Skip to content

Machine learning / Advancing electronic structure

In our research, we are dedicated to advancing electronic structure and machine learning (ML) techniques. We concentrate on two key areas: electronic structure beyond density functional theory and functional properties. Our aim is to bring innovation to these fields by leveraging the power of advanced algorithms and ML methodologies.

Functional Properties from Machine Learning:

Our primary focus lies in the development of ML approaches for speed up of simulations and tailored for predicting functional properties such as energy, force, and more. We have introduced a novel ML force field based on molecular dynamic trajectories, designed to expedite ab initio molecular dynamics (AIMD) simulations. To predict correlation energy at post-Hartree–Fock level, we have created energy density-based descriptors. Notably, our interest extends to feature engineering, where we uncover hidden chemical system information, steering clear of the “black-box” approach.

We also harness ML algorithms to enhance biasing in advanced sampling methods, such as metadynamics. Our ongoing exploration revolves around formulating collective variables using deep learning (DL) techniques for complex chemical systems. To enhance the DL model’s learning capacity, we concurrently develop novel structural representations adaptable to diverse chemical systems. Moreover, we have developed an in-house software suite, collaboratively used with experimentalists, to address real-world chemistry challenges.

Highly accurate electronic structure:

We are interested in highly accurate electronic structure methodologies to tackle demanding systems with multireference character such as transition metal catalysts. Work has included the development of approaches to calculate entire reaction pathways using complete active space methodologies and to study magnetic properties of involved transition metal systems using a variety of methods ranging from Quantum Monte Carlo up to Density Matrix Renormalization Group approaches. Another direction has focused on wavefunction-in-DFT embedding for periodic and nonperiodic systems.

L. Schreder, S. Luber
Propagated (fragment) Pipek–Mezey Wannier functions in real-time time-dependent density functional theory
J. Chem. Phys. 2024, 160, 21, 214117

R. Han, S. Luber, G. L. Manni
Magnetic Interactions in a [Co(II)3Er(III)(OR)4] Model Cubane through Forefront Multiconfigurational Methods
J. Chem. Theory Comput. 2023, 19, 10, 2811–2826

R. Han, J. Mattiat, S. Luber
Automatic purpose-driven basis set truncation for time-dependent Hartree–Fock and density-functional theory
Nat. Commun. 2023, 14, 106

R. Ketkaew, F. Creazzo, S. Luber
Machine Learning-Assisted Discovery of Hidden States in Expanded Free Energy Space
J. Phys. Chem. Lett. 2022, 13, 7, 1797–1805

R. Han, R. Ketkaew, S. Luber
A Concise Review on Recent Developments of Machine Learning for the Prediction of Vibrational Spectra
J. Phys. Chem. A 2022, 126, 6, 801-812

R. Han, S. Luber
Fast Estimation of Møller–Plesset Correlation Energies Based on Atomic Contributions
J. Phys. Chem. Lett., 2021, 12, 22, 5324-5331

R. Han, M.Rodríguez-Mayorga, S. Luber
A Machine Learning Approach for MP2 Correlation Energies and Its Application to Organic Compounds
J. Chem. Theory Comput., 2021, 17, 2, 777–790

R. Han, S. Luber
Trajectory-based machine learning method and its application to molecular dynamics
Molecular Physics, 2020, 118:19-20

R. Han, S. Luber
Complete active space analysis of a reaction pathway: Investigation of the oxygen–oxygen bond formation
J. Comput. Chem., 2020, 41, 1586-1597