Advancing mathematics research with AI-driven formal proof search
Open accessThe DOI matches its Crossref record; 80 of the 80 references it cites were found in a registry or an index; none of the 41 sources its claims cite is checked for support yet.
Abstract
Large language models (LLMs) increasingly excel at mathematics tasks, but their unreliability limits their utility in mathematics research. A mitigation is to use LLMs to generate formal proofs in languages such as Lean, in which the compiler verifies every proof step. We present the first demonstration of this method’s value in solving open problems at scale. We built an artificial intelligence agent for formal proof search that autonomously resolved nine of 353 open Erdős problems, proved 44/492 On-Line Encyclopedia of Integer Sequences conjectures, and is being deployed in combinatorics, optimization, graph theory, algebraic geometry, and quantum optics research. Even a basic agent alternating LLM-based generation with Lean-based verification replicated the Erdős successes. These findings demonstrate the power of formal proof search as an enabler of autonomous mathematical discovery.
Overview
CoChat Summary
TL;DR
The paper reports a large-scale evaluation of AI-driven formal proof search in Lean, using the AlphaProof Nexus framework to attack open research problems. Its full-featured agent autonomously solved 9 of 353 open Erdős problems, proved 44 of 492 OEIS conjectures, and was deployed in several research areas.
Key findings
- The full-featured agent solved 9/353 open Erdős problems, including two questions open for 56 years, at a per-problem inference cost of a few hundred dollars.
- It proved 44/492 OEIS conjectures that manual review found correctly formalized and previously unproven.
- A basic agent alternating LLM generation with Lean verification replicated the Erdős successes but was costlier on the hardest problems.
- AlphaProof in standalone tree-search mode and basic agents using smaller models (Gemini 3.0 Flash, Gemini 3.1 Flash-Lite) solved none of the Erdős problems.
Methods
- The authors built AlphaProof Nexus, an LLM-aided proof-generation framework using Lean compiler feedback, with a basic agent of independent subagents and a full-featured agent adding evolutionary search, Elo-based sketch rating, P-UCB sampling, and AlphaProof as a tool.
- Evaluation used all 353 Lean formalizations of Erdős problems available in Formal Conjectures, terminating search after 3000 episodes if no proof was found; OEIS evaluation used 492 autoformalized open questions with test lemmas guarding against misformalization.
- The full-featured agent used Gemini 3.1 Pro for provers and Gemini 3.0 Flash for rater subagents; comparisons across agents used solve rate versus mean inference cost in USD.
Limitations
- The paper states that successes are concentrated in areas such as combinatorics, convex optimization, and number theory where Lean’s library is mature and tasks decompose into tractable subgoals; most Erdős problems remain out of reach, as do problems requiring extensive new theory.
- The paper states that agents inherit biases of their underlying LLMs and exhibit high search variance, and that reported costs do not capture the full cost of discovery, including identifying tractable problems.
- The paper’s failure analysis found that agents often offloaded core difficulty into a single `sorry` in a helper lemma, and that some top sketches relied on hallucinated literature lemmas; the text also notes misformalization risks, including density-interpretation ambiguities that required correction.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures



![Figure 2 Rater agent prompt (condensed). Elided text is represented by [...] . Text in braces denotes template variables populated at runtime. For example, { player blocks } is replaced by the sketches to be Figure 8 | Full prompt for the basic agent (A). Text in braces denotes template variables populated at runtime. For example, {code} is replaced by the current Lean file.](https://figures.cochat.ai/paper-figures/99176cf508264aaeaf6d185108bab6df/advancing-mathematics-research-with-ai-driven-formal-proof-search-figure-2.png)



Verification
Claims the paper makes
0 Valid · 3 Not valid · 29 Not checked · 127 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Several proofs require sophisticated constructions and the synthesis of distinct mathematical arguments.”
§ Advancing Mathematics Research with AI-Driven Formal Proof Search > 3. Systematic Evaluation on Open Problems
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Upon manual inspection, these lemmas proved to be hallucinations.”
§ Advancing Mathematics Research with AI-Driven Formal Proof Search > Failure Analysis.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The agent also served as a tool for detecting and fixing misformalizations.”
- Also stated in: § 6. Discussion · boosted
“the agents were powerful tools for detecting misformalizations.”
§ Advancing Mathematics Research with AI-Driven Formal Proof Search > AlphaProof Nexus
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 6. Discussion · boosted
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“in Erdős problems #125 and #741(i), the interpretation of”
§ Advancing Mathematics Research with AI-Driven Formal Proof Search > AlphaProof Nexus
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Following the correction of the ambiguity, the agent was still able to resolve the questions.”
§ Advancing Mathematics Research with AI-Driven Formal Proof Search > AlphaProof Nexus
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
98 more claims
Create a free account to see all 159 claims and every source verdict.
- Probably not a substantial claim. 30
- It may state two separate claims as one. 31
- The quote only partly states this claim. 4
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ Advancing Mathematics Research with AI-Driven Formal Proof Search > Conjecture Summary Proof Technique
“741 (ii) (1994) Existence of a basis 𝐴 of order 2 with 𝐴 1 + 𝐴 1 , 𝐴 2 + 𝐴 2 having bounded gaps for all 𝐴 1 , 𝐴 2 such that 𝐴 = 𝐴 1 ⊔ 𝐴 2”
§ 1. Role and Goal You are a world-class mathematician and Lean 4 expert.Your goal is to solve hard mathematical problems by devising a proof strategy and translating it into a Lean 4 proof.# 2. Your Task You will be given Lean code containing theorem statement with a partial proof.You goal is to modify the file to continue proving the statement until there are no sorries left.You will be getting feedback from Lean Compiler after every time you modify code.Compilation errors will be highlighted and you need to keep iterating until the code compiles.CRITICAL: Don't end a session with the code which doesn't compile.You session will be discarded if the final proof doesn't compile.If you can't finish the proof in the current session at least make sure it compiles before you wrap up the session.Put any findings, plans or solutions as comments in the file.Only file content will be passed to the next session.Think like a mathematician: focus on the key insights, proof structure, and creative steps (e.g., constructing an object via 'let').If you get stuck on the main proof, try to gain insights by exploring diverse ideas: study specific cases, or define and attempt to prove interesting generalizations, specializations, or variants of the problem statement as new helper lemmas.Prefer clever mathematical arguments over brute-force casework where possible.Some of these problems are very hard, possibly even open problems in mathematics.But don't be discouraged|approach them with curiosity and persistence.Like George Dantzig, who solved two famous unsolved problems in statistics thinking they were homework, you might solve a problem you think is difficult simply by not knowing it was considered impossible!Believe in your ability to find creative solutions where others might not.CRITICAL: You MUST use tools available to you exhaustively to refine your proof until you either find a proof or are certain the current direction is flawed.Do NOT give up easily, and NEVER put off formalization work to the next session if you have more tool calls available.Your effort in each turn must be maximal: try to solve the problem in full, as if there is no next session.**No Imports:** The execution environment imports 'Mathlib' by default.Do **not** add any 'import' statements in your proof.# Current proof This is the proof to be modified in your session.{code} Make sure that you only propose changes inside sections enclosed by '--EVOLVE-BLOCK-START' and '--EVOLVE-BLOCK-END', or '--EVOLVE-VALUE-START' and '--EVOLVE-VALUE-END' comments. .Text in braces denotes template variables populated at runtime.For example, {player blocks} is replaced by the sketches to be compared. > Erdős #741-(ii)
“Question.Is there a basis 𝐴 of order 2 such that if 𝐴 = 𝐴 1 ⊔ 𝐴 2 then 𝐴 1 + 𝐴 1 and 𝐴 2 + 𝐴 2 cannot both have bounded gaps?”
Citations
Cites
Towards Deep Learning Models Resistant to Adversarial Attacks
Madry, A., Makelov, A., Schmidt, L., Tsipras, D., & Vladu, A. (2017) · arXiv · doi:10.48550/arxiv.1706.06083 · Cited by 6,602
A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play
Silver, D., Hubert, T., Schrittwieser, J., Antonoglou, I., Lai, M., Guez, A., Lanctot, M., Sifre, L., Kumaran, D., Graepel, T., et al. (2018) · Science · doi:10.1126/science.aar6404 · Cited by 3,562
Self-citationThe On-Line Encyclopedia of Integer Sequences
Sloane, N. J. A. (2007) · Lecture Notes in Computer Science · doi:10.1007/978-3-540-73086-6_12 · Cited by 3,225
The extragradient method for finding saddle points and other problems
Korpelevich, G. M. (1976) · Medical Entomology and Zoology · Cited by 2,779
A collection of mathematical problems
Ulam, S. M. (1960) · Medical Entomology and Zoology · Cited by 2,262
Individual choice behavior; a theoretical analysis.
Luce, R. D. (n.d.) · University of Michigan Library Repository · Cited by 1,870
Genetic programming as a means for programming computers by natural selection
Koza, J. (1994) · Statistics and Computing · doi:10.1007/bf00175355 · Cited by 1,685
(2002) · Lecture Notes in Computer Science · doi:10.1007/3-540-45949-9 · Cited by 1,629
Fixed points of nonexpanding maps
Halpern, B. R. (1967) · Bulletin of the American Mathematical Society · doi:10.1090/s0002-9904-1967-11864-0 · Cited by 1,415
A Reductions Approach to Fair Classification
Agarwal, A., Beygelzimer, A., Dudík, M., Langford, J., & Wallach, H. (2018) · arXiv · doi:10.48550/arxiv.1803.02453 · Cited by 1,288
Research works printed but not linked: 10
- 10.1006/jcta.2000.3142 · DOI
- N. J. Sloane The On-Line Encyclopedia of Integer Sequences in Proceedings of the 14th Symposium on Towards Mechanized Mathematical Assistants: 6th International Conference Calculemus ’07 / MKM ’07 (Springer Berlin Heidelberg) (2007) p. 130.10.1007/978-3-540-73086-6_12 · DOI · Paper page
- A. Thakur, G. Tsoukalas, Y. Wen, J. X. Xin, S. Chaudhuri, An in-context learning agent for formal theorem-proving, in Conference on Language Models (2024).
- A. Nagda P. Raghavan A. Thakurta Reinforced generation of combinatorial structures: Ramsey numbers. arXiv.2603.09172 [math.CO] (2026). · arXiv
- D. Alvarez-Melis, T. Jaakkola, S. Jegelka, Structured Optimal Transport, in Proceedings of the Twenty-First International Conference on Artificial Intelligence and Statistics (2018).
- H. Shugart J. M. Altschuler Negative stepsizes make gradient-descent-ascent converge. arXiv preprint arXiv:2505.01423 (2025). · arXiv
- Y. Cai, W. Zheng, Accelerated Single-Call Methods for Constrained Min-Max Optimization, in International Conference on Learning Representations (2023).
- S. Lee, D. Kim, Fast extragradient methods for smooth structured nonconvex-nonconcave minimax problems. Adv. Neural Inf. Process. Syst. (2021).
- S. Lee D. Kim Semi-Anchored Gradient Methods for Nonconvex-Nonconcave Minimax Problems (2024) https://openreview.net/forum?id=rmLTwKGiSP.
- L. Chen, L. Luo, Near-Optimal Algorithms for Makinghe Gradient Small in Stochastic Minimax Optimization. J. Mach. Learn. Res. 25, 1–44 (2024).
Cited by
LeanMarathon: Toward Reliable AI Co-Mathematicians through Long-Horizon Lean AutoformalizationPreprint
Zhang, Y., Sun, Y., Suzuki, T., Lee, J. D., & Liu, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.05400 · Cited by 26
LEAP: Supercharging LLMs for Formal Mathematics with Agentic FrameworksPreprint
Kung, P.-N., Song, L., Hwang, D., Yoon, J., Li, C.-L., Severini, S., Olšák, M., Lockhart, E., Le, Q. V., Gokturk, S. B., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.03303 · Cited by 20
Self-citationSimple Baselines are Competitive with Code EvolutionPreprint
Gideoni, Y., Risi, S., & Gal, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.16805 · Cited by 15
A Domain-Specific Harness for End-to-End Automation of Optimization ResearchPreprint
Kim, H.-T., Ryu, E. K., & Gupta, S. D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.07407 · Cited by 13
Lin, H., Woodruff, D. P., Deng, Y., Mao, J., Zuo, S., & Mirrokni, V. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.15983 · Cited by 9
Metadata
- Authors
- Tsoukalas, G., Kovsharov, A., Shirobokov, S., Surina, A., Firsching, M., Bérczi, G., Ruiz, F. J. R., Suggala, A., Wagner, A. Z., Wieser, E., Yu, L., Huang, A., Horváth, M. Z., Ferraiuolo, A., Michalewski, H., Grosu, C., Lockhart, E., Hubert, T., Balog, M., Kohli, P., & Chaudhuri, S.
- Affiliations
- Google DeepMind (United Kingdom) (GB)
- Aarhus University (DK)
- Google DeepMind, London, UK.
- Google, Mountain View, CA.
- Corresponding author
- Tsoukalas, George
- Kovsharov, Anton
- Shirobokov, Sergey
- Kohli, Pushmeet
- Chaudhuri, Swarat
- ORCID
- Tsoukalas, George
- Kovsharov, Anton
- Shirobokov, Sergey
- Surina, Anja
- Bérczi, Gergely
- Ruiz, Francisco J. R.
- Suggala, Arun
- Wagner, Adam Zsolt
- Wieser, Eric
- Yu, Lei
- Huang, Aja
- Horváth, Miklós Z.
- Ferraiuolo, Andrew
- Michalewski, Henryk
- Grosu, Codrut
- Lockhart, Edward
- Hubert, Thomas
- Balog, Matej
- Kohli, Pushmeet
- Chaudhuri, Swarat
- Venue
- Science
- Published
- 2026-10-08
- Volume
- 394
- Issue
- 6820
- Pages
- 234-239
- ISSN
- 0036-8075 (ISSN-L)
- 1095-9203
- OpenAlex
- W7162098200
- Cited by
- 62
- References
- 76
- Citations by year
- 2026: 61
- FWCI
- 97.09
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- cc-by-nc-nd
- Locations
- Science (publishedVersion)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by-nc-nd)
- arXiv (Cornell University) (cc-by-nc-nd)
- Topics
- Physical Sciences › Computer Science › Computational Theory and Mathematics › Mathematics, Computing, and Information Processing
- Physical Sciences › Computer Science › Software › Logic, programming, and type systems
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
- Keywords
- large language models, Lean theorem prover, AI agents, open problems in mathematics, Erdős problem, proof verification, autonomous reasoning
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-11-21, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-21 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Advancing mathematics research with AI-driven formal proof search. https://app.cochat.ai/papers/10.1126/science.aej2213
- The paper
- Tsoukalas, G., Kovsharov, A., Shirobokov, S., Surina, A., Firsching, M., Bérczi, G., Ruiz, F. J. R., Suggala, A., Wagner, A. Z., Wieser, E., Yu, L., Huang, A., Horváth, M. Z., Ferraiuolo, A., Michalewski, H., Grosu, C., Lockhart, E., Hubert, T., Balog, M., . . . Chaudhuri, S. (2026). Advancing mathematics research with AI-driven formal proof search. Science. https://doi.org/10.1126/science.aej2213
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.