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

[Alessandro Curioni]

Some clarifications:

-the first time you  start an MD after a wfn optimization  you have
positions assigned but not  velocities -  the velocity are assigned
randomly using a Maxwell distribution for the ions  for a temperature T -
the seed for this depends  on timec() so you  will get   different velocity
distributions for  different  runs - Of course this is not true if you
start your dynamics  with T=0 -  all the initial velocities will be zero -
and  you will have  identical dynamics (divergences only on  very long
runs)-  small changes of the temperature (1K)   (this dipends however on
the temperature control chosen)  could remain  hidden for many steps since
the coupled electrons-ions system  has a  finite relaxation time.
Moreover - if you want to have independent trajectories to calculate
ensemble averages - you can  create them using the randomize coordinates
keyword (see manual) followed by a quench bo (see manual) - without having
to randomize the initial coordinates by hand.

Ciao.


Alessandro CURIONI, PhD
Research Staff Member
Computational Biochemistry and Material Science group
IBM Research Division - Zurich Research Laboratory
Saumerstrasse 4
8003 Rueschlikon - Switzerland
e-mail: cur at zurich.ibm.com
www:    www.zurich.ibm.com
Tel: +41-1-7248633
Fax: +41-1-7248958



                                                                           
             Axel Kohlmeyer                                                
             <axel.kohlmeyer at t                                             
             heochem.ruhr-uni-                                          To 
             bochum.de>                Archilles <archilles02 at gmail.com>,  
             Sent by:                  <cpmd-list at cpmd.org>                
             cpmd-list-bounces                                          cc 
             @cpmd.org                                                     
                                                                   Subject 
                                       Re: [CPMD-list] How to rerun the    
             12/08/2004 06:36          same job for statistical purpose?   
             PM                                                            
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           





On Wed, 8 Dec 2004, Archilles wrote:

dear peng,

AP> Dear all,
AP>
AP>
AP> I tried to rerun the same CPMD simulation job several times to obtain
AP> statistical sense of multiple simulations. By "the same CPMD
AP> simulation job ", I mean the identical input file. But what I got is
AP> the identical output files.

well, if you start every time from the same coordinates and
electron structure (i.e. the same RESTART file), you should get
the same trajectory for quite a while until some numerical
noise kicks in and the trajectories will gradually diverge.

if you use the restart written at the end of each simulation
to start the next run, there should be differences.

AP> After changing the temperature by 1 K, each step KS energy remains the
AP> same with two different runs.

that depends on what kind of thermostatting you use.
with TEMPCONTROL IONS for example, the change by 1K will
only affect the simulation, if the instantaneous temperature
with be outside of the given bounds.

it may be helpful, if you'd post your input (or at least the
part up to the coordinates section, if you have a large system).

AP> The trajectory also remains the same for two runs.
AP>
AP> What I expected is that each run are simillar, but quite different
AP> from each other. Maybe each run the program use the same random seed
AP> to start the simulation, which leads to the same results. Is there any

why a random seed? you are doing molecular dynamics, in which the
new positions are determined from the old positions of the atoms
and the forces on them.

AP> way to change it each time? Therefore, I can run the same job more
AP> than once, and get different results to inspect the statistical
AP> properties of the system.

due to ergodicity hypothesis, you only have to run the MD long
enough to sample phase space, povided you are in equilibrium,
and the average over them. of course you could also use different
starting coordinates and thus have independent trajectories, but
those you'd have to create yourself.

regards,
             axel kohlmeyer.

AP>
AP> Thank you very much.
AP>
AP> Peng
AP> _______________________________________________
AP> CPMD-list mailing list
AP> CPMD-list at cpmd.org
AP> http://cpmd.org/mailman/listinfo/cpmd-list
AP>
AP>

--

=======================================================================
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.

_______________________________________________
CPMD-list mailing list
CPMD-list at cpmd.org
http://cpmd.org/mailman/listinfo/cpmd-list