I01 Internat. J. Math. & Math. Sci. (1988) 101-114 VOL. ii NO. ON SOLVABILITY OF BOUNDARY VALUE PROBLEM FOR ELLIPTIC EQUATIONS WITH BITSADZESAMARSKI[ CONDITION J.H. CHABROWSKI The University of Queensland, Department of Mathematics, St. Lucia, 4067, Old, Australia. (Received January 21, 1987) I, ABSTRACT. l,roblem tiLLs l,,)l, I,.m wlue paper we 1,v a line;-r with the solvability of a non-local investigate elliptic equation, which is also known as the boundary the Lilt sadze- Sm, arski condLt ion. We prove existence and uniqueness of a classical solution to this problem. the In tile fnal part ,f" this paper we propose an L’-approach which gives a rise to weak stlutims in a weighted Sobolev space. the B KEY existence of weak sadze--Samarsk [ condi WORDS ANt) Bi tsa,-tz.-.aar.k solutions is INTRODUCTION. PHRASES. modification of the Elliptic equations, non-local problems, cond ions. 35J, 35C, 35R. In recent years several solvability of non-local [1-9]. The crucial point in proving suitable on. 1980 S SUI.JECT CLASSIFICATION CODES. I. a authors have studied the problems for elliptic and parabolic equations The importance of non-local problems appears to have been first noted in the literature by Bitsadze-Samarski. The problem studied in these papers constitutes a direct generalization of the classical boundary value problems. The most significant feature of nonlocal problems is that the boundary condition relates values of a solution on the boundary to its values on some part of the interior of the region. This type of the boundary value problem is often referred to as the boundary value problem with he B[tsadze-Samarskii condition [7],[8]. The problem (2.1),(2.2) discussed in this article arises from the mathematical descri.ption of some processes in a plasma (see paper of non-],’al problems). [6] for full account of physical aspects J.H. CHABROWSKI 102 ’rhc organized is pal,er as In Section 2 we give the follows. uniqueness and existence theorem of the classical solutions of the problem Our mehd is bnsed on lhe aximm principle developed in o,.]),I,.. [5]. papers [.] and Section 3 contains a discussion of the solvability of th non I(,caI problem for harmonic functions in a disc in 2" The results of this section slightly improve the explicit formulae derived by Bisadze see ’1] f’r h,rmoni, functions associated with some non-local [2]) :,,,,I in a general case of a linear elliptic equation we reduce the problem, 2.1,’2.’2) probl,m lhe solvability of the integral equation of the The final sections 4 and 5 are devoted to the study of the second kt,l. non--],,::,l I.,, p’ohle.m l’,,i" a linear elliptic equlion with a paretcr, whose This allows us to remove some princp:] part is in a divergence form. restri,:-ons on appearing in the boundary condtion (2.2). the coefficient On the olher hand ths also suggests further extensions of the solvability of the l,roblem (2 l) (o 2) in a weighted Sobolev space 1,2 (). We adopt 2 here the L-approach to the irichlet problem with L-boundary data from [13] 2. a,,d I0]. UNIOUENESS AND A PRlORl ESTIMATE. We consider a linear equation of the elliptic type n n a..(x) D .u + Z b (x)D u + c(x)u Z i ij i,j=l J i=l [u in O, where 0 is a bounded domain in R p of defined on the boundary (2.1) The purpose of this paper is to n investigate the following non-local problem: and f(x) given continuous functions h find a solution u satisfying the boundary condition u(x) @ where p(x) u(@(x)) (2.2) h(x) on 00, is a given continuous mapping of 00 into O. Throughout this section we make the following assumption (,%) The coefficients of the operator L are bounded in Q and there 0 such that exists a constant n i,j=l aij(x) iCj for all x e 0 and R n .Moret, w.r we assume that 0 satisfies an interior sphere condition at each poknt of o0 (see [11], p. 33-35). The uniqueness of the problem strong mxu,m, PROPOSITION 1. (2.1),(2.2) is a consequence of the principle. Let Ifl(x) 1 on DQ and c(x) _< 0 in 0 md suppose that SOLVABILITY OF BOUNDARY VALUE PROBLEM FOR ELLIPTIC EQUATIONS 103 e ther (a’ -] (b’ c(x ,(x O) at some point x 0 e aO or 0 at some point e O. x] Then the problem (2.1),(2.Z) has at most one solution in It Pt/OOF. 02(0) C(). sufficient to show that if f(x) 0 on Q and h(x) _= 0 on 00 0 is tl,,. only solution of the problem (2.1),(2.2). It is clear that u,d--r-ach of the asstmq,tions (a) or (b) any constant solution must be is then u identially equal to O. / If u 0 then u must be a non--constant solution and by the strong mximum principle ([Ii], Theorem 3.5) we may assume that u ...,:_ It" "xo) m_ax u(x) 0 with x 0 2 e 00. 0 we get a contradiction. Therefore it remains to consider two cases (i) (x 2) < i 0 In the first case #(X 2) e O. and u(x 2) u(@(x2) (ii). we u(x 2) t,((x 2)) u(x2), (x2) In the second case -I _< /(x 2) < O. (ii) which is impossible since have 0 P(x 2) and by the strong maximum principle u takes on a negative minimum at x e 00, that is 3 u(x 3) and we may asstme min u(x) that 0 (x3) 0 since otherwise we get a contradiction. Hence u(x 3) u(#(x 3)) > 0. P(x 3) Now we distinguish two cases either u(x 2) _< [u(x3) or u(x 2) > [u(x3) I. We show that both cases lead to a contradiction. we have which is [ml, ossible. In the second case we have Indeed, in the first case J.H. CHABROWSKI 104 Since both values u(xo) and min2mum :I :.xo) u(x..,)) are negative u attains its negative e O and we arrive at a contradiction. [nsl-,p-tion of the proof of Proposition shows that the following version or’ the maximtun principle holds true. i’ItOI-’OSITION 2. Let Lu (_> O) 0 Then u Cx) . Suppose that 0 (_< O) on 0 in Q and c() u(x) -p(x) in Q, and 0 u((x)) 0 (< O) on 8Q. As an immediale consequence we deduce an a priori estimate THEOIEM 1. d ( Suppose that c(x) 0 and 0 1) o 2) then lu(x)[ PROOF. fl(x) _< a where If u is a solution of the problem d in O and 0 are constants. a sup[f(x)[ :1 + suplh(x)[., for all x e o. Let us define then we have Lv f- c :c suplh() aQ suplf() Q _> f + suplf(x) Q in 0 1 1 0 -, 00 (x) (1 on OQ. llen,:e by Proposition uCx3 on O. _ suplf(x) + l-a sup[h(x)] 2 + l__ suplhCx) Similarly we can establish the inequality u,’x suplfCx) l.. 0 $:S suplh(x) aO on O, ,_-onsderng the auxiliary function 0 _> o SOLVABILITY OF BOUNDARY VALUE PROBLEM FOR ELLIPTIC EQUATIONS ’""’ ItEMA}tE I. 105 i--A suplh(x) oo u":’d’u’lr)l 0 on OO, then any two solutions of the If c m 0 or 0 and (x) problem (’.1),(’2.2) differ by a constant. 3. EX I,’t’I,:NCE 01," CI,ASSICAL SOLUTIONS. We commence by considering a particular case of the non-local problem ’2. 1),,’,.’.’2 which consists of finding a harmonic function u on B(O,I) and satist’ying the boundary condition pu((z)) uz’ where B(O,I on 08(0,1), h(z’ (3.1) an open disc in of radius centred at O, 1 is a constant in the interval [-1,1] and h is a continuous function on B(O,1). The mal,png is given by (z) with 0 is a 5 1, where is N2 , ,(hz) ],(z)] univalent analytic function on B(O,1) such thmt ,(0) O. By virtue of Schwarz’s lemma we have Izl ro I%C.)1 The function 1,2 o(Z) Letting z and (_l(Z)) k(Z) for we have klz for k (3.2) B(O,1). maps disc B(O,1) conformally and univalently onto certain set contained in B(O,1). k _< 1 in B(O,1) and in B(0,1), are harmonic functions we have the u((z)) Since u and 1,2 following representation formula uCz) uC(Z)) Suppose first that -1 u(z) /l n / n [Re I+6 2(I-) for n 1,2 n + z k=l . n 0 ] aB{0, I) 1 t-z t ]h(t)dt mF(z) (3 4) Iterating (3.4) we get 1. U(n(Z)) It follows from (3.3) that [F(n(Z)) Re pk-1 r([k_ l(z)). j" OB(O,1) t+(z) t-n(Z)t h(t) dt (3.5) _< lh(e is) lds and consequently letting n in (3.5) we obtain I06 J.H. CHABROWSKI r uz) n=l pn-lF(-1 (z)) un for,,,1 y on @ =-I. L,’t us now consider the case h(t) dt 2t I "* oB(0, I) Re---.- 2u(0) It follows from (3.4) that (3.6) and tile functional equation (3.4) can be written in the form u(#(z)) u(z) + Re 1 1 Re 2u(0) i= i I OB(O, 1) h(t) dt -z I dB(0,1) t(t-z) "t _1 1 2t z 1 2t ]h(t) dt Re[ (z).z]. Iterating the last equation we obtain 2n-I u(@on(z)) uCz) z + (-i) J Re[}(@j(z)) (3.7) and 2n-2 u(z) 2u(0) U(@Zn_l(z)) + Z j=0 (-])J Re[}(@j(zl) @j(z)].(3.81 It is easy to see that 2 I_(%(z)) for all n series 1,2 (- Z 1) J j =0 n n _< =(_s) 0 [h(e s) [ds B(0,1). and z Re[(j(z)),(z)]__.a Since u(0)+ z (-i) j oJ i. B(0,) converges uniformly on B(0,1). (3.7) and (3.8) we obtain the se u(z, l#j(z)l < Letting lim[ in boh cases Re[(#j(z)) j(z)] and invoking (3.6) we get that u(z) Finally let 1 u’z) By Remark Re 1 4n h(t) dt I z (-I) j + OB(O,I) j=O Re[(@j(z))@j(z,]. i, then u((z)) Re I= ’1 I OB(O, 1) { any two solutions differ by a constant. h(t) dt If z O, then the SOLVABILITY OF BOUNDARY VALUE PROBLEM FOR ELLIPTIC EQUATIONS u((O)): urO)- 0 ICe _1.. n. 107 f(t) dt I 6B(0,1) which g,w".; a necessary and sufficient condition for the solvability of the Hence problem (3.l). a[(z)z]. u(#(z)) () Iterating this functional equation we obtain n--1 j:l we obtain Letting n u(O) + u(z) Z j=l Re[(j(z)) j(z)]. To determine a solution in a unique way we may impose an additional condition u(0) 0 is a given constant. The case / C, where C 1 was considered by Bitsadze in [1] and [2] under the assumption that h is Holder continuous In a general case we reduce the problem (2.1),(2.2) to the Fredholm integral equation of the second kind. THEOHEM 2. . D[j aij Suppose that the assumptions of Proposition 1 hold and that n), D (i,j b i n) and c are Holder continuous on (i Then the pr,blem (2.1),(2.2) admits a unique solution u in PROOF. C2(O) O C(). We try to find a solution in the form ux) - C(oO) where v operator I, and dG f dn dO is v(y) y I dSy G(x.y) f(y) dy, (3.9) O to be determined, G is the Green function for the denotes the conormal derivative. The boundary condition y (2.2) ]ud o tho Fredholm integral equation of the second kind. dn 80 + y y I ,tl(x) G((x),y) f(y) dy. (3.10) (&( a_G,v,x,,y, is continuous function on O Since (O0) c O the kernel p(x) dn 0 x 0. [;y Proposition has only trivial solution, Y the homogeneous equation corresponding to (3.10) tlence by the Fredholm alternative there exists J.H. CHABROWSKI 108 a unqu,. :.,,lution v e c, C(oO’. (2.1), (.’.. ’:’. 4. . ENERGY L"’dO) which by the continuity of the kernel belongs ’,nse,luently the formula (3.9) gives a solution to the problem F.T[MATE. In tlis se-tlOn we Mu + the elliptic equation in the form co,raider n Z I) (a (x) ij i,j:l hu + hu f(x) Dju) n Z b (x) D u + c(x)u + i i=l + in O, (4.1) with the boundary condition (2.2). Throughout this section we assume that /] is a continuous function on bO and assume :o,O C1--mapping O is a b ai3 Diai3 t|mt Further we with the positive Jacobian. n) c and f are Holder continuous on (i, 0 and that O is a bounded domain with the boundary of class C Th, ,,bje,:tive of this section is to show that 2. the problem (4.1),(2.2) hns a unique solution for large values of the parameter A. For small 6 0 we define According to Lemma r(x) 06 14.16 dist(x,O0) belongs to 0 Ix-y[ > ). n {x" rain y. OO [II] in C’(0-06. (p. 355), o if 0 the distance is sufficiently small. Denote by p(x) the extension of the function r(x) into 0 satisfying the following properties p.’.’< -1 I r(x) for x e r(x: 6.1for p(x) tJ 6 e 6 6 06 o p e C2(), 36 p(x) _< ] -4 I> r(x) in 0 for some constant v and finally o0 (x’p(x) 0}. TIIEOREM 3. There exist positive constants A solutioi (:"0)rl C() of the problem (4. I),(2.2) for tn llVu(x)[ r(x)dx + 0 by I h(x) 0<_< d dS X + I f (x)2 06 006 0 O, {x" p(x) C and d such that if u is a 2 lu(x) r(x)dx + sup 0 C( Proof. o in A O 2 I u(x) o08 dS then X dx). We follow the proof of Theorem 5 in [13]. Multiplying (4.1) SOLVABILITY OF BOUNDARY VALUE PROBLEM FOR ELLIPTIC EQUATIONS :u(x) v(x’l Ip(x)-6} on 109 (}6 and integt’at[ng by parts we obtain n dO aij(x) DiP i,j--I 6 .v j u DiP dS . 1 x n Z I 0 Di(aij(x) i,j:l 6 u2(p-6)dx- f (c(x) + A) 6)dx + b.f) Dju.u(p 2 jr Djp)u2dx+ fu(p 5)dx. Applying llolder’s inequality we easily obtain sup f O,6<d 006 + where d and C dS u C x- 2 [Dul 1 2 p dx + f u dx + 0 f u’ p dx + f 0 0 dx (4.2) are positive constants. Similarly I [Iu[ + h p 0 I u" p O. for some (:,, + a P I I h dS + f 4 hat f u( dO where C with 5 dx + I dx + I u 2 oO 0 dSx] f2 dx + f (4.3) dx + O 2dSx (4.4) p dx + sup O<<d I ao I dS + 0 C for some C 2 P dx <_ C 3 u(()) 0 obvious u The estimates (4.2) and (4..4) yield that O. (: u 0 + [ It follows from (2.2) that 0 where <_ C 2 dx 0 > 4 u" dS x O. Since { x. )2 dSx 0 and C 6 dist(,o(})50. C5 > I dx + 0 is a (dO)$ dO cl-mapping u 2 X- I u((x)) 2 dS ,O (4.5) x with the positive Jacobian it is dSx < c6 0 are constants and x dS [0 Dul2 d + O@ is I u2 dx}, a domain containing (4.6) @(dO) Consequently by virtue of the Caccioppoli inequality 110 J.H. CHABROWSKI we have x- 00 for some C O. 7 I u" I Ul O dx 5- i!!f p(x). d] 0 We now observe that 0 where (4.7) [0 u2 p dx + d sup I O<6<d dO u2 d S x’ Choosing d sufficiently small and k sufficiently 0 d large we easily derive the desired estimate from (4.5),(4.S),(4.7) and (4.8). [ep,_-atlng the argument of Theorem 2 we deduce the following TI1EOREM ,1. There exists a positive constant A 0 such that for every c 2 () the problem (4.1),(2.2) admits a unique solution in 5. WEAl( C(). SOLUTIONS. The energy estimate from Section 4 shows that one can expect solutions of the problem (4.1),(2.2) in a weighted Sobolev space defined by wl"(o> {u Wl’oc (o); llDu(x) 2 r(x) dx + I u(x) 2 dx } and equtl,l,e’d with the norm I [{0u(x) 2 r(x) + u(x )2 dx. We r,,-al] briefly that a function u is said to be a weak (generalized) solution of (4.1) if u e n Z I for each v e W ]’2(0) W1,2 loc(O) and it satisifies n a.. D.uD.v + Z b.D.u.v + (c+A)uv]dx I f v dx (5.1) with compact support in O. To proceed further we need some terminology. It follows from the such that for 0 regularity of the botmdary dO that there exists a number 6 6 e (0,o) the domain 0 (defined in Section 4) with the boundary dO 6 6 to each x e 0 there exists a unique o such that e 6v(x where v(x ls the possesses the following property: point x6(xo, of class C I, xs(Xo), dO6 outward normal to dO at x of dO onto O. Xo o), o) The above relation gives a one-to-one mapping SOLVABILITY c)F BOUNDARY VALUE PROBLF FOR ELLIPTIC EQUATIONS It is known that elements of the space have trc:es on lhe boundary O0 (see do not in general, llowever, by Theorem 4 in [13], [12]). ";0) is a solution of (4.1) then there exists a function if u e e 1’2(0), Iii L2(oO) such that O. (x)]" dS x I [u(x 6) lim 60 aQ Thc.r-I’,re, as in the paper [1], we adopt the following L"-approach the l,r,l., I.,’l. ), (’2.2). ,.m Let h e L(OO). 1,2 Wloc( O) A weak solution u of (4 I) is a solution of the non-local problem with the boundary condition (2.2) if h(x)]2 (x) u((x)) I [u(x 8) ]im dS oO 50 in []3] that if u It f,,ll,w. from Theorem the problem (,1.1),(2.’2) the sens of (5.2) O. x W1,2 lot(O) i5.2) is a solution of (with the boundary condition (2.2) understood in ’(O). then u e We mention also that . u((x)) understood n the sense of trace, which is well defined since u e 1-1 ], ,’h,i,. G (see We now ar’" is 1,2(0 Wlo c in a position to establish the existence result in l,2 of tlc proll,,m (4.1), :2.2). Let THEOREM 5 that for" |, Then there exists a positive constant e L(6Q) o 2 2) (4 1) the problem in 1,2 (Q) loc admits such ^ unique a solution. PROOF Let _> m b’t h ii! be a sequence in cl(oo) I such that lim m- OO A le a constant from Theorem 4 and assume that 0 0 Thcor,.m 4 2 u m e C ((J) 0 2- dS (hm-h) x O. For each ^ guarantees the existence of the unique solution C(O) of the problem (4.1),(2.2) with h h Moreover we have m for ea,’h u I O IDUm 12l rdx + I u 2 dx C(l f2 dx + O 0 where C 0 is a constant independent of m. bounded in 1,2(, convering ealy in ’wl’2(O) is f hm2 dS x) Since the sequence u is there exists a subsequence, which we relabel as u m I,2() to a function u. compac’tJy embedded in L2(Q) By Theorem 4.11 in [15], and therefore we may assume that u m i12 J.H. CHABROWSKI converg,:. i u in By ’rheore,. .I 12 (0). It is obvious that u is a weak solution of (4.1). [13] in has u trace a L-(OO) e in the sense of L 2- con,-, g,.nc,, that is 2 I [u (x)] dS x oO To complete the proof we show that (x) Jim Let C (x) ]O). It is easy to show (x) O. u((x)) fl(x) + h(x) a.e. on 0. p(x) is a legitimate test function in (5.1) a,] inlegrating l,y parts we obtain n I1 DiP Ijp aij i,j=l 40 I dS D (a 0 i,j:l i ij Dip {) u dx + n o i=l O O 5.3) Similarly [ :,0 [hm(x) n Z i, j p(X)Um({(x))] + aij Dip I}jp dSx F(um). (5.4) ApI, lyng the estimate (4.6) and the obvious analogue of the energy estimate t, u u P r [u ,’O q we obtain P (<x)) Ilu C Uqi I O0 where k u 1;u) m (x) fl(x)u((x)) C dx] of as m as q f(x))] [hp- hq[" ,mtinuity (@(x))u((x)) F(u 2 inf p,x’), and C the by u m dS C[ I x O IDuP Du q 12 dx 2 j. dS x, 0 is a constant independent of p and q. solutions weak 2 on 0 we may assume Hence that Combining this with the fact that in L (aO). we deduce from (5.3) and (5.4) that + h(x) a.e. on O0 and this completes the proof. REMARK 2. (4.1),(2.2) It is worth noting that the non-local problem of the type equations. The corresponding boundary datum h in [9] belongs to the space H(bO). has been Since H(80) studied in [9] for is a proper subspace deduced from the results of the paper [9]. the higher order elliptic L’(aO), Theorem 5 cannot be SOLVABILITY OF BOUNDARY VALUE PROBLEM FOR ELLIPTIC EQUATIONS 113 REFERENCES. 1. B1TSAIJZE, A.V., S,v:,l ,I;1. l)okl. ’On lhe th.ory of nonlocal boundary value problems’, 3__qO(l) (198.1), 8-10. 2. BI’rSADZE, A.V., ’On a class of conditionally solvable non-local boundary value problems for harmonic functions’, Soviet Mat. Dokl. 3__.1(1) (195), 91-94. 3. B]’rSAI)ZE, A.V. and SAMARSI(I[, A.A., ’On some simple generalizations of lnear elliptic boundary problems’, Soviet Math. Dokl. 10(2) (1969), 398--400. 4. CIIABIIOWSKI, J., ’On non-local problems for parabolic equa!ions’, Nava Math. J., 93 (1984), 109-131. 5. 6. CHAI.I{OWSKI, J., ’On the non--local problem with a functional parabolic equation’, .Funkciala.} Ekvacio.i, 27(1) (1984), 101-123. for SAMAHNEII, A.A., ’Some problems in differential equation theory’, Difr,re:nciai’nye Uravneniya, l__g(ll), (1980), 1925-1935. 7. SEUBACZEVSEI, A.L., ’Solvability of elliptic problems with Bitsadzel)ifferencial’nve Uravneniya, 21(4) Smt’kii boundary conditions’, (1985), 701-706. 8. SI<UI;A’ZEVSEI, A.L., ’Non--local :[liptic problems with a parameter’, Ma!. USSH Sb. 4_9(1) (1984), 197-206. 9. SEIIIIA(’ZEVSKi, A.L., Differencial’nve ’Some non-local elliptic boundary-value problems’, 18(9) (1982), 1590-1599. Uravneniya 10. MIKHAII,OV, V.P., ’Boundary values of elliptic equations in domains with a smooth boundary’, Mot Sb. 101 (143) (1976), 163-188. II. and TIIUDINGER, N.S., ’Elliptic partial differential der Grundlehrnen second blathematischen order’, Wi:;s,:nsehaften o____ ,.4, Springer-Verlag B--rldel-rg-N-w 1983 (Second Edition). GII,BAI/(I, equa| ions I’. of Y---rk- 12. I’OIILSEN, 13. CUABI{OWSEI, J. and THOMPSON, B., ’On the boundary values of the solutions of linear elliptic equations’, Bull. Austral. Math. Sac. 27 (1983), 1-30. 14. KUFNEIt, A.,JOItN, O. and FUIK, S.,’Function spaces’, Leyden, Noordhoff, Prague, Academia, 1977. 15. MEYEI, I?.D., ’Some embedding theorems for generalized Sobolev spaces and applications to degenerate elliptic differential operators’, J__. Math. Mech. 16 (1967), 739-760. Ebb,. Thue, ’Boundary value properties connected with some improl,er I*irichlet integrals’, Math. Stand. 8 (1960), 5-14.