Abstract:Generative models hold the promise of significantly expediting the materials design process when compared to traditional human-guided or rule-based methodologies. However, effectively generating high-quality periodic structures of materials on limited but diverse datasets remains an ongoing challenge. Here we propose a novel approach for periodic structure generation which fully respect the intrinsic symmetries, periodicity, and invariances of the structure space. Namely, we utilize differentiable, physics-based, structural descriptors which can describe periodic systems and satisfy the necessary invariances, in conjunction with a denoising diffusion model which generates new materials within this descriptor or representation space. Reconstruction is then performed on these representations using gradient-based optimization to recover the corresponding Cartesian positions of the crystal structure. This approach differs significantly from current methods by generating materials in the representation space, rather than in the Cartesian space, which is made possible using an efficient reconstruction algorithm. Consequently, known issues with respecting periodic boundaries and translational and rotational invariances during generation can be avoided, and the model training process can be greatly simplified. We show this approach is able to provide competitive performance on established benchmarks compared to current state-of-the-art methods.

What problem does this paper attempt to address?

The paper attempts to address the problem of effectively generating high-quality periodic structures on a limited and diverse dataset when generating periodic material structures. Existing generation methods often struggle to simultaneously meet the requirements of periodicity, symmetry, and invariance, resulting in low-quality generated material structures. Additionally, traditional generation methods typically generate atomic coordinates directly in Cartesian space, which leads to issues with periodic boundaries, translational, and rotational invariance. To solve these problems, the paper proposes a new method—StructRepDiff (Structure Representation Diffusion Model). This method generates material structures in the representation space instead of directly in Cartesian space, thereby avoiding the aforementioned issues. Specifically, the paper utilizes differentiable physics-based structure descriptors to describe periodic systems and combines them with a denoising diffusion model to generate new materials in the representation space. The generated representations are then reconstructed into corresponding Cartesian coordinates through a gradient optimization algorithm, thereby restoring the crystal structure. The main contributions of this method include: 1. Proposing a new method for generating materials in the representation space using a denoising diffusion model. 2. Proposing an effective material representation method that can generate structures in the representation space and demonstrating performance comparable to existing methods. 3. Proposing a gradient-based reconstruction method to reconstruct material structures from the representation space. Through these innovations, the paper aims to improve the quality and efficiency of periodic material generation, providing new tools and methods for research in the field of materials science.

Representation-space diffusion models for generating periodic materials

Crystal Diffusion Variational Autoencoder for Periodic Material Generation

Scalable Diffusion for Materials Generation

Towards Symmetry-Aware Generation of Periodic Materials

Generative Design of Crystal Structures by Point Cloud Representations and Diffusion Model

3-D Inorganic Crystal Structure Generation and Property Prediction via Representation Learning

Physics guided deep learning for generative design of crystal materials with symmetry constraints

Atomic structure generation from reconstructing structural fingerprints

Microstructure reconstruction of 2D/3D random materials via diffusion-based deep generative models

Statistically conditioned polycrystal generation using denoising diffusion models

Denoising diffusion-based synthetic generation of three-dimensional (3D) anisotropic microstructures from two-dimensional (2D) micrographs

Spectroscopy-Guided Discovery of Three-Dimensional Structures of Disordered Materials with Diffusion Models

Efficient Symmetry-Aware Materials Generation via Hierarchical Generative Flow Networks

Generative Inverse Design of Crystal Structures via Diffusion Models with Transformers

Latent Diffusion Models for Structural Component Design

Crystal Structure Prediction by Joint Equivariant Diffusion

WyCryst: Wyckoff Inorganic Crystal Generator Framework

Unified Model for Crystalline Material Generation