1:
2: \documentclass[12pt]{article}
3: \usepackage{amssymb}
4:
5: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
6: %\usepackage{sw20jart}
7:
8: %TCIDATA{TCIstyle=article/art4.lat,jart,sw20jart}
9:
10: %TCIDATA{Created=Wed Dec 01 10:06:51 2004}
11: %TCIDATA{LastRevised=Tue Feb 21 19:52:35 2006}
12: %TCIDATA{Language=American English}
13:
14:
15: \newtheorem{theorem}{{\bf Theorem}}[section]
16: \newtheorem{lemma}{{\bf Lemma}}[section]
17: \newtheorem{proposition}{{\bf Proposition}}[section]
18: \newtheorem{definition}{{\bf Definition}}[section]
19: \newtheorem{corollary}{{\bf Corollary}}[section]
20: \newtheorem{remark}{{\bf Remark}}[section]
21: \newtheorem{example}{{\bf Example}}[section]
22: \newtheorem{assumption}{{\bf Assumption}}[section]
23: \input{tcilatex}
24: \input tcilatex
25: \begin{document}
26:
27: \title{Reflected BSDE with monotonicity and general increasing in $y$, and non-Lipschitz
28: conditions in $z$}
29: \author{Mingyu Xu \thanks{Email: xvmingyu@gmail.com}\\
30: {\small Departement des Math\'ematiques, Universit\'e du Maine,
31: France;}\\{\small Department of Financial Mathematics and Control
32: science, School
33: of Mathematical Science,} \\
34: {\small Fudan University, Shanghai, 200433, China.}\\
35: Draft : March 1st, 2006} \maketitle
36:
37: \textbf{Abstract} In this paper, we study the reflected BSDE with one
38: continuous barrier, under the monotonicity and general increasing condition
39: on $y$ and non Lipschitz condition on $z$. We prove the existence and
40: uniqueness of the solution to these equation by approximation method.
41:
42: \textbf{Keywords} Reflected backward stochastic differential
43: equation, monotonicity, non-Lipschitz
44:
45: \section{Introduction}
46:
47: Nonlinear backward stochastic differential equations (BSDE in short) were
48: firstly introduced by Pardoux and Peng in 1990, \cite{PP90}. They proved
49: that there exists a unique solution $(Y,Z)$ to this equation if the terminal
50: condition $\xi $ and coefficient $f$ satisfy smooth square-integrability
51: assumptions and $f(t,\omega ,y,z)$ is Lipschitz in $(y,z)$ uniformy in $%
52: (t,\omega )$. Later many assumptions have been made to relax the Lipschitz
53: condition on $f$. Pardoux (1999, \cite{P99}) and Briand et al. (2003, \cite
54: {BDHPS}) studied the solution of a BSDE with a coefficient $f(t,\omega ,y,z)$%
55: , which still satisfies the Lipschitz condition on $z$, but only
56: monotonicity, continuity and generalized increasing on $y$, i.e.for some
57: continuous increasing function $\varphi :\mathbb{R}_{+}\rightarrow \mathbb{R}%
58: _{+}$, real number $\mu $ $>0$:
59: \begin{eqnarray}
60: \left| f(t,y,0)\right| &\leq &\left| f(t,0,0)\right| +\varphi
61: (\left| y\right| )\mbox{, }\forall (t,y)\in [0,T]\times
62: \mathbb{R}\mbox{, a.s.;}
63: \label{con-mono} \\
64: (y-y^{\prime })(f(t,y,z) &-&f(t,y^{\prime },z))\leq \mu (y-y^{\prime })^{2}%
65: \mbox{, }\forall (t,z)\in [0,T]\times \mathbb{R}^{d}\mbox{,
66: }y,y^{\prime }\in \mathbb{R}\mbox{, a.s.} \nonumber
67: \end{eqnarray}
68: The case when $f$ is quadratic on $z$ and $\xi $ is bounded was firstly
69: studied by Kobylanski in \cite{K00}. She proved an existence result when the
70: coefficient is only linear growth in $y$, and quadratic in $z$. In \cite
71: {LS98}, Lepeltier and San Mart\'{i}n generalized to a superlinear case in $y$%
72: . More recently, in \cite{LS04}, they and Briand considered the BSDE whose
73: coefficient $f$ satisfies only monotonicity, continuity and generalized
74: increasing on $y$, and quadratic or linear increasing in $z$, i.e.
75: \begin{eqnarray}
76: (y-y^{\prime })(f(t,y,z) &-&f(t,y^{\prime },z))\leq \mu (y-y^{\prime })^{2}%
77: \mbox{, }\forall (t,z)\in [0,T]\times \mathbb{R}^{d}\mbox{,
78: }y,y^{\prime
79: }\in \mathbb{R}\mbox{, a.s.} \nonumber \\
80: \left| f(t,y,z)\right| &\leq &\varphi (\left| y\right| )+A\left|
81: z\right| ^{2}\mbox{, }\forall (t,y)\in [0,T]\times \mathbb{R}\mbox{,
82: a.s.;} \label{con-qua}
83: \end{eqnarray}
84: or
85: \begin{equation}
86: \left| f(t,y,z)\right| \leq g_{t}+\varphi (\left| y\right| )+A\left|
87: z\right| ,\forall (t,y)\in [0,T]\times \mathbb{R}\mbox{, a.s..}
88: \label{con-lin}
89: \end{equation}
90: In the same paper, they studied the case $f(t,y,z)=\left| z\right| ^{p}$,
91: for $p\in (1,2]$, and gave some sufficient and necessary conditions on $\xi $
92: for the existence of solutions.
93:
94: El Karoui, Kapoudjian, Pardoux, Peng and Quenez introduced the notion of
95: reflected BSDE (RBSDE in short) on one lower barrier in 1997, \cite{EKPPQ}:
96: the solution is forced to remain above a continuous process, which is
97: considered as the lower barrier. More precisely, a solution for such
98: equation associated to a coefficient $f(t,\omega ,y,z)$, a terminal value $%
99: \xi $, a continuous barrier $L$, is a triple
100: $(Y_{t},Z_{t},K_{t})_{0\leq t\leq T}$ of adapted processes valued
101: on $\mathbb{R}^{1+d+1}$, which satisfies a square integrability
102: condition,
103: \[
104: Y_{t}=\xi
105: +\int_{t}^{T}f(s,Y_{s},Z_{s})ds+K_{T}-K_{t}-\int_{t}^{T}Z_{s}dB_{s},0\leq
106: t\leq T\mbox{, a.s.,}
107: \]
108: and $Y_{t}\geq L_{t}$, $0\leq t\leq T$, a.s.. Furthermore, the process $%
109: (K_{t})_{0\leq t\leq T}$ is non decreasing, continuous, and the role of $%
110: K_{t}$ is to push upward the state process in a minimal way, to keep it
111: above $L$. In this sense it satisfies $\int_{0}^{T}(Y_{s}-L_{s})dK_{s}=0$.
112: They proved the existence and uniqueness of the solution when $f(t,\omega
113: ,y,z)$ is Lipschitz in $(y,z)$ uniformly in $(t,\omega )$. Then Matoussi
114: (1997, \cite{M97}) consider RBSDE's where the coefficient $f$ is continuous
115: and at most linear growth in $y$, $z$. In this case, he proved the existence
116: of maximal solution for the RBSDE.
117:
118: In \cite{KLQT}, Kobylanski, Lepeltier, Quenez and Torres proved the
119: existence of a maximal and minimal bounded solution for the RBSDE when the
120: coefficient $f(t,\omega ,y,z)$ is super linear increasing in $y$ and
121: quadratic in $z$, i.e. there exists a function $l$ strictly positive such
122: that
123: \[
124: \left| f(t,y,z)\right| \leq l(y)+A\left| z\right| ^{2}\mbox{, with }%
125: \int_{0}^{\infty }\frac{dx}{l(x)}=+\infty .
126: \]
127: In this case, $\xi $ and $L$ are required to be bounded, and $L$ is a
128: continuous process. Recently, in \cite{LMX} Lepeltier, Matoussi and Xu
129: considered the case when $f(t,\omega ,y,z)$ satisfies (\ref{con-mono}) and
130: is Lipschitz in $z$. They proved the existence and uniqueness of the
131: solution by an approximation procedure.
132:
133: In this paper, we study the RBSDEs whose the coefficient $f\ $satisfies the
134: conditions (\ref{con-qua}) or (\ref{con-lin}), when the lower barrier $L$ is
135: uniformly bounded. We prove the existence of a solution, following the
136: methods in \cite{LS04}, and we give a necessary and sufficient condition for
137: the case when $f(t,\omega ,y,z)=\left| z\right| ^{2}$, and its explicit
138: solution.
139:
140: The paper is organized as follows: in Section 2, we present the basic
141: assumptions and the definition of the RBSDE; then in Section 3, we prove the
142: existence of a solution when $f(t,\omega ,y,z)$ satisfies the conditions (%
143: \ref{con-qua}), $\xi $ and $L$ are bounded; in the following section, we
144: consider the case when $f(t,\omega ,y,z)=\left| z\right| ^{2}$, and $\xi $
145: is not necessarily bounded. In this section, we give a necessary and
146: sufficient condition on the terminal condition $\xi $ for $p=2$ and its
147: explicit solution. Finally, in section 5, we study the RBSDE with the
148: condition (\ref{con-lin}), and prove the existence of a solution. At last,
149: in Appendix, we generalize the comparison theorem in \cite{KLQT}, and get
150: some comparison theorems, which help us to pass to the limit in the
151: approximations.
152:
153: \section{Notations}
154:
155: Let $(\Omega ,\mathcal{F},P)$ be a complete probability space, and $%
156: (B_{t})_{0\leq t\leq T}=(B_{t}^{1},B_{t}^{2},\cdots ,B_{t}^{d})_{0\leq t\leq
157: T}^{\prime }$ be a $d$-dimensional Brownian motion defined on a finite
158: interval $[0,T]$, $0<T<+\infty $. Denote by $\{\mathcal{F}_{t};0\leq t\leq
159: T\}$ the standard filtration generated by the Brownian motion $B$, i.e. $%
160: \mathcal{F}_{t}$ is the completion of
161: \[
162: \mathcal{F}_{t}=\sigma \{B_{s};0\leq s\leq t\},
163: \]
164: with respect to $(\mathcal{F},P)$. We denote by $\mathcal{P}$ the $\sigma $%
165: -algebra of predictable sets on $[0,T]\times \Omega $.
166:
167: We will need the following spaces:
168:
169: \[
170: \begin{array}{ll}
171: \mathbf{L}^{2}(\mathcal{F}_{t})= & \{\eta
172: :\mathcal{F}_{t}\mbox{-measurable
173: random real-valued variable, s.t. }E(|\eta |^{2})<+\infty \}, \\
174: \mathbf{H}_{n}^{2}(0,T)= & \{(\psi _{t})_{0\leq t\leq
175: T}:\mbox{predictable process valued in }\mathbb{R}^{n}\mbox{, s.t.
176: }E\int_{0}^{T}\left| \psi
177: (t)\right| ^{2}dt<+\infty \}, \\
178: \mathbf{S}^{2}(0,T)= & \{(\psi _{t})_{0\leq t\leq
179: T}:\mbox{progressively
180: measurable, continuous, real-valued process,} \\
181: & \mbox{s.t. }E(\sup_{0\leq t\leq T}\left| \psi (t)\right|
182: ^{2})<+\infty \},
183: \\
184: \mathbf{A}^{2}(0,T)= & \{(K_{t})_{0\leq t\leq T}:\ \mbox{adapted
185: continuous
186: increasing process, } \\
187: & \mbox{s.t. }K(0)=0\mbox{, }E(K(T)^{2})<+\infty \}.
188: \end{array}
189: \]
190:
191: Now we introduce the definition of the solution of reflected backward
192: stochastic differential equation with a terminal condition $\xi $, a
193: coefficient $f$ and a continuous reflecting lower barrier $L$(in short RBSDE$%
194: (\xi ,f,L)$), which is the same as in El Karoui et al.(1997, \cite{EKPPQ}).
195:
196: \begin{definition}
197: \label{Def}We say that the triple $(Y_{t},Z_{t},K_{t})_{0\leq t\leq T}$ of
198: progressively measurable processes is a solution of RBSDE$(\xi ,f,L)$, if
199: the followings hold:
200:
201: (i) $(Y_{t})_{0\leq t\leq T}\in \mathbf{S}^{2}(0,T)$, $(Z_{t})_{0\leq t\leq
202: T}\in \mathbf{H}_{d}^{2}(0,T)$, and $(K_{t})_{0\leq t\leq T}\in \mathbf{A}%
203: ^{2}(0,T)$.
204:
205: (ii) $Y_{t}=\xi
206: +\int_{t}^{T}f(s,Y_{s},Z_{s})ds+K_{T}-K_{t}-\int_{t}^{T}Z_{s}dB_{s},\;\;0%
207: \leq t\leq T$ a.s.
208:
209: (iii) $Y_{t}\geq L_{t},\;\;0\leq t\leq T.$
210:
211: (iv) $\int_{0}^{T}(Y_{s}-L_{s})dK_{s}=0,$ a.s.
212: \end{definition}
213:
214: \section{The general case of $f$ quadratic increasing}
215:
216: In this section, we work under the following assumptions:
217:
218: \textbf{Assumption 1.} $\xi $ is an $\mathcal{F}_{T}$-adapted and bounded
219: random variable;
220:
221: \textbf{Assumption 2.} a coefficient $f:\Omega \times [0,T]\times \mathbb{%
222: R\times R}^{d}\rightarrow \mathbb{R}$, is such that for some
223: continuous increasing function $\varphi :\mathbb{R}_{+}\rightarrow
224: \mathbb{R}_{+}$, real numbers $\mu $ and $A>0$ and $\forall
225: (t,y,y^{\prime }z)\in [0,T]\times \mathbb{R\times R\times R}^{d}$,
226:
227: $
228: \begin{array}{cl}
229: \mbox{(i)} & f(\cdot ,y,z)\mbox{ is progressively measurable;} \\
230: \mbox{(ii)} & \left| f(t,y,z)\right| \leq \varphi (\left| y\right|
231: )+A\left|
232: z\right| ^{2}\mbox{;} \\
233: \mbox{(iii)} & (y-y^{\prime })(f(t,y,z)-f(t,y^{\prime },z))\leq \mu
234: (y-y^{\prime })^{2}\mbox{;} \\
235: \mbox{(iv)} & y\rightarrow f(t,y,z)\mbox{ is continuous, a.s.}
236: \end{array}
237: $
238:
239: \textbf{Assumption 3.} a barrier $(L_{t})_{0\leq t\leq T}$, is a bounded
240: continuous progressively measurable real-valued process, $b:=\sup_{0\leq
241: t\leq T}\left| L_{t}\right| <+\infty $, $L_{T}\leq \xi $, a.s.
242:
243: Then we present our main result in this section.
244:
245: \begin{theorem}
246: \label{gmq}Under the \textbf{Assumptions} \textbf{1, 2} and \textbf{3}, RBSDE%
247: $(\xi ,f,L)$ admits a maximal bounded solution.
248: \end{theorem}
249:
250: %TCIMACRO{\TeXButton{Proof}{\proof}}
251: %BeginExpansion
252: \proof%
253: %EndExpansion
254: First, notice that $(Y,Z,K)$ is the solution of RBSDE$(\xi ,f,L)$ if and
255: only if $(Y^{b},Z^{b},K^{b})$ is the solution of the RBSDE$(\xi
256: ^{b},f^{b},L^{b})$, where
257: \[
258: (Y^{b},Z^{b},K^{b})=(Y-b,Z,K),
259: \]
260: and
261: \[
262: (\xi ^{b},f^{b}(t,y,z),L^{b})=(\xi -b,f(s,y+b,z),L-b).
263: \]
264: Notice that $(\xi ^{b},f^{b},L^{b})$ satisfies \textbf{Assumption 1, 2 }and $%
265: -2b\leq L^{b}\leq 0$. So in the following, we assume that the barrier $L$ is
266: a negative bounded process.
267:
268: For $C>0$, set $g^{C}:\mathbb{R\rightarrow R}$ be a continuous
269: function, such that $0\leq g^{C}(y)\leq 1$, $\forall y\in
270: \mathbb{R}$, and
271: \begin{eqnarray}
272: g^{C}(y) &=&1\mbox{, if }\left| y\right| \leq C, \label{gfun} \\
273: g^{C}(y) &=&0\mbox{, if }\left| y\right| \geq 2C. \nonumber
274: \end{eqnarray}
275: Denote $f^{C}(t,y,z)=g^{C}(y)f(t,y,z)$; then
276: \begin{eqnarray*}
277: \left| f^{C}(t,y,z)\right| &\leq &g^{C}(y)(\varphi (\left| y\right|
278: )+A\left| z\right| ^{2}) \\
279: &\leq &1_{[-2C,2C]}(y)(\varphi (\left| y\right| )+A\left| z\right| ^{2}) \\
280: &\leq &\varphi (2C)+A\left| z\right| ^{2}.
281: \end{eqnarray*}
282: From the theorem 1 in \cite{KLQT}, there exists a maximal solution $%
283: (Y^{C},Z^{C},K^{C})$ to the RBSDE$(\xi ,f^{C},L)$%
284: \begin{eqnarray}
285: Y_{t}^{C} &=&\xi
286: +\int_{t}^{T}g^{C}(Y_{s}^{C})f(s,Y_{s}^{C},Z_{s}^{C})ds-%
287: \int_{t}^{T}Z_{s}^{C}dB_{s}+K_{T}^{C}-K_{t}^{C}, \label{RBSDE-qb1} \\
288: Y_{t}^{C} &\geq &L_{t},\int_{0}^{T}(Y_{t}^{C}-L_{t})dK_{t}^{C}=0,\mbox{ a.e..%
289: } \nonumber
290: \end{eqnarray}
291: We choose $n\geq 2$ even, and $a\in \mathbb{R}$; applying
292: It\^{o}'s formula to $e^{at}(Y_{t}^{C})^{n}$, we have
293: \begin{eqnarray}
294: e^{at}(Y_{t}^{C})^{n} &=&e^{aT}\xi
295: ^{n}+n%
296: \int_{t}^{T}e^{as}(Y_{s}^{C})^{n-1}g^{C}(Y_{s}^{C})f(s,Y_{s}^{C},Z_{s}^{C})ds-n\int_{t}^{T}e^{as}(Y_{s}^{C})^{n-1}Z_{s}^{C}dB_{s}
297: \label{est-qb1} \\
298: &&-\frac{n(n-1)}{2}\int_{t}^{T}e^{as}(Y_{s}^{C})^{n-2}\left|
299: Z_{s}^{C}\right|
300: ^{2}ds+n\int_{t}^{T}e^{as}(Y_{s}^{C})^{n-1}dK_{s}^{C}-a%
301: \int_{t}^{T}e^{as}(Y_{s}^{C})^{n}ds. \nonumber
302: \end{eqnarray}
303: From \textbf{Assumption 2} and the fact that $n$ is even, we have
304: \begin{eqnarray*}
305: yf(s,y,z) &\leq &yf(s,0,z)+\mu y^{2}, \\
306: y^{n-1}f(s,y,z) &\leq &y^{n-1}f(s,0,z)+\mu y^{n}.
307: \end{eqnarray*}
308: With $0\leq g^{C}(y)\leq 1$, we get
309: \begin{eqnarray*}
310: g^{C}(y)y^{n-1}f(s,y,z) &\leq &g^{C}(y)\left| y\right| ^{n-1}f(s,0,z)+\mu
311: y^{n} \\
312: &\leq &g^{C}(y)\left| y\right| ^{n-1}(\varphi (0)+A\left| z\right| ^{2})+\mu
313: y^{n} \\
314: &\leq &(\frac{1}{n}+\frac{n-1}{n}\left| y\right| ^{n})\varphi (0)+A\left|
315: z\right| ^{2}g^{C}(y)\left| y\right| ^{n-1}+\mu y^{n} \\
316: &\leq &(1+y^{n})\varphi (0)+2CA\left| z\right| ^{2}y^{n-2}+\mu y^{n}.
317: \end{eqnarray*}
318: Substitute it into (\ref{est-qb1}), then
319: \begin{eqnarray*}
320: e^{at}(Y_{t}^{C})^{n} &\leq &e^{aT}\xi ^{n}+\frac{n\varphi (0)}{a}%
321: (e^{aT}-e^{at})+(n\varphi (0)+n\mu -a)\int_{t}^{T}e^{as}(Y_{s}^{C})^{n}ds \\
322: &&+(2nCA-\frac{n(n-1)}{2})\int_{t}^{T}e^{as}(Y_{s}^{C})^{n-2}\left|
323: Z_{s}^{C}\right| ^{2}ds+n\int_{t}^{T}e^{as}(L_{s})^{n-1}dK_{s}^{C} \\
324: &&-n\int_{t}^{T}e^{as}(Y_{s}^{C})^{n-1}Z_{s}^{C}dB_{s}.
325: \end{eqnarray*}
326: Notice that since $K^{C}$ is an increasing process, $n$ is even and $L\leq 0$%
327: , we get immediately
328: \[
329: \int_{t}^{T}e^{as}(L_{s})^{n-1}dK_{s}^{C}\leq 0.
330: \]
331: If we choose $n$ and $a$ satisfying
332: \[
333: n-1\geq 4CA,a=n(\varphi (0)+\mu ),
334: \]
335: then
336: \[
337: e^{at}(Y_{t}^{C})^{n}\leq e^{aT}\xi ^{n}+\frac{n\varphi (0)}{a}%
338: (e^{aT}-e^{at})-n\int_{t}^{T}e^{as}(Y_{s}^{C})^{n-1}Z_{s}^{C}dB_{s}.
339: \]
340: It follows that
341: \[
342: e^{at}(Y_{t}^{C})^{n}\leq E[e^{aT}(\xi ^{n}+\frac{n\varphi (0)}{a})|\mathcal{%
343: F}_{t}]\leq e^{aT}(\left\| \xi \right\| _{\infty }^{n}+1),
344: \]
345: at last we get
346: \[
347: (Y_{t}^{C})^{n}\leq e^{a(T-t)}(\left\| \xi \right\| _{\infty }^{n}+1)\leq
348: (e^{aT}\vee 1)(\left\| \xi \right\| _{\infty }^{n}+1).
349: \]
350: Since $a=n(\varphi (0)+\mu )$, it follows that
351: \[
352: \left| Y_{t}^{C}\right| \leq (e^{(\varphi (0)+\mu )T}\vee 1)(\left\| \xi
353: \right\| _{\infty }^{n}+1)^{\frac{1}{n}}\leq (e^{(\varphi (0)+\mu )T}\vee
354: 1)(\left\| \xi \right\| _{\infty }+1).
355: \]
356:
357: If $C$ is chosen to satisfy $C\geq (e^{(\varphi (0)+\mu )T}\vee 1)(\left\|
358: \xi \right\| _{\infty }+1)$, then we have $\left| Y_{t}^{C}\right| \leq C$,
359: which implies $g^{C}(Y_{t}^{C})=1$, for $0\leq t\leq T$. So, $%
360: (Y^{C},Z^{C},K^{C})$ is the solution of the RBSDE$(\xi ,f,L)$. $\square $
361:
362: \section{The case $f(t,y,z)=\left| z\right| ^{2}$}
363:
364: In this section we consider the case $f(t,y,z)=\left| z\right| ^{2}$, which
365: corresponds to the RBSDE
366: \begin{eqnarray}
367: Y_{t} &=&\xi +\int_{t}^{T}\left| Z_{s}\right|
368: ^{2}ds+K_{T}-K_{t}-\int_{t}^{T}Z_{s}dB_{s}, \label{RBSDE-p21} \\
369: Y_{t} &\geq &L_{t},\int_{0}^{T}(Y_{t}-L_{t})dK_{t}=0. \nonumber
370: \end{eqnarray}
371: Then we have
372:
373: \begin{theorem}
374: Under the assumption $E(\sup_{0\leq t\leq T}e^{2L_{t}})<+\infty $, the RBSDE$%
375: (\xi ,f,L)$ (\ref{RBSDE-p21}) admits a solution if and only if $E(e^{2\xi
376: })<+\infty $.
377: \end{theorem}
378:
379: %TCIMACRO{\TeXButton{Proof}{\proof}}
380: %BeginExpansion
381: \proof%
382: %EndExpansion
383: For the necessary part, let $(Y,Z,K)$ be a solution of the RBSDE (\ref
384: {RBSDE-p21}). By It\^{o}'s formula, we get
385: \begin{eqnarray}
386: e^{2Y_{t}} &=&e^{2\xi
387: }+2\int_{t}^{T}e^{2Y_{s}}dK_{s}-2\int_{t}^{T}e^{Y_{s}}Z_{s}dB_{s}
388: \label{RBSDE-p22} \\
389: &=&e^{2Y_{0}}+2\int_{0}^{t}e^{2Y_{s}}Z_{s}dB_{s}-2%
390: \int_{0}^{t}e^{2Y_{s}}dK_{s}. \nonumber
391: \end{eqnarray}
392: Let for all $n$, $\tau _{n}=\inf \{t:Y_{t}\geq n\}\wedge T$, then $%
393: M_{t\wedge \tau _{n}}=2\int_{0}^{t\wedge \tau _{n}}e^{2Y_{s}}Z_{s}dB_{s}$ is
394: a martingale, and we have
395: \[
396: E[e^{2Y_{\tau _{n}}}]=E[e^{2Y_{0}}-2\int_{0}^{t}e^{2Y_{s}}dK_{s}]\leq
397: E[e^{2Y_{0}}],
398: \]
399: in view of $2\int_{0}^{t}e^{2Y_{s}}dK_{s}\geq 0$. Finally, since $\tau
400: _{n}\nearrow T$, when $n\rightarrow \infty $:
401: \[
402: E[\underline{\lim }_{n\rightarrow \infty }e^{2Y_{\tau _{n}}}]=E[e^{2\xi
403: }]\leq E[e^{2Y_{0}}]<\infty ,
404: \]
405: follows from Fatou's Lemma.
406:
407: Now we suppose $E(e^{2\xi })<+\infty $, set $\widetilde{L}%
408: _{t}=L_{t}1_{\{t<T\}}+\xi 1_{\{t=T\}}$ and
409: \[
410: N_{t}=S_{t}(e^{2\widetilde{L}})=ess\sup_{\tau \in \mathcal{T}_{t,T}}E[e^{2%
411: \widetilde{L}_{\tau }}|\mathcal{F}_{t}],
412: \]
413: where $S_{t}(\eta )$ denotes the Snell envelope of $\eta $ (See El Karoui
414: \cite{E79}), $\mathcal{T}_{t,T}$ is the set of all stopping times valued in $%
415: [t,T]$. Since
416: \[
417: E[\sup_{0\leq t\leq T}e^{2\widetilde{L}_{t}}]\leq E[\sup_{0\leq t\leq
418: T}e^{2L_{t}}+e^{2\xi }]<+\infty ,
419: \]
420: using the results of Snell envelope, we know that $N$ is a supermartingale,
421: so it admits the following decomposition: for an increasing integrable
422: process $\overline{K}$,
423: \[
424: N_{t}=N_{0}+\int_{0}^{t}\overline{Z}_{s}dB_{s}-\overline{K}_{t}.
425: \]
426: Applying It\^{o}'s formula to $\log N_{t}$, we get
427: \[
428: \frac{1}{2}\log N_{t}=\frac{1}{2}\log N_{0}+\frac{1}{2}\int_{0}^{t}\frac{%
429: \overline{Z}_{s}}{N_{s}}dB_{s}-\frac{1}{4}\int_{0}^{t}(\frac{\overline{Z}_{s}%
430: }{N_{s}})^{2}ds-\frac{1}{2}\int_{0}^{t}\frac{1}{N_{s}}d\overline{K}_{s}.
431: \]
432: Set $Y_{t}=\frac{1}{2}\log N_{t}$, $Z_{t}=\frac{\overline{Z}_{t}}{2N_{t}}$, $%
433: K_{t}=\frac{1}{2}\int_{0}^{t}\frac{1}{N_{s}}d\overline{K}_{s}$, then the
434: triple satisfies
435: \begin{equation}
436: Y_{t}=\xi +\int_{t}^{T}Z_{s}^{2}ds+K_{T}-K_{t}-\int_{t}^{T}Z_{s}dB_{s}.
437: \label{RBSDE-qb-s}
438: \end{equation}
439: Thanks to the results on the Snell envelope, we know that $N_{t}\geq e^{2%
440: \widetilde{L}_{t}}$ and $\int_{0}^{T}(N_{t}-e^{2\widetilde{L}_{t}})d%
441: \overline{K}_{t}=0$. The first implies
442: \[
443: Y_{t}\geq \widetilde{L}_{t}\geq L_{t}.
444: \]
445: Obviously, $N_{t}>0$, $0\leq t\leq T$, so $K$ is increasing. Consider the
446: stopping time $D_{t}:=\inf \{t\leq u\leq T;Y_{u}=L_{u}\}\wedge T$, then it
447: satisfies $D_{t}=\inf \{t\leq u\leq T;N_{u}=e^{2L_{u}}\}\wedge T$. By the
448: continuity of $\overline{K}$, we get $\overline{K}_{D_{t}}-\overline{K}_{t}=0
449: $, which implies $K_{D_{t}}-K_{t}=0$. It follows that
450: \[
451: \int_{0}^{T}(Y_{t}-L_{t})dK_{t}=0.
452: \]
453: Now the rest is to prove $Y_{t}\in \mathbf{S}^{2}(0,T)$, $Z_{t}\in \mathbf{H}%
454: _{d}^{2}(0,T)$, and $K_{t}\in \mathbf{A}^{2}(0,T)$. With Jensen's inequality
455: \begin{eqnarray*}
456: Y_{t} &=&\frac{1}{2}\log N_{t}=\frac{1}{2}\log [ess\sup_{\tau \in \mathcal{T}%
457: _{t,T}}E[e^{2\widetilde{L}_{\tau }}|\mathcal{F}_{t}]] \\
458: &\geq &\frac{1}{2}\log [\exp (ess\sup_{\tau \in \mathcal{T}_{t,T}}E[2%
459: \widetilde{L}_{\tau }|\mathcal{F}_{t}])] \\
460: &=&ess\sup_{\tau \in \mathcal{T}_{t,T}}E[\widetilde{L}_{\tau }|\mathcal{F}%
461: _{t}]\geq E[\xi |\mathcal{F}_{t}]\geq U_{t},
462: \end{eqnarray*}
463: where $U_{t}=-E[\xi ^{-}|\mathcal{F}_{t}]$. For all $a>0$, define
464: \[
465: \tau _{a}=\inf \{t;\left| N_{t}\right| >a,\int_{0}^{t}(\frac{\overline{Z}_{s}%
466: }{N_{s}})^{2}ds>a,\left| \int_{0}^{t}\frac{\overline{Z}_{s}}{N_{s}}%
467: dB_{s}\right| >a\}.
468: \]
469: From (\ref{RBSDE-qb-s}), we get for $0\leq t\leq T$%
470: \begin{eqnarray*}
471: 0 &\leq &\int_{0}^{t}Z_{s}^{2}ds=Y_{0}-Y_{t}+\int_{0}^{t}Z_{s}dB_{s}-K_{t} \\
472: &\leq &Y_{0}-U_{t}+\int_{0}^{t}Z_{s}dB_{s}.
473: \end{eqnarray*}
474: Then
475: \[
476: (\int_{0}^{\tau _{a}}Z_{s}^{2}ds)^{2}\leq 3(Y_{0})^{2}+3(U_{\tau
477: _{a}})^{2}+3(\int_{0}^{\tau _{a}}Z_{s}dB_{s})^{2}.
478: \]
479: Taking the expectation, using the Jensen's inequality and $3x\leq \frac{x^{2}%
480: }{2}+\frac{9}{2}$, we obtain
481: \begin{eqnarray*}
482: E(\int_{0}^{\tau _{a}}Z_{s}^{2}ds)^{2} &\leq &\frac{3}{4}(\log
483: N_{0})^{2}+3E(\xi ^{-})^{2}+\frac{1}{2}(E(\int_{0}^{\tau
484: _{a}}Z_{s}^{2}ds))^{2}+\frac{9}{2} \\
485: &\leq &\frac{3}{4}(\log N_{0})^{2}+3E(\xi ^{-})^{2}+\frac{1}{2}%
486: E(\int_{0}^{\tau _{a}}Z_{s}^{2}ds)^{2}+\frac{9}{2},
487: \end{eqnarray*}
488: so
489: \[
490: E(\int_{0}^{\tau _{a}}Z_{s}^{2}ds)^{2}\leq \frac{3}{2}(\log
491: N_{0})^{2}+6E(\xi ^{-})^{2}+9\leq C.
492: \]
493: Since $\tau _{a}\nearrow T$ when $a\rightarrow +\infty $, we get to the
494: limit, and with the Schwartz inequality
495: \[
496: E\int_{0}^{T}Z_{s}^{2}ds\leq (E(\int_{0}^{T}Z_{s}^{2}ds)^{2})^{\frac{1}{2}%
497: }\leq C.
498: \]
499: So $Z\in \mathbf{H}_{d}^{2}(0,T)$. From (\ref{RBSDE-qb-s}), we get for $%
500: 0\leq t\leq T$%
501: \begin{eqnarray*}
502: 0 &\leq &K_{t}=Y_{0}-Y_{t}+\int_{0}^{t}Z_{s}dB_{s}-\int_{0}^{t}Z_{s}^{2}ds \\
503: &\leq &Y_{0}-Y_{t}+\int_{0}^{t}Z_{s}dB_{s}.
504: \end{eqnarray*}
505: Notice that $K$ is increasing, so it's sufficient to prove $%
506: E[K_{T}^{2}]<+\infty $. Squaring the inequality on both sides and taking
507: expectation, we obtain
508: \[
509: E[(K_{T})^{2}]\leq 3Y_{0}^{2}+3E[\xi ^{2}]+3E\int_{0}^{T}Z_{s}^{2}ds\leq C.
510: \]
511: We consider now $Y$; again from (\ref{RBSDE-qb-s}),
512: \[
513: Y_{t}=Y_{0}-K_{t}+\int_{0}^{t}Z_{s}dB_{s}-\int_{0}^{t}Z_{s}^{2}ds,
514: \]
515: so
516: \[
517: (Y_{t})^{2}\leq 4\left( Y_{0}\right) ^{2}+4\left( K_{t}\right) ^{2}+4\left(
518: \int_{0}^{t}Z_{s}dB_{s}\right) ^{2}+4\left( \int_{0}^{t}Z_{s}^{2}ds\right)
519: ^{2}.
520: \]
521: Then by the Bukholder-Davis-Gundy inequality, we get
522: \begin{eqnarray*}
523: E[\sup_{0\leq t\leq T}(Y_{t})^{2}] &\leq &4\left( Y_{0}\right)
524: ^{2}+4E[K_{T}^{2}]+4E[\sup_{0\leq t\leq T}\left(
525: \int_{0}^{t}Z_{s}dB_{s}\right) ^{2}]+4E\left( \int_{0}^{T}Z_{s}^{2}ds\right)
526: ^{2} \\
527: &\leq &4\left( Y_{0}\right) ^{2}+4E[K_{T}^{2}]+CE\left(
528: \int_{0}^{t}Z_{s}^{2}dB_{s}\right) +4E\left( \int_{0}^{T}Z_{s}^{2}ds\right)
529: ^{2}\leq C,
530: \end{eqnarray*}
531: i.e. $Y\in \mathbf{S}^{2}(0,T)$. $\square $
532:
533: \section{The case when $f$ is linear increasing in $z$}
534:
535: In this section, we assume that the coefficient $f$ satisfies
536:
537: \textbf{Assumption 6. }(i) $f(\cdot ,y,z)$ is progressively measurable, and $%
538: E\int_{0}^{T}f^{2}(t,0,0)dt<+\infty $;
539:
540: (ii) for $\mu \in \mathbb{R}$, $\forall (t,z)\in [0,T]\times
541: \mathbb{R}^{d}$
542: and $y,y^{\prime }\in \mathbb{R,}$%
543: \[
544: (y-y^{\prime })(f(t,y,z)-f(t,y^{\prime },z))\leq \mu (y-y^{\prime })^{2};
545: \]
546:
547: (iii) there exists a nonegative, continuous, increasing function $\varphi :%
548: \mathbb{R}^{+}\rightarrow \mathbb{R}^{+}$, with $\varphi (0)=0$, s.t. $%
549: \forall (t,y,z)\in [0,T]\times \mathbb{R\times R}^{d},$%
550: \[
551: \left| f(t,y,z)\right| \leq \left| g_{t}\right| +\varphi (\left| y\right|
552: )+\beta \left| z\right| ,
553: \]
554: where $g_{t}\in \mathbf{H}^{2}(0,T)$;
555:
556: (iv) for $t\in [0,T]$, $(y,z)\rightarrow f(t,y,z)$ is continuous.
557:
558: If $\varphi (x)=\left| x\right| $, then $f$ is linear increasing in $y$ and $%
559: z$. Matoussi proved in \cite{M97} that when $\xi \in \mathbf{L}^{2}(\mathcal{%
560: F}_{T})$ and $L\in \mathbf{S}^{2}(0,T)$, there exists a triple $(Y,Z,K)$
561: which is solution of the RBSDE$(\xi ,f,L)$.
562:
563: Our result of this section is the following:
564:
565: \begin{theorem}
566: \label{exist-zl}Suppose that $\xi \in \mathbf{L}^{2}(\mathcal{F}_{T})$, $f$
567: and $L$ satisfy \textbf{Assumption} \textbf{6} and \textbf{3}, respectively,
568: then the RBSDE$(\xi ,f,L)$ has a minimal solution $(Y,Z,K)\in \mathbf{S}%
569: ^{2}(0,T)\times \mathbf{H}_{d}^{2}(0,T)\times \mathbf{A}^{2}(0,T)$, which
570: satisfies
571: \[
572: Y_{t}=\xi
573: +\int_{t}^{T}f(s,Y_{s},Z_{s})ds+K_{T}-K_{t}-\int_{t}^{T}Z_{s}dB_{s},
574: \]
575: $Y_{t}\geq L_{t}$, and $\int_{0}^{T}(Y_{s}-L_{s})dK_{s}=0$.
576: \end{theorem}
577:
578: First we note that the triple $(Y,Z,K)$ solves the RBSDE$(\xi ,f,L)$, if and
579: only if the triple
580: \begin{equation}
581: (\overline{Y}_{t},\overline{Z}_{t},\overline{K}_{t}):=(e^{\lambda
582: t}Y_{t},e^{\lambda t}Z_{t},\int_{0}^{t}e^{\lambda s}dK_{s}) \label{trans}
583: \end{equation}
584: solves the RBSDE$(\overline{\xi },\overline{f},\overline{L})$, where
585: \[
586: (\overline{\xi },\overline{f}(t,y,z),\overline{L}_{t})=(\xi e^{\lambda
587: T},e^{\lambda t}f(t,e^{-\lambda t}y,e^{-\lambda t}z)-\lambda y,e^{\lambda
588: t}L_{t}).
589: \]
590:
591: If we choose $\lambda =\mu $, then the coefficient $\overline{f}$ satisfies
592: the same assumptions as in \textbf{Assumption 6}, with (ii) replaced by
593:
594: (ii') $(y-y^{\prime })(f(t,y,z)-f(t,y^{\prime },z))\leq 0$.
595:
596: Since we are in the $1$-dimensional case, (ii') means that $f$ is decreasing
597: on $y$. From another part $\overline{\xi }$ still belongs to $\mathbf{L}^{2}(%
598: \mathcal{F}_{T})$ and the barrier $\overline{L}$ still satisfy the
599: assumptions \textbf{Assumption} \textbf{3}. So\textbf{\ }in the following,
600: we shall work under \textbf{Assumption 6' }with (ii) replaced by (ii').
601:
602: Before proving this theorem, we consider an estimate result and a monotonic
603: stability theorem for RBSDEs.
604:
605: \begin{lemma}
606: \label{est-res}We consider RBSDE$(\xi ,g,L)$, with $\xi \in \mathbf{L}^{2}(%
607: \mathcal{F}_{T})$, $g$ and $L$ satisfy \textbf{Assumption} \textbf{6'} and
608: \textbf{3}. Moreover $g(t,y,z)$ is Lipschitz in $z$. Then we have the
609: following estimation
610: \begin{eqnarray*}
611: &&E[\sup_{0\leq t\leq T}\left| y_{t}\right| ^{2}+\int_{0}^{T}\left|
612: z_{s}\right| ds+\left| k_{T}\right| ^{2}] \\
613: &\leq &C_{\beta }E[\left| \xi \right| ^{2}+\int_{0}^{T}g_{s}^{2}ds+\varphi
614: ^{2}(b)+\varphi ^{2}(2T)+1]
615: \end{eqnarray*}
616: where $(y_{t},z_{t},k_{t})_{0\leq t\leq T}$ is the solution of RBSDE$(\xi
617: ,g,L)$. $C_{\beta }$ is a constant only depends on $\beta $, $T$ and $b$.
618: \end{lemma}
619:
620: \begin{remark}
621: The constant $C_{\beta }$ does not depend on Lipschitz coefficient of $g$ on
622: $z$.
623: \end{remark}
624:
625: %TCIMACRO{\TeXButton{Proof}{\proof}}
626: %BeginExpansion
627: \proof%
628: %EndExpansion
629: Since $g\,$is Lipschitz in $z$, by the theorem 2 in \cite{LMX}, the RBSDE$%
630: (\xi ,g,L)$ admits the unique solution $(y_{t},z_{t},k_{t})_{0\leq t\leq T}$%
631: . Apply It\^{o}'s formula to $\left| y_{t}\right| ^{2}$, in view of $%
632: yg(t,y,z)\leq g(t,0,0)\left| y\right| +\beta \left| y\right| \left| z\right|
633: $ and $\sup_{0\leq t\leq T}\left| L_{t}\right| \leq b$, we get
634: \begin{eqnarray*}
635: E[\left| y_{t}\right| ^{2}+\int_{t}^{T}\left| z_{s}\right| ^{2}ds]
636: &=&E[\left| \xi \right|
637: ^{2}+2\int_{t}^{T}y_{s}g(s,y_{s},z_{s})ds+2\int_{t}^{T}L_{s}dk_{s}] \\
638: &\leq &E[\left| \xi \right| ^{2}+2\int_{t}^{T}y_{s}g_{s}ds+2\beta
639: \int_{t}^{T}y_{s}z_{s}ds+2b(k_{T}-k_{t})].
640: \end{eqnarray*}
641: It follows that
642: \begin{eqnarray*}
643: &&E[\left| y_{t}\right| ^{2}+\frac{1}{2}\int_{t}^{T}\left| z_{s}\right|
644: ^{2}ds] \\
645: &\leq &E[\left| \xi \right| ^{2}+\int_{t}^{T}g_{s}^{2}ds+(1+2\beta
646: ^{2})\int_{t}^{T}\left| y_{s}\right| ^{2}ds+2b(k_{T}-k_{t})].
647: \end{eqnarray*}
648: By Gronwall's inequality, we know there exists a constant $c_{1}$ depending
649: on $\beta $ and $T$, such that for $t\in [0,T]$,
650: \begin{equation}
651: E[\left| y_{t}\right| ^{2}]\leq c_{1}E[\left| \xi \right|
652: ^{2}+\int_{0}^{T}g_{s}^{2}ds+b(k_{T}-k_{t})]. \label{est-y1}
653: \end{equation}
654: It follows that
655: \begin{equation}
656: E[\int_{t}^{T}\left| z_{s}\right| ^{2}ds]\leq 2(1+(1+2\beta
657: ^{2})T)c_{1}E[\left| \xi \right|
658: ^{2}+\int_{0}^{T}g_{s}^{2}ds+b(k_{T}-k_{t})]. \label{est-z1}
659: \end{equation}
660:
661: Now we estimate the increasing process $k$ by approximation. Take $z$ as a
662: known process, without losing of generality, we write $g(t,y)$ for $%
663: g(t,y,z_{t})$, here $g(t,0)=g(t,0,z_{t})$ is a process in $\mathbf{H}%
664: ^{2}(0,T)$ in view of linear increasing property of $g$ on $z$.
665:
666: For $m$, $p\in \mathbb{N}$, set $\xi ^{m,p}=(\xi \vee (-p))\wedge m$, $%
667: g^{m,p}(t,u)=g(t,u)-g_{t}+(g_{t}\vee (-p))\wedge m$. We consider RBSDE$(\xi
668: ^{m,p},g^{m,p},L)$,
669: \begin{eqnarray}
670: y_{t}^{m,p} &=&\xi
671: ^{m,p}+\int_{t}^{T}g^{m,p}(s,y_{s}^{m,p})ds+k_{T}^{m,p}-k_{t}^{m,p}-%
672: \int_{t}^{T}z_{s}^{m,p}dB_{s}, \label{appro} \\
673: y_{t}^{m,p} &\geq &L_{t},\int_{0}^{T}(y_{t}^{m,p}-L_{t})dk_{t}^{m,p}=0.
674: \nonumber
675: \end{eqnarray}
676: It is easy to check that $(y^{m,p},z^{m,p},k^{m,p})$ is the solution of RBSDE%
677: $(\xi ^{m,p},g^{m,p},L)$, if and only if $(\widehat{y}^{m,p},\widehat{z}%
678: ^{m,p},\widehat{k}^{m,p})$ is the solution of RBSDE$(\widehat{\xi }^{m,p},%
679: \widehat{g}^{m,p},\widehat{L})$, where
680: \[
681: (\widehat{y}_{t}^{m,p},\widehat{z}_{t}^{m,p},\widehat{k}%
682: _{t}^{m,p})=(y_{t}^{m,p}+m(t-2(T\vee 1)),z_{t}^{m,p},k_{t}^{m,p}),
683: \]
684: and
685: \begin{eqnarray*}
686: \widehat{\xi }^{m,p} &=&\xi ^{m,p}+mT-2m(T\vee 1), \\
687: \widehat{g}^{m,p}(t,y) &=&g^{m,p}(t,y-m(t-2(T\vee 1)))-m, \\
688: \widehat{L}_{t} &=&L_{t}+m(t-2(T\vee 1)).
689: \end{eqnarray*}
690: Without losing of generality, we set $T\geq 1$. Since $\xi ^{m,p}$ and $%
691: g_{t}^{m,p}\leq m$, we have $\widehat{\xi }^{m,p}$ and $\widehat{g}%
692: _{t}^{m,p}\leq 0$. By (\ref{appro}),
693: \[
694: \widehat{k}_{T}^{m,p}-\widehat{k}_{t}^{m,p}=\widehat{y}_{t}^{m,p}-\widehat{%
695: \xi }^{m,p}-\int_{t}^{T}\widehat{g}^{m,p}(s,\widehat{y}_{s}^{m,p})ds+%
696: \int_{t}^{T}\widehat{z}_{s}^{m,p}dB_{s},
697: \]
698: taking square and expectation on the both sides, we get
699: \begin{equation}
700: E[(\widehat{k}_{T}^{m,p}-\widehat{k}_{t}^{m,p})^{2}]\leq 4E[(\widehat{y}%
701: _{t}^{m,p})^{2}+(\widehat{\xi }^{m,p})^{2}+(\int_{t}^{T}\widehat{g}^{m,p}(s,%
702: \widehat{y}_{s}^{m,p})ds)^{2}+\int_{t}^{T}\left| \widehat{z}%
703: _{s}^{m,p}\right| ^{2}ds]. \label{est-k1}
704: \end{equation}
705:
706: In order to estimate the first and the last form on the left side, we apply
707: It\^{o}'s formula to $\left| \widehat{y}_{t}^{m,p}\right| ^{2}$, and get the
708: following with Gronwall inequality,
709: \begin{eqnarray}
710: &&E[\left| \widehat{y}_{t}^{m,p}\right| ^{2}+\int_{t}^{T}\left| \widehat{z}%
711: _{s}^{m,p}\right| ^{2}ds] \label{est-yz1} \\
712: &\leq &c_{2}E[\left| \widehat{\xi }^{m,p}\right| ^{2}+\int_{t}^{T}(\widehat{g%
713: }_{s}^{m,p})^{2}ds+\int_{t}^{T}L_{s}d\widehat{k}_{s}^{m,p}], \nonumber
714: \end{eqnarray}
715: where $c_{2}$ is a constant only depends on $T$. For the third term, let us
716: recall a comparison result of $\widehat{y}_{t}^{m,p}$ in step 2 of the proof
717: of theorem 2 in \cite{LMX},
718: \[
719: \widetilde{y}_{t}^{m,p}\leq \widehat{y}_{t}^{m,p}\leq \overline{y}%
720: _{t}^{m,p},
721: \]
722: where $\widetilde{y}_{t}^{m,p}$ is the solution of BSDE$(\widehat{\xi }%
723: ^{m,p},\widehat{g}^{m,p})$, i.e.
724: \begin{equation}
725: \widetilde{y}_{t}^{m,p}=\widehat{\xi }^{m,p}+\int_{t}^{T}\widehat{g}^{m,p}(s,%
726: \widetilde{y}_{s}^{m,p})ds-\int_{t}^{T}\widetilde{z}_{s}^{m,p}dB_{s},
727: \label{est-bsde}
728: \end{equation}
729: and
730: \[
731: \overline{y}_{t}^{m,p}=ess\sup_{\tau \in \mathcal{T}_{t,T}}E[(\widehat{L}%
732: _{\tau })^{+}1_{\{\tau <T\}}+(\widehat{\xi }^{m,p})^{+}1_{\{\tau =T\}}|%
733: \mathcal{F}_{t}],
734: \]
735: where $\mathcal{T}_{t,T}$ is the set of stoppng times valued in $[t,T]$.
736: Moreover, we have $\sup_{0\leq s\leq T}\overline{y}_{s}^{m,p}=\sup_{0\leq
737: s\leq T}\widehat{L}_{s}$.
738:
739: Since $\widehat{g}^{m,p}$ is decreasing in $y$, we get
740: \[
741: \widehat{g}^{m,p}(s,\overline{y}_{t}^{m,p})\leq \widehat{g}^{m,p}(s,\widehat{%
742: y}_{s}^{m,p})\leq \widehat{g}^{m,p}(s,\widetilde{y}_{s}^{m,p}).
743: \]
744: So to estimate $E[(\int_{t}^{T}\widehat{g}^{m,p}(s,\widehat{y}%
745: _{s}^{m,p})ds)^{2}]$, it is sufficante to get the estimations of $%
746: E[(\int_{t}^{T}\widehat{g}^{m,p}(s,\widetilde{y}_{s}^{m,p})ds)^{2}]$ and $%
747: E[(\int_{t}^{T}\widehat{g}^{m,p}(s,\overline{y}_{s}^{m,p})ds)^{2}]$. First
748: we know that
749: \begin{eqnarray}
750: E[(\int_{t}^{T}\widehat{g}^{m,p}(s,\widetilde{y}_{s}^{m,p})ds)^{2}] &\leq
751: &3E[\left| \widehat{\xi }^{m,p}\right| ^{2}+\left| \widetilde{y}%
752: _{t}^{m,p}\right| ^{2}+\int_{t}^{T}\left| \widetilde{z}_{s}^{m,p}\right|
753: ^{2}ds] \label{est-g1} \\
754: &\leq &c_{3}E[\left| \widehat{\xi }^{m,p}\right| ^{2}+\int_{t}^{T}(\widehat{g%
755: }_{s}^{m,p})^{2}ds], \nonumber
756: \end{eqnarray}
757: in view of estimate result of BSDE(\ref{est-bsde}). Here $c_{3}$ is a
758: constant only depends on $T$. Then with the presentation of $\overline{y}%
759: _{t}^{m,p}$, we have
760: \begin{equation}
761: E[(\int_{t}^{T}\widehat{g}^{m,p}(s,\overline{y}_{s}^{m,p})ds)^{2}]\leq
762: E[2T\int_{0}^{T}(\widehat{g}_{s}^{m,p})^{2}ds+2T\varphi ^{2}(\sup_{0\leq
763: t\leq T}(\widehat{L}_{t})^{+})] \label{est-g2}
764: \end{equation}
765: From (\ref{est-k1}), with (\ref{est-yz1}), (\ref{est-g1}) and (\ref{est-g2}%
766: ), we have
767: \begin{eqnarray*}
768: E[(k_{T}^{m,p}-k_{t}^{m,p})^{2}] &=&E[(\widehat{k}_{T}^{m,p}-\widehat{k}%
769: _{t}^{m,p})^{2}] \\
770: &\leq &c_{4}E[\left| \widehat{\xi }^{m,p}\right| ^{2}+\int_{t}^{T}(\widehat{g%
771: }_{s}^{m,p})^{2}ds+\int_{t}^{T}L_{s}d\widehat{k}_{s}^{m,p}+\varphi
772: ^{2}(\sup_{0\leq t\leq T}(\widehat{L}_{t})^{+})] \\
773: &\leq &c_{4}E[2\left| \xi ^{m,p}\right|
774: ^{2}+4\int_{t}^{T}(g_{s}^{m,p})^{2}ds+2c_{4}b^{2}+\varphi ^{2}(b)] \\
775: &&+4c_{4}(m^{2}T^{2}+\varphi ^{2}(2mT))+\frac{1}{2}%
776: E[(k_{T}^{m,p}-k_{t}^{m,p})^{2}],
777: \end{eqnarray*}
778: where $c_{4}=c_{2}\vee c_{3}\vee (2T)$, which only depends on $T$. It
779: follows that
780: \begin{eqnarray*}
781: &&E[(k_{T}^{m,p}-k_{t}^{m,p})^{2}] \\
782: &\leq &c_{5}E[\left| \xi ^{m,p}\right|
783: ^{2}+\int_{t}^{T}(g_{s}^{m,p})^{2}ds+b^{2}+\varphi
784: ^{2}(b)]+c_{5}(m^{2}T^{2}+\varphi ^{2}(2mT)) \\
785: &\leq &c_{5}E[\left| \xi \right| ^{2}+\int_{t}^{T}g_{s}^{2}ds+b^{2}+\varphi
786: ^{2}(b)]+c_{5}(m^{2}T^{2}+\varphi ^{2}(2mT)),
787: \end{eqnarray*}
788: where $c_{5}=4c_{4}^{2}\vee 8c_{4}$.
789:
790: Now we consider the RBSDE$(\xi ^{p},g^{p},L)$, where $\xi ^{p}=\xi \vee (-p)$%
791: , $g^{p}(t,u)=g(t,u)-g_{t}+g_{t}\vee (-p)$. Thanks to the convergence result
792: in \cite{LMX}, we know that
793: \[
794: (y^{m,p},z^{m,p},k^{m,p})\rightarrow (y^{p},z^{p},k^{p})\mbox{ in }\mathbf{S}%
795: ^{2}(0,T)\times \mathbf{H}_{d}^{2}(0,T)\times \mathbf{A}^{2}(0,T),
796: \]
797: where $(y^{p},z^{p},k^{p})$ is the solution of RBSDE$(\xi ^{p},g^{p},L)$.
798: Moreover, we have $dk_{t}^{p}\leq dk_{t}^{1,p}$, by comparison theorem. So
799: \begin{eqnarray*}
800: E[(k_{T}^{p}-k_{t}^{p})^{2}] &\leq &E[(k_{T}^{1,p}-k_{t}^{1,p})^{2}] \\
801: &\leq &c_{5}E[\left| \xi \right| ^{2}+\int_{t}^{T}g_{s}^{2}ds+b^{2}+\varphi
802: ^{2}(b)]+c_{5}(T^{2}+\varphi ^{2}(2T)).
803: \end{eqnarray*}
804: Then let $p\rightarrow \infty $, thanks to the convergence result in \cite
805: {LMX}, we know
806: \[
807: (y^{p},z^{p},k^{p})\rightarrow (y,z,k)\mbox{ in
808: }\mathbf{S}^{2}(0,T)\times \mathbf{H}_{d}^{2}(0,T)\times
809: \mathbf{A}^{2}(0,T).
810: \]
811: In view \textbf{Assumption 6}-(iii), it follows that
812: \begin{eqnarray*}
813: E[(k_{T}-k_{t})^{2}] &\leq &c_{5}E[\left| \xi \right|
814: ^{2}+\int_{t}^{T}(g(s,0,z_{s}))^{2}ds+b^{2}+\varphi ^{2}(b)] \\
815: &\leq &c_{5}E[\left| \xi \right| ^{2}+2\int_{t}^{T}g_{s}^{2}ds+2\beta
816: ^{2}\int_{t}^{T}\left| z_{s}\right| ^{2}ds+b^{2}+\varphi ^{2}(b)] \\
817: &&+c_{5}(T^{2}+\varphi ^{2}(2T)).
818: \end{eqnarray*}
819: With (\ref{est-z1}), setting $c_{6}=c_{5}\vee (4\beta ^{2}(1+(1+2\beta
820: ^{2})T)c_{1}+2)\vee c_{5}(b^{2}+T^{2})$, we get
821: \begin{eqnarray*}
822: E[(k_{T}-k_{t})^{2}] &\leq &c_{6}E[\left| \xi \right|
823: ^{2}+\int_{t}^{T}g_{s}^{2}ds+b(k_{T}-k_{t})+\varphi ^{2}(b)+\varphi
824: ^{2}(2T)+1] \\
825: &\leq &c_{6}E[\left| \xi \right|
826: ^{2}+\int_{t}^{T}g_{s}^{2}ds+2c_{6}b^{2}+\varphi ^{2}(b)+\varphi ^{2}(2T)+1]
827: \\
828: &&+\frac{1}{2}E[(k_{T}-k_{t})^{2}]
829: \end{eqnarray*}
830: It follows that
831: \[
832: E[(k_{T}-k_{t})^{2}]\leq 2c_{6}E[\left| \xi \right|
833: ^{2}+2\int_{t}^{T}g_{s}^{2}ds+\varphi ^{2}(b)+\varphi
834: ^{2}(2T)+2c_{6}b^{2}+1].
835: \]
836: Consequantly, by (\ref{est-y1}) and (\ref{est-z1}), we obtain
837: \begin{eqnarray*}
838: &&\sup_{0\leq t\leq T}E[\left| y_{t}\right| ^{2}]+E[\int_{0}^{T}\left|
839: z_{s}\right| ds+\left| k_{T}\right| ^{2}] \\
840: &\leq &C_{\beta }E[\left| \xi \right| ^{2}+\int_{0}^{T}g_{s}^{2}ds+\varphi
841: ^{2}(2T)+\varphi ^{2}(b)+1],
842: \end{eqnarray*}
843: where $C_{\beta }$ is a constant only depends on $\beta $, $T$ and $b$. The
844: final result follows from BDG inequality. $\square $
845:
846: The proof of this theorem is step 1 and step 2 of the proof of theorem 4 in
847: \cite{KLQT}, with comparison theorem. So we omit it.
848:
849: With these preparations, we begin our main proof.
850:
851: \textbf{Proof of theorem \ref{exist-zl}. }The proof consists 4 step.
852:
853: \textbf{Step 1. }Approximation. For $n\geq \beta $, we introduce the
854: following functions
855: \[
856: f_{n}(t,y,z)=\inf_{q\in \mathbf{Q}^{d}}\{f(t,y,q)+n\left| z-q\right| \},
857: \]
858: then we have
859:
860: 1. for all $(t,z)$, $y\rightarrow f_{n}(t,y,z)$ is non-increasing;
861:
862: 2. for all $(t,y)$, $z\rightarrow f_{n}(t,y,z)$ is $n$-Lipschitz;
863:
864: 3. for all $(t,y,z)$, $\left| f_{n}(t,y,z)\right| \leq \left| g_{t}\right|
865: +\varphi (\left| y\right| )+\beta \left| z\right| .$
866:
867: Thanks to the results of \cite{LMX}, we know that for each $n\geq \beta $,
868: there exits a unique triple $(Y^{n},Z^{n},K^{n})$ satisfies the followings
869: \begin{eqnarray*}
870: Y_{t}^{n} &=&\xi
871: +\int_{t}^{T}f_{n}(s,Y_{s}^{n},Z_{s}^{n})ds+K_{T}^{n}-K_{t}^{n}-%
872: \int_{t}^{T}Z_{s}^{n}dB_{s}, \\
873: Y_{t}^{n} &\geq &L_{t},\int_{0}^{T}(Y_{t}^{n}-L_{t})dK_{t}^{n}=0.
874: \end{eqnarray*}
875:
876: \textbf{Step 2}. Estimates results. Let $\alpha \geq 0$, be a real number to
877: be chosen later. We set
878: \[
879: U_{t}^{n}=e^{\alpha t}Y_{t}^{n},V_{t}^{n}=e^{\alpha
880: t}Z_{t}^{n},dJ_{t}^{n}=e^{\alpha t}dK_{t}^{n}.
881: \]
882: Then we know that $(U^{n},V^{n},J^{n})$ is the solution of the RBSDE
883: associated with $(\zeta ,F_{n},L^{\alpha })$, where
884: \[
885: \zeta =e^{\alpha T}\xi ,F_{n}(t,u,v)=e^{\alpha t}f_{n}(t,e^{-\alpha
886: t}u,e^{-\alpha t}v)-\alpha u,L_{t}^{\alpha }=e^{\alpha t}L_{t}.
887: \]
888: It is easy to check
889: \[
890: \left| F_{n}(t,u,v)\right| \leq e^{\alpha t}\left| g_{t}\right| +e^{\alpha
891: t}\varphi (\left| u\right| )+\alpha \left| u\right| +\beta \left| v\right| ,
892: \]
893: setting $\psi (u)=e^{\alpha T}\varphi (\left| u\right| )+\alpha \left|
894: u\right| $, with $\psi (u)=0$, we get that $F_{n}$ verifies\textbf{\
895: Assumption 6'}-(iii). Moreover
896: \begin{eqnarray*}
897: uF_{n}(t,u,v) &=&e^{\alpha t}uf_{n}(t,e^{-\alpha t}u,e^{-\alpha t}v)-\alpha
898: u^{2} \\
899: &\leq &ue^{\alpha t}g_{t}+\beta \left| u\right| \left| v\right| -\alpha
900: u^{2}.
901: \end{eqnarray*}
902: And $\sup_{0\leq t\leq T}L_{t}^{\alpha }\leq e^{\alpha T}\sup_{0\leq t\leq
903: T}L_{t}\leq e^{\alpha T}b$. Now we apply It\^{o} formula to $\left|
904: U^{n}\right| ^{2}$ on $[0,T]$, and get
905: \begin{eqnarray*}
906: &&\left| U_{t}^{n}\right| ^{2}+\int_{t}^{T}\left| V_{s}^{n}\right| ^{2}ds \\
907: &=&\left| \zeta \right|
908: ^{2}+2\int_{t}^{T}U_{s}^{n}F_{n}(s,U_{s}^{n},V_{s}^{n})ds+2%
909: \int_{t}^{T}U_{s}^{n}dJ_{s}^{n}-2\int_{t}^{T}U_{s}^{n}V_{s}^{n}dB_{s} \\
910: &\leq &\left| \zeta \right| ^{2}+\int_{t}^{T}e^{2\alpha
911: s}g_{s}^{2}ds+(1+2\beta ^{2}-\alpha )\int_{t}^{T}\left| U_{s}^{n}\right|
912: ^{2}ds+\frac{1}{2}\int_{t}^{T}\left| V_{s}^{n}\right| ds \\
913: &&+\theta e^{2\alpha T}b^{2}+\frac{1}{\theta }(J_{T}^{n}-J_{t}^{n})^{2}-2%
914: \int_{t}^{T}U_{s}^{n}V_{s}^{n}dB_{s},
915: \end{eqnarray*}
916: where $\theta $ is a constant to be decided later. By taking conditional
917: expectation, we get
918: \begin{eqnarray}
919: \left| U_{t}^{n}\right| ^{2}+\frac{1}{2}E[\int_{t}^{T}\left|
920: V_{s}^{n}\right| ^{2}ds|\mathcal{F}_{t}] &\leq &E[\left| \zeta \right|
921: ^{2}+\int_{t}^{T}e^{2\alpha s}g_{s}^{2}ds+\theta e^{2\alpha T}b^{2}|\mathcal{%
922: F}_{t}] \label{est-u1} \\
923: &&+(1+2\beta ^{2}-\alpha )E[\int_{t}^{T}\left| U_{s}^{n}\right| ^{2}ds|%
924: \mathcal{F}_{t}]+\frac{1}{\theta }E[(J_{T}^{n}-J_{t}^{n})^{2}|\mathcal{F}%
925: _{t}]. \nonumber
926: \end{eqnarray}
927: Since
928: \[
929: J_{T}^{n}-J_{t}^{n}=U_{t}^{n}-\zeta
930: -\int_{t}^{T}F_{n}(s,U_{s}^{n},V_{s}^{n})ds-\int_{t}^{T}V_{s}^{n}dB_{s},
931: \]
932: we have
933: \[
934: E[(J_{T}^{n}-J_{t}^{n})^{2}|\mathcal{F}_{t}]\leq 4\left| U_{t}^{n}\right|
935: ^{2}+4E[\left| \zeta \right|
936: ^{2}+(\int_{t}^{T}F_{n}(s,U_{s}^{n},V_{s}^{n})ds)^{2}+\int_{t}^{T}\left|
937: V_{s}^{n}\right| ^{2}ds|\mathcal{F}_{t}].
938: \]
939: Using the same approximation as in Lemma \ref{est-res}, except considering
940: conditional expectation $E[\cdot |\mathcal{F}_{t}]$ instead of expectation,
941: we deduce
942: \[
943: E[(J_{T}^{n}-J_{t}^{n})^{2}|\mathcal{F}_{t}]\leq c_{\beta }E[\left| \zeta
944: \right| ^{2}+\int_{t}^{T}e^{2\alpha s}g_{s}^{2}ds+\psi ^{2}(e^{\alpha
945: T}b)+\psi ^{2}(2T)+1|\mathcal{F}_{t}],
946: \]
947: where $c_{\beta }$ is a constant which only depends on $\beta $, $T$, $b$
948: and $\alpha $. Substitute it into (\ref{est-u1}), set $\alpha =1+2\beta ^{2}$%
949: , $\theta =c_{\beta }$, then we get,
950: \begin{eqnarray*}
951: \left| U_{t}^{n}\right| ^{2} &\leq &2E[\left| \zeta \right|
952: ^{2}+\int_{t}^{T}F_{n}^{2}(s,0,0)ds|\mathcal{F}_{t}]+e^{\alpha T}(\varphi
953: (e^{\alpha T}b)+\varphi (2T)) \\
954: &&+\alpha (e^{\alpha T}b+2T)+1+c_{\beta }e^{2\alpha T}b^{2}.
955: \end{eqnarray*}
956: Recall the definition of $U^{n}$, we get
957: \begin{eqnarray*}
958: \left| Y_{t}^{n}\right| ^{2} &\leq &e^{-2\alpha t}(2E[e^{2\alpha T}\left|
959: \xi \right| ^{2}+\int_{t}^{T}e^{2\alpha s}g_{s}^{2}ds|\mathcal{F}%
960: _{t}]+e^{\alpha T}(\varphi (e^{\alpha T}b)+\varphi (2T)) \\
961: &&+\alpha (e^{\alpha T}b+2T)+1+c_{\beta }e^{2\alpha T}b^{2}).
962: \end{eqnarray*}
963: If we set $M_{t}=(e^{2\alpha T}2E[\left| \xi \right|
964: ^{2}+\int_{t}^{T}g_{s}^{2}ds|\mathcal{F}_{t}]+e^{\alpha T}(\varphi
965: (e^{\alpha T}b)+\varphi (2T))+c_{\beta }e^{2\alpha T}b^{2}+\alpha (e^{\alpha
966: T}b+2T)+1)^{\frac{1}{2}}$, then
967: \begin{equation}
968: \left| Y_{t}^{n}\right| \leq M_{t},\forall t\in [0,T]. \label{est-y2}
969: \end{equation}
970:
971: \textbf{Step 3}. Localisation.
972:
973: First, we know that the sequence $(f_{n})_{n\geq \beta }$ is non-decreasing
974: in $n$, then from comparison theorem in \cite{LMX}, we get
975: \[
976: Y_{t}^{n}\leq Y_{t}^{n+1},\forall t\in [0,T],\forall n\geq \beta .
977: \]
978: Define $Y_{t}=\sup_{n\geq \beta }Y_{t}^{n}$.
979:
980: We now consider the localisation procedure. For $m\in \mathbb{N}$, $m\geq b$%
981: , let $\tau _{m}$ be the following stopping time
982: \[
983: \tau _{m}=\inf \{t\in [0,T]:M_{t}+g_{t}\geq m\}\wedge T,
984: \]
985: and we introduce the stopped process $Y_{t}^{n,m}=Y_{t\wedge \tau _{m}}^{n}$%
986: , together with $Z_{t}^{n,m}=Z_{t}^{n}1_{\{t\leq \tau _{m}\}}$ and $%
987: K_{t}^{n,m}=K_{t\wedge \tau _{m}}^{n}$. Then $%
988: (Y_{t}^{n,m},Z_{t}^{n,m},K_{t}^{n,m})_{0\leq t\leq T}$ solved the following
989: RBSDE
990: \begin{eqnarray*}
991: Y_{t}^{n,m} &=&\xi ^{n,m}+\int_{t}^{T}1_{\{s\leq \tau
992: _{m}\}}f_{n}(s,Y_{s}^{n,m},Z_{s}^{n,m})ds+K_{T}^{n,m}-K_{t}^{n,m}-%
993: \int_{t}^{T}Z_{s}^{n,m}dB_{s}, \\
994: Y_{t}^{n,m} &\geq &L_{t},\int_{0}^{T}(Y_{t}^{n,m}-L_{t})dK_{t}^{n,m}=0.
995: \end{eqnarray*}
996: where $\xi ^{n,m}=Y_{\tau _{m}}^{n,m}=Y_{\tau _{m}}^{n}$.
997:
998: Since $(Y^{n,m})_{n\geq \beta }$ is non-decreasing in $n$, with (\ref{est-y2}%
999: ), we get $\sup_{n\geq \beta }\sup_{t\in [0,T]}\left| Y_{t}^{n,m}\right|
1000: \leq m$. Set $\rho _{m}(y)=\frac{ym}{\max \{\left| y\right| ,m\}}$, then it
1001: is easy to check that $(Y^{n,m},Z^{n,m},K^{n,m})$ verifies
1002: \begin{eqnarray*}
1003: Y_{t}^{n,m} &=&\xi ^{n,m}+\int_{t}^{T}1_{\{s\leq \tau _{m}\}}f_{n}(s,\rho
1004: _{m}(Y_{s}^{n,m}),Z_{s}^{n,m})ds+K_{T}^{n,m}-K_{t}^{n,m}-%
1005: \int_{t}^{T}Z_{s}^{n,m}dB_{s}, \\
1006: Y_{t}^{n,m} &\geq &L_{t},\int_{0}^{T}(Y_{t}^{n,m}-L_{t})dK_{t}^{n,m}=0.
1007: \end{eqnarray*}
1008: Moreover, we have
1009: \[
1010: \left| 1_{\{s\leq \tau _{m}\}}f_{n}(s,\rho _{m}(y),z)\right| \leq m+\varphi
1011: (m)+\beta \left| z\right| ,
1012: \]
1013: and $\left| \xi ^{n,m}\right| \leq m$. From Dini's theorem, we know that $%
1014: 1_{\{s\leq \tau _{m}\}}f_{n}(s,\rho _{m}(y),z)$ converge increasingly to$%
1015: 1_{\{s\leq \tau _{m}\}}f(s,\rho _{m}(y),z)$ uniformly on compact set of $%
1016: \mathbb{R\times R}^{d}$, because $f_{n}$ are continuous and
1017: $f_{n}$ converge increasingly to $f$. And $\xi ^{n,m}$ converge
1018: increasingly to $\xi ^{m}$ a.s., where $\xi ^{m}=\sup_{n\geq \beta
1019: }\xi ^{n,m}$.
1020:
1021: As in \cite{M97}, we can prove that $Y^{n,m}$ converges increasingly to $%
1022: Y^{m}$ in $\mathbf{S}^{2}(0,T)$, and $Z^{n,m}\rightarrow Z^{m}$ in $\mathbf{H%
1023: }_{d}^{2}(0,T)$, $K^{n,m}\searrow K^{m}$ uniformly on $[0,T]$. Moreover, $%
1024: (Y^{m},Z^{m},K^{m})$ solves the following RBSDE
1025: \begin{eqnarray*}
1026: Y_{t}^{m} &=&\xi ^{m}+\int_{t}^{T}1_{\{s\leq \tau _{m}\}}f(s,\rho
1027: _{m}(Y_{s}^{m}),Z_{s}^{m})ds+K_{T}^{m}-K_{t}^{m}-\int_{t}^{T}Z_{s}^{m}dB_{s},
1028: \\
1029: Y_{t}^{m} &\geq &L_{t},\int_{0}^{T}(Y_{t}^{m}-L_{t})dK_{t}^{m}=0,
1030: \end{eqnarray*}
1031: where $\xi ^{m}=\sup_{n\geq \beta }Y_{\tau _{m}}^{n,m}$. Notice that $\left|
1032: Y_{t}^{m}\right| \leq m$, so we have
1033: \[
1034: Y_{t}^{m}=\xi ^{m}+\int_{t}^{T}1_{\{s\leq \tau
1035: _{m}\}}f(s,Y_{s}^{m},Z_{s}^{m})ds+K_{T}^{m}-K_{t}^{m}-%
1036: \int_{t}^{T}Z_{s}^{m}dB_{s}.
1037: \]
1038: From the definition of $\{\tau _{m}\}$, it is easy to check that $\tau
1039: _{m}\leq \tau _{m+1}$, with the definition of $Y^{m}$, $Z^{m}$, $K^{m}$ and $%
1040: Y$, we get
1041: \[
1042: Y_{t\wedge \tau _{m}}=Y_{t\wedge \tau
1043: _{m}}^{m+1}=Y_{t}^{m},Z_{t}^{m+1}1_{\{t\leq \tau
1044: _{m}\}}=Z_{t}^{m},K_{t\wedge \tau _{m}}^{m+1}=K_{t}^{m}.
1045: \]
1046: We define
1047: \[
1048: Z_{t}:=Z_{t}^{1}1_{\{t\leq \tau _{1}\}}+\sum_{m\geq 2}Z_{t}^{m}1_{(\tau
1049: _{m-1},\tau _{m}]}(t),\;\;K_{t\wedge \tau _{m}}:=K_{t}^{m}.
1050: \]
1051: Processes $(Y^{m})$ are continuous, and $P$-a.s. $\tau _{m}=T$,
1052: for $m$ large enough, so $Y$ is continuous on $[0,T]$. It follows
1053: that $K$ is also continuous on $[0,T]$. Furthermore, we have for
1054: $m\in \mathbb{N}$,
1055: \begin{equation}
1056: Y_{t\wedge \tau _{m}}=Y_{\tau _{m}}+\int_{t\wedge \tau _{m}}^{\tau
1057: _{m}}f(s,Y_{s},Z_{s})ds+K_{\tau _{m}}-K_{t\wedge \tau _{m}}-\int_{t\wedge
1058: \tau _{m}}^{\tau _{m}}Z_{s}dB_{s}. \label{st-bsde}
1059: \end{equation}
1060: Finally, we have
1061: \begin{eqnarray*}
1062: P(\int_{0}^{T}\left| Z_{s}\right| ^{2}ds &=&\infty )=P(\int_{0}^{T}\left|
1063: Z_{s}\right| ^{2}ds=\infty ,\tau _{m}=T)+P(\int_{0}^{T}\left| Z_{s}\right|
1064: ^{2}ds=\infty ,\tau _{m}<T) \\
1065: &\leq &P(\int_{0}^{T}\left| Z_{s}\right| ^{2}ds=\infty )+P(\tau _{m}<T),
1066: \end{eqnarray*}
1067: in the same way,
1068: \[
1069: P(\left| K_{T}\right| ^{2}=\infty )\leq P(\left| K_{T}\right| ^{2}=\infty
1070: )+P(\tau _{m}<T).
1071: \]
1072: Since $\tau _{m}\nearrow T$, $P$-a.s., we know that $\int_{0}^{T}\left|
1073: Z_{s}\right| ^{2}ds<\infty $ and $\left| K_{T}\right| ^{2}<\infty $, $P$%
1074: -a.s. Let $m\rightarrow \infty $ in (\ref{st-bsde}), we get $(Y,Z,K)$
1075: verifies the equation.
1076:
1077: \textbf{Step 4}. We want to prove that the triple $(Y,Z,K)$ is a solution of
1078: RBSDE$(\xi ,f,L)$.
1079:
1080: First, we consider the integrability of $(Y,Z,K)$. By (\ref{est-y2}), we
1081: know for $0\leq t\leq T$,
1082: \begin{equation}
1083: \left| Y_{t}\right| \leq M_{t}. \label{est-yn}
1084: \end{equation}
1085: It follows immediately that
1086: \[
1087: E[\sup_{0\leq t\leq T}\left| Y_{t}\right| ^{2}]\leq C_{\beta }E[\left| \xi
1088: \right| ^{2}+\int_{0}^{T}g_{s}^{2}ds+\varphi ^{2}(b)+\varphi ^{2}(2T)+1].
1089: \]
1090: where $C_{\beta }$ is a constant only depends on $\beta $, $T$ and $b$. For $%
1091: K$, notice that $K^{n,m}\searrow K^{m}$, then for each $m\in \mathbb{N}$, $%
1092: 0\leq t\leq T$, we know $0\leq K_{t}^{m}\leq K_{t}^{1,m}$. Obviously, the
1093: coefficient $1_{\{s\leq \tau _{m}\}}f_{n}(s,\rho _{m}(y),z)$ satisfies
1094: \textbf{assumption 6'}, and Lipschitz in $z$, by Lemma \ref{est-res},
1095: \[
1096: E[(K_{T}^{1,m})^{2}]\leq C_{\beta }E[\left| \xi ^{1,m}\right|
1097: ^{2}+\int_{0}^{T}g_{s}^{2}ds+\varphi ^{2}(b)+\varphi ^{2}(2T)+1],
1098: \]
1099: where $\xi ^{1,m}=Y_{\tau _{m}}^{1}$. With (\ref{est-y2}), we have
1100: \[
1101: E[(K_{T}^{1,m})^{2}]\leq 2C_{\beta }E[\left| \xi \right|
1102: ^{2}+\int_{0}^{T}g_{s}^{2}ds+\varphi ^{2}(b)+\varphi ^{2}(2T)+1],
1103: \]
1104: which follows that for each $m\in \mathbb{N}$,
1105: \[
1106: E[(K_{T}^{m})^{2}]\leq 2C_{\beta }E[\left| \xi \right|
1107: ^{2}+\int_{0}^{T}g_{s}^{2}ds+\varphi ^{2}(b)+\varphi ^{2}(2T)+1],
1108: \]
1109: and so does for $K$, i.e. we get $E[(K_{T})^{2}]<\infty $.
1110:
1111: In order to estimate $Z$, we apply It\^{o}'s formula to $\left| Y_{t}\right|
1112: ^{2}$ on the interval $[0,T]$, then
1113: \begin{eqnarray*}
1114: &&\left| Y_{0}\right| ^{2}+\frac{1}{2}E\int_{0}^{T}\left| Z_{s}\right| ^{2}ds
1115: \\
1116: &\leq &E\left| \xi \right| ^{2}+E\int_{0}^{T}g_{s}^{2}ds+(1+2\beta
1117: ^{2})E\int_{0}^{T}\left| Y_{s}\right| ^{2}ds+E[\sup_{0\leq t\leq T}\left|
1118: Y_{t}\right| ^{2}]+E[(K_{T})^{2}].
1119: \end{eqnarray*}
1120: Thanks to the estimates for $Y$ and $K$, there exists a constant $C$ only
1121: depends on $\beta $, $T$ and $b$, such that
1122: \[
1123: E\int_{0}^{T}\left| Z_{s}\right| ^{2}ds\leq CE[\left| \xi \right|
1124: ^{2}+\int_{0}^{T}g_{s}^{2}ds+\varphi ^{2}(b)+\varphi ^{2}(2T)+1].
1125: \]
1126: The last is to check the integral condition. Recall that $%
1127: \int_{0}^{T}(Y_{t}^{m}-L_{t})dK_{t}^{m}=0$, then we have
1128: \[
1129: \int_{0}^{\tau _{m}}(Y_{t}-L_{t})dK_{t}=0,\mbox{ a.s.}
1130: \]
1131: Since $P$-a.s. $\tau _{m}=T$, for $m$ large enough, so
1132: \[
1133: \int_{0}^{T}(Y_{t}-L_{t})dK_{t}=0,\mbox{ a.s.}
1134: \]
1135: i.e. $(Y,Z,K)$ is a solution of RBSDE$(\xi ,f,L)$ in $\mathbf{S}%
1136: ^{2}(0,T)\times \mathbf{H}_{d}^{2}(0,T)\times \mathbf{A}^{2}(0,T)$. $\square
1137: $
1138:
1139: \section{Appendix: Comparison theorems}
1140:
1141: We first generalize the comparison theorem of RBSDE with superlinear
1142: quadratic coefficient, (in view to proposition 3.2 in \cite{KLQT}), to
1143: compare the increasing processes. Assume that \textbf{Assumption 1} and
1144: \textbf{3} hold, and that the coefficient $f$ satisfies:
1145:
1146: \textbf{Assumption 7.} For all $(t,\omega )$, $f(t,\omega ,\cdot ,\cdot )$
1147: is continuous and there exists a function $l$ strictly positive such that
1148: \[
1149: f(t,y,z)\leq l(y)+A\left| z\right| ^{2},\mbox{ with }\int_{0}^{\infty }\frac{%
1150: dx}{l(x)}=+\infty .
1151: \]
1152:
1153: \begin{proposition}
1154: \label{comp1}Suppose that $\xi ^{i}$ are $\mathcal{F}_{T}$-adapted and
1155: bounded, $f^{i}(s,y,z)$, $i=1,2$ satisfy the condition \textbf{Assumption 7}
1156: and $L$ satisfies\textbf{\ Assumption 3}. The two triples $%
1157: (Y^{1},Z^{1},K^{1})$, $(Y^{2},Z^{2},K^{2})$ are respectively solutions of
1158: the RBSDE$(\xi ^{1},f^{1},L)$ and RBSDE$(\xi ^{2},f^{2},L)$. If we have $%
1159: \forall (t,y,z)\in [0,T]\times \mathbb{R\times R}^{d},
1160: $%
1161: \[
1162: \xi ^{1}\leq \xi ^{2},f^{1}(t,y,z)\leq f^{2}(t,y,z),
1163: \]
1164: then $Y_{t}^{1}\leq Y_{t}^{2}$, $K_{t}^{1}\geq K_{t}^{2}$ and $%
1165: dK_{t}^{1}\geq dK_{t}^{2}$, for $t\in [0,T]$.
1166: \end{proposition}
1167:
1168: %TCIMACRO{\TeXButton{Proof}{\proof}}
1169: %BeginExpansion
1170: \proof%
1171: %EndExpansion
1172: From the demonstration of theorem 1 in \cite{KLQT}, we know that for $i=1,2$%
1173: , $(Y^{i},Z^{i},K^{i})$ is the solution of RBSDE$(\xi ^{i},f^{i},L)$ if and
1174: only if $(\theta ^{i},J^{i},\Lambda ^{i})$ is the solution of RBSDE$(\eta
1175: ^{i},F^{i},\overline{L})$ where
1176: \begin{equation}
1177: (\theta ^{i},J^{i},\Lambda ^{i})=(\exp (2AY^{i}),2AZ^{i}\theta
1178: ^{i},2\int_{0}^{\cdot }A\exp (2AY_{s})dK_{s}^{i}) \label{trans1}
1179: \end{equation}
1180: and
1181: \begin{eqnarray*}
1182: \eta ^{i} &=&\exp (2A\xi ^{i}),\overline{L}_{t}=\exp (2AL_{t}) \\
1183: F^{i}(t,x,\lambda ) &=&2Ax[f^{i}(s,\frac{\log x}{2A},\frac{\lambda }{2Ax})-%
1184: \frac{\left| \lambda \right| ^{2}}{4Ax^{2}}].
1185: \end{eqnarray*}
1186: Then we use the approximation to construct a solution. For $p\in \mathbb{N}$%
1187: , we consider the RBSDE$(\eta ^{i},\widetilde{F}_{p}^{i},\overline{L}_{t})$,
1188: where
1189: \[
1190: \widetilde{F}_{p}^{i}(s,x,\lambda )=g(\rho (\theta ))(1-\kappa _{p}(\lambda
1191: ))+\kappa _{p}(\lambda )F^{i}(s,\rho (\theta ),\lambda ).
1192: \]
1193: Here $g(x)=2Axl(\frac{\log x}{2A})$, $\kappa _{p}(\lambda )$ and $\rho (x)$
1194: are smooth functions such that $\kappa _{p}(\lambda )=1$ if $\left| \lambda
1195: \right| \leq p$, $\kappa _{p}(\lambda )=0$ if $\left| \lambda \right| \geq
1196: p+1$, and $\rho (x)=x$ if $x\in [r,R]$, $\rho (x)=\frac{r}{2}$ if $x\in (0,%
1197: \frac{r}{2})$, $\rho (x)=R$ if $x\in (2R,+\infty )$, where $r$ and $R$ are
1198: two constants. Since $\widetilde{F}_{p}^{i}$ are bounded and continuous
1199: function of $(\theta ,\lambda )$, the RBSDE$(\eta ^{i},\widetilde{F}_{p}^{i},%
1200: \overline{L}_{t})$ admits a bounded maximal solution $(\theta
1201: ^{i,p},J^{i,p},\Lambda ^{i,p})$, with $\underline{m}\leq $ $\theta
1202: _{t}^{i,p}\leq V_{0}$. Here $\underline{m}$ and $V_{0}$ are constants given
1203: in Theorem 2 in \cite{KLQT}.
1204:
1205: We know that $\widetilde{F}_{p}^{i}\downarrow \widetilde{F}^{i}$, as $%
1206: p\rightarrow \infty $, where $\widetilde{F}^{i}=F(s,\rho (\theta ),\lambda )$%
1207: . Thanks to the proof of theorem 1 in \cite{KLQT}, it follows that $\theta
1208: _{t}^{i,p}\downarrow \widetilde{\theta }_{t}^{i}$, $J_{t}^{i,p}\uparrow
1209: \widetilde{J}_{t}^{i}$, $0\leq t\leq T$, and $\Lambda ^{i,p}\rightarrow
1210: \widetilde{\Lambda }^{i}$ in $\mathbf{H}_{d}^{2}(0,T)$ and $(\widetilde{%
1211: \theta }^{i},\widetilde{J}^{i},\widetilde{\Lambda }^{i})$ is a solution of
1212: the RBSDE$(\eta ^{i},\widetilde{F}^{i},\overline{L}_{t})$. In addition, $%
1213: \underline{m}\leq $ $\widetilde{\theta }_{t}^{i}\leq V_{0}$. So if we choose
1214: $0<r<\underline{m}$ and $V_{0}<R$, then $\widetilde{F}^{i}=F^{i}$. It
1215: follows that $(\widetilde{\theta }^{i},\widetilde{J}^{i},\widetilde{\Lambda }%
1216: ^{i})$ satisfies the RBSDE$(\eta ^{i},F^{i},\overline{L})$, i.e. $(%
1217: \widetilde{\theta }^{i},\widetilde{J}^{i},\widetilde{\Lambda }^{i})=$ $%
1218: (\theta ^{i},J^{i},\Lambda ^{i})$.
1219:
1220: Since $f^{1}(t,y,z)\leq f^{2}(t,y,z)$, for $(t,x,\lambda )\in
1221: [0,T]\times \mathbb{R}_{^{_{+}}}\mathbb{\times R}^{d}$, we have
1222: $F^{1}(t,x,\lambda )\leq
1223: F^{2}(t,x,\lambda )$. Then for $p\in \mathbb{N}$, $\widetilde{F}%
1224: _{p}^{1}(s,x,\lambda )\leq \widetilde{F}_{p}^{2}(s,x,\lambda )$. Notice that
1225: $\widetilde{F}_{p}^{i}$ is bounded and continuous in $(\theta ,\lambda )$
1226: and $\theta _{t}^{i,p}>0$, by Lemma 2.1 in \cite{KLQT}, it follows that $%
1227: \theta _{t}^{1,p}\leq \theta _{t}^{2,p}$, $J_{t}^{1,p}\geq J_{t}^{2,p}$, $%
1228: dJ_{t}^{1,p}\geq dJ_{t}^{2,p}$, $0\leq t\leq T$. And it follows that for $%
1229: 0\leq s\leq t\leq T$, $J_{t}^{1,p}-J_{s}^{1,p}\geq J_{t}^{2,p}-J_{s}^{2,p}$.
1230: Let $p\rightarrow \infty $, thanks to the convergence results, we get that
1231: \[
1232: \theta _{t}^{1}\leq \theta _{t}^{2},J_{t}^{1}\geq
1233: J_{t}^{2},J_{t}^{1}-J_{s}^{1}\geq J_{t}^{2}-J_{s}^{2},
1234: \]
1235: which implies $dJ_{t}^{1}\geq dJ_{t}^{2}.$ From (\ref{trans}), we know that
1236: \[
1237: Y_{t}^{i}=\frac{\log (\theta _{t}^{i})}{2A},Z_{t}^{i}=\frac{\Lambda ^{i}}{%
1238: 2A\theta ^{i}},dK_{t}^{i}=\frac{dJ_{t}^{i}}{2A\theta _{t}^{i}},
1239: \]
1240: so $Y_{t}^{1}\leq Y_{t}^{2}$ and $dK_{t}^{1}\geq dK_{t}^{2}$, which implies
1241: that $K_{t}^{1}\geq K_{t}^{2}$ in view of $K_{0}^{1}=K_{0}^{2}=0$. $\square $
1242:
1243: From this result, we prove the following comparison theorem when the
1244: coefficient $f$ satisfies monotonicity and general increasing condition in $%
1245: y $, and quadratic increasing in $z$.
1246:
1247: \begin{proposition}
1248: \label{com-zq}Suppose that $\xi ^{i}$ and $f^{i}(s,y,z)$, $i=1,2$ satisfy
1249: the condition \textbf{Asssumption 1} and \textbf{2}, $L$ satisfies \textbf{%
1250: Assumption 3.} The two triples $(Y^{1},Z^{1},K^{1})$, $(Y^{2},Z^{2},K^{2})$
1251: are respectively the solutions of the RBSDE$(\xi ^{1},f^{1},L)$ and RBSDE$%
1252: (\xi ^{2},f^{2},L)$. If we have $\forall (t,y,z)\in [0,T]\times \mathbb{%
1253: R\times R}^{d}$,
1254: \[
1255: f^{1}(t,y,z) \leq f^{2}(t,y,z), \quad \xi ^{1} \leq \xi ^{2},
1256: \]
1257: then $Y_{t}^{1}\leq Y_{t}^{2}$, $K_{t}^{1}\geq K_{t}^{2}$ and $%
1258: dK_{t}^{1}\geq dK_{t}^{2}$, for $t\in [0,T]$.
1259: \end{proposition}
1260:
1261: %TCIMACRO{\TeXButton{Proof}{\proof}}
1262: %BeginExpansion
1263: \proof%
1264: %EndExpansion
1265: First with changement of $(Y,Z,K)$,
1266: \[
1267: (Y^{b},Z^{b},K^{b})=(Y-b,Z,K),
1268: \]
1269: we work with $L^{b}\leq 0$. Since this transformation doesn't
1270: change the monotonicity, then in the following, we assume that the
1271: barrier $L$ is a negative bounded process. As in the proof of
1272: theorem \ref{gmq}, for $C\in \mathbb{R}_{+}$, let
1273: $g^{C}:\mathbb{R}\rightarrow [0,1]$ continuous which
1274: satisfies (\ref{gfun}). Set $f_{i}^{C}(t,y,z)=g^{C}(x)f^{i}(t,y,z)$, $i=1,2$%
1275: , which satisfies \textbf{Assumption 7}, with $l^{i}(y)=\varphi ^{i}(\left|
1276: 2C\right| )$. We consider solutions $(Y^{i,C},Z^{i,C},K^{i,C})$ of the RBSDE$%
1277: (\xi ^{i},f_{i}^{C},L^{b})$ respectively. Using proposition \ref{comp1},
1278: since
1279: \[
1280: f_{1}^{C}(t,y,z)\leq f_{2}^{C}(t,y,z),\mbox{ }\xi ^{1}\leq \xi ^{2},
1281: \]
1282: we get for $t\in [0,T]$,
1283: \[
1284: Y_{t}^{1,C}\leq Y_{t}^{2,C},dK_{t}^{1,C}\geq dK_{t}^{2,C}.
1285: \]
1286: Then by the bounded property of $Y^{i}$, we choose $C$ big enough like in
1287: the proof of theorem \ref{gmq}, which follows immediately
1288: \[
1289: Y_{t}^{1}\leq Y_{t}^{2},dK_{t}^{1}\geq dK_{t}^{2},\forall t\in [0,T].
1290: \]
1291: $\square $
1292:
1293: \begin{proposition}
1294: \label{com-zl1}Suppose that $\xi ^{i}\in \mathbf{L}^{2}\mathbf{(}\mathcal{F}%
1295: _{T})$, $f^{i}(s,y,z)$ satisfy the condition \textbf{Assumption 6}, and $%
1296: L^{i}$ satisfies \textbf{Assumption 3, }$i=1,2$. The two triples $%
1297: (Y^{1},Z^{1},K^{1})$, $(Y^{2},Z^{2},K^{2})$ are respectively solutions of
1298: the RBSDE$(\xi ^{1},f^{1},L)$ and RBSDE$(\xi ^{2},f^{2},L)$. If we have for $%
1299: \forall (t,y,z)\in [0,T]\times \mathbb{R\times R}^{d},$%
1300: \[
1301: \mbox{ }\xi ^{1}\leq \xi ^{2},\,\,\,f^{1}(t,y,z)\leq
1302: f^{2}(t,y,z),\,L_{t}^{1}\leq L_{t}^{2},
1303: \]
1304: then $Y_{t}^{1}\leq Y_{t}^{2}$, for $t\in [0,T]$.
1305: \end{proposition}
1306:
1307: The result comes from the comparison theorem in \cite{LMX} and the
1308: approximation in the proof of theorem \ref{exist-zl}.
1309:
1310: \begin{thebibliography}{99}
1311: \bibitem{BDHPS} Briand, Ph., Delyon, B., Hu, Y., Pardoux, E. and Stoica, L.
1312: (2003) $L_{p}$ solutions of BSDEs, \textit{Stochastic Process. Appl.} 108,
1313: 109-129.
1314:
1315: \bibitem{LS04} Ph. Briand, J.P. Lepeltier and J. San Mart\'{i}n, (2004)
1316: BSDE's with continuous, monotonicity, and non-Lipschitz in $z$ coefficient.
1317: Submitted
1318:
1319: \bibitem{E79} N. El Karoui, (1979) Les aspects probabilistes du contr\^{o}%
1320: le stochastique. Ecole d'\'{e}t\'{e} de Saint-Flour, \textit{Lecture Notes
1321: in Math. }\textbf{876. (}Springer, Berlin), 73-238.
1322:
1323: \bibitem{EKPPQ} N.~El Karoui, C.~Kapoudjian, E.~Pardoux, S.~Peng and M.C.
1324: Quenez, (1997) Reflected Solutions of Backward SDE and Related Obstacle
1325: Problems for PDEs, \textit{Ann. Probab.} \textbf{25}, no 2, 702--737.
1326:
1327: \bibitem{EPQ} N.~El Karoui, S.~Peng and M.C. Quenez, (1997). Backward
1328: stochastic differential equations in Finance. \textit{Math. Finance, }%
1329: \textbf{7, }1-71\textbf{.}
1330:
1331: \bibitem{K00} M. Kobylanski, (2000) Backward stochastic differential
1332: equations and partial differential equations with quadratic growth, \textit{%
1333: Ann. Proba. }\textbf{28}, 558-602.\textbf{\ }
1334:
1335: \bibitem{KLQT} M. Kobylanski, J.P. Lepeltier, M.C. Quenez and S.Torres,
1336: (2002) Reflected BSDE with superlinear quadratic coefficient. \textit{%
1337: Probability and Mathematical Statistics. }Vol. \textbf{22}, 51-83.
1338:
1339: \bibitem{LMX} J.P. Lepeltier, A. Matoussi and M. Xu, (2004) Reflected BSDEs
1340: under monotonicity and general increasing growth conditiond. \textit{%
1341: Advanced in applied probability}, to appear in March, 2005.
1342:
1343: \bibitem{LS98} J.P. Lepeltier and J. San Mart\'{i}n. (1998) Existence for
1344: BSDE with superlinear-quadratic coefficient, \textit{Stochastic and
1345: Stochastic reports,} \textbf{63}, 227-240.
1346:
1347: \bibitem{M97} A. Matoussi, (1997) Reflected solutions of backward
1348: stochastic differential equations with continuous coefficient, \textit{%
1349: Statistic \& Probality Letters} \textbf{34}, 347-354.
1350:
1351: \bibitem{P99} Pardoux, E., 1999. BSDE's, weak convergence and
1352: homogenization of semilinear PDE's in Nonlinear analysis, Differential
1353: Equations and Control, F. H. Clarke \& R. J. Stern Eds , pp. 503-549, Kluwer
1354: Acad. Pub.
1355:
1356: \bibitem{PP90} E. Pardoux and S. Peng, (1990) Adapted solutions of Backward
1357: Stochastic Differential Equations. \textit{Systems Control Lett. }\textbf{14,%
1358: } 51-61.
1359:
1360: \bibitem{RY} D. Revuz and M. Yor, (1991) Continuous artingales and Brownian
1361: motion (Springer, Berlin).
1362: \end{thebibliography}
1363:
1364: \end{document}
1365: