mathematical proof techniques

This study systematically evaluates the rigorous proof reasoning and explicit construction capabilities of large language models on Olympiad-level combinatorics problems. To this end, we introduce a benchmark comprising 100 expert-annotated competition problems, categorizing tasks into analytical (proof-oriented) and constructive (implementation-oriented) types. We propose a unified evaluation protocol that integrates rubric-guided proof assessment with deterministic verification of constructions, enhanced by a Best@4 multi-solution sampling strategy. Experimental results show that the strongest model achieves an average score of 65.4% overall (75.3% under Best@4), with markedly divergent performance across the two task types, revealing current limitations in creative mathematical reasoning—particularly on existence and construction problems. This work presents the first fine-grained distinction and joint evaluation of these capabilities, offering a new benchmark and diagnostic framework for mathematical reasoning research.

This work investigates the goal-directed generation of mathematically meaningful theorems—or lemmas suitable for automated proof—from a given set of axioms. To this end, it introduces a novel approach grounded in the propositions-as-types paradigm, which systematically partitions the space of proof terms according to inductive levels and integrates proof-term enumeration with compression techniques, including separation-based reduction, DAG compression, and combinatory logic. This framework enables the efficient construction and compact representation of proof structures. Experimental evaluation on a fragment of Metamath’s set.mm library demonstrates that the method successfully produces nontrivial and semantically relevant theorems, thereby confirming its feasibility and advantages in the context of automated theorem discovery.

This work addresses a critical gap in existing mathematical formalization benchmarks, which predominantly focus on propositional verification while neglecting the evaluation of explicit solution construction—such as numerical values or algorithms—particularly in applied mathematics. To bridge this gap, the authors propose a construct-and-verify workflow framework that requires agents to first generate concrete solutions and then formally prove their correctness. Building upon this framework, they introduce AMBER, a novel benchmark for applied mathematical reasoning spanning convex analysis, optimization, numerical linear algebra, and high-dimensional probability. Implemented in Lean 4, this benchmark enables the first systematic evaluation of large language models on constructive tasks, revealing that general-purpose reasoning models significantly outperform specialized theorem provers, the latter suffering from “tactic overfitting” that limits their generalization. The study further underscores the pivotal role of instruction-following capability in multi-task formal reasoning.

Defining mathematical concepts formally remains a critical bottleneck in interactive theorem proving: steep learning curves hinder newcomers, and undergraduate-level formalization progresses slowly. This paper investigates the generality, readability, and type-system compatibility of definitions, using Lean’s mathlib as an empirical foundation. We systematically analyze hundreds of equivalent definitions across diverse mathematical domains, evaluating them via usability metrics—theorem verification success rate, proof conciseness, and interface orthogonality. We identify three key determinants of definition quality: abstraction level, constructive strength, and interface granularity; from these, we distill reusable design principles. Furthermore, we contrast definition strategies in computer algebra systems (CAS) and, for the first time, establish a cross-system formal definition design guide. Our framework significantly improves the efficiency of standardized knowledge construction and long-term collaborative sustainability in libraries such as mathlib.

This paper addresses the challenge of structured representation and efficient processing of formal mathematical proofs. Methodologically, it models proofs as logical terms and uniformly characterizes both human expert reasoning and automated inference using syntactic compression trees. It establishes, for the first time, a theoretical connection between proof structuring and syntax-tree compression, modeling the Metamath library as a grammar system generating a compressed giant proof tree. The approach employs condensed separate-term representations, context-free grammar compression, and a customized parse-reconstruct toolchain. Key contributions include: (1) the first grammar-compression-based framework for proof structuring; (2) a scalable proof compression and reconstruction toolkit; and (3) empirical results demonstrating substantial reductions in storage overhead, alongside support for structure-aware retrieval and synthesis.

This work investigates the creative space of mathematical proofs under constraints, with a particular focus on the impact of non-constructive reasoning. We introduce a strategy ablation methodology that integrates our custom-built Meno automated formalization tool with Goedel Prover embeddings to systematically explore both formal and informal proof spaces for foundational theorems from *Analysis I* within the Lean theorem prover. Our experiments successfully generate a novel class of machine-produced proofs, revealing that these proofs cluster along low-dimensional submanifolds in a high-dimensional representation space and significantly diverge from human-constructed proof trajectories. This study provides the first quantitative characterization of the structural differences between machine-generated and human proofs.

Current approaches to automatically verifying natural language mathematical proofs lack modularity and unambiguous structure, especially in advanced mathematics. This work proposes pseudo-formalization (PF), a representation that decomposes proofs into self-contained modules, each explicitly specifying premises, conclusions, and inference steps. We introduce a block verification (BV) algorithm that leverages large language models to independently validate each module. Our method uniquely combines the structural rigor of formal proofs with the expressive flexibility of natural language. Evaluated on both Olympiad-level and research-grade mathematical benchmarks, it substantially outperforms LLM-as-judge baselines, achieving marked improvements in both precision and recall for error detection. We also release ArxivMathGradingBench, an open-source benchmark for evaluating proof verification systems.

This work proposes a systematic formalization of all published mathematical knowledge into a machine-verifiable, continuously evolving structured knowledge base, addressing the challenges of scalability and organization. Taking dilatations of categories in categorical algebra as the first case study, the project integrates interactive theorem proving, dependent type theory, and category theory to frame the complete formalization of mathematics as a universal reasoning benchmark. By constructing a formal prototype of category dilatations, the study demonstrates the feasibility of this approach in expressing complex algebraic structures, thereby establishing both an architectural foundation and a practical pathway toward a large-scale, interconnected, and extensible database of formalized mathematics.

Current evaluations of mathematical proofs generated by large language models predominantly focus on correctness, often overlooking critical dimensions such as clarity, conciseness, insightfulness, and transferability. This work introduces ProofRank, a novel benchmark that systematically defines and quantifies five scalable quality metrics: conciseness, computational simplicity, cognitive simplicity, diversity, and adaptability. Leveraging a dataset derived from mathematical competition problems, the study conducts a comprehensive evaluation of mainstream models using automated proxy metrics. The results reveal substantial variation in proof quality across models and demonstrate that the most correct proofs are not necessarily the highest-quality ones, highlighting a trade-off between correctness and holistic proof quality. These findings underscore the need for future evaluation frameworks to balance practical utility with formal correctness.

This work addresses the absence of high-quality, contamination-free formal benchmarks for research-level mathematical problems, which has hindered accurate evaluation of automated reasoning systems in genuine mathematical discovery. The authors introduce the first continuously evolving Lean 4 formalized mathematics benchmark, comprising 2,615 problems—including 1,029 open conjectures—designed to enable clean evaluation and foster human–AI collaborative verification. By integrating community collaboration, a standardized evaluation subset, and an AI-driven proof auditing pipeline, the benchmark supports rigorous, reproducible assessment. It has already facilitated multiple novel mathematical discoveries, including resolutions of several open conjectures, and provides reproducible baseline results that clearly delineate the current frontier of automated reasoning in research-level mathematics.