Lp solutions of doubly reflected BSDEs under general assumptions✩ Shengjun FAN School of Mathematics, China University of Mining and Technology, Xuzhou, Jiangsu, 221116, PR China

arXiv:1701.04158v1 [math.PR] 16 Jan 2017

Abstract We establish several existence and uniqueness results for Lp (p > 1) solutions of reflected BSDEs with two continuous barriers and generators satisfying a one-sided Osgood condition together with a very general growth condition in y and a uniform continuity condition or a linear growth condition in z. To get existence a necessary and sufficient condition relating the growth of generator with that of barriers is given. We also prove that the solutions can be approximated by the penalization method and by some sequences of solutions of reflected BSDEs with two barriers. Our work strengthens some existing results. Keywords: Doubly reflected backward stochastic differential equation, Existence and uniqueness, Comparison theorem, Stability theorem, Lp solution 2010 MSC: 60H10, 60H30

1. Introduction Backward stochastic differential equations (BSDEs) were first introduced in linear case by Bismut [5] in 1973, and extended to a fully nonlinear version at the first time by Pardoux and Peng [50] in 1990. Later on, as a variation of the notion of nonlinear BSDEs, nonlinear reflected BSDEs (RBSDEs) with one and two continuous barriers were introduced by El Karoui et al. [11] and Cvitani´c and Karatzas [9] respectively. At present it has been widely recognized that these equations have natural connections with many problems in different mathematical fields, such as partial differential equations, mathematical finance, stochastic control and game theory, optimal switching problem and other optimality problems and others (see, e.g. [3], [4] [11], [12], [13], [22], [27], [29], [31], [32], [37], [45], [49], [51], [52], [53], [54], [55] and [56] etc.), and they provide a very useful and efficient tool for studying these problems. In Pardoux and Peng [50], El Karoui et al. [11] and Cvitani´c and Karatzas [9], the existence and uniqueness result of L2 solutions of BSDEs and RBSDEs with L2 data and continuous barriers are proved under the standard assumption that the generator g satisfies the linear growth condition and is Lipschitz continuous with respect to both variables y and z. Many attempts have been made to relax these assumptions, which are too strong for many interesting applications mentioned above. For example, many papers were devoted to solving RBSDEs with less regular barriers, see [4], [22], [23], [25], [40], [44] and [54]; many papers were interested in the existence and uniqueness result of solutions ✩ Supported

by the National Natural Science Foundation of China (No. 11371362) and the Qing Lan Project. Email address: [email protected] (Shengjun FAN)

Preprint submitted to ArXiv

January 17, 2017

for BSDEs or RBSDEs with data that are in Lp (p > 1) and L1 , see [7], [8], [10], [13], [17], [35] and [46] for non-reflected BSDEs, and [1], [3], [30], [39], [40] and [57] for reflected BSDEs; and more papers focused their attention on weakening the linear growth condition and Lipschitz-continuity condition of the generator g with respect to variables y and z, see [6], [7], [8], [13], [17], [18], [19], [20], [21], [33], [35], [36], [37], [46], [47] and [49] for non-reflected BSDEs, and [1], [3], [4], [14], [15], [16], [26], [28], [31], [34], [38], [39], [40], [41], [42], [43], [48], [57], [58] and [59] for reflected BSDEs. Enlightened by these works, particularly by Klimsiak [40] and Fan [16], in this paper we will establish several general existence and uniqueness results for Lp (p > 1) solutions of reflected BSDEs with two continuous barriers under weaker assumptions, which improves considerably some corresponding works. Our approach is based on a combination between existing methods, their refinement and perfection, but also on some new ideas. The rest of this paper is organized as follows. Section 2 contains some notations, definitions, assumptions together with some preliminaries which will be used later. In Section 3, we first establish a key estimate on Lp (p > 1) solutions of penalization equations for reflected BSDEs with one barrier and non-reflected BSDEs, see Proposition 3.3, and then prove a general comparison theorem of solutions for reflected BSDEs with two continuous barriers, which naturally yields uniqueness of the solutions under some very general assumptions, see Proposition 3.5 and Theorem 3.8. In Section 4, by Proposition 4.3 we show the convergence of Lp (p > 1) solutions of the penalization equations for reflected BSDEs with one barrier under several very weaker assumptions. Based on Proposition 3.3, Proposition 3.5 and Proposition 4.3, by Theorem 4.5 (resp. Theorem 4.8) we prove the existence and uniqueness of Lp (p > 1) solutions (resp. the existence of maximal and minimal Lp (p > 1) solutions) for reflected BSDEs with two continuous barriers under the assumption that the generator g satisfies a one-sided Osgood condition with a very general growth condition in y, see (H1) and (H3) in Section 2, a uniform continuity condition in z (resp. a linear growth condition in z), see (H2) (resp. (H2’)) in Section 2, and a generalized Mokobodzki condition which relates the growth of g and that of the barriers, see (H4) in Section 2. This generalized Mokobodzki condition (H4) is also proved to be necessary for existence of a solution in Theorem 4.1. In Section 5, we first establish a general approximation result for Lp (p > 1) solutions for reflected BSDEs with two continuous barriers under some elementary conditions, see Proposition 5.1, and based on it we prove an existence result of the minimal Lp (p > 1) solutions for reflected BSDEs with two continuous barriers under (H4), see Theorem 5.5, where the generator g has a general growth in y and a linear growth in z, but it is interesting that the g may be discontinuous in y as that considered in Fan and Jiang [20] and Zheng and Zhou [59]. In Section 6, we focus our attention on reflected BSDEs with two distinct barriers, which have been extensively investigated in Hamad`ene and Hassani [23], Hamad`ene and Hdhiri [26], Hamad`ene and Hassani [24], Hamad`ene et al. [25] and Bayraktar and Yao [4] by a method involved in a local solution. Different completely from these works, we will directly locate a semi-martingale X· ∈ V p with

− p p > 0 between two distinct and continuous reflecting barriers L· and U· satisfying L+ · , U· ∈ S , see

the notations V p and S p in Section 2 and Proposition 6.1 for details. This means that the Mokobodzki 2

condition holds naturally true in this case, and the generalized Mokobodzki condition holds also true if the generator g has a linear growth in y. Based on this work and some results obtained in previous sections, in Section 6 we establish several general results for Lp (p > 1) solutions of reflected BSDEs with two distinct barriers, see Theorem 6.2, Theorem 6.3 and Proposition 6.4. Finally, we would like to mention that the results obtained in this paper can be regarded as a generalization of Klimsiak [40] in the sense that the conditions required for the generator are greatly relaxed, and also as a generalization of Fan [16] from reflected BSDEs with one continuous barrier to that with two continuous barriers.

2. Notations, definitions, assumptions and preliminaries In the whole paper we fix a real number T > 0 and a positive integer d, and let (Bt )t∈[0,T ] be a standard d-dimensional Brownian motion defined on some complete filtered probability space (Ω, F , P), where (Ft )t∈[0,T ] is the completed σ-algebra filtration generated by (Bt )t∈[0,T ] and F = FT . We assume that if there is not a special illustration, all processes of this paper are defined on Ω × [0, T ], all notions whose definitions are related to some filtration are understood with respect to the filtration (Ft )t∈[0,T ] , and all equalities and inequalities between random elements are understood to hold P − a.s. To avoid ambiguity we stress that writing Xt = Yt , t ∈ [0, T ] we mean that P({ω : Xt (ω) = Yt (ω), t ∈ [0, T ]}) = 1, while writing Xt = Yt for each t ∈ [0, T ] we mean that ∀ t ∈ [0, T ], P({ω : Xt (ω) = Yt (ω)}) = 1. It is clear that they are equivalent if (Xt )t∈[0,T ] and (Yt )t∈[0,T ] are both continuous processes. In what follows, the variable ω in random elements is often omitted as usually done. Denote R+ := [0, +∞), a+ := max{a, 0} and a− := (−a)+ for any real number a. For a set A, we denote by Ac the complement of A and by

1A the indicator function of A. Let sgn(x) represent the sign

of a real number x. For n ≥ 1, the Euclidean norm of an element y ∈ Rn will be denoted by |y|. For p > 0, we define the following spaces: 1∧1/p

• Lp (FT ) the set of all FT -measurable random variables ξ satisfying kξkLp := (E[|ξ|p ]) < +∞; R  T • H the set of all progressively measurable processes X· satisfying P 0 |Xt |dt < +∞ = 1;

p io1∧1/p n hR T < +∞; • Hp the set of all processes X· ∈ H satisfying kXkHp := E 0 |Xt |dt

• S the set of all progressively measurable and continuous processes;  1∧1/p • S p the set of all processes Y· ∈ S satisfying kY kS p := E[supt∈[0,T ] |Yt |p ] < +∞; • M the set of all progressively measurable processes Z· satisfying P • Mp the set of all processes Z· ∈ M satisfying kZkMp 3

R T 0

 |Zt |2 dt < +∞ = 1;

  p/2 1∧1/p RT 2 := E |Zt | dt < +∞; 0

• M the set of all continuous local martingales; i h p/2 < +∞; • Mp the set of all martingales M· ∈ M satisfying E (hM iT )

• V the set of all progressively measurable and continuous processes of finite variation; • V + the set of all progressively measurable, continuous and increasing processes valued 0 at 0; • V p the set of all processes V· ∈ V satisfying E [|V |pT ] < +∞;

• V +,p the set of all processes V· ∈ V + satisfying E [|V |pT ] < +∞. Here and hereafter, for each V· ∈ V and t ∈ [0, T ], |V |t,T represents the random finite variation of V· on the interval [t, T ], and |V |0,T is denoted simply by |V |T . Clearly, if V· ∈ V + , then |V |t,T = VT − Vt . We now recall a definition used in Essaky and Hassani [15]. Definition 2.1. For any two processes K·1 and K·2 in V, we say that: • dK 1 ⊥dK 2 if and only if there exists a progressively measurable set D ⊂ Ω × [0, T ] such that "Z # # "Z T T 2 1 E 1D (t, ω) dKt (ω) = E 1Dc (t, ω) dKt (ω) = 0. 0

0

• dK 1 ≤ dK 2 if and only if for each progressively measurable set D ⊂ Ω × [0, T ], # "Z # "Z T T 1 2 1D (t, ω) dKt (ω) ≤ E 1D (t, ω) dKt (ω) , i.e., Kt1 − Ks1 ≤ Kt2 − Ks2 , 0 ≤ s ≤ t ≤ T. E 0

0

In the rest of this paper, we always assume that ξ is a FT -measurable random variable, V· ∈ V, L· ∈ S and U· ∈ S with Lt ≤ Ut for each t ∈ [0, T ], and that a random function g(ω, t, y, z) : Ω × [0, T ] × R × Rd 7−→ R is progressively measurable for each (y, z), which is usually called a generator. Definition 2.2. By a solution to BSDE (ξ, g + dV ) we understand a pair (Yt , Zt )t∈[0,T ] of progressively measurable processes such that Z Yt = ξ +

T

g(s, Ys , Zs )ds +

t

Z

T

t

dVs −

Z

t

T

Zs dBs , t ∈ [0, T ].

(2. 1)

By a solution to RBSDE (ξ, g+dV, L) we understand a triple (Yt , Zt , Kt )t∈[0,T ] of progressively measurable processes such that  Z T Z T Z T Z T   Yt = ξ + g(s, Ys , Zs )ds + dVs + dKs − Zs dBs , t ∈ [0, T ],     t t t t Lt ≤ Yt , t ∈ [0, T ],  Z T      K· ∈ V + and (Yt − Lt )dKt = 0. 0

4

(2. 2)

¯ By a solution to RBSDE (ξ, g+dV, U ) we understand a triple (Yt , Zt , At )t∈[0,T ] of progressively measurable processes such that  Z T Z T Z T Z T   g(s, Ys , Zs )ds + dVs − dAs − Zs dBs , t ∈ [0, T ],  Yt = ξ +    t t t t Yt ≤ Ut , t ∈ [0, T ],  Z T      A· ∈ V + and (Ut − Yt )dAt = 0.

(2. 3)

sively measurable processes such that  Z T Z T Z T Z T Z T   Y = ξ + g(s, Y , Z )ds + dV + dK − dA − Zs dBs , t ∈ [0, T ],  t s s s s s   t t t t t  Lt ≤ Yt ≤ Ut , t ∈ [0, T ],  Z T Z T     K , A ∈ V + , dK⊥dA and  (Yt − Lt )dKt = (Ut − Yt )dAt = 0. · ·

(2. 4)

0

By a solution to DRBSDE (ξ, g + dV, L, U ) we understand a quadruple (Yt , Zt , Kt , At )t∈[0,T ] of progres-

0

0

Remark 2.3. It is easy to see that

(i) the claim that (Yt , Zt , Kt )t∈[0,T ] is a solution of RBSDE (ξ, g + dV, L) is equivalent to the claim: – (Yt , Zt , Kt , 0)t∈[0,T ] is a solution of DRBSDE (ξ, g + dV, L, +∞). ¯ (ii) (Yt , Zt , At )t∈[0,T ] is a solution of RBSDE (ξ, g + dV, U ) is equivalent to – (Yt , Zt , 0, At )t∈[0,T ] is a solution of DRBSDE (ξ, g + dV, −∞, U ). (iii) (Yt , Zt )t∈[0,T ] is a solution of BSDE (ξ, g +dV ) is equivalent to anyone of the following three claims: – (Yt , Zt , 0)t∈[0,T ] is a solution of RBSDE (ξ, g + dV, −∞); ¯ – (Yt , Zt , 0)t∈[0,T ] is a solution of RBSDE (ξ, g + dV, +∞); – (Yt , Zt , 0, 0)t∈[0,T ] is a solution of DRBSDE (ξ, g + dV, −∞, +∞). (iv) (Yt , Zt , Kt )t∈[0,T ] is a solution of RBSDE (ξ, g + dV, L) is equivalent to ¯ g¯ + dV¯ , U ¯ ¯ ), where ξ¯ := – (Y¯t , Z¯t , A¯t )t∈[0,T ] := (−Yt , −Zt , Kt )t∈[0,T ] is a solution of RBSDE (ξ, ¯· := −L· , g¯(ω, t, y, z) := −g(ω, t, −y, −z). −ξ, V¯· := −V· , U

Definition 2.4. A solution (Yt , Zt )t∈[0,T ] of BSDE (ξ, g + dV ) is called the minimal (resp. maximal) one in some space if for any solution (Yt′ , Zt′ )t∈[0,T ] of BSDE (ξ, g + dV ) in this space, we have Yt ≤ Yt′ , t ∈ [0, T ]

(resp. Yt ≥ Yt′ , t ∈ [0, T ]).

(2. 5)

Similarly, we can define that • A solution (Yt , Zt , Kt )t∈[0,T ] of RBSDE (ξ, g + dV, L) is called the minimal (resp. maximal) one in some space if (2. 5) holds for any solution (Yt′ , Zt′ , Kt′ )t∈[0,T ] of RBSDE (ξ, g + dV, L) in this space.

5

¯ • A solution (Yt , Zt , At )t∈[0,T ] of RBSDE (ξ, g + dV, U ) is called the minimal (resp. maximal) one in some space if (2. 5) holds for any solution (Yt′ , Zt′ , A′t )t∈[0,T ] of RBSDE (ξ, g + dV, U ) in this space. • A solution (Yt , Zt , Kt , At )t∈[0,T ] of DRBSDE (ξ, g +dV, L, U ) is called the minimal (resp. maximal)

one in some space if (2. 5) holds for any solution (Yt′ , Zt′ , Kt′ , A′t )t∈[0,T ] of DRBSDE (ξ, g+dV, L, U )

in this space. In this paper, we will mainly use the following assumptions with respect to the generator, the terminal condition and the barriers, where p > 1. (H1) g satisfies the one-sided Osgood condition in y, i.e., there exists a nondecreasing and concave R du = +∞ such that function ρ(·) : R+ 7→ R+ with ρ(0) = 0, ρ(u) > 0 for u > 0 and 0+ ρ(u) dP × dt − a.e., ∀ y1 , y2 ∈ R, z ∈ Rd ,

(g(ω, t, y1 , z) − g(ω, t, y2 , z))sgn(y1 − y2 ) ≤ ρ(|y1 − y2 |). (H2)

(i) g is continuous in y, i.e, dP × dt − a.e., ∀ z ∈ Rd , g(ω, t, ·, z) is continuous; (ii) g is uniformly continuous in z, i.e., there exists a nondecreasing and continuous function φ(·) : R+ 7→ R+ with φ(0) = 0 such that dP × dt − a.e., ∀ y ∈ R, z1 , z2 ∈ Rd , |g(ω, t, y, z1 ) − g(ω, t, y, z2)| ≤ φ(|z1 − z2 |).

(H2’)

(i) g is stronger continuous in (y, z), i.e., dP × dt − a.e., ∀ y ∈ R, g(ω, t, y, ·) is continuous, and g(ω, t, ·, z) is continuous uniformly with respect to z; (ii) g has a stronger linear growth in z, i.e., there exists two constants µ, λ ≥ 0 and a nonnegative process f· ∈ Hp such that dP × dt − a.e., ∀ y ∈ R, z ∈ Rd ,

|g(ω, t, y, z) − g(ω, t, y, 0)| ≤ ft (ω) + µ|y| + λ|z|. (H3)

(i) g has a general growth in y, i.e, dP × dt − a.e., ∀r > 0, ϕ· (r) := sup |g(·, y, 0) − g(·, 0, 0)| |y|≤r

belongs to the space H; (ii) g(·, 0, 0) ∈ Hp . (H4)

(i) L· , U· ∈ S, ξ ∈ Lp (FT ) and LT ≤ ξ ≤ UT ; (ii) There exists a X· ∈ Mp + V p such that g(·, X· , 0) ∈ Hp and Lt ≤ Xt ≤ Ut for each t ∈ [0, T ].

Remark 2.5. Without loss of generality, we will always assume that the functions ρ(·) and φ(·) defined respectively in (H1) and (H2) are of linear growth, i.e., there exists a constant A > 0 such that ∀ x ∈ R+ , ρ(x) ≤ A(x + 1) and φ(x) ≤ A(x + 1). In order to illustrate our results more clearly, the following several assumptions will also be used.

6

(H1s) g satisfies the monotonicity condition in y, i.e., there exists a constant µ ∈ R such that dP×dt−a.e., ∀ y1 , y2 ∈ R, z ∈ Rd ,

(g(ω, t, y1 , z) − g(ω, t, y2 , z))sgn(y1 − y2 ) ≤ µ|y1 − y2 |. (H2s)

(i) g is continuous in y, i.e, dP × dt − a.e., ∀ z ∈ Rd , g(ω, t, ·, z) is continuous; (ii) g satisfies the uniform Lipschitz condition in z, i.e., there exists a constant λ ≥ 0 such that dP × dt − a.e., ∀ y ∈ R, z1 , z2 ∈ Rd ,

|g(ω, t, y, z1 ) − g(ω, t, y, z2 )| ≤ λ|z1 − z2 |. (H3s) g has a linear growth in y, i.e., there exists a constant µ ≥ 0 and a nonnegative process f· ∈ Hp such that dP × dt − a.e., ∀ y ∈ R, |g(ω, t, y, 0)| ≤ ft (ω) + µ|y|. Remark 2.6. It is clear that assumptions (H1s), (H2s) and (H3s) are respectively stronger than (H1), (H2) and (H3). And, (ii) of (H2) implies (ii) of (H2’). Moreover, the following several assumptions will be used in some technical results of this paper. ¯ such that dP × dt − a.e., (AA) There exists two nonnegative constants µ ¯ and λ ¯ g(ω, t, y, z)sgn(y) ≤ f¯t (ω) + µ ¯|y| + λ|z|, ∀ y ∈ R, z ∈ Rd , where f¯t is a nonnegative process belonging to Hp . (HH)

(i) g is continuous in (y, z), i.e, dP × dt − a.e., g(ω, t, ·, ·) is continuous; (ii) g has a general growth in (y, z), i.e., there exists a constant λ ≥ 0, a nonnegative process f· ∈ Hp and a nonnegative function ψ· (r) ∈ S such that dP × dt − a.e., ∀ y ∈ R, z ∈ Rd , |g(ω, t, y, z)| ≤ ft (ω) + ψt (ω, |y|) + λ|z|, here and hereafter, S denotes the set of nonnegative functions ψt (ω, r) : Ω × [0, T ] × R+ 7→ R+ satisfying the following two conditions: – dP × dt − a.e., the function r 7→ ψt (ω, r) is increasing and ψt (ω, 0) = 0; – for each r ≥ 0, ψ· (r) ∈ H.

(H4L)

(i) L· ∈ S, ξ ∈ Lp (FT ) and LT ≤ ξ; (ii) There exists a X· ∈ Mp + V p such that g(·, X· , 0) ∈ Hp and Lt ≤ Xt for each t ∈ [0, T ].

(H4U)

(i) U· ∈ S, ξ ∈ Lp (FT ) and ξ ≤ UT ; (ii) There exists a X· ∈ Mp + V p such that g(·, X· , 0) ∈ Hp and Xt ≤ Ut for each t ∈ [0, T ].

Remark 2.7. It is easy to verify that, see also Fan [16] for details, (i) (H2)+(H3) ⇒ (HH); (H2’)+(H3) ⇒ (HH); (H1)+(HH)(ii) ⇒ (AA); (HH)(ii) ⇒ (H3); 7

p − p (ii) (H4) ⇒ (H4L) + (H4U); (H4L)(ii) ⇒ L+ · ∈ S ; (H4U)(ii) ⇒ U· ∈ S ; !

p (iii) L+ · ∈ S together with

(iv) U·−

g(t, sup L+ s , 0) s∈[0,t]

t∈[0,T ]

  p − ∈ S together with g(t, − inf Us , 0) s∈[0,t]

∈ Hp implies (H4L)(ii);

t∈[0,T ]

∈ Hp implies (H4U)(ii);

p − p (v) If (H3s) holds, then (H4L)(ii) ⇔ L+ · ∈ S and (H4U)(ii) ⇔ U· ∈ S .

Finally, let us recall some important results on reflected BSDEs with one continuous barrier and non-reflected BSDEs obtained in Fan [16] by virtue of (iv) in Remark 2.3, which will be used. Proposition 2.8. Assume that p > 1, ξ ∈ Lp (FT ), V· ∈ V p and the generator g satisfies assumptions (H1), (H2) and (H3). Then, (i) BSDE (ξ, g + dV ) admits a unique solution (yt , zt )t∈[0,T ] in S p × Mp . (ii) Assume further that (H4L) holds. RBSDE (ξ, g + dV, L) admits a unique solution (yt , zt , Kt )t∈[0,T ] in S p × Mp × V +,p . ¯ (ξ, g + dV, U ) admits a unique solution (yt , zt , At )t∈[0,T ] (iii) Assume further that (H4U) holds. RBSDE in S p × Mp × V +,p . Proposition 2.9. Assume that p > 1, ξ i ∈ Lp (FT ), V·i ∈ V p , and the generator g i satisfies (H1), (H2) and (H3) for i = 1, 2. Assume further that ξ 1 ≤ ξ 2 , dV 1 ≤ dV 2 , and for each (y, z) ∈ R × Rd , dP × dt − a.e., g 1 (t, y, z) ≤ g 2 (t, y, z). We have (i) For i = 1, 2, let (Y·i , Z·i ) ∈ S p × Mp be the unique solution of BSDE (ξ i , g i + dV i ). Then, Yt1 ≤ Yt2 for each t ∈ [0, T ]. (ii) For i = 1, 2, suppose that (H4L) hold for ξ i , Li and X i associated with g i , and that (Y·i , Z·i , K·i ) ∈ S p × Mp × V +,p is the unique solution of RBSDE (ξ i , g i + dV i , Li ). If L1· ≤ L2· , then Yt1 ≤ Yt2 , t ∈ [0, T ].

(2. 6)

Moreover, if L1· = L2· , then dK 1 ≥ dK 2 . (iii) For i = 1, 2, suppose that (H4U) hold for ξ i , Li and X i associated with g i , and that (Y·i , Z·i , Ai· ) ∈ ¯ S p × Mp × V +,p is the unique solution of RBSDE (ξ i , g i + dV i , U i ). If U·1 ≤ U·2 , then (2. 6) holds

true. Moreover, if U·1 = U·2 , then dA1 ≤ dA2 .

Remark 2.10. From the proof in Fan [16], we can stress that in order to obtain (2. 6) in (ii) of Proposition 2.9 we do not need the condition that Z T (Yt2 − L2t ) dKt2 = 0, 0

8

and in order to obtain (2. 6) in (iii) of Proposition 2.9 we do not need the condition that Z

0

T

(Ut1 − Yt1 ) dA1t = 0.

This fact will be sufficiently utilized in Section 3. Remark 2.11. Proposition 2.8, Proposition 2.9 and Remark 2.10 hold still true if we replace (H2) with (H2’), and the word “unique” with the word “maximal (minimal)” in the statements of them.

3. Estimates, comparison and uniqueness of the solutions In this section, we will first establish some uniform estimates on the sequence of Lp (p > 1) solutions of penalized RBSDEs with one continuous barrier and non-reflected BSDEs, which will paly an important role in the proof of our main results. Then, we establish a general comparison theorem for Lp (p > 1) solutions of doubly RBSDEs. Finally, we show the uniqueness of the solution under (H1) and (H2) (ii). Firstly, in view of (iv) in Remark 2.3, the following two lemmas can be regarded as direct corollaries of Lemma 3 and Propositions 1-2 in Fan [16]. Lemma 3.1. Assume that p > 1, ξ ∈ Lp (FT ), V· ∈ V p and the generator g satisfies assumption (AA). Let (Y· , Z· ) ∈ S p × Mp be a solution of BSDE (ξ, g + dV ). Then, there exists a constant C > 0 depending ¯ T such that for each t ∈ [0, T ], only on p, µ ¯, λ,  ! p2 Z T p 2 E  sup |Ys | + |Zs | ds + s∈[t,T ]

"

t

p

≤ CE |ξ| + |V

|pt,T

+

Z

t

T

!p  |g(s, Ys , Zs )|ds Ft  t !p # f¯s ds Ft . Z

T

Lemma 3.2. Assume that p > 1, ξ ∈ Lp (FT ), V· ∈ V p , K· , A· ∈ V +,p , and the generator g satisfies assumptions (H1) with ρ(·), (ii) of (H2’) with f· , µ and λ, and (ii) of (H3). (i) Let (Y· , Z· ) ∈ S p × Mp be a solution of BSDE (ξ, g + dV¯ ) with V¯· = V· + K· , and the following assumption (B1) hold: ¯ · ∈ S p such that g(·, X ¯ · , 0) ∈ Hp and Yt ≤ X ¯ t for each t ∈ [0, T ]. (B1) There exists a X Then, there exists a constant C > 0 depending only on p, µ, λ, A, T such that for each t ∈ [0, T ],  ! p2 !p  Z T Z T E |Zs |2 ds + |KT − Kt |p + |g(s, Ys , Zs )|ds Ft  t t " !p Z T (3. 1) ¯ s |p + ≤ CE sup |Ys |p + |V |pt,T + sup |X fs ds + 1 s∈[t,T ] s∈[t,T ] t !p !p # Z T Z T ¯ s , 0)| ds + + |g(s, X |g(s, 0, 0)| ds Ft . t t

(ii) Let (Y· , Z· ) ∈ S p × Mp be a solution of BSDE (ξ, g + dV ) with V · = V· − A· , and the following assumption (B2) hold: 9

(B2) There exists a X · ∈ S p such that g(·, X · , 0) ∈ Hp and X t ≤ Yt for each t ∈ [0, T ]. Then, there exists a constant C > 0 depending only on p, µ, λ, A, T such that for each t ∈ [0, T ],  !p  ! p2 Z T Z T + |AT − At |p + E |g(s, Ys , Zs )|ds Ft  |Zs |2 ds t t " !p Z T (3. 2) ≤ CE sup |Ys |p + |V |pt,T + sup |X s |p + fs ds + 1 s∈[t,T ] s∈[t,T ] t !p !p # Z T Z T |g(s, 0, 0)| ds Ft . + |g(s, X s , 0)| ds + t t

Let us now establish the following key estimate on Lp (p > 1) solutions of the penalization equation for reflected BSDEs with one continuous barrier and non-reflected BSDEs. Proposition 3.3. Assume that p > 1, ξ ∈ Lp (FT ), V· ∈ V p , the generator g satisfies assumptions (H1), (H2) and (H3), and assumption (H4) holds for L· , U· , ξ and X· . We have

¯ (ξ, g¯n + dV, U ) (i) For each n ≥ 1, let (Y n· , Z n· , An· ) ∈ S p × Mp × V +,p be the unique solution of RBSDE with g¯n (t, y, z) := g(t, y, z) + n(y − Lt )− , i.e.,  Z T Z T Z T Z T Z T  n n n n n  Y dK g(s, Y dV + dA Z ns dBs , t ∈ [0, T ], = ξ + , Z )ds + − −  t s s s t s   t t t t  Z tt  n n − (Y s − Ls ) ds, t ∈ [0, T ], K t := n   0 Z T    n   Y t ≤ Ut , t ∈ [0, T ] and (Ut − Y nt )dAnt = 0. 0

(3. 3)

(Recall (ii) of Remark 2.7 and (iii) of Proposition 2.8). Then, Y n· increases in n, dAn ≤ dAn+1 ,

and there exists a random variable η ∈ L1 (FT ) such that for each n ≥ 1 and t ∈ [0, T ],  ! p2 !p  Z T Z T E  sup |Y ns |p + |Z ns |2 ds |g(s, Y ns , Z ns )|ds Ft  + |K nT − K nt |p + |AnT − Ant |p + s∈[t,T ] t t   ≤ E η Ft . (3. 4)

¯ ·n ) ∈ S p × Mp × V +,p be the unique solution of RBSDE (ξ, g + dV, L) (ii) For each n ≥ 1, let (Y¯·n , Z¯·n , K n with gn (t, y, z) := g(t, y, z) − n(y − Ut )+ , i.e.,  Z T Z T Z T Z T Z T  n n ¯n n n  ¯ ¯ ¯ ¯ g(s, Ys , Zs )ds + dVs + dKt − dAs − Z¯sn dBs , t ∈ [0, T ],  Yt = ξ +   t t t t  Z tt  n n + ¯ ¯ At := n (Ys − Us ) ds, t ∈ [0, T ],   0 Z T     n ¯ ¯ tn = 0.  Lt ≤ Yt , t ∈ [0, T ] and (Y¯tn − Lt )dK 0

(3. 5)

¯ n ≤ dK ¯ n+1 , (Recall (ii) of Remark 2.7 and (ii) of Proposition 2.8). Then, Y¯·n decreases in n, dK

and there exists a random variable η¯ ∈ L1 (FT ) such that for each n ≥ 1 and t ∈ [0, T ],  !p  ! p2 Z T Z T ¯n − K ¯ n |p + |A¯n − A¯n |p + + |K |g(s, Y¯sn , Z¯sn )|ds Ft  E  sup |Y¯sn |p + |Z¯sn |2 ds T t T t s∈[t,T ] t t ≤ E [ η¯| Ft ] .

(3. 6)

10

(iii) For each n ≥ 1, let (Y·n , Z·n ) ∈ S p × Mp be the unique solution of BSDE (ξ, gn + dV ) with gn (t, y, z) := g(t, y, z) + n(y − Lt )− − n(y − Ut )+ (Recall (i) of Proposition 2.8), i.e.,  Z T Z T Z T Z T Z T  n n n n n   Yt = ξ + g(s, Ys , Zs )ds + dVs + dKt − dAs − Zsn dBs , t ∈ [0, T ], t t t Z tt Zt t    Ktn := n (Ysn − Ls )− ds and Ant := n (Ysn − Us )+ ds, t ∈ [0, T ]. 0

0

(3. 7)

¯ n ≤ dK n ≤ dK n and dAn ≤ dAn ≤ dA¯n . Then, for each n ≥ 1 and t ∈ [0, T ], Y nt ≤ Ytn ≤ Y¯tn , dK

And, there exists a random variable η ∈ L1 (FT ) such that for each n ≥ 1 and t ∈ [0, T ],  ! p2 !p  Z T Z T E  sup |Ysn |p + |Zsn |2 ds + |KTn − Ktn |p + |AnT − Ant |p + |g(s, Ysn , Zsn )|ds Ft  s∈[t,T ] t t ≤ E [ η| Ft ] .

(3. 8)

Proof. Let p > 1, ξ ∈ Lp (FT ), V· ∈ V p , the generator g satisfy assumptions (H1) with ρ(·), (H2) with φ(·), and (H3) with ϕ· (r), and assumption (H4) hold for L· , U· , ξ and X· . By representation property of Brownian filtration, we can let (C· , H· ) ∈ V p × Mp be the unique pair of processes such that Xt = XT −

Z

T

Z

dCs −

t

T

Hs dBs , t ∈ [0, T ].

t

(3. 9)

It follows from (ii) of (H2) that dP × dt − a.e., |g(·, X· , H· )| ≤ |g(·, X· , 0)| + φ(|H· |) ≤ |g(·, X· , 0)| + A|H· | + A, from which together with (H4) we know that g(·, X· , H· ) ∈ Hp , and then Z t Z t Z t ˇ t := dVs− ∈ Hp dCs− + g − (s, Xs , Hs )ds + K and Aˇt :=

Z

0

0

0

t

g + (s, Xs , Hs )ds +

Z

t

0

0

dCs+ +

Z

0

t

dVs+ ∈ Hp .

Thus, the equation (3. 9) can be rewritten in the form Xt = XT +

Z

T

g(s, Xs , Hs )ds +

t

Z

T

dVs +

t

Z

t

T

ˇs − dK

Z

t

T

dAˇs −

Z

t

T

Hs dBs , t ∈ [0, T ].

¯ · , Z¯· ) ∈ S p × Mp be the unique solution of the BSDE On the other hand, by (i) of Proposition 2.8, let (X ¯ t = XT ∨ ξ + X

Z

T

¯ s , Z¯s )ds + g(s, X

T

Z

dVs +

t

t

t

T

Z

ˇs − dK

T

Z

t

Z¯s dBs , t ∈ [0, T ],

and (X · , Z · ) ∈ S p × Mp be the unique solution of the following BSDE X t = XT ∧ ξ +

Z

t

T

g(s, X s , Z s )ds +

Z

t

T

dVs −

Z

t

T

dAˇs −

Z

T

t

Z s dBs , t ∈ [0, T ].

It follows from (i) of Remark 2.7 that g satisfies assumption (AA). Then, Lemma 3.1 yields that ¯ · , Z¯· ) ∈ Hp and g(·, X , Z ) ∈ Hp , which together with (H2) leads to g(·, X · · ¯ · , 0)| ≤ |g(·, X ¯ · , Z¯· )| + φ(|Z¯· |) ≤ |g(·, X ¯ · , Z¯· )| + A|Z¯· | + A ∈ Hp |g(·, X 11

(3. 10)

and |g(·, X · , 0)| ≤ |g(·, X · , Z · )| + φ(|Z · |) ≤ |g(·, X · , Z · )| + A|Z · | + A ∈ Hp .

(3. 11)

In what follows, for each n ≥ 1, by (i) of Proposition 2.8, let (Y˙ ·n , Z˙ ·n ) ∈ S p × Mp and (Y¨·n , Z¨·n ) ∈

S p × Mp be respectively the unique solution of the following BSDEs: Z T Z T Z T Z n n ˙n n − ˙ ˙ ˙ Yt = XT ∧ ξ + g(s, Ys , Zs )ds + dVs + n (Ys − Ls ) ds − t

t

dAˇs −

Z

T

(Y¨sn − Us )+ ds −

Z

T

t

T

t

t

Z˙ sn dBs ,

and Y¨tn = XT ∨ ξ +

Z

T

t

g(s, Y¨sn , Z¨sn )ds +

T

Z

dVs +

Z

t

t

T

ˇs − n dK

Z

T

t

t

Z¨sn dBs .

Note that L· ≤ X· ≤ U· . We have, with t ∈ [0, T ] and n ≥ 1, Xt = XT +

Z

T

Z

T

g(s, Xs , Hs )ds +

t

Z

T

dVs + n t

Z

t

T

(Xs − Ls )− ds +

Z

T

t

ˇs − dK

Z

T

Z

T

t

dAˇs −

Z

T

dAˇs −

Z

T

Hs dBs ,

t

and Xt = XT +

g(s, Xs , Hs )ds +

t

Z

T

dVs +

t

Z

T

t

ˇs − n dK

Z

t

T

+

(Xs − Us ) ds −

t

Hs dBs .

t

It then follows from (ii) of (H4) and (i) of Proposition 2.9 that for each n ≥ 1, X t ≤ Y˙ t1 ≤ Y˙ tn ≤ Xt ≤ Ut , t ∈ [0, T ]

(3. 12)

¯ t , t ∈ [0, T ], Lt ≤ Xt ≤ Y¨tn ≤ Y¨t1 ≤ X

(3. 13)

and

which means that for each n ≥ 1, (B1) in Lemma 3.2 holds for X· , the generator g and Y˙ ·n , and (B2) in

Lemma 3.2 holds for X· , the generator g and Y¨·n . Thus, in view of the fact that g satisfies (H1), (H2’) with f· = A, µ = 0 and λ = A, and (H3), by Lemma 3.2 together with (3. 12) and (3. 13) we obtain that there exists a constant C > 0 depending only on p, A, T such that for each t ∈ [0, T ], " i h n n p ¯ s |p + |V |p + |AˇT − Aˇt |p + 1 ≤ CE sup |X s |p + sup |X sup E |K˙ T − K˙ t | Ft t,T n≥1 s∈[t,T ] s∈[t,T ] ! !p # p Z T Z T + |g(s, Xs , 0)| ds + |g(s, 0, 0)| ds Ft t t

(3. 14)

and

sup E

n≥1

with

h

|A¨nT

−



A¨nt |p Ft

i

≤ CE

"

¯ s |p + |V |p + |K ˇT − K ˇ t |p + 1 sup |X s |p + sup |X t,T s∈[t,T ] !p !p # Z T Z T + |g(s, Xs , 0)| ds + |g(s, 0, 0)| ds Ft t t

s∈[t,T ]

K˙ tn := n

Z

0

and A¨nt := n

t

Z

0

t

(Y˙ sn − Ls )− ds, t ∈ [0, T ]

(Y¨sn − Us )+ ds, t ∈ [0, T ].

In the sequel, we will prove (i)-(iii) respectively. 12

(3. 15)

¯ (ξ, g¯n +dV, U ) (i) For each n ≥ 1, let (Y n· , Z n· , An· ) ∈ S p ×Mp ×V +,p be the unique solution of RBSDE

with g¯n (t, y, z) := g(t, y, z) + n(y − Lt )− , i.e., (3. 3).

Firstly, it follows from (iii) and (i) of Proposition 2.9 that for each n ≥ 1, ¯ t , t ∈ [0, T ] and dAn ≤ dAn+1 . ≤X Y 1t ≤ Y nt ≤ Y n+1 t

(3. 16)

Then, in view of (3. 12), by (iii) of Proposition 2.9 with Remark 2.10 we deduce that for each n ≥ 1, X t ≤ Y˙ tn ≤ Y nt , t ∈ [0, T ],

(3. 17)

which means that for each t ∈ [0, T ], |K nT − K nt | = n

Z

T

t

(Y ns − Ls )− ds ≤ n

Z

T

t

(Y˙ sn − Ls )− ds = |K˙ Tn − K˙ tn |.

(3. 18)

Furthermore, it follows from (3. 11) and (3. 17) that for each n ≥ 1, (B2) in Lemma 3.2 holds for X · ,

the generator g and Y n· . Thus, in view of the fact that g satisfies (H1), (H2’) with f· = A, µ = 0 and λ = A, and (H3), by (ii) of Lemma 3.2 we know that for each t ∈ [0, T ] and n ≥ 1,  ! p2 !p  Z T Z T |g(s, Y ns , Z ns )| ds Ft  + |AnT − Ant |p + E |Z ns |2 ds t t " ≤ CE

(3. 19)

sup |Y ns |p + |V |pt,T + |K nT − K nt |p + sup |X s |p + 1 s∈[t,T ] !p !p # Z T Z T + |g(s, X s , 0)| ds + |g(s, 0, 0)| ds Ft . t t

s∈[t,T ]

It then follows from (3. 16)– (3. 19) and (3. 14) that (3. 4) holds true with " ¯ ¯ s |p + |V |p + |AˇT − Aˇt |p + 1 η := C sup |X s |p + sup |X t,T s∈[t,T ] s∈[t,T ] !p !p Z Z T Z T + |g(s, Xs , 0)| ds + |g(s, X s , 0)| ds + t

t

t

T

|g(s, 0, 0)| ds

!p #

for some constant C¯ > 0 depending only on p, A, T . (ii) For each n ≥ 1, let (Y¯·n , Z¯·n , A¯n· ) ∈ S p ×Mp ×V +,p be the unique solution of RBSDE (ξ, g¯n +dV, L)

with g n (t, y, z) := g(t, y, z) − n(y − Ut )+ , i.e., (3. 5).

Firstly, it follows from (i) and (ii) of Proposition 2.9 that for each n ≥ 1, ¯ n ≤ dK ¯ n+1 . X t ≤ Y¯tn+1 ≤ Y¯tn ≤ Y¯t1 , t ∈ [0, T ] and dK

(3. 20)

Then, in view of (3. 13), by (ii) of Proposition 2.9 with Remark 2.10 we deduce that for each n ≥ 1, ¯ t , t ∈ [0, T ], Y¯tn ≤ Y¨tn ≤ X

(3. 21)

which means that for each t ∈ [0, T ], |A¯nT − A¯nt | = n

Z

t

T

(Y¯sn − Us )+ ds ≤ n

13

Z

t

T

(Y¨sn − Us )+ ds = |A¨nT − A¨nt |.

(3. 22)

¯·, Furthermore, it follows from (3. 10) and (3. 21) that for each n ≥ 1, (B1) in Lemma 3.2 holds for X

the generator g and Y¯·n . Thus, in view of the fact that g satisfies (H1), (H2’) with f· = A, µ = 0 and λ = A, and (H3), by (i) of Lemma 3.2 we know that for each t ∈ [0, T ] and n ≥ 1,  !p  ! p2 Z T Z T ¯ Tn − K ¯ tn |p + + |K |g(s, Y¯sn , Z¯sn )| ds Ft  E |Z¯sn |2 ds t t " ¯ s |p + 1 ≤ CE sup |Y¯sn |p + |V |p + |A¯nT − A¯nt |p + sup |X t,T

s∈[t,T ]

+

Z

T

t

¯ s , 0)| ds |g(s, X

!p

+

Z

t

T

!p # |g(s, 0, 0)| ds Ft .

It then follows from (3. 20)– (3. 23) and (3. 15) that (3. 6) holds true with " ¯ s |p + |V |p + |K ˇT − K ˇ t |p + 1 η¯ := C¯ sup |X s |p + sup |X t,T s∈[t,T ] s∈[t,T ] !p !p Z T Z T Z ¯ s , 0)| ds + + |g(s, Xs , 0)| ds + |g(s, X t

(3. 23)

s∈[t,T ]

t

T

t

|g(s, 0, 0)| ds

!p #

for some constant C¯ > 0 depending only on p, A, T . (iii) For each n ≥ 1, let (Y·n , Z·n ) ∈ S p × Mp be the unique solution of BSDE (ξ, gn + dV ) with

gn (t, y, z) := g(t, y, z) + n(y − Lt )− − n(y − Ut )+ , i.e., (3. 7).

It follows from (3. 5) and (3. 3) that, with t ∈ [0, T ] and n ≥ 1, Y nt = ξ +

Z

T

Z

T

t

g(s, Y ns , Z ns )ds+

Z

T

g(s, Y¯sn , Z¯sn )ds+

Z

T

dVs +n

t

Z

T

Z

T

t

(Y ns −Ls )− ds−n

Z

T

(Y¯sn −Ls )− ds−n

Z

T

t

Z

T

Z

T

(Y ns −Us )+ ds−

t

dAns −

Z

T

¯ n− dK s

Z

T

t

Z ns dBs ,

and Y¯tn = ξ +

t

dVs +n

t

t

t

(Y¯sn −Us )+ ds+

t

t

Z¯sn dBs .

Then, (i) of Proposition 2.9 together with (3. 17) and (3. 21) yields that for each n ≥ 1, ¯ t , t ∈ [0, T ], X t ≤ Y˙ tn ≤ Y nt ≤ Ytn ≤ Y¯tn ≤ Y¨tn ≤ X

(3. 24)

which means that for each t ∈ [0, T ], ¯ Tn − K ¯ tn | ≤ |KTn − Ktn | ≤ |K n − K n | ≤ |K˙ Tn − K˙ tn | |K T t

(3. 25)

|AnT − Ant | ≤ |AnT − Ant | ≤ |A¯nT − A¯nt | ≤ |A¨nT − A¨nt |.

(3. 26)

and

¯ = A by (i) ¯ = A and λ Furthermore, note that g satisfies assumption (AA) with f¯· = |g(·, 0, 0)| + 2A, µ

of Remark 2.7. It follows from Lemma 3.1 that there exists a constant C ′ > 0 depending only on p, A, T such that for each t ∈ [0, T ] and n ≥ 1,  ! p2 !p  Z T Z T E + |Zsn |2 ds |g(s, Ysn , Zsn )| ds Ft  t t # " !p Z T p ′ p n n p n n p ≤ C E |ξ| + |V |t,T + |KT − Kt | + |AT − At | + |g(s, 0, 0)| ds + 1 Ft . t 14

(3. 27)

Finally, in view of (3. 24)– (3. 27) and (3. 14)–(3. 15), we can deduce that (3. 8) holds true with " ¯ s |p + |V |p + |K ˇT − K ˇ t |p + |AˇT − Aˇt |p η := C¯ sup |X s |p + sup |X t,T s∈[t,T ] s∈[t,T ] !p !p # Z Z T

+

t

T

|g(s, Xs , 0)| ds

+

t

|g(s, 0, 0)| ds

+1

for some constant C¯ > 0 depending only on p, A, T . The proof of Proposition 3.3 is then complete. Remark 3.4. In view of Remark 2.11, all conclusions of Proposition 3.3 hold still true if we replace (H2) with (H2’), and the expression “the unique solution” with “the maximal (minimal) solution” in its statement. Furthermore, (3. 4), (3. 6) and (3. 8) hold still true if we replace (H2) with (H2’), and the expression “the unique solution” with “any solution” in its statement. In fact, in this case, it is enough to let (X · , Z · ) and (Y˙ ·n , Z˙ ·n ) be respectively the minimal solution of the corresponding equation ¯ · , Z¯· ) be respectively the maximal solution of the instead of the unique solution, and let (Y¨·n , Z¨·n ) and (X corresponding equation instead of the unique solution in the procedure of proof of Proposition 3.3, with omitting the comparisons between the processes indexed with n and n + 1. We now establish a general comparison theorem for Lp (p > 1) solutions of doubly RBSDEs. Proposition 3.5 (Comparison Theorem). Let p > 1, ξ i ∈ Lp (FT ), V·i ∈ V p , Li· , U·i ∈ S and (Y·i , Z·i , K·i , Ai· ) ∈

S p ×Mp ×V +,p ×V +,p be a solution of DRBSDE (ξ i , g i +dV i , Li , U i ) for i = 1, 2. If ξ 1 ≤ ξ 2 , dV 1 ≤ dV 2 , L1· ≤ L2· , U·1 ≤ U·2 , and either   g 1 satisf ies (H1) and (H2)(ii);  dP × dt − a.e., 1 1 2 g 1 (t, Y 2 , Z 2 ) − g 2 (t, Y 2 , Z 2 )
≤ 0 {Yt >Yt } t t t t or

  g 2 satisf ies (H1) and (H2)(ii);  dP × dt − a.e., 1 1 2 g 1 (t, Y 1 , Z 1 ) − g 2 (t, Y 1 , Z 1 )
≤ 0 {Yt >Yt } t t t t

is satisfied, then Yt1 ≤ Yt2 for each t ∈ [0, T ].

Proof. It follows from Itˆ o-Tanaka’s formula that for each t ∈ [0, T ], (Yt1 − Yt2 )+

Z T ≤ (ξ 1 − ξ 2 )+ + sgn((Ys1 − Ys2 )+ )(dVs1 − dVs2 ) t Z T
1 + sgn((Ys − Ys2 )+ ) g 1 (s, Ys1 , Zs1 ) − g 2 (s, Ys2 , Zs2 ) ds Z T Zt T

1 2 + 1 2 sgn((Ys1 − Ys2 )+ ) dA2s − dA1s + sgn((Ys − Ys ) ) dKs − dKs + t Zt T 1 2 + 1 2 sgn((Ys − Ys ) )(Zs − Zs )dBs . + t

Since L1t ≤ L2t ≤ Yt2 , L1t ≤ Yt1 , t ∈ [0, T ] and Z

t

T

sgn((Ys1 − Ys2 )+ ) dKs1 − dKs2




≤ =

RT 0

Z

(Ys1 − L1s )dKs1 = 0, we have T

Zt T t

sgn((Ys1 − Ys2 )+ )dKs1 ≤

Z

t

T

sgn((Ys1 − L1s )+ )dKs1

1{Ys1 >L1s } |Ys1 − L1s |−1 (Ys1 − L1s )dKs1 = 0. 15

Similarly, since Yt1 ≤ Ut1 ≤ Ut2 , Yt2 ≤ Ut2 , t ∈ [0, T ] and Z

t

T

sgn((Ys1

−

Ys2 )+ )

dA2s

−

dA1s




≤ =

Z

Zt

t

T

T

RT 0

sgn((Ys1

(Us2 − Ys2 )dA2s = 0, we have

−

Ys2 )+ )dA2s

≤

Z

T

t

sgn((Us2 − Ys2 )+ )dA2s

1{Us2 >Ys2 } |Us2 − Ys2 |−1 (Us2 − Ys2 )dA2s = 0.

Thus, noticing that ξ 1 ≤ ξ 2 and dV 1 ≤ dV 2 , by virtue of the previous three inequalityies we can get that (Yt1

−

Yt2 )+

Z

≤

T

tZ

+


sgn((Ys1 − Ys2 )+ ) g 1 (s, Ys1 , Zs1 ) − g 2 (s, Ys2 , Zs2 ) ds

t

T

sgn((Ys1 − Ys2 )+ )(Zs1 − Zs2 )dBs , t ∈ [0, T ].

Now, in view of the assumptions of g 1 and g 2 , the rest proof runs as the proof of Theorem 1 in Fan and Jiang [19]. The only difference lies in that in order to deal with the Lp solution we need to use  h i p−1 1 p p E [|XY |] ≤ (E [|X|p ]) p E |Y | p−1 instead of the inequality 
1/2 
1/2   E |Y |2 E [|XY |] ≤ E |X|2

for any FT -measurable random variables X and Y . So the rest proof is omitted here. By virtue of Proposition 3.5, the following corollary follows immediately. Corollary 3.6. Let p > 1, ξ i ∈ Lp (FT ), V·i ∈ V p , Li· , U·i ∈ S and (Y·i , Z·i , K·i , Ai· ) ∈ S p ×Mp ×V +,p ×V +,p

be a solution of DRBSDE (ξ i , g i + dV i , Li , U i ) for i = 1, 2. If ξ 1 ≤ ξ 2 , dV 1 ≤ dV 2 , L1· ≤ L2· , U·1 ≤ U·2 ,

g 1 or g 2 satisfies (H1) and (H2)(ii), and for each (y, z) ∈ R × Rd ,

dP × dt − a.e., g 1 (t, y, z) ≤ g 2 (t, y, z), then Yt1 ≤ Yt2 for each t ∈ [0, T ]. Remark 3.7. We note that in the proof of Proposition 3.5 the following two assumptions are not utilized: Z T Z T 2 2 2 (Ys − Ls )dKs = 0 and (Us1 − Ys1 )dA1s = 0. 0

0

In addition, it follows from Remark 2.3 that Proposition 3.5 and Corollary 3.6 improves Proposition 2.9 if the comparison of dK i and dAi is not considered. Theorem 3.8 (Uniqueness). Let p > 1, ξ ∈ Lp (FT ), V· ∈ V p , L· , U· ∈ S and the generator g satisfy

assumptions (H1) and (H2)(ii). Then DRBSDE (ξ, g +dV, L, U ) admits at most one solution in S p ×Mp ,

i.e, if both (Y· , Z· , K· , A· ) and (Y·′ , Z·′ , K·′ , A′· ) are solutions of DRBSDE (ξ, g + dV, L, U ) in S p × Mp , then dP × dt − a.e., Y· = Y·′ , Z· = Z·′ K· = K·′ and A· = A′· .

Proof. Firstly, it follows from Corollary 3.6 that Yt = Yt′ for each t ∈ [0, T ]. Furthermore, by Itˆ o’s

formula we know that dP × dt − a.e., Z· = Z·′ , and then Kt − At = Kt′ − A′t for each t ∈ [0, T ]. Finally, the conclusion of K· = K·′ and A· = A′· follows from the Ham-Bananch Composition of Sign Measure. 16

4. Existence of the solutions: penalization method In this section, we will prove the existence of Lp solutions for DRBSDEs under the assumptions of (H1), (H2) (resp. (H2’)), (H3) and (H4) by showing the convergence of the sequence of Lp solutions for the penalized RBSDEs with one continuous barrier and the penalized BSDEs. Before that, let us first show that under conditions of (H1), (H2) (ii) (resp. (H2’)(ii)), (H3)(ii) and (H4)(i), (H4)(ii) is necessary to ensure the existence of Lp solutions for DRBSDEs, which is one of our main results. Theorem 4.1 (Necessary of (H4)(ii)). Assume that p > 1, V· ∈ V p , the generator g satisfies (H1), (H2)(ii) (resp. (H2’)(ii)) and (H3)(ii), and that (H4)(i) holds for L· , U· and ξ. If DRBSDE (ξ, g + dV, L, U ) admits a solution (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p , then g(·, Y· , 0) ∈ Hp . That is to say, (H4)(ii) holds true. Proof. We only prove the case of (H2’)(ii). The case of (H2’)(ii) can be proved in a same way. In fact, ¯ := λ (see also ¯ := µ + A and λ it is easy to verify that g satisfies (AA) with f¯· := |g(·, 0, 0)| + f· + A, µ Remark 2.7). It then follows from Lemma 3.1 that " Z !p # T E |g(t, Yt , Zt )|dt < +∞. 0

Then, by (H2’)(ii) together with H¨ older’s inequality we can deduce that " Z !p # " Z !p # " Z T T p p E |g(t, Yt , 0)|dt ≤ 4 E |g(t, Yt , Zt )|dt +4 E 0

0

+(4µT )p E
1, V· ∈ V p , (i) of (H4) holds for L· , U· and ξ, and the generator g satisfies (HH) with f· , ψ· (r) and λ. We have ¯ (i) For each n ≥ 1, let (Y·n , Z·n , An· ) ∈ S p × Mp × V +,p be a solution of RBSDE (ξ, g¯n + dV, U ) with g¯n (t, y, z) := g(t, y, z) + n(y − Lt )− , i.e.,  Z T Z T Z T Z T Z T  n n n n n  Y = ξ + g(s, Y , Z )ds + dV + dK − dA − Zsn dBs , t ∈ [0, T ],  s t s s t s   t t t t t  Z t  Ktn := n (Ysn − Ls )− ds, t ∈ [0, T ],   0 Z T      Ytn ≤ Ut , t ∈ [0, T ] and (Ut − Ytn )dAnt = 0. 0

If

Y·n

n

(4. 2)

n+1

increases in n, dA ≤ dA

1

, and there exists a random variable η ∈ L (FT ) such that for

each n ≥ 1 and t ∈ [0, T ],  ! p2 Z T E  sup |Ysn |p + |Zsn |2 ds + |KTn − Ktn |p + |AnT − Ant |p + s∈[t,T ]

t

≤ E [ η| Ft ] ,

Z

t

T

!p  |g(s, Ysn , Zsn )|ds Ft 

(4. 3)

then there exists a quadruple (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p which solves DRBSDE (ξ, g + dV, L, U ), lim (kY·n − Y· kS p + kZ·n − Z· kMp + kAn· − A· kS p ) = 0,

n→∞

n

and there exists a subsequence {K· j } of {K·n } such that n

lim sup |Kt j − Kt | = 0.

j→∞ t∈[0,T ]

(ii) For each n ≥ 1, let (Y·n , Z·n , K·n ) ∈ S p × Mp × V +,p be a solution of RBSDE (ξ, g n + dV, L) with gn (t, y, z) := g(t, y, z) − n(y − Ut )+ , i.e.,  Z T Z T Z T Z T Z T  n n n n n  Yt = ξ + g(s, Ys , Zs )ds + dVs + dKt − dAs − Zsn dBs , t ∈ [0, T ],    t t t t t  Z t  Ant := n (Ysn − Us )+ ds, t ∈ [0, T ],   0 Z T      Lt ≤ Ytn , t ∈ [0, T ] and (Ytn − Lt )dKtn = 0. 0

If Y·n decreases in n, dK n ≤ dK n+1 , and there exists a random variable η ∈ L1 (FT ) such that (4. 3) holds true for each n ≥ 1 and t ∈ [0, T ], then there exists a quadruple (Y· , Z· , K· , A· ) ∈

S p × Mp × V +,p × V +,p which solves DRBSDE (ξ, g + dV, L, U ),

lim (kY·n − Y· kS p + kZ·n − Z· kMp + kK·n − K· kS p ) = 0,

n→∞

n

and there exists a subsequence {A· j } of {An· } such that n

lim sup |At j − At | = 0.

j→∞ t∈[0,T ]

18

Proof. We only prove the claim (i). The claim (ii) can be proved in the same way. Now we assume that all the assumptions in (i) are satisfied. Since Y·n increases in n, there exists a progressively measurable process Y· such that Ytn ↑ Yt for each t ∈ [0, T ]. By Fatou’s lemma and (4. 3) we can deduce that " # " # sup |Yt |

E

p

=

E

t∈[0,T ]

≤

sup lim inf

t∈[0,T ] n→∞

lim inf E n→∞

"

sup

t∈[0,T ]

|Ytn |p

|Ytn |p

#

"

≤ E lim inf sup

n→∞ t∈[0,T ]

|Ytn |p

#

(4. 4)

≤ E [η] < +∞.

Since dAn ≤ dAn+1 for each n ≥ 1, we know that there exists a progressively measurable and increasing

process (At )t∈[0,T ] with A0 = 0 such that Ant ↑ At for each t ∈ [0, T ], and for each j ≥ n ≥ 1, Ajt − Ant ≤ AjT − AnT , t ∈ [0, T ].

In the above inequality, first letting j → ∞, and then taking the superume with respect to t in [0, T ], finally letting n → ∞, we can obtain that sup |Ant − At | → 0, as n → ∞,

(4. 5)

t∈[0,T ]

which means that A· ∈ V + . On the other hand, note that |An· | ≤ |A· | for each n ≥ 1 and A· ∈ V +,p by a similar proof to (4. 4). From (4. 5) and Lebesgue’s dominated convergence theorem it follows that lim kAn· − A· kS p = 0.

(4. 6)

n→∞

Furthermore, by (4. 3) we can also get that 1

sup |Ytn | ≤ (E [ η| Ft ]) p , t ∈ [0, T ]

(4. 7)

n≥1

and 

sup E 

n≥1

Z

T

0

|Ztn |2 dt

! p2

+ |KTn |p + |AnT |p +

Z

0

T

|g(t, Ytn , Ztn )|dt

!p 

 ≤ E [η] < +∞.

(4. 8)

The rest proof is divided into 6 steps. Step 1. We show that Y· is a c` adl` ag process. For each integer l, q ≥ 1, introduce the following two stopping times: τl

:=

σl,q

:=

  Z t 1 p inf t ≥ 0 : (E [ η| Ft ]) + fs ds + Lt ≥ l ∧ T ; 0   Z t inf t ≥ 0 : ψs (l)ds ≥ q ∧ τl . 0

Then we have, τl → T as l → ∞, σl,q → τl as q → ∞ for each l ≥ 1,

P ({ω : ∃l0 (ω) ≥ 1, ∀l ≥ l0 (ω), τl (ω) = T }) = 1 and P ({ω : ∃l0 (ω), q0 (ω) ≥ 1, ∀l ≥ l0 (ω), ∀q ≥ q0 (ω), σl,q (ω) = T }) = 1. 19

(4. 9)

Now, let us arbitrarily fix a pair of l, q ≥ 1. Since g satisfies (HH) with f· , ψ· (r) and λ, and (4. 7) is satisfied, it follows from the definitions of τl and σl,q that dP × dt − a.e., for each n ≥ 1, |hn;l,q | ≤ 1·≤τl f· + 1·≤σl,q ψ· (l) + λ|Z·n | · with hn;l,q := 1·≤σl,q g(·, Y·n , Z·n ), · "Z T

E

0

#

1t≤τl ft dt ≤ l and E

"Z

(4. 10)

#

T

1t≤σl,q ψt (l)dt ≤ q,

0

(4. 11) n ;l,q ∞ }j=1

from which together with (4. 8), we can deduce that there exists a subsequence {h· j sequence

{hn;l,q }∞ · n=1

hl,q ·

which converges weakly to a process

1

of the

in H . Now, take any bounded linear

functional Φ(·) defined on L1 (FT ). Then there exists a constant b > 0 such that for each h· ∈ H1 and every stopping time τ¯ valued in [0, T ], we have Z τ¯

Z

Φ( hs ds) ≤ b

0

τ¯ 0

hs ds

≤ b h H1 .

L1

R τ¯ Hence, for each stopping time τ¯ valued in [0, T ], Φ( 0 · ds) is a bounded linear functional defined on H1 , which means that

lim Φ

j→∞

Z

τ¯

hns j ;l,q ds

0



=Φ

Z

0

τ¯

 hl,q ds . s

As a result, for every stopping time τ with 0 ≤ τ ≤ σl,q , as j → ∞, Z τ Z τ Z τ 1 hns j ;l,q ds → hl,q g(s, Ysnj , Zsnj )ds = s ds weakly in L (FT ). 0

0

(4. 12)

0

Furthermore, it follows from (4. 8) and Lemma 4.4 of Klimsiak [39] that there exists a process Z· ∈ Mp

and a subsequence of the sequence {nj }∞ ¯ j=1 , still denoted by itself, such that for every stoppingtime τ valued in [0, T ], as j → ∞, Z Z τ¯ nj Zs dBs → 0

τ¯

Zs dBs weakly in Lp (FT ) and then in L1 (FT ).

(4. 13)

0

In the sequel, we define Ktl,q := Y0 − Yt −

Z

t

0

hl,q s ds −

Z

0

t

dVs −

Z

t

dAs +

0

Z

t

Zs dBs , t ∈ [0, T ].

0

Then, in view of (4. 6), (4. 12), (4. 13) and the fact that for each stopping time τ¯ valued in [0, T ], Yτ¯n ↑ Yτ¯ in L1 (FT ), we can deduce that for every stopping time τ with 0 ≤ τ ≤ σl,q , the sequence Z τ Z τ Z τ Z τ n Zsnj dBs dVs − dAns j + g(s, Ysnj , Zsnj )ds − Kτnj = Y0 j − Yτnj − 0

0

0

0

converges weakly to Kτl,q in L1 (FT ) as j → ∞. Thus, since K·n ∈ V +,p for each n ≥ 1, we know that Kτl,q ≤ Kτl,q 1 ∧σl,q 2 ∧σl,q for any stopping times τ1 ≤ τ2 valued in [0, T ]. Furthermore, in view of the definition of K·l,q together with the facts that V· ∈ V p , A· ∈ V +,p , Y·n ↑ Y· and Y·n ∈ S p for each n ≥ 1, it is not hard to check that

K·l,q is a optional process with P− a.s. upper semi-continuous paths. Thus, Lemma A.3 in Bayraktar and

l,q Yao [4] yields that K·∧σ is a nondecreasing process, and then it has P − a.s. right lower semi-continuous l,q

20

l,q paths. Hence, K·∧σ is c` adl` ag and so is Y·∧σl,q from the definition of K·l,q . Finally, it follows from (4. 9) l,q

that Y· is also a c` adl` ag process. Step 2. We show that Yt ≥ Lt for each t ∈ [0, T ] and as n → ∞, sup (Ytn − Lt )− → 0.

(4. 14)

t∈[0,T ]

In fact, it follows from (4. 8) and the definition of K·n that for each n ≥ 1, " Z !p # T E[η] n − E (Yt − Lt ) dt ≤ p . n 0 Hence, by Fatou’s lemma and H¨ older’s inequality, "Z # "Z T − E (Yt − Lt ) dt ≤ lim inf E n→∞

0

T

0

(Ytn

−

#

1

(E[η]) p = 0, n→∞ n

− Lt ) dt ≤ lim

which implies that E

"Z

T

0

−

#

(Yt − Lt ) dt = 0.

Since Y· − L· is a c` adl` ag process, (Yt − Lt )− = 0 and hence Yt ≥ Lt for each t ∈ [0, T ). Moreover,

YT = YTn = ξ ≥ LT . Hence

(Ytn − Lt )− ↓ 0

for each t ∈ [0, T ] and by Dini’s theorem, (4. 14) follows. Step 3. We show the convergence of the sequence {Y·n }. Let τl and σl,q be the sequences of stopping times defined in Step 1. For each n, m ≥ 1, observe that (Y¯· , Z¯· , V¯· ) :=

(Y·n − Y·m , Z·n − Z·m , Z · (g(s, Ysn , Zsn ) − g(s, Ysm , Zsm )) ds + (K·n − K·m ) − (An· − Am · ))

(4. 15)

0

satisfies equation (4. 1). It then follows from (ii) of Lemma 4.2 with p = 2, t = 0 and τ = σl,q that there exists a constant C > 0 such that for each n, m, l, q ≥ 1, # " E

n m sup |Yt∧σ − Yt∧σ |2 l,q l,q 

t∈[0,T ]

≤

CE |Yσnl,q

−

Yσml,q |2

+ sup

+

t∈[0,T ] Z σl,q 0

Z

+ sup

t∈[0,T ]

σl,q

t∧σl,q

|Ytn

−

Z

σl,q

t∧σl,q

(Ysn

−

Ysm ) (dKsn

−

!+

dKsm )

!+

(4. 16)

n (Ysn − Ysm ) (dAm s − dAs )

Ytm | |g(t, Ytn , Ztn )

−

g(t, Ytm , Ztm )| dt



.

Furthermore, by virtue of the definition of K·n and An· with (4. 2) we know that for each t ∈ [0, T ], Z σl,q (Ysn − Ysm ) (dKsn − dKsm ) t∧σ l,q Z σl,q Z σl,q [(Ysn − Ls ) − (Ysm − Ls )] dKsm [(Ysn − Ls ) − (Ysm − Ls )] dKsn − = t∧σ t∧σ l,q Z σl,q Z σl,ql,q (4. 17) (Ysn − Ls )− dKsm (Ysm − Ls )− dKsn + ≤ t∧σl,q

≤

t∧σl,q

m n sup (Yt∧σ − Lt∧σl,q )− |KTn | + sup (Yt∧σ − Lt∧σl,q )− |KTm |. l,q l,q

t∈[0,T ]

t∈[0,T ]

21

and Z

σl,q

t∧σl,q

n (Ysn − Ysm ) (dAm s − dAs ) =

=

Z

σl,q

n [(Us − Ysm ) − (Us − Ysn )] (dAm s − dAs ) t∧σl,q Z σl,q Z σl,q (Us − Ysn )dAm (Us − Ysm )dAns − − s t∧σl,q

t∧σl,q

≤

(4. 18)

0.

Combining (4. 10), (4. 16), (4. 17) and (4. 18) together with H¨ older’s inequality yields that # " E

n m sup |Yt∧σ − Yt∧σ |2 l,q l,q " Z

t∈[0,T ]

≤

CE +C

+C

|Yσnl,q E

E

" "

−

Yσml,q |2

T

+2

0

|Ytn

−

Ytm |

p m − p−1 − L ) sup (Yt∧σ t∧σ l,q l,q

t∈[0,T ]

p n − p−1 − L ) sup (Yt∧σ t∧σ l,q l,q

t∈[0,T ]

+2Cλ E

"Z

0

σl,q

|Ytn − Ytm |2 dt


1t≤τl ft + 1t≤σl,q ψt (l) dt

#! p−1 p

(E [|KTn |p ]) p

#! p−1 p

(E [|KTm |p ]) p

#

1

#! p−1 p  2(p−1) p

(4. 19)

1

 

Z

× E 

0

T

! p2  p1 2 (|Ztn | + |Ztm |) dt  .

Thus, note that Ytn ↑ Yt for each t ∈ [0, T ]. In view of the definitions of τl and σl,q , (4. 7), (4. 8), (4. 11) and (4. 14), by (4. 19) and Lebesgue’s dominated convergence theorem we can deduce that for each l, q ≥ 1, as n, m → ∞, E

"

#

n m sup |Yt∧σ − Yt∧σ |2 → 0, l,q l,q

t∈[0,T ]

which implies that for each l, q ≥ 1, as n, m → ∞, n m | → 0 in probability P. sup |Yt∧σ − Yt∧σ l,q l,q

t∈[0,T ]

And, by (4. 9) and the fact that Y·n increases in n we know that sup |Ytn − Yt | → 0, as n → ∞.

(4. 20)

t∈[0,T ]

So, Y· is a continuous process. Finally, note that |Y·n | ≤ |Y·1 | + |Y· | for each n ≥ 1 and that (4. 4) is satisfied. From (4. 20) and Lebesgue’s dominated convergence theorem it follows that # " lim kY·n − Y· kpS p = lim E

n→∞

n→∞

sup |Ytn − Yt |p = 0.

(4. 21)

t∈[0,T ]

Step 4. We show the convergence of the sequence {Z·n }. Note that (4. 15) solves (4. 1). It follows

from (i) of Lemma 4.2 with t = 0 and τ = T that there exists a constant C ′ > 0 such that for each

22

m, n ≥ 1, 

E ≤

Z

T

0



|Ztn − Ztm |2 dt

! p2  



Z

 C ′ E  sup |Ytn − Ytm |p + sup  t∈[0,T ]

t∈[0,T ]





 +C ′ E  sup  t∈[0,T ]



+C ′ E 

Z

T

0

t

!+  p2

(Ysn − Ysm ) (dKsn − dKsm )

 !+  p2 n   (Ysn − Ysm ) (dAm  s − dAs )

T

Z

T

t



  

! p2  |Ytn − Ytm | |g(t, Ytn , Ztn ) − g(t, Ytm , Ztm )| dt  .

Then, it follows from H¨ older’s inequality together with (4. 18) that for each m, n ≥ 1,  ! p2  Z T E |Ztn − Ztm |2 dt  0

≤

C ′E

"

#

sup |Ytn − Ytm |p + C ′

t∈[0,T ]

1 2

+ (E [|KTm |p ]) +

" Z

E

0

T

E

"

sup |Ytn − Ytm |p

t∈[0,T ]

#! 21   

(|g(t, Ytn , Ztn )| + |g(t, Ytm , Ztm )|) dt

!p #! 21  

from which together with (4. 8), (4. 21) and (4. 13) yields that  ! p2  Z T lim kZ·n − Z· kp p = lim E  |Ztn − Zt |2 dt  = 0. M

n→∞

n→∞

1

(E [|KTn |p ]) 2 ,

 (4. 22)

0

Step 5. We show the convergence of the sequence {K·n }. Let τl and σl,q be the sequences of stopping times defined in Step 1. Since g satisfies (HH), by (4. 10), (4. 11), (4. 8), (4. 20) and (4. 22) we can deduce that there exists a subsequence {nj } of {n} such that for each l, q ≥ 1, Z σl,q n n lim |g(t, Yt j , Zt j ) − g(t, Yt , Zt )|dt = 0. j→∞

0

Then, in view of (4. 9), we have Z t Z t nj nj lim sup g(t, Yt , Zt )dt = 0. g(t, Yt , Zt )dt − j→∞ t∈[0,T ]

(4. 23)

0

0

Combining (4. 6), (4. 20), (4. 22) and (4. 23) yields that P − a.s., for each t ∈ [0, T ], Z t Z t Z t n n n n Zsnj dBs dVs − At j + g(s, Ysnj , Zsnj )ds − Kt j = Y0 j − Yt j − tends to Kt := Y0 − Yt −

Z

t

0

0

0

0

g(s, Ys , Zs )ds −

as j → ∞ and that

n

Z

t 0

dVs − At +

lim sup |Kt j − Kt | = 0.

j→∞ t∈[0,T ]

Hence, K· is a continuous process. 23

Z

t

Zs dBs

0

(4. 24)

Step 6. We show that K· ∈ V +,p and (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p is a solution of RBSDE (ξ, g + dV, L, U ). In fact, by Fatou’s lemma with (4. 24) and (4. 8) we get that # " # " " sup |Kt |

E

p

=

lim sup

E

j→∞ t∈[0,T ]

t∈[0,T ]

≤

n |Kt j |p

≤ lim inf E

sup

j→∞

t∈[0,T ]

n |Kt j |p

#

 n  sup E |KT j |p ≤ E [η] < +∞. j≥1

Hence, K· ∈ V +,p and (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p solves Yt = ξ +

Z

T

g(s, Ys , Zs )ds +

t

Z

T

dVs +

t

Z

T

dKs −

t

T

Z

dAs −

t

Z

T

Zs dBs , t ∈ [0, T ].

t

By Step 2 we know that Yt ≥ Lt for each t ∈ [0, T ], and then Z

T

(Yt − Lt )dKt ≥ 0.

0

On the other hand, in view of (4. 20) and (4. 24), it follows from the definition of K·n that Z

T

(Yt − Lt )dKt = lim

j→∞

0

Consequently, we have Z Furthermore, noticing that Y·n ≤ U· and

Z

T

nj

(Yt

0

n

− Lt )dKt j ≤ 0.

T 0

RT 0

(Yt − Lt )dKt = 0. (Ut − Ytn ) dAnt = 0 for each n ≥ 1 , from (4. 21) and (4. 6)

we can deduce that Yt ≤ Ut for each n ≥ 1, and Z T Z (Ut − Yt ) dAt = lim n→∞

0

T 0

(Ut − Ytn ) dAnt = 0.

Finally, let us show that dK⊥dA. In fact, for each n ≥ 1, we can define the following progressively measurable set Dn := {(ω, t) ⊂ Ω × [0, T ] : Ytn (ω) ≥ Lt (ω)}. Then, from the definition of K·n we know that for each n ≥ 1, # "Z T

E

0

and, in view of

RT

(Ut − Ytn )dAnt = 0, "Z # "Z T n 1Dnc dAt = E E

1Dn dKtn = 0,

0

0

0

T

1{Ytn 1, ξ ∈ Lp (FT ), V· ∈ V p , the generator g satisfies assumptions (H1), (H2) and (H3), and assumption (H4) holds for L· , U· , ξ and X· . Then, DRBSDE (ξ, g + dV, L, U ) admits a unique solution (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p . Moreover, ¯ (i) For each n ≥ 1, let (Y n· , Z n· , An· ) ∈ S p × Mp × V +,p be the unique solution of RBSDE (ξ, g¯n + dV, U ) with g¯n (t, y, z) := g(t, y, z) + n(y − Lt )− , i.e., (3. 3). Then,

lim (kY n· − Y· kS p + kZ n· − Z· kMp + kK n· − K· kS p + kAn· − A· kS p ) = 0.

n→∞

(4. 25)

¯ ·n ) ∈ S p × Mp × V +,p be the unique solution of RBSDE (ξ, g + dV, L) (ii) For each n ≥ 1, let (Y¯·n , Z¯·n , K n with gn (t, y, z) := g(t, y, z) − n(y − Ut )+ , i.e., (3. 5). Then,


¯ ·n − K· kS p + kA¯n· − A· kS p = 0. lim kY¯·n − Y· kS p + kZ¯·n − Z· kMp + kK

n→∞

(4. 26)

(iii) For each n ≥ 1, let (Y·n , Z·n ) ∈ S p × Mp be the unique solution of BSDE (ξ, gn + dV ) with gn (t, y, z) := g(t, y, z) + n(y − Lt )− − n(y − Ut )+ , i.e., (3. 7). Then,

lim (kY·n − Y· kS p + kZ·n − Z· kMp + kK·n − K· kS p + kAn· − A· kS p ) = 0.

n→∞

(4. 27)

Proof. The uniqueness part follows from Theorem 3.8. With regard to (i), combining Proposition 3.3 and Proposition 4.3, in view of (i) in Remark 2.7, we can deduce that there exists a quadruple (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p which solves DRBSDE (ξ, g + dV, L, U ),

lim (kY n· − Y· kS p + kZ n· − Z· kMp + kAn· − A· kS p ) = 0,

n→∞

(4. 28)

and there exists a subsequence {K n· j } of {K n· } such that n

lim sup |K t j − Kt | = 0.

j→∞ t∈[0,T ]

Furthermore, using a similar argument to that in the proof of Theorem 4 of Fan [16], we can obtain that

Z ·

Z ·

n n

lim g(s, Y s , Z s )ds − g(s, Ys , Zs )ds (4. 29)

= 0. n→∞

0

0

Sp

Then, (4. 25) follows from (4. 28) and (4. 29). And, (4. 26) can be proved in a same way.

Y

n t

In the sequel, we prove (iii). Firstly, by Proposition 3.3 we know that for each n ≥ 1 and t ∈ [0, T ], ¯ n ≤ dK n ≤ dK n and dAn ≤ dAn ≤ dA¯n . Then, (4. 25) and (4. 26) yield that ≤ Ytn ≤ Y¯tn , dK lim (kY·n − Y· kS p + kK·n − K· kS p + kAn· − A· kS p ) = 0.

n→∞

(4. 30)

Finally, in view of (3. 8) and (4. 30), a similar argument to that in the fourth step of proof of Proposip tion 4.3 yields that {(Z·n )}∞ n=1 is a Cauchy sequence in M , and then (4. 27) holds true.

25

Remark 4.6. Theorem 4.5 improves (v) of Theorem 6.5 in Klimsiak [40], where the generator g needs to satisfy the stronger assumptions (H1s) and (H2s) than (H1) and (H2) by Remark 2.6. Corollary 4.7. Let p > 1, V·1 , V·2 , ∈ V p and both g 1 and g 2 satisfy assumptions (H1), (H2) and (H3). For i = 1, 2, assume that (H4) holds for ξ i , Li· , U·i and X·i associated with g i , and that (Y·i , Z·i , K·i , Ai· ) ∈

S p × Mp × V +,p × V +,p is the unique solution of DRBSDE (ξ i , g i + dV i , Li , U i ). If ξ 1 ≤ ξ 2 , dV 1 ≤ dV 2 , L1· = L2· , U·1 = U·2 and

dP × dt − a.e., g 1 (t, y, z) ≤ g 2 (t, y, z) for each (y, z) ∈ R × Rd , then dK 1 ≥ dK 2 and dA1 ≤ dA2 . Proof. For each n ≥ 1 and i = 1, 2, let (Y·i,n , Z·i,n ) ∈ S p ×Mp be the unique solution of BSDE (ξ i , gni +dV i )

with gni (t, y, z) := g i (t, y, z) + n(y − Lit )− − n(y − Uti )+ . In view of the assumptions of Corollary 4.7,

it follows from (i) of Proposition 2.9 that for each n ≥ 1, Y·1,n ≤ Y·2,n , and then for each progressively measurable set D ⊂ Ω × [0, T ] and each n ≥ 1, we have # "Z # "Z "Z T T 1,n 1,n E 1D dKt := nE 1D (Yt − L1t )− dt ≥ nE 0

0

and "Z E

0

T

1

1,n D dAt

#

:= nE

"Z

0

T

T

1

2,n D (Yt

0

1

1,n D (Yt

−

Ut1 )+ dt

#

≤ nE

"Z

T

0

1

2,n D (Yt

#

"Z

T

#

"Z

T

− L2t )− dt =: E

−

Ut2 )+ dt

=: E

0

0

1

#

1

#

2,n D dKt

2,n D dAt

.

Since kK·1,n − K·1 kS p + kK·2,n − K·2 kS p → 0 and − A2· kS p → 0 kA1,n − A1· kS p + kA2,n · · as n → ∞ by (iii) of Theorem 4.5, it follows that # "Z # "Z "Z T T 1 2 1D dKt ≥ E 1D dKt and E E 0

0

0

T

1

1 D dAt

#

≤E

"Z

0

T

1

2 D dAt

#

,

which is the desired result. At the end of this section, we put forward and prove a general existence result of Lp solutions for DRBSDEs under the assumptions of (H1), (H2’), (H3) and (H4). Theorem 4.8. Assume that p > 1, ξ ∈ Lp (FT ), V· ∈ V p , the generator g satisfies assumptions (H1), (H2’) and (H3), and assumption (H4) holds for L· , U· , ξ and X· . (i) For each n ≥ 1, let (Y n· , Z n· , An· ) ∈ S p × Mp × V +,p be the minimal solution (resp. the maximal ¯ solution) of RBSDE (ξ, g¯n + dV, U ) with g¯n (t, y, z) := g(t, y, z) + n(y − Lt )− , i.e., (3. 3). Then,

DRBSDE (ξ, g + dV, L, U ) admits a minimal solution (resp. a solution) (Y · , Z · , K · , A· ) ∈ S p × Mp × V +,p × V +,p such that

lim (kY n· − Y · kS p + kZ n· − Z · kMp + kK n· − K · kS p + kAn· − A· kS p ) = 0.

n→∞

26

¯ n ) ∈ S p × Mp × V +,p be be the maximal solution (resp. the minimal (ii) For each n ≥ 1, let (Y¯·n , Z¯·n , K · solution) of RBSDE (ξ, gn + dV, L) with gn (t, y, z) := g(t, y, z) − n(y − Ut )+ , i.e., (3. 5). Then,

¯ · , A¯· ) ∈ S p × Mp × DRBSDE (ξ, g + dV, L, U ) admits a maximal solution (resp. a solution) (Y¯· , Z¯· , K

V +,p × V +,p such that


¯ · kS p + kA¯n· − A¯· kS p = 0. ¯ ·n − K lim kY¯·n − Y¯· kS p + kZ¯·n − Z¯· kMp + kK

n→∞

Proof. We only prove (i), and (ii) can be proved in a same way. In view of Remark 2.11 and Remark 3.4, using a similar argument to that in the proof of Theorem 4.5 we can prove that all the conclusions in (i) of Theorem 4.8 hold true expect the minimal property of the solution (Y · , Z · , K · , A· ) ∈ S p × Mp × V +,p × V +,p of DRBSDE (ξ, g + dV, L, U ) when (Y n· , Z n· , An· ) ∈

¯ S p × Mp × V +,p is the minimal solution of RBSDE (ξ, g¯n + dV, U ) for each n ≥ 1. Now, we will show this property. Indeed, for any solution (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p of DRBSDE (ξ, g + dV, L, U ), it is

¯ not difficult to check that (Y· , Z· , A· ) ∈ S p × Mp × V +,p is a solution of RBSDE (ξ, g¯n + dV¯ , U ) with

V¯· := V· + K· for each n ≥ 1. Thus, in view of the assumption that (Y n· , Z n· , An· ) ∈ S p × Mp × V +,p

¯ is the minimal solution of RBSDE (ξ, g¯n + dV, U ) for each n ≥ 1, (iii) of Proposition 2.9 together with Remark 2.11 yields that for each n ≥ 1, Y nt ≤ Yt , t ∈ [0, T ]. Furthermore, since lim kY n· − Y · kS p = 0, we know that for each n ≥ 1, n→∞

Y t ≤ Yt , t ∈ [0, T ], which is the desired result.

5. Approximation of the solutions In this section, we will establish a general approximation result for the Lp (p > 1) solutions of DRBSDEs under some elementary conditions, and consider the DRBSDEs where the generator g may be discontinuous and have a general growth in y. Proposition 5.1 (Approximation). Assume that for each n ≥ 1, the generator gn satisfies (ii) of (HH)

with the same f· , ψ· (r) and λ. Let p > 1, V· ∈ V p and (i) of (H4) be satisfied for L· , U· and ξ. For n ≥ 1, assume that (Y·n , Z·n , K·n , An· ) ∈ S p × Mp × V +,p × V +,p is a solution of DRBSDE (ξ, gn + dV, L, U ). If Y·n ≤ Y·n+1 , dAn ≤ dAn+1 and dK n ≥ dK n+1 for each n ≥ 1, gn tends locally uniformly in (y, z) to the generator g as n → ∞ in the following sense: d For any sequence {(y n , z n )}∞ such that n=1 in R × R

if y n ≤ y for each n ≥ 1, then

lim (|y n − y| + |z n − z|) = 0,

n→∞

lim gn (t, y n , z n ) = g(t, y, z) dP × dt − a.e.,

n→∞

27

(5. 1)

and there exists a η ∈ L1 (FT ) such that for each n ≥ 1 and t ∈ [0, T ],  ! p2 Z T Z n p n 2 E  sup |Ys | + |Zs | ds + |KTn − Ktn |p + |AnT − Ant |p + s∈[t,T ]

t

T

t

≤ E [ η| Ft ] ,

!p  |gn (s, Ysn , Zsn )|ds Ft 

(5. 2)

then there exists a triple (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p which solves DRBSDE (ξ, g + dV, L, U ) such that lim (kY·n − Y· kS p + kZ·n − Z· kMp + kK·n − K· kS p + kAn· − A· kS p ) = 0.

n→∞

Proof. Since Y·n increases in n, there exists a process Y· such that Ytn ↑ Yt for each t ∈ [0, T ]. In the same way as in the proof of Proposition 4.3, by Fatou’s lemma together with (5. 2) we can deduce that # " E

sup |Yt |p < +∞,

(5. 3)

t∈[0,T ]

1

sup |Ytn | ≤ (E [ η| Ft ]) p , t ∈ [0, T ]

(5. 4)

n≥1

and 

sup E 

n≥1

Z

T

0

|Ztn |2 dt

! p2

+ |KTn |p + |AnT |p +

Z

0

T

!p  |gn (t, Ytn , Ztn )|dt  ≤ E[η] < +∞.

(5. 5)

Furthermore, since dAn ≤ dAn+1 and dK n ≥ dK n+1 for each n ≥ 1, a same argument as that in proving (4. 6) yields that there exist two processes K· and A· in S p such that

lim (kK·n − K· kS p + kAn· − A· kS p ) = 0.

(5. 6)

n→∞

For each positive integer l, q ≥ 1, as in the proof of Proposition 4.3, we introduce the following two stopping times: τl

:=

σl,q

:=

  Z t 1 inf t ≥ 0 : (E [ η| Ft ]) p + fs ds ≥ l ∧ T ; 0   Z t ψs (l)ds ≥ q ∧ τl . inf t ≥ 0 : 0

Then we have P ({ω : ∃l0 (ω), q0 (ω) ≥ 1, ∀l ≥ l0 (ω), ∀q ≥ q0 (ω), σl,q (ω) = T }) = 1.

(5. 7)

Furthermore, since all gn satisfy (HH) with the same f· , ψ· (r) and λ, and (5. 4) is satisfied, in view of the definitions of τl and σl,q , we know that dP × dt − a.e., for each l, q, n ≥ 1,

1t≤σl,q |gn (t, Ytn , Ztn)| ≤ 1t≤τl ft + 1t≤σl,q ψt (l) + λ|Ztn | with E

"Z

0

T

#

1t≤τl ft dt ≤ l and E

The rest proof is divided into 3 steps. 28

"Z

0

T

#

1t≤σl,q ψt (l)dt ≤ q.

(5. 8)

(5. 9)

Step 1. We show the convergence of the sequence {Y·n }. For each n, m ≥ 1, observe that (Y¯· , Z¯· , V¯· ) :=

(Y·n − Y·m , Z·n − Z·m , Z · n (gn (s, Ysn , Zsn ) − gm (s, Ysm , Zsm )) ds + (K·n − K·m ) + (Am · − A· ))

(5. 10)

0

satisfies equation (4. 1). It then follows from (ii) of Lemma 4.2 with p = 2, t = 0 and τ = σl,q that there exists a constant C > 0 such that for each n, m, l, q ≥ 1, " # E

n m sup |Yt∧σ − Yt∧σ |2 l,q l,q 

t∈[0,T ]

CE |Yσnl,q

≤

−

Yσml,q |2

t∈[0,T ] σl,q

Z

+

0

t∈[0,T ]

σl,q

Z

+ sup

+ sup

t∧σl,q

Z

σl,q

t∧σl,q

(Ysn

−

Ysm ) (dKsn

−

!+

dKsm )

(5. 11)

!+

n (Ysn − Ysm ) (dAm s − dAs )

 |Ytn − Ytm | |gn (t, Ytn , Ztn ) − gm (t, Ytm , Ztm )| dt .

RT Furthermore, note that Lt ≤ Ytn ≤ Ut for each t ∈ [0, T ] and n ≥ 1 and that 0 (Ytn − Lt )dKtn = RT (Ut − Ytn )dAnt = 0 for each n ≥ 1. It follows that for each t ∈ [0, T ] and l, q, m, n ≥ 1, 0 Z σl,q Z σl,q [(Ysn − Ls ) − (Ysm − Ls )] (dKsn − dKsm ) (Ysn − Ysm ) (dKsn − dKsm ) = t∧σl,q t∧σl,q Z σl,q Z σl,q (5. 12) (Ysm − Ls )dKsn (Ysn − Ls )dKsm − = − t∧σl,q

t∧σl,q

≤

0.

=

Z

and Z

σl,q

t∧σl,q

(Ysn

−

Ysm ) (dAm s

−

dAns )

=

σl,q

n [(Us − Ysm ) − (Us − Ysn )] (dAm s − dAs ) t∧σl,q Z σl,q Z σl,q m n (Us − Ysn )dAm (Us − Ys )dAs − − s

≤

(5. 13)

t∧σl,q

t∧σl,q

0.

Combining (5. 8), (5. 11), (5. 12) and (5. 13) together with H¨ older’s inequality yields that " # E

m n sup |Yt∧σ − Yt∧σ |2 l,q l,q " Z

t∈[0,T ]

≤

CE

|Yσnl,q

+2Cλ E

−

Yσml,q |2

"Z

0

σl,q

T

+2

0

|Ytn

−

|Ytn

−

Ytm |2 dt

Ytm |


1t≤τl ft + 1t≤σl,q ψt (l) dt

#! p  2(p−1)

p−1 p

 

× E 

Z

0

T

#

(5. 14)

!  p1 2 (|Ztn | + |Ztm |) dt  . p 2

Thus, note that Ytn ↑ Yt for each t ∈ [0, T ]. By the definitions of τl and σl,q , (5. 4), (5. 5) and (5. 9), it follows from (5. 14) and Lebesgue’s dominated convergence theorem that for each l, q ≥ 1, as n, m → ∞, # " E

n m sup |Yt∧σ − Yt∧σ |2 → 0, l,q l,q

t∈[0,T ]

which implies that for each l, q ≥ 1, as n, m → ∞, n m sup |Yt∧σ − Yt∧σ | → 0 in probability P. l,q l,q

t∈[0,T ]

29

And, by (5. 7) and the monotonicity of Y·n with respect to n we know that sup |Ytn − Yt | → 0, as n → ∞.

(5. 15)

t∈[0,T ]

So, Y· is a continuous process. Finally, note that |Y·n | ≤ |Y·1 | + |Y· | for each n ≥ 1 and that (5. 3) is satisfied. From (5. 15) and Lebesgue’s dominated convergence theorem it follows that # " lim kY·n − Y· kpS p = lim E

n→∞

sup |Ytn − Yt |p = 0.

n→∞

(5. 16)

t∈[0,T ]

Step 2. We show the convergence of the sequence {Z·n }. Note that (5. 10) solves (4. 1). It follows

from (i) of Lemma 4.2 with t = 0 and τ = T that there exists a constant C ′ > 0 such that for each m, n ≥ 1, 

Z

E ≤

T

0



! p2  |Ztn − Ztm |2 dt 



Z

 C ′ E  sup |Ytn − Ytm |p + sup  t∈[0,T ]



t∈[0,T ]



 +C ′ E  sup  ′



+C E 

t∈[0,T ]

Z

T

0

Z

T

t

|Ytn

−

T

t

 !+  p2  (Ysn − Ysm ) (dKsn − dKsm )   !+  p2

n (Ysn − Ysm ) (dAm s − dAs )

Ytm | |gn (t, Ytn , Ztn )

−



  

gm (t, Ytm , Ztm )| dt

! p2  .

older’s inequality that for each m, n ≥ 1, Then, in view of (5. 12) and (5. 13), it follows from H¨   #! 12 " # " ! p2 Z T n m p ′ ′ n m p n m 2 E |Zt − Zt | dt  ≤ C E sup |Yt − Yt | + C E sup |Yt − Yt | t∈[0,T ]

t∈[0,T ]

0

· E

" Z

T

0

(|gn (t, Ytn , Ztn )|

+

|gm (t, Ytm , Ztm )|) dt

!p #! 21

,

from which together with (5. 16) and (5. 5) yields that there exists a process Z· ∈ Mp such that  ! p2  Z T lim kZ·n − Z· kpMp = lim E  (5. 17) |Ztn − Zt |2 dt  = 0. n→∞

n→∞

0

Step 3. We show that (Y· , Z· , K· , A· ) ∈ S p ×Mp ×V +,p ×V +,p is a solution of DRBSDE (ξ, g+dV, L, U ).

Since Ytn ↑ Yt for each t ∈ [0, T ], by (5. 1), (5. 17), (5. 5), (5. 8) and (5. 9) we can deduce that there exists a subsequence {nj } of {n} such that for each l, q ≥ 1, Z σl,q n n lim |gnj (t, Yt j , Zt j ) − g(t, Yt , Zt )|dt = 0. j→∞

0

Then, in view of (5. 7), we have Z t Z t n n lim sup gnj (t, Yt j , Zt j )dt − g(t, Yt , Zt )dt = 0. j→∞ t∈[0,T ]

0

0

Combining (5. 6), (5. 15), (5. 17) and (5. 18) yields that Z T Z T Z T Z Yt = ξ + g(s, Ys , Zs )ds + dVs + dKs − t

t

t

30

t

T

dAs −

Z

t

T

Zs dBs , t ∈ [0, T ].

(5. 18)

Since Lt ≤ Ytn ≤ Ut , n ≥ 1 and Ytn ↑ Yt for each t ∈ [0, T ], we have Lt ≤ Yt ≤ Ut for each t ∈ [0, T ]. Furthermore, in view of (5. 16) and (5. 6), it follows that Z

T

(Yt − Lt )dKt = lim

n→∞

0

Z

0

T

(Ytn

−

Lt )dKtn

= 0 and

Z

T

0

(Ut − Yt )dKt = lim

n→∞

Z

T

0

(Ut − Ytn )dAnt = 0.

Finally, let us show that dK⊥dA. In fact, for each n ≥ 1, since dK n ⊥dAn , there exists a progressively measurable set Dn ⊂ Ω × [0, T ] such that "Z T

E

0

1

n Dn dKt

#

"Z

=E

0

T

1

n c dA Dn t

#

= 0.

Then, in view of (5. 6) and the fact that dK ≤ dK n for each n ≥ 1, # "Z # "Z "Z ∞ ∞ T T X X 1∪Dn dKt ≤ 1Dn dKt ≤ E E 0≤E 0

and 0≤E

"Z

0

T

0

n=1

#

1∩Dnc dAt = lim E m→∞

"Z

T

0

n=1

#

1∩Dnc dAm ≤ lim E t m→∞

T

1

n Dn dKt

0

"Z

0

T

#

=0

#

1Dmc dAm = 0. t

Hence, dK⊥dA. Proposition 5.1 is then proved. Remark 5.2. It is clear that the conclusion of Proposition 5.1 holds still true if we replace the expression that Y·n ≤ Y·n+1 , dAn ≤ dAn+1 and dK n ≥ dK n+1 with the expression that Y·n ≥ Y·n+1 , dAn ≥ dAn+1

and dK n ≤ dK n+1 , and replace the expression that y n ≤ y in (5. 1) with the expression that y n ≥ y.

Remark 5.3. From Remark 2.3, it is clear that Proposition 5.1 together with Remark 5.2 improves Proposition 4 in Fan [16] although some ideas of the proof of Proposition 5.1 are lent from there. We now consider the DRBSDEs where the generator g may be discontinuous and have a general growth in y. Let us first introduce the following assumptions introduced by Fan and Jiang [20]: (A1a) g is left-continuous and lower semi-continuous in y, and continuous in z, i.e., dP × dt − a.e., for each (y0 , z0 ) ∈ R1+d , we have

lim

g(ω, t, y, z) = g(ω, t, y0 , z0 )

lim inf

g(ω, t, y, z) ≥ g(ω, t, y0 , z0 );

(−∞,y0 ]×Rd ∋(y,z)→(y0 ,z0 )

and [y0 ,+∞)×Rd ∋(y,z)→(y0 ,z0 )

(A1b) g is right-continuous and upper semi-continuous in y, and continuous in z, i.e., dP × dt − a.e., for each (y0 , z0 ) ∈ R1+d , we have

lim

g(ω, t, y, z) = g(ω, t, y0 , z0 )

lim sup

g(ω, t, y, z) ≤ g(ω, t, y0 , z0 ).

[y0 ,+∞)×Rd ∋(y,z)→(y0 ,z0 )

and (−∞,y0 ]×Rd ∋(y,z)→(y0 ,z0 )

31

˜ ≥ 0 and a process f˜· ∈ Hp such (A2) g has a linear growth in (y, z), i.e., there exist two constants µ ˜, λ that dP × dt − a.e., for each (y, z) ∈ R1+d ,

˜ |g(ω, t, y, z)| ≤ f˜t (ω) + µ ˜|y| + λ|z|. Remark 5.4. It is clear that (HH)(i) ⇔ (A1a) + (A1b), and that (A2) ⇔ (H2’)(ii) + (H3s). Theorem 5.5. Assume that p > 1, V· ∈ V p , g 1 satisfies (H1), (H2’) and (H3), g 2 satisfies (A1a) (resp. (A1b)) and (A2), and that g = g 1 + g 2 . Assume further that (H4) holds true for L· , U· , ξ, X· and g

(or g 1 ). Then, DRBSDE (ξ, g + dV, L· , U· ) admits a minimal (resp. maximal) solution (Y· , Z· , K· , A· ) in S p × Mp × V +,p × V +,p . Proof. We only prove the case of the minimal solution. Another case can be proved in a similar way in view of Remark 5.2. Assume now that p > 1, V· ∈ V p , g 1 satisfies (H1) with ρ(·), (H2’) with f· , µ and λ,

˜ and that g = g 1 + g 2 . Assume further and (H3) with ϕ· (r), g 2 satisfies (A1a) and (A2) with f˜· , µ ˜ and λ,

that (H4) holds for L· , U· , ξ, X· and g 1 . In view of the assumptions of g 1 and g 2 together with the proof of Theorem in Fan and Jiang [20], it is not very hard to prove that for each n ≥ 1 and (y, z) ∈ R × Rd , the following function gn (ω, t, y, z) := gn1 (ω, t, y, z) + gn2 (ω, t, y, z) with gn1 (ω, t, y, z) := inf

u∈Rd

and gn2 (ω, t, y, z) :=

inf

(u,v)∈Rd



g 1 (ω, t, y, u) + (n + 2λ)|u − z|



h i ˜ − z| g 2 (ω, t, u, v) + (n + 2˜ µ)|u − y| + (n + 2λ)|v

is well defined and progressively measurable, dP × dt − a.e., gn increases in n and converges locally

uniformly in (y, z) to the generator g = g 1 + g 2 as n → ∞ in the sense of (5. 1), all gn1 satisfy (H1) with

the same ρ(·), (H2s) with n + 2λ, (H3) with the same ϕ· (r) + µr + 2f· , and dP × dt − a.e., ∀ n ≥ 1,

∀ (y, z) ∈ R × Rd ,

|gn1 (ω, t, y, z) − g 1 (ω, t, y, 0)| ≤ ft (ω) + µ|y| + λ|z|,

(5. 19)

˜ and dP × dt − a.e., ∀ n ≥ 1, ∀ (y, z) ∈ R × Rd , and all gn2 satisfy (H1) with (n + 2˜ µ)x, (H2s) with n + 2λ, ˜ |gn2 (ω, t, y, z)| ≤ f˜t (ω) + µ ˜|y| + λ|z|.

(5. 20)

Then, in view of (5. 19), (5. 20) and (H3) for g 1 , we know that dP × dt − a.e., ∀ n ≥ 1, ∀ (y, z) ∈ R × Rd , |gn (·, y, z)| ≤ ≤

˜ |gn1 (·, y, z)| + |gn2 (·, y, z)| ≤ |g 1 (·, y, 0)| + f· + µ|y| + λ|z| + f˜· + µ ˜|y| + λ|z|

˜ |g 1 (·, 0, 0)| + f· + f˜· + ϕ· (|y|) + (µ + µ ˜)|y| + (λ + λ)|z|.

(5. 21)

That is to say, all gn satisfy (HH) with the same parameters. Note that for each n ≥ 1, gn satisfies (H1), (H2) and (H3) by Remark 2.6 and Remark 5.4 together

with (5. 20). Furthermore, by (5. 21) we know that for each n ≥ 1, gn (·, X· , 0) ∈ Hp and then (H4)

holds for L· , U· , ξ, X· and gn . It then follows from Theorem 4.5 that there exists a unique solution 32

(Y·n , Z·n , K·n , An· ) ∈ S p × Mp × V +,p × V +,p of DRBSDE (ξ, gn + dV, L, U ) for each n ≥ 1. And, noticing that gn increases in n, by Corollary 3.6 and Corollary 4.7 we can deduce that Y·n ≤ Y·n+1 , dAn ≤ dAn+1

and dK n ≥ dK n+1 for each n ≥ 1.

In the sequel, we show that (5. 2) appearing in Proposition 5.1 holds true. In fact, let ˜ g(·, y, z) := g 1 (·, y, 0) − (f· + f˜· ) − (µ + µ ˜)|y| − (λ + λ)|z| and ˜ g¯(·, y, z) := g 1 (·, y, 0) + (f· + f˜· ) + (µ + µ ˜)|y| + (λ + λ)|z|. Then by (5. 19) and (5. 20), g ≤ gn ≤ g¯ for each n ≥ 1, and both g and g¯ satisfy (H1), (H2s), (H3), ˜)|X· | ∈ Hp , g(·, X· , 0) = g 1 (·, X· , 0) − (f· + f˜· ) − (µ + µ g¯(·, X· , 0) = g 1 (·, X· , 0) + (f· + f˜· ) + (µ + µ ˜)|X· | ∈ Hp . Thus, it follows from Theorem 4.5 that DRBSDE (ξ, g + dV, L, U ) and DRBSDE (ξ, g¯ + dV, L, U ) admit ¯ · , A¯· ) ∈ S p × Mp × respectively a unique solution (Y · , Z · , K · , A· ) ∈ S p × Mp × V +,p × V +,p and (Y¯· , Z¯· , K V +,p × V +,p , and by Corollary 3.6 and Corollary 4.7, we know that for each t ∈ [0, T ] and n ≥ 1,

¯T −K ¯ t | ≤ |KTn − Ktn | ≤ |K − K |, and |A − A | ≤ |AnT − Ant | ≤ |A¯T − A¯t |. (5. 22) Y t ≤ Ytn ≤ Y¯t , |K T t T t Furthermore, note that for each n ≥ 1, gn satisfies assumption (AA) with the same f¯· = g 1 (·, 0, 0) + f· +

¯ = λ+λ ˜ since g n satisfies (H1) with the same ρ(·), and (5. 19) and (5. 20) hold f˜· + A, µ ¯=µ ˜ + A and λ 1 ˜ T such true. It follows from Lemma 3.1 that there exists a constant C > 0 depending only on p, A, µ ˜, λ, that for each t ∈ [0, T ] and n ≥ 1,  ! p2 Z T + E |Zsn |2 ds t

≤

"

CE |ξ|p + |V |pt,T

!p  |gn (s, Ysn , Zsn )| ds Ft  t # !p Z T n n p n n p ¯ + |KT − Kt | + |AT − At | + fs ds + 1 Ft . t Z

T

(5. 23)

Finally, combining (5. 22) and (5. 23) yields that (5. 2) in Proposition 5.1 holds true with " ¯T − K ¯ t |p η := C¯ sup |Y s |p + sup |Y¯s |p + |V |pt,T + |K T − K t |p + |K s∈[t,T ] s∈[t,T ] !p # Z T p p 1 ˜ ¯ ¯ +|A − A | + |AT − At | + (|g (s, 0, 0)| + fs + fs ) ds + 1 T

t

t

˜ T. for some constant C¯ > 0 depending only on p, A, µ ˜, λ, Up to now, we have checked all conditions in Proposition 5.1. It follows from Proposition 5.1 that DRBSDE (ξ, g + dV, L, U ) admits a solution (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p such that lim (kY·n − Y· kS p + kZ·n − Z· kMp + kK·n − K· kS p + kAn· − A· kS p ) = 0.

n→∞

(5. 24)

Finally, let us show that (Y· , Z· , K· , A· ) is just the minimal solution of RBSDE (ξ, g + dV, L, U ) in S p × Mp × V +,p × V +,p . In fact, if (Y·′ , Z·′ , K·′ , A′· ) is also a solution of DRBSDE (ξ, g + dV, L, U ) in

S p × Mp × V +,p × V +,p , then noticing that for each n ≥ 1, gn ≤ g and gn satisfies (H1) and (H2), it

follows from Corollary 3.6 that for each n ≥ 1 and t ∈ [0, T ], Ytn ≤ Yt′ . Thus, by (5. 24) we obtain that Yt ≤ Yt′ for each t ∈ [0, T ], which is the desired result. Theorem 5.5 is then proved. 33

Remark 5.6. It follows from Remark 2.3, Remark 2.6, Remark 2.7 and Remark 5.4 that Theorem 5.5 generalizes and unifies some existing results on doubly RBSDEs, RBSDEs with one continuous barrier and non-reflected BSDEs. Example 5.7. Let the generator g := g 1 + g 2 with 1 1t>0 g 1 (ω, t, y, z) = h(|y|) − e|Bt (ω)|·y + (e−y ∧ 1) · |z| sin |z| + √ 4 t and g 2 (ω, t, y, z) = 1y≤0 where, with δ > 0 small enough,

h(x) =

p p 3 |y| + 1y>0 cos y + |y| · |z| + |Bt (ω)|,

   −x| ln x|  

, 0 < x ≤ δ;

h′ (δ−)(x − δ) + h(δ) , x > δ;     0 , other cases.

It is not very hard to verify that g 1 satisfies (H1) with ρ(·) = h(·), (H2’) with f· ≡ 0, µ = 0 and

˜ = 1 for each ˜ = 2 and λ λ = 1, and (H3), and that g 2 satisfies (A1a) and (A2) with f˜· = |B· | + 2, µ p > 1. Thus, if (H4) holds true for g 1 and some ξ, L· , U· , X· and p > 1, then by Theorem 5.5 we know

that DRBSDE (ξ, g + dV, L· , U· ) admits a minimal solution in S p × Mp × V +,p × V +,p . Note that, to our best knowledge, the above conclusion can not be obtained by any existing results. By Corollary 3.6, Corollary 4.7 and the proof of Theorem 5.5, it is not hard to verify the following comparison theorem for the minimal (resp. maximal) Lp solutions of DRBSDEs holds true. Proposition 5.8. Let p > 1 and for i = 1, 2, assume that V·i ∈ V p , g 1,i satisfies (H1), (H2’) and (H3),

g 2,i satisfies (A1a) (resp. (A1b)) and (A2), and that g i := g 1,i + g 2,i . Assume further that for i = 1, 2,

ξ i , Li· , U·i , and X·i satisfy (H4) associated with g i (or g 1,i ), and (Y·i , Z·i , K·i , Ai· ) ∈ S p × Mp × V +,p × V +,p

is the minimal (resp. maximal) solution of DRBSDE (ξ i , g i + dV i , Li , U i ). If ξ 1 ≤ ξ 2 , dV 1 ≤ dV 2 ,

L1· ≤ L2· , U·1 ≤ U·2 , and for each (y, z) ∈ R × Rd ,

dP × dt − a.e., g 1 (t, y, z) ≤ g 2 (t, y, z), then Yt1 ≤ Yt2 , t ∈ [0, T ]. Furthermore, if L1· = L2· and U·1 = U·2 , then dK 1 ≥ dK 2 and dA1 ≤ dA2 . 6. A special case: reflected BSDEs with two distinct barriers In this section, we will focus on RBSDEs with two distinct barriers. These equations have been extensively investigated in Hamad`ene and Hassani [23], Hamad`ene and Hdhiri [26], Hamad`ene and Hassani [24], Hamad`ene et al. [25] and Bayraktar and Yao [4] by a method involved in a local solution. In the present work, we will directly locate a X· ∈ V p with p > 0 between two distinct and continuous

− p reflecting barriers L· and U· (i.e. Lt < Ut , t ∈ [0, T ]) satisfying L+ · , U· ∈ S (see Proposition 6.1).

Based on this work and some results obtained in previous sections, we will establish several results for Lp (p > 1) solutions of RBSDEs with two distinct barriers, which strengthens some known works. 34

− p Proposition 6.1. Let L· , U· ∈ S, L+ · , U· ∈ S for some p > 0, and Lt < Ut , t ∈ [0, T ]. Then, there

exists a X· ∈ V p such that Lt ≤ Xt ≤ Ut , t ∈ [0, T ]. Proof. Let

D := {ω : both Lt (ω) and Ut (ω) are continuous functions on [0, T ], and Lt (ω) < Ut (ω), t ∈ [0, T ]}. Then, P(D) = 1, and for each ω ∈ D, since functions Lt (ω) and Ut (ω) are continuous and strictly separate on [0, T ], we can know that ε0 (ω) := min (Ut (ω) − Lt (ω)) > 0. t∈[0,T ]

For each t ∈ [0, T ] and 0 ≤ t1 ≤ t2 ≤ T , denote the following random variables: − Ht := sup L+ s + sup Us , 0≤s≤t

0≤s≤t

¯ 1 , t2 ) := L(t

sup Ls ,

t1 ≤s≤t2

U (t1 , t2 ) :=

inf

t1 ≤s≤t2

Us .

It is clear that (Ht )t∈[0,T ] is a (Ft )t∈[0,T ] -progressively measurable, nondecreasing and continuous process, ¯ 0 , t))t∈[t ,T ] is a (Ft )t∈[t ,T ] -progressively measurable, nondecreasing and continuous process for each (L(t 0 0 t0 ∈ [0, T ], and (U (t0 , t))t∈[t0 ,T ] is a (Ft )t∈[t0 ,T ] -progressively measurable, non-increasing and continuous process for each t0 ∈ [0, T ]. Furthermore, it is not very hard to verify that ∀ω ∈ D, ∃δ0 (ω) > 0, such that for each 0 ≤ t1 ≤ t2 ≤ T, if t2 − t1 < δ0 (ω), then ¯ L(ω, t1 , t2 ) < Ut2 (ω) and U (ω, t1 , t2 ) > Lt2 (ω).

(6. 1)

In fact, for each ω ∈ D, since Lt (ω) and Ut (ω) are continuous functions defined on [0, T ], we can deduce that there exists a δ0 (ω) > 0 such that for each 0 ≤ r ≤ t ≤ T , if t − r ≤ δ0 (ω), then |Ut (ω) − Ur (ω)| < ε0 (ω)/2 and |Lt (ω) − Lr (ω)| < ε0 (ω)/2. Thus, for each 0 ≤ t1 ≤ t2 ≤ T , if t2 − t1 ≤ δ0 (ω), then ¯ L(ω, t1 , t2 ) − Ut2 (ω)

=

sup (Ls (ω) − Lt1 (ω)) + (Lt1 (ω) − Ut1 (ω)) + (Ut1 (ω) − Ut2 (ω))

t1 ≤s≤t2




inf (Us (ω) − Ut1 (ω)) + (Ut1 (ω) − Lt1 (ω)) + (Lt1 (ω) − Lt2 (ω))

t1 ≤s≤t2

−

ε0 (ω) ε0 (ω) + ε0 (ω) − = 0. 2 2

Hence, (6. 1) holds true. In the sequel, define the following two stopping times: τ10 := inf{t ≥ 0 : Ht ≥ U (0, t)} ∧ T ;

¯ t) ≥ U (0, t)} ∧ T. τ20 := inf{t ≥ τ10 : L(0,

It is clear that τ20 > 0 due to L0 < U0 . Now, we show that if ω ∈ D and τ20 (ω) < T , then ¯ L(ω, 0, t) < U (ω, 0, t), t ∈ [0, τ20 (ω))

(6. 2)

¯ L(ω, 0, τ20 (ω)) = U (ω, 0, τ20 (ω))

(6. 3)

and

35

by considering two different cases τ10 (ω) = τ20 (ω) and τ10 (ω) < τ20 (ω). In fact, in the case of τ10 (ω) = τ20 (ω), it follows from the definition of τ10 that Ht (ω) < U (ω, 0, t), t ∈ [0, τ20 (ω)) and Hτ20 (ω) (ω) = U (ω, 0, τ20 (ω)),

¯ and from the definition of τ20 that L(ω, 0, τ20 (ω)) ≥ U (ω, 0, τ20 (ω)). These three inequalities together with

¯ the basic fact that L(ω, 0, t) ≤ Ht (ω) for each t ∈ [0, T ] yields that (6. 2) and (6. 3) hold true. On the

other hand, in the case of τ10 (ω) < τ20 (ω), by the definition of τ20 we know that (6. 3) holds and (6. 2)

is true for t ∈ [τ10 (ω), τ20 (ω)). Furthermore, if τ10 (ω) 6= 0, then it follows from the definition of τ10 that

¯ 0, t) ≤ Ht (ω) for each Ht (ω) < U (ω, 0, t), t ∈ [0, τ10 (ω)), which together with the basic fact that L(ω,

t ∈ [0, T ] implies that (6. 2) is also true for t ∈ [0, τ10 (ω)). Thus, we have verified that (6. 2) and (6. 3) hold true when ω ∈ D and τ20 (ω) < T .

Based on inequalities (6. 2) and (6. 3), we can show that A ∩ B = ∅ and where

 ω ∈ D : τ20 (ω) = T ∪ A ∪ B = D,

(6. 4)

n o ¯ A := ω ∈ D : τ20 (ω) < T and L(ω, 0, τ20 (ω)) = Lτ20 (ω) (ω)

and

o n B := ω ∈ D : τ20 (ω) < T and U (ω, 0, τ20 (ω)) = Uτ20 (ω) (ω) . In fact, when ω ∈ D and τ20 (ω) < T , there exist two constants s1 (ω), s2 (ω) ∈ [0, τ20 (ω)] such that ¯ ¯ L(ω, 0, τ20 (ω)) = Ls1 (ω) (ω) = L(ω, 0, s1 (ω))

and U (ω, 0, τ20 (ω)) = Us2 (ω) (ω) = U (ω, 0, s2 (ω)). (6. 5)

Then, in order to prove (6. 4), it suffices to show that when ω ∈ D and τ20 (ω) < T , two cases of

s1 (ω) = s2 (ω) = τ20 (ω) and 0 ≤ s1 (ω), s2 (ω) < τ20 (ω) are both impossible to occur. For the first case

of s1 (ω) = s2 (ω) = τ20 (ω), by (6. 3) and (6. 5) we have Lτ20 (ω) (ω) = Uτ20 (ω) (ω), which contradicts with ω ∈ D. For the second case of 0 ≤ s1 (ω), s2 (ω) < τ20 (ω), in view of (6. 2), (6. 3) and (6. 5), we have ¯ ¯ L(ω, 0, τ20 (ω)) = Ls1 (ω) (ω) = L(ω, 0, s1 (ω)) < U (ω, 0, s1 (ω)), and ¯ ¯ 0, s2 (ω)), L(ω, 0, τ20 (ω)) = U (ω, 0, τ20 (ω)) = Us2 (ω) (ω) = U (ω, 0, s2 (ω)) > L(ω, which means that ¯ ¯ L(ω, 0, s2 (ω)) < L(ω, 0, s1 (ω))

and U (ω, 0, s2 (ω)) < U (ω, 0, s1 (ω)).

¯ However, this contradicts with the definitions of L(ω, 0, t) and U (ω, 0, t). Hence, (6. 4) holds true. Furthermore, we define the following stopping times: ¯ 11 , t) ≥ Ut } ∧ T ; σ11 := inf{t ≥ τ11 : L(τ

τ11 := inf{t ≥ τ20 : U (τ20 , t) ≤ Lt } ∧ T ; ¯ 0 , t) ≥ Ut } ∧ T ; τ12 := inf{t ≥ τ20 : L(τ 2

σ12 := inf{t ≥ τ12 : U (τ12 , t) ≤ Lt } ∧ T.

And, for each k ≥ 2, define the following stopping times: 1 1 τk1 := inf{t ≥ σk−1 : U (σk−1 , t) ≤ Lt } ∧ T ;

2 ¯ 2 , t) ≥ Ut } ∧ T ; τk2 := inf{t ≥ σk−1 : L(σ k−1

36

¯ 1 , t) ≥ Ut } ∧ T ; σk1 := inf{t ≥ τk1 : L(τ k

σk2 := inf{t ≥ τk2 : U (τk2 , t) ≤ Lt } ∧ T.

i It follows from (6. 1) that for each ω ∈ D and k ≥ 1, if τk+1 (ω) < T for some i = 1, 2, then i τk+1 (ω) − σki (ω) > δ0 (ω) and σki (ω) − τki (ω) > δ0 (ω).

Consequently, for each ω ∈ D, we can always find a positive integer k0 (ω) such that τk10 (ω)+1 (ω) = τk20 (ω)+1 (ω) = T.

(6. 6)

Now, we can construct the desired process as follows: for each (ω, t) ∈ Ω × [0, T ],   Xt (ω) := U0 1τ10 =0 (ω) + Ht (ω)1τ10 >0 (ω) 1[0,τ10 (ω)] (t) + U (ω, 0, t)1(τ10 (ω),τ20 (ω)] (t) h ¯ τ20 (ω), t)1(τ20 (ω),τ12 (ω)] (t)+ +1A (ω) L(ω, k0 (ω) 

X

k=1

  ¯  2 U (ω, τk2 (ω), t)1(τk2 (ω),σk2 (ω)] (t) + L(ω, σk2 (ω), t)1(σk2 (ω),τk+1 (ω)] (t)

h +1B (ω) U (ω, τ20 (ω), t)1(τ20 (ω),τ11 (ω)] (t)+ k0 (ω) 

X

k=1

  ¯ . 1 L(ω, τk1 (ω), t)1(τk1 (ω),σk1 (ω)] (t) + U (ω, σk1 (ω), t)1(σk1 (ω),τk+1 (ω)] (t)

¯ U and those previous stopping times, it is not very In view of (6. 4), (6. 3) and the definitions of H, L, hard to verify that X· ∈ V. Thus, in order to get the desired conclusion of this proposition, it is enough to prove that for each ω ∈ D, Lt (ω) ≤ Xt (ω) ≤ Ut (ω), t ∈ [0, T ]

(6. 7)

|Xt (ω)| ≤ Ht (ω), t ∈ [0, T ].

(6. 8)

and

In what follows, let us arbitrarily fix a ω ∈ D. We first consider the case when t ∈ [0, τ10 (ω)], which is

divided into two subcases of τ10 (ω) = 0 and τ10 (ω) > 0. For the first subcase, we have X0 = U0 , so (6. 7) holds for t ∈ [0, τ10 (ω)]. On the other hand, it follows from the definitions of τ10 , U (0, 0) and H0 that H0 ≥ U (0, 0) = U0 ≥ −U0− ≥ −H0 ,

which means that (6. 8) holds also true for t ∈ [0, τ10 (ω)]. For the second subcase, we have Xt (ω) =

Ht (ω), t ∈ [0, τ10 (ω)], which means that (6. 8) holds for t ∈ [0, τ10 (ω)]. On the other hand, it follows from

the definitions of τ10 , Ht and U (0, t) that

Lt (ω) ≤ Ht (ω) ≤ U (ω, 0, t) ≤ Ut (ω), t ∈ [0, τ10 (ω)]. That is to say, (6. 7) is also true for t ∈ [0, τ10 (ω)]. Thus, (6. 7) and (6. 8) hold true when τ10 (ω) = T .

When τ10 (ω) < T , we further consider the case of t ∈ (τ10 (ω), τ20 (ω)] with τ10 (ω) < τ20 (ω). In this

¯ τ20 and U that if case, Xt (ω) = U (ω, 0, t) by the definition. And, it follows from the definitions of L, τ10 (ω) < T and τ10 (ω) < τ20 (ω), then ¯ Lt (ω) ≤ L(ω, 0, t) ≤ U (ω, 0, t) ≤ Ut (ω), t ∈ (τ10 (ω), τ20 (ω)], 37

which means that (6. 7) holds when τ10 (ω) < T , τ10 (ω) < τ20 (ω) and t ∈ (τ10 (ω), τ20 (ω)]. Furthermore, by the definition of Ht we have U (ω, 0, t) = inf Us (ω) = − sup [−Us (ω)] ≥ − sup Us− (ω) ≥ −Ht (ω), t ∈ [0, T ]. 0≤s≤t

0≤s≤t

(6. 9)

0≤s≤t

On the other hand, it follows from the definitions of U , τ10 and H that if τ10 (ω) < T and τ10 (ω) < τ20 (ω), then U (ω, 0, t) ≤ U (ω, 0, τ10 (ω)) ≤ Hτ10 (ω) (ω) ≤ Ht (ω), t ∈ (τ10 (ω), τ20 (ω)].

(6. 10)

Combining (6. 9) and (6. 10) yields that (6. 8) holds true when τ10 (ω) < T , τ10 (ω) < τ20 (ω) and t ∈

(τ10 (ω), τ20 (ω)]. Thus, the desired results (6. 7) and (6. 8) holds true when τ20 (ω) = T .

When τ20 (ω) < T , we need to further consider two cases of ω ∈ A and ω ∈ B. In the sequel, we first

consider the case of ω ∈ B and t ∈ (τ20 (ω), τ11 (ω)]. Note by (6. 1) that τ20 (ω) < τ11 (ω) if ω ∈ B, and that U (ω, 0, τ20 (ω)) = Uτ20 (ω) (ω), ω ∈ B.

(6. 11)

When ω ∈ B and t ∈ (τ20 (ω), τ11 (ω)], we have Xt (ω) = U (ω, τ20 (ω), t) by the definition. And, it follows

from the definitions of τ11 and U that if ω ∈ B, then

Lt (ω) ≤ U (ω, τ20 (ω), t) ≤ Ut (ω), t ∈ (τ20 (ω), τ11 (ω)], which means that (6. 7) holds when ω ∈ B and t ∈ (τ20 (ω), τ11 (ω)]. Furthermore, by virtue of (6. 9) we know that if ω ∈ B, then U (ω, τ20 (ω), t) =

inf

τ20 (ω)≤s≤t

Us (ω) ≥ inf Us (ω) ≥ −Ht (ω), t ∈ (τ20 (ω), T ]. 0≤s≤t

(6. 12)

On the other hand, it follows from the definitions of U and H together with (6. 11) and (6. 10) that for each ω ∈ B and t ∈ (τ20 (ω), τ11 (ω)], we have U (ω, τ20 (ω), t) ≤ U (ω, τ20 (ω), τ20 (ω)) = Uτ20 (ω) (ω) = U (ω, 0, τ20 (ω)) ≤ Hτ20 (ω) (ω) ≤ Ht (ω).

(6. 13)

Combining (6. 12) and (6. 13) yields that (6. 8) holds true when ω ∈ B and t ∈ (τ20 (ω), τ11 (ω)]. Thus,

(6. 7) and (6. 8) holds true when ω ∈ B and τ11 (ω) = T .

When ω ∈ B and τ11 (ω) < T , by (6. 1) we know that τ11 (ω) < σ11 (ω), and then we have to consider

the case of t ∈ (τ11 (ω), σ11 (ω)]. When ω ∈ B, τ11 (ω) < T and t ∈ (τ11 (ω), σ11 (ω)], we have Xt (ω) =

¯ ¯ and σ 1 that if ω ∈ B and L(ω, τ11 (ω), t) by the definition. And, it follows from the definitions of L 1

τ11 (ω) < T , then

¯ Lt (ω) ≤ L(ω, τ11 (ω), t) ≤ Ut (ω), t ∈ (τ11 (ω), σ11 (ω)], which means that (6. 7) holds when ω ∈ B, τ11 (ω) < T and t ∈ (τ11 (ω), σ11 (ω)]. Furthermore, it follows ¯ and Ht that if ω ∈ B and τ11 (ω) < T , then from the definitions of L

1 ¯ L(ω, τ11 (ω), t) ≤ sup Ls (ω) ≤ sup L+ s (ω) ≤ Ht (ω), t ∈ (τ1 (ω), T ]. 0≤s≤t

(6. 14)

0≤s≤t

¯ τ 1 and H together with (6. 12) that when ω ∈ B, On the other hand, it follows from the definitions of L, 1

τ11 (ω) < T and t ∈ (τ11 (ω), σ11 (ω)], we have

¯ ¯ L(ω, τ11 (ω), t) ≥ L(ω, τ11 (ω), τ11 (ω)) = Lτ11 (ω) (ω) = U (ω, τ20 (ω), τ11 (ω)) ≥ −Hτ11 (ω) (ω) ≥ −Ht (ω). (6. 15) 38

Combining (6. 14) and (6. 15) yields that (6. 8) holds true when ω ∈ B, τ11 (ω) < T and t ∈ (τ11 (ω), σ11 (ω)].

Thus, (6. 7) and (6. 8) holds true when ω ∈ B and σ11 (ω) = T .

Furthermore, when ω ∈ B and σ11 (ω) < T , by (6. 1) we know that σ11 (ω) < τ21 (ω), and we need to

further consider the case of t ∈ (σ11 (ω), τ21 (ω)]. When ω ∈ B, σ11 (ω) < T and t ∈ (σ11 (ω), τ21 (ω)], we have

Xt (ω) = U (ω, σ11 (ω), t) by the definition. And, it follows from the definitions of τ21 and U that if ω ∈ B and σ11 (ω) < T , then

Lt (ω) ≤ U (ω, σ11 (ω), t) ≤ Ut (ω), t ∈ (σ11 (ω), τ21 (ω)], which means that (6. 7) holds when ω ∈ B, σ11 (ω) < T and t ∈ (σ11 (ω), τ21 (ω)]. Furthermore, by virtue

of (6. 9) we know that if ω ∈ B and σ11 (ω) < T , then U (ω, σ11 (ω), t) =

inf

σ11 (ω)≤s≤t

Us (ω) ≥ inf Us (ω) ≥ −Ht (ω), t ∈ (σ11 (ω), T ]. 0≤s≤t

(6. 16)

On the other hand, it follows from the definitions of U , σ11 and H together with (6. 14) that when ω ∈ B,

σ11 (ω) < T and t ∈ (σ11 (ω), τ21 (ω)], we have

¯ τ11 (ω), σ11 (ω)) ≤ Hσ11 (ω) (ω) ≤ Ht (ω). U (ω, σ11 (ω), t) ≤ U (ω, σ11 (ω), σ11 (ω)) = Uσ11 (ω) (ω) = L(ω,

(6. 17)

Combining (6. 16) and (6. 17) yields that (6. 8) holds true when ω ∈ B, σ11 (ω) < T and t ∈ (σ11 (ω), τ21 (ω)]. Thus, (6. 7) and (6. 8) have been proved when ω ∈ B and τ21 (ω) = T .

Finally, by a very similar argument to the case of ω ∈ B, τ11 (ω) < T and t ∈ (τ11 (ω), σ11 (ω)], and the

case of ω ∈ B, σ11 (ω) < T and t ∈ (σ11 (ω), τ21 (ω)], we can successively check the case of ω ∈ B, τk1 (ω) < T

1 and t ∈ (τk1 (ω), σk1 (ω)], and the case of ω ∈ B, σk1 (ω) < T and t ∈ (σk1 (ω), τk+1 (ω)] for k = 2, 3, · · · , k0 (ω),

and conclude that, in view of (6. 6), the desired results (6. 7) and (6. 8) hold true when ω ∈ B. In the same way, we can prove that the desired results (6. 7) and (6. 8) hold also true when ω ∈ A. Thus, in view of (6. 4), the proof of Proposition 6.1 is completed. In order to state the main results of this section, let us introduce the following assumption with p > 1. (H4’)

(i) L· , U· ∈ S, ξ ∈ Lp (FT ) and LT ≤ ξ ≤ UT ; − p (ii) L+ · , U· ∈ S and Lt < Ut , t ∈ [0, T ].

It follows from Proposition 6.1 that if (H4’) holds for L· , U· and ξ, then there exists a X· ∈ V p such that Lt ≤ Xt ≤ Ut , t ∈ [0, T ], and then (H4) holds true if the generator g satisfies (H3s) or (A2), where a difference from (v) of Remark 2.7 can be seen. Thus, by virtue of Theorem 4.5, Theorem 5.5 and Proposition 5.8 we can obtain the following results, which strengthen some corresponding results proved in Cvitani´c and Karatzas [9], Lepeltier and San Mart´ın [43], Bahlali et al. [2], Hamad`ene and Hassani [23] and Hamad`ene et al. [25]. Theorem 6.2. Assume that p > 1, ξ ∈ Lp (FT ), V· ∈ V p , the generator g satisfies assumptions (H1), (H2) and (H3s), and assumption (H4’) holds for L· , U· and ξ. Then, DRBSDE (ξ, g + dV, L, U ) admits a unique solution (Y· , Z· , K· , A· ) ∈ S p × Mp × V +,p × V +,p .

39

Theorem 6.3. Assume that p > 1, V· ∈ V p and g satisfies (A1a) (resp. (A1b)) and (A2). Assume further that (H4’) holds true for L· , U· and ξ. Then, DRBSDE (ξ, g + dV, L· , U· ) admits a minimal solution (resp. a maximal solution) (Y· , Z· , K· , A· ) in S p × Mp × V +,p × V +,p . Proposition 6.4. Let p > 1 and for i = 1, 2, assume that V·i ∈ V p and g i satisfies (A1a) (resp. (A1b)) and (A2). Assume further that for i = 1, 2, (H4’) holds for ξ i , Li· and U·i , and (Y·i , Z·i , K·i , Ai· ) ∈

S p × Mp × V +,p × V +,p is the minimal (resp. maximal) solution of DRBSDE (ξ i , g i + dV i , Li , U i ). If

ξ 1 ≤ ξ 2 , dV 1 ≤ dV 2 , L1· ≤ L2· , U·1 ≤ U·2 , and for each (y, z) ∈ R × Rd ,

dP × dt − a.e., g 1 (t, y, z) ≤ g 2 (t, y, z), then Yt1 ≤ Yt2 , t ∈ [0, T ]. Furthermore, if L1· = L2· and U·1 = U·2 , then dK 1 ≥ dK 2 and dA1 ≤ dA2 . References [1] Aman, A., 2009. Lp -solution of reflected generalized BSDEs with non-Lipschitz coefficients. Random Oper. Stoch. Equ. 17 (3), 201–219. [2] Bahlali, K., Hamad` ene, S., Mezerdi, B., 2005. Backward stochastic differential equations with two reflecting barriers and continuous with quadratic growth coefficient. Stochastic Process. Appl. 115 (7), 1107–1129. [3] Bayraktar, E., Yao, S., 2012. Quadratic reflected BSDEs with unbounded obstacles. Stochastic Process. Appl. 122 (4), 1155–1203. [4] Bayraktar, E., Yao, S., 2015. Doubly reflected BSDEs with integrable parameters and related Dynkin games. Stochastic Process. Appl. 125 (12), 4489–4542. [5] Bismut, J., 1973. Conjugate convex functions in optimal stochastic control. Journal of Mathematical Analysis and Applications 44, 384–404. [6] Briand, P., Carmona, R., 2000. BSDEs with polynomial growth generators. Journal of Applied Mathematics and Stochastic Analysis 13 (3), 207–238. [7] Briand, P., Delyon, B., Hu, Y., Pardoux, E., Stoica, L., 2003. Lp solutions of backward stochastic differential equations. Stochastic Processes and Their Applications 108 (1), 109–129. [8] Briand, P., Lepeltier, J.-P., San Martin, J., 2007. One-dimensional backward stochastic differential equations whose coefficient is monotonic in y and non-Lipschitz in z. Bernoulli 13 (1), 80–91. [9] Cvitani´ c, J., Karatzas, I., 1996. Backward stochastic differential equations with reflection and Dynkin games. Ann. Probab. 24 (4), 2024–2056. [10] El Asri, B., Hamad` ene, S., Wang, H., 2011. Lp -solutions for doubly reflected backward stochastic differential equations. Stoch. Anal. Appl. 29 (6), 907–932. [11] El Karoui, N., Kapoudjian, C., Pardoux, E., Peng, S., Quenez, M. C., 1997. Reflected solutions of backward SDE’s, and related obstacle problems for PDE’s. Ann. Probab. 25 (2), 702–737. [12] El Karoui, N., Pardoux, E., Quenez, M. C., 1997. Reflected backward SDEs and American options. In: Numerical methods in finance. Vol. 13 of Publ. Newton Inst. Cambridge Univ. Press, Cambridge, pp. 215–231. [13] El Karoui, N., Peng, S., Quenez, M. C., 1997. Backward stochastic differential equations in finance. Mathematical Finance 7 (1), 1–71. [14] Essaky, E. H., Hassani, M., 2011. General existence results for reflected BSDE and BSDE. Bull. Sci. Math. 135 (5), 442–466. [15] Essaky, E. H., Hassani, M., 2013. Generalized BSDE with 2-reflecting barriers and stochastic quadratic growth. J. Differential Equations 254 (3), 1500–1528. [16] Fan, S., 2015. Existence, uniqueness and approximation for Lp solutions of reflected BSDEs under weaker assumptions. arXiv:1510.08587v1 [math.PR] 29 Oct 2015.

40

[17] Fan, S., 2015. Lp solutions of multidimensional BSDEs with weak monotonicity and general growth generators. Journal of Mathematical Analysis and Applications 432, 156–178. [18] Fan, S., Jiang, L., 2010. Uniqueness result for the BSDE whose generator is monotonic in y and uniformly continuous in z. C. R. Math. Acad. Sci. Paris 348 (1-2), 89–92. [19] Fan, S., Jiang, L., 2012. A generalized comparison theorem for BSDEs and its applications. J. Theoret. Probab. 25 (1), 50–61. [20] Fan, S., Jiang, L., 2012. One-dimensional BSDEs with left-continuous, lower semi-continuous and linear-growth generators. Statistics and Probability Letters 82, 1792–1798. [21] Fan, S., Jiang, L., Davison, M., 2013. Existence and uniqueness result for multidimensional BSDEs with generators of Osgood type. Front. Math. China 8 (4), 811–824. [22] Hamad` ene, S., 2002. Reflected BSDE’s with discontinuous barrier and application. Stoch. Stoch. Rep. 74 (3-4), 571– 596. [23] Hamad` ene, S., Hassani, M., 2005. BSDEs with two reflecting barriers: the general result. Probab. Theory Related Fields 132 (2), 237–264. [24] Hamad` ene, S., Hassani, M., 2006. BSDEs with two reflecting barriers driven by a brownian motion and an independent poisson noise and related dynkin game. Electronic Journal of Probability 11 (5), 121–145. [25] Hamad` ene, S., Hassani, M., Ouknine, Y., 2010. Backward SDEs with two rcll reflecting barriers without Mokobodski’s hypothesis. Bull. Sci. Math. 134 (8), 874–899. [26] Hamad` ene, S., Hdhiri, I., 2006. Backward stochastic differential equations with two distinct reflecting barriers and quadratic growth generator. J. Appl. Math. Stoch. Anal., Art. ID 95818, 28. [27] Hamad` ene, S., Lepeltier, J.-P., 2000. Reflected BSDEs and mixed game problem. Stochastic Process. Appl. 85 (2), 177–188. [28] Hamad` ene, S., Lepeltier, J.-P., Matoussi, A., 1997. Double barrier backward SDEs with continuous coefficient. In: Backward stochastic differential equations (Paris, 1995–1996). Vol. 364 of Pitman Res. Notes Math. Ser. Longman, Harlow, pp. 161–175. [29] Hamad` ene, S., Lepeltier, J.-P., Wu, Z., 1999. Infinite horizon reflected backward stochastic differential equations and applications in mixed control and game problems. Probab. Math. Statist. 19 (2, Acta Univ. Wratislav. No. 2198), 211–234. [30] Hamad` ene, S., Popier, A., 2012. Lp -solutions for reflected backward stochastic differential equations. Stoch. Dyn. 12 (2), 1150016, 35. [31] Hamad` ene, S., Zhang, J., 2010. Switching problem and related system of reflected backward SDEs. Stochastic Process. Appl. 120 (4), 403–426. [32] Hu, Y., Tang, S., 2010. Multi-dimensional BSDE with oblique reflection and optimal switching. Probab. Theory Related Fields 147 (1-2), 89–121. [33] Hu, Y., Tang, S., 2015. Multi-dimensional backward stochastic differential equations of diagonally quadratic generators. Stochastic Processes and their Applications. http://dx.doi.org/10.1016/j.spa.2015.10.011. [34] Hua, W., Jiang, L., Shi, X., 2013. Infinite time interval RBSDEs with non-Lipschitz coefficients. J. Korean Statist. Soc. 42 (2), 247–256. [35] Izumi, Y., 2013. The Lp Cauchy sequence for one-dimensional BSDEs with linear growth generators. Statist. Probab. Lett. 83 (6), 1588–1594. [36] Jia, G., 2008. A uniqueness theorem for the solution of backward stochastic differential equations. Comptes Rendus de l’Acad´ emie des Sciences-Series I-Mathematics 346 (7-8), 439–444, comptes Rendus Mathematique. [37] Jia, G., 2010. Backward stochastic differential equations with a uniformly continuous generator and related gexpectation. Stochastic Processes and their Applications 120 (11), 2241–2257. [38] Jia, G., Xu, M., 2014. Construction and uniqueness for reflected BSDE under linear increasing condition. arXiv:0801.3718v1 [math.SG].

41

[39] Klimsiak, T., 2012. Reflected BSDEs with monotone generator. Electron. J. Probab. 17, no. 107, 25. [40] Klimsiak, T., 2013. BSDEs with monotone generator and two irregular reflecting barriers. Bulletin des Sciences Math´ ematiques 137 (3), 268–321. [41] Kobylanski, M., Lepeltier, J. P., Quenez, M. C., Torres, S., 2002. Reflected BSDE with superlinear quadratic coefficient. Probab. Math. Statist. 22 (1, Acta Univ. Wratislav. No. 2409), 51–83. [42] Lepeltier, J.-P., Matoussi, A., Xu, M., 2005. Reflected backward stochastic differential equations under monotonicity and general increasing growth conditions. Adv. in Appl. Probab. 37 (1), 134–159. [43] Lepeltier, J.-P., San Mart´ın, J., 2004. Backward SDEs with two barriers and continuous coefficient: an existence result. J. Appl. Probab. 41 (1), 162–175. [44] Lepeltier, J.-P., Xu, M., 2005. Penalization method for reflected backward stochastic differential equations with one r.c.l.l. barrier. Statist. Probab. Lett. 75 (1), 58–66. [45] Ma, J., Zhang, J., 2005. Representations and regularities for solutions to BSDEs with reflections. Stochastic Process. Appl. 115 (4), 539–569. [46] Ma, M., Fan, S. J., Song, X., 2013. Lp (p > 1) solutions of backward stochastic differential equations with monotonic and uniformly continuous generators. Bull. Sci. Math. 137 (2), 97–106. [47] Mao, X., 1995. Adapted solutions of backward stochastic differential equations with non-Lipschitz coefficients. Stochastic Processes and their Applications 58 (2), 281–292. [48] Matoussi, A., 1997. Reflected solutions of backward stochastic differential equations with continuous coefficient. Statist. Probab. Lett. 34 (4), 347–354. [49] Pardoux, E., 1999. BSDEs, weak convergence and homogenization of semilinear PDEs. In: Clarke, F., Stern, R. (Eds.), Nonlinear Analysis, Differential Equations and Control. Kluwer Academic, New York, pp. 503–549. [50] Pardoux, E., Peng, S., 1990. Adapted solution of a backward stochastic differential equation. Systems Control Letters 14 (1), 55–61. [51] Peng, S., 1997. Backward SDE and related g-expectation. In: El Karoui, N., Mazliak, L. (Eds.), Backward stochastic differential equations (Paris,1995–1996). Vol. 364 of Pitman Research Notes Mathematical Series. Longman, Harlow, pp. 141–159. [52] Peng, S., 1999. Monotonic limit theorem of BSDE and nonlinear decomposition theorem of Doob-Meyer’s type. Probability Theory and Related Fields 113 (4), 473–499. [53] Peng, S., 2004. Nonlinear expectations, nonlinear evaluations and risk measures. In: Stochastic methods in finance. Vol. 1856 of Lecture Notes in Math. Springer, Berlin, pp. 165–253. [54] Peng, S., Xu, M., 2005. The smallest g-supermartingale and reflected BSDE with single and double L2 obstacles. Ann. Inst. H. Poincar´ e Probab. Statist. 41 (3), 605–630. [55] Peng, S., Xu, M., 2010. Reflected BSDE with a constraint and its applications in an incomplete market. Bernoulli 16 (3), 614–640. [56] Rosazza Gianin, E., 2006. Risk measures via g-expectations. Insurance Math. Econom. 39 (1), 19–34. [57] Rozkosz, A., Slomi´ nski, L., 2012. Lp solutions of reflected BSDEs under monotonicity condition. Stochastic Process. Appl. 122 (12), 3875–3900. [58] Xu, M., 2008. Backward stochastic differential equations with reflection and weak assumptions on the coefficients. Stochastic Process. Appl. 118 (6), 968–980. [59] Zheng, S., Zhou, S., 2008. A generalized existence theorem of reflected BSDEs with double obstacles. Statist. Probab. Lett. 78 (5), 528–536.

42