sci.math
[Top] [All Lists]

Re: Complete metric quotient spaces

Subject: Re: Complete metric quotient spaces
From: rusty
Date: Sun, 17 Sep 2006 13:32:24 +0200
Newsgroups: sci.math
On Sat, 16 Sep 2006, William Elliot wrote:

> From: rusty <mr.rusty@xxxxxxx>
> Newsgroups: sci.math
> Subject: Re: Complete metric quotient spaces
> 
> rusty wrote:
> 
> > William Elliot wrote:
> 
> >>> - For a compact metric space, the quotient will be a (compact) metric
> >>> space iff all equivalence classes are closed. (In general for the
> >>> quotient to be metrisable each equivalence class must be closed as the
> >>> inverse image of a closed set under a continuous mapping!)
> 
> > Here is a simple direct proof using an explicit quotient metric in
> > an important special case.
> 
> > Let (X,d) be a *compact* metric space and R in X x X  be a *closed*
> > subset defining an equivalence relation ~.
> 
> Thus R/~ is Hausdorff space.
> 

Amongst other things. 

> > In addition assume that the equivalence relation satisfies the
> > following *regularity* condition: if x ~ y and x_n -> x then
> > there exist y_n ~ x_n such that y_n -> y.
> 
> Is that the condition needed to make well defined the definition
>       [x_n]_n -> [x] when (x_n)_n -> x ?

It is needed to give a poor man's version of the slice structure of 
orbits.

> 
> No, that would be well-defined  for all j, xj ~ yj, (xj)_j -> x, (yj)_j 
>-> y ==> x ~ y

This statement is a consequence of the fact that R is closed in X x X, if 
you thought for a minute. No the regularity condition I wrote is a poor 
man's version of the slice structure of orbits.
 
> > For equivalence classes C_1 and C_2, define D(C_1,C_2) to be the
> > least non-negative number d such that d(x_1,C_2) <= d for all x_1 in
> > C_1 and d(x_2,C_1) <= d for all x_2 in C_2.
> 
> How is this any different than
>       D(C1, C2) = inf{ d(x,y) | x in C1, y in C2 } ?
> It is not as continuous maps on compact sets attain inf.

Well what I define is completely different. I am not sure whether you are 
deliberately trying to be obtuse, but I ask that EVERY point of C_1 be 
within r of some point of C_2 and vice versa. In your definition 
D(C_1,C_2) <= r iff at least one point of C_1 is within r of some point of 
C_2.  

Also what I write down defines a metric whereas what 
you write down does not. In addition it does not have the uniformity 
described below. (I suggest you draw a picture.)

Since for some reason you don't seem to be able to find a proof, here is 
a proof of the triangle inequality. Let x_1\in C_1. By compactness there 
is a point x_2 in C_2 such that

d(x_1,x_2)=d(x_1,C_2) \le D(C_1,C_2). 

Similarly there is a point x_3\in C_3 such that 

d(x_2,x_3) \le D(C_2,C_3). 

But then

d(x_1,x_3) \le d(x_1,x_2) + d(x_2,x_3) \le D(C_1,C_2) + D(C_2,C_3).

In particular 

d(x_1,C_3) \le D(C_1,C_2) + D(C_2,C_3).

Hence 

max_{x_1\in C_1) d(x_1,C_3) \le D(C_1,C_2) + D(C_2,C_3).

Similarly 

max_{y_3\in C_3} d(y_3,C_1) \le D(C_1,C_2) + D(C_2,C_3).

Since D(C_1,C_3) is by definition the maximum of the two quantities on the 
LHS, it follows that

D(C_1,C_3) \le D(C_1,C_2) + D(C_2,C_3)

as required.
 

> 
> > So D(C_1,C_2) <= r iff each point of one class is within a distance r
> > of some point of the other class. It is easy to see that D defines a
> > metric on X/~.
> 
> It is not.  Let S = [0,1] and P = { 0,1 }
> Consider the quotient space, S/P identifying P to a point.
> That is counter example except for lack of regularity.

This is not a counterexample to what I have written.

In this case  the quotient space can naturally be identified as a set with 
both (0,1] and  [0,1). This gives two metrics on the quotient and the metric I 
have defined is in this case the maximum of these two. But the maxixum of 
two metrics is again a metric. 
 
> > To check that the map (X,d) -> (X/~,D) is continuous,
> > suppose x_n -> x. Let C_n = [x_n], C = [x]. If D(C_n,C) -/->0 then,
passing to a
> > subsequence if necessary :
> 
> > /either/ (1) there are points y_n in C_n with inf d(y_n,C) > 0
> > /or/ (2) there are points z_n in C such that inf d(z_n,C_n) > 0
> 
> > (1) Passing to a subsequence if necessary we may assume that y_n ->
> > y. Since R is closed and y_n ~ x_n,  y must lie in C, contradicting
> > d(y,C)= lim d(y_n,C) >0.
> 
> I don't see how regularity applies.  What applies is well-defined.

Regularity is not used here. Who said it had to be?

> 
> > (2) Passing to a subsequence if necessary we may assume that z_n ->
> > z. Since x ~ z and x_n -> x, there exist w_n ~ x_n such that w_n -> z
> > by regularity. Thus d(z_n,w_n) -> 0, contradicting
> > inf d(z_n,C_n) > 0.
> 
> Ok, regularity applies.

Yes, this is where it is needed (it is not required to check that D is a 
metric).

> 
> > Hence D(C_n,C) -> 0 and the map is continuous.  But then there is a
> > continuous 1-1 map of the topological quotient X/~ onto (X/~,D).
> > Since the topological quotient is compact Hausdorff and (X/~,D) is
> > Hausdorff, this map is necessarily a homeomorphism, as required.

In your case it would seem that the possibility of cancelling or 
superseding a  posted message might bring its advantages.

P.S. This post was written remotely with ssh and Pine from a mathematics 
institute in the Canadian Rockies. (I hesitate to post directly from 
the computer provided in my room here.)
-- 
rusty

<Prev in Thread] Current Thread [Next in Thread>
Privacy Policy