Accelerated prediction of RPA correlation energies using range-separation and pseudobands

Accurate and performant quantum chemical modeling approaches will be crucial for the prediction of diverse materials and catalyst properties. Many-body, beyond-DFT approaches such as the random phase approximation (RPA) have excellent accuracy across different material classes and an algorithmic complexity intermediate between DFT and coupled cluster.  However, RPA calculations scale as O(N⁴) and are numerically challenging to fully converge the polarizability, requiring large basis sets. Here, we show two approaches that can be used to accelerate RPA total energy predictions and provide converged results for a diverse test suite of materials. First, we implemented the range-separated RPA approach in BerkeleyGW, as it has been shown to greatly accelerate polarizability convergence with controllable accuracy loss. Second, we show that pseudobands can be used to compress higher energy electronic states in the calculation of the polarizability from which RPA energies are computed. We show that together these approaches can achieve good accuracy for RPA correlation energies at significantly reduced cost.