Sudoku Assumption Method - Where Trial Placement Ends and Logic Begins

/11 min read

Assuming a candidate in Sudoku can be a sound logical step or plain guessing, depending on what you take away from the assumption. This article compares trial placement, proof by contradiction and forcing chains on one grid, explains why "no contradiction appeared" proves nothing, and sets out when assuming is reasonable.

Is the assumption method logic?

The assumption method means placing one candidate in a cell for the sake of argument and following what happens next. Making an assumption does not break any rule of logic. What separates a logical step from a guess is what you take away from it, and only two conclusions are legitimate. If the assumption leads to a contradiction, the assumed candidate can be removed. If two opposite assumptions both lead to the same result, that result is true. Seeing no contradiction, on its own, settles nothing. The guide to HoDoKu, a free program for analyzing and practicing Sudoku, says so directly: a chain that produces no contradiction proves nothing. As long as you keep to these two conclusions and this one prohibition, reasoning from an assumption counts as logic. Writing a trial digit in ink and carrying on as if it were proven is what turns it into guessing.

Trial placement, contradiction and forcing chains

All three methods begin by supposing that some candidate is true. They differ in whether you decide, before you start, what kind of result you are looking for. Trial placement writes the digit on the grid and treats it like any solved cell. If the grid then fills without a contradiction, a puzzle with one solution is solved. But when a contradiction does turn up later, it is easy to lose track of which digits were trial ones and how far back to go. Proof by contradiction follows the assumption as a single line and watches for one thing only. If a contradiction appears, the candidate goes; if not, the whole line is dropped and the grid is restored. A forcing chain assumes, in turn, every candidate left in one cell, or every place one digit can still go. Anything reached on every branch is true whichever branch is right. HoDoKu uses forcing chain as an umbrella term that also covers chains ending in a contradiction. To keep them apart, this article calls the removal-by-contradiction kind proof by contradiction and keeps forcing chain for the kind that relies on branches agreeing.

Trial placement, contradiction and forcing chains compared
MethodWhat you assumeWhat you may concludeLines to follow
Trial placementOne candidate in a cell, written in and played onOnly that the grid is solved, if it fills without contradictionOne (erase everything if it fails)
Proof by contradictionOne candidate in a cellIf a contradiction appears, remove that candidateOne
Forcing chainEvery candidate of one cell, or every place for one digitA result reached on every branchOne per candidate or place

Whatever the method, a line that stalls without a contradiction proves nothing.

Proof by contradiction on a real grid

The grid shows a position reached by placing only naked singles and hidden singles, where neither finds anything more. Cells are named by row and column, so r5c1 is row 5, column 1. Its candidates are 2 and 3. Assume 2 and follow only single deductions. With 2 in row 5, r5c8, whose candidates are 2 and 4, must be 4. Row 5 now has both 2 and 4, so r5c2, with candidates 2, 3 and 4, must be 3. Now look at column 1. A 3 could only go in r4c1 or r5c1. The assumption has put 2 in r5c1, and r4c1 shares a box with the 3 in r5c2. Column 1 has nowhere left for a 3. The assumption reached a contradiction in just two steps, so 2 is removed from r5c1 and the cell is 3. Following the other side, assuming 3, produces nothing new with single deductions. On this grid, only the branch that failed carried any information. The whole argument fits in one sentence: if r5c1 were 2, the 3 in column 1 would have nowhere to go. Being able to explain it that briefly is a good sign that the step is genuine logic.

Assume 2 in r5c1
9
5
3
4
3
7
6
5
3
4
2
8
1
7
6
9
5
8
3
4
1
8
9
3
7
6
2
8
  • Column 1 loses every place for 3
  • Two cells forced by the assumption
  • Cell where 2 is assumed

A position where single-candidate and single-place deductions have run out. Candidates: r5c1 is 2 or 3, r5c8 is 2 or 4, r5c2 is 2 / 3 / 4, r4c1 is 1 / 3 / 5. In column 1, a 3 can only go in r4c1 or r5c1.

Why no contradiction proves nothing

On the same grid, r4c4 has candidates 5 and 9. Assume 5 and follow single deductions only. r6c4 becomes 9, r7c4 becomes 8 and r9c4 becomes 4, then a 9 goes into r4c2 and a 5 into r6c1, and progress stops. Five placements, and not one contradiction. Yet when the puzzle is solved to the end, r4c4 turns out to be 9. The assumption was wrong; the techniques used to follow it simply could not reach the point where the conflict shows. A contradiction can only be found within the reach of the tools you are using to look for it, which is why its absence is never evidence that an assumption is right. That is the sense of HoDoKu's remark that a chain without a contradiction proves nothing. If you take the branch that did not fail as correct and keep writing in ink, the next time you get stuck you will not know where the trial digits began. A line that stalls should be dropped, assumption and all, and the grid put back as it was.

A wrong assumption with no contradiction
9
5
3
4
3
7
6
5
3
4
2
8
1
7
6
9
5
8
3
4
1
8
9
3
7
6
2
8
  • Five cells filled without contradiction
  • Cell where 5 is assumed

r4c4 and r6c4 both hold 5 and 9, r7c4 holds 5 / 8 / 9 and r9c4 holds 4 / 5 / 9. Assuming 5 gives five singles and then stalls with no contradiction. In the actual solution, r4c4 is 9.

Forcing chains: follow both candidates

Now follow both candidates of r4c4. On the 5 branch, r6c4 becomes 9, r7c4 becomes 8 and r9c4 becomes 4. On the 9 branch, r6c4 becomes 5, and r7c4 is again 8 and r9c4 again 4. Since r4c4 must hold one of the two, the 8 in r7c4 and the 4 in r9c4, common to both branches, are true no matter which branch is right. That is a forcing chain conclusion. Neither branch alone said anything, yet laid side by side they give two solved cells. In fact the same result comes from a naked pair: r4c4 and r6c4 share 5 and 9, which uses up both digits in column 4, leaving 8 for r7c4 and 4 for r9c4. Many named techniques can be read as exactly this sort of two-branch argument, folded into a short pattern. Before hunting for a forcing chain, check whether a named technique gives the same result; it usually takes fewer steps and less writing.

Cells that agree on both branches
9
5
3
4
3
7
6
5
3
4
2
8
1
7
6
9
5
8
3
4
1
8
9
3
7
6
2
8
  • Two cells sharing 5 and 9
  • Cells fixed on either branch

Whether r4c4 is 5 or 9, r7c4 becomes 8 and r9c4 becomes 4. The same result follows from the naked pair that uses up 5 and 9 in column 4.

Where is the line between logic and guessing?

The line is not whether you assume, but whether the conclusion still depends on the assumption. A candidate removed by contradiction stays removed when the assumption is withdrawn, because the assumption has been shown to be false. A digit fixed by a forcing chain also survives, because every starting point leads to it. Digits written during trial placement vanish with the assumption, since that was their only support. A simple rule follows: keep on the grid only the conclusions that survive withdrawing the assumption. A second guide is the length of the line and the number of branches. Lines of two to five steps, like the ones above, can be followed mentally and explained afterwards. Once there are three or more branches, or a line runs to a dozen steps or more, following it by hand becomes much harder. HoDoKu calls a chain whose branches split and rejoin, or whose links lean on earlier steps, a forcing net, files it under last resort methods, and says one is hard to find manually unless you are very experienced. Under the same heading it describes brute force, placing a digit and seeing whether a solution follows, as not really a technique. The boundary is better pictured as a gradual slope from short, explainable reasoning to exhaustive search than as a single sharp line.

When to assume and when not to

How far to rely on assumptions depends on why you are solving. If you are practicing techniques, breaking through by assumption costs you the chance to notice the technique the position was asking for. On the other hand, when the named techniques you know have all been tried, picking a two-candidate cell and testing one line of contradiction is a reasonable step for a human solver. The table below sets out how to use an assumption in each situation and why. The common thread is to check, before anything goes on the grid, that the conclusion would survive withdrawing the assumption.

When assuming is reasonable and when to avoid it
SituationHow to use an assumptionWhy
Singles and basic techniques have run outTry one line of contradiction from a two-candidate cellA short line can be followed in your head and explained
Many two-candidate cells remainFollow both branches and keep what they shareA result common to both does not depend on the assumption
You are practicing techniquesLook for a named technique instead of assumingBreaking through by assumption skips the practice
You just want to finishIf you place on trial, write trial digits in another color or smallerYou can see at once what to erase if it fails
You do not know the puzzle has one solutionDo not treat a filled grid as the only answerAnother solution may still exist
The line grows long without a resultDrop the assumption and restore the gridErrors inside a long line are hard to spot and costly to undo

What matters is not whether you assume, but that only conclusions independent of the assumption stay on the grid.

Common mistakes and how to spot them

Most errors with assumptions come from how the conclusion is drawn. The first is treating the branch with no contradiction as proven, as in the r4c4 example. The test is whether the only reason you can give is that nothing went wrong. The second is stacking one assumption inside another. If a second assumption fails while the first is still in force, what you have shown is only that the second candidate is false when the first assumption holds; it is not a removal for the whole grid. Anything removed inside a nested assumption must be restored once the outer assumption is dropped. The third is following one branch and taking for granted that the other ends the same way. A forcing chain only holds when every branch has been followed to the end. The fourth is assuming on a grid whose pencil marks are incomplete or wrong. If the candidate list is wrong, both contradictions and agreements will lead to wrong conclusions. Rechecking the pencil marks once before starting an assumption prevents most of these. The fifth is incomplete cleanup after dropping an assumption. If candidates removed along the trial line are not restored, the next deduction runs on a wrong candidate list. Mark every candidate you remove while following an assumption, and write all marked candidates back when you abandon it.

Techniques to try before assuming

Keeping assumptions as a last resort, for when named techniques stop working, tends to help rather than hinder progress. Before reaching for one, work through naked and hidden pairs, pointing pairs and box line reduction, X-Wing, XY-Wing and coloring. X-Wing, XY-Wing and coloring in particular split into cases on whether a candidate is true or false and keep only what holds in every case, which gives them the same skeleton as a short forcing chain. Coloring can even be read as a way of recording an assumption on the grid with colors, following strong links for a single digit. The related articles cover the order in which to review a stuck grid and the chain of errors that follows a guessed digit. The more puzzles you finish without assuming, the easier it becomes to recognize the positions where an assumption is truly needed.