A NNALES DE L ’I. H. P., SECTION B
S HINTARO N AKAO
On weak convergence of sequences of continuous local martingales
Annales de l’I. H. P., section B, tome 22, n
o3 (1986), p. 371-380
<http://www.numdam.org/item?id=AIHPB_1986__22_3_371_0>
© Gauthier-Villars, 1986, tous droits réservés.
L’accès aux archives de la revue « Annales de l’I. H. P., section B » (http://www.elsevier.com/locate/anihpb) implique l’accord avec les condi- tions générales d’utilisation (http://www.numdam.org/conditions). Toute uti- lisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
On weak convergence of sequences of continuous local martingales
Shintaro NAKAO
Department of Mathematics, Osaka University, Toyonaka, Osaka 560, Japan
Ann. Inst. Henri Poincaré,
Vol. 22, n° 3, 1986, p. 371-380. Probabilités et Statistiques
SUMMARY. -
Let {Mn(t) ~n-1,2,...
be a sequence of Hilbert space valued continuous localmartingales.
Wegive
a necessary and sufficient condition forwhich {
istight
in C in termsof ( ( }. Using
this result weshow the
preservation
of the localmartingale property
under the weak convergence in Cof {
Mots-clés : Continuous local martingale. Weak convergence.
RESUME. -
Soit { Mn(t) ~,~=1,2,...
une suite demartingales
locales conti-nues a valeurs dans un espace de Hilbert. Nous donnons une condition necessaire et suffisante pour
que {
soit tendue dans C a l’aide d’une condition de tensionsur ~ ~ }.
A l’aide de cerésultat,
nous montronsque la
propriete
demartingale
faible locale est conservee par la convergence faible dans C.1. INTRODUCTION
Let H be a real
separable
Hilbert space with innerproduct (,)
andbe an orthonormal basis of H. Consider a
sequence { Mn(t) ~n-1,2,...
Classi,fication AMS : 60 G 44.
Annales de l’Institut Henri Poincaré - Probabilités et Statistiques - 0246-0203
Vol. 22/86/03/371/ 10 ,’~ 3,00/~ Gauthier-Villars 371
372 S. NAKAO
of H-valued continuous local
martingales starting
at0,
where the time interval I is[0, T] ] (0
T +oo)
or[0,
+oo).
Denoteby dia ~ (t)
the vector of all
diagonal components
of( ~ (Mn, ek), (Mn, el) ~ where ( (Mn, ek), (M~‘, el) ~
is thequadratic
variational process of themartingales (Mn, ek)
and(M~‘, el) (cf. [3 ] [6 ] [9 ]).
We thenregard dia ; Mn ~
as an
l1-valued
continuous process in case that H is infinite-dimensional.Rebolledo
[11 ] proved, using Lenglart’s inequality [8],
that in caseof H = f~ the
tightness
in Cof { Mn(t) ~n-1,~9.,_
and thetightness
in C of~ dia ~ M~‘ ~ (t) ~n-1 ~2,... { _ ~ M~‘ ~ {t) ~n= n2,...)
areequivalent
to eachother,
where C is the space of continuoussample
functions. A main purpose of this article is to show the aboveequivalence
holds even if H is infinite- dimensional(Theorem
A in Section2).
By applying
Theorem A we show that any accumulationpoint
of~ ~,~-
~,2,", under the weak convergence in C is also a continuous localmartingale (Theorem
B in Section3).
Moreover thecontinuity
ofthe
mapping dia ( )>
under the weak convergence in C is stated in Theo-rem C in Section 3.
In
Appendix
wegive
anelementary proof
of Theorem A in case of H == [R which is bestadapted
to the situation where all the processes consideredare continuous and which is based on
change
of time.2. TIGHTNESS
OF ~ M’~ ~ AND { dia ~ M~‘ ~ ~
For a real
separable
Banach spaceE,
setCo(I, E) _ ~
w: I ---~E,
w is continuous andv~(o)
=0 }
and we define the usual metric on
Co(I, E) (see [5 ]);
thatis, {
convergesto w if and
only if {
convergesuniformly
to w on eachcompact
interval in I. LetE))
be thetopological 6-algebra
ofCo(I, E).
We denotethe space of all
probability
measures on(Co(I, E), E))) by ~(Co(I, E))
and endow
~(Co{I, E))
with the weak convergencetopology.
Consider thefollowing subspace
of~(Co(I, E)) ;
is continuous local
martingale
for every/
EE’ j,
where E’ is the dual space of E.
Annales de l’lnstitut Henri Poincaré - Probabilités et Statistiques
373
WEAK CONVERGENCE OF CONTINUOUS LOCAL MARTINGALES
For an orthonormal of
H,
define themapping ~):
Co(I, H) -~ Co(I, l2(J)) by
(2.1) ~(w)(t) _ ((w(t),
J for w ECo(I, H), t E I ,
wherefor p
= 1. ?Then it is obvious that
is a
homeomorphism . According
to themapping ~,
we introduce themapping C:
by
We have then
by (2 . 2)
is a
homeomorphism
and
is a
homeomorphism.
We denote
by w(t) =
thesample path
ofCo(I, l2(J)).
Since the0 0
process
(dia ( w ), P) = (( ( P)
is all(J)-valued
continuous process0
for P e
l2{J))),
we can define themapping dia ( ):
in the
following
manner; for P E12(J)))
0
(2. 6) dia ( P )> ==
theprobability
measure onCo(I, ~1(J))
inducedby
0
(dia w X P).
We
get
thefollowing
theorem about thetightness
of afamily
of H-valued continuous localmartingales.
THEOREM A. - Let
Pa E H)) (a
EA).
Then thefollowing
threeconditions are
equivalent
to each other.i )
istight
inCo(I, H) .
ii) { ~(Px) ;
a EA ~
istight
inCo(I, r2(J)) . iii ) { dia ~(Pa) ~ ;
a eA }
istight
inCo(J, l l (J)) .
Vol. 22, n° 3-1986.
374 S. NAKAO
As stated in the introduction Rebolledo
[77] proved
the above theoremusing Lenglart’s inequality
in case of H = ~. So we will prove the above theorem in case of dim H = + oo. For this purpose we estimate the tail behaviorsof { ~(Pa) ;
a EA ~ and { dia ( ~(Pa) ; uniformly
in oc e A.Let K be a subset of P
( p
=1, 2).
It is well known( [4 ])
that K isrelatively compact
if andonly
ifLet I =
[o, T ]
andQa
Elp)) (a
EA). Noting (2 . 7),
it is easy to seethat { Qa ;
a EA }
istight
inCo(I, lp)
if andonly
if for any e > 0 and i== 1,2,...
there exist a
compact
set K ~ lp and03B4(i)
> 0 such thatand
where w =
(wl,
w2,... )
ECo(I, lp).
From now on we denote
by
w =(wi, w2, ... )
thesample path ofCo(I,
We prepare a lemma for the
proof
of Theorem A.0
LEMMA 2.1. - Let I =
[0, T] ]
andl2)) (a
EA).
Then thefollowing
twoproperties
areequivalent
to each other.i )
For any 6 > 0 there exists acompact
set Kc l2
such thatii )
For any £ > 0 there exists acompact
setK c ll
such thatProof
- We show thati ) implies ii).
Assume that(2 .10)
holds. SetSince it holds that
Annales de l’Institut Henri Poincaré - Probabilités et Statistiques
375
~
WEAK CONVERGENCE OF CONTINUOUS LOCAL MARTINGALES
we
get by Chebyshev’s inequality
On the other hand it is easy to see that
00
~r=1 ~2~...
be a sequence such thatE/2.
Thenr = 1
in view of
(2.13)
for any r =1, 2,
... thereexists nr
such thatPutting
we
get
from(2.14)
r
B~
We then set
={
c= (cl,
c2,... )
E c~l1 ~ 203B2/~, 03A3 |cm| ~
Erfor r =
1, 2,
...}. Obviously K
is compact and satisfies from(2.12)
and(2 .15)
Consequently
weget
from(? .10)
We can prove the
implication ii)
-~i)
in the same way as above. D We now return to theproof
of Theorem A.Proofof
Theorem A. It is obvious thati )
andii )
areequivalent
to eachother. We show the
equivalence
betweenii )
andHi).
We may assume that I =[0, T] ]
and dim H = + 00. The Rebolledo’s result[Cor. II,
3 .14 inVol. 22, n° 3-1986.
376 S. NAKAO
~l ~ J) implies
that the condition(2.8) with { ~(Pa) ~
insteadof {
isequivalent
to the condition(2. 8) with { dia ~ ~(Pa) ~ ~
insteadof {
On the other hand we
get by
Lemma 2.1 that the condition(2.9)
with~ ~(Pa) ~
instead isequivalent
to the condition(2.9)
with{ dia ( ~(P~) ~ ~
insteadof { Qa }. Consequently ii )
andiii )
areequivalent
to each other. The
proof
iscompleted.
D3. PRESERVATION
OF LOCAL MARTINGALE PROPERTY
First we show that any accumulation
point
of a sequence of H-valued continuous localmartingales
under the weak convergence is also a H-valued continuous localmartingale.
THEOREM B. -
H))
is closed in~(Co(I, H)).
0
Proof
2014 We may assume H =l2. Let { P,~ ~
convergesweakly
to00 0
P(Pn o o
El2)),
P E-~(Co(I, l2))).
Denoteby Rn
theprobability
measure0
on
Co(I, l2)
xCo(I, ll)
inducedby ((w, dia ~ w ~ ),
Theorem Aimplies {
istight.
Therefore there exists asubsequence { n’ ~
suchthat { R~. ~
converges
weakly
to R. For N =1, 2,
... setFor each i =
1, 2, ...,
s2, ... , sp, ti , t2, ... ,t~
_ s t and bounded continuous functionf
on(12)P
x we haveSince aN is R-a. s. continuous on
and
the
equality (3.1)
holds with R instead ofRn"
Therefore we haveNext we state a
continuity property
of themapping dia ~ )>
definedby (2 : 6).
THEOREM C. -
dia ~ ~ :
1l2))
--~[1))
is continuous.Proof
Within the situation of the aboveproof,
we can show thatAnnales de l’Institut Henri Poincaré - Probabilités et Statistiques
377
WEAK CONVERGENCE OF CONTINUOUS LOCAL MARTINGALES
(Wf - wi, R)
is a continuousmartingale
for any i =1, 2,
.... Therefore R~ , ~ ~
~ ~ 0
is the
probability
measure onCo(I, f)
xCo(I, h)
inducedby ((w, dia ( w ~ ), P).
, ~ ~ ~
0
This
implies that { Rn ~
0 convergesweakly
to R.Hence { dia ~ Pn ~ }
convergesweakly
todia ( P ~
and theproof
iscompleted.
DApplying
the abovetheorems,
we caneasily get
thefollowing
remarkabout the central limit theorem for a sequence of Hilbert space valued continuous local
martingales.
Remark. Let V = I ,2,... be a
nonnegative definite, symmetric
real matrix such
that
i= 1 vii + oo andPw
El2))
be a Wienermeasure such that the mean vector is zero and the covariance function
o o
is
((t
A Consider asequence {Pn} (Pn E l2))).
Then thefollowing
two conditions areequivalent
to each other.o o
i )
convergesweakly
toPw
inCo(I, f2).
o
P,~ ~ ~
convergesweakly
to the distribution of v22t,... )
in
Co(I, [1)
and for anyI, j
=1, 2,
... and t >0 {
the distribution of~ ~ 0
(( (t), P n) on [~ }
convergesweakly
to the 6-distribution atVijt
on ~.Finally
we note that such infinite-dimensional central limit theorem forcurrent valued stochastic processes are
investigated by
Ochi[1 D ] and
IkedaandOchi
[ ~ ].
Vol. 22, n° 3-1986.
378 S. NAKAO
APPENDIX
As stated in the introduction, in the case of H = (~, Theorem A is proved by Rebolledo
as a consequence of a general Aldous theorem for right continuous processes and using Lenglart’s inequality (cf. [7] ] [Il ]). The considered processes Mn may be discontinuous and
only the limits
of { and ( ( M" ~ ~
are assumed continuous. The proof is quite heavy.We give here a direct simple proof adapted to the special case of continuous processes.
LEMMA A.I. - Let M(t) (t E [0, T]) be an R-valued continuous martingale defined
on an usual filtered probability space (S~, ~, P, ~t) starting at 0 with E
[M(T)4
] + ~o.For each finite
partition 0394 = {0 =
so si ... =T}
of [0, T], setThen there exists a universal positive constant C such that
where
and
Proof - It is easy to see that the left hand side of (A. 2)
where [s s Sj+ 1 ( j = 0, 1, ... , l - 1 ). Applying Schwarz’s inequality to the
second part of the right hand side of the ahc»t inequality, we get (A..2). D LEMMA A. 2. - Let M"(t ) (n = 1, 2, ... ) be an R-valued continuous martingale defined
on a filtered probability space (Q, j~, P, with M"(0) = 0. Suppose that there exists a
stochastic process M on (S2, J~, P) such that with probability
one
converges to Muniformly on each compact interval. Further suppose that there exists a constant f3 > 0
such that sup n E [
~4+~]
cx~ (t > 0). Then M is a continuous martingale with E [ ] :r (t > 0) and satisfiesProof - It is obvious that M is a continuous martingale with E [ ~ M(t) ~4 ] oo (t > 0).
Ito formula implies
Annales de l’Institut Henri Poincaré - Probabilités et Statistiques
WEAK CONVERGENCE OF CONTINUOUS LOCAL MARTINGALES 379
We then get from the assumption
For any finite partition A
= ~ 0
= so sl ... Sz =T ~
of [0, T we havewhere t) and IA(M, t) are defined in the same way as (A .1).
Since ( ( (T)2 ~
is uniformly integrable, Lemma A. 1 implies that for any 8 > 0 there exists a positive constant
~ such that
On the other hand it is easy to see that lim an(~, 2) = 0 for any d. Therefore we have
n -~ o0
Combining (A. 4) with (A. 5), we then get (A. 3). 0
Proof of T heorem A in case of H = - First we prove that iii ) implies ii). Let (n = 1, 2, ...) be an R-valued continuous local martingale with M"(0) = 0 defined on a filtered probability space (Om .~n, Pn, We assume
that ( ( M" ~ ~ is
tight in Co([0, + oo), (~).Then for any 11 = 1, 2, ... there exist an extension
(SZ;"
Pn,~r~)
of(Q,
~n, Pn,~t)
anda one-dimensional Brownian motion Bn(t) on
(S~n,
~n, P~" with Bn(0) = 0 such that Mn(t ) = B"( ~ (t )). The tightnessM" ~ ) ~
implies the tightnessof (
Mn~.
Next we prove that ii ) implies iii ). Let Mn(t) (n = 1, 2, ... ) be an [?-valued continuous local martingale defined on a filtered probability space
(Q,
~, P, with Mn(0) = 0. Weassume that with probability
one { M"(t ) ~
converges uniformly on each compact interval.It is sufficient to prove
that { ( Mn ~ ~
is tight in Co( [0, T ], R) (T > 0). For N = 1, 2, ...and n = 1, 2, ... put
It is easy to see that for any e > 0 there exists No such that
(A . 6) P(Sn(N 0) = T) > 1 - E for n = 1, 2, ....
Since {
Mn(t AS,~(N)) ~
is tight in Co( [0, T ], for N = 1, 2, ..., we have by Lemma A. 2that { (
(t A istight
in Co( [0, for N = 1, 2, .... Therefore the tightnessin Co( [0, T ],
of ( ( Mn ~ ~
follows from (A. 6). 0 REFERENCES[1] D. ALDOUS, Stopping times and tightness, Ann. Probability, t. 6, 1978, p. 335-340.
[2] P. BILLINGSLEY, Convergence of probability measures, Wiley and Sons, 1968.
Vol. 22, n° 3-1986.
380 S. NAKAO
[3] C. DELLACHERIE, P. A. MEYER, Probabilities and potential B, North-Holland, 1982.
[4] N. DUNFORD and J. T. SCHWARTZ, Liner operators, Part I, Interscience, 1958.
[5] N. IKEDA, Y. OCHI, Central limit theorems and random currents, to appear in the
Proceedings of the 3rd Bad Honnef Conference on Stochastic Differential Systems.
[6] N. IKEDA, S. WATANABE, Stochastic differential equations and diffusion processes,
North-Holland, Kodansha, 1981.
[7] J. JACOD, J. MEMIN, M. MÉTIVIER, On tightness and stopping times, Stoch. Processes
Appl., t. 14, 1983, p. 109-146.
[8] E. LENGLART, Relation de domination entre deux processus, Ann. Inst. Henri Poincaré,
t. 13, 1977, p. 171-179.
[9] M. MÉTIVIER, Semimartingales, Walter de Gruyter, 1982.
[10] Y. OCHI, Limit theorems for a class of diffusion processes, to appear in Stochastic Processes and their Applications.
[11] R. REBOLLEDO, La méthode des martingales appliquée à l’étude de la convergence
en loi de processus, Bull. Soc. Math. France, t. 62, p. 1-125.
(Manuscrit rcyrr lo ?9 novembre 1985)
COPYRIGHT
Reproduction in whole or in part without the permission of the author or his representative is prohibited (law of March 11, 1957, Article 40, line 1). Such reproduction by whatever means, constitutes an infringement forbidden by Article 4?5 and those following it, of the Penal Code. The law of March 1 l, 1957, lines 2 and 3 of Article 41. authorizes only those copies or reproductions made for the exclusive use of the copyist, and not intended for collective use and such analyses and short quotations as are made for the purposes of an example or illustration.
Toute representation ou reproduction integrale ou partielle. faite sans le consentement de l’auteur ou de ses ayants droit
ou ayants cause, est illicite (loi du 11 mars 1957, alinea 1 er de l’article 40). Cette representation ou reproduction, par quelque precede que ce soit, constituerait une contrefacon sanctionnee par les articles 425 et suivants du Code penal.
La loi du 11 mars 1957 n’autorise, aux termes des alineas 2 et 3 de l’article 41, que les copies ou reproductions stricte-
ment reservees a l’usage prive du copiste et non destinees a une utilisation collective d’une part et d’autre part que les analyses et les courtes citations dans un but d’exemple et d’illustration.
The appearance of the code at the bottom of the first page of an article in this journal indicates the copyright owner’s
consent that copies of the article may be made for personal or internal use, or for the personal or internal use of spe- cific clients. This consent is given on the condition, however, that the copier pay the stated per-copy fee through the Copyright Clearance Center, Inc., Operations Center. 21 Congress St., Salem, Mass. 01970, U.S.A. for copying beyond that permitted by Sections 107 or 108 of the U.S. Copyright Law. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purpose, for creating new collective works,
or for resale.
C; GAUTHIER-VILLARS 1986
Annales de l’Institut Henri Poincaré - Probabilités et Statistiques