DFTB3 within QM/MM simulations

[Jadzia]

Hi everyone,

I'm trying to use DFTB3 as the QM method of a QM/MM simulation of water 
(which is only a test system).

However, my QM water molecules are falling apart. At first some H20 
molecules form H3O+ and OH-, and then it goes on to form H4O++ and O--. All 
this happens between QM atoms only. 

I have adjusted and switched many QM and QMMM parameters (see also my last 
topic here <https://groups.google.com/forum/#!topic/cp2k/EmRoMadjbRw> which 
was intended to get the same simulations running with fully periodic QM/MM 
PBC). As I have found it, it seems that the Hubbard derivative (which is to 
my knowledge the DFTB3_PARAM in CP2K) is related to the problem. If I do 
not specify this parameter for the H and O atoms (it is set to 0 by default 
by CP2K) then my waters are not falling apart. The Hubbard derivatives I 
have used are the ones provided by dftb.org. 

Do you think it is fine to run the simulations without specifying the 
Hubbard derivatives? (i.e. using the default value of 0) 
And do you have any other suggestions or comments on why this problem could 
arise or solved?

If I run QM-only simulations with the same DFTB3 settings and with the 
official Hubbard derivative parameters from dftb.org, then the simulations 
run just fine. Only when enabling QM/MM my QM water falls apart.

However, I now found out that even without Hubbard derivatives I manage to 
let the water fall apart, namely when I change the Poisson settings of the 
QM region. With the following QM Poisson settings the water does not fall 
apart:

    &POISSON
      &EWALD
       EWALD_TYPE SPME
       GMAX 25
       O_SPLINE 5
      &END EWALD
    &END POISSON

The entire input file is attached as well. When changing the Poisson 
settings to the values which seem to be more optimal:

    &POISSON
      &EWALD
       ALPHA .44
       EWALD_TYPE SPME
       GMAX 42 42 43
       O_SPLINE 6
      &END EWALD
    &END POISSON

then the QM water falls apart. Which I find strange since these Poisson 
settings are more "sound" because they are more close to what is 
recommended. E.g. the box size is 42 42 53 Angstrom, and it is recommended 
to use 1 per Angstrom for GMAX. Also, O_SPLINE was set before to 5, which 
can lead to an interpolation problem to my knowledge. 

Looking forward to hearing from you and your recommendations. 

Best wishes,
Jadzia



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170913/d78de025/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cp2k.in.config
Type: application/octet-stream
Size: 3955 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170913/d78de025/attachment.obj>