We propose a stochastic method to reduce the autocorrelation time of a
general Monte Carlo (MC) method and apply it to the variational quant
um Monte Carlo (VMC) simulation of full-core atoms. We achieve a reduc
tion in autocorrelation time of at least a factor of four compared wit
h the standard method. Further, we find an approximate analytic fit to
our results which gives a comparable reduction in autocorrelation tim
e at essentially no cost. Our analytic form is independent of the geom
etry of the system being modeled and, therefore, can be easily applied
to the VMC simulation of solids; it may also prove useful in any MC s
imulation where there are widely varying length scales. Results are pr
esented for C, F, and Si. (C) 1998 American Institute of Physics.