The bonding in many compounds containing heavy atoms is still unclear, especially because of relativistic effects and in particular the spin-orbit coupling (SOC). The latter complicates the possibility of imagining a bridge between the complex electronic structure and simple concepts, such as covalent and ionic bonds, single and multiple bonds, VSEPR structure, etc. In this perspective, we are developing various tools to analyze SOC effects on the chemical bond, either in real space (ELF, QTAIM, conceptual DFT), or in Hilbert space (EBO), using multi-configurational wave functions at the spin-orbit configuration interaction level (SOCI) or the formalism of two-component relativistic DFT calculations.

ELF topological analysis of a gold-chlore bond.

Reference publications:

C. Gomez Pech et al. J. Comput. Chem. 2020, 41, 2055-2065. S. Sarr, J. Graton, G. Montavon, J. Pilmé, N. Galland ChemPhysChem 2020, 21, 240-250.

The team is the leading developer with their collaborators in Toulouse of the so-called QUEST database, the largest dataset of ultra-accurate reference values for excited-state calculations. As the present stage this database includes 1489 aug-cc-pVTZ reference transition energies (731 singlets, 233 doublets, 461 triplets, and 64 quartets) for both valence and Rydberg transitions occurring in molecules containing from 1 to 16 non-hydrogen atoms, as well as accurate data for transition dipoles and excited-state dipoles in a subset of coumpounds. Notably, QUEST includes a significant list of reference values for states characterized by a partial or genuine double-excitation character, known to be particularly challenging for many computational methods. The vast majority of the included values are deemed chemically accurate. This has allowed the team to perform numerous benchmarks of « lower-order » methods.

Reference publications:

A. Chrayteh, A. Blondel, P. F. Loos and D. Jacquemin J. Chem. Theory Comput. 2021, 17, 416-438. P. F. Loos, F. Lipparini, D. A. Matthews, A. Blondel, D. Jacquemin J. Chem. Theory Comput. 2022, 18, 4418-4427. F. Kossoski, M. Boggio-Pasqua, P. F. Loos, D. Jacquemin J. Chem. Theory Comput. 2024, 20, 5655-5678. I. Knysh, F. Lipparini, A. Blondel, I. Duchemin, X. Blase, P. F. Loos, D. Jacquemin J. Chem. Theory Comput. 2024, 20, 8152-8174. P. F. Loos, M. Boggio-Pasqua, A. Blondel, F. Lipparini, D. Jacquemin J. Chem. Theory Comput. 2025, 21, 8010-8033

The team has been working on modelling excited-state absorption (ESA) with TD-DFT. Two directions have been pursued: i) understanding the shortcomings of linear-response schemes for ESA oscillator strengths and finding a criterion to identify a piori the problematic cases; ii) developing a new solvent model adequate for ESA and computationally efficient.

Reference publications:

J. Sirucek, B. Le Guennic, L. Cupellini, D. Jacquemin, J. Chem. Theory Comput., 2026, 22, 502-512. J. Sirucek, B. Le Guennic, D. Jacquemin, B. Mennucci, L. Cupellini, J. Chem. Theory Comput., 2026, 22, 3585-3595

The team has carried out several methodological studies on various aspects of non-adiabatic dynamics simulations. This work contributes to a better understanding of the approximations underlying non-adiabatic dynamics calculations, such as the impact of the level of electronic structure theory used to calculate potential energy surfaces, or in relation to variants of the widely used surface-hopping method. The team has also developed an automated and unbiased protocol for nuclear dimension reduction.

Reference publications:

V. Delmas, A. N. Nardi, I. C. D. Merritt, A. Ferté, I. Fdez. Galván and M. Vacher, J. Chem. Theory Comput. 2025, 21, 6611-6621. T. V. Papineau, D. Jacquemin and M. Vacher, J. Phys. Chem. Lett. 2024, 15, 636-643. I. C. D. Merritt, D. Jacquemin and M. Vacher, J. Chem. Theory Comput. 2023, 19, 6, 1827-1842.

The team has also trained machine learning (ML) algorithm in order to predict the reaction yield of a typical organic reaction. Automatized high-throughput experiments (HTE) has demonstrated the capacity to screen a tremendous amount of reactions in a small amount of time.  To avoid the need to carry out entire reactions screening to reach an efficient model, the use of Active Learning (AL) can help reducing the number of experiments required. Indeed, AL corresponds to a sub-part of supervised ML which uses a loop mechanism using a stopping criterion. We tested one stopping criterion (Stabilization Prediction) on the two chemical reactions datasets, namely the B-H and the Suzuki datasets and evaluate the impact of four main factors: tigated: (i) the number of queries per iteration, (ii) the size of the stop set, (iii) three types of descriptors and (iv) four types of ML estimators as variables to evaluate the stability of the SP method. All development are open to the community through the public gitlab of the team.In the context of a ‘real-time’ AL, we have developed an CEISAM web interface in order to communicate and share information with organic chemists in order to create a model starting from a tiny set of reaction (< 200) for a particular reaction of common interest.

Reference publications

• V. Delmas, D. Jacquemin, A. Blondel, M. Vacher, A. D. Laurent, React. Chem. Eng. 2024, 9, 1206-1215.