Surrogate-based LJ optimization

Driver	Supervisor	Contributors	Stakeholder

Driver	Supervisor	Contributors	Stakeholder
@Owen Madin	@Michael Shirts

Objective	Demonstrate the feasibility of surrogate-based optimization for LJ parameters
Due date	03/01/2022
Status	IN PROGESS

Problem Statement

We would like to perform better LJ optimizations by building multi-surrogate models that approximate the response of physical property simulated from a force field, as a function of force field parameters

Milestones

Milestone	Status

Milestone	Status
Finish preliminary testing for surrogate model building	IN PROGRESS
Finish project outline/plan	IN PROGRESS
Build naive surrogate and perform optimizations for “pure only” training set (experiments 1/2 in the outline)	NOT STARTED
Benchmark optimized parameters for “pure only” training set (experiment 2)	NOT STARTED
Build iteratively refined surrogate (experiment 3) and perform optimization	NOT STARTED
Build surrogate with Bayesian optimization and perform parameter optimization	NOT STARTED
Build surrogate model for “mixture only” training set (experiment 3) and perform parameter optimization	NOT STARTED
Benchmark optimized parameters for “mixture only” training set	NOT STARTED
Write Draft Manuscript	NOT STARTED
Final Manuscript	NOT STARTED

Actions

Notes

References

https://docs.google.com/document/d/1fCVAEqKcZFzZID-XyOh06wF1NMNIUQLneK-dKwmb4eA/edit

Surrogate Optimization collection on OpenFF Zotero.