The Best DFT Functional Is the Ensemble of Functionals
Author Information
Author(s): Rui Yuting, Chen Yuxinxin, Ivanova Elena, Kumar Vignesh Balaji, Śmiga Szymon, Grabowski Ireneusz, Dral Pavlo O.
Primary Institution: Xiamen University
Hypothesis
Can ensembles of density functionals provide better accuracy than individual density functionals in density functional theory (DFT) calculations?
Conclusion
The DENS24 ensembles of density functionals achieve superior accuracy compared to any individual functional, reaching a record-low weighted error of 1.62 kcal mol−1.
Supporting Evidence
- The DENS24 ensemble achieved a weighted error of 1.62 kcal mol−1, significantly lower than the best individual functional's error of 3.08 kcal mol−1.
- The study demonstrates that ensembles can be trained on a limited number of functionals and still maintain good transferability.
- DENS24 ensembles are open-source and can be used for various simulations at an affordable cost.
Takeaway
This study shows that using a group of different methods together can give better results than using just one method by itself.
Methodology
The study proposes a procedure for constructing density functional ensembles (DENS) that combines predictions from multiple density functionals to improve accuracy.
Limitations
The study is limited to the GMTKN55 database and the functionals originally benchmarked on it.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website