Getting Started

Introduction

AniSOAP is a package for creating machine-learnable representations of systems of particles in a way that preserves geometrical information. AniSOAP works similarly to SOAP (Smooth Overlap of Atomic Potentials) in many ways, with the chief difference being AniSOAP’s ability to represent ellipsoidal particles.

First Steps

AniSOAP depends on and utilizes the packages featomic (formerly known as rascaline), metatensor, and ase. It may be helpful to familiarize yourself with these packages before diving into AniSOAP.

Key Concepts

In the AniSOAP representation, particles are treated as multivariate Gaussians (MVGs). For each particle, we can consider its n-body interactions with nearby particles. Each local density field is then decomposed in terms of spherical harmonics and some choice of radial basis functions in order to be compactly represented in vector form.

A single AniSOAP vector represents the environment of a single central ellipsoid. A system of such particles can then be represented, for example, as a matrix of AniSOAP vectors.