Ten advances in mathematics and theoretical computer science
Paper

Ten advances in mathematics and theoretical computer science

2026.08.02
·Service·by Homin.Lee
#AI#Computer Science#Deep Learning#Mathematics#Research

Key Points

  • 1OpenAI has utilized its internal Astra model to resolve ten long-standing open problems in mathematics and theoretical computer science, including Connes’s rigidity conjecture and the construction of non-sofic groups.
  • 2These mathematical proofs were generated by the AI, subsequently formalized into Lean certificates, and refined into manuscripts with human assistance to ensure accuracy and clear attribution.
  • 3This initiative highlights the growing role of AI as a collaborative research tool and emphasizes the company's commitment to supporting the scientific community through accessible and responsible technological advancement.

This paper announces ten major breakthroughs in mathematics and theoretical computer science, achieved using an internal version of OpenAI’s "Astra" model. These results address longstanding open problems—all stagnant for at least a decade—spanning fields such as high-dimensional geometry, group theory, quantum complexity, and extremal combinatorics.

Core Methodology

The research workflow utilized a hybrid human-AI collaboration strategy:
  1. Generation: The Astra model generated the core mathematical arguments to solve each problem. The computational cost for these solutions was approximately $2,000 at Sol API rates.
  2. Preparation: Human researchers collaborated with the model to structure these arguments into formal academic manuscripts.
  3. Verification: The model performed formalization of the arguments using the Lean proof assistant to ensure rigorous, verifiable correctness.
  4. Transparency: The authors provided "reasoning walkthroughs" for each solution, documenting the model’s chain-of-thought process.

Summary of Mathematical Results

  • High-Dimensional Sphere Packing: New upper bounds were established, reaching the Cohn–Elkies threshold.
  • Coding Theory: Exponentially improved bounds for the maximum size of binary and spherical codes at prescribed minimum distances.
  • Group Theory: A construction establishing the existence of non-sofic groups, a central open question.
  • Operator Algebras: A disproof of Connes’s rigidity conjecture regarding the uniqueness of certain groups defined by their von Neumann algebras.
  • Arithmetic Circuit Complexity: Derivation of new lower bounds for computing the permanent, including an arithmetic-formula lower bound of order O(n4/logn)O(n^4/\log n).
  • Quantum Complexity: A theorem providing exponential parallel repetition for general two-player quantum games, generalizing foundational classical complexity principles.
  • Lattice Cryptography: Established polynomial-factor hardness of approximation for the Closest Vector Problem (CVP).
  • Ehrhart’s Volume Conjecture: Determination of the maximum volume for a convex body where the centroid is the unique interior lattice point across all dimensions.
  • Extremal Combinatorics (Ramsey Theory): Resolution of Erdős problem 183, providing a superexponential lower bound for multicolor triangle Ramsey numbers.
  • Extremal Combinatorics (Graph Theory): Resolution of Erdős problems 146 and 180 concerning compactness and degeneracy conjectures.

Ethical Stance

The authors emphasize a commitment to transparency regarding AI-driven research. They argue against claiming sole human authorship for proofs generated by AI, advocating for a model where developers take responsibility for correctness while acknowledging the system's generative role. This initiative aligns with broader efforts to support the scientific community, including the provision of free model access to 100,000 researchers to facilitate future mathematical discovery.