Pruned bases that are compatible with iterative eigensolvers and general potentials: New results for CH3CN

In this paper we demonstrate that when one prunes a product basis for CH3CN using g1(n1)+g2(n2)+⋯+gD(nD)⩽b and appropriate gc(nc) functions it is possible obtain many converged energy levels using a basis that is smaller than the basis obtained with the pruning condition α1n1+α2n2+⋯+αDnD⩽b (Avila and Carrington, 2011) and comparable to the basis built using residual vectors (Garnier et al., 2016). Importantly, the pruning condition g1(n1)+g2(n2)+⋯+gD(nD)⩽b obviates the need to store a Hamiltonian matrix and facilitates the use of iterative eigensolvers because it is compatible with efficient matrix-vector products. Moreover, it may be used in conjunction with a nondirect product quadrature and therefore allows one to use a potential energy surface that is not a simple sum of products.