[molpro-user] large direct MP2 problems

[Kirk Peterson]

Hi all,

As a favor to a colleage, I've been attempting to carrying out what for 
me is a very large MP2 calculation - just over 1400 basis functions.  
The direct SCF runs with no problems (2002.6 on a Opteron, running 
2-way parallel, tcgmsg, 64-bit integer version, fully patched), but I 
have not been able to get the direct MP2 to work.  The default scheme 
(page_dtraf=1) dies with some memory problems (these are all run with 
400 mwords per process, i.e., mem,400,m) and the minimum memory 
algorithm (page_dtraf=0) runs happily for several hours and then dies 
with seemingly an out of disk space error:

  Semi-direct algorithm used
  Total memory available for dtraf:          374.80 MW
  Total memory dtraf will use:               370.41 MW

  Dtraf would need 20 passes for fully direct algorithm

  Number of operators formed in this pass:    22155
Write error in iow_direct_write; fd=82, l=65536, p=2285506560; read 
returns 446464

  ERROR WRITING        65536 WORDS AT OFFSET 2285506560. TO FILE 4  
IMPLEMENTATION=df   FILE HANDLE=    82  IERR=-77824
  ? Error
  ? I/O error
  ? The problem occurs in writew


but there is still about 40GB of space available.

My last try was to run this job fully direct, but in that case I get 
the following error fairly quickly:

  Fully direct algorithm used

  Total memory available for direct transformation:    374.80 MW
  Maximum memory dtraf will use:             373.08 MW
  Memory dtraf would need for single pass: 15509.11 MW

  Dtraf needs 20 passes for fully direct algorithm

  Number of operators formed in this pass:    14535
  # MO in 1st index to be processed:  62  43  62   3
  # MO in 1st index to be kept     :   0   0   0  40

  ERROR EXIT
  CURRENT STACK:      MAIN


Any advice?   I did try the development version as well, but I ran into 
other problems that I will not elaborate upon here.

thanks in advance,

Kirk