We develop and test numerically a Monte Carlo method for fermions on a lattice which accounts for the effect of the fermionic determinant to arbitrary accuracy. It is tested numerically in a 4-dimensional model with SU(2) color group and scalar fermionic quarks interacting with gluons. Computer time grows linearly with the volume of the lattice and the updating of gluons is not restricted to small jumps. The method is based on random location updating, instead of an ordered sweep, in which quarks are updated, on the average, R times more frequently than gluons. It is proven that the error in R is only of order 1/R instead of 1/R 1 2 as one might naively expect. Quarks are represented by pseudofermionic variables in M pseudoflavors (which requires M times more memory for each physical fermionic degree of freedom) with an error in M of order 1/M. The method is tested by calculating the self-energy of an external quark, a quantity which would be infinite in the absence of dynamical or sea quarks. For the quantities measured, the dependence on R-1 is linear for R ≥ 8, and, within our statistical uncertainty, M = 2 is already asymptotic.
ASJC Scopus subject areas
- Nuclear and High Energy Physics