[CP2K-user] Unable to reproduce results of geometry optimisation (PM3) calculations

[Xin Wu]

Hi,

I looked at your input files and found many *inappropriate* options for the 
PM3 calculations, which are too many to be fully enumerated in this short 
reply.

1. %FORCE_EVAL%SUBSYS%CELL: The Diels-Alder reaction is a gas-phase 
chemical reaction, isn't it? The default value of PERIODIC in this section 
is XYZ. Please specify PERIODIC NONE explicitly.

2. %FORCE_EVAL%DFT%QS%SE%COULOMB and
3. %FORCE_EVAL%DFT%QS%SE%EXCHANGE are used for setting up CUTOFF in the 
evaluation of the semiempirical Hamiltonian, e.g. core Hamiltonian plus the 
two electron parts, so that some *numerically* insignificant integrals can 
be safely neglected in a practical calculation and hence saving the 
computation time. Are the values you set in your input files appropriate 
for your system?

4. %FORCE_EVAL%DFT%MGRID is completely irrelevant to any semiempirical 
methods. Please remove the entire section.

5. Your settings for EPS_SCF in both %FORCE_EVAL%DFT%SCF and 
%FORCE_EVAL%DFT%SCF%OUTER_SCF are completely *incompatible* with the 
default settings in Gaussian. If my memory serves me well, the default 
value for SCF convergence in Gaussian is 10^-8 au for geometry optimization 
and it's less for single-point energy evaluation. Please check!

6. Please find the other inappropriate options in the geometry optimization 
section in your input files...

Xin




On Thursday, October 24, 2019 at 1:33:55 PM UTC+2, Salomé Llabrés wrote:
>
> Hi, 
>
>
> I am trying to reproduce a tutorial that uses Gaussian with CP2K, and I 
> have problems reproducing the results originally achieved. 
>
>
> The tutorial describes a fairly simple Diels-Alder reaction using PM3 
> semiempirical. This is an old GROMACS tutorial that uses Gaussian98: 
> http://wwwuser.gwdg.de/%7Eggroenh/EMBO2004/html/qmvacuum.html#top 
>
>
> I was able to reproduce the reactant, transition state and product 
> energies described in the tutorial using Gaussian09. However, when I 
> attempted to reproduce the results with CP2K, I do not obtain the same 
> energy profile. 
>
>
> I am asking for advice on how to fix this problem and obtain reproducible 
> results in CP2K. 
>
>
> I have documented the protocol to run these simple calculations here : 
> https://github.com/bioexcel/cp2K_qmmm_tutorials_for_biological_simulations/tree/master/GMX_DAA
>
>
> These are the results I obtained so far. 
>
>
> *Energy / kJ mol-1*
>
> *Reactants*
>
> *Transition States*
>
> *Product*
>
> GMX Tutorial (Gaussian98)
>
> 0.00
>
> 202.17
>
> -34.13
>
> Gaussian09
>
> 0.00
>
> 200.06
>
> -36.37
>
> CP2K
>
> 0.00
>
> 152.65
>
> -175.81
>
>
> I tried to fine-tune the parameters of the input file but I cannot find 
> the reason behind these different energy profiles. Here it is a list of the 
> different attempts to solve this problem I tried: 
>
>    - The optimiser has a very significant effect on the structure. 
>       - CG gets a similar structure than Gaussian. 
>       - BFGS leads to the structure after SO2 expulsion. This method 
>       cannot be used with the DIMER method.  
>    - SCF convergence: Using MAX_SCF 2000 is useful to converge the SCF in 
>    these simulations. 
>    - GRID CUTOFF convergence doesn't affect the accuracy of these small 
>    systems.
>
> Please find attached input and output files for these files. Also you can 
> find the protocol used in this GitHub environment: 
> https://github.com/bioexcel/cp2K_qmmm_tutorials_for_biological_simulations/tree/master/GMX_DAA
>
>
> Thank you very much in advance, 
>
>
> Salomé Llabrés, PhD 
>
> Applications developer in Data science
>
> EPCC
>
> University of Edinburgh, UK
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20191024/7778a43f/attachment.htm>