A GEOMETRY PUZZLE FROM AN ERDŐS PROBLEM
More points.
Fewer distances.
Every pair of points has a distance. Place the points so those distances repeat. Each board has a proven minimum. Reach it.
Tap a dot to place or remove a point. +2 means two new distances; 0 is free.
BOARD 01
Find the pattern
Your distances
0 pairsOne bar per squared length; taller bars are shared by more pairs. Hover or tap a bar to see its pairs and why they are equal.
Place two points to make your first distance.
How search does on this board
2,000 tries per runA small hint
THE PROBLEM BEHIND THE PUZZLE
Why is the grid stretched?
Points on a square grid share distances very well. Too well: the corners of a square have only two distances between them, a side and a diagonal.
Erdős asked for point sets that keep the total number of distances low while every four points still determine at least three distances. A square grid fails that rule immediately.
Stretch the rows by √2 and every square disappears, yet distances still repeat almost as often. The lattice was already known from work of Moree and Osburn and of Lund and Sheffer. In January 2026 Benjamin Grayzel, working with Gemini, ruled out the last possible exception, which settled Erdős problem #659. Try other stretches and see which ones break.
What does “proven minimum” mean?
For each board and each number of points, an exhaustive branch-and-bound search has checked that no arrangement uses fewer distances. Distance sets only grow as points are added, so a partial arrangement that is already over the limit can be discarded with all its completions. That makes the search small enough to finish.
20 points on the 7 × 7 board need 17 distances. There are 28,277,527,346,376 ways to choose the points. The search rules out 16 or fewer after about 1.7 million nodes, in seconds. A second implementation in C agrees. Earlier in this project the same board was probed with local search and half a million neighbour checks without settling it.
These are minima for subsets of one finite stretched grid. They say nothing about arbitrary points in the plane, and they are not a new theorem.
How is the score exact?
The point labelled (x, y) sits at (x, √2 y). The squared distance between two grid points is dx² + 2dy², always a whole number, so two lengths are equal exactly when two integers are equal. No floating-point tolerance is involved.
Any selection on this grid satisfies the four-point rule; both Python checkers in the repository also enumerate every four-point subset of each full board.
Enter coordinates or verify an export
An export is a certificate for the repository’s exact verifier. Save the file and run:
python3 -m few_distance verify witness.jsonExport needs at least four points. Nothing leaves your browser: no account, server or analytics. Board links store the coordinates in the URL.
Found a pattern worth discussing? Open an issue with a board link ↗
Read further
Grayzel, “Solution to a Problem of Erdős Concerning Distances and Points” (2026) completes the answer to Erdős problem #659. Moree and Osburn count the distances in this lattice. That mathematics is theirs; this project only explores small finite boards.
For the opposite objective, try Oscar Cunningham’s Unique Distancing Problem: make every distance different.