[CP2K-user] [CP2K:14386] Re: Solid state reaction: suitable decoupling

[Jörg Saßmannshausen]

Dear Luca,

thanks for the feedback.

Sorry I was not clearer. 
I was talking about the Poisson solver. 
When you look at the manual:
https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/POISSON.html

you will see there are several options for the POISSON_SOLVER. like MT, 
WAVELET, and PERIODIC for example, and there are is also the PERIODIC keyword. 
So in the past, when I wanted to do a gas phase calculation similar to the one 
you do in say GAMESS, I was using MT and PERIODIC XYZ (the latter for both in 
FORCE_EVAL and SUBSYS). But that would mean I am repeating my gas phase 
calculation in cp2k, which is not what I want to do here. 

Judging from the manual link above, I guess POISSON_SOLVER would be PERIODIC 
as it is better suited in this case.  
Again, the manual suggests that in that case PERIODIC would be set to XYZ as 
that is the default. 
Assuming that is correct, how big do I need to set the unit cell in that case? 
In the x-ray it is a monoclinic space group so in this case I would need to 
set
ABC [angstrom] 10.26100 16.90010 20.89670
ALPHA_BETA_GAMMA 90.0 98.2650 90.0
I guess (values taken from the cif-file).
In this respect, does it make more sense to read in the cell and the atomic 
coordinates from the cif-file, instead of converting that to a xyz file?
Is cp2k 'clever' enough to make the unit cell big enough for the respective 
poisson solver when it is reading the data directly from the cif- file?

As I said, right now it is not about the reaction mechanism, right now it is 
the computation of the final product in the unit cell which we observer. 

I hope it makes better sense now. 

All the best and thanks for your help!

Jörg

Am Sonntag, 13. Dezember 2020, 21:20:38 GMT schrieb Luca:
> Dear Jörg,
> maybe, I did not understand  the question;
> if you "decouple the cell" with MT scheme you are going to model the system
> in gas phase like in GAMESS.
> Anyway, if you do not use periodic system, you have to check that electron
> density goes to zero prior to the cell border.
> When I study a reaction with cp2k, and I need to  characterize the TS, I
> use dimer-method or NEB.
> If I need to model the system in gas phase
> (no periodic system) and the system is "spheric-like",
> I prefer to use  WAVELET.
> I hope it is helpful to you.
> Cheers,
> 
> Luca
> 
> Il giorno domenica 13 dicembre 2020 alle 12:26:24 UTC+1 sassy ha scritto:
> > Dear all,
> > 
> > I am normally calculating molecules in the gas phase, so my knowledge of
> > cp2k
> > is still a but rusty as I don't use it as much as I would love to.
> > 
> > I am currently looking into an observed solid state reaction of the type:
> > 
> > A -> B + C
> > 
> > We got an x-ray of A and we observe the product B + C per x-ray as well as
> > a ~
> > 10% "contamination".
> > If I am using the usual DFT tools like GAMESS-US or Gaussian and put the
> > coordinates of A + B in the input file, they are flying away from each
> > other.
> > That makes sense in the gas phase but not in the solid state.
> > Thus, I thought of calculating the unit cell in cp2k which should mimic
> > better
> > the conditions where we are observing the products.
> > 
> > I was thinking of using PBE-D3 for that, together with the Molopt basis
> > set
> > which was working well for me in other calculations I done before.
> > 
> > I am a bit unsure regarding the affect the size of the cell and which
> > decoupler to use. I usually go for MT but as there has been quite some
> > development recently it seems, I am not sure if that is still the best
> > current
> > advice.
> > 
> > For example, what will happen if for my specific problem I make the cell a
> > bit
> > bigger than what you usually would do for the chosen decoupler?
> > 
> > The overall aim of the project is more ambitious, I want to model the
> > reaction
> > path of this reaction if that is possible. However, that is for the
> > future.
> > 
> > Any suggestions are much appreciated.
> > 
> > All the best from Lyttle Britain.
> > 
> > Jörg