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 which was intended to get the same simulations running with fully periodic QM/MM PBC). As I have found out, 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 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

[Jadzia]

I have contacted one of the developers of DFTB3. He has also much experience with DFTB3 QM/MM water simulations, and he said that even though DFTB3 has a few problems with water, these should not cause them to fall apart. 

Therefore there seem to be two possibilities:

1. Something is wrong with the setup. 
   
2. There is a bug in CP2K. 

I have tried my best to check if something is wrong with the setup, but I couldn't find anything. Also, I have run a lot of simulations with different settings. 

If it is a bug in CP2K I cannot tell, this might be a question for the developers. 

Best wishes,

Jadzia

[Jadzia]

For a further test, I made a smaller water box with only 22 Angstrom box edge length this (before 42 Angstrom). 

I adapted the QM Poisson parameters to:

    &POISSON

      &EWALD

       EWALD_TYPE spme

       ALPHA 0.3

       GMAX 11 11 11

       O_SPLINE 5

      &END EWALD

    &END POISSON

Now the waters are falling apart again. GMAX I have set to half of the box edge lengths, because this worked in the previous test system.

My guess is that the QM/MM electrostatic interaction somehow disturbs the QM density in an unnatural way. The question is can this be fixed?

Best wishes,

Jadzia