HAL Id: hal-00408248
https://hal.archives-ouvertes.fr/hal-00408248
Preprint submitted on 29 Jul 2009
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
L’archive ouverte pluridisciplinaire
HAL, estdestinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de
Self-interacting diffusions IV: Rate of convergence
Michel Benaïm, Olivier Raimond
To cite this version:
Michel Benaïm, Olivier Raimond. Self-interacting diffusions IV: Rate of convergence. 2009. �hal-
00408248�
Self-interacting diffusions IV: Rate of convergence ∗
Michel Bena¨ım
Universit´e de Neuchˆatel, Suisse Olivier Raimond
Universit´e Paris Ouest Nanterre la d´efense, France July 30, 2009
Abstract
Self-interacting diffusions are processes living on a compact Riemannian man- ifold defined by a stochastic differential equation with a drift term depending on the past empirical measure µ
tof the process. The asymptotics of µ
tis governed by a deterministic dynamical system and under certain conditions (µ
t) converges almost surely towards a deterministic measure µ
∗(see Bena¨ım, Ledoux, Raimond (2002) and Bena¨ım, Raimond (2005)). We are interested here in the rate of convergence of µ
ttowards µ
∗. A central limit theorem is proved. In particular, this shows that greater is the interaction repelling faster is the convergence.
∗We acknowledge financial support from the Swiss National Science Foundation Grant 200021-103625/1
1 Introduction
Self-interacting diffusions
Let M be a smooth compact Riemannian manifold and V : M × M → R a sufficiently smooth mapping
1. For all finite Borel measure µ, let V µ : M → R be the smooth function defined by
V µ(x) = Z
M
V (x, y)µ(dy).
Let (e
α) be a finite family of vector fields on M such that X
α
e
α(e
αf )(x) = ∆f(x),
where ∆ is the Laplace operator on M and e
α(f ) stands for the Lie derivative of f along e
α. Let (B
α) be a family of independent Brownian motions.
A self-interacting diffusion on M associated to V can be defined as the solution to the stochastic differential equation (SDE)
dX
t= X
α
e
α(X
t) ◦ dB
tα− ∇ (V µ
t)(X
t)dt.
where
µ
t= 1 t
Z
t0
δ
Xsds is the empirical occupation measure of (X
t).
In absence of drift (i.e V = 0), (X
t) is just a Brownian motion on M but in general it defines a non Markovian process whose behavior at time t depends on its past trajectories through µ
t. This type of process was intro- duced in Benaim, Ledoux and Raimond (2002) (hence after referred as [3]) and further analyzed in a series of papers by Benaim and Raimond (2003, 2005, 2007) (hence after referred as [4], [5] and [6]). We refer the reader to these papers for more details and especially to [3] for a detailed construction of the process and its elementary properties. For a general overview of pro- cesses with reinforcement we refer the reader to the recent survey paper by Pemantle (2007) ([15]).
1The mapping Vx :M → Rdefined by Vx(y) = V(x, y) is C2 and its derivatives are continuous in (x, y)
Notation and Background
Standing Notation We let M (M ) denote the space of finite Borel mea- sures on M, P (M) ⊂ M (M ) the space of probability measures. If I is a metric space (typically, I = M, R
+× M or [0, T ] × M ) we let C(I) denote the space of real valued continuous functions on I equipped with the topology of uniform convergence on compact sets. When I is compact and f ∈ C(I) we let k f k = sup
x∈I| f (x) | . The normalized Riemann measure on M will be denoted by λ.
Let µ ∈ P (M) and f : M → R a nonnegative or µ − integrable Borel function. We write µf for R
f dµ, and f µ for the measure defined as f µ(A) = R
A
f dµ. We let L
2(µ) denote the space of such functions for which µ | f |
2< ∞ , equipped with the inner product
h f, g i
µ= µ(f g) and the norm
k f k
µ= p µf
2. We simply write L
2for L
2(λ).
Of fundamental importance in the analysis of the asymptotics of (µ
t) is the mapping Π : M (M ) → P (M ) defined by
Π(µ) = ξ(V µ)λ (1)
where ξ : C(M ) → C(M ) is the function defined by ξ(f )(x) = e
−f(x)R
M
e
−f(y)λ(dy) . (2)
In [3], it is shown that the asymptotics of µ
tcan be precisely related to the long term behavior of a certain semiflow on P (M ) induced by the ordinary differential equation (ODE) on M (M ) :
˙
µ = − µ + Π(µ). (3)
Depending on the nature of V, the dynamics of (3) can either be convergent
or nonconvergent leading to similar behaviors for { µ
t} (see [3]). When V is
symmetric, (3) happens to be a quasigradient and the following convergence
result hold.
Theorem 1.1 ([5]) Assume that V is symmetric. i.e. V (x, y) = V (y, x).
Then the limit set of { µ
t} (for the topology of weak* convergence) is almost surely a compact connected subset of
Fix (Π) = { µ ∈ P (M ) : µ = Π(µ) } .
In particular, if Fix (Π) is finite then (µ
t) converges almost surely toward a fixed point of Π. This holds for a generic function V (see [5]).
Sufficient conditions ensuring that Fix (Π) has cardinal one are as follows:
Theorem 1.2 ([5], [6]) Assume that V is symmetric and that one of the two following conditions hold
(i) Up to an additive constant V is a Mercer kernel, That is V (x, y) = K(x, y) + C
and Z
K(x, y)f (x)f (y)λ(dx)λ(dy) ≥ 0 for all f ∈ L
2.
(ii) For all x ∈ M, y ∈ M, u ∈ T
xM, v ∈ T
yM
Ric
x(u, u) + Ric
y(v, v) + Hess
x,yV ((u, v), (u, v)) ≥ K( k u k
2+ k v k
2) where K is some positive constant. Here Ric
xstands for the Ricci tensor at x and Hess
x,yis the Hessian of V at (x, y).
Then Fix (Π) reduces to a singleton { µ
∗} and µ
t→ µ
∗with probability one.
As observed in [6] the condition (i) in Theorem 1.2 seems well suited to describe self-repelling diffusions. On the other hand, it is not clearly related to the geometry of M. Condition (ii) has a more geometrical flavor and is robust to smooth perturbations (of M and V ). It can be seen as a Bakry- Emery type condition for self interacting diffusions.
In [5], it is also proved that every stable (for the ODE (3)) fixed point
of Π has a positive probability to be a limit point for µ
t; and any unstable
fixed point cannot be a limit point for µ
t.
Organisation of the paper
Let µ
∗∈ Fix (Π). We will assume that
Hypothesis 1.3 µ
tconverges a.s. towards µ
∗. Sufficient conditions are given by Theorem 1.2
In this paper we intend to study the rate of this convergence. Let
∆
t= e
t/2(µ
et− µ
∗).
It will be shown that, under some conditions to be specified later, for all g = (g
1, . . . , g
n) ∈ C(M )
nthe process
[∆
sg
1, . . . , ∆
sg
n, V ∆
s]
s≥tconverges in law, as t → ∞ , toward a certain stationary Ornstein-Uhlenbeck process (Z
g, Z ) on R
n× C(M ). This process is defined in Section 2. The main result is stated in section 3 and some examples are developed. It is in particular observed that a strong repelling interaction gives a faster conver- gence. The section 4 is a proof section. The appendix, section 5, contains general material on random variables and Ornstein-Uhlenbeck processes on C(M ).
In the following K (respectively C) denotes a positive constant (respec- tively a positive random constant). These constants may change from line to line.
2 The Ornstein-Uhlenbeck process (Z g , Z).
Throughout all this section we let µ ∈ P (M ). For x ∈ M we set V
x: M → R defined by V
x(y) = V (x, y).
2.1 The operator G
µLet g ∈ C(M) and let G
µ,g: R × C(M ) → R be the linear operator defined by
G
µ,g(u, f ) = u/2 + Cov
µ(g, f ), (4)
where Cov
µis the covariance on L
2(µ), that is the bilinear form acting on L
2× L
2defined by
Cov
µ(f, g) = µ(f g) − (µf )(µg).
We define the linear operator G
µ: C(M) → C(M ) by
G
µf (x) = G
µ,Vx(f (x), f ) (5)
= f (x)/2 + Cov
µ(V
x, f ).
It is easily seen that k G
µf k ≤ (2 k V k + 1/2) k f k . In particular, G
µis a bounded operator. Let { e
−tGµ} denotes the semigroup acting on C(M ) with generator − G
µ. From now on we will assume the following:
Hypothesis 2.1 There exists κ > 0 and b λ ∈ P (M ) such that µ << b λ with k
dµdλbk
∞< ∞ , λ and b λ are equivalent measures with k
dλdbλk
∞< ∞ and k
ddλbλk
∞< ∞ , and such that for all f ∈ L
2( b λ),
h G
µf, f i
bλ≥ κ k f k
2bλ. Let
λ( − G
µ) = lim
t→∞
log( k e
−tGµk )
t .
This limit exists by subadditivity. Then
Lemma 2.2 Hypothesis 2.1 implies that λ( − G
µ) ≤ − κ < 0.
Proof : For all f ∈ L
2( b λ), d
dt k e
−tGµf k
2bλ= − 2 h G
µe
−tGµf, e
−tGµf i
bλ≤ − 2κ k e
−tGµf k
bλ. This implies that k e
−tGµf k
λb≤ e
−κtk f k
bλ.
Denote by g
tthe solution of the differential equation dg
tdt = Cov
µ(V
x, g
t)
with g
0= f , where f ∈ C(M ). Note that e
−tGµf = e
−t/2g
t. It is straightfor- ward to check that (using the fact that k
dµdbλk
∞< ∞ )
d
dt k g
tk
bλ≤ K k g
tk
bλwith K a constant depending only on V and µ. Thus sup
t∈[0,1]
k g
tk
λb≤ K k f k
bλ. Now, since for all x ∈ M and t ∈ [0, 1]
d dt g
t(x)
≤ K k g
tk
bλ≤ K k f k
bλ,
we have k g
1k ≤ K k f k
bλ. This implies that k e
−Gµf k ≤ K k f k
bλ. Now for all t > 1, and f ∈ C(M),
k e
−tGµf k = k e
−Gµe
−(t−1)Gµf k
≤ K k e
−(t−1)Gµf k
bλ≤ Ke
−κ(t−1)k f k
bλ≤ Ke
−κtk f k
∞.
This implies that k e
−tGµk ≤ Ke
−κt, which proves the lemma. QED The adjoint of G
µis the operator on M (M ) defined by the relation
m(G
µf ) = (G
∗µm)f
for all m ∈ M (M ) and f ∈ C(M ). It is not hard to verify that G
∗µm = 1
2 m + (V m)µ − (µ(V m))µ. (6)
2.2 The generator A
µand its inverse Q
µLet H
2be the Sobolev space of real valued functions on M , associated with the norm k f k
2H= k f k
2λ+ k∇ f k
2λ. Since Π(µ) and λ are equivalent measures with continuous Radon-Nykodim derivative, L
2(Π(µ)) = L
2(λ) := L
2. We denote by K
µthe projection operator, acting on L
2(Π(µ)), defined by
K
µf = f − Π(µ)f.
We denote by A
µthe operator acting on H
2defined by A
µf = 1
2 ∆f − h∇ V µ, ∇ f i . Note that for f and g in L
2,
h A
µf, g i
Π(µ)= − 1 2
Z
h∇ f, ∇ g i (x)Π(µ)(dx) where h· , ·i denotes the Riemannian inner product on M.
For all f ∈ C(M) there exists Q
µf ∈ H
2such that Π(µ)( Q
µf) = 0 and f − Π(µ)f = K
µf = − A
µQ
µf. (7) Note that if P
tµdenotes the semigroup with generator A
µ, then
Q
µf = Z
∞0
P
tµK
µf dt.
Since there exists p
µt( · , · ) such that P
tµf (x) =
Z
M
p
µt(x, y)f(y)Π(µ)(dy), we have
Q
µf (x) = Z
M
q
µ(x, y)f (y)Π(µ)(dy) where
q
µ(x, y ) = Z
∞0
(p
µt(x, y) − 1)dt.
Then, as shown in [3], Q
µf is C
1and there exists a constant K such that for all f ∈ C(M) and µ ∈ P (M),
k Q
µf k
∞≤ K k f k
∞(8) k∇ Q
µf k
∞≤ K k f k
∞. (9) Finally, note that for f and g in L
2,
Z
h∇ Q
µf, ∇ Q
µg i (x)Π(µ)(dx) = − 2 h A
µQ
µf, Q
µg i
Π(µ)(10)
= 2 h f, Q
µg i
Π(µ).
2.3 The covariance C
µWe let C b
µdenote the bilinear continuous form C b
µ: C(M) × C(M ) → R defined by
C b
µ(f, g) = 2 h f, Q
µg i
Π(µ).
This form is symmetric (see its expression given by (10)). Note also that for some constant depending on µ,
| C b
µ(f, g) | ≤ K k f k × k g k .
We let C
µdenote the mapping C
µ: M × M → R defined by C
µ(x, y) = C b
µ(V
x, V
y).
Then C
µis a covariance function (or a Mercer kernel), i.e. it is continuous, symmetric and P
i,j
λ
iλ
jC
µ(x
i, x
j) ≥ 0.
2.4 The process Z
We now define an Ornstein-Uhlenbeck process on C(M ) of covariance C
µand drift − G
µ. This heavily relies on the general construction given in the appendix.
A Brownian motion on C(M) with covariance C
µis a C(M)-valued stochas- tic process W = { W
t}
t≥0such that
(i) W
0= 0;
(ii) t 7→ W
tis continuous;
(iii) For every finite subset S ⊂ R × M, { W
t(x) }
(t,x)∈Sis a centered Gaussian random vector;
(iv) E [W
s(x)W
t(y)] = (s ∧ t)C
µ(x, y ).
Lemma 2.3 There exists a Brownian motion on C(M ) with covariance C
µ. Proof : Let
d
Cµ(x, y ) :=
q
C
µ(x, x) − 2C
µ(x, y) + C
µ(y, y)
= k∇ Q
µ(V
x− V
y) k
Π(µ)≤ K k V
x− V
yk
where the last inequality follows from (9). Then d
Cµ(x, y ) ≤ Kd(x, y )
and the result follows from Proposition 5.8 and Remark 5.7 in the appendix.
QED
We say that a C(M )-valued process Z is an Ornstein-Uhlenbeck process of covariance C
µand drift − G
µif
Z
t= Z
0− Z
t0
G
µZ
sds + W
t(11) where
(i) W is a C(M )-valued Brownian motion of covariance C
µ; (ii) Z
0is a C(M )-valued random variable;
(iii) W and Z
0are independent.
Note that we can think of Z as a solution to the linear SDE dZ
t= dW
t− G
µZ
tdt.
It follows from section 5.3 in the appendix that such a process exists and defines a Markov process. Furthermore
Proposition 2.4 Under hypothesis 2.1,
(i) (Z
t) converges in law toward a C(M )-valued random variable Z
∞; (ii) Z
∞is Gaussian, in the sense that for every finite set S ⊂ M, { Z
∞(x) }
x∈Sis a centered Gaussian random vector;
(iii) Let π
µdenotes the law of Z
∞. Then π
gis characterized by its variance Var (π
µ) : M (M) → R ,
m 7→ E ((mZ
∞)
2),
and for all m ∈ M , Var (π
µ)(m) =
Z
∞0
Z
M×M
C
µ(x, y )m
t(dx)m
t(dy)dt
= Z
∞0
C b
µ(V m
t, V m
t)dt where
m
t= e
−tG∗µm.
Proof : This follows from Proposition 5.16 in the appendix. Example 5.18 shows that assertion (iii) of this proposition is satisfied. QED
2.5 The process Z
g.
For g = (g
1, . . . , g
n) ∈ C(M )
n, let ˜ M = { 1, . . . , n } ∪ M be the disjoint union of { 1, . . . , n } and M, and C
µg: ˜ M × M ˜ → R be the function defined by
C
µg(x, y) =
C b
µ(g
x, g
y) for x, y ∈ { 1, . . . , n } , C
µ(x, y) for x, y ∈ M,
C b
µ(V
x, g
y) for x ∈ M, y ∈ { 1, . . . , n } . Then C
µgis a Mercer kernel (see section 5.2).
A Brownian motion on R
n× C(M ) with covariance C
µgis a R
n× C(M)- valued stochastic process (W
g, W ) = { (W
tg1, . . . , W
tgn, W
t) }
t≥0such that:
(i) W = { W
t}
t≥0is a C(M )-valued Brownian motion with covariance C
µ; (ii) For every finite subset S ⊂ R × M, { W
tg, W
t(x) }
(t,x)∈Sis a centered
Gaussian random vector;
(iii) E (W
sgiW
tgj) = (s ∧ t) C b
µ(g
i, g
j) and E (W
s(x)W
tgi) = (s ∧ t) C b
µ(V
x, g
i).
Lemma 2.5 There exists a Brownian motion on R
n× C(M) with covariance C
µg.
Proof : Let ˜ d be the distance on ˜ M defined by d(x, y) = ˜
1
x6=yfor x, y ∈ { 1, . . . , n } , d(x, y ) for x, y ∈ M,
d(x, x
0) + 1 for x ∈ M, y ∈ { 1, . . . , n }
where x
0is some arbitrary point in M. This makes ˜ M a compact metric space, and it is easy to show that the function C
µgverifies hypothesis 5.6 (use the proof of Lemma 2.3). The result follows by application of Proposition 5.8. QED
Let now be Z
tg= (Z
tg1, . . . , Z
tgn) ∈ R
ndenote the solution to the SDE dZ
tgi= dW
tgi− (Z
tgi/2 + Cov
µ(Z
t, g
i)) dt, i = 1, . . . , n (12) where (W
g, W ) is as above and Z = (Z
t) is given by (11).
The following result generalizes Proposition 2.4.
Proposition 2.6 Under hypothesis 2.1,
(i) The process (Z
tg, Z
t) converges in law toward a centered R
n× C(M) valued Gaussian random variable (Z
∞g, Z
∞).
(ii) Let π
g,µdenotes the law of (Z
∞g, Z
∞). Then π
g,µis characterized by its variance
Var (π
g,µ) : R
n× M (M ) → R , (u, m) 7→ E (mZ
∞+ h u, Z
∞gi )
2; and for all u ∈ R
n, m ∈ M (M ),
Var (π
g,µ)(u, m) = Z
∞0
C b
µ(f
t, f
t)dt with
f
t= e
−t/2X
i
u
ig
i+ V m
t, and where m
tis defined by
m
tf = m
0(e
−tGµf) + X
ni=1
u
iZ
t0
e
−s/2Cov
µ(g
i, e
−(t−s)Gµf)ds. (13)
Proof : Let G
gµ: R
n× C(M ) → R
n× C(M) be the operator defined by G
gµ=
I/2 A
gµ0 G
µ(14)
where A
gµ: C(M ) → R
nis the linear map defined by A
gµ(f) =
Cov
µ(f, g
1), . . . , Cov
µ(f, g
n) .
Then (Z
g, Z) is a C( ˜ M)-valued Ornstein-Uhlenbeck process of covariance C
µgand drift − G
gµ. It is not hard to verify that under hypothesis 2.1, the assumptions of Proposition 5.16 hold, so that (Z
tg, Z
t) converges in law to- ward a centered R
n× C(M ) valued Gaussian random variable (Z
∞g, Z
∞) with variance
Var (π
g,µ)(u, m) = Z
∞0
C b
µ(f
t, f
t)dt with f
t= P
i
u
t(i)g
i+ V m
tand where (u
t, m
t) = e
−t(Ggµ)∗(u, m). Now (G
gµ)
∗=
I/2 0 (A
gµ)
∗(G
µ)
∗and (A
gµ)
∗u = P
i
u
i(g
i− µg
i)µ. Thus u
t= e
−t/2u and dm
tdt = − (A
gµ)
∗u
t− (G
µ)
∗m
tThus m
tis the solution with m
0= m of dm
tdt = − e
−t/2X
i
u
i(g
i− µg
i)
!
µ − G
∗µm
t(15) Note that (15) is equivalent to
d
dt (m
tf ) = − e
−t/2Cov
µX
i
u
ig
i, f
!
− m
t(G
µf) for all f ∈ C(M), and m
0= m. From which we deduce that
m
t= e
−tG∗µm
0− Z
t0
e
−s/2e
−(t−s)G∗µX
i
u
i(g
i− µg
i)µ
! ds
which implies the formula for m
tgiven by (13). QED
For further reference we call (Z
g, Z) an Ornstein-Uhlenbeck process of co-
variance C
µgand drift − G
gµ. It is called stationary when its initial distribution
is π
g,µ.
3 A central limit theorem for µ t
We state here the main results of this article. We assume µ
∗∈ Fix (Π) satisfies hypotheses 1.3 and 2.1. Set ∆
t= e
t/2(µ
et− µ
∗), D
t= V ∆
tand D
t+·= { D
t+s: s ≥ 0 } . Then
Theorem 3.1 D
t+·converges in law, as t → ∞ , towards a stationary Ornstein- Uhlenbeck process of covariance C
µ∗and drift − G
µ∗.
For g = (g
1, . . . , g
n) ∈ C(M )
n, we set D
tg= (∆
tg, D
t) and D
t+·g= { D
t+sg: s ≥ 0 } . Then
Theorem 3.2 (D
gt+s)
s≥0) converges in law towards a stationary Ornstein- Uhlenbeck process of covariance C
µg∗and drift − G
gµ∗.
Define C b : C(M ) × C(M) → R the symmetric bilinear form defined by C(f, g) = b
Z
∞0
C b
µ∗(f
t, g
t)dt, (16) with (g
tis defined by the same formula, with g in place of f )
f
t(x) = e
−t/2f(x) − Z
t0
e
−s/2Cov
µ∗(f, e
−(t−s)Gµ∗V
x)ds. (17)
Corollary 3.3 ∆
tg converges in law towards a centered Gaussian variable Z
∞gof covariance
E [Z
∞giZ
∞gj] = C(f, g). b
Proof : Follows from theorem 3.2 and the calculus of Var (π
g,µ)(u, 0). QED
3.1 Examples
3.1.1 Diffusions
Suppose V (x, y) = V (x), so that (X
t) is just a standard diffusion on M with invariant measure µ
∗=
λexp(−Vexp (−V)λ).
Let f ∈ C(M ). Then f
tdefined by (17) is equal to (using e
−tGµ∗1 = e
−t/21) = e
−t/2f. Thus
C(f, g) = 2µ b
∗(f Q
µ∗g). (18)
Corollary 3.3 says that
Theorem 3.4 For all g ∈ C(M )
n, ∆
gtconverges in law toward a centered Gaussian variable (Z
∞g1, . . . , Z
∞gn), with covariance given by
E (Z
∞giZ
∞gj) = 2µ
∗(g
iQ
µ∗g
j).
Remark 3.5 This central limit theorem for Brownian motions on compact manifolds has already been considered by Baxter and Brosamler in [1] and [2]; and by Bhattacharya in [7] for ergodic diffusions.
3.1.2 The case µ
∗= λ and V symmetric.
Suppose here that µ
∗= λ and that V is symmetric. We assume (without loss of generality since Π(λ) = λ implies that V λ is a constant function) that V λ = 0.
Since V is compact and symmetric, there exists an orthonormal basis (e
α)
i≥0in L
2(λ) and a sequence of reals (λ
α)
α≥0such that e
0is a constant function and
V = X
α≥1
λ
αe
α⊗ e
α.
Assume that for all α, 1/2 + λ
α> 0. Then hypothesis 2.1 holds with b λ = λ, and the convergence of µ
ttowards λ holds with positive probability (see [6]).
Let f ∈ C(M) and f
tdefined by (17), denoting f
α= h f, e
αi
λand f
tα= h f
t, e
αi
λ, we have f
t0= e
−t/2f
0and for α ≥ 1,
f
tα= e
−t/2f
α− λ
αe
−(1/2+λα)te
λαt− 1 λ
αf
α= e
−(1/2+λα)tf
α. Using the fact that
C b
λ(f, g) = 2λ(f Q
λg ), this implies that
C(f, g) = 2 b X
α≥1
X
β≥1
1
1 + λ
α+ λ
βh f, e
αi
λh g, e
βi
λλ(e
αQ
λe
β).
This, with corollary 3.3, proves
Theorem 3.6 Assume hypothesis 1.3 and that 1/2 + λ
α> 0 for all α. Then for all g ∈ C(M )
n, ∆
gtconverges in law toward a centered Gaussian variable (Z
∞g1, . . . , Z
∞gn), with covariance given by
E (Z
∞giZ
∞gj) = C(g b
i, g
j).
In particular,
E (Z
∞eαZ
∞eβ) = 2
1 + λ
α+ λ
βλ(e
αQ
λe
β).
Note that when all λ
αare positive, which corresponds to what is named a self-repelling interaction in [6], the rate of convergence of µ
ttowards λ is bigger than when there is no interaction, and the bigger is the interaction (that is larger λ
α’s) faster is the convergence.
4 Proof of the main results
We assume hypothesis 1.3 and µ
∗satisfies hypothesis 2.1. It is possible to choose κ in hypothesis 2.1 such that κ < 1/2. In the following κ will denote such constant. Note that we have λ( − G
µ∗) < − κ. Such κ exists when hypothesis 2.1 holds.
4.1 A lemma satisfied by Q
µWe denote by X (M ) the space of continuous vector fields on M , and equip the spaces P (M ) and X (M ) respectively with the weak convergence topology and with the uniform convergence topology.
Lemma 4.1 For all f ∈ C(M ), the mapping µ 7→ ∇ Q
µf is a continuous mapping from P (M ) in X (M ).
Proof : Let µ and ν be in M (M ), and f ∈ C(M ). Set g = Q
µf . Then f = − A
µg + Π(µ)f and
k∇ Q
µf − ∇ Q
νf k
∞= k − ∇ Q
µA
µg + ∇ Q
νA
µg k
∞= k∇ g + ∇ Q
νA
µg k
∞≤ k∇ (g + Q
νA
νg ) k
∞+ k∇ Q
ν(A
µ− A
ν)g k
∞since ∇ (g + Q
νA
νg ) = 0 and (A
µ− A
ν)g = h∇ V
µ−ν, ∇ g i , we get
k∇ Q
µf − ∇ Q
νf k
∞≤ K kh∇ V
µ−ν, ∇ g ik
∞. (19) Using the fact that (x, y) 7→ ∇ V
x(y) is uniformly continuous, the right hand term of (19) converges towards 0, when d(µ, ν) converges towards 0, d being a distance compatible with the weak convergence. QED
4.2 The process ∆
Set h
t= V µ
tand h
∗= V µ
∗. Recall ∆
t= e
t/2(µ
et− µ
∗) and D
t= V ∆
t. Note that D
t(x) = ∆
tV
x.
To simplify the notation, we set K
s= K
µs, Q
s= Q
µsand A
s= A
µs. Let (M
tf)
t≥1be the martingale defined by
M
tf= X
α
Z
t1
e
α( Q
sf)(X
s)dB
sα.
The quadratic covariation of M
fand M
g(with f and g in C(M)) is given by
h M
f, M
gi
t= Z
t1
h∇ Q
sf, ∇ Q
sg i (X
s)ds.
Then for all t ≥ 1 (with ˙ Q
t=
dtdQ
t) , Q
tf(X
t) − Q
1f(X
1) = M
tf+
Z
t1
Q ˙
sf(X
s)ds − Z
t1
K
sf (X
s)ds.
Thus
µ
tf = 1 t
Z
t1
K
sf (X
s)ds + 1 t
Z
t1
Π(µ
s)f ds + 1 t
Z
10
f (X
s)ds
= − 1 t
Q
tf (X
t) − Q
1f (X
1) − Z
t1
Q ˙
sf(X
s)ds
+ M
tft + 1
t Z
t1
h ξ(h
s), f i
λds + 1 t
Z
10
f (X
s)ds.
Note that (D
t) is a continuous process taking its values in C(M) and that D
t= e
t/2(h
et− h
∗). For f ∈ C(M ) (using the fact that µ
∗f = h ξ(h
∗), f i
λ),
∆
tf = X
5i=1
∆
itf (20)
with
∆
1tf = e
−t/2− Q
etf(X
et) + Q
1f(X
1) + Z
et1
Q ˙
sf(X
s)ds
!
∆
2tf = e
−t/2M
eft∆
3tf = e
−t/2Z
et1
h ξ(h
s) − ξ(h
∗) − Dξ(h
∗)(h
s− h
∗), f i
λds
∆
4tf = e
−t/2Z
et1
h Dξ(h
∗)(h
s− h
∗), f i
λds
∆
5tf = e
−t/2Z
10
f (X
s)ds − µ
∗f
.
Then D
t= P
5i=1
D
it, where D
ti= V ∆
it. Finally, note that
h Dξ(h
∗)(h − h
∗), f i
λ= − Cov
µ∗(h − h
∗, f). (21)
4.3 First estimates
We recall some estimates from [3]: There exists a constant K such that for all f ∈ C(M ) and t > 0,
k Q
tf k
∞≤ K k f k
∞k∇ Q
tf k
∞≤ K k f k
∞k Q ˙
tf k
∞≤ K
t k f k
∞. These estimates imply in particular that
h M
f− M
gi
t≤ K k f − g k
∞× t and that
Lemma 4.2 There exists a constant K depending on k V k
∞such that for all t ≥ 1, and all f ∈ C(M )
k ∆
1tf k
∞+ k ∆
5tf k
∞≤ K × (1 + t)e
−t/2k f k
∞, (22)
which implies that ((∆
1+ ∆
5)
t+s)
s≥0and ((D
1+ D
5)
t+s)
s≥0both converge
towards 0 (respectively in M (M) and in C( R
+× M )).
We also have
Lemma 4.3 There exists a constant K such that for all t ≥ 0 and all f ∈ C(M ),
E [(∆
2tf)
2] ≤ K k f k
2∞,
| ∆
3tf | ≤ K k f k
λ× e
−t/2Z
t0
k D
sk
2λds,
| ∆
4tf | ≤ K k f k
λ× e
−t/2Z
t0
e
s/2k D
sk
λds.
Proof : The first estimate follows from
E [(∆
2tf)
2] = e
−tE [(M
eft)
2] = e
−tE [ h M
fi
et]
≤ e
−tZ
et1
k∇ Q
sf k
2∞ds
≤ K k f k
2∞.
The second estimate follows from the fact that
k ξ(h) − ξ(h
∗) − Dξ(h
∗)(h − h
∗) k
λ= O( k h − h
∗k
2λ).
The last estimate follows easily after having remarked that
|h Dξ(h
∗)(h
s− h
∗), f i| = | Cov
µ∗(h
s− h
∗, f ) |
≤ K k f k
λ× k h
s− h
∗k
λ≤ K k f k
λ× s
−1/2k D
log(s)k
λ. This proves this lemma. QED
4.4 The processes ∆
′and D
′Set ∆
′= ∆
2+ ∆
3+ ∆
4and D
′= D
2+ D
3+ D
4. For g ∈ C(M), set ǫ
gt= e
t/2h ξ(h
et) − ξ(h
∗) − Dξ(h
∗)(h
et− h
∗), g i
λ.
Then
d∆
′tg = − ∆
′tg
2 dt + dN
tg+ ǫ
gtdt + h Dξ(h
∗)(D
t), g i
λdt
where for all g ∈ C(M), N
gis a martingale. Moreover, for f and g in C(M), h N
f, N
gi
t=
Z
t0
h∇ Q
esf(X
es), ∇ Q
esg (X
es) i ds.
Then, for all x, dD
′t(x) = − D
t′(x)
2 dt + dM
t(x) + ǫ
t(x)dt + h Dξ(h
∗)(D
t), V
xi
λdt
where M is the martingale in C(M ) defined by M (x) = N
Vxand ǫ
t(x) = ǫ
Vtx. We also have
G
µ∗(D
′)
t(x) = D
t′(x)
2 − h Dξ(h
∗)(D
′t), V
xi
λ.
Denoting L
µ∗= L
−Gµ∗(defined by equation (32) in the appendix), this implies that
dL
µ∗(D
′)
t(x) = dD
′t(x) + G
µ∗(D
′)
t(x)dt
= dM
t(x) + h Dξ(h
∗)((D
1+ D
5)
t), V
xi
λdt + ǫ
t(x)dt Thus
L
µ∗(D
′)
t(x) = M
t(x) + Z
t0
ǫ
′s(x)ds with ǫ
′s(x) = ǫ
′sV
xwhere for all f ∈ C(M ),
ǫ
′sf = ǫ
fs+ h Dξ(h
∗)((D
1+ D
5)
s), f i
λ. Using lemma 5.10,
D
′t= L
−1µ∗(M)
t+ Z
t0
e
−(t−s)Gµ∗ǫ
′sds. (23) For g = (g
1, . . . , g
n) ∈ C(M)
n, we denote ∆
′tg = (∆
′tg
1, . . . , ∆
′tg
n), N
g= (N
g1, . . . , N
gn) and ǫ
′tg = (ǫ
′tg
1, . . . , ǫ
′tg
n). Then, denoting L
gµ∗= L
−Ggµ∗
(with G
gµ∗defined by (14)) we have
L
gµ∗(∆
′g, D
′)
t= (N
tg, M
t) + Z
t0
(ǫ
′sg, ǫ
′s)ds
so that (using lemma 5.10 and integrating by parts) (∆
′tg, D
t′) = (L
gµ∗)
−1(N
g, M )
t+
Z
t0
e
−(t−s)Ggµ∗(ǫ
′sg, ǫ
′s)ds. (24) Moreover
(L
gµ∗)
−1(N
g, M)
t=
N b
tg1, . . . , N b
tgn, L
−1µ∗(M )
t,
where
N b
tgi= N
tgi− Z
t0
N
sgi2 + C b
µ∗(L
−1µ∗(M )
s, g
i)
ds.
4.5 Estimation of ǫ
′t4.5.1 Estimation of k L
−1µ∗(M )
tk
λLemma 4.4 (i) For all α ≥ 2, there exists a constant K
αsuch that for all t ≥ 0,
E [ k L
−1µ∗(M)
tk
αλ]
1/α≤ K
α.
(ii) a.s. there exists C with E [C] < ∞ such that for all t ≥ 0, k L
−1µ∗(M )
tk
λ≤ C(1 + t).
Proof : Since k L
−1µ∗(M )
tk
λ≤ K k L
−1µ∗(M )
tk
λb, we estimate k L
−1µ∗(M )
tk
bλ. We have
dL
−1µ∗(M )
t= dM
t− G
µ∗L
−1µ∗(M)
tdt.
Let N be the martingale defined by N
t=
Z
t0
* L
−1µ∗(M)
sk L
−1µ∗(M)
sk
bλ, dM
s+
bλ