[GAP Forum] StructureDescription

[R. Keith Dennis]

Dear Andries,

Several points:

1.  StructureDescription works quickly (see next point also) with
    groups called from the SmallGroups library.  It tends to be
    much slower with groups described otherwise (e.g., some random
    presentation).

2.  A slightly modified version of the program works substantially
    faster and with less memory; it's my understanding it will be in
    the next version of gap.  (Thanks Burkhard!)

3.  Crashing:  This turned out to be a combination of things:  First,
    some sort of memory problem in our machine.  That's fixed now.
    I can highly recommend the programs that the gap group makes
    available for testing memory.
    Second:  However, in computing large automorphism groups, gap
    continued to crash the machine.  It appears that this was caused
    by some interaction with the operating system.  We switch to 
    Scientific Linux (~ RedHat Enterprise 5) last week & the problem
    seems to have vanished (at least I'm using up to 50+G and the
    machine hasn't crashed yet).

4.  Which group(s) of order 1024?  I'd be intersted in seeing if the
    new version of StructureDescription can handle them.

So for the StructureDescription problem, for my uses,I'm quite happy
at the moment.  The help I received was fantastic.  Thanks everyone!

I still have a problem with computing automorphism groups for groups
of order 256, but I'll send a note about that at lome point later.

Keith


> The GAP manual advertises:
> "The method for StructureDescription exhibits the structure of
>  the given group to some extend using the strategy outlined below.
>  The idea is to return a possibly short string which gives some insight
>  in the structure of the considered group and can be computed
>  reasonably quickly."
>
> (manual typo: s/extend/extent/)
>
> It seems to me that the idea fails. Maybe it is meant to be used only
> on baby groups, say with fewer than 100 elements, but used on
> rather small groups it usually crashes after many hours.
> Reactions are: recursion depth overflow (at recursion depth 5000), or
> memory overflow (needs more than the 3GB allowed on the command line), or
> missing info in Holt/Plesken library.
> This is on a 4GB machine, and Google shows me the complaint of
> R. Keith Dennis who tried groups of order 256 on a 128GB machine,
> and found that GAP crashed, so I need not be surprised that GAP
> fails for me on groups of order 1024.
>
> Other functions reveal the structure of the groups involved rather quickly.
> I wonder what StructureDescription tries to do that is so extremely expensive.
>
> Maybe the present behaviour should be regarded as buggy.
> It does not live up to what the manual promises.
>