A unidetermined contramonad is a 2-monad $T : {\cal C}\to \cal C$ such that

  1. $T$ is contravariant, i.e. a contravariant endofunctor;
  2. the multiplication $\mu_A : TTA \to TA$ is determined as $T\eta_A = T(A\to TA)$.

Q(-1): Is this even a thing? Does this definition already exist under another name?

An example of such monad is the presheaf construction $P : A\mapsto [A°,Set]$; it has the Yoneda embeddings as units, and it is in fact possible to show that $T\eta_A$ acts as a multiplication, in view of the general fact that $$ [[PA°,Set]°,Set] \underset{PP\eta}{\overset{P\eta P}\leftrightarrows} [PA°,Set] $$ is an adjunction ($PP\eta\dashv P\eta P$). (notation: whenever $X$ is a large category, $[X°,Set]$ is the category of small functors).

Q0: Or is it (coming from the adjunction $P\eta\dashv \eta P$)? I'm able to find two natural transformations:

  • $\alpha : F \Rightarrow P\eta(\eta P(F))$, induced by the cowedge $$ Fa\times A(-,a)\to F$$ (or, rather, induced by the action of $F$ on arrows): $\alpha$ mates to $$ \tilde\alpha : Fa \to Set(A(-,a),F) $$ (and $\tilde\alpha$ is invertible, for that matter: Yoneda lemma).
  • $\beta : \Theta \Rightarrow \eta P(P\eta(\Theta))$, induced by the action of $\Theta$ on morphisms: its components are $$ PA(\hom(-,a),F)\to PA(\Theta F,\Theta(\hom(-a,)))$$ who mate to a family of maps $$ \tilde\beta : \Theta(F) \to PA(F, \Theta(\hom(-,a)))$$

Now... who's the unit? Who's the counit?

$P$, being a free cocompletion, is a KZ-monad. As soon as one wants to write down explicitly what this structure is, however, they have to face a few slight inaccuracies in how $P$ was defined;

  • first of all it is not only contravariant, but also partially defined; it is a so-called relative monad, like a monad but not an endofunctor. In this particular case, $P : cat^\text{coop}\to Cat$ is a monad relative to $i^\text{coop} : cat^\text{coop} \to Cat^\text{coop}$, the inclusion of small into locally small categories.

    Q1: Am I wrong if I define a contravariant (total) monad to be a contravariant endofunctor $T : {\cal C}\to \cal C$ which is relative to $1^\text{op} : {\cal C}^\text{op} \to {\cal C}^\text{op}$?

  • second, it seems to satisfy mixed properties of a lax and a colax idempotent 2-monad: in particular, it seems to me that every $P$-algebra $a :PA \to A$ is a left adjoint to the unit (so lax), and yet $\mu \dashv \eta P$ (so colax).

    Q2: Am I committing a mistake? If not, does these mixed properties have to do with the fact that $P$ is contravariant?

  • third it is "unidetermined", i.e. $\mu$ is determined by $\eta$.

    Q3: How does this affect the equations defining a KZ-monad, if at all?

  • 1
    $\begingroup$ Just a couple of quick observations: in your main example, shouldn't $\alpha$ be an isomorphism, and really the inverse of a counit? Here $\eta P_a: Pa \to PPa$ and $P\eta \circ \eta P \cong 1_P$ by a (2-)monad unit constraint. So it's really $\beta$ that's the unit. Second, to elide over "foundational" problems, you could just take your main example to be categories enriched in a quantale, like $2$ or the base $([0, \infty], +)$ of Lawvere metric spaces. I happen to like this topic a lot but have slightly different ways of considering it, which I may get to later. $\endgroup$
    – Todd Trimble
    Oct 20, 2018 at 10:32
  • $\begingroup$ Please, do not hesitate to share any thought on this, were it only a pointer to the literature. I prefer to avoid foundational problem taking my monads not to be endofunctors, but instead internal monoids in the skew-monoidal category $([X,Y],\lhd,J)$, where $F \lhd G\cong Lan_JF\circ G$ for a fixed $J : X \to Y$ (more than often a fully faithful one which is therefore a strong monoidal unit). $\endgroup$
    – fosco
    Oct 20, 2018 at 10:43
  • $\begingroup$ There won't be any pointers to the literature, unless you count any sparse unpublished scribblings of mine on "epistemologies" as "literature". $\endgroup$
    – Todd Trimble
    Oct 20, 2018 at 10:45
  • $\begingroup$ But I'd better consider more carefully your skew-monoidal remark, which sounds interesting... $\endgroup$
    – Todd Trimble
    Oct 20, 2018 at 10:46
  • $\begingroup$ I do consider them literature, and I like the name. I'll have a look at your Lab. :-) so let me say it clearly, you claim that in fact $\mu \dashv \eta P$; moreover, every $P$-algebra works as left adjoint to $\eta$, so $P$ is a lax-idempotent monad. $\endgroup$
    – fosco
    Oct 20, 2018 at 10:48

2 Answers 2


It's a little weird I guess to consider "contravariant monads"; I don't think I've seen them defined before. If one considers such a notion at just the 1-categorical level, then a question is: what should an algebra map $f: A \to B$ mean? The obvious diagram one writes down would be

$$\begin{array}{l} TA & \stackrel{a}{\to} & A \\ \uparrow \; Tf & & \downarrow\; f \\ TB & \stackrel{b}{\to} & B \end{array}$$

but commutativity of such doesn't really fit the examples you have in mind, not even up to isomorphism when we consider 2-categorical structure.

Much more telling is that morally there's both a covariant $T$ and a contravariant $T$ (which I'll denote as $T^\ast$), and what we have is $Tf \dashv T^\ast f$ for all morphisms $f: A \to B$. You are quite right that in your examples we have that $\mu_A = T^\ast \eta_A: TTA \to TA$. So in fact in the examples you are considering, we have an adjoint string

$$T\eta \dashv T^\ast \eta \dashv \eta T$$

and I think that helps to keep things straight. It fits in well with the basic theory of KZ (= lax idempotent) monads (going back to early 70's work of K = Anders Kock) where we have an adjoint string $T\eta \dashv \mu \dashv \eta T$.

To stir the pot even more, there is in your examples a second (covariant) monad which might be denoted $T^o$. By duality this carries a colax idempotent monad structure, and then there are relations between the two such as an "Isbell conjugation" map

$$T \to T^o$$

that is contravariantly adjoint to itself (this time the contravariance is at the 2-cell level). Now I think $T^o(A)$ should be a $T$-algebra and that the Isbell conjugation displayed above will be a $T$-algebra map, although I suspect that in the relative monad setting you are considering, you will not get that all $T^o$-algebras carry $T$-algebra structures (i.e., not all cototal categories are total, a known fact for ordinary category theory with classical foundations) -- although IIRC such a result did hold in the "epistemology" setting (that I was considering many years ago but which begins to fade from memory), where the key 2-monads involved are actual endofunctors.

I'm sorry if this is not completely responsive to all your questions -- the main takeaway is that there are deeper levels to your examples than I think can be reasonably accounted for with only a contramonad formalism, for example the level $T \dashv T^\ast$. Just to say one more thing in anticipation of future discussion: if it helps, I believe you can think of an epistemology as basically equivalent to a Yoneda structure in which all 1-cells are admissible, but that in my brief nLab note I was considering more of an interplay between the Kleisli bicategory $\mathbf{B}$ of $T$ (think profunctors), and its associated bicategory of maps $\text{Map}(\mathbf{B})$ (same objects as $\mathbf{B}$, but the 1-cells are left adjoints. Here $\text{Map}(\mathbf{B})$ gives back the original 2-category of the Yoneda structure, up to a notion of Morita equivalence anyway, and the inclusion $i: \text{Map}(\mathbf{B}) \hookrightarrow \mathbf{B}$ has a KZ right bi-adjoint $P$ which governs essentially all the structure.

  • $\begingroup$ I hope my yesterday's email fits my question in a sufficiently big framework to grasp where this is going :-) there's another side of the story, pretentiously aiming to give a precise connection between "all" the paradigms to do formal CT. It would be nice to fit your epistemologies in the picture! $\endgroup$
    – fosco
    Oct 21, 2018 at 17:24
  • $\begingroup$ I'm still thinking over your email (thanks! it's very interesting). I hope the answer wasn't overly harsh or unfair. There are probably lots of different frameworks that could be considered. $\endgroup$
    – Todd Trimble
    Oct 21, 2018 at 17:28
  • $\begingroup$ As you maybe noticed a few years before me, working with these gadgets is computationally extremely difficult (you lose an "op", you lose a statement). Your answer is a nice birdseye on what I'm trying to axiomatize. :-) $\endgroup$
    – fosco
    Oct 21, 2018 at 17:39

This is just a (too long) comment. As Todd Trimbe observed, there is a slightly incongruence: If $T: \mathcal{C}\to \mathcal{C}^{op}$ it isn't a proper endomorphism, then we have to interpret $T\circ T$ as $T^{op}\circ T: \mathcal{C}\to \mathcal{C}^{op}$ then how we fill the question mark in below?:

$\mu: TT\Rightarrow T: \mathcal{C}\to (?)$.

I only propose a way for make the problem formally coherent.

If $F: \mathcal{A}\to \mathcal{B}, G: \mathcal{A}\to \mathcal{B}^{op}$ in $Cat$ we define a anti-transformation $\alpha: F\Rightarrow^* G $ (marking it by "*") as a family of maps $\alpha_A: F(A) \to G(A)$ such that $\alpha_A= G(f)\circ \alpha_B\circ F(f)\ f: A\to B$, a anti-transformation $\beta: G\Rightarrow F$ has quite similar definition

(exactly is one $\beta: G^{op}\Rightarrow F^{op}$ as above, a anti-transformation is a particular case of dinatural transformation like $F\xrightarrow{ditransf.} G^{op}$)

Then I TRY to define a 2-category $Cat^\pm$ whit objects the small categories, with arrow's $\mathcal{A}\to \mathcal{B}$ of type $(F, +),\ (G, -)$ for $F$ (resp. $G$) a covariant (contravariant) functor from $\mathcal{A}$ to $\mathcal{B}$, with cell's like $(\sigma, s\cdot t): (H, s)\Rightarrow (K, t)$ that is a (usual) transformation $F\Rightarrow G$ if $s\cdot t=1$ or anti-transformation $F\Rightarrow G$ if $s\cdot t=-1$. Now the horizontal composition is defined as $(G, s)\circ (F, 1)= (G\circ F, s),\ (G, s)\circ (F, -1)= (G^{op}\circ F, -s)$, this well defined on arrows, on cell is quite similar (observing that a dual of a transformation (resp. anti-transformation) is still a transformation (resp. anti-transformation)), the vertical composition of cells is defined on components and we have that $(\tau, t)\ast (\sigma, s)= (\tau\ast \sigma, t\cdot s)$.

Edit: I see now that Godement rule failed, this structure neither is a sesquilinear category, but may be enough for working by triple.

Edit: (continue from my previous comment):

If $M$ in a monoid viewing as a one object ($\bullet$) category, let $\Gamma(M):= \bullet\downarrow M$ where a morphism is represented as $x: a\to b$ where $a, b, x\in M, x\circ a=b$, then $M$ is a group iff $M$ is a group. Let $\{1-, 1,\cdot\}$ the multiplicative couple to the additive group $Z_{(2)}$ (module $2$ numbers), and let $I^\pm:= \Gamma(\{+1, -1\})$.

I define $\pm$.category as a small category $\mathcal{C}$ with a isomorphism $\mathcal{C}\cong ||\mathcal{C}||\times I^\pm$, where I call $||\mathcal{C}||$ the category of "essential" object and morphisms of $\mathcal{C}$, then each object of $\mathcal{C}$ is representable as $(X, a)$ where $X\in ||\mathcal{C}||$, $a=\pm 1$, and a morphism like $(f, c): (X, a)\to (Y, b)$ where $f:X\to Y$ in $||\mathcal{C}||$, and $c=a\cdot b$, with composition on $||\mathcal{C}||$ and multiplication of signs, observe that given $(f, c)$ as before we have also $(f, c): (X, -a)\to (Y, -b)$.

There is the natural projection $\pi:\mathcal{C}\to I^\pm$, a $\pm$ functor between $\pm$ categories $F: \mathcal{A}\to \mathcal{B}$ is a "sign-preserving" i.e. a functor $F: ||\mathcal{A}||\to ||\mathcal{B}||$ and we write also $F: \mathcal{A}\to \mathcal{B}$ as the functor defined as $F(X, a)= (F(X), a),\ F(f, c)=(F(f), c)$.

In $Cat\downarrow I^\pm$ we have a product, this is a monoidal product of the category of signed categories and signed functors, exactly $\mathcal{A}\otimes \mathcal{B}:= \mathcal{A}\times_{I^\pm} \mathcal{B}$ as object as $(A, B, a), (A, B)\in ||\mathcal{A}\times \mathcal{B}||,\ a=\pm1$, the unitary object is just $I^\pm$ (consider $\underline{1}\times I^\pm\cong I^\pm$). Then we have the monoidal category $Cat_{\pm}$ of signed categories and signer functors. I define a $Cat_{\pm}$-category $Cat^{\pm}$ with object the some of $Cat$, where objects of $[\mathcal{A},\ \mathcal{B}]$ write as $(F, a): \mathcal{A}\to \mathcal{B}$ are functors $F: \mathcal{A}\to \mathcal{B}$ if $a=1$, or functors $F: \mathcal{A}\to \mathcal{B}^{op}$ if $a=-1$, morphism $(\sigma, s) : (F, a)\to (G, b)$ are usual transformation $\sigma: F\to G$ is $s=1$, and anti-transformation (see mine precedent comment) if $s=-1$. About the composition $[\mathcal{B},\ \mathcal{C}]\otimes [\mathcal{A},\ \mathcal{B}]\to [\mathcal{A},\ \mathcal{C}]$ define $(\tau, t)\circ (\sigma, s)$ where $(\tau, t): (G_1, b_1)\Rightarrow (G_2, b_2)$, $(\sigma, a): (F_1, a_1)\Rightarrow (F_2, a_2)$ as follow:

firs of all observing that $a_1=b_1,\ a_2=b_2$. If $s=t=1$ the definition is just the usual. If $s=-1, t=-1$ then:

if $a_1=b_1=1$ (then $b_2=a_2=-1$) let $(\tau\circ \sigma, -1)$ with components $G_1F_1(A)\xrightarrow{G_1\sigma}G_1F_2(A)\xrightarrow{\tau_{F_2A}} G_2F_2(A)$.

if $a_1=b_1=-1$ (then $a_2=b_2=-1$) let $(\tau\circ \sigma, -1)$ with components $G_1F_1(A)\xrightarrow{\tau_{F_1A}}G_2F_1(A) \xrightarrow{G_2\sigma}G_2F_2(A)$. (I seem that it work well).

The Yoneda $Y_\mathcal{C}:=[(-)^{op}, Set]: \mathcal{C}\mapsto \mathcal{C}^>$ (being natural on morhpism and cell) is a ($Cat_{\pm}$)-"endo"-functor (with obvious consideration about set's universe expansion).

  • $\begingroup$ If $F : A \to B$ and $G : A^o \to B$, you can regard them as parallel functors $F,G : A^o \times A\to B$, and now a dinatural transformation $F \overset{..}\Rightarrow G$ amounts to what you defined, right? $\endgroup$
    – fosco
    Oct 23, 2018 at 18:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.