Project
Exploring speciation and reactivity with machine learning and molecular dynamics
Computational methods have proven powerful for investigating chemical reactivity. Typically, reaction intermediates and transition states are considered in a static fashion, primarily using DFT calculations, whereas the dynamic aspects of a reaction are frequently neglected. Our initial investigations on the dynamics of aromatic substitutions provided insights that contradict the static understanding of these reactions and highlight the importance of molecular dynamics (MD). Within this project, we want to address the dynamic aspects of chemical reactions to uncover new reactivity. Despite the success of theory-driven methods (e.g. DFT, MD) in understanding reactivity, some problems remain challenging to explain, such as speciation in organometallic complexes. With our prior work on Pd(I) dimers we were able to show that data-driven approaches such as machine learning can be useful to address these problems. This project aims to extend our approach to Ni catalysis and identify new ligands capable of stabilizing Ni(I).
Project Details
Project term
March 11, 2025–March 11, 2026
Affiliations
RWTH Aachen University
Institute
Institute of Organic Chemistry
Principal Investigator
Methods
A combination of different quantum chemical software was employed. Gaussian was used for the optimization of stationary points such as intermediates and transition states as well as for Born-Oppenheimer molecular dynamics (BOMD) at the DFT level as a microcanonical ensemble. Involving explicit solvent molecules requires longer simulation times to observe solvent motion and reactive events. While this is not affordable at DFT level, the semi-empirical xTB method is much faster, allowing for molecular dynamics simulations with explicit solvation. To resemble experimental conditions more closely, BOMDs were also run as a canonical ensemble, which is available in the ORCA software (and xTB).
Results
Subproject 1 is focused on dynamic aspects of reaction mechanisms, while Subproject 2 aims at utilizing machine learning to solve speciation problems of nickel catalysts. (1a) Stepwise vs concerted mechanism in nucleophilic aromatic substitution (SNAr) This sub-project focusses on two main aspects: (i) exploring the dynamic aspects (intermediate lifetimes) of the competing concerted and stepwise SNAr reaction pathways and (ii) exploring any underlying trends between structure and intermediate lifetimes. In this reporting period further energy profiles for different combinations of nucleophile, arene and leaving group have been computed. The lifetimes of the found intermediates were then studied by molecular dynamics. Initial lifetimes were obtained from xTB MDs and at the DFT level from BOMDs using a microcanonical ensemble. Although these lifetimes agree on a qualitative level, further evaluation with experimentally more relevant canonical BOMDs is ongoing. An exploratory data analysis was conducted showing no correlation between lifetimes and either activation barrier, electron affinity or Hammett parameters using the preliminary lifetimes as obtained from microcanonical BOMDs. However, further descriptors need to be evaluated and lifetimes updated to reflect the experimentally more relevant canonical ensemble.
(1b) Stepwise vs concerted C-H activation by Co(I) complexes In our ongoing investigation on the mechanism of low valent Co-catalyzed C-H activation, we have explored further NHC ligands. Although more sterically encumbered ligands were found to reduce the barrier for the alternative concerted ligand to ligand H-transfer (LLHT), a two-step one-electron process was still favored. Additionally, we have finalized the necessary method benchmarking on for the open shell system (known to be significantly affected by the utilized DFT functional, especially the amount of HF-exchange), further supporting a two-step, one-electron pathway.
(2a) Ni(I) catalysts for small molecule activation Based on our previous study on CO2 activation by Ni(I), we have conducted a DFT study on potential activation pathways of the greenhouse gas N2O. Different mechanistic pathways were explored and a two-step mechanism via O-atom transfer and subsequent 1,2-shift was found to be favored. As for CO2 activation Ni(I) was found to be more reactive than the corresponding Ni(II) complexes. Despite the similarity to CO2 (isoelectronic), different ligands are reactive for N2O, which is reflected in our computational study that found lower activation barriers for bipyridine ligands. In this context, the O-atom transfer was found rate-limiting for the majority of the explored ligands. However, since this step formally constitutes an electron-transfer, further exploration of open shell variants of this transformation is ongoing before proceeding with the evaluation of the most promising ligand class by our established unsupervised ML workflow.
(2b) Ni(I) NHC dimers During this reporting period, we completed the computational generation of a comprehensive database comprising 450 N heterocyclic carbene (NHC) ligands. For each ligand, different ligation and oxidation states were evaluated, resulting in more than 5,500 distinct molecular structures. From the cleaned and curated Gaussian output files, 235 physicochemical descriptors encompassing steric, electronic, and reactivity-related properties were extracted. The resulting dataset was subjected to our ML workflow, including feature selection, Pearson correlation analysis and threshold-based filtering, followed by k-means clustering. Based on the identified clusters, chiral NHC ligands located within the same cluster as known positive references were selected as promising candidates for Ni(I) dimer formation and are currently evaluated experimentally.
Discussion
For sub-project 1 we could confirm the short intermediate lifetimes, but further trajectories including a thermostat need to be run in order to obtain reliable data for further evaluation. Notably, this phenomenon was found to be disconnected from the free energy profile obtained by static DFT, therefore highlighting the importance of molecular dynamics. In sub-project 2 we have identified stable Ni(I) complexes, that can undergo CO2 insertion at room temperature utilizing an unsupervised ML workflow paired with DFT computational selection based on activation barriers. This work inspired our current study on the activation of N2O, another greenhouse gas, isoelectronic to CO2.
Additional Project Information
DFG classification: 301-02 Organic Molecular Chemistry
Software: Gaussian16, ORCA, CREST
Cluster: CLAIX
Publications
Julian A. Hueffel, Quentin P. Bindschaedler, Francesco Sala,
Franziska Schoenebeck,
Empowering Reactivity Predictions through Noise-Based Data Augmentation,
https://dx.doi.org/10.1021/jacs.5c11632, Sptember 2025
Julian A. Hueffel, Mathilde Rigoulet, Sebastian Wellig, Theresa Sperger, Jas S.Ward, Kari Rissanen, Franziska Schoenebeck,
Discovery of Ni(I) Complexes for CO2 Insertion Enabled by a Machine Learning-Computational-Selection Sequence,
https://dx.doi.org/10.1021/jacs.5c00441, July 2025
Eric Ahrweiler, Aymane Selmani, Franziska Schoenebeck,
Base‐Catalyzed Remote Hydrogermylation of Olefins,
https://dx.doi.org/10.1002/anie.202503573, March 2025
Thesis:
Zivkovic, Filip G.
Fluorinated motifs as structural modifications of amides and related N-carbonyl derivatives,
PhD Thesis, 2026