Coupled-cluster method for nucleonic matter using GPU tensor cores

We show that GPU tensor cores can be a beneficial computational tool for solving coupled-cluster equations with double excitations in momentum space used to describe neutron matter. The momentum conservation greatly reduces the number of non-zero matrix elements between two-neutron states of the system’s potential energy forming blocks along the main diagonal. If the plane wave basis and the number of neutrons are large enough to produces blocks of sizes ≥ 103÷4 then VOLTA tensor cores outperform the respective CPU implementation by an order of magnitude.