Ri-DFT deviations and error with gradient calculation

Just Got Here

2:34:18 AM PST - Wed, Nov 30th 2016

Hey,

I'm new to NWChem and trying to do DFT calculations with the ri-approximation. However, I have a couple of questions and problems.

1) I managed to do single point calculations on water with the pbe96 functional, def2-SVP basis set and def2-SVP as "cd basis" as well. As I compared total DFT energies with calculations without defined cd basis and results from the same methods performed in turbomole, i noticed that the energy obtained in NWChem with defined cd basis is much lower than the others:
- TM without ri: -76.27094
- TM with ri: -76.27104
- NWChem without ri: -76.27264
- NWChem with ri: -76.91946
Now I'm wondering if I'm just not doing it correctly or if the two programs are calculating different things.
My input is:

Title "water ri"

memory 1600 mb

geometry nocenter noautosym noautoz
O     0.0000000    0.0000000   -0.0644484
H     0.7499151    0.0000000    0.5114913
H    -0.7499151    0.0000000    0.5114913
end

basis
* library def2-SVP
end

basis "cd basis"
* library def2-SVP
end

dft
  xc pbe96
  maxiter 5001
end

task dft energy

2) Everytime I try tasks requiring gradient calculations (such as optimizations) with an additional cd basis, I get the following error:

                            NWChem DFT Gradient Module
                            --------------------------


                                     H2O opt



  charge          =   0.00
  wavefunction    = closed shell

 ------------------------------------------------------------------------
 dftg_cdfit_gen: file cdfit does not exist                   0
 ------------------------------------------------------------------------
 ------------------------------------------------------------------------
  current input line :
    28: task dft optimize
 ------------------------------------------------------------------------
 ------------------------------------------------------------------------
 An error occured while trying to read or write to disk space
 ------------------------------------------------------------------------
 For more information see the NWChem manual at http://nwchemgit.github.io/index.php/NWChem_Documentation


 For further details see manual section:                                                                                                                                                                                                   
0:dftg_cdfit_gen: file cdfit does not exist:Received an Error in Communication
--------------------------------------------------------------------------
MPI_ABORT was invoked on rank 0 in communicator MPI COMMUNICATOR 3 DUP FROM 0
with errorcode -1.

NOTE: invoking MPI_ABORT causes Open MPI to kill all MPI processes.
You may or may not see output from other processes, depending on
exactly when Open MPI kills them.
--------------------------------------------------------------------------

This error is taken from the output of a water geometry optimization, but it also occurs with the sample file for nudged elastic band calculations taken from the manual. Does anyone have an idea what could have gone wrong here?

Just tell me if you need additional information on my calculations. Thank you very much!

Forum Vet

11:40:31 AM PST - Wed, Nov 30th 2016

Incorrect basis library name for cd basis

Joachim,
The problem with your poor value of energy when CD fitting is used is due to the fact that the NWChem basis library entry for "def2-SVP" corresponds to an AO basis set. To use for fitting basis set, a more appropriate choice would be "Weigend Coulomb Fitting", as in the example below

basis "cd basis" spherical
 * library "Weigend Coulomb Fitting"
end

As far as the gradient failure is concerned, it looks like your calculation failed to generate a "*.cdfit" file (during the energy part), that is required in the gradient step. Could you please try to run the below (please do not change its content), and then perform the command "ls -l water_ri.*?


start water_ri

Title "water ri"

memory 1600 mb

geometry nocenter noautosym noautoz
O     0.0000000    0.0000000   -0.0644484
H     0.7499151    0.0000000    0.5114913
H    -0.7499151    0.0000000    0.5114913
end

basis spherical
* library def2-SVP
end

basis "cd basis" spherical
* library "Weigend Coulomb Fitting"
end

dft
  xc pbe96
  maxiter 5001
end

task dft gradient

Just Got Here

6:59:59 AM PST - Thu, Dec 1st 2016
Thanks for your help, Edoapra! You're right, I noticed even Turbomole uses the Weigend Coulomb Fitting set in its standard settings, but writes "def2-SVP" on top of the auxbasis file, which was actually the reason i got confused and took def2-SVP into the nwchem input. The gradient calculation works with your proposed input. I've got .cdfit, .db and .movecs files in my output folder. I also got my initial optimization to work. There was a keyword "cgmin" in the dft section, which I had copied from some sample input (I don't really need it). After deleting it, the error disappeared. So the quadratic convergence algorithm seems to be kind of a problem together with ri-dft. Thanks again!

Forum Vet

3:40:50 PM PST - Thu, Dec 1st 2016
Joachim Thank you for figuring out the problem. I have checked into the NWChem trunk svn a fix that prevents gradient from a cgmin scf to try to use CD fitting

Forum >> NWChem's corner >> Running NWChem