#26 Correction of Exchange-Correlation Error (3): Hybrid Functional - Materials Square
MatSQ Blog > Quantum Espresso

#26 Correction of Exchange-Correlation Error (3): Hybrid Functional

2019-12-05 16:45:14

In this weekly tip, following the previous weekly tips #24 DFT+U and #25 van der Waals correction, we will learn how to correct GGA errors using hybrid functional.

 

1. Hybrid Functional

The methods DFT+U and van der Waals correction, which we learned from the previous weekly tip, correct the correlation part of exchange-correlation potential. Meanwhile, the hybrid functional corrects the exchange part.

The hybrid functional, such as PBE0, B3LYP, and HSE, was known as the high-accuracy correction method of exchange-correlation energy [1]. It has higher accuracy than the GGA method as it mixes the Fock exchange to the exchange-correlation functional of GGA when describing electronic structure [2]. In this time, the exchange energy approximation of GGA and HF is mixed in a different method according to the kind of hybrid functional. Unlike the GGA, the hybrid functional has a long-range interaction effect because the Fock exchange describes some part of the long-range interaction.

The hybrid functional not only corrects the total energy of materials but also serves useful in obtaining the correct electronic structure. In particular, it can almost solve the bandgap underestimation, which is the fundamental error of GGA [3]. GGA always estimates a smaller bandgap than it actually is, while Hartree-Fock always estimates larger than it actually is. Therefore, the hybrid functional that mixes them at an optimal ratio can effectively reduce the bandgap estimation error.

 

However, the hybrid functional consume a lot of calculation resource than GGA.

In this weekly tip, we will find out the difference between GGA and hybrid functional regarding electronic structure and calculation time through the density of states (DOS).

 

2. Setting Hybrid Functional in MatSQ

The keyword for hybrid functional can be set at “Scripting option: General.” To use a specific hybrid functional, you should write the desired name of hybrid functional at the “DFT Functional” option in &SYSTEM Namelist. The list of available functional can be retrieved from the Documentation.

 

3. Precautions When Using Hybrid Functional

The calculation with hybrid functional is so complicated that it is not supported in the template mode. Therefore, you should add relevant keywords to the input script in the Manual mode for further calculation setting.

In GGA calculations, the k-point is set to obtain an electronic structure with adequate accuracy while considering efficiency. However, calculation resources are consumed excessively if the GGA k-point setting is applied to exchange energy calculation. Therefore, we should separately set the number of crystal momentum q, which is used for only Hartree-exchange calculation, by adding nqx1, nqx2, and nqx3 keywords under &SYSTEM.

 

Unlike the general GGA calculation, it consumes extremely high memory to calculate the exchange energy in hybrid functional DFT. Moreover, the memory size of the MatSQ server is set according to the number of cores. Hence, it is better to increase the number of cores if the calculation takes too much time or too big memory.

MatSQ provides a server with 48 cores per node. Thus, you should increase the number of nodes if you want to use more than 36 cores. Moreover, the MatSQ credit is consumed in proportion to the number of cores. Therefore, more credit is consumed when using two more nodes compared to using one node, even it takes the same time.

 

4. Calculation Results

We can find out the change of electronic structure when using the hybrid functional through the DOS of the silicon primitive unit cell.

👉 Check the result of calcultaion

Based on the DOS graph obtained by three functional, the entire shape of the DOS graph did not show a significant change. However, the bandgap was improved.

Check the table below for a summary of the bandgap data, calculation time, and the credits used to obtain that data.

When using HSE, we found out that PBE0 the hybrid functional can solve some of the bandgap underestimation problems of PBE. However, the calculation time increased more than 10 times. In GGA, the calculation time is generally proportional to three squares of the number of atoms, but the calculation time might be increased more than five squares when using the hybrid functional.

For example, the silicon primitive unit cell, which is the calculation model used in this example, consisted of just two atoms, but it consumed around 50 min in using the hybrid functional.

Therefore, while hybrid provides better results, you need to be careful of the consumption of computing resources.

We checked the changed electronic structure when using the hybrid functional as well as the precautions of calculation and increased calculation time. Moreover, the hybrid functional DFT can provide improved results, but it needs to be careful because of the high consumption of computing resources.

In MatSQ, you can apply the hybrid functional at the simple structure to obtain an improved DFT result!

 

Thank you for your interest in Materials Square weekly tips until now.
We are ending the weekly tip season 1 with this weekly tip as the last one. However, we will come back season 2 with more content. Thank you!

 

 


[1] Martin, R. M., & Martin, R. M. (2004). Electronic structure: basic theory and practical methods. Cambridge university press.
[2] Paier, J., Marsman, M., Hummer, K., Kresse, G., Gerber, I. C., & Ángyán, J. G. (2006). Screened hybrid density functionals applied to solids. The Journal of chemical physics, 124(15), 154709.
[3] Jain, M., Chelikowsky, J. R., & Louie, S. G. (2011). Reliability of hybrid functionals in predicting band gaps. Physical review letters, 107(21), 216806.

 


Materials Square
Sign up for free & Get $10!
🚀 Go to Materials Square

Creative Commons Attribution-Nonprofit-ShareAlike 4.0

This post is licensed under a Creative Commons Attribution-Nonprofit-ShareAlike 4.0 International License.

Comments

frolep rotrem - 2020-01-12 13:15:01
Very great post. I simply stumbled upon your blog and wanted to mention that I've truly loved browsing your blog posts. In any case I will be subscribing for your rss feed and I hope you write again soon!
Malg Eum Yu - 2020-01-13 10:37:53
Thank you so much!
Djamel - 2020-07-23 12:08:50
I need tutorial for calculation a spontanus polarization of 2D materials whith quantum espresso . Thank you