Re: [ESPResSo-users] Re: problems on runing parallel Espresso

[Axel Arnold]

On Thursday 16 December 2010 04:45:57 Yanping Fan, Liza (Dr) wrote:
> Hi, Alex. Thanks for reply.
>
> I switch skin to 3 or 2, adjust my LJ settings and change P3M accuracy from
> -06 to -05, it showed other problems

Do you mean you used p3m tune accuracy 1e-5 ?? That can't work with your 
length scales. That is absolute force accuracy in length/time^2, and since 
your particles seem to have a diameter of about 10 judging from your box size 
and bond lengths, the accuracy cannot be better than 10^2. All mesh-Ewald 
method are very bad in reaching high accuracies, but these are typically also 
unnecessary.

>
> Either stopped outputting new data or showing this error msg
>
> 0: Script directory: /home/korolev/espresso-2.1.2j_p/scripts
> [compute82:27839] *** Process received signal ***
> [compute82:27839] Signal: Segmentation fault (11)
> [compute82:27839] Signal code: Address not mapped (1)
> [compute82:27839] Failing at address: 0x2aaaef555a78
> [compute82:27839] [ 0] /lib64/libpthread.so.0 [0x39f640e4c0]
> [compute82:27839] [ 1]
> /home/korolev/espresso-2.1.2j_p/obj-Xeon_64-pc-linux/Espresso_bin(P3M_charg
>e_assign+0x247c) [0x47d6ec] [compute82:27839] [ 2]

As long as Espresso crashes in charge assign, it normally means that you 
didn't get the memory for the charge assignment, so the precision is too high.

> Also I don't understand how come different version have different responses
> to the same script. Thanks.

It depends on how much memory is used elsewhere in the code, the compiler 
optimizes differently, all sorts of these things.

Axel

-- 
JP Dr. Axel Arnold Tel: +49 711 685 67609
ICP, Universität Stuttgart      Email: address@hidden
Pfaffenwaldring 27
70569 Stuttgart, Germany