Want to take part in these discussions? Sign in if you have an account, or apply for one below
Vanilla 1.1.10 is a product of Lussumo. More Information: Documentation, Community Support.
Moved the definition of constant functor from cone to a new page constant functor.
Asked a question about notation at constant functor
Suggestion of an addition to constant functor, not important, and not to be carried out by myself: it might add a nice touch to constant functor to add a brief discussion why
the latter giving an example that non-****-ness is not preserved under negation.
This can be seen as a strawman-like argumentation, of course, first describing that someting can be done wrong (which is true for almost everything), and then pointing out that it has not been done wrongly.
This is only to suggest a “sujet”, or to “select a target”, for another, more experienced, nLab author, not to say that this should be added to the constant functor.
Actually those definitions are different in a subtle way that has nothing to do with evil: they give different answers to whether the identity functor of the empty category is constant. See constant function.
Thanks, did not know this aspect yet.
So it seems that that constant function and constant function could instructively correspond with one another, but this would require experience, tact, time and possibly a type-theoretic touch.
It seems to me that an illustrative elementary example is the endofunctor of sending each set-map to the unique set map .
In particular it sends each set to the singleton . It is sometimes called indiscrete partition. I recently had occasion to discuss it, and was unsure what terms to use. is not a constant functor. But any two values of are isomorphic objects of . So in that sense it is essentially constant. I was sort surprised how rarely the term essentially constant functor appears to be used. In particular, it is hardly ever used on the nLab.
In a sense, is a trivial but “pointed” example contrasting a set-theoretic with a category-theoretic point of view: from the former, it seems about as wrong as wrong can be to consider IP a “constant” “function” (in a sense it is set-theoretically essentially the identity function) while from the latter it is, to the contrary, category-theoretically essentially the constant function, essentially constant to the terminal object of the category, even though the members and of two values can be arbitrarily complicated and different sets. One can see this as a reason to omit the topic, but I was recently more or less forced to consider this endofunctor by a larger machinery.
Would you consider calling an “essentially constant functor” (in analogy with e.g. “essentially surjective” functors) acceptable?
Sure, “essentially constant functor” seems like a reasonable terminology. I am surprised if there is any reason to consider your as a functor, though, since as you say categorically it is uninteresting. Can you say anything brief about in what context it arose?
I added some remarks to constant functor.
Can you say anything brief about in what context it arose?
In the context of endofunctors of .
This is a rather tautological answer to your question though; in slightly more detail: in a systematic analysis of certain objects with neither left nor right adjoints of the functor category . Incidentally, IP quite evidently does not have a right-adjoint, despite passing the terminal-object-preservation-test (as it takes singletons to singletons). Saying more would be a bit lengthy, and in particular would require more care on my part, so as to not make false statements, since I have not double checked some of my arguments yet
EDIT: in the above, the “despite” is sort-of-wrong, of course.
The necessary condition for a functor to have a right adjoint is that it preserves colimits, not that it preserves limits such as the terminal object.
Of course this functor, being naturally isomorphic to the constant functor ( being any terminal object), has a left adjoint which is , the constant functor at the empty set which is the initial object.
Let me say what I meant more explicitly: what led you to consider your functor IP as distinct from the constant functor at a one-element set? Having a left or right adjoint is a “categorical” property of a functor (invariant under natural isomorphism), so if that’s the only question you’re asking about functors, then there is no reason to consider any functor that’s essentially constant at a one-element set as different from any other such functor.
The necessary condition for a functor to have a right adjoint is that it preserves colimits, not that it preserves limits such as the terminal object.
Thanks for correcting. I misspoke within the concessive clause in #8, mixing up the two criteria, despite knowing the principles behind these criteria. (Probably the reason was haste, exacerbated by the being-a-right-adjoint-versus-having-a-right-adjoint-issue.)
Needless to say, the claim that IP does not have a right-adjoint is correct, for example since IP maps the terminal (CORRECTED: initial) object to an object not isomorphic to it.
what led you to consider your functor IP as distinct from the constant functor at a one-element set?
What led me to it, was, inter alia, having had to discuss and compare Zermelo’s and von Neumann’s models of the natural numbers.
Zermelo modelled the natural numbers by iterates of the indiscrete-partition-endofunctor.
(Or so I was often told; I never saw an original article of Zermelo where he does so.)
If one would look at Zermelo’s construction purely categorically, it would never get off the ground (or, more precisely, only make it to height ).
Zermelo modelled the natural numbers by iterating a non-constant essentially-constant endofunctor of .
Von Neumann modelled the natural numbers by iterating a non-(essentially-constant), or rather, essentially non-constant, endofunctor of . (The successor operation.)
Some details, probably needlessly.
Zermelo (is said to have) modelled the natural nubmers by the iterates of , for brevity called zermelos, while von Neumann in a letter to Zermelo from the 1920s, which I did see, modelled the natural numbers by the iterates of the endofunctor defined below.
(Here could be taken to be any set, but usually one takes to mean the empty set.)
Nowadays, zermelos appear to be deemed worse than von Neumann’s naturals in various well-known ways (zermelos larger than are not transitive sets, a zermelo larger than does not have the cardinality of the number it is intended to model, zermelos larger than are all isomorphic objects of ). Von Neumann’s naturals do not have any of these defects.
We happend to have and , the agreement of the first iterates being somewhat irritating, but and disagree ever after; and not only are they unequal, but even non-isomorphic ever after, which in of course is equivalent to having unequal cardinalities.
None of the Zermelo natural numbers larger than is a transitive set. (For example, has the element , but it does not have as a subset, whereas has both as an element and as a subset.)
Definition (the endofunctor ). Let consist of the class-function
which sends any set to the set ,
together with the class-function
which sends any set-map to the set-map
with
for each , and .
Remark. The argument of is purposefully written to the left of the functor’s name.
Remark. The tempting definition of letting be , where denotes the usual direct-image function , would not result in a functor, since then for non-surjective maps one has , so then is not an element of .
Remark. With the alternative definition of letting be , we would get an endofunctor of the wide subcategory of consisisting of surjections only. This subcategory does not have an initial object.
Remark. With the above definition, is an endofunctor of , since it is clear that for any set we have , and it can be calculated that for arbitrary set-maps we have an equality of set-maps .
Remark. does neither have a left nor a right adjoint, and this already follows from the trivial necessary conditions of having to preserve terminal resp. initial objects, because increments the cardinality of its argument. More precisely, no extension of the class-function to an endofunctor of can ever have a left or a right adjoint, already for the said reasons.
Needless to say, the claim that IP does not have a right-adjoint is correct, for example since IP maps the terminal object to an object not isomorphic to it.
Of course it doesn’t have a right adjoint (i.e., it isn’t a left adjoint), but (1) this can’t be the reason for it, since typically left adjoints don’t preserve the terminal object, and (2) what do you mean maps the terminal object to an object not isomorphic to it? Isn’t every value of IP a singleton and therefore terminal (you said so yourself in #8)?
(Maybe you meant to say “initial” instead of “terminal”?)
As for “”: a category-theorist wouldn’t particularly care how it’s coded up in ZFC, but as an endofunctor it takes a set to , the coproduct of with a terminal object . A natural numbers object is by definition the initial algebra of that endofunctor.
what do you mean IPIP maps the terminal object to an object not isomorphic
Simply chose the wrong word; I meant to write
an intention which is somewhat visible from the somewhat-wrong definite article “the” in my “the terminal object”. (Needless to say, in there is only generalized-the terminal object.
So are you saying that perhaps one reason that von Neumann’s natural numbers are preferred to Zermelo’s is that von Neumann’s are categorically sensible (as Todd says, they arise from the endofunctor whose initial algebra is the categorical natural numbers object) whereas Zermelo’s are not? I think that’s an interesting observation. Maybe it could be discussed at natural numbers, which is currently lacking any mention of either von Neumann’s or Zermelo’s definition, with a short pointer from constant functor?
Another connection could be drawn to algebraic set theory where the basic operations are a successor operation (which in this discussion is ) and a small power set operation on a poset . One defines membership by iff .
So are you saying that perhaps one reason that von Neumann’s natural numbers are preferred to Zermelo’s is that von Neumann’s are categorically sensible
Yes, or rather: certainly some of the advantages of von Neumann’s definition can be usefully described with the help of category theory. (E.g., without the concept of “isomorphism in ” one would have to take recourse to saying things like “Zermelo’s definition looks like a tally language” or the like; with it, one can say that it remains within the isomorphism class of , or rather, iterates an essentially constant functor.)
I think there are deeper things to work out and say here, especially if one allows oneself to characterize the approaches of Zermelo and von Neumann via a mixture of set-theory and category-theory.
You bring up an important new point, namely to reverse the direction of the exposition and start from category-theoretic considerations, and clearly describe how von Neumann’s definition emerges from them.
I was about to write much more on this, but realized on several occasions that the discussion would quickly veer into where I still feel out of my depth. So I write only little.
Maybe it could be discussed at natural numbers, which is currently lacking any mention of either von Neumann’s or Zermelo’s definition
I agree that it would be nice to have a discussion of it in natural numbers, but this is a touchy subject, and I will not touch it for the moment.
An expository issue is that it is easy to make a mess of it and contradict oneself, e.g. when discussing categorical advantages of von Neumann’s definition, then progressing to ordinals, the total ordering on the class of all ordinals, and so on, then arriving at the concept of limit ordinals—the latter being a concept which could be abhorrent to many readers (being a purely negative definition, via not being a successor; at least this is how I was taught it).
A deeper issue is what the natural category is for to be an endofunctor of.
vs is one of . But there is something unsatisfactory about vs.
There is something to be said for focusing on the wide subcategory of having only surjections as the morphisms. One issue with this is its lack of initial object (robbing one of a canonical point where to start iterating). I have more to say on this, but, as I said, have to learn some things more carefully before doing so.
Again, algebraic set theory studies the recursive structures of set theory from a categorical perspective.
being a purely negative definition, via not being a successor
Certainly there are category theorists who have studied constructive or intuitionistic ordinal analysis; I am not one of them. (Maybe Toby Bartels has, but we don’t see him here as often as we used to.) I don’t know what other definitions are standardly studied, but off the top of my head one could try a, well, limit definition that would say more positively that limit ordinals are precisely fixed points of a certain ordinal operation, namely the one taking a von Neumann ordinal to . (Equivalently, is limit if implies .) Classically I imagine that should be the same; constructively, I haven’t thought about it.
Ordinals make perfect sense constructively, but the notion of “limit ordinal” is indeed harder to define constructively, and also less useful, because however you define it it will no longer be the case that every ordinal is either a successor or a limit. Instead, when working with ordinals constructively you generally have to phrase your definitions so that they work for all ordinals with a single definition rather than splitting into successor/limit cases as is often done classically. For instance, when iterating an endofunctor of a complete poset transfinitely, classically one writes for a successor and for a limit, whereas constructively we write covering all cases at once.
Very useful comment, Mike; thanks.
By the way, I do agree with Todd’s point that you (Peter) would probably be interested in reading some algebraic set theory, particularly Joyal-Moerdijk’s original paper, where they identify “successor” operations such as and algebraically as endomaps of certain “free ZF-algebras”.
Thanks for the comments. This seems a very nice part of the literature, I will look into it.
Some (nLab-)technical remarks, strictly speaking off-topic (if topic is defined by this thread’s name), but clearly a natural outcome of the above discussion:
(It seems not to have one let alone a page of its own, as far as I can tell from the pages on functions, partial functions, the category of sets, etc).
If it indeed does not have a name, would suggest itself.
It could perhaps be useful to have treated systematically somewhere on the nLab, perhaps within some of the above pages.
I think I’ve seen it called .
back in 2013 I started a thread about (not ). Nothing ever came of it but you might find some of the discussion there useful.
From the perspective of finding uses of the categories and (or maybe ) in the nLab this is difficult now because with out explicit names search phrases (say “category set surjection”) turn up way to many hits.
I like myself. BTW, I take back what I said about in the discussion Rod pointed to: sometimes when a category has a standard collection of objects but one is not looking at the “default” morphisms, we do name the category after the morphisms. For example, we name the category of sets and relations , the bicategory of categories and profunctors , and so on. So and now seem fine to me.
I’m not sure quite what I’d put into if it were me. Some of these things like and have a combinatorial significance which could provide interesting material. I might start something stubby sometime in the next few days if no one else does.
The ordering of the preorder reflection of Surj is what set theorists call ≤*. The axiom WISC has an equivalent statement (that works in any Boolean topos) due to François Dorais phrased almost entirely in terms of surjections (or epimorphisms):
for every set X there is a set Y such that for every surjection q of Z on X there is a function s from Y to Z such that qs is a surjection of Y onto X.
Edit: I guess this is really a statement about the fibration over Set with fibre over X the full subcategory of Set/X on the surjections: every fibre has a weakly initial object.
I’m not sure quite what I’d put into if it were me.
To me it seems natural to proceed as follows (this is, overtly—and perhaps overly—mechanical, but nowadays one should be able to pull off such an approach routinely, and it seems a valuable routine to boot, like an exercise or “benchmark”):
to agree upon
a set of (codified, well-studied) definitions of qualities of categories. (Preferably, each quality is in some sense “decidable” from the “usual” presentation of the category under consideration, at least for “sufficiently decidable” categories.) Basic examples of qualities: having, respectively, an initial object, a terminal object, all finite products, all exponentials, all pullbacks, all small limits, being rigid, or even the ill-defined quality of having-an-opposite-category-with-a-“usual name”, etc. I know that there are interdependencies among some of these qualities.
a set of (codified, well-studied) definitions of qualities of embeddings of categories.
(Obvious examples: dense, full, reflective, wide, etc.)
a set of implications of the form
Then analyse
for which qualities it follows that Surj has because of an implication in
for which qualities it follows that Surj has as if by fluke, i.e., in general not every category with is a subcategory of a -category via a -embedding has property , but this subcategory happens to have nevertheless.
Only then try to prove whatever quality of Surj one would like to be sure of.
Maybe Surj is not complex enough to harbour any surprises when proceeding along such a road, but for some qualities I personally feel at a loss to give general reasons, and would have to resort to working from the definitions.
I expect this kind of analysis to have been done in published form on several occasions, perhaps not for , but cannot find an example.
To some extent we do this already, but a general problem there are too many “qualities” around for it to be reasonable to set benchmarks in advance, to be applied to each article whose title names a particular category.
I disfavor writing nLab articles according to checklists and pre-set molds. Personally I would find it very unpleasant both to write and to read articles which follow such mechanical routines “agreed upon” in advance, and so I can’t see myself ever agreeing to agree upon a set and a set .
Most of the time, it seems to me, nLab articles are written in a more organic fashion: if one has something useful or interesting to say to get a ball rolling, then one says it. For example, David Roberts said something interesting in #25. That alone could be a good start. (David? (-: )
I disfavor writing nLab articles according to checklists and pre-set molds. Personally I would find it very unpleasant both to write and to read articles which follow mechanical routines
That seems sensible. What is gained in “guidance” and structure by such a routine can easily be outweighed by the burdensome rigidity of it.
Most of the time, it seems to me, nLab articles are written in a more organic fashion: if one has something useful or interesting to say to get a ball rolling,
A lighweight ball that is probably easy to return is this: Set in its Section 2 quickly tries to dispel worries that this might not be “the” category of sets by a passage which seems to be saying that—under resonable conditions—all categories obtained via ETCS, MK, NBG, SEAR, ZFC are equivalent, in the category-theoretic sense. (Incidentally, isn’t this so essential that it deserves to be said even in Section 1 of Set, which after all allows itself to say more than only a “Definition”?)
Surj should probably try to do something similar, and hasten to say something about (what one might be tempted to call)
, , , , , , , .
The latter notations of course bring to mind the usual reformulations of AC via “pre”-inverses of surjections.
That is another set-theoretic issue that the article-to-be on should maybe quickly address; actually writing such a sweeping passage seems a difficult task though.
That section 2 of Set is pretty circular actually (for instance, “locally small” depends on a background notion of ), and relativizes to different models of set theory. (I don’t mean presentationally different such as the difference between ZFC and NBG; I mean what Lawvere would call “objectively” different in the sense of propositions that can be stated in terms of category theory, e.g., some categories satisfy CH, some don’t.) So there is no one “”, although the section does explain how to characterize among categories considered against the background theory.
Also, not everyone would even agree that is a topos in the Lawvere-Tierney sense; for example I think those with predicativist commitments don’t think so. So I think section 1 comes closer to expressing the complicated truth of the matter (although it seems slightly weird to me to title it “Definition”), and it’s wise to keep such a foundational discussion on a descriptive level, to accommodate the various axes (“axises”) along which people consider the category “”.
, , , , , , ,
Personally, I’d avoid writing all that. Since we agree that there is no one pretty much to the same extent as there is no one : whatever category one is calling , there is a well-defined notion of surjection, and is simply understood relative to that . For the most part the theorems will be the same regardless; if some use is made of AC or something, that can go into the hypotheses. To me it seems pointlessly heavy to draw especial attention to the fact that “” and “” are the same damn thing, insofar as the subject matter of ETCS and SEAR is the same, just presented differently.
So here would be a sample theorem written according to the style I personally favor.
Theorem: Assuming the axiom of choice, every functor maps surjections to surjections, i.e., restricts to a functor .
Proof: If is a surjection, then it admits a section according to the axiom of choice. From and functoriality, we have , so is a retraction and therefore a surjection.
whatever category one is calling , there is a well-defined notion of surjection, and is simply understood relative to that
Which is equally true with “surjection” replaced by any other object of mathematics: group, ring, field, topological space, etc.
Todd #27
I hear you. I plan to do something next week about this.
I made some edits at surjection incorporating #25.
Thanks! I put in a section heading, since it’s not part of the definition of surjection.
One also writes for all sets , but this is a standalone definition, and doesn’t follow from a construction on .
Ugh!! Do people really do that? Is there a way to define that doesn’t special-case (and hence in particular would make sense constructively)?
@Mike: sure, using subquotients, which give, in the presence of LEM, an equivalent preorder to as defined in the piecemeal fashion.
Good. I think we should give that definition, then, not the piecemeal one.
Did so, but left the set-theorists one in, but clarified how they are related.
Thanks! I did some editing; in particular I removed the piecewise proof that implies , since the subquotient one works classically too and is IMHO more elegant. But if you really want it there we can put it back.
(By the way, note that is the same as “ is not inhabited” even constructively.)
1 to 38 of 38