COSMO convergence
Gets Around
9:27:55 AM PST - Fri, Feb 10th 2012
Hi,

I do not converge COSMO calculation. Could you be so kind to help me?


start nw
memory stack 600 mb heap 100 mb global 800 mb
charge 0

geometry 
 C    -0.60565725 1.25747713 0 
 C    -1.39183577 0.104224 0   
 C    -0.78617852 -1.15325313 0
 C    0.60565725 -1.25747713 0 
 C    1.39183577 -0.104224 0   
 C    0.78617852 1.15325313 0  
 H    -1.0766783 2.23541955 0  
 H    -2.47426927 0.18527902 0 
 H    -1.39759097 -2.05014053 0
 H    1.0766783 -2.23541955 0  
 H    2.47426927 -0.18527902 0 
 H    1.39759097 2.05014053 0  
end 

basis cartesian
 * library aug-cc-pvdz
end 

dft 
 CONVERGENCE density 1e-6
 xc pbe0
end 

cosmo 
 dielec 78.4
end 

esp 
 recalculate 
 probe   0.07
 range   0.3
 factor  1
 spacing 0.02
end 

property 
 dipole 
 quadrupole 
end 

task DFT property
task esp

Forum Vet
11:19:08 AM PST - Fri, Feb 10th 2012
Per documentation: http://nwchemgit.github.io/index.php/Release61:Density_Functional_Theory_for_Molecules#ITERA...


Quote:P99 Feb 10th 4:27 pm
Hi,

I do not converge COSMO calculation. Could you be so kind to help me?


start nw
memory stack 600 mb heap 100 mb global 800 mb
charge 0

geometry 
 C    -0.60565725 1.25747713 0 
 C    -1.39183577 0.104224 0   
 C    -0.78617852 -1.15325313 0
 C    0.60565725 -1.25747713 0 
 C    1.39183577 -0.104224 0   
 C    0.78617852 1.15325313 0  
 H    -1.0766783 2.23541955 0  
 H    -2.47426927 0.18527902 0 
 H    -1.39759097 -2.05014053 0
 H    1.0766783 -2.23541955 0  
 H    2.47426927 -0.18527902 0 
 H    1.39759097 2.05014053 0  
end 

basis cartesian
 * library aug-cc-pvdz
end 

dft 
 CONVERGENCE density 1e-6
 xc pbe0
end 

cosmo 
 dielec 78.4
end 

esp 
 recalculate 
 probe   0.07
 range   0.3
 factor  1
 spacing 0.02
end 

property 
 dipole 
 quadrupole 
end 

task DFT property
task esp

Forum Vet
12:21:36 PM PST - Fri, Feb 10th 2012
switch to spherical basis set because of linear dependence
P99,
I strongly recommend you to switch from cartesian to spherical basis in your input,
since you have a few linear dependencies causing numerical problem.
Once you switch to spherical, the dependencies are gone, and the convergence is much smoother.
In other words, the basis set input lines should be

basis spherical
* library aug-cc-pvdz
end


Cheers, Edo

Gets Around
11:22:40 PM PST - Sun, Feb 12th 2012
Unfortunately, neither an increase the number of iterations (up 500) nor a spherical basis are not help.

Forum Vet
11:32:16 AM PST - Mon, Feb 13th 2012
The following input deck will converge without any problems, using an SCF as a starting guess for the DFT, and setting "tolerances tight" to avoid more loose integral and matrix element calculations during the early iterations of DFT.

Bert




start nw
memory stack 600 mb heap 100 mb global 800 mb
charge 0

geometry 
C    -0.60565725 1.25747713 0 

C    -1.39183577 0.104224 0   

C    -0.78617852 -1.15325313 0

C    0.60565725 -1.25747713 0 

C    1.39183577 -0.104224 0   

C    0.78617852 1.15325313 0  

H    -1.0766783 2.23541955 0  

H    -2.47426927 0.18527902 0 

H    -1.39759097 -2.05014053 0

H    1.0766783 -2.23541955 0  

H    2.47426927 -0.18527902 0 

H    1.39759097 2.05014053 0
end
basis spherical
* library aug-cc-pvdz
end
scf
vectors output nw.movecs
end

task scf
dft 
tolerances tight

CONVERGENCE density 1e-6

xc pbe0

vectors input nw.movecs
end
cosmo 
dielec 78.4
end
esp 
recalculate 

probe   0.07

range   0.3

factor  1

spacing 0.02
end
property 
dipole 

quadrupole
end
task DFT property
task esp


Quote:P99 Feb 13th 6:22 am
Unfortunately, neither an increase the number of iterations (up 500) nor a spherical basis are not help.

Gets Around
2:32:29 AM PST - Wed, Feb 15th 2012
Quote:Bert Feb 13th 6:32 pm
The following input deck will converge without any problems, using an SCF as a starting guess for the DFT, and setting "tolerances tight" to avoid more loose integral and matrix element calculations during the early iterations of DFT.


Really this code converges in DFT, and ... then crashes. It would be nice to have the solution of the task, but not the convergence of the DFT per se. The convergence of the DFT requires a spherical basis, the calculation of charges requires a cartesian. As a result, we have not solution.

Forum Vet
6:18:28 PM PST - Wed, Feb 15th 2012
There seems to be something going on in the DFT when it users high symmetry. Will are going to use your test case to resolve the issue. A couple of solutions in the mean time:

1. Turn off symmetry by specifying noautosym in the geometry block, i.e. "geometry noautosym". This will get your calculation to converge at the DFT level with NWChem 6.0.

2. Install the brand new NWChem 6.1. With this code you can get ESP charges with spherical basis sets.

3. Install the brand new NWChem 6.1 with a better convergence algorithm by specifying the cgmin keyword in the dft block. Now you can still use the cartesian basis if you want.

Bert



[QUOTE=P99 Feb 15th 9:32 am][QUOTE=Bert Feb 13th 6:32 pm]The following input deck will converge without any problems, using an SCF as a starting guess for the DFT, and setting "tolerances tight" to avoid more loose integral and matrix element calculations during the early iterations of DFT.


Really this code converges in DFT, and ... then crashes. It would be nice to have the solution of the task, but not the convergence of the DFT per se. The convergence of the DFT requires a spherical basis, the calculation of charges requires a cartesian. As a result, we have not solution.

Gets Around
7:27:16 AM PST - Thu, Feb 16th 2012
Thank you very much.


Forum >> NWChem's corner >> General Topics