Talk:Conglomerate (mathematics)

This article should be deleted

The term "conglomerate" does not have rigorous definition in mathematics. The authors don't define it as an object in a known axiomatic first order theory, and don't construct a special first order theory for it. 7 years ago we discussed this at Mathoverflow, where specialists in logic explained the details. As far as I can see nothing changed since that time. I would nominate this article for deletion. Eozhik (talk) 15:11, 28 March 2019 (UTC)

Concur. Appears to be an element of a WP:fringe theory. The theory, itself, might be notable, but not individual elements. — Arthur Rubin (talk) 18:36, 22 April 2019 (UTC)

Doubts

Here is a message copied from Wikipedia:Articles for deletion/Conglomerate (set theory):

Boris, I already told this: in this article Zhang Jinwen constructs his theory as an extension by definition of GB:

In this article we develope the axiom system of conglomerates denoted as ACG. The system ACG falls into two parts, the first part is just GB, the second part is composed of five groups A,B,C,D,E.

(page 157). He just adds to GB new predicate symbols $scl$, $cog$, with a series of formulas "that define them" (and a new term, "conglomerate"). This is an extension by definition. The new formulas don't cancel the old theorems of GB, everything what was true in GB, is true in the new theory as well. In particular, proper classes still can't belong to anything. If he would define a new relation "X belongs to Y", we could understand this phrase — "proper class belongs to this object" — differently. But he does not do this (and he uses the same symbol for this relation $\in$). This means that he understands this relation in the same sense. Eozhik (talk) 07:04, 25 April 2019 (UTC)

But, as far as I understand, this is not an extension by definition. Yes, axioms of GB still hold, as well as all theorems of GB. But new objects are added, of new types "conglomerate" and "second class"; these are neither sets nor classes (and therefore never appear in axioms, nor theorems, of GB). Accordingly, there cannot be a definition of a conglomerate as "a class satisfying ...", as it should be in any extension by definition. And in fact, the extension is non-conservative (as noted in the article). Yes, he adds to GB new predicate symbols $scl$, $cog$; but they apply not only to classes. Boris Tsirelson (talk) 20:18, 25 April 2019 (UTC)

@Tsirel: Boris, there is no such an option in first order theories, "adding new objects". One can add new symbols to signature (like those $scl$ and $cog$), new letters to alphabet, new formulas to the list of axioms, but not "new objects". There is also a possibility to construct an interpretation (see J. Shoenfield, Mathematical logic, 1967, 4.4.7) of the first order theory in another first order theory (or a model what is the same up to terminology), so that this would look like some kind of extension, but Zhang Jinwen does not do this. And the same with the other autors. This is absurd. Eozhik (talk) 04:41, 26 April 2019 (UTC)

On the other hand, probably, this is an inadvertence of Zhang Jinwen. He should clarify that the first part is not "just GB"; rather, here each axiom of GB is relativized to classes via quantifiers bounded by another predicate symbols $cls$ ("to be a class"). Boris Tsirelson (talk) 20:57, 25 April 2019 (UTC)

Again a hint. Zhang Jinwen should have constructed an accurate first order theory, but he did not do this. That is his problem. Eozhik (talk) 04:41, 26 April 2019 (UTC)

Yes, somewhat a problem, but surmountable. Yes, you are right, formally there is no such an option in first order theories, "adding new objects". On the other hand, a mathematician is not a proof assistant; a human often is able to catch author's idea and interpret the text accordingly. Later, when, and if, the matter becomes well-known and enters textbooks, author's lapsus will be fixed. Don't throw the baby out with the bathwater. You are not writing the referee report to Zhang Jinwen's manuscript. Boris Tsirelson (talk) 05:31, 26 April 2019 (UTC)

No-no! Mathematicians are obliged to be accurate, otherwise what they do is not called mathematics. Moreover, this part of mathematics is called mathematical logic. It is built directly for studying proofs in mathematics, so mathematicians must be especially scrupulous about their constructions here. All the formal requirements must be met. Furthermore, since we are here, "catching someone's idea" is called in Wikipedia "original research". It is not allowed. And since we are humans, presenting somebody's absurd ideas as reliable theories is a violation of moral norms. Eozhik (talk) 06:08, 26 April 2019 (UTC)

After all, every formal theory was created after its intended interpretation. First mathematicians generate a new idea, try it informally, get some intuition, then the idea matures, and is formalized (sometimes with some trials and errors; of course, I also prefer everyone to be always correct, but...). And, I believe, the relativization outlined above (by me) is the right interpretation of the (formally wrong) text of Zhang Jinwen. Boris Tsirelson (talk) 05:50, 26 April 2019 (UTC)

I told about this here:

Boris, I would not protest if the article contained an explanation like "conglomerates are just an idea which did not find a rigor definition yet..." and something like this. But the article does not contain such explanations. And this is a cheating, as I wrote above. Creating a parallel reality like what people do in Russian Wikipedia, see, for example, this thread.

Eozhik (talk) 06:08, 26 April 2019 (UTC)

Well, tastes differ. Yes, mathematicians are obliged to be accurate. But I really do not understand why do you write "absurd idea" (!) when I observe rather understandable lapsus (in the formalization, not in the idea). Clearly, the author, writing "the first part is just GB", did not think that you will interpret it this way. He meant that the classes satisfy GB, but are only a part of his wider universe. And I do not think that the privilege to introduce new objects must be reserved for logicians. Boris Tsirelson (talk) 06:24, 26 April 2019 (UTC)

Absurd because those people don't see absurdities in their constructions. And I dare to say that this grows from their poor understanding of what logic is. If they would be better educated there would not be problems with this:

did not think that you will interpret it this way

And in each field there are some rules. If you introduce new objects in foundations of mathematics you must do this by the rules of foundations of mathematics. Eozhik (talk) 07:33, 26 April 2019 (UTC)

And these are not tastes. These are rules. Eozhik (talk) 07:42, 26 April 2019 (UTC)

Well, now our (different) opinions are voiced clearly; we agree that some authors are not so educated in some branches of mathematics, but we treat this fact differently. I have nothing to add here. Boris Tsirelson (talk) 07:45, 26 April 2019 (UTC)

Original research

By the way, about the original research. @Alexei Kopylov: can you provide the references to the places in the sources you cite where the authors speak about Grothendieck universe, strongly inaccessible cardinals, non-conservative extensions, many-sorted first-order logic, etc.? Eozhik (talk) 06:16, 26 April 2019 (UTC)

•  Done: added pages using a template {{rp}} (personally, I don't like this style, but alternatives are worse). BTW, I never claimed that Grothendieck used the term "conglomerate". I change the name of the section as suggested by D.Lazard. Alexei Kopylov (talk) 20:31, 26 April 2019 (UTC)

Done badly. The question was not what all these terms mean, but how they are related to conglomerates. Among authors who mention conglomerates in their texts, the following people do not speak at all about Grothendieck universe, non-conservative extensions and many-sorted first-order logic:

1. J.Adamek, H.Herrlich, G.Strecker in "Abstract and Concrete Categories: The Joy of Cats",

2. M.S.Osborne in "Basic Homological Algebra",

3. L.Nel in "Continuity theory".

4. V.Laan in "Introduction to category theory"

As far as I can see from the references to Google books the same is true for

5. H.Herrlich, G.Strecker, in "Category theory: an introduction", and

6. G.Preuß in "Theory of Topological Structures: An Approach to Categorical Topology".

From Google books I conclude also that Zhang Jinwen mentiones only the term "non-conservative extension" from this list, but it is seen only that he applies it to abstract systems. It is not seen that he claims what you write

ACG is a non-conservative extension of Quine-Morse set theory and ZF# (Zermelo–Fraenkel set theory with an axiom of existence of inaccessible cardinals).

Where is it written? Eozhik (talk) 04:37, 27 April 2019 (UTC)

• The answer to the last question is Page 170: Theorem 21 (+definition 10). Non-conservative extensions and many-sorted first-order logic are only applied to ACG which is described in Jinwen. So you should not expect to find this term in other references. About Grothendieck universe: I've added a link to a scan of Herrlich & Strecker. You can check that they talk about Grothendieck universe. Nel and Laan name this just "universe" or "universal conglomerate", but is is obvious that this is equal to Grothendieck universe. If you don't think it is obvious, I can rewrite it closer to the sources.Alexei Kopylov (talk) 20:57, 27 April 2019 (UTC)

• As we understood, the only reasonable meaning of the term "conglomerate" is "arbitrary set in ZFC (or in NBG, or in MK), in contraposition to the sets that are subsets of a given Grothendieck universe $U$". It will be honest with respect to the reader to rewrite the article in such a way that this becomes clear (and to remove everything that is not supported by the reliable sources, including non-conservative extensions, many-sorted first-order logic, etc.). Eozhik (talk) 08:33, 28 April 2019 (UTC)

Unreliable source?

Eozhik, you have added "Unreliable source" after two sources. Could you provide reasons why do you believe this sources are unreliable in subsections bellow? Alexei Kopylov (talk) 20:37, 26 April 2019 (UTC)

You forgot also Adamek-Herrlich-Strecker. As I wrote here and here at the talk page, for claiming what is written in this article, in particularly,

As Gödel-Bernays set theory extends ZFC to allow classes, there are extensions that allow conglomerates.

and

ACG is a non-conservative extension of Quine-Morse set theory and ZF# (Zermelo–Fraenkel set theory with an axiom of existence of inaccessible cardinals). Moreover, it can prove the consistency of both of these theories.

it is necessary to built a first order theory (like ZFC, NBG, or MK), where all these propositions are proved. Neither

J.Adamek, H.Herrlich, G.Strecker in "Abstract and Concrete Categories: The Joy of Cats",

nor

Scott Osborne in "Basic Homological Algebra"

nor

Jinwen Zhang in "The axiom system ACG"

do this. We discussed the details with Boris Tsirelson here. Eozhik (talk) 04:56, 27 April 2019 (UTC)

In fact, this applies to all the sources in this article: all of them are unreliable, since nowhere these claims are proved. Eozhik (talk) 05:06, 27 April 2019 (UTC)

• First of all, there is no requirement to build a first order theory. It can be high order theory of anything else. Second, this is not a test for reliability. For example, Adamek et. al. is a secondary source, it may not include all proofs. Third, Osborne does prove what he claims (Proposition 6.13) and Zhang does build a first order theory and does prove what he claims and what I used in the article. Finally, you cannot just say that all sources are unreliable. If a source is published in reliable peer reviewed journal or by well-known scientific publisher you have to provide a good argument why do you think it is unreliable. I suggest for each source that you believe is unreliable you create a separate subsection bellow and put your argument there. Hint: "The source is unreliable because I don't agree with it" is not a good argument. Alexei Kopylov (talk) 21:27, 27 April 2019 (UTC)

• 1. I did this:

  you have to provide a good argument why do you think it is unreliable

2. This:

Zhang does build a first order theory and does prove what he claims

-- is fermatism.

3. There is no requirement to build a formal theory for people who are not related to mathematics. Mathematicians are obliged to give formal definitions and formal proofs. That is the difference.

4. Osborne does not construct a formal theory. As well as the others.

5. We came to the understanding that Herrlich and Strecker is the only source that can be discussed (in particularly, Adamek et al refer to them). The problem with this text is that the notion of conglomerate is introduced there just by renaming, not by constructing a formal theory. The authors just say: "We consider a set $U$ with these properties. What was called "sets" before we now call "conglomerates". By "classes" we now mean (not arbitrary classes, or arbitrary sets in NBG, but) only subsets of $U$. And by "sets" we mean (not arbitrary sets in NBG, but) only elements of $U$". In contrast to what you write here this is not called a non-conservative extension. This even can't be called a model, since they don't construct a formal theory for which this construction could be considered as model. This is renaming. Eozhik (talk) 08:12, 28 April 2019 (UTC)

Puzzling

After a search on MathSciNet I feel astonished. Conglomerates are used in more than 10 publications reviewed (and reviews). No one of these defines this notion, nor cites a source where it is defined (or did I miss something?); all just accept it to be well-known. I guess, J.Adamek, H.Herrlich, G.Strecker (The Joy of Cats...) is the implicit source. The review of this source is linked 528 times from refs and 10 times from reviews. In the end of the review the source is criticized for esoteric terminology (but not conglomerates). What does it all mean?? (The same was just written by me on the AFD page.) Boris Tsirelson (talk) 07:59, 27 April 2019 (UTC)

Detail:

Skornjakov 1973 "Radicals of Ω-rings" (Russian)
Review (by Bergman): "A/z(A) is a conglomerate of members of Σ(S)..."

Hušek 1976 "Lattices of reflections and coreflections in continuous structures" Lecture Notes in Math. Vol. 540
Review (by Kannan): "...the order structure of the conglomerate of all reflections..."

Činčura 1979/80 "Reflective and coreflective subcategories of some category of generalized topological spaces"
Review (by Vorster): "...where L denotes the conglomerate of all coreflective subcategories of C..."

Greve 1980 "How many monoidal closed structures are there in TOP?" Arch. Math.
Review (by Herrlich): "...illegitimate collection (i.e. one which can be mapped onto the conglomerate of all subclasses of a proper class)..."

Börger et al 1981 "Compact and hypercomplete categories" J Pure Appl Algebra
Intro: "...diagrams [...] for which the 'conglomerate' of natural transformations..."

Korostenski et al 1986 "On left-cancellable classes of morphisms" Comm. Algebra
Review (by Pumplün): "...Galois correspondence between the conglomerate of all subclasses of a category..."

Dikranjan et al 1987 "Closure operators. I" Topology Appl.
Summary: "...Galois equivalence between the conglomerate of idempotent and weakly hereditary closure operators of X and the conglomerate of subclasses of M..."

Richter 1990/91 "Algebra ⊂ topology?!" Res. Exp. Math
Review (by Bentley): "...M a conglomerate of mono-sources..."

Šlapal 1995 "A conglomerate of exponential supercategories of the category of finitely generated topological spaces" Aequationes Math

Holgate et al 2016 "Topogenous and Nearness Structures on Categories" Appl Categor Struct
Page 449: "...the ordered conglomerate of all neighbourhood operators..."

Boris Tsirelson (talk) 08:46, 27 April 2019 (UTC)

Boris, you could notice also that the paper by Jinwen Zhang is not mentioned there. Moreover, this publishing house, World Scientific, has a dubious reputation since its publications are not refereed. 528 means that people cite different facts from category theory, not connected to "conglomerates". 10 illustrates the fact that there is some percent of scientists who are very narrow specialists. They don't check the facts even if they lay near (but not in their field). Eozhik (talk) 09:05, 27 April 2019 (UTC)

I wonder, if there is a possibility to let these people know about what happens here. It would be interesting to listen to them. Eozhik (talk) 09:16, 27 April 2019 (UTC)

Yes, I excluded intentionally such authors as Adamek and Zhang from this list. And probably all this shows again that (like it or not) a formal theory is created after its intended interpretation. And this: for category theory, the details of the definition are not important. Note also that most of these works appeared before Adamek (and Zhang). Conglomerates existed during decades as "folklore", and are now in the process of formalization. And surely you know that Cantor worked in the set theory a lot before ZFC. Boris Tsirelson (talk) 09:34, 27 April 2019 (UTC)

Zhang's paper is absent in mathscinet. I am not against if in the article it would be written that this is "folklore", not a rigor theory. The times changed: when Cantor lived there was no necessity to specify this, but now this necessity appeared, since there is a sharp border between "folklore" and "rigor theory". Eozhik (talk) 09:59, 27 April 2019 (UTC)

And details of definitions are important everywhere in mathematics. Eozhik (talk) 10:15, 27 April 2019 (UTC)

I think the best source is Herrlich&Strecke. In the second chapter they said that there are several attempts to build foundation for category theory, each of them with some disadvantages. Then they list the features that they requiry of such foundation (including conglomerates). And they say that these features can be realized. They show how to do it in Appendix. Hope it answer your questions. Alexei Kopylov (talk) 21:42, 27 April 2019 (UTC)

Yes, I see. A pity that the book does not say explicitly (and prove) that consistency of this theory follows from consistency of ZFC plus large cardinal(s). Many set theorists tend to believe in the latter consistency. On the other hand, the book is very close to this point, and probably, an expert in set theory can infer this consequence easily. Boris Tsirelson (talk) 04:34, 28 April 2019 (UTC)

And, really, who needs a special first order theory for conglomerates? Let it be just a convenient (for category theory) special terminology over the theory "ZFC + worldly cardinal". The set V_κ corresponding to a given worldly cardinal κ is the "universe" discussed in the Appendix (linked above)... and everyone should be happy. Boris Tsirelson (talk) 09:10, 28 April 2019 (UTC)

• Boris, people who don't like cheating, they need specification. For example, the phrases like this

  all classes form a conglomerate

  must be supplied with correct explanations of what this exactly means. Because when a reader understands this literally, this becomes lies: not all classes from the point of view of the commonly used set theories, where this term is used, NBG or MK, form a conglomerate in this "terminology". To say nothing about "non-conservative extensions", etc. Eozhik (talk) 09:23, 28 April 2019 (UTC)

Whoever intermixes words taken from different contexts, faces the consequences. (By the way, this is a well-known problem of math on wikipedia; editor "fixes an error" and does not bother to check that his terminology/notation agrees with that in the article; see WT:WPM#"Where triangle's area is triangle's area".) But, of course, it is wise to warn the reader whenever a misunderstanding is expected. This is a pedagogical matter, rather than mathematical. Boris Tsirelson (talk) 10:16, 28 April 2019 (UTC)

About conservative extensions: now, to your liking, it IS an extention by definitions of "ZFC + worldly cardinal" (up to a pedagogical inconvenience that some words are REdefined); and "ZFC + worldly cardinal" is a nonconservative extension of ZFC. Boris Tsirelson (talk) 10:22, 28 April 2019 (UTC)

And the redefinition of "sets" and "classes" is not my "original research", it is taken from the source mentioned. I only added some relation to large cardinals. Boris Tsirelson (talk) 10:29, 28 April 2019 (UTC)

Boris, in which sense the relation $A\subseteq U$ implies the relation $A\in C$ for some "conglomerate" $C$? What is meant by $\in$ here? Eozhik (talk) 10:38, 28 April 2019 (UTC)

Ah, yes, $U$ is a set. Pardon. Eozhik (talk) 11:01, 28 April 2019 (UTC)

And this means that this phrase should also be understood in a very cunning sense:

Every class is a conglomerate.

Since not every class in the sense of NBG is a conglomerate in the sense of this theory (but only those which are called "sets" in NBG). Eozhik (talk) 11:26, 28 April 2019 (UTC)

An ugly terminology, provoking a variety of misunderstandings on very different levels. Eozhik (talk) 11:37, 28 April 2019 (UTC)

No, I think, you are just not used to models. Look at the classical Euclidean models of non-Euclidean geometries: Beltrami–Klein model, Poincaré disk model, Poincaré half-plane model. Do you find it ugly, that "In this model, lines and segments are straight Euclidean segments, whereas in the Poincaré disk model, lines are arcs that meet the boundary orthogonally"? Boris Tsirelson (talk) 18:27, 28 April 2019 (UTC)

Or, closer to set theory, try Constructible universe, Absoluteness#In set theory, Minimal model (set theory), Transitive model etc. Still closer to the business: Large cardinal#Motivations and epistemic status. Boris Tsirelson (talk) 18:34, 28 April 2019 (UTC)

• I have another explanation: the point is that I am a punctual person, attentive to details. The difference between your geometric examples and what we see here is that in non-Euclidean geometries people honestly describe the axiomatic theory, with all the necessary details, so that the question "what does this mean?" does not appear, and after that they construct models (again with a honest explanation of how the objects in the new theory are interpreted in the old theory, so that misunderstandings don't appear). But here the situation is opposite: people pretend that they extend the modern set theories, NBG and MK, by "adding new objects", but when you look closer at what they suggest as "axiomatic theories" it becomes clear that those theories are absurd, because these people simply don't understand the difference between naive set theory and modern axiomatic set theories, trying to construct modern set theory by methods of 19 century. As if they never heard about paradoxes. And moreover, the only reasonable explanation of what they do, as it turns out, is that they don't add something, but on the contrary cut some pieces off NBG and MK. And what looks absolutely scandalous, they use for this a trick that normal people can't expect: they just rename the objects. What becomes visible only after a long study in an appendix, in which no one had guessed to look. And the most puzzling questions like what we discussed, -- "how can a proper class belong to something?" --- disappear just because proper classes are thrown away from their "great theory". This is called "breach of trust". Eozhik (talk) 20:12, 28 April 2019 (UTC)

Your examples from set theory in my opinion are too technical, I don't see how they can be useful for "working mathematician". Eozhik (talk) 20:16, 28 April 2019 (UTC)

(Unindent) You really do not understand that the boundary between sets and classes is not ontological, it shifts whenever we accept new large cardinal (and its first shift was the axiom of infinity). But, well, you want a first order theory anyway; and this is not a problem.

First group of axioms: "just" the axioms of ZFC (only with the word "set" replaced with "conglomerate", if you say it in words).
A new constant: U.
Axiom: U=V_κ for some κ.
Definition: "set" is an element of U.
Second group of axioms: "just" the axioms of ZFC, but relativized to U, which is automatic if you say them by words (the word "set" has already the new meaning); but technically, use bounded (by U) quantifiers. Or instead, one axiom: κ is a worldly cardinal.
Definition: "class" is a conglomerate contained in U.

Did I forgot something? Maybe. Anyway, this is of course my "original research" (until someone will find it somewhere, which will happen inevitably, but I do not know, when). Anyway, it shows that this is not a problem.

About "working mathematician": a working category theorist just uses conglomerates long ago, since he understands intuitively that this is not a problem. But I know, you are dissatisfied. Well, I did my best. And fortunately, category theorists do not need your permission (nor wikipedia's permisson). Boris Tsirelson (talk) 21:08, 28 April 2019 (UTC)

Boris, I understood what can be done for constructing first order theory after seeing how they really define these objects in that Appendix in Herrlich&Strecker. But we can't write this here because this will be an original research. This trick with cutting off parts of NBG or MK is indeed new for me, and I still don't understand why it is necessary. I also was working in category theory and faced some problems in its "foundations". In particular, I needed a proposition that each category has a skeleton. But I resolved this problem inside MK, without this strange trick of cutting off something, see Property 1 at page 5 here. I also did not see that the definition of class can be "shifted" (and I don't understand why this can be useful). Despite all this, as I told before, I would not be against keeping this article, provided it will contain an honest explanation of the essense of the matter so that these traps would disappear. And the essense in my opinion is that this is a trick that cuts off some parts of NBG and MK and renames objects. I am not sure that this will not be an original research. Eozhik (talk) 21:44, 28 April 2019 (UTC)

• @Tsirel: Boris, who made the decision here and why it was made before the article was edited so that this would satisfy all the participants of the discussion? Eozhik (talk) 16:52, 3 May 2019 (UTC)

Do you suspect that the admin "User:78.26" is me?   :-)   Not at all. I know about him not more than is published on his page (available to both of us). Why before? Since the consensus to "keep" is already reached; and now everyone (including you) is free to edit this article (as well us revert other's edit, when appropriate). "Closed as keep" does not mean "no more edits to the article", nor "no more edits to the AFD page"; it means only "no more edits to the closed discussion there". Boris Tsirelson (talk) 17:22, 3 May 2019 (UTC)

@Eozhik: I made that decision. It was "past due" for closure by four days, mostly because I don't think any administrator wanted to read through it, per my close note. Boris Tsirelson is absolutely correct, "keep" does not mean that we keep the article as it is, it just means that the topic is not deleted, and editors may (and of course are roundly encouraged to) improve upon it. Happy editing! 78.26 _{(spin me / revolutions)} 17:31, 3 May 2019 (UTC)

OK, we'll edit the text. I think, Boris (@Tsirel:) should start this, since it was he who first suggested a reasonable formulation. Eozhik (talk) 17:42, 3 May 2019 (UTC)

Hmmm... The current version is mostly due to @Alexei Kopylov:; his opinions and participation is welcome, preferably, before substantial changes in the article. Boris Tsirelson (talk) 18:07, 3 May 2019 (UTC)

Changes proposed and discussed

1

The first phrase "In mathematics, a conglomerate is a collection of classes, just as a class is a collection of sets" is incorrect (and I did not find it in the sources). A member of a conglomerate is generally another conglomerate, and in particular, it may be a class or even a set. Boris Tsirelson (talk) 18:17, 3 May 2019 (UTC)

• "The concept of 'conglomerate' has been created to deal with 'collections of classes'." [1, page 15]

• "Thus we can form the conglomerate of all classes as well as such entities as functions between conglomerates and families of conglomerates." [1, page 16] Boris Tsirelson (talk) 18:24, 3 May 2019 (UTC)

• "One Universe   This approach has been used and developed by Isbell, Mac Lane, and Feferman and is the foundational approach taken for this text. It consists 'essentially' of the addition of one extra stage of flexibility to the Gödel-Bernays-von Neumann approach, as opposed to the addition of a plethora of extra stages as with Grothendieck Universes." [2, page 329] Boris Tsirelson (talk) 18:35, 3 May 2019 (UTC)

• "Universes are of critical importance to formalizing concepts in category theory inside set-theoretical foundations." (Quoted from "Universe (mathematics)".)

I try possible beginning.

In mathematics, conglomerates are members of a universe that contains another universe as a member. This concept has been created to deal with collections of classes, of functions between classes, of families of classes etc. The smaller universe is usually assumed to be a given Grothendieck universe (whose existence is postulated); the larger universe is usually the class of all sets. A single universe is sufficient for most branches of mathematics, but category theory may need a hierarchy of two or more universes. Conglomerates may be used when two universes are sufficient; the infinite hierarchy given by Tarski–Grothendieck set theory serves more demanding situations.

Universes are instrumental when formalizing concepts used in category theory inside set-theoretical foundations. Denoting the smaller universe by U, one translates the three-level terminology of sets, classes and conglomerates into the one-level terminology of the set theory as follows: "set" is translated as "member of U"; "class" is translated as "subset of U"; and "conglomerate" is translated as just "set". This translation is known as "conglomerate conversion".

Remarks, opinions? Boris Tsirelson (talk) 20:02, 3 May 2019 (UTC)

• Boris, I would make several corrections:

In mathematics category theory, conglomerates are members elements of a universe that contains another universe as a members an element. This concept has been created to deal with collections "collections" of classes, of functions between classes, of families of classes etc. The smaller universe is usually assumed to be a given Grothendieck universe (whose existence is postulated); the larger universe is usually the class of all sets. A single universe is sufficient for most branches of mathematics, but category theory may need a hierarchy of two or more universes. Conglomerates may be used when two universes are sufficient; the infinite hierarchy given by Tarski–Grothendieck set theory serves more demanding situations.

Universes are instrumental when formalizing concepts used in category theory inside set-theoretical foundations. Denoting the smaller universe by U, one translates the three-level terminology of sets, classes and conglomerates into the one-level terminology of the set theory as follows: "set" is translated as "member of U"; "class" is translated as "subset of U"; and "conglomerate" is translated as just "set". This translation is known as "conglomerate conversion".

(In MK functions between classes are considered without problem. And I think there is no difference between "family of classes" and "collection of classes".) Eozhik (talk) 04:44, 4 May 2019 (UTC)

Are you sure that there is a term "conglomerate conversion"? Eozhik (talk) 04:15, 4 May 2019 (UTC)

And do they really use an hierarchy of more than two universes? Eozhik (talk) 04:19, 4 May 2019 (UTC)

Also I would change the second paragraph as follows:

The introduction of a new (smaller) universe U entails some changes in terminology: it is considered convenient

• to apply the term "set" only to elements of U,
• to apply the term "class" only to subsets of U,
• to use the term "conglomerate" for all sets (i.e. elements of the bigger universe, not necessary elements U).

So, in particular, if the new theory was constructed as a modification of the theories where the term "class" was used, like Von Neumann–Bernays–Gödel set theory or Morse–Kelley set theory, then proper classes as objects of the old theory are discarded from consideration in the new theory.

It seems to me that would be clearer. Eozhik (talk) 05:07, 4 May 2019 (UTC)

About "conglomerate conversion": you are right; I followed the current article, but in fact, the term in [6, page 6] is rather "conglomerate convention". Boris Tsirelson (talk) 05:14, 4 May 2019 (UTC)

OK. Boris, also in Von Neumann–Bernays–Gödel set theory and in Morse–Kelley set theory, they don't consider a "big universe" that "contains all objects (classes) as elements". Maybe this means that the first paragraph also needs some more corrections. Eozhik (talk) 05:30, 4 May 2019 (UTC)

About "really use an hierarchy of more than two universes": a good question; I do not know; do you?

I did not see this in their considerations. As far as I understand, they use only one universe. (But mentioning Tarski–Grothendieck set theory is useful in this context, I think.) Eozhik (talk) 06:50, 4 May 2019 (UTC)

I only know that

(a) the Tarski-Grothendieck theory provides infinite (moreover, transfinite) hierarchy of universes;

(b) V_ω+ω is the universe of "ordinary mathematics", and is a model of Zermelo set theory (quoted from Von Neumann universe#Applications of V as models for set theories), and nevertheless, the prevailing theory is ZFC (rather than Z), and the prevailing universe is the whole Von Neumann universe (rather than V_ω+ω);

(c) in the (uncountably high) Borel hierarchy, only the first two or three levels are used in "ordinary analysis and probability theory", and nevertheless, the prevailing notion is σ-algebra (including the whole hierarchy).

That is, mathematicians often prefer foundations with great margin, since "economical" foundations tend to make unwanted troubles. Boris Tsirelson (talk) 05:38, 4 May 2019 (UTC)

About relations to NBG and MK. According to the conglomerate convention, only members of U are sets; thus, only members of the power set P(U) are classes; indeed, U is a model of ZFC, and P(U) is a model of NBG and MK. In the spirit of virtualization metaphor, these models are guests, supported by the host (the set theory underlying all that). Sure, one may extend the host by its classes (be it NBG or MK), but this is probably irrelevant to the virtual "machines". Boris Tsirelson (talk) 06:04, 4 May 2019 (UTC)

This is what we see after those changes in terminology:

only members of U are sets; thus, only members of the power set P(U) are classes

Do you think it is correct to say that when we add to NBG or to MK the axiom of existence of a Grothendieck universe we obtain two universes? Initially there were no big universes in NBG and in MK, when we add one universe we obtain one universe, not two. Eozhik (talk) 06:21, 4 May 2019 (UTC)

What about this (a straitforward, non-diplomatic) variant?

In category theory, the term "conglomerate" is used as a synonim of the term "set" in non-conservative extensions of axiomatic set theories, like Zermelo-Fraenkel set theory, Von Neumann–Bernays–Gödel set theory or Morse–Kelley set theory, where the existence of a Grothendieck universe $U$ is postulated (in addition to the other axioms). An example of the non-conservative extension of ZFC is the Tarski–Grothendieck set theory, where not only existence of a single Grothendiech universe, but of an infinite hierarchy of them is assumed.

This concept has been created for the possibility to deal with "collections" of classes, what is desirable in category theory (so that each class can be considered as an element of a "more general collection", a conglomerate). Technicaly this is organized by the following changes in terminology: it is considered convenient

• to apply the term "set" only to elements of $U$,
• to apply the term "class" only to subsets of $U$,
• to use the term "conglomerate" for all sets (not necessary elements of $U$).

So, in particular, if the new theory was constructed as a modification of the theories where the term "class" was used (as a contraposition to the term "set"), like Von Neumann–Bernays–Gödel set theory or Morse–Kelley set theory, then proper classes as objects of the old theory are discarded from consideration in the new theory.

Eozhik (talk) 06:11, 4 May 2019 (UTC)

As far as I understand, in this philosophy they refuse to consider objects which are not sets in MK, i.e. which are "proper classes in the old sense", but (not counting the possible problems caused by the supplementary axiom of existence of $U$) this does not lead to a loss of information about objects of MK since the representation of MK as a model in "MK+Grothedieck universe" means that what can be proved in MK about its usual objects can be proved in "MK+Grothedieck universe" about subsets of $U$. Even the cardinality of sets is not restricted (what one could expect since sets in this theory are elements of a given set $U$) since this is another cardinality, inside the new theory. I would add explanations like this into the text, if customs allow this (what seems doubtful for me, since what we discuss here I perceive as an original research). Eozhik (talk) 07:45, 4 May 2019 (UTC)

By the way, this picture seems to give an explanation of how the notion of cardinality for all "conglomerates", including "proper classes in new sense" (and this means, in old sense as well, since they are presented in this model), can be defined? An interesting theory... Eozhik (talk) 07:55, 4 May 2019 (UTC)

About "straitforward, non-diplomatic", I do not understand you. It seems, for you the question "is a conglomerate REALLY a set or not?" is somewhat political or religious; or, at the very least, ontological. Let me ask you then: is a number (no matter, natural, or real, etc) REALLY a set, or not? Or: is a virtual machine REALLY a finite sequence of bits, or not? For me, this is a wrong way to ask. A real number is a member of a complete ordered field. Nowadays all this is usually interpreted within ZFC, and therefore every number is a set anyway. But this is rather an implementation. Likewise, a file containing an image of a virtual machine is stored on a hard disk and therefore is a finite sequence of bits anyway. But again, this is rather an implementation. See also Equivalent definitions of mathematical structures. A mathematical object has no onthology; its implementation has (more or less...), but implementations may differ. Closer to our business: for me, the phrase '"conglomerate" is a synonym of "set"' (even with light reservations) sounds strange and a bit misleading, just like 'a "real number" is a special case of a "set"' or 'a "virtual machine" is a special case of a "finite sequence of bits"'. My feeling is that you (rather than me) introduce some strange diplomacy. Boris Tsirelson (talk) 09:39, 4 May 2019 (UTC)

Boris, I don't see anything religious in my understanding of sets (and I think, it actually does not matter). Look: there is a necessity to explain the differences between "conglomerates" and "sets" for people who perceive sets as objects of axiomatic set theories like ZFC, NBG and MK. Your phrase

In mathematics, conglomerates are members of a universe that contains another universe as a member.

-- seems confusing for me, because if a person tries to understand this applying his knowledge about, say, MK, that would sound strange for him, since there is no universe in MK. When we add $U$ we obtain only one (small) universe, not two (small and big). There must be another explanation. Eozhik (talk) 09:57, 4 May 2019 (UTC)

OK, to this end my formulation includes "(whose existence is postulated)". Also, "In set theory, universes are often classes" (quoted from "Universe (mathematics)"), thus it is not forbidden to write "the larger universe is usually the class of all sets". A Grothendieck universe must be a set, but the general idea of universe is more permissive. Still, if something can be said more clear, I'll be glad. But for me, your wording is unclear for another reason. Also, the lead must be short; clarifications may follow in sections. Boris Tsirelson (talk) 10:14, 4 May 2019 (UTC)

I don't insist, we can find another formulation. But I don't understand, what can be called "universe" in MK? Eozhik (talk) 10:20, 4 May 2019 (UTC)

"there is no universe in MK"? Not quite so. Note that ZFC does not prove existence of Grothendieck universes (which can be proved, assuming that ZFC is consistent), but it also does not disprove their existence (according to a common belief that ZFC+large cardinal is still consistent). Thus, the number of Grothendieck universes in ZFC is ill defined (and, informally, it means for me that the "real" place of the border between sets and proper classes is left uncertain). And again, "universe" in general (not quite "Grothendieck") is a vague idea; but surely it is reasonable to treat the class of all sets (be it in NBG or MK) as a universe. And in addition, any number of Grothendieck universes may exist (or not) in MK just as in ZFC. Boris Tsirelson (talk) 10:36, 4 May 2019 (UTC)

Boris, the problem is not that Grothendieck universes can exist or not in MK (and because of this we need a supplementary axiom that it exists). The problem is that there is no universe in MK as an object that contains all other objects as elements. What is described in this article

universe is a collection that contains all the entities one wishes to consider in a given situation

-- does not exist in MK. It can be defined, I think, by introducing some supplementary symbols and axioms, but people don't do this. And that is why the reasoning about two universes, big and small, sounds confusing. (As far as I understand, in NBG and even in ZFC the situation is the same.) Eozhik (talk) 10:58, 4 May 2019 (UTC)

There is no "big universe" in first order theories. When we introduce "small universe", it appears, but "big universe" does not appear. Eozhik (talk) 11:03, 4 May 2019 (UTC)

MK is described in Kelley's "General topology" (appendix), NBG in Mendelson's "Introduction to mathematical logic". Eozhik (talk) 11:29, 4 May 2019 (UTC)

I do not think this is a problem, as long as the word "universe" refers to a vague idea rather than a well-defined math notion. My usage conforms to the quote "In set theory and related branches of mathematics, the von Neumann universe, or von Neumann hierarchy of sets, denoted V, is the class of hereditary well-founded sets. This collection, which is formalized by Zermelo–Fraenkel set theory (ZFC), is often used to provide an interpretation or motivation of the axioms of ZFC." from Von Neumann universe. And anyway, "the larger universe" is replaced soon by "the class of all sets". On the other hand, if it exists among sets (which could happen within ZFC), then it can be used as well. And "on the third hand", the underlying theory can be taken to be NBG, and then the class of all sets is a legitimate universe even in your approach, isn't it? I did not write whether I mean ZFC or NBG there. And, after all, my text is intended to be the lead; detailed explanations and clarifications may follow in sections; but the big picture should be reasonably understandable, – not only what conglomerate is (basically), but also, what is it good for (basically). Boris Tsirelson (talk) 17:50, 4 May 2019 (UTC)

Boris, neither ZFC, nor NBG, nor MK have universe in this sense. ZFC doesn't have "big universe" because the class $Set$ of all sets is not an object of ZFC, it is defined in MK (and in NBG), not in ZFC. NBG (and MK) do not have "big universe" because there is no an object in NBG (and in MK) which contains all objects (i.e. all classes) as elements. The class $Set$ of all sets is defined in NBG and in MK, but it does not contain all objects of NBG and MK as elements, only sets (proper classes don't belong to $Set$ as elements). The notion of "big universe" is confusing, it must be eliminated in the text. Eozhik (talk) 18:42, 4 May 2019 (UTC)

It need not contain everything (including classes). I wrote clearly "class of all sets"; and not "big universe" but just "the larger universe". Once again, it seems to be usual enough, to say "the class of all sets is a universe". Boris Tsirelson (talk) 18:51, 4 May 2019 (UTC)

But, well, your proposal is also OK with me (may be up to some remarks). Boris Tsirelson (talk) 19:01, 4 May 2019 (UTC)

When you write about universe you give this link: Universe (mathematics). It is written there that

universe is a collection that contains all the entities one wishes to consider in a given situation

Sets are not all entities that are considered in NBG and in MK, so the class V of all sets can't be a universe in NBG or in MK. There are also proper classes which are not sets. This is confusing. If on the other hand you are speaking about ZFC, then its "universe" is not an object of ZFC. This is also confusing. (By the way, the article that you mention is written badly, it must be corrected as well.) Eozhik (talk) 19:07, 4 May 2019 (UTC)

And by the way, in SOME implementations, a natural number is a set, but a real number is a class, see here. So what? Boris Tsirelson (talk) 09:50, 4 May 2019 (UTC)

I don't understand. Do you mean proper classes? ${\mathbb {N} }$ can be defined as the smallest infinite ordinal number, i.e. a set. After that ${\mathbb {Z} }$ can be defined as a quotient set of ${\mathbb {N} }\times {\mathbb {N} }$, i.e. a set. Then ${\mathbb {Q} }$ can be defined as a quotient set of ${\mathbb {Z} }\times ({\mathbb {N} }\setminus \{0\})$, i.e. again a set. Then real numbers can be defined as Dedekind cuts, i.e. subsets in ${\mathbb {Q} }$, i.e. again sets. Every time we pass from sets to sets. All numbers can be treated as sets. Or you mean something else? Are there theories where this scheme does not work? Eozhik (talk) 10:09, 4 May 2019 (UTC)

Yes, I mean proper class. And I gave you a link (which you did not follow, I guess). Boris Tsirelson (talk) 10:16, 4 May 2019 (UTC)

I did not understand why this can be interesting. Eozhik (talk) 10:22, 4 May 2019 (UTC)

Well, never mind. (Just to show that "to be a proper class" may depend on the implementation.) Boris Tsirelson (talk) 10:36, 4 May 2019 (UTC)

2

Well, let me repeat your text here, as the basis for later discussion. Boris Tsirelson (talk) 19:06, 4 May 2019 (UTC)

---

In category theory, the term "conglomerate" is used as a synonim of the term "set" in non-conservative extensions of axiomatic set theories, like Zermelo-Fraenkel set theory, Von Neumann–Bernays–Gödel set theory or Morse–Kelley set theory, where the existence of a Grothendieck universe $U$ is postulated (in addition to the other axioms). An example of the non-conservative extension of ZFC is the Tarski–Grothendieck set theory, where not only existence of a single Grothendiech universe, but of an infinite hierarchy of them is assumed.

This concept has been created for the possibility to deal with "collections" of classes, what is desirable in category theory (so that each class can be considered as an element of a "more general collection", a conglomerate). Technicaly this is organized by the following changes in terminology: it is considered convenient

• to apply the term "set" only to elements of $U$,
• to apply the term "class" only to subsets of $U$,
• to use the term "conglomerate" for all sets (not necessary elements of $U$).

So, in particular, if the new theory was constructed as a modification of the theories where the term "class" was used (as a contraposition to the term "set"), like Von Neumann–Bernays–Gödel set theory or Morse–Kelley set theory, then proper classes as objects of the old theory are discarded from consideration in the new theory.

---

We can try to express what we need better. How would you correct this? Eozhik (talk) 19:13, 4 May 2019 (UTC)

I agree, "synonim" is a bad word here. We can write

In category theory, the term "conglomerate" replaces the term "set" in non-conservative extensions of axiomatic set theories, like Zermelo-Fraenkel set theory, Von Neumann–Bernays–Gödel set theory or Morse–Kelley set theory...

What do you think? Eozhik (talk) 19:19, 4 May 2019 (UTC)

Yes, the latter sounds better. But still, it may seem а nonsense ("replace of a term does not change the meaning") until reading the second paragraph (and thinking). Also I bother a bit about "extensions ... like ZF, NBG or MK"; the reader that knows already that NBG and MK extend ZF could misinterpret this phrase (even if grammatically it is unambiguous). Boris Tsirelson (talk) 20:30, 4 May 2019 (UTC)

Yes, you are right. What about this variant?

In the part of category theory where extensions of axiomatic set theories assuming the existence of a Grothendieck universe $U$ are considered, the term "conglomerate" is applied to arbitrary sets as a contraposition to the distinguished sets that are elements of $U$.

The most popular axiomatic set theories, Zermelo-Fraenkel set theory, Von Neumann–Bernays–Gödel set theory, and Morse–Kelley set theory, admit non-conservative extensions arising after adding to the system of axioms a supplementary axiom of existence of a Grothendieck universe $U$. (An example of such an extension of ZFC is the Tarski–Grothendieck set theory, where not only existence of a single Grothendieck universe, but of an infinite hierarchy of them is postulated.)

The concept of conglomerate has been created for the possibility to deal with "collections" of classes, what is desirable in category theory (so that each class can be considered as an element of a "more general collection", a conglomerate). Technicaly this is organized by the following changes in terminology: when a Grothendieck universe $U$ is added to the chosen axiomatic set theory it is considered convenient

• to apply the term "set" only to elements of $U$,
• to apply the term "class" only to subsets of $U$,
• to use the term "conglomerate" for all sets (not necessary elements of $U$).

(So, in particular, if the new theory was constructed as a modification of the theories where the term "class" was used as a contraposition to the term "set", like Von Neumann–Bernays–Gödel set theory or Morse–Kelley set theory, then proper classes as objects of the old theory are discarded from consideration in the new theory.)

Eozhik (talk) 04:12, 5 May 2019 (UTC)

Yes, sounds good. A bit long for a lead, but this is probably inevitable in this case. Boris Tsirelson (talk) 04:32, 5 May 2019 (UTC)

Boris, in my understanding, this is not a lead, but almost all what can (and should) be written here. I would just add references and mention that this construction describes a model of ZFC/NBG/MK in the "extended ZFC/NBG/MK". This model, actually, makes this theory reasonable. Here is the extended text as I see it:

In the part of category theory where extensions of axiomatic set theories assuming the existence of a Grothendieck universe $U$ are considered, the term "conglomerate" is applied to arbitrary sets as a contraposition to the distinguished sets that are elements of $U$.

The most popular axiomatic set theories, Zermelo-Fraenkel set theory, Von Neumann–Bernays–Gödel set theory, and Morse–Kelley set theory, admit non-conservative extensions arising after adding to the system of axioms a supplementary axiom of existence of a Grothendieck universe $U$. (An example of such an extension of ZFC is the Tarski–Grothendieck set theory, where not only existence of a single Grothendieck universe is postulated, but of an infinite hierarchy of them.)

The concept of conglomerate has been created for the possibility to deal with "collections" of classes, what is desirable in category theory (so that each class can be considered as an element of a "more general collection", a conglomerate). Technicaly this is organized by the following changes in terminology: when a Grothendieck universe $U$ is added to the chosen axiomatic set theory (ZFC/NBG/MK) it is considered convenient

• to apply the term "set" only to elements of $U$,
• to apply the term "class" only to subsets of $U$,
• to apply the term "conglomerate" to all sets (not necessary elements of $U$).

Formally this gives a model of the initial axiomatic set theory (ZFC/NBG/MK) in the extension of this theory ("ZFC/NBG/MK+Grothendieck universe") with $U$ as the universe.

If the initial axiomatic set theory admits the idea of proper class (i.e. an object that can't be element of any other object, like the class $Set$ of all sets in NBG and in MK), then these objects (proper classes) are discarded from the consideration in the new theory ("NBG/MK+Grothendieck universe"). However, (not counting the possible problems caused by the supplementary axiom of existence of $U$) this in some sense does not lead to a loss of information about objects of the old theory (NBG or MK) since its representation as a model in the new theory ("NBG/MK+Grothendieck universe") means that what can be proved in NBG/MK about its usual objects called classes (including proper classes) can be proved as well in "NBG/MK+Grothendieck universe" about its objects called classes in this theory (i.e. about subsets of $U$, including subsets that are not elements of $U$, which are analogs of proper classes from NBG/MK). At the same time, the passage to the new theory does not give an equivalent theory in this situation, since there can appear some extra propositions about classes which can be proved in "NBG/MK+Grothendieck universe" but not in NBG/MK.

I must say, I don't see what can be added (not counting references). The section about "axiomatic theory of conglomerates" does not have reliable sources. And what is discussed "beyond conglomerates" is logical to discuss beyond this article. Eozhik (talk) 11:27, 5 May 2019 (UTC)

By the way in the article Structure (mathematical logic) to which the quieries model (logic) and model (mathematical logic) are redirected, the formal definition of model in mathematical logic is not given. The same with Model theory. This is bad. (By model I mean the construction from Takeuti's "Proof theory", I.8, but the definition must be presented in different books on mathematical logic.) Eozhik (talk) 12:04, 5 May 2019 (UTC)

"A model of a theory is a structure (e.g. an interpretation) that satisfies the sentences of that theory." (From "Model theory".) Detailed in "Interpretation (logic)#Interpretations of a first-order language". And Wikipedia is not a textbook. Still, if you feel able to do better, try it. Boris Tsirelson (talk) 15:47, 5 May 2019 (UTC)

This definition is too vague. I don't know, maybe later I will find some time for this. Eozhik (talk) 16:58, 5 May 2019 (UTC)

About "what can be added". Maybe, an example of a notion or/and result of category theory that involves conglomerates. Probably it will also show that the power-powerset P(P(U)) is not enough. Boris Tsirelson (talk) 16:03, 5 May 2019 (UTC)

Yes, applications are desirable. I don't know them. If there will appear people who know, that would be good, although as I told before, I don't believe that this notion is really useful. The facts from category theory that I used all are proved without conglomerates (and without difficulties). Eozhik (talk) 16:58, 5 May 2019 (UTC)

If there are no objections, I will put this text into the article instead of what is written now, add references, and forget about this story. Eozhik (talk) 17:01, 5 May 2019 (UTC)

OK with me. A pity that Alexei Kopylov is not with us now; but nothing is lost here, everything can be restored from the history anyway. Being not at all a category theorist, I have no opinion, what is really useful and what is not. However, I see sections "categories of categories" in [1, pages 39-41] and [2, pages 64-66]; conglomerates are needed there; and I'll try to write a bit on this in the article if neither you nor Kopylov do. Boris Tsirelson (talk) 19:23, 5 May 2019 (UTC)

For being with us it is desirable to understand the details. OK, I've made the edits. I removed only the sources where conglomerates are not mentioned or are mentioned in passing. Perhaps, some other sources can be removed as well. Or added. Eozhik (talk) 04:56, 6 May 2019 (UTC)

I added a partition into sections. What is written in "Corollaries" is, in my opinion, an original research. (And that is why I initiated the procedure of deletion.) I forsee accusations if the necessary sources will not be found. Eozhik (talk) 05:11, 6 May 2019 (UTC)

Nice. And, I'll try to source the corollaries. Boris Tsirelson (talk) 05:41, 6 May 2019 (UTC)

Looks good, so far. — Arthur Rubin (talk) 06:21, 6 May 2019 (UTC)

The corollaries are now sourced (to Mac Lane 1969). Boris Tsirelson (talk) 05:11, 7 May 2019 (UTC)

Boris, I can't find the whole text of MacLane's paper, I see only two pages. What is written there, because of what you refer to this text as the source? Eozhik (talk) 05:19, 7 May 2019 (UTC)

I just sent you email. Received? Boris Tsirelson (talk) 05:42, 7 May 2019 (UTC)

I'm not a category theorist, but I would also like to see it, if possible. I have borrowing privileges at CalTech, but they don't seem to have a copy of MacLane's paper, either. — Arthur Rubin (talk) 05:57, 7 May 2019 (UTC)

I just sent email to you, too. Received? Boris Tsirelson (talk) 06:06, 7 May 2019 (UTC)

Спасибо, добрый человек! Received. I see the phrase (at page 195)

From the definition of universe it follows readily that the small sets (with the given membership relation $\in$) themselves satisfy the ZFC axioms for set theory. For that matter, if we take "set" to be small set and "class" to be any subset of the universe $U$, these sets and classes satisfy the usual Gödel-Bernays axioms.

I think, we can treat this as a reference for our claim that this construction gives a model of ZFC/NBG in "ZFC/NBG+Grothendieck universe". As far as I understand, he does not write this directly, I see only an indirect mention (again at page 195):

his $s$, unlike $U$, is not itself a model of ZFC.

But maybe, this is enough. Eozhik (talk) 06:18, 7 May 2019 (UTC)

Received. — Arthur Rubin (talk) 08:33, 7 May 2019 (UTC)

3

Strangely, I do not see the paper [Mac Lane 1969] in the bibliography of [Adamek Herrlich Strecker 1990], nor [Herrlich Strecker 2007]. But i see [Mac Lane, S. Categories for the Working Mathematician. Springer, Berlin–Heidelberg–New York, 2nd ed., 1997] in the former. Maybe this book embraces [Mac Lane 1969]? Boris Tsirelson (talk) 07:21, 7 May 2019 (UTC)

The second edition of Mac Lane's "Categories for the Working Mathematician", of 1998, mentiones [Mac Lane 1969] in the bibliography. Eozhik (talk) 07:35, 7 May 2019 (UTC)

Yes, it does, in Sections I.6, I.7. Boris Tsirelson (talk) 07:34, 7 May 2019 (UTC)

4

Thanks Boris and Eozhik for your work. I like the new sections "Definition", "Corollaries" and "Terminology". The lead section, however, is too narrow now. I like the approach by Herrlich & Strecker: they first describe the properties of conglomerates that are required and then in Appendix give a definition. I think we should follow the same path. First give (probably) informal requirements and then give definition (that we already have). I like Eozhik's text on this page (04:44, 4 May 2019). I will try to combine this text, my old text and other comments, and make a suggestion, unless someone else propose something before me. Alexei Kopylov (talk) 08:17, 9 May 2019 (UTC)

Nice. I got tired, pretending to be a category theorist.   :-)   Boris Tsirelson (talk) 09:48, 9 May 2019 (UTC)

Boris, but you are at least a mathematician. I, on the conrary, don't like this idea. Alexei Kopylov, what you were writing so far was impossible to understand, I am not happy of thinking about this extra work for me and for other people. Eozhik (talk) 13:05, 9 May 2019 (UTC)

Requested move 3 May 2019

The following is a closed discussion of a requested move. Please do not modify it. Subsequent comments should be made in a new section on the talk page. Editors desiring to contest the closing decision should consider a move review after discussing it on the closer's talk page. No further edits should be made to this discussion.

The result of the move request was: Moved to Conglomerate (mathematics), supported by most people (on both sides of the debate) — Martin (MSGJ · talk) 13:38, 20 May 2019 (UTC) — Martin (MSGJ · talk) 13:38, 20 May 2019 (UTC)

---

Conglomerate (set theory) → Conglomerate (category theory) – It's not used in set theory, as there is no proposed axiom system, and a naive approach leads to contradictions. It may be used in category theory. Request move without redirect, as the term Conglomerate (set theory) is misleading. — Arthur Rubin (talk) 17:19, 3 May 2019 (UTC) --Relisting. bd2412 T 17:26, 11 May 2019 (UTC)

• support Formulated in terms of set theory, but used only in (some branches of) category theory. Boris Tsirelson (talk) 17:28, 3 May 2019 (UTC)
• Support the move, as the sources I found were all from the categorical or closely related literature. I can understand the desire for no redirect, but I would only support that option if links to this article are all updated to point to the new title; there is no good reason to break wiki links. --{{u|Mark viking}} {Talk} 17:32, 3 May 2019 (UTC)

  Agree that reasonable uses should be re-pointed to the new name, regardless of whether the redirect is kept. — Arthur Rubin (talk) 20:24, 3 May 2019 (UTC)

• support I agree with Arthur. Eozhik (talk) 17:37, 3 May 2019 (UTC)
• Oppose The target would be better as conglomerate (mathematics) like function (mathematics) or set (mathematics). We don't need more precision in the title as such details are best explained in the body of the article. Andrew D. (talk) 18:00, 3 May 2019 (UTC)

comment Also "conglomerate (mathematics)" looks good for me. Boris Tsirelson (talk) 18:12, 3 May 2019 (UTC)

I can support "conglomerate (mathematics)" as well. --{{u|Mark viking}} {Talk} 18:16, 3 May 2019 (UTC)

comment This term is too rare. It will be an exaggeration to say that it is used in mathematics. Several authors in category theory, and that is all. Eozhik (talk) 08:18, 4 May 2019 (UTC)

comment. I see no references other than in and/or referring to category theory. I would accept a redirect from conglomerate (mathematics) to conglomerate (category theory), but conglomerate (set theory) is just wrong. — Arthur Rubin (talk) 20:01, 4 May 2019 (UTC)

• Comment. I removed the link from class (set theory) as inappropriate, but left the link from von_Neumann–Bernays–Gödel set theory, although I believe the statement to be misleading, as vNBG is a theory, and no theory of conglomerates has been pointed to, as of yet. — Arthur Rubin (talk) 20:21, 3 May 2019 (UTC)
• I have updated the links from Von Neumann–Bernays–Gödel set theory, Conglomerate, Category of small categories, and Glossary of category theory to point to the proposed new title, which I had also made into a redirect for now. GeoffreyT2000 (talk) 19:12, 8 May 2019 (UTC)
  • Thanks for your help. With updated links, a move without redirect is fine with me. --{{u|Mark viking}} {Talk} 19:39, 8 May 2019 (UTC)
• Oppose Yes the main application of conglomerates is category theory, but the notion itself is from the set theory. This is set-theoretic foundation of category theory. Category theory folk do not use this term unless they talk about set theoretic foundation. There is no known contradiction with the right definition, so current title is not misleading. Conglomerate (mathematics) is a good alternative. Anyway redirect is required. Alexei Kopylov (talk) 08:29, 9 May 2019 (UTC)
• Support alternative move to Conglomerate (mathematics). This seems to be the most reasonable alternative, with most of the initial supporters of the proposed move commenting that it is acceptable. Unless there is a competing use of "conglomerate" used as prominently in some other field of mathematics, the rarity of the descriptor is no barrier to the use of "mathematics" as a disambiguation term, compared to things like Conglomerate (company) and Conglomerate (geology). bd2412 T 17:30, 11 May 2019 (UTC)
• Support Conglomerate (mathematics) to avoid needing to determine which subfield the term is part of, since there seem to be no other senses of "conglomerate" in mathematics. -- King of ♥ ♦ ♣ ♠ 20:58, 15 May 2019 (UTC)

---

The above discussion is preserved as an archive of a requested move. Please do not modify it. Subsequent comments should be made in a new section on this talk page or in a move review. No further edits should be made to this section.

What IS the theory?

ZF + MK^P(U)

ZF + vNBG^P(U) (probably implies the former)

ZF + ZF^U (probably still implies the former, but with class being a defined term)

KP + vNBG^P(U)?

Z + MK^P(U) (probably adequate for category theory; but I don't think vNBG^P(U) implies MK^P(U) within Z)

In any case, "set" is an element of U and "class" is a subset of U.

All this is original research on my part, as I cannot find the axioms of ACG, or of any system, in any of the references I have looked at. Category people might not care which implementation is made, but any that we can find in reliable sources should be included if the article is to remain as (set theory). Arthur Rubin (talk) 20:59, 4 May 2019 (UTC)

From [2, page 330]: "Begin with the usual axioms for Zermelo-Fraenkel or Gödel-Bernays-von Neumann set theory."

From [1, page 16]: "Notice that this Axiom of Choice implies an Axiom of Choice for Classes and also the familiar Axiom of Choice for Sets."

I think, the question "What IS the theory" applies to category theory no more than to (say) topology, or algebra. Generally, they are insensitive to reasonable variations of the underlying set theory. In rare cases of such sensitivity, it is stated explicitly. Right? Boris Tsirelson (talk) 04:23, 5 May 2019 (UTC)

And, I guess, ZFC+ZFC^U looks the most natural choice, and is equivalent to ZFC+MK^P(U), which confirms the insensitivity. Right? Boris Tsirelson (talk) 16:11, 5 May 2019 (UTC)

I would think that KP+vNBG^P(U) should be adequate, but category theorists generally don't care. (KP is essentially the minimum required for transfinite recursion over non-set ordinals to be defined.) I sincerely doubt that we need to go to vNBG+ZF^U. Still, at this point, it's original research on my part. — Arthur Rubin (talk) 06:17, 6 May 2019 (UTC)

As an aside, my late mother, and one of her students, were working on topology without AC. I don't know of further progress; her death was about 17 years ago. My father and I had been working on adapting a 1939 Czech paper to work without AC. Topology is significantly different without AC, but I don't know who is studying it, if anyone. — Arthur Rubin (talk) 05:54, 7 May 2019 (UTC)

Oh yes, without AC it should be different. Indeed, $\omega _{1}$ may be a countable union of countable ordinals! Boris Tsirelson (talk) 06:29, 7 May 2019 (UTC)