2
$\begingroup$

For those familiar with (covariant) Galois connections, you may have noticed that they can be viewed as categorical adjunctions. A Galois connection is a pair of maps between posets $X$ and $Y$ $$ f_{\bullet}: X \rightleftharpoons Y: f^{\bullet}$$ such that $f_{\bullet} x \preceq y$ if and only if $x \preceq f^{\bullet} y$. If we view $X$ and $Y$ as small categories, then order-preserving maps are simply functors between such categories as $$Hom(f_\bullet x, y) \cong Hom(x, f^\bullet y)$$ is equivalent to the condition above (naturality is trivial).

In the case of posets, there is an explicit formula to calculate the right adjoint from the left adjoint if one exists (sim. left adjoint from right adjoint).

You can check that $$f^\bullet (y) = \bigvee f_{\bullet}^{-1}(\downarrow y)$$ is a right adjoint (if $f_\bullet$ preserves joins I believe??).

Now here is my question: Are there other areas of mathematics or other examples where an explicit formula for a right adjoint in terms of the left adjoint appears?

References most welcome. Thanks!

$\endgroup$
3
  • 6
    $\begingroup$ (No, it's $f_\bullet$ preserves sups.) Do you know the adjoint functor theorem, which generalizes this formula? ncatlab.org/nlab/show/adjoint+functor+theorem $\endgroup$
    – Todd Trimble
    Dec 27, 2019 at 20:08
  • $\begingroup$ Thanks @ToddTrimble I have read the ncatlab article you sent before but I'm afraid I got lost in some of the technical details. Do you or anyone else know a more readable reference for the Adjoint Functor Theorem? $\endgroup$
    – Hans
    Dec 28, 2019 at 15:09
  • 1
    $\begingroup$ More readable, huh. Well, I first learned it from Categories for the Working Mathematician; you could try there. It's probably in a bunch of places, but I am not intimately familiar with the more recent introductory textbooks. But (since I had a certain amount to do with the writing of the nLab article): which technical details lost you, or where did you first find yourself getting lost? $\endgroup$
    – Todd Trimble
    Dec 28, 2019 at 15:30

2 Answers 2

9
$\begingroup$

Just so that it is recorded here as an answer, here's the formula from the "naive" adjoint functor theorem that directly generalizes the one given in the post for posets:

$$f^\bullet(y) = \lim_{(x,\alpha)\in (f_\bullet \downarrow y)} x$$

where the limit is over the comma category $(f_\bullet \downarrow y)$, whose objects are pairs of $x\in X$ with $\alpha:f_\bullet(x)\to y$.

The subtlety in adjoint functor theorems comes from the fact that this comma category is of the same "size" as the categories $X$ and $Y$, while non-poset categories rarely have limits of the same size as themselves. Thus, the adjoint functor theorems used in practice impose various technical conditions enabling this limit to be replaced by a smaller one; see e.g. the nlab article for details.

$\endgroup$
2
  • $\begingroup$ Thank you . If you wouldn't mind, can you give any intuition as to how the comma category construction comes into play here? $\endgroup$
    – Hans
    Jan 1, 2020 at 2:15
  • 2
    $\begingroup$ The same way that $f_\bullet^{-1}(\downarrow y)$ appears in the poset formula. The object $f^\bullet y\in X$ comes with a counit map $f_\bullet f^\bullet y \to y$ and is universal among such objects equipped with such maps, so to construct it we take the limit over all objects $x\in X$ equipped with such a map $f_\bullet x \to y$. $\endgroup$ Jan 1, 2020 at 16:09
6
$\begingroup$

It is common for the structure of objects in a category to be corepresentable, and so the formula $$Hom(f_\bullet x, y) \cong Hom(x, f^\bullet y)$$ is a formula for the right adjoint.

As an example of how this works, in the category of groups, you have the fact that $$ |G| \cong \hom(\mathbb{Z}, G) $$ where $|G|$ means the underlying set, and that the group operation is given by the natural transformation $$ \hom(\mathbb{Z}, G) \times \hom(\mathbb{Z}, G) \cong \hom(F_2, G) \to \hom(\mathbb{Z}, G) $$ induced by the map $\mathbb{Z} \to F_2$ sending $1 \mapsto xy$, where $F_2$ is the free group on two elements $x$ and $y$. (it is the coproduct of $\mathbb{Z}$ with itself, thus the first isomorphism)

So, if $f^\bullet$ is a group-valued functor, you can compute it as follows:

  • The underlying set of $f^\bullet X$ is given by $\hom(f_\bullet \mathbb{Z}, X)$
  • The group operation is $\hom(f_\bullet \mathbb{Z}, X) \times \hom(f_\bullet \mathbb{Z}, X) \cong \hom(f_\bullet F_2, X) \to \hom(f_\bullet \mathbb{Z}, X)$

and this information suffices to determine the group; the identity and inverse can be given directly by similar formulas if desired.

$\endgroup$
2
  • $\begingroup$ One classical example I am thinking of is the suspension-loopspace adjuction. However, I think it would be impossible to say that the suspension of a space is some formula given by the loopspace of a space. But it is so for posets! I think the obstruction is what @Mike Shulman is talking about concerning small limits. $\endgroup$
    – Hans
    Jan 1, 2020 at 2:14
  • $\begingroup$ I like your answer @Anon, do you know of any more examples in other categories? $\endgroup$
    – Hans
    Jan 1, 2020 at 2:16

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.