Table of Contents
AI Cross-Disciplinary Math: Number Theory Solved Geometry
AI cross-disciplinary math in action: a model used algebraic number theory to crack a geometry problem. What class field towers are, in words a kid can follow.
The most interesting thing about AI cross-disciplinary math right now is a single sentence from OpenAI’s announcement: “These ideas were well-known to algebraic number theorists, but it came as a great surprise that these concepts have implications for geometric questions in the Euclidean plane.”
The problem was about dots on a flat page. The solution came from a branch of mathematics that studies how numbers factor in exotic number systems. Nobody had connected those two things in eighty years. That connection, not the fact that a machine found it, is the part worth explaining to your kid.
Key Takeaways
- The unit distance problem asks how many pairs of points among n points in a plane can be exactly distance 1 apart. Erdős posed it in 1946, and the 2005 reference Research Problems in Discrete Geometry calls it the simplest-to-explain problem in the field.
- Erdős’s own best construction can be understood through the Gaussian integers, numbers of the form a + bi.
- The AI’s proof replaced Gaussian integers with more complicated number fields “with richer symmetries that can create many more unit-length differences.”
- The tools were infinite class field towers and Golod–Shafarevich theory, used to prove the required number fields actually exist.
- Thomas Bloom’s read, reported by Quanta: the result shows “there is a lot more that number theoretic constructions have to say about these sorts of questions than we suspected.”
Why AI cross-disciplinary math surprised the experts
Here is the chain, and I will keep every step checkable.
Start with the geometry. If you want many pairs of points exactly one unit apart, you want a point pattern where the same distance shows up over and over. A square grid does this: most points have four neighbors at distance exactly 1.
Now the number theory. Erdős’s original construction can be described with the Gaussian integers: numbers of the form a + bi where a and b are ordinary integers and i is the square root of −1. Plot a + bi as the point (a, b) and you have a grid. The Gaussian integers behave a lot like ordinary integers, including having unique factorization into primes. That structure is what makes the counting work.
Then the leap. As OpenAI describes it: “The new argument replaces the Gaussian integers by more complicated generalizations from algebraic number theory with richer symmetries that can create many more unit-length differences.”
Read that again, because it is the whole idea. The Gaussian integers give you some symmetry. Richer number systems give you more symmetry. More symmetry means more pairs at the same distance. And more pairs is exactly what the problem was asking for.
The last piece is the hardest and the most technical: you have to prove that the number systems you need actually exist. That is where infinite class field towers and Golod–Shafarevich theory come in. Both are established tools in algebraic number theory; neither had been pointed at discrete geometry.
Arul Shankar, a number theorist who reviewed the result, noted something about how the model got there: “a significant majority of the thoughts are trying to construct a counterexample to the widely believed upper bound, rather than trying to prove it.” It went looking for a break, not a confirmation.
Fields and the tools they lend each other
Borrowing between fields is not new. It is how much of mathematics advances. Here are established examples alongside the 2026 one.
| Problem field | Tools borrowed from | What the borrowing bought |
|---|---|---|
| Discrete geometry (unit distances) | Algebraic number theory (class field towers, Golod–Shafarevich) | Number systems with enough symmetry to beat the grid construction, 2026 |
| Number theory (Fermat’s Last Theorem) | Algebraic geometry and modular forms | Wiles’s 1995 proof, via the modularity of elliptic curves |
| Combinatorics (graph properties) | Linear algebra (eigenvalues) | Spectral graph theory: matrix properties reveal network structure |
| Topology (knot classification) | Statistical physics | Jones polynomial and its physical interpretations |
| Analysis and PDEs | Probability theory | Stochastic methods for solving deterministic equations |
| Cryptography | Number theory (prime factorization) | RSA encryption and most of internet security |
| Condensed mathematics | Formal verification (Lean) | The Liquid Tensor Experiment, completed July 2022 |
The pattern is consistent: progress often comes from someone noticing that machinery built for one purpose describes something else. What is new in 2026 is that the noticing was done by a model with no career incentive, no subfield loyalty, and no sense that geometry people do not read number theory papers.
That last point may be the real advantage. Human mathematicians specialize, attend conferences in their area, and read journals in their field. A model trained on everything has no such boundaries. It is plausible that cross-field connections are exactly where machine search has an edge, though I have not seen that hypothesis tested rigorously.
How to Teach Your Kid About Borrowing Between Fields
The concept is that a tool built for one job can solve a different job, and noticing that is a skill.
Ages 5–8: The Wrong-Tool Game
Set out a few objects: a ruler, a spoon, a rubber band, a paper clip. Name a problem that none of them was designed for, like “get this small toy out from under the couch” or “keep these papers from blowing away.”
Let them pick and try. When they use the ruler as a reaching stick rather than a measuring device, name what just happened: “You used a measuring tool as a grabbing tool. Mathematicians do that with ideas.”
Ages 9–12: Same Shape, Different Subject
Show them that a pattern can show up in unrelated places. The doubling pattern, 1, 2, 4, 8, 16, describes bacteria dividing, a chain letter spreading, paper folding thickness, and compound interest. Have them list four situations that follow it.
Then the real question: if you know how the bacteria story ends, what does that tell you about the chain letter? That is borrowing. The math is the same; only the labels changed.
Ages 13+: Build a Gaussian Grid
Have your teen plot points a + bi as coordinates (a, b) for all integers a and b from −3 to 3. That is a 7 by 7 grid of 49 points. Now count how many pairs are exactly distance 1 apart. They should find 84: 42 horizontal and 42 vertical.
Then the extension: how many pairs are exactly distance √2 apart? (Diagonals: 72 pairs, if they count carefully.) The point is that which distances repeat depends on the structure of the number system you are plotting. Change the number system and you change which distances repeat and how often.
That is the actual mechanism of the 2026 proof, done with graph paper. You are not going to get to class field towers at a kitchen table, and you do not need to. The idea that richer number systems give more repeated distances is the load-bearing insight.
The question to ask: “What else has this same shape?”
What this means for how kids should study
I want to draw a practical conclusion, because this is where the story becomes useful rather than just interesting.
Specialization has a cost. The reason this connection sat undiscovered for eighty years is not that it was beyond human ability. As Science News reported, the tools were already public. Jacob Tsimerman, who reviewed the proof, said he had “briefly worked on this problem and tried to make a counterexample, but failed to make progress.” The tools existed. The bridge did not, because the people who knew the tools were not thinking about the geometry problem.
Breadth is a real asset. A kid interested in two unrelated things is not unfocused. They are the person most likely to notice a connection. That is not a feel-good claim; it is the structure of what happened here.
But breadth without depth notices nothing. The connection required knowing algebraic number theory well enough to know that class field towers exist and what Golod–Shafarevich theory guarantees. Surface familiarity with two fields produces nothing. Real knowledge of two fields is where the value sits.
Bloom’s assessment of the result captures the follow-on effect: “No doubt many algebraic number theorists will be taking a close look at other open problems in discrete geometry in the coming months.” One bridge makes people look for more bridges, and the Erdős problems database is where that search gets tracked. Our piece on why the Erdős problems are falling to AI covers which problems that search is likely to reach.
What to actually do at home
Ask “what else looks like this?” constantly
It is the single question that builds transfer. Traffic and water flow. Populations and compound interest. Sorting cards and organizing a bookshelf. The habit of looking for the shared structure is the habit that produced this result.
Let two interests coexist without merging them prematurely
If your kid loves both chemistry and music, resist the urge to make it a “chemistry of music” project. Let both go deep. The connection, if it comes, comes from depth in each.
Use the graph paper exercise
The Gaussian grid activity above is the closest a parent can get to the actual mechanism of a 2026 mathematical result, using a pencil. That is unusual and worth the twenty minutes.
Teach that surprise is information
The mathematicians quoted in this story all used some version of the word “surprising.” Noga Alon said the fact that the answer is not n^(1+o(1)) “is surprising.” Surprise means a model of the world was wrong, which is the most useful signal available. Reward it in your kid rather than smoothing it over.
What not to do
Do not use this to argue that specialization is obsolete. The proof required deep specialist knowledge in algebraic number theory, and the human follow-up work (Will Sawin showing δ = 0.014) required more of it. The argument is for depth in more than one place, not breadth instead of depth.
What to Watch For Over the Next 3 Months
- Week 4: Your child spontaneously says “that’s like” about two unrelated things. That is transfer, and it is the whole skill.
- Month 2 red flags: Treating school subjects as sealed containers. If your kid thinks math class and science class have nothing to do with each other, the curriculum is teaching the opposite of this story.
- Month 3 self-check: Watch whether Bloom’s prediction holds and more discrete geometry problems fall to number-theoretic methods. That is a testable forecast, and it is being tested right now. The verification side is in how mathematicians check a 125-page AI proof.
Frequently Asked Questions
What are Gaussian integers, in plain language?
Numbers of the form a + bi, where a and b are ordinary whole numbers and i is the square root of −1. Plotted as points (a, b) they form a square grid, and they share useful properties with ordinary integers, including unique factorization into primes. Erdős’s original unit-distance construction can be understood through them.
What is a class field tower?
It is a construction in algebraic number theory where you build an extension of a number field, then an extension of that, and so on. An infinite class field tower is one that never terminates. The AI’s proof used these, along with Golod–Shafarevich theory, to show the number fields required by its construction actually exist.
Why did the connection surprise mathematicians?
Because the problem looked purely geometric and the tools were purely algebraic, and nobody expected the latter to bear on the former. OpenAI’s own description says “it came as a great surprise that these concepts have implications for geometric questions in the Euclidean plane.”
Is cross-field borrowing unique to AI?
No, it is one of the main ways mathematics advances. Wiles proved Fermat’s Last Theorem using algebraic geometry and modular forms; RSA encryption comes from number theory; spectral graph theory applies linear algebra to combinatorics. What was new here is that a model found the bridge.
Should my kid specialize or stay broad?
Both, in that order. The connection in this story required genuine depth in algebraic number theory, not passing familiarity. The argument is for real knowledge in more than one area, which is different from sampling many areas shallowly.
About the author
Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.
Sources
- OpenAI. (2026, May 20). “An OpenAI model has disproved a central conjecture in discrete geometry.” https://openai.com/index/model-disproves-discrete-geometry-conjecture/
- Kakaes, K. (2026, August 3). “Why the legendary Erdős problems are falling to AI.” Quanta Magazine. https://www.quantamagazine.org/why-the-legendary-erdos-problems-are-falling-to-ai-20260803/
- Hulick, K. (2026, June 8). “AI guardrails and the Erdős math problem.” Science News. https://www.sciencenews.org/article/ai-guardrails-erdos-math-problem
- Brass, P., Moser, W., & Pach, J. (2005). Research Problems in Discrete Geometry. Springer. https://link.springer.com/book/10.1007/0-387-29929-7
- Lean Community. (2022, July 14). “Completion of the Liquid Tensor Experiment.” https://leanprover-community.github.io/blog/posts/lte-final/
- Tao, T. (2025, January). “Machine-Assisted Proof.” Notices of the American Mathematical Society, 72(1), 6–13. https://www.ams.org/notices/202501/rnoti-p6.pdf
- Bloom, T. “Erdős Problems.” University of Manchester. https://www.erdosproblems.com/
