QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: Sarvesh Agarwal on October 30, 2012, 11:52

Title: OPTIMIZATION_PROBLEM
Post by: Sarvesh Agarwal on October 30, 2012, 11:52

HELLO Everyone!!
              i have a nanotube structure which has a gas moleclue for adsorbtion, i put it for optimization with DFT (as its necessary to optimize before IV calculation) but the problem is its taking too much time 10-12 days . Is there anyway to reduce it??
Iam using sunfirecluster with 76 cores

Title: Re: OPTIMIZATION_PROBLEM
Post by: zh on October 30, 2012, 12:27

Please post your script file. It will quite useful for looking into your problem.

Title: Re: OPTIMIZATION_PROBLEM
Post by: Sarvesh Agarwal on October 30, 2012, 14:29

thanks for the interest!!! please also tell me the role of SCF iteration and constraint cell!!

Title: Re: OPTIMIZATION_PROBLEM
Post by: zh on October 31, 2012, 05:48

The k-point mesh is not setup properly. Actually, it is quite dense for your supercell. So the calculation is very slow.
"numerical_accuracy_parameters = NumericalAccuracyParameters(
    k_point_sampling=(3, 3, 100),
    )
"

"
vector_a = [17.8336117435, 0.0, 0.0]*Angstrom
vector_b = [0.0, 17.8336117435, 0.0]*Angstrom
vector_c = [0.0, 0.0, 34.10064]*Angstrom
lattice = UnitCell(vector_a, vector_b, vector_c)
"
Because a large supercell is employed in your calculation, you can used 1 x 1 x1 k-mesh (i.e., Gamma point).

Title: Re: OPTIMIZATION_PROBLEM
Post by: Sarvesh Agarwal on October 31, 2012, 07:10

thanks sir!!
         IS SCF iteration necessary or not ?
          but does accuracy depends on this k-points... what is the real function of this parameter do you have any notes related to this??

Title: Re: OPTIMIZATION_PROBLEM
Post by: Sarvesh Agarwal on November 3, 2012, 12:14

hello sir!!
    i did as you suggested me to change the k-points as 1,1,1, and No SCF iteration... but i think still its not able to converge .... i am attaching my log please have a look at this and suggest me something...

Title: Re: OPTIMIZATION_PROBLEM
Post by: Sarvesh Agarwal on November 6, 2012, 10:52

can you please help me with this

Title: Re: OPTIMIZATION_PROBLEM
Post by: Anders Blom on November 6, 2012, 11:31

If you want accurate forces in DFT you need to run the calculation self-consistently.

But there really shouldn't be any need to simulate the entire system like this. All you really want is to know how the molecule sits on the nanotube, so I would take a much smaller piece of the nanotube, with the molecule, constrain all the Carbon atoms, and run with perhaps 100 atoms instead of 300 - it makes a huge difference in calculation time. In principle the Carbon under the molecule would relax too, however, so a better alternative might be to make a small segment of the tube, with periodic conditions in Z, with the molecule, and more k-points in C (still 1x1 in A/B of course), then after the relaxation repeat the tube and remove the repeated molecules, and convert to device.

Note, however, that the device calculation with 300 atoms is also going to be time-consuming.

Title: Re: OPTIMIZATION_PROBLEM
Post by: Sarvesh Agarwal on November 8, 2012, 12:35

thanks sir !!
ur suggestion is really helpful