simulation acceleration of QMMM in CP2K

[Matt W]

Hi,

I guess maybe your time is totally dominated by the grid operations, your 
cell is quite big.

You can make the QM cell smaller (the cell in the QMMM section), so that it 
just fits your QM atoms (allow a bit of spare space).

Matt



On Wednesday, August 16, 2017 at 5:01:37 PM UTC+1, Sun Tao wrote:
>
> Dear CP2K experts:
>
> I am running a system with QMMM method in CP2K. By checking the gaussian 
> electrostatic coupling reference (
> http://pubs.acs.org/doi/abs/10.1021/ct050123f), I am setting the 
> following:
>
>     ECOUPL GAUSS
>     NOCOMPATIBILITY .TRUE.
>     USE_GEEP_LIB 10
>     CENTER NEVER
>
> and I am varying USE_GEEP_LIB with the values of 2, 5, 10 and 12. The 
> problems I have are as the following:
>
> 1. When USE_GEEP_LIB is set to 10, the simulation is fastest. However, I 
> expect the fewer number of levels of grid I use, the faster the simulation 
> would be.
>
> 2. Even if the simulation got accelerated, it is still not enough. The 
> fastest simulation gave me only 4 times speed up compared with pure ab 
> initio MD with DFT for all the atoms.
>
> 3. In the reference, it also mentioned that the cutoff distance would 
> affect the acceleration, while I did not find the corresponding set up in 
> the manual.
>
> I am attaching all my input files. I really want to know how could I set 
> up the input files so that I could get faster simulations? (4 times faster 
> is not good enough for me obviously).
>
> Thanks very much for your help!
>
> Tao
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170817/995774dc/attachment.htm>