TL;DR: A fast generative model for crystals conditioned on space group symmetry; based on Wyckoff positions.
Abstract: Crystal symmetry plays a fundamental role in determining its physical, chemical, and electronic properties such as electrical and thermal conductivity, optical and polarization behavior, and mechanical strength. Almost all known crystalline materials have internal symmetry. However, this is often inadequately addressed by existing generative models, making the consistent generation of stable and symmetrically valid crystal structures a significant challenge. We introduce WyFormer, a generative model that directly tackles this by formally conditioning on space group symmetry. It achieves this by using Wyckoff positions as the basis for an elegant, compressed, and discrete structure representation. To model the distribution, we develop a permutation-invariant autoregressive model based on the Transformer encoder and an absence of positional encoding. Extensive experimentation demonstrates WyFormer's compelling combination of attributes: it achieves best-in-class symmetry-conditioned generation, incorporates a physics-motivated inductive bias, produces structures with competitive stability, predicts material properties with competitive accuracy even without atomic coordinates, and exhibits unparalleled inference speed.
Lay Summary: When we think of crystals, like sparkling gemstones or intricate snowflakes, we often picture something beautiful. This beauty arises directly from their internal symmetry – the highly organized way atoms, like tiny LEGO bricks, arrange themselves. This precise atomic pattern is not just for show; it's fundamental, dictating a crystal's properties: Will it be strong? Conduct electricity? Be transparent?
Scientists are always searching for new materials with amazing properties. But with countless ways atoms could combine, finding useful ones is like searching an infinite haystack. Trying to design new materials without understanding their symmetry rules is like sticking LEGO blocks together without a plan or instructions. While such improvisation might occasionally yield something interesting, it often leads to unstable structures or materials that don't have the desired properties, severely limiting the discovery of truly innovative materials.
This is where our AI tool, WyFormer, comes in. It's designed to be much smarter by learning these fundamental "LEGO instructions" of crystal symmetry. WyFormer uses concepts like "symmetry groups" (the blueprints for atomic arrangement) and "Wyckoff positions" (special spots atoms prefer) to describe crystals in a compact and efficient way – essentially giving the AI a simplified language for these complex structures.
Our AI, using a powerful Transformer architecture (similar to what powers advanced chatbots), learns to generate new crystal "recipes" that automatically follow these symmetry rules. A key feature is its understanding that, much like a cooking recipe, the order in which you list the atomic ingredients doesn't change the final crystal. WyFormer is not only fast but also excels at creating diverse, stable structures and can even predict material properties effectively without needing full 3D atomic details.
By teaching AI the fundamental language of crystal symmetry, WyFormer can help scientists discover novel materials with desired properties much more quickly. This could speed up breakthroughs in all sorts of fields, from new types of batteries and solar cells to stronger alloys and more efficient electronics, offering a powerful assistant to explore the vast world of possible materials.
Link To Code: https://github.com/SymmetryAdvantage/WyckoffTransformer
Primary Area: Applications->Chemistry, Physics, and Earth Sciences
Keywords: material design, machine learning, crystal generation, space group symmetry, Transformer, Wyckoff position, generative model, autoregressive model, permutation invariance
Submission Number: 10229
Loading