From Kurt.Hillig@um.cc.umich.edu  Tue Apr 27 07:07:18 1993
Date: Tue, 27 Apr 93 11:07:18 EDT
From: Kurt.Hillig@um.cc.umich.edu
To: chemistry@ccl.net
Message-Id: <22506386@um.cc.umich.edu>
Subject: Gaussina 9x memory/disk requirements


I've often seen the claim that memory (for in-core) or disk space 
needed by Gaussian 9x scales as N^4/4 (SCF).  I expect a first-order 
approximation to the memory needs can be given by M = A + B N^4, 
where A is the "overhead" and B is a scale factor (words per integral?). 
 
I've occasionally seen some very rough estimates - for example today, 
the claim (paraphrased) 100 basis functions:100 MWords, 200 BF's:200 MW,  
300 BF's:>1GW - but it's hard to fit this to a fourth-order polynomial 
with any accuracy (;-) and it doesn't match M = N^4/4 (i.e. A=0, B=.25)  
particularly well. 
 
Can anyone supply more accurate numbers so I can better advise my users?  
 
    Dr. Kurt Hillig 
   Dept. of Chemistry      I always tell the     khillig@umich.edu 
 University of Michigan     absolute truth    Telephone (313)747-2867 
Ann Arbor, MI  48109-1055    as I see it.    hillig@chem.lsa.umich.edu