Discovering crystal structures with specific chemical properties has become
an increasingly important focus in material science. However, current models
are limited in their ability to generate new crystal lattices, as they only
consider atomic positions or chemical composition. To address this issue, we
propose a probabilistic diffusion model that utilizes a geometrically
equivariant GNN to consider atomic positions and crystal lattices jointly. To
evaluate the effectiveness of our model, we introduce a new generation metric
inspired by Frechet Inception Distance, but based on GNN energy prediction
rather than InceptionV3 used in computer vision. In addition to commonly used
metrics like validity, which assesses the plausibility of a structure, this new
metric offers a more comprehensive evaluation of our model’s capabilities. Our
experiments on existing benchmarks show the significance of our diffusion
model. We also show that our method can effectively learn meaningful
representations.
Discovering crystal structures with specific chemical properties is a key area of focus in material science. The ability to generate new crystal lattices is an important aspect of this research. However, current models have limitations as they only take into account atomic positions or chemical composition, neglecting the relationship between the two.
In order to overcome this limitation, the authors propose a novel approach called the probabilistic diffusion model. This model utilizes a geometrically equivariant Graph Neural Network (GNN) that considers both atomic positions and crystal lattices jointly. This multi-disciplinary approach integrates concepts from both chemistry and physics, enabling a more comprehensive understanding of crystal structures.
To evaluate the effectiveness of their model, the authors introduce a new generation metric inspired by Frechet Inception Distance (FID), a commonly used evaluation metric in computer vision. However, instead of using InceptionV3 as in FID, they use GNN energy prediction. This metric provides a more insightful analysis of the model’s capabilities by assessing the plausibility of the generated crystal structures.
In addition to the new generation metric, the authors also include commonly used metrics like validity, which further assesses the credibility of the generated crystal structures. By incorporating multiple evaluation metrics, they provide a comprehensive evaluation of their diffusion model.
The experiments conducted on existing benchmarks demonstrate the significance of the proposed diffusion model. It showcases the model’s ability to generate meaningful representations of crystal structures with specific chemical properties. The successful application of this approach opens up new avenues in material science research, where the integration of multiple disciplines becomes crucial for advancements.
Overall, this article highlights the importance of considering multidisciplinary concepts in material science. The proposed probabilistic diffusion model that combines chemistry and physics through a geometrically equivariant GNN is a promising step towards generating new crystal lattices with specific chemical properties. The utilization of new generation metrics expands the evaluation process and provides a more comprehensive understanding of the model’s capabilities.