Tech • IA • Crypto
X-UPS Days 2026: "Formal Verification of Proofs" – Sébastien Gouzel 1
5/10TL;DR
Proof assistants like Lean are emerging as tools to verify mathematical results, reduce errors, and standardize rigor, though their adoption is limited by the heavy effort required to formalize existing theory.
KEY POINTS
Growing concern over unreliable math papers
Increasing pressure to publish has led to a rise in papers containing errors or unclear assumptions. Even when authors act in good faith, implicit conventions within specialized communities can cause misinterpretations, especially for newcomers applying results خارج their original context.
Proof assistants as a potential safeguard
Tools such as Lean, based on type theory, allow mathematicians to encode definitions and proofs in a formal language that a computer can verify step by step. Requiring formal verification before publication could significantly reduce incorrect results and ambiguities.
Accessibility without deep technical knowledge
Modern proof assistants are designed so users do not need to understand their internal mechanics. Much like operating systems, they abstract complexity, enabling mathematicians to focus on writing proofs rather than implementing logical foundations.
Example: asymptotic behavior of semi-contractions
A classical result by Kohlberg and Neyman (1983) shows that iterates of certain non-expansive maps behave like translations. A later proof by Karlsson uses subadditivity and geometric arguments to establish convergence toward a linear drift vector, illustrating how nontrivial reasoning can be formalized.
Formalization reveals hidden errors
Translating proofs into Lean has led to the discovery of mistakes not only in others’ work but also in the formalizer’s own publications. The rigid structure forces every assumption and step to be explicit, eliminating reliance on intuition or convention.
Automation remains limited and guided
While Lean includes automation tools that can resolve simple steps, it does not independently discover proofs. Users must guide the process, choosing which lemmas and strategies to apply. This ensures transparency but requires significant user input.
Dependence on large mathematical libraries
Formal proofs rely heavily on extensive libraries like Mathlib, which contain definitions and theorems ranging from basic arithmetic to advanced analysis. Without these, even simple results would require thousands of lines of foundational code.
Time cost remains a major barrier
Formalizing a result can take far longer than writing a traditional proof, especially when prerequisites are not already available in libraries. In cutting-edge research areas, missing groundwork often makes full formalization impractical.
Not all mathematics is equally easy to formalize
Fields with precise, algebraic structures tend to be more straightforward. Areas involving informal reasoning, implicit identifications, or geometric intuition—such as differential geometry—pose greater challenges, though progress is ongoing.
Collaborative and evolving ecosystem
Proof assistant ecosystems are open-source and rapidly evolving. Libraries grow cumulatively, and improvements in tools gradually reduce the effort required. However, updates can temporarily disrupt existing formalizations.
Balancing rigor and practicality
Users can temporarily admit unproven lemmas to advance work, but systems allow checking whether final results depend on such assumptions. This mirrors traditional mathematical practice, where proofs are refined iteratively.
CONCLUSION
Proof assistants offer a promising path toward more reliable and transparent mathematics, but widespread adoption depends on overcoming significant practical and cultural barriers.