131 31 38MB
English Pages 714 [720] Year 1987
PROBABILITY THEORY AND MATHEMATICAL STATISTICS
Volume II PROBABILITY THEORY AND MATHEMATICAL STATISTICS Proceedings of the Fourth Vilnius Conference
Vilnius, USSR, 24-29 June 1985
Edited by
Yu. V. Prohorov, V. A. Statulevicius, V. V. Sazonov and B. Grigelonis
WWNU SCIENCE PRESSICI
Utrecht, The Netherlands
V N U Science Press B V P . O . B o x 2093 3500 G B Utrecht T h e Netherlands
© 1987 V N U Science Press B V First published 1987 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of the copyright owner.
CIP-DATA K O N I N K L I J K E B I B L I O T H E E K , D E N H A A G Probability theory and mathematical statistics: proceedings of the Fourth Vilnius Conference: Vilnius, USSR, 24—29 June 1985/ed. by Yu.V. Prohorov . . . [et al.] — Utrecht: VNU Science Press. — 111. ISBN 90-6764-068-9 (vol. II) ISBN 90-6764-069-7 (set) SISO 517 U D C 519.2(063) Subject headings: probability theory/mathematical statistics.
Printed in G r e a t Britain by J. W. A r r o w s m i t h Ltd., Bristol BS3 2 N T
CONTENTS Preface Mixing properties for/-expansions M. losifescu Non-linear time-domain analysis of Gaussian processes Z.A. Ivkovic Malliavin calculus for discontinuous processes J. Jacod Accompanying laws for processes with independent increments A. Jakubowski Diffusion approximation of multitype branching processes A.Joffe Characterization of queues and its stability estimates V. V. Kalashnikov and S. T. Rachev On the entropy of expansions with odd partial quotients S. Kalpazidou On the rate of convergence in Maruyama's invariance principle S. Kanagawa On a general growth model possessing a subexponential growth rate G. Kersting Stochastic partial differential equations as stochastic space-time models P. Kotelenez Levy-Baxter type theorems and strong sub-Gaussian random processes Yu. V. Kozacenko and V. V. Buldygin Principal component analysis under correlated multivariate regression equations model P.R. Krishnaiah and S. Sarkar Approximation of statistics distribution by convolutions of generalized Poisson measures J. Kruopis
vi
Contents
Some general results in control theory N. V. Krylov
129
On the rate of of distributions of semimartingales K. Kubilius andconvergence R. Mikulevicius
139
On Ito's excursion law, local times and spectral measures for quasidiffusions U. Kuchler
161
Optimum design of general intraclass regression experiments and general analysis of covariance experiments V. G. Kurotschka
167
Limit theorems for functionals of geometric type of homogeneous isotropic random fields N.N. Leonenko
173
On bounds B. Ya. Levit for the minimax risk
203
Asymptotical properties of the local density of measures for semimartingales and some of their applications Yu.N. Lin'kov
211
Generalizations A.I. Martikainen of the law of the iterated logarithm
235
Limit theorems for sums of random variables with a stable limit law J. Mijnheer
253
Estimates and asymptotic expansion of the remainder term in the CLT for randomized decomposable statistics Sh. A. Mirahmedov
263
Multiparametric G.M. Molchan Brownian motion on symmetric spaces
275
Markov maps in noncommutative probability theory and mathematical statistics E.A. Morozova and N.N. Cencov
287
Classes of limit laws for functions of some statistical estimates R. Mukhamedkhanova
311
Topological and probabilistic characterizations of some classes of Banach spaces and operators D.H. Mushtari
327
Contents
vn
Asymptotic expansions of the distributions of sums of i. i. d. Hilbert space valued random variables S. V. Nagaev and V. I. Chebotarev
357
Limit theorems for order statistics based on sums of random variables V. B. Nevzorov
365
Markov additive processes: large deviations for the continuous time case P. Ney and E. Nummelin
377
Hodges-Lehmann efficiency of nonparametric tests Ya. Yu. Nikitin
391
An operator approach in limit theorems for sums of the type n~ "p S„ in Banach spaces R. Norvaisa, V.J. Paulauskas and A. Rackauskas
409
Pointed priors and asymptotics of the a posteriori risk A.A. Novikovandl.N. Volodin
421
Stokes-Boussinesq-Langevin equation and fluctuation-dissipation theorem Y. Okabe
431
The barycenter concept of a set of probability measures as a tool in statistical decision A. Perez
437
Solutions of Bogolyubov equations for infinite three-dimensional systems of particles D. Ya. Petrina, V. I. Gerasimenko and P. V. Malyshev
451
Inequalities for the maximum of partial sums of random variables and the law of the iterated logarithm V. V. Petrov
461
A characterization of a Gaussian vector based on Kagan-Linnik-Rao theorem A. Plucinska
467
Extreme functionals in the space of probability measures S. T. Rachev
471
viii
Contents
A N O D I V : generalization of A N O VA through entropy and cross entropy functions C. Radhakrishna Rao
477
Limit theorems involving restricted convergence and the continuation theory of distribution H.-J. Rossberg
495
On approximation accuracy for distribution functions of the sum of independent random variables using infinitely divisible distributions L. V. Rozovsky
501
On the kinematic dynamo problem in a random flow B.L. Rozovskii
509
On the probability of large excursion of a nonstationary Gaussian process R.Rudzkis
517
Weak convergence of integral type functionals Z. Rychlikandl. Szyszkowski
525
Characterizing the distributions of the random vectors X,, X 2 , X 3 by the distribution of the statistic (X t —X3, X 2 - X 3 ) Z. Sasvari and W. Wolf
535
On large deviations for the probability density of sums of independent random variables L. Saulis
541
On normal approximation in Hilbert space V. V. Sazonov, V. V. Ulyanov and B. Z. Zalesskii
561
Almost-even number-theoretical functions W. Schwarz Limit theorems for randomly indexed sums in a separable Banach space G. Siegel
589
Some properties and applications of Feynman measures in the phase space O. G. Smolyanov and E. T. Shavgulidze
595
581
Contents
ix
On limit theorems for multilinear forms V.A. Statulevicius
609
Characteristic functional and cylindrical measures in DS-groups V. I. Tarieladze
625
On Pareto-type distributions J.L. Teugels
649
Renewal method in the theory of semi-Markov processes on arbitrary spaces J. Tomkd
653
Innovation problem for a class of Ito processes: filtration problem for multi-dimensional diffusion type processes T.A. Toronjadze
659
On Gaussian approximation in Hilbert space, non-identically distributed random variables case V. V. Ulyanov
671
Statistics on the set of the naturals' partitions, limit forms of young random diagrams and asymptotic distributions on the space of positive series A.M. Vershik
683
Bellman inequalities for Markov decision drift processes A.A. Yushkevich
695
Coexistence of low temperature continuous spin Gibbs states M. Zahradnik
703
PREFACE T h e International Vilnius Conferences on Probability Theory and Mathematical Statistics have been held every four years since 1973 and have become traditional. However, their proceedings were never published. Only a limited number of copies of the abstracts of contributions were available, published locally. T h e present publication makes it possible for a wide international scientific community to become familiar with the invited lectures delivered at the 4th Vilnius Conference which was held from 2 4 - 2 9 J u n e , 1985. T h e main topics of the conference were (controlled) random processes and fields, limit theorems of probability theory, and asymptotic methods of statistics. T h e organizing committee of the conference invited all the 45- and 30-minute speakers to present a paper to the Proceedings. W e included in the Proceedings all of the papers received. T h e r e were 453 participants at the conference of whom 150 were invited speakers. W e hope that the 4th Vilnius Conference and its Proceedings will give rise to many important and interesting investigations in the future. Y u . Prohorov
xi
Prob. Theory and Math. Stat., Vol.2, pp. 1-8 Prohorov et al. (eds) 1986 VNU Science Press
MIXING PROPERTIES FOR/-EXPANSIONS Marius Iosifescu Centre of Mathematical
Statistics,
str. Stirbei
Vodâ 174, 77104 Bucharest,
Romania
Abstract—Several c o n d i t i o n s o c c u r r i n g in t h e s t u d y of piecewise m o n o t o n i e t r a n s f o r m a tions a r e discussed. U s i n g the F r o b e n i u s - P e r r o n o p e r a t o r c o n d i t i o n s a r e given u n d e r which either reversed (¿»-mixing o r ^ - m i x i n g hold for / - e x p a n s i o n s .
1. P I E C E W I S E M O N O T O N I C T R A N S F O R M A T I O N S Let T: U-*I= [0, 1] be a continuous m a p , where C / c / i s open and A(£/) = 1 (¿ = Lebesgue measure). Then there exists a finite or countable collection (I„)asA of closed intervals with disjoint interiors such that l j a e / , laz> U, and for any aeA the set Iar\([\U)consists exactly of the endpoints of Ia. Assume that for any aeA the restriction of T to / f l n U is strictly monotonic and extends to a C 1 -function Ta on la with TJIJ = I. A m a p 7" with the above p r o p e r t i e s is c a l l e d a C1
piecewise
monotonic
transformation.
For any a e A let fa denote the function inverse to Ta, thus mapping I onto Ja. F o r any aM = (al,
. . . , a„)eA",
put
/.«=/.; ' ' ' °fa„ (o denotes composition of functions). Clearly ,/„ 0, J 7 r dX = 1 and Pr = r, i.e. p is ^-invariant. If p' with dp' = r' dX is ^-invariant, then r' e BV(I) and p' is absolutely continuous with respect to p. Assume from now on that condition (C) holds. [Remember that, on account of Proposition 3, (C) implies (E m ) for some m.] It is well known (Renyi, 1957; Halfant, 1977) that under (C) there exists a ^-invariant, ergodic probability measure p which is equivalent to X. The density r = dp/dX of p satisfies almost everywhere on /the inequalities 1/C^r^C.
(3)
4
Mixing properties for /-expansions
Moreover, as shown by Rohlin (1961), the dynamical system (I,38, p, T) is exact. Then Theorem 4 implies uniqueness of a probability density h (with respect to p satisfying the equation P-ph = h, the only one thus being h = 1. As to P-p it is easily seen that Pr h =
r
\
heLl(I,
3), p).
[Clearly, on account of (3), the elements of the spaces Ll(I, 38, p) and Ll(I, 38, 1) are identical. We shall simply write 11 - J11 for the norm in any of these spaces whenever confusion is not possible.] Finally, an easy computation shows that assuming in addition condition (BV), the operator Ps satisfies inequality (2) for h e B Vp (I) as P does for h e B V(I), with suitable k, d and D. [The definition of BV-JJ) on analogy of that of BV(I) = BV>(I) is obvious. The remark above concerning Ll(I, p) and Ll{I, 38, X) applies to BVJ1) and BV(I) too.] Thus P-p enjoys properties (i) through (iii) of P. It appears that p is in fact p defined above, and r = reBV(I). Also, s = l and ). l = 1 is the only eigenvalue (of multiplicity 1) of modulus 1 of both P and P. We can therefore state Theorem 5. Assume conditions (C) and (BV) hold. Then for any h e BVp([) and n ^ 1 we have IIP"h -
h dp|| 0 and 0 < q < 1. The proof follows from the inequalities l|, ^ess sup |/i(x)|
for suitable positive constants c, and c 2 not depending on /"". But X
p(/.«) = p(7)= 1,
/c>l,
and sup
£
var /a'(k> < oo
on account of Proposition 1 of Rychlik (1983, p. 73) (take there/= 1). Finally, noting that p({rk} oo to the standard Brownian motion process.
M. Iosifescu
7
4. S O M E C O N C L U D I N G R E M A R K S As has already been mentioned, Renyi's condition (C) was originally introduced for /-expansions. Subsequently, it was dealt with for piecewise monotonic transformations (Adler, 1979; Halfant, 1977). The restrictive assumption Ta{Ia) = / , aeA, is precisely reminiscent of/-expansions, where it is quite natural. In fact, it is not at all essential and one can instead assume that inf
A(TJ/a))>0
aeA
which always holds when A is a finite set. Without such an assumption there can be no absolutely continuous T-invariant measure (cf. Bugiel, 1985). It is certainly interesting to note that, on account of Proposition 1 here, the assumptions in Theorems 1 - 3 of Lasota and Yorke (1973) imply Renyi's condition (C). Thus these theorems d o not improve Renyi's result concerning the existence of a Tinvariant density, save the additional information that this density is a function of bounded variation. The great merit of Lasota and Yorke's paper lies in the new approach it has initiated. The relevance of Ionescu Tulcea and Marinescu's ergodic theorem in this context has been first noted by Keller (1979). This theorem is motivated by and in turn is of fundamental importance in the theory of dependence with complete connections. F o r piecewise monotonic transformations conditions (C) and (BV) ensure the absolute regularity of the label sequence ( a J •))„;>,. Under just (E J for some m and (BV) the only thing which can be asserted is the existence of a natural number k such that (akJ is absolutely regular [cf. Hofbauer and Keller (1982, p. 128)]. It would certainly be interesting to know how (C) and (BV) should be strengthened in this general setting for to get for the label sequence the types of mixing occurring in Theorem 6. Finally, Theorem 6 was stated under stronger assumptions by G o r d i n (1968), who gave no proof. I d o not know of any other instance in which reversed Oy
Suppose
that
in
Ivkovic
^ " H f )
is 0
in
i
F ( t ) = inct)||2 = ) l V t ) H 2 + l h d ( t ) l f = F c Ct) t Fd Ct) the discrete
part
i,J | l M f wherej T T is the spaee of laws of processes with independent increments /PII/ and trajectories belonging to the Skorokhod space D ( R + : R 1 ) ; it is additionally assumed that if k ( k ) .
will denote the scalar product of the vector
Theorem I.
Let
Z ^
x
and
y .
be a sequence of multitype Galton-Watson
processes such that (i)
= I + 1 C n
(ii) lim
= c, ,
n
I = 1
with
lim C n-*»
d ,
= C
ct^ s K
for some
1 J
n+co
a
K ,
i == l,...,d , (iii) lim I (r.-mj n J) 2 p| n ) (r) = 0 , for any L n r:||r||>e /n 1 ^ i = l,... d
e>0
Then, given a sequence 3 ^
the
e
such that
lim
X n (0) = x < n )
,
n
ne
n
= a , a > 0
processes X< n >(t) = e / ^ C n t ]
,
lim x< n > = x Q , x q * 0
will converge weakly to the unique diffusion2 process in R + d starting at Xq with generator given on C functions by: d
L :=
I i=l
(xC)
1
a X
i
d l 1« ^ * 1= 1
X'a,
a2 3x£
.
A. Theorem II.
Let
Zn
33
Joffe
be a multi type critical positive regular
Galton-Watson process with mean matrix ck j .
Let
R
and covariance matrices
be the projection matrix in the Perron-Frobenius
decomposition of / \ / \ X^ '(0) = XQ '
M
M .
Then if
X^(t)
converging to
= e z" , n LntJ
XQ * 0 , ne n
with
a , a > 0 ,
the
sequence ?(n)(t)
=
x(n)(t)
.
n
f
x(n)(x-)(M-I)dAn(T)
n Jo will converge weakly to the unique diffusion in the half space
xR a 0
with diffusion operator
L
given by
Lcp = aXR tt Y a. . ~ — ( a l l 2 .. ij 8x i 8x j and initial condition
3.
tp C -functions on
xR a 0)
Xq .
CONTINUOUS TIME
The model.
We consider a system of
d
types of particles.
denotes the rate of death of a particle type of type P.j(j) = P q -(j 1 -• • .i d ) particles of type Z(t) = ( Z ^ ( t )
i
the probability that when it dies
l,...,d
are created.
Z^(t))
number of particles of type
A.
and j-j
jd
Let
denote the random vector of the l,...,d
in the system at time
t .
We use the following notation
V.£(i)
=
I
oj » = (a „(l)...a „(d)) , in yAi iiiji, m $ JL
' al =
where
a
T
t,lW
denotes the transpose
Multitype branching processes
34
e^ = (6-j j,... ,6 d j)
the
jth
unit vector
a = A(M-I) N+d
the set of
d-dimensional vectors whose components are non-
negative integers. Under those assumptions it can be shown that this system defines a unique Markov process
Z(t)
We consider now a sequence
whose state space is (Z^(-))n>Q
the above notations with superscript n
n
X< )(t) = e n Z^ ^(nt)
of such procseses using
n .
We introduce the sequence
n
(noting that the
.
Z (nt)
are as
Z n (t)
but
with an increasing intensity of jumps which come to the same as accelerating the time). We shall modify the model to deal with population-size dependent multitype branching processes, generalizing the results of Lipow (n) (1977).
Keeping the above notation, we let the quantities
and
p^ '
x e N+ci .
be functions of
We make the following assumptions: (i)
sup sup A ^ ^ e ^ x ) n x
(ii)
(iii)
< °° , lim A ^ n+°°
sup sup n a n ( e ^ x ) < °° x n
= A(x) ;
lim na ( e ^ x ) = C(x) ; n-t-°°
lim sup sup I ||j||2p(n)(j,x) = 0 , I = l,...,d , N-t- n x j:|j|>N
(iv) sup sup a j n ) < - , n x
lim \ [ n ) n+°°
*) = a £ (x) ;
2 (v) Let on
CQ R1^
be the space of and
L
C
be defined on
bounded continuous functions Cn
by
A.
L =
I i=l
(xC(x)). 1
dX
i
35
Joffe
+
§
e
1 xfyix) 3x 1=1 1 1
Then the martingale problem associated with
, a > 0 .
(L,C q ,XQ) , XQ e
has a unique solution (this i s true in particular i f C(x)
and
a(x)
Theorem I I I .
d = 1
or i f
are constants).
Let
Z^(t)
be a sequence of population-size-
dependent multitype branching processes, s a t i s f y i n g the assumptions ( i ) to (v), above, and let
X^(t) = e Z^int)
with
lim c n = a , a > 0 and X ^ ( 0 ) = xi n ^ with lim x j ^ = x n , n / \ u u u x n * 0 . Then the sequence X^ J ( t ) converges weakly to the +d unique diffusion in R starting at xQ with generator L . Theorem IV.
Let
Z(t)
be a c r i t i c a l positive regular multitype
branching process, with parameters
a = A(M-I)
and
o .
Let
R
be the projection matrix in the Perron-Frobenius decomposition of exp (at) .
X ( n ) ( t ) = e n Z(nt)
Then, i f
converging to
Xg , xQ * 0
and
with
X(n)(0)=x^n)
lim ns n = a , a > 0 ,
sequence of martingales
the
t
/(n), X V " ' { T )adx 0 will converge weakly to the unique diffusion in the half space (4.4.10)
xR g 0 ,
?(n)(t) = X(n)(t) - n
with generator
L
given by
Lcp = aXRA I and i n i t i a l condition
XQ
(