2
$\begingroup$

(noting $\otimes$ for composition in distributors, $\phi_f : A \nrightarrow B = B(-,f=)$ and $\phi^f : B \nrightarrow A = B(f-,=)$ the embeddings of a functor $f:A\to B$ in $Dist$, and $Dist(A,B) = [B^{op}\times A, V]$ for the category of distributors from $A \nrightarrow B$)

I am looking at 2 universal properties:

  • The UP of the weighted limit of a functor $D : A \to M $ and weights $J : A \nrightarrow B$ is, naturally for $m :1 \to M$ and $b : 1 \to B$

$$ [1,M] (m,(Lim^JD)b) \simeq [A,V] (J(b,-), M(m,D -)) $$

which can be rewritten

$$ Dist(B,M)(\phi_m\otimes\phi^b, \phi_{Lim^JD}) \simeq Dist(A,M) (\phi_m\otimes \phi^b \otimes J, \phi_D ) $$

  • The UP for the right kan extension in Dist $R : M^{op} \times B \to V$ of $M(-, D=) : M^{op}\times A \to V$ along $J:B^{op} \times A \to V$ is, for $H : M^{op}\times B \to V$

$$Dist(B, M)(H, Ran_{J}\phi_D) \simeq Dist(A,M)(H\otimes J, \phi_D)$$

When special cased to $H$ of the form $\phi_m \otimes \phi^b$, this gives the previous UP. So some $l : B \to M $ representing the right kan extension as $\phi_l \simeq Ran_{J}\phi_D: B \nrightarrow M$ is required to verify more than the UP for $l : B \to M $ to be a weighted limit.

Is there a clever way to highlight the difference between a weighted limit and a representation of a right kan extension in $Dist$, which would shed lights on the difference of nature between the two ?

(Maybe some additional property which says when a weighted limit also a representation.. ? Or some case where one exist and not the other ?)

$\endgroup$
5
  • $\begingroup$ A pointwise right extension along $j \colon A \to B$ in $\mathrm{Cat}$ is precisely a limit weighted by the corepresentable $B(1, j)$, hence a right extension along $B(1, j)$ in $\mathrm{Dist}$. Is this what you are asking? $\endgroup$
    – varkor
    Commented May 2, 2023 at 14:53
  • $\begingroup$ as far as a I understand, a pointwise kan extension is a weighted limit (a specific one, weighted by $\phi_j$). what I am trying to understand is the difference between weighted limits and (representation of) Kan extension in $Dist$, the bicategory of distributors. $\endgroup$
    – nicolas
    Commented May 2, 2023 at 14:56
  • $\begingroup$ (any limit weighted by $P$ has the property that evaluating it at a specific functor $b:X\to B$ is the limit weighted by $\phi^b \otimes P$. we get the limit formula when we choose a functor from $X=1$, to which pointwise extension also evaluate to) $\endgroup$
    – nicolas
    Commented May 2, 2023 at 15:01
  • $\begingroup$ With the corrected definition of weighted limits I don't see how a pointwise kan extension is one.. I will post a new question $\endgroup$
    – nicolas
    Commented May 3, 2023 at 8:37
  • $\begingroup$ Wood defines pointwise extensions in Abstract pro arrows I, as limits weighted by corepresentable distributors. If you look at the definition, it essentially coincides with the classical definition of pointwise right extension, e.g. here. $\endgroup$
    – varkor
    Commented May 3, 2023 at 9:13

1 Answer 1

2
$\begingroup$

I'm not sure I've followed your question exactly, so let me rephrase it to how I've understood it, and you can tell me if I've misunderstood. I shall use different letters to make sure I'm not accidentally using them inconsistently with your conventions.

For a functor $s \colon X \to Z$, I will denote by $s_* \colon X \nrightarrow Z$ the corepresentable distributor $Z({-}_2, s{-}_1)$.

The question essentially appears to be this:

What is the relationship between right (Kan) extensions in the bicategory of distributors, and weighted limits in the 2-category of small categories?

As you say, a right extension in $\mathrm{Dist}$ is defined by the following universal property: Diagram expressing the equation below $$\mathrm{Dist}(Y, Z)(h, \mathrm{Ran}_g f) \cong \mathrm{Dist}(X, Z)(h \otimes g, f)$$

Now suppose we have a distributor (the weight) $\Psi \colon X \nrightarrow Y$ and a functor $s \colon X \to Z$ (the diagram). The $\Psi$-weighted limit of $s$ is a functor $\{ \Psi, s \} \colon Y \to Z$, defined to satisfy $(\{ \Psi, s \})_* \cong \mathrm{Ran}_\Psi (s_*)$. The weighted limit thus satisfies the following universal property: $$\mathrm{Dist}(Y, Z)(h, (\{ \Psi, s \})_*) \cong \mathrm{Dist}(X, Z)(h \otimes \Psi, s_*)$$ In other words, a weighted limit is precisely a corepresentable right extension along a corepresentable distributor.

One reference I find helpful for this topic is Wood's Abstract pro arrows I, though this perspective also appears in other literature by Street, Betti, and others.

$\endgroup$
2
  • $\begingroup$ If that's the case, I used a wrong definition for weighted limit. Which only half surprises me as I was questioning the relevance of this definition. $\endgroup$
    – nicolas
    Commented May 3, 2023 at 6:57
  • $\begingroup$ Thank you for the reference $\endgroup$
    – nicolas
    Commented May 3, 2023 at 6:59

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .