[Search for users]
[Overall Top Noters]
[List of all Conferences]
[Download this site]
Title: | Mathematics at DEC |
|
Moderator: | RUSURE::EDP |
|
Created: | Mon Feb 03 1986 |
Last Modified: | Fri Jun 06 1997 |
Last Successful Update: | Fri Jun 06 1997 |
Number of topics: | 2083 |
Total number of notes: | 14613 |
1478.0. "Mail on SO(2N) from a former MATH noter" by ALLVAX::JROTH (I know he moves along the piers) Fri Aug 09 1991 23:38
Last year I posted a query in the sci.math newsgroup and had all
but forgotten about it. Perhaps any answers were buried under
the flood of postings about 3_door_Monty, or whether 0.9999... is
"really" equal to 1 and what have you. If you've looked there,
you know what I mean - the signal/noise ratio is quite poor.
Thus I was amazed to receive the following mail from a former
DEC math noter and thought I'd share it...
(Now I'll have to go review the Kunneth product formula though!)
From: WASTED::"elrond!gonzo.calcomp.com!fherman" 9-AUG-1991 02:39:26.62
To: <<@elrond:decvax!allvax.enet.dec.com!jroth>>
CC:
Subj: On a conjecture by Jim Roth
Jim,
Back in Oct of 1990, you proposed the following
conjecture on the topology of the even real special orthogonal
groups, namely:
Conjecture:
The special orthogonal group, SO(2n), of Euclidean 2n
space, R^(2n), is homeomorphic to product of n copies of the
(2n-1)-sphere, S^(2n-1).
It turns out to be false, but its still a clever conjecture.
First of all, the topological dimension of both sides, n*(2n-1),
is the same. It is true for n=1, i.e., SO(2) is the just rotations
about the origin in the plane which is just the circle group, S^1.
For n=2, its real close. Namely, SO(4) is homeomorphic (even
group isomorphic) to
(S^3 x S^3)/S^0
where S^0, the zero dimensional circle group, is the group
{1, -1}, the circle of radius of 1 about the origin on the
real line, i.e., {x real: x^2 - 1 = 0}. By the quotient of
S^3 x S^3 by S^0, I mean, mod out by the S^0 antipodal
action on each S^3. As for the group structure on S^3,
identify it with the unit quaternions. The proof of the isomorphism
between (S^3 x S^3)/S^0 and SO(4) requires an some analysis of the
quaternion algebra which I'll skip.
For, n > 1, SO(2n) is not simply connected; its
fundamental group is cyclic of order 2. Therefore the conjecture
needs to be modified if it stands any chance of being true.
Namely,
Modified Conjecture:
SO(2n) = (S^(2n-1) x ... x S^(2n-1)) / S^0
n factors
This is more reasonable since now the topological dimension and
fundamental groups agree. However, again it is flawed. To demonstrate
this, I have to use some results on the cohomology of the classical simple
Lie groups. Specifically, the cohomology of these groups are "thin" in
the sense that (except for some low dimensional exceptions, e.g.,
SO(4), Spin(4) ) the real vector space
dimension of the i-th cohomology group with real coefficents, H^i(G),
of the group, G, is no greater than one. Since the higher cohomology
groups H^i(G), i>1, are the same for all covering groups, the modified
conjecture would imply in particular:
H^(2n-1)(SO(2n)) = H^(2n-1)(S^(2n-1) x ... x S^(2n-1))
Using the Kunneth Product formula on the left hand side yields
a vector of dimension n which produces our promised contradiction.
-Franklin
--
Franklin B. Herman, Ph.D.
Independent Consultant at CalComp, Inc.
USENET: decvax!elrond!fherman@gonzo
T.R | Title | User | Personal Name | Date | Lines |
---|
1478.1 | trying to second-guess reality | HERON::BUCHANAN | object occidented | Mon Aug 12 1991 09:45 | 33 |
| > The special orthogonal group, SO(2n), of Euclidean 2n
> space, R^(2n), is homeomorphic to product of n copies of the
> (2n-1)-sphere, S^(2n-1).
Irrespective of the connectivity issues which Franklin correctly
raises, the conjuecture didn't seem plausible to me, although I haven't got the
Topological Group Theory to prove it. Intuitively, SO(2n) seems much
"richer" than n copies of S^(2n-1).
Here's another conjecture:
SO(n) / S^(n-1) is homeomorphic to SO(n-1)
I've a feeling that this is false, that the special orthogonal groups
have fundamental groups which are alternately trivial and C2, as n is
alternately odd and even. So maybe you need to modify this to say that we're
just considering the components which contain the identity in trying to
establish this result. [This is essentially an analogous modification to the
one that Frnaklin suggested to Jim's original allegation.]
Alternatively, we could conjecture:
{SO(n) / S^(n-1)} / S^(n-2) is homeomorphic to SO(n-2)
as something which could hold for all n. Here the topological dimensions and
the fundamental groups will match.
I don't know *anything* about this area, I'm just trying to guess
intelligently :-).
Andrew.
|
1478.2 | | ALLVAX::JROTH | I know he moves along the piers | Mon Aug 12 1991 21:14 | 49 |
| <<< Note 1478.1 by HERON::BUCHANAN "object occidented" >>>
>> The special orthogonal group, SO(2n), of Euclidean 2n
>> space, R^(2n), is homeomorphic to product of n copies of the
>> (2n-1)-sphere, S^(2n-1).
> Irrespective of the connectivity issues which Franklin correctly
> raises, the conjuecture didn't seem plausible to me, although I haven't got the
> Topological Group Theory to prove it. Intuitively, SO(2n) seems much
> "richer" than n copies of S^(2n-1).
You're intuition is better than mine, I guess :-(
> Here's another conjecture:
>
> SO(n) / S^(n-1) is homeomorphic to SO(n-1)
Actually, this is close but doesn't make real sense - an n-sphere isn't
a "group".
What is true is that the manifold of SO(n)/SO(n-1) can be identified with
the n-1 sphere. This leads to a very nice way of choosing a random
N-dimensional orthogonal rotation matrix uniformly with respect to Haar
measure: Select a set of random k dimensional unit vectors (using, say,
the usual trick of normalizing k-dimensional normal variates) and then
use Householder reflections to successively build up SO(k) from
SO(k-1) and your unit k-vector.
More generally, SO(n)/SO(n-k) is known as a Steiffel manifold -
the manifold of orthonormal k-frames in n-space.
Even more generally, the space of k-dimensional subspaces (relaxing
the orthogonality of the basis vectors) is known as a Grassman
manifold and is SO(n)/(SO(n-k)*SO(k)).
The best reference I know of as introduction to this is, oddly enough,
a paper by a statistician: A.T. James, "The Normal Distribution and
the Special Orthogonal Group". It appeared in 1949 in the American
Journal of Statistics.
A great book about this stuff is "Spaces of Constant Curvature" by
Joseph Wolf. J. F. Adams little book "Lecturs on Lie Groups"
is another really nice introduction...
There's a beautiful blend of analysis, geometry, topology as well
as some combinatorics in this area. I wish I had more time and
talent to understand more than I do...
- Jim
|
1478.3 | | HERON::BUCHANAN | object occidented | Tue Aug 13 1991 10:39 | 16 |
| > You're intuition is better than mine, I guess :-(
You flatter me. The intuitions were larded with misunderstandings
in .1, which I would replace if you hadn't already replied to it.
> Actually, this is close but doesn't make real sense - an n-sphere isn't
> a "group".
Yup. Franklin gave group action to S^0, S^1, S^3 & I unthinkingly
extended it to all S^n. Thanks for re-expressing the statement in a correct
way.
Thanks,
Andrew.
|