10
$\begingroup$

I am reading Fosco's Coend Calculus and Emily Riehl's Categorical Homotopy Theory, Riehl's book motivates it in the following way,

Abstraction 1: Classical limits in terms of cones: Cones from an object $C$ to a system $F:I\to \mathcal{C}$ are thought of as natural transformations from terminal functor, to the collection of natural transformations from the constant functor to the system. Instead of terminal functor, when one has a weight functor $W: I\to \mathbf{Sets}$, a natural transformation $\tau$ from $W$ to the functor $Hom_{\mathcal{C}}(C,F-)$ has as components set maps $\tau_i:Wi\to Hom_{\mathcal{C}}(C, Fi)$. So, each natural transformation gives a more complicated cone (there are different sets of connections from $C$ to $Fi$ for each $i$)

$$Hom_{\mathcal{C}}(C,\textstyle\lim^W F)\cong Hom_{\mathbf{Sets}^{I}}(W, Hom_{\mathcal{C}}(C, F-))$$ Each component natural transformation $\tau\in Hom_{\mathbf{Sets}^{I}}(W, Hom_{\mathcal{C}}(C, F-))$ is a set map $\tau_i:Wi\to Hom_{\mathcal{C}}(C, Fi)$, which assigns to elements of the set $Wi$ cones from $C$ to the system $F:I\to \mathcal{C}$. This is 'weighted' in the following sense,

Abstraction 2: I was thinking that $W:I\to \mathbf{Sets}$ assigns some 'generalised weightage' to the objects in the category $I$, weights measured by sets. By this I mean, when we assign weight in other branches of mathematics we are usually thinking of real numbers, and 'more weight' is based on the ordering of reals. Instead of this order relation we generalise 'more' to mean any set map.

Intuition for Limits: The intuitive way I think of limits are as the object closest from a system, closest in terms of morphisms. By that I mean any other object which can be related to the system must be farther than the limit, that is, there must exist a map from limit to this object. How do I think of weighted limits?

So, I feel weighted limit does the following: It is the object closest to the weight, in the sense that, the collection of maps $\{\lim^W F\to Fi\}$ is closest to the set $Wi$ (again in terms of morphisms) or something like this?

This all seems way too abstract for me, it felt like abstraction of things that are abstract generalizations of the intuitive things. 'Higher abstraction' of some sorts. But how does someone come up with such definitions, and why? and is my intuition correct?

In the below diagram, the weighted limit $Lim^W F$ I guess should be such that the collection of maps from $Lim^W F$ to $Fj$ is closest to $Wj$, closest in the sense that given any other object $C$, there should exist a map from $Hom(C, Fj)$ to $Wj$ enter image description here

$\endgroup$
5
  • 4
    $\begingroup$ I see weighted colimits as natural generalizations of tensor products. Hence, weighted limits as cotensor products. See my preprint on bicategorical colimits for more on this (arXiv:2001.10123). Actually when you change the notation a bit, the analogy becomes very clear. $\endgroup$ Nov 12, 2022 at 3:20
  • 2
    $\begingroup$ @MartinBrandenburg I think you should post an answer here with the detailed explanation of this. $\endgroup$ Nov 12, 2022 at 6:08
  • 4
    $\begingroup$ @SergeiAkbarov the universal property of a weighted colimit is really the universal property of a tensor product. In particular, if you see two modules M,N as Ab-functors on a ring (seeing the ring a as an Ab-category with one object) then the Ab-weighted colimit of M weighted by N is their tensor product as modules. $\endgroup$ Nov 12, 2022 at 6:56
  • $\begingroup$ @SergeiAkbarov At the moment I am not able to write a good expository account on this. Others will do better. $\endgroup$ Nov 14, 2022 at 2:31
  • $\begingroup$ @MartinBrandenburg could you give some intuition as to how tensor products and weighted limits are related? I have read some functor tensor product in terms of a coend formula, say from Riehl's book Categorical Homotopy theory. Coends I interpret as the object closest from the 'double system', i.e., as an average of some sort. $\endgroup$
    – MrPajeet
    Dec 2, 2022 at 19:07

1 Answer 1

11
$\begingroup$

In enriched category theory, weighted limits may be strictly more general than conical limits, in the sense that an enriched category with all conical limits may fail to have all weighted limits. However, in ordinary category theory (and $(\infty, 1)$-category theory too) weighted limits are the same as limits: any weighted limit can be rewritten as a conical limit in a uniform way.

Here is an analogy which might help you understand the meaning of weighted limits: if you have two elements of a group, say $x$ and $y$, you could certainly multiply them together to get another element $x y$... but you could equally well multiply them in the reverse order and get $y x$, or indeed any number of other combinations, such as $y^{-1} x^{-1}$ or $x y x^{-1} y^{-1}$ or whatever; the free group on two elements parametrises all the possible natural operations on two elements of a group. In the same way, a weight $\mathcal{I} \to \textbf{Set}$ parametrises natural operations on a diagram $\mathcal{I} \to \mathcal{C}$ where $\mathcal{C}$ is a complete category, because the category $[\mathcal{I}, \textbf{Set}]^\textrm{op}$ is the free complete category generated by a diagram of shape $\mathcal{I}$.

Specifically, if $\mathcal{C}$ is complete and $F : \mathcal{I} \to \mathcal{C}$ is a diagram, there is a unique (in the appropriate sense) limit-preserving functor $\{ {-}, F \} : [\mathcal{I}, \textbf{Set}]^\textrm{op} \to \mathcal{C}$ such that the composite with the Yoneda embedding $\mathcal{I} \to [\mathcal{I}, \textbf{Set}]^\textrm{op}$ is the given $F : \mathcal{I} \to \mathcal{C}$. Characterising $\{ W, F \}$ explicitly for each individual weight $W : \mathcal{I} \to \textbf{Set}$ leads to the "official" definition of weighted limits, which can then be applied to categories $\mathcal{C}$ not assumed to be complete. Writing each object in $[\mathcal{I}, \textbf{Set}]^\textrm{op}$ as a limit of a diagram built from the representables translates into a uniform way of writing weighted limits of any diagram $\mathcal{I} \to \mathcal{C}$ as conical limits of (other) diagrams in $\mathcal{C}$.

If you really insist on an analogy with real analysis, then perhaps I should warn that thinking of weighted limits as generalised limits is not helpful. It may be better to think of weighted limits as being analogous to functionals defined by integration against a measure ... but even this is a bit of a stretch, in my opinion.

$\endgroup$
5
  • 1
    $\begingroup$ Thank you for the answer! My preferred analogy is with analysis, as I am more familiar with it, also I feel analysis is more intuitive than algebra. I don't seek too precise analogy, something similar to how limits in categories generalize limits in analysis is sufficient $\endgroup$
    – MrPajeet
    Nov 12, 2022 at 18:08
  • 2
    $\begingroup$ In fact, limits in categories do not generalise limits in analysis. $\endgroup$
    – Zhen Lin
    Nov 13, 2022 at 0:46
  • $\begingroup$ They do. See mathoverflow.net/questions/9951/… $\endgroup$ Nov 16, 2022 at 23:57
  • 1
    $\begingroup$ @MartinBrandenburg I do not consider that to be satisfactory. I might even go as far as to say that it is nonsense. The diagram should not depend on the limit point! $\endgroup$
    – Zhen Lin
    Nov 17, 2022 at 8:26
  • $\begingroup$ I see. $\phantom{-}$ $\endgroup$ Dec 2, 2022 at 20:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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