Local behaviour of local times of super-Brownian motion Comportement local des temps locaux du super-mouvement
brownien
Mathieu Merle
D.M.A., Ecole normale supérieure, 45, rue d’Ulm, 75005 Paris, France Received 22 February 2005; accepted 14 June 2005
Available online 7 December 2005
Abstract
Forx0∈Rd\ {0},d3, we study the local behaviour near 0 of the local times(Lyt)of a super-Brownian motionXinitially inδx0. More precisely, ifψ (c)is a suitable normalization, our main result implies that the process(ψ (c)(Lx/ct −L0t), x∈Rd, t0)converges in distribution to a non-degenerate limit asc→ ∞. This allows us to study the local behaviour of the occupation measure ofX, then to recover and to generalise a result of Lee concerning the occupation measure of three-dimensional super- Brownian motion conditioned to hit a distant ball.
©2005 Elsevier SAS. All rights reserved.
Résumé
Pourx0∈Rd\ {0},d3, on étudie le comportement local au voisinage de 0 des temps locaux(Lyt)du super-mouvement brownienXde valeur initialeδx0. Plus précisément, si on noteψ (c)la normalisation adéquate, notre résultat principal implique que le processus(ψ (c)(Lx/ct −L0t), x∈Rd, t0)converge en loi lorsquec→ ∞vers une limite non dégénérée. Ce résultat nous permettra d’étudier le comportement local de la mesure d’occupation deX, puis de redémontrer et généraliser un résultat de Lee concernant la mesure d’occupation d’un super-mouvement brownien tri-dimensionel conditionné à toucher une boule lointaine.
©2005 Elsevier SAS. All rights reserved.
Keywords:Super-Brownian motion; Local time; Occupation measure
1. Introduction and statement of results 1.1. Introduction
The main goal of this work is to study the local behaviour of the occupation measure of super-Brownian motion in dimensiond3.
E-mail address:merle@dma.ens.fr (M. Merle).
0246-0203/$ – see front matter ©2005 Elsevier SAS. All rights reserved.
doi:10.1016/j.anihpb.2005.06.004
Super-Brownian motion is a model for spatial populations undergoing a continuous branching phenomenon, which arises in a variety of different contexts. It was introduced independently by Watanabe (69) and Dawson (75) as the weak limit of branching particle systems. Connections between the wider class of superprocesses and partial differ- ential equations have helped derive some of the basic properties of super-Brownian motion and have also allowed to prove analytic results using probabilistic arguments. More recently, it has been shown that super-Brownian motion appears in scaling limit of a wide range of lattice systems such as the voter model, the contact process, lattice trees or oriented percolation. Therefore, in a way similar to standard Brownian motion, super-Brownian motion can be thought of as a universal object providing information on the asymptotics of many interacting particle systems or statistical mechanics models.
Local times of superprocesses have been studied by many authors (cf. Sugitani [9], Adler and Lewin [1], Krone [4]).
Our main result Theorem 1 gives precise information about the local behaviour of local times of super-Brownian motion, in dimensiond=2 or 3. Let us be more precise. Letx0∈Rd\ {0},Xa super-Brownian motion initially in δx0, andLxt the local time ofX at timet and pointx. As a direct consequence of Theorem 1, ifx∈Rd and if we setψ (c)=√
cwhend=3,ψ (c)=c(ln(c))−1/2whend=2, we will obtain that(ψ (c)(Lx/ct −L0t))t0converges in distribution asc→ ∞to(βa(x)L0
t)t0whereβ is a one-dimensional Brownian motion which is independent ofX anda(x)is a constant depending only onx. Theorem 1 in fact gives a more general statement involving finitely many different values ofx. This allows us to study the local behaviour of the occupation measure ofX(Proposition 2).
Our results are related to a recent paper of Lee [5]. Lee considers a super-Brownian motion started atδcx0 withc large and conditioned on hitting the unit ball. Using analytic methods, he obtains limit theorems for the occupation measure of this ball by super-Brownian motion (see Proposition 1 below). We will show how to recover and to generalise Lee’s results from our main theorem. To do this, we will need to study the local behaviour of the occupation measure ofXunder its canonical measureNx0 (Proposition 5).
Our results on the local behaviour of local times of super-Brownian motion are analogous to the ones obtained by Yor for local times of standard Brownian motion. LetB be a linear Brownian motion started at the origin and letxt denote the local time ofB at levelx and timet. Then Yor [11], proved that(√
c(x/ct −0t))t0, x0converges in distribution to a Brownian sheet asc→ ∞. This should be compared with Theorem 1 below.
After introducing the basic notation, Section 1 first summarises known results which motivate our study, mainly found in [5,9], and [7], then states our results. It also provides an outline of the proof of our main result (Theorem 1) about the local behaviour of local times. Section 2 is devoted to the proof of Theorem 1. In Section 3, we apply Theorem 1 to prove a non-conditional equivalent form of Lee’s result (Proposition 2), then we recover and extend Lee’s result (Proposition 1). Section 4 is devoted to the one-dimensional case.
1.2. Notation
Let MF(Rd) be the space of finite measures on Rd. For μ∈ MF(Rd), f:Rd →R, μ, f is shorthand for
Rdf (x)μ(dx).
We denote byB(Rd)the Borelσ-algebra inRd. The notationB(x, r)stands for the open ball of radiusrcentered atx∈Rd.
We denote byCb(Rd)the space of bounded continuous functions fromRdintoR, and byCb(Rd)+the space of non-negative functions inCb(Rd). IfK is a compact subset ofRd,CK(Rd)is the subset ofCb(Rd)consisting of functions supported onK. Ifn∈N, we letCbn(Rd)be the set of allntimes continuously differentiable functions on Rdwith bounded derivatives of order less thann, andCb∞(Rd):=
n0Cbn(Rd).
We denote byptthe transition density ofd-dimensional Brownian motion, that is fort0,z∈Rd, pt(z)=(2π t )−d/2exp
−|z|2 2t
. Setqt(x)=t
0ps(x)ds, and ifμ∈MF(Rd), μqt(z)=
Rd
qt(z−y)μ(dy).
We consider a super-Brownian motion (Xt, t0)in the space MF(Rd). We will writePμ for the probability measure under whichX0=μ. When there is no confusion we will writeP forPμ,EforEμ. We denote by(FtX)t0 the right-continuous filtration generated byX.
1.3. Basic properties of super-Brownian motion and its local times
In this section we consider a super-Brownian motionXunderPμ, forμ∈MF(Rd).
It is well known (see for example [7], Theorem II.5.9) that for any function φ∈Cb(Rd)+ we can express Eμ[exp(−t
0Xs, φds)]in terms of the solution of a partial differential equation:
Eμ
exp
− t 0
Xs, φds =exp
−
v(t, x, φ)μ(dx)
, (1)
where(t, x)→v(t, x, φ)is continuous onR+×Rdand solves the partial differential equation
∂v(s, x)
∂s =1
2v(s, x)−
v(s, x)2
+φ, (2)
on]0,∞[ ×Rd, with initial valuev(0, x)=0.
(1) is a particular case of theLaplace equationfor the super-Brownian motionX. Note that formula (1) remains valid if φ∈CK(Rd)is not necessarily non-negative, but only then ift is less than a certain explosion timet∗>0.
More precisely, for a fixedT >0, (1) will remain valid fortT for any compactly supported function that is bounded from below by a constant depending onT.
As shown in Chapter IV of [6] by the Brownian snake approach (see also Section II.7 of [7]), for anyx ∈Rd, there exists aσ-finite measureNxonC(R+, MF(Rd))called the excursion measure of super-Brownian motion such that the law of(Xt)t >0underPμis the same as the law of(
Xit)t >0, where
δXi is a Poisson point process with intensity
Nxdμ(x). This fact is called thecanonical decomposition of super-Brownian motion.
Intuitively, if one thinks of super-Brownian motion as the scaling limit of critical branching random walks as it is introduced in [7], the measureXt underNx fort >0 represents the contribution to the population of the descendants at time t of one single individual alive at time 0 at the point x. The canonical decomposition expresses that the super-Brownian motion at timetis obtained by superimposing a Poisson number of such contributions.
It is also well known that the processXsolves amartingale problem(see [7], Theorem II.5.1). For anyφ∈Cb2(Rd), Xt, φ = X0, φ +Mt(φ)+1
2 t 0
Xs, φds, (3)
whereMt(φ)is anFtX-martingale such thatM0(φ)=0 and the quadratic variation ofM(φ)is M(φ)
t= t 0
Xs, φ2 ds.
Sugitani [9] proved that ford3, there exists a random continuous function(t, x)→Lxt from(0,∞)×Rdinto R+such that for anyΨ ∈Cb(Rd),
t 0
Xs, Ψds=
Rd
Ψ (x)Lxt dx. (4)
Lxt is called thelocal timeofXat pointx∈Rdand timet >0.
TakeLx0=0 for everyx∈Rd. The function(Lxt)t0, x∈Rd needs not being continuous at points of the form(0, x), x∈Rd. However Sugitani established ford2 thatLxt is continuous in the pair(x, t )on the set of continuity points forμqt(x). For example, note that this set containsR+×B(0, r)wheneverμ(B(0, r))=0.
Sugitani also proved that ford=1,Lxt is always continuous in the pair(x, t )and even differentiable with respect to the space variable at pointsx∈Rd whereμ({x})=0.
Under the assumption that(t, x)→μqt(x)is continuous in R+×Rd, Sugitani [9] obtained the existence of exponential moments for local times: for anyT >0, there exists a constantK(μ, T ) >0 such that
Eμ exp
rLxt
<∞ (5)
holds for anytT,x∈Rd andr <K(μ, T ). We shall denote byGt,xμ :{z∈C, |z|<K(μ, t )} →Cthe function such that for anyz∈C,|z|<K(μ, t ),
Eμ
exp zLxt
=exp
Gt,xμ (z)
. (6)
For convenience we will writeGt,xδ
x0 =:Gt,xx0 .
IfK is a compact set inRd, and if we only know that(t, x)→μqt(x)is continuous inR+×K (for example if μ(K)=0), it is easy to adapt the proof of (5) in [9] to obtain for anyT >0 the existence of constantsK(μ, T , K) >0, C(μ, T , K) >0 such that
Eμ exp
rLxt
C(μ, T , K) (7)
holds for anytT,x∈K andr <K(μ, T , K). In this case, this clearly allows us to define the functionsGt,xμ for t0, x∈Kin a way such that (6) remains valid forz∈C, |z|<K(μ, t, K).
Let us now discuss thescaling properties of super-Brownian motion. Letc >0, and iff ∈Cb(Rd)+letfcbe the functionfc(x)=f (cx). From (2) we get
v(t, x, fc)=c2v
c2t, cx, c−4f .
The scaling properties of super-Brownian motion easily follow from (1) and that observation. Letμ∈M(Rd). For anyA∈B(Rd),t0, let
X(c)t (A)=c−2Xc2t(cA), μ(c)(A)=c2μ c−1A
.
Then, the law of the process(X(c)t , t0)underPμ(c)is equal to the law of the process(Xt, t0)underPμ. Consequently, for anyφ∈CK(Rd), the law of the process
c−4
c2t
0
Xu, φdu, t0
underPμ(c)is the same as the law of the process t
0
Xu, φcdu, t0
underPμ.
We are now in position to state our main result concerning the local behaviour of local times of super-Brownian motion.
1.4. The main result
Theorem 1.Assumed=2or3. Let ψ (c)=
√
c ifd=3,
√c
ln(c) ifd=2.
Letν∈MF(Rd)such thatν(B(0, ρ))=0for someρ >0, and letXbe a super-Brownian motion inRd with initial valueν. Ifx1, . . . , xkare fixed points inRd\ {0}, the process
Xt, ψ (c)
Lxt1/c−L0t
, . . . , ψ (c)
Lxtk/c−L0t
t0
converges asc→ ∞in the sense of weak convergence in the spaceC(R+, M(Rd)×Rk)to a limiting process which can be written in the form
Xt, βx1
L0t, . . . , βxk
L0t
t0.
Here,(βtx)t0,x∈Rd is a centered Gaussian process independent ofXsuch that cov
βtx, βsy
=a(x, y)(t∧s), anda(x, y)is given by
a(x, y)= 1
4π2
R3 dz 1
|z−x|−|1z|
× 1
|z−y|−|1z|
ifd=3,
1
πx·y ifd=2.
For convenience we will writea(x, x)=:a(x).
Part of the motivation for Theorem 1 came from a recent paper of Lee [5] dealing with asymptotics for the occu- pation measure of super-Brownian motion.
1.5. Lee’s result and extensions
Lee [5] only considers the three-dimensional case, but we will extend his result to the case of dimension 2.
Letd=2 or 3,Xa super-Brownian motion inRd, with initial valueμ=δcx0, wherex0∈Rd,x0=0 andc >0.
LetKbe a compact set inRd, and letφ,ξbe integrable functions onRdsupported onKsatisfying
Kξ(y)dy=0,
Kφ(y)dy=0. We also defineφc,ξcthe functions such that for anyx∈Rd, φc(x)=φ(cx), ξc(x)=ξ(cx).
Note thatφc,ξcare supported onK(c):=c−1K= {y: cy∈K}.
One may think of the particular example whereK= B(0,1), and where we consider the functions φ0=1B(0,1) , ξ0=1B(0,1) ∩{x:xd0}−1B(0,1) ∩{x:xd<0}.
ForT >0, let us introduce the quantities Dφ,T =
T 0
Xt, φdt, Dξ,T = T 0
Xt, ξdt, T 0.
We also set aξ:=
K
K
ξ(y)ξ(z)a(y, z)dydz.
The following result, is proved by Lee in [5] (for the cased=3), by analysing the behaviour ofv(t, x0, ac3φc+ bc7/2ξc)asc→ ∞(cf. (2)).
Proposition 1.Letd=2or3,t >0,x0∈Rd\ {0}and letKbe a compact subset ofRd. Letφandξ be integrable functions on Rd supported on K satisfying
Kφ(y)dy =0,
Kξ(y)dy =0. Under Pδcx
0(·|X hitsK)the random vector
cd−4Dφ,c2t, cd−4ψ (c)Dξ,c2t
converges in distribution asc→ ∞to a non-degenerate limit(D1, D2). Furthermore, if|a|and|b|are small enough the Fourier transform of the limit is
E
exp(iaD1+ibD2)
=1+2|x0|2 4−dGT ,0x
0
ia
K
φ(y)dy−b2aξ 2
,
where the functionGT ,0x0 is determined by(6).
The cased =3 stands as Corollary 1.2 in [5]. The form of the Fourier transform of (D1, D2)implies that the conditional law ofD2knowingD1is Gaussian with mean 0 and varianceaξD1(
φ(y)dy)−1.
In the exampleφ=φ0,ξ =ξ0, Proposition 1 has the following interpretation: conditionally on the event that a super-Brownian motionX started atδcx0 hits the unit ball, the occupation time of the unit ball up to timec2t (that isDφ0,c2t)is of orderc4−d, while the difference between the occupation time of the top half of the unit ball and the occupation time of the bottom half (that isDξ0,c2t) is of order√
cwhend=3, andc(ln(c))1/2whend=2.
In Section 3.2, we will derive Proposition 1 from the following proposition, which is itself a consequence of Theorem 1, as we will see in Section 3.1.
Proposition 2.Letd=2or3. Lett,K,φ,ξ be as in Proposition1, and letν be as in Theorem1. UnderPν, the random vector
cdDφc,t, cdψ (c)Dξc,t
converges in distribution asc→ ∞to
L0t
K
φ(y)dy, Ut
,
where conditionally onL0t,Ut is Gaussian with varianceaξL0t.
As we will explain below, the first componentcdDφc,t indeed converges almost surely. Also note that, from (6), the Fourier transform of(
φ(y)dyL0t, Ut)is simply Eν
exp
iaL0t
K
φ(y)dy+ibUt
=exp
GT ,0ν
ia
K
φ(y)dy−b2aξ 2
.
Let us explain how Proposition 2 follows from Theorem 1. The following approach is also valid in the one- dimensional case, and so we consider the general cased3. Let us first give a simple argument showing thatcdDφc,t convergesPν-almost surely asc→ ∞. By (4), and our assumption on the support ofφ,
cdDφc,t=cd
Rd
φ(cx)Lxt dx=
K
φ(y)Ly/ct dy.
Since(t, x)→Lxt is continuous onR+×B(0, ρ), cdDφc,t −→
c→∞L0t
K
φ(y)dy
almost surely underPν by dominated convergence. See Theorem 5 in [9] for a related statement ([9] only states the convergence in distribution ofcdDφc,t).
Ifφis replaced byξ, the preceding argument is not sufficient. Indeed, since
ξ(x)dx=0, the last step yields cdDξc,t→0 asc→ ∞. Still, we can write
Rd
ξ(y)Ly/ct dy=
Rd
ξ(y)
Ly/ct −L0t dy,
which suggests to focus on the convergence ofψ (c)(Ly/ct −L0t), whereψ (c)is a suitable normalization. This approach is similar to the work of Stroock–Varadhan–Papanicolaou [10] and Yor [11] concerning limit theorems for additive functionals of standard Brownian motion.
The one-dimensional case follows almost immediately from [9] and will be treated briefly in Section 4.Until then we setd=2or3.
1.6. Outline of the proof of Theorem 1
We will first consider the casek=1.We fixx∈Rd\ {0}and a measureνsatisfying the assumption of Theorem 1.
Forα0(α >0 ifd=2)we setgα(z)=∞
0 exp(−αt )pt(z)dt, and fory∈Rd,gαy(z)=gα(z−y). Ifd=3, note that we haveg0(z)=1/(2π|z|).
The key idea is to use the Tanaka formula for local times of super-Brownian motion in dimension less than 3 (see [2], Theorem 6.1). Let y ∈Rd, d 3, μ∈MF(Rd), and let α0 ifd =3 or α >0 ifd =2. Under the assumptionμ, gαy<∞, we havePμ-almost surely,
Xt, gαy
= μ, gyα
+Mt
gαy +α
t 0
Xs, gαy
ds−Lyt, (8)
where Mt(gαy)is an FtX-martingale which is defined in terms of the martingale measure associated with super- Brownian motion. In particular,M0(gαy)=0 andMt(gαy)has quadratic variation
M gyα
t= t 0
Xs, gyα2
ds.
Ifc is large enough so thatx/c∈B(0, ρ), the conditionsν, gα0<∞,ν, gx/cα <∞, will clearly be satisfied.
Thus we can use (8) fory=x/candy=0 to obtain,Pν-almost surely, Lx/ct −L0t =
X0−Xt, gx/cα −gα0 +Mt
gαx/c
−Mt g0α
+α t 0
Xs, gx/cα −gα0
ds. (9)
In what follows we will takeα=0 whend=3. Note that the last term in the right-hand side of the previous formula then vanishes.
In Section 2.1 we will prove the following lemmas:
Lemma 1.Letd=3, T >0. ThenPν-almost surely sup
tT
√ c
X0−Xt, gx/c0 −g00−→
c→∞0.
Lemma 2.Letd=2, T >0. ThenPν-almost surely (a) sup
tT
X0−Xt, c
√ln(c)
gαx/c−g0α−→
c→∞0, (b) sup
tT
t 0
Xs, c
√ln(c)
gαx/c−g0α ds
−→
c→∞0.
From (9), Lemmas 1 and 2 we see that the convergence of (ψ (c)(Lx/ct −L0t))t0 follows from that of (ψ (c)Mt(gx/cα −gα0))t0.
All that remains to do is thus to study the convergence of the martingales Mtx,c:=
⎧⎨
⎩
√cMt((g0x/c−g00)) ifd=3,
√c
ln(c)Mt((gαx/c−g0α)) ifd=2.
Lemma 3.Pν-almost surely, Mx,c
t −→
c→∞a(x)L0t.
This lemma will be proven in Section 2.2.
We then have to discuss the convergence in distribution of the martingaleMtx,c. Using the Dubins–Schwarz theorem (see [8], Theorem V.1.6), we can write
Mtx,c=βx,cMx,ct, (10)
whereβtx,c is a standard Brownian motion. We may and will assume that for suMx,c∞, βsx,c −βux,c = γsx−γux, where γx is a one-dimensional Brownian motion independent ofX. The collection of the laws of the family(X, βx,c)c>0is clearly tight.
In Section 2.3 we will prove
Lemma 4.Suppose that along a subsequencecn ∞we have X, βx,cn (d)
n−→→∞
X, βx . Thenβxis independent ofX.
From the tightness of the laws of(X, βx,c)and Lemma 4, it follows that X, βx,c (d)
n−→→∞
X, βx
(11) with a Brownian motionβxindependent ofX. We know from Lemma 3 thatMx,ct convergesPν-almost surely to a(x)L0t, and the convergence is uniform on every compact time interval by Dini’s theorem. It then follows from (10) and (11) that(Xt, Mtx,c)t0converges in distribution to(Xt, βx
a(x)L0t)t0.
The casek=1 of Theorem 1 now follows from this fact and Lemmas 1 and 2. We will then extend this argument to the general case in Section 2.4.
2. Proof of Theorem 1 2.1. Preliminary reduction
In this section we will prove Lemmas 1 and 2. Recall that the pointx=0 is fixed, and that we have also fixed a measureνsatisfying the assumption of Theorem 1: there existsρ >0 such thatν(B(0, ρ))=0.
We start with a preliminary result providing a uniform bound for the measure of small balls. From [2], Corollary 4.8, we know the following
Proposition 3. Letδ >0be fixed. Ifd=2or3then, for anyμ∈MF(Rd),Pμ-almost surely, lim sup
r0
sup
tδ
sup
y∈Rd
Xt
B(y, r)
Φ(r)−1κμ,
whereΦ(r)=r2(ln(1/r))4−d andκμis a constant depending onμ.
An easy consequence is the following Corollary 1.Pν-almost surely,
lim sup
r0
sup
t0
sup
y∈B0,ρ/2)
Xt(B(y, r) r−2
ln
1 r
d−4
κν.
Proof of Corollary 1. Sinceν(B(0, ρ))=0, it is well known thatPν-almost surely there existsn0(ω)∈Nsuch that supp(Xt)does not intersectB(0,3ρ4)for anyt∈ [0,2−n0]. Providedrρ/4, we then have
sup
t2−n0(ω)
sup
y∈B(0,ρ/2)
Xt
B(y, r) r−2
ln
1 r
d−4
=0,
and thanks to Proposition 3, forPν-almost allω, lim sup
r0
sup
t2−n0(ω)
sup
y∈Rd
Xt B(y, r)
r−2
ln 1
r d−4
κν. Corollary 1 follows. 2
In the following we writeκ0=κ0(x),κ1=κ1(x), . . .for constants that depend on the pointxwhich is fixed.
2.1.1. The cased=3
For convenience we will use the notation hx,c(z):=ψ (c)
g0x/c(z)−g00(z)
=
√c 2π
1
|z−x/c|− 1
|z|
, forz=0, x/c.
We will use the following easy estimates onhx,c: (A1) If|z||2cx|, then|hx,c(z)|√|zc|.
(A2) Letr2|cx|. Then, the maximum of|hx,c|outside the ballB(0, r)is attained at the point rx|x| and its value is (2π√
c r(|xr|−1c))−1.
(A3) If|z| ∧ |z−x/c||2cx| and|z|4|cx|, then|hx,c(z)|2cπ|√xc|. Proof of Lemma 1. Ifcis sufficiently large andz /∈B(0, ρ), using(A2),
hx,c(z) 1 2π√
cρ(ρ/|x| −1/c) −→
c→∞0.
ThusX0, hx,c = ν, hx,cclearly goes to 0 asc→ ∞.
Hence, to get Lemma 1, it is enough to prove thatPν-almost surely, sup
tT
Xt, hx,c=sup
tT
R3
hx,c(z)Xt(dz) −→
c→∞0. (12)
Ift >0 is fixed we know that 0 does not belong to suppXt,Pν-almost surely. Thus the same argument as whent=0 givesXt, hx,c →0 asc→ ∞. The point is that we need this convergence uniformly intT, and we know that there may exist exceptional timestT such that 0∈suppXt.
It will be convenient to cut the domainR3 of the integral in (12) into different areas where we will be able to estimate the integrand. First of all, ifr >0 is fixed, ifz /∈B(0, r)andcis large enough, using again(A2)we have
|hx,c(z)|κ0/√ cso that
R3\B(0,r)
hx,c(z)Xt(dz) κ0
√cXt,1.
Since supt0Xt,1isPν-almost surely bounded, we get sup
tT
R3\B(0,r)
hx,c(z)Xt(dz) −→
c→∞0.
Thus, to get (12) it only remains to prove that for somer1>0, sup
tT
B(0,r1)
hx,c(z)Xt(dz) −→
c→∞0. (13)
From Corollary 1 it follows thatPν-almost surely, there existsr1>0 such that, for anyrr1, sup
t0
sup
y∈B(0,ρ/2)
Xt B(y, r)
2κνr2ln 1
r
. (14)
Clearly we can assumer1ρ/4. Letcbe large enough so thatB(0,4|x|/c)⊂B(0, r1). To get rid of the singularities ofhx,c, we will first deal with the integrals in neighbourhoods of 0 andx/c. First, we have
B(0,|2cx|)
hx,c(z)Xt(dz)
p1
B(0,2p c|x|)\B(0, |x|
2p+1c)
hx,c(z)Xt(dz) .
Using(A1)and (14) we see thatPν-almost surely sup
tT
B(0,2p c|x|)\B(0, |x|
2p+1c)
hx,c(z)Xt(dz) 2κν
2p+1c√ c
|x|
|x| 2pc
2
ln 2pc
|x|
.
It is now clear thatPν-almost surely sup
tT
B(0,|2cx|)
hx,c(z)Xt(dz)
4κν|x| c1/2
p1
pln(2)+ln(c)−ln(|x|)
2−pκ1κνc−1/2ln(c).
Clearly we can obtain an analogous bound for the quantity sup
tT
B(xc,|2cx|)
hx,c(z)Xt(dz) .
Now using(A3)and (14) we get sup
tT
B(0,4|cx|)\(B(0,|2cx|)∪B(xc,|2cx|))
hx,c(z)Xt(dz) 2κν
2c√ c π|x|
4|x| c
2
ln c
4|x|
,
so we finally obtain that sup
tT
B(0,4|x|c )
hx,c(z)Xt(dz)
κνκ2ln(c)
√c . (15)
Let us now consider the integral onB(0, r1)\B(0,4|cx|). LetNbe such that r1
2N 4|x| c r1
2N−1.
Note that, for 1pN,r1/2p2N−p+1|x|/cwhiler1/2p−12N−p+3|x|/c. Once again, using(A2)and (14), we obtain for 1pN
sup
tT
B(0, r1
2p−1)\B(0,r1
2p)
hx,c(z)Xt(dz)
2κνc3/2
|x|(2N−p+1−1)2
2N−p+3|x| c
2
ln c
2N−p+3|x|
2κν
26|x| c1/2 ln
c
|x|
. Therefore,
sup
tT
B(0,r1)\B(0,4|cx|)
hx,c(z)Xt(dz)
N×27κν|x| c1/2 ln
c
|x|
κνκ3(ln(c))2 c1/2 .
The limit (13) now follows from the above and (15), which completes the proof of Lemma 1. 2
2.1.2. The case d=2
Now we considerα >0 and the function hx,c(z):= c
√ln(c)
gαx/c−gα0
(z), forz=0, x/c.
We will need the following estimates onhx,cwhich shall be proven in Appendix A. Whencis large enough, (B1) Ifz∈B(0,|2cx|),
hx,c(z)κ4 c
√ln(c)ln+ 1
|z|
. Ifz∈B(xc,|2cx|),
hx,c(z)κ4 c
√ln(c)ln+ 1
|z−x/c|
. (B2) Ifz∈B(0, c−3/4)\(B(0,|2cx|)∪B(xc,|2cx|)),
hx,c(z)κ5c ln(c).
(B3) Letr > c−7/8. Then ifz /∈B(0, r), hx,c(z)κ6
1 r√
ln(c).
Proof of Lemma 2. We will use a similar method as for proving Lemma 1.
FixT >0. To obtain Lemma 2(a), it is enough to establish thatPν-almost surely, sup
tT
Xt, hx,c−→
c→∞0. (16)
Ifr >0 is fixed andcis sufficiently large we can use(B3)to get sup
tT
R3\B(0,r)
hx,c(z)Xt(dz)
κ6 r√
ln(c)sup
tT
Xt,1.
Since supt0|Xt,1|is finitePν-almost surely, we have sup
tT
R2\B(0,r)
hx,c(z)Xt(dz) −→
c→∞0. (17)
Now, from Corollary 1 we know thatPν-almost surely, there existsr2>0 such that for anyrr2, sup
t0
sup
y∈B(0,ρ/2)
Xt
B(y, r)
2κνr2
ln 1
r 2
. (18)
We can chooser2ρ/4 andclarge enough so thatB(0, c−3/4)⊂B(0, r2). We will first deal with the neighbourhood of singularities 0 andx/c. Using (18) and the estimate(B1)we get for everyp1,
sup
tT
B(0,2p c|x|)\B(0, |x|
2p+1c)
hx,c(z)Xt(dz) κ4
2κνc
√ln(c)
ln 2p+1c
|x|
3|x| 2pc
2
,
and thus sup
tT
B(0,|2cx|)
hx,c(z)Xt(dz)
p1
2κ4κνc
√ln(c)
ln 2p+1c
|x|
3|x| 2pc
2
κ7(ln(c))5/2
c −→
c→∞0.