In the case of systems where the intention is only to perform VMC
calculations, the above optimisation scheme has proved to be
essential. When using the new *u* function from section ,
there is no approximate way to obtain the values of the variational
parameters in the way that the original Fahy prescription allowed functions to be approximately calculated. Therefore some form of
variance minimisation is required to produce the wavefunction. To
achieve state of the art accuracy in these VMC calculations it is
clear that significant effort is required to ensure the quality of the
trial wavefunction before performing the calculation. As the accuracy
of the final VMC calculation is completely controlled by the accuracy
of trial wavefunction, it is necessary to perform the optimisation
procedure over an ensemble of configurations of equivalent size to the
number of moves in the final VMC calculation. If several
re-generations of configurations are required, this leaves the
wavefunction optimisation as the most computationally expensive part
of performing a VMC calculation.

Tue Nov 19 17:11:34 GMT 1996