[Columbus] cidrtfl and parallel CI

[Thomas Mueller]

 Dear Andy,

the parallel as well as the serial CI 
do not care about which code has generated
the DRT file. 

The output files comprise a converged 
calculation (with serial code). 

Geometry optimization is supported with
parallel CI and this will also apply to
nad couplings. 

Make things easier:

 - take some calculation of yours
   and run just a single point calculation
   with it:

   control.run

   niter=1
   hermit
   scf
   mcscf
   ciudg


 - now try to reproduce this stuff with
   the parallel code (just on one node)
   with appropriate mpi setup
   (including GA and install.config)

   niter=1 
   hermit
   scf 
   mcscf
   pciudg
   hosts=my_comp
   nproc=2

   add the appropriate lines to ciudgin
   (nseg0x,nseg1x,nseg2x,nseg3x,nseg4x,nsegd,fileloc)

   runit:

   runc -chaini -m xxxxxx  

   runs the serial portion up to cisrt.x

   runc -chainp -m xxxxxx

   runs the parallel portion

  you find an input example attached.

 Also note, if the Columbus version 
 is much older than summer 2004 the parallel
 setup is queer, since the installation of
  the GA was incorporated into the installation
 script which was a bad decision. 

 
  - extend it to a geometry optimization by
    using standard control.run file for geometriy
    optimization, replace ciden by pciden and ciudg
    by pciudg, add hosts and nproc and runit

    either step by step

      runc -chaini
      runc -chainp
      runc -chains
      runc -chainp
      runc -chains   ...

      or 

      runc 

   - if this works, fine. Now try geom opt. for
     an excited state. 

   - if this works, fine. Now try nadcl.


Regards,
  
  Thomas 




 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: andy.tgz
Type: application/octet-stream
Size: 38038 bytes
Desc: andy.tgz
URL: <http://lists.osu.edu/pipermail/columbus/attachments/20050901/440a19c6/attachment.obj>