Virtualization: species with additive inverses

It would be very useful, for many practical purposes, to have a category “like $\bf Spc$, but where there are additive inverses”. In a serendipitous turn of events, Joyal showed that this is possible.

The framework was established by André Joyal (and more recently extended by Yeong-Nan Yeh, a student of Lawvere at Buffalo) to expand the reach of combinatorial differential and functional calculus.

The idea is similar to how one obtains the integers from the semiring $\mathbb N$ with a “Grothendieck group” construction; a “virtual species” would now be a pair $(F,G)$, interpreted as a formal difference $F-G$ of species, with two of them, say $\Phi=(F,G)$ and $\Phi^\prime=(F^\prime,G^\prime)$ equivalent if and only if $F+G^\prime=F^\prime+G$.

In order for this definition to work as expected, however, it must be the case that the coproduct of species makes it into a traced monoidal category. This seems a pretty strong requirement, but it turns out that it is true!

Theorem (Joyal). The category of finite species (a finite species being an endofunctor of $\bf B$) is traced monoidal with respect to the sum operation. This ensures the presence of a natural family of “trace” morphisms

$$ {\bf B}(X+Z,Y+Z) \to {\bf B}(X,Y) $$

(natural in $X,Y$, and a cowedge in $Z$).

Proof. Finiteness is a fundamental assumption, the result being false for species taking infinite values; note also that

$$ {\bf Cat}({\bf B},{\bf B}) \cong {\bf Cat}({\bf B},{\bf Fin}) $$

thus the sum of finite species is coproduct.

Starting with a bijection $\phi: X + Z \to Y + Z$, and we want to describe a bijection $\psi: X \to Y$. Starting with an element $x \in X$, we can be in two situations:

  1. if $\phi(x) \in Y$, then stop. That will be $\psi(x)$.

  2. if $\phi(x) \in Z$, then feed that back into $\phi$, and keep looping as many times as it takes to finally land in $Y$.

The maps determined in this way is injective between sets of the same cardinality, so it’s a bijection.

So the idea is that if $S_k$ acts on $(F + H)[k] = F[k] + H[k]$, say by $g \mapsto \phi_g$, then apply this algorithm to $\phi_g$.

This allows for the

Definition. A virtual species is an object of the category obtained applying the “categorified Grothendieck group” construction to the traced category $(\bf Spc,+)$ of finite species.

Remark. The category of virtual species is a particular instance of a generalised 2-ring; by this, we mean a category $\cal K$ equipped with

Remark. Note that no 2-ring can have $\oplus=+$, because if $A+B=0$ sum to the initial object, then $A,B=0$. The 2-ring of virtual species is peculiar among 2-rings, in that the sum of species comes from the coproduct operation of regular species, but it’s not a coproduct itself.

Remark. The set $Virt$ of isomorphism classes of virtual species constitutes a commutative ring under the following operations:

Recall from lecture 6 that every virtual species $\Phi$ possesses a unique representation in reduced form expressed as $\Phi = \Phi^+ - \Phi^-$. In this representation, the species $\Phi^+$ and $\Phi^-$ are unrelated, meaning they share no common non-empty molecular subspecies.

The formal power series associated with a virtual species $\Phi = F - G$ are determined by the subtraction of the series of its representative components.

Note that all take value in rings with integer coefficients.

Remark. Operations on ordinary species are extended to the ring of virtual species through linearity and distributivity. Note that new operations, that do not make sense for ordinary species, can be defined and are nontrivial.

Theorem (Yeh 1986). Let $\mathfrak{A}$ be the set of (isomorphism classes) of atoms in $\bf Spc$. The ring $Virt$ of non-weighted virtual species is isomorphic to the ring of formal power series $\mathbb{Z}\llbracket \mathfrak{A} \rrbracket$ in an infinite number of variables, where each variable corresponds to an atomic species (such as $X, E_2, E_3, C_3, \dots$). This identifies that every virtual species has a unique molecular decomposition $\Phi = \sum_{M \in \mathfrak{M}} \phi_M M$, where the coefficients $\phi_M$ are integers. This decomposition serves as a strict generalization of the cycle index series.

Yeh also identified a fundamental relationship between the category of species and permutation groups. There exists a bijection $\Phi$ between the set of isomorphism classes of transitive $S_n$-sets (which are in one-to-one correspondence with conjugacy classes of subgroups of the symmetric group $S_n$) and the set of molecular species of degree $n$. This bijection extends to a ring isomorphism between the Burnside ring $B(S_n)$ of the symmetric group and the ring $\mathcal{V}\mathcal{S}_n$ of virtual species of degree $n$. Under this mapping, the cycle index series of a molecular species corresponds exactly to the Polya cycle polynomial of the associated permutation group.

Remark (Combinatorial Logarithm and Connected Components). Virtual species allow for a generalized definition of “connected structures” even for species that do not naturally possess them in the semi-ring $\bf Spc$. If $F$ is a species such that $F(0) = 1$, there exists a unique virtual species $\Gamma = F^c$ satisfying the equation $F = E^\Gamma = E\circ\Gamma$. This virtual species $\Gamma$ is called the combinatorial logarithm of $F$ and denoted $\log F$. For example, the species $L$ of linear orders does not have an ordinary species of connected components, but it possesses a (strictly) virtual species of connected components $L^c$ in $Virt$.

The existence of subtraction changes the nature of solutions to combinatorial differential equations.