/
2020-04-06 System Object meeting notes

2020-04-06 System Object meeting notes

Owned by Matt Thompson
Last updated: Apr 14, 2021
2 min read

Date

Apr 6, 2020

Participants

  • @Michael Shirts

  • @Matt Thompson

Goals

  • Discuss plans and scope for system object

Discussion topics

Time

Item

Presenter

Notes

Time

Item

Presenter

Notes

???

System Object

@Matt Thompson

  • What is the









interface with QC world?

  • probably not a whole lot of talking here

  • system object will not need to know the optimizer’s objective function

    • just how to calculate energies, if that’s what we care about

    • other kinda edge cases of “I don’t really want energy, I want x other function” could be handled by other functions

thinking about the main purpose: should it be to be define an energy?

  • define how to calculate energy of a configuration of an atomic system

    • key is that it needs to be able to capture forseeable mathematical representations

      • “forseeable” includes things like ML models

  • the “rubber hits the road” is interfacing with libraries (i.e. an MM system) to tell i.e.

  • knows the mathematics that map coordinates to energy, needs to translate that to engines

optimizer gives parameters to system object, system object tells it how to get energies (which could be done by an MM engine, but might not be responsible for that, this could be done with a “simulation object” that includes the openff system object)

provenance:

  • here is a mapping, maybe metadata about how it was generated but also maybe not in scope

  • may not need to have high standards for data provenance

    • for the scope of reproducibility, needs to know where the data came from (i.e. here is this XML file) but not the entire history of everything

    • (however, history of parameters might be useful scientifically interesting, but not necessary for reproducing calculations in this scope)

to get energy, you need

  • coordinates, ways of calculating the energies

Action items

@Matt Thompson better understanding of provenance (data or mathetmical or otherwise) that people want
@Matt Thompson set up a confluence page to host various system object notes, etc.

Decisions

  • should it be a system object and and interoperability exchange point?

    • crux: can conventional FFs be represented in the system object?

  • can we modularize a system object and interoperability?

    • we think yes, but need to be 100% sure

    • make it possible to write writers without having to do that now

  • do we want