Guiding AI Agents to Generate Conjectures by Encoding How Theorems Build on One Another
Organizations: University of Pennsylvania
Abstract
Recently, LLMs have been making headlines, and headway, in mathematical research. Much of this work focuses on proving existing open conjectures. Deciding what to prove is harder and less studied, because of the lack of a clear criterion for what makes a conjecture worth pursuing. We present a method for conjecture generation that builds on the observation that mathematical theorems build on one another: new theorems can modify the setting or hypotheses of earlier theorems, strengthen their conclusions, or extend their scope. We encode these relationships in a new kind of graph, the Theorem Evolution Graph (TEG), which we instantiate as an online search method for LLM agents. An agent builds a TEG in two steps: (1) identifying relevant theorems from papers found by following citations backward and forward from a seed paper; and (2) comparing those theorems to record how one builds on another. LLM judges and an expert mathematician preferred TEG-generated conjectures over those from an unconstrained baseline. LLM judges favored TEG conjectures for significance, ambition, and insight, and favored baseline conjectures for plausibility. Ablations suggest a benefit from constructing and using a TEG even when the available papers are fixed.
Figures & tables
| Variant | Graph supplied to the agent |
|---|---|
| Full | Full graph, including edge descriptions. |
| Unannotated | Same nodes and directed edges, without edge descriptions. |
| Shuffled | Same nodes and number of edges, with edges placed between random node pairs and no descriptions. |
| Similarity | Same nodes, linked by textual similarity using TF-IDF, with at most two links per node. |
| Edgeless | Same nodes, with all edges removed. |
| Seed-only | Only the seed paper’s theorem nodes and the described edges among them. |
| Criterion | TEG | Baseline | TEG preferred (%) | 95% CI | |
|---|---|---|---|---|---|
| Overall preference | 50 | 29 | 63 | [52, 73] | |
| Significance | 103 | 54 | 66 | [58, 73] | |
| Novelty | 59 | 43 | 58 | [48, 67] | |
| Plausibility | 39 | 84 | 32 | [24, 40] | |
| Clarity | 16 | 19 | 46 | [30, 62] | |
| Ambition | 90 | 50 | 64 | [56, 72] |
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
| Role | Model | Budget |
|---|---|---|
| Generating agents (TEG and baseline; all experiments) | Claude Opus 5.5 | $7, 150 steps, 60 min per run |
| Clarity check tool | Claude Opus 5.5 | one call per use |
| Referees, Exp. 1 | GPT-6.1, Grok 4.7, and Claude Sonnet 5.5, one referee each | $3.50, 60 steps per referee |
| Objection checks, Exp. 1 | Rotated across labs: Sonnet checks Grok, GPT checks Sonnet, Grok checks GPT | $2, 40 steps per check |
| Referee, Exp. 2–3 | GPT-6.1 | $3.50, 60 steps |
| Objection checks, Exp. 2–3 | Grok 4.7 (Exp. 2); Claude Sonnet 5.5 (Exp. 3) | $2, 40 steps per check |
| Title | arXiv | Source | Experiments |
|---|---|---|---|
| Local-global principles for torsors over arithmetic curves | 1108.3323v5 | Expert | 1, 2 |
| Transcendental splitting fields of division algebras | 2203.15951v2 | Expert | 1, 2 |
| Index reduction for Brauer classes via stable sheaves (with an appendix by Bhargav Bhatt) | 0706.1072v3 | Expert | 1, 2 |
| Weierstrass preparation and algebraic invariants | 1109.6362v2 | Expert | 1, 2 |
| The Clifford algebra of a finite morphism | 1509.07195v4 | Expert | 1, 2 |
| On rational connectedness and parametrization of finite Galois extensions | 2502.15674v1 | Expert | 1 |
AI Agents for Scientific Discovery
AI agents for scientific discovery use language models, tools, and experiments to generate, test, and refine scientific knowledge.
See all 323 papers