HAL Id: hal-03212113
https://hal.archives-ouvertes.fr/hal-03212113
Submitted on 29 Apr 2021
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires
Toward a Wong-Zakai approximation for big order generators
Rémi Léandre
To cite this version:
Rémi Léandre. Toward a Wong-Zakai approximation for big order generators. Symmetry, MDPI,
2020, 12 (11), pp.1893. �10.3390/sym12111893�. �hal-03212113�
Toward a Wong-Zakai approximation for big order generators
R´ emi L´ eandre
Laboratoire de Math´ ematiques, Universit´ e de Bourgogne-Franche-Comt´ e,
25030, Besan¸con, FRANCE.
email: remi.leandre@univ-fcomte.fr September 11, 2020
1 Introduction
Let us consider a compact Riemannian manifold M of dimension d endowed with its normalized Riemannian measure dx (x ∈ M ).
Let us consider m smooth vector fields X
i(We will suppose later that they are withous divergence). We consider the second order differential operator:
L = 1/2
m
X
i=1
X
i2(1)
It generates a Markovian semi-group P
twhich acts on continuous function f on
M ∂
∂t P
tf = LP
tf ; P
0f = f (2) P
tf ≥ 0 if f ≥ 0. It is represented by a stochastic differential equation in Stratonovitch sense ([3])
P
tf (x) = E[f (x
t(x))] (3)
where
dx
t(x) =
m
X
i=1
X
ii(x
t(x))dw
ti; x
0(x) = x (4) where t → w
itis a flat Brownian motion on R
mClassically, the Stratonovitch diffusion x
t(x) can be approximated by its Wong-Zakai approximation.
Let w
n,itbe the polygonal approximation of the Brownian path t → w
tnfor
a subdivision of [0, 1] of length n.
We introduce the random ordinary differential equation dx
nt(x) =
m
X
i=1
X
i(x
nt(x))dw
tn,i; x
n0(x) = x (5) Wong-Zakai theorem ([3]) states that if f is continuous
E[f (x
nt(x))] → E[f(x
t(x))] (6) We are motivated in this paper by an extension of (6) to bigger order generators.
Let us consider the generator L
k= (−1)
kP
mI
X
i2k. We suppose that the vector fields X
ispann the tangent space of M in all point of M and that they are divergent free. L
kis an elliptic postive essentially self-adjoint operator [1]
which generates a contraction semi-group P
tkon L
2(dx)
Let L
f,kbe the generator on R
m((w
i) ∈ R
m) By [1], it generates a semi- group P
tf,kon C( R
m), the space of continuous functions on the flat space en- dowed with the uniform topology, which is represented by an heat-kernel:
P
tf,k[f ](w
0) = Z
Rm
f (w + w
0)p
f,kt(w) ⊗ dw
i(7) (w = (w
i)) In [7], we noticed that heuristically P
tf,kis represented by a formal path space measure Q
f,ksuch that
Z
E
f (w
kt+ w
0)dQ
f,k(w
.) = P
tk,f(f )(w
0) (8) If we were able to construct a differential equation in the Stratonovitch sense
dx
kt(x) =
m
X
i=1
X
i(x
kt(x))dw
kt,i; x
k0(x) = x (9)
P
tf,k(x) = Z
f (x
kt(x))dQ
f,k(10)
These are formal considerations because in such a case the path measures are not defined. We will give an approach to (11) by showing that some conve- nient Wong-Zakai approximation converge to the semi-group. We introduce, according to [4] and [5] the Wong-Zakai operator
Q
kt[f ](x) = Z
Rm
f (x
i(w)(x))p
f,kt(w) ⊗ dw
i= Z
Rm
f (x(t
1/2kw)(x))p
f,k1(w)dw (11) where
dx
1(w)(x) =
m
X
i=1
X
i(x
s(w))w
ids ; x
0(w)(x) = x (12) As a first theorem, we state:
Theorem 1 (Wong-Zakai)Let us suppose that the vector fields X
icommute.
Then (Q
k1/n)
n(f ) converge in L
2(dx) to P
1kf if f is in L
2(dx)
To give another example, we suppose that M is a compact Lie group G endowed with its normalized Haar measure dg and that the vector fields X
iare elements of the Lie algebra of G considered as right invariant vector fields. This means ,that if we consider the right action on L
2(dg) R
g0f → (g → (f (gg
0)) (13) we have
R
g0[X
if ](.) = X
i[R
g0f ](.) (14) We consider the rightinvariant elliptic differential operator
L
k= (−1)
km
X
i=1
X
i2k(15)
It is an elliptic positive essentially selfadjoint operator. By elliptic theory ([1]), it has a positive spectrum λ associated to eigenvectors f
λ. λ ≥ 0 if λ belongs to the spectrum.
Theorem 2 (Wong-Zakai) Let f = P
a
λf
λsuch that P
λ
|a
λ|
2C
λ< ∞ for all C > 0. Then (Q
k1/n)
n(f ) converges in L
2(dg) to P
1kf .
We refer to the reviews [5], [6] [7] for the study of stochastic analysis without probability.
2 Proof of theorem 1
L
kis an elliptic positive operator. By elliptic theory [?], it has a discete spec- trum λ associated to normalized eigenfunctions f
λ. Since R
Rm
|p
f,k1(w)|
2dw < ∞, Q
k1/nis a bounded operator on L
2(dx). Moreover
Q
k1/nf = X
a
λQ
k1/nf
λ(16) if
f = X
a
λf
λ(17)
The main remark is that we can compute explicitely Q
k1/nf
λ. We put t = 1/n.
Formally
f
λ(x(t
1/2kw))(x) = X
n0
1/n
0!( X
m
X
iw
it
1/2k))
n0(f
λ)(x) (18) Namely, by ellipticity and because the vector fields X
icommute with L
k, we can conclude that the L
2-norm of X
iα11
X
iα22
...X
iαll
f
λhas a bound in λ
Pαi/2kC
Pαiin order to deduce that the series in (18) converges. It is enough to compute
1/n
0! Z
Rm
( X
X
iw
it
1/2k))
n0f
λ(x)p
k,f1(w)dw = B
n0(19)
The main remark is if one of the l
iis not a multiple of 2k, we have Z
Rm
w
l11..w
lmmp
k,f1(w)dw = 0 (20) On the other hand, by using the semi-group properties of P
tk,f, we have
Z
Rm
w
12kl1..w
m2klmp
k,f1(w)dw = (2kl
1)!
l
1! ... (2kl
m)!
l
m! (21)
Therefore, B
n0= 0 if n
0is not a multiple of 2k and is equal because the vector field commute, ,if n
0= 2kl
0to
1 (2kl)
0!
X X
12kl01..X
2kl0
m m
(2kl
1)!
l
1! ... (2kl
m)!
l
m!
(2kl
0)!
(2kl
01)!..(2kl
0m)! f
λ= 1/l
0!(L
k)
l0f
λ(22) We deduce that
Q
k1/nf
λ= exp[−1/nλ]f
λ(23)
and that
(Q
k1/n)
nf
λ= exp[−λ]f
λ(24) such that
(Q
k1/n)
nf = exp[−L]f (25)
if f = P a
λf
λ.
3 Proof of Theorem 2
Let E
λbe the space of eigenfunctions associated to the eigenvalue λ of L
k. Since L
kcommute with the right action of G, E
λis a representation for the right action of G ([2]). Therefore rightinvariant vector fields acts on E
λ. If Z is a rightinvariant vector field, we can consider the L
2norm of Zf
λfor f
λbelonging to E
λ. We remark that (L
k+ C)
1/2kis an elliptic pseudodifferential operator of order 1 (C is strictly ppositive). By Garding inequality [1],
kZf
λk
L2(G)≤ Ckf
λk
L2(G)+ k(L
k+ C)
1/2kf
λk
L2(G)(26) f
λis an eigenfunction associated to (L
k+C)
1/2kand the eigenvalue (λ+C)
1/2k.
Let us consider a polynomial X
iα11
...X
iαll
= Z
l. It acts on E
λand is norm is bounded by ((λ + C)
1/2k+ C)
Pαifor the L
2norm.
From that we deduce that if f
λis an eigenfunction associated to λ of L
kthat the series
X
l
(X
it
1/2kw
i)
ll! f
λ(27)
converges and is equal to f
λ(x(t
1/2kw)(x)) By distinguishing if w is big or not and using (20), we see that if l 6= 2kl
0Z
Rm
(
m
X
i=1
X
it
1/2kw
i)
l0f
λp
f,k1(x)dw = 0 (28) Moreover, by (20) and (21)
1 (2kl
0)!
Z
Rm
(
m
X
i=1
X
it
1/2kw
i)
2kl0f
λp
f,k1(x)dw = t
l0(2kl
0)!
Z
Rm
X X
α1..X
α2kl0
f
λw
2kl1 01..w
2klm 0mp
f,k1(w)dw (29) where 2kl
0jis the number of of α
iequal to j. By using (20) and (21), we recognize in (29)
1t
l0(2kl
0)!
X
αi
X
α1..X
α2kl0f
λ(2kl
01)!
l
01! ... (2kl
0m)!
l
0m! (30)
For l
0= 1, we recognize tL. Let us compute the L
2norm of the previous element. It is bounded by
t
l0(2kl
0)!
X
αi
(λ
1/2k+ C)...(λ
1/2k+ C) (2kl
10)!
l
01! ... (2kl
0m)!
l
m0! (31) For l
0= 1, we recognize tL.
We recognize in the previous sum t
l0(2kl
0)!
X (2kl
0)!
(2kl
10)!...(2kl
m0)!) (2kl
01)!
l
10! ... (2kl
0m)!
l
0m! (λ
1/2k+ C)
2kl0(32) We deduce a bound of the operation given by (29) in
tl0C2kl0
(l0)!
(λ + C)
l0. By the same argument, we have a bound of
tl0
l0!
(L
k)
l0on E
λin
tl0
l0!
C
l0(λ+C)
l0. In order to conclude, we see that on E
λQ
kt= exp[−λt]Id + X
l0>1
t
l0l
0! Q
lλ0,t(33) where Q
lλ0,thas a bound on E
λin C
l0(λ + C)
l0. We deduce that Q
ktacts on E
λby
exp[−λt]Id + t
2Q
λt= R
λt(34) where the norm on E
λof Q
λtis smaller that C exp[Cλt].
But if f = P a
λf
λ(Q
k1/n)
nf = X
a
λ(R
λ1/n)
nf
λ(35)
Moreover
k(Q
kt)f k
L2(G)= Z
G
| Z
Rm
f (x(t
1/2kw)(g)p
f,k1(w)dw|
2dg ≤ C
Z
G
dg Z
Rm
|f (x(t
1/2kw)(g)|
2|p
f,k1(w)|
2dw ≤ C Z
Rm
|p
f,k1(w)|
2dw Z
G
|f (x(t
1/2kw)(g)|
2dg (36)
But Z
G
|f (x(t
1/2kw)(g)|
2dg = kf k
2L2(G)(37) because the vector fields are without divergence. If λ/n < C, the sum
X
λ<Cn
a
λ(R
λ1/n)
nf
λ(38) converges to
X a
λexp[−λ]f
λ(39)
Moreover P
λ>Cn
a
λ(R
λ1/n)
nf
λhas a L
2norm bounded by( P
λ>Cn