateral stationary problem is s i -
multaneously an eigenvalue of the inequality (12) and therefore dI
max <Tu,u> . Further, the greatest bifurcation point d of
llull=l, ue~ 0
(RD), (BC) is s i m u l t a n e o u s l y the greatest eigenvalue of T , i.e. dO =
max <Tu,u> and any u G V realizing this maximum is a n e l -
llutt=t, u e v
genvector of T corresponding to dO . This together with the assump-
tion E(d0) ~ K x V ={0} implies max <Tu,u> < dO . For the
Ilufl=l, u~
details see [6] w h e r e a more general case is considered.

REMARK 3. Let K = {u ~ V; u ~ 0 on FO} . Then E(do,O) ~ K x V =

{0} is fulfilled if there exists U = [u,v] ~ E(d0,0) such that u
changes its sign on F0 .

4. Final remarks.

It is p o s s i b l e to consider also more general inequalities

 234

I ut( ¢ - U) + d v u - V ( ¢ - u) - f ( u , V ) ( 9 - u) dx

+ ¢1 (~) - ~I (u) ~ 0 ,
(13)
IVt( ~ - V) + VV-V (@ - v) - g(u,v) (9 - v) dx

+ ¢2(~) - ¢2(V) ~ 0 for all ¢, 9 6 V , a.a. t ~ 0 ,

where ¢I ' ¢2 are convex proper functionals on V . More general uni-

lateral conditions are included in this formulation. An analogy of T h e o -
rem 3 for ( 1 3 ) w i t h ~2 = 0 is c o n t a i n e d in [63, a d e s t a b i l i z i n g effect
of such unilateral conditions (for ¢I = 0) w i l l be the subject of a
forthcomming paper.

R e f e r e n c e s

[I] DANCER, E. N.: On the structure of solutions of non-linear eigen-

value problems. Ind. Univ. Math. J. 23 (1974), 1069-1076.
[22 DP~BEK, P. and K U ~ E R A , M . : Eigenvalues of inequalities of reaction
-diffusion type and destabilizing effect of unilateral conditions.
36(111), 1 9 8 6 , C z e c h o s l o v a k Math. J. 36 (111), 1986, 116-130.
[3] DR~BEK, Reaction-diffusion systems: Destabilizing
P. a n d K U ~ E R A , M . :
effect of unilateral conditions. To appear.
E4] DR~BEK, P., KU~ERA, M. and M f K O V ~ , M.: Bifurcation points of reac-
tion-diffusion systems with unilateral conditions.
Czechoslovak Math. J. 35 (110), 1985, 639-660.
[5~ K U ~ E R A , M.: Bifurcation points of variational inequalities, Czecho-
s l o v a k Math. J. 32 (107), 1982, 208-226.
[6J KU~ERA, M.: Bifurcation points of inequalities of reaction-diffu-
sion type. To appear.
~7] KU~ERA, M. and N E U S T U P A , J.: Destabilizing effect of unilateral
conditions in reaction-diffusion systems. To a p p e a r in C o m m e n t .
Math. Univ. Carol. 27 (1986), 171-187.
[8] MIMURA, Spatial patterns for an interaction-
M. a n d N I S H I U R A , Y.:
-diffusion equations in morphoaenesis. J. Math. B i o l o g y 7 243-263,
(1979).
 BOUNDARY INTEGRAL EQUATIONS
OF ELASTICITY IN DOMAINS
WITH PIECEWISE SMOOTH BOUNDARIES
V. G. MAZ'YA
Leningrad University, Petrodvoretz. Math. Mech. Faculty
Bibliotechnaya pl. 2, Leningrad, USSR

O. Introduction

In t h e a u t h o r ' s papers [1-~ a method for investigation of b o u n d a r y

integral equations arising in p r o b l e m s of m e c h a n i c s of c o n t i n u u m in do-
mains with pieeewise smooth boundaries was proposed. Traditionally, equa-
tions of the p o t e n t i a l theory are studied directly by m e t h o d s of the
Fredholm and singular integral operator theories. In the case of a non-
smooth boundary this w a y leads to d i f f i c u l t i e s that have not b e e n over-
come until now. In a s e n s e our approach is o p p o s i t e to the traditional
one. It is b a s e d on t h e w e l l - k n o w n fact that solutions of i n t e g r a l equa-
tions can be e x p r e s s e d in terms of s o l u t i o n s of some exterior and in-
terior boundary value problems. These are studied with the h e l p of t h e
theory developed in [4 - 8] and, as a result, theorems on s o l v a b i l i t y of
equations of the p o t e n t i a l theory are o b t a i n e d . For these equations we
can get d i f f e r e n t i a b i l i t y properties and asymptotics of s o l u t i o n s near
singularities of the b o u n d a r y using the same approach ~, 9 - l j. In
[~ our method of c o n s t r u c t i o n of the p o t e n t i a l theory was illustrated
by the example of t h r e e boundary value problems of l i n e a r isotropic
elasticity, namely, the first, the second and mixed, as w e l l as of the
same problems for the L a p l a c e operator under the h y p o t h e s i s that there
exist a finite number of n o n - i n t e r s e c t i n g smooth edges on the b o u n d a r y .
In the p r e s e n t lecture we s t u d y the first two b o u n d a r y value problems
for the L a m ~ system in d o m a i n s with boundary singularities of t h e type
of edges, conic points and polyhedral angles. New results on t h e h a r m o -
nic potential theory are also reported.

1. Domains and function spaces

Let G (i) be a d o m a i n in R3 with compact closure and with the

boundary S , Q(e) = R 3 \G(i) . We suppose that S is t h e u n i o n of a
finite number of "faces" {F) , "edges" {E} and "vertices" {Q) a n d that
openings of all angles are non zero. Confining ourselves only to the
 236

above visual description we r e f e r the r e a d e r to E 8 3 for the e x a c t de-

finition of the c l a s s of d o m a i n s to be c o n s i d e r e d . In any case it con-
tains domains with polyhedral boundaries. We p l a c e the o r i g i n into G (i)-

Let {U} be a s u f f i c i e n t l y small finite covering of G (i) by o p e n

sets s a t i s f y i n g : a) U contains not m o r e than a s i n g l e v e r t e x Q , b)
if U does not c o n t a i n vertices then U intersects not m o r e than a
single edge E . W i t h any v e r t e x Q and any edge E we a s s o c i a t e real
numbers 8Q and YE ' respectively.

By m e a n s of the p a r t i t i o n of u n i t y subordinate to the c o v e r i n g {U}

we d e f i n e the space C~,y
1'S(G(i))"
" with 0 < e < I , B = {60} , Y = {¥E}"

If U intersects no s i n g u l a r i t i e s of S then the c l ' e ( G ( i ) ) - n o r m of

B,y
a function with support in U is e q u i v a l e n t to the n o r m in the u s u a l
HOlder space C I'~ . In the case U N E ~ ~ , U ~ {Q} = ~ and
supp u C U we h a v e

]lul
Ici: (s(i))
sup
x6S(i)
lu x l +
YE
Ir E ( x ) Y E Vu(x) - r E(y) Vu(y) I
sup
x,y 6 G (i) Ix - yI~
where rE(x) is the d i s t a n c e f r o m the p o i n t x to E . If U contains
the v e r t e x Q and supp u C U then

[[Ul
Kg: (G(i))
sup
x 6 G (i)
lu(x)l +

~Q BQ
IPQ(x) ~ _ rE(x)YEvu(x) - pQ(y) R _ r E (x) Y E v u (y) I
sup {E:Q eE} {E:Q E E }
x,y £ G (i) ix - yl e

where pQ(x) = Ix - QI

Replacing here G (i) by G (e) and sup [u(x) i by

x 6 G (i)
sup (I + ixi) lu(x) I we obtain the d e f i n i t i o n of the space
x E G (e)
C8,
¥1'~(G(e)) . By CI'~(S)B,y we denote the s p a c e of t r a c e s on U F of

functions in CI'~(G(i)) or C1,~ (G (e))

B,Y B,Y
Let us i n t r o d u c e another space C0'e(S) of f u n c t i o n s on UF . If
8,¥
U ~ E ~ ~ , U N {Q} = ~ and supp u C U t h e n

IC ~/ sup rE (x) ¥E-~ ]u(x) I +

 237

TE 7E
IrE(x) u(x) - r E(y) u(y) I
+ sup
x,y CUF Ix - yt ~'

If U contains the vertex Q and supp u C U then

BQ YE I
sup ~Q<X) [] rE(x) E -- lu(x) l +
] ]ul I c O ' a ( S ) x e UF {E:Q { E ) {E:Q c-E} rE
8,Y
~Q
IpQ(X) 8Q ~ _ rE(x)YEu(x) - pQ(y) _ rE(y)TEu(y) 1
{E:Q 6 E } {E:Q e E )
sup
x,y~ uF Ix - yl ~

If U intersects no singularities of the boundary and supp u ~ U then

the norm in CB,
0'~(S)
7 is e q u i v a l e n t to the norm in C0'~(S) .

2. Boundary value problems_of elasticity

We consider interior a n d exterior Diriohlet problems:

pAu + (I + ~) V div u = 0 in G (i) , u = g on <J F (D (i))

pAu + (I + p) V div u = 0 in G (e) u = g on UF

and
u(x) : 0((~ + Ixl) -~)
interior a n d exterior Neumann problems:
I (D e))

~AU + (I + p) V div u = 0 in G (i) , Tu : h on <)F , (N (i))

~AU + (I + ~) V div u = 0 in G (e) , Tu = h on <) F , ] (e)

(N )
u(x) = 0((i + I~I) -~) J
where T = T(ax,n) is the m a t r i x with elements Tkj (3x,n) = t~6J a / a n +
in k 8 / S x j + t~nj 8 / 8 x k and n = (n I ,n2,n 3) is an o u t e r normal ~to G (i)

We introduce a collection of reals {6E} which appears in t h e sta-

tement of the next lemma and will be u s e d in t h e sequel• Let ~(z)
(0,2~) be the angle between the tangent half-planes to S at the
point z 6 E from the side of G {i) • We put wE =
inf (7 + In - #(z)I) . Let 6E be the root of the equation
z6E
sin (mE6) + ~ sin mE = 0 with the least positive real part; 6E is
real and decreases as mE increases, I/2 < ~E < I .

We formulate a theorem on solvability of all above mentioned boun-

dary value problems, which is sufficient to justify the boundary inte-
gral equations method.

THEOREM 1. Let {TE} and {SQ} satisfy

 238

0 < I - YE + a < 6 E for all E , (I)

[BQ + {E:Q~ E E } YE - ~ - 3/2[ < EQ J'cr a!~ Q , (2)

where {~Q} is a c o l l e c t i o n o f posi~<Zve number~ in (0,1) which depend

on the tangent cone to S with the vertex Q ~)

Then (i) •
(D (1)) and (D (e)) are ~niqueiy solvable in 1 , aY( G ( i ) )
CB,
and C~'a(G (e)) for all g % clB'a(S) ; (ii) (N (e)) is uniqv~ely ~;ol-
b'Y 1 a 'Y '
vable in C~' (G (e)) fop a~ h c Ct'a(S)~
Y ; (iii) (N (1)) i,~ s o l v a b l e
in CIB:~(G(i~) for all h C CB,y (S) w'tth zero prsnetpal Vector and
principal moment. Its solution is unique up to a r i g i d displacement.

For (D (i)) and (D (e)) this result is c o n t a i n e d in T h e o r e m 11.5

[ 8 ~. The proof for (N (i)) and (N (e)) requires minor t e c h n i c a l
modifications.

3. Solution of (D (i)) and (D (e)) by a s i m p l e ~ a y e r potential

Let Vo be the e l a s t i c simple laye~ ~ p o t e n t i a l with the density o .

T H E O R E M 2. Let {yE} and {6Q} satisfy (I) a n d {2). Then the o p e r a -

tors C60'a(S)
Y 9 ~ ÷ (Vo) IG(e ) 6 CI'a(G
,7 (e)) and CB,Y
0'a(S) 9 a + (Vu) IG(i )

E C I~', ¥~ ( G (i)) are bounded. There exists a bounded inverse V -I

I'~(S) ÷ C0'~(S)
CB,~ B,Y
The first part of the t h e o r e m can be c hecked directly. The second
part follows from T h e o r e m 2 and the identity 2o = Tu (i) - Tu (e) where
(i) (e) (i) (e)
u and u are r e s t r i c t i o n s of Vo to G and G , respec-
tively.

4. S o l u t i o n of (D(i)) • (D(e)) , (N(i)) , (N(e)) BY a double layer

potential

Let W~ be the d o u b l e layer potential with the density T , i.e.

(W~) (x) = ~I I(T(Sy,ny) F(x,y)]*~(y) day ,

S

*)
The numbers ~Q can be d e f i n e d by some spectral boundary value
problems in spherical domains (cf. [ 8 2) but we shall not use it
in what follows. For the problems (D (i)) and (D (e)) it was shown
in [ I ~ that eQ > I/2 . The v a l i d i t y of the last i n e q u a l i t y for
(N (i)) , (N (e)) remains an open question.
 239

where * denotes the a d j o i n t operator and F is the K e l v i n - S o m i g l i a n a

tensor.

If u (i) : W~ then Y satisfies the s y s t e m of s i n g u l a r integral

equations
(- 1 + W)~ : g on <J F . (3)

The solution of (D (e)) c a n be e x p r e s s e d as the sum (WT) (x) +

F(x,0)a + rot F(x,0)b where a , b are unknown constant vectors. Then
the triplet (Y,a,b) satisfies the system

(I + W)~ + [(-,0)a + rot F(.,0)b = g (4)

Representing the solutions of (N(i)) and (N(e)) as W~ we a r r i v e

at the systems
(I + W*)T = h , (5)

(- I + W*)~ = h . (6)

THEOREM 3. Let {TE} and {~Q} satisfy (I) and (2). Then (i) the ope-
rators W and W* are bounded in C41
B
i$ (S) and
o,~ (S)
CB (ii) if
g ~ CB,yI'a(S) then systems (3) and (4) are uniquely solvable in CI'~(S)B,y

and C~1,a (S) x R 3 x R 3 , r e s p e c t i v e l y .

Let us d e s c r i b e a scheme of the p r o o f of the solvability confining

ourselves to s y s t e m (3). Let u (i) and u (e) be the solutions of
(D (i)) and (D (e)) . Then 2g = V ( T u (i) - Tu (e)) . We introduce the
solution v of the p r o b l e m (N (e)) where h : I/2 (Tu (i) - Tu (e))
Since v : I/2(Wv - VTv) on G (e) tben (-I + W) v .= V T v = g on ~ F.
Consequently, the vector-function T = v I UF is a s o l u t i o n of (3). The
inclusion • E CB,y(S) follows directly from Theorem I.

To p r o v e the u n i c i t y of the s o l u t i o n of (3) it s u f f i c e s to e s t a b l i s h

the s o l v a b i l i t y in C0'a(S) of the formally adjoint system (6), w h e r e
h ~ C 0'a
~,y(S) . Let v (~7 be a s o l u t i o n of (N (e)) . We cons.ider the
simple layer potential Vo which coincides with v (e) on G (e) (see
Theorem 2). It r e m a i n s to n o t e that the d e n s i t y ~ satisfies (6).

The above argument contains all e s s e n t i a l points for the p r o o f of

the f o l l o w i n g theorem on s o l v a b i l i t y of s y s t e m s (5) a n d (6).

THEOREM 4. There exists a bounded i~verse: (- I + W*) -I in C0'a(S)

B,Y
System (4) is solvable in C~,y
0'~(S) for all h with zero p r i n c i p a l vec-
tot and p r i n c i p a l moment.

For solution of s y s t e m s of integral equations under consideration

theorems on i n c r e a s i n g smoothness and changing collections {yE} and
 240

{6Q} hold which are similar to t h e theorems of the same kind for solu-
tions of boundary value problems (cf. E~ ). B e f o r e giving an example
let us introduce some function spaces.

Let l be an integer, Z ~ I , 0 < ~ < I If in t h e definition

1,e(G(i) ) we replace V by vI = {61/~x~]~x22~x~3} then we obtain
of C6, X

the definition of the space CB,y

l'~(G(i)) By C£'~(S)
B,y we mean the space
of traces on S of functions in C Z$,y
'~(G~i)). Here we can replace (i) b y (e~

The following assertion which completes Theorem 3 holds.

fy (1) and (2). If g £ C 61 :< (S) l , a y,(S)

~ C8, and l ' aY( S )
~ ~ C6, is a SO-

lutio, Of any system (3) and (4) then T E C [ ~ : y , (S)

Similar complements can be made to T h e o r e m s 2 and 4.

The potential theory for plane boundary value problems can be deve-
loped with the help of the same arguments and even more easily. If b y
S we mean a piecewise smooth contour without cusps and by {Q} we de-
note the set of its angular points then the statements of Theorems I - 5
remain valid up to obvious changes.

For harmonic and hydrodinamic potentials results similar to Theorems

I - 5 can be obtained by the same method.

5. The Fredholm radius of harmonic ~otentials

Here we present a formula for the Fredholm radii of the double

layer harmonic potential W and its formal adjoint W* in certain
H~ider spaces. We shall suppose that S contains no vertices. So the
collection {6Q} in the notation of the function spaces is o m i t t e d .

Let L be a linear operator in a B a n a c h space. The Fredholm radius

r(L) of L is t h e largest radius of a circle Cr = {I £ C : Ill < r}
such that for all I 6 Cr the operator I - IL is F r e d h o l m and
Ind (I - IL) = 0

We introduce the operators

I [ cos(x - y, n(y))
(W~) (x) = ~ ] T(y) ds(y) (7)
s ix - y l 2

(W'T) (x) I
27 I cos(x - y, n(x)) z (y) ds (y) (8)
s Ix - yl 2

where x E <j F .
 241

The following result is proved in [~ :

THEOREM 6. Let ~ 6 (0,1) , 0 < I + a- B < 1 , 1 ~ 0 and

I sin ~(I + a - ~)
R : min sin ((~ - % ( z ) ) il ~ - B))
z 6 tiE
Let W and W~ be operators in C Sl+,Ia( S ) and C B£ +/I ' a ( S ) , respecti-
vely, defined by (7) and (8). Then r(W) = r(W*) = R ,

We note that R > I if and only if

I + ~ - ~ < ~ % In _ %(z)I- for all z ~ <9 E .

The proof of Theorem 6 relies heavily on Theorem 7 where the fol-

lowing notation is used.

Let G = G (i) U G (e) and let u (i) and u (e) be restrictions of

the function u to G (i) and G (e) By C/'a(G) we denote the space
, m , ~G (e) ) . u The
of functions u with u(i) 6 -/'~(G
cB (~)) ' u~e) e C~g~(~) (e) norm
in
in C18'a(G ) is defined as the sum of the norms of and
CB/ ' ~ ( G (i)) and ~,l,~ (G (e) ) . F o r 1 => 1 , 0 < 8 < I + a we put

CA£B '~(G) : {u e C ls' a ( G ) : Au in G , u (i) = u (e) on S}

£B 'a(G) = {u 6 C /B' ~ ( G ) : AU = 0 in G , 6 uOn

(i) 6 uO(ne ) on <9 F}

Clearly, for l _> 0 the operators

~£+l'a(G) 9 u ÷ (1 - ~) % u ( i ) 8u(e) f £
LI : U5 6n (I + I)--~-- C~'a(S) ,

L2 : X£+1 ' a (G) -9 u ~ (I - ~ ) u (i) - (l + ~ ) u (e) # t:~/+q,c~(S)

uB 5
are bounded.

THEOREM 7. Let ~ { (0,1) , 0 < 1 + a - B < I , I £ 0 .

I) If I~I < R then Lk , k = 1,2 , is F r e d h o l m and

ind (Lk) = 0 .

2) If III : R then the range of Lk is n o t closed.

R e f e r e n c e s

[I~ MAZ'YA, V. G.: i n t e g r a l e q u a t i o n s o f the p o t e n t i a l t h e o r y in d o m a i n s

with piecewise smooth boundaries. Uspehi M a t e m . N a u k , v . 3 6 , n 4,
1981 p. 2 2 9 - 2 3 0 .
[~ MAZ'YA, V. G.: On solvability of integral equations in the classical
 242

theory of elasticity in domains with piecewise smooth boundaries.

Republican School-Conf. General Mechanics and Elasticity Theory
(Telavi, 1981), Abstracts of Reports, Tbilisi, ]981, pp. 55-56.
E 3] MAZ'YA, V. G.: The potential theory for t~e Lama system in domains
with piecewise smooth boundaries. Proc. of the Conference on the
Partial Diff. Equations in M e m o r i a m of I. N. Vekua, Tbilisi 1982
(to appear).
[ 4] MAZ'YA, V. G. and PLAMENEVSKI~, B. A.: L -estimates of solutions
of elliptic boundary value problems in p domains w~th edges.
Trudy Moskov. Mat. Obshch. 37 (1978), 49-93; English transl, in
Trans. Moscow Math. Soc. 1980, no. I (37).
E 5] : Estimates of the Green funavions and Schauder esti-
mates of the solutions of e~liptic boundary value problems ~n a
dihedral angle. Sibirsk. Mat. Zh. 19 (1978), 49-93; English transl.
in Siberian Math. J. 19 (]978).
[ 6] : Sahauder estimates of solutions of elliptic boun-
dary value problems in domain~{ with edges on the boundary. Partial
D i f f e r e n t i a l Equations (Proc. Sem. S. L. Sobolev, 1978, no. 2),
Inst. Mat. Sibirsk. Otdel. Akad. Nauk SSSR, Novosibirsk, 1978,
pp. 69-102; English transl, in Amer. Math. Soc. Transl. (2) 123
{1984).
[ 7] : Elliptic boundary val~e problems on manifolds with
singularities. Problemy Mat. Anal., vyp. 6, Izdat. Leningrad.
Univ., Leningrad, ]977, pp. 85-142; English transl., to appear in
J. Soviet Ma%h.
[ 8j : The first boandary value problem for the classical
equations of mathematicat phpsics in domains with piecewise smooth
boundaries (I) Zeitschrift f~r Analysis und ihre A n w e n d u n g e n ,
Bd. 2 (4) 1983, S. 335-359; (If) ibid. Bd. 2 (6) 1983, S.523-551.
93 GRA~EV, N. V. and MAZ'YA, V. G.: On the Fredholm radius for ope-
rators of the double layer potential type on piecewise smooth
boundaries. Vestn, Leningrad Univ. Math. (to appear).
[10] ZARGARYAN, S. S. and ~ Z ' Y A , V. G.: On singularities of solutions
o~" a system of integral equations in potential theory for the ~a-
remba problem. Vestn. Leningrad. Univ. Math. 1983, n I. English
transl.vol. 16 (1984), p. 49-55.
~11~ ZARGARYAN, S. S. and MAZ'YA, V. G.: On the asymptotic behavior of
solutions of integral equations ~n potential theory in a neijhbor-
hood of corner points of the contour. Prikl. Mat. Mekh. vol. 48,
n 1, 1984; English transl, in J. Appl. Math. Mech. 48 (1984).
2~ MAZ'YA, V. G. and PLAMENEVSKI~, On properties of solutions
B. A.:
of three-dimensiona~ problems of elasticity theory and hydrodyna-
mics in domains with isolated singular points. Dinamika Sploshnoj
Sredy ,Novosibirsk, n 50 (1981), p. 99-120 English transl, in
Amer. Math. Soc. Transl. (2) , vol. 123, 1984, p. 109-123.
HIGHER REGULARITY OF WEAK
SOLUTIONS OF STRONGLY
NONLINEAR ELLIPTIC EQUATIONS
C. G. SIMADER
Mathematisches Institut der Universitiit Bayreuth
Postfach 3008, D-8580 Bayreuth, West Ge~r~a,ny

In a b o u n d e d open set G c R N we consider the D i r i c h l e t problem

N
Lu: = -£u- ~ ~igi(~iu)+gO(u) : f , u I = O . (I)
i=I I~G
If f,g are m e a s u r a b l e functions on G, we w r i t e (f,g) : = f f ' g if
f.g 6 L I ( G ) . As usual, a weak solution of (I) is a f u n c t i o n

.1,2
u 6 H0 (G) s u c h that gi($iu)6 LI(G],"
" i = I, .... N

(~oU:=U) and
(H.I)
N
(2)
E(~i iu'~i~)+i=iZ ( g i ( ~ i u ) , ~ i ~ ) + ( g O ( u ) , ~) = (f,~)

holds for all % ~ Co(G).

For the strong nonlinearities gi we a s s u m e

[ 0
gi e C (R) are n o n - d e c r e a s i n g and
(H.2)
i(t).t ~ O for t ~ R, i = O,1 .... N .

Existence of those weak solutions was studied by a considerable number

of a u t h o r s . Observe that the n o n l i n e a r i t i e s depend on d e r i v a t i v e s up
to h a l f the order of the equation. Existence for those problems was
first proved in [3]. Let us e m p h a s i z e that all our considerations hold
true if -A is r e p l a c e d by a strongly elliptic operator of o r d e r 2m and
the n o n l i n e a r i t i e s may depend analogously to (1) up to t h e d e r i v a t i v e s
of o r d e r m. All existence-proves lead to w e a k solutions such that in
addition to (H.I) we get

(H.3) gi($iu).$i u e L] (G) (i = O,1 .... N)

Like as in the c a s e of linear equations two q u e s t i o n s arise: Under

 244

what conditions are the weak solutions unique? Do the weak solutions
have better regularity properties? For s t a r - s h a p e d domains uniqueness
and s t a b i l i t y of weak solutions of (I) was p r o v e d in [4]. This r e s u l t
was considerable generalized to a r b i t r a r y domains with smooth b o u n d a r y
and to v e r y g e n e r a l operators by M. L a n d e s [2].

Concerning higher regularity properties, surprisingly it turns out

that the m e a n w h i l e classical difference quotient method perfectly
works in the u n d e r l y i n g case to gain one more o r d e r of d i f f e r e n t i -
ability. As far as the a u t h o r knows, this m e t h o d goes back to S . A g m o n
(see e.g. [I]). For the n o n l i n e a r i t i e s we a s s u m e

Assume (H.2) and gi e C I (R) , i = 0,1 ..... N.

For i = I ,... ,N we a s s u m e

i) gi' is n o n - d e c r e a s i n g in [0, ~)

(H.4) ii) T h e r e exists a constant C > I such that

gi (-t) i Cg i
(t) for t E R ("nearly odd")
t
iii) Let Gj (t):= o/gi(s)ds. A s s u m e that G i s a t i s f i e s a

h2-condJtion: There exists a K ~> O such that

G i (2t) < KG i(t) for t ~ R.

iv) There exists y > O such that

Igi(t) l<yIgk(t) I for t ~ R and i,k : 1 ..... N

(" i s o t r o p i c " )

An e x a m p l e is g i v e n by

gi(t) := ~itlt[ p-l, ~i e R, P > I (3)

T h e n we can p r o v e the f o l l o w i n g

Theorem. Assume (H.I)-(H.4). Then, for

G ' c c G we h a v e U l G , 6 W 2 ' 2 ( G ' ) , gi'(~iu)-(~i~ku)2L, E L I (G') ,

go' (u).(~ku)2 I e L I ( G ') ( i , k = I ..... N)

• G'

and t h e r e is a c o n s t a n t K = K ( G ' , G , g i) such that

 245

N N
E f (~i~kU) 2+ ~ / gi' (~iu)'(~i[}k u)2 +
i,k:1 G' i,k=1 G'
N
+ k:1E/go'
( $ k U )(u)
2G' < K'(llfll 2(G) 0 (G)
N
+ Z /gi (~iu)'~i u )
i=O G

As m e n t i o n e d above, the proof is done by m e a n s of the d i f f e r e n c e quo-

tient method. We can not give details here. In any case, the proof is
completely elementary although it demands are careful analysis. To see
what is going on, we assume now that the w e a k solution is a r b i t r a r i l y
smooth and sketch how to get the a - p r i o r i - e s t i m a t e of the theorem.
Roughly spoken, one can prove with this m e t h o d all those regularity
properties which can be read of an a - p r i o r i - e s t i m a t e like as in the
theorem.

For this purpose, let ~ ~ Co(G) and for k = I,...,N put ~k@ as a test
function in (2) and integrate at the left hand side by parts, which
leads to

N N
Z (~i~kU,~i@) + ~ (gi' (~iu)~i~kU,~i@)+(go' (U)~kU,@) :
i=I i:I
(4)
: -(f,~k@ )

Let now 6 ~C;(G) such that ~--I in G'. Put @:=~kU.62 in (4) which
gives
N
N
E f (~i$~ u) 2.~2+i=1E/~ u.2~_u3.
{K . ~ i]~~ l +
i=I
N
+ 2 fgi , (~iu).(SiSkU)2 . 2+ N
~ fgi' (;iu) S i ~ k U ~ k U 2 ~ i ~ + (5)
i=I i=I

+ fg O' (u)(~u)2~ 2 : - ( f , ~ k S k U ~ 2 ) - ( f , ~ k U 2 ~ k ~)

The first, third and fifth expression at the left of (5) are that we
have to estimate. E.g. the second admits trivally the e s t i m a t e for
~>O

N
12 E f(~iSkU~).($kU).~i~ I <
i=I
N 22 -I N
2(~i~)2
_< ~- E /(~i~k u) + ~ /(~k u)
i=1 i=I
 246

analogous for the right h a n d side of (5). C u m b e r s o m e seems the fourth

term. To e s t i m a t e it, o b s e r v e gi' (t):£O" For 5 > O we get:

N
12 Z .'<dgi'(3iu).3i[~kU'<) (/gi' ([i u) 3kU~]i~) l
<6)
N
22 ~-I: ,
< 6"Z fgi' (~iu)'(~i~k u) "'~ +e "~gi (~iu)'(3kU)2(~i~)Z
i:1

To e s t i m a t e the s e c o n d expression on the r i g h t h a n d side of (6), we

make use of the following

Lemma: Assume (H.4). Then

i) There is a c o n s t a n t C > 0 such that
Gi(t) < C-Gk(t) for t 6 R and i,k = I , . . . , N

ii) There is a c o n s t a n t C' > 0 such that

gi' (t)s 2 _< C',(gi(t).t+G i(s))

Remark: Property ii) is no s u r p r i s e if we c o n s i d e r e.g. g(t) ::tit] p-1 ,

p > I. Then, g' (t) = p'Itl p-I , g ( t ) ' t = Itl p+1 and G(t) = (p+I)-11tl p+1
By the i n e q u a l i t y a'b < l-]al+l'-1.b i' for I < ,~,l'< ~, I-I+I '-I = I,

we get w i t h I:= + ~ ! and t h e r e f o r e I' = P-tl-

p-1 2

g ' (t)'s 2 < p ' p~ +C ~1 . g (t) "t + 2--~..S(s)

p+]

from which ii) follows. The a s s u m p t i o n s in (H.4)(especially iii))

guarantee L e m m a 2, p r o p e r t y ii) in general.

By m e a n s of the L e m m a we are n o w able to e s t i m a t e the s e c o n d e x p r e s -

sion at the r i g h t h a n d side of (6), f i r s t p o i n t w i s e :

gi' (3iu)'(3kU] 2 ! C' (gi(Diu)'3iu+Gi(3kU))

< C' (gi (3iu)'3iu + C ' G k ( 3 k U ) ) "

If w e o b s e r v e (H.2) and the d e f i n i t i o n of G we conclude

Gk(~kU) ! gk(3kU)3kU- Multiplying by (~i~)2k'integrating, and c o m b i n i n ~
all i n e q u a l i t i e s , we h a v e p r o v e d the d e s i r e d estimate.

References
[ 1] AGMON,S., Lectures on ellip tic boundary value problems, Van Nostrand, Princeton 1965.
[2] LANDES,M., Eindeutigkeit und Stabilitdt schwacher Ldsungen streng nichtlinearer ellip tischer
Randwertprobleme, Thesis, Bayreuth 1983.
[3] SIMADER,C.C~, OberschwacheLdsungendesDirichletproblemsf:drstrengnichlineareelliptische
Differentialgleichungen, Math. Z. 150 (1976), 1-26.
[4] SIMADER,C. G., Remarks on uniqueness and stability o f weak sotutions o f strongly nonlinear elliptic
equations, Bayreuther Math. Schriften 11 (1982), 67-79.
SOME REGULARITY RESULTS
FOR QUASILINEAR PARABOLIC SYSTEMS
J. STARA, O. JOHN
Faculty of Mathematics and Physics, Cha'rl~,~s U~ive~r,s'ity
Sokolovskd 8~ , 186 O0 Prague 8, 6~echoslovakia

We present three results on t h e r e g u l a r i t y of t h e w e a k solutions

of q u a s i l i n e a r parabolic systems of second order. Two of them con-
cern the relation between the regularity of the system and the L i o u -
ville property. The third one is the example of the system for w h i c h
there exists a weak solution of the b o u n d a r y value problem in t h e c y -
linder Q with Lipschitz continuous boundary data which develops the
singularity in t h e interior of Q.
Denote z = [t,xl C R X R n. L e t us c o n s i d e r the system of m e q u a -
tions for m u n k n o w n functions u = [ul,...,u TM] of the form

uti _ D (A~(z,u)D~uJ) = _fi + D g~, i = I,.. .,m,

where we sum over j from 1 to m a n d o v e r ~, B from 1 to n. For the

sake of b r e v i t y we rewrite it at t h e form

(i) u t - div(A(z,u)Du) = -f + d i v g.

i. I n t e r i o r r e g u l a r i t y . L e t Q be a d o m a i n in R X R n. S u p p o s e that
the following assertions o n A, f and g are satisfied:
(i) A = A(z,u) is c o n t i n u o u s on Q X R TM.
(ii) (A(z,u)~,~) > 0 for all [z,u] e Q X R TM, ~ # 0.
(iii) f 6 Ls,Ioc(Q) with some s > n/2 +i, g e Lr,loc(Q) with
r>n+l.
Denote by W2,
o,~oc(Q) the set of all functions belonging to the
L2,1oc(Q) together with their spatial derivatives. R e c a l l that the fu-
nction u 6 w O2,,l1 o c (-Q ) is a weak s o l u t i o n of (I) i n Q if for a l l ~6 D(Q)

f[(u,~ t) - ( A D u , D ~ ) ] d z = f[(f,~) + (g,D~)]dz .

Q Q

Definition I. T h e system (i) is said to be regular if e v e r y

bounded weak solution of (1) is l o c a l l y H~ider contiunuous in the
domain Q.
Definition 2. T h e system (I) has the interior iiouville property
in Q if for e a c h z 0 6 Q e v e r y b o u n d e d w e a k s o l u t i o n of the s y s t e m
 248

w t- d i v ( A ( z 0 , w ) D w ) = 0 in all R X R n is constant.
Theorem I. T h e s y s t e m (1) is r e g u l a r iff it has interior Li-
ouville property in Q.
S k e t c h of the proof. To h a v e the r e g u l a r i t y of the w e a k solution
u in Q, it is s u f f i c i e n t to p r o v e t h a t for each z 0 E Q

lim inf R ~ 0+[R -n-2 / lu(z) - ,RI2dz] = 0 r

Q(z0,R ) Uz 0

where Q(z0,R) = (t0-R2,t0) X Ix;Ix - x01 < R} and U z 0 , R is an i n t e g r a l

m e a n v a l u e of u over Q(z0,R).
Using the blowing-up technique, we o b t a i n this f o r m the i n t e r i o r
Liouville property. (For the d e t a i l s see [ i], [2].)
Remark. (The e l l i p t i c case.) J. D a n ~ e k from Brno proved in his
disertation that Theorem i (modified for e l l i p t i c systems) remains to
be true if we c h a n g e the r e q u e s t of b o u n d e d n e s s of the w e a k s o l u t i o n
(in b o t h the definitions 1 and 2) by the a s s u m p t i o n that the w e a k so-
lutions in q u e s t i o n belong to the s p a c e BMO. Further, he d e s c r i b e d the
nontrivial c l a s s of the s y s t e m s which satisfy BMO-interior Liouville
property.

2. R e g u l a r i t y of the C a u e h y problem. Let ~ be a d o m a i n in R n and

T > 0. D e n o t e Q+ : (0,T) X ~, F : {[0,x] ; x C ~] and Q : Q+ U F. To
the c o n d i t i o n s (i) - (iii) we add the a s s u m p t i o n concerning the i-
nitial function u0, namely:
(iv) u^ • W 1 ~ (~) N L (Q)~ q > n .
u q,±oc
The f u n c t i o n u • W%[~oc(Q)_,_ is a weak s o l u t i o n of Cauehy problem
for the system (I) i n Qwith initial f u n c t i o n u0 i f for each ~ • C~(Q)
w i t h supp ~ C Q

f[(u,9 t) - ( A D u , D ~ ) ] d z : f[(f,~) + (g,Dg)]dz - fu09(0,x)

Q Q

Definition 3. C a u c h y p r o b l e m is regular if its e a c h b o u n d e d so-

lution is l o c a l l y H o l d e r continuous o n Q.
Definition 4.The system (1) has boundary Liouville property on
F if for e a c h z 0 • F e v e r y b o u n d e d w e a k solution of C a u c h y p r o b l e m
for the s y s t e m w t- d i v ( A ( z 0 , w ) D w ) = 0 in the set {[t,x] ; t ~ 0, x• R n]
with initial function u 0 ~ 0 is e q u a l zero i d e n t i c a l l y .
Theorem 2. C a u c h y p r o b l e m for (1) in Q w i t h initial function u0
is r e g u l a r iff the s y s t e m (i) has interior Liouville property in Q +
and b o u n d a r y Liouville property on F.
S k e t c h of the proof. We e x t e n d the c o e f f i c i e n t s a n d the r i g h t
hand side f u n c t i o n of (i) to the c y l i n d e r G : Q u (-Q+). The weak
 249

solution of the C a u c h y problem for (i) in Q can be shifted and p r o -

longed in a" s u i t a b l e manDer to the w e a k solution of the extended sy-
s t e m on t h e w h o l e G. U s i n g now the interior regularity result in G
we o b t a i n the a s s e r t i o n of T h e o r e m 2 immediately. (For the d e t a i l s
see [ 2] .)

3. E x a m p l e . L e t m = n = 3, Q = (0, ~) X B (B is t h e u n i t b a l l in
R3). We obtain the example of the system (1) for w h i c h the b o u n d a r y
value problem with Lipschitzian boundary data on F = [ {0] X B] O
[(0, ~) X ~B] has a solution u which develops the singularity for some
t O > 0 in two steps: a) W e c h o o s e the suitable u and b) to t h i s u we
construct the system for w h i c h u is the w e a k solution of the b o u d a r y
value problem.
In the choice of t h e solution we were inspired by M. Struwe [3]
who constructed the e x a m p l e for the system u ti _ Au i = f i ( t , x , u , D u ) '
i = l,...,m, with fi,growing quadratically in IDul. We set
xi
(2) ui(t,x) = Ixf
x.
i if t < i, i = I,.
# "T'xT ..,3 ,
where ~ is f u n d a m e n t a l solution of the equation w t + Aw = 0. It is
easy to see that u is l o c a l l y Lipschitz continuous on R X R n w i t h ex-
c e p t of t h e h a l f - l i n e p = {[t,0] ~ t ~ i], w h e r e it c e a s e s to be
continuous.
To c o n s t r u c t the system, we m o d i f y the p r o c e d u r e due to M. Gia-
quinta a n d J. Sou~ek° At first we seek the system with bounded and me-
asurable coefficients in the form

(3) w t - div(A(z)Dw) = 0,

did ~
~ where d : D U i- b

Substituting u for w into this s y s t e m we o b t a i n the conditions on b .

under which u is a s o l u t i o n of the system ( 3):

(4) u~ : D bai, i : i,...,3 .

Choosing reasonably the f o r m of b i we get a f t e r tedious calcula-

tions an e x p l i c i t f o r m of c o e f f i c i e n t s A:

<s) : LBhj ÷ d id j
where
 250

i x.x
d • - ~ 6 (a-2) z ~ (6+a)} if t > i.

d , 1 {-6 i[a-2 + (6+a) ~-]

2 -
~4(a-2) + (6+a)q(4-3q)
x.x
±~ 3q
ixl~ (6+a)(l- 2 )} if t < 1 .

Here a > 2 is a r e a l parameter, ~ - IxJ and q = q($) = $-2

2~V>t
£ 2
_ ~-le-~2 ( f e -T d T ) -I.
0
Theorem 3. The function u given by (2) is a weak solution of
the b o u n d a r y value problem for (3) w i t h the c o e f f i c i e n t s given by (5)
and the b o u n d a r y function u0= Trace u on F. T h i s solution is L i p -
schitz continuous with except of the half-line p = {[t,0] ; t ~ i} w h e -
re it c e a s e s to be c o n t i n u o u s .
Remark. Rewriting XlXetX1-2 in the c o e f f i c i e n t s we c a n p a s s to
the d e s i r e d quasilinear system of the type (i). For the details see
[4].

References

[i] J O H N , O . , The i n t e r i o r r e g u l a r i t y and t h e L i o u v i l l e p r o p e r t y f o r

t h e q u a s i l i n e a r p a r a b o l i c s y s t e m s , C o m m e n t . Math, Univ. C a r o l i n a e
23 (1982), 685-690.
[2] JOHN,O., S T A R ~ , J . , On t h e r e g u l a r i t y of t h e weak s o l u t i o n of
Cauchy problem f o r n o n l i n e a r p a r a b o l i c s y s t e m s v i a L i o a v i l l e
p r o p e r t y , C o m m e n t . Math. Univ. C a r o l i n a e 25 (1984), 4 4 5 - 4 5 7 .
[ 3] STRUWE,M., A eounterexample i n r e g u l a r i t y t h e o r y f o r p a r a b o l i c
s y s t e m s , Czech. Math. J o u r n a l 34 (109>, 1984.
[4] S T A R ~ , J . , J O H N , O . , A counterexample.., to a p p e a r in c o m m e n t . Math.
Unive. Carolinae.
CLASSICAL BOUNDARY VALUE PROBLEMS
FOR MONGE-AMPI RE
TYPE EQUATIONS
N. S. TRUDINGER
Centre f o r Mathematical Analysis, Australian [Valio~al University
Ca~berra, A. C T. Australia

This report is concerned with recent work on the solvability of classical

boundary value problems for elliptic Monge-Amp~re type equations with particular
attention to that of the author, P-L. Lions and J.I.E. Urbas [20] on Neumann type
problems. The Dirichlet problem for these equations,

det D2u = f(x,u,Du) in ~ , (I)

u = ~ on ~n , (2)

in convex domains ~ in Euclidean n space ~n , has received considerable

attention in recent years. For the standard Monge-Amp~re equation,

det D2u = f(x) in ~ , (3)

Pogorelev [21,22] and Cheng and Yau [7] proved the existence of a unique convex
solution uE C2(~) ncO'l(~), provided ~ is a uniformly convex C I'I domain in
~n and the functions ~,f ~ CI'I(~) with f positive in ~ Their methods
depended on establishing interior smoothness of the generalized solutions of
Aleksandrov [1]. These results were extended to equations of the more general form
by P-L. Lions [17,18] using a direct PDE approach. Lions' approach led to the
following classical existence theorem of Trudinger and Urbas [26], which we formulate
explicitly for comparison with later results. Here we assume that the function f
in equation (I) belongs to the space CI,1(~×~× ~n) , is positive and non-decreasing
in z , for all (x,z,p) ~ × ~ × ~ n and satisfies the following growth limitations:

f(x,N,p) .< g(x)/h(p) (4)

1 n
for all (x,p) ~x~. n , where N i s some c o n s t a n t and g ~ L1(~2), h ~ Lto c ( l t ) a r e
positive functions such that
 252

fa g < f h
~n
; (5)

f(×,N',p) S K [ d i s t ( x , a e ) ] a ( i + l p l 2 ) ~/2 (6)

for all x~ N, p ~ ~n where N' = max ~ , K, a and g are non-negative constants

am
such that 6 S n+l+ a and N is some neighbourhood of 3~ . Then we have

Theorem 1 [26] Let ~ be a uniformly convex C 1'1 domain in ~n , ~ ~ C I , I ( ~ )

and suppose that f satisfies the above hypotheses. Then there exists a unique
convex solution u ~ C2(~) N C 0'I (~) of the Dirichlet problem (i), (2).
Conditions (4) and (6) were introduced by Bakelman [2] in his treatment of
generalized solutions and they are both sharp [2],[26]. For the special case of
the equation of prescribed Gauss curvature,

det D2u = K(x)(l+lDul2) (n+2)/2 , (7)

conditions (5) and (6) become respectively,

< (8)
I~ Wn ,

K = 0 on 3~. (9)

Moreover condition (8) is necessary for a cO'I(~) solution of equation (7) to exist
[9],[26] while if condition (9) is violated there exist arbitrarily smooth boundary
values ~ for which the classical Dirichlet problem (7), (2) is not solvable, [26].
The above developments shed no light on the global regularity of solutions beyond
being uniformly Lipschitz in ~ . This was an open problem, in more than two
dimensions, for many years and was finally settled, for uniformly positive f,
through the contributions of Ivochkina [lO], who proved global bounds for second
derivatives for arbitrary ~ C3"I(~), 3~E C 3'I, Krylov [14],[15] and Caffarelli,
Nirenberg and Spruck [5] who independently discovered the hitherto elusive global
H~ider estimates for second derivatives. As a particular consequence of this work,
we can infer the following existence theorem for globally smooth solutions of the
classical Dirichlet problem.

Theorem 2 Let ~ be a uniformly convex C 3'I domain in ~Rn~c3'I(~) and

suppose that f ~ C I,I (~x~xRn) is positive and non-decreasing with respect to z,
for all (x,z,p) ~ ~×I~×l~ n and satisfies conditions (4) and (6) with a=O. Then there
 253

exists a unique convex solution u (C3,Y(~) for all y<1 of the Dirichlet problem
(1),(2).
More general results are in fact formulated in [5] , [12] but the condition
6 ~ n+l cannot be improved [26]. The situation with regard to oblique boundary value
problems of the form

5-Du = ¢(x,u) on ~ , (I0)

where B.~ > O on 3~ and v denotes the unit inner normal to 3~ , turned out to
be more satisfactory in that condition (6) is not required for the estimation of
first derivatives. For the case 6 = v, that is for the usual Neumann case,

v-Du = ¢(x,u) on a¢~ , (Ii)

we proved in collaboration with Lions and Urbas in [20], the following existence
theorem,

Theorem 3 Let ~ be a uniformly convex C 3'I domain in ~n and ~ ~ c2"l(~x~)

satisfy

Cz (x'z) ~ YO (12)

for all x,z, ( 3~x~ and some positive constant YO Then if f 6 c l ' l ( ~ x ~ x ~ n)
is positive and non-decreasing with respect to z for all (x,z,p) ( ~x~×~ n
and satisfies condition (5), there exists a unique convex solution u ~ CS'Y{~)
for all y < I of the boundaz~ value problem (i),(ii).
Further regularity of the solutions in Theorems 2 and 3 follows by virtue of
the Schauder theory of linear equations [9], when ~ ,¢ and f are appropriately
smooth. In particular when 3~ ( C~ , ¢ (C~(D~x~) and f ~ C (~x~x~) we
deduce u (C~(~). The proofs of Theorems 2 and 3 both depend, through the method
of continuity as described for example in [9], on the establishment of global
C2'G(~) estimates for solutions of related problems. However the techniques
employed by us to obtain these estimates in the Neumann boundary value case differ
considerably from those used for the Dirichlet problem, particularly with respect
to the estimation of first and second derivatives. For the estimation of sup norms
we make use of the following maximum principle which does include that of
8akelman [2,3] for the Dirichlet problem as a special case.

Theorem 4 [20] Let ~ be a C I bounded domain in ~n and u E C2(~) N CI(~) a convex

solution of the boundary problem (i),(i0) in ~ where f satisfies condition (5),
 254

B'v ~ 0 on ~ and ~ satisfies (12). Then we have the estimate

min {N, +
- sup ~ (x,0)/Y0 - (Bl/Y0+d)R0 }< u < sup ~ (x,0)/Y0 (13)

where d = diam ~,B 1 = suplB I , and R0 is given by

9~

f g = !h
I I<R0

The gradient estimation in the o~lique boundary condition case is a consequence

of convexity as any convex C1(~) function satisfies an estimate

suplDul a c (14)

where C depends on ~0,Bl,lUl0;~, suplB.Du I and ~, provided B'v~80 where

~0 is a positive constant and ~ E CI,I[20]. In contrast, a gradient estimate for

solutions of the Dirichlet problem (i),(2) holds provided $~E C I'I, ~ ~ C1'I(~)
and condition (6) is fulfilled [26].
In both Dirichlet and Neumann problems the global estimation in ~ of second
derivatives is reduced to considerations at the boundary ~ , by means of an approach
which goes back to Pogorelev [21], although its implementation in the Neumann case
[20] is substantially more involved than in the Dirichlet case [9], [5]. The
boundary considerations are different as the Dirichlet problem is handled through
barrier constructions [i0], [5], whereas in [20] we employ different techniques
including a device which necessitates our restriction of the vector ~ to the normal
vector. The consequent estimates may be formulated as follows.

Theorem 5 Let ~ be a C 3'I unifo~cnly convex domain in ~n and t~~ C4(~) NC3(~)
a convex solution of the boundary value problem (I),(ii) where f~ cl'I(~x~X~R n)
is positive and ~e C2'I(~xIR) satisfies (12). Then we have

suplD2ul £ C (15)

where C depends on n,~,f,~ and lUll;~ A similar estimate holds for solutions
of the Dirichlet problem (1),(2) provided ~ E C3'1(~).
We remark that the restriction (12) can be weakened to ~z ~ 0 and the case when
fl/n is convex with respect to p is simpler. We do not know whether one need only
assume ~ e c2'l(~) in the Dirichlet case. Once the second derivatives of solutions
of the boundary value problems (I), (2), (I0) are bounded, we obtain a control on
the uniform ellipticity of equation (I) , and further estimation hence follows from
 255

the theory of fully nonlinear uniformly elliptic equations. In particular interior

C2"a(9) estimates were derived by Calabi [4] for Monge-Amp~re equations and by
Evans [8] and Krylov [13] for general uniformly elliptic equations. Global C2"~(~)
estimates for the Diricblet problem then arose from combination with key boundary
estimates discovered by Krylov [14] and Caffarelli, Nirenberg and Spruck [5]. Global
C2'e(~) estimates for oblique boundary value problems were proved by Lions and
Trudinger [19], with more general results being given by Lieberman and Trudinger [16]
and Trudinger [24]. The global estimates of Krylov [14] and Trudinger[24] are
also applicable to classical solutions of uniformly elliptic Hamilton-Jacobi-Bellman
equations. We may in fact formulate these estimates for general second order
equations of the form

F[u] = F(x,u,Du,D2u) = 0 in f; , (16)

subject to general boundary conditions

G[u] = G(x,u,Du) = 0 on 8~ , (17)

where either G is oblique so that

g .v > 0 (18)
P

for all (x,z,p) ( 3£xlRxlRn, or G is Dirichlet so that

G(x,z,p) = z-~(x) (191

f o r some f u n c t i o n ~ ~ C2'1(~£). Here F(cl'k~x~×An×U),G~cl'ka~x~x~n) where

U i s some open c o n v e x s u b s e t of thelinear space sn of nxn real symmetric
matrices, and F is: (i) elliptic so that the matrix,

Fr = [Fr. ] > 0 , (20)

*J

for all (x,z,p) <~x~×~n , r = [rij ] < U ; and (ii) concave with respect to r
for all ( x , z , p ) ~ ~ x ~ × ~ n , r { U. Then we h a v e

Theorem 6 Let ~ be a bounded C3 ' 1 domain in ~n and u ~ C3(~)nC4(~) a

solution of the boundary value problem ( 1 6 ) , ( 1 7 ) s u c h that D2u(~) c U . Then
we have

[D2u]a;~ 8 C (21)
 256

where a< I and C are positive constants depending only on n,~,]u]2;~ and
the first and.second derivatives of F and G (excluding Frr)~ (and itI3 in
the Dirichlet case).
We remark here that the solution u in ~ e o r e m 6 need only lie in the space
CI'I(~) and the smoothness of a~, G, F, ~ can be reduced, [25]. For application
to Monge-Ampgre type equations the convex set U becomes the set of positive
symmetric matrices.
Finally we note that the sharpness of condition (8) is strikingly demonstrated
by the following result of Urbas [28] concerning extremal domains for the equation
of prescribed Gauss curvature.

Theorem 7 Let ~ be a uniformly aonw~'x domain £n Nn and K(:C I'I(~) be positive

in S2 and satisfy

f K
g2
w
n
(22)

Then there exists a convex solution u ~ C2(~2) of equation (7) in ~'~ l;'~rthermore
the f~nation u is ver'tical at 9~ and is unique up to additive constancs. If'
K is positive in ~ , then the solution u is bounded; if K vanishes o~ 9~
then the solution u approaches infinit;{ at ~)Q .

REFERENCES

[1] A.D. A l e k s a n d r o v , D i r i c h l e t ' s problem f o r t h e e q u a t i o n

DetlIzijll = P ( z i , . . . , Z n Z , X l , . . . , X n ) , V e s t n i k Len2n~£rad Univ. t 3 (1958),
5-24, [Russian].

[2] I. Ya. B a k e l ' m a n , The D i r i c h l e t problem f o r e q u a t i o n s o f blonge-Amp~re t y p e and

t h e i r n - d i m e n s i o n a l a n a l o g u e s , Dokl. akad. Nauk, SS2R 126 ( 1 9 5 9 ) , 923-926,
[Russian].

[3] I. Ya. B a k e l ' m a n , The D i r i c h l e t problem f o r t h e e l l i p t i c n - d i m e n s i o n a l Monge-

Ampere e q u a t i o n s and r e l a t e d p r o b l e m s i n t h e t h e o r y o f q u a s i l i n e a r e q u a t i o n s ,
Proceedings of 2:eminar on Monge-dmp~re Equations and Related Topics, Firenze
1980), Instituto Nazionale di Alta Matematica, Roma, (1982), 1-78.

[4] E. C a l a b i , Improper a f f i n e h y p e r s p h e r e s o f convex t y p e and a g e n e r a l i z a t i o n

o f a t h e o r e m by K. J b r g e n s , M i c h i g a n Math. J . 5 (1958), 105-126.
 257

[51 L. Caffarelli, L. Nirenberg, J. Spruck, The Dirichlet problem for nonlinear

second order elliptic equations, I. Monge-Amp~re equation, Oomm. Appl. Math.
37 (1984), 369-402.

[6] L. Caffarelli, J.J. Kohn, L. Nirenberg, J. Spruck, The Dirichlet problem for
nonlinear second order elliptic equations II. Complex Monge-Amp~re and
uniformly elliptic equations, Comm. P~re Appl. Math. 38 (1985), 209-252.

[71 S.-Y. Cheng, S.-T. Yau. On the regularity of the solution of the n-dimensional
Minkowski problem, C o ~ . Pure AppZ. ~ t h . 29 (1976), 495-516.

[8] L.C. Evans, Classical solutions of fully nonlinear, convex second order
elliptic equations, Comm. P~re Appl. Math. 35 (1982), 333-363.

[91 D. Gilbarg, N.S~ Trudinger, Elliptic partial dif¢~rential equations of second

order, Springer-Verlag, Berlin-Heidelberg-New York-Tokyo, Second Editiom, J983.

[lO] N.M. Ivochkina, Construction of a priori bounds for convex solutions of the
Monge-Amp~re equation by integral methods, Ukrain. Mat. Z. 30 (1978), 45-53,
[Russian].

[ii] N.M. Ivochkina, An apriori estimate of lUi[c2(~) for convex solutions of the

Dirichlet problem for the Monge-Amp~re equations, Zap. Nau~n. Sem. Leningrad.
otdel. Mat. Inst. Steklov. (LOMI) 96 (1980), 69-79, [Russian], English
translation in J. Soviet Math. 2_1 (1983), 689-697.

[12] N.M. Ivochkina, Classical solvability of the Dirichlet problem for the Monge-
AmpSre equation, Zap. Nau$n. Sem. Leningrad. @td~l. Mat. 2nst. Steklov. (LOMI)
131 (1983), 72-79.

[13] N.V. Krylov, Boundedly inhomogeneous elliptic and parabolic equations, izv,
Akad. Nauk. SSSR 46 (1982), 487-523, [Russian]. English t r a n s l a t i o n in Math.
USSR i z v . 200 (1983), 459-492.

[14l N.V. Krylov, Boundedly inhomogeneous elliptic and parabolic equations in a

domain, Izv. Akad. Nauk. SSSR 477 (1983), 75-108, [Russian].

I151 N.V. Krylov, On degenerative nonlinear elliptic equations, Mat. Sb. (N.S.)
120 (1983), 311-330, [Russian].

[16] G.M. Lieberman, N.S. Trudinger, Nonlinear oblique boundary value problems for
nonlinear elliptic equations. Aust. Nat. Univ. Centre for Math. Anal.
Research Report R24 (1984).

[17] P.-L. Lions, Sur les 6quations de Monge-Amp~re I, M a n u s a r i p t a Ma!;h. 41

(1983), 1-44.

Jig] P.-L. Lions, Sur l e s 6quations de Monge-Amp~re I I , Arab. R a t i o n a l M~ch. Anal.

(to appear).

[19] P.-L. Lions, N.S. Trudinger, Linear oblique derivative problems for the
uniformly elliptic Hamilton-Jacobi-Bellman equation, Math. Zeit. to appear).

[20] P.-L. Lions, N.S. Trudinger, J.I.E. Urbas, The Neumann problem for equations
of Monge-Amp~re type. Aust. Nat. Univ. Centre for Math. Anal. Research Report
Rl6 (1985).

121l A.V. Pogorelov, The Dirichlet problem for tile n-dimensional analogue of the
Monge-Amp~re equation, Dok~. Akad. Na~k. ;;SSR 201 (1971), 790-793, [Russian].
English translation in Soviet Math. Dokl. 12 (1971), 1727-1731.
 258

[22] A.V. P o g o r e t o v , The Minkowski multidimensional problem, J . Wiley, New York,

1978.

[23] F. S c h u l t z , A remark on f u l l y n o n l i n e a r , c o n c a v e e l l i p t i c e q u a t i o n s , Pro#.

C e n t r e for, Math. Anal. A u s t . Nat. U n i v . 8 , (1984), 202-207.

[24] N.S. T r u d i n g e r , Boundary v a l u e p r o b l e m s f o r f u l l y n o n l i n e a r e l l i p t i c equations.

Proc. C e n t r e f o r Math. Anal. A u s t . N a t . U n i v . 8 (1984), 6 5 - 8 3 .

[25] N.S. T r u d i n g e r , R e g u l a r i t y o f s o l u t i o n s o f f u l l y n o n l i n e a r e l l i p t i c e q u a t i o n s .
B o l l . Un. Mat. I t a l . , 3-A ( 1 9 8 4 ) , 4 2 1 - 4 3 0 . II A u s t . Nat. U n i v . C e n t r e f o r Math.
A n a l . R e s e a r c h Report R38 (1984).

[26] N.S. T r u d i n g e r , J . I . E . U r b a s , The Dirichl~et p r o b l e m f o r t h e e q u a t i o n o f

p r e s c r i b e d Gauss c u r v a t u r e , B u l l . A u s t z ~ l . Math. Soc. 28 (1983), 217-231.

[27] N.S. T r u d i n g e r , J . I . E . U r b a s , On s e c o n d d e r i v a t i v e e s t i m a t e s f o r e q u a t i o n s o f
Monge-Amp~re t y p e , BulZ. A u s t r a l . Math. S o e . 3 0 ( 1 9 8 4 ) , 321-334.

[28] J . I . E . U r b a s , Some r e c e n t r e s u l t s on t h e e q u a t i o n o f p r e s c r i b e d Gauss

curvature, Proa. Centre f o r Math. Anal. A u s t . N a t . U n i v . 8 ( 1 9 8 4 ) , 215-220.

[29] J . I . E . U r b a s , The e q u a t i o n o f p r e s c r i b e d Gauss c u r v a t u r e w i t h o u t b o u n d a r y

c o n d i t i o n s , J. D i f f e r e n t i a l G e o ~ t ~ (to appear).

[30] J.I.E. U r b a s , The g e n e r a l i z e d D i r i c h l e t p r o b l e m f o r e q u a t i o n o f Monge-Amp~re

type. Ann. L ' I n s t . Henri P o i n a a r g , (to appear).
QUALITATIVE PROPERTIES
OF THE SOLUTIONS
TO THE NAVIER-STOKES EQUA ONS
FOR COMPRESSIBLE FLUIDS
A. VALLI
Dipartimento di Matematica, Universitd di Trento
38050 Povo (Trento), Italy

1. Introduction.

We w a n t to p r e s e n t a new m e t h o d for showing the e x i s t e n c e of a s t a t i o -

nary solution to the equations which describe the m o t i o n of a v i s c o u s

compressible barotropic fluid.

At first it is u s e f u l to r e c a l l some k n o w n results concerning the n o n -

stationary case. The e q u a t i o n s of m o t i o n are

~v
P [~ +(v'V)v - f] = - V[p(@) ] + ~ 5 v + ( ~ + D / 3 ) V d i v v in ]0,T[x~
~p
+ div(pv) = 0 in ]0,T[x ~ ,

V = 0 on ]0,T[x ~,
(NS) I~
Sp : FI~T > 0 (I~Immeas(~)),

v]t:0 = v° in ~ ,

in ~ ,
Plt:0 : Po

where f~ c R 3 is a b o u n d e d domain, with smooth boundary ~; v and p are

the v e l o c i t y and the density of the fluid; p is the p r e s s u r e , which is

assumed to be a k n o w n function of Q; f is t h e (assigned) external force

field; the c o n s t a n t s Z > 0 and { Z 0 are the v i s c o s i t y coefficients;

> 0 is the m e a n density of the fluid, i.e. the t o t a l mass of fluid

divided by I~I; Vo a n d Po are the initial velocity and density.

In the last years it has b e e n proved that:

(i) if v a n d Po- ~ are small enough and f = 0, then problem (NS) has
o
a u n i q u e g l o b a l (in time) solution (Matsumura-Nishida [I]);
 260

(ii) the p r e c e e d i n g result also holds for a sufficiently small f # 0;

moreover, two small solutions are asymptotically equivalent as

t - - + + ~, and consequently if f is p e r i o d i c (independent of t) then
there exists a periodic (stationary) solution (Valli [3]).
It m u s t be u n d e r l i n e d that no o t h e r method is k n o w n for showing the
existence of a s t a t i o n a r y solution, excepting when the v i s c o s i t y coef-
ficients satisfy ~ >>D. In this case Padula [2] p r o v e d that, if f is

small enough, then there exists a stationary solution. Remark, however,

that in g e n e r a l the shear viscosity coefficient D is l a r g e r than the
bulk viscosity coefficient 6. M o r e o v e r , f r o m the m a t h e m a t i c a l point of
view it w o u l d seem only necessary to r e q u i r e that D is p o s i t i v e , with-
out assumptions on the largeness of ~.

The m e t h o d that we w a n t to p r e s e n t here is b a s e d on a " n a t u r a l " linea-

rization of the problem, followed by a f i x e d point argument. The v i s c o -
sity coefficients are only required to satisfy the thermodynamic re-
strictions ~ > 0, ~ ~ 0.

2. The linear problem (L).

Since we are searching for a s o l u t i o n in a n e i g h b o u r h o o d of the e q u i -

librium solution ~ = 5, v = 0, it is u s e f u l to i n t r o d u c e the n e w u n k n o w n
= p - ~ .

The e q u a t i o n s of m o t i o n in the stationary case thus become

I - ~ Av - (~+~/3) ?div v ÷ p1?G = (d+~) [f - "(v'V)v] ÷

+ [Pl - p' (~+~)]?~ in ~,

(S) ~ d i v v + div(v~) = 0 in ~,

vl~ ~ = 0 on ~,

fd = 0

where it is a s s u m e d that Pl a p' (~)> 0.

It is e a s i l y verified that a solution of (S) e x i s t s if w e find a fixed
point of the m a p

¢ : (v,~) ~ (w,~) ,

defined by m e a n s of the solutions of the following linear problem

 261

- p Aw - (~+~/3)Vdiv w + PiVn : (~+7) [ f - ( v ' V ) v ] +

+ [pl-p' (o+~) ]go ~ F i n ~,

(LI div w + div(vq) : 0 in ~,

wl~ ~ = 0 on 8~,

~ = 0

3. A-priori estimates for the solution of (L).

We want to obtain a-priori estimates in Sobolev spaces of sufficiently

large order, in s u c h a way that we can control the behaviour of the non-

linear terms which appear in F. We shall prove that a solution (w,r]) o f

(L) satisfies

(3.1) llwIl3+ II n 112 <

= e I II ~ III
for v I ~ = 0 and llvll 3 <= A small enough. Here [I " [[k is t h e norm in

the Sobolev space Hk(R), and c I depends~in a continuous way on ~, % and

A (but it is independent o f v) .

(a) At first, from well-known results on stokes problem we get

(3.21 il w If3+ II n II2 s c(ll r 111+ II div wll 2)

Hence our aim is t o estimate [] d i v w 112 .

(b) M u l t i p l y i n g (L) I b y w and (L) 2 b y (pl/7)q and integrating in Q o n e

has

I/2
{3.3} [lwH1÷ lldivwll 0 S c ( l l F H _ 1 + l [ v l l a lln[10)
The same argument can be used for estimating all the successive de-

rivatives in the interior of ~, and the tangential derivative

D div w near the boundary ~, obtaining in t h i s way (in l o c a l coor-

T
dinates near SQ)

I/2
(3.4) II D W II1÷ II D div w II 0 s c~ll F I[o÷ II v II 3 II n II 1)
(c) T h e estimate for the normal derivative D div w is o b t a i n e d by ob-
n
serving that on $~

Aw-n ~ Vdiv w -n ,

2
in t h e sense that their difference does not contain D w.
n
 262

Hence by t a k i n g the n o r m a l derivative of (L)2, multiplied by (5) -I

(~+4Z/3), and adding it to the n o r m a l component of (L) I we g e t (in

local coordinates near ~)

(3.5) P i D n ~] + (~+4~/3)/~ mndiV(v~) ~ F-n .

From this equation one easily gets

(3.61 /IDn~l/0~ c(llr II0+ Itvll13/211n II1).

Moreover, going back to (L) I , one has

PiDnrl = ~ £ w - n + ($+~/3)Vdiv w-n + F-n ~ (¢+4D/3)D div w +

n
+ F.n ,

hence from (3.6)

(3.71 Ilmdiv~llo
~e(llFII0*n Ilvl13I/2 [Inllll"
By r e p e a t i n g the same argument for the second order derivatives one

gets

(3.8) 11 div w I12 ~ e(lt F II1+ 11 v 1{~/211 n {12),

hence (3.1) holds if [I v I13~A s m a l l e n o u g h .

4. Existence of the solution of (L).

Though problem {L) is l i n e a r , a n d we k n o w that the a-priori estimate

(3.1) holds, the e x i s t e n c e of a s o l u t i o n w6H3(~), n6H2(~) is not ob-

vious.

In fact, the usual elliptic approximation cannot work in this case.

More precisely, if we add s £~s to (L)

, we m u s t a l s o r e q u i r e a
2
boundary condition (say, D i r i c h l e t or N e u m a n n ) on n • B u t the l i m i t
£
f u n c t i o n ~ is f r e e on 29. H e n c e the s e q u e n c e N c a n o n l y c o n v e r g e in
2
L (~) ( D i r i c h l e t c o n d i t i o n ) , or in HI (~) (Neumann condition), and can-

not converge in H2(~)~

Moreover,.if v ~ 0 (L) is not an e l l i p t i c system in the sense of Ag-

mon-Douglis-Nirenberg (if v = 0 (L) is the Stokes system). Hence the

usual regularization procedures do not w o r k .

One can p r o c e e d in the following way. By a d a p t i n g the m e t h o d of Pa-

 263

dula [2] to p r o b l e m (L), one defines

(4.1) ~ ~ (pl/H)q - (~/~ + I / 3 ) d i v w

and (L) is t r a n s f o r m e d into

I - Aw + 77 = F/H in 9,
(L') " div w = (~/~ + I/3)-<I (plq/%1 - ~) in

wi~ ~ = 0 on ~,
I

(L") ~ ~(~/U + 1/3)-IplD/H + div(vq) = ~(6/H + I/3) -I~ in ~,

L fq = 0

These equations can be solved via a fixed point argument if { >> ~.

Hence the a-priori estimates (3.1) and the continuity method give the
result for any p a i r of v i s c o s i t y coefficients satisfying H > 0 and
~ > 0.

5. E x i s t e n c e of a s o l u t i o n of (S).

We p r o v e at last the e x i s t e n c e of a f i x e d point for the m a p

: (v,o) ~ (w,n).

Taking

K a { (v,o)EH3(~)x H2(~) I vla~, = o, /o : o, II v ll~+ ]1 ~ [I 2 <

= A },

by u s i n g (3.1) one s e e s that

[[w [[3 ÷ In < c~l[

I[ 2 ,~ F[[ t ~ c [ ( [ T,~ I [ 2 ÷ ll([[f[[ '1 ÷ l t v H 22 } ÷

+ II o [I~ ] ~ c(A+I)(if f 111+ A 2)

Choosing A2 = I[ f II 1 << 1 , one has

llw 113 + Itn I12 <= A,

hence ~(K) o K . The set K is c o n v e x and c o m p a c t in X -- H 2 ( ~ ) x H I (~), and

it is e a s i l y seen that the m a p ~ is c o n t i n u o u s in X. The e x i s t e n c e of a
fixed point is n o w a consequence of S c h a u d e r ' s theorem.
 264

References.

[l] A. M a t s u m u r a - T. Nishida, Initial boundary value p r o b l e m s for the

equations of m o t i o n of general fluids, in "Computing methods in
applied sciences and engineering, V", ed. R. Glowinski - J.L. Li-
ons, N o r t h - H o l l a n d Publishing Company, A m s t e r d a m - N e w York-Oxford,
7982. (See also Preprint U n i v e r s i t y of Wisconsin, MRC Technical
Summary Report n ° 2237 (1981)).
[2] M. Padula, E x i s t e n c e and u n i q u e n e s s for v i s c o u s steady compressi-
ble motions, Arch. Rational Mech. Anal., to appear.
[3] A. Valli, Periodic and stationary solutions for c o m p r e s s i b l e Na-
v i e r - S t o k e s equations via a stability method, Ann. Scuola Norm.
Sup. Pisa, (4) 10 (1983), 607-647.
ON GEL'FAND'S METHOD OF CHASING
FOR SILVING MULTIPOINT
BOUNDARY VALUE PROBLEMS
R. P. AGARWAL
Department of Mathematics, National University of Singapore
Singapore 0511

Recently, for multipoint boundary value problems for ordinary

differential equations several constructive methods have been suggested, e.g.
the method of complementary functions and the method of adjoints [1,2], the
integral equations method [3,4], initial adjusting method [12,16], the method
of quasilinearization [5,8] etc. Here, we shall report the formulation of
another practical shooting method, namely the method of chasing for nth order
ordinary linear differential equation

x (n) + n x(n_i)
[ Pi(t) = f(t) (1)
i=l

subject to linearly independent multipoint boundary conditions

n-i
elk x(k)(a i) = Ai, i < i < n (2)
k-0

where a I ~ a 2 ~ ... ~ a n (a I < an). This method is originally developed for

second order differential equations by Gel'fand and Lokutsiyevskii and first
appeared in english literature only recently [9]. Na [ill has briefly
described the method and given different formulations for the different
particular cases of (I), (2). The general systems derived here include the
systems given by Na [ii] as special cases. The power of the method is
illustrated by solving known Holt's problem.

Since the boundary conditions (2) are assumed to be linearly independent,

at the point a i at least one of the Cik , 0 ~ k ~ n-1 is mot zero. Let cij ~ 0
then, at this point a i the boundary condition (2) can be rewritten as

n-I
x(J)(a i) = ~~ dik x(k)(a i) + ~i' i < i < n (3)

ci k Ai
where dik ffi - ~ ; 0 ~ k < n-l, k = j and a i = - -
cij c13
 268

In the differential equation (i), we begin with the assumption that

Pl(t) ~ O, so that

x(n) = - i~2Pi(t)
x(n-i)u + f(t). (4)

Now, for the boundary condition (3) we assume that the solution x(t) of (4)
satisfies (n-l)th order linear differential equation

n-i
x(J)(t) = [ dik (t) x(k)(t) + ~i(t) (5)

where the n functions dik(t); 0 < k ~ n-l, k # j and ai(t) are to be

determined.
Differentiating (~) once, we get

n-l
x(J+l)(t) = [ [dik(t)x(k+l)(t) + d~k(t)x(k)(t)] + ai(t). (6)

N e x t , we s h a l l u s e (5) t o e l i m i n a t e the term x(n-1)(t) from (6), however i t

d e p e n d s on a p a r t i c u l a r v a l u e of j and we need to c o n s i d e r four different
cases :

(i) j = 0, n ) 3 : From (5), we have

n-2
x(n-l)(t) = ~ [ x ( t ) - dik(t)x(k)(t) - ~i(t)]. (7)
i,n-1 k=l

Using (7) in (6) and rearranging the terms, we get

x(n)(t) = [di'n-2(t) + di'n-l(t)] x(t)

d2i,n_l( t )

+ [ i di,n-2(t) + dl,n-l(t)
dil(t) -~lx'(t)
dil(t)

n-2 di, n 2(t) + dl,n_l(t ) di,k_l(t) + dlk(t)]x(k)(t)

+ k=2[[ -d2,n_l(t)
' ~ dik(t) - di,n_l(t}

+ [di,n-2(t) + dl,n_l(t) ~i(t)

~i(t) - di,n_l{t)].
(8)
d2i,n_l(t)
 269

Comparing (4) and (8), we find the system of n differential equations

dl,n_l(t) = -di,n_2(t) + Pn(t)d~,n_l(t)

d~i(t ) = Pn_k(t)dl,n_l(t) - dl,k_l(t ) + Pn(t)di,n_l(t)dik(t); k=n-2,n-3 ..... 2

dil(t) = 1 + Pn(t)dl,n_l(t)dil(t) + Pn_l(t)di,n_l(t) (9)

el(t) = - f(t) di,n_l(t) + Pn(t) di,n_l(t)ei(t).

We also desire that this solution x(t) must satisfy the boundary
condition (3). For this, we compare (13) and (5) at the point a i and find

dik(ai) = dik , I ~ k ~ n-I (I0)

ei(ai) = e i.
In the rest we proceed as for the case J = 0 and obtain the following
systems

(ii) 1 ( J ( n-3

d~,n_l(t) = - di,n_Z(t) - di,j_l(t) dj.,~-l(t) + Pn_j(t)d~,n_l(t)

dlk(t) = - di,k_l(t) - di,j_l(t) dik (t) + (Pn_k(t) + Pn_j(t)dik(t))di,n_l(t)

k = n-2, n-3,..., I; k # J, j + 1

dl,j+l(t) = i - dl,j_l(t)di,j+l(t) + (Pn_j_l(t) + pn_j(t)di,J+l(t))dl,n_l(t)

dio(t ) = - di,j_l(t)dlo(t ) + (Pn(t) + pn_j(t)dlo(t))dl,n_l(t) (ii)

ei(t ) = -di,j_l(t)ei(t) + (pn_j(t)ei(t) - f(t))dl,n_l(t)

dik(a i) = dik; 0 ( k ( n-l, k = j

(12)

ei(a i) = e.

(iii) j - n-2

d~,n_l(t) - i - dl,n_l(t)di,n_3(t) + P2(t)d~,n_l(t)

 270

dlk(t) : - di,k_ l(t) + (Pn_k(t) + P2(t)dik(t))di,n_l(t) - di,n_3(t)dik(t),

l < k< n-3

d~0(t) : - dl,n_3(t)di,0(t) + (Pn(t) + P2(t)di,0(t))di,n_l(t) (13)

a~(t) = - di,n_3(t)=i(t) + (- f(t) + pz(t)ai(t))di,n_l(t)

dik(ai) = dik ; 0 < k < n-l, k # n-2

(14)
~i(ai ) = ~i"

(iv) j = n-I

d~k(t) = - di,k_l(t) - di,n_2(t)dik(t) - Pn_k(t), 1 ~k<n-2

d~o(t) = - di,n_2(t)dio(t) - Po(t) (15)

~(t) = - di,n_2(t)~i(t) + f(t)

dik(a i) = dik; 0 < k < n-2 (16)

a±(ai) = ~i"

For the particular value of j, we integrate the above appropriate system

from the point a i to a n and collect the values of dik(an); 0 < k < n-l, k # j
and =i(an). Thus, (5) provides a new boundary relation at the point a n

n-I
x(k)r~
x(J)(an) = ~ dik(a n) ~ n ) + ~i(an). (17)

Let N be the number of different boundary points i.e. a I < a 2 <...< a N =

a n (n > N ~ 2) and m(aj) represents the number of boundary relations (3)
N
prescribed at the point a J4 and hence =~Im(aj) = n. Thus, in (3) we have m(an)
J
boundary relations at the point a n and to find x(3)(an) , 0 < j < n-i we need
n-m(a n) more new relations (17) i.e. we need to integrate n-m(an) appropriate
differential systems.

Finally, from the obtained values of x(J)(an) , 0 < j < n-I we integrate
 271

backward differential equation (4) and obtain the required solution.

With the help of the following guidelines unnecessary computation can be

avoided : (a) m(an) = max m(aj), otherwise the role of the point an
I<j4N
with the point aj where m(aj) is maximum can be interchanged. (b) We need to
integrate n-m(an) times but not necessarily different differential systems,
specially because differential system does not change as long as in (3) j is
same. In fact, we can have at most n different differential systems.

For the case Pl(t) ~ 0, we rewrite the differential equation (I) as

n
[P(t)x(n-l)] ' = - [ P(t)Pi(t)x(n-i) + P(t)f(t) (18)
i=2

where P(t) = exp(f~iPl(S)ds ).

Assumption that the solution of (18) should satisfy (n-l)th order linear
differential equation

n-i
dij(t)x(J)(t) = ~ dik(t)x(k)(t) + ~i(t) (19)

with di,n_l(t) = P(t) brings the problem in the realm of the foregoing
analysis.

Example. The two point boundary value problem

x" = (2m + I + t2)x (2o)

x(0) = ~, x(~) = 0 (21)

where m and ~ are specified constants, known as Holt's problem [10] is a

typical example where usual shooting methods fall [10,13,14,15]. Faced with
this difficulty Holt [I0] used a finite difference method, whereas Osborne
[13] used multiple shooting method and Roberts and Shipman [14,15] used a
multipoint approach.

For this problem the solution representation (5) reduces to

 272

x(t) = dol(t)x'(t) + a0(t) (22)

and the ease (ill) provides the differential system to be integrated

d6(t) = I - (2m + 1 + t2)d~l(t) (23)

a ~ ( t ) = - (2m + 1 + t 2 ) d o l ( t ) a o ( t )

together with the initial conditions

d01(O) ffi 0, =0(0) = 8. (24)

We use fourth order Runge-Kutta method with step size 0.01 and obtain
dol(t) , e0(t) at t ffi 18.0|. These values are used to calculate x'(18.01) from
(22). The differential equation (20) is integrated backward with the given
x(18.01) = 0 and the obtained value of x'(18.01) using fourth order Runge-
Kutta method with the same step size. The value t = 18.01 has been chosen in
view of restricted Computer capabilities.

The solution thus obtained has been presented in Tables 1-3 for different
choices of m and 8. These tables also contain solutions of the problem
obtained earlier in [10,13,14,15]. For further details of the method and its
applications see [6,7].

References

I. R.P. Agarwal, J. Comp. Appl. Math. 5(1979), 17-24.

2. R.P. Agarwal, J. Optimization Theory and Appl. 36(1982), 139-144.
3. R.P. Agarwal, Nonlinear Analysis : TMA, 7(1983), 259-270..
4. R.P. Agarwal and S. L. Lol, Nonlinear Analysis : TMA, 8~1984), 381-391.
5. R.P. Agarwal, J. Math. Anal. Appl. I07(1985), 317-330.
6. R.P. Agarwal and R. C. Gupta, BIT. 24(1984), 342-346.
7. R . P . Agarwal and R. C. Oupta, Method of chasing for multlpoint boundary
value problems, Appl. Math. C~mp. (to appear).
8. R . E . Bellman and R. E. Kalaba, "Quasillnearlzatlon and Nonlinear
Boundary Value Problems", American Elsevier, New York, 1965.
9. I . S . Berezln and N. P. Zhldkov, Method of Chasing, in "Computing
Methods" (O. M. Blum and A. D. Booth, trans.), Vol. If, Pergamon, Oxford,
1965.
 273

i0. J. F. Holt, Comm. Asso. Comp. Machinery 7(1964), 366-373.

ii. T° Y. Na, "Computational Methods in Engineering Boundary Value Problems"
Academic Press, New York, 1979.
12. T. Ojika and W. Welsh, Intern. J. Comput. Math. 8(1980), 329-344.
13. M. R. Osborne, J. Math. Anal. Appl. 27 (1969), 417-433.
14. S . M . Roberts and J. ~. Shipman, J. Optimization Theory and Applications
7(1971), 301-318.
15. S. M. Roberts and J. S. Shipman "Two-Point Boundary Value Problems :
Shooting ~ t h o d s " Elsevier~ New York, 1972.
16. W. Welsh and T. Ojika, J. Comp. Appl. Math. 6(1980), 133-143.

Table l.m = 0,$ = 1

Present Complementary Finite Solution by Roberts and

t
Solution Functions [8] Difference Osborne [6] Shipman [7]

0 0.9999876 E O0 0. I0000000 E 01 0.i00000 E Ol 0.I000 E 01 0.i0000000 E 01

i 0.2593404 E O0 0.15729920 E 00 0.157300 E 00 0.2593 E 00 0.15729921E 00
2 0.3456397 E-Of 0.46777349 E-02 0.467778 E-02 0.3455 E-01 0.46777350 E-02
3 0.1988532 E-02 0.22090497 E-04 0.220908 E-04 0.1987 E-02 0.22090497 E-04
4 0.4595871 E-04 0.15417257 E-07 0.154175 E-07 0.4590 E-04 0.15417259 E-07
5 0.4125652 E-06 0.15366706 E-If 0.153749 E-II 0.4188 E-06 0.15374602 E-II
6 0.1413020 E-08 -0.73163560 E-15 0.215201 E-16 0.1409 E-08 0.21519753 E-16
7 0.1827268 E-If -0.75311525 E-15 0.418390 E-22 0.1821E-II 0.41838334 E-22
8 0.8863389 E-15 -0.75315520 E-15 0.112244 E-28 0.8825 E-15 0.11224343 E-28
9 0.1605597 E-18 0.413703 E-36 0.1597 E-18 0.41370659 E-36
i0 0.1082885 E-22 0.208844 E-44 0.1058 E-22 0.20895932 E-44
ii 0.2713141 E-27 0.14~078 E-53 0.12279100 E-49
12 0.2521085 E-32 0.135609 E-63 0.13487374 E-49
13 0.8677126 E-38 0.17299316 E-60
14 0.1105113 E-43 -0.25496486 E-65
15 0.5203999 E-50
16 0.9055032 E-50
17 0.5818867 E-64
18 0.4179442 E-72
 274

Table 2. m = i, = ~-i/2

t Present Complementary Finite

Solution Function [8] Differences [5]

0 0.5641878 E 00 0.56418960 E 00 0.5642 E 00

I 0.8285570 E-01 0.50254543 E-01 0.5026 E-01
2 0.7226698 E-02 0.97802274 E-03 0.9782 E-03
3 0.3020138 E-03 0.33550350 E-05 0.3356 E-05
4 0.5431819 E-05 0.18221222 E-08 0.1823 E-08
5 0.3975088 E-07 0.12367523 E-12 0.1482 E-12
6 0.1146879 E-09 -0.29349128 E-13 0.1747 E-17
7 0.1279827 E-12 -0.34242684 E-13 0.2931 E-23
8 0.5456289 E-16 -0.39134491 E-13 0.6912 E-30
9 0.8813160 E-20
I0 0.5361614 E-24
ii 0.1223266 E-28
12 0.1043287 E-33
13 0.3317918 E-39
14 0.3926980 E-45
15 0.1727057 E-51
16 0.2818780 E-58
17 0.1705581 E-65
18 0.1160366 E-73

Table 3. m = 2, B = i~

Present Complementary Finite

Solution Functions [8] Differences [5]

0 0.2500006 E 00 0.25000000 E 00 0.2500 E 00

i 0.2340787 E-01 0.14197530 E-01 0.1420 E-01
2 0.1414359 E-02 0.19141103 E-03 0.1914 E-03
3 0.4411547 E-04 0.49007176 E-06 0.4901 E-06
4 0.6261059 E-06 0.20999802 E-08 0.2101 E-08
5 0.3764660 E-08 -0.36865462 E-13 0.1403 E-13
6 0.9193294 E-I1 -0.72849101 E-13 0.1400 E-18
7 0.8879995 E-14 -0.98795539 E-13 0.2034 E-24
8 0.3334327 E-17 -0.12873356 E-12 0.4224 E-31
9 0.4809239 E-21
I0 0.2641945 E-25
11 0.5493305 E-30
12 0.4302831 E-35
13 0.1265030 E-40
14 0.1391954 E-46
15 0.5719099 E-53
16 0.8757835 E-60
17 0.4990734 E-67
18 0.3216635 E-75
STABILITY AND ERROR ESTIMATES VALID
FOR INFINITE TIME, FOR STRONGLY
MONOTONE AND INFINITELY STIFF
EVOL[7 ON EQUATIONS
O. AXELSSON
Department of Mathematics, University of Nijmegen
Toernooiveld, 6525 El) Nijmegen, Ihe Netherlands
Abstract

For e v o l u t i o n equation with a monotone operator we derive unconditional stability and

error estimates valid for all times. For the @-method, with @ = i/(24~Tg),0<m<1,6>0),
we prove an error estimate 0(T4/3), T + 0, if v ~ I/3, where T is the m a x i m a l time-

step for an a r b i t r a r y choice of the sequence of timesteps and with no further condition

on G and an estimate O(T 2) under some a d d i t i o n a l conditions. The first r e s u l t is an

improvement over the i m p l i c i t midpoint method (@ = ½), for w h i c h an order reduction
to O(~) may occur.

I. I n t r o d u c t i o n

Consider the e v o l u t i o n equation

(I.I) ~ + F(t,u) = 0, t > 0, u(0) = u 0 6 V,

du
V a reflexive Banach space, where 6 =~and F(t,-) : V ÷ V'. Here V' is the space

which is dual w i t h respect to the i n n e r p r o d u c t (-,-) in a H i l b e r t space H, with n o r m

11vll ~ (v,v) s.
We shall assume that F is a m o n o t o n e operator, i.e.
{1.2 ) iF(t,u) - Fit,v), u-v) L P ( t ) flu-v]12 V u, v c v,
where p : (0, ~) ÷ R +, i.e. P(t) ~ 0, t > 0.

A typical example is the p a r a b o l i c evolution equation

(1.3) ut = [ - (a(t,x,Vu)Vu) + g(t,u), t,£ { (0, ~) o ~ , e c IRd ,

with boundary conditions, say u = 0 on DR. Here V = LH ( )J (a C a u c h y prodUct of the

Sobolev space ~I(~)), H = L2(~) and, under certain conditions on a a n d g this is a

parabolic problem, i.e. fulfills (1.2) w i t h pit) > 0.

Other important examples are c o n s e r v a t i v e (hyperbolic) problems for w h i c h (1.2)

is s a t i s f i e d with pit) ~ 0. In the p r e s e n t paper we r e s t r i c t the analysis to the strong-
ly m o n o t o n e case, p(t) ~ P0" > 0.

Classical techniques for the d e r i v a t i o n of d i s c r e t i z a t i o n error estimates for

(1.3), uses a semidiscrete method for the d i s c r e t i z a t i o n in space, namely the
variational form

(ut, ~) + (F(t,u),]) = 0 ~ £ V h c V,
where V h is a finite element space depending on a m e s h p a r a m e t e r h.
This semidiscrete method ("longitudal method of lines") results in a system of

ordinary differential equations (ode) which is "stiff", i.e. components of t h ~ s o l u t i o n

exist, which decay iexponentially) with largely different rates.

 276

The system of ode can be s o l v e d b y m a n y m e t h o d s for stiff ode's. The d i f f i c u l t y

is in p r o v i n g error estimates for the total e r r o r of the form ClhP+c2Tq, p,q > 0,

where • is the time-step. Here C I , C 2 should be i n d e p e n d e n t of h,T. Since the d i m e n s i o n

of the ode d e p e n d s on h, c l a s s i c a l error estimates used in the n u m e r i c a l analysis of

ode, c a n n o t be applied. Furthermore, they p r o v i d e usually only a bound growing with

time t (sometimes even g r o w i n g exponentially - s e e below). We ~ a n t to derive e r r o r

estimates which are v a l i d (i.e. bounded) for all t.

We find it then c o n v e n i e n t to c o n s i d e r a "transversal m e t h o d of lines", i.e. first

discretize the e v o l u t i o n a r y problem (1.3), and more generally (i.I), w.r.t, time. A
convenient time i n t e g r a t i o n method turns out to be the implicit ("one-leg") form of

the 0 - m e t h o d w i t h 0 < O < i / 2(I+~T~, ~ > 0 for some ~, 0 < ~ < i. E r r o r e s t i m a t e s

O((½-O)T + T (3-~)/2)- v a l i d for all t can now be d e r i v e d in the strongly monotone case,

where pit) ~ P0 > 0. w i t h o u t further assumptions the optimal o r d e r we prove is O(T 4/3)

for v = I/3 a n d 8 equal to the upper bound. With some a d d i t i o n a l assumptions we prove
also the o p t i m a l order, O(T2).(At this p o i n t we r e m a r k that there is a m a r k e d differ-

ence in b e h a v i o u r of the i m p l i c i t and e x p l i c i t forms of the 0-method, in p a r t i c u l a r

for v a r i a b l e step-lengths. The latter m e t h o d m a y n o t even converge.)
To i l l u s t r a t e the p r o b l e m s with proving error e s t i m a t e s for t i m e - s t e p p i n g methods,
we c o n s i d e r the Euler (forward) met/~od,

(1.4) v(t+T) = v(t) - T F(t,v(t)), t = 0,~,2T ....

where v is the c o r r e s p o n d i n g approximation to u.
(It is o n l y for n o t a t i o n a l simplicity that we let the time step T be constant.)

Let e(t) = u(t) - v(t) be the error function. Classical error estimates, uses the
two-sided Lipschitzconstant,

(15) L:sup {LIF(t,u) - F(t,v) l[ / llu-vll}, t > 0, u , v ~ v 0 cv

where V 0 contains all functions in a s u f f i c i e n t l y large tube about the s o l u t i o n u. In
the analysis of the Euler f o r w a r d m e t h o d we have to assume that F is t w o - s i d e d Lipschitz-

bounded, i.e. that L < ~, b u t for the i m p l i c i t m e t h o d s to be c o n s i d e r e d l a t e r , we need

only a one-sided bound such as (1.2). From (i.I.) it follows
(1.6) u(t+~) = u(t) - i~ F (t+Ys, u (t+Ts))ds

and f r o m (1.4) and (1.5) we g e t

(1.7) e(t+~) = e(t) - T{m(t,u(t)) - F(t,v(t))} + TR(t,u),
where

R(t,u) Z I01 [F(t,u(t)) - F(t+Ts,u(t+Ts))]ds = f0i [~(t+Ts) - ~(t) ]ds

is the (normalized) local truncation error.

Note that
1 s (2)
(1.8) sup llR(t,u) II = T sup /0 ds I0] (t+~T) llda < ½TD 2
ut __ ,
t>0 t>0
where we use the notation

(1.9) D k = sup I u (k)(t)ll, k = 1,2 . . . .

t
t>0
and we a s s u m e that u t(k)e L ( H ) , i.e. that D k < ~ .
 277

By (1.5) and (1.7) it follows

lle(t+T) ll ~ (l+TL) lle(t)ll + T I IR(t,u)ll, t = 0, T, 2T . . . .
or, by recursion, t/T
lle(t) ll ~ (I+TL) t/TIIe(0)[I + T ~ (I+TL) J-IIIR(t-j~,u)II
j=l
or
I tL
(1.J0) lle(t) II ~ e t L l l e ( 0 )ll + ~ ( e -i) max ]IR(t,u) ll, t = T, 2T . . . .
t>0
Notice that the initial and truncation errors may g r o w as exp(tL).
By (1.8) we have llR(t,u) ll ! C T , where C depends only on the smoothness of the
solution, and not on the Lipschitz constant L. However, in most problems of practical
interest, L is large, so e v e n for moderately large values of t, the truncation error
is a m p l i f i e d by a large factor ~L-lexp(tL).
This is in p a r t i c u l a r true for stiff problems, in w h i c h case the bound (I.I0) (and
the m e t h o d (1.4), even for very small time-steps satisfying TL << i) is p r a c t i c a l l y
useless. This is in fact true for all explicit time-stepping methods.
However, we easily derive the following stability b o u n d for solutions of the
continuous problems valid.
llu(t)-w(t) II ~ e x p ( f ~ - O(s)ds) Ilu(0)-w(0)ll ~ llu(0)-w(0)ll, t > 0.
Here, u, w are solutions of (I.I) corresponding to different initial values, u(0)
and w(0), respectively.
We now face the following problems:
(i) Can we find a numerical time-stepping method for w h i c h a similar stability
bound is valid?
(ii) Can we derive discretization error estimates without a "nasty" large
(exponentially growing)stiffness factor, such as the factor in (i. I0)?
The answer to these problems is affirmative as was p o i n t e d out in [3] and [4] because
the "backward" or implicit Euler method
(1.ii) v(t+T) + TF(t+T)) = v(t), t = 0, T, 2T . . . .
fulfills these conditions.
One finds now the error bound (if e(0) = 0)
(1.12) IIe(t) ll ~ 0 0 1
sup IIR(t,u) ll ~ C T , t > 0,
t>0
where C depends only on P0 and D 2.
This m e t h o d is only first order accurate.
In this r e p o r t we discuss an extension of (1.12) to the class of 8-methods. The
results found complement some of the results in [2].

2. Stabilit~ of the @-method

We shall consider the implicit (also called one-leg) form of the @-method
(2.1) v(t+T) + TF(t,v(t)) = v(t), t = 0, T, 2T . . . . .
v(0) = u 0, where t = 8t + (i-@) (t+T) = t + (I-@)T and v(t) = @v(t) + (l-@)v(t+T) ,
0 < @ < I. For @ = 0 and @ = 1 we get the Euler b a c k w a r d (i.e. the Rothe m e t h o d (see
[7]) ,for evolutionary partial differential equations and Euler forward methods,
respectively.
 278

When F is monotone, i.e. satisfies (i .2) , it will follow that the nonlinear
equation (2.1) has a unique solution v(t+~) in V, if 8 < i.
A s is w e l l k n o w n the implicit form of the @-method can be w r i t t e n as an Euler
backward (implicit) step (t + t = t + (I-8)Y).
(2.2) v(~) + T(l-e)F(t,v(~)) = v(t),
followed by an Euler forward (explicit) step (~ + t+T)
(2.3) v(t+T) + ~ 8 k(t) = v(t),
where k(t) = F(t,v(t)).
(2.2) follows if we m u l t i p l y (2.1) by (I-0) and define v( ) as a linear function
in each interval [t,t+T]. Then v(t) = v(t). (2.3) follows if we s u b t r a c t (2.2)
from (2.1).

In practice we p e r f o r m errors, such as i t e r a t i o n and r o u n d - o f f errors w h e n

solving (2.2) and also round-off errors when computing v(t+T) from (2.3).
(In tJ~e p a r a b o l i c e v o l u t i o n equation, we also get space d i s c r e t i z a t i o n errors, when
solving (2.2).) We shall assume that these errors are Tr@(t) and Tso(t) , respectively,
where [Ir6(t) ll < CI, IIss(t) ll < C2, t > 0, and Ci, i = 1,2 are constants, independent
of T. We get then the p e r t u r b e d equations
(2.4.1) v(t) + Y(l-@)F(t,v(t)) = v(t) + ~rg(t).
(2.4.b) v(t+T) + T 8 k(t) = v(t) - Ts 0 (t),
~<~) : F(l:,v(~)),
which are the equations the c o m p u t e d a p p r o x i m a t i o n s v actually satisfy.
Multiplying (2.4.a) by 8 and s u b t r a c t i n g (2.4.b), m u l t i p l i e d by (I-9), we get
r~
(2.5) v(~) = {1-8)v(t+~) + 8v(t) + Te(t) = v(t) + Ta(t),
where ~(t) = 8rs(t) + (l-8)s6(t).

By s u m m a t i o n of (2.4.a) and (2.4.b), we find

(2.6) v(t+T) ~ ~m(t,v({)) : v(k) + xS(t),
where 8(t) : rs(t) - se(t).
For the u n p e r t u r b e d equations we have
(2.5') v([) : v(t)
and
(2.6') v(t+7) + 7F([,v([)) : v(t),
respective ly.
Let the difference be e(t) = v(t) - v(t).
We find then from (2.5), (2.5') and (2.6), (2.6'),
(2.7) e (t) = e (t) + 7@ (t] ,
(2.8) e(t+T) - e(t) + T[F(t,V({)) - F(t,v(t))] = IS(t),
respectively.
We shall assume that D(t) > p(t) > P0 > 0 in (1.2).
T a k i n g the inner product by (2.8) with e([), we find then, by (1.2) and (2.7),
(e (t+T) e(t), e(t) + T~(t)) + Tp0 I le(t) + T~(t) ll 2 < T(8, e(t) + T~(t)).
By use of the a r i t h m e t i c - g e o m e t r i c mean inequality, we find
 279

• (~,e([>) i ½o~1~Tle[i 2 + ½~%lte(~)Tl2, and

(e(t+T) -- e(t), e(t)) + ½TP0 I le(t) + T~(t) I]2 --< ½p--ITI0 181''12 - (e(t+T) - e(t), ~ ) .
By use of the inequality, IIa+bll 2 ! ½11all 2 - llbll 2 and the arithmetic-geometric
inequality once more we get
(2.9) (e(t+T) - e ( t ) , g(t)) + ~<%llg(t) ll 2 i ½ p ~ l l B I I 2
+ ½T~lle(t+T)-e(t)ll2 +--½T2--~(I+p0<I+V)II~II2,
where O < ~ < I. The chosen value of u will be specified later.
An elementary computation (see [I] shows that
(e(t+T) - e(t), e(t)) : ½[I le(t+z) l! 2 + (1-2e)lle(t+T) - e(t) ll 2 - l]e(t) il 2 ]
and
l le(t)[[ 2 = (i-8) IIe(t+T) ll 2 + 8 I le(t) ll 2 - (i-0)6 I le(t+T) - e(t) l[ 2.
Using these identities in (2.9), we find
(2.10) [I + ½~pe(1-8)]IIe(t+T)II 2 + [i-2e - ½1p0(l-e)8 - T~][le(t+T) - e(t) II 2
[i- ½7P0e]lle(t)ll2 + P01Tlleli2 + 2T2-~IIG[12 ,
I+4
where we have assumed t~at T L i is small enough so that po I ~ i.
We shall now choose 8 ~ 80, where 80 is the largest number L i, for which the factor
of the second term of (2.10), 1 - 28 - ½TP0 (i-0)8 - T v ~ 0.
We find then 80 : ½ - IO(TV) l, T ÷ 0.
By recursion, it now follows from (2.10),
lle(t) ll2 qt/Tlle(0) 2 + -i (t/T)-1 (t/T)
Tp 0 g q -J-IEi+½(l-e)TP0]-isup y2(s)
j=1 s>0
where
(2.11) ¥2(s) : ]le(s) 2+2pol-~li~(s) l12 '
and

q = (1 - ½e~p ~ / [1 + ½(1-8)TP0]
Since e < ½, we ha'~ q < 1, and we find

lle(t) ll 2 < q t / T ! l e ( 0 )
ll2 + p0212 + (i-8)Tp 0] sup y2(s), Vt > 0.
S>0
Hence, the 8-met~hod is unconditionally stable (independent of the stiffness and
of "r), if 0 < 80.
We collect the result found in
Theorem 2.1. (Stability.) If (i.I) is strongly monotone~ i.e. p(t) >--P0 > 0 in (1.2),
and if 8 < 80, where 80 is the largest number < i, for which
i--28-½TP0(l-8)8-T ~ >_ 0, 0 _< V < i, then
ile(t) II 2 <_ gt/Tl[e(0) I12 + P0 ~'
2[2 + (I-8)<P0 ] sup T2(s) gt > 0
where y(s) satisfies (2.11). S>0

Here e(t) = v(t) - v(t) is the perturbation error, v(t) is the solution of the
perturbed equations (2.4.a,b), and v(t) is the solution of the unperturbed 8-method (2.
Corollary 2.1. If e(0) = 0, then
(2.12) lle(t) ll <_p01 [2+(l-8)TPo ]½ su~Iy(s) l, Vt > O.
This generalizes the stability part of (1.12) to the implicit class of e-methods.
 280

3. Truncation errors
It remains to consider the truncation errors for the @-method. For the solution
u of (I.i) we have
(3.1) u(t) = u(t) + T ae(t),
where, by an elementary computation,
(l-e)s
1
~8(t) = -(I-@)T f0 ds / fl(t-or)d~
-8s
Hence
(3.2) sup II%(t) ll = ½e(~-e)~ D2.
t>0
Similarly,
(3.3) u(t+T) + TF(t,u(t)) = u(t) + rSo(t),
where,
(3.4) ~o(t) = ~-1(u(t+1) - u(t) - ~u(t))
1 S
= T /0 ds fl-O ~](t+oT)do

= T f01 ds f½S fi(t+OT)dd + T I 0 ds f -0fi(t+~T)do

= T f0½ ds [f0 ~(t + (½+~)T)do - fi(t+(½-O))T]do + T folds /l_Ofi(t+oT)do.½

Hence, if ut(3)e L (H), i.e. suplIut(3) (s) II < ~, then

1 2 s>0
(3.5) sup IIBe(t) II < ~ T D 3 + ~I½-81D 2.
t>0
Let the time-discretization error, E ( t ) = u(t) - v(t). By (2.5'), (3.1) and (2.6'),
(3.3) and using the estimates in section 2, we get by Corollary (2.1), for the strong-
ly monotone case,

(3.6) [IE(t)l[ < p-112 + (1-e)zp ]½ sup lYe(t)I, 0 < e < e 0,

- 0 0 t>0
where y~(t) = II~e(t) II 2 + 2poTl-v I {os(t) ll 2.

Hence, by (3.2) and (3.5),

(3.7) 1 2D 3 +
IYo(t) I = ~-~ Tt½_elD2 + / p-0-/ 2 T(3-V)/28 ( l _ 0 ) D 2 .

With 0 = I/(2+~T V) <_ 00 (i.e. with ~ a large enough positive number) , (3.7) implies

Iy0(t) i = lo(r2)l + io(~i+~)J + I0(~(3 ~)/2)I, ~ +0

Its order is highest, namely OfT 4/3) , if we choose v = i/3.

We collect these results in
Theorem 3.1. (Discretization error) The discretization error of the e-method with
@ = I/(2+~V) <-- 80' 6 > 0, where O 0 is defined in Theorem 2.1, satisfies

ilE(t)[l < P I[2 + (l-@)Tpo]½ sup Iye(t) i : 10(TI+v) i, if 0 < v <-3

0 !
t>0 V t > 0,
1
10(y(3-V)½)I, if7< V < i
for any solution u of a strongly monotone problem (i.i), for which u~3)e L (H).
Its order is highest, IIE(t) li = !O(y4/3) I, if v = I/3.
 281

Remark 3.1. It follows readily from (3.4), that Theorem 3.1 remains valid if we
replace the regularity requirement, u t(3) H, with the weaker requirement that u (2)
(2)
is H61dercontinuous with exponent ~. In fact it suffices that u t is H61der-
continuous in the interior of each interval (t,t+Y).

Remark 3.2. Theorem 3.1 remains valid for any choice of timesteps Tk, constant or
variable, for which T k < CT, for some positive constant C.
In some problems we have to adjust the timesteps to get convergence o r fast
enough convergence, because some derivative of u of low order can be discontinuous
at certain points. It may for instance happen that F in (1.1) is discontinuous for
certain values of t.
In such cases we w a n t to adjust the timesteps so that those values of t become
stepping-points. Hence the result in Theorem 3. i, although not of optimal order as
we shall see, is of particular importance for cases where we have to change the time-
steps in an irregular fashion.
We shall now p r e s e n t an optimal order, O(~ 2) , result, but valid only if the
time~teps are essentially constant.
Consider first the equations (2.4.a,b) for variable parameters 8 = 8 k and T = Yk'
k = 0,I . . . . . For the solution of (I.i) we get then truncation errors ~ = R(tk,Tk,@ k)
and S k = S(tk,Tk,0 k) , defined by
(3.8.a) u(t) + Tk(l-Sk)F(t,u(t)) = u(~) - Tk~ ,

(3.8.b) U(tk+~) + Tk0 k F(t,u(t)) = u(t) + TkSk,

where { = t k + (1-@k)T k and t k is the k'th stepping point.

Guided by a trick in [5] for the implicit midpoint method (i.e. (2.2), (2.3)
with @ = ½) we define
(3.9) u(t k) = u ( ~ ) - Tk~.
Then (3.8.a,b) takes the form
(3.10.a) u(t) + Tk(1-@k)F(t,u(t)) = 6(~),

(3.10.b) u ( ~ + I) + Tk0kF(t,u(t) = u(t)+ Tk~@(tk),

where

(3.11) ~8(~) = Sk - ~ k + l ~ + i / T k

Multiplying (3.10.a) by 8 k and subtracting (3.10.b), multiplied by (i-8 k) , yields

(3.12.a) u(t) = @kU(tk) + (l-@k) + ( l - e k ) U ( ~ + l ) - (l-Sk)~@(tk)

Summation of (3.10.a) and (3.10.b) yields

(3.12.b) u(tk+ I) + TkF(t,u(t)) = u(t k) + Tk~Stk).

We define u(~) = u(~) (we let 0 < @k' so ~ < t ~ ~ + i ) , so by (3.12.a),

(3.13) u(t) = ~(tk) + ~k~@(tk) , where ~0(tk) = -(l-0k)~0(tk).

Note that (3.13) and (3.12.b) have the same form as (3.1) and (3.3), respectively.
 282

To estimate ~8 and ~@, we need to estimate Bk and S k. By (3.8.a) and (3.8.b) we find

(3.14) ~ = (t-@k) [~(t) - ~0

rl d(tk + (]-Sk)<kS)ds]
I ii +
= (l-Sk)2Z k I 0 ds s U(tk (l-Sk)~ka(dg)
and
(3.15) S k = Ok[/~ u(tk+Tk-OkYkS)ds - ~(t)]
1 i1
= 8~k f0 ds s u(tk+Tk-SkTk O)dO-

By (3.11), (3.14) and (3.15) we find

li~o(tk) ll =
~k11(%~k)2
-
2 2
(l_Sk+l) Yk+iI ½D 2
+ O(T~)D3
if e~ ~ Tk ~ CT for some p o s i t i v e cohstants c,C.

Hence !I~@(tk) II = O(72), < ÷ 0, k = 0,I .... if (SkTk)2 - (l-Sk+l)2T~+l : 0(T 3) or

8k
Tk+l 1 ~ + O(T2)
-Sk+ 1 k
Remark 3.3. Since for stability reason, Bk < ½, we see that if
@k
(3.16) Tk+ I = l_Sk+l Tk,

then {Tk } is a decreasing sequence. Frequently, in practice we want to choose small

steps in ~le initial (transient period) and then larger steps, i.e. contrary to (3.16).
Consider now for simplicity the case "[k = T, 8 k : Q = I/(2+~7), C > 0, k = 0,i,...
Let E(t) = u(t) - v(t). It follows as before (see 3.6) and Theorem 2.1) that

llE(t) II 2 ! qt/~[iE(0) [i 2 + p02- [2 + (l-8)Tp0] O(<2)D3

We have E(0) = u(0) - v(0) = u(0) - u(0) = ~0R0 , i.e. by (3.9) and (3.14),
tl~(o) ll = O(T2)D2.
Similarly, by (3.1),
flu(t) - v(t) l I ~ flu(t) - u(t) l I + ]lu(t) - v(t) l I = O(T2)D2 + llE(t) II.
We collect these results in
8k
Theorem 3.2. If Tk+ I l_Sk+l lk + O(~2)'

@k : i/(2+6kTk)' 6 > ~ > 0, k = 0,1 ..... then the @-method (2.2), (2.3) has a
k 2 (3)
discretization error OCT ), valid at all stepping points tk,if u t ~ L ( H ) and if
(i.i) is strongly monotone.

Remark 3.4. In [2], it is proven an optimal order, O(T 2) estimate, valid for arbitrary
variable time-steps, if in addition to the assumptions in Theorem 3.1, we assume that
= i, that {l~F/~tll is not large and that the Gataux derivative ~F/~u exists and
satisfies: II~F/$u u(2)
iIt is of the same order as D 3 (i.e. not larqeo for smooth
solutions). Note that for a linear problem u t = Au with constant operator A, we have
~F/~u u (2)= A3u = u (3) For a more general parabolic problem, we have typically that
t t "
sup [[ ~F/~u U t(2)1l is of the order o f s u p [lU t(3) I[ when the solution (and its
t>t 0 t~t 0
 283

derivatives) is smooth for t ~ to, b e c a u s e then u has e s s e n t i a l l y components along

the e i g e n f u n c t i o n s corresponding to the s m a l l e s t eigenvalues of the J a c o b i a n 3F/$u.

In the results presented in the p r e s e n t paper, we h a v e h o w e v e r not even assumed the

existence of the Jacobian.

4. C o n c l u s i o n s
o
In [6] was shown b y c o n s i d e r i n g the p r o b l e m u(t) + l(u-g(t)) = g(t) , t > 0

for i very large, that the a c c u r a c y of the a p p r o x i m a t e solutions obtained o f t e n are

unrelated to the c l a s s i c a l order of the m e t h o d used.

For the i m p l i c i t midpoint method (i.e. (2.1) with @ = ½), this e r r o r o r d e r

reduction is e a s i l y seen to be c a u s e d by that the d a m p i n g factor q in q ~ e o r e m 2.1

approaches the value -I. For (almost) constant steplength this causes a cancellation
effect a n d the g l o b a l error remains O(T2), but for I and/or 7 variable this is not

the case and the o r d e r is only O(1) in general.

We have shown that by c h o o s i n g @ = I/(2+~), ~ > 0, 0 < ~ < l, a h i g h e r order
(at least O(~4/3)) can b e achieved. This is due to the d a m p i n g wi~h a factor q,

where lql ~ e/(l-8) for i large.

Under additional assumptions and with ~ = i we can also get an error 0(~2). Hence

the error order is never w o r s e that for the i m p l i c i t midpoint rule.

It is a n t i c i p a t e d t h a t a similar modification of higher order Lobatto type

implicit Runge-Kutta methods can give a less severe order reduction ~lan if they are
not m o d i f i e d (cf [63 a n d [4]).
 284

References

i. O. Axelsson, Error estimates for Galerkin methods for quasilinear parabolic

and elliptic differential equations in divergence form, Numer. Math. 28,
1-14 (1977).

2. O. Axelsson, Error estimates over infinite intervals of some discretizations of

evolution equations, BIT 24 (1984) , 413-424.

3. G. Dahlquist, Error analysis for a class of methods for stiff nonlinear initial
value problems, Numerical Analysis (G.A. Watson, ed.), Dundee 1975, Springer-
Verlag, LNM 506, 1976.

4. R. Frank, J. Schneid and C.W. Ueberhuber, The concept of B-convergence, SIAM J.

Numer. Anal. 18(1981) , 753-780.

5. J. Kraaijevanger, B-convergence of the implicit midpoint rule and the trapezoidal

rule, Report no. 01-1985, Institute of Applied Mathematics and Computer Science,
University of Leiden, The Netherlands.

6. A. Prothero and A. Robinson, The stability and accuracy of one-step methods,

Math. Comp. 28(1974), 145-162.

7. K. Rektorys, The Method of Discretization in Time and Partial Differential Equations.

D. Reidel Publ. Co., Dordrecht-Holland, Boston-U.S.A., 1982.
RECES I' RESULTS
IN THE APPROXIMATION
OF FREE BOUNDAR S
F. BREZZI
Instituto d i A n a l i s i N u m e r i c a del C. N. R., Universita d i P a v i a
C.so Carlo Alberto, 5 - 2 7 1 0 0 Pavia, Italy

§ I. We p r e s e n t here a short survey on r e s u l t s recently obtained in

the approximation of free boundaries. For e x a m p l e s o f free boundary
problems that are interesting in physics and e n g i n e e r i n g we r e f e r for
instance to 111,171,181. Here we s h a l l stay at a very abstract level,
without considering, essentially, the nature of the free boundary pro-
blem u n d e r c o n s i d e r a t i o n nor the type of discretization which is employ
ed. In t h a t we a r e r a t h e r following 131 o r t h e f i r s t part Of 191. For
practical cases in which the following results a p p l y we r e f e r to 141,
110i and 191.
In t h e n e x t section we p r e s e n t the framework in which the theory
will be d e v e l o p e d and i n the third section we s h a l l present some ab-
stract results, most of them w i t h o u t proof. The p r o o f s can be f o u n d in
the corresponding references.

§ 2. For the sake o f simplicity we s h a l l consider the following "model"

situation. We a r e g i v e n a bounded domain D i n ~n with piecewise Lips-
chitz boundary (to fix the ideas). We a r e a l s o given a function u(x) in
C°(D). The f u n c t i o n u(x) will be t h e solution of our free boundary pro
blem. The n a t u r e of the problem itself is immaterial at this stage.
We assume t h a t

(I) u(x) ~ 0 VxeD

and we assume t h a t the continuous free boundary F is characterized by

(2) F:= D N ~ ( D +)

where

(3) D+:= {xix~D , u(x)>O};

We assume , f i n a l l y that we have c o n s t r u c t e d a sequence {Uh(X)~ , f o r

O<h~h of "approximating solutions" which converges to u(x) in C°(D):
0 • - -

Again, the procedure employed to construct {u h } i s irrelevant at the

moment. We s e t

(4) E p ( h ) : = iIu-u h IiLp (D) I sp ~ .

and we r e m a r k t h a t we have a l r e a d y assumed

 286

(5) lim E (h) = 0

h+O
We would l i k e to construct a " d i s c r e t e free boundary" Fh as in (2) and
then to estimate the distance of Fh from F in terms of Ep(h), defined
in (4). In order to make our l i f e even e a s i e r , we assume t h a t , as in
(I),

(6) Uh(X)~O Vx ~D, Vh~h o

and we s e t , as a f i r s t trial,

(7) D+h::{xlx ~D' Uh(X)>O}

(8) Fh:=DN@(D~)

Unfortunatly, e l e m e n t a r y examples show t h a t F h can be v e r y f a r from F

even f o r u h v e r y c l o s e t o u. For i n s t a n c e i f D = I - 1 , 1 1 and u ( x ) = ( x ) +
(that is u(x)=O f o r x<O and u ( x ) = x f o r x>O) we have F : { O } . I f now
Uh(X)=U(x)+hS(x+1) the Fh={-1} no m a t t e r how s m a l l is h o r how b i g
is s. It should be c l e a r now t h a t the setting (I)...(8) does n o t a l -
low the p r o o f o f any bounds on t h e d i s t a n c e o f F h from F. In t h e n e x t
section weshall present a few r e m e d i e s t h a t (under suitable additional
assumptions) have been p r o p o s e d t o i m p r o v e t h e s i t u a t i o n .

§ 3. The f i r s t trial in this direction has been done a b o u t t e n y e a r s

ago i n 121. Assume t h a t g(h) is a function o f h such t h a t

(9) E (h)<g(h) O<h~hl~h 0

(10) l i m g(h)=O
h+O
and s e t

(11) D+g={xlx
h
~ D' Uh(X)>g(h) } "

Then we have j21.

Theorem 1-Under t h e above a s s u m p t i o n s we h a v e , for h<h 1

+
(12) Dh,g C D+,

and for all x ~ D +, t h e r e exists h2>O such t h a t xED + for h<h 2.

' h,g - -

The p r o o f is i m m e d i a t e . We p o i n t out that, in other words D+

' h,g
converges t o D+ " f r o m inside". At o u r k n o w l e d g e t h e o r e m 1 i s still
among t h e b e s t r e s u l t that one can o b t a i n without additional informa
tion.
A tipical additional information that one can g e t , i n many c a s e s ,
is the behaviour of u(x) in D+, near t h e f r e e boundary. To f i x the
ideas, assume from now on t h a t F is a smooth ( s a y , CI ) surface. For
 287

any p o i n t x of F one can look at the r e s t r i c t i o n o f u along the d i r e -

c t i o n normal to F and p o i n t i n g i n t o D+. I f

(13) u(x)~c(x-x) s for Ix-x}~a (s ~ ~ )

with c,a and s i n d e p e n d e n t o f x we s h a l l say t h a t "u(x) grows l i k e ds

near F" (d i s for distance to the f r e e b o u n d a r y ) . For i n s t a n c e i n the
particular case o f a n i c e o b s t a c l e p r o b l e m , i t has been proved i n 151
t h a t u ( x ) grows l i k e d 2 near F. I f one can p r o v e a s i m i l a r p r o p e r t y
f o r the a p p r o x i m a t i o n s Uh(X) , one can use such an i n f o r m a t i o n t o g e t
e s t i m a t e s on the d i s t a n c e o f F h from F. T h i s has been done in 141 f o r
the case o f a n i c e o b s t a c l e problem w i t h a p i e c e w i s e l i n e a r f i n i t e
element approximation that satisfies the d i s c r e t e maximum p r i n c i p l e .
The r e s u l t in 14J i s e s s e n t i a l l y that the d i s t a n c e o f F h from F be-
haves l i k e (E ( h ) ) ½. T h i s i d e a has then been e x t e n d e d i n It01 t o the
one-phase S t e f a n problem in s e v e r a l d i m e n s i o n s , and i n I l l l to para-
bolic variational inequalities o f o b s t a c l e t y p e . The m a j o r drawback
of this technique, however, i s that it is often very difficult to
prove growth p r o p e r t i e s for the d i s c r e t e solutions Uh(X), w h i l e the
behaviour of u(x) itself is e a s i e r to a n a l y z e (see J 8 [ , 1 7 I f o r seve-
r a l growth p r o p e r t i e s proved on the c o n t i n u o u s p r o b l e m ) .
A new and i n t e r e s t i n g s e t o f r e s u l t s has been t h e n o b t a i n e d i n
19 I by c o m b i n i n g , somehow, t h e two p r e v i o u s t e c h n i q u e s : to change
D+h i n t o a s u i t a b l e Dh,g
+ and to use some g r o w t h p r o p e r t y on u ( x ) .
In o r d e r to g i v e the f l a v o u r o f t h i s p r o c e d u r e we s h a l l p r e s e n t
here two r e s u l t s in t h i s direction. More d e t a i l e d results and
examples can be found in [9[.

Theorem 2 191 With t h e n o t a t i o n s and a s s u m p t i o n s (I)...(II), if

u(x) has the growth p r o p e r t y (13) for some s>O, t h e n t h e r e e x i s t
c>O and h3>O such t h a t for all h with O<h~h 3

(14) dist (Fh,g,F)~(cg(h))I/s,

where

(15) F h , g : = D N B ( hD,+g )
and (14) means

(16) V x h E F h , g 3 x EF such t h a t IX-Xhl~(cg(h)) I/s

P r o o f - Let Rh e F h , g . T h i s i m p l i e s Uh(Xh)=g(h)>E ( h ) , and t h e r e f o -
re Xh ( D+" From (13) one has now t h a t , f o r h small enough,

(17) U(Rh)aClR-XhIs
 288

for some x EF. Since Uh(Xh)=g(h) one has from the t r i a n g l e i n e q u a l i t y

(18) U(Rh)~g(h)+E~(h)<29(h)
and (16) follows from (17) and (18).

Theorem 3 Under the same assumptions of theorem 2, i f

(19) g(h)>E (h) sp/(1+sp)

P
then there e x i s t c>O and h4>O such that

(20) meas(D+~Dh,g)~C(g(h))
+ I/s

where in (20) the symbol A indicates as usual the symmetric difference

of sets.

The proof can be found in 191.

We would l i k e to conclude with a somehow phylosophical remark (see 131).

In general, condition (13) implies that the global r e g u l a r i t y of u(x)
in ~ i s , at best
(21) u ~cS(O).
I f one uses, f o r i n s t a n c e , f i n i t e element methods in o r d e r to appro-
ximate u, one has, at best (see e . g . 161):

(22) E~ (h)= flu-Uh IIc o~chr flu II r O~rsmin(k+1 ,s)

C
where k is the degree of the p o l y n o m i a l s . Using now theorem 2, say,
w i t h g(h)=2E (h) one has from (14)
(23) d i s t (Fh,g,F)~c(E (h))I/s

and from (23) and (22):

(24) d i s t (Fh,g,F)cch r/s (at best).

One can now make the following observations.

I) Since r~s in (24), the error cannot beat the mesh size.

2) For s "big" one need a "big" k (that i s , polynomials of high

degreee) so that the case r=s can be achieved in (22) (and hence
in (24)).
3) For s "small" the error that one gets from (24) is surprising good.
Unfortunately, for i r r e g u l a r u, (22) is often d i f f i c u l t to prove
in practical cases.
 289

References

ill C. Baiocchi - A. C a p e l o : " V a r i a t i o n a l and q u a s i v a r i a t i o n a l inequa-

lities. A p p l i c a t i o n s to f r e e boundary p r o b l e m s " , J. W i l e y , 1984.

i21 c. Baiocchi G. P o z z i : " E r r o r e s t i m a t e s and f r e e - b o u n d a r y conver

gence f o r a f i n i t e difference discretization of a parabolic vari
ational inequality", RAIRO Numer. Anal. 11, 4 (1977), 315-340.
!31 F. B r e z z i : " E r r o r e s t i m a t e s in the a p p r o x i m a t i o n o f a f r e e bounda
ry u, Math. P r o b l . in S t r u c t u r a l A n a l y s i s , ( G . Del P i e r o - F . Mauceri
Eds.), S p r i n g e r , CIS, Courses and L e c t . n. 288 (1985), 17-23.

!41 F. Brezzi - L. C a f f a r e l l i : " C o n v e r g e n c e o f the d i s c r e t e f r e e

boundaries f o r f i n i t e element a p p r o x i m a t i o n s " , RAIRO Numer. Anal.,
17 (1983), 385-395.

i5I L. C a f f a r e l l i : " A remark on the H a u s s d o r f f measure of a f r e e

boundary, and the convergence of c o i n c i d e n c e sets'~ B o i l . U.M.I.
(5), 18 A (1981), 109-113.

161 P. C i a r l e t : " T h e Finite Element Method f o r E l l i p t i c Problems'~

N o r t h - H o l l a n d (1978).
171 A. F r i e d m a n : " V a r i a t i o n a l P r i n c i p l e s and Free Boundary Problems",
Wiley, New York, (1982).

181 E. Magenes, E d i t o r ( 1 9 8 0 ) , " F r e e Boundary Problems'~ 2 v o l . ,

Proc. Sem. ( P a v i a , 1979), I s t i t u t o N a z i o n a l e di A l t a M a t e m a t i c a ,
Roma.

191 R.H. N o c h e t t o : " A note on the a p p r o x i m a t i o n of f r e e boundaries

by f i n i t e element methods", To appear in R . A . I . R . O . Anal.Numer.

1101 P. P i e t r a - C. V e r d i : " C o n v e r g e n c e o f the approximated f r e e - b o u n

dary f o r the m u l t i d i m e n s i o n a l one-phase Stefan p r o b l e m " , ( t o ap-
pear i n Computational Mechanics).
1111 P. P i e t r a - C. V e r d i : " C o n v e r g e n c e o f the a p p r o x i m a t e f r e e boun-
dary f o r the m u l t i d i m e n s i o n a l one-phase Stefan p r o b l e m ~ ( t o ap-
pear).
FINITE ELEMENT SOLIYrION
OF A NONLINEAR DIFFUSION
PROBLEM WITH A MOVING BOUNDARY
L. CERMAK and M. ZLAMAL
Computing Center of the Technical University in Brno
Obrdncf~ mir~u 21, 602 O0 B~'no, Czechoslovakia

In r e c e n t years two-dimensional process simulators for m o d e l l i n g

and s~mulation in the d e s i g n of V L S I semiconductor devices have ap-
peared (see, e.g., Maldonado [2]). The underlying mathematical problem
consists in s o l v i n g numerically the following boundary value problem:
~u
(i) 8~ - V . [ D ( u ) V u l in ~(t), 0 < t < T,
~(t) = {(x,y)I~(y,t) < x < L0, 0 < y < B} ,

8u
(2) ~--nl = 0, 0 < t < T, F(t) = {(x,y) Ix = ~(y,t),
8~(t)-r(t)
0 < y < B} ,

(8) D(u)~ = y~n u on r(t), 0 < t < T ,

(4) u(x,y,0) = u*(x,y) in n(0).

Here u is the unknown concentration of an i m p u r i t y , D(u) is the con-
centration dependent diffusion coefficient (0 < d 0 ~ D ( u ) ~ d; z ~ u ~ O~
1 ~u
~(y,t) is a g i v e n function (0 ~ ~ ~ ~ L0) , ~ is the d e r i v a t i v e in the
direction of the o u t w a r d normal, ¥ is c o n s t a n t , $n is the rate of t h e
motion of ~(t) in the direction of the o u t w a r d normal a n d u* is the
given initial concentration.
If u is a s u f f i c i e n t l y smooth solution of (i) - (4) (we r e m a r k
t h a t we do n o t k n o w any result from which existence of a s o l u t i o n of
(i) - (4) follows), t h e n by m u l t i p l y i n g (I) by v 6 H l ( ~ ( t ) ) and inte-
grating over G(t) we get

(5) ~t e (0,T) ?Su

~-~, v ~
JL2(~(t) ) + a(u,t;u,v) = 0 ~v 6 V(t) ~ Hl(n(t)~
here
a(w,t;u,v) :: fD(w)Vu.Vvdxdy - y f~nUVdxdy.
n(t) r(t)
We use (5) f o r d e f i n i n g the semidiscrete solution. First we c o n s t r u c t
a suitable moving triangulation of ~ { t ) . We consider the one-to-one
mapping of the r e c t a n g l e Q = (0,L0) X (0,B) on ~(t):

(6) x = F(a,8,t) ~ ~(S,t) + ~[ l - L 0 1 ~ ( B , t ) ] , y = 8.

We cover ~(0) by triangles completed along F(0) by c u r v e d elements in a

manner described in Z l ~ m a l [4]. Let Pk = (xk,Yk), k = l,...,d, be the
 292

nodes of this triangulation and let Qk = (~k'6k) be their inverse

immages in the mapping (6), i.e.

_ x k- ~(6k,0)
- ' 8k = Yk "
~k I_L~I (~k,0)

The triangulation T(t) of ~(t) is determined for t > 0 by the nodes

Pk(t) = (Xk(t)lYk), xk(t) = F(mk,Bk,t). The elements of T(t) are again
triangles or curved elements. As shape functions we use linear polyno-
mials. We denote by Vh(t) C HI(£(t)) the set of all trial functions
and by Wk(X,y,t), k = l,...,d, the basis functions of Vh(t).
The semidiscrete solution is assumed in the form
d
U(x,y,t) = Z Uk(t)wk(x,y,t)
k=l
and determined by

(7) ~t e (0,T) (~t U'V)L2(g(t))+ a(U,t;~,v) = 0 ~v e Vh(t),

U(x,y,0) : U~(x,y) .

U ~ C Vh(0 ) is a suitable approximation of u ~.

If we denote by ~(t) the d-dimensional vector (Ul(t),..., Ud(t))T
by M(t), R(t) and K(U,t) the d X d matrices
d 8Wk d
M(t) : {(wj,Wk)L2(~(t))}j,k=l, R(t) = [(wj,~-~--)L2(~(t))}j,k= l,

K(U,t) : {a(Utt;wj,wk)]~,k:l

then the matrix form of (7) is

M(t)U + [R(t) + K(U~t)]U = 0 ,
48)
~(o) : u~
Here U = d U and the matrices M and K are standard mass and stiffness
matrices, respectively. The matrix R is unsymmetric.
We discretize (8) in time. For simplicity, we use a uniform parti-
tion of (0,T) : t i : iAt, i = 0,1,...,q. In the sequal U l, Ml,...
means ~(ti), M(ti) , . . . . Now we set t : ti+ 1 in (8), replace _O i+l by

At-IA~i, A~i = ~i+l_ _~ , and linearize the nonlinear term in (8). We

get
d
= ..i i+l
(9) si+iAui_ + At[R i+l + K(~,ti+l)J_'~+l : ~, ~i k=iZ~kWk ,

U° = U~

For practical computations it is necessary to do one more step: to

replace curved elements by triangles and to compute all matrices
 293

numerically. Also, we could apply the Crank-Nicholson approach for

solving (8) or, more generally, the 8-method. We have restricted
ourselves to justify the procedure defined by (9).
We consider a family {T~] of triangulations of ~0from which a
family [Th(t)] of the triangulations of ~(t) is constructed as de-
0
scribed above. Let hK0 be the greatest side of an element K06 T h and
h = max
K0eT~ hK0"

W e consider a family ITS} such that h - 0 and the minimum angle condi-
tion is satisfied. We have proved the following main results:
I. Let
b(w,t;v,v) a 0 vv E V(t), t E (OAT>
where
1
b(w,t;u,v) : a(w,t;u,v) - ~ £(t)I$ n u v d r "

Then for At sufficiently small,

At ~ At~ where At~ does not depend
I+l i+l ~i
on h and on the index i, the matrices M + At(R + K(U ,t~.~)] ,
i : 0,...,q-l, of the szstems (9) are regular so that U I, i : l,...,q
are uniquely determined. Furthermore, the scheme (9) is unconditionally
stable in the L2-norm, i.e. for At ~ At 0 we have
(i0) max ]IuiI[L2 ~ c~uOII
iSiSq (a i ) L2(n 0 )

where C does not depend on At and on h.

2. Let the form b be uniformly V(t)-elliptic, i.e.
b(w,t}v,v) Z b011vll2 vv 6 V(t), t E (0,T)
Hl(a(t))
and let 6n ~ 0. Then for At sufficiently small, At ~ At 0 where At 0
does not depend on h, there holds

(ii) max IIu i- q ui 1/2

iSi<q UI~L2(£1) + {Ati:l
Z llui- llHl(£i)] -<

<- c(llu 0- U 011 + h +At) .

L 2 (~0)

The proof starts from the variational formulation

U i+l
(U i+l- 5 i ,V)L2(£i+ I) - A t ( ~ ,GI+Iv)L2( ai+i )+ Ata(~i,ti+l;

(12) oi+l,v) = 0
vv 6 ~zi+l i -- 0 ..,q-i
~h ' '" '
uO = U~~ •

Here G is a computable function defined on each element by means of the

 294

map which maps uniquely this element on the reference one. From (I 2)
the existence and an unconditional stability of the scheme can be
proved. The error estimatinq is b a s e d on the Ritz apDroximation
6 Vh(t ) defined by
d(u,t;u - ~,v) = 0 ~ v ~6 V h ( t )
where d(w,t;u,v) = b(w,t~u,v) - ~r(t)/~nUVdr" U - ~ is e s t i m a t e d using

a technique which in a c a s e that the boundary does not move is e s s e n -

tially that of Wheeler [ 3], Dupont, Fairweather, Johnson [ i] a n d ZI~-
mal [ 5] . I n s t e a d of estimating ~{(u-~)we estimate Dt(u-~) where the
operator D t is d e f i n e d b y D t -- G~-~ + ~-~ . If the boundary does not

move, G -- 0 a n d D t -- a__
%t "

R e f e r e n c e s

[ i] T. Dupond, G.Fairweather and JoP. Johnson, Three.Lzv@~ Gal@~z~ ~ 6 ~ 0 ~ for

Pa/tabo~i6 Eqb~tions° SIAM J. Numer. Anal. 11 (1974), 392-41Oo
[ 2] C.D. Saldonado, ROMANS II, A T~go-Dim£~io*~ P~0c@~6 Simu2z~tor for Mod~in 9
and Simulation in the Design of VLSI Devices. Applied Physics A 31 (1983) ,
119-138.
[ 3] M.F. Wheeler, A p r ~ o ~ Lo Err0r Est/mat@w$ for G ~ / ~ Approximations to
P~abo~6 Pa~l ~ff~r~nti~ Equ~oY~5. SIAM J. Numer. Anal. iO (1973),
723-759.
[ 4] S. Zl~mal, Cu/tu£d E l ~ e n ~ in the F i ~ £ Element Method. SIAM J. N~mer. Anal.
iO (1973), 229-240.
[5] M. Zl~mal, Finite Eleme~ Method~ for Nonlinear Parabalic Equations. R.A.I.R.O.
Anal~ Numer. 11 (1977), 93-107.
ANAINSIS OF THACKER'S METHOD
FOR SOLVING THE L ARI D SHALLOW
WATER EQUATIONS
J. DESCLOUX, R. FERRO
EPFL Department of Mathematics
-

CH 1015 Lausanne, Switzerland

I . INTRODUCTION.

In t h e i r s i m p l e s t form, the shallow water equations read

at~(x,t) = - b(x)vH(x,t) + fRU(x,t), x 6 Q, t ~ o, (I.I)

atH(x,t) = - #.O(x,t), x 6 Q, t ~ o, (I.2)

U(x,t).n(x) = o, x E 8~, t ~ o, (1.3)

~(X,O) = ~o(X), H(x,o) = Ho(X), x ~ ~. (1.4)

Here , ~ c ~2 is a bounded open domain with C= boundary a~ and closure ~. ~ is the

outwards unit normal to a~. ~(x,t) = (O](x,t),U2(x,t)) is a two components vector
related to the average horizontal velocity. H(x,t) is the height of the surface of
the basin. Up to a constant factor, b(x) is the depth of the basin. We shall assume
that b ( . ) i s a C ~ ( ~ ) s t r i c t l y positive function, f which represents the intensity of
the Coriolis forces is taken to be constant. ~o and Ho are given i n i t i a l conditions.
R is the (-~/2) rotation operator acting in ~ , i . e . R(Xl,X2) = ( ~ , - x l ) . The tan-
gential vector ~ at a~ is given by ~ = - R~.

Equations ( l . l ) , ( l . 2 ) can be easily set in the framework of the theory of semigroups.

L e t ~ = (L2(~))3 be the Hilbert space with scalar product

((u,g),(v,hl~:
~ ~ b-~
I ~(x).~(x) + g(x)h(x) (].5)

and associate norm II.J~where ~ = (uz,u2), ~ = (vl,v2). We define the operator L

with domainS(L) by the relations:

~(L) = {(u,g) (~ v.u 6 L:~(~), u-n = o on a~, g 6 HI(~)} (1.6)

L(u,g) = ( - b v g , - v . u ) , (1.7)

where Hk(c~) i s the c l a s s i c a l Sobolev space of order k. One can v e r i f y t h a t L is a

skewadjoint operator, i . e . the adjoint of L is -L which implies in particular for
(u,g) and (v,h) i n ~ ( L ) the following relation:

(L(u g ) , ( v , h ) ~ = - ((u,g),L(v,h)~f,. (1.8)

 296

I t follows that L is the infinitesimal generator of a conservative group so that Pro-

blem ( I . I ) - ( I . 4 ) possesses a unique solution with the property:

l l ( ~ ( - , t ) , H ( . , t ) l ~ = constant; (I.9}

furthermore , from ( I . 2 ) , ( I . 3 ) , one deduces immediately the law of mass conservation

H(x,t) = constant. (l.lO)

Remarque I . I : In [4], we give some results concerning the r e g u l a r i t y of solutions of

Problems ( l . l ) - ( l . 4 ) .

The purpose of this paper is to analyse a numerical method proposed by Thacker I l l ,

[2] for solving Problems ( l . l ) - ( l . 4 ) ; more exactly, we shall consider in fact two va-
riants of Thacker's scheme.

2. DISCRETIZATION.

We consider a sequence {~h} of standard triangulations of ~, as shown in the figure.

h denotes the maximum length of the sides of the triangles o f ~ h. We assume that a l l
angles of a l l triangles of a l l triangularizations are bounded from below by some po-
s i t i v e constant. For a p a r t i c u l a r ~ h , l e t N be equal to three times the total number
of nodes minus the number of nodes belonging to the boundary 8~, ~h w i l l denote the
i n t e r i o r of the union of the triangles o f ~ h .

Let us consider a fixed t r i a n g u l a r i z a t i o n ~h. For each node Pk" l e t Ak be the polygon
formed by the triangles containing Pk and let ~k be the measure of Ak. Zk w i l l denote
the set of indices j such that PjG aAk: in the figure, Z2 = {2,3,4,5}, Z6= {7,8,g,I0,
I I } ; c l e a r l y , k ~ Zk i f and only i f Pk~ a~. For some node Pk, l e t Pj ~ aAk, i . e .
j ~ Zk; l e t P~ 6 aAk be the node preceding Pj with respect to the trigonometric orien-
tation and l e t P~ ~ 8Ak be the node following Pj; we define the vector:

~ j k = P;PB ; (2.1)

in the figure, we have, for example, ~8,6 = PTPs, ~s,2 = PfP2" ~22 = P ~ " Furthermore,
we introduce at Pk e a~ the approximate tangent and normal vectors:

= ]
~k ~ Qkk , Nk = R+Tk• (2.2)

For a f u n c t i o n Q defined on ~, l e t ¢o be the continuous, piecewise l i n e a r ( w i t h res-

pect t O . h ) f u n c t i o n defined on Qh and equal to ¢ a t the n o d e s ; f o r a v e c t o r f u n c t l o n 4 ,
+
we d e f i n e , in the same way, componentwise i t s i n t e r p o l a n t ¢o" By Greens's f o r m u l a .
we have the i d e n t i t i e s :
 297

l
1
~k A{k•
v¢o - Z ¢(Pj)
2~k j~Z k (R~jk) '
(2.3)

_]_l ~ -> ->

V'~o = ~l Z ->¢(Pj).(R~ " (2.4)
~k Ak j6Z k Jk)'

clearly, the right members of (2.3),(2.4) define a natural approximation of v¢(Pk)

and ~-#(Pk) respectively.

With the help of (2.3),(2.4), we now define a space semi-discretization of Problem

( l . l ) - ( l . 4 ) . For all nodes Pk' Hk(t)is an approximation of H(Pk,t); for interior nodes
Pk G Q' Ok(t)= (Ub(t),Uk2(t)) is an approximation of O(Pk,t); for boundary nodes
Pk G 8Q' UTk(t) is an approximation of the tangential componentof ~(Pk,t), i.e. of
~(Pk,t).~(Pk). Method IS is then defined by the relations:

Uk
b-k Ok(t )=-½ jCZk
~ Hj(t)(R~jk) + ~k
Uk fRUk(t)' Pk ~ a ; (2.5)

P~k OTk(t ) = _ ½ ~ Hj(t)(R~jk)'~ k , Pk ~ a~ ; (2.6)

bk jeZk

l j~Zk
~k Hk(t) = - ~ z ~j(t).(R~jk) , Pk ~ ~ ; (2.7)

Ok(°) = Oo(Pk)" Pk G ~; UTk(O) = Oo(Pk).~(Pk), Pk C am ; (2.8)

Hk(O ) = Ho(Pk), Pk ~ ~ ; (2.9)

here the "dot" represents the time derivative, bk = b(Pk) and in (2.7),

~j(t) = UTj(t)~ j if Pj ~ a~.

By choosing any fixed order, all the unknown function Ukl(t), Uk2(t), UTk(t) and
Hk(t) can be set in a single vector w(t) of dimension N. Then Problem (2.5)-(2.9) can
be written in the compact form

Dw(t) = Aw(t); w(o) = wo ; (2.10)

whereD isadiagonal matrix with diagonal elements of the form Uk or uk/bk. Because of
property (I.8) for L, one could expect that A is an antisymmetric matrix. Due to dif-
f i c u l t i e s at the boundary, which seem inherent to the problems and impossible to over-
come in a natural way, A is only "almost" antisymmetric. By inspection of the figure,
one can show:
 298

~o

x~ P~

Lemma 2.1: Let Pk and Pj be two different nodes belonging to a same triangle. We sup-
pose that at most one of them belong to the boundary B~. Then

~jk + ~kj = ~" (2.11)

As easily seen from (2.5)-(2.7), A would be exactly antisymmetric i f (2.11) would

hold when i f both Pk and Pj belong to @~. There are several ways to modify Scheme
(2.5)-(2.7) for obtaining the desired property of antisymmetry. One of them consists
in remplacing in (2.5),(2.6) ~jk by "~kj; by Lemma2.1, (2.5) is not modified whereas
(2.6) becomes

P~k OTk(t) = ½ ~ Hj(t)(R~jk)'~k , Pk G @~. (2.12)

bk j6Z k

Methed llS~ is then defined by Relations (2.5),{2.12),(2.7),(2.8),(2.9) and can be

written as

D~(t) = Bw(t), w(o) = wo (2.]3)

where B is an antisymmetric matrix of order N.

For a vector v ~ ~N and a matrix G of order N, l e t

Itvlt : (i N tvilZ) i/z {IGII : sup,, IIGvll

• v~" llvll "
 299

By using the smoothness of a~ and the angle property of the sequence {~h} , one can
verify:

Lemma 2.2: There exist three constants cz,c2,c3, independent of h such that

a) lID-112 (A-B)D-IlZll ~ ci, b) c2h-z ~ lID- 112BD-Zlzl{ sc'3h-~

By using Lemma2.2a, the antisymmetry of B, Relations (2.7),(2.9) in connection with

the d e f i n i t i o n (2.2) of ~k' one can deduce the following properties of Methods IS
and IIS.

proposition 2.1:
a) I f w is solution of (2.13), then IIDZ/2w(t)([2 = IIDllZwoIl2

b) There is a constant c. independent of t and of h, such that i f w is solution of

(2.10), then ilD112w(t)II2 ~ eCtIIDl/2woll z .

c) I f w is solution of (2.10) or of (2.13), then z ~kHk(t) = ~ PkHo(Pk)-

Pk~fi Pk6fi

Remark 2.1: Since the sum is a l l Pk'S is equal to three times the area of ~k, Propo-
s i t i o n 2.1 appears to be, up to a factor 3, the discrete counterpart of Properties
( I ~ 9 ) , ( I . I 0 ) of the exact solution.

We now turn to the time discretisation. Let c > o be the time increment and set
t n = n~. We shall apply to (2.10) and (2.13) the two-step method, sometimes called
"leap-frog" scheme. Method I is then defined by the relations:

T
w z / z = wo + ~ D-IAWo (2,14)

wz = wo + "~D-IAWl/2 (2.15)

Wn+ z = Wn_1+ 2-~D-ZAWn , n = 1 , 2 , 3 . . . . (2.16)

Method I I is defined by replacing in (2.14)-(2.16). A by B.

The s t a b i l i t y analysis of Method I I is t r i v i a l since D~B or D-I/~BD -I/z have a pure

imaginary spectrum. By using Strang's Lemma [3] in connection with Lemma2.2a, we
easily deduce the s t a b i l i t y of Method I from that of Method I I .

Proposition 2.2: There exist a function ~: Co,l) -~ ~R and a constant c, both indepen-
dent of h,~ and n such that i f T < I/liD- z/2 BD-Z/21[, we have:

a) {ID-l/ZwnII2 < ~(~ II D- i/z BD-i12)ii ) IIDI/2woII for Method I I,

b) ([D1/~Wni12< ~(~ {p-I/2BD-V2H)eCtn(IDi/2 Woll for Method I.

 300

Remark 2.2: Methods I and I I s a t i s f y a law o f mass c o n s e r v a t i o n as Methods IS and IIS

(see P r o p o s i t i o n 2 . 1 c ) .

Remark 2.3: Proposition 2.2a,b prove the s t a b i l i t y of both Method I and I f ; however
Method I I appears to be "more" stable than Method I.

Remark 2.4: wn can be written as a vector of order N with components UknI, Ukn2, UTkn
and Hkn. I f the Coriolis term f = o, then for Methods I and I I , i t is possible to
compute Uknl, Ukn~, UTkn only at even values of n and Hkn at odd values of n which
reduces the computer time and the storage requirements by a factor 2; real Thacker's
scheme, which is somewhat more d i f f i c u l t to analyse, keeps this property even for
f ~ o. In fact, i f f = o, Method I is identical to Thacker's scheme ( l O ) , ( l l ) , ( l l ' )
in [ 2 ] p.683.

Remark 2.5: Thacker [2] has remarked that his scheme can be considered, to some ex-
tend, asalumped version of a Galerkin method. In [4], we b r i e f l y analyse the effect
of "lumping" on s t a b i l i t y .

Remark 2.6: The s t a b i l i t y c o n d i t i o n ~ < I / l I D - I / Z BD-Z/211 in P r o p o s i t i o n 2.2, i m p l i e s ,

by Lemma 2.2b t h a t • = O(h).

3. ERRORESTIMATES

Our estimates w i l l be based on the standard consistency*stability argument. S t a b i l i t y

has already been analyzed in Section 2.

We begin with a classical study of consistency by assuming that the components of the
solution of Problem ( l . l ) - ( l . 4 ) belong toC°([o,T]; C3(~)). We f i r s t associate to this
solution a vector u(t) 6 ~N in the following way; l e t w(t) the solution of the semi-
d i s c r i t i z e d problem (2.10); we set: u i ( t ) = U~(Pk,t) i f wi(t) = Uk~, ~ = 1,2;

ui(t) = ~(Pk,t)'~(Pk) if wi(t) = UTk(t); u i ( t ) = H(Pk,t ) i f w i ( t ) = Hk(t).

Clearly, the time discretization which is of order two, w i l l induce errors of size
O(~Z), which, by Remark 2.6, can be written as O(h2). Let us define for T > o:

~i (T) = max {{DI/2(G(t)-D'IAu(t)){[, Eli (T) = max llDl/2(u(t)-D'iBu(t))((,

o~t~I o~t~T

Ri(zI) (T) = max IIDz/2(u(tn)-wn)ll i f wn is obtained by Method l ( I I ) .

o~tn~T

For i = I , I I , Ri(T ) is the error of Method i , whereas, by using ( l . l ) , ( l . 2 ) , ~i(T) is

the space consistency error.
 301

In the following, we shall say that for a node Pk, Ak is s~mmetric, i f for each
Pj G a~k' there exists P~ ~ aAk which is symmetric to Pj with respect to Pk" Clearly
i f Pk G ~ , Ak cannot be symmetric. The basic difference schemes defined by (2.3),
(2.4) are of order 2, with respect to h, i f Ak is symmetric; otherwise there are only
of order I. We shall say that the sequence of triangularizations ~ h } possesses Pro-
perty G i f there exists a constant c, independent of h such that for a l l Pk ~ Be one
has l#k-~(Pk) I ~ ch2 ; Property G implies a certain regularity in the d i s t r i b u t i o n
of the nodes on the boundary. Elementary but tidious calculations allow to establish:

Lemma 3.1: For any fixed T > o and i = I , I I , we have: a) ~i(T) = O(hZl2); b) ~i(T) =
O(h) i f Property G is satisfied; c) ~i(T) = O(h3/2) i f Ak is symmetric for a l l Pk G
and i f Property G is satisfied.

From Lemma 3.1 f o l l o w s immediately.

P r o p o s i t i o n 3.1: Let T > o and ( ~ ( o , I ) be f i x e d numbers. For each t r i a n g u l a r i z a t i o n

~h' ~ is chosen in such a way t h a t o < • < ~/IID-Z/2BD-Z/zlI. Then f o r i = I , I I :
a) Ri(T) = O(h z/2); b) Ri(T) = O(h) i f P r o p e r t y G is s a t i s f i e d ; c) Ri(T) = O(h 3/z) i f
Ak is symmetric f o r a l l Pk ~ Q and i f Property G is s a t i s f i e d .

Remark 3.1: Suppose, that, instead of (2.2), we set ~k = ~(Pk ) (exact tangent vector).
Then: a) We loss the exact mass conservation property for both Methods I and II /see
Remark 2.2); b) Proposition 3.1 remains v a l i d for Method I; Proposition 3.1a remains
valid for Method I f .

We now turn to an e r r o r a n a l y s i s under weaker r e g u l a r i t y assumptions. We s h a l l sup-

pose t h a t the components of the s o l u t i o n o f Problem ( I . I ) - ( I . 4 ) belong to C ° ( [ o , T ] ;
H2(Q)). For s i m p l i c i t y , we s h a l l f u r t h e r m o r e assume t h a t ~ i s convex so t h a t ~h c ~.
The d i f f i c u l t y here comes from the f a c t t h a t the time d e r i v a t i v e of the s o l u t i o n is
not a continuous f u n c t i o n of the space v a r i a b l e . Let ~kn = (Ukn1,Ukn2),Hkn be the
approximate s o l u t i o n obtained by Method I or I I corresponding to the exact s o l u t i o n
~(Pk,tn), H(Pk,tn); here we set ~kn = UTkn~k i f Pk ~ ~ " Let Vh be the space o f con-
tinuous piecewise l i n e a r f u n c t i o n s on Qh corresponding t o ~ h . We d e f i n e the f u n c t i o n
~h(x,t) = (Uh1(x,t),Uh2(x,t)), Hh(X,t), f o r x ~ Qh and t = t n , in the f o l l o w i n g way:
U h B ( . , t n) E Vh, ~ = 1,2, H h ( . , t n ) E Vh; ~ h ( P k , t n ) = ~kn, Hh(Pk,tn) = Hkn f o r a l l
PkE~.

The main t r i c k will c o n s i s t in i n t r o d u c i n g f u n c t i o n s Yh(X,t) = ( Y h l ( X , t ) , Y h z ( x , t ) ) ,

Z h ( X , t ) belonging to Vh f o r f i x e d t and which are, f o r f i x e d t , Cl@ment's approxima-
t i o n s o f the exact s o l u t i o n ' s components; f o r the n o t i o n of Cl~ment's a p p r o x i m a t i o n ,
see [ 7 ] . Corresponding to the exact equation ( I . 2 ) , we have f o r a l l Pk ~ ~ the f o l l o -
wing i d e n t i t y :
 302

~kZh(Pk(tn+l ) - ~kZh(Pk,tn_z) + • N Y(P~,tn)-(RQ4k)jj (3.1)

j~Z k
= {~kZh(Pk,tn+ z) - ~kZh(Pk~tn_z) - 2~A~ H(x,tn)} (3.2)

+ ~ { JGZk
~ ~ ( P j , t n ) . (R~jk) - 2A~ ~ , ~ ( X , t n ) } . (3.3)

The "time" e r r o r term (3.2) can be e a s i l y estimated by using the f a c t t h a t the opera-
t i o n s o f time d e r i v a t i v e and Cl6ment's approximation commute. The "space" e r r o r term
(3.3) can be handled by remarking the f o l l o w i n g i d e n t i t y which is a d i r e c t consequen-
ce of ( 2 . 4 ) :
J~ZkZ ~ ( P j , t n ).(R~jk) - 2A~ ~- ~(X,tn) : o.

Similarly to (3.1)-(3.3), we can write an equation corresponding to ( l . l ) ; i t is

s l i g h t l y more complicated to handle because of the boundary condition (1.3) and of
the presence of the function b(x). With Proposition 2.2, this allows to get error es-
timates between Cl~ment's approximation and the solution of Method I or I I . The final
result is contained in the following proposition:

Proposition 3.2: Let T > o and ~ ~ ( o , l ) be fixed numbers. We suppose: a) each compo-
nent of the exact solution belong to C°([o,T]; Hz(~)); b) Property G is satisfied;
c) for each t r i a n g u l a r i z a t i o n , ~ is chosen such that o < z < ~/IID-I/zBD-I/ZII. Then for
both Methods I and I I we have:

max_{ S (b@x) l~(X'tn) - ~h(X'tn )12 + {H(x'tn) - Hh(X'tn )Iz)}1/2 = O(h).

°~tn~T ah

Remark 3.2: In [ 4 ] , we give some numerical r e s u l t s .

Remark 3.3: In order to compute the spectrum of the operator L defined in Section l,
one could think of using the same space discretization as in Method I or I I ; however
this generates spurious eigenvalues. For a proper treatment of this problem, see [5],
[6].

References
[ 1] W.C.THACKER,Irregular Grid Finite-Difference Techniques: Simulations of Oscillations in Shallow
Circular Basins, Journal of Oceanography, VoL 7, 1977, 284-292.
[2] WC THACKER Comparison of Finite-Element and Finite-Differences Sehemes, Part I: One-Dimension-
al Cravity WaveMotion, Part H: Two-Dimensional Gravity WaveMotion, Journal of Oceanography,
vol. 8, 1978, 676-689.
[3 ] G.STRANG, Accurate Partial Difference Methods, Numerische Mathematik, 6, 1964, 37-46.
[4] J.DESCLOUX, R.FERRO, On Thacker's scheme for solving the tinearized shallow water equations.
Report. Departement de Math6matiques. Ecole Polytechnique F6dgrale de Lausanne, 1985.
[51 M.LUSKtN, Convergence of a Finite Element Method for the Approximation of Normal Modes o f the
Oceans, Math. Comp. 33, 1979, 493-519.
[6] J.DESLOUX, M.LUSKIN, J. RAPPAZ, Approximation of the Spectrum of Closed Operators." The
Determination of Normal Modes of a Rotating Basin, Math. Comp. 36, 1981, 137-154.
[7] Ph.CLEMENT, Approximation by Finite Element Functions using Local Regularizations, RAIRO 9,
1975, 77-84.
THE CONVERGENCE OF A NEW METHOD
FOR CALCULATING LOWER B O U N D S
TO EIGENVALUES
F. GOERISCH and J. ALBRECHT
Institut fiir Mathematik, Technische UniversiEit Clausthal
Erzstrafie 1, D 3392 Clausthal-Zellerfeld, West Germany

The relationship between an inclusion theorem due to N. J. L e h m a n n [3]

and one recently proposed [2] is i n v e s t i g a t e d here. The theorem due to
N. J. L e h m a n n yields better bounds to the e i g e n v a l u e s , whereas the n e w
theorem is in g e n e r a l considerably easier to a p p l y . It is shown that
sequences of b o u n d s to e i g e n v a l u e s can be o b t a i n e d with the u s e of the
new theorem, and that these sequences converge to the b o u n d s provided
by L e h m a n n ' s theorem. This fact is illustrated by m e a n s of n u m e r i c a l
results f o r the following eigenvalue problem:
A2¢ = -tA~ in ~, % = "~n = 0 on 3~, ~:={(x,y) e~2: Ix}<~, Iyt<~},
which occurs in the calculation of b u c k l i n g stresses of c l a m p e d plates
under compression.

§I The two inclusion theorems are first stated, in a v e r s i o n which

deviates somewhat from that presented in the original papers [2], [3],
but which is e s p e c i a l l y well suited for p r a c t i c a l applications. The
following assumptions and definitions are required for t h i s purpose:

Assumptions
AI D is a r e a l vector space. M a n d N are symmetric bilinear forms
on D; M(f,f) > 0 for all f • D, f ~ O.
A2 There exist sequences (li) i 6 ~ and (#i) i • ~ such that
li•~, ~ieD, M(~i,%k) = 6ik for i,k•~,
M(f,~i) = l i N ( f , ~ i ) for all f e D, i•~,

N(f,f) = i=[ l l i ( N ( f , ~ i ) ) 2 for all f • D.

A3 X is a r e a l vector space; T: D + X is a l i n e a r operator; b is a

symmetric bilinear f o r m on X. b(f,f) > O for all f e X and
b(Tf,Tg) = M(f,g) for all f,g e D.
A4 p•~, p > O; ne~, v i • D for i=1,...,n.

Definitions
DI Matrices A
o a n d A I are defined by
 304

Ao:=(M(vi,vk))i,k=1,..., n , A1:=(N(vi,vk))i,k=1,..., n •
D2 If A is a s y m m e t r i c matrix of o r d e r n, with the p r o p e r t y that
Ao-2pAI+p2A is p o s i t i v e definite, ~i(A) denotes the i-th s m a l l e s t
eigenvalue of the e i g e n v a l u e problem (Ao-PA1)z = ~ ( A o - 2 p A I + p 2 A ) z.

T h e two i n c l u s i o n theorems, whose relationship is to be i n v e s t i g a t e d ,

yield inclusion intervals for the e i g e n v a l u e s of the e i g e n v a l u e p r o b l e m
M(f,~) = lN(f,~) for all f 6 D. (1)
The t h e o r e m s are as follows:

Theorem I (N. J. L e h m a n n [3])

L e t u i 6 D be such that M(f,ui) = N ( f , v i) for all f 6 D, i = 1 , . . . , n ; let
the m a t r i x A 2 be d e f i n e d by A 2 : = ( M ( u i , u k ) ) i , k = 1 , .... n ' and let

A o - 2 p A I + p 2 A 2 be p o s i t i v e definite. Moreover, suppose that q e ~ , q ~ n,

~ q ( A 2) < O.
The i n t e r v a l [p-p(1-~q(A2))-1,p) then c o n t a i n s at l e a s t q e i g e n v a l u e s I)
of the e i g e n v a l u e problem (1).

Theorem 2 ([2])
L e t w i e X be such that b ( T f , w i) = N ( f , v i) for all f e D, i=I ..... n; let
the m a t r i x A2 be d e f i n e d by A 2 : = ( b ( w i , w k ) ) i , k = 1 .... ,n ' a n d let

Ao-2pA1+p2A2 be p o s i t i v e definite• Moreover, suppose that q e ~ , q ~ n,

~q(A2 ) < O.
The i n t e r v a l [ p - p ( 1 - ~ q ( A 2 ) ) - 1 , p ) then c o n t a i n s at least q e i g e n v a l u e s I)
of the e i g e n v a l u e problem (I).

If the a s s u m p t i o n s of t h e o r e m I are s a t i s f i e d , and if w is d e f i n e d by

1
w i := Tu i for i = 1 , . . . , n , the a s s u m p t i o n s of t h e o r e m 2 are also ful-
filled because of A2 = A2" Thus, theorem I is an i m m e d i a t e consequence
of t h e o r e m 2.
The importance of t h e s e t h e o r e m s is due to the fact that t h e y p r o v i d e
a m e a n s of c a l c u l a t i n g accurate lower b o u n d s to the e i g e n v a l u e s of
problem (I). If the e i g e n v a l u e s of (I) are a r r a n g e d in a n o n - d e c r e a s i n g
order,

if p is a l o w e r b o u n d to the e i g e n v a l u e ip+q (p,q6~) and if, for

example, the a s s u m p t i o n s of t h e o r e m 2 are s a t i s f i e d , then
p-p(1-Bq(A2))-1 is a lower b o u n d to Ip . For an a p p r o p r i a t e choice of
the q u a n t i t i e s involved, this b o u n d to i is very a c c u r a t e , even if p
P
is o n l y a c o m p a r a t i v e l y rough lower b o u n d to 1
p+q

I) E i g e n v a l u e s are a l w a y s counted according to t h e i r m u l t i p l i c i t y .

 305

It is o f t e n d i f f i c u l t , or e v e n impossible, to e x p l i c i t l y give the

elements u. r e q u i r e d in t h e o r e m I; by m e a n s of t h e o r e m 2, in contrast,
l
i n c l u s i o n i n t e r v a l s for the e i g e n v a l u e s can be d e t e r m i n e d with c o m p a r -
ative ease - p r o v i d e d that X, b and T have been a p p r o p r i a t e l y chosen
(compare §3). H o w e v e r , the r e s u l t s thus o b t a i n e d c a n n o t be b e t t e r than
t h o s e w h i c h w o u l d be p r o v i d e d by t h e o r e m 1, as is n o w shown:

Lemma I
Let the a s s u m p t i o n s of t h e o r e m I a n d 2 be f u l f i l l e d .
Then p-p(1-~q(A2))-1 ~ p - p ( 1 - ~ q ( A 2 ) ) -1
~[~9[! S i n c e b(Tui,wk) = N(ui,vk) = M(ui,uk) for i,k=1 ..... n, it
follows that b ( w i - T u i , W k - T U k ) = b(wi,wk) - M(ui,uk) ; hence, the matrix
A2-A2 is p o s i t i v e semidefinite. W i t h the use of the c o m p a r i s o n theorem,
one o b t a i n s ~q(A2) ~ ~q(A2). The a s s e r t i o n can n o w be i m m e d i a t e l y
deduced.

§2 On the b a s i s of t h e o r e m 2, a s e q u e n c e of i n c l u s i o n intervals [Tm,p)

w i l l n o w be c o n s t r u c t e d in such a m a n n e r that (Tm) m e l ~ c o n v e r g e s , and
the i n t e r v a l [lim Tm,P) c o i n c i d e s w i t h the c o r r e s p o n d i n g inclusion
m-~oo
interval f r o m t h e o r e m I. For this p u r p o s e , the f o l l o w i n g additional
assumptions and definitions are required.

Assumptions
A5 W. e X for i = 1 , . . . , n and w* e X for ie3~,
1 1
b ( T f , w i) = N ( f , v i) for all feD, i=I .... ,n,
b(Tf,w[) = 0 for all f e D , iel~.
The matrix (b(w[.,w*)) is r e g u l a r for all meiN.
• k i,k=l , .... m

A6 X o := { g e X : b(Tf,g) = O for all f e D } ; for all g e X ° a n d all g e ] R

with g > 0 there exist numbers me]q, Cl,...,Cme]R such that
m m
b ( g - [ c . w * , g - [ c w*) < e.
i=I i i i= I i ± --
Rem_ark- If b(f,f) > O holds for all f e X with f ~ O, that is, if
(X,b(.,.)) is a p r e - H i l b e r t space, the a s s u m p t i o n A6 s t a t e s p r e c i s e l y
t h a t the s u b s p a c e s p a n n e d by {wE: ielN} is d e n s e in X o-

Definitions
D3 A2 :=(b(wi'wk))i,k=1,...,n ;

Fm : : ( - b ( w i ' w ~ ) ) i : 1 , . . ,n;k:l,.
. . . ,m ' Gm ::(b(wi'wk))i,k:1 ,...,m'

A2, m := i 2 - F m G m I F m' for all mel~.

The inclusion intervals [Tm,P) can n o w be given:

 306

Theorem 3
Let m , q e 3 N with q < n; let the matrix Ao-2pA1+p2A2,m be p o s i t i v e defi-
nite, and let ~q(A2,m) < O.
)-1
If "rm i s defined b y "rm := p - p ( 1 - t X q ( A 2 , m) , the interval [ r m , p)
contains at least q eigenvalues of the e i g e n v a l u e problem 11).
Proof:_ Let F m G mI=- (dik)i=1, .... n;k=1 .... ,m " The a s s e r t i o n follows imme-

diately from t h e o r e m 2, if the w. o c c u r r i n g there are d e f i n e d by

1
^ m
W i := W i + k ! i d i k W ~ for i=I ..... n.

The f o l l o w i n g result concerning the c o n v e r g e n c e of the sequence

(Tm)mel~ is n o w obtained:
Theorem 4
Let the a s s u m p t i o n s of t h e o r e m I be satisfied. If T is d e f i n e d by
) -I m -I
"rm := p-p(1-gq(A2, m) f o r m ~ l N , t h e n l i m "rm = p - p ( 1 - ~ q ( A 2 ) )
m.+eo
......
Proof: Let F mG-I
m = (,(m)
Ctik ) i = l , . . . , n ; k = l , . . . , m for melt. Then
^ m (m) , ,
b(Tui-wi-k[Idik= W k , W j) = O (2)

for i=l,...,n,j=l,...,m, melt. Let ee~ with s > O. Since Tu -w. e X

1 1 o
for i=l,...,n, there exist numbers laiN, and C i k e ] R for i=l,...,n,
k=1,... ,i such that
1 1
bCTui-q-k!leikw, Tui-wi-!lCiWp ^ ,

for i=1,... ,n. With the use of (2), it can be shown that
m
^ m} _(m) ,.*, T u . - w . - ~. d ! m ) w *) < e
b(Tui-wi-k__laik w 1 1 k= I iK K --

for i=1,...,n and all m61~ with m > i. By means of the C a u c h y - S c h w a r z

inequality, it follows that
m
Ib(Tui-wi-k! d l k ) W { ' Tuj-w.-3 ~ ~jk
a(m) Wk;
*" [ _< e
1 " k=l

for i,j=l,...,n and all m¢l~ with m > i. Hence,

~lim b ( T u i _ ~ i _ k -~ l a~(m)
m÷ , ~ w~ , ~ d ~ )w)*
T u .3- w .3- k=1 = O

for i,j=l,...,n. F r o m the e q u a t i o n

m
^ m (m) , T u . - w . - ~. d!m)w*))
(b(Tui-wi-k!idik Wk' 3 3 k=l 3 K K i,j=1,...,n = A2,m-A2

it f o l l o w s that A 2 , m - A 2 is p o s i t i v e semidefinite for me]g, and that

lim A 2 , m = A 2 . This gives lim ~q(A2,m) = ~ q ( A 2) , f r o m which the asser-
 307

tion follows immediately.

~9~[~i The sequence (Tm) m • ~ is n o n - d e c r e a s i n g .

§3 The p r a c t i c a l application of the results presented is n o w illus-

trated with the u s e of an e x a m p l e . - The quantities D, M, N, X, T, b
occurring in the a s s u m p t i o n s AI a n d A3 are defined in the following
manner (~2) (~) d e n o t e s the Sobolev space defined in [4]):

D := { f 6 )(~): f(x,y) = f(y,x) = f(-x,y) for (x,y)6~},

M(f,g) := f A f £ g d x d y , N(f,g) :: f ( ~
3g
~
3f
+ ~
~_~)dxdy for f,g e D ,

X := {f • L 2 ( ~ ) : f(x,y) = f(y,x) = f(-x,y) for (x,y) e ~ } ,

Tf := -£f for f e D, b(f,g) := / f g d x d y for f,g e X,

where ~ := { ( x , y ) 6 ~ 2 : Ixl<~, lyl<~}.

In this case, the eigenvalue problem (I) is the w e a k f o r m of the
eigenvalue problem

~(x,y) = ~(y,x) = #(-x,y)On for (x,y) e ~. ~ (3)

For specifying the quantities occurring in A4 anO AS, a number r e~ is
first chosen; 0, n, vi, wi' w~1 are then defined as follows:
p := 10, n := 9 r ( r + 1 ) ,

vi(x,y) := c o s S + 1 ( x ) c o s t + l ( y ) + cosS+] (y)cost+l(x) for i=1 .... ,n,

where s,te~ are determined by r > s > t, i=~s(s-1)+t,
^

w. := V. for i=1,...,n,
l l
w;(x,y) :: c o s h ( i x ) c o s ( i y ) + cosh(iy)eos(ix) for i em.
The assumptions AI, A3, A4 a n d A5 are obviously fulfilled, and the
proofs of A2 a n d A6 p r o c e e d in a n a l o g y with the corresponding proofs
in [4], p. 472 a n d [I], respectively.
By means of the C a u c h y - S c h w a r z inequality, it f o l l o w s from the compar-
ison theorem that the eigenvalues of the p r o b l e m
-A~ = %~ in ~, ~ = O on ~,
~(x,y) = ~(y,x) = ~(-x,y) for (x,y) e ~,
are lower bounds to the corresponding eigenvalues of (1); h e n o e 0=IO
is a l o w e r bound for the second eigenvalue of (I).
The first six t e r m s of the sequence (Tm) m E ~ calculated for q=1 with
t h e u s e of theorem 3 are compiled in the first six r o w s of t a b l e 1 for
various values of n. Since at l e a s t one eigenvalue of (I) is c o n t a i n e d
in e a c h of the intervals [Tm,p), the numbers Tm are lower bounds to
the lowest eigenvalue of (q). An u p p e r bound A to t h i s eigenvalue,
which has been determined with the use of the functions v I .... ,v n b y
 308

means of the Rayleigh-Ritz method, is g i v e n in the last r o w of table I,

for the respective value of n.
By v i r t u e of t h e o r e m 4, the sequences (Tm)me~ converge to the c o r r e -
sponding bounds which would result from theorem I. In o r d e r to a p p l y
theorem I, h o w e v e r , it w o u l d be n e c e s s a r y to d e t e r m i n e the exact
solution u i of the b o u n d a r y value problem
~u i
A2ui = - A v i in ~, u i - Sn - 0 on ~,

which is n o t an e a s y task.

n = 1 n = IO n = 21

TI 5.049 5.057 5.057

Y2 5.250 0 5.265 8 5.265 9
~3 5.283 52 5.302 42 5.302 46
T4 5.284 556 5.303 564 5.303 602
~5 5.284 582 0 5.303 587 3 5.303 625 3
T6 5.284 582 21 5.303 587 37 5.303 625 40

A 5.333 333 34 5.303 662 26 5.303 626 22

Table I Bounds to the lowest eigenvalue of (3)

The method based on t h e o r e m 3 has also been applied with great success
to m a n y other eigenvalue problems involving partial differential
equations.

The authors gratefully acknowledge the support of t h i s w o r k by the

Deutsche Forschungsgemeinschaft.

References
[I] C o l a u t t i , M.P.: Su un t e o r e m a di c o m p l e t e z z a c o n n e s s o al m e t o d o di
W e i n s t e i n p e r il c a l c o l o d e g l i a u t o v a l o r i . A t t i A c c a d . Sci. T o r i n o ,
CI. Sci. Fis. Mat. Nat. 97 (1962/63), 171 -191
[2] G o e r i s c h , F. a n d H. H a u n h o r s t : E i g e n w e r t s c h r a n k e n for E i g e n w e r t -
aufgaben mit partiellen Differentialgleichungen. Z. A n g e w . Math.
Mech. 67 (1985), 129 - 135
[3] L e h m a n n , N.J.: O p t i m a l e EigenwerteinschlieBungen. Numer. Math. 5
(1963), 246 - 272
[4] R e k t o r y s , K.: V a r i a t i o n a l m e t h o d s in m a t h e m a t i c s , s c i e n c e a n d
engineering. Dordrecht-Boston: D. R e i d e l P u b l i s h i n g C o m p a n y 1977
BIFURCATION ANALYSIS OF STIMULATED
BRILLOUIN SCATTERING
V. JANOVSKY, I. MAREK, J. NEUBERG
Faculty of Mathematics and Physics, Charles University
Malostranskg ndm. 25, 110 O0 Prague, Czechoslovakia

I. Introduction

Our problem is m o t i v a t e d by t h e following physical effect (Stimu-

lated Brillouin Scattering): A laser b e a m of a g i v e n frequency is t a r -
geted on a m a t e r i a l sample. If the laser intensity is s m a l l then the
beam penetrates without being affected. If the intensity is a b o v e a
threshold then the sample acts like a mirror and reflects some energy
back (Stokes' wave). This is d u e to s t i m u l a t e d pressure (acoustic) wave
in the sample. The frequences of all three waves (i.e. laser and Stokes
and pressure) are ~ o u p l e d .

L e t us a c c e p t that Stimulated Brillouin Scattering can be m o d e l l e d

by the following initial value problem: Find complex-valued functions
EL , ES , p (the s l o w l y varying amplitudes of laser and Stokes and
pressure waves) of time t ~ 0 a n d one spatial variable 0 ~ x ~ l
such that

EL = E L - i E s P ' ES = - E~ - i E L ~ , cp = p' - p - iELE S (1.1)

(Notation: E = ~ E , E' = ~a E , E is c o m p l e x conjugate, i =

(-I) I/2 , c is a p o s i t i v e constant) with boundary condition

p(/,t) = ES(0,t) = 0 , EL(/,t) = ae i@ (1.2)

for t ~ 0 ; a and @ are real parameters. At t = 0 , an i n i t i a l

condition (compatible with (1.2)) is p r e s c r i b e d .

1.3 REMARK. The sample occupies the interval 0 ~ x ~ 1 . Laser light

2
of intensity a is focused at the p o i n t x = I and propagates in ne-
gative direction of the x - a x i s . Note that functions E L ~ ae i~ , ES
p ~ 0 are a steady state solution to (1.1), (1.2). This trivial solu-
tion c o r r e s p o n d s to the s i t u a t i o n w h e n no s t i m u l a t i o n of p r e s s u r e w a v e s
2
occurs which is e x p e c t e d if a is less t h e n the t h r e s h o l d . P h y s i c a l
 310

significance of the model (1.1), (1.2) is d i s c u s s e d e.g. in [I].

1.4 OBSERVATION. If EL , ES , p solve (1.1). t h e n , for each real

constant y , the functions ELeIX , ES , p e iX solve (1.1), too. It
implies that we can assume ~ ~ 0 and a ~ 0 without loss of genera-
lity.

Our aim is b i f u r c a t i o n analysis of trivial solution to steady state

problem (1.1), (1.2) with respect to variations of parameter a . We
mention dynamical stability of bifurcated solutions, too.

2. Covariance of the equations ~overnin~ the steady state

Let us introduce the operator F of the steady state: Homogenising

the boundary condition (1.2) by substitution EL := E L + a , we define
F = F(U,a) at U = (EL, ES, p) and at a 6 ~1 as follows:

F(U,a) = (E L - i E s P , - E'S - i E L P - iap, p' - p - i a E S - i E L E S)

Then F acts (e.g.) on the real linear space X = {U = ( E L , E S , P) : EL,

ES, p are complex-valued functions of x , continuously differentiable
on 0 ~ x ~ £ , satisfying boundary condition EL(Z) = ES(0) = p(£) = 0}.
The range of F(.,a) is in the real linear space Y = {U = (EL,Es,P) :
EL, ES, p are complex-valued, continuous functions of x , 0 ~ x ~ £}.
In the usual topology, there is c o m p a c t imbedding of X into Y .

Let us define suitable linear transformations on Y : If U =

(E L , E S, P) E Y then

MBU : (EL, ei~Es , e-iBp) for each B 6 ~1 '

-- j+1
TjU = (EL' (-I)3Es' (-I) p) for j = 1,2 .

Let {M~,TI,T2} denote the group generated by M E , T1 , T2 (~ 6 RI ).

The covariance property of the steady state operator follows from

2.1 LEMMA. If U C X then MsF(U,a) = F(M U,a) and T/F(U,a)~ =

F(TjU,a) for each ~ 6 ~1 and j = 1,2 .

Proof. This can be done by a straightforward calculation.

As a direct consequence, we obtain

2.2 PROPOSITION. If r 6 {MB,TI,T2} then FF(U,a) = F(FU,a) for each

U E X .
 311

3. Bifurcation analysis

We resume the steady state problem: Given a value of a ~ 0 , find

U @ X such that F(U,a) = 0 . Obviously, U0 ~ 0 is trivial solution
for each value of a .

Let L(a) : X + Y be Fr6chet derivative of the operator F(.,a) :

X ÷ Y at U0 . If there is a b i f u r c a t i o n from U0 at a 6 ~I then
the kernel Ker L(a) is n o n t r i v i a l . Direct calculations yield

3.1 LEMMA. If a ~ 0 then Ker L ( a ) is n o n t r i v i a l if and only if

a 6 K , where K = {a ~ 0 : 4a 2 _ I ~ 0 , t g ~£
2 _ w , = (4a 2 - I)½}.

The set K can be a r r a n g e d as an i n c r e a s i n g sequence {aj}~= I ,

0 < a I < aj < aj+~ , lim a. = + ~ .
, j÷+~ 3

3.2 LEMMA. If aj 6 K then Ker L ( a j ) is s p a n n e d by vectors ~I =

(O'Vo'-iWo) ' ~2 = (O'-iVo'Wo) from X , where V0 and W0 are the
mx
following real functions of x : v 0 = e x/2 sin ~- , w0 =

(-I) j+1 e x/2 sin m(£ -2 x) ; m = (4a - 1) I/2

Let us make some remarks on bifurcation equation: By virtue of Fred-

holm alternative, space Y can be decomposed as a direct sum of kernel
and range of the operator L(aq) : X + Y . Thus each solution U to
3
F(U,a) can be written as U = ~ y i ~ i + U~ where y~s are real coordina-
tes, ~s span the kernel and U~ belongs to the range. According to
Liapunov-Schmidt reduction (see e . g . ~3]), if U and a are suffici-
ently close to U0 and a. respectively then U can be identified
3
with coordinates Yl ' Y2 of the projection into Ker L(aj) . The co-
ordinates satisfy bifurcation equation H(YI' Y2' a - aj) = 0 , where
H is germ of a mapping H : ~2 x ~I ÷ R2 "

Following standard routines, Taylor expansion of H can be fo~d.

In o u r particular case, we have calculated that

+
H(Y1 ' Y2' a - aj) = pj(a- aj) Y2 - q j (Yl + Y2 ) Y2
(3.3)
a j ) 2 [ yl ]
+ terms of the 4 th order
sj(a - Y2

2 ~£
where PJ ' qJ ' si are real constants, namely pj = £ + 2 cos ~- ,
£

qi = ~( ' - l " J cos -f- (v (£) - v u ) ( v u + w o) dx ; for e, v 0 , w0 see 3.2.

0
Both pj and qj are positive.
 312

3.4 OBSERVATION. Group {Ms,T1,T2} leaves the kernel Ker L(aj) in-
variant. The matrix representation of {M6,TI,T2} on Ker L(aj) is
0(2) group of 2 x 2 orthogonal matrices. Following Sattinger [2],
the b i f u r c a t i o n equation is c o v a r i a n t under 0(2) symmetry group.

3.5 THEOREM. The bifurcation equation H(Yl' Y2' a - a k) = 0 is

O(2)-equivalent to

[a --
ak --
( 2 + y~)3 [Yl] = 0
Yl Y2

(for the notion o f O(2)-equival~nce, see [4]).

Proof. We just quote E4], L e m m a 5.18 (our p o i n t s 3.3 and 3.4 v e r i f y

the a s s u m p t i o n s ) .

We c o n c l u d e that each a. E K is a p o i n t of s u p e r c r i t i c a l bifurca-

3
tion. Bifurcation diagrams are 2 - D manifolds (symmetric, in a sense,
about a-axis).

4c. Stability

Let us r e s u m e the initial value problem (1.1), (1.2) where we sub-

stitute EL := E L + a : Given a ~ 0 and U in ~ X , find U = (EL,Es, p)
such that U = F(d,a) for t > 0 and U = U in at t = 0 . Asymptotic
stability (as t -~ + = ) of the above problem is u n d e r question, assu-
ming that U in is a s t e a d y state solution being subjected to a s m a l l
perturbation. In [51 we h a v e tackled this question by m a k i n g u s e of
Principle of L i n e a r i s e d Stability.

We just quote our results (which are not surprising anyway) :

Trivial solution U0 is s t a b l e as 0 ~ a < aI (i.e. up to the first
bifurcation point). Beyond this point, it l o o s e s stability. Nontrivial
steady state solutions which emanate from bifurcation points a2,a3,...
are u n s t a b l e . On the o t h e r hand, the b r a n c h emanating from aI is sta-
ble.

The above facts are illustrated by the following numerical experi-

ments.

EXAMPLE I. Data: £ = c = I , a = I (i.e. a < a I) ; U in is j u s t

the trivial solution U0 being perturbed by G a u s s i a n "noise" ( 6-corre-
lated, dispersion = I , mean value = 0 ). T i m e interval: 0 ~ t ~ 0.3 .

Results are presented in F i g u r e 1 : At each time, the (numerical)

solution U is p r o j e c t e d onto Ker L ( a I) and then on yl,Y2-plane.
 313

Point S is projection of U in . It is apparent that solution creeps

towards the origin, i.e. U0 . The second graph indicates velocity of
the motion in time.

Y2 !

0
Q)
\
>

Figure I

EXAMPLE 2. Data: £ = c = I , a = 5 (i.e. a 2 < a < a 3) ; U in is

a (numerical) steady state solution on the 2nd branch which is randomly
perturbed as above. Time interval: 0 ~ t ~ 10 .

Legend to Figure 2 : U is projected onto Ker L(al) again. S

is the position of projected U in . Note that projection of all steady
solutions on the first branch at a = 5 would be a circle centred at
origin O and passing through C . We observe U oscillating around
a point on this circle for large t .
 3t4

Y2

o C
o
I
> Yl

Figure 2

R e f e r e n c e s

[lj STARUNOV, V. S. and FABELINSKIJ, I. L.: Uspekhi fiz. nauk. 98, NO 3

(1969).
[~ SATTINGER, D. H.: Group representation theory and branch points of
nonlinear functional equations. SIAM J. Math. Anal.,8, (1977), 2,
pp. 179-201.
E3~ CHOW Shui-Nee and HALE, J. K.: Methods of Bifurcation Theory.
Springer, N. Y., 1982.
~] GOLUBITSKY, M. and SCHAEFFER, D.: Imperfect bifurcation in the pre-
sence of symmetry. Commun. Math. Phys., 67 (1979), pp. 205-232.
[5~ HAJEK, M. and JANOVSKY, V. and NEUBERG, J.: On stability of Stimu-
lated Brillouin Scattering. Technical Report KNM MFF No 076/85,
Charles U n i v e r s i t y of Prague, 1985.
SUPERCONVERGENCE RESULTS FOR
LINEAR TRIANGULAR ELEMENTS
M. Kl~t2EK
Mathematical Institute, Czechoslovak Academy of Sci,enees
115 67 Prague 1, Czechoslovakia

The a i m of the p a p e r is to p r e s e n t several superconvergence pheno-

mena which have been observed and analyzed when employing the standard
linear elements to s e c o n d order elliptic problems. We s h a l l illustrate
them in t h e i r simplest form solving the m o d e l problem:

- Au = f in ~ C R2 , (I)
u = 0 on 3~ ,
where ~ is a c o n v e x polygonal domain and u is s u p p o s e d to be s m o o t h
enough.
Let {Th} be a r e g u l a r family of t r i a n g u l a t i o n s of ~ , i.e., ZI~-
mal's condition on the m i n i m a l angle of t r i a n g l e s is f u l f i l l e d . The dis-
crete analogue of (1) w i l l consist in f i n d i n g uh C Vh such that

(VUh,VVh)0,[~ = (f,vh)0, ~ ~ vh £ Vh , (2)

where
I
Vh = {Vh 6 H0(~) I Vhl T 6 P] (T) ~T C Th} .

It is k n o w n E15,39~ that the e r r o r estimates

Cph 2 I l u 1 1 2 , p , a if p ~ E2, ~) ,
- (3)
tlu UhtlO'P'~ ~ < Ch211n h I l l u l l 2 , ~ , f l if p = ~ ,

llVu - ? U h l I O , p , ~ G Chllull2,p, ~ if p ~ ~,~] , (4)

are o p t i m a l . Nevertheless, we can improve the o r d e r convergence (in some

norm IIl'lll w h i c h is c l o s e to I I-110,p, e ) by a s u i t a b l e post-proces-
sing ~ , and this we call the superconvergence. The p o s t - p r o c e s s i n g
should be e a s i l y computable and the n o r m Ill. Ill m a y be e.g. a discrete
analogue of l l.llO,p, ~ , or II1" III = l l'll0,P,e0 for ~0 C C fl (i.e.
~0 c n ), or 111"111 = ll'110,p, a , etc. We introduce several examples
where ~ is a r e s t r i c t i o n operator to s o m e subset of ~ , an a v e r a g i n g
and an i n t e g r a l smoothing operator. L e t us e m p h a s i z e that many supercon-
vergence phenomena are very sensitive to the m e s h geometry (therefore,
uniform, quasiuniform or p i e c e w i s e uniform triangulations are m o s t l y
 316

employed). In this paper, we a s s u m e for b r e v i t y that each Th is uni-

form, i.e., any two adjacent triangles of Th form a parallelogram.

Let Nh be the set of n o d a l points of Th . Then the use of the

expansion theorem for l i n e a r elements by E3~ yields (cf. (3))

max
x e Nh
lu(x) - Uh(X) I ~ Ch4ttullc4(~ ) , (s)

provided Th consists of e q u i l a t e r a l triangles. We m e n t i o n that the

(stiffness) matrix arising from (2), w h e n taking the standard Courant
basis functions, is the same as for the w e l l - k n o w n 7-point finite diffe-
rence scheme (see e.g. E3~ , p. 91)

2 (6Uo . Ul .- U2 . - U3 . U4 , U5 U6) = h 2 f0 + h 4 A f 0 / 1 6

with the r a t e of c o n v e r g e n c e 0(h 4)

Remark I. Using (I), (2), (5), a n d the affine one-to-one mapping F

between any uniform triangulation Th and a triangulation Th consis-
t i n g of e q u i l a t e r a l t r i a n g l e s , one e a s i l y o b t a i n s an a n a l o g u e of (5) for
-1
Th = F (T h) , indeed, b u t for o t h e r e q u a t i o n . F o r i n s t a n c e , the t r i a n -
gulation sketched in Fig. I guarantees the nodal superconvergence for
the e q u a t i o n - Aft + 02fi/OxOy = f .

\ / //!,
Fig. 1 Fig. 2

Remark 2. A c o n v e n i e n t c o m b i n a t i o n of l i n e a r and b i l i n e a r e l e m e n t s m a y
4
give the 0(h )-superconvergence at n o d e s for the p r o b l e m (I) on t r i a n -
gulations consisting of r i g h t - a n g l e d triangles. Let {u i} and {v i} be
the Courant piecewise linear basis functions over the triangulation of
Fig. I and 2, r e s p e c t i v e l y , and let {t i} be the standard basis functi-
ons for b i l i n e a r rectangular elements. Put

w i = ti/2 + ui/4 + vi/4

and d e n o t e by Wh the linear hull of {w i} (dim W h = d i m Vh). Now, the

matrix arising from (2), if w e replace Vh by Wh , is the same as for
the 9-point difference scheme over square meshes [35~, p. 90; and it is
thus easy to d e r i v e the r a t e 0(h 4) at n o d e s employing the basis {w i}
The n e x t table shows the v a l u e s of the m a x i m u m error over all n o d e s for
various choices of b a s i s functions when u(x,y) = y(y - I) sin ~x is
the exact solution of (1) on the u n i t square ~ = (0,1) x (0,1)
 317

h-1 Vi . .
(Vl+Ul)/2 .
ti. (tl+Vl)/2 Wi
4 1.2069 E-2 1.2069 E-2 1.2962 E-2 6.0703 E-4 1.6832 E-4
8 3.1027 E-3 3.1027 E-3 3.1589 E-3 1.3156 E-4 1.0307 E-5
16 7.8126 E-4 7.8126 E-4 7.8478 E-4 3.5250 E-5 6.4092 E-7
32 1.9567 E-4 1.9567 E-4 1.9589 E-4 8.7640 E-6 4.0006 E-8

Further we present superconvergence results for the gradient of

u h ~ V h . According to ~ ,26~, the tangential component of Vu h is a
superconvergent approximation to the tangential component of Vu at
midpoints of sides. Denoting by Mh the set of these midpoints, we may
then define a recovery operator for both the components of the gradient
by the relation (see L4,8,9,11,26,28,30,31,33,40~)

VUh(X) I
= ~(VUhIT1 + VUhlT2 ) , x ~ Mh ~ ~ , (6)

where T 1 , T 2 6 Th are those adjacent triangles for ~ T 4 / ~

which x 6 TI ~ T2 (note that VUlT i is constant).
As shown in ~ 1 , 3 0 ~ ,

max I IVu(x) - V~h(X) II ~ Ch211n hi llu113,=, ~ Fig. 3

x E Mh~

or even 0(h 2) for the discrete L2-norm [26~ (cf. (4)). For a three-di-
mensional analogue of (6), see [5~.
Note that the sampling at centroids of the bilinear elements leads
to the superconvergence of the gradient [24]. This is not true for the
linear elements. However, a weighted averaging scheme between four ele-

ments, 3 Ch ~0 ~..... ~ /
"-J
VUh(X) I
= ~(3VUhl T + ~ VUhiTi
)~ , x e ,
i=I
yields [26j
h( E I Ivu(xl _ ~VUh(Xl 112)~ ~Ch211ull3,~
x E C h ~ ~0 Fig, 4

Here Ch is the set of centroids of all T 6 T h , ~0 C C.~, and TI,T2,

T 3 £ Th are the triangles adjacent to that triangle T 6 Th for which
x 6 T . Using (6), one can define a discontinuous piecewise linear field
V~Uh which recovers the gradient of u even at any point of ~0 C C
(see [36~). By the following averaging at nodes x 6 N h we may determine
a continuous piecewise linear field _V~uh over the whole domain ~ :

I ~ VUhl T , x6 N h ~ ~ , ~ 8 ~
T ~ {x}¢~
VUh(X) = 0 , x 6Y , (7) X
3
~(i~l~Uhl
, _ Ti-~Uhl %1 x~ N h ~ ( ~ ~),
Fig. 5
 318

where Y is the set of v e r t i c e s of ~ , Ti and T3 form a parallelo-

g r a m for e v e r y i = 0,1,2 , and T 1N T 2 N T3 = {x} when x6N h~ (8~-Y)
- see Fig.5. In this c a s e the g l o b a l superconvergence estimate reads E23~ :

llvu ?~hIIO,p, ~ ~ Ch211n hll-2/PIluIl3,p,~ , p E {2,~} . (8)

F o r the g e n e r a l i z a t i o n of the s c h e m e (7) to e l l i p t i c s y s t e m s w i t h non-
homogeneous boundary conditions of s e v e r a l types, we refer to E20]. If
8~ is s m o o t h then a local 0 ( h 3 / 2 ) - s u p e r c o n v e r g e n c e in ~0 C C ~ can be
achieved ~O,21J in the L 2 - n o r m ( Th are not u n i f o r m n e a r the b o u n d a r y
~ ).
Consider now triangulations as m a r k e d in Fig. I or 2 and the s m o o t h -
ing p o s t - p r o c e s s i n g operator

Uh(X) = ~h-2 IU h ( X + y) dy ,
Dh
where Dh = (-h,h) x (-h,h) . If n0 C C ~ and 8~ is a g a i n smooth then
(see [37,38])
fin- ~hIl1,% =< Ch 3/2
IIull3,~
which is, in fact, a superconvergent estimate for the g r a d i e n t .
Another t y p e of an i n t e g r a l smoothing operator which yields a super-
convergent approximation for 7u as w e l l as for u e v e n on i r r e g u l a r
meshes is p r e s e n t e d in [3]. In [I~ a least squares smoothing of Vu h
is p r o p o s e d to o b t a i n a better approximation to 7u . Related papers with
superconvergence of l i n e a r e l e m e n t s further include ~,6,7,12,13,17,18,
25,29,3~, see a l s o the s u r v e y p a p e r s ~O,22,27].
Let us n o w t u r n to s u p e r c o n v e r g e n t approximations to the b o u n d a r y
flux q = ~8u
18~ ( n is the o u t w a r d unit normal to 8~ ). S e t t i n g

q h : n.v~hl 8~ '
where ~u h is g i v e n by (7), we i m m e d i a t e l y get f r o m (8) t h a t

I lq - q h I l 0 ~ 8 ~ ~ ch2 IIn hl I lull3 ~,~ ,

i.e., the c o n t i n u o u s piecewise linear function qh approximates q

b e t t e r t h a n the p i e c e w i s e constant function qh = n ' v u h I ~ "
Another continuous piecewise linear approximation qh to the b o u n -
dary flux q can be d e f i n e d w i t h the h e l p of G r e e n ' s formula

I qh Vh ds = (7Uh,VVh)0, ~ - (f,vh)0, n ~v h e Uh ,
8Q
where
u h = {v h e H 1(n) [ Vhl T e PI(T) ~T e Th} .
This technique suggested by E16], p.398, is b a s e d on some ideas of [14~.
Numerical t e s t s of the p r e s e n t e d superconvergent schemes can be
f o u n d in ~,6,11,19,21,23,24,26,36J .
 319

R e f e r e n c e s

[ I] ANDREEV, A. B.: Superconvergence of the gradient for linear trian-

gle elements for'elliptic and parabolic equations. C. R. Acad.
Bulgare Sci. 37 (1984), 293-296.
E 2] ANDREEV, A. B., EL F~TRI , M. and LAZAROV, R. D.: Superconvergence
of the gradient in &he fimite element method for some elliptic and
parabolic problems (Russian). Variational-Difference Methods in
Math. Phys., Part 2 (Prec. Conf., Moscow, 1983), Viniti, Moscow,
1984, 13-25.
E 3] BABU~KA, I. and MILLER, A.: The post-processing in the finite ele-
ment method, Part I. Internat. J. Numer. Methods Engrg. 20 (1984),
1085-1109.
[ 4] CHEN, C. M.: Optimal points of the stresses for triangular linear
element (Chinese). Numer. Math. J. Chinese Univ. 2 (1980), 12-20.
[ 5] CHEN, C. M.: Optimal points of the stresses for tetrahedron linear
element (Chinese). Natur. Sci. J. Xiangtan Univ. 3 (1980), 16-24.
~ CHEN, C. M.: Finite Element Method and Its Analysis in Improving
Accuracy (Chinese). Hunan Sci. and Tech. Press, Changsha, 1982.
7J CHEN, C. M.: Superconvergence of finite element approximations to
nonlinear elliptic problems. (Proc. China-France Sympos. on Finite
Element Methods, Beijing, 1982), Science Press, Beijing, Gordon
and Breach Sci. Publishers, Inc., New York, 1983, 622-640.
8] CHEN, C. M.:~An estimate for elliptic boundary value problem and
its applications to finite element method (Chinese). Numer. Math.
J. Chinese Univ. 5 (1983), 215-223.
L 9] CHEN, C. M.: WI, ~ -interior estimates for finite element method
on regular mesh. J. Comp. Math. 3 (1985), I-7.
O] CHEN, C. M.: Superconvergence of finite element methods (Chinese).
Advances in Math. 14 (1985), 39-51.
I] CHEN, C. M. and LIU, J.: Superconvergence of the gradient of trian-
gular linear element in general domain. Preprint Xiangtan Univ.,
1985, 1-19.
~ CHEN, C. M. and THOM~E, V.: The lumped mass finite element method
for a parabolic problem. J. Austral. Math. Soc. Ser. B 26 (1985),
329-354.
3~ CHENG, S. J.: Superconvergence of finite element approximation for
Navier-Stokes equation. (Prec. Conf., Bonn, 1983), Math. Schrift.
Me. 158, Bonn, 1984, 31-45.
[143 DOUGLAS, J., DUPONT, T. and WHEELER, M. F.: A Galerkin procedure
for approximating the flux on the boundary for elliptic and parabo-
lic boundary value problems. RAIRO Anal. Num~r. 8 (1974), 47-59.
[15] FRIED, I.: On the optimality of the pointwise accuracy of the fi-
nite element solution. Internat. J. Numer. Methods Engrg. 15
(1980), 451-456.
[16] GLOWINSKI, R.: Numerical Methods for Nonlinear Variational Prob-
lems. Springer Series in Comp. Physics. Springer-Verlag, Berlin,
New York, 1984.
~ EL HATRI, M.: Superconvergence of axisymmetrical boundary-value
problem. C. R. Acad. Bulgare Sci. 36 (1983), 1499-1502.
~8] EL HATRI, M.: Superconvergence in finite element method for a dege-
nerated boundary value problem (to appear), 1984, I-6.
9] HINTON, E. and CAMPBELL, J. S.: Local and global smoothing of dis-
continuous finite element functions using a least squares method.
Internat. J. Numer. Methods Engrg. 8 (1974), 461-480.
~0] HLAV~EK, I. and K~f~EK, M.: On a superconvergent finite element
scheme for elliptic systems, I. Dirichlet boundary conditions, II.
Boundary conditions of Newton's or Neumann's type (submitted to
Apl. Mat.), 1985, 1-29, 1-17.
~I] K~f~EK, M. and NEITTAANM~KI, P.: SuperconVergence phenomenon in
the finite element method arising from averaging gradients. Numer.
 320

Math. 45 (1984), 105-116.

[22] K~f~EK, M. and NEITTAANMAKI, P.: On superconvergence techniques.
Preprint No. 34, Univ. of Jyv~skyl~, 1984, 1-43.
E23~ K~f~EK, M. and NEITTAANM~KI, P.: On a global superconvergence of
the gradient of linear triangular elements. Preprint No. 85/4,
Univ. Hamburg, 1985, 1-20.
[24] LASAINT, P. and ZLI~AL, M.: Superconvergence of the gradient of
finite element solutions. RAIRO Anal. Num~r. 13 (1979), 139-166.
[253 LEVINE, N.: Stress ampling points for linear triangles in the fi-
nite element method. Numer. Anal. Report 10/82, Univ. of Reading,
1982.
[26] LEVINE, N.: Superconvergent recovery of the gradient from piecewise
linear finite element approximations. Numer. Anal. Report 6/83,
Univ. of Reading, 1983, 1-25.
~7~ LIN, Q.: High accuracy from the linear elements. Proc. of the Fifth
Beijing Sympos. on Differential Geometry and Differential Equa-
tions, Beijing, 1984, I-5.
~8] LIN, Q. and LU, T.: Asymptotic expansions for finite element appro-
ximation of elliptic problem on polygonal domains. Comp. Methods
in Appl. Sci. and Engrg. (Proc. Conf., Versailles, 1983), North-
-Holland Publishing Company, INRIA, 1984, 317-321.
~9~ LIN, Q. and LU, T.: Asymptotic expansions for finite element eigen-
values and finite element solution. (Proc. Conf., Bonn, 1983),
Math. Schrift. No. 158, Bonn, 1984, 1-10.
E30] LIN, Q., LU, T. and SHEN, S.: Asymptotic expansion for finite ele-
ment approximations. Research Report IMS-11, Chengdu Branch of
Acad. Sinica, 1983, I-6.
~I~ LIN, Q., LU, T. and SHEN, S.: Maximum norm estimate, extrapolation
and optimal point of stresses for the finite element methods on
the strongly regular triangulations. J. Comput. Math. 1 (1983),
376-383.
~2] LIN, Q. and WANG, J.: Some expansions of the finite element appro-
ximation. Research Report IMS-15, Chengdu Branch of Acad. Sinica,
1984, 1-11.
~3] LIN, Q. and XU, J. Ch.: Linear elements with high accuracy. J.
Comp. Math. 3 (1985), 115-133.
~34] LIN, Q. and ZHU, Q. D.: Asymptotic expansion for the derivative of
finite elements. J. Comp. Math. 2 (1984), 361-363.
[35] MICHLIN, S. G. and SMOLICKIJ, Ch. L.: Approximation Methods for
Solving Differential and Integral Equations (Russian). Nauka, Mos-
cow, 1965.
~6] NEITTAANM~KI, P. and K~f~EK, M.: Superconvergence of the finite
element schemes arising from the use of averaged gradients. Accu-
racy Estimates and Adaptive Refinements in Finite Element Computa-
tions, (Proc. Conf., Lisbon, 1984), Lisbon, 1984, 169-178.
~7] OGANESJAN, L. A., RIVKIND, V. J. and RUCHOVEC, L. A.: Variational-
-Difference Methods for the Solution of Elliptic equations (Rus-
sian). Part I (Proc. Sem., Issue 5, Vilnius, 1973), Inst. of Phys.
and Math., Vilnius, 1973, 3-389.
~8] OGANESJAN, L. A. and RUCHOVEC, L. A.: Variational-Difference
Methods for the Solution of Elliptic Equations (Russian). Izd.
Akad. Nauk Armjanskoi SSR, Jerevan, 1979.
~9~ RANNACHER, R. and SCOTT, R.: Some optimal error estimates for pie-
cewise linear finite element approximations. Math. Comp. 38 (1982),
437-445.
~40] ZHU, Q. D.: Natural inner superconvergence for the finite element
method. (Proc. China-France Sympos. on Finite Element Methods,
Beijing, 1982), Science Press, Beijing, Gordon and Breach Sci.
Publishers, Inc., New York, 1983, 935-960.
MIXED FINITE ELEMENT IN 3D IN H(div)
AND H(curl)
J. C, NEDELEC
Ecole Polytechnique, Centre de Mathdmatiques Appliqudes
91128 Palai~eau, France

I. INTRODUCTION.

Frayes De Venbeke first i n t r o d u c e the mixed finite element. Then P.A. Raviart
and J,M. Thomas does some mathematics on these element in 2D and others do also :
F. Brezzi V. B abuska ...
In 1980 we introduce a family of some mixed finite element in 3D and we use them for
solving Navier Stokes equations.
In 1984 F, Brezzi, J. Douglass and L.D. Marini introduce in 2D a new family of mixed
finite element conforming in H(div). That paper was the starting point for building
new families Of finite element in 3D,

II. FINITE ELEMENT IN H(div).

Notations.

K is a tetrahedron
~K its boundary
n the normal
f a face which area is I d
2
f r
a is an edge which lenght is J~ ds

curl u = V ^ u u = (ul, u2, u3)

H(curl) = {u E L2(~)) 3 ; curl u E (L2(~)) 3 }

div = V. u
H(div) = {u E (L2(~)) 3 ; div u @ L2(~) }

Spaces of polynomials.

Pk = polynomials of degree less or equal to k

~k = " homogeneous of degree k
x!
D k = (Pk_l)3 + PNk_ l r

Sk = {p E (pk) ; (r.p) ~ 0 }
r =

I
x2
x3

= (Pk_|) 3 ~ Sk
 322

dim Sk = k(k + 2)

dim ~k = (k + 3)(k2 + I) k

= (k + 3)(k + 2) k
dim
k
2

We are now able to introduce the finite element conforming in H(div).

Definition. We define the finite element by

I) K is a tetrahedron
2) P = (Pk)3 is a space of polynomials
3) The set of degrees of freedom which are

(3.1) (p . n)q dy ; V q E pk(f) ;

f

(3.2) jf (p . q) dx ; V q E ~ k - 1
K
we h a v e the

Theorem.

The above finite element is unisolventand conforming in H(div). The associate in-
terpolation operator H is such that

div ~p = 9" div p ; V p E H(div) ,

where ~* is the L 2 projection on Pk-1 "

When k = I, the corresponding element has no interior moments and 12 degrees of free-
dom. Its divergence is constant.

Proposition. For a tetrahedron "regular enough" which diameter is k, we have

II p - ~p N(L2(K))3 < c h k+1 []p ][(Hk+I(K))3 ;

II D(p - ~p) [l(e2(K))3 < c hk Itpll(Hk+1(K))3

We are not going to prove this theorem. But we can recall that a finite element
is said to be conforming in a functional space if the interpolate of an element of
this space belong to this space.

In our case, the conformity in H(div) is equivalent to the continuity of the

normal composent at each interface. This property is clearly true for our finite ele-
ment since the unknowns on the face are

I (p . n) q dy ; V q E Pk(f)
f
and p.n is also Pk(f).
 323

III. FINITE ELEMENT IN H(curl).

A finite element is conforming in H(curl) if the tangential components are con-

tinue at the interface of the mesh.
We introduce the corresponding finite element.

D~finition.

|) K is a tetrahedron
2) P = (Pk)3 is the space of polynomials
3) The degrees of freedom are the following moments

3.1) la(p T) q ds ; V q • Pk(a)

3.2 ) (p t q) dy ; V q E ~k_|(f) and tangent to the face f

3.3) ~K (p . q) dx ; V q E %_ 2

We have the

Theorem.

The above finite element is unisolvent and conformlng in H(curl). Moreover if

H is the corresponding interpolation operator and If* the interpolation operator as-
sociate to the H(div) finite element introduce previously for degree k-! we have

curl ~[p = If* curl p

IV. APPLICATION TO THE EQUATION OF STOKES.

The Stokes'equation is usually written in the (u,p) variable in a bounded domain

of R 3 as
- ~ Au + grad p = f , in
div u = 0 in

ul r = o

We introduce the vector potential ~ as

- A~ = curl u , in
div ~ = 0 , in

^nit = o
Then the Stokes equation can be written in the (~,~) variables where

to = curl u
We introduce
H(div O) = { v E (L2(~)) 3 ; div v E 0 , v.nlr = 0 }

H = { ~ E H(curl) ; div ~ = 0 ; ~niF = 0 }

 324

Then a variational formulation of the Stokes equation is

Ir (curl w.curl ~)dx = IO (f.curl ~)dx ; V ~ E ff

I (w.H)dx - [ (curl ~.curl ll)dx : 0 ; V H E H(curl)

Let C h be a mesh covering ~ .

We can introduce some finite element spaces

Wh = { toh c H(curl) ; ~0hlK E (pk)3 ; V K ~ C k }

Wh= 0 { wh E Wh ; ~ohr, nIF = o }

v h = { v h E H(div) ; Vhl K E (pk_l) 3 ; V K C Ch }

U h = V h N H(div 0)

The approximate problem become then

I (curl Wh. Vh)dX = I (f'Vh)dX ; V Vh ~ Uh ;

I (Wh'~h)dX - I (Uh. CUrl Hh) dx = 0 ; V TIh E W h

We can also use a vector potential ~h"

This goes like that

@h = { Oh E HI(12) ; 0hiE @ Pk+l ; V K e Ch }

o n H~ (g%)
@h = Oh

We have the

Theorem.

If the transgulation is regular,for every v h E Uh,there exist a unique

~h @ Wh0 such that

curl ~h = v h

I (~h " grad 0h)dX = 0 ; V eh @ e h

o
and we have also
ll~hllH(curl) < c llVhll(L2(~)) 3 •

This theorem can be use to transfer the above approximate problem in one in (~,c0) and
also to find a local basis in the space U h.
 325

BIBLIOGRAPHY

F. BREZZI, On the existence, uniqueness and approximation of saddle point

problems ausing from Lagrangian multip]iers. RAIRO 8 : 129 - 151 (1974).

F. BREZZI, J. DOUGLASS & L.D. MARINI, Two families of mixed finite elements
for second order elliptic problems. To appear in Numerische Mathematik.

P.G. CIARLET, The finite el~ment method for elliptic problems. North Holland
Amsterdam (1978).

P.G. CIARLET & P.A. RAVIART, A mixed finite element method for the biharmonic
equation. Mathematical aspects in finite element method (C de Boor ed.) pp. 125 -
145 Academic Press New York (1974).

M. FORTIN, An analysis of the convergence of mixed finite element method.

RAIRO 11 : 341 - 354 (1977).

J.C. NEDELEC, Mixed finite element in ~3. Numerische Mathematik 35 : 315 - 341
(1980).

J.C. NEDELEC, Elements finis mixtes inaompressibles pour l'equation de Stokes

dans ~ . Numerische Mathematik 39, 97 - 112 (1982),

P.A. RAVIART & J.M. THOMAS, A mixed finite element method for 2nd order elliptic
problems, In Dold A Eckmann B (eds), Mathematical aspects of finite element
methods. Lecture Notes 606 Springer Berlin (1977).

J.M. THOMAS, Thesis Paris (1977).

J.C. NEDELEC, A new family of mixed finite element i n ~ 3 (~ parattre).

FREE BOUNDARY PROBLEMS FOR
STOKES' FLOWS A N D FINITE ELEMENT
METHODS
J. A. NITSCHE
Institut fiir angewandte Mathematik, Albert-Ludwigs-Universitgt
Freiburg i'm Breisgau, West Germany
Abstract:
In two dimensions a 5tokes" flow is considered symmetric to the abscissa
n = 0 and periodic with respect to T. On the free boundary II'II= 8(T) the
conditions are: (i) the free boundary is a streamline, (ii) the tangential
force vanishes, (iii)the normal force is proportional to the mean curvature
of the boundary. By straightening the boundary, i.e. by introducing the var-
iables x = T, Y = qlS(~'), the problem is reduced to one in a fixed domain.
The underlying differential equations are now highly nonlinear: They con-
sist in an ellipticsystem coupled with an ordinary differential equation for
8. The analytic properties of the solution as well as the convergence of the
proposed finite element approximation are discussed.

I_~.In accordance to the restrictions formulated in the abstract the problem under consid-
eration is: We ask for the free boundary n = 8(T), l-periodic in T, such that there exists a
solution pair U = (UI,Uz) and P with the properties:

(iI) In the domain ~ = { (T,n) l lni < 5(I~) } the system of differential

equations
(I. I) 6iklk = Fi

hold true with

(I.2) 6ik = Uilk ÷ Ukli - P6ik.

(i 2 ) In the domain EZ the incompressibility condition

(i.3) V, U = U11~ + U2t~ = 0

holds true.
 328

(iiI) The free boundary rl = -+ 80[) is streamline, i. e.

(I.4) U2-SIUI = 0 forri=-+5([).

(ii2) On the free bouncJary the shear-force vanishes, i.e.

(i.s) 6iktink ~ 0
with t = (tl,tz) and n = (nl,n,~) being the tangential resp. normal
unit vectors.

(ii 3) The normal-force is proportional to the mean curvature, i. e.

(I. 6) 6ik hi rlk -- KH .

We will consider fluid motions only "not too far" from LI° : (1,0). Together with pO = 0 and
5 ° = I the trine (U°,P°,5 °} is a solution to the problem stated above with F ° = O. - The
main idea of our analysis is the "straigthening" of the free boundary, quite often used. This
consists in introducing new variables
(I. 7) x = [ , Y = n 1 5([)
Since we are looking for solutions {U,P,5} near to {Uo,P°,5 ° } we replace U, P and 5 - de-
pending on [,q - by (1+ul,u2), p and 1+s depending on x,y. This leads to a nonlinear problem
in the new variables but now in the fixed domain

(I. B) Q+ = { (x,y) I lyl < I } .

Because of our setting all functions are assumed to be l-periodic in x. For functions F resp.
in the new variables f symmetric with respect to y = O, i. e. f1(x,-y) = fi(x,y) and f2(x,-y) =
-fz(x,y), the solution also will be symmetric to y = 0. Hence we can restrict ourselves to
the unit square

(1.9) Q = { (x,y)I O < x , y < 1 }.

The condition of symmetry implies the boundary conditions

u2(x,O) = 0 ,
(1.10)
uiLy(x,O) = 0
By linearizing, i.e. by spitting into linar and nonlinear terms, we get from (1.1)the system

a~(2ulfx-p) + 8y(ui19+u21~) : ~x~11 ÷ ay~-12 ÷ fl ,

(1.11)
8x(ully+uzt~) + 89(2u2t~l-p) = cn~:21 + c~y~-22 + f2 .
 329

Here Zik = ~-ik(u,p,s)are at least quadratic in their arguments, for example it is

(1.!2) Y-lZ = -2yslu11~ + 2(1+s)-l(1+y2sl2)UZly - (l+s)-lystu21y + yslp.
In the new variables condition (I. 3) becomes
UIIx + U21y =: I)
(1.i3)
= (1+S)-1(yslU|ly+ SUZlg)
The boundary condition (I. 4) may be used as defining relation for s --s(x):

ss : (t+ul)-lu2
(1.14)
=: u2 + P
(I. 5) leads to a boundary condition of the type
(1.15) Ullu + u21x = T1 .

The mean curvature H of the free surface depends on the second derivative 8 ~j resp. s ~I
This quantity may be computed from (1.14).In this way (I. 6) leads to the second boundary
condition of the type
(1.16) 2U2Iu - P + KU21x = I"2 .

The Ti = Ti(y,p,s) are at least quadratic in their arguments. Similar to the ~-ikthey depend
only on the functions themselves and their first derivatives. 81nce s is assumed to be
l-periodic we have Ss ~ = 0. Here S w resp. later ,I'j'ware abbreviations defined by
I
(1.17) J'w = j'w(x,1)dx, ,i'J"w = J'j"w(x,y)dxdy
0 Q
In view of the boundary condition (1.10) we get from (1.13) J'SD = -J'uz . Therefore the
quantity

1{ = J'J'o ~P
(1.18)
=: ¥(u,p,s)
will be zero. Hence we may replace in (1.13)the right hand side D by
(1.19) B = D - I{ .

In the new variables we have the

Problem:
Given the vector f defined in Q (I. 9) and l-periodic in x. Find u, p, s l-periodic in
x, fulfillingthe differential equations (1.11),(1.13) in Q, and the boundary condi-
tion (1.10)on y=O as well as (1.14),(1.15),and (1.16)on y=1.
 330

2__,.The idea of proving the existence of a solution of the problem as well as deriving a fi-
nite element m e t h o d in order to approximate this solution is as follows: We consider
the quadruple In = {ul,uz,p,s} as an element of a linear space 111 equipped with an appropri-
ate norm. The geometric boundary condition (1.101)has to be imposed on uz. Obviously Ul as
well as s are defined up to a constant only. Therefore we nomalize ul, s according to J'J'uv =
0, J's = O. The correspondent restriction of the space III will be denoted by "111 . $imilar-
ily we consider the octuple n = {:EI1,~-1z,£zI,~-2,z,D,P,TI,T2} as an element of a linear
space 11, also equipped with a norm. By (1.12), (1.13) etc. the mapping A - III ~ II is de-
fined. The mapping B " 11 ~ "III which associates the solution of the boundary value
problem to the right hand sides is constructed by the natural weak formulation of the
problem: If m E "III is the solution then with any p = {v,q, r} E 'III the variational equa-
tions hold:

a(m,p) + b(m,p) = L1(n,p) + F(f,p)

(2.2) b(p,m) = Lz(n,p)
c(m,ii ) - j'uzrI = j'pr I
Here Li, F, a, b, c are bilinear functionals; especially a(.,.),b(.,.),and c(.,.)are defined by

a(m,p) =S£[2UII~VII~+(UIIy+UzI~)(VIIy+VZI~)÷2UZlyV21y]-K£U21xY2
(2.3) b(m,p) =S.f q {Ull~ + uzl~l}
c(m,p) = .I" sir t
The standard inf-sup condition is valid for the form b(.,.), because of Korn's second ine-
quality a(.,.) may be extended to a bounded and coercive bilinear form in the 8obolev space
HI(Q)xHKQ). In connection with the normalisation of ul and s uniqueness of the mapping B
is guaranteed.

5._,.Since the mapping A is nonlinear we will work with HGIder-spaces: We equip the spaces
"III and 11 in the following way with norms, in these topologies they are Banach-spaces:
For p = {v,q,r}e "III we define

mpm:: InpI.m
(3.1)
= ~;'IIviIIcI.~(Q
) + IIqIIco.x(Q) + Ilrflc2.~(1)
 331

Here II.IIck.~(.
) denote the usual HOlder-norms with A E (0,1], l is the unit interval

For w = {~11,~-12,~.21,~22,D,P,ThT2} EII we define

IIIIll.= Ill
(3.2)
= 7~-iknCo.~(Q)+ H6HCo.~(1)+ HPIIcI.~(Q) + 7gTiIIco,x(l)
NOW we consider elements p in the ballB6("III) := {II I P E "III ^ Ilpll_~ 5} with 6 < 5o < I
and 6o fixed. Obviously the two estimates are valid:

(3.3)
n.=ii. _< c6fl1¢- =iHm
Here "c" denotes a numerical constant depending only on 6o which may differ at different
places.
It can be shown: The mapping B is bounded, i.e. for m = Bn the estimate

(3.4) HI,,HI _< + Zuf Uco.,,(Q )

is valid. Thus the Banach Fixed Point Theorem leads to: For IIfiflsufficiently small and 6
chosen approprlately the mapping
(3.5) T := B IR
possesses an unique fixed point in the ball B6('lll). It turns out that the quantity ¥ (1.18)
vanishes. This implies that the fixed point corresponds to the solution of the original
problem.

4~ NOW let "111h be an appropriate finite element approximation space. By restricting in

(2.2) the elements p = ph E "111hand looking for the solution mh E "1111hthe mapping Bh and
consequently also Th (see (3.5)) is defined.
It can be shown: Under certain conditions concerning the approximation spaces, especially
the Brezzi condition is needed, the mapping Bh is bounded, i. e. an inequality of the type
(3.4) holds true. This finallyleads to almost best error estimates: Let m E "111 and mh E
111h be the solution of the analytic problem resp. the finite element solution then

(4,1) |m - mh I .< Cinf{ | m - Phi I ph E "111h} .

 332

The proofs and the complete bibliographywillappear elsewhere. Here we refer only to

Bemelmans, J. (1981a)
GIeichgewichtsfiguren zSher FI(~ssigkeitenmit Oberflachenspannung
Analysis I,241-282 (1981)

Bemelmans, J. (1981b)
L!quid Drops in a viscous Fluidunder the Influenceof Gravity and 8urface Tension
Manuscripta math. 36, 105-123 (1981)

Bemelmans, J. and A. Friedman (1984)

Analiticity for the Navier-Stokes Equations Governed by 8urface Tension on the Free
Boundary
J. of Diff.Equat. 55, 135-150 (1984)

Nitsche, J. A.
8chauder Estimates for Finite Element Approximations on second Order EllipticBoundary
Value Problems
Proceedings of the Special Year in Numerical Analysis, Lecture Notes #20, Univ. of
Maryland, Babuska, I.,T.,-P.Liu, and J. Osborn eds., 290-343 (1981)

8chulz, F. (1982)
Ober eIIiptische Monge-Amperesche Differentialgleichungen mit einer Bemerkung zum
Weylschen Einbettungsproblem
Nachr. Akad. Wiss. G6ttingen, IIMath.-Phys. KIasse 1981, 93-108 (1982)
ENCLOSING METHODS FOR PERTURBED
BOUNDARY VALUE PROBLEMS
IN NONLINEAR DIFFERENCE EQUATIONS
J. W. SCHMIDT
Technical University QFDresden
Mommsenstr. 13, Dresden, DDR

i. In t h e lecture nonlinear equations Fa(Z) = 0 are considered de-

pending on an input parameter vector a which m a y be s u b j e c t e d to
errors, shortly a 6 A. In o r d e r to s t u d y the influence of the input
a E A on t h e solutions z , by m e a n s of m o n o t o n e e n c l o s i n g m e t h o d s
a
intervals are c o n s t r u c t e d c o n t a i n i n g for e a c h a E A at l e a s t o n e
solution za . Such a type of m e t h o d s c a n be d e v e l o p e d if the o p e r a -
tors Fa possess some monotony properties, see S C H M I D T / S C H N E I D E R
[l].

2. T h e F D M - d i s c r e t i z a t i o n of the boundary value problem

u'' = 2~0sinh u - ~(t~, u(0) = p, u(~) = q- (2.1)

appearing in i n n e r electronics is c h o s e n as a m o d e l problem. Let the

net d e n s i t y ~ be a i v e n bv

~(t) = ~(t,a I ..... a 7 ) : (2.2)

i0 -a2 t2 ~ 1 0 8 e _ ~ 4 t 2 a 8 e _ ~ 7 ( t _ ¥)2
i0 e + + 105+ ~610
i 3 5

In g e n e r a l the parameter vector a = (~0" ... "a7 )T is a f f e c t e d with

errors,

a = a ± e, e = ( e 0 , . . . , e 7 )T . (2. 3)

This vector interval represents the set A.

Applying the common finite difference method to (2.2) (h = y / ( N + i)

step size, t i = i h nodes, ~i approximation to u ( t i ) ) one gets the
following system of e q u a t i o n s

Fa(Z) = F:(z) + Fa(Z) (2.4)

with
 334

(F (z)) i = -~i_1% 2~ i- ~i+]+ ~0h2e~ i (2.5)

(Fa(Z)) i = _ ~ o h 2 e -~i_ h 2 ~ ( t i , a l .... ~ 7 ) (i = I,...,N)

and ~0 = p' ~N+I = q" H e r e the i-th c o m p o n e n t of a v e c t o r z is

written as (z) i = ~i' and so on. O b v i o u s l y , the o p e r a t o r s Fa are
offdiagonally antitone, the d e r i v a t i v e s DF~ are isotone a n d the d e r i -
vatives DF a are a n t i t o n e if s 0 > 0. T h e s e properties are e s s e n t i a l
in w h a t follows.

3. L e t R,S be f i n i t e d i m e n s i o n a l linear spaces partially ordered by

c l o s e d cones. Thus these cones are n o r m a l and r e g u l a r , too. For a
continuous operator

F : D = [Yl,Xl] C R ~ S (3.17

a mapping AF : D X D ~ L(R,S) is c a l l e d an i s o t o n e - a n t i t o n e divided

difference operator if

F(x)-F(y) ~ AF(x,y)(x-y) for Yl ~ y S x S Xl~ (3.2)

AF(x,y) ~ AF(u,v) for Yl ~ v S x ~ u S x I (3. 3)

(it For F = F + + F- the m a p p i n g

AF(x,y) = DF+(x) + DF-(y)

is a d i e i d e d difference operator if DF + is i s o t o n e and DF- is a n t i t o n e ,

see [10].

(ii) If, in a d d i t i o n , F is o f f d i a g o n a l l y antitone

AF(x,y) = d i a g DF+(x) + diag DF-(y)

is a d i v i d e d difference operator, see [ i0] .

(iii) In i n t e r v a l mathematics the m a x i m a l derivative

AF(x,y) = ( m a x akFi(z))
y~zSx

is w i d e l y used being also a divided difference operator, see [7].

 335

4. It is a s s u m e d that for a n v o p e r a t o r

Fa : D = [xl,y I] C R ~ S, a E A (4.1)

an i s o t o n e - a n t i t o n e divided difference operator AF a e x i s t s . Because,

in g e n e r a l , F a a n d ~F a a r e n o t e x p l i c i t l y a v a i l a b l e , b o u n d s of t h e i r s
a r e used. S u p p o s e t h e r e e x i s t m a p p i n g s U , V : D -- S such that

U(z) ~ F a (z) S V(z) for z 6 D, a e A . (4.2)

The bounds U and V are assumed to b e sharp in the following sense,

F a (z) S 0 for a l l a 6 A implies V(z) ~ 0 , (4.3)

F a (z) ~ 0 for all a 6 A implies U(z) ~ 0 , (4.4)

valid for e v e r y z 6 D. F u r t h e r , for AF a let e x i s t an u p p e r bound

B : D X D - L(R,S) characterized by

~Fa(X,V) ~ B(X,V) for y I S V S x ~ x I, a 6 A, (4.5)

B(x,v) S B(u,v) for Yl ~ v ~ y ~ x S u S x I. (4.6)

Now, the iterative process c a n be f o r m u l a t e d .

Method [ i] : D e t e r m i n e Xn+l,Yn+ 1 such that

U ( x n) + B ( x n , Y n ) ( X n + I- x n) = 0 , (4.7)

V ( y n) + B ( X n , Y n ) ( Y n + I- yn ) = 0, n = 1,2, . . . .

If ~F a is t a k e n a c c o r d i n g to (i) or (ii) o n e g e t s a N e w t o n - t y p e
m e t h o d or a J a c o b i - N e w t o n - t y p e method, respectively°

5. M o n o t o n e enclosina theorem: Let x l , Y 1 6 R, Yl S Xl be such that

V ( y I) ~ 0 ~ U ( x 1) . (5.1)

Suppose that the linear operators B(x,y) are invertible and that

B ( x , y ) -I a 0 f o r Yl ~ y ~ x S x I. (5.2)

Then the sequence (x n) a n d (yn) are w e l l - d e f i n e d by (4.7), a n y of

the operators F a, a 6 A, possesses a zero z a 6 [Yl,Xl], and for such
zeros the monotone enclosing

Yl S''-< Y n - i < Yn -< Za < Xn < X n - i <''-< Xl' n = 1,2... (5. 3)

 336

is v a l i d .
A proof s h a l l he s k a t c h e d . The operator T,

T(z) = z - B(xl,Yl)-iFa(Z), z e R

is i s o t o n e since for Yl S y ~ x S x I one gets

-i
T(x) - T(y) = x - y - B ( X l , Y I) {Fa(X) - Fa(Y)},

Fa(X) - Fa(Y) S B(x,y)(x - y) S B ( x l , Y l ) ( X - y)

implying T(y) S T(x). Further, T ( x l) S x I and T ( y I) ~ Yl hold.

Thus, a fixed-point t h e o r e m of K a n t o r o v i c h assures z a = T(z a) for
some vector z a 6 [Yl,Xl] and hence Fa(Z a) = 0 follows.

Next, beinning w i t h Yn S Za <

- Xn' Fa(Z a) = 0, V ( y n) < - 0 <
- U ( x n) one
gets immediately Xn+ 1 ~ Xn, further Xn+ l <
_ z a b e c a u s e of

-1
Xn+ I- z a = x n- z a- B ( X n , Y n) {U(x n) - F a ( Z a ) } ,

U ( x n) - Fa(Z a) S F a ( X n) - Fa(Z a) ~ B ( X n , Y n ) ( X n- Za)~

a n d U ( X n + 1 ) ~ 0 in c o n s e q u e n c e of

F a ( X n) a U ( X n) = B(Xn,Vn)(Xn-Xn+l) ~ Fa(Xn)- Fa(Xn+l), a 6 A.

Analogously Yn ~ Y n + l ~ Za and V ( Y n + l ) -< 0 is d e r i v e d .

6. The a s s u m p t i o n (5.2) c a n be w e a k e n e d as follows: There esists a

mapping G E L(S,R) w i t h ker G = {0} a n d

G ~ 0, G B ( X l , Y I) ~ I ,

see [ I] .

7. In the m o d e l p r o b l e m (2.4),(2.5) let the i n p u t p a r a m e t e r s be

s 0 : i ± ~0' e I : 2.8 ± el, ~2 : 2321.385 ± ~2" ~3 : 11/3 ± £3'

~4 = 1 1 2 1 . 9 1 8 ± S4, ~5 = 2 ± e5, ~6 : 20/3 ± e6, ~7 : 8 6 9 " 9 1 5 7 ± ~ 7

and y = 0.25717, p = arsinh(~(0)/2) and q = arsinh(~(y)/2). In o r d e r

to d e m o n s t r a t e the d i f f e r e n t influence of t h e s e p a r a m e t e r s on the
respective components of the zeros some t y p i c a l examples are g i v e n
computed by the N e w t o n - t y p e method (4.7),(i). The d i m e n s i o n in all
= .. )T
c a s e s is N = 30. F u r t h e r , the n o t a t i o n x* (~l' "'~30 = lim Xn,
= )T
y* (ni~...,~30 = l i m V n is used.
 337

ExamDle I: e0; 0.01, el: Q(i%0) Example 2: El= 0.01, el= 0(i#l)

5 19.943.. 19.963.. 19.949.. 19.957

9 -12.509.. -12.489.. -12.500.. -12.498..
I0 0.453.. 0.473.. 0.4625.. 0.4646..
Ii i0.618.. 10.638 10.6282.. 10.6286..
15 12.197.. 12.217.. 12.20699609 12.20699609
20 13.420.. 13.440.. 13.43032446 13.43032446
25 18,155.. 18.175.. 18.16511803 18.16511803

Example 3: g5 : 0 "01 " c.:0(i#5)

1 Example 4: ~ 6 = 0 "01, E i : 0(i%6)

i ni ~i ~i ~i

5 19.953080.. 19.953055.. 19.95308310 19.95308310

9 -12.505.. -12.493.. -12.49914874 -12~49914874
I0 0.413.. 0.513.. 0.46357969 0.46357969
II 10.602.. 10.654., 10.62844138 10.62844139
15 12.201.. 12.211.. 12.206994.. 12.206997..
20 13.428.. 13.431.. 13.429.. 13,431.,
25 18.165105.. 18.165130.. 18.163.. 18.166..

References

[i] SCHMIDT,J.W., SCHNEIDER,H.. Enclosina methods in ~erturbed non-

linear operator equations. C o m D u t i n a 32, i-ii (1984).
[2] COLLATZ,L., Au~aaben monotoner Art, Arch. Math. (Basel)3, 366-376
(1952).
[3] NICKEL,K., The c o n s t r u c t i o n of a p r i o r i bounds f o r t h e s o l u t i o n of
a two p o i n t boundary v a l u e problem w i t h f i n i t e e l e m e n t s , Computing
23, 247-265 (1979).
[4] SCHR~DER,H., Operator I n e q u a l i t i e s , New York-London: A c a d e m i c Press
1980.
[5] SPREUER,H., A method f o r t h e c o m p u t a t i o n of bounds f o r o r d i n a r y
l i n e a r boundary v a l u e p r o b l e m s , J . Math. A n a l . A p p l . 81, 99-133
(1981).
[61WILDENAUER,P., A new method f o r a u t o m a t i c a l c o m p u t a t i o n of e r r o r
bounds f o r n o n l i n e a r boundary v a l u e p r o b l e m s , c o m p u t i n g 34,
131-154 (1985).
 338

[7] ALEFELD,G., ~uadratisch konuergente Einschlie~ung ~Jon LBsungen

nichtkon~exer Gleichungssysteme, z . Angew. M a t h . Mech. 5 4 ,
335-345 (1974).
[8] ORTEGA,J.M., RHEINBOLDT,W.C., Monotone i t e r a t i o n s for nonlinear
equations with a p p l i c a t i o n to Gauss-Seidel methods, SIAM J .
Numer. Anal. 4, 171-190 (1967).
[9] SCHMIDT,J.W., LEONHARDT,H., Eingrenzung yon L~sungen mit Hilfe
der Regula f a l s i , Computing 6, 318-329 (1970).
[ i0] SCHMIDT,J.W., SCHNEIDER,H., Monoton einschlieBende Verfahren bei
additiv zerlegbaren Gleichungen, z. Angew. Math. Mech, 63, 3-11
(1983).
[ii] SCHNEIDER,N., Monotone EinsehlieBung dureh Verfahren vom Regula
falsi-Typ, Computing 26, 33-44 (1981).
[12] T~RNIG, W., Monoton konvergente I t e r a t i o n s v e r f a h r e n zur L~sung
nichlinearer Differenzen-Randwertprobleme, Beitr~qe Numer. M a t h .
4, 245-257 (1975).
[13] T~RNIG.W.,Monoton e i n s e h l i e ~ e n d konvergente I t a r t i o n s p r o z e s s e
vo~ Gauss-Seidel-Typ, Math. Meth. Appl. Sci. 2, 489-503 (1980).
[14] GRO~MANN,CH.. KRATZSCHMAR.M., Monotone Diskretisierung und adap-
tive G i t t e r g e n e r i e r u n g f~r Zwei-Punkt-Randwertaufgaben, Z. Angew
Math. Mech. 65, T264-266 (1985).
[15] GRO~MANN,CH., KR~TZSCHMAR,M., ROOS,H.-G., Uniformly enclosing
discretization methods of high order for boundary value problems
Math. Meth. Appl. Sci. (submitted 1985).
ERROR ESTIMATES FOR FINITE ELEMENT
METHODS FOR SEMH,INEAR P A R A B O H C
PROBLEMS WITH NONSMOOTH DATA
V. THOMI~E
Department of Mathematics, C~talmers University of Technology
S-41296 G6teborg, Sweden

We shall sur,~ey s o m e recent work on the numerical solution of the

semilinear initial boundary value problem

(I) ut ~ u = f(u) in QxI, l=(O,t*],

u=O on @i]xl,

u(O) = v i n ~,

where i~ a b o u n d e d domain i n IRd w i t h a sufficiently smooth boundary @f/s

and f is a smooth function on R for which we assume for simplicity

that f and f' arc bounded. Such an assumption is n o r m a l l y reasonable

only i£ the solution o£ ( I ) is known a p r i o r i to be b o u n d e d , ~u~_<B,

say, but if this is the case f may be m o d i f i e d if necessary for lul>B

to satisfy our assumption~ without changing the solution of (I).

i i
For spatial discretization of (I), let S h C H O = Ho(i~) be a family

of finite-dlmensional spaces parantetrized by a sall positive

parameter h and let the semidiscrete solution Uh:I-dS h b e defined by

(2) (Uh, t~K)+(VUh,VK) = (f(u h),K) , for K~Sh, t~I,

U h ( O ) = v h ~ Sh,

where (.,.) is the standard inner product in L2(i]).

In order to discuss the error i n (2) ~e assume that S h is such

that the corresponding linear elliptic problem admits an O ( h r) error

 340

estimate in L2=L2(O). More p r e c i s e l y , we assume that the elliptic

projection PI' i.e. the orthogonal projection o n t o Sb w i t h respect to

the D i r i c h l e t inner product (Vv, vw), satJs£ies, for some r_>2 and some

constant M,
I r
(3) IIPI,~-~II _< MhrllVllHr , for ~,~HoNH ,

where ||.il denotes the norm i n L 2. It is then well known t h a t i£ u is

su£ficJently smooth on t h e closed interval I, and i £ the discrete

initial data vh are suitably chosen, then

IlUh(t)-u(t)ll < C(u,H) h r, for tfi.

To g u a r a n t e e t h a t u is smooth enough f o r this result, both

smoothness o f v and c o m p a t i b i i t y conditions betv~een v and t h e

differential equation at @~ f o r t=O a r e n e c e s s a r y . For i n s t a n c e , in

the linear homogeneous case (£-=0 i n (1)) it was shown i n Bramble,

Schatz, ThorPe and IJahlbin [3] that

IlUh(t) - u(t)ll ~ ChrllVllHr £or veD((-~)r/2), tEI,

which thus requires ~J~#[@~=O for j<r/2. Such requirements are not

al~rays satis£ied in p r a c t i c e and it is t h e r e £ o r e o£ i n t e r e s t to

analyze the error for nonsmooth or i n c o m p a t i b l e data. Note that the

solution of (1) will always be smooth for positive time. For the

linear homogeneous e q u a t i o n t h i s may be e x p r e s s e d b y s a y i n g t h a t the

Laplacian generates an a n a l y t i c semigroup E(t)=exp(ht) and that

u(t) =E(t) v sati s£ies

(4) U E(t) vllhjg <_ Ct -(#-a) /2llvll.H ~ where IIvll.

H a :ll(-~) a/2vll.

For t h e linear homogeneous e q u a t i o n t h e nonsmooth d a t a s i t u a t i o n

has been i n v e s ± i g a t e d in Blair [2], Hel£rich [5], Bramble, Schatz,

Thom~e and N a h l b i n [ 3 ] and l a t e r papers ( c f . Thom~e [ ? ] ) . In this

case, it ~r~y be s h o w n using the smoothness property (4) that i£ ~ h is

chosen as PO v , the L2 projection of v onto S h. then

(5) llUh( t) -u( t) ll _< Cha+°t-a/2llvn. Ha, £or O < a < a + a < r .
- _ _

In particular, optimal order convergence is attained for t positive

even i£ v is o n l y in L 2. i~ s i m i l a r result showing O ( h r) con,Jergence

 341

for positive time without initial regularity is known a l s o for the

linear inhomogeneous p r o b l e m , c£. Thom~e [ 7 ] .

In the semilinear situation the following result has been p r o v e d

i n Johnson, Larsson, Thom~e and Wahl bi n [ 6 ] .

Theorem I . Let u be a s o l u t i o n of (I) with llvll_<p. [Issume f u r t h e r that

(3) is satisfied (with r-_'2) and l e t u h he t h e solution o£ (2) with

Vh=PoV. Then t h e r e e x i s t s a constant C=C(p, PD such t h a t

IlUh( t ) - u { t ) ll < Ch2(t-l+[log(h2/t) [) ~ for tEI-.

The above result thus shog~ that for r=2 the error in the

semilinear case is essentially o£ the same order as for the linear

homogeneous equation. For r}2, however, the result o£ Theorem I is

~eaker than the case a=O o£ (5). The reason ~1hy t h e above argument

fails to yield higher order convergence than second is related to the

lack o£ integrahility of the right hand side of (5) for o}2, a=O. In

spite o£ t h i s , it m a y be shown t h a t an a n a l o g u e o£ ( 5 ) holds, in the

sense t h a t the convergence rate in L2 a t positive time is almost two

p o ~ e r s o£ h h i g h e r than the order of regularity of the initial data

(up to the optimal order O(hr)).

It m a y he shown t h a t Theorem 1 i s , in fact, essentially sharp in

t h e sense t h a t an e s t i m a t e of the form

(G) ffuh(tO)-U(to)ll < C(p,M, to) ha , lu(x,t) [ _< B~

cannot hold for any o) 2 and to) O~ regardless of the value o£ r. {Note

that the requirement that u Js hounded is more stringent than

boundedness of Ilvll.) We s h a l l sketch an example to indicate this.

Consider thus the problem

(7) ut - Uxx = £(u) for xEJ=[O,l], tEI,

u(O,t) = u(l,t) = O,

u(x,O) = v(x) ,

where £ ( u ) = u2 f o r ~u I _< B. Let h=l/N, xj=jh, J n = ( X n , Xn+l ) and

consider the semidiscrete analogue using the finite dimensional space

Sh = { ~ e C { J ) ; ~j e Nr-1 for n=O,...,N-1; ~(O)=~(I)=O}.

n
For t h e initial values ~ choose

v(x) = VN(X) = 'P(Nx),

 342

where ~ i s a not identically vanishing function of the form

~r+ 1
~(x) = ~=I ~.sinj nJx,

which is orthogonal t o /~r-i on J , Note t h a t vN i s then orthogonal to

Sh, and a l s o t h a t , i n d e p e n d e n t l y of N,
_r+l
IIvNllLc ° ~ Z:1 I~j I = P ,

where p may be chosen s m a l l e r t h a n B. The e x a c t s o l u t i o n o£ (7) is

then also smaller than B i n modulus on I - = [ O j t * ] with t* suitably

small~ i n d e p e n d e n t l y o£ N. S i n c e Vh=VN,h=PoVN=O~ by t h e c o n s t r u c t i o n

of VN, v~ have uh(t)_--O on I and hence e( t) fuh( t) -u( t) =-u( t) , Using

c o m p a r i s o n theorems and some F o u r i e r s e r i e s arguments~ one may show

for u=u N that

lie(t) II = I l U N ( t ) l l ~ C/N 2 = C h 2 .

Hence an i n e q u a l i t y such as ( 6 ) is not possible for 0}2. lie may t h i n k

of this as an example o£ n o n l i n e a r i n t e r a c t i o n of Fourier modes.

lie s h a l l now b r i e f l y consider the discretization of equations

such as ( I ) and (2) w i t h respect to the time v a r i a b l e . Consider thus a

s e m i l i n e a r problem of the form

(8) du/dt + flu = £(u) for teI,

u(O) = v~

where fl i s a positive definite selfadjoint linear but not n e c e s s a r i l y

bounded o p e r a t o r i n a Hilbert space H, and where f is bounded t o g e t h e r

with its Fr6chet d e r i v a t i v e .

For the approximate solution of (8) ~ introduce a time step k

and let gn~S be the approximation of U(tn), tn=nk , defined by a scheme

o£ the form

(9) Un+ I = EkU n + kF(k~y n) , n=O,l,2 I,..

UO = v.

Here Ek=r(k~l) where r(k) is a rational function which is such that for

some p_> 1,

(10) r(X) = e -A + O(k p+I) as A-K),

and such that

(11) ~r(A) ~ ( I for X ~ O.

 343

Further F(k,~) is such that (9) is consistent with (8). More

precisely, assume for small k, with II,ll the norm in H,

(12) IlF(k,O) - F(k,9) l! ~ C i t e - *ll

and
-1
(13) Iltt (F(k,l~) - f(ft))ll 5" C k ( l l t t t t t l ÷ l) for tt.<'D(tt),

A simple example is provided by the linearized backvrard Euler

method,

- U n + I - U n) I k ÷ tiUn+ 1 = f(U n ) '

which is o£ this form vnth r(A) = I/(I+)O and F(k,~°)=Ekf(~°) and which

satisfies (IO) with p=l, as ~ell as (II), (12) and (13).

We first recall a nonsmooth data error estimate by Baker, Bramble

and Thom~_'e [I] (see also [7]) £or the linear homogeneous equation, £~0

i n (8) and the corresponding discrete schenx-= (9) with F(k,~)=O:

IIUn - u ( t n) ll ~ c k P t n P l I v l l for v~H, tn~I.

This result may be combined with the corresponding result £or

discretization in space of (I) to yield error bounds for totally

discrete schemes o£ order O ( h r + k p) for t positi-ve without smoothness

assumptions on the initial data.

In the semilinear situation we have the £ollowing nonsnvooth data

error estimate by Crouzeia and Thom~e [4],

Theorem 2. Under our present assumptions we have

IIUn - U(tn) ll < C ( p ) k ( t n l l o g ( t n + l / k ) + (log(tn+l/k)) 2} £or Ilvll_<p.

This result may again be combi ned with Theorem 1 concerning

discretization in space to show an e s s e n t i a l l y O(h2÷k) convergence

result £or the complete disoretization o£ (I), without any other

requirements £or the initial data than v~L2(Q).

In the same way as £or the semidiscrete equation, the

nonlinearity limits the order o£ convergence possible in the case o£

non-smooth data. Thus, in particular, one may show by an example that

tot a Runge-Xutta type method o£ order o£ accuracy p) l and i£ s)l then

it is not possible to show

IIOn - Uttn) ll < C ( p ) k s £or Ilvll<_p, tn=t)O.

 344

Re£erenoes.

1. G.A. Baker, J.H. Bramble a n d V. T h o ~ e , Single step Galerkin

approximations £or parabolic problems. Math. Comp. 31 (1977),

818-847.

2. J. Blair, ftpproximate s o l u t i o n o£ elliptic and parabolic boundary

value problems, Thesis, Univ. o£ Cali£ornia, Berkeley, 1970.

3. J.H. Bramble, fl.H. S c h a t z , V. Thom~e a n d L.B. Qahlbin, Some

conwergence estimates £or semilinear GaIerkin type approximations

£or parabolic equations. SII~M J. Numer. ~Inal. 14 ( 1 9 7 7 ) , 218-241.

4. M. C r o u z e i x and V. Thom~e, On t h e d i s c r e t i z a t i o n in t i m e o£

semilinear p a r a b o l i c e q u a t i o n s w i t h non-smooth i n i t i a l data. To

appear.

5. H.P. Hel£rich, Fehlerabsch~tzungen £iir das Gaierkinver£ahren zur

Losung won Evolutionsqleichungen. Manuscr. Math. 13 (1974),

219-235.

6. C. Johnson, S. L a r s s o n , •. Tbomee and L . B . Wahlbin, Error

e s t i m a t e s £ o r s e m i d i s c r e t e a p p r o x i m a t i o n s o£ s e m i l i n e a r p a r a b o l i c

equations ~th non-smooth d a t a . To a p p e a r .

7. V. Thom~e, Galerkin Finite Element Methods £or Parabolic

Problems. Lecture Notes in Mathematics no. 1054, Springer-Uerlag,

1984.
SINGULARITIES IN TWO- AND
THREE-DIMENSIONAL ELLIPTIC
PROBLEMS AND FINITE ELEMENT
METHODS FOR THEIR TREATMENT
J. R. WHITEMAN
Brunel University
Uxb~idg~ England
I. INTRODUCTION
The effective use of finite element methods for treating elliptic
boundary value problems involving singularities is well recognised.
As a result considerable e f f o r t has b e e n e x p e n d e d by m a t h e m a t i c i a n s
and engineers in d e v e l o p i n g special finite element techniques which
can p r o d u c e accurate approximations to the solutions of problems
involving singularities.
The w o r k of m a t h e m a t i c i a n s has b e e n m a i n l y in the c o n t e x t of two-
dimensional Poisson problems. It has e x p l o i t e d and relied on k n o w n
theoretical results concerning the r e g u l a r i t y of solutions of weak
forms of problems of this type, and has p r o d u c e d significant finite
element error e s t i m a t e s for this limited class of problems. Comparable
progress has not been made in the finite element treatment of three-
dimensional problems involving singularities, mainly on a c c o u n t of the
lack of t h e o r e t i c a l results for the t h r e e - d i m e n s i o n a l case. This is
particularly relevant to the case of t h r e e - d i m e n s i o n a l re-entrant
vertices.
In this paper we p r e s e n t a survey of t h e finite element treatment
of singularities. This is first done in the c o n t e x t of a model two-
dimensional Poisson problem and estimates for various norms of the er-
ror are given. Some finite element techniques for s i n g u l a r i t i e s are
then described, taking into account their effects on c o n v e r g e n c e rates
and accuracy. In problems with singularities the a p p r o x i m a t i o n of sec-
ondary quantities by retrieval from a p p r o x i m a t i o n s to the solutions
(primary quantities) is of great importance,and so this is also treated
here. F i n a l l y Poisson problems with singularities in three d i m e n s i o n s
are p r e s e n t e d and the s t a t e - o f - t h e - a r t for this case is c o n t r a s t e d with
that for two dimensions.

2. POISSON PROBLEMS INVOLVING SINGULARITIES

2.1. Two D i m e n s i o n a l Poisson Problems.
Let ~ C R 2 be a simply connected polygonal domain with boundary
8~. We consider first the much studied model problem in which the
scalar function u(x) satisfies
- k[u(x)] : f(E), ~ E ~, (2.1)
u(x) : 0, x E 3~,
 346

where f E L2(~) . A w e a k f o r m of (2.1) is d e f i n e d in the u s u a l S o b o l e v

s p a c e ~I(~), a n d for this u 6 ~i(~) satisfies
a(u,v) = F(v), ~ v E ~i(~), .2)
where
a(u,v) E / VuVv d ~ , u,v e ~l(~), (2.3)
and
F(v) m f fv d~, v e ~i(~). (2.4)

Problem (2.1) is t r e a t e d by c o n s i d e r i n g the w e a k form (2.2), w h e r e the

bilinear f o r m has the i m p o r t a n t properties that it is c o n t i n u o u s ,
symmetric and elliptic on ~!(~), see C i a r l e t [ i] .
For the f i n i t e element solution of (2.1) the r e g i o n ~ is
partitioned quasi uniformly into t r i a n g u l a r elements ~e in the u s u a l
manner a n d the G a l e r k i n method is a p p l i e d to (2.2). Conforming trial
and test f u n c t i o n s are e m p l o y e d ahd the solution u 6 ~i(~) is
approximated by u h 6 S h, w h e r e Sh c ~i(~) is a finite d i m e n s i o n a l
s p a c e of p i e c w i s e polynomial functions of d e g r e e p, (p a I), and u h
satisfies
a ( u h , v h) = F ( v h) ~ v h e S h. (2.5)
The w e l l k n o w n b e s t a p p r o x i m a t i o n property of the G a l e r k i n solution
g i v e s the i n e q u a l i t y

Hu - UhIll, n S IIu - Whll i,~ ~ w h E S h, (2.6)

where ]IVl]l,~ is the e n e r g y norm ]IVvi]L (~) Since (2.6) h o l d s for all
w h 6 S h, we m a y take the i n t e r p o l a n t 2 Uh sh to u for w h in (2.6)
and, using approximation theory, it f o l l o w s that
Uu - Uh~l, ~ ~ C h ~ U J k + l , ~ , (2.7)

where p = min(p,k), whilst C is a c o n s t a n t . Throughout the p a p e r all

constants in the e s t i m a t e s are d e n o t e d by C.
The a c t u a l v a l u e of u is thus d e p e n d e n t both o n the c h o i c e of p
and on the r e g u l a r i t y of the s o l u t i o n u of (2.2). U n d e r the c o n d i t i o n
that f 6 L2(~ ) the r e g u l a r i t y of u is d e t e r m i n e d by the s h a p e of 8~.
If ~ is a c o n v e x polygon, then u e ~ i ( ~ ) N H2(~), so t h a t k = i in
(2.7) a n d

11u - UhIll, ~ S C hluJ2, ~ (2.8)

In this case, see S c h a t z [2],

IIu - Uh[IL2(n ) S C h 2 1 u 1 2 , ~ (2.9)

sO that there is an 0(h) c o n v e r g e n c e gain through changing f r o m the
1 - n o r m to the L 2 - n o r m . T h e a b o v e two e s t i m a t e s are optimal in that
they are the b e s t that c a n be o b t a i n e d by a p p r o x i m a t i n g from S h a
 347

function with the r e g u l a r i t y of u. It has also b e e n shown, see

Nitsche [3] .and C i a r l e t [2], that for this case the L ~ - n o r m of the
e r r o r has 0(h 2) c o n v e r g e n c e .
As has b e e n s t a t e d in S e c t i o n i, p r o b l e m s with boundaries having
re-entrant corners, and thus c o n t a i n i n g boundary singularities, are
of m a i n interest here. We thus c o n s i d e r again problem (2.1), b u t now
in the s i t u a t i o n where ~ is a n o n - c o n v e x polygonal domain with
interior angles ej, 1 N j N M, w h e r e
0 < ~i S ~2 ~ . . . ~ ~ < ~m ~ . . . ~ ~M ~ 2~.
In this case the s o l u t i o n u of (2.2) is s u c h that u E ~ i ( ~ ) _ H2(e),
a n d it has b e e n s h o w n b y G r i s v a r d [4] that over ~ u c a n be w r i t t e n as
M
U = E a.x.(r.)u (r.,e) + w, (2.10)
j=m 3 3 3 J 3 J
where (rj,Sj) are l o c a l p o l a r c o o r d i n a t e s centred on the jth c o r n e r of
8n, the Xj are s m o o t h c u t - o f f functions for the corners, w E H2(~) and
• ~/aj ~8j
uj(rj,ej) = rj sin eJ

The regularity of u is c l e a r l y determined by the t e r m in the s~immation

in (2.10) associated with the M th corner. In f a c t u E H l + ~ / e M - S ( ~ ) for
every e > 0, see a l s o S c h a t z and Wahlbin [5].
Since a M > ~ and u E H I + ~ / ~ M - ~ ( ~ ) , it f o l l o w s from (2.7) that
(~I~M-~)
Iiu - UhlIl, ~ ~ C h IuII+~/~M_~ , (2.11)

and, see S c h a t z [2], that

2(~/aM-E)
flu - uhliL2(n ) S C h lUll+~/~M_C (2.12)

Whereas the c o n v e r g e n c e g a i n in the c h a n g i n g f r o m the 1 - n o r m to the

L2-norm is 0(h) for the c a s e w h e r e ~ is a c o n v e x p o l y g o n , (2.8),(2.9),
the gain is less for the r e - e n t r a n t case.
Estimates of the type (2.11) and (2.12), being global, reflect
the w o r s t b e h a v i o u r of the s o l u t i o n over ~. The s i t u a t i o n may not be so
bad locally, in p a r t i c u l a r a w a y f r o m the c o r n e r s where from (2.10)
u 6 H 2. T h u s we n o w c o n s i d e r L~-estimates. Suppose that at the jth
v e r t e x z. of @~ the i n t e r s e c t i o n of s w i t h a d i s c c e n t r e d at z. and
3 3
c o n t a i n i n g no o t h e r c o r n e r is ~j and t h a t n 0 ~ ~ \ ( j U I ~ j ) . It has b e e n
s h o w n by S c h a t z and Wahlbin [5] that

min(p+l,2~/ M)-e
llu - UhllL~(90) < C h (2,13)
 348

~/~M-~
~u - UhIIL~(eM) ~ C h (2.14)

Similar estimates were discussed by Oden and O ' L e a r y [6]. It should

be emphasised that all the above estimates are based on q u a s i - u n i f o r m
meshes and piecewise pth order polynomials.
Specific examples of estimates (2.11) - (2.14) are those where
the region ~ contains a slit, ~ = 2~, for which
M

u e H3/2-~(~), flu - UhUl,~ =0(hi/2-~)' flu - UhllL2(~) = 0(h l-c)

Jlu - UhllL (~0) : 0(hl-~), flu - UhlIL~(~M) : 0(h I/2-~)

and where the region is L-shaped, ~M = 3~/2, for which

u E H5/3-E(~), ItU - Uhl11,a = 0(h 2/3-s), flu - UhllL2 (~) £ 0(h 4/3-e)

flu - UhllL~(~ ) = 0(h4/3-~), flu - UhIIL~(~M) = O(h 2/3-~)

2.2 Techniques for Singularities.

The error estimates of Section 2.1 indicate the d e t e r i o r a t i o n fran
the optimal state caused by the presence of the singularity. On
account of the practical importance of singularities, much effort has
been expended in producing special finite element techniques fro
treating singularities, and a c o n s i d e r a b l e literature now exists. The
approaches fall mainly into three classes; a u g m e n t a t i o n of the trial
ans test spaces with functions having the form of the dominant part of
the singularity, use of singular elements, use of local mesh r e f i n e m e n ~
These techniques and their effects are now reviewed briefly.
Since for problems of type (2.2) with re-entrant corners the form
of the singularity is known, use of this can be made by augmenting the
space S h with functions having the form of the singularity. The
solution u of (2.3) is in this case a p p r o x i m a t e d by u h E AugS h. The
techniques, p g o p o s e d by Fix [7] and used by Barnhill and W h i t e m a n [8]
and Stephan and W h i t e m a n [9], enables estimates as for problems with
smooth solutions to be obtained. It does, however, have the disadvan-
tage of producing a system of linear equations in which the coefficient
matrix has a more complicated structure than normal.
The technique of employing singular elements involves in elements
near the singularity the use of local functions which approximate
realistically the singular behaviour. Elements of this type have been
 349

proposed by Akin [ I0] , B l a c k b u r n [ ii] and S t e r n and Becker [ 12], and

their use can lead to s i g n i f i c a n t increase in accuracy of u h. O ' L e a r y
[13], specifically for the S t e r n - B e c k e r element, has proved that use
of the e l e m e n t produces no improvement in the rate of c o n v e r g e n c e in
the error estimate. The increase in accuracy must therefore be pro-
d u c e d by r e d u c t i o n in size of the c o n s t a n t in the estimate.
Local mesh r e f i n e m e n t near a s i n g u l a r i t y was o r i g i n a l l y performed
on an ad-hoc basis without theoretical backing. In recent years error
analysis has been p r o d u c e d which indicates the g r a d i n g which a mesh
should have near a corner in order that the effect of a s i n g u l a r i t y
m a y be nullified. Examples of such local m e s h r e f i n e m e n t are g i v e n by
Schatz and W a h l b i n [5] and B a b u s k a and O s b o r n [14]. Another approach
is to use adaptiv£ mesh refinement involving a-posteriori error
estimation.
With adaptive mesh refinement the region ~ is p a r t i t i o n e d
initially and the local error in each e l e m e n t is estimated. If, for a
particular element, this is g r e a t e r than a p r e s c r i b e d tolerance, the
element is subdivided thus c a u s i n g the local refinement, see B a b u s k a
and R h e i n b o l d t [15], [16]. Hierarchical finite elements have recently
been incorporated into the technique, Craig, Zu und Zienkiewicz [17],
as have m u l t i g r i d methods Bank and S h e r m a n [18] and Rivara [19].

2.3 Retrieved Quantities.

As has b e e n stated in S e c t i o n I, for problems involving boundary
singularities the a p p r o x i m a t i o n of s e c o n d a r y (retrieved) quantities is
most important. Specifically the c o e f f i c i e n t s
a~ in (2.10) of singular
3
terms are of p r a c t i c a l significance, so that ways m u s t be found of
approximating these accurately. Apart from the o b v i o u s approach of
using c o l l o c a t i o n or least squares methods to fit terms to c a l c u l a t e d
results, it is often possible to e x p l o i t the m a t h e m a t i c s of the o r i g i n ~
problem. An i m p o r t a n t case is that of a p r o b l e m containing a slit,
a M = 2~, a n d here use can be made of the "J-integral" concept to pro-
duce an integral expression for the aM, see D e s t u y n d e r et ai.,[20] .
This integral can be a p p r o x i m a t e d using the c a l c u l a t e d solution u h. For
piecewise linear test and trial functions on a mesh with local refine-
ment, 0(h) estimates are given in [20] for the a b s o l u t e value in the
error in the a p p r o x i m a t i o n to the s i n g u l a r i t y coefficient aM .
Fro problem (2.2), when a singularity is present, the i n t e g r a n d of
the "J-integral" involves derivatives of the s o l u t i o n u. Thus the
accuracy of the a p p r o x i m a t i o n to the integral, and hence to the singu-
larity coefficient, depends on the errors in the g r a d i e n t S of u h.
 350

A possibility exists here of e x p l o i t i n g superconvergenee properties

in the e s t i m a t i o n of errors in gradients of Uh, provided local esti-
mates can be obtained. To date the error estimates have d e p e n d e d on
the global r e g u l a r i t y of the solution u, see Levine [21].

2.4 Three Dimension Poisson Problems.

We consider a g a i n problems of the type (2.1), except that now
C R 3 is a p o l y h e d r a l domain. The weak forms and the finite element
method for the t h r e e - d i m e n s i o n a l case can be described similarly,
again with ~ C R 3. S i n g u l a r i t i e s can in this case occur on account of
r e - e n t r a n t edges and vertices. The d e c o m p o s i t i o n of the three dimen-
sional weak solution c o r r e s p o n d i n g to (2.10) has been shown, e.g. by
Stephan [ 22] to have the form
M N
u : ~ a + ~ fk Vk + w (2.15)
j:l jXjuj k:l ~k
(vertices) (edges)

w h e r e w 6 H2(~), ×j(rj) and ~k(~k ) are cut-off functions r e s p e c t i v e l y

for the vertices and edges, w h i l s t the uj and v k are functions
a s s o c i a t e d also r e s p e c t i v e l y with vertices and edges. For an edge the
v k have the two dimensional form for any plane o r t h o g o n a l to the edge
a s s o c i a t e d with the appropriate two dimensional problem, w h i l s t the
b k are functions of z k.
The singular function u for each v e r t e x is found by solving a
3
Laplace-Beltrami eigenvalue p r o b l e m on that part of the surface of the
unit ball c e n t r e d on the v e r t e x cut off by the faces of the vertex.
W h e n the v e r t e x is such that the e i g e n v a l u e p r o b l e m is separable (has
a single coordinisation), there are special cases w h e n the p r o b l e m can
be solved exactly, see W a l d e n and K e l l o g g [23]. When this is not so,
for e x a m p l e for a v e r t e x made up from three m u t u a l l y o r t h o g o n a l
planes, Beagles and W h i t e m a n [24], a numerical a p p r o x i m a t i o n to the
eigenvalue must be o b t a i n e d with the result that the singular function
will not be known exactly.
C l e a r l y this lack of knowledge of the exact singular functions is
very important from the finite e l e m e n t point of view, and in
p a r t i c u l a r means that the error analysis of S e c t i o n 2.1 cannot in
general be transferred d i r e c t l y to the t h r e e - d i m e n s i o n a l singular
case. All the s i n g u l a r i t y m e t h o d s d e s c r i b e d in S e c t i o n 2.1 are affect-
ed, although all are used in the t h r e e - d i m e n s i o n a l context. The
augmentation technique is obviously adversely affected, although
B e a g l e s and W h i t e m a n [25] have d e v i s e d the technique of non-exaet
a~gmentation, w h e r e b y the trial and test function sDaces in the
 351

Galerkin procedure are augmented with the non- exact singular

functions. As far as we are aware no method of the "J-integral"
type exists for three-dimensional Poisson problems.
The above indicates that the state-of-the-art fro treating
three-dimensional singularities with finite element methods is f a r
less advanced than that for the two-dimensional case. This arises more
from limitations in t h e theory of three-dimensional Poisson problem
than from the finite element methods themselves.

R e f e r e n c e s
| I] Ciarlet P.G., The f i ~ e Eleme~ Method for Elliptic Problems. Nort-Holland,
Amsterdam, 1979.
[ 2] Schatz A., An introduction to the analysis of the ~r~or i n the f i n i t e element
method for second order 611iptic boundary value problems, pp. 94-139 of P.R.Tur-
her (ed.) Numerical Analysis Lancaster 1984. Lecture Notes in Mathematics 129,
Springer-Verlag, Berlin, 1985.
[3] Nitsche J.A., [~-convergence of f i ~ e ~ e n t approxima~o~, m ~ e m ~ c a l
~ p e c ~ of finite eleme~/t method. Lecture Notes in Mathematics 606, Springer-
Verlag, Berlin, 1977.
[4] Grisvard P., BehaviouJ% of the solutio~ of an ~ p t i c bound~J%y ual~e problem
in a polygonal or polyhedral domain, pp. 207-274 of B. Hubbard (ed.), Numerical
Solution of Partial Differential Equations III, SYNSPADE 1975. Academic Press,
New York, 1976.
[5] Schaltz A. and Wahlbin L., Maximum no~u~ @~tim~@~ in the f i ~ e element method
on plane po~gonal d o ~ . Parts I and II. Math. Comp. 32, 73-109, 1978, and
Math. Comp. 33, 465-492, 1979.
[6] Oden J.T. and O'Leary J., Some r e m ~ on f i ~ e e l ~ e ~ approximatio~ of
crack problems and an a ~ s i s of hybrid methods, j. struct. Mech. 64, 415-436,
1978.
[7] Fix S., Higher order R~leigh Ritz approxim~onS, j. Math. Mech. 18, 645-657,
1969.
[ 8] Barnhill R.E. and Whiteman J.R., E~or a ~ y s ~ of Gal~r~in metho~ for
Dirichlet problems containing boundary s i n g u l a r i t i ~ . J. Inst. Math. Applies.15,
121-125, 1975.
[9] Stephan E. and Whiteman J.R., S i n g ~ of the Laplacia~ ~ corne~ and
edges of three dimensional domains and their treatment with f i n i t e element
method6. Technical Report BICOM 81/i, Institute of Computational Mathematics,
Brunel Uninersity, 1981.
[ I0] Akin J.E,, Generation of element~ with singulariti~. Int. J. Numer. Method.
Eng. IO, 1249-1259, 1976.
[Ii] Blackburn W.S., C ~ c u ~ o n of stress i n t ~ i ~ y facito~ a/t c2u~ck tips u~ing
special f i ~ e elements, pp. 327-336 of J.R. Whiteman (ed.), The Mathematics of
Finite Elements and Applications. Academic Press, London, 1973.
[ 12] Stern M. and Becker E., A confor~ng crack tip element with q~dratic variation
in the sing~£ar field~. Int. J. Nurser. Meth. Eng. 12, 279-288, 1978.
I 13] O'Leery J.R., An @/t%or ~ s ~ for s i ~ r finite el@ghoul,s. TICOM Report 81-4,
Texas Institute of Computational Mechanics, University of Texas at Austin, 1981. ~
[ ~4] Babuska I. and Osborn J., [ i ~ Z eleme~ methods for the s o l , o n of p ~ b l ~
rough i ~ p ~ ~t~. pp. 1-18 of P.Grisvard, W.Wendland and J.R.Whiteman (eds.),~
Singu~e~ and C o ~ t n ~ i v e Method~ for Thei% Tre~2me~t. Lecture Notes in
Mathematics 1121, Springer Verlag, Berlin, 1985.
[ 15] Babuska I. and Rheinboldt W.C., E~or ~ t i ~ for a~ptive f i ~ e element
eomp~t~tions. SIAM j. Num. Anal. 15, 736-754, 1978.
[ 16] Babuska I. and Rheinboldt W.C., Reli~le ~ o r ~ti~on and mesh ~apt~tion
for the finite eleme~ ~ethod. pp. 67-108 of J.T. Oden (ed.), Computational
Methods in nonlinear Mechanics. North-Holland, Amsterdam, 1979.
 352

[ 17] Craig A.W., Zhu J.Z. and Zienkiewicz O~C., A-posteri%oai error estimation,
adaptive mesh r e f i n e m e ~ and m~itigrid m~hods ~ i n g hierarchical f i ~ t e
eleme~%~ bas~s, pp. 587-594 of J~R. Whiteman (ed.), The Mathematics of Finite
Elements and Applications V., MAFELAP 1904. Academic Press, London, 1985.
[ 18] Bank R.E. and Sherman A.H., The u~e of adaptive g~d refineme~ for badly
behaved elliptic par~i~ differe~i~ equation, pp. 18-24 of computers in
simulation xxII. North Holland, Amsterdam, 1980.
[ 19] Rivara M.C., ~yna~6 i m p l e m e ~ o n of the h-version of ~ e fin~e element
method, pp~ 595-602 of J.R. Whiteman (ed.), The Mathem. of Finite Elements and
Applic. V., MAFELAP 1984, Academic Press, London, 1985.
[ 20] Destuynder P, Djaoua M~ and Lescure S., On ~umeric~ metho~ for fracture
mechanics, pp. 69-84 of P~ Grisvard, W.L. Wendland and J.R. Whiteman (ads.),
Singularities and constructive methods for their treatment. Lecture Notes in
Mathematics 1121, Springer Verlag, Berlin, 1985.
[21] Levine N., Supe~eonvergenee recovery of the gradient from piecewise linear
fini~ element approximations. Technical Report 6/83, Dept. of Mat/~ematics,
University of Reading, 1983.
[ 22] Stephan E . , A modified Fix method for the mixed boundary value problem of the
L a p l a ~ n in a po~hedral domain. Preprint Nr. 538. Fachbereich Mathematik,
T.H. Darmstadt, 1980.
[23] Walden H. and Kellogg R.B.,Numerical determi~a~on of the fun~ental ~igen-
value for the Laplace op~ator on a sphercial domain. J . E n g i n e e r i n g
Mathematics 11, 299-318, 1977.
[ 24] Beagles A.E. and Whiteman J.R., Treatmen~ of a re-emtr~t vertex in a three
dimen&ional Poi~son problem, pp. 19-27 of P.Grisvard, W.H. Wendland and
JoR. Whiteman (ads.), Singularities and Constructive Methods for Their
Treatment. Lecture Notes in Mathematics 1121, Springer Verlag, Berlin, 19S5.
[ 25] Beagles A%E. and Whiteman J.R., Finite el@Inent t~ea~ent of b o u ~ u ~
si~ularities by a r g u m e ~ o n with non-exact singular fun~io~. Technical
Report BICOM 85/I, Institute of Computational Mathematics, Brunel University,
1985.
SOME NEW CONVERGENCE RESULTS
IN FINITE EI MENT THEORIES
FOR ELLIPTIC PROBLEMS
A. 2ENiSEK
Computing Center of the Technical University
Obrdnc~, mi~u 21, Brno, Czechoslovakia

THE LINEAR PROBLEM

We c o n s i d e r the following variational problem: Find u E W such

that
a(u,v) : L(V) ~ L~(v) + LF(v) Vv E V (i)
where
a(v,w) : ffkij v,i w,j dx, (2)

Ln(v) : f/vf dx, LF(V) : f vq ds, (3)

F2

V = {v @ HI(2): v : 0 on FI; m e s l F 1 > O], W : z + V ; (4)

is a b o u n d e d domain in E 2 w i t h a boundarv r : F 1 U F 2 (F I N F 2 :
= ~); z @ HI(~), z = u on FI, ~ E L2(FI) is a g i v e n function. In (2)
and in w h a t follows the summation convention over repeated subscripts
is a d o p t e d a n d v, = 8 v / ~ x . The f u n c t i o n s k = k (x) are b o u n d e d
i~ i 13 13
and measurable in ~ D ~ a n d s a t i s f y

k i j ( x ) ~ i ~ j ~ C~i~ i Vx e ~ D ~ V~i,~ i E El, (5)

where C > 0, a n d f E L2(~), q E L2(F2). Assumption (5) and Friedrichs'

inequality imply that the form a(v,w) is V - e l l i p t i c . Thus, according
to the Lax-Mil~ram lemma, problem (i) has just one solution.
Problem (i) is a p p r o x i m a t e d by the problem: Find u h E W h such
that

ah(Uh[V) : Lh(V) ~ L~(v) + L~(v) Vv E V h (8)

where V h is a f i n i t e element approximation of V a n d W h : z h + Vh, zh

being a finite element approximation of z. T h e forms ah(v,w), Lh(V)
approximate the forms a(v,w), L(v) in the following way: The sets Q
and F2 a p p e a r i n g in (2), (3) a r e substi, t u t e d by ~h a n d Fh2 a n d the
 354

obtained forms ah(v,w), ~h(V) are then computed numerically.

In t h e C i a r l e t ' s and Raviart's theory and its m o d i f i c a t i o n s (see
[ i],[2],[6]) the solution u of (I) is a s s u m e d to be s u f f i c i e n t l y
smooth, u 6 Hn+l(2) (n h i), and the m a x i m u m rate of c o n v e r g e n c e is
proved:

1[u - Uhl11,~h < Ch n, (7)

where u is the Calderon's extension of u into H n + I ( E 2 ).

The p r o b l e m of c o n v e r g e n c e of u h to ] (when u e HI(2) only) was
solved recently in [8]. In this section we p r e s e n t outlines of con-
siderations of [8].

Theorem i. L e t the b o u n d a r y r of the d o m a i n ~ be p i e c e w i s e of

class C n+l. Let rh a p p r o x i m a t e F piecewise by arcs of d e g r e e n. Let
kij,f ~ w~n~) a n d let the q u a d r a t u r e formula used on the standard
triangle T O for calculation of a h ( v , w ) a n d L~(v) be of d e g r e e of p r e -
cision 2n - 2. L e t q E cn(u), where U is a d o m a i n containing r 2, a n d
let the q u a d r a t u r e formula u s e d o n [ 0,i] for c a l c u l a t i o n of L~(v)
be of d e g r e e of p r e c i s i o n 2n - i. Let u be so s m o o t h that there ex-
ists a function z 6 H2(2) such that z = ~ on F I. T h e n

iiu - Uh]ll, ~ h 0 if h ~ 0. (8)

Proof. The assumptions concerninm F a n d ah(v,w) imply, according

to [2] and [7], that the forms ah(v,w) are u n i f o r m l y Vh-elliptie.
Thus we h a v e similarly as in [ i] , [2]:

ILh(W) - ah(u,w) I
ILl - UhLll,~h < C sup +
<weV h 1,9 h

+ inf II~ - vll I ah + s u m . (9)

v~W h ' we% h I~w~II

For simplicity we p r o v e Theorem 1 only in the case F1 = F,

: 0 and n = i, i.e. ~h has a polygonal boundary a n d we use line-
ar t r i a n g u l a r finite elements. As F is p i e c e w i s e of class C 2 each
boundary triangle satisfies (for sufficiently small h) one of the
following possibilities:

a) T ~ Tid , b) T i d C T, (i0)

where Tid is the ideal boundary triangle (see [9]) whose approxima-
tion is T.
In o r d e r to e s t i m a t e the first t e r m on the right-hand side of (9)
let us d e f i n e a function ~ C V associated w i t h w C V h in the following
 355

way: Q(Pi ) : w(P.) at the v e r t i c e s P of the t r i a n g l e s of the trian-

l 1
gulation T h of the d o m a i n Qh ~ the f u n c t i o n w is c o n t i n u o u s on ~,
linear on each interior triangle of Th, equal to zero on Tid T
and linear on T in the case (10a) and, finally, equal to Zl~mal's i-
deal "linear" interpolate of w on Tid in the case (10b) (see [9]).
Thus we can write, according to (I):

Lh(W) - ah(u,w) : Lh(W) - L(~) + a(u,w) - ah(u,w). ll)

We have

L(w) : fff~dx : fffw dx + ~ ~ff (Q - w)f dx - ff wf d x ~

Qh b QTid T-Tid J
where the sum is taken over all b o u n d a r y triangles (lOb). Expresslng
similarly a(u,Q) - ah(u,w) we o b t a i n from (ii):

ILh<W) - ah(u,w), ~ JL~(w) - L~(w)l +

b T-Tid

According to [2, Theorem 4 . 1 . 5 ] ,

ILh(W) - Lh(W)l ~ ChllWlJl,ah. (13)

Denoting the first sum in (12) b r i e f l y by S 1 we have, according to

the Cauchy inequality and the b o u n d e d n e s s of k :
z3

The d e f i n i t i o n of Q and the proof of [9, T h e o r e m 2] imply

IIw - ~l]l,Ti d _< Chllwll2,Ti d S ChilWlll, T

because w is linear on T and (10b) holds. Thus

ISll S ChllWIIl,~h. (lq)

As m e s ( T - Tid) = 0(h 3) we have

~llwll2 S Ch3~ max w 2 .

b 0'T-Tid b

Further, similarly as [7,(39)] we can prove

3
Jtwi Z 2
TeT h i=l
 356

w h e r e PTi (i = 1,2,3) are the v e r t i c e s of T. U s i n g t h e s e r e s u l t s and

the fact t h a t w,i are c o n s t a n t s on e a c h T E T h w e can e a s i l y find for
the s e c o n d s u m S 2 o n the r i g h t - h a n d side of (12>:
r
IS21 . ~wll-I
1,~ h _< C ~ uIIwll
~ 2 T _ Tld IlwlI-2
0, 0~ h

+ b~lWl~'T-Tid lwl-21'~h~ 1/2 ~ c h l / 2 (25

According to (12) - (15), the first t e r m on the r i g h t - h a n d side of

(9) is o(hl/2).
As to the s e c o n d t e r m on the r i g h t - h a n d side of (9) we can find
a set {Vh], w h e r e v h E Wh, such that

Jlu - Vhlll,~h 0 if h ~ 0. (16

The following p r o o f of (16) h o l d s also in the c a s e m e s l F 1 < meslF:

The set G = C~([) A V is d e n s e in V (see [3]). Thus for e v e r y ~ > 0
we can find v C E G s u c h that IIu - v Ill,~ < e. Let v E and v~ be the
Calderon's extensions of v e into HI(E 2) and H 2 ( E 2 ), r e s p e c t i v e l y . We
have

IIu - l h V lll,~h~ II~ - ~ lll,~h +

+ lJ~ - ~ I I l , ~ h _ ~ + 11~ - IhVel;l,~h (17

where IhV s is the i n t e r p o l a t e of v< in Wh, i.e. the p i e c e w i s e linear

function which has the same f u n c t i o n values at the v e r t i c e s of T 6 T
h
as the f u n c t i o n v . The p r o p e r t i e s of the C a l d e r o n ' s extensions, the
absolute continuity of the L e b e s g u e integral a n d the f i n i t e element
interpolation theorem imply that for all h ~ ho(e) the r i g h t - h a n d
side of (17) is b o u n d e d by Ke, w h e r e K does not d e p e n d o n e. As
IhV e e V h = W h r e l a t i o n (17) i m p l i e s (16).
The set {Vh} a p p e a r i n g in (16) is b o u n d e d . Thus, according to [2,
Theorem 4.1.4], the t h i r d t e r m o n the r i g h t - h a n d side of (9) is 0(h).
This finishes the p r o o f in our s i m p l e case. The g e n e r a l case F 1 C F,
n a 1 is c o n s i d e r e d in [8] .

SOME NONLINEAR PROBLEMS

Let the f o r m a(u,v) appearing in (i) be now n o n l i n e a r in u, li-

near in v, s t r o n g l y monotone and Lipschitz continuous a n d let
a(0,v) = 0 for all v E HI(s). In a d d i t i o n , let the forms a h ( v , w ) be
uniformly strongly monotone and uniformly Lipschitz continuous in X h
(the f i n i t e e l e m e n t approximations of H I ( ~ ) ) , i.e. let
ah(v,v - w) - a h ( w , v - w) > Clv - wl 2
- l,~h '
 357

lah(v,z) - ah(w,z)[ S KIiv - wiJl,~hilZl11,~h

Vv,w,z E X h C Hl(~h) Vh (0,h 0 )

where the positive constants C, K do not depend on v,w,z a n d h. Fi-

nally, let the forms ah(v,w) be u n i f o r m l y Limschitz continuous. Un-
der these assumptions the abstract error estimate has again the form
(9) (see [5]).
A typical f o r m a(u,v) satisfying all assumptions presented in
this section is g i v e n bv r e l a t i o n (2) w i t h
2
kij = b ( x , ( V v ) )6ij (18)
where 6ij is the Kronecker delta and where the function b(x,n) has
the following properties (see [4]):
a) T h e functions b(x,n), 8 b ( x , n ) / O x i, 8b(x,q)/Sn are con-
tinuous in ~ x [ 0,~) , w h e r e ~ D ~.
b) There exist constants c I > o, c 2 > 0 such that

c I ~ b ~ c2, 18b/~xil ~ c2, 0 ~ ~b/8~ ~ c2 in ~ × [ 0 ~ ) ,

l~[(~b/~n)(x,~2), ~2(Sb/~n)(x,~2) ~ c2 Vx e ~, ¥~ E E 1 .

The functions (18) w i t h properties a), b) appear in m a n y physical

and technical applications.
Now we generalize the result introduced in T h e o r e m i:

Theorem 2. Let the f o r m a(u,v) appearing in (i) be d e f i n e d by (2)

and (18) a n d let the function b(x,~) have properties a), b). Let
the a s s u m p t i o n s of T h e o r e m 1 be satisfied with n : I. T h e n the solu-
tions u and u h of p r o b l e m s (i) a n d (6) e x i s t a n d are unique and re-
lation (8) holds. In a d d i t i o n , if u E H2(~) then the rate of c o n v e r -
gence is g i v e n by (7), where n : I.

Proof. As n : 1 we c o n s i d e r only linear triangular elements. We

again restrict ourselves to the case F 1 = F, u = 0. T h e e x i s t e n c e
and uniqueness of u a n d u h is p r o v e d in [4]. The first property b)
allows us to r e p e a t (ii) - (15). Thus the first t e r m on the right-
-hand side of (9) is o(hl/2). Also relation (16) remains unchanged,
only the analysis of the third t e r m on the right-hand side of (9) is
different: As Vv : c o n s t , on T @ T h for all v E V h = W h we c a n w r i t e

lah(Vh,W) - ah(Vh,W) l ~ ~ ] m e s ( T ) b ( P T , g h i T) -
TeT h

- ff b ( X , g h ) d X l . l ( V V h l T . V W i T ) l
T
 358

where v h are the functions from (18) and gh = (VVh)2. We used one-
-point integration formula with the centre of gravity PT of T ~ T h.
Using the p r o p e r t i e s of the function b(x,~) we see, according to [2,
Theorem 4.1.5], that the absolute value of the difference on the
right-hand side is b o u n d e d b y Ch mes(T). Thus the r i g h t - h a n d side of
the last inequality is b o u n d e d by chllwll and r e l a t i o n (8) is valid.
i,~ h
The error estimate in the case u ~ H2(~) is d e r i v e d in [5] where
also more general forms a(v,w) are considered.

References

[ i] C I A R L E T P.G., RAVIART P.A., The c o m b i n e d effect of curved bounda-

ries and numerical integration in i s o p a r a m e t r i c finite element
methods. In: The M a t h e m a t i c a l F o u n d a t i o n s of the Finite E l e m e n t
M e t h o d w i t h A p p l i c a t i o n s to Partial Differential E q u a t i o n s (A.K.
Aziz, Editor), Academic Press, N e w York, 1972, pp. 409-474.
[2] CIARLET P.G., The Finite E l e m e n t Method for Elliptic Problems.
North-Holland, Amsterdam, 1978.
[3] D O K T O R P., On the density of smooth functions in c e r t a i n sub-
spaces of S o b o l e v space. C o m m e n t a t i o n e s M a t h e m a t i c a e Universita-
tis Carolinae 14 (1973), 609-622.
[4] F E I S T A U E R M., On the finite element approximation of a cascade
flow problem. (To appear).
[~ F E I S T A U E R M., ~ENf~EK A., Finite element methods for n o n l i n e a r el-
llptio problems. (To appear).
[6] ~ENf~EK A., N o n h o m o g e n e o u s boundary conditions and curved triangu-
lar finite elements. ApI. Mat. 26 (1981), 121-141.
[7] ~ENf~EK A., Discrete forms of Friedrichs' i n e q u a l i t i e s in the fi-
nite e l e m e n t method. R.A.I.R.O. Anal. num4r. 15 (1981), 265-286.
[8] ~ENf~EK A., How tO avoid the use of Green's t h e o r e m in the
Ciarlet's and Raviart's theory of v a r i a t i o n a l crimes. (To appear).
[9] ZLAMAL M., Curved elements in the finite element methods. I. SIAM
J. Numer. Anal. IO (1973), 229-240.
MATHEMA CAL SOLUTION OF DIRECT
A N D INVERSE PROBLEM
FOR TRANSONIC C A S C A D E FLOWS
P. BOLEK, J. FOI~T, K. KOZEL, J. POLASEK
National Research Institute f o r Machine Design
11000 Prague 9- B6chovice, Czechoslovakia

The w o r k deals w i t h numerical solution of d i r e c t and inverse

problem of transonic cascade flows b a s e d on p o t e n t i a l model. Gover-
ning e q u a t i o n of a direct problem is full p o t e n t i a l equation, gover-
ning e q u a t i o n of an inverse problem is e q u a t i o n for M a c h number in
hodograph plane (~,~), #-velocity potential, b-stream function.
Both equations are p a r t i a l differential equations of second order, mi-
xed e l l i p t i c - h y p e r b o l i c type. In the s o l u t i o n of d i r e c t problem one
can c o n s i d e r discontinuity of the first d e r i v a t i v e s along some cur-
ves c a l l e d shock waves, in the inverse p r o b l e m one must find c l a s s i -
cal solution.
Numerical solution of both problems is b a s e d on using finite dif-
ference method and J a m e s o n ' s rotated difference scheme. The s y s t e m of
difference equations is solved iteratively using succesive line rela-
x a t i o n method.
The w o r k presents results of numerical solution of t r a n s o n i c
flows in cascade of c o m p r e s s o r and turbine type and one example of
numerical solution of inverse problem.

I: Direct problem
A steady irrational isoentropic flow is fully described by the
quasilinear partial differential equation of m i x e d elliptic-hyperbo-
lic type for a v e l o c i t y potential:
(a 2_ ~2)# - 2# ~ ~ + (a 2_ #2)# : 0 , (i)
x xx x y xy y yy
where ¢ is v e l o c i t y potential and a : a(¢ 2 + #2).
x y
We assume the e x i s t e n c e of w e a k shock w a v e s as curves of d i s c o n t i n u i -
ty of the first d e r i v a t i v e s #x,#y. The weak solution is a s s u m e d in a
class K(n), where ~ is a d o m a i n of s o l u t i o n (see [ i]).
The mathematical formulation of transonic cascade flows is some
combination of Dirichlet's, Neuman's and periodic boundary value pro-
blem. On the inlet b o u n d a r y we p r e s c r i b e a Dirichlet's condition
(w : w ~ , on p r o f i l e contour a Neuman's condition of n o n - p e r m e a b i l i t y
(%#\~n : 0) and on the outlet boundary also a Neuman's condition
(w = w2 ), where ~2 is a c o n s t a n t determined uniquely by the value of
 362

circulation of v e l o c i t y around the o n e p r o f i l e of the c a s c a d e y. Po-

tential % still satisfies a Kutta-Youkovski condition on the trai-
ling e d g e of the p r o f i l e . The value of y, u n k n o w n in a d v a n c e , is d e -
termined during iteration process of the n u m e r i c a l solution.
Equation (i) is p o s s i b l e to l o c a l l y transform to the form

(I - M 2 ) ~ s s + ~nn = 0 (2)

that is s i m i l a r to e q u a t i o n (i), M-Mach number, M = M(¢~), M-given

function, s - streamline direction, n - normal.
Consider (x,y) coordinate system and regular orthogonal grid.
Jameson's concept of stable difference scheme is b a s e d on c e n t r a l
difference approximation of second order for ~ using ~
ss %xx' xy, yy
in e l l i p t i c point (I M 2 < 0) a n d b a c k w a r d a p p r o x i m a t i o n of f i r s t
order for ~ss in h y p e r b o l i c point (I - M 2 < 0). Central approximation
of s e c o n d order in b o t h cases is u s e d for • (details see [ i]).
nn
The system of d i f f e r e n c e equations is s o l v e d by a S L O R m e t h o d .
It is s o l v e d in o n e ste~ of i t e r a t i o n for g r i d points ivinq on line
x~ : const., s u c c e s i v e in the d i r e c t i o n of flow. The relaxation Dara-
1
- m e t e r is c h o s e n 1.7 for all m e s h p o i n t s in line x i = const., if all
this points does not lie on p r o f i l e contour and if t h e i r local Mach
number in c o m p u t e d iteration is less than i; a n d e q u a l to 1 in o t h e r
cases.

II: Inverse problem

Solving inverse problem of transonic flow o v e r an a i r f o i l or
through a cascade the following governing equation in h o d o g r a p h plane
has been used
AM~+ BM + cM~ + DM2 ~ 0, Ca)

B=M
c : -(i + 3~I--M 2 + - y3-~
- M 4 )p3-~
~-i '

D = -(i + ~M2p -I)

M-Mach number, ~ - velocity potential, ¢ - stream function.

Smooth solution is c o n s i d e r e d in this case due to r e g u l a r i t y of t r a n s -
formation (x,y) ~ (¢.#). Boundary value problem is b a s e d on eq.
(3) and Dirichlet's conditions for an a i r f o i l or c o m b i n a t i o n of D i r i -
chlet's, Neuman's and periodicity conditions for a c a s c a d e .
The details are described in [2]. Numerical solution of the pro-
blem is a s i m i l a r to the solution of eq. (i). Knowing M(#,~) we find
angle ~ (oriented angle of the flow in (x,y) system)
 363

f p ~-i M M-IdT
G0

and then streamline coordinates ("zero" streamlines)

@ @
cos
q(M) dT, y(@,~) = y0(@0,~)+@ f ~ s i n 0 d~,
x(~,~) = X(~o,~) + f
@o 0
q = (U 2 + V 2) : Fi (M) , Fi - given function.

IiI: N u m e r i c a l results
Fig. I shows the iso-Mach lines of transonic flows c a l c u l a t i o n
for compressor cascade with u p s t r e a m Mach number M ~ = 0.83. We can
see the typical choked fows with so called closed sonic line (M = i).
It means that first end of the sonic line is situated on lower profi-
le surface and the other end is situated on the upper profile surface.
Fig. 2 shows the iso-Mach lines of transonic flows c a l c u l a t i o n
for turbine cascade with u p s t r e a m Mach number M~= 0.337 and d o w n s t r e a m
Mach number M2= 0.803. Small supersonic region (M > i) is situated near
lower profile surface. This cascade is more cambered and there-
fore the p r o b l e m of numerical solution of transonic flows through
this cascade is very complicated. The comparisons of our numerical
results and e x p e r i m e n t a l data is published in [4].
Fig. 3 shows results of inverse p r o b l e m for given Mach number
along upper (Mh) and lower (M d) profile surface (fig. 3a)~ fig. 3b
showes g e o m e t r y of found cascade c o r r e s p o n d i n g given d i s t r i b u t i o n
of Mach number along profile surface and o t h e r parameters.

References
[i] F o ~ t J . , K o z e l K . , N u m e r i c a l S o l u t i o n of P o t e n t i a l T r a n s o n i c
Flow P a s t Blade C a s c a d e s , Strojnick~ ~as. 35 (1984), 3 (in czech).

[2] Bolek P., M a t h e m a t i c a l S o l u t i o n of I n v e r s e Problem of T r a n s o n i c

Potential Flow P a s t A i r f o i l and Through a Cascade, T h e s i s ~VUT,
Prague, 1982 (in czech).

[3] Jameson A . , N u m e r i c a l C o m p u t a t i o n of T r a n s o n i c Flows With Shock

Waves, S y m p o s i u m T r a n s o n i c u m II, C~tingen 1975, Springer Verlag,
1976.

[4] Fort J., Kozel K., C a l c u l a t i o n of T r a n s o n i c Flow Through

Compressor and T u r b i n e Cascades Using R e l a x a t i o n Method f o r
Full Potential Equation, International C o n f e r e n c e on Numerical
Methods and Applications, Sofia, 1984 (Proceedings).
 364

[5] Fogt J., Kozel K., Numerical S o l u t i o n of t h e I n v i s e i d S t a -

t i o n a r y T r a n s o n i c Flow P a s t an I s o l a t e d A i r f o i l and Through
a Cascade, Applmath I, Bratislava 1984 (Proceedings).
[6] KoZel K., Pol~ek J . , Numerical S o l u t i o n of T w o - D i m e n s i o n a l
and T h r e e - D i m e n s i o n a l S n v i s c i d T r a n s o n i c Flow, P r o b l e m e u n d
Methoden der Mathematischen Physik, 8. Tagung, 1983, in
Teubner Texte zur Mathematic, band 63, Leipzig.
 365

Fig. 1 : Compressor cascade. Iso-Mach lines of c o m p u t e d flow

field, increment AM = 0.05, M ~ = 0.83

. . . . . . . . . . A ~

1 f(-'YJ~ltkl I'/1 /

D4

07

Fig. 2 : Turbine cascade. Iso-Mach lines of c o m p u t e d flow field,

increment A M = 0.05, Moo = 0,337, M 2 = 0.809
 366

4.2 ~

= 30 * /~ - 4G. ~r °
4~0
M~ = 9.84o M z = O. C~8

0.~¸

Fig. 3a: Inverse problem. D i s t r i b u t i o n of Mach number along

• . °

Fig, 3b: Inverse problem. Cascade geometry for given distri-

b u t i o n of Mach number along profile.
EINIGE ANWENDUNGEN
DER MEHRDIMENSIONALEN
APPROXIMATIONSTHEORIE ZUR
LOSUNGSEINSCHLIEt UNG
BEI RANDWERTAUFGABEN
L. COLLATZ
Inst. f~r Angewand~ Mathematik der Universit~t Hamburg
Bundesstre~fie 55, D-2000 Hamburg 1~ West Germany
S u m m a r y . In t h i s s u r v e y l e c t u r e we s u m m a r i z e a t f i r s t s o m e e l s e w h e r e
d e s c r i b e d m e t h o d s for i n c l u s i o n of s o l u t i o n s of l i n e a r a n d n o n l i n e a r
b o u n d a r y v a l u e p r o b l e m s , a n d a p p l y t h e m to c e r t a i n t h r e e d i m e n s i o n a l
p r o b l e m s . In s i m p l e c a s e o n e c a n c h e c k w i t h a i d of m u l t i v a r i a t e A p p r o -
x i m a t i o n - t h e o r i e , w h i c h d e g r e e of a c c u r a c y c a n b e r e a c h e d . A n u m e r i c a l
three-dimensional e x a m p l e s h o w s ~his.
A b s t r a c t . In d i e s e m U b e r s i c h t s v o r t r a g w e r d e n z u n ~ c h s t a u c h s c h o n a n d e r n -
o r t s b e s c h r i e b e n e M e t h o d e n zur E i n s c h l i e B u n g d e r L ~ s u n g e n y o n l i n e a r e n
und nichtlinearen Randwertaufgaben z u s a m m e n g e s t e l l t u n d d a n n an v e r s c h i e
denen dreidimensionalen Aufgaben getestet. Die GUte der erreichten
N ~ h e r u n g k a n n in e i n f a c h e n F ~ ! l e n m i t H i l f e d e r m u l t i v a r i a t e n A p p r o x i -
m a t i o n s t h e o r i e b e u r t e i l t w e r d e n , w i e es an e i n e m e i n f a c h e n B e i s p i e l
v o r g e f U h r t wird.

I. E i n f H h r u n g und Zielsetzung
Vorgelegt sei eine Operatorgleichung der Form
Tu = r. (1.1)
T sei e i n g e g e b e n e r (evtl. nichtlinearer) Operator, der einen Bereich
D eines halbgeordneten Banachraumes K I in e i n e n halbgeordneten Banach-
raum R 2 abbildet, r 6 R 2 ist g e g e b e n und u6D gesucht. H~ufig ist R I = R 2.
F~r die Praxis ist folgender Ordnungsbegriff n~t~ lich: F~r zwei in
einem Bereich B des n-dimensionalen Punktraumes R n definierte reellwer-
tige Funktionen g,h bedeute
g < h, daS g(x) < h(x) ist fur alle x 6 B. (1.2)
Dabei bezieht sich das Zeichen < auf die klassische Ordnung reeller
Zahlen. Wenn g < h gilt, kann m a n das Intervall I=[g,h] einfUhren als
die Menge p(x) der Funktionen I = {p(x), g < p < h}.

Hat man eine N~herungsl~sung z(x) f u r u(x) berechnet, so i n t e r e s s i e r t

sich der Anwender daf~r, wieviele der vom Computer ausgedruckten Dezi-
malen richtig sind und garantiert werden k~nnen, d.h. es soll ein ge-
n~gend kleines Intervall I=[v,w] angegeben werden, welches mit Sicher-
heit eine L~sung u yon (1.1) enth~it mit
v < u < w (1.3)
Die in d e n A n w e n d u n g e n auftretenden Probleme sind h~ufig so k o m p l e x e r
Natur, d a b es fur d e n M a t h e m a t i k e r in d e r R e g e l zu s c h w i e r i g ist, ein
solches Intervall anzugeben. Trotzdem wurden hier in n e u e r e r Zeit F o r t -
schritte erzielt, u n d es soll h i e r beschrieben werden, wie man bei ver-
schiedenen einfachen Modellen Z u m Ziel kommen kann. Dabei sollen beson-
 368

ders dreidimensionale Aufgaben betrachtet werden.

II. M e t h o d e der Berechnun ~ und Zusammenstellung der Grundlagen

I. O p e r a t o r e n monotoner Art. Der Operator T heiBt "von m o n o t o n e r Art"
oder "inversmonoton" (SCHRODER [62]), falls
aus Tf < Tg folgt: f < g f~r alle f,g 6 D. (2.1)
FUr weitreichende Klassen linearer und nichtlinearer elliptischer und
parabolischer Gleichungen und Integralgleichungen, aber auch fur g e -
wisse F~lle hyperbolischer Gleichungen wurde monotone Art bewiesen
(vgl. COLLATZ [68],[81], WALTER [70], BOHL [74], SCHRODER [80] u . a . ) .

FUr viele Anwendungen ist a u c h ein anderer Monotoniebegriff wichtig :

T heist synton (antiton), wenn Tf<Tg aus f<g (f>g) folgt fur alle f,g6D.

2. L i n e a r e und nichtlineare Approximation. Sei T e i n O p e r a t o r von mono-

toner Art; man versucht zwei F u n k t i o n e n v(x), w(x) aufzustellen mit
Tv < r < Tw; (2.2)
dann gilt die Einschlie~ung (1.3). F~r die N u m e r i k liBt m a n v und w
noch von Parametern a ,b abh~ngen:
v=v(x,a)=v(x,a I ..... ap) , a n a l o g w : w ( x , b I ..... bq)
und bestimmt die a ,b so, dab das Intervall m6glichst klein wird.

3. L i n e a r e und nichtlineare Optimierung. Auf dem Computer bestimmt man

die a ,b aus der semi-infiniten Optimierung:

-6<w(x,b) - v(x,a) < 6; Tv(x,a) < r(x) < Tw(x,b) ; 6 = Min. (2.3)

4. Iterationsverfahren. Brauchbare Schranken v,w bekommt man h~ufig mit

Hilfe eines Iterationsverfahrens, ausgehend von passenden Startelementen
v0,w 0 wie es schon oft ausf~hrlich dargestellt worden ist (BOHL [74],
SCHRODER [80], COLLATZ [68] u.a.).

5. S i n g u l a r i t ~ t e n . Diese erfordern i.a. eine besondere Beachtun@, wie

es h i e r in III n ~ h e r behandelt wird. Wenn man den Typ und die Lage der
Singularit~ten kennt, so k a n n man sie "abspalten". (vgl. TOLKSDORF [85],
DOBROWOLSKI [85], WHITEMAN [85]). Hier ist n o c h viel Forschung n~tig.

6. S c h a u d e r s c h e r Fixpunktsatz. Der Operator T sei "monoton zerlegbar",

d.h. er lasse sich als Summe eines syntonen Operators T I und eines an-
titonen Operators T 2 schreiben. Wenn der O p e r a t o r T ~berdies vollstetig
ist, so l~Bt sich unter gewissen Voraussetzungen mit Hil~e des genann-
ten S a t z e s die Existenz yon mindestens einer L6sung u im I n t e r v a l l
[v0,w 0] n a c h w e i s e n , vgl. die in 4. g e n a n n t e Literatur.

7. M______ultivariate T s c h e b y s c h e f f Approximation. Eine gegebene stetige re-

ellwertige Funktion f(x)6C(B) soll durch Funktionen ~ einer gegebenen
 369

Menge WcC(B) im T s c h e b y s c h e f f s c h e n Sinne, d.h. bezUglich der Maximum-

norm "m~glichst gut" approximiert werden; man fragt nach der "Minimal-
abweiehung" p(f,W)=inf I l~-f{ ! .
~6W
Wenn es zwei (normalerweise endliche) Punktmengen M I , M 2 yon B g i b t mit
der E i g e n s c h a f t : FUr kein Paar ~,~6W ist
~(x) - ~(x) > 0 f~r alle x6M1; ~(x)-~(x)<0 f~r a l l e x6M 2 (2.4)
so h e i B t die Vereinigung M von M I und M 2 eine H-Menge ("Haar-Menge")
M = M l U M 2. N u n sei h(x) eine Funktion aus W und es g e b e eine H-Menge M,
f~r d i e der Fehler £=h-f auf M # 0 ist; ferner soll es k e i n e Funktion
~6W geben mite. (h-~)>0 auf M. Dann bestehen fdr p die Schranken
(COLLATZ-KRABS [73], MEINARDUS [67])
m1=i~f Ecl± p(f,m £ ]IE1i = m2 (2.5)
M
Dies gibt einen Anhaltspunkt f~r d i e G ~ t e der erreichbaren Ann~hrung
f~r f in d e r Klasse W: I s t m I "nahe" an m2, so h a t m a n m i t h n a h e z u die
beste Ann~herung an f erreieht; f~r m 1 = m 2 ist h e i n e beste Approximation
Ist m I gr~Ber als die g e w ~ n s c h t e Genauigkeit, so mug man zu e i n e r ande-
ren F u n k t i o n e n k l a s s e W ~bergehen (z.B. mehr Parameter verwenden).

8. A l g o r i t h m u s f~r H - M e n g e n . In (2.4) s i n d ~ u/id ~ h e i i e b i g e ElemenZe

a u s W, sodas die NachplWifung bei nichtlinearer Approximation schwie-
rig werden kann. Bei vorliegender N~erung h~eW gendgt es a b e t , h~als
q zu n e h m e n ~nd (2.4) nut f~r beliebiges ~ W zu prdfen (?h.Defer~-
J.P.Thiran, Exch3nge AIKoritbm for multivariate Polynomials, Intern.
Ser. Numerical ~ath.59(~982) "115-q28. Dann ist H = M I U M 2 e i n e H-Menge,
wenn die Ungleichunge,
~(x) - ~(x) > 0 fur x£M1; -[~(x) - ~(x)] > 0 fGr x g M 2 (2.6)
keine LSsung ~(x)6W zulassen. Falls W ein linearer Unterraum v o n C(B)
ist, kann man zum Nachweis, dab e i n e H-Menge vorliegt, den GauSschen
Algorithmus benutzen (ein B e i s p i e l in Nr. IV). Wenn eine H-Menge vor-
liegt, gelten die Schranken (2.5) und man kann die G U t e der N~herung
beurteilen.

III. Einige Testbeispiele

I. E i n e nichtlineare Anfangs~Randwertaufgabe. Ein sehr einfaches Modell
der Navier-Stokesschen Gleichungen ist d i e e i n d i m e n s i o n a l e Gleichung
(Burger's Gleichung) f~r e i n e Funktion u(x,t) :
~u ~u 82
Tu = ~ + u - ~-~ -9. u = 0 in B
8x 2 ~9 .! 9

mit dem Bereich B = {(x,t), 0 < x < ~, t > 0}

und den Anfangs- und Randbedingungen, Fig. I:

u(x,0) = sin x f~r 0 < x < z, u(0,t) = u(z,t) = 0 f~r t > 0.
 370

FUr die N~herungsfunktion w(x,t) wird der Ansatz gemacht:

n
u ~ w(x,t) = sin x + Z a. ~j (x,t)
j=1 3
mit ~I = t sin x, ~2 = tx(~-x), ~3 = tx2(~-x) ,~4=t2 x (n-x) ,~5=~szn(2x).
t

Bei Diskresisierung mit den Schrittweiten 1/32, (bzw. /20) in x-, bzw.
t-Richtung erh~it man im Intervall 0<t<0.2 die Fehlerschranken
v:0.1 lw-u[5 v=0.5i
n=; 0.1013 o.--6-g~gI
n=3 0.0450 0.0420 I
n=4 0.0408 0.0377 I
n=5 0.0171 0.0252]
FUr den gr6Beren Wert u=0.5 fallen die Fehlerschranken ungHnstiger aus
als fur ~=.0.I. Ich danke Herrn Dipl.Math. Uwe Grothkopf fur die nume-
rische Durchrechnung auf dem Computer.

2. EinschlieBung einer Ableitun~ bei einem unbeschr~nkten dreidimensio-

nalen Bereich. (Elektrostatisches Potential u(x,y,z) zwischen einer Ku-
gel (Potential u=l) und zwei Ebenen mit dem Po~sential u=0, Fig.2) . Ge-
sucht ist die elektrische Feldst~rke ~ im Punkte P=(0,0,1), ~=3u/3z(P)
r
Au
Randwertaufgabe: %x+Uzy+Uzz=0 in (Izl<3,1rl>l) mit r=(x,y,z) /

u = 0 auf r I und F 2 (Ix!=3) und u=1 auf F3(Irl=l ) /~%/ L%'~

Ansatz fur eine N~herungsfunktion
a0 m I I ~.~ ~ ' !
u ~ v = a + ~ + 1 2 1I a i ( .
I r - (=0 , 0 , z i ) +Ir-(0,0,-zi) ~) ~=l~k~J'm
mit v(P)=1~0<zi<l oder zi>3) / ~P ~/
Nichtlineare Optimierung: v(r)~u(t) ~v
auf F I F 2 F 3, (~) (p)=Min.
EinschlieBung fur ~ (berechnet von Herrn Dipl.Math. J~rg Haarmeyer)
m=Anzahl der Parameter un~ere Schranke fur N obere Schranke fHr
4 - 1.358 022 - 1.357 97
8 - 1.358 022 - 1.358 016

3. Sin@ul~re Kante bei einer dreidimensionalen Aufgabe. Modell einer

Temperaturverteilung u(x,y,z) in einem Raumteil B, Fig. 3
mit Au=0 in B = {(x,y,z), Izl<1; [xl<1 fNr 0<y<1, und-~<x<0 fur -1<y~0.}
und u=1 auf Grund- und Deckfl~che und auf den beiden ~
an die z-Achse grenzenden Reehteekrandflaehen; u=z2 i !i ~
fur x=-I und y=l;und u linear in x, bzw. y auf den ,
S /
restlichen zwei Randfl~chenst~cken, Fig. 3 (ausfdhr- !
liehe Behandlung an anderer Stelle). ~.3

IV. H-Mengen bei einer dreidimensionalen Randwertauf@ab e

Als ~hnliches Beispiel fHr eine dreidimensionale Temperaturverteilung
 371

u(x,y,z) werde gew~hlt; Fig. 4

-Au=1 in B = { ( x , y , z ) 0<x<2, 0<y<2, x+y<3, [zI<1}; u=0 auf 3B
A n s a t z f~r die N ~ h e r u n g s f u n k t i o n e n Vn:
u ~ v = - ~z 2 + aivi, m i t den die P o t e n t i a l g l e i c h u n g erf~llenden
n Z i=I
2 2 ~ 2
Polynomen: v~=1,2v2=x+y , v3=x-y , v4=x +y-zz ,
v5=x3+y3-3/x+ly) z , v 6 = x y ( x + y ) - ( x + y ) z 2,...

M i t Ek = Vk - u l a u t e t die O p t i m i e r u n g s a u f g a b e :
-6 < Ek < ~ in B, ~ = Min.

Ich d a n k e H e r r n D i p l ° - M a t h . Zheng Tsinghua aus S c h a n g h a i f~r die D u r c h -

f~hrung der T e c h n u n g auf e i n e m C o m p u t e r u n d fdr Fig. 4 u n d 5; m a n er-
h~It:
Verwendeter A n z a h l k der 6
polynomgrad parameter
I 2 0.25
2 4 0.153
4 9 0.0382
6 16 0.0145
8 25 0.0049~
M a n h a t d a m i t die E i n s c h l i e B u n g ]u-v25J~0.00493.

2. A l ~ o r i t h m u s zur T e s t ung der H - M e n @ e n e i g e n s c h a f t . Um ein einfaches

Beispiel zu h a b e n , w e r d e n die P u n k t e PI,P2,...,P5 wie in Fig. 5 ausge-
w~hlt; Die Koordinaten x,y,z der P u n k t e sind in d e m f o l g e n d e n Schema
angegeben. Daneben s t e h e n die F a k t o r e n vj der z u g e h 0 r i g e n T e r m e aj,vj.
B e i den P~ b e d e u t e t ein "+"-Zeichen (bzw. e i n "-"-Zeichen, da~ Pv zu
M I (bzw. M2) geh6rt. Die e r s t e Zeile des S c h e m a s ist zu l e s e n als
I - a I + 0 • a2 + 0 • a3 - 2 a4 > 0
Da h i e r lineare Approximation vorliegt, k a n n man wie b e i m G a u B s c h e n
Eliminationsverfahren Unbekannte eliminieren, indem man Ungleichungen
mit p o s i t i v e n Faktoren multipliziert und a d d i e r t , wie es in der S p a l t e
"Operation" der f o l g e n d e n Tabelle angegeben ist. D a n n stellen die Un-
gleichungen (9)(10) einen Widerspruch dar, die P u n k t e PI,...,P5 bilden
eine H - M e n g e .

Bei d e m A n s a t z alv1+a2v2 gen~gen PI,P2,P3 w i e es im S c h e m a der ge-

strichelte Rahmen andeuten soll, zum Nachweis der H-Mengeneigenschaft,
in d e m (1) u n d (7) m i t al>0, (-2+r)a1>0 einen Widerspruch ergeben. Bei
Hinzunahme v o n a 3 v 3 und P4 m i t (I) (8) e r g e b e n a1>O, (-4+4s-s2)a?>0 ei-
nen Widerspruch.
 372

Punkte _ x y zI a I •I a2(x+y) ~3xY a4(x2+y2-2z 2)

(I) PI (+) 0 0 III I 0 :! 0 -2
(2) P2 (-)0~r< 2 r 0 0 II -I -r :i 0 _r 2
(3) P3[t~ . . . . . . . . 210 - I I !!0 2
47i sl s I --i. . . . . -2s i-s 2 2-2s 2
(5) P5(+) 2 II I I 3 .......... 3

16) 2- (4) +s z. (5)

Operation -2+s 2
2!i[_s2 -4s+3sZi 0 4-s 2
(7) 2- (2)+r- (3) 0 2r-2r 2
(8) (4s-3s 2) • (3) +2. (6) -- 0 8+8s-8s 2
(9)

lot
(2-r). (I)+(7)

[ -4+4r-2r

16s-lOs 2
=2s(8-5s)~0
2
=-2-2 (l-r) 2<0

Widerspruch.

Literatur
BOHL, E. [74] Monotonie, L~sbarkeit und Numerik bei Operatorgleichun-
gen. Springer, 1974, 255 S.
COLLATZ, L. [52] Aufgaben monotoner Art, Arch.Math.Anal.Mech.3 (1952)
366-376.
COLLATZ, L. [68] Funktional Analysis und Numerische Mathematik, Sprin-
ger, 1968, 371 S.
COLLATZ, L. [81] Anwendung yon Monotonies~tzen zur EinschlieBung der
L~sungen von Gleichungen; Jahrbuch Uberblicke der Mathematik 1981,
189-225.
COLLATZ, L. [85] Inclusion of regular and singular solutions of cer-
tains types of integral equations, Intern.Ser.Num.Math. 73 (1985)
93-102.
COLLATZ, L. - W. KRABS [73] Approximationstheorie, Teubner, Stuttgart,
1973, 208 S.
DOBROWOLSKI, M. [85] On quasilinear elliptic equations in domains with
conical boundary points, Bericht Nr. 8506, Juni 1985, Univers. d.
Bundeswehr M~nchen, 16 S.
MEINARDUS, G. [67] Approximation of functions, Theory and numerical
methods, Springer, 1967, 198 p.
SCHRODER, J. [62] Invers-monotone Operatoren, Arch. Rat.Mech.Anal. 10
(1962), 276-295.
SCHRODER, J. [80] Operator inequalities, Acd. Press (1980), 367 p.
TOLKSDORF, P. [85] On the Dirichletproblem for quasilinear equations
in domains with conical boundary points, erscheint in Comm. Diff.
Equ.
WALTER, W. [70] Differential and Integral Inequalities, Springer (1970)
352 p.
WHITEMAN, J.R. [85] Singularities in two- and threedimensional elliptic
problems and finite element methods for their treatment, erscheint
in Proc. Equadiff 6, Brno 1985.
A POSTERIORI ESTIMATIONS
OF APPROXIMATE SOLIYrIONS FOR SOME
TYPES OF BOUNDARY VALUE PROBLEMS
R. KODNAR
Institute of Applied Mathematics, Comenius University
Mlynskd dolina, 842 15 Bratislava, Czechoslovakia

I. Motive

When any approximate method is employed, it is e£ importance to

know an estimation of the error involved in the approximate solution.
In general, there exist two kinds of such estimations (i) a priori
and (ii) a postericri. The a priori assessments are obtained from the
qualitative properties of the problem. They ussually possess an as-
ymptotic character, are pessimistic and are used chiefly in theoret-
ical considerations. The a posteriori assessments are carried out on
the basis of already constructed approximate solution. Among the ways
of construction of a posteriori estimations, a significant role is
played by s.c. counter-direction methods. They are based on the fol-
lowing simple considerations:
Let A be a linear positively definite operator in a real Hilbert
space H. Then the problem to find a generalized solution of the equa-
tion
Au=f, f~H (1)
and minimization of the functional
F(u)=[U,U~A-2(U, f)H (21
are equivalent ~I~ . In (2) [ 'JA denotes a scalar product in the spa-
ce H A of the generalized solution to equation (1). It holds
m~ F(ul=-Iluoll2 ,
HA

llun~ II~=F( un >+ IIUo II A2

where u n is an approximate solution constructed by the variational
method and I! II
A is the norm in H A . Usually the numbers d can be con-

strutted greater than flu011~ but close to it (lower bound estimation

of F(u0)). Then with their aid we get
 374

IIun-u o lIA~(F(un)+a )I/2. (3)

Such a construction of numbers d is possible, for example, in
the following case:
Let us assume that there exists a functional P1(u) such that
inf F 1(u)=lluOll~ .
HA
Then it is possible to select d=F1(u), u~H A. With the selection of u
being suitable, it can be achieved that d is very near to llUoll~.
Inequality (3) explicitly gives the error estimation of the ap-
proximate solution. In the case of equations with worse operators
than those mentioned in (I), it is not always possible to obtain an
estimation in the form of (3), but one gets estimations of some other
types. In literature several constructions of the lower estimations
d are described. But the majority of them is applicable for the sin-
gle special problems only.
Functions in the whole text are real.

2. Basic no~ions

Let Q be a bounded domain in Rm with the boundary ~Q which sat-

isfies several conditions of smoothness. Let on ~ be given a bound-
ary value problem such that uoEV is its weak solution if
~v~V: ( (V,Uo))--<v, f> +Z(v,h)-((v,w)). (4)
Nearby ((,)) is a bounded bilinear form o n Hk(Q), k~-1, v(Hk(Q), < ,>
is a duality on V, wEHk(Q),

Z(v,h)= ~ Z I ~ hpldS,
p--1 1~I % ~n pl

1
f~V +, h p l ~ L m ( + Q p ) , n i s a d i r e c t i o n an o u t w a r d n o r m a l t o Q. L e t

~u,v~V: ((u,v))=((u,v)). (6)

Theorem I. Let
~u~V,~v~Hk(Q) - ((u-v,u-v))~-0. (7)
Let v1~Hk(Q) be such that
~u~V: ((u,v I ) )=<u, f~ +Z(u,h)-((u,w)). (8)
 375

Then
~Vn~V: a 2 llUo-Vnll~=((Vn,V n) )+2 <Vn, f>-2Z(Vn,h)+
(9)
+2((Vn,W))+((Vl,Vl)),
holds.
Proof. Immediately follows from (5),(6),(7),(8).
If the informations on regularity of solution of the starting
boundary value problem are avalaible construction of the lower esti-
mation may be simplified. Consider a linear boundary value problem
Au=f in Q, (10)
Biu=O on ~Q,
where A is a differential operator of 2k-th order. Denote
K=[u~C (2k'll(~laC(2k)(Q), AueL2(Q)~,

DA=[ulu~K, u fulfills all the boundary conditions from (I0)~

Suppose that A,f,Q and the boundary conditions in (10) are such, that
the solution u 0 of the problem (10) belongs to D A. Let (('))I be a
symmetrical bilinear form such that
}u~K: ((u,u))igO, ((u,u))1=O~>u=0, (11)
#u~K,~V~OA: ((u,v))1=(Au,v)52. (~2)
Remark I. It can be shown that such a form exists for the most of
the boundary value problem with Laplace and also with blharmonio op-
erator. Form (('))I to the given problem is not uniquely defined.
Theorem 2. u 0 minimizes in DA the functional

F1(u)=((u,u))1-2(u,f)se.
Let v£K be such that Av=f. Then
F1(uo)g -((v,v))I.
Proof. ~ U~DA,~V(K: ((v-u,v-u))1~0. Then
((u,u))1-2(u,f)52g -((v,v))1-2(u,f)52+2((u,v))1•
From it follows
~U~DA,~V~K: F1(u)~ -((v,v))l-2(Av-f,u)L2.

Denote J(v)=((v,v)). J(v) is a functional defined on Hk(Q). Let

the assumptions of Theorem I be fulfilled. Then for v~Hk(Q) fulfill-
ing (8) there is J(v)~((Uo,Uo)). Let
oa=[vlv~k(Q) and fulfil (a)j.
Then ((uO,Uo))= min J(v).
Dj
Lemma 1. The minimizing sequence for the functional J(v) converges
to the solution u 0 of the equation (4) in the following sense
 376

((Un-Uo,Un-Uo))----o. [z,O
Proof. Denote the minimizing sequence ~Unl I . Then Dj~zn=Un-U 0. It
holds ~ u ~ V : ((Zn,U))=O. From that
((Un-Uo,Un-Uo))=((Un,Un))-2((Zn,Uo))-((Uo,Uo)) =
=((un,Un))-((Uo,Uo))-
Remark 2. Procedure formulated by Theorem 1,2 is a generalization of
Trefftz method.

3. The construction of minimizin~ sequence

Denote
U={v~Hk(Q)I~u~V: ((u,v))=<u,f~+Z(u,h)-((u,w)~.
Lemma 2. The set U is convex and closed.
Proof. By direct verifying.
Lemma 3. The functional J(v) is convex on Hk(Q) and its minimum on U
is attained at u O.
Proof. Convexity follows from differentiability.
Corollary I. Relations
vEU: J(u)~J(v),
Vv~U: J~(u,v-u)~O
are equivalent. The given problem can be solved by means of varia-
tional inequalities. The obtained minimizing sequence converges by
given way to the solution u O.
Thus we get further counter-direction methods to variational
method of the solution of the primary problem.

4. Several special cases

Let A be a linear, positively definite operator A,(H,H). Let be

given further Hilbert space H I with scalar product (,)I" Let
A=T~, (13)
where T is operator TE(DT(H,HI) , B is positively definite operator
B~(H I,H I ). T~E (DT*CH I,H) is operator adJoint to T. Assume that
OT)D A, DT*D BTD A.
Operator T*BT is thus defined at least on D A. Let the problem Au=f
have solution Uo6DA, while f=T~g, where g~DT~. Denote
wo=Tuo, w=Tu, U~DT,
 377

G(w)=(Bw,w)1-2(g,w) 1 , w~H 1.
Theorem ~. w 0 minimizes G on TDTCH I. If T~v=f, then
1
a(Wo)= - -~-(v,v) I, (14)
where a is a constant from positively definiteness of operator B.
Remark ~. In case A = A & a n d Dirichlet boundary conditions we get from
Theorem 3 the principle of the method of unharmonic residue [I ] .
Corollary 2. UI={V~DT, IT~v=f } is convex and closed set. (v,v) I is a
convex functional on H I. If there is Wo~U I and a~=1 (from (14)), then
W 0 minimizes functional (v,v) I on U I •

Example. Consider a boundary value problem

/~u=f, U IDQ= YH~u [~Q=0, f6L2(Q) (15)

Let the problem (15) have the solution u0%C(1)(~)~C(4)(Q). Denote:

H=L2( Q)--HI ,
DA={Ulu~C(4)(Q)NC(1)(~), ~ u ~ H , u fulfil the boundary
conditions from (15) I'
DT={Ulu~C(1)(~)~C(4)(Q), u fulfil the boundary
conditions from (15)},
Define
~u,v~DT~. (u,v)1~, Z IDiuDiv dQ_-(u,v)E2 .
Ill-~2 Q
96
Then there is in (13) T=T = ~ , B is an identic operator. On lower es-
timations of minimum of the functional
U£DA: (~,~u,u)L2-2(u, f)L2
Theorem 3 and Corollary 2 can be used.
Remark 4. Problem (15) is a mathematical model of clamped plate. Sim-
ilarly the mathematical models of further kinds of boundary of plate
and web we may investigated [2] .
Remark 5- Procedures from sections 2,3,4 may also be used for non-
linear problem of the special type:
Uo£V is called the solution of the problem if
f(x,u0+w)£L2(Q),
~v~V: ((v,u o) )+ If(X,Uo(X)+W(X) )v(x) dQ--Z(v,h)-((v,w)).
Q
At the same time it is supposed that function f: ~×RI-~ R I is contin-
uous and for fixed x£Q
f(x, rl)~f(x,r 2) ~ r l , r 2 ~ R I, r1~r 2
holds.
 378

All the rest notations are the same as in (4).

5. Slobodyanskii procedure

In ~3] Slobodyanskii proposed procedure to get lower bound as-

sessment. Generalization of this procedure for further, even non-
linear problems, is described in ~4] °

REFERENCES

[I] Michlin! S.G. : Variationsmethoden der Mathematischen Physik.

Berlin: Akademie Verlag 1962.
[2] Kodn~r I R~ : A posteriori estimation of approximate solution for
mathematical models of plate and web. (in preparation).
3] Slobod~anskii ~.G. : On corvension of the problem of minimum of
a func~iohal into the problem of maximum. Doklady Akad. Nauk
rl
SSSR 91, Nr.4 (1953) (Russian).
[4J D~ubek, J., R°Kodngr, M.Skaloud: Limit state of the plate ele-
ments of steel structures. ~sel-Boston-Stuttgart: Birkh~user
Verlag 1984, 298p.
NONLINEAR DYNAMICS SYSTEMS -
BIFURCATIONS, CONTINUATION
METHODS, PERIODIC SOLUTIONS
M. KUBitEK and M. HOLODNIOK
Department of Mathematics and Computer Centre, Prague Institute of Chemical
Technology
166 28 Praha 6, Czechoslovakia
[. INTRODUCTION

Let us consider a system of autonomous nonlinear ordinary differential equations

(nonlinear dynamic system)
y'= f ( y , p ) (1)
where "= d / d r , y ~ Rn , p E Rm are parameters, f : Rn x R m ~ R n , f ~ C1
Steady s t a t e s o l u t i o n s of (1) are defined by
f(y,p) = 0 (2)
A set S(f) ~ Rn x Rm , S ( f ) : (y,p) , f(y,p) = 0 , i s c a l l e d " s o l u t i o n diagram" [1£]~
sometimes also " b i f u r c a t i o n diagram" [ l O ] . S o l u t i o n diagram i s mostly considered f o r
one parameter, i . e . Pl o n l y , w h i l e values of remaining parameters P 2 ' ' Pm are f i x e d .
Continuation a l g o r i t h m s have been developed in the l a s t ten years f o r an automatic
computation of such s o l u t i o n diagrams [ e . g . , 1 4 , 24, 9, lO, l l ] . Stability of steady
s t a t e s o l u t i o n s can be determined on the basis of eigenvalues ~ of the Jacobian m a t r i x
J = [gf/~Yt • I f an eigenvalue A crosses the imaginary a x i s i n complex plane (by va-
r y i n g some parameter, e . g . , PI ) a b i f u r c a t i o n occurs in generic cases. Several review
papers surveying numerical methods f o r l o c a t i o n of b i f u r c a t i o n p o i n t s appeared r e c e n t l y
[22, 20, 19, ?]. Four i t e r a t i v e a l g o r i t h m s f o r the e v a l u a t i o n of Hopf b i f u r c a t i o n p o i n t s
have been published in [ 5 ] .
Nain purpose of t h i s paper i s to discuss p e r i o d i c behaviour observed in two t y p i c a l
mathematical models of chemical and engineering systems, review computational methods
f o r c o n t i n u a t i o n and b i f u r c a t i o n of p e r i o d i c s o l u t i o n s .
The f i r s t model i s w e l l known Lorenz model [21] of lhe flow in the l a y e r of l i q u i d
heated from below (the Rayleigh - B6nard problem). The system
Yi = - 6 Y l + 5Y2 ' Y2 = - YlY3 + rYl - Y2 ' Y3 = YlY2 - bY3 (3)
is obtained by a reduction of the system of Navier Stokes equations and the equation
describing heat transfer. The dimensionless parameters p = (r,d ,b) correspond to :
6 - Prandtl number, r - reduced Rayleigh number, b is related to a wave-number of
the convective structure . A detailed description of behaviour of the model can be
found in the Sparrow's book [27] , a structure of periodic solutions was discussed
in [7] .
 380

The second model describes behaviour of two well mixed reaction cells with linear
diffusion coupling and the '~russelator" reaction kinetic scheme. [he model is used
as a standard model system for the discussion of dissipative structures in nonlinear
chemical systems [23] . It can be written in the form

Yi = A - (B + l)y I + Y~Y2 + D(Y3 - Yl )

Y2 = BYl - Y~Y2 + O(Y4 - Y 2 ) / Y

(4)
y3 = A - (B + l)y 3 + Y~Y4 + O(Yl - Y3 )
2
y4 = By 3 - y3y4 + O(y 2 ~ y 4 ) ~
Here p = (D,B,A,~) , A and B are constant concentrations, O and 0/~ define the
intensity of mass exchange between the cells, Yl ' Y2 and Y3 ' Y4 are dimensionless
concentrations of reaction intermediates in the first and second cell, respectivelly.

2. CONTINUATION OF PERIODIC SOLUTIONS

We shall present here a short description of an algorithm for the continuation

(and computation) of periodic solutions based on the shooting method together with
a continuation along the arclength of the solution locus. Oetailed description of the
algorithm is presented in [4] .
A periodic solution with the period T fulfils
y(T) - y(O) = 0 (5)

Considering shooting method we choose initial conditions

y(0) = x (6)
x e Rn , and the v a l u e o f the p e r i o d T . Then the system (i) can be numerically in-
tegrated for fixed p from t = 0 to t = T , the results of integration
y(T) : ~f(x,T,p) (7)
are dependent on the choice of x , T and p . Inserting (7) into (5) we obtain a sys-
tem of n nonlinear equations
F(x,T,p) : '~(x,T,p) - x : 0 (s)
with n + 1 unknowns x , T and m parameters p . We have to fix one variable
except T (or add some "normalization" equation). Let us fix xk for some k , in
such a way that xh actually exists on the trajectory of the k-th component of the
wanted periodic solution Yk(t) , t ~ [O~T] . To continue periodic solutions in depen-
dence on one parameter, say Pl ' we can use standard continuation algorithm OERPAR
[14, 19] for continuation of solutions of n equations (8) for n unknowns Xl,... ,
Xk_l, Xk+l,... , Xn, T and one parameter Pl ' This continuation algorithm requires
an evaluation of the functions F in (8) and of the Jacobi matrix ~F/@x, @F/~T,
F/@Pl " Elements of the Jacobi matrix can be determined on the basis of variational
differential equations for variational variables
V(t) = 9 y / @ x , q(t) = ~ y / g p l , (9)
V is n by n matrix and q is n by 1 , i.e.,
 381

V" = JV , V(O) = I , q" = Jq + 8 f / @ P l ' q(O) = 0 (lO)

The elements of the Jacobi matrixof the system (8) are~then defined as
@F/@x = V(T) - I , @F/ @T = f(y(T),p) , @ F / ~ P l = q(T) (ll)

The continuation routine can proceed until the fixed value of xk "disappears" from
the course of the periodic solution. To avoid this disappearance, the algorithm ex-
changes xk adaptively.
Several solution diagrams obtained by the continuation algorithm are presented
in Figs i - 3 .
The stability of the computed periodic solution can be determined on the basis
of characteristic multipliers, i.e., of eigenvalues ~ of the monodromy matrix
M : @]o'/gx : V(T) (12)
One multiplier is always equal to 1 because (i) is autonomous.~If all remaining multi-
pliers lie inside the unit circle, the periodic solution is stable, if at least one
of them lies outside, then the periodic solution is unstable.
The use of the above described continuation algorithm is limited by the applica-
bility of the shooting method. If the initial value problems are unstable, i.e., there
are multipliers of the order 105 or higher, the integration and thus the simple shoot-
ing method usually fails. In such cases the multiple shooting method can be success-
fully used as, e.g., for the Hedgkin - Huxley model of the conduction of the nervous
impulse, where ~ 4 ~ 1 0 9 [6 ].

3. BIFURCATIONS OF PERIOOIC SOLUTIONS

Bifurcation of periodic solutions occurs when a multiplier crosses the unit circle
when varying a parameter. It can happen in three qualitatively different ways, i.e.~
when (~: 1 ~ ~ : - i ; I~l : 1 , ~ s / i, s = i, 2, 3, 4. The cases ~3 : 1
and ~4 = 1 are of special interest [e.g., 8] .

3.1 LIMIT POINTS AND SYMMETRY BREAKING BIFURCATION POINTS ( (~= l)

The monodromy matrix M has ~ : l as an eigenvalue (of multiplicity two) and,

therefore, the Jacobi matrix @F/ 8 x has two zero eigenvalues. It means that no
unique dependence of the periodic solutions on a parameter exists in the neighbourhood
of this point. The bifurcation point can be either limit (turning) point (cf.~ e.g.~
point denoted L.P. in the Fig. i) or symmetry breaking bifurcation point when there
exists an inherent symmetry in the system (of. point denoted SB in the fig. 1). Both
bifurcation points can be determined by using shooting (or multiple shooting) method
and methods for steady-state bifurcations. Either methods which use evaluation of the
determinant of a matrix [13, 18, 19] or method without evaluation of the determinant
[ e.g., l, 26] can be used.

3.2 PERIOD - DOUBLING BIFURCATIONS

When the monodromy matrix has ~ = - I as an eigenvalue, then the so called

 382

$8 ,.- ...............
/
/
/
7.0
i
/
/
/
I 6.5 ., ~-. >. -~
/
I
& ..... .I #-~-L.R ..- .... ~.-.

6.0 L.P. " * " - - ' - ~ - -- -- "

",-'%,
"',"'-

I I l I I I .

1.2 1.3 1.4

4. ! , , ,
1,2 1.3 1,& 15 D

FIG. 1 : Solution diagram of periodic solutions of (4), A = 2, B = 5.9, = 0.I [25}.

A~r- amplitude of Yl " - period-doubling bifurcation point, o - symmetry
eaking bifurcation point, stable, ---- unstable.

A1 B

3
....... T# ~':%~..... 5.6
l I I
,' I ! 3T
I I I
P I
sJ I I 3T
s I
.I S I 5,4
I i
i i
I s
I i

5.2
l-laRl~" !
0.05 0.06 O 0.04 ' 0105
FIG. 2 : Solution diagram of periodic so- FIG. 4 : Bifurcation diagram of periodic
h t i o n s of (4), A = 2, B = 5.5, solutions of (4), tori bifurcation
= O.l [25]. T - tori bifurca- points. A = 2, @ = O.l.
tion point, H~B.P. - two mutually o - point of higher degeneration.
symmetric Hopf bifurcation points.
Further see legende to Fig. l.

period-doubling bifurcation occurs, i.e., a branch of periodic solutions with appro-

ximately double period (asymptotically) branches off the original branch of periodic
solutions (cf. points denoted • in the Figs i - 3).
DETERMINATION OF PERIOD - DOUBLING BIFUP.CATION POINTS ( ~ = - I ) . Four itera-
tion algorithms forlcomputation of period-doubling bifurcation points have been publi-
 383

26O
30G
A2 A2

.::% E E - C E =.-- ~
240 % _ - -

s s
, s

220 ...~
I I ~ i ,,- i ~ 26£ t I ~ L
300 320 340 360 460 L4~3 500
FIG. 3 : Solution diagram of periodic solutions of (3), ~ = 4, b : 16 [ 7 ].
A2 - a m p l i t u d e of y? . • - period doubling bifurcation point, o - symmet-
ry breaking bifurcation point, - - s t a b l e , ---- unstable.

shed and are compared in [3]. We shall summarize very briefly two of them.
Let the characteristic polynomial of the monodromy matrix M be

P(~4) = ( - 1 ) n det(M - c x ~ I ) = ~z~n + a l c ~ n-I . . . . . an_l~4 + an (i3)

The c o e f f i c i e n t s a] can be computed by using standard software. ~z4 = - 1 i s the r o o t

of (13) if
n
Fn+l(X,T,p) : 1 + ~ (-1)ia i = 0 (14)
i=l
As a r e s u l t we obtain n + 1 n o n l i n e a r equations (8) (14) f o r n + 1 unknowns
X l , . . . , Xk_l, X k + l , . . . , Xn, T, Pl " Newton method i s used to solve t h i s system.
P(~4) must have one r o o t equal to u n i t y , therefore, i t can be decomposed i n t o
the form

P(~) = (d~ + 1)(d~ - 1)(~4 n-2 + b l ~ n - 3 + ... + bn_ 2) + C~4 + D (15)

The c o e f f i c i e n t s bl, ..., bn_2, C, D can be evaluated r e c u r r e n t l y . If we determine

the p e r i o d i c s o l u t i o n where C = 0 and 0 = 0 , we have a p e r i o d doubling b i f u r c a t i o n
p o i n t . We can use
Fn+l(X,T,p) = D : 0 (16)

instead of (14) and solve the system (8) (16) again by Newton method (C = 0 automati-
c a l l y f o r the s o l u t i o n ) .
A number of p e r i o d - d o u b l i n g b i f u r c a t i o n p o i n t s has been s u c c e s s f u l l y computed in
t h i s way. Some of them are r e p o r t e d in Figs 1 - 3 .
CASCADE OF PERIOD - DOUBLING BIFURCATIONS. FEIGENBAUMSEQUENCE. High accu-
racy of computed p e r i o d - d o u b l i n g b i f u r c a t i o n p o i n t s enables t o t e s t the v a l i d i t y of
Feigenbaum's r e s u l t s also f o r more complicated and continuous dynamic systems. Several
p e r i o d - d o u b l i n g b i f u r c a t i o n p o i n t s of the Lorenz model (3) are presented i n the Table 1
Results i n the Table correspond t o a cascade of p e r i o d - d o u b l i n g b i f u r c a t i o n s ~ c f .
Fig. 3. The values of the parameter r ( = p l ) a t the i n d i v i d u a l b i f u r c a t i o n p o i n t s
form a Feigenbaum sequence #
, r j t , [ 2 ] . The values
 384

TABLE 1 : A cascade of period-doubling bifurcation points in the Lorenz model ( 3 ) ,

= 16, b : 4, k : I~ x k = 3.82038.

x2 x3 Tj rj 6/]
1 20,90946 273,34849 0.30218 356,93391
2 16.85987 246,64055 0.63009 338.06197 4.9740
3 21.19530 259,36006 1.26750 334.26789 4.7313
4 17.29002 244.99724 2.53818 333.46599 4.6824
5 17.24223 244.70901 5.07771 333.29472 4.6707
6 17.25889 244.74356 10.15599 333.25806

6j = (r]+ 1 - rj)/(rj - r j _ 1) (17)

are presented in the Table, too. We can observe a very good convergence to a limit,
which is approximately ~*~4.6692 [2]
OIRECTION OF EMANATING BRANCHES. Let us have a period-doubling bifurcation
point (x*,T~,p ~) , determined, e.g., by the algorithms described above. Let us seek
periodic solutions with the period approximately eq,a] to 2T~ in the ne]ghbourhood
of (x*,T*,p ~) . Therefore, we define a nonlinear system
G(x,T,D) = "~['l~F(x,T,p), T, p ] - x : 0 (18)
for the unknowns Xl,... , Xk_l, Xk,... , Xn, T and the parameters p . The system (18)
has a bifurcation (crossection) point at (x~,T~,p ~) . There are two branches which
intersect at this point. One branch is a branch of "composed" periodic solutions
obtained by a composition of two original periodic solutions on a known branch. The
second branch is the bifurcated branch of solutions with a double period. Directions
of the branches can be evaluated by the algorithm described in [17] . Let us note that
we need second derivatives of G (computed,e.g., by finite differences). Directions
of branches resulting for the first period-doubling bifurcation point from the Table 1
are presented in the Table 2 together with starting points used for the continuation
of the bifurcated branch. More detailed description will be presented in [ 1 6 ] .

TABLE 2 : Directions of branches emanating from the first bifurcation point in the
Table i.

Direction on Starting point

original branch bifurcated branch ( A x2 = 0.I)

dx 2 dx 3 x 2 = 21.0095 (+)
- - = 0.1262 : 2. 9431
dr x 3 = 273.643 (+)
dx 2
dx 3 dT
T = 0.30617 (+)
--= l.OO42 : 0.56E-7
dr
(period = 2T)
dx 2
dT dr
r = 356. 93 (-)
-- = - 0.43[-3 : - 0.17E-3
dr dx 2
sign of the change of individual variables for starting
continuation (direction parameters).
 385

3,3 TORI BIFURCATIONS

The monodromy matrix M has the eigenvalues

(~1,2 = a t ib , a 2 + b2 = 1 , d~,2 / 1 , s = 1, 2, 3, 4 (19)

D e c o m p o s i t i o n o f the c h a r a c t e r i s t i c polynomial (13) g i v e s

p((z~) = (~4 2 OJ~ + 1)(~ n-2 + bld~n-3 + ..- + bm_ 2) + C~4 + D , (20)

where ~= 2a and ( ( ~ 2 _ ~jdL4 + 1) = (Cz4 - z ~ ] ) ( ( ~ - /~2) . The c o e f f i c i e n t s

bl,... , bn_2, C, 8 can be a g a i n e v a l u a t e d r e c u r r e n t l y similarly as above. Two ad-
ditional equations
Fn + l ( x , T , p , ~ ) = C = 0 , Fn+2(x,T,p,~) = D = 0 (21)

have t o be f u l f i l l e d at the tori bifurcation point. As a r e s u l t we have n + 2 equa-

tions (7) (21) for n + 2 unknowns X l , . . . , Xk_l, Xk+l,... , Xn, T, p , ~ . The Newton
method can be used t o s o l v e t h i s nonlinear system.
A m o d i f i e d method makes use o f t h e f a c t that (za = 1 i s an e i g e n v a l u e o f M ,
i.e., the decomposition of P(~) i s i n t h e form

p(¢z4) = ( ~ 3 _ (2a + 1 ) ~ 2 + (2a + i ) ~ - 1)(d~ n-3 + b l ~ n-2 . . . . . bn_ 3) +

+ C ~ 2 + Dd~4 + E (22)
Coefficients bl,... , bn_3, C, D, E can be e v a l u a t e d r e c u r r e n t l y and the Newton
method i s used f o r t h e s o ] u t i o n o f the n + 2 by n + 2 nonlinear system ( 8 ) ( 2 1 ) .
E = 0 automatically at the resulting tori bilurcation point.
Resulting tori bifurcation points for the model (4) are shown in the Table 3,
cf. Fig. 2 for 8 = 5.5. The parameter 0 ( = pl ) has been considered as a bifur-
cation parameter. If we continue tori bifurcation points in dependence on another
parameter of the problem, here, e.g., B ( = p2 ) , we obtain so called bifurcation
diagram [19] . Results of one such continuation are presented in Fig. 4 . The points
where __.r~3 : i or 4 = i are denoted 3T or 4T , respectively. The curve ends
at the point where ~i = (~2 = ~ 3 = 1 .

4. DISCUSSION ANO REMARKS

The algorithm for the continuation of periodic solutions can be used also for
parabolic partial differential equations when these are transformed into a set of or-
dinary differential equations by using a semidiscretization (method of lines) [12] .

TABLE 3 : Resulting tori bifurcation points for (4), A : 2, ~= 0.1, k = 1, x 1 : 2.

B x2 x3 x4 T a D

5.3 2.43727 1.90236 2.49646 7.15820 - 0.01894 0.048626

5.4 2.45219 1.92098 2.50011 7.62984 - 0,36244 0.051152
5.5 2,47887 1,92659 2.52457 8.32408 - 0.54139 0.052495
5,6 2.51478 1.92259 2.56374 9.19639 - 0.12304 0.053030
 386

An algorithm for evaluation of Hopf bifurcation points in parabolic equations has been
published recently [15] , The algorithms for evaluation of period-doubling and tori
bifurcation points can be easily used for most autonomous dynamic systems of lower or-
der, say n -~ 20 . Of course, the use of the algorithms is limited by the applicabi-
lity of the shooting method (stability). Simple modifications of the algorithms can be
used also for a nonautonomous system with a time-periodic right hand side.

REFERENCES

i. 8ecker K.H., Seydel R. : Lect. Notes in Math. 878, Springer Verlag, Berlin 1981,
p. 99.
2. Feigenbaum M.J. : J. Star. Phys. 6, 669 (1979).
3. Holodniok M., Kub$Sek M. : Computation of period doubling bifurcation points
in O.O.E. Preprint, Tech. Univ. Munchen, M-8486 (1984).
4. Holodniok M., Kubi~ek M. : J. Comput. Phys. 55, 254 (1984).
5. Holodniok M., Kubi~ek M. : Appl. Math. Comput. 15, 261 (1984).
6. Holodniok M., KubiSek M. : Continuation of periodic solutions in ordinary diffe-
rential equations with application to the Hodgkin-Huxley model, Inter. Symp. of
Numerical Analysis~ Madrid, 17. - 19. 9. 1985 .
7. Holodniok M., Kubi~ek M., Marek M. : Stable and unstable periodic solutions in
the Lorenz model. Preprint, Tech. Univ. Munchen, M-9217 (1982).
8. Iooss G., Joseph 0,0. : Elementary stability and bifurcation theory. Springer
Verlag, New York 1981.
9. Jepson A.O., Keller H.8. : in [20] , p. 219.
i0. Keller H.B. : in "Applications of Bifurcation Theory", Ed. by P. Rabinowitz,
Academic Press, New York, 1977, p. 359.
11. Keller H.B. : in "Recent Advances in Numerical Analysis", Ed. by C. de Boor,
G.H. Golub, Academic Press, New York 1978, p. 73.
12. Knedlik P., Holodniok M., Kubi~ek M., Marek M. : Periodic solutions in reaction-
-diffusion problems, 7th CHISA Congress, Prague 1984.
13. Kubi~ek M. : Appl. Math. Comput. I, 341 (1975).
14. Kobi~ek M. : ACM TOMS 2, 98 (1976).
15. Kubi~ek M., Holodniok M. : Chem. Eng. Sci. 39, 593 (1984).
16. Kubi@ek M., Kli~ A., Holodniok M. : in preparation.
17. Kubi~ek M., K l i ~ A. : Appl. Math. Comput. 13, 125 (1983).
18. Kubi~ek M., Marek M. : Appl. Math. Comput. 5, 253 (1979).
]9. Kubidek M., Marek M. : Computational methods i n b i f u r c a t i o n t h e o r y and d i s s i p a -
t i v e structures. Springer Verlag, New York 1983.
20. Kupper I., Mittelmann H.D., Weber H., Eds. : Numerical methods for bifurcation
problems. Birkhauser, Basel 1984.
21. Lorenz E.N. : J. Athmosph. Sci. 20, 130 (1963).
22. Mittelmann !~.O., Weber H., Eds. : Bifurcation problems and their numerical solu-
tion. 8irkhauser, Basel 1980.
23. Nicolis G., Prigogine I. : Self-organization in nonequilibrium systems. J. Wiley,
New York 1977.
24. Rheinboldt W.C., Burkardt J.V. : ACM TOMS 9, 215 (1983).
25. Schreiber I., Holodniok M., Kubi~ek M., Marek M. : J. Stat. Phys., to be published
26. Seydel R. : Numer. Math. 32, 51 (1979).
27. Sparrow C. : The Lorenz Equations : Bifurcations, Chaos and Strange Attractors.
Springer Verlag, Berlin 1982.
THE ROTHE METHOD FOR NONLINEAR
HYPERBOLIC PROBLEMS
E. MARTENSEN
Mathematisches Ins~tut II, Universitdt Karlsruhe
7500 Karlsruhe 1, West Germany

The ROTHE method or the h o r i z o n t a l method of lines, if it is a p p -

lied to p a r a b o l i c as w e l l as to h y p e r b o l i c evolution problems, reduces
these problems to a s e q u e n c e of e l l i p t i c problems. That from a former
point of view, such an a p p r o a c h has appeared more natural in the c a s e
of p a r a b o l i c t h a n of h y p e r b o l i c problems, may serve as an e x p l a n a t i o n
for the c o n s i d e r a b l e delay of time in s t u d y i n g the m e t h o d for b o t h clas-
ses of p r o b l e m s . So after ROTHE [11] has introduced his method in the
early thirties of o u r century, numerous parabolic differential equation
problems, linear as w e l l as n o n l i n e a r ones, have been t r e a t e d by it s u c -
cessfully; the n a m e s of L A D Y S H E N S K A J A , REKTORYS, NE~AS, and KA~UR may
stand here for m a n y others (references, for instance, m a y be s e e n from
the b o o k of R E K T O R Y S [10]). On the o t h e r hand, efforts for applying the
ROTHE method to h y p e r b o l i c problems firstly have been started during
the last d e c a d e . Results have been given mainly for c e r t a i n linear prob-
lems of m a t h e m a t i c a l physics, so as for the w a v e equation [1,2,5], the
continuity equation [3], and the M A X W E L L equations [4]; recently the
vibrating string problem with discontinuous data has been completely
s o l v e d by the R O T H E method [6]. Further linear hyperbolic problems have
been investigated by REKTORYS [10]. With regard to n o n l i n e a r hyperbolic
problems, however, one is s t a n d i n g at the v e r y b e g i n n i n g . First results
of M U N Z [8,9] concerning the q u a s i l i n e a r scalar conservation equation,
especially have shown the R O T H E method as a s u i t a b l e tool for approxi-
mation of s h o c k s and rarefaction waves.

In the following we shall consider the C A U C H Y problem for the

BURGERS equation
1
u t + ~ (uZ) x = 0 (x,t) e ~ × (0,~) (I)

where the initial values

u(x,O) = u (x) xe ~ (2)

o

are assumed to b e piecewise continuous with at m o s t a finite number of

discontinuities and existing limits for x - ¥ ~. The ROTHE method for a
fixed chosen time step length h > 0 leads to the o r d i n a r y differential
equation
 388

h
u + ~ (u2) ' = u ° x e • (3)

which for given u ix), x ~ R, has to b e solved successively according

o
to the next time step. On the solutions u(x), x ~ ~, of (3), there
are imposed piecewise continuity with at m o s t a finite number of dis-
continuities and existing limits for x - ¥ =: furthermore, the square
[u(x)] 2 , x G ~, is asked as a piecewise continuously differentiable
function (as a consequence of the foregoing, for the derivative there
may occur at most a finite number of discontinuities). Without mention-
ing in detail, the following assertions will concern to solutions of
(3) having at least these properties.

Theorem I (Behaviour at the infinity). For a solution u(x), x C ~,

of (3) it h o l d s

lim u(x) = l i m U (X) (4)

x~¥== x~¥= o

Proof follows immediately from (3) in c o n n e c t i o n with the second

L' H O S P I T A L rule:

0 = lim h[u(x)]~ llm {h

= ~x [u(x)] z } = lim Uo(X) - lim u(x)

Remark. The proof of Theorem I makes only use of the conservation

property of the underlying partial differential equation (I). Thus the
accordance of the limits (4) W i l l be obtained analogously for other
hyperbolic problems when they are given in c o n s e r v a t i o n form. For in-
stance, this holds for the EULER equations.

Theorem 2 (Global uniqueness). There exists at m o s t one continuous

solution u(x), x @ ~, of (3).

Proof. Assuming that there exist two different continuous solutions

u(x),v(x), x G ~, so the continuous function w(x) := u ( x ) - v ( x ) , x ~ ~,
does not vanish everywhere. Note that because of Theorem I it h o l d s

lim w(x) = lim u(x) - lim v(x) = 0 (5)

x ~ x~+~ x~+~

Let now x oe ~ be a point with w ( x o) + 0. If w ( x ) has at least one zero

in the open interval (-~,Xo) , then for continuity there exists a maxi-
mum zero in this interval and we denote it b y a < Xo; if, however, there
are no zeroes in (-~,Xo) , we put a = -~. Analogously let b > x O denote
the minimum zero for w(x) in (Xo,~) or stand for ~, respectively. To-
gether with (5) we get

lim w(x) = lim w(x) = 0 (6)

x~a x-b

Observing the continuity and piecewise continuous differentiability

 389

of [U(X)]Z [v(x)] 2 " x e ~, it follows from (3) and (6) by improper

integration that
b b b
w(x) dx = {u(x)- v(x)} dx = - ~ ~[u(x) - [v(x) dx
a a a

: - 7h ~
LU(X ]2 - [ v ( x ) ] 2 h (x)(u(x) + v(x) ) = 0
a

This is a contradiction to w ( x ) + 0, xe (a,b).

Theorem 3 (Local unlqueness). Let (a,b) ~ R be an arbitrary finite

or infinite open interval and let the above ordinary differential equa-
tion problem be formulated analogously for (a,b) instead of ~. Let
further u(x),x~(a,b), be a pg~ti~_~99~ous ( ~ g ~ Z ~ _ ~ i ~ 9 ~ )
solution of (3) which has a ~ 9 ~ [ ~ _ ~ for x - a ( 9 ~ 9 ~ Z 9 _ ~
for x -- b ) . Then there does not exist another continuous solution of (3)
with the same limit for x - a (x- b).

Proof only for the first case. Assume that there exists a contin-
uous solution v(x), x e (a,b), different from u(x), x 8 (a,b), but with
the same limit for x - a. Then the difference w(x):= u(x) - v(x), x ~ (a,b),
forms a continuous function satisfying

lim w(x) = 0 (7)

x~a
Next we are able to find a point x ° ~ (a,b) with properties

w ( x o) } 0 u ( z o) + v ( x o) ~ 0 (8)

Indeed, if u(x) + v(x), x ~ (a,b), has no zeroes, from continuity and

lim {u(x) + v(x)} = 2 lim u(x) > 0

x~a x~a
it follows that u ( x ) + v ( x ) > 0, x e (a,b), and so it is trivial to find
x e (a,b) satisfying (8)~ if, however, u(x) + v(x), x e (a,b), has a zero
o
x ~ (a,b), so this zero immediately fulfills the second condition in
o
(8) and the first condition follows from v ( x o) = - u ( x o) as

w(x ) = u(x ) - v(x ) = 2u(x ) > 0

o o o o
Now we denote by a* < x the maximum zero for w(x) in the open interval
o
( a , x o) if there exists a zero at all, otherwise we put a* = a. So in any
case when observing (7), we get

lim w(x) = 0 (9)

x~a *
Then by improper integration, it follows from (3) and (9) that
 390

XO XO XO

W(X) dx = {U(X) - V(X)} dx = -g ~

a* a* a*

= - w(x)(u(x) +v(x)) = - W(Xo)(U(X o) + v ( x °

here because of {8), we have the contadiction, that the left hand side
has the sign of w ( x O) + 0 w h i l s t the right hand side either has the oppo-
site sign or vanishes.

Remark I. T h e o r e m 3 gives a hint how to proceed for solving the

differential equation (3) uniquely. So if starting at some point with
a positive or negative initial value, one has to integrate to the right
or to the left, respectively. On the other hand, the sign of the exact
solution analogously indicates the direction of the characteristics.
So it turns out that local uniqueness for the ROTHE solution is assured
by integrating into the direction of characteristics.

Remark 2. As it can be seen from the example Uo(X) : I, x ~ ~ , the

sign condition in Theorem 3 plays a significant role. So the solution
u ( x ) : I, x e R, is the only one of (3) w~th limit I for x - -~, but
there exist an infinite number of further solutions with limit I for
x ~ ~ indeed, with an arbitrary real constant C, such a solution u{x),
x e ~, may be obtained as the inverse of the monotonously decreasing
function

x(u) = - h {u+ ln(u- I)} + C u e (I ~)

We shall make use of the foregoing theorems when discussing the

following four examples.

ExamD!e I (MUNZ [8]). If u (x), x ~ ~ , is the step function with

o
value 2 for negative or I for posieive x, respectively, the exact solu-
tion u(x,t),(x,t) ~ ~ x [0 ~), of the evolution problem (I) and (2) is
B
given as a shock wave at x : 7t with value 2 left or I right of the
shock, respectively. Assume that for an arbitrary time step a ROTHE
solution exists which, for convenience, will be denoted by u(x),x e R[
o
besides the general properties mentioned above let this solution be
monotonously nonincreasing with lower bound ] let it have the value 2
for x C (-~,0), and let it be continuous for x e (0,~). Note that for
such solution the limits for x - ¥ ~ exist and that everything holds for
the given initial function. The next ROTHE step u(x), x e ~, then may
be computed from (3) as a continuous so]ution with value 2 for x e (-m,0)7
for x 8 [0,~) the solution follows by means of the initial condition
u(0) : 2 in connection with the lower function Uo(X ) and the upper
 391

function 2. This especially yields

I £ U o ( X ~ ~ u(x) , x e (0,-) (I0)

From the differential equation (3) together with (I0) it follows that
u' (x) < 0, x @ (0,~)[ so u(x) x @ ~, is monotonously nonincreasing and
because of (10), it has the lower bound I. Theorem 2 as well as the
first case of Theorem 3 say that there is no further continuous solu-
tion, so the next ROTHE step is well-defined. Finally by induction, all
ROTHE solutions are uniquely determined. Because of Theorem I, for every
ROTHE solution the limit 2 for x --~ or ] for x - ~ is obtained, respec-
tively.

Example 2 (MUNZ [8]). Here u (x), x ~ R, is considered as a step

o
function with value I for negative or 2 for positive x, r e s p e c t i v e l y .
x
The exact solution is a rarefaction wave with values ~ f o r t ~ x ~ 2t,
0 < t < ~ and value I left or 2 right of the wave, respectively. As it
turns out quite similarely to Example I, the ROTHE method again can
be carried out uniquely.

Example 3 (MARTENSEN [7]). The initial values Uo(X) , x C ~, are

given as -1 for negative or ] for positive x, r e s p e c t i v e l y . The exact
x
solution is a rarefaction wave with values ~ f o r -t ~ x ~ t, 0 < t < ~ and
value -I left or I right of the wave, respectively. Evidently Theorem 3
is not applicable with respect to b o t h the infinities. If beginning
with the first time step, the ROTHE solutions u(x), x G ~, are further
asked to be continuous, monotonously increasing, and skew-symmetric
with respect to the origin, then such solutions can be constructed
successively by means of a fixed point method. Uniqueness is now as-
sured by Theorem 2. As a secondary result it turns out that all the
ROTHE solutions (contrarily to their squares) are not from each side
differentiable at the origin.

Example 4 (MUNZ [9]). If u (x) , x ~ ~ , h a s t h e value 2 for nega-

o
rive or -I for positive x, respectively, the exact solution is obtained
I
as a shock wave at x :~t with value 2 left or -1 right of the shock,
respectively. For the piecewiese continuous ROTHE solution u(x), x G ~,
beginning with the first time step, the further supposition is made
that the square [u(x)] 2 , x ~ ~, remains continuous when passing through
a discontinuity~ in such a way there is made use of the conservation
property governing the ROTHE differential equation (3). In particular,
with a well-defined discontinuity x* 8 ( 0 , ~ ) , the ROTHE solution u(x),x e ~,
is obtained with constant value 2 for x ~ (-~,0), as a monotonously de-
creasing solution of the differential equation (3) for x e [0,x*] satis-
fying the initial condition u{0)= 2 and the free boundary condition
 392

u(x*) = I, and w i t h constant value -I for x ~ (x*,~). Here Theorem 3

leads to local uniqueness for the left interval (-~,x*) as w e i l as for
the right one (x*,~)~ furthermore by m e a n s of T h e o r e m 3, this ROTME
solution turns out to be the o n l y one with exactly one d i s c o n t i n u i t y
whilst a contlnuous solution does not exist, with regard to the com-
plete ROTHE method, the d i s c o n t i n u i t i e s form a monotonously increasing
sequence.

For the e x a m p l e s mentioned before numerical computations have been

done by standard methods, where the results have shown a high accuracy
in c o m p a r i s o n with the e x a c t solutions [7,8,9]. Recently for s u c h non-
linear hyperbolic problems the L1-convergence of the R O T H E method with
respect to any compactum in the upper (x,t)-plane has been proved [9].
The pointwise convergence, however, remains still as an o p e n q u e s t i o n .

References

[I] G e r d e s , W.[ M a r t e n s e n , E.: Das R o t h e v e r f a h r e n fur die r ~ u m l i c h

eindimensionale Wellengleichung. Z A M M 58 (1978) T 3 6 7 - T 3 6 8
[2] H a l t e r , E.: Das R o t h e v e r f a h r e n fur das A n f a n g s - R a n d w e r t p r o b l e m
der W e l l e n g l e i c h u n g im A u ~ e n r a u m . D i s s e r t a t i o n , K a r l s r u h e 1979
[3] H a l t e r , E.: The c o n v e r g e n c e of the h o r i z o n t a l line m e t h o d fo~ the
c o n t i n u i t y e q u a t i o n w i t h d i s c o n t i n u o u s data. ZAMP 35 (1984)
715-722
[4] M a r t e n s e n , E.: T h e c o n v e r g e n c e of the h o r i z o n t a l line m e t h o d for
M a x w e l l ' s e q u a t i o n s . Math. M e t h o d s Appl. Sci. I (1979) 1 0 1 - 1 1 3
[5] M a r t e n s e n , E.: T h e R o t h e m e t h o d for the w a v e e q u a t i o n in s e v e r a l
s p a c e d i m e n s i o n s . Proc. Roy. Soc. E d i n b u r g h 84A (1979) 1-18
[6] M a r t e n s e n , E.: The R o t h e m e t h o d for the v i b r a t i n g s t r i n g c o n t a i n -
ing c o n t a c t d i s c o n t i n u i t i e s . Meth. Verf. math. Phys. 26 (1983)
47-67
[7] M a r t e n s e n , E.: A p p r o x i m a t i o n of a r a r e f a c t i o n w a v e by d i s c r e t i z a -
tion in time. A p p l i c a t i o n s of M a t h e m a t i c s in T e c h n o l o g y , V. B o f f i
and H. N e u n z e r t eds. S t u t t g a r t : T e u b n e r 1984, 195-21]
[8] Munz, C.-D.: U b e r die G e w i n n u n g p h y s i k a l i s c h r e l e v a n t e r S t o B w e l -
l e n l ~ s u n g e n mit d e m R o t h e v e r f a h r e n . D i s s e r t a t i o n , K a r l s r u h e 1983
[9] Munz, C . - D . : A p p r o x i m a t e s o l u t i o n of the R i e m a n n p r o b l e m for the
B u r g e r s e q u a t i o n by the t r a n s v e r s a l m e t h o d of lines. To a p p e a r in
ZAMP
[10] R e k t o r y s , K.: T h e M e t h o d of D i s c r e t i z a t i o n in T i m e and P a r t i a l
Differential Equations. Dordrecht/Boston/London: Reidel Publishing
C o m p a n y 1982
[11] R o t h e , E.: Z w e i d i m e n s i o n a l e parabolische Randwertaufgeben als
Grenzfall eindimensionaler Randwertaufgaben. M a t h . A n n . 102 (1930)
650-670
SOME SOLVED A N D UNSOLVED
CANONICAL PROBLEMS
OF DIFFRACTION THEORY
E. MEISTER
Technical U n i v e r s i t y D a ~ n s t a d t
Schlosgartenstr. ~ D 6100 - Da'r~'~stadt, West G e r m a n y
i. Introduction
Mathematical diffraction theory is concerned with the following
boundary value problem in case of an incoming or primary time-harmonic
wave-field Re[~pr(~)e-i~t] :
Given an obstacle ~ C Rn; n = 2 or 3; with boundary F = D~. Find the
scattered field #sc(~) in ~a:= R n- ~, s.th.

(l.i) (A + k2)~ (x) = 0 for x e

sc -- a
with a wave-number k = kl+ ik 2 E C++- {0} fulfilling a boundary
condition
(l.2a) Bl[~s c(X)]
: = IF
- ~sc(X)IF-- = f(x)_ of Dirichlet-type
or

(l.2b) B2[~sc(X)]
IF:_ = (~-~ + i p(x))~sc(X)]
F _ _ = g(x)_
Neumann (p ~ 0)
of { }_ type.
Impedance (p ~ 0)
In the case of edges E and/or vertices V C F existing the "edge
condition"
(1.3) ~sc(X)_ = 0(i ) and V~sc(X)_ @ Lloc2 (~a)
should hold. Besides this the scattered field should be "outgoing",
i.e. "Sommerfeld's radiation conditions" should hold

(1.4) ~sc(~) = ~(e-k2r), (~ - i.k)~se(~) : ~(e-k2r/r ~)

as r = Ixl ~
For smooth compact boundaries F this problem has completely been
solved, e.g. by the boundary integral equation method (BEM) (c.f.e.g.
COLTON-KRESS (1983) [2]) or by means of Sobolev space methods (c.f.
e.g. LEIS (1985) [ ii]). Generalizations to piecewise smoothly bounded
domains were carried out by GRISVARD (1980) [6] and COSTABEL (1984)
[4], e.g.

2. The S o m m e r f e l d Half-Plane Problem

There are a number of "canonical diffraction problems" with
domains whose boundaries extend to infinity and having corners and
 394

cusps. The most famous one is the "Sommerfeld half-plane problem",

the first diffraction problem having been treated in a mathematically
rigorous way (1896) [15].
Applying the well-known representation formula for outgoing
solutions of the Helmboltz equation (1 .i ) the Sommerfeld half-plane
problems leads to the following integral or integro-differential
equations (of the first kind) of the Wiener-Hopf type:

(2.1) 01 ~.(1M0 )(klx-x't) I(x')dx' = -4i.¢pr(X,0) for x ~ 0

in the case of the Dirichlet problem and

d2 ~#
(2.2) (---r + k 2) f~H! I )(klx-x' I) Q(x')dx' = 4i pr (x,0) for x > 0
dx z 0 U %y
in the case of the Neumann problem with the unknown jumps

(2,3) I(x') := ~ y sc (x',+0) %ysc (x',-0) for x' > 0

and

42.4) Q(x') := ~sc(X',+0) - ~sc(X',-0) for x' a 0 #

respectively.
The theory of such equations, but of the second kind, in LP(R+)
or --~'P(R+)-spaces__ for m e NO, 1 ~ p ~ ~ has been developed by
M.G. KREIN (1958/62) [9], E.Gerlach (1969) [5] and, combined with
other integral operators than 1-convolutions, by G.THELEN (1985) [17].
To solve the equations (2.1) or 42.2) on the h a l f - l i n e , or more
directly the original boundary value problem, one applies a one-di-
mensional Fourier transform to the scattered wave function

(2.5) Ssc(l,y) := f~eilX~ (x,v)dx, I 6 R, y ~ 0 .

_~ sc

The usual, or S'-distributional Fourier transform technique leads to

the following "function-theoretic Wiener-Hopf equations".in the case
of a damping medium, i.e. Im k = k 2 > 0, and an incomi~q Dlane wave:

(2.6) E_(1) + ~i ~+(~)/~x2 k2 = [i(l + k cos @)]-i

and
(2.7) O (1) + ~1 Q+(I) ~ I 2 - k 2 = -k sin 8 [l + k cos 8] -1

respectively, for the Dirichlet and Neumann case with the unknown
F-transforms E_, ~_ being holomorphic for Im ~ < k 2 and I+, Q+ being
holomorphic for Im I >-k 2 cos 8. The equations 42.6) and (2.7) are
equivalent to "non-normal Riemann boundary value problems on a line"
parallel to the real l-axis.
The well-known steps of factorization of y(Z):= ~ l z- k 2 into
 395

-i
y+(l).¥_(l), the multiplication of (2.6) and (2.7) by y_ and by y_
-i
respectively, then additive decomposition of y_.[X + k cosS] and
y -1.IX + k cosS] "I in the x-strip gives after rearrangement and
application of Liouville's theorem the explicite solutions to eqs.
(2.6) and (2.7) as

(2.8) ~+(x) = 2v9~ cosel2.y+(x)[x + k cose] -1

and
-i
(2.9) ~+(l) = -2i%/~ sin@/2.y+ (X)[% + k cosS] -I
for Im I > -k 2 cos8

These functions being known allow to calculate ~ (x,y) in both

sc
cases after applying an inverse F-transform and shifting the line of
integration in the complex l-plane to get all informations relevant,
i.e. the edge behaviour an the far field in the geometrically different
regions.
This functiontheoretic method has been applied successfully to a
big number of canonical problems in microwave theory and to other
diffraction problems, e.g. for systems of parallel semi-infinite pla-
tes (A.E.Heins (1948) [7]), or cascades of such (J.F. Carlson, A.E.
Heins (1946/50) [ i]), or cylindrical semi-infinite pipes ( e . g . L . A .
Vajnshtejn (1948) [18]).
The "canonical mixed Sommerfeld half-plane problems", where there
are given different boundary conditions on the faces 6+ of the semi-
infinite screen 6 := {(x,y) E R2: y = 0, x ~ 0}, may b~ transformed by
the same Fourier technique into a 2X2-functiontheoretic system of
Wiener-Hopf equations

(2.10) --_$(X) = K(X)$+(X)=

-- + --~(X) for -k 2 cos8 < Im X < k 2
with the known 2x2-function matrix

(2.11) K(X):= (~X-k)/(~+k) i)

= -i ~(l+k)/(X-k)
and the unknown 2xl-function-vectors

(2.12) L(I) . - ,

I (V77k - $+(X, -0) )

_~.(X) : = - ~ ^'~+(X,+0)Iv77[ "

The matrix K(X)
=
- or a closely related one - has been factorized
into [~_(X)] -I K+(X) only (1982/83) by A.E.Heins [8], (1981) by

A.D.Rawlins [14] and (1981/85) by the present author [ 12] , indepen-

dently by different methods. Now the solution of the mixed Sommerfeld
 396

problem may be written down explicitely and gives full information on

the behaviour of ~sc' VCsc as r ~ 0 and r - ~, respectively, which is
now different at the edge compared to the one-boundary-condition-prob-
lems. The corresponding mixed boundary value problems for systems of
parallel semi-infinite plates or a tube are unsolved up to now due to
the lack of a known explicit factor\ration of the 2×2-function matrices
involved (c.f.e.g. the authors paper (1984/85) [ 12] !).
The Sommerfeld half-plane problems have been generalized to the
so called "Quarter-plane Problems of Diffraction Theory" where the
half-plane, i.e. the screen 6 C R 3, is replaced by a screen ~ C R3
which is the quarter-plane R ++
2 := {(x,y,z) E R3: z = 0, x a 0, z ~ 0}
with two semi-infinite lines as edges meeting in the corner E at the
origin. Like for an arbitrary plane screen Z C R 2 the 2-dimensional
xy
F-transform applied to the unknown scattered field ~ (x), x @ R 3,
SC - - - -

leads to the following "Two-dimensional Wiener-Hopf functional equa-

tions"

(2.13) y-l(Xl,~2)~ (~i,~ 2) - ~R2\ ~(XI~X2,0) = - ~pr,R2\~(Xl,~2,0)

and
(2.14) y(ll,12)Qz(ll,l 2) - ( *)R2\Z(II,12,0 ) = - ( ~ z C p r ) R 2 \ Z ( l l , 1 2 , 0 )

where y(ll,12) := J l~ + I~ - k 2 and the indices Z and R2\Z refer to

the 2D-F-transforms of the restrictions to Z and R2\ Z, respectively.
Up to now there exists no explicit factor\ration of the multi-
^
plication operator y with respect to the complementary projectors PZ'
A
QZ := I - PZ in spaces FLP(R 2) or FwS'P(R2), s > 0, 1 < p ~ 2 (~) . But
there exists now a very general theory for "general Wiener-Hopf or
Toeplitz operators" of the form
(2.15) P2AIPI x u = v E p2 Y
for bijective continuous operators A : X - Y acting between two Banach-
spaces X,Y with bounded projectors PICS~(X), P2 @ ~ ( Y ) . This theory by
F.-O.Speck (1983/85) [16] gives necessary and sufficient conditions for
the general invertibility and Fredholm property of operators of type
(2.15) in dependance on factor\ration properties of q w.r.t. (Pi,P2).

3. Canonical Transmission Problems

Another big class of canonical diffraction problems exists given
by the following specification:
Given a primary time-harmonic wave-field Re[~pr(~)e-i~t] and a region

~i C R n, n = 2 or 3, and finitely many disjoint regions ~2,...,~N C R n,

s. th. ~ ~. = Rn. Then one looks for a scattered field ~ (x),
j=l 3 " sc --
x 6 R n, s. th. % sc <x)I~3e
- . C2(~j) N CI(~j\ {0} ) and
 397

(3.1) (A + k~)#sc(~) = 0 in ~j, j = I,...,N I

fulfilling the "transmission conditions"

(3.2a) ~sc,3"(x)-- - ~sc,t (x)- = Fjl(X)--

and on 8~jA 8~ 1 #
8# 8~
(3.25) ~j,~'----~(x)
onj --
+ ~.x" ~ - -0n- ~I( x )-- = Gjl(X)
--

with prescribed data Fil, Ghl from the primary field on the common
boundary parts 8~3, n 8nl.
Additionally the edge conditions ~j(x) = ~(x) la ' = 0(I) and V~j~
L~oc(n
j ) _ and the radiation condition for #l(X) ~s Ixl = r - ~ have to
hold.
Again in the case of smoothly bounded domains with compact bounda-
ries 8~j this "transmission or interface problem" has been solved by
the boundary integral method and in the case of two-dimensional polygo-
nal domains by M.Costabel and E.Stephan (1985) [3].
In the special case of two different media (i.e. N = 2) and a
plane interface (i.e. ~ i = 8~2= Rxy or = R I) the problem is elementary
x
and gives, for a plane wave as the primary wave-function, the well-
known relations from Snellius' law and the reflection and transmission
coefficients explicitely. The corresponding "two-dimensional Sommerfeld
half-plane problems with two media" are unsolved up to now - as far as
an explicit representation is concerned - due to the unknown matrix
factors of the 2X2-Wiener-Hopf function matrices involved here having
two different square roots ~ 2- k~ and ~ to be taken into
account [12].
A very important canonical transmission problem is the so-called
" D i e 1 ectrlc
" W edge Problem " , i.e. the case of ~i = R ~2 and ~o= r 2 \ R ~2 in
R or the corresponding "Dielectric Octant Problem" in R~-space: This
has been generalized to the "Four-Quadrant-Transmission-Problem" in R 2
with the four quadrants filled with different media. Applying 2D-
Fouriertransformation the restrictions of the unknown scattered field
mav be represented by the ID - F - transformed Cauchv-data on the semi-
infinite lines, the boundaries of the quadrants. For $i(~i,~2) one
gets e.g.

(3.3) ~I(AI,X2) : [iX2.t¢(1)(XI)

I + g&(1)(X I) + i X l . ~ )(X2 ) +

^(i) 2 2 2)-1 for Im X I, Im 12> -61,-62

+ g2 (~2)] ' (XI + ~2 - kl
2
with 81 + 622 < (Jmkl)2
Due to the transmission conditions (3.2) the total sum of all nu-
 398

merators of the Sj(l ,I ) is a known function Z(Ii,12). Dividing by

21 2
the known N(11,12,k ) with an appropriate k ~ C+ one arrives at the
"Four-part Wiener-Hopf functional equation"
4 2 2 Z(II,~ 2 )
(3.4) ~ (i + ~ - ~)Pj*(II,~2) :
j:l I 12 122~ N(ll,12,k 2)

holding for a pair of strips of C 2. Here we have

(3.5) %j(11,1 2) :: Pj$(Xl,X 2) :: (F2xQj:F~I~)(Xl,X 2)

It has been shown (e.g. by N.Latz (1968) [ I0]) that in the case of
Im k > 0 the auxiliary k may be chosen in such a way that eq. (3.4) is
3
uniquely solvable in FLP(R2), I < p S 2, for any ~pr(~) 6 LP(R2). The
present author has derived quite recently (1984) [13] a 4X4-system of
integral equations for the Fourier-cosine transforms of the normal
derivatives on the bounding semi-axis's of the four quadrants Qj. This
system is uniquely solvable in the case of Im k. > 0 and Ik - k I and
3 3 v
Ipj- small by Banach's fixed point theorem in the spaces (Lq(R+)) 4
for 2 ~ q < ~, but the general case of four different wave numbers k
3
is still unsolved.

References

[1 ] CARLSON,J.F., A.E.HEINS, The reflection o f electromagnetic waves by an infinite set o f plates, h

Quart.Appl. Math. 4(1946),313-329, Ih 5(1947),82-88, llh 8(1950),281-291.
[2] COLTON,D., R.KREIS, IntegralEquations in Scattering Theory, J.Wiley, New York et al. 1983.
[3] COSTABEL,M., E.STEPHAN, A direct boundary integral equation method for transmission problems,
Journ. Math, Appl. 106(1985), 367-413.
[4] COSTABEL,M., Starke Elliptizitat yon Randintegraloperatoren erster Art., Habil.-srift=preprint
Nr. 868, FB Mathematik TH Darmstadt, Dez. 1984.
[5] GERLACH,E., Zur Theorie einer Kiasse yon Zntegrodifferentialgleichungen, Dissert. TU Berlin 1969.
GRISVARD,P., Boundary value problems in non-smooth domains, Univ. of Maryland, MD 20742
Lecture Notes 19(1980).
[7 ] HEINS,A.E., The radiation and transmission properties o f a pair o f semi-infinite parallel plates, I. Quart.
Appl. Math. 6(1948) 157-166 Ih 215-220.
[8] HEINS,A.E., 7he Sommerfeld Half-Plane Problem Revusited I: The Solution o f apair o f Coupled
Wiener-HopflntegralEquations, Math. Meth. Appl. Sci. 4(1982), 74-90, Ih 5(1983), 14-21.
[9] KREIN,M. G., Integral Equations on the Half-Plane with Kernels Depending upon the Difference o f the
Arguments, Amer. Math. Soc. Transl. 22(1962), 163-288.
[ 10] LATZ,N., Untersuchungen uber ein s~lares Ubergangswertorob~lem ctus den Wh~orie der Beugung
elektromagnetische Wellen an dielektrischen Keilen, Dissert. U ~aaroruc~:en *vo~.
[11 ] LEIS,R., Lectures on initial-boundary value problems in mathematical physics, B.G.Teubner-J.Wiley,
Stuttgart-New York (1985 to appear).
[ 12] MEISTER,E., Some multiple-part Wiener-Hopfproblems in mathematical physics, Banach Center
Public. (to appear 1985) = Preprint No. 600, FB Math. TH Darmstadt, Maa 1981.
[ 13 ] MEISTER,E., Integral equations for the Fourier transformed boundary values for the transmission
problems for right angled wedges and octants, Math. Meth. Appl. Sci. 7.
[14] RAWLINS,A.D., The explicit Wiener-Hopf factorization o f a special matrix, Z. Angew.Math.Mech.
61(1981),527-528.
[ 15 ] SOMMERFELD, A., Mathematische Theorie der Diffrak tion, Math. Ann. 47(1896), 317-374.
[16] SPECK, F.-O., General Wiener-Hopffactorization methods, Res. Notes in Math., vol. 119, Pitman,
Boston et al. 1985
[ 17 ] THELEN-ROSEMANN-NIEDRIG,G., Zur Fredholmtheorie singularer Integro-Differentialoperatoren
aufder Halbachse, Diss. TH Darmstadt 1985.
ENTROPY COMPACTIFICATION
OF THE TRANSONIC FLOW
J. NECAS
Faculty of Mathematics and PAysics, Charles University
Malostranskg ndm. 25, 110 O0 Prague 1, Czechoslovakia

i. I n t r o d u c t i o n
L e t us c o n s i d e r a compressible, irrotational,steady, adiabatic,
isentropic and inviscid fluid in a b o u n d e d , simply connected domain
C R n, n = 2,3, with LiDschitz boundary. The relation between the
presure D and the density p is

(1.1) P-- : (2E-) ~, 1 < ~ < 2,

PO %
where quantities with the zero index correspond to the speed ~ = 0. If
is the v e l o c i t y vector, then the c o n d i t i o n of the i r r o t a t i o n a l flow is

(1.2) rot ~ = 0

The flow satisfies the c o n t i n u i t y equation

(1.3) div (p~) = 0

and the E u l e r equation of m o t i o n :

(1.4) ~ grad ~ = - ~ grad p .

p
This implies for the p o t e n t i a l of the v e l o c i t y is s a t i s f i e d the
equation

(1.5) div (pVu) : 0 ,

where
1
(1.6) p : p(iVul 2) : Po(l _ ~x-I I V u l 2 ) ~ - i
2a 0
and a is the speed of the sound. If the M a c h number defined as

M d~t IVUla is < 1 ~ IVul 2 < ~ ,

the flow is s u b s o n i c and the equation (5) is e l l i p t i c . In the o p p o s i t e

case the flow is s u p e r s o n i c and the equation (5) is h y p e r b o l i c . A flow
with subsonic and supersonic regions is c a l l e d transonic.
It is i m p o r t a n t to u n d e r l i n e that the e q u a t i o n (5) d o e s not contain
an information about the b e h a v i o u r of the e n t r o p y on the shock
surfaces. The entropy condition across the shock: IVui is d e c r e a s i n g .
 400

T h i s c a n be f o r m u l a t e d , for e x a m p l e , in the f o r m

(1.7) Au S K < ~ ;

we shall consider in the n e x t o n l y p h y s i c a l speed, i.e. such that

vga 0
(i .8) IVul

We suppose the b o u n d a r y 0n = F I U F 2 , u = 0 on F 1 and p ( I V u l 2 ) O0v

u = g

o n P2"
The transonic flow problem was considered numerically by many
authors. We mention h e r e a b o o k b y R. G l o w i n s k i [I] ; t h e r e are e x c e l e n t
numerical results by m a n y authors: M.O. Bristeau, R.Glowinski,
J. P e r i a u x , P. P e r r i e e r , O. P i r o n n e a u , G.Poirer, M.Feistauer, A.Jameson,
K.Kozel, J.Pol~ek, M.Vav~incov~. They used entropy conditions of the
type (7), u p w i n d i n g iterations and viscosity approximations. We s h a l l
do the same in the next. The e n t r o p y condition (7) is c o m p a c t i f y i n g ,
which follows f r o m some s l i g h t g e n e r a l i s a t i o n of the r e s u l t b y F . M u r a t
[2]. M o r e c o m p l e t e discussion of the r e s u l t is in M . F e i s t a u e r , J.Ne~as
[3], M . F e i s t a u e r , J.Madel, J.Ne~as [4], J . N e ~ a s [5]. A j u s t i f i c a t i o n
o f the f i n i t e e l e m e n t aDD~o×ima+ion is disollssed in P h . G . C i a r l e t ,
J.Mandel, J.Ne~as [6].

2. F o r m u l a t i o n of the p r o b l e m , compactness bv e n t r o p y
W e look for a w e e k solution to the e a u a t i o n (1.5), i.e. for
u e wl'~(n), such t h a t for v 6 V = {v 6 W I , 2 ( n ) ; v : 0 on FI}

(2.1) Ip(IVul2)VuVvdx = IgvdS, g 6 L~(0n) .

n 0n
If F 1 = 0, w e s u p p o s e JgdS = 0.
an
W e c a n g i v e also D i r i c h l e t data on a part F t C F2, w h e r e FtC {x 6 0n;
(~,$) < 0}. For an i l l u s t r a t i o n , let us c o n s i d e r a paralel flow:
= (u,0,0), u = u(xl). F o r w = a ~ l e t us c o n s i d e r w 6 W 1 ' ((0,I)),
, ~-i ,I/(~-i)
w(l ) = 0, lw' I2 < x--21, s a t i s f y i n g with u(s) = (1 - -~--s)

(2.2) (~(w'2)w') ' = 0 in (0,i) ,

(2. 3) W ' ( 0 ) ~ ( W ' ( 0 ) 2) = A

(2.4) W'' < K, K > 0 .

SO f i r s t A 6 [0,0.57] which is c l e a r f r o m teh Fig. i; the g e n e r a l

solution of (2.2), (2. 3), (2.4) is s k e t c h e d o n the Fig. 2 and is unique,
the c a u c h y data in the o r i g i n e being prescribed.
 401

0.57

0.,91v2 1211
Figure 1

........v. ~ ~ xl Let us m e n t i o n that a u n i q u e n e s s

of the e n t r o p x solution is p r o -
Cauchy 2 / I/ bably not true in m o r e d i m e n s -
data / ions. T h e e x i s t e n c e of the
/
/ solution will follow f r o m some
/ "~ p o s t e r i o r i " conditions
/
g i v e n b y an i d e a l c o m p u t e r .
Figure 2

2.5. Definition. 2a~ 2a~

Let h 6 cl([0,s0]), x-~-f < s 0 < ~-t-f_, h(s) > 0 in (0,s 0] a n d let it
satisfy here the m o n o t o n y condition: h(s) + 2sh'(s) > 0. A t r a n s o n i c
f l o w is c a l l e d h-entropic if ~ e D + (~) : (~ ~ 0)

(2.6) - fh(IVul2)VuV~dx S K f~dx, K 6 R1 .

(2.6') Examples: 2a~

(I) h(s) ~ i, SO < ~-i "

- / V u V ~ d x ~ K I~dx ~ & u S K, M = ~ ,

6a~
(2) h(s) = sp(s), so < ~ ,
M = ~3 : entropy by v i s c o s i t y ,

(3) h(s) = -p(s)in (i - ~ a ~ s),

2a~ 1
SO < ~ TO' ~0 = [ ~ i n (i - ~0) 1(i - 6T 0) ;

M = 1.91 : Hugoniot's entropy,

6a~
(4) h(s) = -p'(s)s, So < ~ i "
natural entropy, M = 2.23. o
2a~
In a f o r m a l way: the m o n o t o n y condition for h and s > ~ is s u f f i c i e n t
and necessary, for the s o l u t i o n satisfies the e n t r o p y condition o n the
shock surface.
 402

2.7. Theorem (F.Murat). Let {G n} be a s e q u e n c e of f u n c t i o n a l s defined

on WI'2(~), G n = G. Let for h 6 D+(Q). <Gn.h) z 0. Then G n - G in
[WI'D(Q)] '. vp > 2.

Idea of the proof: ~ c ~ c ~, ¢ 6 D+(~), ¢(X) : 1 in ~i" Let

h E D(~), supp h C ~i" There is

(2.8) -ithllc([) ¢ ~ h S ~hHc(~) ¢ '

and it follows

(2.9) l(Gn,h) I -< (Gn,¢) Uh~c(~)'

hence G n is a sequence of Radon measures. Let u n be d e f i n e d by

(2.10) ~(Vu Vh + u h ) d x = (Gn,h) , ~h 6 W I ' 2 ( ~ )

n n

n 2 C ~2 C nl, q > n, h q w ~ ' q ( ~ i ) .

Then
l,q .

(2.11) ,,-Au n + u n = G n " 6 tWO (~i)] '

and
(2.12) llUnllwl,q , ~ C(~ 2) .
(a 2 )
Because wl'q(~ )~ ~ C(% ) ~ {u n} is c o n v e r g e n t in w l ' q ' ( ~ 2) to u. An
usual interpolation technic as well as the e s t i m a t e
1

(2.1 3) I \ ~ 2 V(u n- u)Vhdxl ~ ( ~\~2f [V(u n- u) 12dx) 2

1 1
( f IVhl2dx) 2 g ( / I v ( u - u) 12dx) 2
~\ ~2 ~\ a2 n
1 i 1
• (\~21vhlPdx) p I ~ \ a 2 t2 P

gives the result.

2.14 Theorem,
Let Eh= {u;llUlIwi,2(Q ) S C, IVuJ 2 ~ s o ,

v~ 6 D+(~) : - fh(IVul2)VuV~dx S K I~dx]

n
T h e n Eh i s compact in wl~2(a).

Idea of the proof: put (Gn,h) = K fhdx + f h ( I V U n 12 )

• VunVhdx, (G,h) = K fhdx + fh(IVul2)VuVhdx.

 403

It f o l l o w s from the theorem 2.7: <G n - G, u n - u) ~ 0, p r o v i d e d u n ~ u.

For the p a i r i n g (G n- G, u n- u) we use the Leray-Lions trick from the
theory of m o n o t o n e operators, a

3. S o l u t i o n of the t r a n s o n i c problem by use of the a l t e r n a t i n ~

functional
The equation (i .5) is the E u l e r ' s equation to the functional
2
IVul
( 3.1 ) ~(u) = ~1 /( f p(t)dt)dx - fguds.
0 8~
2a~
Let IVul 2 g ,~,Ll, a n d d e f i n e w E WI'2(~) by

(3.2) /p(IVul2)VwVhdx = fghds

for h 6 ~ '2(Q), h = 0, o n F I. (In the c a s e FI= 0, Ft= 0 9 we suppose

Iwdx = 0. )

Define the alternating functional by

( 3. 3) ~(u) = ~(u) - ¢(w(u)) .

Because p'(s) S 0, we h a v e w i t h c > 0 (for d e t a i l s see [ 3] t [4] , [5])

(3.4) clIu - w(u)ll21 2 ~ ~(u).

W ' (~)

So u is a s o l u t i o n of the transonic problem iff ~(u) = 0. B u t if V is

the s p a c e od solutions in w l ' 2 ( e ) t h e n we h a v e

3.4. Theorem
The alternating functional attains on E h N V N {u = u 0 on F t} its
minimum in some point u. If %(u) = 0, t h e n u is a h - e n t r o p i c solution
of t h e t r a n s o n i c problem.

Figure 4 Figure 5
~(u) ¢(u)

on m u s t fall
in the ap-
physical
propriete
[ t I / solution energy hole
v v -

~2fu I _ u 3 u 2 uI u3 u2
in q e n e r a l
a nOn-physical solution
 404

4. Viscosit~ method
Let us consider a complet system of gas: p - pressure, p-density,
T -temperature, ~ - velocity vector, provided:

(4.1) 0 < T 1 ~ T S T 2 < ~, T 6 WI'2(Q),

(4.2) 0 < Pl S p, p in p 6 L2(e),

(4. 3) p : RpT, R is the gas-constant,

2a~
(4.4) ~ e [WI'2(~)] 3 ]v12 <

(4.5) u : ~(T), I = A(T), ~,A q C(R+), A = -82~, 8 < I, FI: 0,

Ft: 0,
(4.6) div(pJ) : 0 in n, pv.v : g on ~ ,

(4.71 pv.Vv + Vp : V(A div ~) + 2 div(~e), in e~

avi avj
e = {eij(v)}, 2eij(v) = ~ + ax i ,

2a 2
(4.8) ~ : T ° on ~ , nv~°~ <_ c , I~°I _< ....
[WI,2(~)] 3 ,-Z '
-+

(4.9) cvdiv(p.ln x--~r-f_.v) = k + in ~

p

E(v) = l(div v) 2 + 2~tree, Cv, k E R +1 •

(4.10) Cvg In T~_----~- k ~1 ~-~

ST : h on ~n~ RhULI(~ ) _< c .
p

Let 8 n > 0, and take ~n = U.Bn, An: lBn, kn= kS n. We let go Bn ~ 0 and
look for an optimal control problem: let the cost functional• where
= e + Vu, be
1
~-i ~ 2)x-i
(4.11) Z(~) = li~12dx + lip - PO ( I - 21vl ]2dx
e ~ 2a 0
and let us look for I(v) - 0, "~ posteriori" entropy condition:
p q WI,I(~),

(4.12) vp.~ ~ -K .

4.13 Remark
fhdS a k :IVTI2 @
~'-'T---T_ dx + I dx .
an

Open problem: how to estimate the pressure p? If I(v) < ~, then

 405

L 2 estimate of p follows.

4.14 Definiton
A sequence {Un}, f U n d X = 0, IVUn 12 < s O is h - entropic, if
~ E D+(e):

(4.1S) - ./h(IVUnl2)VUnV~dx _< K f~dx + (Rn,~)

where

R ~ 0 in [WI'2(£)] I
n

4.16 Theorem
~n
Let T , p , Pn' v be a sequence to solutions of (4.1) - (4.10)
n
with I(~n) n- 0 and satisfying Vpn ~n ~ -K. Suppose, Without the loss
of generality, that u n - u. Then {Un} is h(s) = p(s)s - entropic,
u n ~ u, and u is a h-entropic solution to the transonic problem,
6a~
PDovided IVUn 12 S s o < ~ .
~n
Idea of the proof: Let ~ E D+(~), multiply 44.7) by v ~ and
integrate by parts. We get

(4.17) ~fPnV~ ~~v~V i ~n3 d X = - ~1 S P n i~nl 2 vjn ~%~ dx =

= _ f~Pn n ~n na~
vig~x - /In(div ~n)2~dx - Slndiv v .v _----~x -
%x i ~ Q l~x i

- 2 fUneij(~n)eij(~n)~dx - 2 f~neij(~n)v~ ~0xj

dx S

S K S~dx - fl div ~nvn "4~-2--dx- 2~SUneij(~n)v?i ~-9--dx =

~ n i ~x i ~xj

de_f
- K f~dx + (S ,~)
n

But we nave, because of 4.13

1
(4.18) USnn[wI, 2 ( a ) ] , S eB~ I

and replacing in £fPnl~nl2v nj ~ia~ , Pn by o0(l - ~x-ll~nl2)~-&

and then ~n by Vu , the result follows, s

n

Solving the system 44.1)-(4.10) with the cost functional (4.11),

 406

the i n e q u a l i t y (4.12) can be e x p e c t e d , h u t m u s t be supposed. For to

cancel this c o n d i t i o n , let us f i r s t formulate (4.6) in the w e a k
sense:

(4.19) fpv i ~ dx = fg~dS .

Any approximate solution to (4.1 9) s a t i s f i e s in fact

(4.20) fpv ~--~--dx = fg~dS + (R,~> ,

i ~x i 0~

where the t e r m R r e p r e s e n t s small m a t e r i a l sources in ~ and a s m a l l

f l u x of the m a t e r i a l through the %Q. If w e c h o o s e R in an a p p r o p r i a t e
way, we get a u t o m a t i c a l l y (4.1 5) p r o v i d e d the c o s t f u n c t i o n a l
1
(4.21) J(v) -- fl~12dx + /[ p - P0(l - _ ~ I ~ 1 2 ) ~-I ] 2 d x +
~ 2a 0
TO 2
+ f[ in T in ] dx
P PO
1 PO
tends to zero; here T O = ~ p--~

For to p r e c i s e the c o n d i t i o n s , let B ~ 0, 8 n > 0 X n , P n , k n as b e f o r e

_1 n '

and p u t a n - 0 in the w a y that ~nSn 2 - ~. Let us s u p p o s e T O > p0 -I

Put in (4.20)
1-~
(4.22) <Rn,~) = fpn vn ~~ dx - ;~ n n _ _ x +
± ~ n vi O x i
Tn
in .......
+ /p (i - Pn )v n D~ dx + /f ~dx +
n TO - ~i ~ n
In
~-i
o0

+ (Gn '~> + (Hn'~) f

where

I1 <_ K~n, l---fIG II ~ ~ 0

llfn L ~ ( a ) an n [ W ± , 2 ( ~ ) ] ,

~HnII[WI,2(~)] , ~ 0 and ( H n , ~ ) > 0 v ~ E D+(e).

4.23 T h e o r e m .
Let h(s) = -~(s)in (i - x-! s) (see e x a m p l e s (2.6')) a n d T , Pn'
2a~ n

Pn' ~ n a s e q u e n c e of s o l u t i o n s to ( 4 . 1 ) - ( 4 . 1 0 ) , w i t h J ( v-~n) ~ 0 a n d
 407

with (4.20), (4.22). Let us suppose without the loss of generality

u n ~ u. Suppose JVUn j2 < s o ( s e e (2.6'), (3)). Then {u n} is h-entropic
and u is a h-entropic solution to the transonic problem.

4.24 Remark
The term /pl-~v i @--~--dx gives "upwinding" in the continuity
z
equation.

Idea of the proof to the theorem 4.23: We have from (4.9) as

before

(4.25) 6n ~/ i--
T 2 iVTnJ2dx + 8n /IVvnl 2dx < c <

Take ~ 6 D+(2). It follows from (4.9)

T JVT j2
(4.26) -c / Pn in n v n 0~ dx = + k / ~ n ~dx -
v ~-i i ~--x-~. n
~ Pn l n
VT E (un)
- kn / ~-~ V~dx + f ~ ~dx .
n ~ n
i i
Multiply (4.26) by and add to (4.20).
cv TO
in
P0
We get

(4.27) i__ f (Pn - Pnl-en)vn ~ dx . . 1 . f f.n ~ d. x 1 (Gn,~) _

an i i an an

k IVTn 12 E (~n)
_ 1 (Hn,~) _ n f ~dx - i f n _ ~ ~dx +
an an Q ~T ~ Q Tn
n
k VT
_~n / n
+ an T ~ n V~dx

and the result follows as in the theorem 4.16.

References

[i] R.Glowinski, N u m e r i c a l Methods for Nonlinear Variational P~oblems,

Springer-Verlag, 1984.

J2] F. M u r a t , i~injection du cone p o s i t i f de H-I dans W-1"q est

compacte #our tout q < 2, J. Math. Pures. Appl. (9) 60 (1981),
 408

309-322.

[3] Mo Feistauer, J. Ne~as, On t h e S o l v a b i l i t y of T r a n s o n i c P o t e n t i a l

Flow Problems, Zeitschrift f~r Analysis und ihre Anwendungen,
Bd. 4 (4) 1985, 305-329.

[4] M. Feistauer, J. Mandel, J. Ne~as, Entropy Regularization of the

Transoni6 P o t e n t i a l Flow Problem, Comm. Math. Univ. Carol. 25
(~ 1984, 431-443.

[5] J. Ne~as, Compacit~ par e n t r o p i e d ' ~ c o u l e m e n t s des f l u l d e s ~ U n i v .

Pierre et Marie Curie, Paris VI, 1985.

[6] Ph.G. Ciarlet, J. Mandel, J. Ne~as, On t h e convergence of F i n i t e

Element Approximations of t h r Transonic P o t e n t i a l Flow Problem,
to appear.
THE GLOBAL EXISTENCE OF WEAK
SOLUTIONS OF THE MOLLIFIED SYSTEM
OF EQUATIONS OF MOTION OF VISCOUS
COMPRESSIBLE FLUID
J. NEUSTUPA
Faculty of Mechanical Engineering, Czech Technical University
Suchbdtarova 4, 166 07 Prague ~ Czechoslovakia
I. I n t r o d u c t i o n

It is k n o w n that weak solutions of the N a v i e r - S t o k e s eauations

for incompressible liquid exist on a t i m e interval of an a r b i t r a r y
lenght (see e.g. [ 3] [ I0]). No a n a l o g o u s result has been deri-
ved in t h e c a s e of e q u a t i o n s of m o t i o n of v i s c o u s compressible fluid
till now. Only the existence of solutions of s u c h equations local in
time w a s proved (see e.g. [ I] , [5], [8], [9] ) and if some
theorems about the global in time existence of s o l u t i o n s appeared,
they contained assumptions of the type "the initial conditions are
small enough" (see e.g. [4]), "the flow is o n e - d i m e n s i o n a l " ([2]),
etc. We study the e x i s t e n c e of w e a k solutions of t h e e q u a t i o n s of
motion of v i s c o u s compressible fluid on a time interval of a g i v e n
lenght in this paper, but the system of e q u a t i o n s we deal with is
rather modified in a c o m p a r i s o n with a full general system of e q u a -
tions governing the motion of v i s c o u s compressible fluid. The modi-
fication consists in t h e following points:

a) W e assume the d y n a m i c viscosity coefficient p to be a p o s i t i v e

constant.
b) W e do n o t take the e n e r g y equation into account a n d we u s e the
relation between the pressure p and the d e n s i t y

(1.1) o = c.p

instead of it. c and x are constants such that c > 0, ~ 6 (1~6).

The tilda over p~ r e p r e s e n t s a certain regularization (mollifi-
cation). Its e x a c t meaning is e x p l a i n e d in t h e p a r a g r a p h 2., but
we c a n w r i t e in a d v a n c e that ~(x) is an a v e r a g e of p~ c o n s i d e -
red with a proper smooth weight function on a n e i g h b o u r h o o d Bh(X)
of x (where the r a d i u s h of this neiqhbourhood m a y be a r b i t r a r i -
ly small).
c) We u s e the m o l l i f i c a t i o n denoted bv ~ a l s o in some terms in the
Navier-Stokes eauations for the system we d e a l with has the form
 410

(1.2) p,t + (pu.) = 0 ,

] ,J

( i . 3) (Oui),t + (Oujui),j = - e.( ),i + "3 U u j , j i + ~uitjj

(i = 1,2,3).

U = ( U l , U 2 , U 3) has a physical meaning of the v e l o c i t y of the m o v i n g

fluid. In [7], R. R a u t m a n n used the similar mollification in the
Navier-Stokes equations for the incompressible liquid in o r d e r to
prove the g l o b a l in t i m e existence of s t r o n g solutions in t h r e e - d i -
mensional domains. The notion of the v e l o c i t v of the fluid at the
point x is u s u a l l y introduced by m e a n s of an a v e r a g e of the v e l o c i -
ties of all particles of t h e fluid contained in a s m a l l neighbour-
hood of x. So if h is s m a l l enough, u. is a l m o s t the s a m e as u
l 1
from the p o i n t of v i e w of m e c h a n i c s . T h e s y s t e m (i. 3) e x p r e s s e s the
2nd N e w t o n law of m e c h a n i c s applied to p a r t i c l e s moving along the
integral curves of t h e flow field U.
We shall use the Rothe method. We can give only a brief outli-
ne of t h e w h o l e procedure here. Details may be f o u n d in [6] .

2. F o r m u l a t i o n of an i n i t i a l - b o u n d a r y value proble ~

Assume that ~ is a b o u n d e d region in R 3 w i t h the boundary of

the c l a s s C 2+(a) for some a 6 (0.i). L e t us c h o o s e h > 0 and put

~h : {x • R3; dist(x,~) < h} .

A s s u m e t h a t h c a n be c h o s e n so s m a l l that 5~ h is a l s o of the class

C 2+(~) . P u t

I~I 2
Wh(~) = K h exp (- h2_l~l 2 ) for ~ • R 3, I~I < h ,

~h(~) = 0 for ~ 6 R 3, I~I a h .

L e t K h be c h o s e n so t h a t the integral of ~h o v e r R 3 is e q u a l to i.
If f • L l ( ~ h ) , put

(2.1) f(x) = ] ~h(X - y)f(y)dy .

~h
If f is d e f i n e d in ~h X R 1 t h e n we d e n o t e by f the function regu-
larized in t h e s p a c e variable only. If the regularization ~ is
applied to a n y function def in the space variable on ~ o n l y (like for
example components of the v e l o c i t y or their approximations), we d e a l
 411

w i t h thi~ function as .if it is d e f i n e d on ~h and is i d e n t i c a l l y

equal to zero on Zh - ~"
We shall solve the e q u a t i o n (1.2) on ~h X (0,T) and the s y s t e m
(i. 3) o n ~ X (0,T) ( w h e r e T is a g i v e n p o s i t i v e number). We consider
the b o u n d a r y condition

(2.2) u i 8~ ~ 0 (i : 1,2,3)

and the i n i t i a l conditions

(2.3) Plt : 0 : P0 '

(2.4) (PUi)It : 0 : P0U0i (i : 1,2,3) ,

where P0' U0 : (u01,u02,u03) are g i v e n functions such that

P0 • H I ( Q h ) ' P0 ~ 0, U 0 • ~ 1 ( ~ ) 3 .

We s h a l l call by the w e a k solution of (1.2), (i. 3), ( 2 . 2 ) , (2. 3),

(2.4) the c o u p l e of f u n c t i o n s U,p such that

U ~ ( U l , U 2 , U 3) • L 2 ( 0 , T ; ~ I ( ~ ) 3 ) ,
(2.5)
p • L(0,T;HI(~h)),~ p ~ 0 ,

T
1
(2.6) f /{PUi~i, t + pujUi~ i j + C ( p ~ ) ~ i , i ~uj,j~i, i -
0 Q

- Uui,j~ i ' j } d x d t : - / ~u0i(~i t : 0 )dx

for all ~ - (~I,~2,~3) • C~(~ X (0,T)) 3 such t h a t ~i[SQ 0

~i Tt [: ------0 (i : 1,2,3)~

T
(2.7) f f{P¢,t + puj¢,j}dxdt : - f p0(¢]t : 0)dx
0 ~h ~h
for all ¢ E C (~h X (0,T)) such that ¢[t = T ~ 0.

By m e a n s of a s i m i l a r m e t h o d as it is u s e d in [ i] in the c a s e
of the N a v i e r - S t o k e s equations for the incompressible liquid, it c a n
be p r o v e d t h a t if U,p s a t i s f y (2.5), (2.6)~ (2.7) then p.U is a.e.
in (0,T) equal to a c o n t i n u o u s function f r o m (0,T) into H - I ( Q ) 3 .
H e n c e we c a n u n d e r s t a n d under (PUi)[t:0 (i:i,2,3) in ~ . 4 ) limits
as t - 0+ of the c o m p o n e n t s of this function. Similarly, it m a y be
s h o w n that p is a.e. in <0,T) equal to a c o n t i n u o u s function from
(0,T) into H l ( ~ h ) * (the dual of H I ( Q h ) ) . It g i v e s a reasonable sence
to the initial condition (2. 3).
 412

3. The time discretization

Let m be a natural number. Put T = T/m, t k = k.T (k = -1,0,1,...

...,m). Denote p ( 4 ) = P0' ui(0) = u0i (i = 1,2,3) and let p(k), u(k) =
= (u[k) (k) ~k)
, u2 ~ u ) denote an approximation of a solution on k-th time
layer. A discrete version of (2.5), (2.6) and (2,7), which we use in
the followinq, is: We look for p(0) p(1) p(m) q Hl(~h)
p (k) >
- 0 (k = 0,1, • ..,m) and U (I) ,. ..,U (m) 6 ~I(~) 3 so that

(3.1)k i{p(k-l) (k). (k-2) (k-l)%. (k-l)~ ( k ~ ) u ~ k - l ) #

O ui ~i - p ui l - Tp J i,j

- ~uj,j i,i + T~ui~j i,j ~ux = 0

for all # ~ (~i,~2,~3) 6 ~ ( ~ ) 3 and k = l,...,m ,

( 3.2)k p(k) _ p(k-l) + T(p (k) Uj

qk),J,j = 0

for k = 0,1,...,m.
We can further proceed in such a w a v that we successively solve
(3.2)_ (for the unknown p(0)), (3.1) 1 and (3.2) 1 (for the unknowns
(i),~(i)), . (m)
U ..., ( 3.1 )m and (3.2) m (for the unknowns u , p(m)). It
can be done using standart methods of the functional analysis and the
theory of the partial differential equations. The following inequal-
ities may be also derived:

( 3. 3) I[~P'I( k ~ ) u ( k ) u ( k ) + i ~ p(s-2)(u(S)_u!S-l))(u~S)_u .(s-I ))+

i i ~ s =i i 1 1
k
1 ~ . (s) ~ (s) (s). I p(k)~dx
+ ~T~S~I[Uj,j)2+_ TUS:ILU~_,~ ui'j]dx + ~ ~h

C
~ P0U0iU0i dx + ~ I P0dx (k = l,...,m) ,
Qh
(3.4) g (k)a2. k
P Hl(~h ) + s=0~~ p(S) - p(S-l)~2Hl(~h) _<

1
K 1 exp(4~NU011L2(~ ) 3 + / ~P0u0iu0idx +

+ -c ~ pXdx) Ilp01121 (k : 0,i, ..,m)

x-i ~h 0 " H (~h)

for an appropriate positive constant KI, independent on k.

4. An approximate solution of (2.6) t (2.7) and the limit

process for m - +~

Put
 413

(4.1) mp(t) = p(k) for t e (tk,tk+l > (k : -i,0,1~...,m-l) ,

(4.2) mu(t) = U (k+l) for t 6 (tk,tk+l> (k : 0,1,...,m-l) .

It follows from (3.3) and (3.4) that the sequence {mp] (resp.
{mu}) is uniformly bounded in L~(O,T; Hl(~h )) (resp. in the spa-
ce L2(0,T; hl(Q) 3)) and that {mpjmul2} is uniformly bounded in
L~(0,T; LI(~)). Usinq the Holder inequality, it can be also easi-
ly shown that {mpmu} is uniformly bounded in L°°(0,T) LI2/7(Q) 3) and
in L2(0,T; W3/2i (~)3). There exist subsequences (denoted by
{mp}, { m u } again) and functions p, U so that mp _ p weakly _ *
in L~(0,T; HI(Qh)), mu ~ U weakly in L2(0,T; HI(~)3), mp mU
p U weakly - * in L~(0,T; LI2/7(Q) 3) and weakly in the sDace
L2(0,T; W 1 (Q)3) Bv means of other estimates of m mU m
3/21 i p and p
in ~ (0.T~ W3/2(~ 3, H-I(~)3) and ~ ( 0 , T ; Hl(~h ), L2(~h )) (see
e.g. [3] or [ i0] for the definition of these spaces), we can Drove
that even mp p stronqlv in L2(0,T; L2(~h )) and m m U
p U stronqlv in L2(0.T~ L2(Q)3).
The functions mp, m U satisfy (2.6), resp. (2,7) with some
errors E., resp. E 2. It is shown in [6] that E 1 : 0(T I/2 ) and E 2 :
: 0(~i/2~ for T -- 0+ (i.e. m ~ +~ ). These relations toqether with
the tYPeS of converqences mentioned above are sufficient to Drove
that ~, U satisfy (2.5)~ (2.6)~ (2.7).
If we use ( 3. 3) and ( 3.4), we can also derive the estimate

(4. 3) lJp~ ~(0,T; Hl(~h) ) N K 1 exD ( f ~1 P0U0iu0idx +

c f 4 dx)'JlP0112
+ ~ Qh Hl(~h)

and the energy inequality

1 c
(4.4) f PUiUi}t=tl dx + ~-/-[ ~hf ~It:tldX +

tz )2
+ f f{~(uj + dxdt <
0 ~ '3 Uui'jui'j} -

1 c
<- f ~P0U0iU0i dx + ~ / P0 dx
Qh

(@or every t I E <0,T)),

While the estimate (4.3) depends on the parameter h (used in
 414

the r e q u l a r i z a t i o n in (1.2) and (i. 3)) a c c o r d i n q to the d e p e n d a n c e

of K 1 on h, the energy inequality (~.4) is q u i t e independent on h.
But in s p i t e of this fact. we are n o t able to D r o v e that if h ~ 0+,
we can g e t a s o l u t i o n of (1.2)~ (i. 3) w i t h o u t the m o l l i f i c a t i o n yet.

References

[i] ITAYA,N., On t h e Cauchy problem f o r t h e s y s t e m of f u n d a m e n t a l

e q u a t i o n s d e s c r i b i n g t h e movement of c o m p r e s s i b l e v i s c o u s
f l u i d s , K o d a i Math., Sem. Rep. 23, 1971, 60-120.
[2] KAZHIKOV,A.V., SHELUKIN,V.V., Unique g l o b a l s o l u t i o n i n t i m e of
initial-boundary v a l u e problems f o r o n e - d i m e n s i o n a l e q u a t i o n s
o f a v i s c o u s g a s , Prikl° Math. Mech. 41, 1977, 282-291.
[3] L I O N S , J . L . , Quelques m~thodes de r ~ s o l u t i o n des probl~mes aux
l i m i t e s non l i n ~ a i r e s . Dunod° P a r i s . 1969.
14] MATSUMURA.A.. N I S H I D A , T . . I n i t i a l - b o u n d a r y v a l u e problems ~ o r
t h e e o u a t i o n s o~ m o t i o n o~ c o m p r e s s i b l e v i s c o u s and h e a t - c o n -
d u c t i v e f l u i d s , C o m m on Math. P h y s i c s 89. 1983, 445-464.
[5] N A ~ H . J . . L e ~robl~me de Cauehu Dour l e s ~ o u a t i o n s d i f f . d'un
f l u i d e g ~ n e r a l , Bull. Soc. Math. F r a n c e 90, 1962, 487-497.
[61 N E U S T U P A , J . , The g l o b a l weak s o l v a b i l i t y of an i n i t i a l - b o u n d a r y
v a l u e problem of t h e N a v i e r - S t o k e s t y p e f o r t h e c o m p r e s s i b l e
f l u i d , to appear.
[7] RAUTMANN,R., The u n i q u e n e s s and r e g u l a r i t y of t h e s o l u t i o n s of
N a v i e r - S t o k e s p r o b l e m s , Funct. Theor. Meth. for PDR, Proc.
conf. Darmstadt 1976, Lecture N. in M a t h . , 561, 1976, 378-393.
[8] SOLONNIKOV,V.A., S o l v a b i l i t y of t h e i n i t i a l boundar~ v a l u e ~ro-
blem f o r t h e e q u a t i o n s of m o t i o n of a v i s c o u s c o m p r e s s i b l e
f l u i d , J. S o v i e t Math. 14, 1980, 1120-1133.
[9] T A N I , A . , On t h e f i r s t initial-boundary v a l u e problem o f
c o m p r e s s i b l e v i s c o u s f l u i d m o t i o n , P u b l . RZMS K y o t o U n i v . 13,
1977, 193-253.
[10] TEMAM,R., Navier-Stokes equations, N o r t h - H o l l a n d Publ. Comp.,
Amsterdam - New York - Oxford, 1979.
BIFURCATIONS N E A R A DOUBLE
EIGENVALUE OF THE RECTANGULAR
PLATE PROBLEM WITH A DOMAIN
PARAMETER
Z. SADOVSK'f
Institute of Construction and Architecture of thz Slovak A cademy of Sciences
D(ebravskd cesta, 842 20 Bratislava, Czechosh)vakia

Let us consider the b i f u r c a t i o n problem of the F ~ p p l - K ~ r m ~ n equa-

t i o n s of a t h i n elastic rectangular plate having the length a and w i d t h
b when the aspect ratio ~:a/b varies near a value e:~ yielding a
c
double buckling load of the plate. In a s u i t a b l e non-dimensional for-
mulation, the governing equations will refer to a c o m m o n domain. On
the o t h e r hand, there appears a small perturbation parameter say O in
the e q u a t i o n s which we introduce as ( M a t k o w s k y et al. 1980 [ 1 ])

O= i__ _ i_ (i)
2 2 "

Starting from a sample boundary conditions we d e f i n e a variational

solution to the b o u n d a r y value problem using energy spaces H, V g i v e n
as c e r t a i n subspaces of the S o b o l e v space W~(f~) . An introduction of
suitable equivalent norms in H,V leads to the o p e r a t o r equations

w %w+~2M2w-~(~-~) Lw- < ~ ~)c<w,~) = 0 <2a)

~C c

- ~ + OAI~ - @ 2 A 2 ~ _ i ( ~ _ 8) B ( w , w ) = 0 , (2b)
2 ac

where I is t h e load parameter and weH, ~V refer to the p l a t e deflec-

tion and A i r y stress function, respectively. The operators L,B,C are
essentially those introduced by B e r g e r 1967 [2] for a p l a t e with a
definite domain. In a d d i t i o n , we h a v e obtained linear bounded and self-
adjoint operators MI, M 2 a n d AI, A 2 acting from H into itself and from
V into itself, respectively. Eq.(2b) can be u n i q u e l y solved for ~ =
~e(@,w). If @ is s u f f i c i e n t l y small, ~:~ m a y be e a s i l y found in the form
of p o w e r series in @. Substituting this solution into (2a) for @ and
introducing the small load p a r a m e t e r x,

c
i = (3)
gC
 416

where Ic c o r r e s p o n d s to the d o u b l e buckling load of t h e p l a t e , we ar-

rive at the r e s u l t i n g equation

F(w, ×,@) = w - ~M I - XC(I+ X)( i _ 8) Lw

c
+ 1 C(w,B(w,w))+ h.o.t.= 0 (4)
2~ 4
c
to be s o l v e d for [w,x]@ ] n e a r the o r i g i n of the space HXRXR. Obviously,
w=w*(×,@) ~ 0 is a l w a y s a solution to Eq. (4).
Let H c be t h e e i g e n s u b s p a c e of the d o u b l e eigenvalue Ic and Pc the
operator of the o r t h o g o n a l projection of H o n t o H c. We a s s u m e that the
following hypotheses hold:

(HI) For U~Hc, B u,u = 0 only if u = 0.

(H2) The operator PcMI restricted to ueH c has only simple eigenvalues.

(H3) F(W,x,8 ) commutes with respect to the g r o u p

S = {I,S,-I,-S}

of o p e r a t o r s on H, w h e r e S possesses the action of one of the

operators Sx, Sy, Sxy

Sx: H ~ H, u(x,y) ~ (SxU)(x,y) = u(l-x,y)

Sy: H ~ H, u(x,y) ~ (.SyU)(x,y) = u(x,l-y)

Sxy ~ SxSy ~ SyS x

and H c is s p a n n e d by ~ l e H + , ~2eH ; II~il! = i, i = 1,2 w h e r e

H+ ~ {u~H: Su~u], H- ~ {u~H: Su=-u}.

The hypothesis (HI) is c o m m o n l y used.(H2) implies a transversal split-

t i n g of t h e d o u b l e eigenvalue I c into simple ones appearing as a re-
sult of p e r t u r b a t i o n . Actually we can show that near ~ = ~ the e i g e n -
values 18i of the linearized equation are of the f o r m iAS" = ~c + ~i~ +
+ h.o.t., i=I,2 with X{~ ~. By the h y p o t h e s i s (H3) r e s t r i c t i o n s upon
the b o u n d a r y and load conditions are imposed allowing for the occurence
of the assumed symmetry of Eq. (4).
The study of Eq. d4) is c o n s t r u c t i v e - via the L i a p u n o v - S c h m i d t reduc-
tion and the implicit function theorem. Assuming

w = ~i~i + ~2~2 • ~ (5)

with ~ eR~ ~E H c and following Vanderbauwhede 1982 [3] we obtain
that the b i f u r c a t i o n equations admit the form

G i (~,X,~) = ~i H i ( < ' X ' ~ ) = 0, i=i,2 6 11

with
 417

Hi( ±<i' ± ~2 '×'0) = Hi( ~i' C2 '×'8]' i=1,2. (7)

Thus, a Z 2 e Z2 symmetry of Eqs. (6) is p r e s e n t . Moreover, we may dis-

tinguish one-mode and coupled-mode solutions to Eq. {4). T h e o n e - m o d e
solutions w e H @ or w ~ H - correspond to s o l u t i o n s of

~'2 = 0, HI(~I,0,X,@) = 0 (8)

and
~i = 0, H2(0,¢9,X,8) = 0, (,9)
respectively, while coupled-mode solutions correspond to s o l u t i o n s of

HI( ~,X,O) = 0, H2 (<,X,O) = 0. (i0)

Departing from Chow-Hale 1982 [4] , we e s t i m a t e the small solutions

to Eqs. 6 by an a p r i o r i b o u n d b a s e d on (HI) and then s c a l e E q s . ( 6 )
((8-10))by

= B~, X = 2 sinv , @ = u2 cos~ , (ii)

where 8:~(v,~), the a n g l e determines a direction in the (×,@) p a r a -

meter plane and u is a n e w small p a r a m e t e r . The scaled equations are

gi ( 8,~,~) ~ 8ihi ( ~ , % ~ ) = 0, i-1,2 (12)

where
hl( S,v,u) -sin ~ + I cos v + a8~ * b ~2 + h.o.t.
= -- 2 '
c
l'
h2( B,v,U) = -sin v + 7--2 cosy + b~ + cS~ + h.o.t. J

'C
Due to (HI) it is a > 0, c > 0 and we a s s u m e that b > 0, too.
Definition I. (Golubitsky-Schaeffer 1979 [5] ) The b i f u r c a t i o n pro-
blem (6) is n o n - d e g e n e r a t e if b/a ~ I, b/c ~ 1 and b 2 ~ ac.
According to G o l u b i t s k y - S c h a e f f e r [5] , the r a t i o s b/a, b/c r e p r e -
sent the m o d a l p a r a m e t e r s of b i f u r c a t i o n problem (6). The lines of de-
neracy divide the s t u d i e d positive quadrant of b/a, b / c p l a n e into s i x
regions within each of w h i c h the local features of b i f u r c a t i n g solutions
to E q s . ( 6 ) are t o p o l o g i c a l l y equivalent. The n o n d e g e n e r a t e cases of
Eqs.(6) were analysed in [5] by m e a n s of the s i n g u l a r i t y theory. Our
study comprises the d e g e n e r a t e cases while the e m p l o y e d tools are sim-
pler.
Letting Z ~ 0, the r e d u c e d equations g( 8,v,0) = 0 the s c a l e d e q u i v a -
l e n t s of (8-10))can be e a s i l y solved for 8=8o. If b 2 ~ ac, the n o n - t r i -
vial solutions appear at v v a l u e s forming an o p e n subinterval within
the c o n s i d e r e d interval (-~/2, 3~/2) of ~ values, with end points dif-
fering f r o m -~/2, 3~/2. If b 2 = ac, the r e d u c e d s y s t e m is s o l v a b l e only
at a c e r t a i n ~e(-~/2, 3~/2), say v =v 9, h a v i n g then a continuum of so-
 418

lutions g i v e n by the e q u a t i o n

o 2 o 2 s X~ s
a ( 81) + b( B2 ) = sin ~ Ic cosu (13)

Successive continuations of the s o l u t i o n s to ~ ~ 0 by the i m p l i c i t

function theorem s u c c e d Qnly w i t h i n the open set of r e g u l a r points in
the p a r a m e t e r plane. Thus, the d e s c r i p t i o n of the s o l u t i o n set n e e d s
to be c o m p l e t e d in a b e i g h b o u r h o o d of the s i n g u l a r p o i n t s {potential
bifurcation curves). The f i r s t step in the a n a l y s i s which appears to
be c r u c i a l is the c h o i c e of (~l,e) or (~2,@) or (~i' ~2 ) as a new pa-
rameter plane if (8)f(9) or (i0), r e s p e c t i v e l y , is solved. In this w a y
we o b t a i n t h a t the small solutions to Eq. (4) form a c o n n e c t e d set con-
sisting of the trivial, one-mode and c o u p l e d - m o d e solution subsets.
The d e s c r i p t i o n of the s o l u t i o n set to E q . ( 4 ) yields immediatelypri-
mary and s e c o n d a r y bifurcation curves at c r o s s - s e c t i o n s of t r i v i a l and
one-mode or o n e - m o d e and c o u p l e d - m o d e solution subsets, respectively.
N o w the q u e s t i o n on the p o s s i b l e additional bifurcation curves and the
conditions of t h e i r appearing or n o n - e x i s t e n c e is to be a n s w e r e d .
Theorem 2. S u p p o s e (HI-3) hold and the c o e f f i c i e n t s a, b, c of bi-
furcation equations (6) of Eq. (4) satisfy:

(i) b 2 @ ac, b >0

(ii) b 2 = ac, b >0

In the case (ii) let f u r t h e r be

lim T(8°) ' @O ~ 0 N lim T(8°), ~ 0,

o o
81 ~0 82 ~ 0
where T(~ °) is d e f i n e d by

T ( ~ c ) = b h l ' ~ u ~ ° ' v s ' 0 ) - ah2 , uu ~o ' s , 0 )

o v e r the e l l i p s e (13)(the circles in the s u b s c r i p t positions de-

note a t o t a l d i f f e r e n t i a t i o n of T ( 8 0 ) w i t h respect to ~i or B2 ]
and e i t h e r
(a) it h o l d s
o
a 81 o o
T(ff°), Q = - _ __ T(8°),@=~ O, ¥ ~o : 81> 0, 82 > 0
b 8o
2
or
(b) for c e r t a i n it is

T(80~),@= T(BO~), (2~= 0,

T(,B°*), ~ = (~ --~&)
8i 2
T(BO*), ~ 0
~2
 419

and
lira T(~ °) ~ lira T(Bo).
o o
BI --0 82 ~0

Then near the origin of the(x~8) parameter plane, the b i f u r c a t i o n dia-

gram of Eq.(4) consists of four distinct b i f u r c a t i o n curves: two pri-
mary and two secondary b i f u r c a t i o n curves, and in addition in the case
(ii)(b) of a unique curve of limit points. A crossing of the primary
and secondary b i f u r c a t i o n curves changes the number of solutions to
Eq.(4) by two and four, respectively. A crossing of the limit-point
curve changes the number of solutions by eight.
Proof. In order to study the set of b i f u r c a t i o n curves, we solve the
system consisting of the scaled b i f u r c a t i o n equations (12) together
w i t h the condition of v a n i s h i n g of the c o r r e s p o n d i n g J a c o b i a n
Jg = Jg(8,v,~). At solutions to the reduced system (~= 0 the value of
the c o r r e s p o n d i n g Jacobian is always zero but one of Eqs. (12>, say
gl = O, and the equation Jg = 0 may be uniquely solved for B1 ~ % ( B 2 , ~ ) ,
= ~(B2,u)near such solution. Substituting B,l~ for 81,~ in the re-
m a i n i n g equation g2 = 0 we get an e q u a t i o n the small solutions ~=B~(~]
of w h i c h can be studied by Newton's polygon method. The p r i m a r y and
secondary b i f u r c a t i o n curves c o r r e s p o n d to the tripple roQts while the
limit-point curve to the simple root of the remaining equation (gl = 0
or g2 = 0). We note that T(B°,)is a p o l y n o m i a l of second degree in
(B~)2 and (~B~) 2.
C 0 r o l l a r ~ 3. There exist two pairs of one-mode (one from H t and the
other from H-) and in the cases (i), (ii)(a) no or two pairs while in
the case ii b no, two or four pairs of c o u p l e d - m o d e solutions to
Eq. (4), 8=0 near w=0 and any l> lc sufficiently close to ic.
Equation F ( w , x , 0 ) = 0 describes an important p r o b l e m of plate having
a double buckling load. A direct calculation of the b u c k l e d states of
the plate bifurcating at the buckling load may be p e r f o r m e d e l i m i n a t i n g
one of the unknowns and then applying the Newton p o l y g o n method to the
remaining equation.
A necessary part of the analysis of the studied b i f u r c a t i o n p r o b l e m
is the i n v e s t i g a t i o n of stability of b i f u r c a t i n g solutions. Following
the concept of linearized stability, a solution w = w ~ ( ~ , ~ ) to Eq. (4) is
stable if the e i g e n v a l u e s p of the e i g e n v a l u e problem

F'(w ~, X,~ )~ -pq, = 0 (14)

are positive. It is well known that the trivial solution is stable at
any 8 and I> 0 less than the first p o s i t i v e eigenvalue I 8+ of the li-
nearized equation. For A> X~ the trivial solution is always unstable.
 420

M c L e o d and S a t t i n g e r 1973 [61 showed that the L i a p u n o v - S c h m i d t reduc-

tion contains informations required for the stability analysis of bifur-
cating solutions. Later Sattinger 1979 [7] has shown that the stabi-
lity of a o n e - p a r a m e t e r family of b i f u r c a t i n g solutions is determined,
to the lowest order, by the eigenvalues of the Jacobian m a t r i x of redu-
ced bifurcation equations. Sattinger's theorem fails e.g . if b 2 ~ ac,
since then the Jacobian m a t r i x of the reduced equations has always one
zero eigenvalue. The n o n - c o n f o r m i n g degenerate cases can be treated by
the following theorem:
T h e o r e m 4. Suppose (HI-3) hold and the coefficient b of bifurcation
equations (6) of Eq. (4) satisfies b> 0. Then, the linearized stability
of any o n e - p a r a m e t e r family of isolated solutions to Eq. <4) bifurcating
at the double eigenvalue kc is s u f f i c i e n t l y close to the b i f u r c a t i o n
point d e t e r m i n e d by the J a c o b i a n of b i f u r c a t i o n equations (6)(positive
value implies stability>.
Proof~ F ' ( w ~ , × , 8 ~ is an analytic and symmetric p e r t u r b a t i o n of the
operator F'(0,0,0>. The spectrum of F ' ( 0 , 0 , 0 ) is discrete and non-nega-
tive with zero as a double eigenvalue having a p o s i t i v e isolation dis-
tance. A p p l y i n g the L i a p u n o v - S c h m i d t reduction to Eq. (14> we arrive at
an e i g e n v a l u e p r o b l e m in R 2 yielding the p e r t u r b a t i o n of zero eigenva-
lue. Now if we consider the e i g e n v a l u e p r o b l e m for the Jacobian m a t r i x
of b i f u r c a t i o n equations (6"> evaluated at w=w ~ we see that this equa-
tion differs from the former one only in the h i g h e r - o r d e r terms. Justi-
fying the p e r t u r b a t i o n technique in both cases and comparing the pertur-
bation equations we conclude the assertion.
Let us note that the stability analysis sometimes fails to indicate
the e n e r g e t i c a l l y p r e f e r r e d e q u i l i b r i u m path of the plate and direct
comparison of energy levels of buckled states is necessary. Such situa=
tion is encountered if ~ 0 , b>a, h>c, since then at I = i c there bifur-
cate two different pairs of stable solutions to Eq. <4 >.

References
i. B.J. Matkowsky, L.J. Putnick and E.L. Reiss,"Secondary states of rectangular pla-
~es," SIAM J.Appl.Math. 38(19801, 38-51.
2. M.q. Berger,"0n von K/rm/n's equations and the buckling of a thin elastic plate,
I, The clanped plate", Cc~n.Pure Appl. Math, 2~i1967~ ~687-719.
3. A. Vanderbauwhede, "Local Bifurcation and Symmetry," Pitman, London, 1982.
4. S.N. Chow and J.K. Hale, "Methods of Bifurcation Theory", Springer-Verlag,
Berlin, 1982.
5. D. Schaeffer and M. Golubitsky, "Boundary conditions and mode jumping in the buck-
ling of a rectangular plate", Ccmm.Math.Phys. 69(1979), 209-230.
6. J.B. McLef~ and D.H. Sattinger,"Loss of stability and bifurcation at a double eigen-
value", J. Funct. Anal. 14 1973 , 62-84.
7. D.H. Sattinger, "Group Theoretic Methods in Bifurcation Theory", Lecture Notes in
Math., 762, Springer-Verlag, Berlin, 1979.
DELAY MAKES PROBLEMS
IN POPULA ON MODELLING
K. SMiTALOVA
Department oJ'Applied Mathematics, Comenius University
M[ynskd dotina, 842 15 Bratislava, Czechoslovakia

T h e basic population model x' = ax, although it is very simple,

plays an important role in the h i s t o r y of modelling. Contrary to mathe-
matical models in physics, it is not d e s i g n e d to d e t e r m i n e quantities.
Here the m a t h e m t i c s only exhibits a tendency. The p u r p o s e of such a
model is not prediction, but insight [2].
The same is true for m o s t population models. Important are the qua-
litative properties, such as s t a b i l i t y of e q u i l i b r i u m states, existence
of periodic solutions, etc.

Intuitively it is clear that the h i s t o r y of d e v e l o p m e n t of a m o d e ~

led b i o l o g i c a l system is very important for his state in the future.
Hence the use of d e l a y e d differential equations seems to be appropriate.
Moreover, using the d e l a y one can d e s c r i b e age and also the spatial
structure of the p o p u l a t i o n [3]. Generally speaking, a model involving
the d e l a y allows more complex types of b e h a v i o r than models without
delay.

By m o d e l l i n g usually the f o l l o w i n g assumptions are made:

i) D i f f e r e n t histories determine different solutions.

2) Suitable choice of the h i s t o r y allows the s o l u t i o n to a t t a i n

any p r e s c r i b e d value at a g i v e n time.

The first p r o b l e m is a p r o b l e m of i n j e c t i v i t y of the shift opera-

tor (or s o l u t i o n map) [ i]. Take e.g. the e q u a t i o n

x'(t) = -ax(t - i)[i - x(t)]

which arises in a n a t u r a l w a y from a m o d e l

y'(t) = a(y - y2)

of s e l f - l i m i t e d population. Then for every initial function ~ satis-

fying the c o n d i t i o n ~(0) = i we have x(t) ~ I for t ~ 0.
 422

Let C(-l,O) be the space of c o n t i n u o u s mappings from [-I,0] to

the set R of r e a l numbers. For the equation

x'(t) : f(x(t - 1)) (i)

define an o p e r a t o r T : C(-l,O) ~ C(-1,0) by

T(~)(t) = x (t + 1)~ t • [-i,0]

where x~ d e n o t e s the solution of (1) d e t e r m i n e d by the initial function

@ C(-I,0). Such an o p e r a t o r T is c a l l e d shift operator or solution
map for t h e equation (1 >. The above quoted example exhibits an e q u a t i o n
for w h i c h the shift operator is not injective. The following result sta-
tes that this is a c t u a l l y the typical case for e q u a t i o n (i).

Theorem i. L e t C be the m e t r i c space of c o n t i n u o u s mappings from R

to R, equipped with the m e t r i c p(f,g) = min{l, suplf(x) - g(x)1}. Let
x
H C C be the set of those mappings f • C, for w h i c h t h e r e are i n i t i a l
functions ~ % $ with ~(0) = ~(0), generating the same solution x (t) =
= x (t) of (1) for t ~ O. T h e n C \ H is n o w h e r e dense in C.
In o t h e r words, the s h i f t operator for the solution (i) is g e n e r i -
cally not injective.

Proof. The set M of t h o s e f 6 C, w h i c h attain strong local maxima

at c e r t a i n points, is c l e a r l y open and d e n s e in C. H e n c e it s u f f i c e s to
show t h a t M C H.
Choose, for a n v g i v e n f E M, points a < c < b with f(a) = f(b) <
< f(c) = max{f(t); t 6 [a,b]}. For any d > 0 choose a s e t M(d) of
points

a = a 0 < a I <...< ak = c = b k <...< b I < b0 = b

(k is s u i t a b l e integer) such that f(a i) = f(b i) for a n y i, a n d

if(t) - f(s)l < d whenever t, s • [ a j , a j + I] or t, s • [ b j + l , b j]

for some j. L e t ~(d), ~(d) 6 C(-1,0) be such that ~(d)(i/k - I) = a ,

l
¢(d)(i/k - i ) = bi, and ~(d), %(d) be linear on e v e r y of the intervals
[i/k - i ~(i + l ) / k - i], i = O,...,k .

We follow the above construction for e v e r y d = l/n, n = 1,2, . . . .

Without loss of g e n e r a l i t y we m a y assume t h a t M(i/n) C M(i/(n + i))
 423

lim ~(i/n) : 9 and lim %(i/n) : %

n-~ n~

uniformly. Moreover, @(0) = %(0) and f(@(t)) : f(%(t)) for e v e r y

t E [-i,0], i.e. ~ and % are the desired initial functions, q.e.d.

Modifying the a b o v e argument for e q u a t i o n with continuously differ-

entiable right side, we obtain the next

1
Theorem 2. L e t C be the set of c o n t i n u o u s l y differentiable functi-
ons f r o m R to R, w i t h the u s u a l cl-netric. L e t H 1 be the set of t h o s e f
f r o m C I, f o r w h i c h there are initial functions ~ # % with 9(0) : %(0),
generating the same solution of (i). T h e n both

Int H 1% # and Int (CI\H I) %

Remark. Theorem 2 c a n be g e n e r a l i z e d to the e q u a t i o n

x'(t) = f(x(t)~x(t - I)) (2)

where f is c o n t i n u o u s l y differenatiable. We c o n j e c t u r e that also Theorem

1 can be g e n e r a l i z e d to t h i s case, although weave not able to g i v e any
proof.

Next consider the second problem. Given some ([,[) @ R 2, where

> 0, d o e s there exist an i n i t i a l function ~ 6 C(~ ,0) such that
x ([) : ~, where x is the solution of the equation (I)? T H i s is the
problem of p o i n t w i s e completeness of (1). In [4] we r e c e n t l y gave a
sufficient condition for p o i n t w i s e completeness, which for (i) can be
formulated as f o l l o w s :

Theorem 3. L e t t , x E R be given. If f @ C s a t i s f i e s the L i p s c h i t z

condition

If(x) - f(y) i S L I x - yl for any x,y C R

and

L . ~< 1 (3)

then there is a f u n c t i o n ~ E C(-I,0) such that x (~) = x.

Note that theorem is true also for x E R n.

 424

The following exalnple s h o w s that the c o n d i t i o n (2) in g e n e r a l

cannot be o m i t t e d .

Example. Let f(x) ~ 0 for x ~ I, a n d f(x) = i - x for x > i .

Let ~ @ C(-I,0) be an i n i t i a l function. Let x be the c o r r e s D o n d i n q
solution of (1). Then x~(t) is n o n - i n c r e a s i n g in [0,~), hence we ha-
v e x (t) a x (1), for t 6 [0,1]. Since x~(t) S i for t > i, w h e r e
~(t) = x (J) = const for t E [-i,0] , we h a v e x (2) S i. T h i s is t r u e
for a r b i t r a r y 9.

The abo~e quoted prcperties cf d e l a y e d differential equations does

not seem to be g o o d for m o d e l l i n g . Theorem ! e.g. indicates a structural
non-stability for m o d e l s involving the equation (1). But also the well-
known Volterra's - Lotka's predator-prey differenatial model is s t r u c t u -
rally unstable though its r o l e among the k n o w n models is of g r e a t impor-
tance.

References

[i] H A L E , J . , Theory of Funtional D i f f e r e n t £ a l Equations, Springer,

Berlin - Heidelberg - New York 1977.

[2] H I R S C H , M . W . , The dynamical approach to d i f f e r e n t i a l equations,

Bull. Amer. Math. Soc. ii, No. 1 (1984), 1-64.

[3] MAC DONALD,N., Time Lags in B i o l o g i c a l Models, L e c t u r e N o t e s in

Biomathematics 27, Springer, Berlin - H e i d e l b e r g - N e w Y o r k 1978.

[4] SMfTALOV~,K., On a problem concerning a f u n c t i o n a l d i f f e r e n t i a e

equation, M a t h . S l o v a c a 30 (1980), 239-242.