25 Internat. J. Math. & Math. Sci. VOL. 18 NO. (1995) 25-32 A STABILITY RESULT FOR PARAMETER IDENTIFICATION PROBLEMS IN NONUNEAR PARABOLIC PROBLEMS STANISLAW MIGRSKI Ja,gclloni, University ul. Naxwjki 11, 30072 Cracow (Received Hatch 15, 1993) ABSTRACT. A sequence f idmttificatim lrollems of coecients in the parallic equation witlt n<mlinear boundary conditims is cmsidered. The parameter (index of an element of the sequence) appears in the cost flnctionals as well as },mndary data. It is pr,vcd that tlte optimal solutions exist anltha.t under Some continuous cvergmice f the cost functima.ls and the convergence of the data, the sets of optinml solutims cmverge in sne smse to the set <f <ptinml solutions of the linit pr,bln,: KEY WORDS AND PtIRASES. Invm’se 1,r,l,lmt, system i(lelit-ific.ation, stability of sflutions, 1,arabolic differential eqmtion, n,mlinom" l,mnda.ry condition. 1992 AMS SUBJECT CLASSIFICATION. 35R3{), 35I(60. INTRODUCTION. In the recent years tlmre has been. an increasing interestin the imrameter identification (or inverse) problems inw,lvig differential equation constraints. Such problems arise in particular in 1. the coefficient estimation for partiM differentiM equati,ms (for example in [2-3], [14], [17-18])s wellas in the theory of structural optimization. The identification problems consist in dctennining of unknown parameters (coefficients) from known ol,scrvations of the modelled processes. In this paper we investigate a class of identification probhems fi.r the second order nonlincar parabolic system: u’= -A(t). in fl x + fl(u) g g u(O) o in where fl C R" with boundary F, 0 < T < +oo, operator A(t) has the form in (0, T), (1.1) r x (O,T), (1.2) , fl o (,, (1.3) is a maxi,nal monotone gral)h in R x R, the a) and o,,(,) "(*’10,. " is the conormal derivative sociated to A(t). Above, v is the unit outward normal vector to P. Given the set of admissible parameters and the cost fltnctional defined on the space 26 S. W (see N()tati()n) ()f th(.: s()ltti(ns (.()(1.1) MIG6RSKI (1.3), we ;re itereste(l i, tl,e f()ll(,wig l)a.runeter i(lentifictti()u () , . Here a([ where u(a, ) denotes tle weak s()luti()u i c()rresI)(),[ing to the (lath. n., g an[ i what fi)llows, we Sll)p()se theft tle gral)h fl is fixed. Our [dm is to l)r()ve the existence of ()l)tinml s()lt.i()us (i.e.n.n clenet which realizes the niuimnu) t() () ud to sl()w tle stability ()f ()t)tiuM s()lti()ns u(l(.r IWrtnrl)ati()us of the well tm of tl,., ,’,,st ftutcti,,,tM ft. As in(li,’n.te,l it, [I], [4-5] ,m,l [I0-11] the st.,d,ility g mt(l : this kind i)lays an important role in It sltould be noticed that a c()nxptu(.,ss of a(hnissible subset of lint&meters is the cruciltl tssunl>tion in the identific[,tion l)rol)lems (c()ml)tre e.g. [5]). Sinee the cost fimction is not c()nvex in generd) the uniq,eness of (.)i)tinml solutions is not guarant(I. Therefot the stability is understood in the sense of continnity of mttltivtd,ed mtpl)ing. We note that the widely kn()wn q)pron.(’h to the pn.rmneter identifictttion i)rol)lems used alto) i,, n,,m(,ric,,l methods (see [4-51, [141) is th,, outp,t le,tst sqnm’es formuh,tion ([10-11], [13]). In tiffs al)l)roa.ch the cost fimcti()md to I)e miniaturized h.s the where C: W Z is a.n observation ol)erttor defined on the sl)a.ee of sohtions, za is the desired element (target) in the space of ol)servations Z. Such cases tre also included in the fi’e of the paper. Next, it should I)e underlined that the identification of coefficients in partiM differentid equttions is, in genend, m mst,,.I,le l)r,.,1)lem ([1G-17], [13]). This is d,e to the theory of hom()genization ([8], [18]) which slows thnt ,)perators with highly oscillatory coefficients c "rel)lace(l" by very diflhrent ones a.n(l still giving the same response. Finally, we i)oint out that tle l)rol,lems (:)f the type (1.1) (1.3) occur in many mathematical models of l)henomenu studied in physics. For exnml)le, eq,ation (1.1) (lesebes the change of pressure during the flow of viscous fluids in porous media or it governs the heat distribution in a body occupying the w)lume ft. It is nt,nd to consider such problenm not only with the classical (Dirichlet and/or Neumam) I>oun(htry conditions, but tlsc) with the more gener ones. The boundary condition (1.2) includes some particuhtr cses e.g. the Signorini condition, the Stefan-Boltzmn heat radiation law, the .Newt()n’s law, the natural convection, the MidlisMenten law. For these and other importmt examples of the condition (1.2) which appem-s in mechanics, biology md chemistry, we refer to [12], [7], [9] m(! the bibliography in them. We recall that the evolution vtrittional iuequdities can be formulated in the fi)rm (1.1) (1.3) (see [6], [9], [12], [15]). For the general identification theory i)resented in an’abstract rammer we refer The remainder of this note is divided in three parts. In Section 2 we give a sult on the continuous dependence of solution to b(,un,htry vdue problem (1.1)- (1.3) on the data. With this background, in Section 3, we show that the problem (P) h a s()lution. The lt section is devoted to the stability of optimal solutions xvith spect to variations in the given data and the cost fimetional. Notation. Let fl be a bounded open subset of R" with Lipsehitz continuous boundm’y F. For H La(fl), by [[.[[, [.[, the norms in V and H, respectively. By V’ we denote the dnal to V. Following e.g. [15], we define the Banach spaces: Y La(0, T; V), H La(0,T; H), Y’ La(O, T; V’) (the spaces of the square s,mmmble fi,nctions defined on (0, T) with the vMues, we denote 27 NONLINEAR PARABOLIC PROBLEMS with rcp:ct t< t, is un<lert<,<,l il the dit, ri],tti<,lml st D O’ . will ]e del<te<l Iy v Tim ,,,,, ii 1,’(A, B)= .s,,.+,{,(,,, B)" Thrttghtt this l>l>er 2. dvJd L(F). Givel, cl:I ets A alld B hl a Banach space X, we dc6no t.l,, ,,,,,,.,,ti,,,, ,,f ,,t 1,y ,I(.,:,A)= +,,.1{11.,- ,,ll.x- ,, A} st.anl ht" tl,e im,er l)rdtct tl,e distance f,,,cti,,,, and eile a. s,ati cnx’tmtit ver a (.5) A}. t,l,;te! sdscrilt.s is a.dlt.etl. CONTINUOUS DEPENDENCE ON TIIE DATA. The goal of this sectin is t study the qtestin of contitmous dependence f solutions t) (1.1) (1.3) on the data aij, 9 m,l :. First we give tire existence result n tire slutions to tltis ltllem. To this enl, we athqt the fllwi,g is a vea.k s,,h,ti,,n t,o DEFINITION 2.1. (see [7]) A fimcti,,n u if there exists a fiutctin w L() such t.l,t ,,,(,, t) /(,,.(-, t)) alld < .,,’(t),,, > ,,..,. ,,,, (1.1) (1.3) if (2. ) C, ,.,, +,,(t;.,,(t),,,)+ < .,,(t),, >v=< /(t),., >r w, e v,,,.c. (0,T), -(()) :, where we have set a(t;z,v) aij(x,t),lx, We need the following hypotheses (H,) (Hz) Vz,,, e V, a.e. e (0, T). (2.2) (1.1) (1.3): the coefficients {aij}, i,j 1,...,n are functions fi’om C(Q) such that a’ii aij(:t,t)ij, V e R" a.e. in Q, for some constant a > 0, is a maximM (multivalued) m,,notone graph in R / R which satisfies the condition 0 fl(0), ,,n tltc data ,f the 1-,rollem (2.3) It is well known (see e.g. [61) that flis a subdifferential of a proper, convex, l.s.c, fimction PROPOSITION 2.1. In addition to the hypotleses (H)- (Hs) we sume that j() Then the problem (1.1) (1.3) has a unique weak solution u 6 W. Morover, the following L(fl). estimate holds: II-IIw + I1,,’11<=) ( + Ilgll) + I1), (2.4) LZ() is a selection of fl xvhich apl)ears in (2.1) and c R i independent of g d The proof of this result can be found in [7], Proposition 1, where the overMl hypothes on where w 6 the coefficients aij were more restrictive. A careful look in that proof can convince the reader that it is still true for the ce of coefficients satisfying (H). Let {ai}, k 6 N, be a sequence in C() satisfying (2.3) uniformly with respect to k and let (1.3) corresponding to {aj }, {(g, )}, we have LEMMA 2.1. L’(). It Under the ubove ct m, ttttions, aii in let us a.ssume that C(), i,j, stisfies (Hz) d j() S. 28 (9,, t,) as MIG6RSKI (9, t2) in L2(E) x H, +c, tlen k "t- ,t i V f3 C’(0, T; H) w,akly in W, a,l . k +c, where L u(-i.,g,) is tiilu, s<lti (still it (1.1)- (1.3) corrt,sp,,n,ling t, ai.i, g a,l ,s tl, sesc f Defiit, im 2.1) N,te that iy h,wer senicott,iudty ,fj, it fi,llows tha.t j() L(). Tlwrefore, Pr,l,<,it.i<,u 2.1, we k,w tl,t t.l,we exists a mdqe solti,n to (1.1) (1.3) c,,rresp,nding to aij, y alibi Sulstr;t.ctiug tw, .qmt.ios which Itre satisfied by u and u, we obtain PR()OE. fr, m . < .,,()- .,,’(.),.,, > + ,(;.,,(.,,,) + < ,,,(.)- ,,,(),, >r= < .()- (.), >r fi,r all v V, (0, T), where a.e. s ’iv,,’k L"(E) are sm’h tlat ,,,(,} e z(.,,(,}), .,,,{,) f(u(,}) .. ,,,- and a(.; .,.), a(.;., .) are of the form (2.2) wit, h he coecients ai, a}, respectively. the flmcti,n v aul intgrating lmth sides we get: Taking u(s) u(:) ,,(s;..(.s)-,,(),,,(s)-,,(s))ds + lu(t)-,,(t) + + < ,,() ,,,(),,,() ,() >t. d. k(; ,(), ,,/) ,).d] + gl 1 + (1 ,,(; ,,(),,,() ,,()) (.), ,) ) r ds. Hence, from (H) and (t12) we c)l.)tain: + fo’ ’ o,,() o.,,,,(s) (,’ ,,)0,, !1.0.i Uing the continuity of the trace operator from V to 0.,() 0. Id + L(P) mad tim inequMity 2ab N a+ b we have lug(t)-,,(t)l +, /c., II,,(s)- u(s)llas N I1"t) "ollc.( Ilull +c: I -1 + I() where ci, 1,2 are positive constructs indepen(lent of k. l’om this, (2.4) and from the hypotheses we deduce that u,u in C(O,T;H), as k-l-c. On the other hand, owing to the estimate (2.4), which is tmifortn with respect to k, ut u we obtain weakly in The uniqueness of solutions of problem (1.1)- (1.3) implies that the whole sequence u converges to u in the sense of (2.5). This completes the proof of the lemma. 29 NONLINEAR PARABOLIC PROBLEMS EXISTEN(:E IIES[!I:I’ 3. IN I1)I);NT1FI(IATIt)N. Using tile rcstlt ,f I.(lna 2.1, w’. ,’,1 11, till,wig ln’,x.’’ t.l .xisl.,lc, ,f sltt, in ,,f t.lm i,lettificati lrl,l,n (D). Tlis is given i I,’.t tl: TItE()I/EM 3.1. almissillc ltra,mt,’.rs satisti,b L,t tle cst 5mctimal l" 1 tsstttll.is R 1., ,t" W, Sttlltst’ t.lat the set f lrlsit.im 2.1 ltll. w,kly squ,nt.ially ltwer scmicmtimns m . Then the lrolhn (’P) has a sllt, in. PR()O. a,l let (Ha) XXhe ,l,l,ly tlc dir,ct (’t,l,l ,f c.ad’ths ,f’ variat.i,,ts (see e.g. 1,e t itiizig s,,le,,’ t’r, scl tl,t , {a} lint Sin(.,c <1 .<t.I is c,)ml)act, a ,u ,:/(u(,.,., .q, 9)) tl(:’.r(.: exists a. 3d swh that (,. a(). It id’{,.’l(,,.(-.,9,)) -. std,s(qw.ce of tle seqteltCe Ad} satisfy ),).. {a,}, tl)ll()ws fl’<)m L(unnla. 2.1 that in [1]). Let 9, rehd)eled again as {a,}, I)a.rticflar R.EMARI, 3.1. In general, v,,itl,nt coavxity a.ss,nqtins, we do not expect uniqueness of the ol)timal soluti,,n in ide,tifica.ti,, ([1], [16]). 4. STABILITY RESULT. In this section we give the rcsflt f this. lml)er on the dependence (hence also the stability) of the optimal clemnts for the ln’,,1,h:nn (’P) on the data a.s well as on the cost functional. We consider tlm sequence (inlexed by the lm.rmneter k N) of the identification problems: () m,in find a*={,,iS} in sothat where u(a,g,k) are the s,,luti,)ns in to(1.1) (1.3) corresponding to the perturbed data and the are We show that the set of optima.1 solutions to (Pk) functionals. perturbed gk, k k converges in sone sense, to the set of optimal solution to (P). We need the following cmtimos convergence of 5mctionals. Let (A’,r) be a topological R. space and k :,l" DEFINITION 4.1. We sa.y tha.t a. sepencc of functionals {}, k N, is sequential t,:, and we write continuously convergent (shortly, C,-com,erges) C,,(r- X) lim if for every x ,V and for every {xk} C ,V whi,:h r-c,,nvcrges t,, x, the sequence k(x) converges t,, :(.,). For each k N, we denote by S, $ the sets of optimal elements to the proMen (P), (P), respectively, i.e. ith the above notation we have MIG6RSKI S. 30 li, l, wl.,rc h*(.,-)is arg’ I,y If (,t.3) is ,t, vcrifie,! t,l.u’e exist g > 0 and a sequeuce c,t.ri,lict, i. + s,:l t.lt I,:,, }, k < h*(S.,.,S), V k. (’,lrly, there exist a. S,. ,,f c,mpactuess ,ff N. sucl t.lt < d(,.L.,S ), V In view (4.3) {}, (1.5). 1.,fi,,1 i We PR()()F. $,, $ M, we !,.,1,’, :,. c N. tl.t tluc cxist, (4.4) ulscqucnce of {a. }, that we will d:nt,e iu the ame way, wh tlu.t fi,r some a* M. ,. u. Let ,ow u(a., g., )and u* ..(.*, g, ),h.,n,,te the s,:,hti,,ns t,, (1.1)- (1.3) which )a,l ("*,9, ), r"sl’cctively. Fr,,m (4.1) and (4.8) c,,rresl,oul to the triph,s (..,., 9,., .,. 2.1 gives u* w,a.ldy iu W, u. d C’,:,-’nv’rg,s Since the fimctiouals J, t J(.,,’) . as +. k wc i,,:h,.(,,L ). u(.,9,,,.) and Le us fix au arbitrary M. Let ’u,. (1.1) (1.3) with the inlicated data. Frcm the ccmt, iuts (4.6) u u(a,9 ) be the lepeudeuce ou the data, solutions we conclude that u. Hence, as above, ’u wcaklyiu W, as we have J(,t) Since a.,. are in In the limit, S,,,, we +oo, one gets from (4.6), (4.7) s(,,(,,’,,j,)) _< s(,,(,,,,)), v ,, e as k,, , we lu.w ,* ,(,,Z.,,s) now from (4.5), liu Jr,,. (u,,.). we ,:,btain tha.t proof of the theorem. (4.7) have From the arbitrariety of a But kv +. (4.8) . S and this implies that II,L --*11. c,.,tra,licts (4.4). This proves (4.3) (4.8) ,! completes the 31 NONLINEAR PARABOLIC PROBLEMS ’-FI; c,verg’n’e f’,v (4.2) lulls, itt’’, f’w tl,’ ’tu’c f" f’uct, iu;,l ff t.h’ fiwn (1.). t" l’rtuled listrilnttetl ols.l’vatis il "H cv.rgilg t za, ,1 tiffs xve hav that(4.2), is satisfied with eml,ed,ling is c,nl,;,ct. (c,,l,a,v,’ [15]). in..L k +, tlmn usig tlu cnxqatct,t,ss ,f tlt tx’ace ’nl,llitg fi’m W int L2(E), we easily gt REhIARK 4.1. ()te c g’u’ra,lize t.l" ’,slts lres,nt,.l al,w to t,le case when A(t) is ol)erat{w tf tim t) 0 0 a litt:reti;l ()tlr t.lu:vy, with sme niuw chngcs, cnt ltntlle ittx;’rs, lnlh’tns inwlving the idetifica.tion of REIARK 4.2. A futher gcter;,liza.t.i f nr r,slts can lm oltained considering the ilentificatim prolh-.s fr (1.1) (1.a) witl the ixl lnmtlary cmditicns (1.2), where gi Pi (0, T) ani Pi are the (lisj,int parts of F. The exact fi)nmdatim f the results with olviCnts mliticatims is h,f’t t, the: ilfl.:rested reader. instea(l ,,f _ wlfih: tim mth)t" was visiting the Scmla Normale Sul)eriore di Pisa and was S,tl)l))rtel I)y tlu: Istitut Naziniale (li Alta Matematica Francesco Severi, Rome, Italy. Tiffs w()rk was 1)a,rtially tim(led by a National Research Project in ACKN()WLEDGEIvlENT. TI,= l,al)(’r ws wtitt,n Mathematics "Eqmzioni Differenziali e Ca.lc()h) dell(: \’aria.zioni" f the Ministero (lell’Universit,4 e (ldla Ricerca Scientifica e Tecn)hgia., 40% c(mtra.cts. REFERENCES AHMED,N.U. Qptinfizat, iot a,t____[I Idett, ifica.t.i,n ,f Systems Gv,,ened Space, Pitman London- Melbourne, 1988. on Bana(:lt R,s(’,.rch N()U:s i,t M,.t.hema.tics Ew’l,,tion Eq,,ations 184, Longman, Boston BALAKRISHNAN, A.V. hlentificati()t ,f distrilmted parameter systmns: non-computational aspects, Proceedings IFIP \;Vorkig C)nference, 1R(me, 1976, A.R.uberti, Ed., Lecture Notes in Control and hfl)rnm.tion Sc.ienccs, \&)l.1, Springer-Verlag, Berlin, 1978, 1-10. BAMBERGER, A., CHAVENT, G. and LAILLY, P. About the stability of the inverse problem in 1-D wave equati(ms apl)licati(ms to the interpretatim of seismic profiles, Appl. Ma.tl. O!)t.i. 5 (1979), 1-47. BANKS, H.T. On a va.riati()lal a.1)l)r()ach t() some pa.rmneter estimation prol)lems, ICASE Rep. No. 85-32, NASA Langley Res. Ctr., Ha.nl)t(m, VA, June, 1985. BANKS, H.T. and ILES, D.W. On COml)actness of a(lmissil)le parameter sets, Proceedings IFIP W()rking Conf(.rell(’e, Gainsville, 1986, I.Lasiecka, R.Triggiani Eds., Lecture Notes in Control and 142. Info,ma.tim Sci:n(’es, Vol.97, Springer-Vcrlag, Berlin, 1987, 130- 32 S. BARBU, V. B,,ulary c<,xtt.rl lw,i,l’t. (1982), 125-1-t3 ()ptil. 2f.._)) 9. 10. BREZIS, It. MIGRSKI Pr(l,h.l(:s wi/l ,tli’tr st.;t: .lti<t, lil;,t(’r,lx, .l_. F,I,II,. Pir(.s SIAM ,l. Ct.rl A!!)I. 5_.j.l (1972), 1-168. C()L()NIUS, F. (l KUNISCII, K. ()ttl)t h:ast squares stal)ility fi)r estimation of the (liihsi()n (’()(Ili(:i(,nt in (,llil)tic C(luti() Pr()(-(.(li IFIP W()rki ()Itferece, Gainsvillc, 1986, l.Lnsi(..(-k, R.Triggia.6 E(Is., L(.wtu’c N()tes in C(mtr()l a(l Inf()nmS(’i(ttc(s, V(,I.97, Si)rig(.r-V(wlg, B(’rli, 1987, 185-195. ti(m 11. C()L()NIUS, F. ii(l KUNISC, II, B()n,lary \rl,, l"r()l,h.,s, K. 3_ Stld.)ility f()r Pm’a.n(.:t(r Estimation in Tw() P()it R__,,i),(: A__.’(.w. hl,tl. 37() (1986), 1-29. 1"2. Paris, 1972. 13. KRAVARIS, C,. m! SEINFELD, J.II. hl(:ntifica.i()t ()f l)a.rameters in (listril)ute.(! I)armnct(:r systcuis I,y r(:g, da.rizati,,n, SIAM J. C()tt.r()l ()!)t,io 23 (1985), 217-241. 14. KUNISCII, K. I,h.:(.ificiti() l(l Estila.tii)n ()f l-’ara.neters in Abstract Ca.u(,hy Problemu, Matlmnuai(’I C,>t.r()l Tl,.,)ry. ll;,la,q C(’t,’.r Pl)li(’ati,,n, V,)1.14., PWN, Warsaw, 1985, 277-298. 15. LIONS, J.L. Dm()(l-Gttli(:r K’.t.l,o(h,s (I(,_2 rSs()lti(, (h,s l)r()l)l(,n(:s ____x linfit.es Villrs, P,ris, 1969. 16. LIONS, J.L. 17. MUR.AT, F. Contre-cx,:tq,h:s p(,tr (livers l)rolA(,m(:s (nile (’(mtroh: intervient ficients A,. M;,t. I>un. A )121I. 11__2 (1977), ,19-68. 18. MURAT, F. Control in c,),:ffi(’i(,nts, S,st(:s ,i(.___[l Pergamon Press, 1989, 8()8-812. ()n lin(:.aires S,)ne Aspc.cts (,t’ Mo(h:lli,g Pr,)l,lems in Distri]mted Parameter Systems, Procee(ligs IFIP W(,rking (-_’.,)fi’(,.’e, Rome, 1976, A.R.uberti, E(l., Lecture N)tcs in Contr()l and hff(>rnn.ti(,n S:i(,(:(:s, V()I.I, Sl)ringer-Verhtg, Berlin, 1978, 11-41. C,()tr()I (hms les co(,f- Encvc!()l)e.lia M.G.Singh E(I.,