We propose improved versions of the standard diffusion Monte Carlo (DMC)
and the lattice regularized diffusion Monte Carlo (LRDMC) algorithms.
For the DMC method, we refine a scheme recently devised to treat
nonlocal pseudopotential in a variational way. We show that such
scheme-when applied to large enough systems-maintains its effectiveness
only at correspondingly small enough time-steps, and we present two
simple upgrades of the method which guarantee the variational property
in a size-consistent manner. For the LRDMC method, which is
size-consistent and variational by construction, we enhance the
computational efficiency by introducing: (i) an improved definition of
the effective lattice Hamiltonian which remains size-consistent and
entails a small lattice-space error with a known leading term and (ii) a
new randomization method for the positions of the lattice knots which
requires a single lattice-space.
