I ntn. J. Math. & Math. Sci. Vol. 3 No. 3 (1980) 549-558 549 GENERALIZED RANDOM PROCESSES AND CAUCHY’S PROBLEM FOR SOME PARTIAL DIFFERENTIAL SYSTEMS MAHMOUD M. EL-BORAI Faculty of Science King Abdulaziz University P.O. Box 1540 Jeddah SAUDI ARABIA (Received February i, 1979 and in revised form October 22, 1979) ABSTRACT. the form D In this paper we consider a parabolic partial differential system of t H t L(t,x,D) H t. The generalized stochastic solutions responding to the generalized stochastic initial conditions H Ht, cor- are given. Some properties concerning these generalized stochastic solutions are also obtained. KEY WORDS AND PHRASES. Generalized Stochastic Solutions, Strongly Parabolic Systems. 1980 MATHEMATICS SUBJECT CLASSIFICATION CODES: 60H15, 35R60. 1. INTRODUCTION. Consider the system D u t Lu (i.i) M. M. EL-BORAI 550 . where --, Dt D kl k I + k + Euclidean space E (-i) k k D I L-- Ikl k n n ...D 1 D Lk(t,x) < 2b )X r r D k 1 n r kn, t (0,T), T>0, x is an element of the n-dimensional n’ and (Lk(t,x) Ikl <- 2b) is a family of square matrices of order N. We assume that (i.i) is a strongly parabolic system on 0,T], t every x En 7 En’ {(t,x): Gn+I and every (t,x) 6Gn+I Lk(t,x) o Re Ikl aN), (a I in the sense that for every complex vector a 2b k where k ’lal 2 aS + olk k n n I + a, 2 2b n and is a positive constant 2)b (see [I]). In the above inequality and in-the following, we denote the scalar product of two N-vector functions u and v by the bracket notation (u,v). As usual, we denote by Cm functions defined on E n 0 < m < =, the set of all real-valued (En) which have continuous partial derivatives of order up to and including m (of order < set of all matrices (a) (h ector functions h with compact support, r Lk(t,x), Ikl < if m I N. i, =). N) h By Cmo (En’ N) we denote the such that every h r is in We assume that the elements of the 2b, satisfy the following conditions: They are bounded on Gn+I and satisfy a Holder condition of order with respect to x, (0 < a < i). (b) For every x En Cm(En they are continuous functions in t [0, T]. 551 GENERALIZED RANDOM PROCESSES AND CAUCHY’S PROBLEM (c) (u functions. 0, T], they are C (E) n For every t in u I N) satisfy the initial condition u(x, o) where UoN), (Uol u o [u r u or We say that u is of the class S(E UrC2b(En) Let u n) o 6 C(E (1.2) (x), n) are bounded on E if for each t (0,T), n N]. r--1 Dtur C(En) and N. i It has been proved [ 2] that, under conditions (a) and (b), there exists a fundamental matrix Z(t,0,x,y) of the system (i.i) such that I u(t x) Z(t 0 x y) u (y) dy, dy o dYl dy (1.3) r E n represents the unique solution of the Cauchy problem (i.i), (1.2) in the class S(En). Let (V r r N) be a family of Gaussian random measures in the i 2]. sense of Gelfand and Vilenkin defined on E Ig r (s)12 1 We say that gr gr Let be a complex-valued function is of the class Kr if the integral dF (s) exists, where F is a positive measure such that r r E for any two Borel sets B Fr(B I B 2) [Vr(BI) Vr(B2)] I and B 2 o the real line [r i, N and E (.) denotes the expectation of (.)]. Let H be an N-vector of generalized stochastic processes, which associates with every h Co (En, N) an N-vector of random variables defined by Hr(h) (h)) (HI(h), H(h) E gro(S) 1 d Vr(S), (1.4) 552 M. M. EL-BORAI gro(S) I (I r(x,s), h(x)) dx, E n where (I on E I, r r N) is a family of N-vectors of continuous functions n+l" It is assumed also that all the components of I dependently of s. Clearly, gro r are bounded on E n in- Kr. is of the class The theoretical development in section 2 exhibits the use of formula (1.3) in order to integrate (i.i) when the initial condition is an N-vector of general- ized stochastic processes, which is defined by (1.4). Also, some essential properties are derived in section 3. 2. GENERALIZED STOCHASTIC SOLUTIONS. An N-vector w(t,x,s) of functions is said to be of the class C(En+I, N) if, for each t in (0,T), the components of w(t,x,s) represent continuous functions En+I and of (x,s) on they are bounded on E n the generalized stochastic vector H [Sr(t,x,s) Co (En, N), Ht(h) family H t(h) S C’ C(En+l (gl We say that is of the class V if there exists a N), r N-1 such that, for each h in 1, can be represented in the form J E g t independently of s. g(t,s) dV(s), 1 gN )’ gr(t’s) I h(x))dx, (Sr (t ,x, s) E n H (h) t (Hit(h)- N--H’-t(h)) (h) H rt )| gr E It is clear that, for each t in (0,T), given by grl E Kr. (t,s)dV (s). r 1 The expectation of .IHrt’12 is GENERALIZED RANDOM PROCESS AND CAUCHY’S PROBLEM I IHrt 12 E E If d d--{ Dtgr(t,s)__ Hrt (h) . Ikl for each t in (0,T), then we define E d V r (s) I Dt E 1 gr(t,s) At,s) Agr (t’s) gr(t s) d Vr(S) 1 and l.i.m, denotes limit in the mean, Dtgr(t,s) At 2 dFr(S) 0. 1 IklDk Lk,, (-1) < 2b adjoint matrices to At I t->o . E gr(t + lim L are C Kr Agr(t’ s) I l.i.m t/0 Agr(t,s) e Fr(S). d by d- Hrt(h) Let 2 1 exists and belongs to d where Igr(t,s) 553 (Lk, Ikl - < where 2b). (Be, Ikl < 2b) is the family of Since the coefficients of the operator (En) functions, it follows that, for every h in C (E o n is also in C o _(En,N). We call H dH system (1.1) if H t and t d-- dH Co (En,N) h are of the class V and h) t H t(h t) and t in (0,T). H (h) o (2.1) We assume that H(h) (2.2) where H is defined by (1.4). THEOREM i: The Cauchy problem (2.1), (2.2) has a unique generalized stochastic solution H PROOF: t in the class V. Let (S (t,x,s) r t a generalized stochastic solution of the t d--{-- for every h in N), L h r i, system (I.i) with the initial conditions" N) be a family of solutions of the 554 M. M. EL-BORAI S (O,x,s) r I (x,s), r N. i r (1.3), one gets Using formula Sr(t,x,s) Ir(Y,S) | Z(t,O,x,y) E (2.3) dy. n According to the properties of the fundamental matrix Z, we find r N. 1, Sr E C(En+I’ N), Set, J Ht(h)-- E g(t,s) dV (s) 1 and gr(t’s) I E where Sl(t,x,s), D t [ E are defined by (2.3). N), (Sr(t,x,s) Sr C(En+I, Since I h(x))dx (DtSr(t,x,s) h(x))dx E n n (Sr(t,x,s), h *(x) t dx. E The last formula proves that D grl t 6 Kr. Now we already have d dt Ht(h) (Sr(t E where d H t 1 E We also have H (h) o x,s), h (x)) dx dV(s) t n is of the class V. . E g(O,s) dV(s) 1 N), Using again the properties of Z, we get is of the class V. t o (En’ n SN(t,x,s) it follows that H h(x))dx with hl E C (Sr(t,x,s), H t (h) GENERAL RANDOM PROCESSES AND CAUCHY’S PROBLEM 555 where I gr(O’s) E (Ir(X,S), h(x)) dx. n Thus the existence of the generalized stochastic solution H condition H H is proved. o To prove the uniqueness of H H(h) then E [Hrol 2 t(h) [gr(S) 12 it is sufficient t o Co (En,N) 0 for every h in I E with the initial (2.1) with the initial condition H (h) to show that the only solution of 0 is H t dF(s) and t in (O,T). O, and hence gro (s) If O on E Ho O, 1 1 Therefore, gro(S) I (Ir(X,S) h(x)) dx O, E which is true for any arbit=mryh in Since d H t(h) H t(h t), Co (En, N), and hence Ir(X,S) 0 on En+I. it follows that therefore, I (DtS r(t,x,s) L S (t,x,s), h(x)) dx r O, E which implies DtSr(t,x,s) L Sr(t,x,s). (2.4) We also have s r (O,x,s) The uniqueness of the problem o. (2.5) (2.4), (2.5) gives s r (t,x,s) 0, (2.6) 556 M. M. EL-BORAI (O,T), (x,s) E t Using (2.6), one gets H t(h) En+I, I (r O, for every h in C N). =o (En’N) and t in (O,T). This completes the proof. A CONVERGENCE THEOREM. 3. (h Let h m m h ), m-- i, 2, m N I’ G is a bounded open domain of E Wr E L2(G ), r w (x)) r N and L i, xG If H THEOREM 2: dx (3.1) 0 It is assumed that w (x) r 0 N. 1 where r (G, N), where (G) denotes the set of all Lebesgue 2 measurable square integrable functions on G. for o Suppose that n lim | (h (x) m J r where be a sequence in C t(hm) I gm(t’s) dV(s), then l.i.m. m+ where gm(t’s) (gml(t,s) r(t, s) is defined by PROOF: t m) J q(t,s) dV(s), gmN(t,s)), I (Sr(t’x’s) gmr (t,s) J H (h (Sr(t,x,s), h (x)) dx, q m (ql qN w(x))dx, and the family (Sr, r I, N) (2.3). A straight forward application of the Cauchy Schwarz inequality establishes that lira I (Sr(t,x,s), hm(X) I (Sr(t,x,s), dx G G w(x))dx (3.2) GENERALIZED RANDOM PROCESSES AND CAUCHY’ S PROBLEM According to the conditions imposed on the family (Ir(X,S) r i 557 N) and according to the properties of the fundamental matrix Z, we can find a constant A such that Igm < (t,s) (3.3) A, r for all m, s, t (O,T) and r For any positive integers N. i and m, we have E I Hrt(h )12 Hrt(hm) Igmr (t,s)- gr (t,s) 2 dFr(S). By a standard argument based on (3.2) and (3.3), the righthand side of (3.4) can m) Thus, H (h be shown to go to zero. t is a Cauchy sequence. We deduce also that lim I Igmr (t,s) r(t,s) 2 dFr(S) O. The last argument leads to the fact that there exists a stochastic process R (t) such that E r IR (t)l 2 E lim Following Doob and that < IHrt (hm) r(t) 12 R 0 3], we find J r (t s) R (t) r r(t,s) dV (s) r J| (Sr (t ,x, s) w(x))dx. This completes the proof. COROLLARY: w (x) r 0 for x For vector functions (w w N) where in C G), there exists a sequence (h) m (En, l.i.m. m/ H (hm) o w I H o(w) w , L2(Q) and N) such that M. M. EL-BORAI 558 i.i.m. H m/= t (hm)=H t (w). The proof can be deduced directly by using theorem 2. (Compare []). REFERENCES [i] C.D. Idelman, Parabolic Systems. P. 87, Moscow 1964, (Mir Publishers). [2] N.Ya. Vilenkin, I.M. Gelfand, Generalized Functions, Vol. 4, 1964 (Translated). [3] J.L. Doob, Stochastic Processes. [] M.M. Ei-Borai, On Stochastic Differential Equation in a Hilbert Space, Egyptian Statistical Journal Vol. 1__7, No. 2, 1973. New York 1953.