Mathematical Notes, vol. 76, no. 4, 2004, pp. 590596.

Translated from Matematicheskie Zametki, vol. 76, no. 4, 2004, pp. 635640.
Yu. Shamarova.
c
Original Russian Text Copyright
2004
by E.

Constructing a Brownian Sheet with Values

in a Compact Riemannian Manifold
Yu. Shamarova
E.
Received March 5, 2004

Key words: Brownian motion, Brownian sheet, Laplace operator, compact Riemannian man-

ifold, compact Lie group, cylindrical function.

x
In this paper, we propose a new method of constructing a two-parameter random eld WM
(s, t) ,
x M , with values in a compact Riemannian manifold M possessing the property that the random
x
x
( , t) and WM
(s, ) are Brownian motions on the manifold M with parameters t
processes WM
and s , respectively, issuing from the point x . (By a Brownian motion on a manifold M with
parameter t we mean the diusion process generated by the operator (t/2)M , where M is
the Laplace operator on the manifold M .) For the case in which the manifold is a compact Lie
group, the two-parameter random eld constructed in the paper coincides with the Brownian sheet
dened by Malliavin [1] in 1991. (Malliavin called this random eld a Brownian motion with values
in C([0, 1], M ) , which is the set of continuous functions dened on the closed interval [0, 1] and
taking values in M .) Nevertheless, for the case in which the manifold is a compact Lie group, the
method proposed in the present paper essentially diers from that used in Malliavins paper.

x
1. FIRST STEP IN THE CONSTRUCTION OF THE RANDOM FIELD WM

Suppose that M is a d-dimensional compact Riemannian manifold without boundary isometrically embedded in Rm . By a Brownian sheet with values in Rm we mean the family of m
independent standard Brownian sheets. Suppose that Wt,s is an n-dimensional Brownian sheet.
Consider Wt,s as a process taking values in the space C([0, 1], Rm ) . We denote this process by the
symbol Wt . We introduce the following notation: if E is a locally convex space, then E t denotes
C([0, t], E) ; if y C([0, 1], Rm ) is a continuous function, then Wy denotes the distribution of
y
)t = (t) + Wty .
the process Wty = y + Wt . If C([0, 1], Rm ) , then we dene the process (W
Suppose that

Wy is the distribution of this process and Ey , is the expectation with respect to the

y
measure

Wy . Further, U (M ) denotes the -neighborhood of the manifold M . We consider W
for functions y and satisfying the conditions: y(0) M , (0) = 0 . The goal of this section
is to prove the existence of a limit (given below) with respect to the family of bounded continuous cylindrical functions, where by a cylindrical function C([0, 1] [0, 1], Rm ) R we mean a
function f for which there exists a nite collection of points 1 , . . . , n , 1 , . . . , k and a function
f: Rnk R such that

f () = f((1 , 1 ), (1 , 2 ), . . . , (n , k )).
This limit denes the measure

WyM , ,s,t :

Ey , {f ()I{(Wy )t (s)U (M )} }
f ()

WyM , ,s,t (d) = lim
.
y
0


Wy {(W
)t (s) U (M )}
C([0,s],Rm )t
590

0001-4346/2004/7634-0590

c

2004
Springer Science+Business Media, Inc.

(1)

A BROWNIAN SHEET WITH VALUES IN A RIEMANNIAN MANIFOLD

591

Before proving the existence of such a limit, we consider the process

z
s
(W
,s )t = (t) + Bt ,

where : [0, 1] Rm is a continuous function satisfying the condition (0) = 0 and Bts is a
Brownian motion with parameter s issuing from the point z . The results obtained for this process
will be used for a subsequent construction.
z
z
Some results for the process (W
,s )t . Suppose that W ,s denotes the distribution of the
z
process (W
,s )t and Ez , ,s is the expectation with respect to this distribution.

Lemma 1. The limit


C([0,t],Rm )

f ()WzM , ,s,t (d) = lim

Ez , ,s {f ()I{(Wz , s )t U (M )} }
z ) U (M )}
Wz ,s {(W

,s t

with respect to the family of continuous bounded cylindrical functions, exists and defines the measure WzM , ,s,t in the integral on the left.
Sketch of the proof. Let us nd a function f: Rk+1 R and a nite set of points 1 , . . . , k
such that
f () = f((1 ), . . . , (k ), (t)).
We have



C([0,t],Rm )

f ()WzM , ,s,t (d)

= lim

f ()I{ : (t)U (M )} Wz ,s (d)

C([0,t],Rm )
Wz ,s { : (t) U (M )}



1
W
= lim W
P (1 , 0, dx1 )
PW (2 1 , x1 , dx2 )
0 P (t, 0, U (M z (t))) Rm
Rm


PW (t k , xk , dxk+1 )

U (M (t)z)

f(x1 + z + (1 ), . . . , xk + z + (k ), xk+1 + z + (t)),

where



1
|z x|2
dz.
exp
P ( , x, dz) =
2s
(2s )m/2
W

Since the function in the integrand is bounded, it suces, by Lebesgues theorem, to prove that
the following limit exists:

lim

(U (M (t)z))

= lim

f(x1 + z + (1 ), . . . , xk+1 + z + (t))PW (t k , xk , dxk+1 )


U (M (t)zxk )

PW (t, 0, U (M z (t)))
f(x1 + z + (1 ), . . . , xk+1 + xk + z + (t))PW (t k , 0, dxk+1 )
PW (t, 0, U (M z (t))

By M1 we denote the manifold M (t) z xk and by M2 the manifold M (t) z . Further,

suppose that


1
1
lU (M1 ) ,
lU (M2 ) ,
=
=
volmd ()
volmd ()
MATHEMATICAL NOTES

Vol. 76

No. 4

2004

 YU. SHAMAROVA
E.

592

where l is the Lebesgue measure on Rm . We can easily see that the proof of the existence of this
limit can be reduced to that of the existence of the limit



|xk+1 xk |2
g(x
)
exp

(dxk+1 )
k+1
2s(tk )
Rm


,
lim

|xk+1 |2
0
exp

(dx
)

k+1
m
2st
R
where g : R R is another symbol for the function f introduced to indicate the dependence on
the last variable solely. We can easily show that, as 0 , the measures and converge
weakly to the surface measures on M1 and M2 , respectively.

Lemma 2. The limit (1) with respect to the family of continuous bounded cylindrical functions
exists.

Wy ( : (t) ) is the transition probability
Sketch of the proof. Suppose that PW (t, y , ) =

for the measure

Wy , where y C([0, 1], Rm ) . Further, suppose that the function
f: C([0, s], Rm )k+1 R
and the nite set of points 1 , 2 , . . . , k satisfy the relation
f () = f((1 ), (2 ), . . . , (k ), (t)).
Let the symbol s denote the coordinate mapping. The proof is carried out by using the following
formula from [2, p. 204]:





W
0


f ()W (d) =
P (1 , 0, dw1 )
PW (2 1 , w1 , dw2 )
C([0,s],Rm )t
C([0,s],Rm )
C([0,s],Rm )


f(w1 , . . . , wk+1 )PW (t k , wk , dwk+1 )

s1 (U (M (t)y(s)))

and applying Lemma 1 to the measure in the last integral.

2. ASYMPTOTICS IN t FOR AN INTEGRAL OF SPECIFIC FORM
Proposition 1. Let i be an isometric embedding of the manifold M in Rm and g C 2 (M ) .
Then




 t
|z y|2
t
1
M (dz) = g(y) + g(y) c(y) scal(y) M g(y) + tR(t, y),
g(z) exp
d/2
2t
8
2
(2t)
M
where |R(t, y)| < Kt1/2 , K is a constant independent of y , scal(y) is the scalar curvature at the
point y , and the function c(y) is of the form
  2 i 2
(0),
c(y) =
xk xl

k ,l

where the xk are the normal coordinates in a neighborhood Uy of the point y which are specified by
the homeomorphism of the neighborhood Uy onto a neighborhood of zero U in Rd . Independently
of the local coordinates, c(y) can be written as

1
1
c(y) = M M |y |2 y scal(y)
2
3
and, therefore, c(y) depends only on the embedding i .
MATHEMATICAL NOTES

Vol. 76

No. 4

2004

A BROWNIAN SHEET WITH VALUES IN A RIEMANNIAN MANIFOLD

593

Sketch of the proof. We have obtained a more exact asymptotic expression in comparison with
that obtained for an integral of similar form in [3]. The idea of the proof is the same.

Corollary 1. Suppose that g C 2 (M ) . Then the following asymptotics is valid :



|zy|2
M (dz)
g(z)
exp

2t
M
t


= g(y) M g(y) + tR1 (t, y),

2
2
M (dz)
exp |zy|
2t
M
where |R1 (t, y)| < K1 t1/2 , and K1 is a constant independent of y .
older function of H
older order ,
Corollary 2. Suppose that g C 2 (M ) , y M , and is a H
1/3 < < 1/2 , such that (0) = 0 . Suppose that PrM is the projection mapping onto the
manifold M along the subspaces normal to the manifold and defined in a suitable neighborhood of
the manifold M (t, y) = PrM (y + (t)) . Then the following asymptotics is valid :



|zy(t)|2
M (dz)
g(z)
exp

2t
M
t


= g(y + M (t)) M g(y) + tR2 (t, y),

|zy(t)|2
2
M (dz)
exp
2t
M
where |R2 (t, y)| < K2 t31 and K2 is a constant.
x
3. SECOND STEP IN THE CONSTRUCTION OF THE RANDOM FIELD WM

Suppose that f is a continuous bounded cylindrical function on C([0, s], Rm )1 and : R M

is a function which is the trajectory of the Brownian motion on M such that (0) = x . Suppose
that P1 = {0 = t0 t1 tn = 1} is a partition of the interval [0, 1] . If E is a locally
convex space, then to each E 1 we can assign a nite sequence of n elements
(1 , 2 , . . . , n ) E t1 E t2 t1 E tn tn1 ,
where j is dened on the interval [0, tj tj1 ] by the formula j (t) = (tj1 + t) . We dene
the function ti1 ti on the interval [0, ti ti1 ] as follows:
ti1 ti (t) = (ti1 + t) (ti1 ).
Let us dene the measure

WxM ,,s,P1 by the formula




WxM ,0t1 ,s,t1 (d1 )
f ()

WxM ,,s,P1 (d) =
t
m
1
m
C([0,s],R )
C([0,s],R ) 1

(t )


WM1 ,1t t ,s,t2 t1 (d2 )

1 2
t
t
m
2
1
C([0,s],R )


(tn1 tn2 )


WMn1
(dn )f (1 , 2 , . . . , n ).

,t
tn ,s,tn tn1
n1

C([0,s],Rm )tn tn1

WxM ,,s,P1 is well dened.

It is readily veried that i (ti ti1 )(0) M , so that the measure

Further, let P2 = {0 = s0 s1 sk = 1} be a partition of the interval [0, 1] . Now, suppose
W
that s is a time parameter. Instead of the symbol

WxM ,,s,P1 , we shall write

M ,s,P1 . Let us
x
dene the measure WM ,P1 ,P2 by the formula


x


WxM ,s1 ,P1 (d1 )
f ()WM ,P1 ,P2 (d) =
C([0,1],Rm )1
C([0,1],Rm )s1

(s )


WM1 ,s21s1 ,P1 (d2 )

s
s
C([0,1],Rm ) 2 1


(sn1 sn2 )


WMn1

,sn sn1 ,P1 (dn )f (1 , . . . , n ).

C([0,1],Rm )sn sn1

MATHEMATICAL NOTES

Vol. 76

No. 4

2004

 YU. SHAMAROVA
E.

594

Theorem 1. For each x M , if the meshes of the partitions P1 and P2 tend to zero, then
the sequence of measures WxM ,P1 ,P2 is weakly convergent to the measure WxM with respect to the
family of continuous bounded cylindrical functions. The measure WxM regarded as the distribution
of a process with values in C([0, 1], M ) , possesses a transition probability at time t coinciding
with the distribution of the Brownian motion with parameter t on the manifold issuing from the
point x .
Sketch of the proof. Suppose that
h

WxM ,,s,P1



 

= h WM ,s,P1 =

C([0,1],Rm )

=
C([0,t1 ],Rm )

WxM ,0t1 ,s,t1 (d1 )

C([0,tn tn1 ],Rm )

h()WxM ,,s,P1 (d)

(t )

C([0,t2 t1 ],Rm )

WM1 ,1t

(tn1 tn2 )
,s,tn tn1 (dn )h(1 ,
n1 tn

WMn1
,t

1 t2

,s,t2 t1 (d2 )

2 , . . . , n ).

Suppose that there exists a function f: C([0, 1], Rm ) R such that f () = f((t)) . Then


1
f ()

W
(d)
=
f(w)

W
M ,s,P1
M ,s,P1 s (dw)
m
1
m
C([0,s],R )
C([0,1],R )

f(w)W
=
M ,s,P1 (dw).
C([0,1],Rm )

This yields


C([0,1],Rm )1

f ()WxM ,P1 ,P2 (d)

=


C([0,1],Rm )

C([0,1],Rm )

Consider the integral

C([0,1],Rm )

WxM ,s1 ,P1 (dw1 )

1
Ww
M ,s2 s1 ,P1 (dw2 )

C([0,1],Rm )

WMn2
,sn1 sn2 ,P1 (dwn1 )

w

WMn1
,sn sn1 ,P1 (dwn )f (wn ).


C([0,t],Rm )

g()WzM , ,s,t (d),

where the function g C([0, t], Rm ) is such that there exists a function g C(R) for which
g() = g((t)) . As a result of simple calculations, we obtain


g()I{ : (t)U (M )} () Wz ,s (d)
C([0,t],Rm )
z
g()WM , ,s,t (d) = lim
0
Wz ,s { : (t) U (M )}
C([0,t],Rm )



|x1 z(t)|2
g(x1 ) M (dx1 )
exp

2ts
M


.
=

|x1 z(t)|2

exp

(dx
)
M
1
2ts
M
First, suppose that the function f is such that there exists a function p : Rm R and numbers
t, s [0, 1] for which f () = p((t, s)) . The integral

f ()WxM ,P1 ,P2 (d)
C([0,1],Rm )1

MATHEMATICAL NOTES

Vol. 76

No. 4

2004

A BROWNIAN SHEET WITH VALUES IN A RIEMANNIAN MANIFOLD

595

is of the form









|xn1 xn2 |2
|xn xn1 |2
|x1 x|2
exp

exp

exp

dx
dx
dxn
1
n1 M
2s1 t1
2s1 tn1
2s1 tn
M
M







2
2
|
x1 x|2
n1 xn2 |
n xn1 |
exp 2s
exp |x2s
d
x1
d
xn1 M exp |x2s
d
xn
M
M
1 t1
1 tn1
1 tn






2
|y1 x1 |2
n2 xn1 +xn2 |
exp 2s
exp |yn1 y2s
dy1
dyn1
M
M
2 t1
2 tn1







2
|
y1 x1 |2
n2 xn1 +xn2 |
d
y1
d
yn1
exp 2s
exp |yn1 y2s
M
M
2 t1
2 tn1



xn +xn1 |2
dyn
exp |yn yn1
2s2 tn
M




xn +xn1 |2
exp |yn yn1
d
yn
2s2 tn
M






|un1 un2 zn1 +zn2 |2
|u1 z1 |2
du
dun1
exp

exp

1
2sk1 t1
2sk1 tn1
M
M







2
|
u1 z1 |2
n2 zn1 +zn2 |
d
u1
d
un1
exp 2s
exp |un1 u
2sk1 tn1
M
M
k1 t1



2
n1 zn +zn1 |
dun
exp |un u2s
M
k1 tn



2
n1 zn +zn1 |
d
un
exp |un u2s
M
k1 tn






|vn1 vn2 un1 +un2 |2
|v1 u1 |2
exp

exp

dv
dvn1
1
2sk t1
2sk tn1
M
M







2
|
v1 u1 |2
n2 un1 +un2 |
exp 2s
exp |vn1 v2s
d
v1
d
vn1
M
M
k t1
k tn1



2
un +un1 |
p(vn ) dvn
exp |vn vn1
2sk tn
M



,
2
|
vn vn1 un +un1 |
d
v
exp

n
2sk tn
M
where ti = ti ti1 , sj = sj sj1 , and, to simplify the notation, instead of M (dz) we use
dz . We have also assumed that tn = t and sk = s . Let us denote this integral by I(P1 , P2 , p) .


Lemma 3. The integral I(P1 , P2 , p) converges to exp st

p if the meshes of |P1 | and |P2 |

M
2
tend to zero.
Sketch of the proof. Using Corollaries 1 and 2 of Proposition 1, we obtain the following asymptotics for the integral I(P1 , P2 , p):
I(P1 , P2 , p)(x) = p(x)


st
M p(x) + O(s2 t2 ) + O |P2 |, |P1 | .
2

Hence we see that the following limit exists:

lim

|P1 |0, |P2 |0

I(P1 , P2 , p)(x) = p(x)


st
M p(x) + O(s2 t2 ) = (Qst p)(x) .
2

Further, we verify that Q + = Q Q . Let s and t satisfy = st . Let us nd a s and

[0,t+t]
[0,s+s]
, P2
, p)
t such that + = (s + s)(t + t) , and consider the integral I(P1
for partitions of the closed intervals [0, s + s] and [0, t + t] resulting from supplementing the
corresponding partitions of the closed intervals [0, s] and [0, t] by points of partitions of the closed
intervals [t, t + t] , [s, s + s] . We can easily see from the structure of the integral I(P1 , P2 , p)
that

[0,t] [0,s]

[0,t+t]
[0,s+s]
[t,t+t]
[s,s+s]
, P2
, p) = I P1 , P2 , I(P1
, P2
, p) .
I(P1
MATHEMATICAL NOTES

Vol. 76

No. 4

2004

596

YU. SHAMAROVA
E.

Passing to the limit in this expression and taking = st + ts + st into account, we nd

that Q is a semigroup which, by the proof above, satises
(Q p)(x) = p(x)
Thus, we nd that

M p(x) + O( 2 ).
2



Q = exp M .

2

For a function f depending on at several points, such as at points i [0, s] and j [0, t] ,
the integral I(P1 , P2 , p) will be of the same form. The convergence examined above occurs on
each square [i1 , i ] [j1 , j ] . Each of operators of the form


i
j
M
exp
2
acts on the corresponding variable of the function p dened as
f () = p(11 (1 , 1 ), . . . , kl (l , l )),
where ij is dened on [0, i i1 ] [0, j j1 ] by the formula
ij (s, t) = (i1 + s, j1 + t).

x
Corollary 3. Suppose that M is a compact Lie group. Then the random field WM
regarded as
a process with values in C([0, 1], M ) , coincides with the Brownian motion constructed in [1].

Sketch of the proof. The proof follows from Theorem 1 and Theorem 2.15 from [4] (Theorem 2.15
from [4] was also proved in Lemma 3.3 from [5]).

REFERENCES
1. P. Malliavin, in: Diusion Processes and Related Problems in Analysis (Evanston, IL, 1989 ), vol. 1,
Birkh
auser, Boston, MA, 1990, pp. 1731.
2. N. Ikeda and Sh. Watanabe, Stochastic Dierential Equations and Diusion Processes, North-Holland,
Amsterdam, 1981.
3. O. G. Smolyanov and H. von Weizs
acker, and O. Wittich, Canad. Math. Soc. Conference Proc., 29
(2000), 589602.
4. B. K. Driver and V. K. Srimurthy, Ann. Probab., 29 (2001), no. 2, 691723.
5. V. K. Srimurthy, Probab. Theory Related Fields, 118 (2000), no. 4, 522546.
M. V. Lomonosov Moscow State University

MATHEMATICAL NOTES

Vol. 76

No. 4

2004