Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2011.14715 · Journal of Chemical Theory and Computation

Machine Learning-Assisted Selection of Active Spaces for Strongly Correlated Transition Metal Systems

Pavlo Golub, Andrej Antalík, Libor Veis, Jiří Brabec

Published Sep 27, 2021• 44 citations
arXivHTMLPDF

Abstract

Active space quantum chemical methods could provide very accurate description of strongly correlated electronic systems, which is of tremendous value for natural sciences. The proper choice of the active space is crucial but a nontrivial task. In this article, we present a neural network-based approach for automatic selection of active spaces, focused on transition metal systems. The training set has been formed from artificial systems composed of one transition metal and various ligands, on which we have performed the density matrix renormalization group and calculated the single-site entropy. On the selected set of systems, ranging from small benchmark molecules up to larger challenging systems involving two metallic centers, we demonstrate that our machine learning models could predict the active space orbitals with reasonable accuracy. We also tested the transferability on out-of-the-model systems, including bimetallic complexes and complexes with ligands, which were not involved in the training set. Also, we tested the correctness of the automatically selected active spaces on a Fe(II)-porphyrin model, where we studied the lowest states at the DMRG level and compared the energy difference between spin states or the energy difference between conformations of ferrocene with recent studies.

DOI
10.1021/acs.jctc.1c00235

Related papers

SC1MC-2020: A database of transition metal complexes for training ML models to predict one-and two-site entropies

Pavlo Golub, Pavel Beran, Libor Veis +1 more

We introduce a new database SC1MC-2020 of artificial mono transition metal complexes involving reference data (HF energy, orbitals, and integrals), and one- and two-site entropies obtained from corresponding DMRG calculations. The purpose of this database is to include all important metal - ligand interactions of strongly correlated systems and obtain as much as possible information about the electronic correlation structure, which could be exploited by machine learning models to estimate these important information almost immediately for any similar systems with high degree of transferability.

•Jan 15, 2021•2 citations

Density Matrix Renormalization Group with Dynamical Correlation via Adiabatic Connection

Pavel Beran, Mikuláš Matoušek, Michał Hapka +2 more

The quantum chemical version of the density matrix renormalization group (DMRG) method has established itself as one of the methods of choice for calculations of strongly correlated molecular systems. Despite its great ability to capture strong electronic correlation in large active spaces, it is not suitable for computations of dynamical electron correlation. In this work, we present a new approach to the electronic structure problem of strongly correlated molecules, in which DMRG is responsible for a proper description of the strong correlation, whereas dynamical correlation is computed via the recently developed adiabatic connection (AC) technique which requires only up to two-body active space reduced density matrices. We report the encouraging results of this approach on typical candidates for DMRG computations, namely, n -acenes ( n = 2 → 7), Fe(II)–porphyrin, and the Fe 3 S 4 cluster.

arXiv:2108.12803•Journal of Chemical Theory and Computation•Nov 11, 2021•49 citations

SC1MC-2022: A database of transition metal complexes for training ML models to predict one-site entropies and mutual information

Pavlo Golub, Pavel Beran, Andrej Antalík +1 more

We introduce a new version of the database SC1MC (SC1MC-2022), obtained by extension of the recent SC1MC-2020, which includes artificial mono transition metal complexes. The database involves reference data used as inputs for training of machine learning models, one- and two-site entropies, and mutual information obtained at the DMRG level for canonical and split-localised orbitals. The purpose of this database is to obtain as much as possible information about the electronic correlation structure, which could be exploited by machine learning models to estimate these important information without a significant computational cost for any similar type of systems with some degree of transferability.

arXiv:2101.06090•Jan 15, 2021•3 citations

Fast and Accurate Excitation Energies from Density Matrix Renormalization Group Calculations Improved by Machine Learning

Pavlo Golub, Libor Veis

An estimation of optoelectronic properties is a crucial necessity and an ongoing challenge in modern functional material science. This involves accurate accessing of ground and excited states in systems with complex electronic structure that are often too computationally expensive for accurate many-body methods. Building on our previous work on ground states [J. Phys. Chem. Lett. 2025, 16, 3295-3301], we present an efficient and cost-effective approach for evaluating electronic excitations in functional materials, combining the density matrix renormalization group method as a complete active space solver with machine learning techniques. We demonstrate its performance on π-electron-correlated systems, namely polycyclic aromatic compounds. The transferability and effectiveness of the derived machine learning model are demonstrated on a number of challenging examples with up to 34 π-electrons.

arXiv:2609.12616•Sep 11, 2026•0 citations

RLEASE: Reinforcement Learning Efficient Active Space Engine

Etinosa Osaro, Abhishek Mitra, Andrew J. Jenkins +6 more

Selecting the active space for multireference electronic-structure calculations is a long-standing bottleneck that often requires expert chemical intuition and costly trial-and-error. We introduce RLEASE (Reinforcement Learning Efficient Active Space Engine), a low-cost method for automatic, geometry-dependent active-space selection. A neural network predicts per-orbital diagnostic scores ($\hat{s}_{1}$) from inexpensive Hartree-Fock orbital descriptors, and a learned threshold partitions orbitals into active and inactive sets. The threshold policy is optimized with proximal policy optimization, using the discrepancy between sc-NEVPT2 energies computed with the selected active space and DMRG reference energies as the reward. After training, the same RLEASE-selected active spaces can be used with multireference perturbation theory or composite coupled-cluster energy estimators. Despite being trained on a small set of molecules and geometries, RLEASE transfers to chemically diverse test systems, producing compact active spaces and competitive potential-energy surfaces relative to established entropy-based selectors. Because deployment requires only inexpensive orbital descriptors and neural-network inference, RLEASE enables high-throughput multireference workflows without molecule-specific retraining or target-system pilot DMRG calculations.

arXiv:2606.07879•Jun 5, 2026•1 citation

TNASS: Tensor Network Active Space Selection with the Entanglement Feature

Angus Mingare, Isabelle Heuzé, Peter V. Coveney

The quality of multi-scale modelling techniques in molecular electronic structure calculations, such as embedding and subspace methods, relies upon the chosen active space. The automation of active space selection is vital for ensuring the accuracy, reproducibility, and scalability in such calculations. In this work, we introduce Tensor Network Active Space Selection using the Entanglement Feature. Through the isolation of strongly correlated electrons, this method provides a scalable foundation for embedding methods in multi-scale modelling. By representing the purities of all possible orbital partitions as a Matrix Product State, our method isolates regions of strong electron correlation without requiring manual preselection of target atoms or the calculation of expensive high-order density matrices. The results demonstrate that this approach leads to lower ground state energies and more accurate dipole moments than other fully automated selection schemes such as those based solely on single-orbital entropy or the selection of spatial orbitals around the HOMO/LUMO gap.

physics.chem-phCross-listedarXiv:2608.03645•Aug 5, 2026•0 citations
Active space quantum chemical methods could provide very accurate description of strongly correlated electronic systems, which is of tremendous value for natural sciences. The proper choice of the active space is crucial but a nontrivial task. In this article, we present a neural network-based approach for automatic selection of active spaces, focused on transition metal systems. The training set has been formed from artificial systems composed of one transition metal and various ligands, on which we have performed the density matrix renormalization group and calculated the single-site entropy. On the selected set of systems, ranging from small benchmark molecules up to larger challenging systems involving two metallic centers, we demonstrate that our machine learning models could predict the active space orbitals with reasonable accuracy. We also tested the transferability on out-of-the-model systems, including bimetallic complexes and complexes with ligands, which were not involved in the training set. Also, we tested the correctness of the automatically selected active spaces on a Fe(II)-porphyrin model, where we studied the lowest states at the DMRG level and compared the energy difference between spin states or the energy difference between conformations of ferrocene with recent studies.
We introduce a new database SC1MC-2020 of artificial mono transition metal complexes involving reference data (HF energy, orbitals, and integrals), and one- and two-site entropies obtained from corresponding DMRG calculations. The purpose of this database is to include all important metal - ligand interactions of strongly correlated systems and obtain as much as possible information about the electronic correlation structure, which could be exploited by machine learning models to estimate these important information almost immediately for any similar systems with high degree of transferability.
The quantum chemical version of the density matrix renormalization group (DMRG) method has established itself as one of the methods of choice for calculations of strongly correlated molecular systems. Despite its great ability to capture strong electronic correlation in large active spaces, it is not suitable for computations of dynamical electron correlation. In this work, we present a new approach to the electronic structure problem of strongly correlated molecules, in which DMRG is responsible for a proper description of the strong correlation, whereas dynamical correlation is computed via the recently developed adiabatic connection (AC) technique which requires only up to two-body active space reduced density matrices. We report the encouraging results of this approach on typical candidates for DMRG computations, namely, n -acenes ( n = 2 → 7), Fe(II)–porphyrin, and the Fe 3 S 4 cluster.
We introduce a new version of the database SC1MC (SC1MC-2022), obtained by extension of the recent SC1MC-2020, which includes artificial mono transition metal complexes. The database involves reference data used as inputs for training of machine learning models, one- and two-site entropies, and mutual information obtained at the DMRG level for canonical and split-localised orbitals. The purpose of this database is to obtain as much as possible information about the electronic correlation structure, which could be exploited by machine learning models to estimate these important information without a significant computational cost for any similar type of systems with some degree of transferability.
An estimation of optoelectronic properties is a crucial necessity and an ongoing challenge in modern functional material science. This involves accurate accessing of ground and excited states in systems with complex electronic structure that are often too computationally expensive for accurate many-body methods. Building on our previous work on ground states [J. Phys. Chem. Lett. 2025, 16, 3295-3301], we present an efficient and cost-effective approach for evaluating electronic excitations in functional materials, combining the density matrix renormalization group method as a complete active space solver with machine learning techniques. We demonstrate its performance on π-electron-correlated systems, namely polycyclic aromatic compounds. The transferability and effectiveness of the derived machine learning model are demonstrated on a number of challenging examples with up to 34 π-electrons.
Selecting the active space for multireference electronic-structure calculations is a long-standing bottleneck that often requires expert chemical intuition and costly trial-and-error. We introduce RLEASE (Reinforcement Learning Efficient Active Space Engine), a low-cost method for automatic, geometry-dependent active-space selection. A neural network predicts per-orbital diagnostic scores (s^1\hat{s}_{1}s^1​) from inexpensive Hartree-Fock orbital descriptors, and a learned threshold partitions orbitals into active and inactive sets. The threshold policy is optimized with proximal policy optimization, using the discrepancy between sc-NEVPT2 energies computed with the selected active space and DMRG reference energies as the reward. After training, the same RLEASE-selected active spaces can be used with multireference perturbation theory or composite coupled-cluster energy estimators. Despite being trained on a small set of molecules and geometries, RLEASE transfers to chemically diverse test systems, producing compact active spaces and competitive potential-energy surfaces relative to established entropy-based selectors. Because deployment requires only inexpensive orbital descriptors and neural-network inference, RLEASE enables high-throughput multireference workflows without molecule-specific retraining or target-system pilot DMRG calculations.
The quality of multi-scale modelling techniques in molecular electronic structure calculations, such as embedding and subspace methods, relies upon the chosen active space. The automation of active space selection is vital for ensuring the accuracy, reproducibility, and scalability in such calculations. In this work, we introduce Tensor Network Active Space Selection using the Entanglement Feature. Through the isolation of strongly correlated electrons, this method provides a scalable foundation for embedding methods in multi-scale modelling. By representing the purities of all possible orbital partitions as a Matrix Product State, our method isolates regions of strong electron correlation without requiring manual preselection of target atoms or the calculation of expensive high-order density matrices. The results demonstrate that this approach leads to lower ground state energies and more accurate dipole moments than other fully automated selection schemes such as those based solely on single-orbital entropy or the selection of spatial orbitals around the HOMO/LUMO gap.