Abstract
In this paper, we consider a stochastic optimal control problem, in which the cost function is defined through a reflected backward stochastic differential equation in sublinear expectation framework. Besides, we study the regularity of the value function and establish the dynamic programming principle. Moreover, we prove that the value function is the unique viscosity solution of the related Hamilton–Jacobi–Bellman–Isaac equation.
Similar content being viewed by others
References
Peng, S.G.: \(G\)-Expectation, \(G\)-Brownian motion and related stochastic calculus of Itô type. In: Stochastic Analysis and Applications, Abel Symposium, vol. 2, pp. 541–567. Springer, Berlin (2007)
Epstein, L.G., Ji, S.L.: Ambiguous volatility and asset pricing in continuous time. Rev. Financ. Stud. 26, 1740–1786 (2013)
Epstein, L.G., Ji, S.L.: Ambiguous volatility, possibility and utility in continuous time. J. Math. Econ. 50, 269–282 (2014)
Hu, M.S., Ji, S.L., Peng, S.G., Song, Y.S.: Backward stochastic differential equations driven by \(G\)-Brownian motion. Stoch. Proc. Appl. 124, 759–784 (2014)
Hu, M.S., Ji, S.L., Peng, S.G., Song, Y.S.: Comparison theorem, Feynman–Kac formula and Girsanov transformation for BSDEs driven by \(G\)-Brownian motion. Stoch. Proc. Appl. 124, 1170–1195 (2014)
Soner, H.M., Touzi, N., Zhang, J.F.: Wellposedness of second order backward SDEs. Probab. Theory Relat. Fields 153, 149–190 (2012)
Hu, M.S., Ji, S.L.: Dynamic programming principle for stochastic recursive optimal control problem driven by a \(G\)-Brownian motion. Stoch. Proc. Appl. 127, 107–134 (2017)
Fouque, J.-P., Pun, C.S., Wong, H.Y.: Portfolio optimization with ambiguous correlation and stochastic volatilities. SIAM. J. Control. Optim. 54, 2309–2338 (2016)
Matoussi, A., Possamaï, D., Zhou, C.: Robust Utility maximization in non-dominated models with 2BSDEs. Math. Finance 25(2), 258–287 (2015)
Neufeld, A., Nutz, M.: Robust utility maximization with Lévy processes. Math. Finance 28(1), 82–105 (2018)
El Karoui, N., Pardoux, E., Quenez, M.C.: Reflected Backward SDEs and American Options. In: Rogers, L.C.G., Talay, D. (eds.) Numerical Methods in Finance, pp. 215–231 (1997)
Li, H.W., Peng, S.G., Soumana Hima, A.: Reflected solutions of BSDE driven by \(G\)-Brownian motion. Sci. China Math. 61(1), 1–26 (2018)
Wu, Z., Yu, Z.Y.: Dynamic programming principle for one kind of stochastic recursive optimal control problem and Hamilton–Jacobi–Bellman equation. SIAM J Control Optim. 47(5), 2616–2641 (2008)
Peng, S.G.: Nonlinear expectations and stochastic calculus under uncertainty. arXiv:1002.4546v1
Denis, L., Hu, M.S., Peng, S.G.: Function spaces and capacity related to a sublinear expectation: application to \(G\)-Brownian motion paths. Potential Anal. 34, 139–161 (2011)
Cohen, S.: Quasi-sure analysis, aggregation and dual representations of sublinear expectations in general spaces. Electron. J. Probab. 17, 1–15 (2012)
Peng, S.G.: A generalized dynamic programming principle and Hamilton–Jacobi–Bellmen equation. Stoch. Stoch. Rep. 38, 119–134 (1992)
Peng, S.G.: Backward stochastic differential equation and application to optimal control. Appl. Math. Optim. 27, 125–144 (1993)
Buckdahn, R., Nie, T.Y.: Generalized Hamilton–Jacobi–Bellman equations with Dirichlet boundary and stochastic exit time optimal control problem. SIAM. J. Control. Optim. 54(2), 602–631 (2016)
Peng, S.G.: Backward stochastic differential equations-stochastic optimization theory and viscosity solutions of HJB equations. In: Yan, J., Peng, S., Fang, S., Wu, L. (eds.) Topics on Stochastic Analysis, pp. 85–138. Science Press, Beijing (1997). (in Chinese)
Hu, M.S., Wang, F.L.: Stochastic optimal control problem with infinite horizon driven by \(G\)-Brownian motion. ESAIM Control Optim. Calc. Var. 24(2), 873–899 (2018)
Buckdahn, R., Li, J.: Stochastic differential games with reflection and related obstacle problems for Isaacs equations. Acta Math. Appl. Sin. English Ser. 27(4), 647–678 (2011)
Crandall, M.G., Ishii, H., Lions, P.L.: User’s guide to viscosity solutions of second order partial differential equations. Bull. Am. Math. Soc. 27(1), 1–67 (1992)
Song, Y.S.: Some properties on \(G\)-evaluation and its applications to \(G\)-martingale decomposition. Sci. China Math. 54, 287–300 (2011)
Acknowledgements
Li’s research was supported by the German Research Foundation (DFG) via CRC 1283. Wang’s research was supported by the National Natural Science Foundation of China (Nos. 11601282, 11871310 and 11871458) and the Natural Science Foundation of Shandong Province (No. ZR2016AQ10). The authors wish to thank the editor and the referees for their valuable comments and constructive suggestions which improved the presentation of this manuscript.
Author information
Authors and Affiliations
Corresponding author
Additional information
Hélène Frankowska.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Reflected G-BSDE
The definition of reflected G-BSDEs Given an obstacle process \(\{S_t\}_{t\in [0,T]}\), a terminal value \(\zeta \in L^{\beta }_G(\varOmega _T)\) with \(\zeta \ge S_T\) for \(\beta >2\) and generator \(f(t,\omega ,y,z):[0,T]\times \varOmega \times \mathbb {R}\times \mathbb {R}^{d}\rightarrow \mathbb {R}\), a triple of processes \((Y,Z,A)\in \mathfrak {S}_G^{2}(0,T)\) is called a solution of reflected G-BSDE with data \((\zeta , f,S)\) if the following properties hold:
-
(i)
\(Y_t=\zeta +\int _t^T f(s,Y_s,Z_s)\hbox {d}s -\int _t^T Z_s \hbox {d}B_s+(A_T-A_t)\);
-
(ii)
\(Y_t\ge S_t\), and \(\{-\int _0^t (Y_s-S_s)\hbox {d}A_s\}_{t\in [0,T]}\) is a non-increasing G-martingale,
where \(\mathfrak {S}_{G}^{2}(0,T)\) is the collection of processes (Y, Z, A) such that \(Y\in S_G^{2}(0,T)\), \(Z\in M_G^{2}(0,T)\) and \(A\in S_G^{2}(0,T)\) is a continuous non-decreasing process starting from origin.
The well-posedness of reflected G-BSDEs Consider the following assumption:
-
(H1)
there exists a constant \(\beta >2\) such that for any y, z, \(f(\cdot ,\cdot ,y,z)\in M_{G}^{\beta }(0,T)\);
-
(H2)
there exists a constant \(L_1>0\) such that \( |f(t,y,z)-f(t,y^{\prime },z^{\prime })|\le L_1(|y-y^{\prime }|+|z-z^{\prime }|); \)
-
(H3)
there exists a constant c such that \(\{S_t\}_{t\in [0,T]}\in S_G^\beta (0,T)\) and \(S_t\le c\) for each \(t\in [0,T]\);
-
(H3’)
\(\{S_t\}_{t\in [0,T]}\) has the following form:
$$\begin{aligned} S_t=S_0+\int _0^t b_s\hbox {d}s+\sum _{i,j=1}^{d}\int _0^t \gamma ^{ij}_sd\langle B^i, B^j\rangle _s+\sum _{j=1}^{d}\int _0^t \kappa _s^j \hbox {d}B_s^j, \end{aligned}$$where the processes \(b_s,\gamma ^{ij}_s=\gamma ^{ji}_s\in M_G^\beta (0,T)\) and \(\kappa _s^j\in H_G^\beta (0,T)\), \(1\le i,j\le d\).
Lemma A.1
([12]) Assume that f satisfies (H1)-(H2) for some \(\beta >2\) and let (H3) or (H3’) hold. Then, the reflected G-BSDE has a unique solution \((Y,Z,K)\in \mathfrak {S}_{G}^{2}(0,T)\).
Lemma A.2
([12]) Let \(\zeta ^{\nu }\in L_{G}^{\beta }(\varOmega _{T})\), \(\nu =1,2\) and \(f^{\nu }\), \(S^{\nu }\) satisfy (H1)-(H3) for some \(\beta >2\). Assume that \((Y^{\nu },Z^{\nu },K^{\nu })\in \mathfrak {S}_{G}^{2}(0,T)\), \(\nu =1,2\) are the solutions of the reflected G-BSDE corresponding to data (\(\zeta ^{\nu }\), \(f^{\nu }\), \(S^{\nu }\)). Set \(\hat{Y}_{t}=Y_{t}^{1}-Y_{t}^{2}\), \(\hat{S}_{t}=S_{t}^{1}-S_{t}^{2}\) and \(\hat{\zeta }=\zeta ^1-\zeta ^2\). Then, there exists a constant \(\hat{C}\) depending on T, G, \(\beta \), c and \(L_1\) such that
where \(\varPsi _{t,T}=\sum _{\nu =1}^2\hat{\mathbb {E}}_t[\sup _{s\in [t,T]}\hat{\mathbb {E}}_s[1+|\zeta ^\nu |^2+\int _t^T |f^{\nu }(r,0,0)|^2\mathrm{d}r]]\).
Lemma A.3
([12]) Let \(\zeta ^{\nu }\in L_{G}^{\beta }(\varOmega _{T})\), \(\nu =1,2\) and \(f^{\nu }\), \(S^{\nu }\) satisfy (H1), (H2), (H3’) for some \(\beta >2\). Assume that \((Y^{\nu },Z^{\nu },K^{\nu })\in \mathfrak {S}_{G}^{2}(0,T)\), \(\nu =1,2\) are the solutions of the reflected G-BSDE with data (\(\zeta ^{\nu }\), \(f^{\nu }\), \(S^{\nu }\)). Set \(\bar{Y}_{t}=(Y^1_t-S_t^1)-(Y^2_t-S_t^2)\) and \(\hat{S}_{t}=S_{t}^{1}-S_{t}^{2}\). Then, there exists a constant \(\bar{C}\) depending on T, G, \(\beta \) and \(L_1\) such that
where \(\hat{\lambda }_{s}=|f^{1}(s,Y_{s}^{2},Z_{s} ^{2})-f^{2}(s,Y_{s}^{2},Z_{s}^{2})|\), \(\bar{\lambda }_s^{\nu ,0}=|f^{\nu }(s,0,0)|+|b^{\nu }_s|+\sum \nolimits _{i,j=1}^{d}|\gamma _s^{\nu ,ij}|+\sum \nolimits _{j=1}^d|\kappa _s^{\nu ,j}|\) and \(\hat{\rho }_s=|b^1_s-b^2_s|+\sum \nolimits _{i,j=1}^{d}|\gamma _s^{1,ij}-\gamma _s^{2,ij}|+\sum \nolimits _{j=1}^d|\kappa _s^{1,j}-\kappa _s^{2,j}|\).
Appendix B: The Complement Proofs
The following maximal inequality for G-martingale has been firstly established by Song [24].
Lemma B.1
Assume \(\alpha \ge 1\) and \(\delta >0\). Set
Then, we have
Proof
The proof is immediate from the definition of conditional G-expectation and Theorem 3.4 in [24]. \(\square \)
Now, we are going to state the proof of Lemmas 3.2 and 3.3.
Proof
It is sufficient to prove the second inequalities in both cases, since the first ones can be proved similarly. For convenience, we omit superscripts t. We fist prove the second inequality in Lemma 3.2. Set \(\hat{X}_s=X_s^{\xi ,u}-X_s^{\xi ',u'}\), \(\hat{\varPhi }(X_T)=\varPhi (X^{\xi ,u}_T)-\varPhi (X^{\xi ^{\prime },u^{\prime }}_T)\) and \(\hat{\lambda }_s=f(s,X^{\xi ^{\prime },u^{\prime }}_s,Y^{\xi ^{\prime },u^{\prime }}_s,Z^{\xi ^{\prime },u^{\prime }}_s,u^{\prime }_s) -f(s,X^{\xi ,u}_s,Y^{\xi ^{\prime },u^{\prime }}_s,Z^{\xi ^{\prime },u^{\prime }}_s, u_s)\). Applying Lemma A.2 and Lemma B.1 yields that
where \(C_2\) is a generic constant depending on T, G, c, L and n (may vary from line to line), and \( \varPsi _{t,T}^2\le C_2(1+|\xi |^3+|\xi ^{\prime }|^3)\). From Lemma 2.1, we could get the desired result.
Then, we prove the second inequality in Lemma 3.3. Applying G-Itô’s formula (see Theorem 6.5 of Chap. III in [14]) to \(l(s,X_s^{\xi ,u}) \) yields that
where \(b^{\xi ,u},\gamma ^{\xi ,u}\) and \(\kappa ^{\xi ,u}\) are given by
Denote by \(C_3\) a generic constant depending on T, G, n and L, which may vary from line to line. Then, recalling Lemma A.3, we deduce that
where
Set \(\bar{\varPsi }_{t,T}=\sup \limits _{s\in [t,T]}(1+|X^{\xi ,u}_s|+|X^{\xi ^{\prime },u^{\prime }}_s| +|X^{\xi ,u}_s|^2+|X^{\xi ^{\prime },u^{\prime }}_s|^2)\) and \(\hat{X}_s=X_s^{\xi ,u}-X_s^{\xi ',u'}\). Then, recalling assumptions (A1), (A2\(^{\prime }\)) and Lemma B.1, we derive that
Consequently, in spirit of Lemma 2.1, we get
which, together with \( |Y^{\xi ,u}_t-Y^{\xi ^{\prime },u^{\prime }}_t|^2\le 2(|l(t,\xi )-l(t,\xi ^{\prime })|^2+|\bar{Y}_t|^2), \) implies the inequality (ii). The proof is complete. \(\square \)
Finally, we are ready to state the proof of Eq. (7).
Proof
For readers’ convenience, we shall give the sketch of the proof. For simplicity, we omit the superscripts (t, x).
From the proof of Lemma 4.4 in [12], it suffices to prove that
For simplicity, set \(\varTheta ^{N,u}_r=(X_r^{u},Y_r^{N,u},Z_r^{N,u})\). Now, recalling Lemma 4.3 in [12], we derive that for each \((t,x)\in [0,T]\times \mathbb {R}^n\) and \(u\in \mathcal {U}^t[t,T]\),
where \(\tilde{S}_s^{N,u}=e^{N(s-T)}(\varPhi (X^{u}_T)-l(s, X^{u}_s))+\int _s^T Ne^{N(s-r)}(l(r, X^{u}_r)-l(s, X^{u}_s))\hbox {d}r\).
In spirit of Lemma B.1 and using a similar analysis as Equation (4.3) in [12], we conclude that
Next, we shall deal with the term \(\tilde{S}_s^{N,u}\). Let \(C_{5}>0\) be a generic constant. It follows from Equation (4.4) in [12] that for each fixed \(\varepsilon ,\delta >0\) and \(\beta >2\),
Noting that for each \(u\in \mathcal {U}^{t}[t,T]\) and any integer \(\rho >0\), we have
where \(A={\{\sup \limits _{s\in [t,T]}|X^{u}_s|\le \rho \}}\). Denote \(t^{\rho }_i=\frac{i}{\rho }(T-t)\), \(i=0, \ldots , \rho \). Then, we deduce that, for each \(\varepsilon \le \frac{T-t}{\rho }\),
Consequently, letting \(\varepsilon \rightarrow 0\) and sending \(\rho \rightarrow \infty \) we obtain that
which, together with Eq. (8) and (4.5) in [12], implies that for each \(t<\delta <T\)
By the definition of \(Y^{1,u}\), we have
It is easy to check that for some \(\beta >2\),
By Lemma B.1, it follows that
Note that \((Y_T^{1,u}-l(T, X^{u}_T))^- =0\). Then, it holds that
Thus, by a similar analysis as the above we derive that for each integer \(\rho >0\),
where \(m(\cdot )\) is a nonnegative continuous function satisfying \(\lim _{\delta \rightarrow 0}m(\delta )=0\).
On the other hand, it follows from Lemmas 2.1 and B.1 that
Therefore, by the above analysis, it holds that for each integer \(\rho >0\)
Sending \(\delta \rightarrow 0\) and then \(\rho \rightarrow \infty \) yields the desired result. \(\square \)
Rights and permissions
About this article
Cite this article
Li, H., Wang, F. Stochastic Optimal Control Problem with Obstacle Constraints in Sublinear Expectation Framework. J Optim Theory Appl 183, 422–439 (2019). https://doi.org/10.1007/s10957-019-01546-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10957-019-01546-3
Keywords
- Sublinear expectation
- Reflected backward stochastic differential equations
- Dynamic programming principle