Sudoku

The route from an American magazine to a global daily puzzle

Sudoku is often linked to older mathematical grids, especially Latin squares, because both organize symbols under non-repetition constraints. That comparison is useful for understanding the structure, but the documented modern publishing history is much more recent. Cornell University's Mathematics Department identifies Howard Garns as the designer of a puzzle published by Dell Magazines in 1979 under the title Number Place. Its familiar regions and one-of-each-symbol requirement established the recognizable form played today.

Japanese publisher Nikoli encountered Number Place in an American puzzle magazine and introduced it to Japanese readers in 1984. Nikoli's own history says the initial Japanese title expressed the idea that each number must remain single. The phrase was eventually shortened to Sudoku. In 1986 the publisher adopted symmetrical clue placement as a construction convention. Symmetry is not one of the logical rules, but it gives a handcrafted grid balance and became part of Nikoli's editorial character.

The next major shift concerned distribution rather than mechanics. Retired New Zealand judge Wayne Gould wrote software capable of generating puzzles and brought them to British newspapers, with publication beginning in 2004. The format spread rapidly through newspapers in 2005 and then moved naturally into browsers and mobile devices. Digital play introduced candidate toggles, undo histories, highlighted peers, daily seeds, and leaderboards while leaving the central deduction unchanged.

Sudoku now ranges from a quiet pencil puzzle to a timed international discipline. The World Puzzle Federation runs championships and a Grand Prix that include classic grids and variants. Those events demonstrate the depth available to expert solvers, but their pace is not the standard every player must adopt. A daily browser puzzle remains complete on its own: a finite set of givens, one intended solution, and a trail of conclusions that can be enjoyed without competitive pressure.

Rules, units, peers, and candidates

The board has nine horizontal rows, nine vertical columns, and nine outlined boxes. Solvers often call any row, column, or box a unit. A cell belongs to three units at once, and all other cells in those units are its peers. The printed or locked digits are givens. Every empty cell must receive one digit, and a completed board is legal only when every unit contains 1 through 9 exactly once.

This vocabulary makes explanations shorter and checks more reliable. Suppose an empty cell shares a row with 1, 3, and 8, a column with 2, 4, and 9, and a box with 5. Those peers rule out seven digits, leaving 6 or 7 as candidates. Neither candidate is a commitment. A pencil mark simply records that a value still survives all three filters. When another placement removes 7, the cell becomes forced and 6 can be entered with a complete reason.

A properly constructed standard puzzle has one intended solution. Uniqueness matters because it lets deductions determine the finish rather than leaving an arbitrary choice between equally legal boards. If two candidates seem interchangeable, continue searching outside that cell. A row in another box may settle one of them, or a candidate pattern may eliminate it. Guessing can fill space for a while, but it hides the useful question: which restriction have I not yet used?

A repeatable workflow from first scan to final check

Begin with completion pressure. Scan for units containing seven or eight givens, because their missing set is small. If a row lacks only 2 and 5, test its two empty cells against their columns and boxes. One conflict may decide both positions. Continue through dense units, but do not remain in reading order; a nearly completed box near the bottom may be much more productive than the next row at the top.

Next switch from cells to digits. Pick one digit and locate every copy already on the board. Within each box that lacks it, use those occupied rows and columns to cross out impossible positions. If only one square remains in a box, place the digit there. This box scanning method is especially effective early because givens create broad lines of exclusion. Repeat it for each digit rather than assuming the most visually common one is the only useful target.

When direct placements slow down, add candidates. For each unsolved cell, subtract the digits already present among its peers. A cell with one remaining candidate is a naked single. A digit that has only one possible cell inside a unit is a hidden single. The second situation can be overlooked because the winning cell may display several marks; you find it by counting locations for a digit, not candidates in a cell.

After every confirmed placement, perform maintenance before hunting for something new. Remove that digit from all peers and inspect any cell or unit that changed. This small habit creates cascades: one placement makes a single, that single locks a digit into a box, and the box then completes a row. Entering several remembered moves before updating candidates risks preserving impossible notes and missing the actual chain.

If the board stalls, rotate your viewpoint. Review cells with two candidates, digits with only two locations in a unit, and intersections between a box and a line. Use a check button as a diagnostic after a reasoned sequence, not as a slot machine for testing nine answers. At the end, verify every unit rather than relying only on a celebratory animation; the final audit reinforces the same constraint awareness used throughout the solve.

Singles, locked candidates, and useful subsets

Singles are the foundation rather than a beginner trick to abandon. A naked single has only one legal candidate in its cell. A hidden single is the sole location for a particular digit within a row, column, or box. Advanced patterns usually work by removing candidates until a new single appears, so accurate single detection remains important on the hardest grid.

Locked candidates exploit an overlap. If every possible 8 in the upper-middle box lies on row two, the box's eventual 8 must occupy that row. Therefore, no cell elsewhere on row two can contain 8. The reverse direction also works: if all candidates for 8 in a row lie inside one box, remove 8 from the box's other rows. You gain an elimination without knowing the exact square.

A naked pair appears when two cells in one unit have the same two candidates and no others. Those cells must contain the pair in some order, which means both digits can be erased from every other cell in the unit. A naked triple follows the same principle across three cells whose combined candidates contain only three digits. The cells need not each show all three; what counts is that the group's available values are confined to the same number of positions.

Hidden pairs invert the search. If two digits occur as candidates nowhere except the same two cells in a unit, those cells are reserved for the digits. Other candidates can be removed from them. Hidden subsets are easiest to spot by tracking how often each digit appears in a unit. They are also easy to misread, so verify the entire row, column, or box before deleting anything.

Patterns for demanding boards

Hard puzzles often require you to connect candidate locations across multiple units. The X-Wing is a well-known example. If candidate 4 appears in exactly the same two columns in two different rows, the four cells form a rectangle. Each row must place its 4 in one of those columns, so the columns will receive their 4s at the rectangle corners. Candidate 4 can then be removed from other cells in those columns. Extra candidate positions in either defining row invalidate the pattern.

Candidate chains offer a more general language. A strong link means that one of two positions must hold a digit; a weak link means both positions cannot hold it. Alternating these relationships can show that a candidate seen by both ends of a chain is impossible, or that assuming one value would force a duplicate. Keep chains short while learning and mark each implication. A conclusion you cannot retrace is too fragile to use.

Some apps and books present dozens of named formations. Names help solvers discuss a shape, but memorizing a catalog is less valuable than asking three questions: where can this digit still appear, which positions are linked, and what candidate can be removed in every possible outcome? A valid technique ends with a concrete placement or deletion. Merely noticing a symmetrical arrangement does not change the board.

Backtracking is a legitimate computer algorithm and can be used carefully by a human, but it is different from a logical solve. If a recreational puzzle is designed for deduction, exhaust local and linked patterns before branching. When you do explore a branch for study, record the assumption explicitly and reverse every consequence if it creates a contradiction. Otherwise an invisible guessed value can contaminate the remainder of the session.

Common mistakes and productive practice

The most damaging mistake is checking only one unit. A digit may look perfect in a box while duplicating a value several cells away in its column. Stale pencil marks cause a quieter version of the same problem: a candidate eliminated three moves ago remains visible and prevents you from seeing a single. Other frequent errors include treating two cells that share a pair plus extra candidates as a naked pair, or deleting a candidate from an entire line when locked candidates apply only to part of it.

Build an error-resistant cadence: observe, state the reason, enter the digit, and update its peers. When a contradiction appears, locate the most recent step whose reason cannot be restated. Check for a mistap before dismantling a whole region. Undoing to an explainable position teaches more than repeatedly asking the interface whether a random answer is correct.

Practice can target process rather than speed. On one board, solve without a clock and name every single. On another, focus on candidate maintenance. Choose an intermediate grid and highlight every box-line interaction even if a direct single is available elsewhere. These constrained sessions make a technique recognizable under normal conditions. Reviewing a completed board also reveals which notes were unnecessary and where a shorter deduction existed.

The activity can exercise sustained visual attention, orderly note keeping, and logical elimination in the ordinary sense of practicing those tasks. It should not be sold as medical treatment or as a test of intelligence. Enjoyment, fewer unforced errors, and clearer explanations are sound measures of progress. Speed can follow once the workflow is stable, but a fast incorrect grid is not a stronger solve.

Use difficulty as a training dial. Easy puzzles offer many singles and are ideal for learning the interface. Medium puzzles introduce locked candidates and pairs. Hard puzzles ask you to preserve accurate notes over longer chains. Move upward when you can finish the current level consistently and explain your eliminations. If fatigue turns every cell into visual noise, stop and return later; a fresh scan often reveals a restriction you had ceased to notice.