publications list


Computation of extremal eigenvalues of large Hermitian and Hamiltonian matrices in quantum and classical physics

E.V. Tsiper

School of Computational Sciences, George Mason University, Fairfax, VA 22030-44\ 44
Center for Computational Materials Science, Naval Research Laboratory, Washingt\ on, DC 20375
Large Hermitian eigenvalue problems appear frequently in electronic structure theory. The underlying quantum many-body problem is often referred to as "incomputable" since it involves Hilbert spaces whose dimensionality grows exponentially with the number of interacting particles. This complexity is, in fact, the source of the idea of quantum computation, since a moderately large quantum many-body system possesses an amount of information greater than any conventional computer can handle. Yet, exact diagonalization of large matrices using conventional computers is an invaluable tool that helps us understand many physical phenomena.

In this presentation I will review the Lanczos recursion and related Krylov-subspace methods that allow us to compute extremal eigenvalues of ultra-large Hermitian matrices. I will also discuss a complementary classical problem that leads to Hamiltonian (non-Hermitian) eigenvalue equation, and an intricate relation that exists between its eigenvalues and the eigenvalue differences, or excitation energies, of the quantum system. I will present a minimum principle and TDHF-Lanczos recursion that generalize the Rayleigh minimum principle and the Hermitian Lanczos recursion to the Hamiltonian case. The presentation will be accompanied by several examples from solid state physics and quantum chemistry.

Colloquium, George Mason University, October 27, 2005
presentation: PDF