Monoidal Categories and Mac Lane's Coherence Theorem

Intuition Beginner

A great deal of mathematics is about combining two things into one. You can pair two sets into a set of pairs. You can join two lists into one longer list. You can run two processes side by side and treat them as a single bigger process. Each of these takes a pair of objects and returns a single combined object, always in the same fixed way. A monoidal category is the setting that names this "combine two into one" operation and lays down the rules it must obey to behave well.

The first rule is about grouping. If you combine three things, you can combine the first two and then bring in the third, or hold the first and combine the last two. These two routes should land you in matching places. The second rule is about a neutral ingredient: there is a do-nothing object that, when combined with anything, gives back that thing unchanged. The empty list joined to a list returns the same list; a one-element set paired with a set is a relabelled copy of that set.

Here is the catch that makes the subject interesting. The two grouping routes do not give the identical result on the nose; they give matching results joined by a fixed, reversible comparison. A monoidal category is a consistent way to combine two objects into one, together with rebracketing comparisons that never bite you: no matter how you regroup, the comparisons always agree.

Visual Beginner

Picture a workshop where every job takes two parts and bolts them into one assembly. The bolting station is the combine operation. Feed it parts $A$ and $B$ on the left, and it returns the single assembly "$A$-with-$B$" on the right. There is also a blank part, the do-nothing piece: bolting it onto any assembly returns that assembly, just re-stamped.

Now suppose you have three parts to bolt together. You can bolt the first pair, then add the third; or bolt the last pair, then add the first. The two finished assemblies are not stamped identically, but the workshop guarantees a fixed reversible adapter that converts one into the other. That adapter is the rebracketing comparison.

feature	the combine operation	the do-nothing object	the rebracketing adapter
takes in	two objects	nothing	one regrouping of three objects
returns	one combined object	a neutral object	the matching regrouping
list example	join two lists	the empty list	regroup a triple join
set example	pair two sets	a one-element set	regroup a triple pairing

The picture's point is that the adapters are not arbitrary. Every chain of adapters between the same two regroupings agrees, so you never have to track which regrouping you used. That guarantee is the heart of the subject.

Worked example Beginner

Work with finite lists of letters and the join operation, which glues two lists end to end. The do-nothing object is the empty list, written $[]$.

Step 1. Combine two lists. Join $[c,a]$ and $[t]$ to get $[c,a,t]$. The combine operation here is concatenation, and it takes the pair of lists to one list.

Step 2. Check the do-nothing object. Join the empty list $[]$ on the left of $[c,a,t]$: you get $[c,a,t]$ back. Join it on the right: again $[c,a,t]$. The empty list changes nothing on either side.

Step 3. Try the two groupings of three lists $[c]$, $[a]$, $[t]$. Group the first pair first: join $[c]$ and $[a]$ to get $[c,a]$, then join $[t]$ to get $[c,a,t]$. Group the last pair first: join $[a]$ and $[t]$ to get $[a,t]$, then join $[c]$ on the left to get $[c,a,t]$.

Step 4. Compare. Both routes give $[c,a,t]$. For lists the two groupings produce the very same list, so the rebracketing adapter is the do-nothing comparison.

What this tells us: joining lists combines two into one, the empty list is neutral, and regrouping a triple changes nothing about the final list. Lists are the simplest case where the rebracketing adapter is as plain as possible. In richer settings the adapter is a genuine reversible comparison rather than the do-nothing one, but it always behaves with the same reliability.

Check your understanding Beginner

Formal definition Intermediate+

A monoidal category is a category $\mathcal{C}$ equipped with a tensor product functor $\otimes :\mathcal{C}\times \mathcal{C}\to \mathcal{C}$, a unit object $I\in \mathcal{C}$, and three natural isomorphisms — the associator ${\alpha }_{X,Y,Z}:(X\otimes Y)\otimes Z\stackrel{\sim }{\to }X\otimes (Y\otimes Z)$, the left unitor ${\lambda }_X:I\otimes X\stackrel{\sim }{\to }X$, and the right unitor ${\rho }_X:X\otimes I\stackrel{\sim }{\to }X$ — subject to two coherence axioms ^{[Mac Lane 1998]}. The pentagon axiom requires the diagram of four ways of reassociating $((W\otimes X)\otimes Y)\otimes Z$ to commute: $$ \alpha_{W,X,Y\otimes Z}\circ\alpha_{W\otimes X,Y,Z} =(\mathrm{id}W\otimes\alpha{X,Y,Z})\circ\alpha_{W,X\otimes Y,Z}\circ(\alpha_{W,X,Y}\otimes\mathrm{id}_Z). $$ The triangle axiom relates the associator to the unitors: $$ (\mathrm{id}X\otimes\lambda_Y)\circ\alpha{X,I,Y}=\rho_X\otimes\mathrm{id}_Y\colon (X\otimes I)\otimes Y\to X\otimes Y. $$ Notation introduced here ($\otimes$, $I$, $\alpha$, $\lambda$, $\rho$, and below $c$, $\mathrm{Mon}(\mathcal{C})$) is recorded in _meta/NOTATION.md; the natural-isomorphism language is that of 41.01.02.

Standard examples: $(\mathbf{Set},\times ,\{\ast \})$ with cartesian product; $({\mathbf{Vect}}_k,{\otimes }_k,k)$ and $(R\text{-}\mathbf{Mod},{\otimes }_R,R)$ with the algebraic tensor product; $(\mathbf{Ab},{\otimes }_{\mathbb{Z}},\mathbb{Z})$; the endofunctor category $([\mathcal{C},\mathcal{C}],\circ ,\mathrm{Id})$ under composition, whose monoids are the monads of 41.05.01; chain complexes $(\mathbf{Ch}(R),\otimes ,R[0])$ with the Koszul sign in the differential; graded vector spaces with the tensor grading; and cobordism categories, where $\otimes$ is disjoint union and the structure underlies topological quantum field theory.

A monoidal category is strict when $\alpha ,\lambda ,\rho$ are all identities, so $(X\otimes Y)\otimes Z=X\otimes (Y\otimes Z)$ and $I\otimes X=X=X\otimes I$ on the nose. The endofunctor example is strict.

Definition (monoid object). A monoid in a monoidal category $\mathcal{C}$ is a triple $(M,\mu ,\eta )$ with $M\in \mathcal{C}$, a multiplication $\mu :M\otimes M\to M$, and a unit $\eta :I\to M$ satisfying the associativity law $\mu \circ (\mu \otimes {\mathrm{id}}_M)=\mu \circ ({\mathrm{id}}_M\otimes \mu )\circ {\alpha }_{M,M,M}$ and the unit laws $\mu \circ (\eta \otimes {\mathrm{id}}_M)={\lambda }_M$ and $\mu \circ ({\mathrm{id}}_M\otimes \eta )={\rho }_M$. Monoids and their morphisms form a category $\mathrm{Mon}(\mathcal{C})$. In $(\mathbf{Set},\times )$ a monoid object is an ordinary monoid; in $(\mathbf{Ab},{\otimes }_{\mathbb{Z}})$ it is a ring; in $(R\text{-}\mathbf{Mod},{\otimes }_R)$ it is an $R$-algebra; in $([\mathcal{C},\mathcal{C}],\circ )$ it is a monad, recovering the slogan of 41.05.01 that a monad is a monoid in the category of endofunctors.

Definition (monoidal functor). A lax monoidal functor $F:\mathcal{C}\to \mathcal{D}$ between monoidal categories is a functor together with coherence cells ${\varphi }_{X,Y}:F(X)\otimes F(Y)\to F(X\otimes Y)$ (natural) and ${\varphi }_0:I_{\mathcal{D}}\to F(I_{\mathcal{C}})$ making three diagrams commute: the associativity hexagon relating $\varphi$ to the two associators, and the two unit squares relating ${\varphi }_0$ to the unitors. It is strong when ${\varphi }_{X,Y}$ and ${\varphi }_0$ are isomorphisms, and strict when they are identities. A monoidal natural transformation $\theta :F\Rightarrow G$ between lax monoidal functors is a natural transformation whose components are compatible with the coherence cells: ${\theta }_{X\otimes Y}\circ {\varphi }_{X,Y}^F={\varphi }_{X,Y}^G\circ ({\theta }_X\otimes {\theta }_Y)$ and ${\theta }_I\circ {\varphi }_0^F={\varphi }_0^G$.

Definition (braided and symmetric). A braiding on a monoidal category is a natural isomorphism $c_{X,Y}:X\otimes Y\stackrel{\sim }{\to }Y\otimes X$ satisfying the two hexagon axioms, which express compatibility of $c$ with $\alpha$ on both triple tensor patterns. The category is symmetric when $c_{Y,X}\circ c_{X,Y}={\mathrm{id}}_{X\otimes Y}$. The example $({\mathbf{Vect}}_k,\otimes )$ is symmetric via $v\otimes w\mapsto w\otimes v$; representations of a quantum group are braided but not symmetric.

Definition (rigid / compact closed). An object $X$ has a (left) dual $X^{\ast }$ when there are maps ${\mathrm{ev}}_X:X^{\ast }\otimes X\to I$ and ${\mathrm{coev}}_X:I\to X\otimes X^{\ast }$ satisfying the two zigzag (triangle) identities. A monoidal category in which every object has a dual is rigid; a symmetric rigid category is compact closed. In $({\mathbf{Vect}}_k,\otimes )$ the finite-dimensional spaces are exactly the dualizable ones, and the composite $I\to X\otimes X^{\ast }\to X^{\ast }\otimes X\to I$ recovers the dimension (categorical trace).

Definition (closed monoidal). $\mathcal{C}$ is closed when each functor $-\otimes Y$ has a right adjoint $[Y,-]$, the internal hom, giving the tensor-hom adjunction $\mathcal{C}(X\otimes Y,Z)\cong \mathcal{C}(X,[Y,Z])$. In $R\text{-}\mathbf{Mod}$ this is ${\mathrm{Hom}}_R(X{\otimes }_{R}Y,Z)\cong {\mathrm{Hom}}_R(X,{\mathrm{Hom}}_R(Y,Z))$.

Counterexamples to common slips Intermediate+

• The associator is data, not a property. Writing $(X\otimes Y)\otimes Z=X\otimes (Y\otimes Z)$ is correct only in a strict monoidal category. In $({\mathbf{Vect}}_k,\otimes )$ the two iterated tensor products are distinct vector spaces joined by the canonical isomorphism $\alpha$; conflating them discards the coherence content the pentagon controls.

• The pentagon and triangle are not redundant. One might hope a single coherence axiom suffices, but the pentagon governs reassociation of four factors and the triangle ties the unit into the associativity, and neither follows from the other together with naturality. Kelly showed certain a priori unit conditions are consequences, but the pentagon and triangle are the irreducible pair.

• Braided is strictly weaker than symmetric. A braiding need not satisfy $c_{Y,X}\circ c_{X,Y}=\mathrm{id}$; the braid-group representations arising in $\mathbf{Rep}(U_q\mathfrak{g})$ have $c$ and its inverse genuinely different, which is exactly what makes them sensitive to knotting. Assuming symmetry collapses the braid group to the symmetric group and erases the link-invariant content.

Key theorem with proof Intermediate+

The signature result is that coherence is automatic: the two axioms force every formal diagram of associators and unitors to commute.

Theorem (Mac Lane coherence; one-object form). Let $W(\mathcal{C})$ be the free monoidal category on one generating object built from a monoidal category $\mathcal{C}$: its objects are the bracketed words in a single letter and the unit $I$, and its only arrows are the formal composites of $\alpha ,\lambda ,\rho$ and their inverses, tensored and composed. Then between any two objects of $W(\mathcal{C})$ there is at most one arrow ^{[Mac Lane 1963]}. Equivalently: in any monoidal category, every diagram built solely from components of $\alpha ,\lambda ,\rho$, their inverses, identities, $\otimes$, and $\circ$ commutes.

Proof. Each bracketed word $w$ has an underlying length $|w|$, the number of letter-occurrences (the unit $I$ contributes $0$). Reassociating and removing units never change the length, so all words of a fixed length $n$ are connected by formal arrows, and words of different lengths by none. Fix $n$ and let $r_n$ be the right-normalized word $x\otimes (x\otimes (\cdots \otimes x))$ with all brackets to the right and no units. For each word $w$ of length $n$ define a canonical arrow ${\kappa }_w:w\to r_n$ as any formal composite of $\alpha$'s (rightward) and unit-deletions $\lambda ,\rho$ that reaches $r_n$; existence is by induction on the bracketing, peeling the outermost left grouping and applying one $\alpha$ or, at a unit, one unitor. It remains to show ${\kappa }_w$ is independent of the choices, for then any two formal arrows $w\to w^{\prime }$ both equal ${\kappa }_{w^{\prime }}^{-1}\circ {\kappa }_w$ and so coincide. Independence reduces to confluence of the rewriting "push a bracket right / delete a unit": the only local ambiguities are the pentagon (two ways to reassociate four factors) and the triangle (two ways to delete a unit adjacent to a bracket), and these are exactly the two axioms, which assert the competing rewrites agree. By Newman's lemma a terminating, locally confluent rewriting system is confluent, and the length-decreasing-then-rightward system terminates, so all canonical arrows coincide. Hence $\mathcal{C}(w,w^{\prime })$ in $W(\mathcal{C})$ has at most one element. $\square$

Bridge. This theorem builds toward strictification — the statement that every monoidal category is monoidally equivalent to a strict one — and appears again in the Master tier, where the free monoidal category is replaced by its category of words and the equivalence is constructed explicitly. The foundational reason coherence holds is that the pentagon and triangle are precisely the local confluence conditions of the bracket-pushing rewriting system, so once those two diagrams commute, Newman's lemma propagates commutativity to every diagram; this is exactly the mechanism by which a finite axiom list controls an infinite family of equations. The result generalises the list example of the Beginner tier, where reassociation was already invisible because concatenation is strictly associative, to the case where the associator is a genuine non-identity isomorphism. Putting these together, the central insight is that coherence lets one work as if every monoidal category were strict — dropping brackets and unit factors without comment — and the bridge is that this licence is what makes monoidal categories the practical foundation under tensor categories, quantum groups, and the topological quantum field theories cross-referenced below.

Exercises Intermediate+

Advanced results Master

Theorem (strictification). Every monoidal category $\mathcal{C}$ is monoidally equivalent to a strict monoidal category ${\mathcal{C}}^{\mathrm{str}}$ ^{[Mac Lane 1998]}. The strictification is the category whose objects are finite lists of objects of $\mathcal{C}$ and whose morphisms are morphisms of the iterated tensor products taken in the right-normalized bracketing; tensor product is concatenation of lists, strictly associative and strictly unital, and the strong monoidal functor $\mathcal{C}\to {\mathcal{C}}^{\mathrm{str}}$ sending an object to its length-one list is an equivalence whose coherence cells are the canonical reassociation isomorphisms of the coherence theorem.

Theorem (free monoidal category; universal property). The free monoidal category $\mathcal{F}(1)$ on one object is the category of bracketed words in a single letter with the formal $\alpha ,\lambda ,\rho$ as its only isomorphisms; by the coherence theorem it is equivalent to the discrete strict monoidal category $(\mathbb{N},+,0)$, the simplex-style bracketing category collapsed by coherence. For any monoidal category $\mathcal{C}$ and object $X\in \mathcal{C}$ there is a unique-up-to-monoidal-isomorphism strong monoidal functor $\mathcal{F}(1)\to \mathcal{C}$ sending the generator to $X$, the evaluation at $X$. The strictification $2$-functor $(-{)}^{\mathrm{str}}:\mathbf{MonCat}\to \mathbf{StrMonCat}$ is left $2$-adjoint to the inclusion, and coherence is the statement that the unit of this $2$-adjunction is a componentwise monoidal equivalence.

Theorem (Mac Lane coherence with units, full form). In the free monoidal category on any set of generators, every diagram of formal arrows — composites of instances of ${\alpha }^{\pm 1},{\lambda }^{\pm 1},{\rho }^{\pm 1}$ under $\otimes$ and $\circ$ — commutes; equivalently, the canonical functor from the free monoidal category to its strictification is faithful on formal arrows ^{[Kelly 1964]}. Kelly's refinement isolates the minimal axiom set: the pentagon, the triangle, and the derived equalities ${\lambda }_I={\rho }_I$ and the compatibility of $\lambda ,\rho$ with $\alpha$ at the unit follow, so only the pentagon and triangle are independent.

Theorem (Joyal-Street; braided coherence and the braid group). A braided monoidal category satisfies a coherence theorem in which formal diagrams commute up to the action of the braid groups $B_n$: the free braided monoidal category on one object has $\mathrm{Hom}(n,n)=B_n$, and the hexagon axioms are precisely the relations making the braiding a representation of $B_n$ compatible with tensoring ^{[Joyal-Street 1993]}. The Yang-Baxter equation $(c\otimes \mathrm{id})(\mathrm{id}\otimes c)(c\otimes \mathrm{id})=(\mathrm{id}\otimes c)(c\otimes \mathrm{id})(\mathrm{id}\otimes c)$ is the hexagon read on three factors, and a symmetric category is the quotient $B_n\twoheadrightarrow S_n$. The Eckmann-Hilton argument is the degenerate case: a one-object braided category with a single arrow forces the two compositions to agree and be commutative, the categorical source of the commutativity of ${\pi }_2$.

Synthesis. Putting these together, coherence is the assertion that the pentagon and triangle generate all equations among the structure isomorphisms, and strictification is its operational payoff: every monoidal category is monoidally equivalent to a strict one, so one may compute as if brackets and units were invisible. The foundational reason both hold is that the free monoidal category on one object is, after coherence, the bracketing category collapsed to $(\mathbb{N},+,0)$, and the strictification $2$-functor realizes this collapse as the unit of a $2$-adjunction whose components are equivalences. This generalises the monad picture of 41.05.01: a monad is a monoid in the strict monoidal endofunctor category, and the present unit explains why "strict" was available there for free, while in $({\mathbf{Vect}}_k,\otimes )$ or $\mathbf{Rep}(U_q\mathfrak{g})$ the associator is a genuine non-identity that coherence nonetheless renders harmless. This is exactly the structure under tensor categories, quantum groups, and operads: the braided refinement is dual to the symmetric one in that the braid group $B_n$ surjects onto $S_n$, the Yang-Baxter equation is the hexagon on three strands, and the rigid/compact-closed duals supply the evaluation, coevaluation, and trace that a topological quantum field theory assigns to cobordisms. The central insight is that "tensor with coherent associativity", "monoid in a monoidal category", "strictifiable structure", and "the algebra under TQFT and quantum groups" name one phenomenon, and the bridge is the coherence theorem that lets each viewpoint be used without bookkeeping.

Full proof set Master

Proposition 1 (the unitors are determined; ${\lambda }_I={\rho }_I$). In any monoidal category the triangle and pentagon axioms imply ${\lambda }_I={\rho }_I:I\otimes I\to I$, and more generally $\lambda$ and $\rho$ are compatible with $\alpha$ at the unit.

Proof. Apply the triangle axiom $({\mathrm{id}}_X\otimes {\lambda }_Y)\circ {\alpha }_{X,I,Y}={\rho }_X\otimes {\mathrm{id}}_Y$ with $X=Y=I$. Whisker by ${\lambda }_I$ and use naturality of $\lambda$ at ${\lambda }_I:I\otimes I\to I$, which gives ${\lambda }_I\circ ({\mathrm{id}}_I\otimes {\lambda }_I)={\lambda }_I\circ {\lambda }_{I\otimes I}$ reduced via the coherence rewriting to a single canonical arrow $I\otimes (I\otimes I)\to I$. Naturality of $\rho$ at ${\lambda }_I$ identifies ${\lambda }_I\circ ({\rho }_I\otimes {\mathrm{id}}_I)$ with the same canonical arrow. Since ${\lambda }_I$ is invertible, cancel it to obtain ${\rho }_I\otimes {\mathrm{id}}_I={\mathrm{id}}_I\otimes {\lambda }_I$ followed by ${\alpha }_{I,I,I}$, and reading both as unit-deletions $I\otimes I\to I$ forces ${\rho }_I={\lambda }_I$. The general compatibility follows because every diagram of unitors and associators commutes by the coherence theorem, of which this equality is the smallest instance. $\square$

Proposition 2 (coherence by rewriting; the one-object theorem). In the free monoidal category $W$ on one object, any two formal arrows between fixed objects are equal.

Proof. Equip the formal arrows with the rewriting system whose rules push an associator rightward, $(a\otimes b)\otimes c⇝a\otimes (b\otimes c)$, and delete a unit, $I\otimes a⇝a$ and $a\otimes I⇝a$. Each rule strictly decreases the pair (number of unit-occurrences, weighted left-bracket depth) in lexicographic order, so the system terminates: every word rewrites to the right-normalized unit-free word $r_n$ of its length $n$. The critical pairs are: two associators applicable to a four-factor word, whose joinability is the pentagon; an associator and a unit-deletion sharing a redex, whose joinability is the triangle (and Proposition 1); and two unit-deletions, joinable by Proposition 1. All critical pairs are joinable, so by Newman's lemma the terminating system is confluent, giving a unique normal form $r_n$ and a unique canonical arrow ${\kappa }_w:w\to r_n$. Any formal arrow $f:w\to w^{\prime }$ satisfies ${\kappa }_{w^{\prime }}\circ f={\kappa }_w$ (both are canonical $w\to r_n$, hence equal), so $f={\kappa }_{w^{\prime }}^{-1}\circ {\kappa }_w$ is forced. Thus ${\mathrm{Hom}}_W(w,w^{\prime })$ has at most one element. $\square$

Proposition 3 (strictification is a monoidal equivalence). The functor $L:\mathcal{C}\to {\mathcal{C}}^{\mathrm{str}}$, $X\mapsto [X]$ (the length-one list), is a strong monoidal equivalence.

Proof. Define ${\mathcal{C}}^{\mathrm{str}}$ with objects finite lists $[X_1,\dots ,X_n]$ and ${\mathcal{C}}^{\mathrm{str}}([X_1,\dots ,X_m],[Y_1,\dots ,Y_n])=\mathcal{C}(X_1\otimes \cdots \otimes X_m,Y_1\otimes \cdots \otimes Y_n)$, the tensors taken right-normalized; composition uses composition in $\mathcal{C}$. Concatenation of lists is strictly associative and strictly unital with unit the empty list, so ${\mathcal{C}}^{\mathrm{str}}$ is strict monoidal. The functor $L$ is fully faithful since ${\mathcal{C}}^{\mathrm{str}}([X],[Y])=\mathcal{C}(X,Y)$, and essentially surjective since $[X_1,\dots ,X_n]\cong [X_1\otimes \cdots \otimes X_n]$ via the canonical coherence isomorphism. By 41.01.02 a fully faithful essentially surjective functor is an equivalence. Its coherence cells ${\varphi }_{[X],[Y]}:[X]\otimes [Y]=[X,Y]\to [X\otimes Y]$ are the canonical isomorphisms; the lax hexagon and unit squares commute because every such diagram of associators and unitors commutes by Proposition 2. Hence $L$ is a strong monoidal equivalence. $\square$

Proposition 4 (rings, algebras, and monads as monoid objects). Monoid objects in $(\mathbf{Ab},{\otimes }_{\mathbb{Z}},\mathbb{Z})$ are rings, in $(R\text{-}\mathbf{Mod},{\otimes }_R,R)$ are $R$-algebras, and in $([\mathcal{C},\mathcal{C}],\circ ,\mathrm{Id})$ are monads.

Proof. In $(\mathbf{Ab},{\otimes }_{\mathbb{Z}})$ a monoid $(M,\mu ,\eta )$ has $\mu :M{\otimes }_{\mathbb{Z}}M\to M$ a bilinear map and $\eta :\mathbb{Z}\to M$ an element $1=\eta (1)$; the associativity law is associativity of $\mu$ and the unit laws say $1$ is two-sided, so $(M,+,\mu ,1)$ is a ring, and morphisms of monoid objects are ring homomorphisms. The $R$-module case is identical with ${\otimes }_R$, where the unit $R\to M$ makes $M$ an $R$-algebra (the image of $R$ is central by the unit laws together with $R$-bilinearity). In $([\mathcal{C},\mathcal{C}],\circ ,\mathrm{Id})$ the category is strict, so the associativity and unit laws of a monoid object read $\mu \circ \mu T=\mu \circ T\mu$ and $\mu \circ \eta T=\mathrm{id}=\mu \circ T\eta$ — the monad axioms of 41.05.01. Each identification is an isomorphism of categories $\mathrm{Mon}(\mathcal{C})\cong \{\text{rings}/R\text{-algebras}/\text{monads}\}$, because the data and laws correspond term by term. $\square$

Connections Master

• Natural transformations and functor categories 41.01.02. The associator, unitors, braiding, and the coherence cells of a monoidal functor are all natural isomorphisms in the sense built there, and the pentagon and triangle are equations between pasted naturality squares. The strictification proof invokes the characterisation of equivalences from 41.01.02 — fully faithful plus essentially surjective — to upgrade the canonical comparison functor to a monoidal equivalence, so the entire coherence apparatus is an application of functor-category theory to the special $2$-category of monoidal categories.

• Monads as monoids in endofunctors 41.05.01. A monad is exactly a monoid object in the strict monoidal category $([\mathcal{C},\mathcal{C}],\circ ,\mathrm{Id})$, and Proposition 4 places the monad axioms, the ring axioms, and the $R$-algebra axioms under one $\mathrm{Mon}(\mathcal{C})$ theorem. The whiskered composites $T\mu$, $\mu T$ that appear in the monad laws of 41.05.01 are the tensor product of this monoidal category, so the present unit supplies the foundational home for the "monoid in a monoidal category" slogan that 41.05.01 used informally.

• Adjunctions and the tensor-hom adjunction 41.03.01. A closed monoidal category is one where each $-\otimes Y$ has a right adjoint $[Y,-]$, an instance of the adjunctions of 41.03.01; the internal hom and the tensor-hom isomorphism are the defining adjunction unit and counit. Duals in a rigid category are a special adjunction $X⊣X^{\ast }$ internal to the monoidal structure, and the categorical trace is the composite of unit and counit, tying the rigid theory here to the adjoint-functor theory of 41.03.01.

• Representation theory, TQFT, and quantum groups. Symmetric monoidal functors out of a cobordism category are topological quantum field theories, and the rigid/compact-closed duals supply the evaluation and coevaluation a TQFT assigns to caps and cups; braided monoidal categories of representations of a quantum group $U_q\mathfrak{g}$ produce the Reshetikhin-Turaev link invariants via the Yang-Baxter braiding. The fusion and modular tensor categories that organise rational conformal field theory and anyonic quantum computation are monoidal categories with the rigidity, braiding, and coherence developed in this unit, which is the foundational layer the representation-theory and TQFT corpus specialises.

Historical & philosophical context Master

The definition of a monoidal category and the coherence problem were posed together by Saunders Mac Lane in Natural associativity and commutativity (1963) ^{[Mac Lane 1963]}, where the pentagon and hexagon first appear as the diagrams that must be imposed and the coherence theorem is proved by reducing all formal equalities to these. Mac Lane's question — when does "all diagrams of constraints commute" follow from finitely many — was answered in his paper for associativity and commutativity, and the unit conditions were sharpened immediately afterward by G. M. Kelly in On Mac Lane's conditions for coherence of natural associativities (1964) ^{[Kelly 1964]}, who showed several proposed axioms were redundant and isolated the pentagon and triangle as the independent pair. The systematic account, including strictification and the monoid-object language, was consolidated in Mac Lane's Categories for the Working Mathematician (1971; 2nd ed. 1998) ^{[Mac Lane 1998]}, Chapters VII and XI.

The braided refinement was introduced by André Joyal and Ross Street in Braided tensor categories (1986 preprint; published 1993) ^{[Joyal-Street 1993]}, who proved the braided coherence theorem identifying the endomorphisms of the free braided monoidal category with the braid groups and connecting the hexagon to the Yang-Baxter equation. This coincided with Vladimir Drinfeld's quantum groups, whose representation categories are braided but not symmetric, and with the Reshetikhin-Turaev construction of link and 3-manifold invariants from modular tensor categories. The Eckmann-Hilton argument, that two compatible unital binary operations coincide and are commutative, is the degenerate one-object case and the categorical explanation of the commutativity of higher homotopy groups; the rigid and compact-closed structures became the axiomatic backbone of the topological and categorical quantum field theories developed by Atiyah, Segal, and Baez-Dolan.

Bibliography Master

@article{MacLane1963,
  author  = {Mac Lane, Saunders},
  title   = {Natural associativity and commutativity},
  journal = {Rice University Studies},
  volume  = {49},
  number  = {4},
  year    = {1963},
  pages   = {28--46}
}

@book{MacLane1998,
  author    = {Mac Lane, Saunders},
  title     = {Categories for the Working Mathematician},
  edition   = {2},
  publisher = {Springer},
  series    = {Graduate Texts in Mathematics 5},
  year      = {1998}
}

@article{Kelly1964,
  author  = {Kelly, G. Max},
  title   = {On Mac Lane's conditions for coherence of natural associativities, commutativities, etc.},
  journal = {Journal of Algebra},
  volume  = {1},
  year    = {1964},
  pages   = {397--402}
}

@article{JoyalStreet1993,
  author  = {Joyal, Andr\'e and Street, Ross},
  title   = {Braided tensor categories},
  journal = {Advances in Mathematics},
  volume  = {102},
  number  = {1},
  year    = {1993},
  pages   = {20--78}
}

@book{EGNO2015,
  author    = {Etingof, Pavel and Gelaki, Shlomo and Nikshych, Dmitri and Ostrik, Victor},
  title     = {Tensor Categories},
  publisher = {American Mathematical Society},
  series    = {Mathematical Surveys and Monographs 205},
  year      = {2015}
}

@article{ReshetikhinTuraev1991,
  author  = {Reshetikhin, Nicolai and Turaev, Vladimir},
  title   = {Invariants of 3-manifolds via link polynomials and quantum groups},
  journal = {Inventiones Mathematicae},
  volume  = {103},
  year    = {1991},
  pages   = {547--597}
}