Sander Dahmen, Johannes Hölzl, and I have formalized Ellenberg and Gijswijt’s solution to the cap set problem in Lean. Our paper about this project will appear at ITP 2019. Here are some slides from a talk I gave about this formalization.
The cap set problem is a combinatorial question, easy to ask but difficult to answer. Briefly, it asks about the growth rate of the cardinality of cap sets, subsets of a group that contain no three-term arithmetic progression, as the dimension of the underlying space increases. Ellenberg and Gijswijt discovered a solution in 2016; their celebrated proof appeared in the Annals of Mathematics in 2017. The approach that Ellenberg and Gijswijt take is an extremely clever application of elementary methods, which makes it a good candidate to formalize.
Very little cutting-edge mathematics is formalized in proof assistants. Reasons for this are well-documented throughout the literature: the costs are too high; the tools and infrastructure are not ready; the general expertise isn’t there. The Lean Forward project aims to change this, by spurring collaboration between tool developers, formalizers, and mathematicians. Our project is a step toward these goals. Admittedly, our choice of a formalization target is uniquely accessible. But this gives us a first data point for Lean Forward: at least one modern, noteworthy paper in mathematics can be formalized with a reasonable amount of time and effort.
Lean Forward emphasizes that collaboration between people from different areas is needed to bring formalized mathematics to the mainstream. Our project is an example of exactly this kind of collaboration. We need mathematicians, computer scientists, engineers, and people in between to come together to make lofty, ambitious formalization projects into reality.