The Lean Theorem Prover (System Description)
By Leonardo de Moura, Soonho Kong, Jeremy Avigad, Floris van Doorn, Jakob von Raumer. 2015
733 citations recorded by the source corpus.
Lean is a new open source theorem prover being developed at Microsoft Research and Carnegie Mellon University, with a small trusted kernel based on dependent type theory. It aims to bridge the gap between interactive and automated theorem proving, by situating automated tools and methods in a framework that supports user interaction and the construction of fully specified axiomatic proofs. Lean is an ongoing and long-term effort, but it already provides many useful components, integrated development environments, and a rich API which can be used to embed it into other systems. It is currently being used to formalize category theory, homotopy type theory, and abstract algebra. We describe the project goals, system architecture, and main features, and we discuss applications and continuing work.
Subfields: Artificial Intelligence, Program Analysis & Verification
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Mohit Bansal
- Yejin Choi
- Percy Liang
- Luke S. Zettlemoyer
- Hannaneh Hajishirzi
- Jianfeng Gao
- Yoshua Bengio
- Kyunghyun Cho
In Program Analysis & Verification
- Sean Welleck
- Markus Norman Rabe
- Yuhuai Wu
- Rahul Saha
- Christian Szegedy
- Stanislas Polu
- Kshitij Bansal
- Zhangir Azerbayev
Citations in the corpus
- Solving Quantitative Reasoning Problems with Language Models (2022, methodology): “This can be achieved using specialized programming languages that facilitate the simulation of logical and mathematical thinking using a computer, such as Coq [20], Isabelle [21], HOL4 [22], Lean [23], Metamath [24] and Mizar [25].”
- Llemma: An Open Language Model For Mathematics (2023, background): “Interactive proof assistants such as Lean (de Moura et al., 2015), Isabelle (Wenzel et al., 2008), and Coq (Paulin-Mohring, 1989a;b) express mathematics in programming languages that allow for verification.”
- Formal Mathematics Statement Curriculum Learning (2022, methodology): “In this work, we target the miniF2F (Zheng et al., 2021) benchmark, which consists of 244 validation and 244 test formalized statements of mathematical problems from various competitions.”
- Autoformalization with Large Language Models (2022, methodology): “These projects relied on interactive theorem provers, such as Isabelle [50], Coq [11], HOL Light [22], and Lean [12], which are essentially programming languages that enable users to enter their statements and proofs in a formal language, and which can then be checked automatically for correctness.”
- INT: An Inequality Benchmark for Evaluating Generalization in Theorem Proving (2020, methodology): “In Figure 1 we illustrate proofs of the same statement with LEAN [6] and INT assistant.”
- LeanDojo: Theorem Proving with Retrieval-Augmented Language Models (2023, methodology): “The dataset consists of formal definitions, theorems, and proofs written in Lean [1].”
- HyperTree Proof Search for Neural Theorem Proving (2022, positive): “An example of a proof in Lean [4], an interactive theorem prover, is given in Figure 1 and the corresponding proof hypertree3 is illustrated in Figure 5 of the Appendix.”
- ProofNet: Autoformalizing and Formally Proving Undergraduate-Level Mathematics (2023, background): “Formal statements are expressed in the Lean 3 theorem prover [de Moura et al., 2015], and depend on Lean’s mathlib [mathlib Community, 2020].”
- A Survey of Deep Learning for Mathematical Reasoning (2022, neutral): “For Lean, LeanStep (Han et al., 2022) provides a dataset of proof-steps from Lean’s mathematical library along with auxiliary tasks, while Lean-Gym (Polu et al., 2023) provides an interactive REPL.”
- Developing Bug-Free Machine Learning Systems With Formal Mathematics (2017, methodology): “…proof assistant (Gordon, 1979; Gordon & Melham, 1993; Harrison, 1996; Nipkow et al., 2002; Owre et al., 1992; Coq Development Team, 2015-2016; de Moura et al., 2015), which consists of (a) a programming language, (b) a language to state mathematical theorems, and (c) a set of tools for…”
- NaturalProofs: Mathematical Theorem Proving in Natural Language (2021, background): “[47]), in which mathematics is written in a verifiable formal language that resembles source code, such as Mizar [41], Lean [7], or Metamath [26].”
- Logic-LM: Empowering Large Language Models with Symbolic Solvers for Faithful Logical Reasoning (2023, methodology): “These works demonstrate the proficiency of LLMs in translating a considerable fraction of mathematical problems into formal specifications defined in tools like SymPy (Meurer et al., 2017), Isabelle/HOL (Paulson, 1994), and Lean (de Moura et al., 2015).”
- LIME: Learning Inductive Bias for Primitives of Mathematical Reasoning (2021, methodology): “We demonstrate that LIME pretrained models provide significant gains across four large mathematical reasoning benchmarks: IsarStep (Li et al., 2021), HOList Skip-tree (Rabe et al., 2021), MetaMathStep (Polu & Sutskever, 2020), and LeanStep (de Moura et al., 2015).”
- Draft, Sketch, and Prove: Guiding Formal Theorem Provers with Informal Proofs (2022, background): “Interactive theorem proving Modern verification systems for mathematics are centered around interactive theorem provers (ITPs), such as Isabelle (Paulson, 1994), Lean (Moura et al., 2015), Coq (Barras et al.”
- Proof Artifact Co-training for Theorem Proving with Language Models (2021, background): “L EAN Lean is an interactive theorem prover and functional programming language (de Moura et al., 2015).”
- LLMSTEP: LLM proofstep suggestions in Lean (2023, methodology): “Interactive proof assistants such as Lean [5], Isabelle [17], and Coq [14] enable the verification of mathematics and software using specialized programming languages [2, 12].”
- HOList: An Environment for Machine Learning of Higher-Order Theorem Proving (extended version) (2019, methodology): “We are not aware of a comprehensive coverage of fundamental mathematics in Isabelle, HOL4, or Lean.”