OpenMM-ML 1.8

A major focus of this release is on ML/MM simulations. New features related to them include:

• ML/MM simulations are often much faster than before.
• The link-atom method is now supported for simulations where a molecule spans the ML-MM boundary.
• It introduces a new architecture for supporting alternative embeddings. Currently, mechanical embedding is still the only available embedding. This is in preparation for adding electrostatic and polarizable embeddings in future releases.

This release also adds built in support for more models: a variety of pretrained AIMNet models, custom AIMNet models provided by the user, MACE-POLAR-1, and MACELES-OFF.

What's Changed

• set gpu_preprocessing=True in FeNNix by @mitkotak in #153
• Support for generic and potential-specific customizable embeddings by @epretti in #152
• Add code of conduct and contributor guide by @epretti in #154
• Feature supply aimnet2 custom model by @fjclark in #156
• Add More AIMNet Models by @fjclark in #157
• Refactor embedding and interpolation code by @epretti in #158
• [bug] MACE model dtype precision is applied to model as well as inputs. by @patrickwrowe in #162
• Add MACELES-OFF pretrained model by @epretti in #163
• Add CI on self-hosted runners by @epretti in #165
• Add MACE-POLAR-1 pretrained models by @epretti in #164
• Added UMA example by @peastman in #166
• Add link-atom method for molecules crossing ML-MM boundary by @epretti in #161
• Use setParticles() to restrict atoms by @peastman in #167
• Update version number to 1.8 by @peastman in #168

New Contributors

• @mitkotak made their first contribution in #153
• @fjclark made their first contribution in #156
• @patrickwrowe made their first contribution in #162

Full Changelog: 1.7...1.8