This thread is locked.Only browsing is available.
Top Page > Browsing
Difference in total energy between version 3.6 and 3.7
Date: 2013/08/01 19:18
Name: Bob Schoeters   <bob.schoeters@uantwerpen.be>

Dear Prof. Ozaki,

before I started using OpenMX 3.7 I started running some tests, to see if I could proceed with the calculations I was doing but now with the new version.

I compared the total energy + the forces computed by both versions and saw a difference.
Both calculations were done using exactly the same inputfile (so with the same pseudopotentials). The system considered consists of 3 unit cells of a SiGe core-shell nanowire of about 2nm.

Version 3.6 gives: |Maximum force| (Hartree/Bohr) = 0.000837613655

*******************************************************
Total Energy (Hartree) at MD = 1
*******************************************************

Uele = -488.416677253263

Ukin = 498.666362793599
UH0 = -20767.418393882697
UH1 = 3.342455266330
Una = -650.232402107206
Unl = 195.596132325449
Uxc0 = -369.494148477977
Uxc1 = -369.494148477977
Ucore = 19890.981590572967
Uhub = 0.000000000000
Ucs = 0.000000000000
Uzs = 0.000000000000
Uzo = 0.000000000000
Uef = 0.000000000000
UvdW = 0.000000000000
Utot = -1568.05255198751138

Version 3.7: |Maximum force| (Hartree/Bohr) = 0.000816178047

*******************************************************
Total Energy (Hartree) at MD = 1
*******************************************************

Uele = -488.415484052568

Ukin = 498.672542744951
UH0 = -20767.419154943644
UH1 = 3.342344175158
Una = -650.236186219205
Unl = 195.598465760279
Uxc0 = -369.488766114339
Uxc1 = -369.488766114339
Ucore = 19890.981590572963
Uhub = 0.000000000000
Ucs = 0.000000000000
Uzs = 0.000000000000
Uzo = 0.000000000000
Uef = 0.000000000000
UvdW = 0.000000000000
Utot = -1568.037930138175

I am aware that the absolute value of the total energy has little meaning, but since I use the same pseudopotential and except some upgrades the same code, I find this strange. The difference in energy looks small in Hartree, but is very large if you are for example interested in formation energies.

Could you explain the reason(s) for these differences?

Now I am in doubt whether to continue this study using version 3.6 or recalculating everything everything using 3.7, since the speed up is remarkable.

Kind regards,
Bob Schoeters
メンテ
Page: [1]

Re: Difference in total energy between version 3.6 and 3.7 ( No.1 )
Date: 2013/08/01 23:25
Name: T. Ozaki

Hi,

At least, there are two places where methods have been changed for the numerical
integration in Ver. 3.7 compared to Ver. 3.6. The first one is how a set of FFT
grids is generated based on the keyword 'scf.cutoffenergy'. Please check whether
you used the same number of grids along a-, b-, and c-axes for both the calculations.
The second one is how the exchange-correlation energy is evaluated. In Ver. 3.7,
the Levedev grid for spherical integrals has been introduced for efficiency and accuracy.

These changes must actually cause the difference.

Of course, you have to use the same version (either Ver. 3.6 or Ver. 3.7) for
evaluation of the formation energy so that the error related to the numerical
integration can be cancelled out.

Regards,

TO
メンテ
Re: Difference in total energy between version 3.6 and 3.7 ( No.2 )
Date: 2013/08/02 16:53
Name: Bob Schoeters  <bob.schoeters@uantwerpen.be>

Dear Prof. Ozaki,

thank you for the quick and clear response.

Regards,
Bob
メンテ

Page: [1]