Search This Blog

Wednesday, January 30, 2008

Dalton problems 2: Too long input record (ERI)


My CC calculations stopped with the forrtl: severe (22): input record too long, unit 9, file /tmp/niewod/RhbCCS/betaccs_rhb631+gd/CCSD_IAJB error message.


I changed four things and the job finished properly.

  • First I changed a machine, but both were Itanium.
  • Next I lowered the print level from 3 to 2.
  • In the original jobfile I requested a direct calculation in a CC part, in the new one I put .DIRECT keyword in a main part (for all calculations).
  • Finally I requested more memory (1900mb).
I got the proper results but the calculations end with error 32. It means forrtl: severe (32): invalid logical unit number, unit -10001, file unknown I reran jobs with new commands at the first machine, and it worked. Next I upgraded print level to 3 and job finished with success. So the issue could be cause by small amount of requested memory or direct/non-direct calculation in a HF,MP2 or CC calculation.


Monday, January 21, 2008

Dalton problems 1: Direct and NonDirect HF

Short introduction

Once, I tried to create a website with notes on my Dalton's problems. I have had not time to upgraded it because Quantum Chemistry (and Dalton) has become much less important for me. Recently, I've decided that the best way to save my notes would be to add them as entries in my English blog. So they are.


I found difference between direct and non-direct Hartree-Fock results. The direct calculation didn't converge when non-direct did (look into the files below).


The reason of my problems was a very sharp convergence criteria. The screening in the direct SCF gives round-off errors, which makes it impossible to converge to 1.0D-10. (The default screening is 1.0D-14, which is usually safe, unless users ask for very sharp convergence,as I did!) Disable screening in the direct SCF, resolve my problem. I had to add: **INTEGRALS *TWOINT .IFTHRS 20 When convergence was set to 1.0D-09 the difference between direct and non-direct results disappear and both calculations converged in 28 iterations, with (nearly) the same energy:

  • direct_rhb631+gd.out: -551.031670938128
  • nodirect_rhb631+gd.out: -551.031670938119

I would like thanks Kenneth Ruud and Hans Jorgen Aa. Jensen for help.