Continuous differentiability of the value function of semilinear parabolic infinite time horizon optimal control problems on \(L^2(\Omega)\) Under control constraints (Q2128616)

Let \(\Omega \) be an open, connected and bounded subset of \(\mathbb{R}^{d}\) with Lipschitz continuous boundary \(\Gamma \), \(Y=L^{2}(\Omega )\), \( V=H_{0}^{1}(\Omega )\), \(U=L^{2}(0,\infty ;\mathcal{U})\), where \(\mathcal{U}\) is a Hilbert space which will be identified with its dual, \(W(0,T)=\{y\in L^{2}(0,T;V)\); \(\frac{dy}{dt}\in L^{2}(0,T;V^{\ast })\), \(W_{\infty }=W(0,\infty )\), and the set of admissible controls \(U_{ad}\subset \{u\in U:\left\Vert u(t)\right\Vert _{\mathcal{U}}\leq \eta \), for a.e. \(t>0\}\), where \(\eta \) is a positive constant. The main part of the paper is devoted to the analysis of the stabilization problem for an abstract semilinear parabolic equation formulated as an infinite horizon optimal control problem under control constraints \(\mathcal{V}(y_{0})=\min_{(y,u)\in W_{\infty }\times U_{ad}}J(y,u)=\min_{(y,u)\in W_{\infty }\times U_{ad}}\frac{1}{2} \int_{0}^{\infty }\left\Vert y(t)\right\Vert _{Y}^{2}dt+\frac{\alpha }{2} \int_{0}^{\infty }\left\Vert u(t)\right\Vert _{\mathcal{U}}^{2}dt\), subject to the semilinear parabolic equation \(y_{t}=\mathcal{A}y+\mathcal{F}(y)+Bu\) in \(L^{2}(I;V^{\ast })\), \(y(x,0)=y_{0}\) in \(Y\). The authors assume that the operator \(\mathcal{A}\) with domain \(D(\mathcal{A})\subset Y\) and range in \(Y\) generates a strongly continuous analytic semigroup \(e^{\mathcal{A}t}\) on \(Y \) and can be extended to \(\mathcal{A}\in L(V,V^{\ast })\), \(B\in \mathcal{L}(\mathcal{U},Y)\), and there exists a stabilizing feedback operator \(K\in \mathcal{L}(Y,\mathcal{U})\) such that the semigroup \(e^{(\mathcal{A}-BK)t}\) is exponentially stable on \(Y\), the nonlinearity \(\mathcal{F}:W_{\infty }\rightarrow L^{2}(I;V^{\ast })\) is twice continuously Fréchet differentiable, with second Fréchet derivative \(\mathcal{F}^{\prime \prime }\) bounded on bounded subsets of \(W_{\infty }\), \(\mathcal{F}(0)=0\), \( \mathcal{F}:W(0,T)\rightarrow L^{1}(0,T;\mathcal{H}^{\ast })\) is weak-to-weak continuous for every \(T>0\), where \(\mathcal{H}\) is Hilbert space which embeds densely in \(V\), and finally \(\mathcal{F}^{\prime }( \overline{y})\in \mathcal{L}(L^{2}(I;V),L^{2}(I;V^{\ast }))\). The authors prove that associated to each local solution \((\overline{y}(y_{0}),\overline{ u}(y_{0}))\) to this problem, there exists a neighborhood of \(U(y_{0})\) such that the local value function \(\mathcal{V}:U(y_{0})\subset Y\rightarrow \mathbb{R}\) is continuously differentiable, if \(y_{0}\) is sufficiently close to the origin in \(Y\).\ In a second theorem, the authors prove that if \(( \overline{y}(y_{0}),\overline{u}(y_{0}))\) denotes a global solution to this problem for \(y_{0}\in D(\mathcal{A})\) with sufficiently small norm in \(Y\), and if there exists \(T_{y_{0}}>0\) such that \(\mathcal{F}(\overline{y})\in C([0,T_{y_{0}});Y)\), then the following Hamilton-Jacobi-Bellman equation holds at \(y_{0}\): \(\mathcal{V}^{\prime }(y)(\mathcal{A}y+\mathcal{F}(y))+ \frac{1}{2}\left\Vert y\right\Vert _{Y}^{2}+\frac{\alpha }{2}\left\Vert \mathbb{P}_{\mathcal{U}_{ad}}(-\frac{1}{\alpha }B^{\ast }\mathcal{V}^{\prime }(y))\right\Vert _{Y}^{2}+\left\langle B^{\ast }\mathcal{V}^{\prime }(y), \mathbb{P}_{\mathcal{U}_{ad}}(-\frac{1}{\alpha }B^{\ast }\mathcal{V}^{\prime }(y))\right\rangle _{Y}=0\), where \(\mathbb{P}_{\mathcal{U}_{ad}}\) is the projection operator on \(\mathcal{U}_{ad}=\{v\in \mathcal{U}:\left\Vert v\right\Vert _{\mathcal{U}}\leq \eta \}\). The optimal feedback law is given by \(\overline{u}(0)=\mathbb{P}_{\mathcal{U}_{ad}}(-\frac{1}{\alpha }B^{\ast } \mathcal{V}^{\prime }(y))\). For the proofs, the authors mainly use properties of analytic semigroups. They also use properties of general minimization problems.\ Consider the minimization problem \(\min f(x)\), \(e(x,q)=0\), \(x\in C \), where \(C\) is a closed and convex subset of a real Hilbert space \(X\), \( f:X\rightarrow \mathbb{R}^{+}\) is twice continuously differentiable in a neighborhood of the local solution \(x_{0}\) to the preceding problem associated to a nominal reference parameter \(q_{0}\in P\) a normed linear space, \(e:X\times P\rightarrow W\) is continuous and twice continuously differentiable with respect to \(x\), with first and second derivatives which are Lipschitz continuous in a neighborhood of \((x_{0},q_{0})\). The authors introduce the Lagrangian \(\mathcal{L}:X\times P\times W^{\ast }\rightarrow \mathbb{R}\) associated to the above problem and defined as \(\mathcal{L} (x,q,\lambda )=f(x)+\left\langle \lambda ,e(x,q)\right\rangle _{W^{\ast },W}\) and they assume that \(0\in inte^{\prime }(x_{0},q_{0})(C-x_{0})\), where \(int\) denotes the interior in the \(W\) topology. This implies the existence of a Lagrange multiplier \(\lambda _{0}\in W^{\ast }\) such that: \(\left\langle \mathcal{L}(x_{0},q_{0},\lambda _{0}),c-x_{0}\right\rangle _{X^{\ast },X}\geq 0\), \(\forall c\in C\), \(e(x_{0},q_{0})=0\). Finally, defining the operator representation \(A\in L(X,X^{\ast })\) of \(\mathcal{L}^{\prime \prime }(x_{0},q_{0},\lambda _{0})\) defined as \(\left\langle Ax_{1},x_{2}\right\rangle _{X^{\ast },X}=\mathcal{L}^{\prime \prime }(x_{0},q_{0},\lambda _{0})(x_{1},x_{2})\), then the operator \( E=e(x_{0},q_{0})\in \mathcal{L}(X,W)\), the authors assume the existence of \( \kappa >0\) such that: \(\left\langle Ax,x\right\rangle _{X^{\ast },X}\geq \kappa \left\Vert x\right\Vert _{X}^{2}\), \(\forall x\in kerE\). Under further hypotheses, the authors prove a stability result, that is the existence of a neighborhood \(U=U(x_{0},\lambda _{0})\subset X\times W^{\ast }\), of a neighborhood \(N=N(q_{0})\subset P\), and of a constant \(\mu \) such that for all \(q\in N\) there exists a unique \((x(q),\lambda (q))\in U\) satisfying \( \mathcal{L}(x(q),q,\lambda (q))+\partial I_{C}(x(q))\), in \(X^{\ast }\), \( e(x(q),q)\), in \(W\), and \((x(q_{1}),\lambda (q_{1}))-(x(q_{2}),\lambda (q_{2}))_{X\times W^{\ast }}\leq \mu \left\Vert q_{1}-q_{2}\right\Vert _{P}\) , \(\forall q_{1},q_{2}\in N\). For the proof, the authors use the implicit function theorem of Dontchev for generalized equations and an existence result for a minimization problem involving generic operators.