Abstract

Anthony Rasmussen, Stephen Jeffrey and Sean Smith

A common way to infer dynamical information for a reacting chemical system is to time-propagate an initial wavepacket, then Fourier filter the evolved wavefunction. The crucial step of acting with the quantum mechanical time-propagation operator involves exponentiating a large non-Hermitian matrix. In recent years, Newton polynomial expansions of the matrix exponential have proven numerically stable but expensive. To reduce this expense, we have recently devised a new scheme that performs the time-propagation and chemical analysis entirely within a Krylov subspace. For some benchmark systems, our method seems relatively more efficient in terms of execution time and storage economy.