[CPMD-list] How to rerun the same job for statistical purpose?

[Axel Kohlmeyer]

On Thu, 9 Dec 2004, Archilles wrote:

dear peng,

a few remarks to your post.

- there are more ways to achieve what you want:
  a) you could run a normal MD and use constraints
     to keep your reaction from happening and then
     start an MD from several selected, decorrelated
     coordinate snapshots along the trajectory.
  b) you could use metadynamics to determine the
     free energy profiles for both reaction pathways.

- 200 hours (is this serial or parallel?) is not exactly
  excessive for cpmd calculations of larger systems. i know
  of simulations that have been running over a large number
  (20-200) processors for weeks and months.

- there are some problems with your input:
  a) your supercell is much too small. you are using SYMMETRY 0
    yet your electron density is not zero at the box border
    (in fact it 'wraps around', as can be easily seen from 
    creating a DENSITY.cube file, and then visualize it ). 
    please see the manual for the requirements of the HOCKNEY
    poisson solver.
  b) i assume you use pseudopotentials from the goedecker-hartwigsen
    library. a plane wave cutoff of 70.0 is (far) too small for those.
    AFAICT, you need at least 120ry, probably 150ry (thanks to the
    oxygen being rather 'hard'). 
  c) a high GC-CUTOFF of 5.0e-6 is only needed for low CUTOFF calculations
    (e.g. for vanderbilt ultra-soft pseudopotentials  with 25ry cutoff). 
    for high cutoff calculations a much lower value can/should be used.
  

AP> Thank you all for your help.
AP> 
AP> After reading all the replies, I think I have the following options to
AP> do what I want:
AP> 1) Run a certain LONG time, I will get good enough statistical
AP> simulation results
AP> 
AP> 2) Change the default random number generator DURAND to a different
AP> one, then recompile the CPMD code.
AP> 
AP> 3) Use RANDOMIZE COORDINATES combined with QUENCH BO keywords to
AP> randomize the optimized initial position
AP> 
AP> Since I am interested in selectivity of epoxide ring openning (two
AP> possible positions), once the ring openned, I cannot afford to run it
AP> until the reaction reverse and happen again: it take about 80 hours to
AP> see the ring openning, and even after 200 hours running the system
AP> remained as product. Here I would like to try RANDOMIZE COORDINATES  
AP> with QUENCH BO  method.
AP> 
AP> Here is the input file I am using. With "# New Line", it means I just
AP> added these lines.
AP> 
AP> Thanks again for all of you.
AP> 
AP> Best regards,
AP> Peng

[...]

-- 
=======================================================================
Dr. Axel Kohlmeyer   e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.