Talk:Whitney embedding theorem

[First topic]
I assume that there are at least two versions, with different conditions, yielding 2n and 2n+1 as the dimension. Can anyone authoratively clarify this? --Pjacobi 01:46, August 27, 2005 (UTC)
 * 2n is stronger, but often it is proved only for 2n+1 sinse it is much easier. Tosha 01:04, 23 September 2005 (UTC)

Are the hyperbolic plane and hyperbolic n-space smooth and second countable? In other words, are such manifolds (the hyperbolic plane being a surface, or 2-manifold) covered by this theorem? Kevin Lamoreau 04:20, 7 November 2005 (UTC)
 * Technically, yes. But the smooth structure of hyperbolic n-space is identical to that of Euclidean n-space, so the Whitney theorem is unnecessary; you can already smoothly embed Hn in Rn, let alone R2n. This embedding is going to screw up the curvature, but curvature isn't a property of smooth manifolds, so we're not worried about that here. Melchoir 06:14, 24 February 2006 (UTC)
 * Thanks for your reply, Melchoir. What I am actually interested in is knowing what dimensions of Euclidean space hyberbolic n-spaces of various n (particularly hyperbolic 1-, 2- and 3-space) are "naturally" contained in.  In other words, if I were to take the hyperboloid model of the hyperbolic plane or the equivilent model for hyperbolic 1- or 3-space and converted the graph from the Minkowski space R1,n (n being the dimension of the hyperbolic space in question) to the infinate-dimensional Euclidean space R∞, first, could I make such a conversion without altering the graph beyond the fact that I'm graphing it in a different space, second, would the resulting graph lie entirely within a "hyperplane" (although with an infinite codimension) of Euclidean m-space of finite m and third, could anyone tell me what m would be for n (the dimension of the hyperbolic space) = 1, 2 and 3.  I thought at first that saying a manifold could be embedded in a cetain space was equivalent to saying that it could be graphed in that space and be as much "itself," if you know what I mean, as a sphere is in Euclidean 3-space or as any n-sphere is in Euclidean (n + 1)-space.  Perhaps that is the case for an isometric embedding (not merely a smooth embedding), but perhaps not.  If that's not the case is there any term for (or always equivilent to) a manifold being able to be so contained in a certain space.  In any case that's what I'm really interested in.  I know it may not matter much in terms of any mathematical application, but I just want to know what dimension of Euclidean space hyperbolic 1-, 2- and 3-space can be "naturally" contained in. Kevin Lamoreau 05:16, 1 May 2006 (UTC) - edited by myself, Kevin Lamoreau 20:17, 1 May 2006 (UTC)


 * It does sound like you're asking about isometric embeddings. Well, 1-dimensional flat and hyperbolic space are geometrically and in every way the same. But at 2 dimensions and greater, my intuition says that it's impossible to isometrically embed H^n in Euclidean space of any dimension; it would be hard to keep the image from circling around and hitting itself. I'm sure there are theorems to this effect; you may get a more enlightened response at Reference desk/Mathematics. Melchoir 20:26, 4 June 2006 (UTC)

The 2-dimensional hyperbolic plane admits an isometric embedding into 6-dimensional Euclidean space. See: Blanusa (Monatshefte Math. 59 (1955) 217-229)  More generally, any compact Riemann manifold admits *some* isometric embedding into some (very high dimensional) Euclidean space. This is called the Nash Embedding Theorem and it is a "big deal". Anyhow, for the embedding of the 2-dimensional hyperbolic plane in R^6, here is an explicit formula from a sci.math.research post of Dave Rusin's:

-- Now I will describe the embedding of R^2  into  R^6. This is not exactly what one would call an "obvious" construction, but there are certain patterns to it which suggest how Blanusa might have been led to it. We will send the point (u,v) to a point with six coordinates (x1, ..., x6) which are functions of u  and  v  of the special forms

x1 = x1(u)

x2 = f1(u) sin( v psi1(u) )

x3 = f1(u) cos( v psi1(u) )

x4 = f2(u) sin( v psi2(u) )

x5 = f2(u) cos( v psi2(u) )

x6 = v

I will describe the functions x1, f1, f2, psi1, psi2 (of one variable each) in stages.

Let [x] denote the integer part of x.

The functions psi1  and  psi2  are periodic functions of |u| (period = 2) which on [0,2]  are exponentials of linear maps: psi1(u) = exp( 2*[ (|u|+1)/2 ] + 5 ) psi2(u) = exp( 2*[ ( |u| )/2 ] + 6 ) There are discontinuities in the psi_i  at certain integers but other parts of the construction will keep the x_i  smooth.

Define two functions phi_i  via certain normalized antiderivatives: writing F(x) = sin( pi x )/exp( sin^{-2}(pi x) ) we have phi1(u) = { (1/A) integral( F(x), x=0 to x= u+1 ) }^(1/2) phi2(u) = { (1/A) integral( F(x), x=0 to x= u  ) }^(1/2) where A = integral( F(x), x=0 to x=1 ) = 0.141327... These functions phi_i  are non-negative, periodic, and satisfy phi1^2 + phi2^2 = 1 and phi1(u) = phi2(u+1). You can think of the phi_i as being very smooth versions of |sin(pi u)|  and  |cos(pi u)|.

Now set   f_i(u) = sinh(u) phi_i(u)/psi_i(u)   for  i=1,2  and define x1  to be an antiderivative of  1-(f1')^2-(f2')^2  having  x1(0)=0.

Loosely speaking the mapping x1  sends lines in  R^2  far enough away, and the the coordinates x2, x3, x4, x5 allow the points in these line to spin around in four perpendicular directions, with enough spinning to account for the fact that the images of lines are supposed to grow very long. The last coordinate x6  merely adds a motion in another perpendicular direction to separate points so that these curves don't self-intersect.

The metric which R^2  inherits from this embedding into  R^6  comes out to ds^2 = du^2 + cosh(u)^2 dv^2, from which one finds the curvature to be constant and negative, making  R^2  into the hyperbolic plane.

dave --


 * Note: I placed lines above to clearly demarcate the writing of Dave Rusin (i.e., more than just formulas).


 * Incidentally, the Nash embedding theorem states that any Riemannian n-manifold -- including the noncompact ones -- has an isometric embedding into some Euclidean space, and it gives an upper bound on the required dimension as a function of n. (This bound has been improved somewhat over the years.)  For a large number of specific cases, the least possible embedding dimension is not known.


 * Also, it's worth mentioning that the Nash-Kuiper C^1 embedding theorem implies that if one requires that the isometric embedding be only once continuously differentiable, then the hyperbolic plane embeds isometrically in 3-space.Daqu (talk) 09:05, 31 January 2009 (UTC)

Assumptions
This article, and Whitney immersion theorem, assumes a "second-countable... manifold". As explained at Topological_manifold, the word "manifold" implies second-countable. I'll remove the second-countable bit. Melchoir 06:04, 24 February 2006 (UTC)
 * It is "usually required" but not always required!, it is better to revert it Tosha 21:38, 4 March 2006 (UTC)
 * I think that by "usually required," the author of that section meant that most mathematicians have being Hausdorff and second-countable as a criteria for a topological space to be considered a topological manifold. If every manifold is (at least) a topological manifold, than most mathematicians may indeed believe that the word manifold implies second-countability.  But I'm not sure that are not any non-topological manifolds (differential manifolds are topological, but people may sometimes speak of topological manifolds as those which are not differential), though.  I just thought I'd add what I just did to try to acheive greater clarity. Kevin Lamoreau 19:05, 22 April 2006 (UTC)
 * On sentence of the Manifold, reads, "It is customary to require that the space be Hausdorff and second countable [to be deemed a topological manifold]." Also, the section Manifold begins by saying, "In topology, an n-manifold is a second countable Hausdorff space in which every point has a neighborhood homeomorphic to an open Euclidean n-ball, :$$\mathbf{B}^n = \{ (x_1, x_2, ..., x_n) | x_1^2 + x_2^2 + ... + x_n^2 < 1 \}.$$"  So it seems pretty clear that the general consensus of mathemeticians is that all manifolds are second-countable.  I hope that information helps. Kevin Lamoreau 19:21, 22 April 2006 (UTC)

That is all true but it is also ok to leave it in the formulation (just in case), we could put it in "".--Tosha 22:02, 22 April 2006 (UTC)
 * Second-countable is very commonly assumed for manifolds. The digression in the intro was distracting from the real content so I removed most of it. - Gauge 04:06, 28 June 2006 (UTC)


 * The above statements are misleading. Yes, most topologists studying manifolds are interested in studying manifolds that are assumed to be second countable and Hausdorff.  But in careful mathematical writing, these assumptions are mentioned explicitly, not tacitly presumed by the mere use of the word manifold.


 * There are, by the way, some extremely interesting surfaces that are not second countable, in particular the Pruefer surface. Unlike the "long line" or surfaces built by using the long line, the Pruefer surface does not depend on the Axiom of Choice.Daqu (talk) 09:05, 31 January 2009 (UTC)

What type of embedding does the Whitney embedding theorem apply to?
Inspired by a realisation that there is more than one type of embedding (with definitions that are not equivilant, although satisfaction of some of those definitions imply satisfaction of others, i.e. all isometric embeddings are smooth), I thought I'd ask which type of embedding, as defined in the article of that name in Wikipedia, the Whitney embedding theorem applies to. In other words, in what way does the Whitney embedding theorem state that any smooth, second-countable (if that's not redundant) m-dimensional manifold can be embedded in Euclidean $$2m$$-space? That information ought to be added to the first sentence of this article, as well as the defining sentences of all articles on embedding theorems in Wikipedia. It would only take the addition of one adverb for this article, although if the answer is that the Whitney Embedding Theorem only guarantees the existance of a smooth embedding (as opposed to an isometric embedding) of a smooth, second-countable m-dimensional manifold in Euclidean $$2m$$-space, that probably ought to be specified somewhere in the article. I'd be happy to do make these editions myself, as I consider myself pretty good at coming up with encyclopidia-appropriate language when I know what I want to be (I don't care that much when I'm asking for information in discussion pages), but if someone else wants to go ahead and do that him/herself that's fin with me. Kevin Lamoreau 20:41, 1 May 2006 (UTC)
 * Smoothly. (done) Melchoir 20:21, 4 June 2006 (UTC)
 * Thanks. Kevin Lamoreau 19:21, 5 June 2006 (UTC)


 * It's not true that "all isometric embeddings are smooth", since smooth means infinitely diffferentiable. For example, a square curve, of side equal to π/2, parametrized by arclength in the plane is an isometric embedding of the unit circle, and this isometric embedding is not even once-differentiable.Daqu (talk) 16:47, 27 February 2009 (UTC)


 * What are you responding to? It looks like you maybe put your comment in the wrong place?  There's no isometric anything in the comment above yours.  This article is about the Whitney embedding theorem, which makes no comment about metrics. Rybu (talk) 03:00, 2 March 2009 (UTC)


 * The first sentence of this section includes the clause "i.e., all isometric embeddings are smooth."Daqu (talk) 02:49, 31 March 2009 (UTC)

Technically, the strong Whitney embedding theorem (SWET) only applies to connected 2nd countable Hausdorff manifolds. If you allow disconnected manifolds, then consider the two point space (as a 0-dimensional manifold) and find an embedding into R^0.

The proof of the SWET is done in several steps. You need the weak embedding theorem (embeds in 2n+1), then you consider n-dimensional manifolds with n>2, because this gives 2n>4 and so you can approximate a map from a 2-disc to R^{2n} by an embedding (by the weak embedding theorem). But just having the disc is not enough. You can not remove a single double point! Whitney went to a lot of trouble in his paper to show how you have to artifically introduce extra "local" double points so that you can use it to cancel with the "natural" ones. You handle 0, 1 and 2-dimensional manifolds seperately.

Anyhow, this article needs a lot of cleaning up.

connected (?)
Q: Any reason for the connected hypothesis??

A: Consider the case where the manifold is 0-dimensional. A 2-point set does not embed in a 1-point set.

some recent edits
Hi Nbarth, one of your recent edits had a mistake in it. The mobius band isn't the only non-orientable manifold that embeds in R^3, for example. All compact connected non-orientable manifolds with non-empty boundary embed in R^3. I also brought back the comment that the strong Whitney embedding theorem is best-possible in dimensions that are a power of 2. Maybe it doesn't deserve to be in the front matter of the article but it should be somewhere. Your revision had it deleted and replace with some redundant info on the weak immersion theorem. Redundant as in the sense that it was already mentioned elsewhere in the article. Anyhow the spirit of your revisions seem okay to me modulo execution and one mistake. Rybu (talk) 16:02, 6 May 2009 (UTC)

Can the theorem be extended to infinite dimensional Banach manifolds?
That is, is it true that an infinite dimensional Banach manifold is embeddable in its corresponding Banach space? Are there known counterexamples? — Preceding unsigned comment added by 193.188.47.46 (talk) 07:59, 23 October 2017 (UTC)

Errors
The introduction to the article contains this passage:

"*The weak Whitney embedding theorem states that any continuous function from an $n$-dimensional manifold to an $m$-dimensional manifold may be approximated by a smooth embedding provided $m > 2n$. Whitney similarly proved that such a map could be approximated by an immersion provided $m > 2n − 1$. This last result is sometimes called the Whitney immersion theorem."

This has the inequalities exactly backwards. It is the dimension of the larger manifold (into which the embedding or immersion occurs) that must be on the higher side of the inequality (not the dimension of the manifold that is being embedded or immersed).

I hope someone knowledgeable about this subject will fix these serious errors. 2601:200:C082:2EA0:6D35:62ED:104D:48D0 (talk) 04:45, 4 March 2023 (UTC)