search

QMCPACK / qmcpack

Main repository for QMCPACK, an open-source production level many-body ab initio Quantum Monte Carlo code for computing the electronic structure of atoms, molecules, and solids with full performance portable GPU support
http://www.qmcpack.org
Other
297 stars 138 forks source link

Integration test: 3-body Jastrow #573

Closed jtkrogel closed 6 years ago

ye-luo commented 6 years ago

For the moment, we have the H2O example testing JeeI. But I prefer to have an optimization test with J3. @jtkrogel Could you transfer the ownership of this issue to me?

jtkrogel commented 6 years ago

Is PBC not distinct? I guess I would expect differences in distance calculations to be picked up elsewhere. Transferring ownership.

ye-luo commented 6 years ago

Correct. BC is handled within distance tables.

prckent commented 6 years ago

I think we need a J3 test in PBC splines. Something like a single primitive cell of NiO for consistency with the performance tests, and a system where the J3 contribution is non-negligible. This will be essential to test any GPU J3 implementation unless the yet to be found GPU J3 creator also writes a Gaussian basis implementation (CPU bypass or GPU) based on @anbenali 's yet to be merged CPU-side rewrite :-)

ye-luo commented 6 years ago

nightly tests shows all the CPUs are fine and GPUs fail as inteded. I gave up NiO mainly because of the not small h5 size even for a small cell. On monoO, the energy gain from J3 is distinguishable. @prckent can we close this ticket?

prckent commented 6 years ago

To test J3 sensibly we need a system with non-negligible J3 contribution. The test for O is a start but would not let us find e.g. a numerical problem due to optimizations. For NiO we need the smaller cell for the performance tests, as discussed. Then the J3 test can run if the wavefunction is available.