Chainmail math 5: operadic right-modules and toroidal decomposition, the theory
This post is under construction. See the previous post in this series for context and content of the following post, or the bottom of the first in the series for a table of contents.
Symmetric sequences, operads, and free functors
Let \(\Sigma\) denote the groupoid \(\Sigma := \coprod_{n \in \NN} \Sigma_n\); this is equivalent to the core of the category of finite sets.
Definition 6. Fix \(\cC\) a cocomplete symmetric monoidal closed category. The *category of symmetric sequences in \(\cC\) is the functor category \(\cC^\Sigma\). These possess a symmetric monoidal product called the *composition product, computed on objects by
$$\cO \circ \cO'(n) := \coprod_{n_1 + \cdots + n_k = n} \cO(k) \otimes \bigotimes_{i} \cO'(n_i).$$
An operad is a monoid in symmetric sequences.
We could have equivalently replaced \(\Sigma\) with \(\mathbf{Fin}^{\simeq}\); we use the skeleton for the sake of the notation \(\cO(n)\).
I’ll cut to the chase; there is a free forgetful adjunction between pointed symmetric sequences and operads, given by the free monoid construction with respect to \(\circ\). There’s likewise a free-forgetful adjunction between symmetric sequences and left-modules over a given operad, as well as right-modules. Left-modules are a generalization of algebras, which are precisely left modules concentrated in degree 0.
If we view the weight grading of a symmetric sequence \(X_\bullet\) as being the “number of elements,” then we may view right-modules over \(\cO\) as an algebraic structure wherein, given an element of \(X_\bullet\) with \(n\) elements, and \(n\) elements of \(\cO\) with \(n_1,\dots,n_k\) elements, we produce an element of \(X_\bullet\) with \(\sum_i n_i\) elements.
We use this perspective to formalize toroidal composition. But first, we sketch the notion of a operadic congruence generated by a sub-symmetric sequences.
Congruences, quotients, and operads
This section will be far from self contained.
Let \(\cC\) be a finitely complete category.
A
The quotient \(X / R\) is, if it exists, the coequalizer of the diagram \(R \rightrightarrows X\).
In particular, any morphism \(S \rightarrow X\) has a kernel pair, defined as the pullback of the cospan \(S \rightarrow X \leftarrow S\), and this supports the structure of a congruence. If all quotients exist for kernel pairs, then every coequalizer is expressed as the quotient of its kernel pair; this is a nonabelian version of the axiom in abelian category that cokernels of kernels agree with kernels of cokernels.
If in fact the morphism property of being a coequalizer of some parallel pair is pullback-stable, then \(\cC\) is called regular; regular categories are the categories with pullback-stable image factorizations, yielding a form of the first isomorphism theorem: every morphism \(f:X \rightarrow Y\) factors canonically as a coequalizer followed by a monomorphism.
Claim. Let \(G \hookrightarrow X \times X\) be a subobject, for \(X\) a monoid. Then, the poset \(P\) of congruences on \(X\) containing \(G\) (ordered by containment) has a unique minimal element.
It suffices to prove that \(P\) is cofiltered and its opposite poset satisfies the conditions of Zorn’s lemma. In fact, we can prove a stronger condition: we show that \(P\) has arbitrary nonempty limits, i.e. it has arbitrary nonempty intersections.
Toroidal composition
There are symmetric sequences \(A^\vee\) of atoroidal toroidal weaves, \(A^{\Sym}\) of atoroidal weaves, \(W^\vee\) of toroidal weaves, and \(W^{\Sym}\) of weaves, in sets. Toroidal composition provides two maps of symmetric sequences:
$$W^\vee \circ W^\vee \xrightarrow{\mu^\vee}W^\vee.$$
$$W^{\Sym} \circ W^\vee \xrightarrow{\mu} W^{\Sym}.$$
We’d like to realize the first as the composition map of an operad, and the second as a structure map of an operadic right module. To do the first, we need to choice a unit. This is not too hard:
Proposition 1. \(\cW^{\vee}(1)\) is contractible; that is, there is a contractible space of one-component toroidal weaves.
To prove this, note that there is a deformation retract of \(\cW^{\vee}(1)\) onto the realization given by the core curve of the solid torus; this is a translation followed by a rotation.
Let \(\eta \in \Hom(\cO_{\operatorname{triv}}, W^\vee) \simeq W^{\vee}(1) \simeq *\) be the for \(W^\vee\).
Proposition 2. The maps \(\mu^\vee, \eta\) endow on \(W^{\vee}\) the structure of an operad.
This is more-or-less obvious, so I’ll skip it. The following proposition is also obvious, and I’ll also skip it
Proposition 3. The map \(\mu\) endows \(W^{\Sym}\) with the structure of a right module over \(W^\vee\).
These summarize the existence of algebraic structure on weaves; there’s both an operadic composition within toroidal weaves, and an operadic right-action of toroidal weaves on weaves. We next describe toroidal decomposition, which amounts to giving presentations of \(W\) and \(W^\vee\) as quotients of free objects.
Toroidal decomposition
Note that \(\cA^\vee\) is contractible by Proposition 1, so \(A^\vee\) is uniquely pointed. There is a free operad \(\Fr_{\Op}\prn{A^\vee}\) on \(A^\vee\) as a pointed object; this is the operad parameterizing trees labelled with atoroidal toroidal weaves, modulo the one-component weave \(O\) acting as unit. The operad structure on \(W^\vee\) is given by a structure map \(\Fr_{\Op}\prn{W^\vee} \rightarrow W^\vee\). This induces a map
$$\Fr_{\Op}\prn{A^\vee} \rightarrow \Fr_{\Op}\prn{W^\vee} \rightarrow W^\vee.$$
In the other direction, the previous discussion of toroidal decomposition yields a map of operads \(W^\vee \rightarrow \Fr_{\Op}\prn{A^\vee}\), after choosing a (non-canonical) decomposition of each fully toroidal weave into a tree of 2-component weaves. We’d like to realize these as maps within a split coequalizer diagram in operads; to do so, we need the technology of operadic congruence envelopes.
TODO: put discussion of operadic congruence envelopes here.
Now, let \(r^\vee_{M} \in \Fr_{\Op}\prn{A^\vee}^{\times 2}\) be the relation between the two realizations of \(M_3\) within \(\Fr_{\Op}\prn{A^\vee}\); these are both trees of height 2 with one leaf node \(M_2\) and the other \(O\), with the first making one choice of position for the \(M_2\) and the other making the other. Similarly, let \(r^\vee_O \in \Fr{|Op}\prn{A^\vee}^{\times 2}\) be the relation between the two realizations of \(3 \cdot O\).
Toroidal decomposition for toroidal weaves can be phrased as follows:
Theorem 4. The operadic cofork diagram
$$\langle r_M, r_O \rangle \rightrightarrows \Fr_{\Op}(A^\vee) \rightarrow \cW^\vee$$
is split by toroidal decomposition; hence \(\cW^\vee\) is the quotient of the operad of trees labelled by \(A^\vee\) by the realizations of \(M_3\) and \(3 \cdot O\).
Similarly, note that \(A \circ W^\vee\) is the free right \(W^\vee\)-module on \(A\). Let \(r_M\) and \(r_0\) be the relations in \(A \circ W^\vee\) corresponding with the realizations of \(M_3\) and \(3 \cdot O\). Toroidal decomposition for weaves can be phrased as follows:
Theorem 5. The right \(W^\vee\)-module cofork diagram
$$\langle r_M, r_O \rangle \rightrightarrows A \circ W^\vee \rightarrow W^\Sym$$
is split by toroidal decomposition; hence \(W^{\Sym}\) arises by affixing a single atoroidal weave to a toroidal weave, modulo the pairs of such decompositions for \(M_3\) and \(3 \cdot O\).
We then may pass from \(W^{\Sym}\) to \(W\) by taking \(\Sigma\)-orbits.
The various functors employed in this construction are summarized in the following chart: