ATLAS: Autoformalized Textbook Library At Scale (github.com)

ATLAS, a new initiative from Facebook Research, has successfully autoformalized a vast library of textbook mathematics using Lean 4, translating informal statements and proofs into formal Lean code. This library currently encompasses a wide array of topics, including analysis, algebra, geometry, and more, covering around 630,999 lines of code with a significant 71.3% of selected statements formalized. Powered by the AutoformBot pipeline, ATLAS aims to supply reusable formal components that can enhance both human and machine-driven formalization efforts in mathematics. The significance of ATLAS lies in its potential to revolutionize how mathematical knowledge is documented and verified, making it more accessible for rigorous computational use. As this project is intended to be a living resource, the team emphasizes ongoing efforts to expand the corpus, refine the generated content, and align it with Mathlib conventions. The library also features evaluation metrics and a visualizer to facilitate exploration and comparison between informal statements and their formal counterparts, underscoring its utility for both researchers and educators within the AI/ML community. External contributions are encouraged, highlighting a collaborative approach to advancing mathematical formalization.