Our group investigates the design of next-generation cathode materials for Li-ion and Na-ion batteries using advanced first-principles simulations. Standard density-functional theory (DFT) often fails for transition-metal compounds due to strong self-interaction errors, so we employ extended Hubbard-corrected approaches (DFT+U+V) with parameters computed fully from first principles. This methodology captures the delicate balance between electron localization and orbital hybridization, enabling accurate predictions of structural, electronic, magnetic, and electrochemical properties. We have demonstrated that DFT+U+V provides voltages and oxidation-state transitions in remarkable agreement with experiment for phospho-olivine and spinel-type cathodes. Our work establishes a reliable framework for exploring complex transition-metal oxides and systematically discovering novel cathode materials, without relying on empirical fitting.

Our group has developed a fully first-principles framework for modeling spin-wave spectra, based on time-dependent density-functional perturbation theory (TDDFPT) with nonempirical Hubbard functionals. This approach, implemented in a general noncollinear formulation, goes beyond empirical models by directly probing the dynamical spin susceptibility while treating Hubbard corrections self-consistently. It enables accurate predictions of magnon dispersions in complex transition-metal compounds, as demonstrated for benchmark systems such as NiO and MnO, where our results are in excellent agreement with experiments. We are now applying these advanced techniques to altermagnets, a recently discovered class of materials that combine compensated magnetic order with spin-splitting in the electronic structure. By exploring their magnon spectra from first principles, we aim to uncover the fundamental mechanisms governing their spin excitations and to assess their potential for magnonic and spin-caloritronic applications.

Core-level spectroscopies such as x-ray absorption near-edge structure (XANES) and x-ray photoelectron spectroscopy (XPS) are powerful tools for probing the electronic structure of materials. In our group, we develop and apply first-principles methods based on density-functional theory with extended Hubbard functionals to model these spectroscopies in complex transition metal compounds. By determining Hubbard parameters fully from first principles and employing efficient Lanczos-based algorithms, our approach provides a reliable and accurate description of spectral features across a broad energy range, at a computational cost much lower than that of hybrid functionals. We apply these methods to technologically relevant systems such as perovskite oxides and related transition-metal compounds, where conventional functionals often fail to capture key electronic properties. Our simulations not only reproduce experimental spectra with high fidelity but also offer microscopic insight into the role of electronic localization, orbital hybridizations, and chemical substitutions. This enables us to interpret and predict spectroscopic signatures in materials central to energy, catalysis, and electronic applications.

In our group, we develop and apply advanced Hubbard-corrected approaches such as DFT+U and DFT+U+V, where the Hubbard parameters are not fitted but computed fully from first principles. Using density-functional perturbation theory (DFPT) with monochromatic perturbations, we achieve highly efficient and automated calculations of both on-site (U) and inter-site (V) interactions, with tight control over precision and convergence. This framework allows us to capture the subtle balance between electron localization and orbital hybridization, enabling quantitatively predictive simulations of structural, electronic, magnetic, and electrochemical properties. We also develop HP, an open-source implementation of DFPT for Hubbard parameters within the Quantum ESPRESSO distribution. Designed for efficiency and portability, HP makes advanced Hubbard corrections accessible across computational platforms, from laptops to supercomputers. Together, these developments establish a robust foundation for exploring transition-metal and rare-earth compounds with predictive accuracy, paving the way for reliable first-principles design of functional materials.

Machine learning is transforming computational materials science by bridging the gap between accuracy and efficiency. In our group, we develop data-driven models that complement and accelerate first-principles simulations of complex materials. A central focus is the prediction of Hubbard parameters for DFT+U+V, which are essential for accurately describing transition-metal and rare-earth compounds. Using equivariant neural networks with atomic occupation matrices as descriptors, we capture the electronic structure, chemical environment, and oxidation states of materials. Trained on high-quality data from first-principles linear-response calculations, our models predict on-site U and inter-site V values with only a few percent error, while avoiding the heavy computational cost of density-functional perturbation theory. This enables high-throughput exploration of complex materials with unprecedented speed. Beyond Hubbard parameters, we also explore machine learning interatomic potentials that approach first-principles accuracy at the cost of classical force fields. Our work emphasizes graph-based neural network architectures, which naturally capture long-range interactions, charge transfer, and complex bonding networks — features that are difficult to model with traditional approaches. Together, these developments open new avenues for accelerated discovery and design of functional materials for energy and electronic applications.

In a close collaboration with the group of Prof. Laura Grigori (EPFL/PSI), we develop next-generation numerical algorithms that exploit the randomization techniques and mixed-precision capabilities of modern and emerging GPUs. Our goal is to improve the efficiency of materials science simulations, particularly those based on density-functional theory. By carefully leveraging lower-precision arithmetic while controlling accuracy, we design methods that maintain rigorous error bounds for critical operations. Randomization provides an additional advantage, enabling scalable algorithms with simplified communication patterns and reduced computational cost—while still offering strong probabilistic guarantees. Our work directly targets complex electronic-structure codes that use plane-wave basis sets and pseudopotentials, such as Quantum ESPRESSO, DFTK, and SIRIUS. These widely used codes will benefit from our optimizations on cutting-edge mixed-architecture supercomputers, including next-generation systems like Alps at CSCS.