Penalization method for a nonlinear Neumann PDE via weak solutions of reflected SDEs

20  Download (0)

Full text

(1)

El e c t ro nic J

o f

Pr

ob a bi l i t y

Electron. J. Probab.18(2013), no. 102, 1–19.

ISSN:1083-6489 DOI:10.1214/EJP.v18-2467

Penalization method for a nonlinear Neumann PDE via weak solutions of reflected SDEs

Khaled Bahlali

Lucian Maticiuc

Adrian Z˘ alinescu

§

Abstract

In this paper we prove an approximation result for the viscosity solution of a system of semi-linear partial differential equations with continuous coefficients and nonlinear Neumann boundary condition. The approximation we use is based on a penalization method and our approach is probabilistic. We prove the weak uniqueness of the solution for the reflected stochastic differential equation and we approximate it (in law) by a sequence of solutions of stochastic differential equations with penalized terms. Using then a suitable generalized backward stochastic differential equation and the uniqueness of the reflected stochastic differential equation, we prove the existence of a continuous function, given by a probabilistic representation, which is a viscosity solution of the considered partial differential equation. In addition, this solution is approximated by solutions of penalized partial differential equations.

Keywords: Reflecting stochastic differential equation; Penalization method; Weak solution;

Jakubowski S-topology; Backward stochastic differential equations.

AMS MSC 2010:60H99; 60H30; 35K61.

Submitted to EJP on November 28, 2012, final version accepted on October 19, 2013.

SupersedesarXiv:1308.2173.

1 Introduction

LetGbe aC2 convex, open and bounded set fromRd, and for(t, x)∈[0, T]×G¯ we consider the following reflecting stochastic differential equation (SDE for short)

Xs+Ks=x+ Z s

t

b(Xr)dr+ Z s

t

σ(Xr)dWr, s∈[t, T], withKa bounded variation process such that for anys∈[t, T],Ks=

Z s t

∇`(Xr)d|K|[t,r]

and|K|[t,s] = Z s

t 1{Xr∈∂G}d|K|[t,r] , where the notation|K|[t,s]stands for the total vari- ation ofKon the interval[t, s].

Support: PHC Volubilis MA/10/224 & Tassili 13MDU887; IDEAS 241/05102011; POSDRU/89/1.5/S/49944.

Université de Toulon (IMATH) and Aix-Marseille Université (CNRS, LATP), France.

E-mail:bahlali@univ-tln.fr.

“Alexandru Ioan Cuza” University of Iasi and “Gheorghe Asachi” Technical University, Iasi, Romania.

E-mail:lucian.maticiuc@uaic.ro

§“Alexandru Ioan Cuza” University of Iasi and “Octav Mayer”Mathematics Institute of the Romanian Academy, Romania. E-mail:adrian.zalinescu@gmail.com

(2)

The coefficients b and σ are supposed to be only bounded continuous on Rd and σσuniformly elliptic. The first main purpose is to prove that the weak solution(X, K) is approximated in law (in the space of continuous functions) by the solutions of the non-reflecting SDE

Xsn =x+ Z s

t

[b(Xrn)−n(Xrn−πG¯(Xrn))]dr+ Z s

t

σ(Xrn)dWr, s∈[t, T], where πG¯ is the projection operator. Since for n → ∞ the term Ksn := nRs

t(Xrn− πG¯(Xrn))dr forces the solutionXn to remain near the domain, the above equation is called SDE with penalization term.

The case where b and σare Lipschitz has been considered by Lions, Menaldi and Sznitman in [11] and by Menaldi in [14] where they have proven thatE(sups∈[0, T]|Xsn− Xs|) −→ 0, as n → ∞. Note that Lions and Sznitman have shown, using Skorohod problem, the existence of a weak solution for the SDE with normal reflection to a (non- necessarily convex) domain. The case of reflecting SDE with jumps has been treated by Łaukajtys and Słomi´nski in [8] in the Lipschitz case; the same authors have extended in [9] these results to the case where the coefficient of the reflecting equation is only continuous. In these two papers it is proven that the approximating sequence (Xn)n is tight with respect to the S-topology, introduced by Jakubowski in [6] on the space D R+,Rd

of càdlàgRd-valued functions. Assuming the weak (in law) uniqueness of the limiting reflected diffusionX, they prove in [9] thatXn S-converges weakly toX. We mention that (Xn)n may not be relatively compact with respect to the Skorohod topologyJ1.

In contrast to [9], we can not simply assume the uniqueness in law of the limitX, and the weak S-convergence ofXntoXis not sufficient to our goal. In our framework, we need to show the uniqueness in law of the couple(X, K)and that the convergence in law of the sequence(Xn, Kn)to(X, K)holds with respect to uniform topology.

The first main result of our paper will be the weak uniqueness of the solution(X, K), together with the convergence in law (in the space of continuous functions) of the penal- ized diffusion to the reflected diffusionX and the continuity with respect to the initial data.

Subsequently, using a proper generalized BSDE, we deduce (as a second main result) an approximation result for a continuous viscosity solution of the system of semi-linear partial differential equations (PDEs for short) with a nonlinear Neumann boundary con- dition

∂ui

∂t (t, x) +Lui(t, x) +fi(t, x, u(t, x)) = 0, ∀(t, x)∈[0, T]×G,

∂ui

∂n (t, x) +hi(t, x, u(t, x)) = 0, ∀(t, x)∈[0, T]×∂G, ui(T, x) =gi(x), ∀x∈G, i= 1, k,

whereLis the infinitesimal generator of the diffusionX, defined by L= 1

2 X

i,j

(σ(·)σ(·))ij(·) ∂2

∂xi∂xj

+X

i

bi(·) ∂

∂xi

,

and∂ui/∂nis the outward normal derivative ofuion the boundary of the domain.

Boufoussi and Van Casteren have established in [3] a similar result, but in the case where the coefficientsbandσare uniformly Lipschitz.

We mention that the class of BSDEs involving a Stieltjes integral with respect to the continuous increasing process|K|[t,s] was studied first in [16] by Pardoux and Zhang;

(3)

the authors provided a probabilistic representation for the viscosity solution of a Neu- mann boundary partial differential equation. It should be mentioned that the continuity of the viscosity solution is rather hard to prove in our frame. In fact, this property essentially uses the continuity with respect to initial data of the solution of our BSDE.

We develop here a more natural method based on the uniqueness in law of the solution (X, K, Y)of the reflected SDE-BSDE and on the continuity property. Similar techniques were developed, in the non reflected case, in [2], but in our situation the proof is more delicate. The difficulty is due to the presence of the reflection processKin the forward component and the generalized part in the backward component.

Throughout this paper we use different types of convergence defined as follows: for the processes(Yn)n andY, byYn −−−→

u Y we denote the convergence in law with re- spect to the uniform topology, by Yn−−−→

J1

Yn we mean the convergence in law with respect to the Skorohod topologyJ1 and byYn −−−→

S Y we understand the weak con- vergence considered in S-topology.

The paper is organized as follows: in the next section we give the assumptions, we formulate the problem and we state the two main results. The third section is devoted to the proof of the first main result (proof of the convergence in law of(Xn, Kn)to(X, K) asn→ ∞and the continuous dependence with respect to the initial data). In Section 4 the generalized BSDEs are introduced, the continuity with respect to the initial data is obtained and we prove the approximation result for the PDE introduced above.

2 Formulation of the problem; the main results

Let G be aC2 convex, open and bounded set fromRd and we suppose that there exists a function`∈Cb2(Rd)such that

G={x∈Rd :`(x)<0}, ∂G={x∈Rd:`(x) = 0}, and, for allx∈∂G,∇`(x)is the unit outward normal to∂G.

In order to define the approximation procedure we shall introduce the penalization term. Letp:Rd→R+ be given byp(x) = dist2(x,G)¯ .

Without restriction of generality we can choose`such that h∇`(x), δ(x)i ≥0, ∀x∈Rd, whereδ(x) :=∇p(x)is called the penalization term.

It can be shown thatpis of classC1onRdwith 1

2δ(x) = 1

2∇(dist2(x,G)) =¯ x−πG¯(x), ∀x∈Rd,

whereπG¯(x)is the projection ofxonG.¯ It is clear thatδis a Lipschitz function.

On the other hand,x7→dist2 x,G¯

is a convex function and therefore

hz−x, δ(x)i ≤0, ∀x∈Rd, ∀z∈G.¯ (2.1) LetT >0and suppose that:

(A1) b:Rd→Rdand σ:Rd→Rd×d0 are bounded continuous functions.

Remark 2.1. In fact we can assume that the functionsbandσhave sublinear growth but, for the simplicity of the calculus, we will work with assumption (A1).

(4)

(A2) the matrixσσis uniformly elliptic, i.e. there existsα0>0such that for allx∈Rd, (σσ) (x)≥α0I.

Moreover, there exist some positive constantsCi,i= 1,2,α∈R,β ∈R+andq≥1 such that

(A3) f, h: [0, T]×Rd×Rk →Rk and g:Rd→Rk are continuous functions and, for all x, x0∈Rd,y, y0∈Rk,t, t0∈[0, T],

(i) hy0−y, f(t, x, y0)−f(t, x, y)i ≤α|y0−y|2,

(ii) |h(t0, x0, y0)−h(t, x, y)| ≤β(|t0−t|+|x0−x|+|y0−y|), (iii) |f(t, x, y)|+|h(t, x, y)| ≤C1(1 +|y|),

(iv) |g(x)| ≤C2(1 +|x|q).

(2.2)

Let us consider the following system of semi-linear PDEs considered on the whole space:









∂uni

∂t (t, x) +Luni (t, x) +fi(t, x, un(t, x))− h∇uni(t, x), nδ(x)i

−h∇`(x), nδ(x)i hi(t, x, un(t, x)) = 0 uni (T, x) =gi(x), ∀(t, x)∈[0, T]×Rd, i= 1, k ,

(2.3)

and the next semi-linear PDE considered with Neumann boundary conditions:













∂ui

∂t (t, x) +Lui(t, x) +fi(t, x, u(t, x)) = 0, ∀(t, x)∈[0, T]×G,

∂ui

∂n (t, x) +hi(t, x, u(t, x)) = 0, ∀(t, x)∈[0, T]×∂G, ui(T, x) =gi(x), ∀x∈G, i= 1, k,

(2.4)

whereLis the second order partial differential operator L= 1

2 X

i,j

(σ(·)σ(·))ij(·) ∂2

∂xi∂xj

+X

i

bi(·) ∂

∂xi

,

and, for anyx∈∂G

∂ui

∂n (t, x) =h∇`(x),∇ui(t, x)i is the exterior normal derivative inx∈∂G.

Our goal is to establish a connection between the viscosity solutions for (2.3) and (2.4) respectively. The proof will be given using a probabilistic approach. Therefore we start by studying an SDE with reflecting boundary condition and then we associate a corresponding backward SDE. Since the coefficients of the forward equation are merely continuous, our setting is that of weak formulation of solutions.

For(t, x)∈[0, T]×G¯ we consider the following stochastic differential equation with reflecting boundary condition:

(i) Xst,x+Kt,xs =x+ Z s

t

b(Xrt,x)dr+ Z s

t

σ(Xrt,x)dWr,

(ii) Kst,x= Z s

t

∇`(Xrt,x)d|Kt,x|[t,r] , (iii) |Kt,x|[t,s]=

Z s

t 1{Xrt,x∈∂G}d|Kt,x|[t,r] , ∀s∈[t, T],

(2.5)

(5)

where|Kt,x|[t,s] is the the total variation ofKt,xon the interval[s, t]1.

We denote bykt,xs the continuous increasing process defined bykt,xs :=|Kt,x|[t,s]. It follows that

kst,x= Z s

t

∇`(Xrt,x), dKrt,x

. (2.6)

Using the penalization termδ we can define the approximation procedure for the re- flected diffusionX.

Under assumption(A1)we know that (see, e.g., [7, Theorem 5.4.22]), for eachn∈ N, there exists a weak solution of the following penalized SDE

Xst,x,n=x+ Z s

t

b(Xrt,x,n)−nδ(Xrt,x,n) dr+

Z s t

σ(Xrt,x,n)dWr, ∀s∈[t, T]. (2.7) Let

Kst,x,n:=

Z s t

nδ(Xrt,x,n)dr ,

kst,x,n:=

Z s t

∇`(Xrt,x,n), dKrt,x,n

,∀s∈[t, T].

(2.8)

We mention that (see, e.g., [7]) the solution process(Xst,x,n)s∈[t,T]is unique in law under the supplementary assumption(A2).

Here and subsequently, we shall denote byV andVn: Vst,x:=x+

Z s t

b(Xrt,x)dr+ Z s

t

σ(Xrt,x)dWr,

Vst,x,n:=x+ Z s

t

b(Xrt,x,n)dr+ Z s

t

σ(Xrt,x,n)dWr, ∀s∈[t, T].

(2.9)

Hence (2.5) and (2.7) become respectively

Xst,x+Kst,x=Vst,x and Xst,x,n+Kst,x,n=Vst,x,n, ∀s∈[t, T]. Definition 2.2. We say that Ω,F,P,{Fs}s≥t, W, X, K

is a weak solution of (2.5) if Ω,F,P,{Fs}s≥t

is a stochastic basis, W is a d0-dimensional Brownian motion with respect to this basis,X is a continuous adapted process andKis a continuous bounded variation process such thatXs∈G¯,∀s∈[t, T], and system (2.5) is satisfied.

The main results are the following two theorems. The first one consists in establish- ing the weak uniqueness (in law) of the solution for (2.5) and the continuous dependence in law with respect to the initial data.

Theorem 2.3. Under the assumptions (A1−A2), there exists a unique weak solution (Xst,x, Kst,x)s∈[t,T] of SDE (2.5). Moreover,

(Xt,x,n, Kt,x,n)−−−−→

u (Xt,x, Kt,x) and the application

[0, T]×G¯ 3(t, x)7→(Xt,x, Kt,x) is continuous in law.

1For 0 s < t T, the total variation ofY on [s, t]is given by|Y|[s,t](ω) = sup

nn−1

P

i=0

|Yti+1(ω) Yti(ω)|o

,where∆ :s=t0< t1<· · ·< tn=tis a partition of the interval[s, t].

(6)

Once this result for the forward part is established we then associate a BSDE involv- ing Stieltjes integral with respect to the increasing processkt,x in order to obtain the probabilistic representation for the viscosity solution of PDE (2.3).

The next result provides the approximation of a viscosity solution for system (2.4) by the solutions sequence of (2.3).

Theorem 2.4. Under the assumptions(A1−A3), there exist continuous functionsun : [0, T]×Rd→Rd andu: [0, T]×G¯ →Rd such thatunis a viscosity solution for system (2.3),uis a viscosity solution for system (2.4) with Neumann boundary conditions and, in addition,

n→∞lim un(t, x) =u(t, x), ∀(t, x)∈[0, T]×G.¯

3 Proof of Theorem 2.3

We shall divide the proof of this Theorem into several lemmas. First of all we recall that the existence of a weak solution is given, under assumption(A1), by [12, Theorem 3.2].

For the simplicity of presentation we suppress from now on the explicit dependence on(t, x)in the notation of the solution of (2.5) and (2.7).

We first prove an estimation result for the solutions of the penalized SDE (2.7).

Lemma 3.1. Under assumption (A1), for any q ≥ 1, there exists a constant C > 0, depending only ond, T andq, such that

E

sups∈[t,T]|Xsn|2q +E

sups∈[t,T]|Ksn|2q

+E|Kn|q[t,T]≤C, ∀n∈N. (3.1) Proof. Without loss of generality we can assume that0∈G. From Itô’s formula applied for|Xsn|2it can be deduced that

|Xsn|2+ 2 Z s

t

hXrn, dKrni=|x|2+ 2 Z s

t

hXrn, b(Xrn)idr+ 2 Z s

t

hXrn, σ(Xrn)dWri +

Z s t

|σ(Xrn)|2dr, s∈[t, T]. Writeτm:= inf{s∈[t, T] :|Xsn| ≥m} ∧T,m∈N, and by the above,

|Xs∧τn m|2+ 2 Z s∧τm

t

hXrn, dKrni ≤C+|x|2+C Z s∧τm

t

|Xrn|dr+ 2 Z s∧τm

t

hXrn, σ(Xrn)dWri, s∈[t, T]. Here and in what followsC >0will denote a generic constant which is allowed to vary from line to line.

Therefore

|Xs∧τn m|2+ Z s∧τm

t

hXrn, dKrni q

≤C

1 +|x|2q +C

Z s∧τm

t

|Xrn|2dr q

+C

Z s∧τm

t

hXrn, σ(Xrn)dWri

q

, and

Esupr∈[t,s]

|Xr∧τn m|2+ Z r∧τm

t

hXun, dKuni q

≤C

1 +|x|2q +CE

Z s∧τm

t

supu∈[t,r]|Xun|2qdr+CEsupr∈[t,s]

Z r∧τm

t

hXun, σ(Xun)dWui

q

.

(3.2)

(7)

By Burkholder-Davis-Gundy inequality we deduce

Esupr∈[t,s]

Z r∧τm

t

hXun, σ(Xun)dWui

q

≤CE

Z s∧τm

t

|Xun|2|σ(Xun)|2du

q/2

≤CE

Z s∧τm

t

|Xun|2du

q/2

≤C

1 +E Z s∧τm

t

supu∈[t,r]|Xun|2qdr

, and (3.2) yields

Esupr∈[t,s∧τ

m]|Xrn|2q ≤C

1 +|x|2q+E Z s

t

supu∈[t,r∧τ

m]|Xun|2qdr

, ∀s∈[t, T], since from (2.1) applied forz= 0∈G, we have

Z s t

hXrn, dKrni=n Z s

t

hXrn, δ(Xrn)idr≥0.

From the Gronwall lemma,

Esupr∈[t,s∧τm]|Xrn|2q≤C

1 +|x|2q

, ∀n∈N. Takingm→ ∞it follows that

Esupr∈[t,T]|Xrn|2q ≤C, ∀n∈N. (3.3) Once again from (3.2) and (3.3) we obtain

E Z T

t

hXrn, dKrni

!q

≤C

1 +|x|2q .

We have that there existsε >0such that the ballB¯(0, ε)⊂G, and, forz=ε|KKnsn−Ktn s−Ktn| ∈ G, inequality (2.1) becomes

ε|Kvn−Kun| ≤ Z v

u

hXrn, dKrni, ∀t≤u≤v≤T, and by the definition of total variation ofKn,it follows that

εqE

|Kn|q[t,T]

≤EZ T t

hXrn, dKrniq

≤C .

Lemma 3.2. Under assumption(A1)the sequence(Xsn, Ksn, kns)s∈[t,T] is tight with re- spect to theS-topology.

Proof. In order to obtain theS-tightness of a sequence of integrable càdlàg processes Un, n ≥ 1, we shall use the sufficient condition given e.g. in [10, Appendix A] which consists in proving the uniform boundedness for:

CVT(Un) +E sup

s∈[t,T]

|Usn| ,

where

CVT(Un) := sup

π m−1

X

i=0

Eh

E[Utni+1−Utni/Fti] i

(3.4)

(8)

defines the conditional variation ofUn, with the supremum taken over all finite parti- tionsπ:t=t0< t1<· · ·< tm=T.

Using Lemma 3.1, we deduce that there exists a constantC >0such that for every n∈N

CVT(Kn) +E sup

s∈[t,T]

|Ksn|

≤E

|Kn|[t,T] +E

sup

s∈[t,T]

|Ksn|

≤C.

Sincekn is increasing andl ∈Cb2 Rd

, then there exist constantsM, C >0 such that for everyn∈N

CVT(kn) +E sup

s∈[t,T]

|kns|

≤2E(kTn) = 2EZ T t

h∇`(Xrn), dKrni

≤2E sup

s∈[t,T]

|∇`(Xsn)| · |Kn|[t,T]

≤2ME

|Kn|[t,T]

≤2M C.

By the definition of Vn, assumption (A1) and the fact the conditional variation of a martingale is0, we obtain for eachn∈N,

CVT(Vn)≤CVT

Z · t

b(Xrt,x,n)dr

+ CVT

Z · t

σ(Xrt,x,n)dWr

= CVT Z ·

t

b(Xrt,x,n)dr

≤E Z T

t

b(Xrt,x,n) dr

!

≤(T−t)M ≤C

Therefore (see also Lemma 3.1), there existsC >0such that for everyn∈N CVT(Xn) +E

sup

s∈[t,T]

|Xsn|

≤CVT(Vn) + CVT(Kn) +E sup

s∈[t,T]

|Xsn|

≤C.

Lemma 3.3. Under the assumptions(A1−A2), the uniqueness in law of the stochastic process(Xs)s∈[t,T]holds.

Proof. Let Ω,F,P,{Ft}t≥0, W, X, K

be a weak solution of (2.5) andf ∈C1,2 [0, T]×G¯ . We apply Itô’s formula tof(s, Xs):

f(s, Xs) =f(t, x) + Z s

t

∂f

∂r+Lf

(r, Xr)dr− Z s

t

h∇xf(r, Xr),∇`(Xr)idkr +

Z s t

h∇xf(r, Xr), σ(Xr)dWri.

(3.5)

Since σσ is supposed to be invertible, we deduce, using Krylov’s inequality for the reflecting diffusions (see [18, Theorem 5.1]), that for anys∈[t, T],

E Z s

t

∂f

∂r +Lf (r, Xr)

1{Xr∈∂G}dr

≤C Z s

t

Z

G

det (σσ)−1∂f

∂r +Lfd+1

1∂G drdx d+11

= 0.

(9)

Thus, equality (3.5) becomes f(s, Xs) =f(t, x) +

Z s t

∂f

∂r +Lf

(r, Xr)1{Xr∈G}dr− Z s

t

h∇xf(r, Xr),∇`(Xr)idkr +

Z s t

h∇xf(r, Xr), σ(Xr)dWri , P-a.s.

Therefore

f(s, Xs)−f(t, x)− Z s

t

∂f

∂r +Lf

(r, Xr)1{Xr∈G}dr is aP-supermartingale wheneverf ∈C1,2 [0, T]×G¯

satisfies h∇xf(s, x),∇`(x)i ≥0,∀x∈∂G.

From [21, Theorem 5.7] (applied with φ = −`, γ := ∇φ and ρ := 0) we have that the solution to the supermartingale problem is unique for each starting point (t, x), therefore our solution process(Xs)s∈[t,T]is unique in law.

Remark 3.4. Following the remark of El Karoui [5, Theorem 6] we obtain the unique- ness in law of the couple(X, K), since the increasing process kdepends only on the solutionX (and not on the Brownian motion). The uniqueness is essential in order to formulate the issue of the continuity with respect to the initial data.

Lemma 3.5. We suppose that the assumptions(A1−A2)are satisfied. Then (i) (Xn, Kn)−−−−→

u (X, K), (ii) kn−−−→

u k.

Proof. (i)First we will prove the convergence:

(Xn, Kn)−−−→

S (X, K). (3.6)

We shall apply [9, Theorem 4.3(iii)]. We recall that we have the uniqueness of the weak solution. For anyn∈N,s∈[t, T], letHsn:=x∈G¯ and the processesZsn := (s, Ws). Our equation can be written as

Xsn =Hsn+ Z s

t

(b, σ) (Xrt,x,n), dZsn

−Ksn, ∀s∈[t, T].

The processesZn satisfy the(UT)condition (introduced in [20]), since for any discrete predictable processesUn,U¯n of the formUsn=U0n+Pk

i=0Uin, respectivelyU¯sn = ¯U0n+ Pk

i=0inwith|Uin|, U¯in

≤1, E

Z q 0

Usnds+ Z q

0

sndWs

2

≤2E

Z q 0

Usnds

2

+ 2E

Z q 0

sndWs

2

≤2q2+ 2E Z q

0

sn

2ds≤2q(q+ 1).

Therefore the assumptions of [9, Theorem 4.3] are satisfied and thus we obtain that Xn−−−→

S X.

Using once again [9, Theorem 4.3(ii)] and definition (2.9) we deduce that Xtn1, Xtn2, . . . , Xtnm, Vn

−−−→(Xt1, Xt2, . . . , Xtm, V),

(10)

for any partition t =t0 < t1 <· · · < tm =T. The above convergence is considered in law, on the space Rdm

× D([0, T],Rd)endowed with the product between the usual topology on Rdm

and the Skorohod topologyJ1. Hence

(Xn, Vn)−−−→

S (X, V), since(Xn, Vn)nis tight.

It is known that the spaceD([0, T],Rd)of càdlàg functions endowed withS-topology is not a linear topological space, but the sequential continuity of the addition, with respect to theS-topology, is fulfilled (see Jakubowski [6, Remark 3.12]). Therefore

Kn=Vn−Xn−−−→

S V −X =K.

In order to obtain the uniform convergence4of the sequence(Xn, Kn)nwe remark that, sinceVn,V are continuous andVn−−−→

J1 V, this convergence is uniform in distribution:

Vn −−−→

u V.

Using the Skorohod theorem, there exists a new probability space Ω,ˆ Fˆ,Pˆ

on which we can define random variablesV ,ˆ Vˆn such that

Vˆ ====law V, Vˆn====law Vn, ∀n∈N, and

sup

s∈[t,T]

|Vˆsn−Vˆs|−−−→a.s. 0.

LetXˆnbe the solution of the equation Xˆsn+

Z s t

nδ( ˆXrn)dr= ˆVsn, s∈[t, T], X¯n be the solution of

sn+ Z s

t

nδ( ¯Xrn)dr= ˆVs, s∈[t, T], and denote

sn:=

Z s t

nδ( ˆXrn)dr, K¯sn:=

Z s t

nδ( ¯Xrn)dr.

It is easy to prove (see, e.g., [8, Lemma 2.2] or [22, Lemma 2.2]) that

sn−X¯sn

2

sn−Vˆs

2+ 2 Z s

t

( ˆVsn−Vˆs)−( ˆVrn−Vˆr), d( ˆKrn−K¯rn) , therefore

sup

s∈[t,T]

sn−X¯sn

2≤ sup

s∈[t,T]

sn−Vˆs

2+ 4 sup

s∈[t,T]

sn−Vˆs

|Kˆn|[t,T]+|K¯n|[t,T]

. (3.7)

Since( ˆXn,Kˆn)==== (Xlaw n, Kn)and|Kn|[t,T]is bounded in probability by inequality (3.1),

|Kˆn|[t,T] is bounded in probability. Applying [8, Theorem 2.7], it follows that|K¯n|[t,T] is also bounded in probability.

4We are thankful to professor L. Słomi´nski for his useful suggestion in the proof of this part.

(11)

But

sup

s∈[t,T]

sn−Vˆs

−−−→prob 0, therefore, from (3.7),

sup

s∈[t,T]

sn−X¯sn

2 prob

−−−−→0. (3.8)

On the other hand, letXˆ be the solution of the Skorohod problem Xˆs+ ˆKs= ˆVs, s∈[t, T].

It can be shown (see the proof of [12, Theorem 2.1] or the proof of [17, Theorem 4.17]) that

sup

s∈[t,T]

sn−Xˆs

2 prob

−−−→0, therefore, from (3.8),

sup

s∈[t,T]

sn−Xˆs

2 prob

−−−→0.

SinceKˆs= ˆVs−Xˆs,Kˆsn= ˆVsn−Xˆsn, s∈[t, T], ( ˆXn,Kˆn)−−−→prob

u ( ˆX,K),ˆ and

( ˆXn,Kˆn)==== (Xlaw n, Kn), the conclusion follows.

(ii)In order to pass to the limit in the integral Z s

t

h∇`(Xr), dKri, we apply the stochastic version of Helly-Bray theorem given by [23, Proposition 3.4]. For the convenience of the reader we give the statement of that result:

Lemma 3.6. Let (Xn, Kn) : (Ωn,Fn,Pn) −→ C [0, T],Rd

be a sequence of random variables and(X, K)such that

(Xn, Kn)−−−→

u (X, K). If(Kn)nhas bounded variation a.s. and

sup

n∈NP

|Kn|[0,T]> a

−→0, asa−→ ∞, thenKhas a.s. bounded variation and

Z T 0

hXrn, dKrni−−−→

u

Z T 0

hXr, dKri, asn−→ ∞.

Returning to the proof of Lemma 3.5, the conclusion(ii)follows now easily, sincek andknare defined by (2.6) and (2.8) respectively.

Remark 3.7. Let the assumptions(A1−A2)be satisfied. Then the weak solution(Xst,x)s∈[t,T] is a strong Markov process. Indeed, taking into account the equivalence between the existence for the (sub-)martingale problem and the existence of a weak solution for re- flected SDE (2.5) (see [5, Theorem 7]), we obtain that the weak solution(Xst,x)s∈[t,T] is a strong Markov process since the uniqueness holds (see [5, Theorem 10]). In our situation, this equivalence can be obtained by using Krylov’s inequality for reflecting diffusions.

(12)

The following result will finalize the proof of Theorem 2.3.

We extend the solution process to[0, T]by denoting

Xst,x:=x, Kst,x:= 0, ∀s∈[0, t). (3.9) Lemma 3.8. We suppose that the assumptions(A1−A2)are satisfied and let(Xst,x, Kst,x)s∈[t,T]

be the weak solution of (2.5). Then(Xst,x, Kst,x)s∈[t,T] is continuous in law with respect to the initial data(t, x).

Proof. Let(t, x)∈[0, T]×G¯ be fixed and(tn, xn)→(t, x), asn→ ∞. We denote (Xsn, Ksn) := (Xstn,xn, Kstn,xn).

We will prove first that the family(Xn, Kn)is tight as family ofC([0, T],Rd×Rd)-valued random variables.

Applying Itô’s formula for the process Xsn −Xrn, where r is fixed and s ≥ r, we deduce

|Xsn−Xrn|2= 2 Z s

r

hXun−Xrn, b(Xun)idu−2 Z s

r

hXun−Xrn, dKuni+ Z s

r

|σ(Xun)|2du + 2

Z s r

hXun−Xrn, σ(Xun)dWuni

≤2 Z s

r

hXun−Xrn, b(Xun)idu+ Z s

r

|σ(Xun)|2du+ 2 Z s

r

hXun−Xrn, σ(Xun)dWuni, sinceXun, Xrn ∈G¯ and

Z s r

hz−Xun, dKuni= Z s

r

hz−Xun,∇`(Xun)idknu ≤0, ∀0≤r≤s, ∀z∈G.¯ Therefore, using thatb, σare bounded functions andG¯ is a bounded domain,

E

|Xsn−Xrn|8

≤C|s−r|4+CE

supv∈[r,s]

Z v r

hXun−Xrn, σ(Xun)dWuni4

≤C|s−r|4+CEZ s r

|Xun−Xrn|2|σ(Xun)|2du2

≤C|s−r|4+C|s−r|2≤Cmax

|s−r|4,|s−r|2 .

(3.10)

ConcerningK, we remark first that Ksn−Krn=

Z s r

b(Xun)du+ Z s

r

σ(Xun)dWun−(Xsn−Xrn). Hence

E

|Ksn−Krn|8

≤CE

|Xsn−Xrn|8

+CEZ s r

b(Xun)du8

+CE

supv∈[r,s]

Z v r

σ(Xun)dWun8

≤Cmax

|s−r|4,|s−r|2

+C|s−r|8+CEZ s r

|σ(Xun)|2du4

≤Cmax

|s−r|8,|s−r|2 .

(3.11)

(13)

Observe that the constants in the right hand of the inequalities (3.10) and (3.11) do not depend on(t, x). Therefore, applying a tightness criterion (see, e.g. [17, Cap. I]) we deduce that the family(Xt,x, Kt,x)is tight (viewed as a family ofC([0, T],Rd×Rd)-valued random variables) with respect to the initial data(t, x).

Taking into account the above conclusion and the Prokhorov theorem, we have that if(tn, xn)→(t, x), asn→ ∞, then there exists a subsequence, still denoted by(tn, xn), such that

Xn :=Xtn,xn −−−→

u X, Kn :=Ktn,xn −−−→

u K, asn→ ∞.

It remains to identify the limits, i.e.X ====law Xt,xandK====law Kt,x. By the Skorohod theorem, we can choose a probability space Ω,ˆ F,ˆ Pˆ

(which can be taken in fact as [0,1],B[0,1], µ

whereµis the Lebesgue measure), and( ˆXn,Kˆn,Wˆn), ( ˆX,K,ˆ Wˆ)defined on this probability space, such that

( ˆXn,Kˆn,Wˆn)==== (Xlaw n, Kn, Wn), ( ˆX,K,ˆ Wˆ)==== (X, K, Wlaw ) and

( ˆXn,Kˆn,Wˆn)−−→a.s. ( ˆX,K,ˆ Wˆ), asn→ ∞.

It is not difficult to see that Wˆn,FtWˆn,Xˆn

and W ,ˆ FtW ,ˆ Xˆ

are Brownian motions.

We now define,

sn:=x+ Z s

t

b( ˆXrn)dr+ Z s

t

σ( ˆXrn)dWˆrn and Vˆs:=x+

Z s t

b( ˆXr)dr+ Z s

t

σ( ˆXr)dWˆr, s∈[t, T].

Arguing as in the proof of [1, Proposition 12] (see also [19, Chapter III-3] for more details), it can be shown thatRs

t σ( ˆXrn)dWˆrn andRs

t b( ˆXrn)drconverge in probability to Rs

t σ( ˆXr)dWrand respectivelyRs

t b( ˆXr)dr. Sinceσandbare bounded, we deduce using the Lebesgue theorem that this convergence holds inLq( ˆΩ)for eachq≥1. Therefore,

E sup

s∈[t,T]

sn−Vˆs

q

→0, asn→ ∞, IfVn is defined by

Vsn :=x+ Z s

t

b(Xrn)dr+ Z s

t

σ(Xrn)dWrn thenXsn+Ksn=Vsn, P-a.s. And it is not difficult to see that

(Xn, Kn, Wn, Vn)==== ( ˆlaw Xn,Kˆn,Wˆn,Vˆn) on C([0, T],Rd×Rd×Rd0×Rd) and

sn+ ˆKsn= ˆVsn, a.s.

which yields, passing to the limit, that

s+ ˆKs= ˆVs, a.s.

Then the coupled process( ˆXs,Kˆs)is a solution of (2.5) corresponding to the initial data (t, x). Taking into account the uniqueness in law of the solution (Xst,x, Kst,x)s∈[t,T](see Remark 3.4) we deduce that the whole sequence(Xsn, Ksn)s∈[t,T] converges to the pro- cess(Xst,x, Kst,x)s∈[t,T], and therefore the continuity with respect to(t, x)follows.

(14)

4 BSDEs and nonlinear Neumann boundary problem

Let us now consider the processes(Xst,x,n, kst,x,n)t≤s≤T and(Xst,x, kt,xs )t≤s≤T given by relations (2.5) - (2.8), for(t, x)∈[0, T]×G¯.

For the proof of Theorem 2.4 we associate the following generalized backward stochastic differential equations (BSDEs for short) on[t, T]:

Yst,x=g(Xt,xT )+

Z T s

f(r, Xrt,x, Yt,xr )dr− Z T

s

Ut,xr dMrXt,x− Z T

s

h(r, Xrt,x, Yt,xr )dkrt,x, (4.1) and respectively the BSDE corresponding to the solution of (2.7)

Yst,x,n=g(Xt,x,nT )+

Z T s

f(r, Xrt,x,n, Yt,x,nr )dr− Z T

s

Ut,x,nr dMrXt,x,n

− Z T

s

h(r, Xrt,x,n, Yt,x,nr )dkt,x,nr ,

(4.2)

where

MsXt,x :=

Z s t

σ(Xrt,x)dWr, MsXt,x,n:=

Z s t

σ(Xrt,x,n)dWr (4.3) are the martingale part of the reflected diffusion processXt,x andXt,x,n respectively.

We assume for simplicity that the processes(Xst,x,n, Kst,x,n)s∈[t,T] and(Xst,x, Kst,x)s∈[t,T] are considered on the canonical space.

We recall that the coefficientsf, gandhsatisfy assumption (A3). Then, given the pro- cesses(Xst,x,n, kt,x,ns )s∈[t,T] and(Xst,x, kst,x)s∈[t,T], this assumption ensures (see [16]) the existence and the uniqueness for the couples(Yst,x,n, Ust,x,n)s∈[t,T] and(Yst,x, Ust,x)s∈[t,T] respectively. Arguing as in [3], one can establish the following result.

Proposition 4.1. Let the assumptions(A1−A3)be satisfied.

Let (Yst,x,n, Ust,x,n)s∈[t,T] and (Yst,x, Ust,x)s∈[t,T] be the solutions of the BSDEs (4.2) and (4.1), respectively. Then

Yt,x,n, Mt,x,n, Ht,x,n

−−−−−−−→

S×S×S Yt,x, Mt,x, Ht,x , where

Mst,x,n:=

Z s t

Ut,x,nr dMrXt,x,n, Hst,x,n:=

Z s 0

h(r, Xrt,x,n, Yt,x,nr )dkt,x,nr ,

Mst,x:=

Z s t

Ut,xr dMrXt,x, Hst,x:=

Z s 0

h(r, Xrt,x, Yt,xr )dkt,xr (4.4) andMXt,x,nandMXt,xare defined by (4.3).

Moreover, we have that lim

n→∞Ytt,x,n=Ytt,x.

Remark 4.2. The solution process(Yst,x)s∈[t,T] is unique in law. Indeed, following [4, Theorem 3.4], it can be proven that, since the coefficientsband σsatisfy the assump- tions(A1−A2)and the solution process has the Markov property, there exists a deter- ministic measurable functionusuch that the solutionYst,x=u(s, Xst,x),s∈[t, T]dP⊗ds a.s. The conclusion follows by Lemma 3.3 and the uniqueness (as a strong solution) of Y.

In the following, we extendXt,x, Kt,xto[0, T]as in (3.9) and(Yt,x, Ut,x)by denoting Yst,x:=Ytt,x, Ust,x:= 0 and MsXt,x:= 0,∀s∈[0, t).

Figure

Updating...

References

Related subjects :