Abstract:The crystal diffusion variational autoencoder (CDVAE) is a machine learning model that leverages score matching to generate realistic crystal structures that preserve crystal symmetry. In this study, we leverage novel diffusion probabilistic (DP) models to denoise atomic coordinates rather than adopting the standard score matching approach in CDVAE. Our proposed DP-CDVAE model can reconstruct and generate crystal structures whose qualities are statistically comparable to those of the original CDVAE. Furthermore, notably, when comparing the carbon structures generated by the DP-CDVAE model with relaxed structures obtained from density functional theory calculations, we find that the DP-CDVAE generated structures are remarkably closer to their respective ground states. The energy differences between these structures and the true ground states are, on average, 68.1 meV/atom lower than those generated by the original CDVAE. This significant improvement in the energy accuracy highlights the effectiveness of the DP-CDVAE model in generating crystal structures that better represent their ground-state configurations.

What problem does this paper attempt to address?

The paper mainly addresses two key issues: 1. **Improving Crystal Structure Generation Models**: The paper proposes a new crystal structure generation framework called Diffusion Probabilistic Crystal Diffusion Variational Autoencoder (DP-CDVAE). This model combines Variational Autoencoder (VAE) with Diffusion Probabilistic (DP) models, aiming to improve the quality of generated crystal structures by adopting a diffusion probabilistic approach instead of the traditional score matching method. 2. **Enhancing the Ability to Generate Crystal Structures Close to Ground State**: The paper particularly focuses on the similarity between the generated crystal structures and the ground state structures obtained through Density Functional Theory (DFT) calculations. The study finds that compared to previous methods (such as the original CDVAE model), the crystal structures generated by DP-CDVAE are energetically closer to their corresponding ground state configurations, indicating that these structures are more likely to represent actual ground state structures. Specifically, the DP-CDVAE model consists of two parts: the VAE part is responsible for predicting the lattice parameters of the crystal structure and the number of atoms in the unit cell; the diffusion part uses a diffusion probabilistic method to denoise fractional coordinates and predict atomic coordinates. The paper demonstrates that the DP-CDVAE model performs comparably to the original CDVAE model in reconstruction tasks, while significantly outperforming it in generating ground state structures. For example, in the carbon structure generation task, the energy difference between the structures generated by DP-CDVAE and those optimized through DFT is reduced by an average of approximately 68.1 meV/atom, indicating the advantage of DP-CDVAE in generating crystal structures closer to the ground state.

Diffusion probabilistic models enhance variational autoencoder for crystal structure generative modeling

Diffusion probabilistic models enhance variational autoencoder for crystal structure generative modeling

Con-CDVAE: A method for the conditional generation of crystal structures

An effective method for generating crystal structures based on the variational autoencoder and the diffusion model

Deep learning generative model for crystal structure prediction

Data-driven discovery of 2D materials by deep generative models

Variational Autoencoding Molecular Graphs with Denoising Diffusion Probabilistic Model

Data-driven discovery of novel 2D materials by deep generative models

Crystal Diffusion Variational Autoencoder for Periodic Material Generation

Vector Field Oriented Diffusion Model for Crystal Material Generation

Generative Inverse Design of Crystal Structures via Diffusion Models with Transformers

Crystal Structure Prediction by Joint Equivariant Diffusion

Latent Space Diffusion Models of Cryo-EM Structures

Data-Driven Score-Based Models for Generating Stable Structures with Adaptive Crystal Cells

Self-Supervised Generative Models for Crystal Structures

VQCrystal: Leveraging Vector Quantization for Discovery of Stable Crystal Structures

Equivariant Diffusion for Crystal Structure Prediction

PCVAE: A Physics-informed Neural Network for Determining the Symmetry and Geometry of Crystals

Crystal Structure Prediction Using Generative Adversarial Network with Data-Driven Latent Space Fusion Strategy

Scalable Diffusion for Materials Generation