Exploring speciation and reactivity with machine learning and molecular dynamics

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. The xTB package was used to conduct molecular dynamics simulations with explicit solvation. While a BOMD approach using DFT is our preferred approach for molecular dynamics in gas phase, solvation can only be accounted for implicitly. Involving explicit solvent molecules would require simulation times of hundreds of picoseconds to observe solvent motion and reactive events. Although computationally affordable for up to 5 ps (up to 80 atoms), BOMD becomes unfeasible for longer simulation times. However, the xTB package utilizes a rapid semi-empirical method that is capable to dynamically study systems up to 1000 atoms. Therefore, using xTB gives the unique opportunity to study mechanisms dynamically in solution at reasonable accuracy, even for larger systems.

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 We previously investigated the mechanism of nucleophilic aromatic substitution of different arenes, nucleophiles and leaving groups. This study primarily focused on determining whether the reaction occurs via a concerted or a step-wise process for a given set of reactants and suggested that the lifetime on the observed intermediate of the step-wise process is frequently short and comparable to that of a transition state. In addition to a detailed evaluation of the potential energy pathways by means of static DFT, dynamic aspects were studied using xTB with explicit solvation. For further reliability and plausibility of the determined lifetimes from xTB dynamics, we also conducted Born-Oppenheimer molecular dynamics (BOMD) at the DFT level. For each reaction profile 100 trajectory calculations were run in order to be able to determine statistically relevant lifetimes of the intermediate. These MD calculations at the DFT level qualitatively confirmed our results obtained at the xTB level and indicated lifetimes in the same order of magnitude. No boundary between step-wise and concerted cases was found and instead a continuous increase of intermediate lifetimes was observed.

(1b) Stepwise vs concerted mechanism in nucleophilic aromatic substitution In contrast to traditional Brønsted acid-catalyzed Friedel Crafts reactions, which typically proceed with low regioselectivity, we had observed high selectivities under reaction conditions that furnish highly reactive adamantyl triflate in situ. We explored the mechanism by means of DFT, which confirmed that the observed preference for adamantylation in para-position is thermodynamically motivated. Dynamics calculations were then utilized to study the proposed reactive intermediate, i.e. the in situ formed adamantyl triflate. To this end the starting materials were surrounded by 52 dichloromethane molecules and constrained within a sphere. Pre-equilibration was performed using GFN2-xTB, where the solute was fixed and only the explicit solvent was optimized. From the obtained solvated starting materials 50 trajectories were then computed using GFN2-xTB. These confirmed that already at room temperature the adamantyl triflate bond is cleaved and the adamantyl cation is liberated as a reactive species. Further reaction of the adamantyl cation with the arene could not be observed within the available simulation times, showcasing the extended lifetimes of the cation under the explored reaction conditions.

(2a) Nickel^(I) catalysts for CO² insertion In this reporting period we completed our in silico dataset of bidentate phosphine ligands, and their corresponding Ni^(II) and Ni^(I) complexes. Utilizing an unsupervised machine learning workflow, we aimed to identify stable Ni^(I) complexes that can undergo CO2 insertion at room temperature. Two ligands selected as positive and negative reference points based on literature reports regarding the stability of their Ni^(I) complexes. For feature selection we chose those descriptors that led to the biggest separation of the reference ligands. K-means clustering then was used to group the entire dataset according to the selected features and identified one cluster of 21 ligands that were grouped together with the positive reference ligand. These candidates were then additionally vetted in terms of their activation barriers for CO² activation. This was necessary since our machine learning workflow was dedicated to identifying stable Ni^(I) complexes, but does not distinguish the corresponding reactivity. We hence searched for transition states for CO² insertion (both inner-sphere and outer-sphere mechanism) at the Ni^(I)-Ph complexes for each of the identified 21 ligands, and evaluated the corresponding activation barriers. The most promising candidates were then tested experimentally and showed catalytic activity towards CO² activation at room temperature, furnishing the corresponding benzoic acids after acid quenching.

(2b) Nickel(I) dimers Continuing our efforts to develop a general database for nickel speciation we have started to build a ligand library for NHC-ligands in the last reporting period. While the ultimate aim is to utilize this extended database for the identification of novel Ni^(I) dimers, a secondary focus is to establish a general and extensive library for NHC ligands, which as opposed to phosphine ligands currently does not exist and would be of interest for data-driven approaches more generally. So far, we have included a range of diverse NHC ligands with focus on those that are likely synthesizable. For this we have conducted a detailed literature search to include NHC ligands that have been synthesized and have then strategically modified these structures with different substituents that would still allow for an analogous synthesis compared to the known parent ligand.

Discussion

For subproject 1 our findings reaffirmed the observation of short intermediate lifetimes. 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 subproject 2 we applied an unsupervised ML workflow to identify Nickel(I) complexes for CO² insertion and built databases for nickel speciation using different ligands.

Additional Project Information

DFG classification: 301-02 Organic Molecular Chemistry
Software: CREST, Gaussian16
Cluster: CLAIX

Publications

Thesis:

Avetik Gevondian, ‘Site-selective, orthogonal functionalization of organogermanes, and “batch-forbidden” transformations enabled in cyclic flow mode’, Dr. rer. nat., 2024

Marvin Mendel, ‘Modular Pd(I) cross-coupling strategies and original Ni(I) metalloradical catalysis’, Dr. rer. nat., 2024