The 200-Terabyte Math Proof: Why 7,825 Matters
Quick answer: The 200-terabyte math proof solved the Boolean Pythagorean triples problem. In 2016, Marijn Heule, Oliver Kullmann and Victor Marek proved with SAT-solving software that the numbers 1 to 7,824 can be colored red and blue with no one-color Pythagorean triple (like 3, 4, 5), but 1 to 7,825 cannot. So no such coloring of all whole numbers exists. The verified proof was almost 200 terabytes, with a 68-gigabyte certificate. In September 2026, two short human-readable proofs of the infinite statement appeared on arXiv, one reportedly found by an AI system.
Facts checked November 22, 2026.
In 2016, Nature called it the largest math proof ever: almost 200 terabytes, enough to fill about 200 laptop hard drives. It answers a question you can explain to a 10-year-old, about coloring numbers red and blue. Here is the problem, why 7,825 is the magic number, how a computer solved it, how anyone can trust a proof no human can read, the surprising 2026 sequel, and a puzzle where you can try it yourself.
What Is the Boolean Pythagorean Triples Problem?
Can you color every positive whole number either red or blue so that no Pythagorean triple \(a^2 + b^2 = c^2\) has all three numbers the same color?
For example, \(3^2 + 4^2 = 5^2\), so 3, 4 and 5 cannot all be red, and cannot all be blue. The same goes for 5, 12, 13 and for 6, 8, 10. The trouble is that numbers belong to many triples at once: 5 appears in both \((3, 4, 5)\) and \((5, 12, 13)\), so choices interact. Ronald Graham, a leading combinatorialist, asked the question in the 1980s and offered a 100-dollar prize for the answer.
The Answer at a Glance
| Fact | Detail |
|---|---|
| Result | \(\{1, \ldots, 7824\}\) can be colored with no one-color triple; \(\{1, \ldots, 7825\}\) cannot |
| So the full answer | No: the whole numbers cannot be colored this way |
| Researchers | Marijn Heule (UT Austin), Oliver Kullmann (Swansea University), Victor Marek (University of Kentucky) |
| Announced | arXiv, May 3, 2016; best paper at the SAT 2016 conference |
| Computer | About 2 days on 800 cores of the Stampede supercomputer (Texas Advanced Computing Center), about 4 CPU-years |
| Proof size | Almost 200 terabytes (DRAT format), compressed to a 68-gigabyte certificate |
| Prize | Graham paid the 100 dollars to Heule in May 2016 |
Why Is 7,825 the Magic Number?
7,825 is the smallest \(N\) for which every red-and-blue coloring of 1 to \(N\) contains a one-color Pythagorean triple. The result has two halves:
- 7,824 works: the researchers found explicit colorings of 1 to 7,824 with no one-color triple.
- 7,825 fails: they proved that no coloring of 1 to 7,825 works. Adding 7,825 adds new triples, such as \(5180^2 + 5865^2 = 7825^2\), and with them no valid coloring survives.
Because any coloring of all whole numbers would also color 1 to 7,825, the finite result settles the infinite question: the answer to Graham’s problem is no.
Among the numbers 1 to 7,825 there are 9,472 Pythagorean triples. That count includes primitive triples like \((3, 4, 5)\) and multiples like \((6, 8, 10)\); the multiples are not redundant, because 3 and 6 can have different colors. There are \(2^{7825} \approx 3.63 \times 10^{2355}\) ways to color the numbers, a number with 2,356 digits, far too many to check one by one.
Try It: Color the Numbers
Click a number to switch it between red and blue. The puzzle counts how many Pythagorean triples are all one color. With small ranges it is easy to reach zero; the 2016 proof shows that once the range reaches 7,825, zero becomes impossible.
Some numbers, such as 1 and 2, belong to no Pythagorean triple at all, so their colors never matter. Others join the puzzle only as the range grows: 7 first matters at 25, because \(7^2 + 24^2 = 25^2\). The “find” button uses a quick random search, which easily succeeds for small ranges; it proves nothing about 7,825, where no valid coloring exists.
How the Computer Solved It
The team turned the question into a giant logic puzzle, a Boolean satisfiability (SAT) problem. Each number \(n\) gets a true-or-false variable \(x_n\) (true for blue, false for red). Each triple gives two rules, called clauses. For \((3, 4, 5)\):
\[ (x_3 \lor x_4 \lor x_5) \quad \text{and} \quad (\lnot x_3 \lor \lnot x_4 \lor \lnot x_5). \]The first says “at least one of 3, 4, 5 is blue,” the second “at least one is red.” Together they forbid a one-color triple. With 9,472 triples there are 18,944 clauses. For 1 to 7,824 the formula is satisfiable (SAT); for 1 to 7,825 it is unsatisfiable (UNSAT).
- Shrinking the search. Symmetry and number theory cut the \(10^{2355}\) colorings down to just under a trillion cases. For example, swapping every red and blue gives an equivalent coloring, which halves the work immediately.
- Cube-and-conquer. A “look-ahead” solver split the problem into a huge number of sub-problems (cubes), and fast “conflict-driven” solvers then worked on them in parallel across 800 cores.
- Recording every step. To show that no coloring of 1 to 7,825 works, the solvers logged every deduction in a format called DRAT. That log is the almost 200-terabyte proof.
How Was the 200-Terabyte Proof Verified?
No human will ever read 200 terabytes. Instead, the proof was verified by a separate, much simpler proof-checking program. The idea is that it is far easier to write a small checker you trust than a fast solver you trust. The 68-gigabyte certificate is not a zipped copy of the proof; it lists the sub-problems, from which anyone with about 30,000 hours of processor time can rebuild and recheck the full proof. In follow-up work published in 2017 and 2019, researchers including Luís Cruz-Filipe and Peter Schneider-Kamp checked the result with a verified checker built in the Coq proof assistant, including the link between the coloring problem and its SAT encoding. That is the same idea behind proof assistants like Lean.
The deeper complaint is different. As Nature put it, the computer showed that the coloring is impossible but did not give an underlying reason. Mathematicians still hope for a human explanation of why 7,825 is where it breaks.
2026: From 200 Terabytes to 7 Pages
The story took a twist in September 2026. Two preprints on arXiv gave new, short proofs that every 2-coloring of the positive integers contains a one-color Pythagorean triple:
- James Leng, “A new proof of two-color partition regularity of Pythagorean triples” (September 28, 2026). The abstract says the proof “was found by ChatGPT Astra,” an AI system.
- William J. Wesley, “An elementary proof of the 2-regularity of Pythagorean triples” (September 30, 2026), 7 pages, which says it needs “neither lengthy computation nor difficult analytic machinery.”
Two cautions. These are preprints, not yet peer-reviewed as of November 22, 2026. And their abstracts claim the infinite statement, not the exact threshold: the 2016 computer proof is still what pins the answer at 7,825. If they hold up, they answer the old complaint that the computer showed that the coloring fails without showing why.
Bigger Proofs Since 2016
| Year | Result | Proof size |
|---|---|---|
| 2014 | Erdős discrepancy problem, case \(C = 2\) (Konev and Lisitsa) | About 13 gigabytes |
| 2016 | Boolean Pythagorean triples | Almost 200 terabytes |
| 2017 | Schur number five equals 160 (Heule; published 2018) | About 2 petabytes |
| 2020 | Keller’s conjecture in dimension 7 (Brakensiek, Heule, Mackey and Narváez) | SAT proof, independently checked |
| 2023 | Packing chromatic number of the square grid is 15 (Subercaseaux and Heule) | SAT proof |
For the human side of the record books, including the 15,000-page classification of finite simple groups, see the longest math proofs ever.
Why It Matters for AI and Math
The 2016 proof used specialized automated reasoning, not today’s language models, but it shows a split that now runs through AI mathematics: finding a proof and verifying it are different jobs. SAT solvers searched a vast space and left a checkable record. In 2026, a language-model system is credited with finding a short human-readable proof of the same theorem. Either way, trust comes from checking: a small kernel in Lean, a verified SAT checker, or careful peer review. See can we trust AI math proofs?
Common Misconceptions
“The proof is 200 terabytes because the answer is complicated.” No. The answer is one number, 7,825. The size comes from recording every step that rules out all the colorings.
“Nobody checked it.” It was checked by an independent proof-checking program, and the compressed certificate is public.
“It proves something about all numbers by checking infinitely many.” It only needed 1 to 7,825: if those cannot be colored, no larger set can either.
“The computer tried every coloring.” No. Logical deductions ruled out huge groups of colorings at once; checking \(10^{2355}\) colorings one by one would be impossible.
Watch: The 200-Terabyte Proof
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 200-terabyte math proof?
It is the 2016 computer proof by Heule, Kullmann and Marek that the numbers 1 to 7,825 cannot be colored red and blue without creating a one-color Pythagorean triple.
What is the Boolean Pythagorean triples problem?
It asks whether the positive integers can be split into two colors so that no triple with a squared plus b squared equal to c squared is all one color. The answer is no.
Why is the number 7,825 important?
The numbers 1 to 7,824 can be colored with no one-color Pythagorean triple, but 1 to 7,825 cannot.
How was the 200-terabyte proof verified?
A separate proof-checking program verified the DRAT proof, and a 68-gigabyte certificate lets anyone rebuild and recheck it.
Is the 200-terabyte proof the largest proof ever?
It was called the largest in 2016. Heule’s 2017 proof that Schur number five is 160 is about 2 petabytes, ten times larger.
Is there a short proof of the Boolean Pythagorean triples problem?
In September 2026, James Leng and William Wesley posted short preprints proving that every 2-coloring has a one-color Pythagorean triple; Leng credits ChatGPT Astra with finding his proof. They were not yet peer-reviewed as of November 2026.
Keep Exploring
- Longest Math Proof Ever: 15,000 Pages to 2 Petabytes
- Can We Trust AI Math Proofs? How Lean Verifies Them
- Lean Theorem Prover: Lean 4 and Mathlib Explained
Sources and Further Reading
- Heule, Kullmann and Marek: Solving and Verifying the Boolean Pythagorean Triples Problem via Cube-and-Conquer (arXiv, May 3, 2016)
- UT Austin Computer Science: Marijn Heule uses Texas Stampede supercomputer to crack 200-terabyte math proof
- Wikipedia: Boolean Pythagorean triples problem
- Wikipedia: Marijn Heule
- James Leng: A new proof of two-color partition regularity of Pythagorean triples (arXiv, Sept 28, 2026)
- William J. Wesley: An elementary proof of the 2-regularity of Pythagorean triples (arXiv, Sept 30, 2026)
Related to This Article
More math articles
- The Ultimate HiSET Math Formula Cheat Sheet
- Demographic influences
- Input-price changes cause substitution
- 7th Grade Math Worksheets for West Virginia (WVGSA): 95 Free Printable PDFs
- 5th Grade NYSE Math FREE Sample Practice Questions
- The Muscular System
- 7th Grade WVGSA Math Worksheets: FREE & Printable
- Self-incrimination
- Presidential Communication: The Bully Pulpit and Its Limits
- How to Solve Multi-Step Word Problems


What people say about "The 200-Terabyte Math Proof: Why 7,825 Matters - Effortless Math"?
No one replied yet.