On the solution existence of nonconvex quadratic programming problems in Hilbert spaces (Q1696189)

The aim of this paper is to provide existence results for the following problem (QP): \[ \begin{cases} \min f(x):=\frac{1}{2}\langle x,Tx\rangle +\langle c,x\rangle\\ \mathrm{s.t.}\;x\in {F}, {\mathrm{where}}\; F:=\{x\in {\mathcal{H}}:\; g_i(x):=\frac{1}{2}\langle x,T_i x\rangle +\langle c_i,x\rangle+\alpha_i\leq 0,\; i=1,,2,\dots, m\}, \end{cases} \] where \(\mathcal{H}\) is a Hilbert space, \(T:\mathcal{H}\to \mathcal{H}\) is a continuous self-adjoint operator, \(T_i\) is a positive semidefinite continuous linear self-adjoint operator on \(\mathcal{H},\) \(c,c_i\in \mathcal{H},\) and \(\alpha, \alpha_i\) are real numbers, \(i=1,2,\dots, m.\) Despite the feasible set is always convex, the problem (QP) is not, in general, a convex problem. The main results of the paper are the content of Theorem 2 and Theorem 5; both provide a sufficient condition for the existence of an optimal solution on a nonempty set \(F\). The assumptions of Theorem 2 are the following: \(f\) is bounded from below over \(F,\) \((v\in 0^+F,\langle v,Tv\rangle=0)\Rightarrow \langle c_i,v\rangle =0,\forall i: T_i\neq 0,\) and \(\langle x,Tx\rangle\) is a Legendre form (i.e., \(Q\) is weakly lower semicontinuous, and, if \(x_k\) weakly converges to \(x_0\) and \(Q(x_k)\) converges to \(Q(x_0)\), then \(x_k\) strongly converges to \(x_0\)). Here, \(0^+F\) denotes the recession cone of \(F.\) The second assumption can be dropped in case \(m=1.\) In Theorem 5, the assumption that \(\langle x,Tx\rangle\) is a Legendre form is replaced by the request that \(T, T_i\) (\(i=1,2,\dots, m\)) are compact with closed range.