next up previous contents
Next: Performing DMC Calculations on Up: Diffusion Quantum Monte Carlo Previous: The Fixed-Node Approximation

DMC With Non-Local Pseudopotentials

The evaluation of non-local pseudopotentials within the QMC solid codes is dealt with in detail in chapter gif. Optimising wavefunctions with respect to these non-local pseudopotentials is discussed in chapter gif. In chapters gif and gif, DMC calculations involving a non-local pseudopotential to describe silicon are reported. The use of such non-local pseudopotentials within DMC is a problem that is yet to be completely resolved. Here we discuss the approximations used later in the thesis to try and deal with this problem.

Consider Eq.(gif), in which a pseudopotential is used to describe the valence electrons in a system;

equation1540

We can divide the full valence Hamiltonian, tex2html_wrap_inline6529 , into a local and non-local part

equation1547

where tex2html_wrap_inline6531 includes the kinetic energy, the local part of the pseudopotential and the Coulomb interaction, and W includes the non-local part of the pseudopotential. The diffusion-drift equation for f, Eq.(gif), can be split into two terms representing the local and non-local parts of the Hamiltonian

  equation1556

The first three terms on the left hand side of Eq.(gif) can be interpreted as a local diffusion, drifting and branching process. However, the fourth term represents the operation of the non-local pseudopotential on the unknown wavefunction, tex2html_wrap_inline5833 , producing non-local branching. In the DMC calculations described in chapters gif and  gif, we have used the ``locality approximation'' which was introduced by Christiansen[38, 39, 40] and recently applied by Mitas[41, 42] to the problem of the unknown wavefunction in Eq.(gif). In this approximation, the non-local pseudopotential acts not on the unknown wavefunction, tex2html_wrap_inline5833 , but on the guiding wavefunction, tex2html_wrap_inline6541 ,

equation1576

One is then free to impose the fixed-node approximation in the same way as before. In fact, when using the above local model potential, it is important that the fixed node approximation is applied as this steers the random walk away from the nodes of tex2html_wrap_inline5833 , where there will generally be divergences in the local model potential. Without the fixed node approximation, these divergences would cause large fluctuations in the population of diffusing particles.

It should be noted that within the ``locality approximation'', it is no longer generally true that the DMC estimate of the energy is an upper bound to the true groundstate energy. It must always be less than the VMC energy and it has been shown, [42] that the DMC energy converges quadratically to the exact groundstate energy as the guiding function tex2html_wrap_inline6541 approaches the true groundstate wavefunction.


next up previous contents
Next: Performing DMC Calculations on Up: Diffusion Quantum Monte Carlo Previous: The Fixed-Node Approximation

Andrew Williamson
Tue Nov 19 17:11:34 GMT 1996