Blog Archive

Saturday, November 21, 2026

From the Closed Structure on SymMonCat to Tensor Products of PROPs

From-Symmetric-Monoidal-Categories-to-Tensor-Products-of-PROPs (8)

From Symmetric Monoidal Categories to Tensor Products of PROPs

Abstract. A PROP describes operations with finitely many inputs and outputs. We construct the tensor product of symmetric monoidal categories that represents bilinear functors. For PROPs, its interchange law says that applying one operation along rows and the other down columns gives the same composite in either order. The Eckmann–Hilton argument computes the tensor square of the unital magma PROP, giving the PROP for commutative monoids. Tensoring the monoid PROP with its opposite yields the PROP for bimonoids, whose models in vector spaces are bialgebras. Tensoring with the PROP for copying and discarding freely turns the given tensor into a categorical product. This cartesianization is left biadjoint to the inclusion of cartesian categories into symmetric monoidal categories. On Lawvere theories the external tensor recovers the Kronecker tensor.

How can the same compatibility law turn two multiplications into one commutative product and also allow a noncommutative bialgebra? Both calculations start with a rectangle. We apply one operation along its rows and the other down its columns. Reversing the direction of one operation changes the boundary of the rectangle and hence the equation it expresses.

Give four input wires the row order (x1,x2,y1,y2). To operate on the columns, we need the order (x1,y1,x2,y2). The symmetry in a symmetric monoidal category supplies the crossing of the middle two wires. What makes the two ways of composing operations agree? We can trace that equality back to a functor of two variables.

The four factors in a bilinear functor

Write ⊗ for the tensor inside a symmetric monoidal category, I for its unit, and σ for its symmetry. We suppress associators and unitors and keep the permutations of factors visible. We work with ordinary categories, using small categories as inputs to an external tensor product and small skeleta for categories of finite sets.

Given symmetric monoidal categories B and C, let [B,C]⊗ be the category of strong symmetric monoidal functors and monoidal natural transformations. Such a functor F comes with coherent isomorphisms F(b)⊗F(b′)≅F(b⊗b′) and IC≅F(IB). A monoidal natural transformation respects these isomorphisms.

The functor category has a pointwise tensor, given by (F⊗G)(b)=F(b)⊗G(b). To exhibit F⊗G as a strong monoidal functor we have to exchange the two middle factors. Writing Fb for F(b), its comparison is

Fb⊗Gb⊗Fb′⊗Gb′→1⊗σGb,Fb′⊗1Fb⊗Fb′⊗Gb⊗Gb′→F2⊗G2F(b⊗b′)⊗G(b⊗b′).

Now take a strong symmetric monoidal functor L:A→[B,C]⊗ and set H(a,b)=L(a)(b). For fixed a, the functor H(a,−) has the structure of L(a). For fixed b, the functor H(−,b) is strong symmetric monoidal because evaluation at b preserves the pointwise tensor and its unit.

There is one more piece of information in L. Its comparison L(a)⊗L(a′)→L(a⊗a′) is a monoidal natural transformation. To see what that means, arrange the four values of H in a square.

(H(a,b)H(a,b′)H(a′,b)H(a′,b′))

Start with the four factors tensored in row order. Combine each row, then combine the two results. Alternatively, exchange the middle two factors, combine each column, and then combine those results. Monoidality of the comparison says that these two maps to H(a⊗a′,b⊗b′) agree.

A bilinear functor H consists of a bifunctor with these two strong symmetric monoidal structures, natural in all variables, satisfying this mixed square and its unit conditions. In particular H(IA,b) and H(a,IB) are identified with IC, compatibly with the other variable's tensor. The two identifications at (IA,IB) agree. These conditions let us reconstruct L by currying. Gurski, Johnson and Osorno, §§2–3

For a concrete instance, equip finite sets and bijections with disjoint union. The bifunctor H(S,T)=S×T is bilinear for this monoidal structure by distributivity in each variable. A pair (s,t) in (S1⊔S2)×(T1⊔T2) belongs to a unique block Si×Tj. The two expansions give precisely the same four blocks. Passing from one order to the other moves S1×T2 past S2×T1.

The tensor product representing bilinear functors

Write ⊠ for the tensor between symmetric monoidal categories. The universal bilinear functor A×B→A⊠B sends (a,b) to a⊠b and (s,t) to s⊠t. Precomposition with this functor and currying give natural equivalences of categories

[A⊠B,C]⊗≃Bilin⁡(A,B;C)≃[A,[B,C]⊗]⊗.

The arrows in the middle category are natural transformations monoidal in each variable. The right-hand side identifies the internal hom. This gives a closed symmetric monoidal structure in the bicategorical sense. One can work with permutative categories, in which associativity and the unit laws are strict. With strong symmetric monoidal functors as morphisms, their inclusion into symmetric monoidal categories is a biequivalence. Gurski, Johnson and Osorno, Theorem 1.2 and §§4–6

In permutative models we can construct A⊠B by generators and relations. Every bilinear functor has an underlying bifunctor, so we begin with the free permutative category on A×B. There is one letter a⊠b for each pair of objects. Its objects are the finite ordered tensor words

(a1⊠b1)⊗⋯⊗(an⊠bn).

Tensoring concatenates words, and the empty word is the unit I. Each pair of arrows s:a→a′ and t:b→b′ gives a one-letter arrow s⊠t:a⊠b→a′⊠b′. These arrows already compose as they do in A×B, so (s′⊠t′)∘(s⊠t)=(s′∘s)⊠(t′∘t) and 1a⊠1b=1a⊠b. We can tensor these arrows and permute letters using the symmetry.

Consider the two-letter word (a⊠b)⊗(a⊠b′) and the single letter a⊠(b⊗b′). Bilinearity gives an isomorphism H(a,b)⊗H(a,b′)≅H(a,b⊗b′) in the target. We need an arrow in the source that a functor can send to this isomorphism. Doing the same in both variables and for the empty word, we adjoin the following four families of arrows and their inverses. The subscript records the entry held fixed.

λa:(a⊠b)⊗(a⊠b′)→∼a⊠(b⊗b′),ρb:(a⊠b)⊗(a′⊠b)→∼(a⊗a′)⊠b,ua:I→∼a⊠IB,vb:I→∼IA⊠b.

Which composites of generators should agree? First, require all four families to be natural in their entries. For example, if t:b→b′, we can apply t once after combining a column, or apply it to each letter before combining. Naturality gives

(1a⊗a′⊠t)∘ρb=ρb′∘((1a⊠t)⊗(1a′⊠t)).

At fixed a, impose the associativity, unit and symmetry axioms for (λa,ua). Thus three letters with the same a can be combined in either parenthesization to give the same arrow to a⊠(b1⊗b2⊗b3). The same axioms apply to (ρb,vb) at fixed b.

Now put xij=ai⊠bj for i,j∈{1,2}. From the row-ordered word x11⊗x12⊗x21⊗x22 we want a map to (a1⊗a2)⊠(b1⊗b2). Combining the rows first applies λa1⊗λa2. To combine the columns first we must move x12 past x21, using τ=1⊗σx12,x21⊗1. Bilinearity imposes the relation

ρb1⊗b2∘(λa1⊗λa2)=λa1⊗a2∘(ρb1⊗ρb2)∘τ.

There are also rectangles with an empty direction. For two empty rows, inserting a unit in each row and then combining them gives ua1⊗a2=ρIB∘(ua1⊗ua2). Two empty columns give vb1⊗b2=λIA∘(vb1⊗vb2). When both directions are empty, the two arrows I→IA⊠IB agree, so uIA=vIB. Two parallel expressions now represent the same arrow of A⊠B precisely when we can pass from one to the other using the relations inside tensor products and composites.

Given a bilinear H with permutative target C, send a word to ⨂iH(ai,bi), send s⊠t to H(s,t), and send λ,ρ,u,v to the corresponding structure maps of H. Every defining equation holds under this assignment by the axioms of H, so it gives a strict symmetric monoidal functor H^:A⊠B→C. These assignments determine the extension uniquely since the generators produce every arrow. Conversely, any strict symmetric monoidal functor F out of the quotient gives a bifunctor by H(a,b)=F(a⊠b) and H(s,t)=F(s⊠t). The images of λ,ρ,u,v give H its bilinear structure. These two constructions are inverse.

A transformation monoidal in each variable extends by tensoring its components on the letters. For a strong symmetric monoidal functor out of the presentation, its tensor and unit comparisons supply a monoidal isomorphism with the strict extension of its restriction. This gives the first equivalence displayed above. Gurski, Johnson and Osorno, Definition 4.4 and Propositions 5.3 and 6.8

The tensor unit is (FinBij,⊔), the groupoid of finite sets and bijections under disjoint union. It is free symmetric monoidal on one object. A strong symmetric monoidal functor out of it is determined up to monoidal isomorphism by the image X of a singleton. It sends an n-element set to X⊗n and a permutation to the corresponding symmetry. Evaluation at the singleton thus gives [FinBij,C]⊗≃C. Interchanging the two variables of the universal pairing gives symmetry for ⊠, and both bracketings of a threefold tensor represent trilinear functors. Gurski, Johnson and Osorno, §§8–12

Reading the universal pairing as a rectangle

For arrows s:a→a′ and t:b→b′, bifunctoriality gives two factorizations of s⊠t.

(s⊠1b′)∘(1a⊠t)=s⊠t=(1a′⊠t)∘(s⊠1b).

To read these paths as operations, fix objects a,b and write X=a⊠b. Bilinearity identifies a⊗m⊠b⊗n with X⊗mn, including X⊗0=I. Expanding the a-power first gives m rows of n wires. Expanding the b-power first gives n columns of m wires. Write τm,n for the symmetry that puts the row-ordered inputs into column order. For a 2×3 rectangle it sends the ordered list of wire labels

(11,12,13,21,22,23)⟼(11,21,12,22,13,23).

Suppose s:a⊗m→a and t:b⊗n→b. Their induced operations on X are s¯=s⊠1b and t¯=1a⊠t under the power identifications. What does 1a⊗m⊠t do? After splitting into m rows, naturality of the comparison in t identifies it with m parallel copies of t¯. Splitting into columns gives the corresponding description for s⊠1b⊗n. Using row order to identify the source of s⊠t with X⊗mn, its two factorizations now read

s¯∘t¯⊗m=s⊠t=t¯∘s¯⊗n∘τm,n.

This is the interchange law. Bifunctoriality supplied the equality. Bilinearity and its mixed square supplied the row and column readings of that equality. Every input wire occurs once on each side.

For the 2×3 rectangle, give the two rows the labels (u1,u2,u3) and (v1,v2,v3). The formula becomes

s¯(t¯(u1,u2,u3),t¯(v1,v2,v3))=t¯(s¯(u1,v1),s¯(u2,v2),s¯(u3,v3)).

The letters label the six inputs of an arrow X⊗6→X. This use of variables makes sense in any symmetric monoidal category. Forming a tuple from six arrows with a common source Y would also require a map copying Y into six factors.

Passing to one-colour PROPs

A one-colour PROP is a strict symmetric monoidal category with objects 0,1,2,…, tensor on objects given by addition, and chosen generator 1. An arrow m→p describes an operation with m inputs and p outputs. A PROP map preserves the generator, tensor and symmetries strictly. A model in a symmetric monoidal category C is a strong symmetric monoidal functor to C.

Take PROPs P and Q. Every generating object m⊠n of their external tensor is isomorphic to the tensor power (1⊠1)⊗mn, so choosing these powers again gives a one-colour PROP. For s:m→p in P and t:n→q in Q, the two readings both start with an m×n array and end with a p×q array.

PathFirst operationSecond operation
Rows firstApply t to the m rows to get m×q.Apply s to the q columns to get p×q.
Columns firstApply s to the n columns to get p×n.Apply t to the p rows to get p×q.

At each stage the symmetry puts the factors into the indicated order. For instance, on m×k row-ordered wires the map applying s down all columns and returning the output to row order is τk,p∘s¯⊗k∘τm,k. Thus both routes give s⊠t:mn→pq under the chosen power identifications. Juxtaposing the two operations inside one PROP, s¯⊗t¯, has boundary m+n→p+q. Multiplication of arities comes from the rectangle, and addition of arities comes from juxtaposition.

This gives a presentation of P⊠Q. Take the coproduct of the two PROPs on the same colour, so they already share the wires and permutations. Impose the rectangle equations for mixed pairs of operations, including empty inputs or outputs. Equivalently, a pair of PROP maps P→R and Q→R extends uniquely to P⊠Q→R precisely when their images satisfy interchange. It suffices to check generating operations in each factor. Tensoring and composing rectangles extends the equations to all operations. Hackney and Robertson, §§3.2–3.3 and Theorem 35

P⊠Q≅(P⨿Q)/∼int,

where ∼int is the congruence generated by those equations under composition and tensoring.

If AlgQ⁡(C) denotes the category of models and monoidal natural transformations, the same construction gives

AlgP⊠Q⁡(C)≃AlgP⁡(AlgQ⁡(C)).

The inner model category carries the pointwise tensor from C. An operation s:m→p of P must be a map of Q-models from the m-fold tensor of a model to its p-fold tensor. Commuting with t:n→q is exactly the homomorphism condition for this map. A morphism of the resulting models is a single arrow X→X′ preserving both structures. We can therefore compute the tensor product by writing out the rectangles for its generators.

Two computations with the same interchange law
The Eckmann–Hilton argument computes the tensor square of the unital magma PROP

Let U be the PROP of unital magmas. It is generated by a multiplication 2→1 and a unit 0→1, subject only to the left and right unit laws. What does the tensor square U⊠U force these generators to satisfy? On its generating wire X we have two multiplications μ,ν:X⊗2→X. Interchange between the two units identifies them. Write e:I→X for their common value. Interchange between the two multiplications equates the following two arrows from X⊗4 to X.

μ∘(ν⊗ν)=ν∘(μ⊗μ)∘(1⊗σX,X⊗1).

Label the four input wires a,b,c,d in row order. The equality reads μ(ν(a,b),ν(c,d))=ν(μ(a,c),μ(b,d)). The Eckmann–Hilton argument uses this same rectangle three times, each time inserting e in different slots. For example, the first row below precomposes both sides with 1X⊗e⊗e⊗1X:X⊗2→X⊗4.

Inputs in row orderRows firstColumns firstConsequence
(x,e,e,y)μ(x,y)ν(x,y)A common product m=μ=ν
(e,x,y,e)m(x,y)m(y,x)Commutativity
(x,y,e,z)m(m(x,y),z)m(x,m(y,z))Associativity

Write PCMon for the PROP of commutative monoids. Sending its multiplication and unit to m and e gives a PROP map PCMon→U⊠U. Conversely, interpret both copies of U in PCMon using its generating multiplication and unit. Both paths around the binary rectangle multiply the same four inputs, and the remaining rectangles hold by the unit laws. This gives a map back. The two maps are inverse on generators, so the calculation identifies the PROPs themselves. In particular, it holds for unital magmas satisfying interchange in any symmetric monoidal category. Aguiar and Mahajan, Proposition 6.29 and Remark 6.30

U⊠U≅PCMon≃(FinSet,⊔).

The last description makes the wires explicit. A function between finite sets assigns each input to one output. Multiply the inputs in each fibre, taking the unit for an empty fibre. Associativity and commutativity make this independent of parentheses and fibre order. Composing functions composes these fibrewise products. Baez, Universal Algebra and Diagrammatic Reasoning, p. 42

A monoid and a comonoid give a bimonoid

Take the PROP M for associative unital monoids, with μ:2→1 and η:0→1. In Mop the reversed generators are a comultiplication Δ:1→2 and a counit ε:1→0. What happens to the rectangle when the second operation points outwards?

Start with two input wires. One path multiplies first and comultiplies the result. The other first comultiplies each input, making four wires in the order (x1,x2,y1,y2). To multiply corresponding outputs we need the order (x1,y1,x2,y2). The crossing in the bialgebra axiom is precisely this change of order.

Δ∘μ=(μ⊗μ)∘(1⊗σX,X⊗1)∘(Δ⊗Δ).

There are three more pairs of generators, and each contributes a familiar axiom.

PairInterchange equation
η,Δ Δ∘η=η⊗η
μ,ε ε∘μ=ε⊗ε
η,ε ε∘η=1I

Together these say that Δ and ε are monoid homomorphisms, with the tensor-product monoid structure on X⊗X. We have therefore computed M⊠Mop≅PBimon, the PROP for bimonoids. Its models in (Vectk,⊗k,k) are precisely the unital, counital k-bialgebras. Baez, p. 58

For instance, on the group algebra k[G] put Δ(g)=g⊗g and ε(g)=1 on basis elements, then extend linearly. On g,h∈G, both paths give gh⊗gh. This calculation allows gh and hg to differ. The additional Hopf structure comes from the inverse in G, which gives S(g)=g−1. The same construction on the monoid algebra k[N]=k[t] gives Δ(t)=t⊗t. An antipode would require S(t)t=1, which has no solution in k[t].

Free cartesianization by copying and discarding

In the group algebra example we chose a comultiplication on one object. Now suppose we want any two arrows u:Z→X and v:Z→Y to be able to use the same input. Their tensor u⊗v takes two copies of Z as input. To pair them into an arrow out of Z we need a way to copy Z. To project from X⊗Y onto X we also need a way to discard Y. How can we adjoin these operations so that the given tensor becomes a categorical product?

Consider what copying, discarding and permuting wires can do. Each of n outputs selects one of the m inputs, so the wiring is a function [n]→[m], where [n]={1,…,n} and [0]=∅. For example, the constant function [2]→[2] with value 1 describes the wiring (x,y)↦(x,x). Both outputs use the first input and the second input is discarded. These wirings form the PROP D=(FinSetop,⊔). It is the PROP for cocommutative comonoids, generated by a copy δ:1→2 and a discard !:1→0. Coassociativity makes repeated copying independent of parentheses, cocommutativity makes the two copies interchangeable, and the counit laws say that discarding either copy leaves the other wire unchanged.

Tensoring an operation with a wiring

Let f:Z→W be an operation in P and w:m→n a wiring in D. Tensoring their sources and their targets gives an arrow in C=P⊠D of type

f⊠w:Z⊠m⟶W⊠n.

In D, n is the disjoint union of n single wires. Strong monoidality in the second variable gives the identifications

Z⊠n≅(Z⊠1)⊗n,W⊠n≅(W⊠1)⊗n.

These identifications are natural in the first variable. Thus f⊠idn:Z⊠n→W⊠n acts as (f⊠id1)⊗n, one arrow on each copy. For a binary f:2→1, the arrow f⊠id3:6→3 applies f to three supplied pairs. With the wiring d3:1→3 that triples its input, f⊠d3:2→3 uses a single pair three times.

For the wiring dn:1→n that copies one input n times, consider the pair (f,dn):(Z,1)→(W,n) in P×D. We can apply f first, passing through (W,1), or dn first, passing through (Z,n). Since composition is computed coordinate by coordinate, we have

(idW,dn)∘(f,id1)=(f,dn)=(f,idn)∘(idZ,dn).

Now apply the functor ⊠:P×D→C. It sends (W,1) to W⊠1. Since it preserves composition, the path through (W,1) is sent to (idW⊠dn)∘(f⊠id1). Applying the same rule to the other path gives a second factorization of the very same arrow f⊠dn:Z⊠1→W⊠n.

f⊠dn=(Z⊠1→f⊠id1W⊠1→idW⊠dnW⊠n)=(Z⊠1→idZ⊠dnZ⊠n→f⊠idnW⊠n).

The upper path applies f to one bundle and copies the output. The lower path copies the input and then applies f to each copy, since the monoidal comparison reads its last arrow as (f⊠id1)⊗n. In the presentation of C, this equality is exactly the interchange relation imposed on the coproduct P⨿D.

At n=2, idZ⊠d2 gives the copy map δZ⊠1 after identifying Z⊠2 with (Z⊠1)⊗2. At n=0, Z⊠0≅I, so idZ⊠d0 gives the discard !Z⊠1 and f⊠id0 acts as idI. With the same identifications at W, interchange gives

f⊠d2=δW⊠1∘(f⊠id1)=(f⊠id1)⊗2∘δZ⊠1,f⊠d0=!W⊠1∘(f⊠id1)=!Z⊠1.

Copying a whole bundle X⊗Y puts its factors in the order X,Y,X,Y. Copying each factor separately gives X,X,Y,Y, so we exchange the middle two wires. Discarding the bundle discards both factors. Thus

δX⊗Y=(1X⊗σX,Y⊗1Y)∘(δX⊗δY),!X⊗Y=!X⊗!Y.

The wirings themselves satisfy both equations. Copying before or after a wiring selects the same input for each output copy, and there is only one wiring to 0. Composition and tensoring preserve these equations, so copy and discard are natural for every arrow of C.

Why the tensor is a product

Take any h:Z→I. Its output is the empty bundle, whose discard is !I=1I. The discard equation now gives h=!I∘h=!Z. Thus every object has exactly one arrow to I, which makes I terminal.

Now take u:Z→X and v:Z→Y. To use the same input for both, copy it and apply one arrow to each copy. This gives the candidate pairing

⟨u,v⟩:Z→δZZ⊗Z→u⊗vX⊗Y.

To read off either component, discard the other factor. The projections are

πX: X⊗Y→1X⊗!YX,πY: X⊗Y→!X⊗1YY.

Projecting the pairing onto X discards the output of v. The discard equation moves that discard back to Z, and the counit law cancels it against the copy. Applying this calculation to each projection gives

πX∘⟨u,v⟩=(u⊗(!Y∘v))∘δZ=(u⊗!Z)∘δZ=u,πY∘⟨u,v⟩=((!X∘u)⊗v)∘δZ=(!Z⊗v)∘δZ=v.

For uniqueness, take any h:Z→X⊗Y. The copy equation for h moves the copy from its input to its output, giving

⟨πX∘h,πY∘h⟩=(πX⊗πY)∘(h⊗h)∘δZ=(πX⊗πY)∘δX⊗Y∘h=h.

In the last step δX⊗Y supplies two whole copies. The projections keep X from the first and Y from the second. Discarding one copy of each factor leaves the other unchanged by the counit laws, so (πX⊗πY)∘δX⊗Y=1X⊗Y. Taking components and forming the pairing are therefore inverse maps in the defining bijection for a product,

C(Z,X⊗Y) ≅ C(Z,X)×C(Z,Y).

For a:Z′→Z the copy equation also gives ⟨u,v⟩∘a=⟨u∘a,v∘a⟩, so this bijection is natural in Z.

Conversely, suppose ⊗ is a product and I is terminal. Take δZ=⟨1Z,1Z⟩ and let !Z be the unique arrow to I. For every h:Z→W, both δW∘h and (h⊗h)∘δZ have components (h,h), and !W∘h and !Z agree by terminality. Both ways of making three copies have components (1Z,1Z,1Z), swapping two copies preserves the diagonal, and discarding one copy gives 1Z. The diagonal on a product duplicates both coordinates, giving the formula for δX⊗Y above. Thus a symmetric monoidal category is cartesian exactly when its objects carry cocommutative comonoid structures that are natural in every arrow and compatible with the tensor and unit. Di Liberti, Loregian, Nester and Sobociński, Theorem 2.22 and Corollary 2.23

Free cartesianization as a reflection

Let E be a category with finite products. Taking its binary product × as tensor and its terminal object 1E as unit makes it symmetric monoidal. This defines the inclusion U below. In the other direction, we can tensor any small symmetric monoidal category A with the wiring PROP D=(FinSetop,⊔).

U:Cart⟶SymMonCat,U(E)=(E,×,1E),L:SymMonCat⟶Cart,L(A)=A⊠D.

The arrows in Cart are finite-product-preserving functors, and its 2-cells are natural transformations. In SymMonCat we use strong symmetric monoidal functors and monoidal natural transformations.

Fixing the singleton 1 in the universal pairing gives the unit of cartesianization, a strong symmetric monoidal functor

ηA:A⟶U(L(A)),a⟼a⊠1,(f:a→b)⟼(a⊠1→f⊠id1b⊠1).

The product calculation applies to this general A just as it did to a PROP. The wirings ida⊠d2 and ida⊠d0 copy and discard a⊠1. Each a⊠n is canonically a tensor of n copies of a⊠1. Interchange makes copying and discarding natural, and the bilinear comparisons extend them compatibly to tensor words. Hence the tensor in L(A) is a product and its unit is terminal. On a strong symmetric monoidal functor H:A→B, cartesianization acts by H⊠idD:L(A)→L(B).

Why is L left adjoint to U? Suppose F:A→U(E) is strong symmetric monoidal. Its structure maps already identify F(a⊗b) with F(a)×F(b) and F(IA) with 1E. We seek a finite-product-preserving functor F―:L(A)→E such that the composite

A→ ηA U(L(A))→ U(F―) U(E)

is monoidally isomorphic to F. Once that identification is fixed, F― must send a⊠1 to F(a) and a⊠n to F(a)×n up to the product comparison. In particular, the copy a⊠1→a⊠2 must go to the diagonal F(a)→F(a)×F(a). The discard a⊠1→a⊠0 must go to the unique map F(a)→1E.

We can now construct the extension. A wiring w:n→m selects one input for each of its m outputs. Use those same projections to define F(a)×n→F(a)×m. For f:a→b, apply F(f) on each coordinate. These two prescriptions commute because at each output both composites apply F(f) to the input chosen by w. With no outputs the composites agree by terminality. Concatenating the lists of factors gives the monoidal comparison in the wiring variable. The product comparisons of F give the comparison in the other variable, coordinate by coordinate. Both orders of comparison have the same projections, so the assignment (a,n)↦F(a)×n is bilinear. The universal property of ⊠ gives F―, unique up to a unique isomorphism compatible with the chosen identification.

A monoidal natural transformation α:F⇒F′ extends on a⊠n by taking the product of n copies of αa. Thus restriction along ηA gives an equivalence of hom-categories,

Cart(L(A),E)≃SymMonCat(A,U(E)),G⟼U(G)∘ηA.

This equivalence is natural in both variables up to coherent isomorphism. That is the meaning of the left biadjunction L⊣U. Passing to isomorphism classes of functors on both sides gives the ordinary adjunction on hom-sets. Krähmer and Mahaman, Clones from comonoids, §1, on the left adjoint

Why is this a reflection? When E already has finite products, take F to be the identity of U(E). Its extension is the counit εE:L(U(E))→E. It sends e⊠n to e×n. Going back, e↦e⊠1 preserves products because a strong monoidal functor between cartesian tensors preserves the terminal object and the projections obtained by discarding a factor.

The composite on E sends e to e×1≅e. The other composite and the identity of L(U(E)) have the same restriction up to isomorphism along the unit. By uniqueness of extension they too are isomorphic. Thus the counit is an equivalence L(U(E))≃E. This makes Cart a reflective sub-2-category of SymMonCat. For example, disjoint union is already a product in FinSetop, so D⊠D≃D.

From commutative monoids to matrices

Take the PROP (FinSet,⊔) for commutative monoids from the Eckmann–Hilton calculation. Taking E=Set in the universal property shows that the cartesianization has the same set-valued models, namely commutative monoids. In the original PROP the singleton has only its identity endomorphism. After cartesianization, copying a wire and multiplying the copies gives a new unary operation X→δXX⊗X→μX, the square x↦x2. More generally, an output in m variables is a monomial x1a1⋯xmam. Copying allows repeated variables, discarding allows exponent zero, and the empty product gives the unit.

An arrow m→n is now a tuple of n such monomials. Recording their exponents gives one row per output and one column per input. For example, (x,y)↦(x2y,y) gives (2101). Every matrix of natural numbers occurs. Evaluating on the free generators of the commutative monoid Nm distinguishes different exponent vectors, and substituting monomials multiplies their exponent matrices. Hence

(FinSet,⊔)⊠D≃Mat⁡(N).
Tensor products of Lawvere theories

A single-sorted Lawvere theory is a cartesian category whose objects are the finite powers of a chosen object, up to isomorphism. In the PROP presentation, n is the nth power of the generating wire and an arrow m→n is an n-tuple of operations in m variables. If L and T are such theories, associativity and T⊠D≃T give (L⊠T)⊠D≃L⊠T. Their tensor is thus cartesian. For any cartesian target, product-preserving functors out of L⊠T correspond to bifunctors on L×T preserving finite products separately in each variable. This is the Kronecker tensor product of Lawvere theories. Hyland and Power, §3

Each row below gives the set-valued models of such a tensor product. Here k is a field and all module actions are unital left actions. Bohmann and Szymik, §2 · Lucyshyn-Wright, §3.2, on affine operations

Two theoriesWhat the mixed operations doSet-valued models
Groups and groupsEckmann–Hilton identifies the two products, and the inverse for that product is unique.Abelian groups
Pointed sets and affine k-spacesEvery affine combination of copies of the chosen point returns that point, since its coefficients sum to 1. It serves as an origin.k-vector spaces
G-actions and H-actionsThe unary actions of g and h commute, giving the action of (g,h). (G×H)-sets
G-actions and k-vector spacesEach g acts by a linear automorphism. Extending linearly gives the action of the group algebra.k[G]-modules
R-modules and S-modulesThe two additions coincide and the two scalar actions commute, so (r⊗s)x=r(sx) defines the combined action. (R⊗ZS)-modules

No comments:

Post a Comment

Popular Posts