Alien Math · Toys
Three shape puzzles
Paint the plane so no two points exactly 1 apart share a colour. Find one tile that every shape 1 across fits inside. Carve the biggest sofa that turns a corner. Each one is easy to state, and each one is still open. Play with them here, one step at a time.
All three are famous open problems, and nothing here moves any of them forward. The pictures, the exhaustive colouring check, the fitting search and the carving are ours; every bound and record on this page belongs to the people named at the bottom.
Toy 1 · Colours, distance 1
Paint the plane, one rule
Give every point on an endless flat sheet a colour. The one rule: two points exactly 1 apart must never match. How few colours can you get away with? Start small.
Three colours fail on seven points
These seven points sit so that every line drawn is exactly 1 long. Click a point to change its colour. Then let the computer try every way to colour them.
With three colours there are 37 = 2,187 ways to colour the seven points, and every one puts two matching colours at the ends of some line. So the plane needs at least four colours.
Seven colours always work
Tile the floor with hexagons a bit less than 1 across, coloured so a hexagon’s colour only repeats far away. A stick of length 1 can’t have both ends in one hexagon, and it can’t reach the next hexagon of the same colour. Drag the stick around. Then change the tile size and watch it break.
Four was beaten in 2018
Aubrey de Grey found a set of 1,581 points that four colours can’t handle. Others have since shrunk it; the smallest known, as of 2021, has 509 points.
The hexagon floor
John Isbell’s seven-colour hexagon floor is the best known way to paint the plane. Nobody has done it with six.
Five, six or seven
Those are the only answers left. The seven-point shape above is the Moser spindle, from 1961, which settled “at least four”.
What the model leaves out: the computer only checks seven points here, and the sticks are dropped on a finite patch of floor. The bound of five needs a much bigger graph and a computer search we didn’t rerun.
Toy 2 · One tile for every shape
One tile fits all
Take any flat shape whose widest stretch is 1: no two of its points are more than 1 apart. Is there one tile, as small as possible, that every such shape fits inside after you turn and slide it? Henri Lebesgue asked in 1914. Here is the classic answer: a regular hexagon exactly 1 across between opposite sides.
The computer turns the shape through 60° in small steps and, at each turn, works out whether some slide keeps it inside all six walls. The hexagon’s three pairs of walls are each exactly 1 apart, and a shape 1 across is never wider than 1 in any direction, so one turn always works. The disk that is 3% too wide is the control: it is wider than 1 in every direction, so no turn can work, and the search says so.
Trim two corners
Julius Pál showed the hexagon (area 0.8660) still covers everything after cutting two corners off along lines touching its inner circle, leaving 0.8453. Press “Show Pál’s trim”.
Shaved to 0.8441
A century of shaving followed. Philip Gibbs’s 2018 tile has area 0.8440935944.
The floor rises to 0.834
No tile can be smaller than 0.832 (Brass and Sharifi, 2005). A September 2026 preprint by Zeng claims a computer-checked 0.834. The true answer is somewhere between.
What the model leaves out: curved shapes are drawn with a few hundred straight sides, and a fit counts when the shape is inside to the computer’s rounding. Only the plain hexagon is searched here; the trimmed one is drawn, with its area worked out from the cut lines.
Toy 3 · A hallway, a corner, a couch
The sofa and the corner
A hallway 1 wide turns a right-angle corner. What’s the biggest flat sofa you can slide and turn around it? Leo Moser asked in 1966. A unit square works (area 1). A half-disk of radius 1 turns on the spot (area 1.571). You can do better by carving.
Start with a long block. Turn the hallway around it a little at a time, while its inner corner swings along a half-circle, and cut away everything the walls cross. What’s left can make the turn by construction. The slider sets how far apart the two rounded ends sit. John Hammersley worked out this family in 1968: the best split gives area 2.2074. The uncarved block is the control; drive it and watch the walls bite.
Gerver’s sofa
Joseph Gerver found a smoother sofa built from 18 curved and straight pieces, with area about 2.2195. Hammersley’s is close behind.
Probably the end
In November 2024 Jineon Baek posted a long argument that Gerver’s sofa is the biggest possible. It is a preprint; checking it is under way.
Both ways round
If the sofa must turn left and right, Dan Romik found the best known one, with area about 1.6450.
What the model leaves out: the sofa is carved on a grid of squares 1/150 wide and the hallway turns in 180 small steps, so the counted area runs slightly above the exact formula. The sofa here follows Hammersley’s path; Gerver’s better shape uses a different path that we didn’t rebuild.
Prior art
Who got there first
Paint the plane
- The question is the Hadwiger–Nelson problem, posed around 1950 by Edward Nelson. The seven-point spindle is from William and Leo Moser (1961); the seven-colour hexagons are credited to John Isbell (Hadwiger–Nelson problem).
- Aubrey de Grey, The chromatic number of the plane is at least 5 (2018), with a 1,581-point graph. Later searches, including the Polymath16 project, shrank it; the smallest known, as of 2021, has 509 points.
One tile fits all
- Lebesgue posed it in a 1914 letter; Pál’s hexagon and two-corner trim are from 1920 (Lebesgue’s universal covering problem).
- Philip Gibbs, An upper bound for Lebesgue’s covering problem (arXiv 1810.10089, 2018): area 0.8440935944. Peter Brass and Mehrbod Sharifi, lower bound 0.832 (2005).
- Zeng, arXiv 2609.01284 (September 2026): a computer-checked lower bound of 0.834. A separate preprint, arXiv 2606.04458, gets 0.833 for convex tiles only.
The sofa
- Leo Moser posed it in 1966; John Hammersley’s family and its 2.2074 are from 1968 (Moving sofa problem).
- Joseph Gerver, On moving a sofa around a corner, Geometriae Dedicata (1992). Jineon Baek, Optimality of Gerver’s sofa (arXiv, 2024).
- Dan Romik, Differential equations and exact solutions in the moving sofa problem, Experimental Mathematics (arXiv 1606.08111, 2016).
The code behind every number runs in your browser: plane.js (the spindle check and the hexagon floor), cover.js (the fitting search and Pál’s areas), sofa.js (the carving and the drive).
Next click down
The sofa is the closest of the three to finished: if Baek’s argument holds up, the answer is Gerver’s 2.2195. The plane is the furthest: going from “at least 5” to “at least 6” would need a new idea, and the search for small five-colour graphs is still a good place to look.
More toys: three toys that shouldn’t work. Back to the Alien Math wing or the Lab.