El e c t ro nic
Journ a l of
Pr
ob a b il i t y
Vol. 16 (2011), Paper no. 66, pages 1815–1843.
Journal URL
http://www.math.washington.edu/~ejpecp/
Self-interacting diffusions IV: Rate of convergence
∗Michel Benaïm
Institut de Mathématiques Université de Neuchâtel,
Rue Emile-Argand 11, CH-2000 Neuchâtel, michel.benaim@unine.ch
Olivier Raimond
Laboratoire Modal’X, Université Paris Ouest, 200 avenue de la République 92000 Nanterre, France
olivier.raimond@u-paris10.fr
Abstract
Self-interacting diffusions are processes living on a compact Riemannian manifold defined by a stochastic differential equation with a drift term depending on the past empirical measureµt of the process. The asymptotics ofµt is governed by a deterministic dynamical system and under certain conditions(µt)converges almost surely towards a deterministic measureµ∗(see Benaïm, Ledoux, Raimond (2002) and Benaïm, Raimond (2005)). We are interested here in the rate of convergence ofµt towardsµ∗. A central limit theorem is proved. In particular, this shows that greater is the interaction repelling faster is the convergence.
Key words:Self-interacting random processes, reinforced processes.
AMS 2010 Subject Classification:Primary 60K35; Secondary: 60H10, 62L20, 60F05.
Submitted to EJP on July 29, 2009, final version accepted July 18, 2011.
∗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 mapping1. For all finite Borel measureµ, letVµ:M→Rbe the smooth function defined by
Vµ(x) = Z
M
V(x,y)µ(d y). Let(eα) be a finite family of vector fields onM such thatP
αeα(eαf)(x) = ∆f(x), where∆is the Laplace operator onM andeα(f)stands for the Lie derivative of f alongeα. Let(Bα)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)
d Xt=X
α
eα(Xt)◦d Bαt − ∇(Vµt)(Xt)d t
whereµt=1tRt
0δXsdsis the empirical occupation measure of(Xt).
In absence of drift (i.e V = 0), (Xt) 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 trajectory through µt. This type of process was introduced in Benaim, Ledoux and Raimond (2002) ([3]) and further analyzed in a series of papers by Benaim and Raimond (2003, 2005, 2007) ([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 processes with reinforcement we refer the reader to the recent survey paper by Pemantle (2007) ([16]).
Notation and Background
We letM(M)denote the space of finite Borel measures onM,P(M)⊂ M(M)the space of proba- bility measures. IfIis a metric space (typically,I=M,R+×Mor[0,T]×M) we letC(I)denote the space of real valued continuous functions onI equipped with the topology of uniform convergence on compact sets. The normalized Riemann measure on M will be denoted byλ.
Let µ ∈ P(M) and f : M → Ra nonnegative or µ−integrable Borel function. We write µf for R f dµ, and fµ for the measure defined as fµ(A) = R
Af dµ. We let L2(µ) denote the space of functions for which µ|f|2 < ∞, equipped with the inner product 〈f,g〉µ = µ(f g) and the norm kfkµ=p
µf2. We simply write L2 forL2(λ).
Of fundamental importance in the analysis of the asymptotics of(µt)is the mappingΠ:M(M)→ P(M)defined by
Π(µ) =ξ(Vµ)λ (1)
1The mappingVx:M→Rdefined byVx(y) =V(x,y)isC2and its derivatives are continuous in(x,y).
whereξ:C(M)→C(M)is the function defined by ξ(f)(x) = e−f(x)
R
Me−f(y)λ(d y). (2)
In[3], it is shown that the asymptotics ofµtcan be precisely related to the long term behavior of a certain semiflow onP(M)induced by the ordinary differential equation (ODE) onM(M):
µ˙=−µ+ Π(µ). (3)
Depending on the nature ofV, the dynamics of (3) can either be convergent or nonconvergent lead- ing to similar behaviors for{µt}(see[3]). WhenV is symmetric, (3) happens to be aquasigradient and the following convergence result holds.
Theorem 1.1 ([5]). Assume that V is symmetric, i.e. V(x,y) =V(y,x). Then the limit set of {µt} (for the topology of weak* convergence) is almost surely a compact connected subset of
Fix(Π) ={µ∈ P(M):µ= Π(µ)}.
In particular, ifFix(Π) is finite then(µt) converges almost surely toward a fixed point ofΠ. This holds for a generic functionV (see[5]). Sufficient conditions ensuring thatFix(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: For some constant C, V(x,y) =K(x,y) +C,
and for all f ∈L2,
Z
K(x,y)f(x)f(y)λ(d x)λ(d y)≥0.
(ii) For all x ∈M,y∈M,u∈TxM,v∈TyM
Ricx(u,u) +Ricy(v,v) +Hessx,yV((u,v),(u,v))≥K(kuk2+kvk2)
where K is some positive constant. HereRicx stands for the Ricci tensor at x andHessx,y is the Hessian of V at(x,y).
ThenFix(Π)reduces to a singleton{µ∗}andµt→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 ofΠhas a positive probability to be a limit point forµt; and any unstable fixed point cannot be a limit point forµt.
Organisation of the paper Letµ∗∈Fix(Π). We will assume that
Hypothesis 1.3. µtconverges a.s. towardsµ∗.
In this paper we intend to study the rate of this convergence. Let
∆t=et/2(µet−µ∗).
It will be shown that, under some conditions to be specified later, for all g= (g1, . . . ,gn)∈C(M)n the process
∆sg1, . . . ,∆sgn,V∆s
s≥t converges in law, as t → ∞, toward a certain stationary Ornstein-Uhlenbeck process(Zg,Z)onRn×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 convergence. The section 4 is a proof section.
In the following K (respectively C) denotes a positive constant (respectively a positive random constant). These constants may change from line to line.
2 The Ornstein-Uhlenbeck process ( Z
g, Z ) .
For a more precise definition of Ornstein-Uhlenbeck processes on C(M) and their basic properties, we refer the reader to the appendix (section 5). Throughout all this section we letµ∈ P(M)and g= (g1, ...,gn)∈C(M)n. Forx ∈M we setVx :M →Rdefined byVx(y) =V(x,y).
2.1 The operatorGµ
Leth∈C(M)and letGµ,h:R×C(M)→Rbe the linear operator defined by
Gµ,h(u,f) =u/2+Covµ(h,f), (4) where Covµis the covariance on L2(µ), that is the bilinear form acting onL2×L2defined by
Covµ(f,h) =µ(f h)−(µf)(µh). We define the linear operatorGµ:C(M)→C(M)by
Gµf(x) =Gµ,Vx(f(x),f) = f(x)/2+Covµ(Vx,f). (5) It is easily seen that kGµfk∞≤(2kVk∞+1/2)kfk∞. In particular,Gµ is a bounded operator. Let {e−t Gµ}denote the semigroup acting on C(M) with generator−Gµ. From now on we will assume the following:
Hypothesis 2.1. There exists κ > 0 such that µ << λ with kdµdλk∞ < ∞, and such that for all f ∈L2(λ),〈Gµf,f〉λ≥κkfk2λ.
Let
λ(−Gµ) = lim
t→∞
log(ke−t Gµk)
t .
This limit exists by subadditivity. Then
Lemma 2.2. Hypothesis 2.1 implies thatλ(−Gµ)≤ −κ <0.
Proof : For all f ∈L2(λ), d
d tke−t Gµfk2λ=−2〈Gµe−t Gµf,e−t Gµf〉λ≤ −2κke−t Gµfkλ.
This implies thatke−t Gµfkλ≤e−κtkfkλ. Denote by gt the solution of the differential equation d
d tgt(x) =Covµ(Vx,gt)
with g0 = f ∈C(M). Note that e−t Gµf = e−t/2gt. It is straightforward to check that (using the fact that kdµdλk∞ < ∞) d
d tkgtkλ ≤ Kkgtkλ with K a constant depending only on V and µ. Thus supt∈[0,1]kgtkλ≤Kkfkλ. Now, since for allx ∈M andt∈[0, 1]
d d tgt(x)
≤Kkgtkλ≤Kkfkλ, we havekg1k∞≤Kkfkλ. This implies thatke−Gµfk∞≤Kkfkλ. Now for allt>1, and f ∈C(M),
ke−t Gµfk∞ = ke−Gµe−(t−1)Gµfk∞ ≤ Kke−(t−1)Gµfkλ
≤ Ke−κ(t−1)kfkλ ≤ Ke−κtkfk∞. This implies thatke−t Gµk ≤Ke−κt, which proves the lemma. QED TheadjointofGµ is the operator onM(M)defined by the relation
m(Gµf) = (Gµ∗m)f for allm∈ M(M)and f ∈C(M). It is not hard to verify that
Gµ∗m= 1
2m+ (V m)µ−(µ(V m))µ. (6)
2.2 The generatorAµ and its inverseQµ
LetH2be the Sobolev space of real valued functions onM, associated with the normkfk2H=kfk2λ+ k∇fk2λ. Since Π(µ) and λ are equivalent measures with continuous Radon-Nykodim derivative, L2(Π(µ)) =L2(λ). We denote byKµthe projection operator, acting onL2(Π(µ)), defined by
Kµf = f −Π(µ)f. We denote byAµthe operator acting onH2 defined by
Aµf = 1
2∆f − 〈∇Vµ,∇f〉.
Note that for f andhinH2 (denoting〈·,·〉the Riemannian inner product onM)
〈Aµf,h〉Π(µ)=−1 2 Z
〈∇f,∇h〉(x)Π(µ)(d x). For all f ∈C(M)there existsQµf ∈H2such thatΠ(µ)(Qµf) =0 and
f −Π(µ)f =Kµf =−AµQµf. (7)
It is shown in[3]thatQµf isC1 and that there exists a constantK such that for all f ∈C(M)and µ∈ P(M),
kQµfk∞+k∇Qµfk∞≤Kkfk∞. (8) Finally, note that for f andhin L2,
Z
〈∇Qµf,∇Qµh〉(x)Π(µ)(d x) =−2〈AµQµf,Qµh〉Π(µ)=2〈f,Qµh〉Π(µ). (9)
2.3 The covarianceCµg
We letCbµdenote the bilinear continuous formCbµ:C(M)×C(M)→Rdefined by Cbµ(f,h) =2〈f,Qµh〉Π(µ).
This form is symmetric (see its expression given by (9)). Note also that for some constantKdepend- ing onµ,|Cbµ(f,h)| ≤Kkfk∞× khk∞.
We let Cµ denote the mapping Cµ : M × M → R defined by Cµ(x,y) = Cbµ(Vx,Vy). Let ˜M = {1, . . . ,n} ∪M andCµg: ˜M×M˜ →Rbe the function defined by
Cµg(x,y) =
Cbµ(gx,gy) for x,y ∈ {1, . . . ,n}, Cµ(x,y) for x,y∈M,
Cbµ(Vx,gy) for x∈M, y ∈ {1, . . . ,n}. ThenCµ andCµgare covariance functions (as defined in subsection 5.2).
In the following, when n= 0, ˜M = M and Cµg = Cµ. When n ≥ 1, C(M˜) can be identified with Rn×C(M).
Lemma 2.3. There exists a Brownian motion onRn×C(M)with covariance Cµg. Proof : Since the argument are the same forn≥1, we just do it forn=0. Let
dCµ(x,y) := p
Cµ(x,x)−2Cµ(x,y) +Cµ(y,y)
= k∇Qµ(Vx−Vy)kΠ(µ) ≤ KkVx−Vyk∞
where the last inequality follows from (8). ThendCµ(x,y)≤K d(x,y). Thus dCµ satisfies (30) and we can apply Theorem 5.4 of the appendix (section 5). QED
2.4 The process(Zg,Z)
LetGµg:Rn×C(M)→Rn×C(M)be the operator defined by Gµg=
In/2 Agµ 0 Gµ
(10) where In is the identity matrix onRn andAgµ :C(M)→Rn is the linear map defined byAgµ(f) =
Covµ(f,g1), . . . , Covµ(f,gn) .
SinceGµgis a bounded operator, for any lawν onRn×C(M), there exists ˜Z= (Zg,Z)an Ornstein- Uhlenbeck process of covarianceCµg and drift−Gµg, with initial distribution given byν (using Theo- rem 5.6). More precisely, ˜Z is the unique solution of
¨ d Zt = dWt−GµZtd t d Ztgi = dWtgi −
Ztgi/2+Covµ(Zt,gi)
d t, i=1, . . . ,n (11) where ˜Z0is aRn×C(M)-valued random variable of lawν and ˜W= (Wg,W)is aRn×C(M)-valued Brownian motion of covariance Cµg independent of ˜Z. In particular, Z is an Ornstein-Uhlenbeck process of covarianceCµ and drift−Gµ. Denote byPg,t µthe semigroup associated to ˜Z. Then Proposition 2.4. Assume hypothesis 2.1. Then there existsπg,µthe law of a centered Gaussian variable inRn×C(M), with varianceVar(πg,µ)where for(u,m)∈Rn× M(M),
Var(πg,µ)(u,m) := E(mZ∞+〈u,Z∞g〉)2
= Z ∞
0
Cbµ(ft,ft)d t with ft =e−t/2P
iuigi+V mt,and where mtis defined by mtf =m0(e−t Gµf) +
Xn
i=1
ui Z t
0
e−s/2Covµ(gi,e−(t−s)Gµf)ds. (12)
Moerover,
(i) πg,µ is the unique invariant probability measure ofPt.
(ii) For all bounded continuous function ϕ on Rn × C(M) and all (u,f) ∈ Rn × C(M), limt→∞Pg,µt ϕ(u,f) =πg,µϕ.
Proof : This is a consequence of Theorem 5.7. To apply it one can remark thatGµg is an operator like the ones given in example 5.11.
The varianceVar(πg,µ)is given byVar(πg,µ)(ν) =R∞
0 〈ν,e−sGµgCµges(Gµg)∗ν〉dsforν= (u,m)∈Rn× M(M) =C(M˜)∗. ThusVar(πg,µ)(u,m) =R∞
0 Cbµ(ft,ft)d t with ft =P
iut(i)gi+V mt and where (ut,mt) =e−t(Gµg)∗(u,m). Now
(Gµg)∗=
I/2 0 (Agµ)∗ (Gµ)∗
and(Agµ)∗u=P
iui(gi−µgi)µ. Thusut=e−t/2uandmt is the solution withm0=mof d mt
d t =−e−t/2 X
i
ui(gi−µgi)
!
µ−(Gµ)∗mt. (13) Note that (13) is equivalent to
d
d t(mtf) =−e−t/2Covµ X
i
uigi,f
!
−mt(Gµf) for all f ∈C(M), andm0=m. From which we deduce that
mt=e−t G∗µm0− Z t
0
e−s/2e−(t−s)G∗µ X
i
ui(gi−µgi)µ
! ds which implies the formula formt given by (12). QED
An Ornstein-Uhlenbeck process of covarianceCµg and drift −Gµg will be called stationarywhen its initial distribution isπg,µ.
3 A central limit theorem for µ
tWe state here the main results of this article. We assumeµ∗∈Fix(Π)satisfies hypotheses 1.3 and 2.1. Set∆t=et/2(µet−µ∗), Dt=V∆t andDt+·= (Dt+s)s≥0. Then
Theorem 3.1. Dt+· converges in law, as t → ∞, towards a stationary Ornstein-Uhlenbeck process of covariance Cµ∗ and drift−Gµ∗.
Forg∈C(M)n, we setDtg= (∆tg,Dt)andDt+·g = (Dgt+s)s≥0. Then
Theorem 3.2. Dt+·g converges in law towards a stationary Ornstein-Uhlenbeck process of covariance Cµg∗ and drift−Gµg∗.
DefineCb:C(M)×C(M)→Rthe symmetric bilinear form defined by
Cb(f,h) = Z ∞
0
Cbµ∗(ft,ht)d t, (14) with (ht is defined by the same formula, withhin place of f)
ft(x) =e−t/2f(x)− Z t
0
e−s/2Covµ∗(f,e−(t−s)Gµ∗Vx)ds. (15) Corollary 3.3. ∆tg converges in law towards a centered Gaussian variable Z∞g of covariance
E[Z∞giZ∞gj] =Cb(gi,gj).
Proof : Follows from theorem 3.2 and the calculus ofVar(πg,µ)(u, 0). QED
3.1 Examples
3.1.1 Diffusions
Suppose V(x,y) = V(x), so that (Xt) is just a standard diffusion on M with invariant measure µ∗= λe x p(−exp(−VV)λ).
Let f ∈C(M). Sincee−t Gµ∗1=e−t/21, ft defined by (15) is equal toe−t/2f. Thus
Cb(f,g) =2µ∗(fQµ∗g). (16) Corollary 3.3 says that
Theorem 3.4. For all g ∈ C(M)n, ∆gt converges in law toward a centered Gaussian variable (Z∞g1, . . . ,Z∞gn), with covariance given by
E(Z∞giZ∞gj) =2µ∗(giQµ∗gj).
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 caseµ∗=λandV symmetric.
Suppose here thatµ∗ = λ and that V is symmetric. We assume (without loss of generality since Π(λ) =λimplies thatVλis a constant function) thatVλ=0.
Since V is compact and symmetric, there exists an orthonormal basis (eα)α≥0 in L2(λ) and a se- quence of reals(λα)α≥0 such thate0is a constant function and
V =X
α≥1
λαeα⊗eα.
Assume that for allα, 1/2+λα> 0. Then hypothesis 2.1 is satisfied, and the convergence of µt
towardsλholds with positive probability (see[6]).
Let f ∈C(M) and ft defined by (15), denoting fα = 〈f,eα〉λ and ftα = 〈ft,eα〉λ, we have ft0 = e−t/2f0and forα≥1,
ftα = e−t/2fα−λαe−(1/2+λα)t
eλαt−1 λα
fα
= e−(1/2+λα)tfα.
Using the fact thatCbλ(f,g) =2λ(fQλg), this implies that
Cb(f,g) =2X
α≥1
X
β≥1
1
1+λα+λβ〈f,eα〉λ〈g,eβ〉λλ(eαQλeβ). This, with corollary 3.3, proves
Theorem 3.6. Assume hypothesis 1.3 and that 1/2+λα > 0 for all α. Then for all g ∈ C(M)n,
∆gt converges in law toward a centered Gaussian variable (Z∞g1, . . . ,Z∞gn), with covariance given by E(Z∞giZ∞gj) =Cb(gi,gj).
In particular,
E(Z∞eαZ∞eβ) = 2
1+λα+λβλ(eαQλeβ).
When allλαare positive, which corresponds to what is named a self-repelling interaction in[6], the rate of convergence ofµt towardsλis bigger than when there is no interaction, and the bigger is the interaction (that is largerλα’s) faster is the convergence.
4 Proof of the main results
We assume hypothesis 1.3 and µ∗ satisfies hypothesis 2.1. For convenience, we choose for the constantκin hypothesis 2.1 a constant less than 1/2. In all this section, we fix g = (g1, ...,gn) ∈ C(M)n.
4.1 A lemma satisfied byQµ
We denote byX(M)the space of continuous vector fields on M, and equip the spacesP(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→ ∇Qµf is a continuous mapping fromP(M) in X(M).
Proof : Letµandν be inM(M), and f ∈C(M). Seth=Qµf. Then f =−Aµh+ Π(µ)f and k∇Qµf − ∇Qνfk∞ = k − ∇QµAµh+∇QνAµhk∞
= k∇h+∇QνAµhk∞
≤ k∇(h+QνAνh)k∞+k∇Qν(Aµ−Aν)hk∞. Since∇(h+QνAνh) =0 and(Aµ−Aν)h=〈∇Vµ−ν,∇h〉, we get
k∇Qµf − ∇Qνfk∞≤Kk〈∇Vµ−ν,∇h〉k∞. (17) Using the fact that (x,y) 7→ ∇Vx(y) is uniformly continuous, the right hand term of (17) con- verges towards 0, whend(µ,ν)converges towards 0, d being a distance compatible with the weak convergence. QED
4.2 The process∆
Setht=Vµt andh∗=Vµ∗. Recall∆t=et/2(µet−µ∗)andDt(x) =V∆t(x) = ∆tVx. Then(Dt)is a continuous process taking its values inC(M)andDt=et/2(het−h∗).
To simplify the notation, we set Ks = Kµs, Qs =Qµs andAs =Aµs. Let(Mtf)t≥1 be the martingale defined byMtf =P
α
Rt
1 eα(Qsf)(Xs)d Bαs. The quadratic covariation ofMf andMh(with f andhin C(M)) is given by
〈Mf,Mh〉t= Z t
1
〈∇Qsf,∇Qsh〉(Xs)ds.
Then for allt≥1 (with ˙Qt= d tdQt) ,
Qtf(Xt)−Q1f(X1) =Mtf + Z t
1
Q˙sf(Xs)ds− Z t
1
Ksf(Xs)ds.
Thus
µtf = 1 t
Z t
1
Ksf(Xs)ds+1 t
Z t
1
Π(µs)f ds+1 t
Z 1
0
f(Xs)ds
= −1 t
Qtf(Xt)−Q1f(X1)− Z t
1
Q˙sf(Xs)ds
+ Mtf t +1
t Z t
1
〈ξ(hs),f〉λds+1 t
Z 1
0
f(Xs)ds.
For f ∈C(M)(using the fact thatµ∗f =〈ξ(h∗),f〉λ),∆tf =P5
i=1∆itf with
∆1tf = e−t/2 −Qetf(Xet) +Q1f(X1) + Z et
1
Q˙sf(Xs)ds
!
∆2tf = e−t/2Meft
∆3tf = e−t/2 Z et
1
〈ξ(hs)−ξ(h∗)−Dξ(h∗)(hs−h∗),f〉λds
∆4tf = e−t/2 Z et
1
〈Dξ(h∗)(hs−h∗),f〉λds
∆5tf = e−t/2 Z 1
0
f(Xs)ds−µ∗f
! .
ThenDt=P5
i=1Dit, whereDit=V∆it. Finally, note that
〈Dξ(h∗)(h−h∗),f〉λ=−Covµ∗(h−h∗,f). (18) 4.3 First estimates
We recall the following estimate from[3]: There exists a constantK such that for allf ∈C(M)and t>0,
kQ˙tfk∞≤ K tkfk∞.
This estimate, combined with (8), implies that for f andhinC(M),
〈Mf −Mh〉t≤Kkf −hk∞×t and that
Lemma 4.2. There exists a constant K depending onkVk∞such that for all t≥1, and all f ∈C(M) k∆1tfk∞+k∆5tfk∞≤K×(1+t)e−t/2kfk∞, (19) which implies that((∆1+ ∆5)t+s)s≥0 and((D1+D5)t+s)s≥0 both converge towards0(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≥0and all f ∈C(M), E[(∆2tf)2] ≤ Kkfk2∞,
|∆3tf| ≤ Kkfkλ×e−t/2 Z t
0
kDsk2λds,
|∆4tf| ≤ Kkfkλ×e−t/2 Z t
0
es/2kDskλds.
Proof : The first estimate follows from
E[(∆2tf)2] =e−tE[(Meft)2] =e−tE[〈Mf〉et] ≤ Kkfk2∞. The second estimate follows from the fact that
kξ(h)−ξ(h∗)−Dξ(h∗)(h−h∗)kλ=O(kh−h∗k2λ).
The last estimate follows easily after having remarked that
|〈Dξ(h∗)(hs−h∗),f〉|=|Covµ∗(hs−h∗,f)| ≤Kkfkλ× khs−h∗kλ. This proves this lemma. QED
4.4 The processes∆0 and D0
Set∆0= ∆2+ ∆3+ ∆4andD0=D2+D3+D4. For f ∈C(M), set εtf =et/2〈ξ(het)−ξ(h∗)−Dξ(h∗)(het−h∗),f〉λ. Then
d∆0tf =−∆0tf
2 d t+d Ntf +εtfd t+〈Dξ(h∗)(Dt),f〉λd t where for all f ∈C(M),Nf is a martingale. Moreover, for f andhinC(M),
〈Nf,Nh〉t= Z t
0
〈∇Qesf(Xes),∇Qesh(Xes)〉ds.
Then, for all x,
d D0t(x) =−D0t(x)
2 d t+d Mt(x) +εt(x)d t+〈Dξ(h∗)(Dt),Vx〉λd t
whereM is the martingale inC(M)defined byM(x) =NVx andεt(x) =εVtx. We also have Gµ∗(D0)t(x) = D0t(x)
2 − 〈Dξ(h∗)(D0t),Vx〉λ. DenotingLµ∗= L−Gµ∗ (defined by equation (32) in the appendix (section 5)),
d Lµ∗(D0)t(x) = d D0t(x) +Gµ∗(D0)t(x)d t and we have
Lµ∗(D0)t(x) =Mt(x) + Z t
0
ε0s(x)ds withε0s(x) =ε0sVx where for all f ∈C(M),
ε0sf =εsf +〈Dξ(h∗)((D1+D5)s),f〉λ. Using lemma 5.5,
D0t=L−µ∗1(M)t+ Z t
0
e−(t−s)Gµ∗ε0sds. (20) Denote ∆tg = (∆tg1, . . . ,∆tgn), ∆0tg = (∆0tg1, . . . ,∆0tgn) , Ng = (Ng1, . . . ,Ngn) and ε0tg = (ε0tg1, . . . ,ε0tgn). Then, denoting Lµg∗=L−Gg
µ∗ (withGµg∗ defined by (10)) we have Lµg∗(∆0g,D0)t= (Ntg,Mt) +
Z t
0
(ε0sg,ε0s)ds
so that (using lemma 5.5 and integrating by parts) (∆0tg,D0t) = (Lµg∗)−1(Ng,M)t+
Z t
0
e−(t−s)G
g
µ∗(ε0sg,ε0s)ds. (21) Moreover
(Lµg∗)−1(Ng,M)t=
Nbtg1, . . . ,Nbtgn,L−µ∗1(M)t
, where
Nbtgi =Ntgi− Z t
0
Nsgi
2 +Cbµ∗(L−µ∗1(M)s,gi)
ds.