Coherence for Categorified Operadic Theories

Let X \in \ensuremath{\mbox{\upshape{\textbf{Set}}}}, and consider the subcategory \I of \ensuremath{\mbox{\upshape{\textbf{Set}}}} whose objects are finite subsets of X and whose morphisms are inclusions. This is a directed poset, and thus a filtered category. X is the colimit of the inclusion of \I into \ensuremath{\mbox{\upshape{\textbf{Set}}}}.

See [m+m], Corollary VII.6.5.

See [johnstone], Lemma 1.7.

See [johnstone], Lemma 1.8.

Let A be a K-algebra. We define a \mathcal{T}_K-algebra F_A as follows:

• F_A {\bf n} = A^n for all n \in \natural;

• If k \in \mathcal{T}_K({\bf n}, 1) = K_n, then F_Ak = \hat k;

• if (k_1, \dots, k_n) : {\bf m} \to {\bf n} in \mathcal{T}_K, then F_A (k_1, \dots, k_n) is the unique arrow A^m \to A^n such that the diagram

  \xymatrix{ & A^m \ar[ddl]_{\widehat k_1} \ar[ddr]^{\widehat k_n} \ar@{..>}[d] \\ & A^n \ar[dl]^{\widehat{\delta^1_n}} \ar[dr]_{\widehat{\delta_n^n}} \\ A & \dots & A }

  commutes.

Let f : A \to B be a morphism of K-algebras. Then the diagram

commutes for all n \in \natural and all k \in K_n. By the universal property of B^m, the diagram

commutes for all n, m \in \natural and all (k_1, \dots, k_m) : {\bf n} \to {\bf m} in \mathcal{T}_K. Hence F_f = (f^n)_{{\bf n}} is a natural transformation F_A \to F_B, and hence a morphism of \mathcal{T}_K-algebras. This defines a functor F_{(-)} : \ensuremath{\mbox{\upshape{\textbf{Alg(K)}}}} \to \ensuremath{\mbox{\upshape{\textbf{Alg(T_K)}}}}; we wish to show that it is an equivalence.

For every \mathcal{T}_K-algebra G, we may define a K-algebra A by setting A = G{\bf 1} and \hat k = G({\bf n} \stackrel{k}{\longrightarrow} {\bf 1}) for all k \in K_n and all n \in \natural. Then F_A is isomorphic as a \mathcal{T}_K-algebra to G, and hence the functor F_{(-)} : \ensuremath{\mbox{\upshape{\textbf{Alg(K)}}}} \to \ensuremath{\mbox{\upshape{\textbf{Alg(T_K)}}}} is essentially surjective on objects. We shall show further that it is full and faithful. Let A and B be K-algebras, and let \alpha_n : F_A \to F_B be a morphism between their associated \mathcal{T}_K-algebras. Since the diagram

commutes for all n \in \natural and all i \in \{1, \dots, n\}, it must be the case that \alpha_n = \alpha_1^n for all n. Hence, the diagram

must commute for all n \in \natural and all k \in K_n. So \alpha_1 is a K-algebra morphism, and \alpha_n = F_{\alpha_1}. Hence F_{(-)} is full. Suppose F_f = F_g; then (F_f)_{\bf 1} = (F_g)_{\bf 1}, so f = g. Hence F_{(-)} is faithful; and hence it is an equivalence of categories.

Since every object in a Lawvere theory is a copower of {\bf 1}, a Lawvere theory \mathcal{T} is entirely determined (up to isomorphism) by the hom-sets \mathcal{T}({\bf n}, {\bf 1}), and thus by K_\mathcal{T}. The theorem follows straightforwardly.

See [a+r], Theorems 3.18 and 1.5, and Remarks 3.4(4) and 3.6(6).

Using the symbol * for the unique object, let P_n = P(*, \dots, * ; *), where the input is repeated n times. The rest of the conditions follow trivially from the definition of a multicategory.

Trivial.

We shall construct a functor K_{(-)} : {\ensuremath{\mbox{\upshape{\textbf{FP-Operad}}}}} \to \ensuremath{\mbox{\upshape{\textbf{Clone}}}}, and show that it is bijective on objects, full and faithful.

If P is a finite product operad, let K_P be the following clone:

• (K_P)_n = P_n for all n \in \natural,

• composition is given by composition in P: if p \in P_n and p_1, \dots, p_n \in P_m, then p\bullet(p_1, \dots, p_n) \in (K_P)_m is f \cdot (p\circ(p_1, \dots, p_n)) \in P_m, where

  \begin{array}{lccl} f : &{\underline{nm}} &\to & {\underline{m}} \\ & x & \mapsto & ((x - 1) \mod m) + 1, \end{array}

• for all n \in \natural and all i \in {\underline{n}}, the projection \delta^i_n is f^i_n \cdot 1, where

  \begin{array}{lccl} f^i_n : &{\underline{1}} &\to & {\underline{n}} \\ & 1 & \mapsto & i. \end{array}

It is easily checked that K_P satisfies the axioms for a clone given in Definition 1.2.1.

On morphisms, K_{(-)} acts trivially: morphisms of clones and of finite product operads are simply maps of signatures commuting with the extra structure, and K_{(-)} preserves the underlying map of signatures.

Let K be a clone. Let P_K be the finite product operad for which

• (P_K)_n = K_n for all n \in \natural,

• 1 = \delta^1_1,

• p \circ (p_1, \dots, p_n) = p \bullet (p_1 \bullet (\delta^1_m, \dots, \delta^{k_1}_{\sum k_i}), \dots, p_n \bullet (\delta^{k_1 + \dots + k_{n-1} + 1}_{\sum k_i}, \dots, \delta^{\sum k_i}_{\sum k_i})) for all n and k_1, \dots, k_n \in \natural, all p \in K_n, and all p_1 \in K_{k_1}, \dots, p_n \in K_{k_n},

• f \cdot p = p \bullet (\delta^{f(1)}_m, \dots, \delta^{f(n)}_m) for all n, m \in \natural, all f : {\underline{n}} \to {\underline{m}} and all p \in K_n.

We will show that K_{P_K} = K for all K \in \ensuremath{\mbox{\upshape{\textbf{Clone}}}}, and that P_{K_P} = P for all P \in {\ensuremath{\mbox{\upshape{\textbf{FP-Operad}}}}}. Let K be a clone. Then (K_{P_K})_n = (P_K)_n = K_n for all n \in \natural. If n, m \in \natural, k \in K_n and k_1, \dots, k_n \in K_m, then the composite k\bullet(k_1, \dots, k_n) in K_{P_K} is given by the composite f \cdot (k\circ(k_1, \dots, k_n)) in P_K, where f is given by (?) above. This in turn is given by the composite

in K. By the associativity law for clones, this is equal to

which in turn may be simplified to k \bullet (k_1, \dots, k_n) as required. For every n \in \natural and every i \in {\underline{n}}, the projection \delta^i_n in K_{P_K} is given by f^i_n \cdot 1, where f^i_n is defined in (?): this in turn is given by 1 \circ (\delta^i_n) = \delta^1_1 \bullet (\delta^i_n) = \delta^i_n. Hence K_{P_K} = K.

Conversely, let P be a finite product operad we shall show that P_{K_P} = P. For every n \in \natural, the set (P_{K_P})_n is equal to P_n. The unit element is given by 1 = \delta^1_1 = f^1_1 \cdot 1 = 1. If p \in P_n and p_i \in P_{k_i} for i = 1, \dots, n, then the composite p\circ(p_1, \dots, p_n) is given by p \bullet (p_1 \bullet (\delta^1_m, \dots, \delta^{k_1}_{\sum k_i}), \dots, p_n \bullet (\delta^{k_1 + \dots + k_{n-1} + 1}_{\sum k_i}, \dots, \delta^{\sum k_i}_{\sum k_i})) in K_P, which in turn is given by (after simplification) p \circ (p_1, \dots, p_n) in P. Hence P = P_{K_P}, and K_{(-)} is bijective on objects. The reasoning above also suffices to show that K_{(-)} is well-defined on morphisms and full (since preserving a finite product operad structure amounts exactly to preserving the associated clone structure). Since the morphisms of both categories are simply maps of signatures with extra properties and K_{(-)} commutes with the forgetful functors to \ensuremath{\mbox{\upshape{\textbf{Set}}}}^\natural, then K_{(-)} is faithful. Hence K_{(-)} is an isomorphism of categories, and {\ensuremath{\mbox{\upshape{\textbf{FP-Operad}}}}} \cong \ensuremath{\mbox{\upshape{\textbf{Clone}}}}.

Let (A, (\hat{\phantom{\alpha}})) be a P-algebra. Since the elements of the finite product endomorphism operad \mathop{\rm{End}}(A) are endomorphisms of A, and composition is given by composition of morphisms, then K_{\mathop{\rm{End}}(A)} = \mathop{\rm{End}}(A), the endomorphism clone of A. Since the functor K_{(-)} : {\ensuremath{\mbox{\upshape{\textbf{FP-Operad}}}}} \to \ensuremath{\mbox{\upshape{\textbf{Clone}}}} is an isomorphism, a morphism of finite product operads P \to \mathop{\rm{End}}(A) is exactly a map of clones K_P \to \mathop{\rm{End}}(A). Hence an algebra for P is exactly an algebra for K_P. A morphism between P-algebras is a morphism between their underlying objects that commutes with \hat p for every p \in P_n and every n \in \natural; this is true iff it commutes with \hat k for every k \in (K_P)_n and every n \in \natural.

[, C] [, C] FP(, C)FP(, C) FP(, C)FP(, C) Certainly Q^* restricts in this way, since Q preserves finite products. \ensuremath{\mbox{\upshape{\textbf{FP}}}}(\Csmall, \ensuremath{\mathcal{C}}) and \ensuremath{\mbox{\upshape{\textbf{FP}}}}(\Dsmall, \ensuremath{\mathcal{C}}) are full subcategories of [\Csmall, \ensuremath{\mathcal{C}}] and [\Dsmall, \ensuremath{\mathcal{C}}], so if we can show that Q_! restricts to a functor \ensuremath{\mbox{\upshape{\textbf{FP}}}}(\Csmall, \ensuremath{\mathcal{C}}) \to \ensuremath{\mbox{\upshape{\textbf{FP}}}}(\Dsmall, \ensuremath{\mathcal{C}}), then it is automatically left adjoint to the restriction of Q^*.

Let X : \Csmall \to \ensuremath{\mathcal{C}} preserve finite products. We must show that Q_! X : \Dsmall \to \ensuremath{\mathcal{C}} preserves finite products. We shall proceed by showing that Q_! X preserves terminal objects and binary products.

Recall that

for all b \in \Dsmall. Hence, using 1 for the terminal objects in \Dsmall and \ensuremath{\mathcal{C}},

since X preserves finite products and the colimit of a diagram D over a category with a terminal object 1 is simply D 1.

Now, let b_1, b_2 \in \Dsmall. Then

(?) is the definition of Q_!; (?), (?) and (?) are from the definition of products; (?) and (?) are applications of the Density Formula; (?), (?) and (?) use the distributivity of products over colimits in C (since C is cartesian closed), and (?) uses the fact that X preserves finite products.

So Q_! preserves terminal objects and binary products, and hence all finite products.

# s_i1#1 s_i # r_j1#1 r_j #_s_i1#1_s_i #_r_j1#1_r_j We will make use of Beck’s theorem to prove monadicity: precisely, we shall make use of the version in [catwork] VI.7.1, which states that U is monadic if it has a left adjoint and it strictly creates coequalizers for U-absolute coequalizer pairs. Recall that a functor G: \ensuremath{\mathcal{C}} \to \ensuremath{\mathcal{D}} strictly creates coequalizers for a diagram \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B} in \ensuremath{\mathcal{C}} if, for every coequalizer e: GB \to E of Gf and Gg in \ensuremath{\mathcal{D}}, there are a unique object E' in \ensuremath{\mathcal{C}} and a unique arrow e' : B \to E' such that GE' = E and Ge' = e, and moreover that e' is a coequalizer of \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B}.

Let \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B} be a U-absolute coequalizer pair in \ensuremath{\mbox{\upshape{\textbf{Alg(T)}}}}, and e: U B \to E be the coequalizer of \xymatrix{UA \ar@<0.5ex>[r]^-{Uf} \ar@<-0.5ex>[r]_-{Ug} & UB}. We wish to extend E to a functor E': T \to \ensuremath{\mathcal{C}}. Define E' to be equal to E on objects. On arrows, we shall define E' using the universal property of E and the U-absolute property of \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B}.

For each arrow \phi : s_1 \times \dots \times s_n \to r_1 \times \dots \times r_m in T (where s_i, r_j \in S), consider the diagram

in \ensuremath{\mathcal{C}}, where e: UB \to E is a coequalizer for \xymatrix{UA \ar@<0.5ex>[r]^-{Uf} \ar@<-0.5ex>[r]_-{Ug} & UB}.

Since \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B} is a U-absolute coequalizer pair, \prod e_{s_i} : \prod UB s_i \to \prod E s_i is a coequalizer. Since f and g are T-homomorphisms, (?) serially commutes, so (\prod e_{r_j})\phi factors uniquely through \prod e_{s_i}. Define E'\phi to be this map, as shown (and note that E'\phi = E\phi if \phi is in T'). This definition straightforwardly makes E' into a functor T \to \ensuremath{\mathcal{C}}. Since E is a T'-algebra, and products in T are the same as products in T', we may deduce that E': T\to C preserves finite products, and thus is a T-algebra. Clearly, E' is the unique extension of E to a T-algebra such that e is a T-algebra morphism. It remains to show that e is a coequalizer map for \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B} in \ensuremath{\mbox{\upshape{\textbf{Alg(T)}}}}.

Suppose \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B \ar[r]^-{d} & D } is a fork in \ensuremath{\mbox{\upshape{\textbf{Alg(T)}}}}. Then \xymatrix{UA \ar@<0.5ex>[r]^-{Uf} \ar@<-0.5ex>[r]_-{Ug} & UB \ar[r]^-{Ud} & UD } is a fork in \ensuremath{\mbox{\upshape{\textbf{Alg(T’)}}}}, so Ud factors through e; say Ud = he. We must show that h is a T-homomorphism. As before, take \phi : s_1 \times \dots \times s_n \to r_1 \times \dots \times r_m in T, and consider the diagram

We must show that the curved square on the far right commutes. Now (D\phi) \circ (\prod d_{s_i}) = (D\phi) \circ (\prod h_{s_i}) \circ (\prod e_{s_i}), and (\prod Ud_{r_j})\circ(UB\phi) = h\circ e\circ(UB\phi) = h \circ (E\phi) \circ (\prod e_{s_i}), since e is a T-algebra homomorphism. But (\prod Ud_{r_j})\circ \phi = \phi \circ (\prod d_i), so D\phi \circ (\prod h_{s_i}) \circ (\prod e_{s_i}) = h \circ (E\phi) \circ (\prod e_{s_i}). And \prod e_{s_i} is (regular) epic, so h \circ (E\phi) = (D\phi) \circ (\prod h_{s_i}).

So h is a T-algebra homomorphism. Hence U strictly creates coequalizers for U-absolute coequalizer pairs, and hence is monadic.

Each category mentioned is a category of algebras for some \natural-sorted theory, and the monadicity of each adjunction mentioned is obtained by a simple application of Lemma 2.5.3. For instance, symmetric operads are algebras for the theory presented by

• operations: one of the appropriate arity for each composition operation in Definition 2.2.5, and an operation \sigma \cdot - for each n \in \natural and each \sigma in S_n.

• equations: one for each instance of the axioms in Definition 2.2.5, and an equation (\sigma \cdot -)\circ (\rho \cdot -) = \sigma\rho \cdot - for each \sigma, \rho \in S_n and every n \in \natural.

We proceed by checking the triangle identities. We require to show that

commute. For (Proof), we proceed by induction on trees. We shall suppress all subscripts on natural transformations in the interest of legibility. For the base case:

For the inductive step, let \Phi be a signature, \phi be an n-ary element of \Phi, and \tau_1, \dots, \tau_n be trees labelled by \Phi. Then:

Hence \epsilon{\ensuremath{F_{\rm pl}}} \circ {\ensuremath{F_{\rm pl}}}\eta = 1{\ensuremath{F_{\rm pl}}}, as required.

For (Proof), let P be a plain operad, and let p be an n-aryarrow in P.

So {\ensuremath{U^{\rm pl}}}\epsilon \circ \eta{\ensuremath{U^{\rm pl}}} = 1_{\ensuremath{U^{\rm pl}}}, as required.

As before, we proceed by checking the triangle identities. First, let P be a plain operad, p be an n-ary arrow in P, and \sigma \in S_n. Then (\sigma, p) is an element of \ensuremath{\mathcal{S}} \times P.

Now let P' be a symmetric operad, and p' be an n-ary arrow in P'.

So the triangle identities are indeed satisfied, and (\ensuremath{\mathcal{S}} \times -)\dashv {\ensuremath{U^{\Sigma}_{\rm pl}}}.

Q, \pi_1, \pi_2 are given by the diagram

As a right adjoint, {\ensuremath{U^{\rm pl}}} preserves pullbacks; we take the standard construction of pullbacks in \ensuremath{\mbox{\upshape{\textbf{Set}}}}^\natural as subobjects of products, in which case h is an inclusion map. An element of ({\ensuremath{U^{\rm pl}}} Q)_n is then a pair (e_1, e_2) of n-ary strongly regular {\ensuremath{U^{\Sigma}}} P trees such that \epsilon(e_1) = \epsilon(e_2). Hence, Q is a signature of plain-operadic {\ensuremath{U^{\Sigma}}} P-equations satisfied by P. Conversely, let E be a signature of plain-operadic {\ensuremath{U^{\Sigma}}} P-equations satisfied by P, and let (e_1, e_2) be an element of E_n: then \epsilon(e_1) = \epsilon(e_2) and so (e_1, e_2) is an element of ({\ensuremath{U^{\rm pl}}} Q)_n.

See [acc] 20.15.

Suppose \xymatrix{W \ar@<0.5ex>[r]^-{fe} \ar@<-0.5ex>[r]_-{ge} & Y \ar[r]^-{i} & A } is a fork. Then ife = ige, so if = ig since e is epi. So \xymatrix{X \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & Y \ar[r]^-{i} & A } is a fork, and hence i factors uniquely through h.

Let \ensuremath{\mathcal{C}} have all kernel pairs, and \xymatrix{A \ar@<0.5ex>[r]^-{f} \ar@<-0.5ex>[r]_-{g} & B \ar[r]^-{e} & C } be a coequalizer diagram in \ensuremath{\mathcal{C}}. Let \xymatrix{W \ar@<0.5ex>[r]^-{p} \ar@<-0.5ex>[r]_-{q} & B} be the kernel pair of e. We will show that \xymatrix{W \ar@<0.5ex>[r]^-{p} \ar@<-0.5ex>[r]_-{q} & B \ar[r]^-{e} & C } is a coequalizer diagram. Since ef = eg, we may uniquely factor (f,g) through W:

Suppose \xymatrix{W \ar@<0.5ex>[r]^-{p} \ar@<-0.5ex>[r]_-{q} & B \ar[r]^-{h} & D } is a fork. hp = hq, so hpi = hqi, so hf = hg. By the universal property of e, we may factor h uniquely through e. So \xymatrix{W \ar@<0.5ex>[r]^-{p} \ar@<-0.5ex>[r]_-{q} & B \ar[r]^-{e} & C } is a coequalizer diagram, as required.

Let \epsilon' be the unit of the adjunction {\ensuremath{F_{\Sigma}^{\rm pl}}} \dashv {\ensuremath{U^{\Sigma}_{\rm pl}}}. This adjunction is monadic, so \epsilon_{{\ensuremath{U^{\Sigma}_{\rm pl}}} P} is regular epi by Lemma 2.7.7. By Lemma 2.7.9, \epsilon_{{\ensuremath{U^{\Sigma}_{\rm pl}}} P} is the coequalizer of its kernel pair, i.e.

is a coequalizer diagram.

The adjunction {\ensuremath{F_{\rm pl}}} \dashv {\ensuremath{U^{\rm pl}}} is monadic, so if \epsilon' is its counit, then \epsilon'_Q : {\ensuremath{F_{\rm pl}}}{\ensuremath{U^{\rm pl}}} Q \to Q is (regular) epi by Lemma 2.7.7. Hence, by Lemma 2.7.8, _# ’_Q1_#1 ’_Q

is a coequalizer. But \pi_{1} \epsilon'_Q = \overline{{\ensuremath{U^{\rm pl}}} \pi_1}, and similarly \pi_{2} \epsilon'_Q = \overline{{\ensuremath{U^{\rm pl}}} \pi_2}. Hence, by Corollary 2.7.6, {\ensuremath{U^{\Sigma}_{\rm pl}}} P is the plain operad generated by {\ensuremath{U^{\Sigma}}} P, satisfying all plain-operadic equations true in P. Since {\ensuremath{U^{\Sigma}}} P = {\ensuremath{U^{\rm pl}}} {\ensuremath{U^{\Sigma}_{\rm pl}}} P, the n-ary operations of {\ensuremath{U^{\Sigma}_{\rm pl}}} P are exactly the n-ary operations of P.

As above.

We will show

1. if t is a \Phi-term, then \mathop{\rm{term}}(\mathop{\rm{label}}(t)\cdot \mathop{\rm{tree}}(t)) = t;

2. if (f \cdot \tau) is a finite product tree, then f = \mathop{\rm{label}}(\mathop{\rm{term}}(f \cdot \tau)) and \tau = \mathop{\rm{tree}}(\mathop{\rm{term}}(f \cdot \tau)).

(?) Let t be a \Phi-term. Let f = \mathop{\rm{label}}(t), and \tau = \mathop{\rm{tree}}(t). Then \mathop{\rm{term}}(f \cdot \tau) is \mathop{\rm{term}}(\tau, (x_{f(1)}, \dots, x_{f(n)})). We proceed by induction.

• if t = x_i, then \mathop{\rm{term}}(f\cdot \tau) is \mathop{\rm{term}}(|, (x_i)), which is x_i.

• if t = \phi (t_1, \dots, t_n), where each t_i has arity k_i, then

  \begin{aligned} \mathop{\rm{term}}(f \cdot \tau) &= &\mathop{\rm{term}}(\phi \circ (\mathop{\rm{tree}}(t_1), \dots, \mathop{\rm{tree}}(t_n)), (x_{f(1)}, \dots, x_{f(n)})) \\ & = & \phi (\mathop{\rm{term}}(\mathop{\rm{tree}}(t_1), (x_{f(1)}, \dots, x_{f(k_1)})), \dots, \\ & & \phantom{\phi (} \mathop{\rm{term}}(\mathop{\rm{tree}}(t_n), (x_{f((\sum_{i=1}^{n-1} k_i) + 1)}, \dots, x_{f(\sum_{i=1}^n k_i)}))) \\ & = & \phi (t_1, \dots, t_n) \\ & = & t. \end{aligned}

(?) Let \tau be an n-ary plain-operadic tree in \Phi, and f a function of finite sets with codomain {\underline{m}}. Let t = \mathop{\rm{term}}(f \cdot \tau). We proceed as usual by induction on \tau.

• If \tau = |, then t = x_{f(1)}; then \mathop{\rm{tree}}(t) = | = \tau and \mathop{\rm{label}}(t) is the function {\underline{1}} \to {\underline{m}} sending 1 to f(1), i.e. \mathop{\rm{label}}(t) = f.

• If \tau = \phi \circ (\tau_1, \dots, \tau_n), then

  \begin{aligned} t &= &\mathop{\rm{term}}(\phi \circ (\tau_1, \dots, \tau_n), (x_{f(1)}, \dots, x_{f(\sum k_i)})) \\ &= &\phi (\mathop{\rm{term}}(\tau_1,(x_{f(1)}, \dots, x_{f(k_1)})), \dots, \mathop{\rm{term}}(\tau_n, (x_{f((\sum_{i=1}^{n-1} k_i) + 1)}) \dots, x_{f(\sum_{i=1}^n k_i)}))) \end{aligned}

  By induction, \mathop{\rm{var}}(t) = (x_{f(1)}, \dots, x_{f(\sum k_i)}), so \mathop{\rm{label}}(t) = f, and \mathop{\rm{tree}}(t) = \phi \circ(\tau_1, \dots, \tau_n) = \tau as required.

In Lemma 2.8.2, we factored every \Phi-term t into a strongly regular tree \mathop{\rm{tree}}(t) and a labelling function \mathop{\rm{label}}(t). By definition, t is linear iff \mathop{\rm{label}}(t) is a bijection, which occurs iff t = \mathop{\rm{term}}(\sigma \cdot \tau) for some plain-operadic tree \tau and some bijection \sigma. Hence, the linear terms and permuted trees are in one-to-one correspondence. Similarly, strongly regular terms and plain-operadic trees are in one-to-one correspondence.

By definition, the presentation is linear iff \pi_1, \pi_2 factor through the signature of linear \Phi-terms. By Lemma 2.8.3, this signature is isomorphic to {\ensuremath{T_{\Sigma}}} \Phi, so we are done.

We will consider the linear case; the strongly regular case is proved analogously.

If Q is linear, then there exists a linear presentation \xymatrix{E \ar@<0.5ex>[r]^-{} \ar@<-0.5ex>[r]_-{} & {\ensuremath{F_{\rm fp}}} \Phi} for Q. We may regard E as a subobject of the signature of \Phi-equations. By assumption, E consists only of linear equations; by Lemma 2.8.3, every (s,t) \in E is (\mathop{\rm{term}}(\sigma_1 \cdot \tau_1), \mathop{\rm{term}}(\sigma_2 \cdot \tau_2)) for some pair (\sigma_1 \cdot \tau_1, \sigma_2 \cdot \tau_2) of permuted trees. So the diagram \xymatrix{E \ar@<0.5ex>[r]^-{} \ar@<-0.5ex>[r]_-{} & {\ensuremath{F_{\rm fp}}} \Phi} in FP-Operad is the image under {\ensuremath{F_{\rm fp}^{\Sigma}}} of a diagram

in \ensuremath{\mbox{\upshape{\textbf{-Operad}}}}. This diagram has a coequalizer: call it P. The functor {\ensuremath{F_{\rm fp}^{\Sigma}}} is a left adjoint, and thus preserves coequalizers: hence, Q is the image under {\ensuremath{F_{\rm fp}^{\Sigma}}} of P.

Now suppose Q = {\ensuremath{F_{\rm fp}^{\Sigma}}} P for some symmetric operad P. We may take the canonical presentation of P:

and apply {\ensuremath{F_{\rm fp}^{\Sigma}}} to it:

Since {\ensuremath{F_{\rm fp}^{\Sigma}}} is a left adjoint, it preserves coequalizers, so the transpose of this parallel pair is a presentation for Q. Take this transpose:

where \eta' is the unit of the adjunction {\ensuremath{F_{\rm fp}}} \dashv {\ensuremath{U^{\rm fp}}}, and the bars refer to transposition with respect to the adjunction {\ensuremath{F_{\Sigma}}} \dashv {\ensuremath{U^{\Sigma}}}.

This is in precisely the form required for Lemma 2.8.4.

See [hohc] 6.2.4. Let 1 be the terminal plain operad; algebras for 1 are monoids. Since 1 is terminal, every plain operad P comes equipped with a map !: P \to 1. This induces a cartesian map of monads T_! : T_P \to T_1, and T_1 is the free monoid monad.

This follows from the facts that wide pullbacks are products in slice categories and that the functor f^* : \Acat/B \to \Acat/A induced by a map f : A \to B is product-preserving.

See [c+j1], Theorem 2.6.

This can be established by a straightforward check of the definition.

“If” is obvious: we concentrate on “only if”. It is enough to show that \sigma^{-1} is a Q-transformation, which is to say that

commutes for all (q,a_\bullet) \in Q \circ A, and this follows from the fact that {\sigma_{h(q,a_\bullet)}} \circ \phi = \psi \circ {h(q,\sigma_{a_\bullet})}.

This follows from Lemmas 2.5.1 and 2.5.3, via an application of the argument of Theorem 2.5.4.

This is established in [acc], Proposition 20.24: however, we shall provide a proof for the reader’s convenience. We shall establish the axioms listed in Definition 3.0.2.

?. Take an algebra map

Applying axiom ? to the factorization system (\ensuremath{\mathcal{E}}, \ensuremath{\mathcal{M}}), we obtain a decomposition f = m e, where e: A \to I and m: I \to B. We wish to lift this to a decomposition of f as an algebra map. In other words, we need a map i: TI \to I making the diagram

commute, such that (I,i) is a T-algebra. Since T preserves E-arrows, T e \mathop{\bot} m, and we may obtain i by applying this orthogonality to the diagram

It remains to show that (I,i) is a T-algebra. For the unit axiom, consider the diagram

The top squares commute by naturality, and the outside triangles commute since (A,a) and (B,b) are T-algebras. Hence the diagram

commutes if the dotted arrow is either 1_I or i\eta_I. By orthogonality, i\eta_I = 1_I.

For the multiplication axiom, observe that the diagrams

both commute. So the diagram

commutes if we take the dotted arrow to be either i\mu_I or i(Ti). By orthogonality, i\mu_I = i(Ti).

?. The image under U of an isomorphism in \ensuremath{\mathcal{C}}^T is an isomorphism in C. The class E contains all isomorphisms in C, so \overline \ensuremath{\mathcal{E}} = U^{-1}(\ensuremath{\mathcal{E}}) contains all isomorphisms in \ensuremath{\mathcal{C}}^T. By similar reasoning, \overline \ensuremath{\mathcal{E}} is closed under composition with isomorphisms, and \overline \ensuremath{\mathcal{M}} also satisfies these conditions.

?. We wish to show that \overline \ensuremath{\mathcal{E}} \mathop{\bot} \overline \ensuremath{\mathcal{M}}. Take T-algebras

and algebra maps

where the first two maps are in \overline \ensuremath{\mathcal{E}} and \overline \ensuremath{\mathcal{M}} respectively. Suppose that ge = mf. Now, e \in \ensuremath{\mathcal{E}} and m \in \ensuremath{\mathcal{M}}, so e \mathop{\bot} m, and there is a unique map t in C such that

commutes. We wish to show that t is a map of T-algebras.

Consider the diagram

We wish to show that the middle square commutes: the assumptions tell us that all other squares commute. Recall that T e \mathop{\bot} m, and apply orthogonality to the square

Now,

and mti = gi by definition of t, so ti = u by uniqueness. Similarly,

and j(Tf) = j(Tt)(Te) by definition of t, so j(Tt) = u by uniqueness. Hence j(Tt) = ti, and t is a map of T-algebras.

By construction, t is unique. So e \mathop{\bot} m in \ensuremath{\mathcal{C}}^T, so \overline \ensuremath{\mathcal{E}} \mathop{\bot} \overline \ensuremath{\mathcal{M}}. All the axioms are satisfied, and so (\overline \ensuremath{\mathcal{E}}, \overline \ensuremath{\mathcal{M}}) is a factorization system on \ensuremath{\mathcal{C}}^T.

See again [acc] section 20, in particular Definition 20.21 and Proposition 20.30.

Immediate.

This is a straightforward application of Lemma 2.9.13.

First we show that \eta is a morphism of monads in the sense of Street (Definition 1.4.1). One axiom corresponds to the outside of the diagram

commuting; all the inner segments commute (the top right triangle by the unit axiom for monads), so the outside must commute. The other axiom corresponds to the diagram

and this commutes trivially. Hence, \eta is a morphism of monads 1 \to T.

Now suppose that \alpha : 1 \to T is a morphism of monads. From the unit axiom for monad morphisms, the diagram

must commute, so \eta = \alpha.

If T is isomorphic to the identity monad, then by Lemma 4.3.2 the isomorphism concerned must be \eta, so \eta must be invertible. It is readily checked that if \eta is invertible, then \eta^{-1} must be a morphism of monads, so if T is trivial then it is isomorphic to the identity monad.

Let n \in \natural, and t_1, t_2 \in {\ensuremath{\mbox{\upshape Wk_(P)}}}_n. The objects of {\ensuremath{\mbox{\upshape Wk_(P)}}}_n are the elements of ({\ensuremath{F_{\Sigma}}} \Phi)_n, by construction. Since \phi_n factors through a full functor \xymatrix{ {\ensuremath{\mbox{\upshape Wk_(P)}}}_n \ar@{(>->}[r] & (D_*P)_n } and (D_*P)_n is the discrete category on P_n, there is an arrow t_1 \to t_2 in {\ensuremath{\mbox{\upshape Wk_(P)}}}_n iff \phi(t_1) = \phi(t_2). Since this functor is also faithful, such an arrow must be unique. Hence {\ensuremath{\mbox{\upshape Wk_(P)}}}_n is a poset; it is readily checked that it is also an equivalence relation.

First note that {\ensuremath{\mbox{\upshape Wk_F_^pl (F_^pl P)}}} is well-defined: {\ensuremath{F_{\Sigma}^{\rm pl}}} is a left adjoint, and hence preserves colimits, so {\ensuremath{F_{\Sigma}^{\rm pl}}} \phi is a regular epi in \ensuremath{\mbox{\upshape{\textbf{-Operad}}}}.

{\ensuremath{\mbox{\upshape Wk_F_^pl (F_^pl P)}}} is defined by its universal property, so it is enough to show that the Cat-operad {\ensuremath{F_{\Sigma}^{\rm pl}}} ({\ensuremath{\mbox{\upshape Wk_(P)}}}) also has this property. Specifically, it is enough to show that if

is the bijective-on-objects/full-and-faithful factorization of \phi, then in the diagram

the arrow {\ensuremath{F_{\Sigma}^{\rm pl}}} b is bijective on objects and the arrow {\ensuremath{F_{\Sigma}^{\rm pl}}} f is levelwise full and faithful (note that D_* {\ensuremath{F_{\Sigma}^{\rm pl}}} = {\ensuremath{F_{\Sigma}^{\rm pl}}} D_*). But this follows straightforwardly from the explicit construction of {\ensuremath{F_{\Sigma}^{\rm pl}}} in Section ?.

A {\ensuremath{F_{\Sigma}^{\rm pl}}} \phi-weak {\ensuremath{F_{\Sigma}^{\rm pl}}} P-category is a category A and a morphism {\ensuremath{\mbox{\upshape Wk_F_^pl (F_^pl P)}}} \to \mathop{\rm{End}}(A) of symmetric Cat-operads. By Theorem 4.4.8, this is equivalent to a morphism {\ensuremath{F_{\Sigma}^{\rm pl}}} ({\ensuremath{\mbox{\upshape Wk_(P)}}}) \to \mathop{\rm{End}}(A) in \ensuremath{\mbox{\upshape{\textbf{Cat--Operad}}}}, which is equivalent by the adjunction {\ensuremath{F_{\Sigma}^{\rm pl}}} \dashv {\ensuremath{U^{\Sigma}_{\rm pl}}} to a morphism of plain Cat-operads {\ensuremath{\mbox{\upshape Wk_(P)}}} \to {\ensuremath{U^{\Sigma}_{\rm pl}}} \mathop{\rm{End}}(A). This is exactly a \phi-weak P-category.

See [catwork] XI.1.

See [catwork], p. 257.

permperm Let F_{\sigma\cdot v}(A_1, \dots, A_n) = F_v(A_{\sigma(1)}, \dots, A_{\sigma(n)}), and similarly on morphisms. Then F_{\sigma\cdot v} has the required type. We may decompose {\mbox{\upshape{can}}}_M(\sigma\cdot v, \rho\cdot w) as \mbox{\upshape{perm}}_M(\sigma,\rho)\ {\mbox{\upshape{can}}}_M(v, w), where \mbox{\upshape{perm}}_M(\sigma,\rho) : F_{\sigma\cdot v} \to F_{\rho\cdot v} is a composite of \taus.

Equation 4.6.2 and Lemma 4.6.4 together imply that the diagram

commutes. It remains to show that F_{\sigma \cdot v} is unique with this property.

Suppose that F_{\sigma\cdot v} is not unique for some \sigma \cdot v, and that there exists some natural transformation G: (\sigma \cdot v)_N(FA_1, \dots, FA_n) \to F((\sigma\cdot v)_M (A_1, \dots, A_n)), composed of tensor products of components of F_0 and F_2, such that G \neq F_{\sigma\cdot v}. Suppose further that \sigma\cdot v and G have been chosen to be a minimal counterexample, in the sense that of all such counterexamples, \sigma may be written as a product of the smallest number of transpositions. If no transpositions are used, then we have a contradiction, because then \sigma = 1_n, and Lemma 4.6.4 tells us that G = F_v. But suppose \sigma = t_1t_2\dots t_m where each t_i is a transposition: then t_1 \cdot G is a natural transformation (\sigma\cdot v)_N (FA_{t_1 1}, \dots, FA_{t_1 n}) \to F((\sigma\cdot v)_M(A_{t_1 1}, \dots, A_{t_1 n})), and thus a transformation (t_1\sigma\cdot v)_N (FA_\bullet) \to F((t_1\sigma\cdot v)_M(A_\bullet)). But t_1\sigma = t_2t_3\dots t_m, and thus (by minimality of \sigma), it must be the case that t_1 \cdot G = F_{t_1\sigma\cdot v} = t_1\cdot F_{\sigma\cdot v}. Hence G = F_{\sigma \cdot v}.

The 1-cells of {{\ensuremath{\mbox{\upshape Wk_(P)}}}} are the same as those of {\ensuremath{F_{\Sigma}}} \Phi; hence, (\hat{\phantom{\alpha}}) is entirely determined on 1-cells by a map of signatures \Phi \to {\ensuremath{U^{\Sigma}}} \mathop{\rm{End}}(\ensuremath{\mathcal{C}}), which we have given. On 2-cells, Theorem 4.6.3 and the uniqueness property of 2-cells in {{\ensuremath{\mbox{\upshape Wk_(P)}}}} tell us that if \delta_1, \delta_2 are 2-cells in {{\ensuremath{\mbox{\upshape Wk_(P)}}}}, then \widehat{\delta_1 . \delta_2} = \hat \delta_1 \otimes \hat \delta_2 = \hat \delta_1 \hat . \hat \delta_2, and \widehat{\delta_1 \delta_2} = \hat{\delta_1} \hat{\delta_2} wherever \delta_1, \delta_2 are composable. Hence, (\ensuremath{\mathcal{C}}, (\hat{\phantom{\alpha}})) is a well-defined {{\ensuremath{\mbox{\upshape Wk_(P)}}}}-algebra.

Because there is at most one 2-cell \tau_1 \to \tau_2 for any pair of 1-cells \tau_1, \tau_2 in {{\ensuremath{\mbox{\upshape Wk_(P)}}}}, all diagrams involving these commute. In particular, the axioms for a symmetric monoidal category are satisfied. The 2-cells in \mathop{\rm{End}}(\ensuremath{\mathcal{C}}) are natural transformations, so \alpha, \lambda, \rho and \tau (as images of 2-cells in {{\ensuremath{\mbox{\upshape Wk_(P)}}}} under the map (\hat{\phantom{\alpha}}) : {{\ensuremath{\mbox{\upshape Wk_(P)}}}} \to \mathop{\rm{End}}(\ensuremath{\mathcal{C}})) are natural transformations. All 2-cells in {{\ensuremath{\mbox{\upshape Wk_(P)}}}} are invertible, so \alpha, \lambda, \rho and \tau are all natural isomorphisms. Hence (\ensuremath{\mathcal{C}}, \otimes, I, \alpha, \lambda, \rho, \tau) is a symmetric monoidal category.

Let RS\ensuremath{(\ensuremath{\mathcal{C}}, \otimes, I, \alpha, \lambda, \rho, \tau)} = \ensuremath{(\ensuremath{\mathcal{C}}, \otimes', I', \alpha', \lambda', \rho', \tau')}. Their underlying categories are equal, both being \ensuremath{\mathcal{C}}.

Their underlying categories are the same. As above, (\check{\phantom{\alpha}}) is determined on objects by the values it takes on . and e: these are \otimes = \hat . and I = \hat e respectively. So (\check{\phantom{\alpha}}) = (\hat{\phantom{\alpha}}) on objects. If \delta : \tau_1 \to \tau_2, then \check \delta is the unique canonical map from \check \tau_1 \to \check \tau_2, which, by an easy induction, must be \hat \delta. So (\check{\phantom{\alpha}}) = (\hat{\phantom{\alpha}}), and hence (\ensuremath{\mathcal{C}}, (\hat{\phantom{\alpha}})) = SR(\ensuremath{\mathcal{C}}, (\hat{\phantom{\alpha}})).

Let (G, G_0, G_2) = RS(F, F_0, F_2). Then G is the underlying functor of S(F, F_0, F_2) which is F, and G_0, G_2 are both the canonical maps with the correct types given by Theorem 4.6.5: that is to say, they are F_0 and F_2 respectively.

Let (G, \gamma) = SR(F, \phi). Then G is the underlying functor of R(F, \phi), which is F. Let (F, F_0, F_2) = R(F, \phi). Each component of \gamma is then by definition the correct component of the canonical map arising from F_0, F_2 in the process described in Theorem 4.6.5. By the “uniqueness” part of the Theorem, this must be the corresponding component of \phi. Hence \gamma = \phi.

Lemmas 4.6.8 and 4.6.9 show that R and S are bijective on objects; Lemmas 4.6.10 and ? show that R and S are locally bijective on morphisms. Hence, R and S are a pair of mutually inverse isomorphisms of categories.

The proof is exactly as for the proof of existence of the factorization systems on Cat-Operad and Cat-\Sigma-Operad given in Example 3.0.10.

We adopt the notation for elements of P introduced in Example 2.3.4. Let f be the unique function {\underline{2}} \to {\underline{1}}, and let t : {\underline{2}} \to {\underline{2}} be the permutation transposing 1 and 2. Then \epsilon(f \cdot \left[1 \atop 1\right]) = [2] \in P_1. Let (\ensuremath{\mathcal{C}}, (\hat{\phantom{\alpha}})) be a Q-algebra. We shall write \hat \left[1 \atop 1\right](A, B) as A \otimes B. We may impose a symmetric monoidal category structure on \ensuremath{\mathcal{C}}, where the symmetry map is the image under (\hat{\phantom{\alpha}}) of the unique map \delta : \left[1 \atop 1\right] \to t \cdot \left[1 \atop 1\right] in Q_1. All diagrams in Q_1 commute, so in particular, the following diagram commutes:

The two unlabelled arrows are mutually inverse. Applying (\hat{\phantom{\alpha}}) to the entire diagram, and evaluating the resulting functors at A \in \ensuremath{\mathcal{C}}, we see that the following diagram commutes:

and hence \tau_{A A} = 1_{A \otimes A}.

It is clear that the action we have defined is strict and associative on objects and that 1_P acts as a unit: we must show that the action on arrows is associative. Let f_i^j:(p_i^j, a_{i\bullet}^j) \to (q_i^j, b_{i\bullet}^j), \sigma \in Q_n, and \tau_i \in Q_{k_i} for j = 1, \dots, k_i and i = 1,\dots,n. We wish to show that h'(\sigma\circ(\tau_\bullet), f_\bullet^\bullet) = h'(\sigma, h'(\tau_1, f_1^\bullet), \dots, h'(\tau_n, f_n^\bullet)).

The LHS is

The RHS is

where each h'(\tau_i, f_i^\bullet) is

So the equation holds if the following diagram commutes:

The triangles all commute because all \deltas are images of arrows in Q, and there is at most one 2-cell between any two 1-cells in Q. *+[o][F-]2 commutes by the definition of h'(\tau_i, f_i^\bullet), and *+[o][F-]1 commutes by naturality of \delta.

We want a sequence (\psi_\bullet) of natural transformations:

Let \psi_i be given by <-0.7>*!(-2,-1.2)1^# 1 <-0.7>*!(-2,-1.2)1^#1 <-0.7>*!(-2,-1.2)1^k_i <-0.7>*!(-2,-1.2)1^k_i <-0.7>*!(-2,-1.2)1^n <-0.7>*!(-2,-1.2)1^n

We must check that \psi satisfies (?) and (?) from Lemma 2.9.11. For (?):

For (?), consider the following diagram:

(?) is the outside of the diagram. *+[o][F-]1 commutes by the triangle identities. *+[o][F-]2 commutes by naturality of \eta. *+[o][F-]3 commutes since (F, \pi) is a P-functor. *+[o][F-]4 commutes by naturality of \delta. *+[o][F-]5 is the definition of \psi. Hence the whole diagram commutes, and (G, \psi) is a P-functor.

To see that (F, G, \eta, \epsilon) is a P-equivalence, it is now enough to show that \eta and \epsilon are P-transformations, since they satisfy the triangle identities by hypothesis.

Write (GF, \chi) = (G, \psi) \circ (F, \pi). We wish to show that \eta is a P-transformation (1,1) \to (GF, \chi). Each \chi_{q, a_\bullet} is the composite

Applying the definition of \psi, this is

The axiom on \eta is the outside of the diagram

*+[o][F-]1 commutes by naturality of \eta, *+[o][F-]2 commutes by naturality of \pi^{-1}, and *+[o][F-]3 commutes since G\pi \circ G\pi^{-1} = G(\pi \circ \pi^{-1}) = G1 = 1G. The triangle commutes by the triangle identities. So the whole diagram commutes, and \eta is a P-transformation. By Lemma 4.2.8, \eta^{-1} is also a P-transformation. Similarly, \epsilon and \epsilon^{-1} are P-transformations.

Let F: \ensuremath{\mbox{\upshape{\textbf{st}}}}(\ensuremath{A}) \to A be given by F(p, a_\bullet) = h(p, a_\bullet) and identification of maps. This is certainly full and faithful, and it is essentially surjective on objects because \delta^{-1}_{1_Q}: h(1_P, a) \to a is an isomorphism. By Lemma 5.1.3, it remains only to show that F is a weak P-functor.

We must find a sequence (\phi_i : h_i (1 \times F^i) \to Fh') of natural transformations satisfying equations (?) and (?) from Lemma 2.9.11. We can take (\phi_i)_{q, (p_\bullet, a_\bullet^\bullet)} = (\delta_{q \circ(p_\bullet)})_{a_\bullet^\bullet} for q \in Q_n and (p_1, a^1_\bullet), \dots, (p_n, a^n_\bullet) \in \ensuremath{\mbox{\upshape{\textbf{st}}}}(\ensuremath{A}). For (?), we must show that

All 2-cells in this equation are instances of \delta. Since there is at most one 2-cell between two 1-cells in Q, the equation holds.

For (?) to hold, we must have

Since st(A) is a strict monoidal category, \delta' = 1. Apply this observation, and the definitions of F, \phi and h'; then (?) becomes

Since there is at most one arrow between two 1-cells in Q, this diagram commutes. So (F,\phi) is a weak P-functor, and hence (by Lemma 5.1.3) an equivalence in Wk-P-Cat.

For each (A,h) \in \ensuremath{\mbox{\upshape{\textbf{Wk-P-Cat}}}}, we construct an initial object A \mathop{\stackrel{\scriptstyle{(F', \psi)}}{\longrightarrow}} \ensuremath{\mbox{\upshape{\textbf{st}}}}(\ensuremath{A}) of the comma category (A\downarrow U'), thus showing that st is functorial and that {\rm\textbf{st} } \dashv U' (and that (F', \psi) is the component of the unit at A). Let (B, h'') be a strict P-category, and (G, \gamma): A \to U'B be a weak P-functor. We must show that there is a unique strict P-functor H making the following diagram commute:

(F', \psi) is given as follows:

• If a \in A, then {F'}(a) = (1, a).

• If f: a \to a' in A then {F'}f is the lifting of h(1,f) with source (1, a) and target (1, a').

• Each \psi_{(p, a_\bullet)} is the lifting of (\delta_{1_Q})_{h(p, a_\bullet)}: h(p, a_\bullet) \to h(1, h(p, a_\bullet)) to a morphism h'(p, F'(a)_\bullet) = (p, a_\bullet) \to (1, h(p, a_\bullet)) = F'(h(p, a_\bullet)).

For commutativity of (Proof), we must have H(1, a) = G(a), and for strictness of H, we must have H(p, a_\bullet) = h''(p, H(1, a)_\bullet). These two conditions completely determine H on objects.

Now, take a morphism f:(p, a_\bullet) \to (p', a_\bullet'), which is a lifting of a morphism g: h(p,a_\bullet) \to h(p',a_\bullet') in A. Then Hf is a morphism h''(p, Ga_\bullet) \to h''(p', Ga_\bullet'): the obvious thing for it to be is the composite

and we shall show that this is in fact the only possibility. Consider the composite

in st(A). Composition in st(A) is given by composition in A, so this is equal to the lifting of \delta_{1_Q} \circ g \circ \delta_{1_Q}^{-1} = h(1, g) to a morphism (1, h(p, a_\bullet)) \to (1, h(p', a_\bullet')), namely F'g. So f = \psi^{-1} \circ F'g \circ \psi, and Hf = H\psi^{-1} \circ HF'g \circ H\psi. By commutativity of (Proof), HF' = G and H\psi = \gamma, so Hf = \gamma^{-1} \circ Gg \circ \gamma as required.

This completely defines H. So we have constructed a unique H which makes (Proof) commute and which is strict. Hence (F', \psi): A \to U'\,\ensuremath{\mbox{\upshape{\textbf{st}}}}(\ensuremath{A}) is initial in (A\downarrow U'), and so {\rm\textbf{st} } \dashv U'.

We shall construct an invertible Cat-\Sigma-operad transformation \eta:\alpha \to \beta. We form the \eta_ns as follows: for all p \in P_n, let \gamma\alpha(p) = \gamma\beta(p). Since \gamma is levelwise full, there exists an arrow (\eta_n)_p: \alpha(p) \to \beta(p) such that \gamma_n((\eta_n)_p) = 1_{\gamma\alpha(p)}. Since \gamma is levelwise full and faithful, this arrow is an isomorphism. Each \eta_n is natural because, for all n \in \natural and f: p \to q in P_n, the image under \gamma of the naturality square

is

which commutes since \gamma\alpha = \gamma\beta. Since \gamma is faithful, the naturality square commutes, and \eta_n is natural. It remains to show that the collection (\eta_n)_{n \in \natural} forms a Cat-\Sigma-operad transformation, in other words that the equations #_n #_1#1_n #1_ _k_i_k_i

hold, for all n, k_1 \dots k_n \in \natural and every \sigma \in S_n. As above, it is enough to show that the images of both sides under \gamma are equal, and this is trivially true by definition of \eta.

Let Q be the weakening of P with respect to \phi: {\ensuremath{F_{\Sigma}}} \Phi \to P. By the triangle identities, we have a commutative square

By functoriality of the factorization system, this gives rise to a unique map \chi: Q \to {\ensuremath{\mbox{\upshape Wk(P)}}} such that

commutes. We wish to find a pseudo-inverse to \chi.

Since \Sigma-Operad is monadic over \ensuremath{\mbox{\upshape{\textbf{Set}}}}^\natural, a regular epi in \Sigma-Operad is a levelwise surjection by Theorem 3.0.11. So we may choose a section \psi_n of \phi_n : ({\ensuremath{F_{\Sigma}}} \Phi)_n \to P_n for all n \in \natural. So we have a morphism \psi : {\ensuremath{U^{\Sigma}}} P \to {\ensuremath{U^{\Sigma}}} {\ensuremath{F_{\Sigma}}} \Phi in \ensuremath{\mbox{\upshape{\textbf{Set}}}}^\natural. We wish to show that

commutes. This follows from a simple transpose argument:

This induces a map

We will show that \omega is pseudo-inverse to \chi. Now,

commutes. So \xymatrix{Q\ar@<0.5ex>[r]^-{1_Q} \ar@<-0.5ex>[r]_-{\chi\omega} & Q \ar@{(>->}[r] & P} is a fork. By Lemma 5.3.1, \chi\omega \cong 1_Q, and similarly \omega\chi \cong 1_{{\ensuremath{\mbox{\upshape Wk(P)}}}}. So Q \simeq {\ensuremath{\mbox{\upshape Wk(P)}}} as a symmetric Cat-operad, as required.

Let \phi: {\ensuremath{F_{\rm pl}}}{\ensuremath{U^{\rm pl}}} P \to P be the component at P of the counit of the adjunction {\ensuremath{F_{\rm pl}}} \dashv {\ensuremath{U^{\rm pl}}}. Let \epsilon be the counit of the adjunction {\ensuremath{F_{\Sigma}^{\rm pl}}} \dashv {\ensuremath{U^{\Sigma}_{\rm pl}}}.

By Theorem 4.4.8, there is an isomorphism {\ensuremath{F_{\Sigma}^{\rm pl}}} ({\ensuremath{\mbox{\upshape Wk_(P)}}}) \cong {\ensuremath{\mbox{\upshape Wk_F_^pl (F_^pl P)}}}, and by Theorem 5.3.2, there is an equivalence {\ensuremath{\mbox{\upshape Wk_F_^pl (F_^pl P)}}} \simeq {\ensuremath{\mbox{\upshape Wk(F_^pl P)}}}. Hence {\ensuremath{F_{\Sigma}^{\rm pl}}} ({\ensuremath{\mbox{\upshape Wk(P)}}}) \simeq {\ensuremath{\mbox{\upshape Wk(F_^pl P)}}}.

\ensuremath{\mbox{\upshape{\textbf{Multicat}}}}_X is monadic over \ensuremath{\mbox{\upshape{\textbf{Multigraph}}}}_X by Lemma 2.5.1 and Theorem 2.5.3, and an object of \ensuremath{\mbox{\upshape{\textbf{Multigraph}}}}_X can be considered as an object of \ensuremath{\mbox{\upshape{\textbf{Set}}}}^Y, where Y = X \times X^*, and X^* is the free monoid on X: for each x \in X, and each sequence x_1, \dots, x_n \in X^*, there is a set of funnels x_1, \dots, x_n \to x. Hence, by 3.0.11, every regular epi in \ensuremath{\mbox{\upshape{\textbf{Multicat}}}}_X is locally surjective.

The objects functor O: \ensuremath{\mbox{\upshape{\textbf{Cat}}}} \to \ensuremath{\mbox{\upshape{\textbf{Set}}}} has both a left adjoint D and a right adjoint I. Hence O and I preserve products, and hence by Lemma 2.5.1 they induce an adjunction

Since O_* is a left adjoint, it preserves colimits, and in particular regular epis: hence, every regular epi in \ensuremath{\mbox{\upshape{\textbf{Cat-Multicat}}}}_X must be locally surjective on objects.

The symmetric case is proved analogously.

The proof is exactly as for Theorem 5.3.2.

See [bkp], Theorem 3.13.

See [bkp], Theorem 5.12.

The construction is given in Power’s paper [power], and the universal property of the algebra constructed is proved in Lack’s paper [lack].

Cat-Operad is monadic over \ensuremath{\mbox{\upshape{\textbf{Cat}}}}^\natural via one of the special monads of Theorem 6.1.8, and hence, for every plain \ensuremath{\mbox{\upshape{\textbf{Cat}}}}-operad P, the pseudo-morphism classifier of P is none other than {\ensuremath{\mbox{\upshape Wk(P)}}}. Hence, if A is a category, then a strict map of Cat-operads {\ensuremath{\mbox{\upshape Wk(P)}}} \to \mathop{\rm{End}}(A) is precisely a weak map P \to \mathop{\rm{End}}(A), or equivalently a \bar T_P-pseudo-algebra structure on A.

Recall our construction of the finite product operad whose algebras are commutative monoids in Example 2.3.4. From this, we may deduce that if A is a set, then an element of T_P A is a function A \to \natural assigning each element of A its multiplicity: in other words, a multiset of elements of A. Let (A, a, \alpha, \beta) be a pseudo-algebra for T' in Cat. Then we have a binary tensor product:

where x^1y^1 is the function A \to \natural sending x and y to 1 and all other objects of A to 0. The tensor is defined analogously on morphisms. The components of \alpha and \beta give us associator, symmetry and unit maps, and it can be shown that they satisfy the axioms for a monoidal category. However, since the function x^1y^1 is equal to the function y^1x^1 for all x, y \in A, it must be the case that x \otimes y = y \otimes x.