You are on page 1of 7

This article was downloaded by: [Michigan State University]

On: 31 January 2015, At: 12:15


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Stochastics: formerly Stochastics


and Stochastics Reports
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/gssr18

A trotter product formula for


random matrices
a
Marc A. Berger
a
Department of Mathematics , The Weizmann Institute of
Science , Rehovot, 76100, Israel
Published online: 04 Apr 2007.

To cite this article: Marc A. Berger (1988) A trotter product formula for random
matrices, Stochastics: formerly Stochastics and Stochastics Reports, 23:2, 79-84, DOI:
10.1080/17442508808833483

To link to this article: http://dx.doi.org/10.1080/17442508808833483

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information
(the “Content”) contained in the publications on our platform. However, Taylor
& Francis, our agents, and our licensors make no representations or warranties
whatsoever as to the accuracy, completeness, or suitability for any purpose
of the Content. Any opinions and views expressed in this publication are the
opinions and views of the authors, and are not the views of or endorsed by Taylor
& Francis. The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and Francis
shall not be liable for any losses, actions, claims, proceedings, demands, costs,
expenses, damages, and other liabilities whatsoever or howsoever caused arising
directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan,
sub-licensing, systematic supply, or distribution in any form to anyone is
expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Stochasaics, Vol. 23, pp. 79-84
Photocopying permitted by license only
% 1988 Gordon and Breach, Science Publishers, Inc.
Printed in Great Britain

A Trotter Product Formula for


Random Matrices
MARC A. BERGER
Department of Mathematics, The Weizmann Institute of Science,
Downloaded by [Michigan State University] at 12:15 31 January 2015

Rehovot 76100, Israel

(Received I September 1986; in final form 19 May 1987)

This note contains a generalization of the Trotter product formula to the setting of
multiple linear systems of stochastic differential equations. From the result it follou~s
that the solution of a system in Stratonovich form in a Lie algebra lies in the
corresponding Lie group.
KEY WORDS: Stochastic differential equation, Trotter product formula.
Mathematics subject classification (1980): 60F05, 60H20.

Let a,, . . . , a, be real d x d matrices. Trotter's product formula


asserts that

lim [exp (tlna,) +. . . x exp (t/na,)ln


n-u,
= exp
( x ).
t
l:j
aj
(1)

This well known result extends to semi-groups on a Banach space


with generators aj, (e.g. Pazy [6, Cor. 3.5.5]), and can be thought of
as an "alternating step" method for integrating evolution equations.
Thus one runs the semi-group exp(tu,) for time tln, then the semi-
group exp(ta,) for time tln, etc., cycling around from a, back to a,.
Given a system A =(a,, ay',. . . ,a$)) of p + 1 real d x d matrices, let
@,(s, t) denote the fundamental matrix,
80 M. A. BERGER

for the Stratonovich linear stochastic system with coefficient matrices


a, and ay', ..., 02). Here ( e l , .. .,8,) is a p-dimensional Brownian
motion. Denote OA(t):=QA(O,t). Observe that @,(R), 3,>0, has the
same distribution as @,,(I), where

If B = (a,, ah1',. . . ,OF)) is another system like A, denote

A + B: = ( a A+ a,, a y )+ abl', . . . , a y ) + OF)).


Downloaded by [Michigan State University] at 12:15 31 January 2015

If Aj, 1 5 j g m , are all systems like A then the analogue to ( 1 ) would


be to run the stochastic semi-groups cDA, over successive time
intervals

i.e. to run the time-ordered product

It is understood that the same Brownian motion (O,, . . . ,9,) is used


for each OAj.This can always be achieved by enlarging p and setting
02:= 0 for some j, k. The following theorem, then, is a generalization
of (1).

where

and [., . ] denotes the Lie commutator, [b,c ] := bc -cb.


TROTTER PRODUCT FORMULA 81
The convergence "3"in (2) is that of weak convergence of
measures on C([O, 11: GL,). This result is easily understood in terms
of the Campbell-Hausdorff Theorem for Lie algebras. The proof of
this theorem is based on a central limit theorem for random matrices
in GL,.

CLT Let Din, 1S i g n , be a triangular array of square-integrable


d x d matrices which are i.i.d. for each fixed n. Assume that
a, a"', . . . ,dP)
are (non-random) matrices for which

lim n QDin- I ) = a;
n-+
Downloaded by [Michigan State University] at 12:15 31 January 2015

30

P
lim n E((Dinx-x, y))* = ((dk'x, y))', V x , y E Rd. (4)
n+m k= 1

Assume in addition that


limn [ ~ D ~ , - I / ~ = O , QE>O.
n-tcc jDin21(>e

Then

where A =(a, a''), . . . , P ) .


This result follows from Stroock and Varadhan [7, Thm. 2.41. An
early version of it appears in Grenander [5, Thm. 4.4.21. In Berger
[3] it is proved using the semi-group product formula of P. Chernoff
[4]. These latter two references only obtain convergence in distri-
bution in (6) for each fixed t.
An example of a triangular array Din satisfying the above hypo-
theses is

where U , T/',, W,, are random d x d matrices satisfying IEJ uiI2, IE\ 1/;12,
[E/wn('
< cc and
M. A. BERGER
LEUi = 0, lim n EKn= 0,
n+ m

lim n E ~ W , , / ~ = O .
n+ m

In order for the Din here to be i.i.d. for each fixed n, it suffices that
the triples (U,, 6,T/I/In) be i.i.d, for each fixed n.
Regarding condition (5): it is clear that if

lim n E / D ~ , - I I ~ + ' = O
n'cc
Downloaded by [Michigan State University] at 12:15 31 January 2015

for some 6 > 0, then (5) follows.

Proof of Theorem First of all (2) is to be recast from Stratonovich


to It8 form. For A =(a,, ay),. . . ,OF))
let YA(s,t) denote the fundamental
matrix
P f
+
f

YAs, t) = 1
k=l s
o$'YA(s,U) dOk(u) +J aAYA(s,u) du,
s

for the It8 system. Using the transformation rule between It8 and
Stratonovich stochastic differential equations (e.g. Arnold [I, Sec.
10.21) one sees that (2) is equivalent to

where

This fits the CLT above with

Note that YAj(s,t) and YAj(s- t, 0) have the same distribution and
TROTTER PRODUCT FORMULA 83
o) are independent if (s, t ) n(u, o) = 0. Thus it
that YAj(s,t ) and YAk(u,
suffices to establish

1
lim - E[(YAl(t)x .. ' x YArn(t)x-x,
y)I2
t-0 t
Downloaded by [Michigan State University] at 12:15 31 January 2015

1
lim - E [ ( Y A l ( tx) . . . x YAm(t)x-x,
y ) I 4 =0, V X ,y E Rd
r-0 t

These follow respectively from the identities

x ... x Y ~ , ( ~ ) - I ]
EIYAl(t)
84 M. A. BERGER

and
Downloaded by [Michigan State University] at 12:15 31 January 2015

It follows from Berger [2] that

det @,(t) = nrn

j= 1
det @,l(t),

where A , , . . . ,A,, 3 are as in the theorem. Thus if OAj(t)E GLd for all
t , then so does @,(~)EGL,. The same holds immediately for SL,.
More generally, it follows now from the theorem above that if
. . ,@Am are all processes on the same Lie group G c GL,, then
0,lies in this same group.

References
[I] L. Arnold, Stochastic Dijerential Equations: Theory and Applications, Wiley-
Interscience, New York, 1974.
[2] M. A. Berger, A remark on stochastic fundamental matrices, Ann. Prob. 11 (1983),
215-216.
[3] M. A. Berger, Central limit theorem for products of random matrices, Trans.
Amer. Math. Soc. 285 (1984), 777-803.
[4] P. Chernoff, Note on product formulas for operator semi-groups, J. Func. Anal. 2
(1968), 238-242.
[5] U. Grenander, Probabilities on Algebraic Structures, Wiley, Stockholm, 1963.
[6] A. Pazy, Semigroups of Linear Operators and Applications t o Partial Differential
Equations, Springer-Verlag, New York, 1983.
[7] D. W. Stroock and S. R. S. Varadhan, Limit theorems for random walks on Lie
groups, Sankya Ser. A (3) 35 (1973), 277-294.

You might also like