My PhD thesis - Condensed Matter Theory - Imperial College London
My PhD thesis - Condensed Matter Theory - Imperial College London
My PhD thesis - Condensed Matter Theory - Imperial College London
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
CHAPTER 4.<br />
ERRORS IN QMC SIMULATIONS<br />
The Ewald interaction v E is the correct interaction to use for the Hartree energy, 6<br />
but not for the exchange-correlation energy. The reason for this is that the entire<br />
exchange-correlation hole (a total charge deficit of one electron) is contained within<br />
the simulation cell; there should be no images of the exchange-correlation hole.<br />
The Ewald interaction, which includes the effects of image charges, is therefore<br />
inappropriate for the calculation of U XC .<br />
4.2 Fixed-node errors<br />
After many time steps, the walkers in a conventional DMC simulation are distributed<br />
according to the fixed-node density Ψ T Ψ FN<br />
0 rather than the desired density Ψ T Ψ 0 .<br />
Any fixed-node estimate of the ground-state energy must be variational: E FN<br />
0 ≥<br />
E 0 . The equality holds only when the nodes are exact; while this may be achievable<br />
for a one-electron system, it is almost impossible in many-electron calculations.<br />
An indication of the difficulty in correctly guessing the nodal surface is the high<br />
dimensionality of that surface: (Nd − 1), for a system of N electrons moving in d<br />
dimensions. It is not possible to deduce the nodal surface from the condition that<br />
the wave function be zero when two electrons coincide: this defines a surface of only<br />
(N − 1)d dimensions.<br />
The fixed-node approximation is uncontrolled; the size of the error it introduces<br />
cannot be calculated analytically. It is not surprising that a great deal of time and<br />
effort has been devoted to overcoming this problem, with limited success.<br />
The release-node algorithm of Ceperley and Alder [12] uses separate populations<br />
of positive and negative walkers which are allowed to cross the nodes of the trial<br />
wave function. The problem with this method is that both walker populations grow<br />
geometrically in time, leading to exponentially-increasing statistical fluctuations;<br />
the method becomes a race to obtain convergence to the ground state before the<br />
6 Although the Hartree energy defined by equation (4.24) depends on the value of ξ, the total<br />
energy does not, because of the corresponding terms in the ion-ion energy. The exchange-correlation<br />
energy defined by the same equation does not depend on ξ.<br />
65