QMCPACK
Enery density app level testing
Proposed changes
Add application level testing for batched vmc and dmc energy.
This brought me up hard against some issues with check_stats.py and the summing tests of the dmc.dat over steps in a block versus the energy density over particles in blocks. I had to figure out just what could be expected here accuracy wise. On the side of the dmc.dat it's less than you would think. I'd still say its not completely settled but I think a reasonable tolerance for these sums is order of the error between the two different ways to get the energy avg of a block from the dmc.dat.
I still think that the energy density sums should match closer to the dmc.dat single average route but they don't by any better a tolerance than the two different ways to get the avg block energy from dmc.dat does. I'm not sure how seriously this should really be taken?
The new estimators follow the rule that we accumulate weighted measurements and we average only once before data is committed. Obviously the output of dmc.dat does not take this route and I added a test in check_stats.py to look at the error this causes, it can be surprising large and so I needed to drop the tolerance of the relative error check for the comparisons to the dmc blocks. I added a new element to the dmc.dat so that the error introduced for block sums can be considered.
this paper PHYSICAL REVIEW E 83, 066706 (2011) got me thinking about how we should probably think about the sizes of blocks when collecting observables with DMC a bit more. I haven't run any tests to see how correlated our samples are in the EnergyDensityTests but I get the impression that is an issue.
What type(s) of changes does this code introduce?
Delete the items that do not apply
- New feature
- Testing changes (e.g. new unit/integration/performance tests)
- Documentation or build script changes
- Other (please describe):
Does this introduce a breaking change?
- No
What systems has this change been tested on?
sdgx server
Checklist
Update the following with an [x] where the items apply. If you're unsure about any of them, don't hesitate to ask. This is simply a reminder of what we are going to look for before merging your code.
-
- [ ] I have read the pull request guidance and develop docs
-
- [ ] This PR is up to date with the current state of 'develop'
-
- [ ] Code added or changed in the PR has been clang-formatted
-
- [ ] This PR adds tests to cover any new code, or to catch a bug that is being fixed
-
- [ ] Documentation has been added (if appropriate)
The new estimators follow the rule that we accumulate weighted measurements and we average only once before data is committed. Obviously the output of dmc.dat does not take this route and I added a test in check_stats.py to look at the error this causes, it can be surprising large.
- "weighted measurements" did you mean weight multiplied quantities?
- "it can be surprising large", I'm wondering if there is a bug. @jtkrogel Does "check_stats.py" do weighted average or not?
In the past, I was able to get the energy density sum to match (I believe the dmc.dat data) to high precision (given the number of decimal places written to dmc.dat, that is).
If high agreement is there, then the weighting has been done in an identical fashion. Note, this really should be the same between a scalar estimator (the total energy) and a "collectable" one (the energy density) at each step.
Sum rules should be easy to enforce...
If you want to check more closely, just run with 1 step per block. Then each energy entry in scalar.dat should match each entry in dmc.dat. The volume integral of each energy density entry in stat.h5 should match each of the scalar.dat and dmc.dat values.
Also, the definition of the weighting should be unambiguous. See e.g. equation 3.53 in Foulkes 2001 and equation 24.21 in David's book "Interacting Electrons".