Cartesian atomic cluster expansion for machine learning interatomic potentials

Bingqing Cheng

2024-07-30

Abstract:Machine learning interatomic potentials are revolutionizing large-scale, accurate atomistic modelling in material science and chemistry. Many potentials use atomic cluster expansion or equivariant message passing frameworks. Such frameworks typically use spherical harmonics as angular basis functions, and then use Clebsch-Gordan contraction to maintain rotational symmetry, which may introduce redundancies in representations and computational overhead. We propose an alternative: a Cartesian-coordinates-based atomic density expansion. This approach provides a complete set of polynormially indepedent features of atomic environments while maintaining interaction body orders. Additionally, we integrate low-dimensional embeddings of various chemical elements and inter-atomic message passing. The resulting potential, named Cartesian Atomic Cluster Expansion (CACE), exhibits good accuracy, stability, and generalizability. We validate its performance in diverse systems, including bulk water, small molecules, and 25-element high-entropy alloys.

Computational Physics,Machine Learning,Chemical Physics

What problem does this paper attempt to address?

The paper attempts to address the problem of efficiently and accurately simulating atomic-scale interactions in the fields of materials science and chemistry using machine learning interatomic potentials (MLIPs). Specifically, the authors propose a new method called Cartesian Atomic Cluster Expansion (CACE), which operates directly in the Cartesian coordinate system, avoiding the use of traditional spherical harmonics.

### The main contributions of the paper include:
1. **Atomic environment representation in Cartesian coordinates**: Traditional methods often use spherical harmonics to describe atomic environments, but these methods are complex when dealing with rotational symmetry. CACE performs all operations directly in the Cartesian coordinate system, simplifying the computational process.
2. **Compact invariant feature representation**: CACE can generate a set of compact and polynomially independent invariant features, significantly reducing the number of linearly dependent features compared to previous methods like Atomic Cluster Expansion (ACE) and Moment Tensor Potentials (MTP).
3. **Efficient radial channel coupling**: CACE employs a trainable radial channel coupling mechanism, further improving the model's efficiency and accuracy.
4. **Low-dimensional element embedding**: By introducing low-dimensional element embedding, the model can better capture the relationships between different chemical elements and improve cross-element generalization capabilities.
5. **Message passing mechanism**: CACE integrates two message passing mechanisms, one based on the relative positions between atoms and the other a recursive edge embedding scheme, further enhancing the model's expressive power.

### Experimental validation:
- **Liquid water system**: CACE performs excellently on the liquid water dataset, outperforming or matching existing methods in terms of energy and force prediction errors, especially showing outstanding stability in high-temperature conditions.
- **Small molecules (ethanol and 3BPA)**: For organic small molecules like ethanol and 3-(benzyloxy)pyridin-2-amine (3BPA), CACE's energy prediction is comparable to NequIP, but its force prediction error is slightly higher. However, its stability remains very good.
- **High-entropy alloy system**: For high-entropy alloys composed of 25 metal elements, CACE demonstrates an excellent learning curve and low error, particularly excelling in element extrapolation, being able to predict unseen elements (such as rhenium and osmium) well.

In summary, CACE not only shows good performance across various systems but also has significant advantages in cross-element and temperature range generalization capabilities.