S. Sorella
Collaborators: M. Calandra F. Becca L. Capriotti and A. Trumper
GFMCSR: an effective remedy for the sign problem disease.
The ''Stochastic Reconfiguration'' scheme for the stabilization of the
sign problem in the Green Function Monte Carlo technique (GFMC) is
discussed. This method stabilizes the average sign of the walkers
during a GFMC simulation by introducing a systematic ''bias'' that can
be reduced and eventually vanishes in a well defined theoretical
limit. This is achieved when all the independent correlation
functions characterizing the Hilbert space of a strongly correlated
system are set to be statistically equal before and after the
``Stochastic Reconfiguration''. This exact limit is prohibitive for a
physically large Hilbert space, However a considerable reduction of
the mentioned ''bias'' can be easily achieved, by considering only few
correlation functions in the ''Stochastic Reconfiguration''. The
method is applied and tested for three lattice hamiltonians, which
still represents a challenge in computational physics due to the
''sign problem'':
- the frustrated hard core bosons,
- the antiferromagnetic Heisenberg model with frustration,
- the t-J model: i.e. the strongly correlated electrons in a lattice.
It will be shown that for all three models the accuracy of the ground
state energy per site and of some correlation functions is physically
acceptable even for large number of lattice sites and electrons. This
property allows us to draw some interesting physical conclusions about
the ground state properties of the considered models.
NATO ASI, Cornell Theory Center,
Last modified: Thu Jul 2 10:26:13 EDT 1998