Multideterminant diffusion Monte Carlo applied to Molecules and Solids

In the past decade, fixed-node Diffusion Monte Carlo using a single Slater-Jastrow determinant as a trial wavefunction has proven to reproduce systematically (within 50meV) the energies of a wide range of molecules and solids. While it has been demonstrated for molecular systems that the path to chemical accuracy (<1Kcal/mol) relys on improving the accuracy of the nodal surface, which can be achieved using a multideterminants trial wavefunction from MCSCF or Selected CI, no such calculations have been attempted on a solid. In this talk, we will discuss the effect of using a multi-determinant selected CI trial wavefunctions, generated with Configuration Interaction using a Perturbative Selection made Iteratively (CIPSI),in diffusion Monte Carlo (DMC) on the energies of molecules and solids. This first study will allow us to analyze the feasibility, advantages and size consistency issues when using large multideterminants expansion in QMC at the era of exascale computers.