Longest Math Proof Ever: 15,000 Pages to 2 Petabytes
Quick answer: The longest proof written by humans is the classification of finite simple groups, commonly estimated at about 15,000 journal pages across more than 500 articles by around 100 mathematicians; its last major gap closed in 2004. Computer-generated proofs are far larger: almost 200 terabytes for the Boolean Pythagorean triples problem (2016) and about 2 petabytes for Schur number five (2017). On October 26, 2026, the company Axiom announced a Lean formalization of the classification, about 43.1 million lines of code checked by Lean’s kernel.
Facts checked November 22, 2026.
Most proofs you meet in school fit on a page. Euclid’s proof that there are infinitely many primes takes about five lines. So why do some proofs run to thousands of pages, or to petabytes of computer output? Here are the longest proofs in mathematics, why they got so long, and how mathematicians (and now AI) check them.
What Counts as the Longest Proof?
There is no single way to measure a proof. Human proofs are measured in journal pages, computer proofs in terabytes of recorded steps, and formal proofs in lines of code for a proof assistant. These sizes are not directly comparable: a page of mathematics carries ideas, a SAT certificate records millions of tiny logical deductions, and Lean code spells out every definition. So the honest answer has three parts:
- Longest human proof: the classification of finite simple groups, about 15,000 pages.
- Largest computer proof: Schur number five, about 2 petabytes (2017).
- Largest formal proof: the 2026 Lean formalization of the classification, about 43.1 million lines.
The Longest Proofs at a Glance
| Proof | Year completed | Size | Type |
|---|---|---|---|
| Classification of finite simple groups | 2004 | 10,000+ pages (about 15,000 by common estimates) | Human, about 100 authors |
| Geometric Langlands conjecture | 2024 | 800 to 1,000 pages in five papers | Human, nine authors |
| Robertson–Seymour graph minor theorem | 1983–2004 | About 500 pages in about 20 papers | Human |
| Feit–Thompson odd order theorem | 1963 | 255 pages (one paper) | Human |
| Fermat’s Last Theorem (Wiles, Taylor–Wiles) | 1995 | About 130 journal pages in two papers | Human |
| Four color theorem | 1976 | Hundreds of pages plus a computer case check | Computer-assisted |
| Kepler conjecture | 1998 (formal proof 2014) | Hundreds of pages plus gigabytes of computation | Computer-assisted |
| Boolean Pythagorean triples problem | 2016 | 200 terabytes | Computer (SAT solver) |
| Schur number five equals 160 | 2017 | About 2 petabytes | Computer (SAT solver) |
| Lean formalization of the classification | 2026 | About 43.1 million lines of code | Formal, AI agents with human guidance |
Page counts for multi-paper proofs are estimates; they depend on which papers you count.
The Classification of Finite Simple Groups: 15,000 Pages
A group is a collection of operations that can be combined by precise rules, like the rotations of a square: doing one rotation and then another gives another rotation. A simple group is one that cannot be broken down into smaller pieces in the technical sense (it has no nontrivial proper normal subgroup). As an analogy, finite simple groups play the role for groups that prime numbers play for whole numbers. The classification, nicknamed the Enormous Theorem, says every finite simple group is one of these:
- cyclic groups of prime order (one family),
- alternating groups \(A_n\) for \(n \ge 5\) (one family),
- groups of Lie type over finite fields (16 families),
- or one of exactly 26 sporadic groups that fit no pattern.
The largest sporadic group, the Monster, has about \(8.08 \times 10^{53}\) elements. Proving that the list is complete took more than 500 journal articles by around 100 mathematicians over about half a century. The final gap, the “quasithin” case, was closed in 2004 by Michael Aschbacher and Stephen Smith in a 1,221-page two-volume work. Because no single person could check it all, Daniel Gorenstein, Richard Lyons and Ronald Solomon began a “second-generation” proof that streamlines the argument into a multi-volume series that is still being published.
The 2026 Breakthrough: 43 Million Lines of Lean
On October 26, 2026, the AI company Axiom announced “a complete formalization in Lean of the first-generation proof” of the classification, with every logical step checked by Lean’s kernel. According to Axiom, the project has about 43.1 million lines of Lean, 2.05 million declarations and material from 715 papers and books, and took 24 days. The team split the theorem into about 100 milestones and had many AI agents, guided by mathematicians, formalize different parts. Its public repository describes the proof as having no unfinished steps (“sorry-free”) and using only Lean’s standard axioms.
What Errors Did It Find?
Formalizing forces every step to be spelled out. Axiom reported an issue in a published argument for the Unbalanced Group Theorem, one of the intermediate results, and proved it by an alternative route. St Andrews group theorist Colva Roney-Dougal, who was not involved, told IEEE Spectrum she was “really, really surprised” and that the formal proof clears up several mistakes in the human literature. None of this means the classification was wrong; it means some published steps needed repair.
Has Anyone Independently Checked It?
Lean’s kernel checks that each step follows logically. Whether the formal statements faithfully match the theorem mathematicians mean is a separate question, and as of November 22, 2026, no named outside team had published a full independent audit. The figures above are Axiom’s own. Nine days later, IEEE Spectrum reports, Axiom also announced a roughly 16-million-line formalization of the Stacks Project, a huge algebraic geometry reference. For how that kind of checking works, see can we trust AI math proofs? and the Lean theorem prover explained.
How Long Would It Take You to Read?
Pick a proof and a reading speed. Serious mathematics is slow: a careful reader may manage only a few pages a day.
Computer Proofs: Terabytes and Petabytes
Some proofs are long because a computer checks a huge number of cases.
- Four color theorem (1976). Kenneth Appel and Wolfgang Haken reduced the problem to checking about 1,900 configurations by computer, the first major theorem proved this way. A 1996 proof by Robertson, Sanders, Seymour and Thomas cut this to 633, and Georges Gonthier verified it in the Coq proof assistant in 2005. See the four color theorem explained.
- Kepler conjecture (1998). Thomas Hales proved that the grocer’s pyramid is the densest way to stack oranges, using hundreds of pages and extensive computation. Referees said they were “99% certain,” so Hales led the Flyspeck project, which completed a formal proof in 2014.
- Boolean Pythagorean triples (2016). Can you color every whole number red or blue so that no Pythagorean triple such as \(3^2 + 4^2 = 5^2\) is all one color? Marijn Heule, Oliver Kullmann and Victor Marek showed it works up to 7,824 and is impossible from 7,825 on. The SAT-solver proof was almost 200 terabytes. In September 2026, two short preprints (one credited to an AI system) gave new proofs of the infinite statement; see the 200-terabyte math proof.
- Schur number five (2017). Heule showed that the numbers 1 to 160 can be split into five groups with no group containing \(x\), \(y\) and \(x + y\), but 1 to 161 cannot. The proof, published in 2018, was about 2 petabytes.
How Do Computers Check Enormous Proofs?
| Tool | Its job | Example |
|---|---|---|
| SAT solver | Searches an enormous space of cases and records each deduction | Schur number five |
| Certificate checker | A small program that rechecks every recorded step | Boolean Pythagorean triples |
| Proof assistant | Checks formal proofs from definitions and axioms | Lean, Coq (Rocq), HOL Light |
| AI agent | Writes or helps write the formal proof code | Axiom’s 2026 formalization |
Nobody reads these proofs line by line. Trust rests on a small checker that is easy to inspect, not on the fast solver or AI that found the proof.
Why Do Some Proofs Get So Long?
- Many cases. Classification results must rule out every possibility, one family at a time.
- New machinery. Wiles’s proof of Fermat’s Last Theorem rests on decades of theory about elliptic curves and modular forms, so the “real” length is far more than its 130 pages. See Fermat’s Last Theorem explained.
- Brute force. A SAT solver may need trillions of steps to rule out every coloring, even when the statement fits on one line.
- Short proofs can be hard to fill in. Grigori Perelman’s proof of the Poincaré conjecture (2002–2003) was three preprints, but complete expositions by other mathematicians run to several hundred pages.
Long Proofs That Are Still Disputed
Length can hide errors. Shinichi Mochizuki’s proposed proof of the abc conjecture, about 600 pages of “inter-universal Teichmüller theory,” was published in 2021, but Peter Scholze and Jakob Stix argued in 2018 that a key step fails, and most number theorists do not accept it as proved. Formal verification is one way such disputes could finally be settled.
Common Misconceptions
“A longer proof is a better proof.” No. Mathematicians prize short, clear proofs; length is usually a cost, not a virtue.
“Computer proofs are not real proofs.” They are real, but their trust rests on checked software. Formal verification in Lean or Coq makes that trust explicit.
“Finding an error in a long proof means the theorem is false.” No. The 2026 formalization found problems in intermediate arguments, but repaired them; the classification itself stands.
“Nobody understands the classification.” Experts understand its strategy well; what no single person has done is check every page, which is exactly what the second-generation proof and the 2026 Lean formalization address.
Watch: The Biggest Proofs in Math
If you learn better by watching, these videos are worth your time.
Check Your Understanding
Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.
Answer each question and press Submit to see your result right away.
Frequently Asked Questions
What is the longest math proof ever written?
The classification of finite simple groups, often called the Enormous Theorem, is the longest proof written by humans: well over 10,000 journal pages across roughly 500 articles, completed in 2004.
What is the longest computer proof?
The proof that Schur number five equals 160, found by Marijn Heule in 2017, is about 2 petabytes. The 2016 Boolean Pythagorean triples proof is almost 200 terabytes.
Can a 15,000-page proof be compared with a 2-petabyte proof?
Not directly. Pages measure human-written explanation, while petabytes measure machine-recorded logical steps, and Lean lines measure formal code.
Has the classification of finite simple groups been checked by computer?
Yes, according to Axiom, which on October 26, 2026 announced a Lean formalization of the first-generation proof with about 43.1 million lines checked by Lean’s kernel. No full independent outside audit had been published as of November 22, 2026.
How long is the proof of Fermat’s Last Theorem?
Andrew Wiles’s 1995 paper and his follow-up with Richard Taylor fill about 130 journal pages, but they rely on decades of earlier theory.
Why are some math proofs so long?
Some must check many cases, some build new theory, and computer proofs may test trillions of possibilities.
Keep Exploring
- Can We Trust AI Math Proofs? How Lean Verifies Them
- Lean Theorem Prover: Lean 4 and Mathlib Explained
- Four Color Theorem: What the 2026 Proof Changes
Sources and Further Reading
- Axiom: The Classification of Finite Simple Groups, formalized in Lean
- Heule, Kullmann and Marek: Solving and Verifying the Boolean Pythagorean Triples Problem (arXiv, 2016)
- IEEE Spectrum: Startup Makes Headway Towards Formalizing All of Math With AI (Nov 18, 2026)
- Wikipedia: List of long mathematical proofs
- Plus Magazine: The enormous theorem
Related to This Article
More math articles
- How to Use Area Models and the Distributive Property to Multiply?
- CHSPE Math Formulas
- Rounding Numbers Practice — Free Interactive Drills
- How to Find the Domain and Range of Quadratic Functions
- How to Find Scalar Multiplication of Vectors?
- FTCE Math Flashcards (Free Online: Formulas, Terms & Concepts)
- Miranda rights
- The Digestive System
- How to Understand Random Sampling and Variation in Samples?
- How to Get a PhD in Math


What people say about "Longest Math Proof Ever: 15,000 Pages to 2 Petabytes - Effortless Math"?
No one replied yet.