I ntern. J. h. & Nath. Sci. Vol. 2 No. 4 (1979) 669-676 669 A NOTE ON LOCAL ASYMPTOTIC BEHAVIOR FOR BROWNIAN MOTION IN BANACH SPACES MOU-HSlUNG CHANG Department of Mathematlcs The University of Alabama in Huntsville Huntsville, Alabama 35807 U.S.A. (Received July 26, 1978) ABSTRACT. In this paper we obtain an integral characterization of a two-slded upper function for Brownian motion in a real separable Banach space. where B This characterization generalizes that of Jaln and Taylor n. [2] The integral test obtained involves the index of a mean zero Gaussian measure on the Banach space, which is due to Kuelbs [3] The special case that when B is itself a real separable Hilbert space is also illustrated. KEY WORDS AND PHRASES. Gaussian measur on B-space, abstract Wienzr spaces, covariane operators, Bwnian motion in B-space, upper and lower functions, integral test. AMS(MOS) SUBJECT CLASSIFICATION (1970) CODES. Primary 60J65, 60G17. M. CHANG 670 INTRODUCTION i. Let B be a real separable Banach space with norm be the topological dual of B. I’II such that that B contains a Hilbert space H II" II is a measurable norm on H II II As a consequence, the B norm of [i] is weaker than Furthermore, it is also shown in [I] that let F lxll. sup xeK [I I11 Thus B. is the extension of the to B and we shall say that canonical normal distribution on H H. in the sense H* -H through a restriction map we have the relation that B* generated by and let B* is a mean zero Gaussian measure on If B then it is well known from [i with norm II’II If K denotes the unit ball of H in the norm I’II, , The definition of index of is n I, is due to Kuelbs [3], where n sup {k: I fl llfj II and It is known from [3] that n Hilbert space then n that (t)+ . lower function for denoted by L, if e A function P(I IW(t+v) u+v < 6) Let tl/2(t)+0 as II’II, with respect to the norm with 0 < H; F (i < j < k)}. on B equals the multiplicity of the maximal eigenvalue I as t+0 and such that orthogonal in {W(t): 0 _< t < oo} denote -Brownian denote the class of functions from (0,) to [0, =) such Let DEFINITION i. fk fl’ is finite and if B itself is a I exists and of the covariance operator for motion in B. B*; fkE W(t-u) I. is called an upper function for if given t > 0, there exists a < 21/2(u+v)I/2(u+v) In this case, we say {W(t): U. ll t+0. t > {W(t): > 0 F for all u, v > 0 EU. 0} with respect to the norm is called a t > 0} 671 ASYMPTOTIC BEHAVIOR FOR MTION IN BANACII SPACES d, In the case that B e {W(t): < < } [(t)]d+2--e-2(t)/2 is an upper function for d-dimension standard Brownian [2] have shown that 0 motion if O+ a d-dimensional Euclidean space, Jain and Taylor . with respect to the Euclidean norm t dt < 11"112 if and only This integral test for two-sided growth in is the same as that for one-slded growth in d+2. In the case that B is an infinite dimensional real separable Banach space, Kuelbs (t), a nonnegative, non-decreasing t l’II I respect to if P dt < > <_ t < on B if and only if tl/m(t) F for only a bounded set of t’s) Based on the results of Jain-Taylor and Kuelbs, it is very natural to g is in U (Definition i) for equivalent norm [I-II I } is called one-sided upper function with where the (I IW(t) II I -Brownian motion {W(t): 0 l’I, ll with respect to some equivalent norm [(t)]nl e-2(t)/2 [3] has shown that continuous function defined for large values of t, is a one-sided upper function for that {W(t): on B if and only if 0 _< 10+ t < } with respect [(t)]nl+2 t main purpose of this note is to verify this conjecture. e notation a(h)~ b(h) means lid a(h) i. conjecture to some -2(t)/2 dt < . The Throughout this note c will stand for a positive number whose value may change from line to line. 2. d t The i. h-0 b(h) M IN RESULTS The ol]owing useful estimates have been used repeatedly in [3] and they can be verified by the argument similar to that in d-dimensional Euclidean space d [4, p. 222]. LEMMA 2. Let {W(t): 0 < t < Banach space B having norm P sup 0_< t I < 2P t2< sup 0< t<h h IIW(t) II II’II- be Brownian motion in a real separable Then for all %, h > 0 Iw(t# > %h I/2) W(tl) _< 4P > %h I/2) < (I IW(h) ll>%hl/2). () M. CHANG 672 We have the following integral test for a two-slded upper function {W(t): 0 < for oo}: t < Let {W(t):0 < t <} 5e -Brownian motion in a real THEOREM 3. .II separable Banach space B having norm denote the index of lx such that sup 0+ F and U with respect to II.II I Let n II.II I I on B if and only if e-2(t)/2 dt<. 2 t PROOF. ,e. and assume Then there is an equivalent norm I [ (t) ].n.!+. II" II, (2) is due to The construction of the equivalent norm which is defined to be lxll where =x rllll lQxll), n (x) 1 Y e j j=l Q(x) x (xeB) (x)ej and (x), ej (-) denotes the linear function Consider the sequence a e m as m-. Let u i n, i io{ ’h Note that for each i, u n,i -m/log m m > 2. (i an’ Vn, i + vn,i a n Then log n m am+I an’ mn,i {: lW(t+vn i) W(t-u F {: lw(t+vn W(t-u n,i i n, i) n, i i + (log m) -i 0 < i < log n. If u,v > 0 are sufficiently small we can choose an n sufficiently large such that a Assume that the integral (2) diverges. fj (’)/F. a n+l < u+v < a Define II i > 21/2an i/2(an)r} II > 21/2anI/2(an) } and n [3], ASYMPTOTIC BEHAVIOR FOR MOTION IN BANACH SPACES 673 Since {W(t) :0 _< t < oo} I dimensional Brownian motion in divergence of the integral (2) implies that H B is standard n log n n=2 F(F i=0l Fn, i and infinitely many of n’ i Consequently infinitely many of E n,i occurr with probability one. Now assume that the integral (2) converges. i < log n such that u n,i < u < u occur. n,i+l and v < v Thus eeL. Let us choose a suitable n,i-I Then lll > 2112 (u+v) ll2(u+v) F) l/2W(t-u) lll > 21/2-an+l (an+l) F) P(I IW(t+v) W(t-u) P(llW(t+v) < --< P(O < t sup < t < u 1 2 n, i+l+Vn, i-i + P (0 _< t < sup t < I 2 {zW(t): 0 < Since B H, t < Un, i+l+Vn, i-i } is standard n lz(W(t 2) W(tl))l I IQ(W(t2) W(tl))ll I > zI/2 112 an+ 1 (an+l)) > z.i/2 112 I an+ (an+l)F). dimensional Brownian motion in by the same argument as those in Theorem 3.1 [2], we conclude that the first term in the right hand side of the above inequality being zero for infinitely many n and i. As for the second term in the above inequality, we have sup P(0_<tl<t 2 _<Un, i+l+Vn, _< 4P < C i_l lQ(W(t2 (I IQW(1) II >(u 2an+I n, i+l+Vn, i-i exp{-e 2an+l W(tl))ll > 1/2 F2} n, i+l+Vn, i-i 2 (an+l) (an+l)F) 1/2 21/2 an+ I (an+ I) F) P(Ano i) M. CHANG 674 Where 1/2 sup B Q(dx), A2(x) n l]x]t and {A } we choose (i be such that P(An, i) _< C exp {-6 n + )/2F + 3/log (i n)an+I, if we have 2(an+l)} exp {-2(an+l)} {-qb2 (an+l) < C exp for every xeB. 3, p. 253]. (i + 2/log n)a Un,i+l+Vn,i_l Aj(x)I sup The last inequality comes from Since is a sequence in B* such that j } Thus log n 7. Y. P(An, i) < c Y. (log n) exp n--2 i--0 {-2(an+l)} n=2 {-2(an+l)} [(an+l)]nl < c exp n=2 < since the integral (2) converges , (see Lemma 2.12 of [2]). From Lemma 2.15 (i) of [2] we conclude that P(An,i, i.o.) 0. Thus P( sup t-Un, i+l--<tl <t2<t+Vn, i-ll-- lW(t2) W(tl)l- II for all i and n sufficiently large. Thus < z,,1/2 1/2 an+I (an+l)F) eU. In case that B is a real separable Hilbert space, then n I equals the multiplicity of the maximal eigenvalue of the covariance operator for V. We have the same result as those of Theorem 3. THEOREM 4. Let {W(t): 0 < t < } be v-Brownian motion in a real separable Hilbert space H with norm II’II, and suppose e. Then i 675 ASYMPTOTIC BEHAVIOR FOR MOTION IN BANACH SPACES II’II is in U with respect to the given norm [(t) ]nl+2 S04 Where n e t _2(t) / 2 if and only if dt < oo, (3) . I denotes the multiplicity of the maximal eigenvalue of the covariance operator for } and {v } be the same as those in n, i n,i If u,v > 0 are sufficiently small we choose Let sequence {a PROOF. the proof of Theorem 3. n }, {u n sufficiently large such that such that u n,i < u < u an+4+/- _< u+v and v < v n,i+l < an and then fix i _< log n If the integral (3) diverges, n,i-I then we proceed as those in Theorem 3 and conclude that L. Now if the integral (3) converges define B n, i sup {: t-Un, i+l<--tl<t2<--tqVn, i IW(t2) W(tl) 21/2an+11/2 > (an+l) F}. Then e(Bn,i) < 4e(llW(1) ll < C[(an+l) _< C[(an+I) 2an+I > (un, i+l+Vn, i-i nf2 exp { (- (an+1 an+l2 (an+l) F2 ni 2 exp 1/2 (Un, i+l+Vn, i_l) % _2 (an+l) } Where % is the maximal eigenvalue of the covarience of that % F2 [3]. n, i+l+Vn ,i-1 and it is known The last inequality comes from [3] and the fact that a u ) n+l 1+3 (log n) -1. Thus log n Y. n=2 i=O P(Bn,i < C log n Y. Y [ n=2 i= 0 I)_ (an+ n C Y. n=2 exp {- 2 (an+l) } 2(an+1 )} < since From Lemma 2.15 (i) we have P(Bn, i, [(an+I)] the integral (3) converges. I nl-2 exp -i- i.o.) 0. 676 M. CHANG That is P(B c n, i) i for sufficiently large i and n. Thus U. REFERENCES i. Gross, L. Abstract Wiener Spaces Proc. Fifth Berkeley Symposium on Math Statistics and Probability, (1965) 31-42. 2. Jain, N. C. and Taylor, S. J. Local Asymptotic Laws for Brownlan The Annals of Probabillty Vo.l. i, No. 4 (1973) 527-549. 3. Kuelks, J. Sample Path Behavior The Annals of Probability, 4. Motion. for Brownlan Motion in Banach Spaces. No. 2 (1975) 247-261. Vol. 3 Taylor, S. J. Exact Asymptotic Estimates of Brownlan Path Variation. D_uke Math. J.. 39, (1972) 219-241.