Toward Accurate Hydrogen Bonds by Scalable Quantum Monte Carlo

Single-determinant (SD) fixed-node diffusion Monte Carlo (FNDMC) gains popularity as a benchmark method scalable to large noncovalent systems, although its accuracy limits are not yet fully mapped out. We report on an interesting example of significant SD FNDMC accuracy variations in middle-sized hydrogen-bonded dimer complexes, formic acid (FA) vs methanediol (MD), distinct by the maximum bond order (2 vs 1). While the traditional SD FNDMC schemes based on bias cancellation are capable of achieving benchmark (2%) accuracy for MD, this has not been the case for FA. We identify the leading systematic error source in energy differences and show that suitably designed Jastrow factors enable SD FNDMC to reach the reference accuracy for FA. This work clearly illustrates the varying accuracy of the present-day SD FNDMC at the 0.1 kcal/mol scale for a particular set of systems but also points out promising routes toward alleviation of these shortcomings, still within the single-reference framework.