Const Qty & EXT_RESTART

[Rolf David]

Hi again,

Problem solved. At least found.

I was not in the MD restart but in the wavefunction restart.

There is a difference between a wavefunction.restart created by a square 
number of MPI processes and created by other number.

One created by a normal amount of MPI processes can be "correctly" read by 
a square number of MPI processes (let say 128 to 144)
One created by a square amount of MPI processes can be "correcly" read by 
a  square number of MPI processes (let say 144 to 36)
One created by a square amount of MPI processes cannot be  "correctly" read 
by a square number of MPI processes  (wfn converged to a different one, 
hence a drift) (let say 64 to 128)

On our HPC, a non-square MPI processes don't lead to poor performance 
(scaling 128 to 144 is *as expected)* but lead to an incompatibility 
between restart.wfn (and also accuracy I guess)

Maybe it's how the decomposition of matrices is done. Or integrals.

Anyway, if i'll stick to a square number of MPI as "specified in the 
output" even if in term of pure performance it's the same compared to 
non-square (quid of accuracy?) ,  it is good.


On Wednesday, March 18, 2015 at 1:07:25 PM UTC+1, Rolf David wrote:
>
> Good morning everyone,
>
> I have a QM/MM simulation which runs for 0.5ps (1000steps) > Everything 
> looks 'normal' (well not so much good, but not exploding). Const Qty is 
> fluctuating up and down but not in a direction or the other.
>
> I want to restart the calculation for another 0.5ps to reach 1ps > I use 
> my last.restart, put it in EXT_RESTART section. Everything is read, 
> coordinate, kin energy, temp, pot energy, const qty. but Const Qty goes up. 
> And only up.
>
> In fact the picture attached shows it. You see at precisely 1000 something 
> goes nuts in the Const Qty. So I'm inclined to think there is something 
> wrong in my 'restart' job. 
>
> My restart input is the same as my first run except for renaming the 
> project (0-1000 to 1000-2000)
>
> &EXT_RESTART
>   RESTART_DEFAULT TRUE
>   RESTART_FILE_NAME run_0-1000.restart
> &END EXT_RESTART
>
> If you have some idea of what i'm doing wrong (for the restart), maybe 
> RESTART_DEFAULT FALSE and set up RESTART_POS, RESTART_VEL, RESTART_QMMM, 
> RESTART_THERMOSTAT ?
>
> Thanks
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150331/a6a7ac76/attachment.htm>