Top Page > Browsing
Can I optimize the "Optimized PAO"?
Date: 2017/03/27 05:26
Name: Pui-Wai (Leo) Ma   <leo.ma@ukaea.uk>

Dear Prof. Ozaki,

I would like to work on a system with ~1500 tungsten atoms. I can only use W9.0-s2p1d1. I cannot increase the number of orbitals due to the limitation of memory. I tested the accuracy of energy by calculating the <111> crowdion formation energy using 129 atoms cell, in comparison to 128 atoms perfect lattice. It seems the result is a bit far from the VASP result.

Is it possible to use existing W9.0.pao to have further optimization on the local atomic orbitals, and achieve a better result? Thank you for your help.

Best regards,
Leo Ma

Dr. Pui-Wai (Leo) Ma
Senior Scientist
Culham Centre for Fusion Energy (CCFE)
Abingdon, Oxfordshire OX14 3DB, UK
Tel. +44 (0)1235 465194
Fax. +44 (0)1235 466435
email: leo.ma@ukaea.uk
メンテ
Page: [1]

Re: Can I optimize the "Optimized PAO"? ( No.1 )
Date: 2017/03/27 18:09
Name: Pui-Wai (Leo) Ma  <leo.ma@ukaea.uk>

Dear Prof. Ozaki,

I am sorry. I think I was asking a wrong question. Please correct me if my understanding is wrong.

1) If I write sn>n = sn, it means I am using n primitive orbitals as if they are n optimized orbitals.

2) If I write sm>n, it means I am constructing n optimized orbitals from m primitive orbitals using coefficients from the PAO file.

3) If I do orbital optimization, it is optimizing the contraction coefficients.

Then, it comes to something I don't understand. In the file Wbcc_opt.dat, it is using W7.0p-s15>1p15>2d10>1f10>1, and in W_opt.dat, it is using W7.0p-s15>2p15>1d15>1. In the example, W7.0-s3p2d2f1 and W7.0-s3p3d3f2 are used. I have 3 questions.

1) When W7.0-s3p2d2f1 and W7.0-s3p3d3f2 are used, they are only using primitive orbitals, so no contraction coefficient is used?

2) If I use contraction, in W7.0.pao, there are 2 set of optimized contraction coefficients, which set is used?

3) If I use contraction, should I use W7.0-sn1>m1pn2>m2dn3>m3fn4>m4, where
n1 <= 15, m1 <=2,
n2 <= 15, m2 <=2 or 1 ??,
n3 <= 10 or 15 ??, m3 <=1,
n4 <= 10, m4 <=1

Thank you very much for your help.


Best regards,
Leo Ma
メンテ
Re: Can I optimize the "Optimized PAO"? ( No.2 )
Date: 2017/03/31 06:58
Name: T. Ozaki

Dear Leo,

> 1) If I write sn>n = sn, it means I am using n primitive orbitals as if they are n
> optimized orbitals.
> 2) If I write sm>n, it means I am constructing n optimized orbitals from m primitive
> orbitals using coefficients from the PAO file.
> 3) If I do orbital optimization, it is optimizing the contraction coefficients.

Yes, you are right.

> 1) When W7.0-s3p2d2f1 and W7.0-s3p3d3f2 are used, they are only using primitive
> orbitals, so no contraction coefficient is used?

No. They are already contracted. The pao files on the database 2013 store the contracted
orbitals. So, when you specify as W7.0-s3p2d2f1, you will use the contracted orbitals
generated by the pre-orbital optimization.


> 2) If I use contraction, in W7.0.pao, there are 2 set of optimized contraction
> coefficients, which set is used?

Two optimized orbitals are merged in W7.0.pao based on a subspace rotation technique.
When you specify W7.0-s1p1d1, the orbitals correspond to optimized ones for a single atom.
When you specify W7.0-s2p2d2, the first ones (s1p1d1) are equivalent to those
of W7.0-s2p2d2, and the second s-, p-, and d-orbitals are optimized ones for the bcc bulk.
Therefore, W7.0-s2p2d2 are optimum for both the single atom and bulk so that
the basis functions can be very transferable for a wide range of environments.


> 3) If I use contraction, should I use W7.0-sn1>m1pn2>m2dn3>m3fn4>m4, where
> n1 <= 15, m1 <=2,
> n2 <= 15, m2 <=2 or 1 ??,
> n3 <= 10 or 15 ??, m3 <=1,
> n4 <= 10, m4 <=1

Unless you generate your own optimized orbitals, you do not need to be bothered by
the issue.


By the way, did you check the results for the <111> crowdion formation energy using 129
atoms cell, in comparison to 128 atoms perfect lattice, using the proper sized basis
functions in OpenMX? If the result is not well compared to reliable values by other codes,
something in the calculation must be wrong or the quality of pseudopotential may not be reliable.
So, I think what you should do first is to reproduce the result by setting
a proper parameters. I guess that W7.0-s2p2d2 might be enough to reproduce the result.
Once you obtain the comparable result, you can start to consider how the computational effort
can be reduced without significantly loosing the accuracy.

Regards,

TO

P.S.
I guess that you are working with Duc and Sergei.
Please give them my best regards.
メンテ
Re: Can I optimize the "Optimized PAO"? ( No.3 )
Date: 2017/03/31 01:11
Name: Pui-Wai (Leo) Ma  <leo.ma@ukaea.uk>

Dear Prof. Ozaki,

Thank you very much for your reply. Surely, I will pass your regards to them.

I am currently checking the 53 atoms vacancy case. The convergent result for formation energy is about 10% smaller than result using VASP. I am using W9.0-s3p2d2f1. I will try to do contraction to see if I can get the same result.

I shall check the convergent result for <111> crowdion later. I only worked on s2p1d1, but the result is 30% smaller than VASP, as expected.

Contraction is important to me, since I am aiming at a system with 1500 atoms or larger. I will need you help again in near future.

By the way, do you have plan to implement AM05 functional? http://dft.sandia.gov/functionals/AM05.html

Best regards,
Leo Ma
メンテ
Re: Can I optimize the "Optimized PAO"? ( No.4 )
Date: 2017/03/31 07:01
Name: T. Ozaki

Dear Ma,

How about using W7.0 instead of W9.0?
I wonder W7.0 is enough to get convergent results for bulks.
If so, memory requirement can be reduced compared to the case of W9.0.

We are not planning to implement AM05 at this moment.

Regards,

TO
メンテ
Re: Can I optimize the "Optimized PAO"? ( No.5 )
Date: 2017/03/31 17:50
Name: Pui-Wai (Leo) Ma  <leo.ma@ukaea.uk>

Dear Prof. Ozaki,

I got better results on vacancy and crowdion. It seems W7.0 is also ok. I am going to do further checking. I shall discuss with you privately later. Thank you for your help.

Best regards,
Leo Ma
メンテ

Page: [1]

Title (must) Move the thread to the top
Your Name (must)
E-Mail (must)
URL
Password (used in modification of the submitted text)
Comment (must)

   Save Cookie