This document details the execution procedure for adding Sage RC2 (openff-2.0.0-rc.2) optimizations to datasets produced with the Optimization Benchmarking Protocol - Season 1 .

All steps assume the deployed conda environment is active, e.g. with:

4. Optimization execution

Using your chosen approach for executing optimizations in Optimization Benchmarking Protocol - Season 1 | 4. Optimization execution , you will run MM optimizations using the 1:4 season specification.

Multi-node approaches

A. Multi-node, persistent server, long-lived manager

To submit and execute the MM stage using the QM stage data as input:

You can export the data from this dataset with:

B. Multi-node, persistent server, short-lived managers

Same approach as (A) above.

C. Multi-node, no server/manager setup

If your cluster uses e.g. SLURM, create a submission script like the following, called submit_molecules.sh:

For the MM stage, using 1 thread and 6GiB for each:

You could batch multiple files into each submission with e.g. xargs:

D. Single-node, persistent server

Same approach as (A) above.

E. Single-node, no server/manager setup

To execute the MM stage using the QM stage data as input:

Because thread-parallelism may not scale as well as simply performing many optimizations at once with your machine, consider running multiple execute commands at once, giving each a batch of files to work on. You can do this in one shot with xargs. For example, for the MM step, we might choose 1 thread per command, and 8 commands: