From: Dave Rusin
Subject: Re: group compactification
Date: Sun, 21 Feb 1999 17:20:38 -0600 (CST)
Newsgroups: [missing]
To: adamp@math.uchicago.edu
[Message from adamp, edited to fit messages which follow]
>Consider the map G->bG from a discrete (nonabelian) group
>to its group compactification.
>Is it always a monomorphism?
>
>By a group compactification I mean the initial
>homomorphism G->bG from a discrete group G to a compact group bG such that
>the image of G is dense in bG. I think it is usually called the Bohr
>compactification. The compactification map is not required to be a
>homeomorphism onto its image. Existence and uniqueness, up to isomorphism,
>are obvious. I would like to know if it is a monomorphism. If G is not
>discrete I think there are examples when this compactification has a
>kernel. I don't know if this is the case when G is nonabelian discrete. In
>the abelian case G can be seen as a module over integers Z. Every cyclic
>Z-module embeds into a circle which is compact and injective as a Z-module
>hence the Bohr compactification can not have a kernel in this case.
>
>Thanks,
>Adam Przezdziecki
==============================================================================
From: Dave Rusin
Subject: Re: group compactification
Date: Sat, 20 Feb 1999 23:45:07 -0600 (CST)
Newsgroups: [missing]
To: adamp@math.uchicago.edu
Forgive me for being dense, but what is bG? The compactifications of
general topological spaces I have seen -- one-point cpt., Stone-Cech cpt. ---
come equipped with 1-to-1 maps from X to X*; while it's not clear to
me which compactifications can be given a group structure, and while
the map from X to X* need not be a homeomorphism onto its image, in general,
it nonetheless does seem difficult to envision a good notion of
"compactification" which would have a kernel!
dave
==============================================================================
From: Adam Przezdziecki
Subject: Re: group compactification
Date: Sun, 21 Feb 1999 16:13:54 -0600 (CST)
Newsgroups: [missing]
To: rusin@math.niu.edu (Dave Rusin)
Sorry for being unclear. By a group compactification I meant the initial
homomorphism G->bG from a discrete group G to a compact group bG such that
the image of G is dense in bG. I think it is usually called the Bohr
compactification. The compactification map is not required to be a
homeomorphism onto its image. Existence an uniqueness, up to isomorphism,
are obvious. I would like to know if it is a monomorphism. If G is not
discrete I think there are examples when this compactification has a
kernel. I don't know if this is the case when G is nonabelian discrete. In
the abelian case G can be seen as a module over integers Z. Every cyclic
Z-module embeds into a circle which is compact and injective as a Z-module
hence the Bohr compactification can not have a kernel in this case.
If bG is the one-point compactification of G then G embeds in bG but the
action of G on bG fixes the point at "infinity" hence the group structure
does not extend to bG.
If bG is the Stone-Cech compactification then the isolated points of bG
are precisely those that came from G hence bG is not homogeneous
(homeomorphisms don't act transitively on bG) so bG again does not admit
any group structure compatible with the topology.
Adam
[previous letter quoted -- djr]
==============================================================================
From: Torsten Ekedahl
Subject: Re: group compactification
Date: 22 Feb 1999 12:30:03 -0600
Newsgroups: sci.math.research
Adam Przezdziecki writes:
> Consider the map G->bG from a discrete (nonabelian) group
> to its group compactification.
> Is it always a monomorphism?
A compact group is "residually linear" in the sense that any
non-identity element can be mapped to a non-identity element under a
homomorphism to some GL_n(C). (This follows from the Peter-Weyl
theorem.) Hence any group that can be embedded into a compact group is
residually linear. There are lots of non-residually linear groups. For
instance any finitely generated residually linear group is residually
finite and there are lots of f.g. non-residually finite groups.
==============================================================================
From: Joerg Winkelmann
Subject: Re: group compactification
Date: 22 Feb 1999 12:30:10 -0600
Newsgroups: sci.math.research
Adam Przezdziecki wrote:
> Consider the map G->bG from a discrete (nonabelian) group
> to its group compactification.
> Is it always a monomorphism?
> By a group compactification I mean the initial
> homomorphism G->bG from a discrete group G to a compact group bG such that
> the image of G is dense in bG. I think it is usually called the Bohr
> compactification. The compactification map is not required to be a
> homeomorphism onto its image. Existence and uniqueness, up to isomorphism,
> are obvious. I would like to know if it is a monomorphism. If G is not
> discrete I think there are examples when this compactification has a
> kernel. I don't know if this is the case when G is nonabelian discrete. In
> the abelian case G can be seen as a module over integers Z. Every cyclic
> Z-module embeds into a circle which is compact and injective as a Z-module
> hence the Bohr compactification can not have a kernel in this case.
Every compact group has a faithful linear representation on a Hilbert
space, namely the space of L^2 functions on this group and
the union of invariant finite-dimensional subspaces in this Hilbert
space is dense. Using this, it follows that given a compact group K
and an element g in K there is a finite dimensional representation f
of K such that f(g) is not trivial. (f depends in general on g, unless
K is assumed to be a Lie group.)
Now let G be a finitely generated discrete group.
If G can be embedded into a compact group K, then for every g in G
there exists a finite dimensional representation f with f(g) non-trivial.
However, there are countable groups such that the intersection
of the kernels of all finite dimensional representations
is non-trivial.
Thus for these groups the Bohr compactification can not be
injective.
AFAIK examples of such finitely generated groups can be found
using lattices in universal coverings of real semisimple algebraic
groups.
Maybe this is contained in:
Raghunathan, M. S.(F-IHES)
Torsion in cocompact lattices in coverings of ${\rm Spin}(2,\,n)$.
Math. Ann. 266 (1984), no. 4, 403--419.
HTH
Joerg Winkelmann
--
jwinkel@member.ams.org
Mathematisches Institut der Universitaet Basel
http://www.cplx.ruhr-uni-bochum.de/~jw/
==============================================================================
From: flor
Subject: Re: group compactification
Date: Tue, 23 Feb 1999 20:27:57
Newsgroups: sci.math.research
The original question refers to a situation which was treated long ago. A
group admitting an injective homomorphism into some compact group is
called "maximally almost periodic". The original question thus amounts to
the following problem: Is every discrete group maximally almost periodic?
Questions of this kind were studied by J. von Neumann and E. Wigner in
their 1940 paper "Minimally almost periodic groups". The simplest example
of a group which is not maximally almost periodic may well be the
following: let G be the group of affine selfmaps of the real line, under
composition. (This is a two-dimensional, noncommutative Lie group.) It is
easily seen that in this group, all nontrivial translations form just one
conjugacy class. Any homomorphism of G into a compact group thus maps all
translations into a single conjugacy class; call it C. Every conjugacy
class in a compact group is closed; so C is closed. Now let c be the image
of any nontrivial translation. Not only c is an element of C but every
power c^n, n/=0, also is in C. In every compact group, the closure of the
set of all positive powers of any element contains the unit element. Hence,
our conjugacy class C contains the unit element which, of course, is
conjugate only to itself. So C ={e}; that ist, all translations are mapped
into e, under every homomorphism of G to a compact group. This proof did
not assume that the homomorphism should be continuous; hence it works for G
with the discrete topology.