Here is a natural first reaction when you meet set difference: "If 7 - 3 = 4 and 3 - 7 = -4 for numbers, and 'subtraction is subtraction,' then A - B and B - A should be the same thing up to some flip — or at least related in a clean way." The flip part is wrong; the "related" part is a little right. Let us pin down what actually happens.

A - B and B - A are different sets for almost every pair (A, B). The only universal statement that connects them is that they are disjoint — they share no element — and together they make up the symmetric difference A \triangle B. That is as deep as the link goes.

The definition, which is where the asymmetry comes from

A - B = \{ x \mid x \in A \text{ and } x \notin B \}

The definition does two different things with A and B. It asks for membership in A and non-membership in B. Those two roles are not interchangeable. If you swap them, you get a condition that points at different elements:

B - A = \{ x \mid x \in B \text{ and } x \notin A \}

Now the membership test is against B and the non-membership test is against A. Nothing in this flip is "a sign change" or "a reversal" — both operations are positive in what they keep; they just keep different things.

A quick numerical check

A = \{1, 2, 3\}, \ B = \{2, 3, 4\}.

Different sets. Not even close to equal.

Why: 1 passes the first test (in A, not in B) but fails the second (not in B). 4 does the opposite. The overlap \{2, 3\} fails both tests — it is in both sets, so it is in neither difference.

Why "subtraction is subtraction" is a trap

Number subtraction looks like set difference, but they are analogies, not the same operation.

The one structural parallel: number subtraction is not commutative, and set difference is not commutative. That is the only part the analogy gets right.

The one case where they agree

If A - B = B - A, then every element of A - B is also in B - A. But elements of A - B are in A and not in B, while elements of B - A are in B. The only way both conditions can hold is if A - B is empty. Similarly, B - A must be empty. Which means every element of A is in B and every element of B is in A — that is, A = B.

A - B = B - A \iff A = B.

So "order does not matter" only when the two sets are actually identical.

What A - B and B - A have in common: they are disjoint

An element of A - B is in A but not in B. An element of B - A is in B but not in A. Can the same element appear in both differences? No — the first requires "not in B," the second requires "in B." Both cannot hold. Therefore:

(A - B) \cap (B - A) = \varnothing.

Their union is the symmetric difference — the elements that belong to exactly one of the two sets:

(A - B) \cup (B - A) = A \triangle B.

This is the deepest statement you can make about the pair: they are the two halves of the symmetric difference, cleanly split by which set the element originally belonged to.

The Venn picture makes it obvious

Two Venn diagrams showing A minus B and B minus A shading opposite crescentsTwo Venn diagrams side by side each containing overlapping circles A and B. In the left diagram only the left crescent of A is shaded, representing A minus B. In the right diagram only the right crescent of B is shaded, representing B minus A. The shaded regions never overlap. A not-equal sign sits between the two diagrams. A − B A B ≠ B − A A B
$A - B$ shades the left crescent; $B - A$ shades the right crescent. The two crescents never touch. Together they make the symmetric difference, but neither contains a single element of the other.

Once you see the two crescents, the question "are they the same?" answers itself. They are in different parts of the diagram. They can only coincide by both being empty — the A = B case.

A cardinality identity worth knowing

|A - B| = |A| - |A \cap B|
|B - A| = |B| - |A \cap B|

From these,

|A - B| - |B - A| = |A| - |B|.

So the sizes of the two differences differ by |A| - |B|. If A is larger than B, then A - B is larger than B - A by exactly that amount. If |A| = |B|, the two differences have the same cardinality — but still are not equal as sets (unless A = B outright).

Why: the overlap A \cap B sits inside both A and B. Subtract the overlap from |A| and you get the "only A" count; subtract it from |B| and you get the "only B" count. The overlap cancels when you compare.

A worked example to cement the asymmetry

Let $A = \{2, 4, 6, 8, 10\}$ and $B = \{3, 6, 9, 12\}$. Compute $A - B$ and $B - A$, then verify the disjointness and union identities.

Step 1. Find the overlap.

A \cap B = \{6\}

Why: 6 is the only number in both lists.

Step 2. Compute A - B.

A - B = \{2, 4, 8, 10\}

Why: drop 6 from A since it is also in B. Everything else in A stays.

Step 3. Compute B - A.

B - A = \{3, 9, 12\}

Why: drop 6 from B. Everything else in B stays.

Step 4. Check disjointness.

(A - B) \cap (B - A) = \{2, 4, 8, 10\} \cap \{3, 9, 12\} = \varnothing \quad \checkmark

Step 5. Check symmetric-difference union.

A \triangle B = (A - B) \cup (B - A) = \{2, 3, 4, 8, 9, 10, 12\}

Every element is in exactly one of A, B — as the "exactly one of" reading of symmetric difference predicts.

Result. A - B = \{2, 4, 8, 10\} and B - A = \{3, 9, 12\} are disjoint sets, different sizes, and together they make A \triangle B.

The right mental slogan

Replace "subtraction is subtraction" with a slogan that survives set theory: set difference is a direction. A - B points from A away from B. B - A points from B away from A. Turning the arrow around gives you a different set, the same way turning around on a road puts you on a different view. The symmetric difference is the union of both viewpoints — it forgets the direction entirely.

Once you adopt the "direction" metaphor, the non-commutativity is obvious rather than puzzling. The unexpected thing would be if it were commutative.

Related: Set Operations · Symmetric Difference · A − B vs A △ B Side-by-Side · A − B vs A ∩ B′