Module: pines
| Description | Usage |
|---|---|
| permutationally invariant atomic-level featurization for neural net based CVs. | |
| Authors: N. Herringer and S. Dasetty |
Direct native dependencies: core, tools, colvar
See module dependencies for the distinction between native dependencies and shortcut requirements.
Details
This module implements PINES (Permutationally Invariant Networks for Enhanced Sampling), a permutationally invariant featurization of the atomic configuration that is discussed in the paper cited below. The PINES action computes a vector of switching functions of selected interatomic distances, sorted within blocks of atom-group pairs in the spirit of the PIV collective variable. The elements of this vector are intended to be used as input features of a neural network that learns the collective variables (for example with PYTORCH_MODEL), in systems such as solutions where molecules of the same species are interchangeable.
Installation
This module is not installed by default. Add --enable-modules=pines to your './configure' command when building PLUMED to enable these features.
Actions
The following actions are part of this module
| Name | Description | Tags |
|---|---|---|
| PINES | Calculate the permutationally invariant PINES features of a configuration. | COLVAR |
References
More information about this module is available in the following articles: