Enhancing robustness and efficiency of density matrix embedding theory via semidefinite programming and local correlation potential fitting

Density matrix embedding theory (DMET) is a powerful method for treating strongly correlated systems. In order to apply DMET to large-scale systems, the correlation potential fitting procedure often becomes a crucial computational bottleneck. This paper proposes the L-DMET method to simultaneously improve the robustness and efficiency of correlation potential fitting, via a semidefinite programming reformulation and a local correlation potential fitting procedure. The performance of L-DMET is demonstrated using the 2D Hubbard model and the hydrogen chain.