ParmFit Tutorials

Get to know ParmFit’s parameterization workflow by solving a few simple examples.

Before you start: install AmberTools and Gaussian 16

Every case needs AmberTools. Every NCAA and MetalAA case, and Refinement Cases 2 and 3, also need Gaussian 16 for their RESP charges — the Refinement cases also for their QM reference; only Refinement Case 1 runs without it. MAPLE looks for Gaussian only when a run reaches its first Gaussian job, so a missing g16 stops the run partway through.

The Systems

Before You Start

  • Requirements. Software and input needs are summarized in the ParmFit hub’s Requirements section.
  • Model. Every run in these tutorials uses the same checkpoint — UMA-S-1.1, selected with #model=uma(size=uma-s-1p1) (the Refinement launchers add task=omol).
  • Hardware. Every excerpt and timing in these tutorials comes from runs on an RTX 3060 Laptop GPU and an 8-core AMD CPU — sufficient for every run shown here. Every Gaussian job used the defaults qm_nproc=8 and qm_mem=24 (GB); set lower values in the sidecar config on a smaller machine (see General Settings).