Abstract:We consider the problem of generating periodic materials with deep models. While symmetry-aware molecule generation has been studied extensively, periodic materials possess different symmetries, which have not been completely captured by existing methods. In this work, we propose SyMat, a novel material generation approach that can capture physical symmetries of periodic material structures. SyMat generates atom types and lattices of materials through generating atom type sets, lattice lengths and lattice angles with a variational auto-encoder model. In addition, SyMat employs a score-based diffusion model to generate atom coordinates of materials, in which a novel symmetry-aware probabilistic model is used in the coordinate diffusion process. We show that SyMat is theoretically invariant to all symmetry transformations on materials and demonstrate that SyMat achieves promising performance on random generation and property optimization tasks. Our code is publicly available as part of the AIRS library (<a class="link-external link-https" href="https://github.com/divelab/AIRS" rel="external noopener nofollow">this https URL</a>).

What problem does this paper attempt to address?

The paper attempts to address the problem of how to generate periodic material structures with physical symmetry. Specifically, existing molecular generation methods have extensively studied symmetry-aware molecular generation, but these methods fail to fully capture the different symmetries of periodic materials. Therefore, this paper proposes a new periodic material generation method—SyMat, which can capture the physical symmetry of periodic material structures. ### Main Issues 1. **Challenges in Generating Periodic Materials**: - Periodic materials have complex 3D periodic structures, and existing deep generative models struggle to effectively capture their physical symmetry. - Ideally, the generative model should maintain invariance to the symmetry transformations of periodic materials (including permutation, rotation, translation, and periodic transformations). 2. **Limitations of Existing Methods**: - Existing generative models are mainly applied to non-periodic molecules and cannot be directly applied to the generation of periodic materials. - Some existing generative models (such as variational autoencoders and generative adversarial networks) violate the physical symmetry of materials when generating periodic materials. ### Solution 1. **SyMat Method**: - **Atom Type and Lattice Generation**: Generate the set of atom types, lattice lengths, and lattice angles through a variational autoencoder model. - **Atom Coordinate Generation**: Generate atom coordinates using a score-based diffusion model, which employs a new symmetry-aware probabilistic model during the coordinate diffusion process. 2. **Symmetry Awareness**: - **Permutation Invariance**: Achieved by generating a set of atom types rather than specific atom type vectors. - **Rotation and Translation Invariance**: Achieved by generating lattice lengths and lattice angles rather than directly generating lattice vectors. - **Periodic Transformation Invariance**: Achieved by using a periodic-invariant edge distance score function in a multi-graph representation. ### Experimental Results - **Random Generation Task**: SyMat performs excellently in the random generation of periodic materials. - **Property Optimization Task**: SyMat also achieves satisfactory results in generating periodic materials with target properties. ### Summary This paper proposes a new periodic material generation method, SyMat, which successfully captures the physical symmetry of periodic materials through a variational autoencoder and a score-based diffusion model, thereby performing well in both generation tasks and property optimization tasks.

Towards Symmetry-Aware Generation of Periodic Materials

Representation-space diffusion models for generating periodic materials

WyCryst: Wyckoff Inorganic Crystal Generator Framework

Symmetry-adapted modeling for molecules and crystals

A Dataset Generation Framework for Symmetry-Induced Mechanical Metamaterials

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

Symmetry Based Material Optimization

Efficient Symmetry-Aware Materials Generation via Hierarchical Generative Flow Networks

Symmetry-based computational search for novel binary and ternary 2D materials

Crystal Diffusion Variational Autoencoder for Periodic Material Generation

Symmetry Engineering in Low-Dimensional Materials

Symmetry adaptation for self-consistent many-body calculations

Symmetry-Constrained Generation of Diverse Low-Bandgap Molecules with Monte Carlo Tree Search

Material symmetry recognition and property prediction accomplished by crystal capsule representation

Symmetry Enhanced Network Architecture Search for Complex Metasurface Design

SLICES-PLUS: A Crystal Representation Leveraging Spatial Symmetry

Symmetry-guided data-driven discovery of native quantum defects in two-dimensional materials

Atomic structure generation from reconstructing structural fingerprints

Periodic rhomboidal cells for symmetry-preserving homogenization and isotropic metamaterials

Symmetry-Informed Geometric Representation for Molecules, Proteins, and Crystalline Materials

MolSym: A Python package for handling symmetry in molecular quantum chemistry