How do grid points from an abstract number system end up as many pairs exactly one unit apart on a flat plane?

This explores the bridge in the recent unit distance breakthrough: how points that start life as a lattice inside an abstract number system become points in the ordinary plane with an unusually large number of pairs exactly one unit apart. The collection covers this result in two notes, so it can show the outline of the mechanism but not a line-by-line proof. The background is a question Erdős asked: among n points in the plane, how many pairs can sit at distance exactly 1? His own answer used the square grid. Pick a distance that can be written as a sum of two squares in many different ways, then rescale so that distance becomes 1. That gives slightly more than n pairs, and the excess shrinks so slowly that Erdős conjectured nothing could do better by a fixed power of n. What made OpenAI's unit distance counterexample succeed? describes how OpenAI's counterexample broke that ceiling.

The idea behind the bridge, stated loosely, is that a number field (the rational numbers extended with extra algebraic numbers) has a ring of integers that forms a lattice. In fields with the right symmetry (the 'CM fields' named in How many unit distances can points in a plane have?), you can send each element to a point in the complex plane, which is the ordinary flat plane. Distance there is the complex absolute value. So two lattice points end up exactly one unit apart whenever their difference lands at absolute value 1. The problem then becomes algebraic: build a field with many elements of size exactly 1, while keeping the lattice points packed tightly enough that n points capture many such pairs.

The notes identify the key new ingredient: letting the degree of the field grow without limit. A higher degree means more dimensions in the lattice, and so more room for elements of size 1. The usual cost is that the field's 'size' measures (the class number and the discriminant) blow up and destroy the count. Golod–Shafarevich towers are a classical tool from the 1960s that produce infinite chains of fields in which these measures stay under control. In the construction, a fixed prime that splits in the field keeps them in check as the degree rises What made OpenAI's unit distance counterexample succeed?. None of the tools were new. What was new was turning the degree up, a setting nobody had pushed. The follow-up construction writes out the exponent the original left unstated: more than n^1.014 unit distance pairs, stated as 1.014114/C for an absolute constant C How many unit distances can points in a plane have?. The power looks small, but Erdős's conjecture said no fixed power above 1 was possible.

The surprising part is how the result came about. The breakthrough is a new combination of old machinery found by an AI lab, and a second group then made it explicit and checkable. This matches Tao's argument that opaque machine tools can contribute rigorous mathematics when their output passes through reliable validation, such as human re-derivation or proof checkers Can opaque machine learning models help prove new mathematics?. It also connects to the point that formalizing one theorem means rebuilding a whole web of definitions behind it Can autoformalization work on individual statements alone?, and class field towers are exactly that kind of deep stack. If you want the real mechanics (why CM fields specifically, and how the counting argument converts lattice density into pair counts), the collection doesn't yet go that far. Its two notes are the doorway, and the papers behind them are the next step.