DFT+U calculation problems

[Xiaohui Yang]

Dear, all

I did some calculations with DFT+U functional recently, they all obtained from the CP2K program. 

At first, I calculated the geometry optimization of the structre, and obtained the total energy directly from the last optimized structure (let's say E1). Then, I ran a single point energy calculation with the same optimized structure and got the total energy E2. 

However, those two total energies (E1 & E2) I collected from the exact same structure have significant difference, E2-E1 = 1.2 eV.  

When checking their output results,  I find that the total energy difference could be attributed to two parts, DFT+U energy and  Core Hamiltonian energy. The input and output files are attached below. 

As far as I know, the total energy calculated from the same settings and structures should be the same. I have no idea about these results. Could you please help me check what is wrong with my calculation? Any comments will be greatly appreciated. 

Best Regards,

Xiaohui

[Krack Matthias (PSI)]

Dear Xiaohui

 

It seems the two DFT+U runs converged to different states for the same atomic configuration, since band gaps and spin densities are quite different. The Hubbard U term promotes an electron localisation, in your case of the Ti 3d electrons. Different 3d orbital occupation patterns can result in different total energies, especially for large U values. You can check the 3d occupations by activating the &PLUS_U print section in &FORCE_EVAL/&DFT/&PRINT.

 

Best regards

 

Matthias

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To post to this group, send email to cp...@googlegroups.com.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.

[Xiaohui Yang]

Dear Matthias,

Many thanks for your reply! I will check the 3d occupations. As you mentioned, the result converged to different states for the same configuration, I am wondering if the SCF threshold is too large in this case. The previous EPS_SCF was 1.0E-7, I am going to try a smaller threshold for a more accurate result.

Thanks again for your help!

Best Regards,

Xiaohui

在 2018年4月1日星期日 UTC+8下午11:13:16，Matthias Krack写道：