Talk:Banach–Tarski paradox

From Wikipedia, the free encyclopedia

WikiProject Mathematics
Mathematics Portal
This article is within the scope of WikiProject Mathematics, which collaborates on articles related to mathematics.
Mathematics rating: B Class Mid Priority  Field: Foundations, logic, and set theory
One of the 500 most frequently viewed mathematics articles.

Contents

Can this construction (perhaps the one using five pieces) be shown in a picture? Can it be animated on a computer screen (in simulated 3D)?

I find pictures much more intuitive than symbolic manipulation. David 16:10 Sep 17, 2002 (UTC)

All Banach-Tarski constructions involve non-measurable pieces, so I don't think a useful picture (or animation) would be possible. --Zundark 16:46 Sep 17, 2002 (UTC)

What about the series of pictures in Scientific American magazine some years ago that showed how a ball can be sliced up and rearranged to become something else? (I forget what it was.) I believe that example involved non-measurable pieces as well. I think non-measurable pieces can be visualized, just not realized in nature because they may be infinitely thin or whatever. In any case, the pictures were interesting. David 17:05 Sep 17, 2002 (UTC)

A piece which is infinitely thin has Lebesgue measure 0. Non-measurable pieces are much worse than this, and cannot even be explicitly described. I still maintain that a useful picture of a Banach-Tarski dissection is not possible, especially as it's impossible to even specify such a dissection (rather than merely prove one exists). I can't comment on the Scientific American pictures, as I haven't seen them. --Zundark 18:27 Sep 17, 2002 (UTC)

The paradoxical decomposition of the free group in two generators, which underlies the proof, could maybe be visualized by depicting its (infinite) Cayley graph and showing how it consists of four pieces that look just like the whole graph. AxelBoldt 18:41 Sep 17, 2002 (UTC)

The following text has been copied from Talk:Banach Tarski Paradoxical Decomposition:

How about a full name for Hausdorff so it can be linked?

Is this "doubling the interval" thingie related to the fact that on the Real line there are the same number of points in any interval of any length? Or am I simply showing my ignorance? Seems we need an article on infinity. --Buz Cory

---

I've put in Felix Hausdorff's first name, but there's no article for him yet.

Doubling the unit interval would be impossible if the doubled interval didn't have the same number of points as the original. But there's more to it than that, because only countably many pieces are used, whereas breaking it up into individual points would involve uncountably many pieces.

Zundark, 2001-08-09


[edit] Clarifications requested

Can you please clarify (in the article) the following basic points:

  • First, the article should explain what it means by saying that the two balls are of the "same size" as the original, since it implies later in the article that "size" is not meant in the "usual" sense of measure theory.
  • Second, it should clearly distinguish the theorem from the "ordinary" fact that any infinite set (e.g. two spheres) can always be identified bijectively with a similarly infinite subset (e.g. one sphere) of itself.
  • Third, there seems to be in a contradiction: in the introduction, it talks about a "solid ball" (which sounds like a sphere-bounded volume) whereas the proof talks about S2 (which usually denotes a spherical surface).

Steven G. Johnson 04:07, 22 Mar 2004 (UTC)

  • It means size as in Lebesgue measure. Which part of the article do you think implies something different?
    • It says something about the pieces not being measurable, so combined with the fact that it never defines "size" I found this confusing. The main point is that it should clearly define "size". Steven G. Johnson
  • It says at the beginning that the decomposition is into finitely many pieces. This clearly distinguishes it from a decomposition into uncountably many pieces. If you don't think this is clear enough, perhaps you could add a clarification yourself.
    • I don't think the article as it stands is clear enough, simply because people might be confused into thinking that the mere identification of two spheres with one is the paradoxical part (most people aren't familiar with the fact that infinite sets can be identified with subsets of themselves). Steven G. Johnson
  • The proof talks about S2, but at the very end it shows how to extend this to the ball (with a slight fudge). This is also explained at the beginning of the proof sketch, so I don't think it requires further clarification.

--Zundark 10:13, 22 Mar 2004 (UTC)

I've tried to clarify the above points; please check. I'm still a bit confused by your saying that "size" is meant in the sense of Lebesgue measure, since later in the article it talks about there being no non-trivial "measure" for arbitrary sets. I guess the point is that the pieces are not measurable, but their combination is? Steven G. Johnson 21:53, 22 Mar 2004 (UTC)

The revised definition is much more clear, thanks! It is great to have a formal definition of what is actually being proved. Steven G. Johnson 03:12, 23 Mar 2004 (UTC)

Thanks, I think we should not push too hard on these clarifications, it is very nice article, I belive further clarifications might make too havy. Tosha 04:58, 23 Mar 2004 (UTC)


As I understand the definitions of S(a) in the section about paradox decomposition, we mean S(a) = \{af | f \in F_2\}. Now note that there exist strings in F2 which start with "aa..." which means that by aS(a − 1) we would get the whole free group back and not only the part without S(a − 1) as indicated by the picture. To point this out: aS(a^{-1}) = \{ax | x \in S(a^{-1})\} = \{aa^{-1}x | x \in F_2\} = F_2 --74.236.150.135

If the first letter is a − 1, then the second letter can't be a (because we're talking about words in reduced form). So aS(a − 1) doesn't contain any words starting with a. Note that the picture shows aS(a − 1) as the complement of S(a), not the complement of S(a − 1). --Zundark 10:38, 27 November 2006 (UTC)

[edit] end of proof

I removed the last part from the proof, it is a sketch, not a prrof and I think such details should not be covered. Tosha 01:49, 5 May 2004 (UTC)

By the way, there is a new page Hausdorff paradox. I don't feel qualified to work on the content; but it is clearly very close to this page. If this is becoming a featured article candidate, perhaps including tha material might make this page more complete.

Charles Matthews 07:57, 5 May 2004 (UTC)


[edit] Curiosity

I came across an interesting anagram of "BANACH TARSKI".

It's "BANACH TARSKI BANACH TARSKI". —Ashley Y 10:45, 2004 Jul 9 (UTC)

LOL! Loisel 00:29, 1 May 2007 (UTC)

[edit] BTP & arguments against axiom of choice

The usual argument against the idea that the BTP genuinely undermines the plausibility of the axiom of choice, is that the axiom of choice allows one to construct non-measurable subsets, which it is wrong to regard as "pieces" of the original ball: instead they interleave with each other to an infinitesimally fine degree, allowing a trick rather similar to Russell's Hotel to be carried out. What's puzzling is the intrusion of a paradox of the infinite into what at first glance appeared to be a statement of geometry.

I'm not applying an edit directly, because this issue has ramifications elsewhere, and I haven';t time to properly formulate the text right now. Changes are needed:

  • cut it up into finitely many pieces -- very misleading description
  • it is a paradox only in the sense of being counter-intuitive. Because its proof prominently uses the axiom of choice, this counter-intuitive conclusion has been presented as an argument against adoption of that axiom. -- needs counterargument here
  • Changes needed in AofC page ---- Charles Stewart 20:07, 1 Sep 2004 (UTC)
I agree, but it was better, it is all result of the second edit of ArnoldReinhold, so I have reverted it and added all later changes. Tosha 04:06, 2 Sep 2004 (UTC)
I think I've maybe not made my complaint clear: the article begins with an description of the TBP that is couched in intuitive but contentious terms: people who say the resolution of the paradox is that our intuitions about cutting up solids and applying spatial transformations only applies to measurable connected subsets (as indeed I do) will object to the way the topic is framed; the reversion hasn't changed much. I'm not going to have much time for wikiing in the next ten days, but I plan on applying some changes then. ---- Charles Stewart 09:07, 2 Sep 2004 (UTC)

I wonder if replacing "cut it up into" with "divide it into" would be clearer, since the division isn't a knife cut style topological division. I also think the last sentence of the introduction starting "actually, as explained below" is confusing and arguably not neutral, since the previous sentence provides both points of view. Any objections to replacing that phrase and deleting that sentence? Warren Dew 23:53, 12 March 2007 (UTC)

[edit] Hausdorff paradox

The proof here is closer to Hausdorff paradox I think to move it there and leave this page with no proof. Tosha

[edit] Picture?

I can provide an illustration of Step 1 of the proof sketch, namely the paradoxical decomposition of F2. I envision something like the picture at free group, but with the sets S(a − 1), aS(a − 1) and S(a) marked and labelled. Is there interest? --Dbenbenn 01:35, 6 Dec 2004 (UTC)

I think it wold be great (with colors?)... Tosha 02:50, 6 Dec 2004 (UTC)

Done! Is it clear, or can it be improved? --Dbenbenn 04:56, 13 Dec 2004 (UTC)

Very nice I think Tosha 07:18, 13 Dec 2004 (UTC)

[edit] A simple picture

The Banach-Tarski "paradox": A sphere can be decomposed and reassembled into two spheres the same size as the original.
The Banach-Tarski "paradox": A sphere can be decomposed and reassembled into two spheres the same size as the original.

Anyone want this picture on the page?

  • Looks good. I'll put it in. Eric119 23:34, 5 Feb 2005 (UTC)


The fact that the free group can be so decomposed follows from the fact that it is non-amenable. I think we should put this in - It makes the discussion a little more transparent.

[edit] Similar Result

I remember reading somewhere about a similar result, the name of which I can't remember. I don't know if the two are related theoretically, but they remind me of each other. The other result says that a set of points exists in euclidean space such that the projections of the set onto different planes can produce any set of 2 dimensional images you want. Does anyone know the name of this, or has anyone even heard of it at all? It seems like a link from this article to one on this other result might be useful. --Monguin61 10:00, 10 December 2005 (UTC)

If anyone else is interested, the idea I was remembering was the digital sundial. Wikipedia's entry on sundials has an external link to the patent of an actual digital sundial, and as of now they are available for purchase from www.digitalsundial.com. The design of the physical device was the result of the work of one K. Falconer, and the idea is described, presumably in more detail, in his book on fractal geometry. --Monguin61 01:23, 15 December 2005 (UTC)

[edit] Is the Paradox related to the axiom of choice?

Banach–Tarski paradox is indeed counter intuitive but I'm not sure that what makes it counter intuitive is related to the axiom of choice. I would say that by using the axiom of choice one can get a counter intuitive result from another one which is already counter intuitive :

Think about a subset of the ball as something "covering" some part of the ball. The intuition tells us that when you move by a rotation a subset of the ball into the ball you will "uncover" some points and also "cover" some points which where "uncovered" before applying the rotation, but that these two phenomena will compensate each other. In fact this is not true in the non measurable world and this is the heart of the Banach–Tarski paradox.

More precisely the intuition tells us that it is impossible to find a subset X of the unit sphere (the sphere bounding the ball), apply a rotation to it and get a subset strictly included in X.

Similarly intuition tells us that it is impossible to find a subset X of the unit sphere which is strictly included in the sphere, apply a rotation to X and get something which stricly contains X.

This is nevertheless true :

Using the same notation as in the article take some x lying in the sphere (not on the rotation axis) and consider the subset of the sphere X = S(a − 1).x. If one applies the rotation a − 1 to it we get something stricly included in X and if one apply the rotation a to X one get something that strictly contain X. This is due to the fact that

a.S(a^{-1})=S(a^{-1})\cup S(b)\cup S(b^{-1}) and the way H is acting on the unit sphere.

You should realize that the Banach-Tarski paradox is only somewhat more strange than the fact that e.g. applying a shift to the left by 10 units of length to the set of all integers greater than 10 on the real line (and labeling corresponding points with their new values) produces a strictly bigger set - the set of all positive integers.

Hi, new to wikipedia so not totally sure how to talk in here, apologies. Just to pick up on this; the set of all positive integers is not a bigger set than the set of all integers greater than 10. Intuitively it seems to be, but there is a bijection between the two sets given by the function, i.e. f(x) = x - 10. So you can pair off one number from each set together: (1,11), (2,12), (3,13),... etc. In this way, the two sets have the same size. So the Banach-Tarski result is actually much stronger. Bobmonkey 19:27, 13 November 2006 (UTC)
To expand on that, it's only with the Axiom of Choice that you can create those nonmeasurable sets that seem to become 'three times larger' when rigidly rotated. Without the Axiom of Choice, those sets don't exist, and the problem doesn't arise. Warren Dew 19:53, 12 March 2007 (UTC)
Um, that's a little confused. Axioms aren't something you use to create sets, nor do they have any bearing on whether sets exist. Without the axiom of choice you can't prove that the sets exist. This is an epistemological rather than an ontological limitation. --Trovatore 20:08, 12 March 2007 (UTC)
Without the Axiom of Choice, you can't prove those sets exist. If you can't prove they exist, why would you assume that they do? As someone who is not a fan of the Axiom of Choice, I would assume the opposite, that they don't exist. Problem solved. Warren Dew 23:03, 12 March 2007 (UTC)
Whether they exist or not is independent of whether I can prove it, or of whether you assume it. It has nothing to do with you me at all. --Trovatore 23:19, 12 March 2007 (UTC)
Hey. Without the Axiom of Choice, you cannot show that nonmeasurable sets exist. This whole Banach-Tarski argument was originally intended (I believe) to show that the Axiom of Choice (AC) can be rejected as it does come up with such seemingly nonsensical consequences. Although nearly all mathematicians nowadays use AC, some do not accept it.
This is why the existence of these sets is in contention; if you do not accept AC, such sets cannot be shown to exist, so you do not accept the existence of such sets. I, on the other hand, assume AC is true, and so accept the existence of these sets. So it does make a big difference as to which mathematical perspective you look at this from!
I see your point that in that such sets existing ought to not be a matter of opinion, but whether or not you allow AC gives two different mathematical "worlds" - one where such sets can be shown to exist, and one where there is no reason at all to assume the existence of such sets. In this second case, you would assume that they do not exist, unless you could find a way to prove that they do without using AC. The article on Zermelo-Fraenkel set theory gives more information about all this. I hope this makes sense... :)Bobmonkeyofdoom 22:03, 4 April 2007 (UTC)
No, you're conflating epistemology and ontology here. Axioms don't give you worlds; they make assertions about those worlds. Now formalists will not accept the existence of those worlds at all, and that's fine; but even from the formalist point of view it's not accurate to speak about the axiom "creating" the sets. What the axiom does is enable a formal proof of a formalization of the claim that they exist; that formalization, for the formalist, is strictly speaking an uninterpreted string.
Of course this sort of syntax-only formalism is not the only alternative to realism; there are more nuanced intermediate views, like some sort of fictionalism, where the objects are considered not to exist in reality, but to be used by humans as a convenient guide to thought. Even in that view you shouldn't speak of assertions creating objects, though; it's a bad metaphor and therefore a bad guide to thought. --Trovatore 22:38, 4 April 2007 (UTC)
I agree with everything you say here! I don't ever talk about axioms "creating" sets (or at least I didn't mean to give that impression), just that whether or not such sets exist is predicated on whether or not you accept AC as true. Is this a fair comment? I bow to your superior knowledge as having a PhD in Set Theory :) But as AC is not dependent on ZF, can you not either accept or not accept the existence of nonmeasurable sets based on whether you accept AC?Bobmonkeyofdoom 00:07, 5 April 2007 (UTC)
Yes, that's fine. Except of course that it's not an if-and-only-if -- AC guarantees that there are nonmeasurable sets, but ~AC certainly doesn't guarantee that there aren't. --Trovatore 00:46, 5 April 2007 (UTC)

[edit] endashes

Someone recently moved this article from Banach–Tarski paradox to Banach-Tarski paradox (substituting the endash with a hyphen). Really that's the name I'd prefer too; I don't like all these Unicodes that look almost like ASCII characters and can easily be mistyped. But the current WP standard seems to be endashes when the names of two workers are combined, so I moved it back. --Trovatore 23:57, 4 January 2006 (UTC)

I moved it back to a hyphen. I don't know what "current WP standard" you're referring to, but I've never seen an en dash used in this type of case anywhere else. In particular, in TeX it's terribly easy to make an en dash (just type --), but even in TeX people use just a hyphen. dbenbenn | talk 12:01, 15 January 2006 (UTC)
Well, the case has been made that endashes are the correct way, so that we can know that Banach and Tarski are two people. I have come round to agreeing, despite the potential inconveniences. Birch–Swinnerton-Dyer then uniquely parses to two people. So, please don't move back like that. No one should get fanatical about format issues, but making a stand on them is just going to become a time-sink and distraction. Charles Matthews 19:30, 15 January 2006 (UTC)
The question is, which use is more common? It isn't up to us to determine whether an en dash is "correct". And in my experience, Banach-Tarski paradox is always written with a hyphen. I have never seen it written with an en dash. (Whereas mathematical papers written in TeX always use an en dash in something like Birch–Swinnerton-Dyer conjecture, precisely because using a hyphen would be ambiguous.) dbenbenn | talk 09:37, 16 January 2006 (UTC)
I have moved the page back to a hyphen for the aforestated reasons. What is more intuitive to type? wiki/Banach-Tarski paradox or wiki/Banach%E2%80%93Tarski_paradox?Kyle McInnes (talk) 15:40, 20 October 2006 (UTC)
It's irrelevant what's easier to type, as long as there's a redirect. Why do people keep bringing up the typing thing? That's what redirects are for. --Trovatore 17:28, 20 October 2006 (UTC)
Apologies. Thanks for creating the redirect. Kyle McInnes (talk) 17:41, 20 October 2006 (UTC)
No prob. For future reference, redirects are automatically created by page moves; it was there before this latest move-and-move-back. --Trovatore 18:59, 20 October 2006 (UTC)
The claim is that the endash is standard, in text prepared going through a professional copy-editing process. But, as I say, I don't think time should be spent on warring over such issues. Charles Matthews 10:14, 16 January 2006 (UTC)
As far as I understand, it should be with endash. (One can do anything in his TeX-file but for encilopedia should follow standards, also responsible editor would change it for publication). Tosha 16:17, 19 January 2006 (UTC)

[edit] Layman's section

This is a popular paradox, and I believe that there are (relatively) many laypersons who may have a passing interest in what this thing is all about. In my opinion, it would be nice if there was a section here that would require very little background, intuitively explaining what the thing is about and wouldn't require much time to read (no equations!).

Sketch of how such a section could be done:

  • Size comparison. Explain that if there is a one-to-one correspondence (a bijective function) between two sets, they aren't necessarily of the same size in the ordinary sense. Example about [0, 1] <--y=2x--> [0,2]. This is why measure theory is necessary. Measure theory gives a notion of size (measure) that works much like Riemann integrals do (I assume the layperson knows some basics about integrals, if he doesn't, all explanation is futile), but is more general. Even so, the "problem" with measure theory is that not all sets are measurable, just like not all functions can be integrated. These non-measurable sets are weird beasts for which our intuition does not work. (Are all/most fractals measurable? Would be good example either way, as they're also famous "weird geometry" beasts) See also Cantor set for a set with infinite number of points but measure of 0 (infinite amount of "Cantor dust" is still just dust; similarly, a curve has measure of 0 in two dimensions, because it does not have area, except for weird beasts).
  • Show that a set of points can easily be duplicated, if there are no restrictions as to what you can do with it. E.g. take a unit cube, cut it in two at x=0.5, then (translate and) multiply x of each point in the set with two, giving two exactly same sets of points as the original cube. This works the same way as one-to-one correspondence way of comparing set size: you might think that scaling x this way would leave "gaps" in the end result, but this is not so. Infinite sets are weird enough as they are. But the Banach-Tarski paradox is more paradoxical than just this.
  • The paradox in Banach-Tarski paradox is that if you take a (solid) sphere, cut it in 5 pieces (although non-measurable, weird beast pieces), then move and rotate these to their places, you get two exact copies of the original sphere. Being exact copies, they of course have the same measure, too. Note that there is no scaling here, and no carving of the curve to infinite number of pieces, or anything. All the "funny stuff" is in what the 5 pieces look like. Being non-measurable, such pieces of course can't exist in the real world, as all real world objects are measurable (we can't even carve things up to infinite precision, much less somehow invoke the axiom of choice).
  • Of the five pieces, one is for "fixing up" a small (countable) amount of gaps in the actual construction done with the other four pieces. The idea is to split the sphere in 4 parts in such a way that you can reassemble the original sphere from two of them, thus getting two original spheres. This reassembly is possible, because two of the pieces, which are supposed to be one-quarter of the size of the original sphere, actually become 3/4ths of the sphere when rotated by about 70.5 degrees. By performing this rotation on both of them, we get two 3/4ths of the sphere, and still have two 1/4 pieces to match. Combining these, we get two full spheres. However, this is merely an intuitive description, as the pieces are in fact non-measurable, and thus a notion that they are "quarter" or "3/4th" of the original sphere has no meaning. The construct thus escapes from the common fact that rotation preserves size by rotating pieces that do not have size. It just so happens, that when these pieces are combined to again produce measurable objects, the total size has magically doubled in the process.
  • For details, see the actual proof.

I don't know how exactly to phrase this, and my knowledge of the matter is rather superficial (I'm pretty much a layperson myself!), so I didn't dare actually try to modify the article.

The goals I was thinking of are:

  • Do not require much knowledge. (high school math is enough)
  • Provide (non-obscure!) references for further reading, to connect all this to a larger background.
  • Be clear in what is paradoxical and counterintuitive. This means explicitly pointing such things out.
  • Be clear in what is not paradoxical or counterintuitive. This means explicitly stating what part the paradox is restricted to. "At least I don't have to think about these parts" makes for easier comprehension.
  • Be clear in what the Banach-Tarski paradox is about, as opposed to what other (semi-)counterintuitive results there are.
  • Try to keep reader's interest instead of merely stating the facts.

Too much to ask? Maybe. But in my opinion, an explanation like stated above would exist in an ideal encyclopedia entry, as not all interested readers know much of anything about modern math. (and an ideal encyclopedia is for everyone, right? :-)

It may be that some of the above would be better to put in a general article about geometrical paradoxes (and a link from here to that), but I think that merely referring to the articles about measure theory, related paradoxes, etc. would scare many away. Reading a layperson's section should require just about the same amount of effort as reading a popular science magazine article, and chasing hyperlinks in search of comprehensible and relevant information is a far cry from that.

130.233.22.111 16:35, 6 February 2006 (UTC)

Frankly... YES PLEASE!! I have not got a clue what this article is on about. I agree the subject of the article is counter-intuative but I cannot comprehend any of the evidence for it therefore, as a layperson, this article is gibberish, sorry. 0.999... I can get my head round but this is utterly inpenetrable, sorry.AlanD 19:55, 20 August 2007 (UTC)

[edit] On hyphenated hyper-2-tuples

I went ahead and moved it (no admin assistance required). You can't move an article onto another real article, but you can move article A onto article B when B is a redirect to A, unless B has some other history. Not sure what the rule is if B has at some point been a redirect to C. --Trovatore 17:13, 9 March 2006 (UTC)

[edit] Doing the decomposition physically

In a recient article in the Journal of Symbolic Logic, Trevor M. Wilson titled "A continuous movement version of the Banach—Tarski paradox: A solution to de Groot's Problem", it was proved that the dissaembly and reassembly of the balls can be performed by continuously and rigidly moving the pieces without the pieces ever intersecting at any point in time. Should this fact be mentioned and referenced? --Ramsey2006 22:46, 13 October 2006 (UTC)

The pieces themselves are still infinitely convoluted and cannot actually be constructed physically (atoms are of a finite size, after all). So this still doesn't get anywhere near doing the decomposition "physically". But the fact that they can be translated, rotated, and reassembled without ever intersecting may be worth the mention.—Tetracube 06:01, 15 October 2006 (UTC)

Done. I also removed a confused paragraph about it not being a real paradox, which in addition to being dubious was also in a completely inappropriate section. --Trovatore 06:40, 15 October 2006 (UTC)

I've changed the intro sentence to say "finitely many infinitely convoluted non-overlapping point sets", rather than "finitely many pieces" -- I think the core of the paradox is that the term "pieces" is misleading, since the normal common-sense intuitions regarding volume and solidity no longer apply to point sets of the type required. Although this wording is slightly awkward, I think it's appropriate in this case, because it avoids the far greater confusion that the use of the simple but misleading word "pieces" would generate, and the meaning is fully explained a couple of paragraphs below. -- Gigacephalus 08:25, 13 November 2007 (UTC)
I like the use of "point sets", at least in conjunction with the rest of the edits you did. I may break the first sentence into two in order to allow people to get a bit futher into the article before taking on much math.Warren Dew 21:37, 13 November 2007 (UTC)

[edit] Improving the text

This is a fantastic article, with quite accessible proof of the Banach–Tarski theorem about doubling the ball! Unfortunately, there are some inaccuracies and OR statements scattered around. For example, after reading Banach and Tarski's paper, I didn't get the impression that they intended it to somehow undercut the axiom of choice, as the text had claimed (hence I've removed that sentence). What they say is that

Le rôle que joue cet axiome dans nos raisonnements nous semble mériter l'attention

and go on remarking how for two key results of their paper, the proof of the first uses the axiom of choice in a much weaker way then the proof of the second.

I also regret that a clear explanation of the difference between one and two dimensional Euclidean spaces (where a paradoxical decomposition of this type is impossible) and the higher dimensional cases, related to amenability of the Euclidean group in low dimensions and non-amenability in high dimensions, is missing. While it is more relevant for the general theory of paradoxical decompositions, it seems prudent to have at least one section on this in the present article. Arcfrk 02:54, 3 September 2007 (UTC)

I've made substantial revisions, in particular, adding references to amenability in a few places. Arcfrk 09:50, 3 September 2007 (UTC)

[edit] Banach-Tarski and Hilbert's Hotel

I have a feeling like the Banach-Tarski paradox is something like the continuum equivalent to Hilbert's Hotel, where it's possible to put additional guests into a hotel that already has all rooms occupied. The hotel creates something (space for new guests) from seemingly nothing by exploiting its infinite nature, which seems counter-intuitive since such exploits don't work in the physical (finite) world - just like the Banach-Tarski thing. Can someone with a deeper understanding of how the proof works elaborate on this? wr 87.139.81.19 (talk) 12:45, 23 November 2007 (UTC)

Not much there. Hilbert's Hotel requires no skill, this requires ... skill. Charles Matthews (talk) 15:53, 23 November 2007 (UTC)

[edit] exposition of Pawlikowski result could be misleading

The article currently says

In 1991, Janusz Pawlikowski proved that the Banach-Tarski paradox follows from ZF plus the Hahn-Banach theorem. The Hahn-Banach theorem doesn't rely on the full axiom of choice but can be proven using a weaker version of AC called the ultrafilter lemma. So Pawlikowski proved that the set theory needed to prove the Banach-Tarski paradox, while stronger than ZF, is weaker than full ZFC.

This phrasing, while it doesn't actually say so, kind of makes it sound as though it wasn't known before 1991 that full AC is not needed to prove Banach–Tarski. That is surely not the case. It's obvious from even a high-level description of the proof that the key use of AC is to choose elements from equivalence classes of points in R3 -- that is, essentially reals -- which means it follows from the existence of a wellordering of the reals. And it must have been known since the sixties that it's consistent with ZF that the reals can be wellordered but some larger set (say, the powerset of the reals) cannot.

Any suggestions on how to rephrase, while still getting the valid part of the point across? One possibility would be to find a reference from earlier that specifically mentions that BT is weaker than full AC -- does anyone have one? --Trovatore (talk) 20:44, 20 December 2007 (UTC)

I see what you mean, I went overboard slightly, sorry. A good place to look for cites of earlier work is probably Pawlikowski's paper itself--does anyone have access? As mentioned in the edit summary I got the cite from the Hahn-Banach talk page (plus Polish wikipedia), though I googled a little more info about the result before making that edit.

Btw what inspired this is I've been wondering for a while whether BT follows from ACA0, which is sort of plausible because Brown and Simpson proved that H-B for separable spaces (which would include R3, I'd expect, but I haven't examined the Brown&Simpson result) follows from WKL0 which I think is even weaker. ACA0 is basically the system of Weyl's "Das Kontinuum" according to Feferman. Warning: what follows after this is total OR and possibly nonsense. But basically these systems are about the minimum needed for doing functional analysis and therefore for doing quantum mechanics. What I'm getting at is that maybe it's hard to axiomatize physics in a way that doesn't lead to BT. So instead of being a weird artifact of set theory, BT becomes in some sense a theorem of physics, a somewhat disturbing notion. 75.62.4.229 (talk) 12:48, 21 December 2007 (UTC) (reworded slightly 18:47, 21 December 2007 (UTC))

I found and added a pdf link to Pawlikowski's paper (which is just one page long) and a related one. Still not sure how to describe the result's significance. 75.62.4.229 (talk) 21:55, 22 December 2007 (UTC)

[edit] no general meaning of explicit construction

The article had claims about the lack of existence of an explicit construction. The proof of the Banach-Tarski decomposition begins by establishing a free action of F2 on the unit sphere, and then asks for a set containing exactly one point from each orbit of the group action. Although last step requires some form of choice, it would still be considered "explicit" by many mathematicians I have met; they would consider the entire proof an explicit construction, although not a canonical one. While it would be possible to make those claims precise by referring to the definability of the decomposition, it isn't accurate to simply say that the construction is nonexplicit, because there is no generally accepted meaning of an explicit construction in mathematics. I rephrased a few sentences to remove the issue. Also, it isn't particularly accurate to talk about 'algorithms' when discussing constructions in set theory; it confuses set-theoretic constructions with computable constructions. — Carl (CBM · talk) 15:54, 14 January 2008 (UTC)

These fairly charged statements had been introduced into the article just a few hours earlier, and frankly, I don't quite like most of the changes. Maybe we should just revert to the earlier version? Arcfrk (talk) 20:44, 14 January 2008 (UTC)
I hope the words are not charged, but neutral. The article before was biased towards the point of view that R is well-orderable. The BT construction is not at all explicit, because it requires a choice of an element from uncountably many sets. There is no algorithm to decide which points are in the sets, nor is there any procedure which allows the sets to be generated by any process other than transfinite induction to the continuum. The construction fails in a model where all subsets of R are measurable, and such models are exactly the same as far as computations or explicitly definable sets are concerned. I wanted to make the article neutral towards different set theory models, especially with regard to models which are probabilistically intuitive, meaning ones where you can consistently talk about choosing real numbers at random. Every time somebody talks about choosing a real numbers at random, they are implicitly imagining a universe where every subset of R has Lebesgue measure.Likebox (talk) 05:56, 15 January 2008 (UTC)
("Charged" may not be the best term to use, but it appears that if you eliminated any bias, you've reintroduced a different form of bias and probably in larger quantities.) I am not sure that I get your point concerning algorithmic definability of the BT decomposition (which is not at all the same as "explicitness"), but then I don't have a PhD in Set Theory. The distinctions will likely be lost on the vast majority of the readers. Unfortunately, your edits removed a more plain word discussion of the meaning of the paradox and replaced them with quite technical and fairly polemical statements. The lead is not a good place for that! Arcfrk (talk) 06:15, 15 January 2008 (UTC)
Sorry, I might not have done a very good job. Perhaps "algorithmic" is better than explict? Just add the language that you feel is best, I was being bold, and perhaps I tilted the article the other way.Likebox (talk) 07:40, 15 January 2008 (UTC)
I think generally that it's not too controversial that there's no explicit example, but the problem is that the statement itself has no precise agreed meaning. The problem with "algorithmic" is a different one. In most formulations all computable functions from the reals to the reals are continuous, so that there isn't a computable partition of the ball into five nonempty pieces at all, never mind whether you can rearrange them paradoxically. --Trovatore (talk) 07:45, 15 January 2008 (UTC)
You are right, algorithmic is ridiculous. Maybe the right phrase is "not constructible by joining and intersecting regions"? All I wanted the reader to get is that the sections are scatterings of points, which are constructed by a transfinite induction, not visualizable regions defined by any normal geometric operations.Likebox (talk) 08:01, 15 January 2008 (UTC)
Well, it's certainly true that you can't do it with any countable collection of anything that could intuitively be described as "knife cuts" (that's a weak consequence of the fact that the pieces can't be Borel sets), and it would be worthwhile to get that point across. But it's tricky to think of any wording that conveys that intuitive idea, and still actually means something mathematically. --Trovatore (talk) 19:29, 15 January 2008 (UTC)
You're right. I give up. The only ways I can think of to phrase it are convoluted.Likebox (talk) 20:51, 15 January 2008 (UTC)
Ok, one last try, then I give up.Likebox (talk) 21:20, 15 January 2008 (UTC)

[edit] Edit on 2008-3-11

I undid an edit by Likebox. There are three specific parts that were undone:

  • The claim that the pieces are impossible to describe. This is simply false - the proof of the theorem gives a clear description of what the pieces are.
  • The claim that an infinite number of choices is required. Any proof, being finite, will only invoke the axiom of choice finitely many times.
  • That Alain Connes dislikes the theorem. This may be worth discussing somewhere, but not at the end of the very first paragraph. The vast majority of working mathematicians accept the axiom of choice, the existence of nonmeasurable sets, etc. A specific quote from Connes would establish context much better, in any case, than a generic reference to a book which is fundamentally about noncommutative geometry, not about the Banach-Tarski paradox. I would look up the quote by Connes if a more specific reference is given.

— Carl (CBM · talk) 15:42, 11 March 2008 (UTC)

"Infinitely many choices" doesn't mean "infinitely many uses of AC", it just means "at least one use of AC". (If there were only finitely many choices, AC wouldn't be needed.) --Zundark (talk) 16:06, 11 March 2008 (UTC)
Indeed, but the article already says that AC is required, lower in the lede: "What makes the paradox possible in set theory is the axiom of choice, which allows the construction of nonmeasurable sets, collections of points that do not have a volume in the ordinary sense and require an uncountably infinite number of arbitrary choices to specify." So unless something more than "AC is required" is intended, this doesn't need to also be in the first paragraph, which is already too long and detailed. — Carl (CBM · talk) 16:24, 11 March 2008 (UTC)
P.S. I do think a case could be made that each invocation of AC is a single choice - the choice of a set of representatives. — Carl (CBM · talk) 16:26, 11 March 2008 (UTC)
The article should clearly and quickly distinguish between a "theorem" like this and "all triangles have 180 degrees" or "there are infinitely many primes". Unlike the others, this "theorem" is a matter of opinion, and it's well understood that there is no right opinion.
I don't think anybody could sanely think that "uncountably many choices" means a proof with uncountably many invocations of the axiom of choice. But, kudos man, just thinking about a proof of uncountable length makes my brain get get all woozy!
I put in the caveats for the reader whose stomach becomes upset by counterintuitive results which contradict a correct intuition. This stuff is a bunch of set theoretic foolishness which has nothing to do with three dimensional geometry. It's really about the growth-rate at infinity for groups, which is a geometric notion on groups. The space-geometry part is just "the reals are an ordinal!" again, and there's a hundred years of this out there and its really getting tired. As for Alain Connes, bless his soul, his book explicitly advocates thinking of all subsets of R as measurable and you don't find many people fighting this fight. But it should be hashed and rehashed for the benefit of students, who need to be alerted as to when they are being sold a bill of goods.Likebox (talk) 09:07, 12 March 2008 (UTC)
Actually, I can see Likebox's point, but he's wrong in every detail. If he could propose an edit which deals with his concerns, but is not mathematically incorrect, it might be included. — Arthur Rubin (talk) 13:02, 12 March 2008 (UTC)
The problem with presenting things in a first-principles way is that the details look unfamiliar and the arguments need to be checked carefully because it is easy to make a mistake. I do it anyway, because the benefit is that it is much better pedagogically. In general, I don't think there is a good way of classifying people into "correct" and "incorrect". It is more useful to classify arguments into correct and incorrect. But everybody makes mistakes and I've made more than my share. In a forum like this, I don't worry so much about not getting corrected.Likebox (talk) 17:04, 12 March 2008 (UTC)
The article does already discuss the issue of choice and measurability - see the third and fourth paragraphs of the lede.
I still don't buy the "uncountably many choices", though. For one thing, the issue of countability/uncountability is a red herring - AC is needed to choose from countable collections of sets as well. But more importantly, the key property of the axiom of choice is that all the choice are made at the same time rather than one after another. The issue isn't whether it's possible to choose representatives separately, but whether it's possible to simultaneously select a representative from each of the sets in the collection. Saying "uncountably many choices" has a connotation that the choices are made one after another. — Carl (CBM · talk) 13:46, 12 March 2008 (UTC)
Yes, I know that the idea here is that the choices are made "at the same time", but that is the whole philosophical point. You shouldn't carelessly think of sets as already selected from a predefined universe of sets, that's leads to Cantor/Russell paradoxes. To actually make a theory, you need to think of them as part of a universe that's constructed step by step. In the step by step construction, choice functions are only "born" after uncountably many steps. The whole question revolves around whether you are going to view sets as preexisting, or constructed symbolically by a computational step-by-step procedure. The first point of view is, in my opinion, outdated.
Oh thank goodness, Likebox is here once again to save us from our "outdated" ways of thinking about mathematics. Can a 'computational' "proof" of the B/T Paradox be far behind? —Preceding unsigned comment added by 65.46.253.42 (talk) 20:13, 12 March 2008 (UTC)
The article discusses choice, but it doesn't do it with enough feeling in my opinion. The distinction between countable and uncountable choice is the essential one. Countable choice is not a big deal unless you're studying set theory, because you end up using it without thinking. It's really only choice for the continuum which is a problem. Uncountable choice on sets which are as big as the continuum is controversial because it's only purpose seems to be to prevent you from defining Lebesgue measure properly. This is central to how you view the real numbers, and it gums up any natural discourse about random objects.
I fortuitously stumbled on a respected source that says this (as an inconsequential aside)--- Alain Connes, and I think that justifies inclusion.Likebox (talk) 17:04, 12 March 2008 (UTC)
Could you give the page number for the reference to Connes? If he mentions it as an aside, why would that mean it should be included here? (In any case, as I pointed out, it is included here.) The axiom of choice in not controversial in the present day; it's accepted and used by the vast majority of mathematicians including noncommutative geometers. I'm interested to see what Connes has to say.
Also as an aside, in the canonical "constructed step by step" model of set theory, L, choice functions are all definable, and thus each can be constructed via a single application of comprehension. They aren't "born after an uncountable number of steps" any more than the entire set of real numbers is. — Carl (CBM · talk) 17:38, 12 March 2008 (UTC)
The pages is 51-52 and around there, and the discussion is very nice. It's an aside in the sense that it is outside the main line of development of the book.
You are right--- the step by step model has a choice function, but the construction of the step-by-step model makes it clear that it isn't constructing all of the universe. The reason is that L constructs "ordered" objects, the stuff that you can define iteratively in terms of text descriptions that are not "random". The intuition I have in this case is that it does not construct all of R, it "misses points", the random points.Likebox (talk) 20:41, 12 March 2008 (UTC)
Yep. And even some not-all-that-random ones. Like 0# for example. --Trovatore (talk) 20:49, 12 March 2008 (UTC)

[edit] Refimprove tag

I tagged this article with "Refimprove" to call for improved referencing. More use of in-line citations (footnotes) would be appropriate. For example, the first theorem stated is implied to be quote from the 1924 paper that appears in the bibliography below, but it would be appropriate to include a footnote at the location of the quote and to cite the article and its page number. doncram (talk) 22:11, 11 March 2008 (UTC)

No, that would not be appropriate. A careful exposition of all material appearing in this article is contained in the book of Wagon in the bibliography. The reference to Banach–Tarski paper (a primary source) is included here both for its historical value and because the full paper is freely available. It is not helpful to cite random pages, and I find it most ironic that an article with an abundance of well chosen sources has been tagged as "insufficiently referenced". Arcfrk (talk) 23:09, 11 March 2008 (UTC)
That isn't a direct quote - note the lack of quotation marks. It's simply a statement of the theorem, set off from the main text. — Carl (CBM · talk) 23:28, 11 March 2008 (UTC)
I'm sorry if the tag was inappropriate for this article. I only visited because CBM has been engaging in a discussion elsewhere with me, with a degree of bullying that I was beginning to experience as a personal attack, over the fact that I had chosen to put copied text into quotation marks. He was essentially taking the position there that full referencing of quotations is wrong. I wondered what articles CBM works on, and looked at this one that CBM has contributed to. Reading the article it seemed ambiguous to me, but perhaps I projected too much about CBM in assessing that this article was poorly referenced. sincerely, doncram (talk) 17:01, 12 March 2008 (UTC)

[edit] The Banach-Tarski paradox is, indeed, a theorem

I noticed Arcfrk removed this sentence:

Because there are points of view in mathematics which reject constructions of this type, this theorem is not considered to be true in an absolute sense, but only true relative to a particular system of axioms for set theory

While there are a very few mathematicians who dislike the axiom of choice, the consensus in the field is that the axiom of choice is a perfectly valid axiom and theorems proved with it are not "suspect" or "contingent" (no more than any other theorem). The lede does cover the minority point of view: "The existence of nonmeasurable sets, such as those in the Banach–Tarski paradox, has been used as an argument against the axiom of choice, although most mathematicians accept that nonmeasurable sets exist." That is about as much weight as the minority viewpoint should be given here, as it is exceedingly uncommon among contemporary mathematicians. — Carl (CBM · talk) 11:39, 13 March 2008 (UTC)

The fact that the axiom of choice is accepted and used without question does not mean the results which absolutely require choice are uncontingent. They are not "suspect" in the sense that they are going to lead to logical problems, they are "suspect" in the sense that they are useless and silly. Nobody worries about BT when constructing measures, because mathematics evolved a machinery of measurable sets and random forcing which allows you to sidestep it, so that in every way it is more productive to think of it as false. But it took fifty years to acquire BT immunity, and the machinery evolved in a hostile environment, and that makes it cumbersome and alien to a student.
This has nothing to do with disliking choice on sets. It has to do with whether the real numbers are a small enough collection to be productively described as a well-orderable set at all. Nearly all theorems are absolutely true, meaning that once you understand the proof, it can be easily translated from any one reasonable axiomatic system to any other. But a very few theorems, like BT, are true only relative to a particular axiomatization. When you claim to prove a counterintuitive geometrical theorem and that theorem is not true in an absolute sense, then this theorem is just an opinion. While I personally think it is an outdated opinion, I respect that it is a majority opinion, and it should be presented in the way that the majority believes is best. But it should not be presented as an unqualified absolute mathematical truth.Likebox (talk) 14:58, 13 March 2008 (UTC)
There's no benefit in discussing "useless" and "silly" here. I don't follow your second paragraph above; all formalized theorems are only true relative to their axioms being true. In any case, the Banach-Tarski paradox is considered by nearly the entire mathematics community to be an (absolute) theorem, and Wikipedia is not the place to argue for a change in that attitude. If the opinion of the mathematics community begins to change, then this article should reflect that.
The fact that the theorem uses the axiom of choice is explicitly covered in the lede; I don't see the need to give it more weight than the reasonable weight it is already given. — Carl (CBM · talk) 15:51, 13 March 2008 (UTC)
There are different axiomatizations of the same objects, and nearly all of the time, you can translate theorems back and forth between different formalizations. For example, you can translate any theorem in finite set theory to the original Peano axioms and vice versa. There are axiomatizations of the reals as a complete ordered field, and you can translate most theorems into this formalization. Usually the objects that you have in mind when discussing geometry can be approached from the point of view of either second order logic or the theory of countable sets, and the different axiomatizations and points of view give results which are consistent with each other and with set theory. Those theorems which do not change from axiomatization to axiomatization are absolute, and they are not wedded to any particular formal system.
The Banach Tarski theorem is in no way accepted by nearly all the mathematical community as an absolute theorem. What is true is that is a theorem of ZFC, and by social convention a ZFC theorem is accepted. This is to avoid squabbling over foundations when trying to evaluate correctness of a proof. My guess of the breakdown in opinion in the professional community would be: 45% really true because ZFC is reality, 45% true as a matter of social convenience, 5% false, 5% no truth value. But I didn't take a formal poll, and I might be wrong.


You _of course_ have some kind of citation or reference to back up this astounding claim, I trust? —Preceding unsigned comment added by 65.46.253.42 (talk) 20:27, 13 March 2008 (UTC)
Of course not--- I didn't think it was an astounding claim.Likebox (talk) 23:15, 13 March 2008 (UTC)
Just to elaborate--- the Banach Tarski paradox is well known to not be an absolute theorem, in the sense that alternate axiomatizations have every subset of R measurable. The question is to what extent working mathematicians have a bias towards one axiomatization or another. From my experience, most of them don't care about questions of axiomatization that are not directly relevant to their work, so that means that they don't have a strong opinion one way or another. But having said that, the breakdown should follow the usual platonism/formalism/constructivism proportions, which I think are in the ratio 45/45/10.
Here's an unrepresentative informal poll I found [1].
And I don't see the Banach-Tarski paradox mentioned there once. As has been pointed out, you are of course welcome to your (wrong) opinions about the nature of mathematics, but they don't belong in Wikipedia unless they can be substantiated (by something other than a PHPnuke forum thread) and are deemed, by editorial consensus, to be relevant.
The point is that mathematicians, like any other human community, are aware that sometimes living in a community takes some compromise, and choice is not the most horrible compromise. These compromises, however, are sometimes mistinterpreted as a consensus about mathematical truth, when they are not.Likebox (talk) 18:20, 13 March 2008 (UTC)
I defer to Carl, who is an expert on set theory and logic, concerning the perceptions of AC within those fields. As a general impression, I cannot recall ever personally encountering a mathematician who was afraid to use Zorn's lemma, well-orderness principle, and a few other workhorses of mathematical proofs dealing with infinite sets (and my sample size is in the hundreds, if not thousands). Contrary to what has been stated above, their use is nearly universal: much of basic analysis, algebra, and geometry uses them without explicitly acknowledging this fact, which is a clear sign how fringe is the view that Likebox is trying to peddle. Concerning axiomatizations: as far as I know, the consistency of neither ZF nor ZFC has been proved. In this sense, all of mathematics is on shaky foundation. I do not, however, consider either prudent or practical to insert a disclaimer of this nature into every single mathematical book, paper, and encyclopaedic article.
On the subject of compromises: some time ago, Likebox and I worked out a compromise that we will keep the beginning of the article friendly to people just trying to learn a bit of interesting mathematics, without having to endure a mile long list of qualifiers (which I deeply cared about), and he can insert his onerous disclaimers elsewhere. However, he mistook this compromise for universal agreement with his misinformed interpretations, which it was not. Arcfrk (talk) 00:13, 14 March 2008 (UTC)
It is unfortunate that you are taking this as an attempt on my part to insert my own interpretation into an article. It isn't. First of all, I don't have any problem with choice or Zorn's lemma, they are fine. I am only saying that when you prove a theorem like this one, every sensible mathematician knows that it does not have the same truth value as "integer multiplication is commutative". It is a matter of axioms and opinions, and it is only a social compact that makes you think of it as true. This is the last I am saying on the subject.Likebox (talk) 01:46, 14 March 2008 (UTC)


"This is the last I am saying on the subject" -- Oh now President Likebox, I think we both know from experience that's just not true Dolores Landingham (talk) 18:26, 14 March 2008 (UTC)
I didn't mean to sound imperious--- sorry. I just don't have anything more to say--- I'm repeating myself.Likebox (talk) 14:29, 15 March 2008 (UTC)
Re Arcfrk: The status of AC in mathematics and its status in logic are both interesting philosophical questions and (in my opinion) have somewhat different answers. In mathematics outside of axiomatic set theory, as far as I can tell, the perception that there is any issue with AC is very small, and apparently diminished during the 20th century, possibly because contemporary mathematicians learned AC earlier in their education. Within axiomatic set theory, the study of determinacy is the main area in which axioms incompatible with choice are of interest. But this study is far removed from ordinary mathematics at the present time.
It is true that the consistency of ZF has not been proved in the same way the consistency of Peano arithmetic has been (and such a proof is very unlikely). However, the consistency of ZFC is equivalent to the consistency of ZF by work of Goedel. — Carl (CBM · talk) 02:59, 14 March 2008 (UTC)
Just a note on determinacy: I don't think that many set theorists who work in this area do so because they think the axiom of determinacy is actually true. Among those who are realists, and therefore think the question has a true or false answer, I'd bet the large majority think the axiom of choice is true. But the axiom of determinacy has some very interesting inner models, like L(R) for a fairly minimal example, and some interesting stuff happens in those models.
From this point of view, inner models satisfying AD are models that simply omit the "problematic" sets of reals like the ones that witness Banach–Tarski. Inside L(R) you can't do a paradoxical decomposition, because the sets making up the decomposition just aren't elements of L(R). L(R) is allowed to think that all sets of reals are Lebesgue measurable, not because it disagrees with V about what it means to be Lebesgue measurable, but just because it can't see the counterexamples.
There are lots of interesting questions about how closely an inner model of ZF+AD can resemble V. For example, which cardinals have to collapse? Steve Jackson has done a bunch of stuff on this. --Trovatore (talk) 04:58, 15 March 2008 (UTC)