[CPMD-list] Restart Option for Wave Function & Geometry Optimizatio

[Axel Kohlmeyer]

On Tue, 24 Apr 2007, Shashidhar G wrote:

SG> Dear All
SG> 
SG> I'm running CPMD on IBM, AIX, power 4 ( 54 Nos. of 4-Way SMP nodes and 1 No.
SG> of 32-Way SMP node) for Si256 atoms.
SG> Here I'm attaching my input file for Wave Function optimization. I ran it
SG> with 16 processor (8 X 2), but I didn't achieved the self consistence
SG> criterion for wave function optimization and program stops due to wall clock
SG> time exceeded.
SG> 
SG> I would like to Restart the calculation from last run. Here I'm giving my
SG> 1. Input File for Wave Function Optimization
SG> 2. Input File for Restart the Wave Function Optimization.
SG> 
SG> Please Suggest me If I'm in wrong direction, and any thing is incorrect in
SG> both files 1 & 2 .
SG> 
SG> -----------------------------------------------------------
SG> 1. Input For Wave Function Optimization.
SG> ------------------------------------------------------------

[...]

on top of ari's (very valid) concerns, here is a typo
that will make your performance very poor:

SG> 
SG> &ATOMS
SG> *SI.OUT1 KLEIMAN-BYLANDER

this should read KLEINMAN-BYLANDER.
due to the missing 'N' cpmd assumes gauss-hermite
integration which is very costly for the large number
of particles that you have.


SG>    LMAX=P LOC=S
SG>    256

generally, i was wondering, whether you have tried this
on a smaller system first. since the pseudopotential looks
as if it was newly created (why?), it would be a very good
idea to first validate it with less demanding calculations.

SG> -------------------------------------------------------------------------------------------------------------------
SG> 2. Input For Restarting the optimization of Wave Functions From Previous Run
SG> -------------------------------------------------------------------------------------------------------------------

another way to have a better chance to restart
(note that with your setup, you have to rename
your RESTART.1 file to RESTART) from an interrupted
run would be to use the STORE and RESTFILES flags.
see the manual for details.

SG> &CPMD

SG>   BFGS

this has no effect for a wavefunction optimization.

what however would be helpful in your case, is to
use a different algorithm to optimize the wavefunction.
a pre-conditioned conjugate gradient with linesearch
(PCG MINIMIZE) is much more robust than the default
ODIIS (which is faster for systems where it works well).

[...]

SG> Please give your suggestion if I'm going in wrong direction

hope this helps,
    axel.

SG> 
SG> With Regards.
SG> SHASHIDHAR . G
SG> Project Intern,
SG> Centre for Development of Advanced Computing(CDAC)-Pune,
SG> Pune-411 007. INDIA.
SG> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.