Concavity of eigenvalue sums and the spectral shift function

From MaRDI portal



Abstract: It is well known that the sum of negative (positive) eigenvalues of some finite Hermitian matrix V is concave (convex) with respect to V. Using the theory of the spectral shift function we generalize this property to self-adjoint operators on a separable Hilbert space with an arbitrary spectrum. More precisely, we prove that the spectral shift function integrated with respect to the spectral parameter from −infty to lambda (from lambda to +infty) is concave (convex) with respect to trace class perturbations. The case of relative trace class perturbations is also considered.


Let us denote by \(S\) the set of trace class operators and \(\xi(\lambda; A_0+ V,A_0)\) the spectral shift function for the pair of operators \((A_0+ V,A_0)\). Theorem: Let \(A_0\) and \(V\) be selfadjoint operators on a separable Hilbert space and \(V\in S\). Then for an arbitrary real-valued nonincreasing function \(f\) of bounded total variational \[ g(V)= \int_R f(\lambda)\xi(\lambda; A_0+ V,A_0) d\lambda \] is concave, i.e., for any \(V_1,V_2\in S\) one has \(g(\alpha V_1+ (1-\alpha)V_2)\geq \alpha g(V_1)+ (1-\alpha)g(V_2)\) for all \(\alpha\in [0,1]\).











This page was built for publication: Concavity of eigenvalue sums and the spectral shift function

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1589980)