Combining real-space and local range separation—The MH24 locally range-separated local hybrid functional DOI Creative Commons
Toni M. Maier

The Journal of Chemical Physics, Год журнала: 2024, Номер 161(21)

Опубликована: Дек. 3, 2024

In this work, the development of a new general-purpose exchange–correlation hybrid functional based on recent locally range-separated local approach is presented. particular, functional, denoted as MH24, combines non-empirical treatment admixture long-range exact exchange with real-space separation for exact-exchange governed by mixing function (LMF) and empirical LYP-based correlation to enable flexible description same- opposite-spin effects. The nine parameters MH24 model have been optimized using state-of-the-art super-self-consistent-field approach, which exploits sensitivity specific properties, such core ionization potentials, electron affinities, atomization energies, in regions real space LMF into core, valence, asymptotic part. functionals are shown be able simultaneously provide good accuracy valence properties well affinities noble gas dimer dissociation curves, while satisfying multiple known constraints related functionals. thus major step toward more sophisticated models.

Язык: Английский

Data-Driven Improvement of Local Hybrid Functionals: Neural-Network-Based Local Mixing Functions and Power-Series Correlation Functionals DOI Creative Commons
Artur Wodyński,

Kilian Glodny,

Martin Kaupp

и другие.

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Янв. 13, 2025

Local hybrid functionals (LHs) use a real-space position-dependent admixture of exact exchange (EXX), governed by local mixing function (LMF). The systematic construction LMFs has been hampered over the years lack physical constraints on their valence behavior. Here, we exploit data-driven approach and train new type "n-LMF" as relatively shallow neural network. input features are meta-GGA character, while W4-17 atomization-energy BH76 reaction-barrier test sets have used for training. Simply replacing widely "t-LMF" LH20t functional n-LMF provides LH24n-B95 functional. Augmented DFT-D4 dispersion corrections, LH24n-B95-D4 remarkably improves WTMAD-2 value large GMTKN55 suite general main-group thermochemistry, kinetics, noncovalent interactions (NCIs) from 4.55 to 3.49 kcal/mol. As found limited flexibility B95c correlation disfavor much further improvement NCIs, proceeded replace it an optimized B97c-type power-series expansion. This gives LH24n LH24n-D4 3.10 kcal/mol, so far lowest rung 4 in self-consistent calculations. perform moderately well organometallic transition-metal energetics leaving room improvements that area. Compared complete neural-network like DM21, present more tailored just LMF flexible but well-defined human-designed LH retains possibility graphical analyses gain deeper understanding. We find both recent x-LMF suppress so-called gauge problem hybrids without adding calibration required other LMFs. plots show this can be traced back values small-density region between interacting atoms NCIs n- x-LMFs low t-LMF. also trained covalent bonds deteriorating binding energies. current enables fast efficient routine calculations using n-LMFs Turbomole.

Язык: Английский

Процитировано

1
An evaluation of local double hybrid density functionals DOI Creative Commons

Nóra Kovács,

Martin Kaupp, Artur Wodyński

и другие.

Chemical Physics Letters, Год журнала: 2025, Номер unknown, С. 142048 - 142048

Опубликована: Март 1, 2025

Язык: Английский

Процитировано

0
Combining real-space and local range separation—The MH24 locally range-separated local hybrid functional DOI Creative Commons
Toni M. Maier

The Journal of Chemical Physics, Год журнала: 2024, Номер 161(21)

Опубликована: Дек. 3, 2024

In this work, the development of a new general-purpose exchange–correlation hybrid functional based on recent locally range-separated local approach is presented. particular, functional, denoted as MH24, combines non-empirical treatment admixture long-range exact exchange with real-space separation for exact-exchange governed by mixing function (LMF) and empirical LYP-based correlation to enable flexible description same- opposite-spin effects. The nine parameters MH24 model have been optimized using state-of-the-art super-self-consistent-field approach, which exploits sensitivity specific properties, such core ionization potentials, electron affinities, atomization energies, in regions real space LMF into core, valence, asymptotic part. functionals are shown be able simultaneously provide good accuracy valence properties well affinities noble gas dimer dissociation curves, while satisfying multiple known constraints related functionals. thus major step toward more sophisticated models.

Язык: Английский

Процитировано

1