Winter school covering advanced topics in electronic structure theory.

Porous materials serve diverse applications like molecular separations and catalysis, offering energy-efficient methods for capturing greenhouse gases and valuable noble gases including xenon. Despite its importance in medicine and nuclear processes, xenon extraction remains challenging due to its low atmospheric concentration (0.087 ppm) and inertness. In this work, we present two complementary machine learning (ML) models: (1) an ML interatomic potential (MLIP) obtained by training E3-equivariant graph neural network (GNN), Allegro, on a PBE-D4 theory-level dataset; and (2) an NMR-ML model obtained by training the invariant GNN, SchNet, on a PBE-(SVP,TZVP)/BHandHLYP-SVP theory-level dataset. The MLIP model enables accurate, data-efficient, and transferable large-scale molecular dynamics simulations and the NMR-ML model enables predicting Xe NMR chemical shifts, providing microscopic interpretations of experimental ¹²⁹Xe NMR observations.

Porous organic cages (POCs) in porous liquids (PLs) offer promising platforms for selective Xe capture. Understanding Xe binding and dynamics in these systems is crucial for material design. While DFT-based molecular dynamics (MD) accurately models these interactions, it is computationally prohibitive for large systems. Machine learning interatomic potentials (MLIPs), particularly neural networks like Allegro, provide a scalable alternative. We present an MLIP trained on DFT-D4 theory-level data for PLs and POCs with Xe, covering over 1.8 million atoms across varied structures. Furthermore, we train an invariant SchNet model to predict ¹²⁹Xe NMR magnetic shielding tensors, offering microscopic insights into static NMR spectra and dynamic relaxation behaviors.

Porous organic cages (POCs) in porous liquids (PLs) offer promising platforms for selective Xe capture. Understanding Xe binding and dynamics in these systems is crucial for material design. While DFT-based molecular dynamics (MD) accurately models these interactions, it is computationally prohibitive for large systems. Machine learning interatomic potentials (MLIPs), particularly neural networks like Allegro, provide a scalable alternative. We present an MLIP trained on DFT-D4 theory-level data for PLs and POCs with Xe, covering over 1.8 million atoms across varied structures. Furthermore, we train an invariant SchNet model to predict ¹²⁹Xe NMR magnetic shielding tensors, offering microscopic insights into static NMR spectra and dynamic relaxation behaviors.

Porous materials serve diverse applications like molecular separations and catalysis, offering an energy-efficient method for capturing greenhouse gases (CO₂ and CH₄) and valuable noble gases (Xe, Ar, and Kr). Xenon — vital in optics, medicine, and nuclear fission processes — poses extraction challenges due to its low atmospheric abundance (0.087 ppm by volume) and inertness, driving high commercial costs. The recent development of porous liquids (PLs) with cavities formed by porous organic cages (POCs) has shown promise in addressing these challenges. Here, we present an MLIP model constructed using Allegro, trained on DFT-D4 level energies, forces, and virials of structures comprising 600 to 1170 atoms. The dataset encompasses 1.8 million atoms with 12.5 million data points. The MLIP will be applied to provide microscopic interpretation of experimental ¹²⁹Xe NMR observations via an ML model for magnetic shielding tensors.

Porous materials serve diverse applications like molecular separations and catalysis, offering an energy-efficient method for capturing greenhouse gases and valuable noble gases. Here, we present an MLIP model constructed using the local equivariant deep NN architecture, Allegro. This model was trained on energies, forces, and virials computed at the DFT-D4 level with PBE functional, from a dataset comprising over 1600 structures of 600 to 1170 atoms (H, C, N, O, F, Cl, and Xe). The dataset encompasses 1.8 million atoms with 12.5 million data points. The MLIP will be applied to provide microscopic interpretation of experimental ¹²⁹Xe NMR observations via an ML shielding tensor model trained on the same dataset.

Porous materials serve diverse applications like molecular separations and catalysis, offering an energy-efficient method for capturing greenhouse gases and valuable noble gases. Here, we present an MLIP model constructed using Allegro, trained on energies, forces, and virials at the DFT-D4 level with PBE functional. The dataset comprises over 1600 structures of 600 to 1170 atoms, encompassing 1.8 million atoms and 12.5 million data points covering TBA-type, HAP-type, and DCT-type PLs as well as CC3 POCs with Xe atoms inside or near the cavities. The MLIP will be paired with an ML model for magnetic shielding tensors to provide microscopic interpretation of experimental ¹²⁹Xe NMR observations at both static spectral and dynamic relaxation levels.

Recently, Machine Learning (ML) methods have penetrated almost all research areas in materials modelling and high-throughput materials screening. The ML triumph has so far mainly focused on developing surrogate models for the potential energy surface with superior computational efficiency while retaining first principles accuracy. The approach to learn observable properties directly is just emerging and is challenged by several issues. This event is meant to support the development of a new collaborative, international network connecting different fields of research and integrating the young researchers community with the help of a scientifically diverse, interactive workshop.

Porous materials serve diverse applications, including molecular separations and catalysis, facilitating the efficient capture of greenhouse gases (CO₂ and CH₄) and valuable noble gases (Xe, Ar, and Kr). Porous liquids (PLs) with porous organic cages (POCs) offer a promising solution to xenon isolation challenges. We introduce precise and data-efficient MLIP models developed using Allegro, trained on DFT-level energies, forces, and virials in structures comprising 600 to 1170 atoms. The dataset comprises 1.7 million atoms and 12 million data points. The MLIP models enable simulations of large-scale porous liquids under realistic physicochemical conditions, facilitating microscopic interpretation of experimental Xe NMR data at both static and dynamic levels.

Various applications benefit from porous materials, such as molecular separations and catalysis, enabling the efficient capture of greenhouse gases (CO₂ and CH₄) and valuable noble gases (Xe, Ar, and Kr). Porous liquids (PLs) with cavities formed by porous organic cages (POCs) show promise in addressing xenon isolation challenges. Here, we present accurate and data-efficient MLIP models built using Allegro, a local equivariant deep NN architecture, trained on DFT-level data covering structures with 600 to 1170 atoms (H, C, N, O, F, Cl, Xe), totaling 1.6 million atoms with 11.2 million data points. The MLIP models enable simulating large-scale porous liquids at realistic physicochemical conditions and provide microscopic interpretation of experimental Xe NMR data at both static and dynamic levels.

Diverse applications benefit from porous materials, such as molecular separations and catalysis, enabling the efficient capture of greenhouse gases and valuable noble gases. Porous liquids (PLs) incorporating cavities formed by porous organic cages (POCs) show promise in addressing xenon isolation challenges. In this work, we present accurate and data-efficient MLIP models built using Allegro, trained on DFT-level data covering structures with 600 to 1170 atoms (H, C, N, O, F, Cl, Xe), totaling 1.7 million atoms with around 12 million data points. The MLIP models enable simulating large-scale porous liquids at realistic physicochemical conditions and provide microscopic interpretation of experimental ¹²⁹Xe NMR data at both static and dynamic levels.

Numerous studies have been conducted on inorganic oxy-hydroxy-fluoride compounds with the aim of determining their physicochemical properties. Their crystalline structures are of particular interest, yet conventional diffraction techniques are unable to distinguish between O and F atoms. Solid-state NMR spectroscopy (ss-NMR), coupled with DFT calculations, brings forth accurate modeling of the crystal structure. Herein, we present results for oxy-trifluoride of niobium and tantalum MOF₃ (M = Nb, Ta) and the Hexagonal-Tungsten-Bronze phase of TiOF₂ (HTB-TiOF₂). Their structures, which show correlated disorder, were precisely modelled using X-Ray Powder Diffraction, ¹H and ¹⁹F MAS ss-NMR and DFT calculations.

In 4 days of lectures and hands-on sessions, this workshop covers ab-initio molecular dynamics, machine learning force fields, structure prediction, and phonons. During the hands-on sessions, participants learn to perform ab-initio simulations using the Vienna Ab-initio Simulation Package (VASP), with an opportunity to meet the VASP development team directly.

Inorganic oxyfluorides are heteroanionic inorganic compounds that present interesting physical properties. Due to similar anion size, oxide and fluoride anions can occupy the same crystallographic sites. Since O and F atoms have quasi-similar scattering factors, structural modeling of disordered inorganic oxyfluorides is challenging using X-ray diffraction techniques. The high sensitivity of solid-state NMR spectroscopy to local environment makes it an ideal tool for disordered solids. In this study, the structures of NbOF₃ and TaOF₃ were revised using X-ray powder diffraction and ¹⁹F MAS ss-NMR. Ten possible 2×2×1 supercells were built and DFT optimized. The resulting agreement between experimental and theoretical ¹⁹F isotropic chemical shifts validates these ten models as accurate descriptions of the disorder in the studied structures.

Cette communication illustre, par quelques exemples tirés de nos travaux récents, l'apport de la RMN du solide et de la modélisation par DFT des paramètres RMN, à la description structurale de fluorures inorganiques désordonnés. La RMN de ¹⁹F a permis d'identifier et quantifier les divers environnements du fluor dans une anatase hydroxyfluorée lacunaire cationique. Les structures des isotypes NbO₂F et TaO₂F d'une part, et NbOF₃ et TaOF₃ d'autre part, ont été modélisées à l'aide de supermailles permettant de satisfaire les désordres corrélés qui les caractérisent. Les paramètres RMN de ¹⁹F calculés à partir de ces modèles, en très bon accord avec les paramètres expérimentaux, valident ces modèles structuraux.

Tantalum and niobium oxyfluorides are heteroanionic inorganic compounds that present interesting physical properties. Structural modeling of inorganic disordered oxyfluorides is challenging since their anionic disorder is difficult to characterize using conventional diffraction techniques — O and F atoms are indistinguishable due to quasi-similar scattering factors. Solid-state NMR spectroscopy (ss-NMR) sensitivity to short-range environmental effects, coupled with DFT calculations, brings forth accurate structural solutions. In this study, the structure of NbOF₃ and TaOF₃ was precisely modeled using X-ray powder diffraction, ¹⁹F and ¹H MAS ss-NMR and DFT calculations, unambiguously establishing the existence of one-dimensional strings of correlated O/F disorder along <100> and <010> directions.

De nombreux matériaux cristallisés présentent un désordre de nature chimique (sites cristallographiques occupés à la fois par des atomes de natures différentes ou des lacunes) ou topologique et des propriétés physiques intéressantes. Pour mieux comprendre et améliorer ces propriétés, une description structurale de ces matériaux est nécessaire. La sensibilité de la Résonance Magnétique Nucléaire (RMN) à l'environnement local du noyau sondé en fait une sonde idéale pour les solides désordonnés. Des informations structurales précises peuvent en être extraites en confrontant résultats expérimentaux et calculs DFT issus de modèles structuraux, des supermailles reflétant à la fois les caractères périodiques et désordonnés du matériau. Ma thèse vise à décrire aussi précisément que possible la structure de (halogéno)-(hydroxy)-(oxy)-fluorures inorganiques désordonnés d'intérêt.