Alternating Inference Chains (AIC)

An Alternating Inference Chain (AIC) is a sequence of Sudoku candidate inferences in which Strong and Weak Links alternate.

AICs generalize much of advanced Sudoku chain logic.

Instead of restricting the chain to:

• one digit, as in an X-Chain;
• or a sequence of bivalue cells, as in an XY-Chain;

an AIC can move through different digits, cells and houses as long as every transition obeys the required inference.

The core reading is:

if this candidate is false, the next must be true; if that candidate is true, the next must be false; continue until the endpoints prove something useful.

AICs are not guesswork. They are compact proofs built from candidate relationships that are already present in the grid.

Prerequisites: Strong and Weak Links

AICs only make sense if Strong and Weak Links are precise.

Strong Link

For candidates A and B:

A false → B true

A common example is a conjugate pair: a digit has exactly two legal positions in a house.

If one is false, the other must be true.

Weak Link

For candidates A and B:

A true → B false

Candidates are weakly linked when they cannot both be true.

Examples include:

• the same digit in two cells that see each other;
• two different candidates in the same cell.

One subtle but important rule

A Strong Link can also be used in the weak direction, because if A and B cannot both be true, A true still implies B false.

A merely Weak Link cannot be upgraded into a Strong Link.

This asymmetry is fundamental when building chains.

Why the links must alternate

Suppose we start from candidate A being false.

A Strong Link lets us infer:

A false → B true

From B being true, a Weak Link lets us infer:

B true → C false

From C being false, another Strong Link gives:

C false → D true

So the logical rhythm becomes:

false → true → false → true → ...

That is the “alternating inference” in AIC.

If two consecutive steps do not support the required truth transition, the chain is invalid no matter how attractive the candidate geometry looks.

AIC notation

Many Sudoku solvers write:

• = for a Strong Link;
• - for a Weak Link.

A simple schematic chain might look like:

A = B - C = D - E = F

Read it from left to right as an implication path.

Do not treat the punctuation as decoration.

Every symbol must correspond to a real candidate relationship in a cell or house.

What makes an AIC useful?

A chain is only useful when its endpoints prove something outside the chain or create a contradiction that forces a conclusion.

Two endpoint patterns are especially important.

Same-digit endpoints

If an AIC begins and ends with Strong Links on the same digit, at least one endpoint must be true.

Therefore any outside candidate for that digit that sees both endpoints can be eliminated.

This is the classic AIC endpoint elimination.

Different-digit endpoints in cells that see each other

If the two endpoints use different digits and the endpoint cells see each other, the chain can prove cross-eliminations:

• the start digit cannot appear in the end cell;
• the end digit cannot appear in the start cell.

The exact endpoint rule must be derived from the chain, not guessed from visual symmetry.

AIC Type 1

A common taxonomy calls a chain with the same digit at both ends an AIC Type 1.

The chain proves:

at least one endpoint contains digit X.

Any other X candidate that sees both endpoints is false.

This resembles an X-Chain endpoint rule, but the interior of an AIC can switch digits and use a much wider range of link types.

Step-by-step AIC Type 1 example

AIC Type 2

An AIC Type 2 starts and ends with Strong Links on different digits in endpoint cells that see each other.

The chain proves a pair of cross-eliminations.

Suppose the endpoints are:

• candidate X in cell A;
• candidate Y in cell B;
• and A sees B.

Then the chain can establish that:

• Y cannot be in A;
• X cannot be in B.

This follows because each endpoint possibility forces the opposite endpoint state needed to eliminate the conflicting candidate.

Step-by-step AIC Type 2 example

AICs and Nice Loops

Nice Loop terminology predates or overlaps with the modern AIC language used by many solvers.

The underlying inference logic is closely related.

A useful practical distinction is:

• AIC: emphasize an open chain and what its endpoints prove;
• Nice Loop: emphasize a chain that closes back on itself and what the closure forces.

Different programs and communities classify borderline cases differently.

For solving correctness, the important question is not the label.

The important question is:

does every Strong/Weak transition remain valid, and does the endpoint or loop conclusion logically follow?

Discontinuous Nice Loops

A loop can return to its starting cell/candidate in a way that creates a contradiction.

For example, the loop may effectively prove:

if A is true → ... → A is false

Then A must be false.

Or it may prove:

if A is false → ... → A is true

Then A must be true.

This is a discontinuous loop because the closure creates the contradiction that produces the deduction.

The result may be:

• one candidate elimination;
• or a forced placement.

Continuous AIC / Nice Loops

A loop is continuous when the closure obeys the normal propagation rules instead of generating a start/end contradiction.

These loops can be powerful because weak relationships along the loop can become effectively locked by the full cycle.

That may eliminate:

• extra copies of a digit from a house represented by a weak link;
• extra candidates from a cell represented by a weak link between two digits.

Continuous loops are conceptually harder than ordinary endpoint AICs, so learn open chains first.

Cannibalistic AICs

Sometimes an AIC eliminates a candidate that also appears inside the chain itself.

That can feel illegal at first.

It is not.

A chain is a proof, not a permanent set of simultaneously true candidates.

If the endpoint logic proves that an internal candidate is false, eliminating it is valid even though that candidate helped express the inference.

Such chains are often called cannibalistic AICs.

AIC vs X-Chain

An X-Chain:

• uses one candidate digit throughout;
• alternates Strong and Weak Links for that digit;
• commonly eliminates the digit from cells seeing both strong endpoints.

An AIC:

• can switch candidate digits;
• can move through both cell links and house links;
• supports a broader range of endpoint conclusions.

So an X-Chain is best understood as a restricted single-digit member of the broader alternating-inference family.

AIC vs XY-Chain

An XY-Chain uses a sequence of bivalue cells.

Inside each bivalue cell, the two candidates provide the internal either/or relationship; between cells, matching candidates create the external links.

An AIC is more general.

It does not require every participating cell to be bivalue and can transition through conjugate pairs, cell links and other valid candidate structures.

If you are comfortable with XY-Chains, AICs are the natural next step because the notation becomes more general while the implication mindset stays the same.

AIC vs Simple Coloring

Simple Coloring normally follows one digit through a connected network of conjugate pairs and assigns two polarities.

An AIC:

• may use several digits;
• does not require a full two-color network;
• follows one explicit inference path.

Coloring is often easier to visualize.

AICs are more flexible.

What is a Grouped AIC?

A Grouped AIC allows one node in the chain to represent a group of candidates rather than one individual candidate.

For example, several candidate copies of X inside one box/row intersection may collectively behave as one side of a Strong Link.

The logic is still alternating inference, but the nodes become more abstract.

Grouped AIC deserves its own later Guide because it introduces a significant jump in notation and recognition complexity.

This page establishes the non-grouped foundation first.

How to find AICs without tracing random chains

Blindly following every possible candidate relationship is inefficient.

Use a target-driven workflow.

1. Find dense link structures

Look for:

• conjugate pairs;
• bivalue cells;
• candidate digits with sparse positions;
• intersections where strong relationships connect.

2. Start from a plausible endpoint

Good endpoints often have a potential common target or a useful interaction with another candidate in a nearby cell.

3. Alternate deliberately

At every step ask:

• what truth state do I currently know?
• do I need a Strong or Weak inference next?
• does the candidate relationship actually support it?

4. Stop when the endpoints prove something

A long chain that produces no elimination is not useful.

Do not keep extending it merely because more links exist.

5. Prefer the shortest clear proof

The same elimination may have several descriptions:

• X-Chain;
• XY-Chain;
• Empty Rectangle;
• Turbot Fish;
• AIC.

Use the shortest representation that remains rigorous.

Common AIC mistakes

Mistake 1: alternating symbols without alternating inference

Writing = - = - does not make a valid chain.

Each link must support the required truth transition.

Mistake 2: treating a Weak Link as Strong

From A true → B false you cannot automatically infer A false → B true.

Only a genuine Strong Link supports that direction.

Mistake 3: assuming chain candidates are all true together

An AIC expresses conditional inference, not one simultaneous candidate assignment.

Mistake 4: eliminating from a candidate that sees only one endpoint

For a same-digit Type 1 endpoint elimination, the target must see both endpoints.

Seeing one is not enough.

Mistake 5: building unnecessarily long chains

Long chains are harder to verify and easier to misread.

If a shorter named pattern proves the same result, use it.

Mistake 6: confusing a contradiction proof with guessing

An AIC does not choose an arbitrary branch and hope it works.

Every inference is logically forced by Strong/Weak relationships, and the conclusion is proven for all relevant endpoint possibilities.

When should you learn AICs?

AICs make the most sense after you are comfortable with:

1. candidates;
2. Strong and Weak Links;
3. X-Chains;
4. XY-Chains;
5. Simple Coloring or similar link-network reasoning.

If those concepts are still difficult, AIC notation can feel much more complicated than the underlying logic really is.

FAQ

What does AIC stand for in Sudoku?

AIC stands for Alternating Inference Chain. The chain alternates Strong and Weak candidate relationships to propagate true/false implications.

Is an AIC the same as a Nice Loop?

They are closely related inference frameworks, and terminology varies between solvers. AIC language often emphasizes open endpoint chains, while Nice Loop terminology emphasizes closed loops and propagation rules.

Does an AIC have to use more than one digit?

No. A single-digit alternating chain can be expressed in AIC language, although a specialized name such as X-Chain may be clearer.

Is an XY-Chain an AIC?

Conceptually, XY-Chains fit inside the wider alternating-inference framework, but they have a more specific structure based on bivalue cells. Keeping the specialized name is useful because it tells you how to search for the pattern.

Are AICs trial and error?

No. An AIC is a deductive implication proof. Each transition is justified by an existing Strong or Weak Link, and the endpoint conclusion follows logically.

What should I learn after AIC?

The natural extensions are Grouped AICs, ALS-based chains, continuous loops and more complex forcing structures.

What to learn next

Once ordinary AICs are comfortable, the next major step is grouped inference: allowing several candidates to act as one node. That opens the door to Grouped AIC, ALS chains and many expert-level deductions.