AI-powered DFT methods
Fudan University, Shanghai, 20048, China
Density Functional Theory (DFT) is currently the most widely used and successful method for electronic structure calculations. However, as the exact functional is unknown, all DFT computations require using some forms of Density Functional Approximations (DFAs). We believe that, on one hand, it is necessary to develop increasingly accurate DFAs that incorporate the basic essence of physics; one the other hand, it is important to develop statistical models, using machine learning (ML) techniques, to correct the intrinsic errors of the existing DFAs. With more accurate data being accessible, new physics and new insights are expected to emerge. To this end, our group is dedicated to developing the doubly hybrid functionals of the XYG3 type [1-5], while also advancing multiple ML methods for accurate predictions of energies and properties [6-10]. This talk will introduce some of the effort and progress made by our group in developing AI-powered DFT methods.
[1] Y. Zhang, X. Xu, W. A. Goddard III, Proc. Nat. Acad. Sci., USA 2009, 106, 4963.
[2] Y. Zhang, X. Xu, Y. Jung, W. A. Goddard III, Proc. Nat. Acad. Sci., USA 2011, 108, 19896.
[3] N. Q. Su, Z. Zhu, X. Xu, Proc. Nat. Acad. Sci., USA 2018, 115, 2287.
[4] Y. Wang, Y. Li, J. Chen, I.Y. Zhang, X. Xu, JACS Au 2021, 1, 543.
[5] Y. Wang, Z. Lin, R. Ouyang, B. Jiang, I.Y. Zhang, X. Xu, JACS Au 2024, 4, 3205.
[6] J. Wu, X. Xu, J. Chem. Phys. 2007, 127, 214105.
[7] I. Y. Zhang, J. Wu, X. Xu, Chem. Commun., 2010, 46, 3057.
[8] Y. Zhou, J. Wu, X. Xu, J. Comput. Chem., 2016, 37, 1175.
[9] Y. Chen, W. Yan, Z. Wang, J. Wu, X. Xu, J. Chem. Theory Comput. 2024, 20, 9500
[10] W. Yan, X. Lai, Y, Chen, W. Zhang, J. Wu, X. Xu, J. Am. Chem. Soc. 2025, 147, 47044