No label defined (Q4501638)

The problem considered in this paper is \[ \text{maximize }\int_0^1u(t,x(t)) dt\qquad \text{subject to } \int_0^1x(t) dt\leq a,\quad x(t)\in F(t,\rho_\varepsilon(t)), \tag{1} \] where \(\varepsilon\in(0,\varepsilon_0]\), under the following main structural assumptions: (a) \(\rho_\varepsilon\) is measurable and maps \([0,1]\) into a given separable complete metric space \(\Gamma\); (b) \(u:[0,1]\times \mathbb R^n_+\to \mathbb R\), where \(\mathbb R^n_+\) is the non-negative orthant of \(\mathbb R^n\), is measurable with respect to \(t\), continuous with respect to \(x\), and with the property that for every \(\delta>0\) there is \(\eta\in L^1[0,1]\) such that \(|u(t,x)|\leq\delta|x|\) if \(|x|\geq\eta(t)\); (c) the values \(\int_0^1u(t,x(t))dt\) are uniformly bounded whenever \(x\) is in a family of functions which are bounded by a fixed integrable function; (d) \(F:[0,1]\times\Gamma\to{\mathcal C}(\mathbb R^n_+)\), where \({\mathcal C}(\mathbb R^n_+)\) is the family of closed subsets of \(\mathbb R^n_+\), endowed with the Kuratowski convergence, is measurable with respect to \(t\), continuous with respect to \(\gamma\), and with the property that for every \(\varepsilon>0\) there is a measurable selection \(y_\varepsilon\) of \(F(\cdot,\rho_\varepsilon(\cdot))\) such that \(\int_0^1y_\varepsilon(t)dt\leq a\), namely problem (1) is feasible, in other words, the integral of the set-valued map \(F(\cdot,\rho_\varepsilon(\cdot))\) forms the feasible set for the optimization problem (1); (e) the family of functions \(\ell_\varepsilon(t)=\min\{|x|: x\in F(t,\rho_\varepsilon(t))\}\) is uniformly integrable; (f) \(\rho_\varepsilon\) converges narrowly to a Young measure \(\mu\) as \(\varepsilon\to 0\). Now consider the variational limit problem \[ \text{maximize }\int_0^1 \int_{\mathbb R^n_+} u(t,x)\xi(t)(dx) dt\qquad\text{subject to } \int_0^1\int_{\mathbb R^n_+} \xi(t)(dx) dt\leq a,\quad \xi(t)\in M(t), \tag{2} \] where \(M(t)\) is the family of probability measures on \(\mathbb R^n_+\) selectionable with respect to \(\mu(t)\) from \(F(t,\mu(t))\), that is \(M(t)={\mathcal S}(F(t,\mu(t))\). A solution of (2) is a Young measure \(\xi\) with values in \({\mathcal P}(\mathbb R^n)\), the space of probability measures on \(\mathbb R^n\). The main result of the paper shows that problem (2) is a variational limit of (1) as \(\varepsilon\to 0\), that is: Theorem. Assume also that there is an open neighborhood of \(a\), contained in the feasible set of \((1)\) for \(\varepsilon>0\) sufficiently small. Then \(a\) is in the feasible set of \((2)\), and the optimal value \(\text{val }(\varepsilon)\) of \((1)\) converges to \(\text{val }(0)\), the optimal value of \((2)\), as \(\varepsilon\to 0\). Furthermore, for every sequence \(\varepsilon_i\to 0\) as \(i\to\infty\) there is a subsequence \(\varepsilon_j\) such that \(x_{\varepsilon_j}\), the corresponding optimal solution of \((1)\), which clearly exists, converges narrowly to an optimal solution \(\xi\) of \((2)\). Moreover, if \(\xi\) is a fixed optimal solution of \((2)\), then there exist selections \(y_\varepsilon\) of \(F(\cdot,\rho_\varepsilon (\cdot))\), with \(\int_0^1y_\varepsilon(t)dt\leq a\), that is feasible for problem \((1)\), which converge narrowly to \(\xi\) as \(\varepsilon\to 0\), and such that \[ \lim_{\varepsilon\to 0} \Biggl|\text{val}(\varepsilon)- \int_0^1u(t,y_\varepsilon(t)) \Biggr|=0. \] These selections can be modified in order to generate near optimal solutions of \((1)\). The paper contains also a long list of comments which show how to extend the main result in different directions, and indicates how the result is placed with respect to previous analogous theorems in the wide literature on the subject. The paper ends with an interesting example for which the variational limit problem can be explictly solved. For the entire collection see [Zbl 0921.00020].