Abstract
The existence and uniqueness of the mild solution for semilinear elliptic partial differential equations associated with stochastic delay evolution equations are obtained by means of infinite horizon backward stochastic differential equations. Applications to optimal control for an infinite horizon are also given.
Similar content being viewed by others
1 Introduction
In this article we consider infinite horizon stochastic delay evolution equations of the form
where
W is a cylindrical Wiener process in a Hilbert space Ξ. A is the generator of a \(C_{0}\) semigroup in another Hilbert space H, and the coefficients F and G are assumed to satisfy Lipschitz conditions with respect to the appropriate norms.
Our approach to optimal control problems for stochastic delay evolution equations is based on backward stochastic differential equations (BSDEs), which were first introduced by Pardoux and Peng [1]: see [2–4], as general references. To be more precise, we consider the following forward-backward system:
If we assume ψ satisfies suitable conditions, then there exists a unique continuous adapted solution of (1.2) with λ large enough, denoted by \((X(\cdot ,x),Y(\cdot,x),Z(\cdot,x))\) in \(H\times R\times \Xi\). We define a deterministic function \(v: {\mathcal{C}}\rightarrow R\) by \(v(x)=Y(0,x)\), where \({\mathcal{C}}\) denotes the space of continuous functions from \([-\tau,0]\) to H, and it turns out that v is unique mild solution of the generalization of the nonlinear elliptic partial differential equation:
where
\(\{e_{i}\}_{i=1}^{\infty}\) denotes a basis of Ξ, 10 denotes the character function of \({\{0\}}\), \(\overline {\nabla_{x}\phi(x)}\), \(\overline{\nabla_{x}^{2}\phi(x)}\) denote the extensions of \(\nabla_{x}\phi(x)\), \(\nabla_{x}^{2}\phi(x)\), respectively (see Lemma 2.1 and Lemma 2.2). \(\overline{\nabla_{0}v(x)}\) is defined by \(\overline{\nabla_{0}v(x)}G(x)=\overline{\nabla_{x}v(x)}(G(x)1_{0})\). In this paper, we call (1.3) the nonlinear stationary Kolmogorov equation.
We consider a controlled equation of the form
The control process u takes values on a measurable space \((U,\mathcal{U})\). The aim is to minimize an infinite horizon cost functional of the form
over all admissible controls. Here q is function on \({\mathcal{C}}\times U\). We define the Hamiltonian function relative to the above problem: for all \(x\in\mathcal{C}\), \(z\in \Xi\),
Then, under suitable conditions, we eventually show that v is the value function of the control problem.
Stochastic optimal control problems have been studied by many authors. In [5–9], the authors proved there exists a direct (classical or mild) solution of the corresponding Hamilton-Jacobi-Bellman (HJB) equation, by which the optimal feedback law is obtained. Gozzi [8, 9] showed there exists unique mild solution of the associated HJB equation, where the diffusion term only satisfies weaker nondegeneracy conditions.
The viscosity solution methods have been successfully applied to stochastic optimal control problems (see [10–15] and references therein). Lions [13] proved that there exists a unique viscosity solution for a general class of fully nonlinear second order equations in an infinite dimensional Hilbert space. In [15], the existence and uniqueness of the viscosity solution for general unbounded second order partial differential equation were shown. Optimal control problems for stochastic differential equations with delay are considered in [10, 14, 16, 17]. Chang et al. [17] found that the value function for an optimal control problem of a general stochastic differential equation with a bounded memory is the unique viscosity solution of the HJB equation.
BSDEs were known to be useful tools in the study of stochastic optimal control problems; see, for example, [18, 19]. The optimal control problems for stochastic systems in infinite dimensions have been considered in [20–25]. Using Malliavin calculus and BSDEs, Fuhrman and Tessitore [22] showed that there exists a unique mild solution of nonlinear Kolmogorov equations and found that the mild solution coincides with the value function of the control problem. In Fuhrman and Tessitore [23], the existence and uniqueness of the mild solution for semilinear elliptic differential equations in Hilbert spaces were obtained by means of infinite horizon BSDEs in Hilbert spaces and Malliavin calculus, moreover, the existence of optimal control is proved by the feedback law. Fuhrman et al. [21] considered the optimal control problems for stochastic differential equations with delay and the associated Kolmogorov equations, the existence and uniqueness of the mild solution for the Kolmogorov equations was proved and the existence of optimal control was obtained. In Fuhrman et al. [20], the optimal ergodic control of a Banach valued stochastic evolution equation was studied and the optimal ergodic control was obtained by the ergodic BSDEs.
The main result of this paper is the proof of existence and uniqueness of the mild solution of (1.3) and (1.4). Some authors considered the Kolmogorov equations associated with stochastic evolution equations (see [22, 23]) and with stochastic delay differential equations (see [21]). However, as far as we know, there are few authors who concentrated on (1.3) and (1.4), for example, [26] and [27] for nonlinear parabolic partial differential equations. In this paper we want to extend the results of [23] to stochastic delay evolution equations in Hilbert spaces. Thanks to Lemma 2.1 and Lemma 2.2, we can consider the optimal control problem of (1.1) and the associated nonlinear Kolmogorov equations (1.3) and (1.4).
The plan of the paper is as follows. In the next section we introduce the basic notations and two basic lemmas. Section 3 is devoted to proving the regularity of the mild solution for infinite horizon stochastic delay evolution equation. The forward-backward system is considered and (4.10) is proved in Section 4. In Section 5, the mild solution of Kolmogorov equation (1.3) is considered. Finally, applications to optimal control for an infinite horizon are presented in Section 6.
2 Preliminaries
We list some notations that are used in this article. We use the symbols Ξ, K, and H to denote real separable Hilbert spaces, with scalar products \((\cdot,\cdot)_{\Xi}\), \((\cdot,\cdot)_{K}\), and \((\cdot,\cdot)_{H}\), respectively. Let \(|\cdot|\) denote the norm in various spaces, with a subscript if necessary. Let \({\mathcal {C}}=C([-\tau,0],H)\) denote the space of continuous functions from \([-\tau,0]\) to H, endowed with the usual norm \(|f|_{C}=\sup_{{\theta\in[-\tau,0]}}|f(\theta)|_{H}\). \(L(\Xi,H)\) denotes the space of all bounded linear operators from Ξ into H; the subspace of Hilbert-Schmidt operators, with the Hilbert-Schmidt norm, is denoted by \(L_{2}(\Xi,H)\).
Let \((\Omega,{\mathcal{F}},P)\) be a complete space with a filtration \(\{{\mathcal{F}}_{t}\}_{t\geq0}\) which satisfies the usual condition, i.e., \(\{{\mathcal{F}}_{t}\}_{t\geq0}\) is a right continuous increasing family of sub σ-algebra of \(\mathcal{F}\) and \({\mathcal{F}}_{0}\) contains all P-null sets of \(\mathcal{F}\). A cylindrical Wiener process defined on \((\Omega,{\mathcal{F}},P)\), and with values in a Hilbert space Ξ, is a family \(\{W(t),t\geq0\}\) of linear mappings \(\Xi\rightarrow L^{2}(\Omega)\) such that for every \(\xi, \eta\in\Xi\), \(\{W(t)\xi,t\geq0\}\) is a real Wiener process and \({E}(W(t)\xi\cdot W(t)\eta)=(\xi,\eta)_{\Xi}t\). In the following, \(\{W(t),t\geq0\}\) is a cylindrical Wiener process adapted to the filtration \(\{{\mathcal{F}}_{t}\}_{t\geq0}\).
Now let us define several classes of stochastic processes with values in a Banach space F:
-
\(L^{p}_{{\mathcal{P}}}(\Omega; L^{q}_{\beta}(F))\), defined for \(\beta\in R\) and \(p,q\in[1,\infty)\), denotes the space of equivalence classes of processes \(\{Y(s), s\geq 0\}\), with values in F, such that the norm
$$|Y|^{p}=E\biggl(\int^{\infty}_{0}e^{q\beta s} \bigl\vert Y(s)\bigr\vert ^{q}\, ds\biggr)^{\frac{p}{q}} $$is finite and Y admits a predictable version.
-
\({\mathcal{K}}^{p}_{\beta}\) denotes the space \(L^{p}_{{\mathcal {P}}}(\Omega; L^{2}_{\beta}(F))\times L^{p}_{{\mathcal{P}}}(\Omega; L^{2}_{\beta}(L_{2}(\Xi,F)))\). The norm of an element \((Y,Z)\in{\mathcal{K}}^{p}_{\beta}\) is \(|(Y,Z)|=|Y|+|Z|\).
-
\(L^{p}_{{\mathcal{P}}}(\Omega;C([0,T];F))\), defined for \(T>0\) and \(p\in[1,\infty)\), denotes the space of predictable processes \(\{Y(s), s\in[0,T]\}\) with continuous paths in F, such that the norm
$$|Y|^{p}=E\sup_{s\in[0,T]}\bigl\vert Y(s)\bigr\vert ^{p} $$is finite. Elements of \(L^{p}_{{\mathcal {P}}}(\Omega;C([0,T];F))\) are identified up to indistinguishability.
-
\(L^{q}_{{\mathcal{P}}}(\Omega;C_{\eta}(F))\), defined for \(\eta\in R\) and \(q\in[1,\infty)\), denotes the space of predictable processes \(\{Y(s), s\geq 0\}\) with continuous paths in F, such that the norm
$$|Y|^{q}=E\sup_{s\geq0}e^{\eta qs}\bigl\vert Y(s)\bigr\vert ^{q} $$is finite. Elements of \(L^{q}_{{\mathcal {P}}}(\Omega;C_{\eta}(F))\) are identified up to indistinguishability.
-
Finally, for \(\eta\in R\) and \(q\in[1,\infty)\), we defined \({\mathcal{H}}^{q}_{\eta}\) as the space \(L^{q}_{{\mathcal{P}}}(\Omega; L^{q}_{\eta}(F))\cap L^{q}_{{\mathcal {P}}}(\Omega;C_{\eta}(F))\), endowed with the norm
$$|Y|_{{\mathcal{H}}^{q}_{\eta}}=|Y|_{L^{q}_{{\mathcal{P}}}(\Omega; L^{q}_{\eta}(F))}+|Y|_{L^{q}_{{\mathcal {P}}}(\Omega;C_{\eta}(F))}. $$
We say \(f\in C^{1}({\mathcal{C}};H)\) if f is continuous, Fréchet differentiable and \(\nabla_{x}f:{\mathcal{C}}\rightarrow L({\mathcal{C}},H)\) is continuous.
We say \(g\in{\mathcal{G}}^{1}({\mathcal{C}};H)\) if g is continuous, Gâteaux differentiable with respect to x on \({\mathcal{C}}\) and \(\nabla_{x}g:{\mathcal{C}}\rightarrow L({\mathcal{C}},H)\) is strongly continuous.
Let \(F_{c}\) be the vector space of all simple functions of \(z1_{[a,c)}\) where \(z\in H\), \(c\in(a,b]\) and \(1_{[a,c)}:[a,b]\rightarrow R\) is the character function of \([a,c)\). It is clear that \(C([a,b],H)\cap F_{c}=\{0\}\). Form the direct sum \(C([a,b],H)\oplus F_{c}\) and give it the complete norm
Now let us give two basic lemmas that will be used in the following sections (see Lemma 2.1 and Lemma 2.2 in [26]).
We say \({f}:C([a,b],H)\oplus F_{c}\rightarrow K\) satisfying (V) if \(\{x^{n}\}_{n\geq1}\) is a bounded sequence in \(C([a,b],H)\) and \(y+z1_{[a,c)}\in C([a,b],H)\oplus F_{c}\) such that \(x^{n}(s)\rightarrow y(s)+z1_{[a,c)}\) as \(n\rightarrow \infty\) for all \(s\in[a,b]\), and \(\sup_{s\in[a,b]}|(x^{n}(s)-y(s),h_{i})|\leq |(z,h_{i})|\) for all \(i\in N\), where \(\{h_{i}\}_{i\geq1}\) is a basis of H, then \(f(x^{n})\rightarrow f(y+z1_{[a,c)})\) as \(n\rightarrow\infty\).
Lemma 2.1
Let \(f\in L(C([a,b];H);K)\). Then, for every \(c\in (a,b]\), f has a unique continuous linear extension \(\overline{f}:C([a,b],H)\oplus F_{c}\rightarrow K\) satisfying (V). Moreover, the extension map \(e:L(C([0,T],H),K)\rightarrow L(C([0,T],H)\oplus F_{c},K)\), \(f\rightarrow\overline{f}\) is a linear isometry.
We say \({f}:[C([a,b],H)\oplus F_{c}]\times[C([a,b],H)\oplus F_{c}]\rightarrow R\) satisfies (W) if \(\{x^{n}\}_{n\geq1}\), \(\{y^{n}\}_{n\geq1}\) are bounded sequences in \(C([a,b],H)\) and \(x+z_{1}1_{[a,c)}, y+z_{2}1_{[a,c)}\in C([a,b],H)\oplus F_{c}\) such that \(x^{n}(s)\rightarrow x(s)+z_{1}1_{[a,c)}\), \(y^{n}(s)\rightarrow y(s)+z_{2}1_{[a,c)}\) as \(n\rightarrow \infty\) for all \(s\in[a,b]\), and \(\sup_{s\in[a,b]}|(x^{n}(s)-x(s),h_{i})|\leq |(z_{1},h_{i})|\), \(\sup_{s\in[a,b]}|(y^{n}(s)-y(s),h_{i})|\leq |(z_{2},h_{i})|\) for all \(i\in N\), then \(f(x^{n},y^{n})\rightarrow f(x+z_{1}1_{[a,c)},y+z_{2}1_{[a,c)})\) as \(n\rightarrow\infty\).
Lemma 2.2
Let \(\beta:C([a,b],H)\times C([a,b];H)\rightarrow R\) be a continuous bilinear map. Then, for every \(c\in(a,b]\), β has a unique continuous bilinear extension \(\overline{\beta}:[C([a,b],H)\oplus F_{c}]\times [C([a,b],H)\oplus F_{c}]\rightarrow R\) satisfying (W).
3 The forward equation
In this section we consider the system of stochastic delay evolution equations:
We make the following assumptions.
Hypothesis 3.1
-
(i)
The operator A is the generator of a strongly continuous semigroup \(\{e^{tA}, t\geq0\}\) of bounded linear operators in the Hilbert space H. We denote by M and ω two constants such that \(|e^{tA}|\leq Me^{\omega t}\), for \(t\geq0\).
-
(ii)
The mapping \(F: {\mathcal{C}}\rightarrow H\) is measurable and satisfies, for some constant \(L>0\),
$$ \bigl\vert F(x)\bigr\vert \leq L \bigl(1 + \vert x\vert _{C} \bigr), \qquad \bigl\vert F(x)-F(y)\bigr\vert \leq L|x-y|_{C}, \quad x,y\in \mathcal{C}. $$ -
(iii)
The mapping \(G: {\mathcal{C}}\rightarrow L_{2}(\Xi,H)\) is measurable and satisfies, for every \(x,y\in\mathcal{C}\),
$$ \bigl\vert G(x)\bigr\vert _{L_{2}(\Xi,H)}\leq L\bigl(1+\vert x\vert _{C}\bigr),\qquad \bigl\vert G(x)-G(y)\bigr\vert _{L_{2}(\Xi,H)} \leq L|x-y|_{C} $$(3.2)for some constants \(L>0\).
-
(iv)
\(F(\cdot)\in C^{1}({\mathcal{C}},H)\), \(G(\cdot)\in C^{1}({\mathcal{C}},L_{2}(\Xi,H))\).
We say that X is a mild solution of equation (3.1) if it is a continuous, \(\{{\mathcal{F}}_{t}\}_{t\geq0}\)-predictable process with values in H, and it satisfies, P-a.s.,
We define \(G^{N}(t,x)\) from \({\mathcal{C}}\) to \(L_{2}(\Xi;H)\) by
where \(\{e_{i}\}_{i\geq1}\) denotes a basis of Ξ. We denote by \(X^{N}(\cdot,x)\) the solution of (3.3) with respect to \(G^{N}\).
We first recall two well-known results of (3.1) on a bounded interval.
Theorem 3.2
(Theorem 3.2 in [26])
Assume that Hypothesis 3.1 holds. Then for all \(q\in[2,\infty)\) and \(T>0\) there exists a unique process \(X\in L^{q}_{\mathcal{P}}(\Omega,C([0,T];H))\), a mild solution of (3.1). Moreover,
for some constant C depending only on q, T, τ, L, ω, and M.
Theorem 3.3
(Theorem 4.2 in [26])
Assume that Hypothesis 3.1 holds and let \(u: {\mathcal{C}}\rightarrow R\) be a Borel measurable function such that \(u(\cdot)\in{C}^{1}({\mathcal{C}},R)\) and
for some \(C>0\), \(m\geq0\) and for every \(x\in{\mathcal{C}}\). Then the joint quadratic variation on \([0,T']\),
for every \(x\in{\mathcal{C}}\), \(i=1,2,\ldots,d\) and \(0\leq T'< T\).
By Theorem 3.2 and the arbitrariness of T in its statement, the solution is defined for every \(s\geq 0\). To stress the dependence on the initial data, we denote the solution by \(X(x)\). We have the following result.
Theorem 3.4
Assume that Hypothesis 3.1(i)-(iii) holds and the process \(X(\cdot,x)\) is a mild solution of (3.1) with initial value \(x\in {\mathcal{C}}\). Then, for every \(q\in[1,\infty)\), there exists a constant \(\eta(q)\) such that the process \(X_{\cdot}(x)\in{\mathcal {H}}^{q}_{\eta(q)}\). Moreover, for a suitable constant \(C>0\), we have
where the constants \(\eta(q)\) depending only on q, τ, L, ω, and M.
If we assume that Hypothesis 3.1(iv) also holds true, then the map \(x\rightarrow X_{\cdot}(x)\) belongs to \(C ^{1}( {\mathcal{C}},{\mathcal {H}}^{q}_{\eta(q)})\) and for every \(h\in{\mathcal{C}}\), the process \(\nabla_{x}X_{s}(x)h\), \(s\in[0,+\infty)\) solves, P-a.s., the following equation:
Moreover, \(|\nabla_{x}X_{s}(x)h|_{{\mathcal{H}}^{q}_{\eta(q)}}\leq C|h|_{C}\) for a suitable constant \(C>0\).
Proof
We define a mapping Φ from \({\mathcal{H}}^{q}_{\eta}\times \mathcal{C}\) to \({\mathcal{H}}^{q}_{\eta}\) by the formula
We will prove that, provided η is suitably chosen, \(\Phi(\cdot ,x)\) is a contraction in \({\mathcal{H}}^{q}_{\eta}\), uniformly in x, or in other words, there exists \(c<1\) such that for every \(x\in\mathcal{C}\),
For simplicity, we treat only the case \(F=0\), the general case being handled in a similar way. We will use the so called factorization method; see [28], Theorem 5.2.5. Let us take \(q>1\) and \(\alpha\in(0,1)\) such that
by the stochastic Fubini theorem,
where
By \(\sup_{-\tau\leq l\leq0}|e^{(s+l)A}x(0)|\leq Me^{\omega s}|x|_{C}\), we see that the process \(e^{(s+\cdot)A}x(0)\), \(s\geq0\), belongs to \({\mathcal{H}}^{q}_{\eta}\) provided \(\omega+\eta<0\). Next let us estimate \(\Phi'(X_{\cdot})\), since
setting \(q'=\frac{q}{(q-1)}\),
Applying the Young inequality for convolutions, we have
and we obtain
If we start again from (3.8) and apply the Hölder inequality, we obtain, for \(\eta<0\),
and
So we conclude that
On the other hand, by the Burkholder-Davis-Gundy inequalities, for some constant \(c_{q}\) depending only on q, we have
which implies
so that
for suitable constants \(C_{1}\), \(C_{2}\). Applying the Young inequality for convolutions, we obtain
This shows that \(|Y|_{L^{q}_{\mathcal {P}}(\Omega;L^{q}_{\eta}(H))}\) is finite provided \(\eta<0\) and \(\omega+\eta<0\), so the map is well defined.
If \(X_{\cdot}^{1}\), \(X_{\cdot}^{2}\) are processes belonging to \({\mathcal{H}}^{q}_{\eta}\) and \(Y^{1}\), \(Y^{2}\) are defined accordingly, similarly we find that
By the inequalities (3.9) and (3.10), we obtain an explicit expression for the constant c in (3.7) and it is immediate to find that \(c<1\) provided \(\eta<0\) is chosen sufficiently small. We fix such a value \(\eta(q)\). The first result is a consequence of the contraction principle. We get the estimate (3.5) also by the contraction property of \(\Phi(\cdot,x)\).
Now let us study the regular dependence of the solution on the initial datum. Firstly, we show that the map \(x\rightarrow X_{\cdot}(x)\) belongs to \({\mathcal{G}}^{1}({\mathcal{C}},{\mathcal {H}}^{q}_{\eta(q)})\). By the parameter depending contraction principle (see [29]), it suffices to prove that
We split the proof into several steps.
Step 1. It is clear that Φ is continuous.
Step 2. The directional derivative \(\nabla_{X}\Phi(X,x)\) in the direction \(N\in{\mathcal{H}}^{q}_{\eta (q)}\) satisfies
Moreover, the mappings \((X,x)\rightarrow \nabla_{X}\Phi(X,x;N)\) and \(N\rightarrow \nabla_{X}\Phi(X,x;N)\) are continuous. We only prove this claim in the special case \(F=0\). For fixed \(x\in{\mathcal{C}}\), for all \(s\geq0\), we define
By a similar procedure, we show that, for \(\frac{1}{q}<\alpha<\frac{1}{2} \) and a suitable constant \(c_{q}\),
where
Thus, for a suitable constant c,
and setting
we find that
From Hypothesis 3.1(iii) and (iv) it follows that \(|\nabla_{x}G(x)h|\leq L|h|\), which implies
Moreover,
Since \(e^{rA}\nabla_{x}G(x)h\) is continuous in x, then, by the dominated convergence theorem, it follows that \(E\int^{\infty}_{0}e^{r\eta(q)q}|Y^{\varepsilon}(r)|^{q}\,dr\rightarrow0\).
In a similar way we find that the mappings \((X,x)\rightarrow \nabla_{X}\Phi(X,x;N)\) and \(N\rightarrow \nabla_{X}\Phi(X,x;N)\) are continuous.
Step 3. It is easy to show that the directional derivative \(\nabla_{x}\Phi(X,x;h)\) in direction \(h\in{\mathcal{C}}\) is the process given by
and the mappings \((X,x)\rightarrow \nabla_{x}\Phi(X,x;h)\) and \(h\rightarrow \nabla_{x}\Phi(X,x;h)\) are continuous.
From the parameter depending contraction principle (see [29]), it follows that (3.6) holds true. The final estimate is a trivial consequence of (3.6).
Now we have to prove that the map \(x\rightarrow X_{\cdot}(x)\) belongs to \(C^{1}({\mathcal{C}},{\mathcal{H}}^{q}_{\eta (q)})\). For simplicity, we set \(F=0\). For every \(x,y,h\in{\mathcal{C}}\), by (3.6) we have
By a similar procedure, we find that, for some constant \(c_{q}\), it may vary from line to line,
Letting \(\eta(q)\) be sufficiently small such that \(c_{q}(\int_{0}^{\infty}s^{-2\alpha}e^{2(\omega+\eta (q))s}\,ds)^{\frac{q}{2}} (\int_{0}^{\infty}e^{(\eta(q)+\omega) s}\times s^{q'(\alpha-1)}\,ds)^{\frac{q}{q'}} (\int_{0}^{\infty}e^{(\eta(q)+\omega)s }\,ds+1)< 1\), we find that
Since \(x\rightarrow X_{\cdot}(x)\) is continuous from \({\mathcal{C}}\) to \({\mathcal{H}}^{q}_{\eta(q)}\), if we assume \(x\rightarrow y\), then there exists a subsequence \(\{x_{n}\}\) such that \(X_{\cdot}(x_{n})\rightarrow X_{\cdot}(y)\), P-a.s. As \(\nabla_{x}G(x)\) is continuous in x and \(|(\nabla_{x}G(X_{\sigma}(x))-\nabla_{x} G(X_{\sigma}(y)))|^{2p}_{L({\mathcal{C}};L_{2}(\Xi,H))}\leq(2L)^{2p}\), by the dominated convergence theorem, we have
Let us assume that there exists a subsequence \(\{x_{m}\}\) such that
Then there exists a subsequence \(\{x_{m_{k}}\}\) of \(\{x_{m}\}\) and a constant \(\varepsilon>0\) such that
On the other hand, by a similar procedure, there exist a subsequence \(\{ x_{m_{k_{l}}}\}\) of \(\{x_{m_{k}}\}\) such that
It is a contradiction. Thus we obtain
The proof is finished. □
4 The backward-forward system
In this section we consider the system of stochastic differential equations, P-a.s.,
We make the following assumptions.
Hypothesis 4.1
-
(i)
The mapping \(\psi: {\mathcal{C}}\times K\times L_{2}(\Xi,K)\rightarrow K\) is continuous and, for some \(L>0\), \(\mu\in R\), and \(m\geq1\),
$$\begin{aligned}& \bigl\vert \psi(x,y_{1},z_{1})-\psi(x,y_{2},z_{2}) \bigr\vert \leq L|y_{1}-y_{2}|+L|z_{1}-z_{2}|, \\& \bigl\vert \psi(x,y,z)\bigr\vert \leq L\bigl(1+|x|_{C}^{m}+|y|+|z| \bigr), \\& \bigl\langle \psi(s,x,y_{1},z)-\psi(s,x,y_{2},z),y_{1}-y_{2} \bigr\rangle _{K}\geq\mu|y_{1}-y_{2}|^{2} \end{aligned}$$for every \(x\in{\mathcal{C}}\), \(y,y_{1},y_{2}\in K\), \(z,z_{1},z_{2}\in L_{2}(\Xi,K)\).
-
(ii)
The map \(\psi(\cdot,\cdot,\cdot)\in{C}^{1}({\mathcal {C}}\times R\times L_{2}(\Xi,R); R)\) and there exist \(L>0\) and \(m\geq0\) such that
$$\bigl\vert \nabla_{x}\psi(x,y,z)h\bigr\vert \leq L|h|_{C}\bigl(1+\vert x\vert _{C}+|y| \bigr)^{m}\bigl(1+\vert z\vert \bigr) $$for every \(x,h\in{\mathcal{C}}\), \(y\in R\), and \(z\in L_{2}(\Xi,R)\).
For the backward-forward system (4.1) we have the following basic result (see Proposition 3.11 and Proposition 5.1 in [23]).
Theorem 4.2
Assume that Hypotheses 3.1 and 4.1 hold. There exists a constant \(\eta (q)\) such that, for \(p\in(2,+\infty)\), β, and q satisfying
and for every \(\lambda>\hat{\lambda}=-(\beta+\mu-L^{2}/2)\), the following hold:
-
(i)
For every \(x\in{\mathcal{C}}\), there exists a unique solution in \({\mathcal{H}}^{q}_{\eta(q)}\times{\mathcal{K}}^{p}_{\beta}\) of (4.1), which will be denoted by \((X(\cdot,x),Y(\cdot,x),Z(\cdot,x))\). Moreover, \(Y(x)\in L^{p}_{\mathcal{P}}(\Omega;C_{\beta}(R))\).
-
(ii)
The maps \(x\rightarrow X(x)\), \(x\rightarrow(Y(x),Z(x))\), \(x\rightarrow Y(x)\) belong to \({\mathcal{G}}^{1}({\mathcal{C}};{\mathcal {H}}^{q}_{\eta(q)})\), \({\mathcal{G}}^{1}({\mathcal{C}};{\mathcal {K}}_{\beta}^{p})\), and \({\mathcal{G}}^{1}({\mathcal{C}};L^{p}_{\mathcal {}P}(\Omega;C_{\beta}(R)))\), respectively. Moreover, for every \(h\in {\mathcal{C}}\), \((\nabla_{x}Y(s,x)h,\nabla_{x}Z(s,x)h)\) solves the equation, P-a.s.,
$$\begin{aligned}& \nabla_{x}Y(s,x)h-\nabla_{x}Y(T,x)h+ \lambda\int^{T}_{s}\nabla_{x}Y( \sigma,x)h\, d\sigma +\int^{T}_{s} \nabla_{x}Z(\sigma,x)h\, dW(\sigma) \\& \quad = -\int^{T}_{s}\nabla_{x}\psi \bigl( X_{\sigma}(x),Y(\sigma,x),Z(\sigma,x)\bigr)\nabla_{x}X_{\sigma}(x)h \, d\sigma \\& \qquad {}-\int^{T}_{s}\nabla_{y} \psi\bigl( X_{\sigma}(x),Y(\sigma,x),Z(\sigma,x)\bigr)\nabla_{x}Y( \sigma,x)h\, d\sigma \\& \qquad {}-\int^{T}_{s}\nabla_{z} \psi\bigl( X_{\sigma}(x),Y(\sigma,x),Z(\sigma,x)\bigr)\nabla_{x}Z( \sigma,x)h\, d\sigma, \quad s\in[0,+\infty). \end{aligned}$$(4.3) -
(iii)
For every \(x,h\in\mathcal{C}\), \(t\geq0\) there exists a suitable constant \(c>0\) independent of x and h such that
$$\begin{aligned}& E\sup_{s\geq0}e^{p\beta s}\bigl\vert \nabla_{x} Y(s,x)h\bigr\vert ^{p} +E\biggl(\int ^{+\infty}_{0}e^{2\beta\sigma}\bigl\vert \nabla_{x} Y(\sigma ,x)h\bigr\vert ^{2} \biggr)^{\frac{p}{2}} \\& \qquad {}+ E\biggl(\int^{+\infty}_{0}e^{2\beta\sigma} \bigl\vert \nabla_{x} Z(\sigma ,x)h\bigr\vert ^{2} \biggr)^{\frac{p}{2}} \\& \quad \leq c|h|^{p}_{C}\bigl(1+|x|_{C}^{m^{2}} \bigr)^{p}. \end{aligned}$$(4.4)
Theorem 4.3
Under the assumptions of Theorem 4.2, we have
Proof
Setting \(\nabla_{x}Y^{\lambda}(s)=e^{-\lambda s}\nabla_{x}Y(s)\) and \(\nabla_{x}Y^{\lambda}(s)=e^{-\lambda s}\nabla_{x}Y(s)\), by the Itô formula, (4.3) is equivalent to
For every \(T>0\), we can show that, for a suitable constant \(c_{p}>0\) depending only on p,
On the other hand, for every \(\varepsilon>0\), letting T be large enough, by estimate (4.4) we see that
Therefore,
The proof is finished. □
Corollary 4.4
Assume that Hypotheses 3.1 and 4.1 hold. Then the function \(v^{N}(x)=Y^{N}(0,x)\) belongs to \({C}^{1}({\mathcal{C}};R)\) and there exists a constant \(C>0\) independent of N such that \(|\nabla_{x} v^{N}(x)h|\leq C|h|_{C}(1+|x|_{C}^{m^{2}})\) for all \(x,h\in{\mathcal{C}}\).
Moreover, for every \(x\in {\mathcal{C}}\), we have
Proof
By Theorem 4.2 and Theorem 4.3, it is easy to find that \(v^{N}\in{C}^{1}({\mathcal{C}};R)\), and its property is a direct consequence of Theorem 4.2. By a similar procedure to Theorem 6.1 in [23], we obtain \(Y^{N}(s,x)=\upsilon^{N}(X^{N}_{s}(x))\), P-a.s. for all \(s\geq0\).
For every \(T>0\), we consider the joint quadratic variation of \(Y^{N}(\cdot,x)\) and the Wiener process \(We_{i}\) on an internal \([0,T]\). By the backward stochastic differential equation we get
From Theorem 3.3 it follows that
By the arbitrariness of T, we obtain
□
Theorem 4.5
Let us assume that Hypotheses 3.1 and 4.1 hold true. In addition, we assume that
Then, for every \(p>2\), we have
in particular, we find that
Proof
The proof is very similar to Theorem 4.3, so we omit it. □
Now we are in a position to prove the main result of this section.
Theorem 4.6
Assume that Hypotheses 3.1 and 4.1 hold, and let (4.7) hold true. Then the function \(v(x)=Y(0,x)\) belongs to \(C^{1}({\mathcal{C}};R)\) and there exists a constant \(C>0\) such that \(|\nabla_{x} v(x)h|\leq C|h|_{C}(1+|x|_{C}^{m^{2}})\) for all \(x\in{\mathcal{C}}\) and \(h\in {\mathcal{C}}\). Moreover, for every \(x\in {\mathcal{C}}\), we have
Proof
By a similar procedure to Corollary 4.4, we can show the statements except (4.10).
It follows from (4.5) and (4.9) that
Consequently,
Since
we see that, for every \(T>0\), there exists a subsequence \(\{N_{k}\}\) such that
By (4.6) and (4.11) we deduce that
By the arbitrariness of T, we have
The proof is finished. □
5 Mild solution of the Kolmogorov equation
Let \(X(\cdot,x)\) denote the unique solution of (3.3). We denote by \({\mathcal {B}}(\mathcal{C})\) the set of measurable functions \(\phi:{\mathcal{C}}\rightarrow R\) with polynomial growth. The transition semigroup \(P_{s}\) is defined for arbitrary \(\phi\in{\mathcal {B}}(\mathcal{C})\) by the formula
We study a generalization of the Kolmogorov equation of the following form:
where
\(\{e_{i}\}_{i=1}^{\infty}\) denotes a basis of Ξ.
Definition 5.1
We say a function \(v: {\mathcal{C}}\rightarrow R\) is a mild solution of the nonlinear Kolmogorov equation (5.1), \(v\in{\mathcal{G}}^{1}( {\mathcal{C}};R)\), if there exist some constants \(C>0\), \(q\geq0\) such that
and the following equality holds true, for every \(x\in{\mathcal{C}}\) and \(T\geq0\):
Theorem 5.2
Assume that Hypotheses 3.1 and 4.1 hold, and let (4.7) hold true. Then there exists \(\hat{\lambda}\in R\) such that, for every \(\lambda>\hat{\lambda}\), the nonlinear stationary Kolmogorov equation (5.1) has a unique mild solution v. The function v coincides with the one introduced in Theorem 4.6.
Proof
(Existence) Let v be the function defined in Theorem 4.6. Then v has the regularity properties stated in Definition 5.1. It remains to prove that equality (5.3) holds true. By Theorem 4.6 we have
Thus
Applying the Itô formula to the backward equation in (4.1) gives
Taking the expectation and applying (5.4) we obtain the equality (5.3).
(Uniqueness) Let v be a mild solution of (5.1), by (5.3) we have, for every \(x\in{\mathcal{C}}\), \(0\leq s\leq T\),
By the Markov property of X, we obtain
then by a change of variable, we have
where
Now let \(T>0\) be fixed, then by the representation theorem (see Proposition 4.1 in [22]), there exists \(\widetilde{Z}\in L^{2}_{\mathcal {P}}(\Omega\times[0,T],\Xi)\) such that \(E[\xi|{\mathcal{F}}_{s}]=v(x)+\int^{s}_{0}{\widetilde{Z}}(\sigma )\,dW(\sigma)\), \(s\in[0,T]\). Therefore, by the Itô formula, we find that
By Theorem 3.3 and Theorem 4.6 we have \(\langle v(X_{\cdot}(x)), We_{i}\rangle_{[0,T']}=\int^{T'}_{0}\overline{\nabla_{0}v(X_{\sigma}(x))}G(X_{\sigma}(x)) e_{i}\, d\sigma\) for every \(T'\in[0,T)\). Hence, \(e^{\lambda\sigma}{\widetilde{Z}}(\sigma)=\overline{\nabla _{0}v(X_{\sigma}(x))}G(X_{\sigma}(x))\), P-a.s., and equality (5.5) can be rewritten as
By the arbitrariness of T, we see that the pairs \((Y(s,x),Z(s,x))\) and \((v(X_{s}(x)), \overline{\nabla_{0}v(X_{s}(x))}G(X_{s}(x)))\), \(s\geq0\), solve the same backward stochastic differential equation in (4.1). By uniqueness, we have \(Y(s,x)=v(X_{s}(x))\), \(s\geq0\). Setting \(s=0\) we show \(Y(0,x)=v(x)\). The proof is finished. □
6 Application to optimal control
In this section we study the controlled state equation:
The solution of the above equation will be denoted by \(X^{u}(s,x)\) or simply by \(X^{u}(s)\). Our aim is to minimize the cost functional
over all the admissible control system.
We formulate the optimal control problem in the weak sense following the approach of [30]. By an admissible control system we mean \((\Omega,{\mathcal {F}},\{{{\mathcal{F}}}_{t}\}_{t\geq0},P,W,u,X^{u})\), where \((\Omega,{\mathcal{F}},\{{{\mathcal {F}}}_{t}\}_{t\geq0},P)\) is a filtered probability space satisfying the usual conditions, W is a cylindrical P-Wiener process with values in Ξ, adapted to the filtration \(\{{{\mathcal{F}}}_{t}\}_{t\geq0}\). u is an \({{\mathcal{F}}}_{t}\)-predictable process with values in U, \(X^{u}\) denotes a mild solution of (6.1). An admissible control system will be briefly denoted by \((W,u,X^{u})\) in the following.
We define in a classical way the Hamiltonian function relative to the optimal control problem: for every \(x\in\mathcal{C}\), \(z\in \Xi\),
and the corresponding, possibly empty, set of minimizers
We are now ready to formulate the assumptions we need.
Hypothesis 6.1
-
(i)
A, F, and G verify Hypothesis 3.1, moreover, G satisfies (4.7).
-
(ii)
\((U,\mathcal{U})\) is a measurable space. The map \(g: {\mathcal {C}\times} U\rightarrow R\) is continuous and satisfies \(|g(x,u)|\leq K_{g}(1+|x|_{C}^{m_{g}})\) for suitable constants \(K_{g}>0\), \(m_{g}>0\), and every \(x\in \mathcal{C}\), \(u\in U\). The map \(R: {\mathcal{C}\times} U\rightarrow\Xi\) is measurable and \(|R(x,u)|\leq L_{R}\) for suitable constants \(L_{R}>0\) and every \(x\in \mathcal{C}\), \(u\in U\).
-
(iii)
The Hamiltonian ψ defined in (6.3) satisfies the requirements of Hypothesis 4.1 (with \(K=R\)).
-
(iv)
We fix \(p>2\), q, and \(\beta<0\) satisfying (4.2), and such that \(q>m_{g}\).
We are in a position to prove the main result of this section.
Theorem 6.2
We assume that Hypothesis 6.1 holds true and λ verifies
and suppose that the set-valued map Γ has non-empty values and it admits a measurable selection \(\Gamma_{0}: {\mathcal{C}}\times\Xi\rightarrow U\). Let υ denote the function in the statement of Theorem 4.6. Then for all admissible control systems we have \(J(u)\geq\upsilon(x)\) and the equality holds if and only if
Moreover, the closed loop equation
admits a weak solution \((\Omega,{\mathcal{F}},\{{{\mathcal {F}}}_{t}\}_{t\geq0},P,W,X)\) which is unique in law and setting
we get an optimal admissible control system \((W,u,X)\).
Proof
We consider (6.1) in the probability space \((\Omega,{\mathcal{F}}, P)\) with filtration \(\{{\mathcal{F}}_{t}\}_{t\geq0}\) and with an \(\{{\mathcal{F}}_{t}\}_{t\geq0}\)-cylindrical Wiener process \(\{W(t), t\geq0\}\). Let us define
and
Let \(P^{u}\) be the unique probability on \({\mathcal {F}}_{[0,\infty)}\) such that
We note that under \(P^{u}\), the process \(W^{u}\) is a Wiener process. Let us denote by \(\{{\mathcal {F}}^{u}_{t}\}_{t\geq0}\) the filtration generated by \(W^{u}\) and completed in the usual way. Relatively to \(W^{u}\), (6.1) can be rewritten
In the space \((\Omega,{\mathcal{F}}_{[0,\infty)},\{{\mathcal {F}}^{u}_{t}\}_{t\geq0},P^{u})\), we consider the system of forward-backward equations
Applying the Itô formula to \(e^{-\lambda s}Y^{u}(s)\) and writing the backward equation in (6.7) with respect to the process W we get
Recalling that R is bounded, we have, for all \(r\geq1\) and some constant C,
It follows that
We see that the stochastic integral in (6.8) has zero expectation. If we take the expectation with respect to P in (6.8), we obtain
By Theorem 4.2, \(Y^{u}(\cdot,x)\in L^{p}_{\mathcal {P}}(\Omega;C_{\beta}(R))\), so that
By the Hölder inequality we see that, for suitable constants \(C>0\),
By Theorem 3.4 we obtain \(E^{u}\sup_{s\geq t}e^{\eta(q) qs}|X^{u}_{s}|^{q}<\infty\), by a similar process, we find that
for suitable constants \(C>0\), and
Since \(Y^{u}(0,x)=\upsilon(x)\) and \(Z^{u}(s,x)=\overline{\nabla _{0}v(X_{s}^{u}(x))}G(X^{u}_{s}(x))\), P-a.s. for a.a. \(s\in[0,\infty)\) we have
Thus adding and subtracting \(E\int^{\infty}_{0}e^{-\lambda \sigma}g(X^{u}_{\sigma},u(\sigma))\,d\sigma\) and letting \(T\rightarrow\infty\), we conclude that
It implies that \(J(u)\geq\upsilon(x)\) and that the equality holds true if and only if (6.5) holds true.
(Uniqueness) Let X be a weak solution of (6.6) in an admissible set-up \((\Omega,{\mathcal{F}},\{{\mathcal {F}}_{t}\}_{t\geq0}, P,W)\). We define
Since R is bounded, the Girsanov theorem ensures that there exists a probability measure \(P^{0}\) such that the process
is a \(P^{0}\)-Wiener process and
We denote by \(\{{\mathcal {F}}^{0}_{t}\}_{t\geq0}\) the filtration generated by \(W^{0}\) and completed in the usual way. In \((\Omega,{\mathcal{F}}_{[0,\infty)},\{{\mathcal{F}}^{0}_{t}\} _{t\geq0},P^{0})\), X is a mild solution of
and
Note the joint law of X and \(W^{0}\) is uniquely determined by A, F, G, and x. Taking into account the above formula, we conclude that the joint law of X and \(\rho(T)\) under \(P^{0}\) is also uniquely determined, and consequently so is the law of X under P. This completes the proof of the uniqueness part.
(Existence) Let \((\Omega,{\mathcal{F}}, P)\) be a given complete probability space. \(\{W(t),t\geq0\} \) is a cylindrical Wiener process on \((\Omega,{\mathcal{F}}, P)\) with values in Ξ, \(\{ {\mathcal{F}}_{t} \}_{t\geq0}\) is the natural filtration of \(\{W(t), t\geq0\}\), augmented with the family of P-null sets. Let \(X(\cdot)\) be the mild solution of
and by the Girsanov theorem, let \(P^{1}\) be the probability on Ω under which
is a Wiener process (notice that R is bounded). Then X is the weak solution of (6.6) relative to the probability \(P^{1}\) and the Wiener process \(W^{1}\). The proof is completed. □
References
Pardoux, E, Peng, S: Adapted solution of a backward stochastic differential equation. Syst. Control Lett. 14, 55-61 (1990)
Briand, P, Hu, Y: Stability of BSDEs with random terminal time and homogenization of semilinear elliptic PDEs. J. Funct. Anal. 155, 455-494 (1998)
El Karoui, N, Mazliak, L (eds.): Backward Stochastic Differential Equations. Pitman Research Notes in Mathematics Series, vol. 364. Longman, Harlow (1997)
Pardoux, E: BSDEs, weak convergence and homogenization of semilinear PDEs. In: Clarke, FH, Stern, RJ (eds.) Nonlinear Analysis, Differential Equations and Control, pp. 503-549. Kluwer Academic, Dordrecht (1999)
Barbu, V, Da Prato, G: Hamilton-Jacobi Equations in Hilbert Spaces. Pitman Research Notes in Mathematics, vol. 86. Pitman, London (1983)
Cannarsa, P, Da Prato, G: Second-order Hamilton-Jacobi equations in infinite dimensions. SIAM J. Control Optim. 29(2), 474-492 (1991)
Cannarsa, P, Da Prato, G: Direct solution of a second-order Hamilton-Jacobi equations in Hilbert spaces. In: Da Prato, G, Tubaro, L (eds.) Stochastic Partial Differential Equations and Applications. Pitman Research Notes in Mathematics, vol. 268, pp. 72-85. Pitman, London (1992)
Gozzi, F: Regularity of solutions of second order Hamilton-Jacobi equations and application to a control problem. Commun. Partial Differ. Equ. 20, 775-826 (1995)
Gozzi, F: Global regular solutions of second order Hamilton-Jacobi equations in Hilbert spaces with locally Lipschitz nonlinearities. J. Math. Anal. Appl. 198, 399-443 (1996)
Elsanousi, I: Stochastic control for system with memory. Dr. Scient. thesis, University of Oslo (2000)
Lions, PL: Viscosity solutions of fully nonlinear second-order equations and optimal stochastic control in infinite dimensions. Part I. The case of bounded stochastic evolutions. Acta Math. 161(3-4), 243-278 (1988)
Lions, PL: Viscosity solutions of fully nonlinear second order equations and optimal stochastic control in infinite dimensions. Part II. Optimal control of Zakai’s equation. In: Da Prato, G, Tubaro, L (eds.) Stochastic Partial Differential Equations and Applications II. Lecture Notes in Mathematics, vol. 1390, pp. 147-170. Springer, Berlin (1989)
Lions, PL: Viscosity solutions of fully nonlinear second-order equations and optimal stochastic control in infinite dimensions. III. Uniqueness of viscosity solutions for general second-order equations. J. Funct. Anal. 86(1), 1-18 (1989)
Øksendal, B, Sulem, A: A maximum principle for optimal control of stochastic systems with delay, with applications to finance. In: Mendaldi, JM, Rofman, E, Sulem, A (eds.) Optimal Control and Partial Differential Equations - Innovations and Applications, pp. 1-16. IOS Press, Amsterdam (2000)
Świȩch, A: ‘Unbounded’ second order partial differential equations in infinite dimensional Hilbert spaces. Commun. Partial Differ. Equ. 19(11-12), 1999-2036 (1994)
Chang, MH, Pang, T, Pemy, M: Optimal control of stochastic functional differential equations with a bounded memory. Preprint (2006)
Chang, MH, Pang, T, Pemy, M: Stochastic optimal control problems with a bounded memory. In: Zhang, X, Liu, D, Wu, L (eds.) Operations Research and Its Applications. Papers from the Sixth International Symposium, ISORA ’06, 8-12 August, Xinjiang, China, pp. 82-94. World Publishing Corporation, Beijing (2006)
El Karoui, N, Peng, S, Quenez, MC: Backward stochastic differential equations in finance. Math. Finance 7(1), 1-71 (1997)
Peng, S: A generalized dynamic programming principle and Hamilton-Jacobi-Bellman equation. Stoch. Stoch. Rep. 38, 119-134 (1992)
Fuhrman, M, Hu, Y, Tessitore, G: Ergodic BSDEs and optimal ergodic control in Banach spaces. SIAM J. Control Optim. 48(3), 1542-1566 (2009)
Fuhrman, M, Masiero, F, Tessitore, G: Stochastic equations with delay: optimal control via BSDEs and regular solutions of Hamilton-Jacobi-Bellman equations. SIAM J. Control Optim. 48(7), 4624-4651 (2010)
Fuhrman, M, Tessitore, G: Nonlinear Kolmogorov equations in infinite dimensional spaces: the backward stochastic differential equations approach and applications to optimal control. Ann. Probab. 30(3), 1397-1465 (2002)
Fuhrman, M, Tessitore, G: Infinite horizon backward stochastic differential equations and elliptic equations in Hilbert spaces. Ann. Probab. 32, 607-660 (2004)
Masiero, F: Infinite horizon stochastic optimal control problems with degenerate noise and elliptic equations in Hilbert spaces. Appl. Math. Optim. 55, 285-326 (2007)
Zhou, J, Liu, B: Optimal control problem for stochastic evolution equations in Hilbert spaces. Int. J. Control 83(9), 1771-1784 (2010)
Zhou, J, Liu, B: The existence and uniqueness of the solution for nonlinear Kolmogorov equations. J. Differ. Equ. 253, 2873-2915 (2012)
Zhou, J, Zhang, Z: Optimal control problems for stochastic delay evolution equations in Banach spaces. Int. J. Control 84(8), 1295-1309 (2011)
Da Prato, G, Zabczyk, J: Ergodicity for Infinite-Dimensional Systems. Cambridge University Press, Cambridge (1996)
Zabczyk, J: Parabolic equations on Hilbert spaces. In: Stochastic PDEs and Kolmogorov Equations in Infinite Dimensions. Lecture Notes in Math., vol. 1715, pp. 117-213. Springer, Berlin (1999)
Fleming, WH, Soner, HM: Controlled Markov Processes and Viscosity Solutions. Applications of Mathematics, vol. 25. Springer, New York (1993)
Acknowledgements
This work was partially supported by the National Natural Science Foundation of China (Grant No. 11401474), Shaanxi Natural Science Foundation (Grant No. 2014JQ1035), and the Fundamental Research Funds for the Central Universities (Grant No. 2452015087).
Author information
Authors and Affiliations
Corresponding author
Additional information
Competing interests
The author declares to have no competing interests.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Zhou, J. The existence and uniqueness of the solution for nonlinear elliptic equations in Hilbert spaces. J Inequal Appl 2015, 250 (2015). https://doi.org/10.1186/s13660-015-0764-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-015-0764-7