Curium

🦀💎 Crystal structure and symmetry for humans 💎🦀

Curium is a library for working with crystal informatics in Rust. I may implement a larger set of functionality later, but for now it provides a complete description of space groups and symmetry operations, useful for quickly generating space group tables or computing various symmetry-related geometric quantities.

The library has two major goals:

• 🔥 Blazing speed. Right now, a large amount of performance-critical calculation—file I/O, distance checks, generating space group operations, enumerating Wyckoff positions—is done with painfully slow amounts of Python code due to its dominance in the scientific landscape. It's my hope that a Python wrapper for this library will remove this bottleneck, just as LLM tokenizers have been rewritten in Rust to remove that bottleneck and other libraries like numpy use low-level code to get C speed with Python syntax.
• 🔬 An API that makes understanding crystals easier. Existing libraries are often written for specific workflows and intended for use by experienced materials scientists. My hope is that Curium can be accessible to people like me—AI researchers trying to understand this exciting domain without deep background in the subject. Many crystal libraries have weakly-typed APIs, poor documentation, and enormous libraries of extremely specific algorithms without any kind of entry point for a user. My goal is to make working with crystals natural for the user, even (especially!) if the user is a novice.

I think both of these goals are nicely summarized by my first work—implementing a description of symmetry operations. Comparing with SymmOp, the pymatgen equivalent:

• 🔥 Curium can store symmetry operations compactly: due to more granular numeric types and sophisticated mathematical representations, a Curium SymmOp is about 10x smaller than the equivalent pymatgen SymmOp in memory.
• 🔥 Curium avoids using floating-point arithmetic to represent symmetry operations. This means a single Curium SymmOp is actually more like 2 Python numbers, because space isn't wasted storing decimal places that aren't even correct in the first place.
• 🔬 Curium can represent generalized affine transformations (via the Isometry type), but for symmetry operations in groups Curium implements the algorithm given in ITA to turn those matrix operations into geometric operations. This means users of Curium can easily figure out what it represents beyond a list of numbers.
• 🔬 With SymmOps defined entirely using integers, there's never any numerical stability issues, which can create challenging bugs. If you want to know if two operations are equal, just compare them directly!

Contributing

✨ Contributions are always welcome! ✨

With the project in such an embryonic stage, it's probably best to contact me directly if you're interested in building something within Curium. You can find me on here or, for more rapid communication, find me on Discord at pollardsrho.

Acknowledgements

• This would not be possible without the International Tables for Crystallography Volume A. It's an exceptional reference, and my hope with Curium is to make crystallography code as clear as the book's exposition.

• I'm part of the Machine Learning and Evolution Laboratory @usccolumbia at the University of South Carolina. This code is, for the time being, purely my own—if it's bad, blame me.

• The authors of pymatgen have set out the roadmap for what Curium aims to do. Curium's API, at least in the broad strokes, is heavily indebted to their work in that library. GEMMI, PyXtal, and spglib are other libraries that I've looked at for inspiration.