The axiom schema of separation (also called restricted comprehension or specification) states: for any set A and any first-order formula φ(x), the collection {x ∈ A : φ(x)} is a set. By requiring that new sets be carved out of an already-existing set A, separation avoids Russell's paradox: the paradoxical 'R' would require A to be the universal set, which ZFC never asserts exists. Separation is technically a schema — one axiom for each first-order formula φ — and is one of the primary tools for constructing subsets within ZFC.
Practice applying separation to construct specific sets: intersections A ∩ B = {x ∈ A : x ∈ B}, the set of even numbers within ℕ, and relative complements. Verify that each construction starts from an existing set. Then revisit Russell's paradox and identify exactly why separation prevents it.
From your overview of ZFC, you know that set theory needed a disciplined replacement for naive comprehension — the intuitive but contradictory principle that any property defines a set. Russell's paradox showed that the "set of all sets that don't contain themselves" leads to contradiction: R ∈ R ↔ R ∉ R. The fix is to never form a set from scratch using a property alone; instead, you must always carve a new set out of an existing one. This is the Axiom Schema of Separation (also called *Aussonderung*, restricted comprehension, or the specification schema): for any set A and any first-order formula φ(x) (possibly with parameters), the collection {x ∈ A : φ(x)} is a set.
The word "schema" is important: separation is not a single axiom but an infinite family of axioms, one for each formula φ. This is necessary because first-order logic cannot quantify over formulas (that would require second-order logic), so ZFC must include one axiom per formula as a template. In practice you use separation without thinking about which instance you're invoking: when you write A ∩ B = {x ∈ A : x ∈ B}, you are applying the instance of separation where φ(x) is the formula x ∈ B. Similarly, the relative complement A \ B = {x ∈ A : x ∉ B}, the set of even naturals {n ∈ ℕ : ∃k (n = 2k)}, and the kernel of a function {x ∈ A : f(x) = 0} all use separation with different choices of φ.
The key structural feature of separation is the restriction to an existing set A. This is precisely what blocks Russell's paradox. To form the Russellian set R, you would need φ(x) to be x ∉ x, and you would need A to be the "set of all sets." But ZFC never asserts such a universal set exists — and in fact, separation itself (combined with other axioms) proves it cannot exist. If a universal set V existed, then by separation you could form {x ∈ V : x ∉ x}, which would be the Russellian R. Since R ∈ R ↔ R ∉ R is a contradiction, the existence of V must be false. Separation thus both enables the construction of subsets and participates in the proof that no universal set exists.
Separation interacts with the other ZFC axioms in a division of labor. The axiom of pairing gives you small sets {a, b}; the power set axiom gives you the set of all subsets of a given set; the union axiom gives you the union of a family of sets. Separation's role is to filter: given any of these sets, you can cut out the subcollection satisfying any property you can express in first-order logic. This makes separation the primary tool for intersection, relative complement, and carving out structured subsets — the bread-and-butter operations of mathematical practice. The Axiom Schema of Replacement (which you'll study next) extends this by allowing the output to be a new set formed by applying a function, not just a subset cut from an existing one.