Quantum Monte Carlo Beyond Fixed-Node/Phase Approximations Using Extended Configuration Spaces and Residuals

Electronic structure quantum Monte Carlo (QMC) that is based on sampling particle configurations has a number of desired properties such as direct access to interparticle correlations and pointwise information about deviations of the local energy from an estimated eigenvalue. The price for these advantages is the fixed-node/phase approximation that is typically used to eliminate the well-known fermion sign/complex value problems. We explore possibilities of recently introduced spinor-based QMC formulation to find improvements over the fixed-node/phase results. One direction is a spinor-based released-node method that enables one to correct part of the fixed-node error by tuning the fixed-phase sampling of configurations in overcomplete continuous spin representation. Another strategy is based on the use of residuals froma variational treatment to improve the trial function as well as to increase efficiency of the estimators. An analysis of results from these calculations enables us to identify the types of fixed-node biases commonly present in trial functions and to estimate corresponding improvements in accuracy and efficiency.