Learning to Discover Interesting Mathematics

Hugging Face Daily Papers Papers
large-language-models mathematics theorem-proving self-expanding-library proof-difficulty intrinsic-interest mathlib-overlap

Summary

This paper defines intrinsic interestingness for mathematical theorems using proof-to-statement length ratio and trains a 27B model to predict proof difficulty, enabling the generation of more interesting theorems and self-expanding mathematical libraries with reduced overlap with existing knowledge like Mathlib.

Recently, Large Language Models (LLMs) have been increasingly able to solve advanced mathematical problems, including many that have been open for decades. This opens the door to expansion of mathematical knowledge at unprecedented scale. Yet, while LLMs may be able to conjecture and prove more and more theorems, it remains open whether this new mathematical knowledge is interesting or useful. We define intrinsic interestingness of a theorem as the ratio between the length of its proof and the length of its statement. We show that this correlates strongly with an extrinsic measure of the downstream utility of a theorem. We identify the difficulty of a proof conditioned on a set of premises as a useful primitive for computing these metrics, and train a 27B model that predicts proof difficulty more accurately than frontier general-purpose models. Optimizing for our metric creates a model capable of producing more interesting theorems, while also reducing substantial or full overlap with Mathlib from 91.9% to 30.6%, showcasing the creation of more out-of-distribution math. We show that our system can generate candidate theorems, select the most interesting among them, and iteratively build on a self-expanding mathematical library. These metrics provide a practical and quantifiable signal for ranking conjectures and guiding proof search within formal mathematical libraries. Our framework provides a path towards self-expanding, machine-verified mathematical libraries that can choose worthwhile statements without relying on human-supplied targets.
Original Article
View Cached Full Text

Cached at: 09/25/26, 11:48 PM

Paper page - Learning to Discover Interesting Mathematics

Source: https://huggingface.co/papers/2609.28603

Abstract

Recently,LargeLanguageModels(LLMs)havebeenincreasinglyabletosolveadvancedmathematicalproblems,includingmanythathavebeenopenfordecades.Thisopensthedoortoexpansionofmathematicalknowledgeatunprecedentedscale.Yet,whileLLMsmaybeabletoconjectureandprovemoreandmoretheorems,itremainsopenwhetherthisnewmathematicalknowledgeisinterestingoruseful.Wedefineintrinsicinterestingnessofatheoremastheratiobetweenthelengthofitsproofandthelengthofitsstatement.Weshowthatthiscorrelatesstronglywithanextrinsicmeasureofthedownstreamutilityofatheorem.Weidentifythedifficultyofaproofconditionedonasetofpremisesasausefulprimitiveforcomputingthesemetrics,andtraina27Bmodelthatpredictsproofdifficultymoreaccuratelythanfrontiergeneral-purposemodels.Optimizingforourmetriccreatesamodelcapableofproducingmoreinterestingtheorems,whilealsoreducingsubstantialorfulloverlapwithMathlibfrom91.9%to30.6%,showcasingthecreationofmoreout-of-distributionmath.Weshowthatoursystemcangeneratecandidatetheorems,selectthemostinterestingamongthem,anditerativelybuildonaself-expandingmathematicallibrary.Thesemetricsprovideapracticalandquantifiablesignalforrankingconjecturesandguidingproofsearchwithinformalmathematicallibraries.Ourframeworkprovidesapathtowardsself-expanding,machine-verifiedmathematicallibrariesthatcanchooseworthwhilestatementswithoutrelyingonhuman-suppliedtargets.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2609\.28603

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.28603 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.28603 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.28603 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Learning to Discover Interesting Mathematics

arXiv cs.LG

This paper introduces a framework for LLMs to discover and prove interesting mathematical theorems by optimizing for a metric based on proof difficulty, leading to more novel and useful mathematical knowledge with reduced overlap with existing libraries.

PriorProof: A Point-in-Time Measure of Technique Novelty for Formal Proofs

arXiv cs.AI

PriorProof introduces a method to measure the novelty of proof techniques in formal mathematics by analyzing the dependency footprint of Lean proof terms against a prior built from an earlier snapshot of Mathlib. The method agrees with human raters on 69.7% of pairs and provides interpretable score gaps.