Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!uwm.edu!cs.utexas.edu!ginosko!gem.mps.ohio-state.edu!rpi!batcomputer!kahn
From: kahn@batcomputer.tn.cornell.edu (Shahin Kahn)
Newsgroups: comp.arch
Subject: Re: 1000000x1000000 Matrix (was: li
Message-ID: <9125@batcomputer.tn.cornell.edu>
Date: 23 Oct 89 19:44:41 GMT
References: <9118@batcomputer.tn.cornell.edu> <46500084@uxe.cso.uiuc.edu>
Organization: Theory Center, Cornell U., Ithaca NY
Lines: 37

In article <46500084@uxe.cso.uiuc.edu> mcdonald@uxe.cso.uiuc.edu writes:
>>1) Is your 1000000x1000000 matrix dense?  If not, how sparse is it?
>It is dense. Totally dense. No element is (in principle) zero.
 
Wow!

>>2) How did you solve it?
>Are you kidding? All I get is uproarious laughter. But, now

There are out-of-core packages that seem to do a good job.  If you can
get the disk space!

>but he does say that the methods are well known ones for sparse
>matrices.

He probably doesn't have the storage problem.  If he does statistics,
I have only seen very sparse ones in that field,, so far.
(A side comment,,  an application also with huge sparse matrices
is 'animal science'!  An obvious one, right?  I say that only to
point out that real supercomputing applications come up all the time
as people start tackling more realistic problems, etc...)

>>4) What is the application?  (Its not a complex matrix, is it?!)
>Mine is quantum mechanics of vibrations. No, it is real symmetric.
 
How come the tricks used in ab-Initio Quantum Chem cant be used?

>on the Illiac IV, which suited it perfectly. Only problem
>was, the results only proved that classical mechanics won't
>give correct results.

I find this a fascinating subject.
Here's a real supercomputing application with use and need for
everything, it seems.
 
*What* is an ideal architecture for dealing with a 1,000,000x1,000,000
dense matrix??