Recently, we advanced the GW methodology to deep core excitations as measured by X-ray photoelectron spectroscopy (XPS), combining exact numeric algorithms in the real frequency domain with partial self-consistency and relativistic corrections [1,2,3]. We successfully benchmarked our core-level GW approach for molecular 1s binding energies [2] and its Bethe-Salpeter-equation extension for molecular K-edge transition energies [4]. However, its application to complex materials is computationally very demanding. We approach this challenge by i) reducing the scaling with respect to system size and ii) combining the XPS predictions with machine learning (ML) models. A scaling reduction by one order of magnitude has already been achieved by introducing an analytic continuation of the screened Coulomb interaction in our algorithm. We show that the revised algorithm yields mean absolute deviations of less than 10 meV with respect to the conventional implementation. Furthermore, we present a quantitatively accurate ML model for XPS predictions of disordered materials. Our kernel ridge regression ML model is based on a comprehensive data base of density functional theory (DFT) and GW data. We applied our combined DFT-GW-ML approach to materials containing carbon, hydrogen and oxygen and showed that we obtain qualitative and quantitative agreement with experiment, resolving spectral features within 0.1 eV of reference experimental spectra [5]. [1] JCTC, 14 (2018), 4856 [2] JPCL, 11 (2020), 1840 [3] JCP, 153 (2020), 114110 [4] JCTC, 18 (2022), 1569 [5] arXiv:2112.06551 (2021)

Simulation of excited states of proteins is challenging, especially in photoactive proteins, due to the complexity of creating realistic models (protonation states, electrostatic interactions, hydrogen bond network, solvent effects, etc.) and the non-adiabatic coupling between electronic and nuclear degrees of freedom. In addition, the overall computational cost frequently imposes serious constraints on the description of excited states. In this talk, I will present the recent advances of electrostatic potential fitting electronic QM/MM embedding in simulating the vibrational effects on absorption spectra. For this purpose, we developed analytic second derivatives of the QM/MM energy, from which we extract protein vibrations.[1,2] As a remarkable feature, we managed to completely eliminate the scaling of QM equation on the MM system size, which was introducing one of the main computational bottlenecks in such computations.[3] This allows a routine calculation of second derivatives at the QM/MM level. To illustrate this, I will show several examples of the cryptochrome flavoprotein, a blue-light sensor involved in oxidation-reduction reactions, from which we have computed the infrared spectra,[4] vibrationally resolved spectra,[5] chemical reactivity, etc. \newline \newline [1] K. Schwinn, N. Ferré, M. Huix-Rotllant, J. Chem. Theory Comput.~\textbf{16}, 3816 (2020). \newline [2] M. Huix-Rotllant, N. Ferré, J. Chem. Theory Comput.~\textbf{17}, 538 (2021). \newline [3] K. Schwinn, N. Ferré, M. Huix-Rotllant, J. Chem. Phys.~\textbf{151}, 041102 (2019). \newline [4] M. Huix-Rotllant, K. Schwinn, N. Ferré, Phys. Chem. Chem. Phys.~\textbf{23}, 1666 (2021). \newline [5] K. Schwinn, N. Ferré, M. Huix-Rotllant, Phys. Chem. Chem. Phys.~\textbf{22}, 12447 (2020).

Spectroscopy is a fundamental tool in materials research and characterisation and has consequently become a major objective of machine learning tasks. Here, deep learning based on neural networks (NNs) is a particularly powerful approach. NNs are universal approximators with the ability to represent almost arbitrarily complex relationships, and we have reviewed the performance of different NN architectures in spectral prediction tasks [1]. Despite early successes, data-driven spectroscopy remains challenging. To advance the current state-of-the-art we must address the issues of raw data availability, material representation and its invertibility, as well as model interpretability, uncertainty and scalability [2]. Deep learning spectroscopy has developed into an innovative and dynamic research field, with the potential to usher in a new era in materials characterisation. \newline \newline [1] K. Ghosh, A. Stuke, M. Todorovi\'c et al, Adv. Sci ~\textbf{6}, 1801367 (2019). \newline [2] H. Kulik et al, Electron. Struct. in press https://doi.org/10.1088/2516-1075/ac572f (2022).

Single-layer graphene (SLG) exhibits unique optoelectronic properties, including large mobility and broadband linear absorption, tunable via electrical control of the Fermi level (E$_F$) [1] Many optoelectronic applications has been proposed to exploit the ultrafast dynamics (∼fs-ps) of its charge carriers, including broadband photodetectors, nonlinear frequency converts, saturable absorbers and modulators [2], providing the building blocks for next-generation data communication systems. Therefore, the investigation of the relaxation dynamics of SLG charge carriers has been driven by both fundamental scientific interest in the intriguing electron-electron and electron-phonon interactions of SLG and by the need to guide the route for technological development. Here, I discuss the relaxation dynamics of SLG charge carriers as unveiled by time-resolved pump-probe spectroscopy, from the visible to the terahertz range and by time-resolved photocurrent measurements [3-5]. The interband absorption of a femtosecond pump pulses is used to induce strongly nonequilibrium charge carriers distributions in SLG, which thermalize to Fermi-Dirac (FD) distribution at elevated temperature through carrier-carrier scattering, acting on a sub-100 femtoseconds time scale. The thermalization is followed by a picosecond cooling phase, where different phonon systems exchange energy with the hot electrons, including SLG acoustic and optical phonons, and substrate phonons. The interplay of the different the cooling pathways and its dependence on the sample quality (defect density and mobility), on the substrate, and on the probing configuration is explored by studying the ultrafast response of standard SiO$_2$-supported SLG, compared with that of technological relevant systems such as suspended SLG and SLG encapsulated by hBN or WSe$_2$, providing a unique benchmark for the different theoretical models proposed for hot electrons cooling. Among the proposed cooling pathways, predictions of supercollision theory [6], near-field coupling to hyperbolic phonons of the encapsulant [7], heat diffusion [8] and scattering with strongly coupled optical phonons (SCOP) [9]. Specifically, the investigation of high-quality graphene, in which extrinsic decay mechanisms are suppressed, allows to identify the intrinsic limit to the cooling dynamics of hot electrons in the scattering with graphene SCOP. Unprecedentedly large tunability of the out-of-equilibrium properties [5,10] and non linearities [11] of SLG is finally demonstrated by electrical control of the Fermi level E$_F$ with ionic liquid gating, and rationalized in terms of quenching of the SCOP emission due to saturation of the phase space. \textbf{References} \newline [1] A. C. Ferrari, \textit{et al.}, Nanoscale\textbf{7}, 4598–810 (2015). \newline [2] F. Bonaccorso, Z. Sun, T.A. Hasan, A.C. Ferrari, Nat. Photon. \textbf{4}, 611 (2010). \newline[3] K.-J. Tielrooij \textit{et al.}, Nat. Nanotechnol. \textbf{13}, 41-46 (2018). \newline[4] E.A.A. Pogna\textit{et al.}, ACS Nano \textbf{15}, 11285-11295 (2021). \newline[5] E. A. A. Pogna \textit{et al.}, ACS Nano, XX, XXX (2022). https://doi.org/10.1021/acsnano.1c04937 \newline[6] J. C. W. Song, M. Y. Reizer, L. S. Levitov, Phys. Rev. Lett. \textbf{109}, 106602 (2012). \newline[7] A. Principi, M. B. Lundeberg, N. C. H. Hesp, K. -J. Tielrooij, F.H.L. Koppens, M. Polini, Phys. Rev. Lett. \textbf{118}, 126804 (2017). \newline[8] A. Block \textit{et al.}, Nat Nanotechnol. \textbf{16}, 1195–1200 (2021). \newline[9] Tomadin A., Brida D., Cerullo G., Ferrari, A.C. and Polini M., Phys. Rev. B \textbf{88}, 035430 (2013). \newline [10] L. Ghirardini, E.A.A. Pogna, G. Soavi \textit{et al.}, 2D Materials \textbf{8}, 035026 (2021). \newline[11] G. Soavi \textit{et al.} ACS Photon. \textbf{6}, 2841-2849 (2019).

In a previous study [1], we developed a $\mathbf{k}$-dependent self-energy that describes many-body renormalization effects due to the scattering of electrons with magnons and Stoner excitations. We implemented this $GT$ self-energy, which contains infinitely many spin-flip ladder diagrams ($T$-matrix formulation), within the full-potential linearized augmented-plane-wave method and applied it to the elementary ferromagnets Fe, Co, and Ni. The $GT$ self-energy was shown to have a profound impact on the electronic band structure giving rise to strong spin-dependent lifetime effects and emergent dispersion anomalies in valence and conduction bands, in particular, predicting a band anomaly in iron that was later confirmed experimentally. In the present work [2], we refine the method by combining $GT$ with the $GW$ self-energy, thus combining electron-magnon and electron-plasmon scattering processes. Applying this double-counting-free $GWT$ self-energy to Fe, Co, and Ni, we find that the inclusion of $GW$ rectifies a number of deficiencies we had found with $GT$ alone: (i) The $d$ bandwidth, overestimated in $GT$, is corrected; (ii) Band broadening is less extreme; (iii) The violation of causality inherent to $GT$ is healed. Furthermore, the shape of the band anomaly in iron improves with respect to experiment. The $GWT$ spectral functions are computed to be in an overall better agreement with the experimental results than those obtained with the $GW$ or $GT$ techniques alone, even showing partial improvements over local-spin-density approximation dynamical mean-field theory. Moreover, $GWT$ magnetic moments and exchange splittings are closer to experiment than when calculated with LSDA, $GW$, or $GT$. The performed analysis provides a basis for applying the $GWT$ technique to a wider class of magnetic materials. \newline [1] M. C. T. D. M\"uller, S. Bl\"ugel, and C. Friedrich, Phys.~Rev.~B 100, 045120 (2019). \newline [2] D. Nabok, S. Bl\"ugel, and C. Friedrich, npj Comput. Mater. 7, 178 (2021).

All-optical spectroscopy of matter is a well established technique to access properties of electrons and quasi-particles in matter. The description of these optical properties is usually done employing ab initio methods such as linear response time-dependent density functional theory (TDDFT) or Green functions based methods. Recent advances in experimental techniques have opened up the possibility to develop novel types of spectroscopies based ultrashort and intense laser pulses, allowing for instance to access non-equilibrium dynamics of electronic systems or to allow for exploring non-equilibrium phase diagram of correlated materials. The strong-field electronic dynamics in solids have received a lot of attention, in particular due to the experimental observation of high-harmonic generation in solids. The dynamics associated with strong laser fields requires a non-perturbative description of the electronic dynamics. One possible way of describing this dynamics by ab initio methods is to use real-time TDDFT, which will be introduced. In this talk, I will first discuss the standard optical spectroscopies and how they are usually modeled using ab initio methods. Then, I will present how now novel experimental methods can be simulated and what are the remaining challenges posed by recent experimental advances.

Scientific and technological breakthroughs are often a result of interdisciplinary efforts in which diverse areas of expertise led to the necessary insights, understanding, and solutions that enabled us to excel. In this aspect, materials characterization, and experts therein, play an equally important role. In various fields, materials characterization is indispensable to fundamental and exploratory studies, to process development and control, as well as to process monitoring. To sustain this role, characterization methods must be able to answer to the needs of future materials and technology, in which the materials library keeps expanding, device architectures gain in complexity, broad time- and length-scales need to be addressed, etc. Keeping up with this evolution requires pushing the analytical performance to (or beyond) the limits in view of spatial- and time-resolution, sensitivity, and reliability, and establishing reproducible and accurate measurement schemes that ultimately yield quantitative information. This presentation will reflect on these aspects and the related challenges in materials characterization, particularly those arising from advanced semiconductor materials and applications, with focus on spectroscopy methods. Next to a brief overview of the spectroscopy facilities available at imec, we will describe how they can support the development of novel materials with well-controlled properties. Eventually, this presentation might spark the listeners interest to engage in interdisciplinary research efforts tackling these open challenges.

Moore’s law, transistors, memories, Graphical and Central Processing Units (GPU, CPU), nanoelectronics power consumption, data storage and data transfer speed, 5G and 6G communications, sensors, embedded radars, holography, artificial intelligence, health monitoring, ... are parts of the bubbling technologies invading our lives. It is often perceived that these innovations are the sole result of ground-breaking circuit and architecture designs, which is overshadowing how intimately all these are relying on material and process technology innovations. Being able to identify the most suited material(s) in terms of performances, how to deposit and integrate them at the nanometer scale into functional devices, with a high degree of control, reproducibility, and reliability, is not only a high technology value problem but also a process of incredible complexity which is intertwining numerous fields of expertise and complex research patterns. The identification of the right material(s) that sometimes contain up to 4 of 5 different chemical elements from the 118 possible ones of the periodic table is only a first step in the process. The next one consists in including the constraints linked to their deposition, etching, compatibility with the gas and thermal window used in nanoelectronics and controlling their interfaces. Ultimately, this process relies on the need to build a complex multi-scale modelling that account for material growth constraints, stochasticity effects and interface variability to engineer and unlock technical challenges in nanoelectronics. All of these are a daunting task that is often ignored in enthusiastic material research claims. In this talk, some examples will be discussed on how this problem is being approached and how first-principle and multi-scale modelling techniques are being used to enable material implementation in nanoelectronics.

In a strongly nonlinear regime, photoinjection of charge carriers can be temporally confined to a single half-cycle of an ultrashort laser pulse. This extreme temporal confinement is instrumental in attosecond metrology: the sudden change of a medium’s electric and optical properties can be used as a gate for time-resolved measurements, providing attosecond temporal resolution without the necessity to generate attosecond light pulses. In return, attosecond metrology provides real-time insight into light-matter interaction, including the process of producing charge carriers through absorption of light, as well as the subsequent light-driven motion of electrons and holes. In my talk, I will review our recent experimental and theoretical research on ultrafast photoexcitation of solids.

Electronic and optical properties of materials are affected by atomic motion through electron-phonon interaction. This interaction has two noticeable effects on the electronic structure: variation of band gaps with temperature, but also band-gap renormalization even at zero temperature, due to zero-point motion. Ignored in most calculations of the electronic structure, zero-point effects have been evaluated recently from first principles for several materials. Many of these calculations relied on the adiabatic approximation, reasonably valid for materials without infrared activity, but eagerly applied to other materials. \newline I report first on the first-principles evaluation of band edge zero-point renormalization (ZPR) beyond the adiabatic approximation for 30 materials [1]. When light elements (e.g. oxgen) are present, the band gap renormalization is often larger than 0.3 eV and can be bigger than 1 eV. This effect cannot be ignored if accurate band gaps are sought. In particular, it is useless to go beyond G$_0$W$_0$ without including ZPR effects in such materials. For infrared-active materials, global agreement with available experimental data is obtained only when nonadiabatic effects are taken into account. They even dominate zero-point renormalization for many materials. A generalized Fr\"ohlich model (multiphonon, anisotropic and degenerate electronic extrema) represents accurately non-adiabatic effect. Its accuracy is assessed against first-principles results. This model describes the essential physics and accounts for more than half the total ZPR for a large set of materials, especially for valence band edges, despite its neglect of interband electronic transitions, Debye-Waller contributions and acoustic phonon contributions present in the full first-principles approach. \newline The Fr\"ohlich model is actually a workhorse for the large polaron community, used for decades. For the same set of materials, and using its generalized version, the domain of validity of its major hypothesis, namely the large extent of the electronic cloud with respect to the interatomic distance (large polaron), is tested, as well as the possible breakdown of perturbation theory due to self-trapping effects. Most materials exhibit large polaron behaviour as well as validity of perturbative treatment. However, especially for the valence band, there is also a non-negligible number of materials for which perturbative treatment cannot be applied and/or for which the self-trapping region has a dimension close to the atomic repetition distance. \newline \newline [1] A. Miglio, V. Brousseau-Couture, E. Godbout, G. Antonius, Y.-H. Chan, S.G. Louie, M. Côté, M. Giantomassi and X. Gonze. npj Computational Materials, 6, 167 (2020). \newline [2] B. Guster, P. Melo, B.A.A.Martin, V. Brousseau-Couture, J. de Abreu, A. Miglio, M. Giantomassi, M. Côté, J.M. Frost, M.J. Verstraete and X. Gonze, Phys. Rev. B, 104, 235123 (2021)

Excitonic effects have proved to be crucial for the quantitative description of optical absorption [1]. This picture has been confirmed for core excitations [2]. The description of excitons can be conducted, within the many-body Green's functions formalism, via the solution of the Bethe-Salpeter equation (BSE), which takes the electron-hole interaction, in the presence of all the other electrons, automatically into account. More recently, we have put a lot of effort into extending the applicability of the BSE towards: i) different spectroscopies (like electron and photon scattering [3-5]), ii) unconventional picture (excitonic band-structure [6]), and iii) pictorial tools (of the electron density evolution). In this seminar, I will illustrate the achievements and problems of the BSE in tackling these new spectroscopies and tools, with particular emphasis on a recent approach for the description of Resonant Inelastic X-ray Scattering (RIXS).