Optimality conditions and duality theory for minimizing sums of the largest eigenvalues of symmetric matrices (Q1321653)

Let \(A\) be an \(n\times n\) real symmetric matrix and let \(k\in \{1,\dots,n\}\). The paper concerns the functions \(f_ k(A)= \sum^ k_{i=1} \lambda_ i\) and \(g_ k(A)= \sum^ k_{i=1} |\mu|_ i\), where \(\lambda_ 1,\dots,\lambda_ n\) denote the eigenvalues of \(A\) ordered in the form \(\lambda_ 1\geq\cdots\geq \lambda_ n\) and \(\mu_ 1,\dots,\mu_ n\) denote the eigenvalues of \(A\) ordered by \(|\mu_ i|\geq\cdots\geq |\mu_ n|\). The sums \(f_ k(A)\) and \(g_ k(A)\) are characterized as support functions in terms of the Frobenius inner product \(\langle A,B\rangle= \text{tr}(AB^ T)\). Subdifferential properties are considered for \(f_ k(A)\) and \(g_ k(A)\) as well as for the composite functions \(f_ k(A(x))\) and \(g_ k(A(x))\) where \(A(x)\) smoothly depends on the \(m\)-real-vector \(x\). Finally the minimization problem of the composite functions [cf. \textit{J. Cullum}, \textit{W. E. Donath} and \textit{P. Wolfe}, Nondiffer. Optim., Math. Program. Study 3, 35-55 (1975; Zbl 0355.90054)] is discussed. The subdifferential results provide necessary conditions for \(x\) to be a local minimizer; in case \(x\) does not satisfy the necessary conditions, a descent direction is generated.