Mosco approximation of integrands and integral functionals (Q2565017)

The authors discuss the approximation of minimization problems \(\min_{x\in X}f(t,x)\) where \(f:T\times X\to R^+_0\cup \{+\infty\}\) and the parameter \(t\) is choosen at random in \(T\); \((T,\mathcal A,\mu)\) is a probability space, \(X\) is a Banach space. This formulation contains both stochastic approximation and approximations of functionals of the calculus of variations \[ \int_Tf(t,u(t)) dt,\qquad u\in L^1(T,X),\tag{1} \] where \(f\) is a normal convex integrand. The main idea is to find a suitable sequence \((f_n)_{n\in N}\) of ``simpler'' functions in such a way that \[ \inf_{x\in X}f_n(t,X)\;\text{``converges'' to} \inf_{x\in X} f(t,x) \] \[ \text{Argmin \(f_n\) ``converges'' to Argmin } f \] The topology they adopt is the natural one in this kind of problems, i.e., Mosco convergence [\textit{U. Mosco}, Adv. Math. 3, 510-585 (1969; Zbl 0192.49101)]. As it is well known, Mosco topology is an extension of Kuratowski topology, strictly connected to epi-convergence or \(\Gamma\)-convergence. The approximating functions they take into consideration are of the type \[ g(t,x)=\sum_{k=1}^p1_{A_k}(t) g_k(t) \leqno(2) \] where \(\{A_k:\;k=1,\dots,p\} \) is an \(\mathcal A\)-measurable partition of \(T,\) each \(g_k\) is Lipschitzian and \(1_{A_k}\) denotes the probabilistic indicator of \(A_k.\) The main result deals with integral functionals of type (1). Precisely, they prove the existence of a sequence \((f_n)_{n\in N}\) of normal, proper, convex integrands of type (2) such that \[ f_n\longrightarrow f,\quad f_n^*\longrightarrow f^*,\quad I(f_n)\longrightarrow I(f),\quad I(f^*_n)\longrightarrow I(f^*) \] with respect to Mosco convergence, relative to the natural topologies of the respective spaces, where \(*\) denotes the Fenchel conjugate. This result is strictly connected with a convergence theorem of \textit{A. Salvadori} [Sémin. Anal. Convexe, Univ. Sc. Tech. Languedoc 15, Exp. No. 5, 25 p. (1995; Zbl 0615.49004)]. Among the auxiliary results they prove, we mention the approximation of a multivalued function \(F:T\to \mathcal F(X),\) with closed values, by means of a sequence of simple measurable multifunctions, with respect to Wijsman topology. Finally, we mention that the results of the present paper are new even in the particular case \(x\) is a finite dimensional space.