1-category equipped with relations

A 2-category equipped with proarrows is a 2-category together with a 2-category of “proarrows” which are intended to generalize the arrows of $K$ in the same way that profunctors generalize the functors in Cat. Since profunctors are a categorification of relations, it is natural to think of decategorifying such equipments to give a structure on a 1-category that equips it with “relations”. We call this structure a *1-category equipped with relations*.

Recall that a *2-category equipped with proarrows* (aka “proarrow equipment” or “equipment”) can be defined as a certain sort of double category, with $\mathcal{V}(\underline{K}) = K$. If, in such a double category, any two squares with the same boundary are equal, we say that it is is a **(1,2)-category equipped with proarrows**, or a **(1,2)-category proarrow equipment**. This is equivalent to requiring that the 2-category of proarrows (and hence also the underlying 2-category of arrows) is locally posetal, i.e. a (1,2)-category.

For example, if $V$ is any quantale, then $V Cat$ is naturally a (1,2)-category equipped with proarrows. In particular, taking $V=\mathbb{2}$, we have a (1,2)-category proarrow equipment whose objects are preorders.

A **1-category equipped with relations** is a (1,2)-category equipped with proarrows, regarded as a double category $\underline{K}$, together with an involution $\underline{K}^{h op} \cong \underline{K}$ which is (isomorphic to) the identity on objects and (vertical) arrows. Here $\underline{K}^{h op}$ denotes the horizontal opposite of a double category obtained by reversing the horizontal (pro-)arrows but not the vertical ones. We also call this structure a **relation equipment** or a **1-category proarrow equipment**.

In particular, the definition implies that we have an involution $K \cong K^{co}$ which is the identity on objects and arrows, which for a (1,2)-category means that $K$ is actually (equivalent to) a 1-category. Note though that the 2-category of proarrows (which we now call “relations”) is still (like Rel) a (1,2)-category, not necessarily a 1-category.

For example, for any quantale $V$, the sub-2-category of $V Cat$ consisting of the *symmetric* $V$-categories (those where $A(x,y) = A(y,x)$) is a 1-category equipped with relations. In particular, for $V=\mathbb{2}$, we have the relation equipment $\underline{Rel}$ of sets, functions, and binary relations.

In general, we can think of a relation equipment as generalizing some of the properties of $\underline{Rel}$. For instance, internal relations in any regular category also form a relation equipment.

It is proven in

- Carboni, Kelly, Wood, “A 2-categorical approach to change of base and geometric morphisms, I” (PDF)

that a (1,2)-category is a cartesian bicategory precisely when it is a cartesian object in a suitable 2-category of proarrow equipments (where we make a bicategory $M$ into an equipment by taking the proarrows to be those of $M$ and the arrows to be the “maps” in $M$, i.e. the morphisms having right adjoints). Here is a rough sketch of the argument, using the double-category description of equipments.

Let $\underline{K}$ be a 1-category equipped with relations, which is a cartesian object in the 2-category of relation equipments (that is, it is a **cartesian relation equipment**). Then $\mathcal{H}(\underline{K})$ is a cartesian bicategory.

That $\underline{K}$ is a cartesian object means, in particular, that it is a pseudomonoid in the 2-category of equipments. By lifting the coherence data from arrows to representable proarrows, it follows that $\mathcal{H}(\underline{K})$ is a monoidal 2-category. Being a cartesian object also gives a cartesian product on objects and proarrows, with diagonals $\Delta\colon X\to X\times X$, and lifting these arrows to representable proarrows $\Delta_\bullet$ and $\Delta^\bullet$ gives each object a commutative monoid and comonoid structure. Now for any proarrow $\phi\colon X\to Y$, the square

$\array{X & \overset{\phi}{\to} & Y\\
^\Delta \downarrow & \Downarrow & \downarrow^\Delta\\
X\times X& \underset{\phi\times \phi}{\to} & Y\times Y}$

in $\underline{K}$ induces 2-cells, i.e. inequalities, $\Delta_\bullet \phi \le (\phi\times\phi)\Delta_\bullet$ and $\phi \Delta^\bullet \le \Delta^\bullet(\phi\times\phi)$.

A bicategory of relations is a (1,2)-category which is a cartesian bicategory, and which also satisfies some additional conditions. We can also construct this structure starting from a relation equipment.

Let $\underline{K}$ be a relation equipment satisfying the hypotheses of the previous theorem, and suppose in addition that every proarrow $\phi\colon x\nrightarrow y$ in $\underline{K}$ can be written as $f_\bullet g^\bullet$ for some (vertical) arrows $f$ and $g$. (That is, “tabulations” in a certain sense exist.) Then $\mathcal{H}(\underline{K})$ is a bicategory of relations.

We first verify the axiom $\Delta^\bullet \Delta_\bullet = 1$. Since $\Delta^\bullet \Delta_\bullet$ is the restriction of $1_{X\times X}$ along $\Delta$ on both sides, it suffices to show that

$\array{X & \overset{1_X}{\to} & X\\
^\Delta\downarrow &\Downarrow& \downarrow^\Delta\\
X\times X& \underset{1_{X\times X}}{\to} & X\times X}$

is a cartesian 2-cell in $\underline{K}$. But if we have any other square

$\array{A & \overset{\phi}{\to} & B\\
^{(f,g)}\downarrow &\Downarrow& \downarrow^{(h,k)}\\
X\times X& \underset{1_{X\times X}}{\to} & X\times X}$

then $(f,g)$ factoring through $\Delta$ means that $f=g$, and likewise $h=k$. Composing the given square with the projection

$\array{X\times X & \overset{1_{X\times X}}{\to} & X\times X\\
\downarrow &\Downarrow & \downarrow\\
X& \underset{1_X}{\to} & X}$

(which comes from being a cartesian object in $Equipments$), we obtain a square

$\array{A & \overset{\phi}{\to} & B \\
^f\downarrow &\Downarrow & \downarrow^g\\
X& \underset{1_X}{\to} & X}$

which factors the given square through the putative cartesian one. The factorization is unique since all 2-cells are unique.

We now verify the Frobenius axiom $\Delta^\bullet \Delta_\bullet = (1\times \Delta_\bullet)(\Delta^\bullet \times 1)$. Since $\Delta$ is associative, we have a square

$\array{X & \overset{1_X}{\to} & X\\
^\Delta\downarrow && \downarrow^\Delta\\
X\times X & \Downarrow & X\times X\\
^{1\times \Delta}\downarrow && \downarrow^{\Delta\times 1}\\
X\times X\times X & \underset{1_{X\times X\times X}}{\to} & X\times X\times X}$

and therefore a square

$\array{X\times X & \overset{\Delta^\bullet \Delta_\bullet}{\to} & X\times X\\
^{1\times \Delta}\downarrow && \downarrow^{\Delta\times 1}\\
X\times X\times X & \underset{1_{X\times X\times X}}{\to} & X\times X\times X}$

and it suffices to show that this is a cartesian 2-cell. So suppose given a square

$\array{A & \overset{\phi}{\to} & B\\
^{(f,g,g)}\downarrow & \Downarrow & \downarrow^{(h,h,k)}\\
X\times X\times X & \underset{1_{X\times X\times X}}{\to} & X\times X\times X.}$

The fact that $g$ and $h$ appear twice is equivalent to saying that the left and right boundaries of this square factor through $1\times\Delta$ and $\Delta\times 1$, respectively. Now by assumption, $\phi = u_\bullet v^\bullet$ for some $u\colon C\to B$ and $v\colon C\to A$. Thus our square is equivalent to one

$\array{C & \overset{1_C}{\to} & C\\
^{(f v,g v,g v)}\downarrow & \Downarrow & \downarrow^{(h u,h u,k u)}\\
X\times X\times X & \underset{1_{X\times X\times X}}{\to} & X\times X\times X.}$

But this is just a 2-cell in the vertical category $K$, which is a 1-category; hence we have $(f v,g v, g v) = (h u, h u, k u)$ and thus $f v = h u = g v = k u$. Calling their common value $m$, we thus have a composite square

$\array{C & = & C & = & C\\
^{(m,m)}\downarrow && \downarrow^{m} && \downarrow^{(m,m)}\\
X\times X & \underset{\Delta^\bullet }{\to} & X & \underset{\Delta_\bullet}{\to} & X\times X}$

(since $\Delta m = (m,m)$) which gives us the desired factorization. The other Frobenius axiom is, of course, dual.

If $\underline{K}$ is a relation equipment satisfying the hypotheses of the theorem, then $\mathcal{H}(\underline{K})$ is an allegory.

It is shown here that any bicategory of relations is an allegory.

Other attempted axiomatizations of the same idea “something that acts like the category of relations in a regular category” include:

A comparison of “regular proarrow equipments” with “regular fibrations of subobjects” is in

- Finn Lawler,
*Fibrations of predicates and bicategories of relations*, arXiv

Revised on June 29, 2015 15:11:15
by Mike Shulman
(108.247.150.134)