2020-06-30 QCFractal Users Meeting notes

Updates from MolSSI

• Nothing this week. Hoping to have new docker images built today. Should have new versions of psi4, qcengine, Will also try to roll in rdkit, openmm, openforcefield.

Manager status

• Queue is currently slowed to a trickle of error cycled jobs. All managers/workers should be off.

• Jobs that continue failing after error cycling will be opened for scientific review. This will be tracked on qca-dataset-submission issue tracker

ESP Grids on final molecules

• No problem. I'll bring up the question of storing ESP points at tomorrow's QCFractal meeting at 8 AM Pacific [1]. The big questions are:

  • Can we afford to store this info in QCA (and for how many mols)?

    • BP – How big is a grid?

    • HJ – Large grids are 250 points per heavy atom. (8 bytes each) → 2kB / heavy atom.

    • (Av mol of 20 heavy atoms → 40 kB)

  • Is there a standard grid-spacing scheme that we can use, so that we don't end up with multiple grids per mol?

    • HJ – FCC grid seems like a clear choice.

    • SB – AM1-BCC originally used FCC, so it makes sense for us to use it as well.

    • HJ – I made a grid.dat file that I fed to psi4 (created by respyte package).

    • SB – We could make a standalone implementation of grid gen pretty easily. Just a few knobs – Inner and outer radius, etc.

    • JW – Would separate final submission be burdensome for queue management?

    • BP – I had thought about making a general framework for these two-step processes, but haven’t yet. This would require a different dataset definition (like OptimizedPropertyDatast)

    • TG – Would ESP calcs be a new “driver”/”service”/”procedure”? Like analogous to hessian driver.

    • BP – Would need to think about which of the above ESP grid gen would be. Probably a subcategory of the generic “property” driver.

  • Does QCEngine/psi4 already support this?

    • (General) We’ll need to implement grid generation on the manager-side, and just have the job spec provide the options for it.

    • TG – Psi4 has some grid generation functionality, but it’s not clear if it’s suitable for our needs. http://www.psicode.org/psi4manual/1.1/cubeprop.html#

    • (General) – We’re not sure if this covers our needs

    • HJ will research this method and deermine whether it’s appropriate for our needs. If not, HJ will bring a recommendation for how to generate grids to the next QCF meeting (use numpy? Package RESPYTE? other?)

  • What would be the work required/timeline for implementation in QCArchive?

    • DD – If this is like a torsiondrive, it would live on the server. Otherwise it could live on the managers.

    • BP – Probably wouldn’t be a service, but it would be a procedure. We should ask Lori Burns about how this would extend the schema.

    • (General) This would be, at the fastest, on the order of one QCF + QCNG release cycle

    • BP – We had aimed for this to be roughly monthly, but it’s slowed since DGS left. We have a big release scheduled at the end of July. QCNG may be at a different pace than that.

    • (General) – Likely roll-out timeline for this is 1-3 months.

User questions

Other

• Send nice quotes to BP for grant review visit. Possibly also cool pictures of what QCA enabled.