A monoidal category is a category C equipped with a bifunctor ⊗: C × C → C (the tensor product), a unit object I, and natural isomorphisms for associativity (A ⊗ (B ⊗ C) ≅ (A ⊗ B) ⊗ C) and left/right unit laws (I ⊗ A ≅ A ≅ A ⊗ I), all satisfying Mac Lane's coherence conditions (the pentagon and triangle axioms). Examples include (Set, ×, {*}), (Vect, ⊗, k), (Ab, ⊗_Z, Z), and (Cat, ×, 1). Mac Lane's coherence theorem guarantees that every diagram built from the associator and unitors commutes, so one may work as if ⊗ were strictly associative and unital.
Start with (Set, ×, {*}) and verify the associator and unitor isomorphisms explicitly. Then move to (Vect_k, ⊗_k, k) and confirm the same axioms hold. State the pentagon and triangle axioms and check them for these examples. Appreciate the coherence theorem by constructing a diagram with multiple paths and verifying they agree.
You already know that a category has objects and morphisms, and that many categories come equipped with a notion of "combining" objects — sets have cartesian product, vector spaces have tensor product, groups have direct product. A monoidal category makes this notion of combination precise: it is a category C equipped with a bifunctor ⊗: C × C → C, a unit object I, and carefully chosen natural isomorphisms that say ⊗ is associative and I is a unit — but only up to isomorphism, not on the nose.
The three structural isomorphisms are the associator α_{A,B,C}: A ⊗ (B ⊗ C) → (A ⊗ B) ⊗ C, the left unitor λ_A: I ⊗ A → A, and the right unitor ρ_A: A ⊗ I → A. These must satisfy two coherence axioms: the pentagon axiom (which says that the five ways to reassociate A ⊗ B ⊗ C ⊗ D all agree) and the triangle axiom (which relates the associator and unitors when one argument is I). These axioms are not arbitrary — they are the minimal conditions needed to prevent contradictions when you combine multiple reassociations.
Mac Lane's coherence theorem is the punchline: provided the pentagon and triangle axioms hold, *every* diagram built from the associator and unitors commutes automatically. This means you can work as if ⊗ were strictly associative and I were a strict unit — you can write A ⊗ B ⊗ C without parentheses and suppress unit objects without worrying about which path through the diagram you took. In practice, mathematicians routinely use this without comment. What coherence does *not* say is that the associator is the identity morphism; in most examples it is a genuine, non-trivial isomorphism.
The examples of monoidal categories span mathematics: (Set, ×, {∗}) with the cartesian product and any one-element set; (Vect_k, ⊗_k, k) with the tensor product and the ground field; (Ab, ⊗_Z, Z) with the abelian tensor product; (Cat, ×, 1) with the product of small categories. Crucially, a monoidal category need not be symmetric — the tensor A ⊗ B and B ⊗ A may not be naturally isomorphic. A braiding or symmetry is additional structure, not part of the basic definition. This distinction matters in representation theory and quantum groups, where the braiding captures physically meaningful data about the exchange of particles.