Just in case anyone is wondering, the “Anonymous Coward” who performed the last edits at Karoubi envelope was me (it was an accident; I’m on vacation and working from a computer different from my own).

]]>After googling a little bit, it seems that all the hits I got pointed to Karoubi envelope as a synonym of idempotent-splitting completion (even for $Ab$-enriched categories). So what can I say? It would be nice if Karoubi envelope in the literature on $Ab$-enriched categories meant the “true Karoubi envelope” as Zoran put it, but for whatever reason people don’t seem to adopt the “true” concept under that name. Does Borceux treat the general enriched case?

]]>No, I meant only finite direct sums for the $Ab$-enriched case.

]]>closure under absolute colimits, which includes direct sums and splitting of idempotents

Not only *finite* direct sums ?

I expected the answer from you – and was lucky to get it quickly!

let me try to summarize what I think you are saying

Yes, you summarized what I said that I *suspected* that it was true, but was not sure. Otherwise I would not ask (I am trying to compare many external references, but it is hard to quickly find explicit all what I need in a due comparison).

Zoran, I expect you’re right that the situation is a bit of a mess, but let me try to summarize what I think you are saying and then propose a remedy:

The nLab says that the Karoubi envelope is the Cauchy completion for categories considered

*qua*$Set$-enriched categories, butUsually in mathematics, when the term “Karoubi envelope” comes up, it is understood as pertaining to $Ab$-enriched categories, where Cauchy completion means something else (closure under absolute colimits, which includes direct sums and splitting of idempotents).

I think that sounds right to me, particularly the second point. I think merging idempotent completion into Karoubi envelope is therefore a bad idea; we should have them as separate entries. (I can check once this comment is posted, but I prefer the term idempotent-splitting completion and hope that at least there’s a redirect for that. Edit: yes there is, but the redirect is to Karoubi envelope.)

Of course Cauchy completion, for which we actually have multiple entries, is the most general (pertaining to all forms of enrichment), and I think deserves a stand-alone entry.

]]>The entries Karoubian category, , Karoubi envelope, Cauchy completion confuse me quote a lot in the case of $Ab$-enriched categories (what is the default case for Karoubian envelope in the practical mathematics). If understood the terminology, the $n$Lab says that the Cauchy completion is the general case of enriched categories and Karoubi envelope is
the default for usual categories. This seems to say that in the case of Ab-enriched categories that the underlying Set-category of the Karoubian completion is the Karoubian completion of the underlying Set-category. But this seems not to be so. Borceux’s monograph’s proof of the existence of completion takes Yoneda embedding, notes that it is Karoubi complete, takes the representables and their split subobjects where one particular splitting is chosen for each such object. So, in the case of a ring viewed as a $Ab$-enriched category with one object, one would look just at projective ideals (+with splitting within the ring), while the true Karoubian envelope is the category of finitely generated projectives. Thus the Karoubi completion of $Ab$-enriched categories is completion under idempotents *and* direct sums. $Ab$-enriched category does not need to have the direct sums so it is not obvious how the general discussion implies we have to take retracts of finite coproducts ?

I moved a bunch of redirects pointing to various synonyms of “idempotent completion” from *Karoubian category* to *idempotent completion*. Okay?

I also made some slight cosmetic edits in the former entry.

]]>I added some more material to Karoubian category (e.g., answering a question about a converse statement).

Thanks!

Since $p:X\to X$ is idempotent iff $\id_X - p$ is idempotent, this is the same as saying every idempotent has a kernel.

Ah, this was precisely the point I was missing.

]]>Of course, linking between pages is the easiest thing in the world. If you see pages that ought to be linked, just put them in (as I just did, cross-linking between Karoubian category and Karoubi envelope).

While I was at it, I added some more material to Karoubian category (e.g., answering a question about a converse statement).

]]>Oh, I think that’s just a French-ism. It seems to me people close to algebraic geometry like saying ‘kernel’ instead of ‘equaliser’.

Ah, I see.

Hm... it's strange that there weren't even any links between these two pages.

??? Karoubi envelope links directly to Cauchy complete category, in more than one place. And I see that Cauchy complete category links back to Karoubi envelope as well.

Sorry, I meant between the pages Karoubian category and Karoubi envelope.

]]>Hm… it’s strange that there weren’t even any links between these two pages.

??? Karoubi envelope links directly to Cauchy complete category, in more than one place. And I see that Cauchy complete category links back to Karoubi envelope as well.

]]>This I do understand; if one can think of Ker(id, p) as the kernel of p then one can also see it as the image, since it is canonically isomorphic to Coker(id, p). What i don’t understand is in what sense Ker(id, p) is the kernel of p? (Why is a Karoubi category one where all idempotents “have a kernel”?) I thought the kernel is the object represented by the functor Y -> Ker(Hom(Y,X) -> Hom(Y,X)), or the equaliser Ker(p, 0) when C admits zero morphisms.

Oh, I think that’s just a French-ism. It seems to me people close to algebraic geometry like saying ‘kernel’ instead of ‘equaliser’.

]]>The equaliser of an idempotent and the identity is (as an object) also the coequaliser of the same pair.

This I do understand; if one can think of Ker(id, p) as the kernel of p then one can also see it as the image, since it is canonically isomorphic to Coker(id, p). What i don't understand is in what sense Ker(id, p) is the kernel of p? (Why is a Karoubi category one where all idempotents "have a kernel"?) I thought the kernel is the object represented by the functor Y -> Ker(Hom(Y,X) -> Hom(Y,X)), or the equaliser Ker(p, 0) when C admits zero morphisms.

But doesn’t this article duplicate what is at Cauchy complete category and Karoubi envelope?

Hm... it's strange that there weren't even any links between these two pages. I suppose we will have to do some reorganization.

]]>The equaliser of an idempotent and the identity is (as an object) also the coequaliser of the same pair. So you can think of it either as a quotient or as a subobject!

But doesn’t this article duplicate what is at Cauchy complete category and Karoubi envelope?

]]>Added the definitions of Karoubian category and Karoubi envelope that appear in (an exercise in) SGA 4.

A stupid question: why do they call that difference kernel the *image* of p? In what sense is it the image?