search

QMCPACK / miniqmc

QMCPACK miniapp: a simplified real space QMC code for algorithm development, performance portability testing, and computer science experiments
Other
26 stars 35 forks source link

Allows changing acceptance ratio via -R. #177

Closed ye-luo closed 6 years ago

ye-luo commented 6 years ago

The default value is unchanged 0.5 as before.

prckent commented 6 years ago

What is motivation for adding this option?

Perhaps a wider discussion is needed about what the "fake" QMC in miniqmc is actually doing: VMC w/wo drift, where 50% acceptance is fine, or a fake DMC where the acceptance is 99%+ but we need drift and an "energy" (laplacian) measurement? These are not the same. Since we care about DMC performance more than VMC, are we able to properly capture DMC work?

ye-luo commented 6 years ago

The current default fake run is 50% VMC with drift. I add this option to capture DMC as a 99% VMC with drift. I think both are important.

prckent commented 6 years ago

I would rather our default was fake DMC since this is what we are audited on.

ye-luo commented 6 years ago

I don't want to change the default in this PR. For developing algorithms, assuming 99% is worse than 50%. Optimization may go into extreme cases.

prckent commented 6 years ago

OK.

I prefer something other than -R since there is already -r. Otherwise no objections on this PR.

ye-luo commented 6 years ago

I went through the list of arguments. -a, -r have been used already. So I picked -R.

prckent commented 6 years ago

Case sensitivity will trip people up. Change -r to something else or pick a different input label for the ratio. Why not -rcut for -r (isn't this the pp cutoff? what is rmax?)

ye-luo commented 6 years ago

rmax is for NLPP cutoff. The current argument handling only allows 1 charactor. adjusting rmax is a rarely used option. @prckent how about change it to -x and use -r for acceptance ratio?

prckent commented 6 years ago

Good with me (-r, -x ).