81 cells, 9 boxes — and 27 hidden units

Count the little squares and you get 81: nine rows of nine. That’s the answer most people want. But sudoku players and mathematicians count differently. They see the grid as a stack of 27 “units” (sometimes called houses): 9 rows, 9 columns and 9 boxes — the 3×3 blocks. Each of those 27 units must contain the digits 1 to 9 exactly once.

The clever part is that every single cell belongs to three units at once — one row, one column and one box — and through them it’s linked to 20 other cells (its “peers”) that it can never share a digit with. So the real answer to “how many squares” depends on whether you mean the cells you can see, or the invisible web of constraints connecting them.

The secret geometry that makes it a puzzle

Here’s the strategy hiding in the square formation. Strip away the 3×3 boxes and a sudoku is just a Latin square — fill each row and column with 1–9, which is almost trivially easy. The boxes are what turn it into a genuine puzzle: they add a third overlapping constraint, and every clever solving technique is really about exploiting where those constraints intersect.

That’s why expert methods all read the geometry rather than the numbers. A “hidden single” works because a digit can only fit one cell in a unit. A “pointing pair” works because if a digit in a box must lie on one row, it can be erased from the rest of that row. “Box-line reduction” uses the same overlap in reverse. None of these are about arithmetic — they’re about the way the rows, columns and boxes are stitched together. The squares are the strategy.

How to actually play

The rules are gloriously simple, and despite all the digits there is no maths involved — the symbols could just as easily be nine colours. You start with a grid that has some cells already filled in (the “clues” or “givens”), and your job is to fill every empty cell so that each row, each column and each 3×3 box contains every digit from 1 to 9 exactly once. A properly made puzzle has exactly one solution, reachable by logic alone — never guessing. Newspapers typically give around 25 clues; the difficulty comes from how cleverly those clues are placed, not how many there are.

How computers — and AI — crack it instantly

A computer solves any valid 9×9 sudoku in milliseconds, and it doesn’t need fancy machine learning to do it. Sudoku is a textbook constraint-satisfaction problem, and the standard approach combines two ideas: constraint propagation (repeatedly ruling out impossible digits, exactly like a human does) and backtracking (when stuck, try a candidate, and if it leads to a contradiction, rewind and try another). Between them they sweep the grid clean almost instantly.

So why are machines so good at it? Because the problem is finite, rigid and small: 81 cells, nine options each, rules with no ambiguity — ideal territory for systematic search. (Interestingly, sudoku on an enormous n×n board is a genuinely “NP-complete” hard problem, but the everyday 9×9 version is tiny by comparison.) AI built on neural networks can learn to solve them too, but for plain sudoku that’s rather like using a sledgehammer: the old-fashioned logic is already unbeatable.

Sudoku by the numbers

The grid is small but the maths is vast. There are 6,670,903,752,021,072,936,960 — about 6.7 sextillion — valid completed grids, which collapse to roughly 5.5 billion truly distinct ones once you account for rotations and relabelling. And the fewest clues a solvable puzzle can possibly have is 17: that no 16-clue puzzle exists was finally proven in 2012 by a year-long exhaustive computer search.

For all its Japanese name, sudoku is largely an American invention: architect Howard Garns published the modern form as “Number Place” in 1979. It was Japanese puzzle company Nikoli that renamed it sūdoku — short for a phrase meaning “the digits must stay single” — before it swept the world in the mid-2000s.

Sources

• “Mathematics of Sudoku” (Wikipedia) & Felgenhauer–Jarvis (2005) — 81 cells, 27 units, 6.67×10²¹ grids, 5.47bn essentially different.
• McGuire, Tugemann & Civario (2012) — proof that the minimum number of clues is 17.
• Sudoku history & solving-algorithm references — Howard Garns / Nikoli; constraint propagation and backtracking; NP-completeness of generalised sudoku.

Related