Multivariable Aumann integrals and controlled wave equations (Q1842057)

The paper is concerned with convexity and other properties of joint marginal integrals of multifunctions in several variables. Some applications to control problems for the one-dimensional wave equation with Darboux boundary conditions are given at the end of the paper (bang- bang results, existence of optimal controls without convexity conditions). More precisely, if \(F\) is a measurable multifunction from \(\mathbb{R}^n\) into \(2^{\mathbb{R}^m}\), the set of joint marginal integrals of \(F\) in the directions \(v_1,\dots, v_k\) (\(v_i\) for \(i= 1,\dots, k\) are unit vectors in \(\mathbb{R}^n\)) is defined by \[ \int_{(v_1,\dots, v_k)} Fd\lambda= \Biggl\{\int_{(v_1,\dots, v_k)} fd\lambda\mid f\in L^1,\;f(x)\in F(x)\text{ a.e.}\Biggr\}, \] where \(\int_{(v_1,\dots, v_k)} fd\lambda\) denotes the vector function from \(\mathbb{R}^{n- 1}\) into \(\mathbb{R}^k\) of integrals of \(f\) along lines: \(\int_{v_i} fd\lambda= g_i\) with \(g_i(x_i')= \int^\infty_{-\infty} f(x_i'+ \lambda v_i) d\lambda\), \(x_i'\) being any point in \(v^\perp_i\). It is proved that, if \(F\) is an integrably bounded measurable multifunction with nonempty and closed values, then, for any family \(v_1,\dots, v_k\), the set \(\int_{(v_1,\dots, v_k)} Fd\lambda\) is a closed, bounded convex subset of \([L^1(\mathbb{R}^{n- 1}; \mathbb{R}^m)]^k\) and \[ \int_{(v_1,\dots, v_k)} \overline{\text{co }} Fd\lambda= \int_{(v_1,\dots, v_k)} Fd\lambda= \int_{(v_1,\dots, v_k)}\text{extr } Fd\lambda. \] The existence of solutions for nonconvex control problems for the wave equation is proved thanks to an extension of the Lyapunov convexity theorem. The method is closely related to the one used by the reviewer [Appl. Math. Optimization 26, No. 1, 39-62 (1992; Zbl 0768.49004)].