Selection of optimization series

[Jiani]

Hello developers,

    I recently tried to use QMCPACK for some periodic systems containing about 100 electrons, and encountered some small problems when optimizing the wavefunction. 

    I considered 3-J items when optimizing, but I noticed that when selecting different optimization series, the DMC results under the same timestep will be slightly different. According to my understanding, after a certain imaginary time, their results should converge. 

    How should I choose the most appropriate optimization series? Can this difference be eliminated by DMC timestep extrapolation?

Best Regards

Jiani

[Paul R. C. Kent]

Hi Jiani

I suggest you do some experiments on a small system to explore the following: In the absence of pseudopotentials, indeed fixed node DMC will give the same result when time step extrapolated, regardless of Jastrow. Only the efficiency will vary. In the case of pseudopotential use, an error is incurred in the non-local pseudopotential contribution to the total energy which depends on the accuracy of the trial wavefunction and the pseudopotential evaluation scheme (locality, one of the t-moves variants). In practice this means you will get slightly different answers with slightly different Jastrow factors, e.g. by doing independent Jastrow optimizations with different seeds. However the difference is usually small for "good" trial wavefunction optimizations. If the different is large then check the calculations carefully. Obviously an intentionally bad Jastrow or a poorly chosen Jastrow -- e.g. far too small cutoff -- will give larger differences. This dependence of the DMC result on the quality of trial wavefunction when pseudopotentials are used is a key reason to do a good job optimizing the wavefunctions in VMC before moving to DMC.

-- Paul