The configuration interaction method, which is well-known as the shell-model calculation in the nuclear physics community, plays a key role in elucidating various properties of nuclei. In general, these studies require a huge number of shell-model calculations to be repeated for parameter calibration and quantifying uncertainties. To reduce these computational costs, we propose a new workflow of shell-model calculations using a method called eigenvector continuation (EC). It enables us to efficiently approximate the eigenpairs under a given Hamiltonian by previously sampled eigenvectors. We demonstrate the validity of EC as an emulator of the shell-model calculations for a valence space, where the dimension of parameters is relatively large compared to the previous studies using EC. We also discuss its possible applications to the quantification of theoretical uncertainty, using an example of Markov chain Monte Carlo sampling for a simplified problem. Furthermore, we propose a new usage of EC: preprocessing, in which we start the Lanczos iterations from the approximate eigenvectors, and demonstrate that this can accelerate the shell-model calculations and the subsequent research cycles. With the aid of the eigenvector continuation, the eigenvectors obtained during the parameter optimization are not necessarily to be discarded, even if their eigenvalues are far from the experimental data. Those eigenvectors can become accumulated knowledge. In order to enable efficient sampling of shell-model results and to demonstrate the usefulness of the methodology described above, we developed a new shell-model code, ShellModel.jl. This code is written in Julia language and therefore flexible to add extensions for the users purposes.
Download