10th Users' Conference of IT4Innovations

5 Nov 2026, 10:00 → 6 Nov 2026, 14:00 Europe/Prague

atrium (IT4Innovations)

Branislav Jansik (IT4Innovations), Karina Pešatová (IT4I), Tomáš Kozubek (IT4Innovations), Vít Vondrak (IT4Innovations, VSB-Technical University of Ostrava)

10^th Users' Conference of IT4Innovations will take place on 5 and 6 November 2026. All of our users, as well as research and project partners from various organisations, research institutions, and industry, are welcome to attend the Conference.

The submission deadline for contributions is September 20, 2026.

Registration is open until October 26, 2026.

Attendees will discover more about our future upgrade plans, listen to talks given by our prominent users, and can engage in discussions during the Users' Council meeting and a poster session.

Contribution types: Users' talks/Posters

Users' talks

Selected talks by our prominent users will be presented at scheduled times during the whole conference. Each talk is expected to take a maximum of 20 minutes (with discussion included):

• IT4I Project(s) (5 minutes): Dedicate approximately 5 minutes to provide insights into your computational approach and methods. Discuss software utilisation, job sizes, parallelisation techniques, and the extent of computational resources employed. Sharing your computational experience will provide valuable context for the audience.
• Research (10 minutes): Allocate around 10 minutes to delve into your research endeavours. Highlight key findings, address challenges encountered, and emphasise the significance of your work. This segment should offer a concise yet comprehensive overview of your research area.
• Discussion (5 minutes): Please reserve 5 minutes to engage with questions and discussions from the audience.

Posters

Please note that the required poster size is A1 portrait orientation. 

If you would like IT4Innovations to print and display your poster at the conference, please ensure you send your printing file to training@it4i.cz by October 25.

 

This event is supported by the Ministry of Education, Youth and Sports of the Czech Republic through dedicated funding for the Large Research Infrastructure project “e-INFRA CZ – LM2023054”.

 

All presentations and educational materials of this course are provided under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

Branislav Jansík.pdf

Ctirad Červinka.pdf

Dominik Legut.pdf

Markéta Palocýnová.pdf

Matúš Kaintz.pdf

Michal Kubíček.pdf

Michal Svatoš.pdf

Ondřej Pejcha.pdf

Pavel Polishchuk.pdf

Peter Klaver.pdf

Petr Ptáček.pdf

Radim Vavřík.pdf

Ryszard Kukulski.pdf

Štěpán Sklenák.pdf

Zaira Carboni.pdf

Thursday 5 November

Registration and networking

Barbora NG’s Inauguration

12:30 Lunch

Keynote I

1 From Defects to Devices: Design Guidelines for High-Performance Diamond-based Solar Cells and Single-dopant Diodes

As global energy demand rises and the transition towards carbon-neutral energy systems accelerates, developing highly efficient energy-harvesting and electronic devices is becoming increasingly important. Diamond is an exceptional platform for nanoengineered optoelectronics owing to its ultra-wide band gap, outstanding carrier mobility, and high thermal conductivity. Crucially, its wide band gap enables unique defect-engineering strategies that are difficult to realise in conventional semiconductors.

With this aim, we present a unified first-principles framework that combines density functional theory with GW corrections, Bethe-Salpeter equation calculations, and carrier transport modelling coupled to a device-level Poisson solver to investigate two novel device architectures. The first is an intermediate-band photovoltaic device based on boron-vacancy-boron (BVB) cluster defects. The BVB defect introduces multiple intermediate bands within the diamond band gap, enabling the absorption of sub-bandgap photons and providing a promising route towards overcoming the Shockley-Queisser efficiency limit of conventional solar cells. The second architecture is a phosphorus-based PN junction in which both regions are doped with the same dopant species. In this design, the p-type region is formed through phosphorus-vacancy (PV) complexes, enabling a simplified single-dopant device concept that could reduce manufacturing complexity, minimise cross-contamination and streamline fabrication processes.

Our results show that BVB defects provide strong intermediate-band optical functionality while preserving the favourable transport and thermal properties of diamond. We further establish practical design guidelines for photovoltaic operation, including optimal crystal orientation, absorber thickness and junction doping profiles. For the phosphorus-based diode, we demonstrate that PV defects support impurity-band transport and high conductivity, making them suitable for degenerate p-type applications such as tunnel diodes and asymmetric junctions. Furthermore, its temperature-dependent Seebeck coefficient anisotropy and sign-reversal offer novel thermal management pathways.

The outcomes of this investigation constitute key technological guidelines that can be promptly applied to the design and fabrication of next-generation, high-performance diamond optoelectronic devices.

This work was supported by the project “The Energy Conversion and Storage”, funded as project No. CZ.02.01.01/00/22_008/0004617 by Programme Johannes Amos Commenius, call Excellent Research. This work was supported by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID:90254). The access to the computational infrastructure of the OP VVV funded project CZ.02.1.01/0.0/0.0/16_019/0000765 “Research Center for Informatics” is also gratefully acknowledged.

Speaker: Matúš Kaintz (Czech Technical University in Prague)

Users' Talks I

2 Linear absorption of single-layer graphene deposited on quartz: density-functional tight binding for sensing applications

Fine-tuning of applications based on 2D materials requires advanced characterization of their optical and electronic properties in conditions that are compatible with industry standards. While studies can be realized in free-standing conditions or at low temperature, identifying the contribution of the substrate in ambient conditions of laser processing are in high demand. Moreover, currently available optical measurements show some dependence on the chosen substrate. In these conditions, ab-initio modeling appears as an important path to gain knowledge for the development of applications.
In this work, we used the density functional tight-binding (DFTB) formalism to study the optical response of a flake of graphene deposited on a sample of quartz. After preparing the system with an optimal separation, the effect of lattice temperature on the optical response was estimated using an ensemble average of multiple configurations extracted from a quantum force-field (Born-Oppenheimer) molecular dynamics with a Nosé-Hoover thermostat. The computational results were compared with experimental measurements of transmission spectra, from 200 nm up to 2.5 µm, using a commercial monolayer graphene sample deposited on fused silica.
Beyond the role of shape and size of the flake on its optical response, it was found that its deposition on a quartz substrate can affect its Fermi energy and linear absorption spectra. Our computational protocol is a step forward in simulating challenging systems under realistic conditions, while providing intuitive insights for the design of nanoelectronic applications based on usage of organic 2D materials.

Speaker: Prof. Thibault Derrien (FZU Institute of Physics)

3 Spin-lattice models in cubic symmetry - multiscale bridge in magnetic properties

We present an approach based on the Néel model to build a classical spin-lattice Hamiltonian for cubic crystals capable of describing magnetic properties induced by the spin-orbit coupling like magnetocrystalline anisotropy and anisotropic magnetostriction, as well as exchange magnetostriction. Taking advantage of the analytical solutions of the Néel model, we derive theoretical expressions for the parametrization of the exchange integrals and Néel dipole and quadrupole terms that link them to the magnetic properties of the material. This approach allows us to build accurate spin-lattice models with the desired magnetoelastic properties. We also explore a possible way to model the volume dependence of magnetic moment based on the Landau energy. This feature allows us to consider the effects of hydrostatic pressure on the saturation magnetization. We apply this method to develop a spin-lattice model for BCC Fe and FCC Ni, and we show that it accurately reproduces the experimental elastic tensor, magnetocrystalline anisotropy under pressure, anisotropic magnetostrictive coefficients, volume magnetostriction, and saturation magnetization under pressure at zero temperature. [1] In addition, we have constructed spin-lattice model for the antiferromagnetic cubic compounds. Here we develop a methodology to integrate magnetic properties into interatomic potentials for cubic antiferromagnets by adding a magnetic Hamiltonian which includes both the Heisenberg exchange and Néel model. [2] We apply this approach to NiO by constructing two potentials: one based on the Born model of ionic solids and another using a reference-free modified embedded atom method. Both potentials include magnetoelastic interactions and are validated against Density Functional Theory calculations, showing excellent agreement in mechanical and magnetic properties at zero temperature. These models
enable large-scale simulations of magnetoelastic phenomena in antiferromagnets and open avenues for molecular dynamics studies involving coupled electric and magnetic fields in metal oxides. We extent our study for several properties at finite temperatures.

References:
[1] P. Nieves, J. Tranchida, S. Arapan, D. Legut, Phys. Rev. B {\bf 103}, 094437 (2021).
[2] I. Korrnienko, P. Nieves, J. Šebesta, R. Iglesias, J. Chem. Phys. {\bf 165}, 024701 (2026).

Speaker: Dominik Legut (IT4I)

4 DFT computational viewpoint of the Alpha oxygen – a unique oxidation active site

The Alpha oxygen represents a highly reactive, extra-framework oxygen species stabilized within zeolite matrices via the low-temperature catalytic cleavage of either nitrous oxide or dioxygen. Characterized by an exceptionally high oxidation potential, this species is capable of mediating the direct, selective partial oxidation of methane to methanol under ambient conditions. Consequently, the Alpha oxygen serves as a highly attractive paradigm in the field of heterogeneous oxidation catalysis.[1]

Despite its technological significance, experimental elucidation of its precise coordination geometry, structural dynamics, and reactivity remains constrained. To resolve these ambiguities, density functional theory (DFT) simulations were employed to model the mechanistic pathways governing the Alpha oxygen formation, stabilization, and interfacial reactivity. This presentation provides an overview mapping the structural, electronic, and catalytic properties of this potent oxidant.[1]

Reference

[1] Stepan Sklenak and Jiri Dedecek: "Alpha oxygen - a unique oxidation active site from a quantum chemical viewpoint", Chem. Soc. Rev., 2026, 55, 2120-2144 .

Speaker: Stepan Sklenak (J. Heyrovsky Institute of Physical Chemistry)

Accessing EuroHPC Systems

Convener: Branislav Jansik (IT4Innovations)

Services from the High-Level Support Team (HLST)

15:40 Coffee Break

Keynote II

5 Multi-dimensional radiation-magneto-hydrodynamical simulations of the lives and deaths of interacting binary stars

Majority of stars in the Universe exist in gravitationally bound binaries and hierarchical multiple systems. Interactions in binary star systems, such as transfer of mass between components, mergers, and explosions, significantly alter stellar evolutionary pathways. The resulting systems can range from mundane to exotic and include gravitational-wave sources and a variety of astrophysical brightenings including cosmologically relevant Type Ia supernovae.

The key phases of single and binary star evolution are often not directly observable and too complex for analytic treatment, necessitating numerical work. We review results from our group at the Institute of Theoretical Physics, Charles University, obtained with multidimensional hydrodynamic, magnetohydrodynamic, and radiation-hydrodynamic simulations using codes developed with the help of ERC Starting Grant and computation time generously allocated at IT4I.

First, we developed a radiation module for the moving-mesh code JET. It is based on mixed-frame grey flux-limited diffusion with implicit time integration. Second, we performed 3D hydrodynamic and magnetohydrodynamic adaptively and statically refined simulations on spherical and Cartesian meshes of the post-dynamical inspiral phase of common envelope evolution. Third, our 3D smoothed particle hydrodynamics simulations showed that a companion grazing a red supergiant on an eccentric orbit can eject envelope material.

We conclude with the computational challenges ahead and our plans for GPU-accelerated codes built around binary-star geometry within our ongoing ERC Consolidator Grant.

Speaker: Ondřej Pejcha (Institute of Theoretical Physics, Faculty of Mathematics and Physics, Charles University)

Users' Talks II

6 Using HyperQueue to run multiple ATLAS workloads at IT4Innovations

The distributed computing system of the ATLAS experiment at the Large Hadron Collider (LHC) uses computing resources of IT4Innovations for several years. So far, only selected workloads were used, i.e. those most suitable for many-core environment. The computing needs of the experiment do change and with that there is a need to be able to run more workloads. The HyperQueue, a tool designed to simplify execution of large workflows at HPCs, allows us to do that. Its schedulers were updated to allow full utilization of available resources by ATLAS jobs which request different number of CPU cores or memory. This work started in 2025 and it is now concluded.

Speaker: Dr Michal Svatoš (FZU, AV CR)

ATL-SOFT-SLIDE-2026-614.pdf

7 FrAMonC – Fragment-based Ab Initio Monte Carlo simulations of structural and thermodynamic properties of amorphous molecular materials

Delicate interplay of non-covalent interactions (NCI) governs the properties and phase behavior of molecular materials, being unprecedentedly rich already for the simplest molecules, such as water. Existing first-principles methods enable to model properties and polymorphism of molecular solids at least qualitatively [1]. However, those methods still frequently fail to yield at least semi-quantitative predictions [2], and are completely inapplicable for modelling properties of liquids or the solid-liquid equilibrium. Immense computational complexity inherently related to large-scale ab initio simulations of liquids and glasses has impeded this field of research for long [3]. Current work presents a newly developed fragment-based ab initio Monte Carlo approach for highly accurate predictions of structural and thermodynamic properties of amorphous molecular materials that require minimum empiric inputs [4]. A sophisticate nested Monte Carlo sampling that combines multiple potentials is a key feature enabling to reach a high computational accuracy at a reasonable computational cost. Since the best description of the underlying non-covalent interactions can be obtained with the coupled-clusters theory, aim of this work is to incorporate the coupled-clusters accuracy into simulations of properties such as thermal expansivity or compressibility of liquids. A suitable composite combination of force fields, semi-empiric quantum-chemical methods [5] and fragment-based ab initio many-body expansion [6] for the individual nested Monte Carlo layers represents a unique development in this field. Adoption of the many-body expansion of the cohesive energy and modern low-scaling formulations the coupled-clusters theory prove to be viable, indeed imparting the coupled-clusters accuracy to the amorphous realm of molecular liquids and glasses. A crucial feature of the computational protocol, greatly improving its efficiency, robustness and the applicability range is a ∆-machine learning scheme that is used to correct the outputs of the cheap and fast quantum-chemical models up to the standards of the coupled-clusters theory. The architecture of equivariant message-passing neural networks [7] is used to exploit the vast libraries of molecular interactions generated during the initial full quantum chemical sampling of the bulk materials. Most of the expensive quantum-chemical evaluations of molecular interactions energies can be then bypassed in latter bulk simulation stages using thence trained potentials. Validation of the computational methodology is presented for three small model systems, but future development for material design of pharmaceuticals or organic semiconductors is expected.

References
1. Červinka, C.; Štejfa, V.; Pokorný, V.; Touš, P.; Růžička, K. Cryst. Growth Des. 2023, 23, 9011.
2. Touš, P.; Červinka, C. Cryst. Growth Des. 2023, 23, 4082.
3. Červinka, C. J. Mol. Liq. 2025, 417, 126600.
4. Červinka, C. J. Chem. Theory Comput. 2025, 21, 9868.
5. Ludík, J.; Kostková, V.; Kocian, Š.; Touš, P.; Štejfa, V.; Červinka, C. J. Chem. Theory Comput. 2024, 20, 2858.
6. Červinka, C.; Beran, G. Chem. Science 2018, 9, 4622.
7. Schütt, K.; Unke, O.; Gastegger, M. Proc. Mach. Learn. Res. 2021, 139, 9377.

Financial support from the Czech Science Foundation (23-05476M) is acknowledged. Computational resources were supplied under the project e-INFRA CZ (ID:90254).

Speaker: Ctirad Cervinka

8 Small dust particle bombardment of a fuzz covered W surface

We present large scale molecular dynamics simulations of medium and high velocity W dust particle (10 - 40 nm diameter) impacts on fuzz covered W surfaces, as will likely occur in future fusion energy reactors. Results show that the commonly held intuition, that fuzz structures are fragile, is wrong. Nanorods are ductile due to their small dimensions and they undergo extensive plastic deformation under impact of medium velocity dust particles. A fuzz layer absorbs the energy and momentum of medium velocity small dust particles more gradually over a much greater thickness (up to several hundred nm) and a much longer time scale (up to over 1 ns) than a solid W surface. Impacts of high velocity dust particles do not cause extensive plastic deformation. Instead, dust particles liquefy on first impact on a nanorod, along with part of the nanorod. If the radius of the high velocity dust particle is large enough, the particle will smash through the nanorod, fragment into many smaller liquid droplets and these continue to move deeper into the fuzz layer. The amount of W that reflects back to the vacuum from a high velocity particle impact, as potential high-Z fusion plasma pollutant, ranges from 1/100th to 1/10th of the volume of the dust particle. High velocity particles also lose their energy and momentum over a much greater depth (hundreds of nm) and over a much longer time (hundreds of ps) than on a solid W surface. From a fusion energy application point of view, the results are encouraging, as they show that W fuzz is not fragile but is instead a better absorber for small W dust particles than a W solid surface.

Speaker: Peter Klaver (VSB Technical University of Ostrava)

Poster Session

9 From chemical ordering to mass transport: Machine-learning atomistic simulations of cubic TiAlON

Titanium aluminium nitride (TiAlN) is a benchmark protective coating. Incorporating oxygen can delay thermally induced phase transformations by up to 300 °C, but this enhanced thermal stability comes at the cost of a lower elastic modulus. Understanding how chemical disorder, vacancies, and local atomic reconstructions control this trade-off and influence high-temperature mass transport requires simulations beyond the length and time scales accessible to density-functional theory. We therefore developed an atomic cluster expansion machine-learning potential for cubic TiAlON through iterative active learning focused on structural, mechanical, and diffusion-related configurations. The potential reproduces DFT energies and forces with RMSE values of 4.08 meV/atom and 0.178 eV/Å, respectively. Independent validation gives normalized mean absolute errors of 2.2–3.5% for the elastic moduli, an energy RMSE of 5.8 meV/atom, and an RMSE of 0.25 eV for point-defect formation and reconstruction energies. Monte Carlo simulations show that chemically random TiAlON configurations are both higher in energy and softer than structures exhibiting moderate Ti/Al partitioning. Warren–Cowley parameters connect the developing chemical short-range order with an increase in elastic modulus. At higher vacancy concentrations, this ordering becomes coupled to metal-vacancy clustering and the displacement of predominantly Al atoms towards tetrahedral sites, producing tetrahedrally reconstructed, vacancy-rich aggregates. Nanosecond-scale molecular-dynamics simulations at 2000 K reveal that these reconstructed motifs remain dynamic, forming transient aggregates and enabling Al motion between neighbouring tetrahedral sites. Together, the results identify chemical short-range order and vacancy-mediated reconstruction as coupled mechanisms governing the mechanical response and high-temperature atomic transport in cubic TiAlON.

Speaker: Pavel Ondračka (MUNI)

10 Effects of Hydrogen Occupation on the Magnetic and Electronic Properties of LaNi5- and LaNiSn-Based Hydrides: A DFT Study

Solid-state hydrogen storage has become an attractive alternative to conventional hydrogen storage options, such as compressed gas or cryogenic liquid hydrogen, due to its higher operational safety, lower energy demands, and high volumetric hydrogen density. This type of storage is based on the incorporation of hydrogen atoms into the lattice of a host compound, forming metal hydrides.

LaNi$_5$ represents a well-known candidate for solid-state hydrogen storage, providing favourable storage capacity, practical plateau pressure, and ease of activation. Its crystal structure contains multiple interstitial sites available for hydrogen occupation. Using Density Functional Theory (DFT) calculations, we investigated the influence of different hydrogen lattice-site occupations and hydrogen content on the magnetic and electronic properties of LaNi$_5$-based hydrides. Pristine LaNi$_5$ exhibits weak ferrimagnetic behaviour. We found that there is no direct relationship between the total hydrogen content and the resulting magnetic properties. Instead, the magnetic properties are strongly governed by the type of occupied hydrogen site, which is related to the hybridization between Ni $3d$ and H $1s$ states.

The effect of hydrogen incorporation was also investigated in the related LaNiSn system, particularly in the stoichiometric intermetallic LaNiSn compound, for which the available information in the literature remains limited. Bader charge analysis was employed to investigate the redistribution of electron density upon hydrogen incorporation and to provide insight into the changes in the chemical environment of La, Ni, Sn, and H. The obtained results contribute to a better understanding of the relationship between hydrogen-site occupation, electronic structure, and magnetic properties in LaNi$_5$- and LaNiSn-based hydrides.

Computational resources were provided by the e-INFRA CZ project (ID: 90254), supported by the Ministry of Education, Youth and Sports of the Czech Republic. These resources were utilized through IT4Innovations National Supercomputing Center, MetaCentrum as well as CERIT Scientific Cloud. This work was created as part of the project No. CZ.02.01.01/00/22_008/0004631 "Materials and technologies for sustainable development'' within the Jan Amos Komenský Operational Programme, financed by the European Union and from the state budget of the Czech Republic.

Speaker: Kateřina Dočkalová

11 Use of open data for training machine-learning interatomic potentials

With machine learning and its use in science rapidly growing in popularity, the need for high-quality training data is increasing. Most researchers however train their models either on their own data or on curated databases. With the growing emphasis on open science, a large amount of data from other researchers is now openly available, but such data often come without any guarantee of quality, thus its suitability for machine learning is uncertain.

In this work, we assess the quality of the data in NOMAD, the largest open materials simulation database, and its practical applicability for training machine-learning interatomic potentials for atomistic simulations. We present a workflow designed to tackle several challenges associated with the NOMAD data: automatically filtering out results with low numerical accuracy, deduplicating structures in the training data, and combining results coming from multiple DFT implementations with different total energy offsets.

With this workflow, we have successfully trained silicon-based potentials using only simulations from NOMAD as training data. The resulting potential predicts phase stability at a level comparable to state-of-the-art potentials, while also accurately describing large-scale atomic systems, even at high temperatures. This demonstrates that using open data can significantly reduce the time and costs required to generate suitable training datasets for machine-learning interatomic potentials.

Speakers: Pavel Ondračka (MUNI), Šimon Kratochvíl

12 Roofline Benchmarking and CPU-to-CPU Variability of Barbora NG

Barbora NG is the latest CPU-based supercomputer operated by IT4Innovations, consisting of 144 nodes, each equipped with two Intel Xeon 6952P processors. Benchmarking its sustainable performance is therefore an important step in assessing its energy efficiency and identifying potential performance bottlenecks of this new HPC platform.

In this work, we utilize a Roofline benchmark with a set of predefined arithmetic intensities covering both memory-bound and compute-bound regimes. The benchmark is instrumented with the MERIC Toolkit, enabling monitoring of core and uncore frequencies together with power draw. This provides a more detailed analysis of performance behavior and potential bottlenecks beyond the conventional metrics of computational performance in FLOP/s and memory throughput in GB/s. The benchmark results and MERIC measurements are further used to quantify CPU-to-CPU variability in both performance and power draw across the Barbora NG HPC platform.

Speaker: Daniel Velicka (IT4Innovations, VSB-TUO)

13 Modeling of High-Quality Electron Beam for a Compact EUV Free-Electron Laser

Laser-plasma accelerators (LPAs) can produce electron beams with high peak current and low emittance, making them attractive drivers for compact free-electron lasers (FELs). However, the large angular divergence and energy spread of LPA beams make efficient beam transport and FEL amplification challenging.

In this work, we simulate the generation of an LPA electron beam with the particle-in-cell code OSIRIS, using shock injection as the injection scheme. The beam is then used to design a compact beamline for FEL operation in the extreme ultraviolet (EUV) regime, in which the FEL reaches saturation within a single undulator module. Full start-to-end simulations, from plasma acceleration through beam transport to FEL radiation, were performed to evaluate the performance of the complete system.

Our results show that shock injection can deliver electron beams of sufficient quality to drive a compact FEL, supporting the development of LPA-based light sources and their applications. These progress are critical for future FEL facilities at ELI ERIC in the Czech Republic and for the EuPRAXIA project.

Speaker: Mihail Miceski

14 Multi-Level Compression for Direct Volume Rendering

Efficient visualization of volumetric data produced by large-scale simulations requires compression to limit memory usage. We propose a lossy, multi-level compression technique designed for direct volume rendering of large scalar fields. Our approach represents the input volume as a hierarchy of grids containing polynomial coefficients, which adaptively approximate the fine-scale features found in the original voxel data. To enable efficient parallel construction, we employ Morton-order processing together with OpenVDB, and convert the resulting structure to NanoVDB for efficient sampling. We evaluate the proposed data structure by integrating it into Cycles renderer and testing it on several representative CFD and astrophysical datasets. The results demonstrate compression rates of 7.1× to 173× relative to uncompressed NanoVDB, with compression taking approximately 7 seconds per billion voxels. These results indicate that multi-level polynomial compression provides a practical balance between reducing memory consumption, preserving visual fidelity, and maintaining random access performance for large volumetric datasets in a Monte Carlo renderer.

Speaker: Markéta Faltýnková

15 Effective solution of differential equations on highly parralel architectures

Ordinary differential equations (ODEs) are used for simulating and modelling real-world problems. Such problems consist of thousands to millions of equations. Nowadays, the graphical processing units are massively parallel devices that are used to accelerate many computations. The aim of this paper is to show the possibilities of solving such systems on highly parallel systems, namely GPUs.
This paper implements the Modern Taylor Series Method (MTSM) for linear problems together with state-of-the-art methods: Euler and four-stage Runge-Kutta. The implementation is done using C++ with OpenMP offloading to GPU, which compiles code for both CPU and GPU and allows the parallel utilization of GPU resources. The limitation to linear problems allows for straightforward computation of higher-order derivatives, allowing the MTSM for arbitrary precision limited only by the precision of used data types.
The modelled problem for testing methods is a signal travelling through a telegraph line, which can scale to an arbitrary number of equations. The telegraph line problem is then used to analyse the behaviour on GPU in comparison to CPU behaviour and evaluate whether GPU acceleration was beneficial.
The results show that the GPU architectures can be beneficial.

Speaker: Petr Bém

16 Simulation Model of the Mitochondrial Ribosome

Mitochondrion is a cellular organelle responsible for several crucial biochemical pathways. One of the most important is the respiratory chain followed by oxidative phosphorylation (OxPHOS), which is essential to produce cellular energy in the form of ATP. The reactions of the respiratory chain are carried out by respiratory chain complexes (RCCs). The genetic information for the synthesis of several essential RCC subunits is encoded in the mitochondrial genome (mtDNA). Mitochondrial ribosomes (mitoribosomes) are biomacromolecular complexes located within mitochondria. Mitoribosomes are responsible for performing the final step of mtDNA expression - mitochondrial translation which takes place isolated from the cytosolic translation. Proper mitoribosomal function directly affects one of the central processes of cellular metabolism. Therefore, a detailed understanding of mitoribosomal behaviour is of great importance with potential applications in cancer therapy and longevity.

Here we present a full all-atom simulation model of the human mitoribosome in explicit solvent. To the best of our knowledge, this is the first [K(2.1][RK2.2]simulation model of a mitoribosome completed under such conditions. We use the simulation model to describe the dynamics of the selected mitoribosomal parts. We focus on comparison of the human mitoribosome with a bacterial ribosome, as these two ribosomes share a lot of structural similarities and evolutionary origin.
The simulation model was validated against the available cryo-EM model using basic analytical metrics RMSD, RMSF, and radius of gyration. According to RMSD and RMSF values we have revealed the most static and the most flexible regions of the mitoribosome.

Overall, our model opens a way for studies of mitochondrial ribosome dynamics at atomic detail.

Speaker: Rudolf Kvasňovský (University of Chemistry and Technology)

17 Anharmonic Effects in Tin Studied by MLIP

The $\alpha$–$\beta$ phase transition in tin represents a challenging problem for atomistic simulations due to the small free-energy difference between the two phases and the importance of finite-temperature effects. Accurate evaluation of these effects requires simulations beyond the time and length scales typically accessible by first-principles methods. Machine-learned interatomic potentials (MLIPs) provide an efficient alternative, enabling large-scale molecular dynamics simulations with near-DFT accuracy at a fraction of the computational cost of DFT.

In this work, an Atomic Cluster Expansion (ACE) potential trained on density functional theory data is employed to investigate the thermodynamic properties of $\alpha$- and $\beta$-Sn. The temperature dependence of their free energies is studied using both the quasi-harmonic approximation and thermodynamic integration. Comparison of these approaches allows us to evaluate the role of anharmonic contributions and their influence on the predicted phase stability.

Anharmonic free energies are calculated using the Calphy workflow based on molecular dynamics and thermodynamic integration. The free energies of both phases are obtained by reversible switching between the ACE description and suitable reference systems, followed by temperature integration over the investigated temperature range. This approach provides access to the full anharmonic free energy beyond the quasi-harmonic approximation and enables a direct comparison of the thermodynamic stability of $\alpha$- and $\beta$-Sn.

Speaker: Petr Sestak (Academy of Sciences of CZ, Institute of Physics of Materials)

18 Momentum-Resolved Energy-Loss Spectroscopy of Molybdenum Disulfide: First-Principles Interpretation of Experimental Spectra

Electron energy-loss spectroscopy (EELS) is an experimental technique used in transmission electron microscopy to probe single-particle and collective excitations, including excitons and plasmons. Momentum-resolved EELS goes beyond studying the energies of these excitations by revealing their dispersion as a function of momentum transfer, an important physical quantity. Experimental spectra reveal these effects, however, their full interpretation requires a quantitative theoretical description. For this reason, we employ computationally demanding first-principles calculations based on density functional theory (DFT) and many-body perturbation theory (MBPT). We focus on molybdenum disulfide, a representative of transition-metal dichalcogenide, with important optoelectronic properties, including an indirect-to-direct band-gap transition. Its ground-state properties are calculated using DFT and MBPT, while the energy-loss function is currently computed using time-dependent DFT. This work demonstrates how high-performance computing can connect experiment and theory, helping to uncover the underlying electronic properties of low-dimensional materials.

Speaker: Michal Štulpa (Department of Condensed Matter Physics, Masaryk University)

19 Laser Wakefield Acceleration in Discharge-Generated Plasma Channels

Laser-plasma accelerators (LPAs) have demonstrated the capability to produce high-energy, high-quality electron beams over compact distances. A key challenge is to maximise the coupling efficiency between the laser pulse and the accelerating plasma structure. One promising approach is to guide the laser pulse through a preformed plasma channel, maintaining the laser focus over extended propagation distances and thereby enabling sustained electron acceleration. This technique has enabled the acceleration of electrons to energies up to 10 GeV through laser guiding over several centimetres.
In this work, we investigate laser guiding in plasma channels generated by high-voltage electrical discharges in gas-filled capillaries for high repetition rate applications. We examine the influence of both the plasma channel properties and the laser parameters on guiding performance. Furthermore, three-dimensional Particle-In-Cell (PIC) simulations are employed to study the laser-plasma interaction and the resulting electron beam acceleration within these guided plasma structures.

Speaker: Alex Whitehead (ELI-ERIC (ELI-Beamlines Facility))

20 Radiation resistant refractory high entropy alloys

Refractory High Entropy Alloys (RHEAs) such as the BCC NbMoTaW system represent a promising class of structural materials for advanced nuclear reactors due to their exceptional physical and mechanical properties. In nuclear environments, structural materials face severe irradiation damage, which is driven by the accumulation of microscopic irradiation defects such as displacement-induced vacancies and transmutation-produced helium (He) atoms. Because He has an extremely low solubility in solids, it aggregates into bubbles, causing severe material degradation such as swelling, blistering, surface roughening, and embrittlement. Since He bubble formation is directly dictated by point defect energetics and kinetics, this study utilizes Density Functional Theory (DFT) to investigate the formation, and migration of monovacancies, as well as interstitial and substitutional He defects formation in the NbMoTaW matrix. The DFT calculations revealed that the formation energies for all investigated point defects exhibit wide distributions driven by the severe lattice distortions and large differences in the local chemical environments characteristic of HEAs. Furthermore, He atoms strongly prefer occupying substitutional positions over interstitial ones, with the octahedral position being energetically favored over the tetrahedral position among the interstitial sites. Crucially, the formation energy distributions of the point defects show significant overlap, indicating a strong tendency for defect recombination that potentially suppresses and reduces the overall formation of He bubbles. Additionally, the migration energies display a wide distribution, revealing that vacancies follow distinct, preferential pathways through the alloy matrix. Ultimately, these results give preliminary insights into the behaviour of point defects in NbMoTaW RHEA and their further impact on He bubbles formation and diffusion.

Speaker: Mr Oussama Madyouni (Faculty of Electrical Engineering - CTU in Prague)

21 Blood Cell Membrane Models for Nanoparticle Delivery Systems

Medical drug delivery is a rapidly developing field, including a wide range of nanocarrier used to increase the drug's amount in the target cells and decrease its potential side effects. However,a precise targetting of the delivery to other organs than liver remains a challenge. A novel polymer-based i-Bind nanocarrier has been shown to hitchhike on blood cells, with different affinities to red and white blood cells.

All atom molecular dynamics simulations have been used to compare the red and white blood cell model membranes and their interactions with the carrier nanoparticle polymer and its i-Bind coating.

The i-Bind coating interacted with both membranes in a non-specific manner, with a preference for hydrogen bonds with the phospholipid phosphate groups in both studied membrane models. However, i-Bind coating showed preferential binding to the white blood cell membrane. This is in contrast with experimentally measured data, suggesting that i-Bind-lipid interactions are not the most significant contribution to the overall behaviour, and that blood cell surface proteins and the total number of each cells in blood must also be taken into account.

Speaker: Petra Čechová (Regional Centre of Advanced Technologies and Materials, Czech Advanced Technology and Research Institute (CATRIN), Palacký University Olomouc: Olomouc, CZ)

22 Hardware drift and the accuracy of molecular properties from quantum computers

Quantum chemistry on quantum hardware requires estimating expectation values from repeated circuit measurement. Recent hardware results suggest this choice is less innocent than it looks: beyond a certain point, additional shots stop improving accuracy, plausibly due to hardware-induced drift of parameters.¹ Existing adaptive stopping strategies monitor the raw output distribution and assume the device is stationary during a run, an assumption that has not been tested on real hardware.²

Using a few small molecules as a benchmark, we measure how drift propagates into the electronic energy and into atomic forces. Both are obtained from the same measured reduced density matrices, but the former is a direct expectation value, whilst the latter requires further classical processing of those matrices, so the two are expected to respond differently to drift. Building on this, we develop a stopping criterion that terminates measurement once further shots no longer improve accuracy and evaluate it against fixed-budget and distribution-based alternatives on identical recorded shot streams. All measurements will come from the VLQ quantum computer based in Ostrava.

_{1. Olarte Hernandez, R. O.; Ziems, K. M.; Kjellgren, E.; Kongsted, J.; Coriani, S.; Sauer, S. P. A. Analytical Nuclear Gradients and Hessians on Quantum Hardware via Orbital-Optimized VQE with Error Mitigation. arXiv:2608.08758, 2026.}

_{2. Bisicchia, G.; Bocci, A.; Pimentel, E.; Brogi, A. StableShots: Online Shot Stopping for Quantum Circuit Execution. arXiv:2606.22170, 2026.}

Speaker: Przemyslaw Grenda (Charles University)

23 Bringing Regional Traffic Simulation to HPC

Regional traffic simulation supports mobility analysis and traffic forecasting by generating detailed vehicle movement data under different demand and road network scenarios. Producing these datasets requires large numbers of vehicles to be simulated and routes to be recalculated as traffic conditions change. Repeating the simulations with different settings creates two computational challenges: reducing the cost of individual route calculations, and processing large numbers of routing requests efficiently.

We use RUTH, a mesoscopic traffic simulator developed at IT4Innovations, to run regional simulations on high-performance computing (HPC) systems. Our Central Bohemian case features a road network of 194,585 nodes and 478,517 road segments and simulates up to 936,000 vehicles during the morning and afternoon period.

To reduce the cost of individual rerouting requests in this large use case, we are evaluating spatial and road class filters that restrict the road segments searched. Spatial filters limit the search to an area around the origin and destination, while road class filters exclude lower-priority roads from parts of the search. As an initial evaluation, we tested the filters on a representative sample of 50,000 origin–destination pairs from the Central Bohemian region. The ellipse filter reduced the median query time from 916.4 ms to 422.0 ms while still finding routes for 96.94% of the pairs. The hierarchical filter reduced the median query time further, to 107.1 ms, but found routes for only 86.38% of the pairs. This illustrates a trade-off. Tighter filters speed up queries but leave a small fraction unresolved, which still requires a less restricted search.

To process the large number of routing requests, RUTH uses the Asynchronous Communication and Execution library (ACE) to distribute them across HPC compute nodes through MPI or ZeroMQ. In weak scaling experiments on Karolina, we increased the workload from 1 million requests on 1 node to 16 million requests on 16 nodes while keeping the number of workers and requests per node constant. Elapsed time increased by only 1.8%, corresponding to approximately 98% weak scaling efficiency. Both backends therefore handled sixteen times as many routing requests with little increase in elapsed time when given proportionally more compute nodes.

These results address the two sides of the routing challenge. They reduce the cost of individual queries while scaling the overall workload with the available compute nodes. Next, we plan to implement and validate less restricted fallback searches for unresolved queries, evaluate the filters within full regional simulations, and assess how well the combined approach scales to larger geographical areas and vehicle populations.

Speaker: Pavlína Smolková (IT4Innovations (VSB-TUO))

24 First principles study of chalcogen vacancy effect on the optoelectronic and photocatalytic properties of transition metal dichalcogenides monolayers

In working conditions, chalcogen vacancies spontaneously occur in two-dimensional transition metal dichalcogenides (TMDCs) monolayers, affecting their optoelectronic and photocatalytic properties. To study how chalcogen vacancies affect such properties, we use quantum mechanical calculations considering prototypical MX2 (M = Mo, W, X = S and Se) TMDCs monolayers. Structural optimisations show that M-X bond lengths about a vacancy are different compared to the bond lengths in the pristine structure. Band structure calculations reveal that the introduction of vacancies produce electronic states about the Fermi level, hence resulting in the reduction of the band gap. Work function and electrostatic potential calculations show that the introduction of vacancies induce an asymmetry in the electrostatic potential facilitating the charge separation; such feature is absent in a pristine monolayer. All the considered defective structures are capable of performing hydrogen evolution reaction, while co-catalyst is required to perform oxygen evolution reaction when used for water splitting. WS2 and WSe2 defective monolayers can serve as an efficient photocatalytic material for reducing CO2 into useful chemical products. The presented results show that vacancy-containing TMDCs monolayers own photocatalytic capabilities compared to the pristine counterparts, thus showing that de-
fective TMD monolayers have prospective applications and should not be regarded as flawed products to be discarded. Finally, the results might constitute guidelines for the experimental synthesis of vacancy-engineered MX2 monolayers for optoelectronic devices and photocatalytic applications.

Speaker: Tahir wahab (Czech Technical University in Prague)

25 HPC-Enabled Real-Time Monitoring of Aseptic Gene Therapy Manufacturing Using Computer Vision

Maintaining consistent quality in pharmaceutical manufacturing is especially challenging in the manual production of gene therapy medicinal products, where operator interactions within isolator environments play an important role. Continuous monitoring can provide additional information about the manufacturing process and support its documentation, evaluation, and traceability. This work focuses on developing a computer vision system to support real-time monitoring and analysis of manufacturing processes in an isolator environment.

The core of the work is the development and training of deep learning models for detecting laboratory objects and DataMatrix codes. Dedicated image datasets were created for both tasks, containing images acquired under varying conditions representative of the isolator environment. YOLO-based models are used to detect selected laboratory objects and DataMatrix codes in camera images. DataMatrix code detection is combined with code reading to retrieve the encoded information. The three-dimensional position of detected laboratory objects can be determined using the camera system. The obtained 3D coordinates provide a basis for subsequent tracking of individual objects and monitoring their movement within the isolator.

Training and optimizing detection models require processing large image datasets and repeated evaluation of different model configurations. High-performance computing resources are therefore used to accelerate computationally demanding training and experimentation, allowing multiple model configurations and training parameters to be efficiently evaluated. The trained models can subsequently be integrated into the real-time monitoring system for continuous processing of image data acquired within the isolator environment.

By combining laboratory object detection, DataMatrix code detection and reading, 3D position estimation, and real-time image processing, the proposed system provides information about the objects present in the monitored area, their identification, and their spatial position. The obtained information can serve as a basis for subsequent object tracking and movement monitoring, contributing to automated documentation and traceability of the manufacturing process. Further development of the system will focus on computer vision-based monitoring of the process for gene therapy medicinal products, including the recognition and analysis of operator activities.

Speaker: Nikola Cudráková

26 A Comprehensive Stability Classification and Dynamical Analysis of a Brain-Inspired Coupled Rössler-Type System

A deeper understanding of how parameter variations and initial-condition uncertainty influence the behavior of nonlinear dynamical systems is essential for reliable stability and sensitivity analysis. Nonlinear systems can exhibit substantially different responses under changes in system parameters and initial conditions, making it important to systematically identify the factors governing these changes and assess the robustness of the observed behavior. This study investigates a three-dimensional coupled Rössler-type system using a combination of parametric and ensemble-based analyses. The effects of different system parameters, coupling strengths, and initial conditions are examined to characterize dynamical behavior, stability, sensitivity, and extreme responses. The results demonstrate that coupling strength significantly influences the organization of system responses, while (c_1) emerges as an important parameter governing stability and response characteristics. The study provides a systematic framework for understanding parameter-dependent behavior and initial-condition sensitivity in coupled nonlinear dynamical systems.

Speaker: Haseeba Sajjad (IT4Innovations)

27 Biased-exchange metadynamics of nascent peptide-accompanied rearrangements of the ribosome catalytic center

Nascent peptides can regulate gene expression while they are still being synthesized, using interactions within the ribosomal exit tunnel to interrupt elongation. The Vibrio export-monitoring peptide (VemP) links protein export to translational control. Its compact C-terminal helix lies next to the peptidyl-transferase center (PTC), where peptide bonds are formed, and constrains conserved nucleotides implicated in aminoacyl-tRNA accommodation. Cryogenic-electron microscopy resolves two alternative conformations of nucleotide U2506, referred to as A and B, within the arrested complex. Previous studies suggested that these states support distinct peptide-ribosome interactions, but their relationship to peptide relaxation and PTC remodeling remains unresolved.
We ask which PTC rearrangements accompany reduced VemP helicity and how they differ between the two identified U2506 conformational states.
We used bias-exchange well-tempered metadynamics to sample the C-terminal 11-residue helical VemP segment in a ribosomal tunnel model initialized separately from A and B. The peptide remained linked to P-site tRNA. Analysis of unbiased replica configurations related peptide helicity, extension, and tunnel position to nucleotide fluctuations and structural changes of the PTC related to the release of translational arrest.
Reduced VemP helicity was associated with peptide extension in both systems, but substantially greater U2584/U2585 remodeling in A. These nucleotides also exhibited larger conformational fluctuations in A. Within this ensemble, lower helicity was associated with peptide displacement away from the PTC and U2584/U2585 conformations closer to the substrate-accommodated reference. U2506 nevertheless retained distinct conformations, and steric overlaps with the reference substrate persisted. Our findings suggest that VemP relaxation can accompany selective PTC remodeling without an A-site substrate, while complete restoration of substrate-compatible geometry requires additional rearrangements that likely happen before the VemP segment is fully synthesized.

Speaker: Josef Cikhart

28 Tuning magnitude and direction of lattice thermal conductivity in transition metal dichalcogenide heterobilayers

We investigate the nanoscale mechanisms determining in-plane lattice thermal conductivity (LTC) of pristine and W-doped MX$_2$-M$^{\prime}$X$^{\prime}_2$ transition metal dichalcogenide heterobilayers in both phonon and relaxon bases. For the W-doped systems, we evaluate the anharmonic force constants using machine-learning potentials trained on first-principles data. Pristine heterobilayers exhibit isotropic LTC with preserved ordering across temperature. Under nearest-neighbour chalcogen substitution the LTC follows the ordering Te < Se < S; relaxon analysis further links the LTC and viscosity to harmonic phonon descriptors including phonon mode layer localisation and cophonicity. A sufficiently large mass contrast is required to induce layer localisation of the transport-relevant phonon modes, leading to enhanced LTC. The viscosity, which correlates positively with LTC, is associated with the low-cost cophonicity descriptor. Doped heterobilayers exhibit reduced and anisotropic LTC across all temperatures, while the aforementioned descriptors no longer correlate with the LTC. We find that the direction of maximum LTC can in principle be modulated by composition and temperature, although future work is required to establish robustness of the LTC magnitude and anisotropy trends. Thanks to its general formulation the analysis protocol can be readily extended to other van der Waals heterostructures, and the descriptors may be implemented in high-throughput engines to identify promising layered materials with tailored thermal transport characteristics.

This work was co-funded by the European Union under the project “Robotics and advanced industrial production” (reg. no. CZ.02.01.01/00/22_008/0004590), by the Czech Science Foundation project No. 24-12643L, and by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID: 90254).

Speaker: Mr Elliot Perviz (Czech Technical University)

29 Case study: Six months of work from heat flow simulation to the arrow of time experiment.

Learning with humility & Celebrating 10th Users' Conference of IT4Innovations

First FTA: meeting the Karolina computer with a simple code for heat flow in iron. Second FTA: testing the VLQ system as the first user, creating a quantum Hello World, and repeating the IBM arrow of time test. Third FTA: quantum simulations of iron atoms, changed to carbon because of a 24-qubit hardware limit. This connected work led to the final patent application.

Speaker: Ladislav Ničman

30 Privacy-preserving federated learning with light-weight attention improved CNNs for automated leukemia detection across distributed medical imaging

This research work describes a lightweight, secure, and interpretable federated learning framework for automatic leukemia classification, which identifies and addresses various problems regarding clinical data security and collaborative model building among partnering healthcare organizations. This framework employs a distributed learning paradigm that allows a number of healthcare facilities to work together to build a high predictive performance classification model while training the model without exchanging sensitive information about patient data, thus ensuring data privacy and methodological reproducibility. The proposed framework employs a lightweight attention-enhanced convolutional neural network (CNN) for the automated classification of leukemia cells to one of the four categories: benign, early, pre-leukemic, and pro-leukemic at only 0.14 s/batch. The global model at 3 clients achieves 95.70% test accuracy while at 5 clients and increased training rounds achieve 96.56% on test set on a weighted aggregation method. Additionally, for increased clinical interpretability and transparency explainable methods are used in this study.

Speaker: Muhammad Zeerak Awan (VSB - Technical University of Ostrava)

31 Influence of the N/P ratio on Lipid Nanoparticle's structure and behaviour

Lipid nanoparticles (LNPs) are drug delivery vehicles that are widely used and FDA-approved. Their main purpose is to safely and efficiently deliver a drug — most commonly a nucleic acid — to target cells, where subsequent processes such as vaccination or gene silencing take place. Despite their widespread use, encapsulation and delivery efficiency remain relatively low; improving this efficiency requires a better understanding of LNPs' inner organization, behavior, protein corona, and other related properties. One property increasingly discussed in the current literature [1] is the N/P ratio: the molar ratio between the ionizable lipid's nitrogen-containing head group (N) and the phosphate groups of the RNA backbone (P). This N/P ratio can influence particle size, internal organization, and consequently encapsulation efficiency.

Molecular dynamics simulations make it possible to observe the system dynamically over time, revealing how LNP behavior, inner organization, and related properties change across different N/P ratio values.
[1] Yulin Mo, Alexander F. A. Keszei, Shagun Kothari, Heyi Liu, Anni Pan, Paige Kim, Jiachuan Bu, Albert Kamanzi, David L. Dai, Mohammad T. Mazhab-Jafari, Juan Chen, Sabrina Leslie, Gang Zheng; Lipid-siRNA Organization Modulates the Intracellular Dynamics of Lipid Nanoparticles. Journal American Chemical Society 26 March 2025; 147 (12): 10430–10445. https://doi.org/10.1021/jacs.4c18308

Speaker: Adéla Chadalíková (CATRIN, VŠB-TUO)

32 Polypyridine Iron Complexes in Elektro- & Photocatalytic CO2 Reduction

Carbon dioxide (CO2) emissions of the industrial and power sector have been identified as one of the primary driving-forces of anthropogenic climate change.1 Hence, carbon capture technologies and the subsequent utilisation of captured CO2 are expected to surge dramatically w.r.t. their industrial importance. One promising utilisation is the reduction of CO2 to CO (CO2RR), a major component of industrially precious syngas.
In model studies2,3, iron complexes with the polypyridine ligand, DBPy-PyA, have proven to be active, affordable and stable scaffolds for the CO2RR under electrochemical as well as photochemical conditions. This warrants further investigation into this catalyst class to gauge its potential.
DBPy-PyA=(1-([2,2’-bipyridin]-6-yl)-N-([2,2’-bipyridin]-6-ylmeth-yl)-N-(pyridin-2-ylmethyl) methanamine
Thus, we engage in a theoretical study that aims to identify and tabulate key indicators for the performance of the polypyridine iron complexes, starting from their binding free energy of CO2 vs. H+ and CO, as well as substituent effects on the polypyridine skeleton.
This project (CO2RedFePP) has received funding from the European Union’s Horizon Europe research and innovation programme under the Marie Skłodowska-Curie grant agreement No 101209173.

Speaker: Florian Lemken (Institute of Organic Chemsitry Biochemistry, Czech Academy of Sciences)

33 Regional climate projections of non-CO2 forcers and their impact on climate and air quality using RegCM5-Chem

Despite the well-established role of CO₂ in climate change, significant uncertainties remain regarding the impacts of non-CO₂ radiative forcers, including aerosols, methane, and tropospheric ozone, due to their complex atmospheric interactions and strong regional variability. This presentation assesses the climate and air quality effects of these non-CO₂ radiative forcers using high-resolution regional climate simulations with the RegCM5-Chem model, which couples the atmospheric chemistry and its feedback on the climate system. Simulations were performed for mid-century conditions under two future climate scenarios, driven by the EC-Earth3-AerChem global climate model. The results suggest that ambitious emission reduction policies can bring substantial air-quality benefits.

Speaker: Natália Machado Crespo (Charles University)

34 Predicting polymorphs of pharmaceutical co-crystals via crystal structure prediction

Accurately investigating the polymorphism of a given compound is an important part of material science in various areas. The number of polymorphs a compound can form and the relationship between them greatly influences the applicability of the compound for industrial usage for example. However, polymorphism has even greater implications for the field of pharmaceuticals and drug forms, both in terms of patient safety and finances or intellectual property, as patents usually extend only to specified polymorphs. The phase behaviour of an active pharmaceutical ingredient (API) by itself can be complex enough, but other solid forms are also permitted for use as the basis for drug forms, including salts and co-crystals.
In co-crystals, two or more different chemical species bind together to form a single crystal structure. In terms of pharmaceutical co-crystals, they consist of either multiple APIs or more commonly of an API and some other excipient molecule commonly referred to as a co-former. These co-formers, like the API, must be safe to consume by humans. The usage of co-crystals over pure solid API forms can be facilitated by several factors – namely that a co-crystal is regarded as a different solid form and therefore does not constitute patent litigation and more importantly, while the medicinal effect of the API stays the same, the co-crystals solid form properties are different than that of the original pure API, such as solubility and bioavailability. In other words, these co-crystal forms offer immense flexibility in how a solid drug form is constructed, provided a given co-former truly forms a co-crystal with the target API.
In this work, the polymorphism potential of two known carboxylic acid co-crystals of the API agomelatine is studied via the methods of crystal structure prediction (CSP) in combination with a computational thermodynamic stability assessment for selected generated solid phases. In the basic setup of CSP, the immense phase space available in crystalline (co-)crystals is searched through efficiently using a particular structure generation methodology, followed by a cheap calculation of their energy. The best structures found during this process are then subjected to more rigorous re-optimizations and calculations to determine their stability in relation to each other. In our case, we chose the mol-CSPy program for structure generation, which performs a quasi-random algorithm to pack molecules it treats as rigid bodies into periodic structures based on predetermined space group symmetry. Since the molecules within these solid forms are flexible, we introduced a pre-generation step, identifying combinations of different conformations of API and co-former molecules that have a higher likelihood of forming dense co-crystals before performing a full detailed rigid body search only within these best-performers. Afterwards, we re-optimized the best individual structures using periodic density functional theory (DFT-D) calculations. To compute the relative thermodynamic stability, we used an approach combining DFT-D and density functional tight binding calculations within the so-called quasi-harmonic approximation to calculate solid-state vibrational contributions to the total thermodynamic properties, allowing us to rank the structures by stability at temperatures higher than 0 K.

Speaker: Petr Touš

35 Spectroscopy Enhanced by Two-Dimensional Materials for the Detection of Free Heme Associated with Intravascular Hemolysis

Two-dimensional materials are well established as highly effective substrates for Surface-Enhanced Resonance Raman Spectroscopy (SERRS) due to their strong chemical enhancement mechanisms and high molecular absorption, which enables the detection of biomolecules with improved sensitivity and selectivity. The iron-protoporphyrin ring complex, also known as heme, is the essential functional core of human hemoproteins. The release of labile heme into the bloodstream serves as a critical indicator of hemolysis and a driver of vascular complications in hemolytic disorders. However, current constraints in accurately quantifying labile heme within complex biological environments make it challenging to establish its exact contribution to vascular pathogenesis. To address this issue our goal is to conduct a comprehensive first-principles computational study of the Raman enhancement of labile heme after adsorption on pristine and boron-doped graphene substrates.

Towards this goal, we perform preliminary benchmark studies of the isolated heme, graphene substrates, and adsorbed heme-graphene systems to establish the validity of our methods. We calculate the phonon modes and raman spectra of the most stable conformational states and analyze the raman shift of the relevant heme vibrational modes. In future work, we plan to build toward a more realistic model by incorporating pathologically relevant axial ligands and solvent interactions of the labile heme, to ultimately provide experimentally verifiable (resonance) Raman spectra which could aid in the early identification of hemolysis.

This work was co-funded by the European Union under the project “Robotics and advanced industrial production” (reg. no. CZ.02.01.01/00/22_008/0004590), by the Czech Science Foundation project No. 24-12643L, and by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID: 90254).

Speaker: Angarag Munkhbat

36 Thermoelectric properties of defective scandium nitride nanostructures from first principles

The thermoelectric properties of scandium nitride (ScN) thin films vary significantly with the experimental conditions [1,2]. In this context, it is well established that structural defects and chemical impurities play a crucial role, affecting the thermoelectric performance of ScN nanostructures [3,4]. Nevertheless, the underlying transport mechanisms and how they are affected remain unclear. Reported measurements show large variability, but strategies to enhance efficiency often yield conflicting results [4,5]. Recent progress has demonstrated the possibility of significantly reducing thermal transport in ScN thin films [6,7]; however, achieving an effective improvement in thermoelectric efficiency requires that this reduction is not neutralized by a decrease in the power factor.
In this presentation, using the Landauer-Büttiker approach, we analyze how different types of imperfections alter electronic and thermal transport in the ScN lattice. We analyze two types of defects: those that are commonly present and those that can be intentionally introduced to tailor transport properties. Among the latter, we examine the effects of the intentional substitution of scandium atoms with other metals, such as niobium, strontium and tungsten [8]. Among the former, we identify two dominant classes with opposing effects: (i) contiguous nitrogen vacancies, which enhance electrical conductivity but reduce the magnitude of the Seebeck coefficient (S), and (ii) oxygen substitutions coupled with nearby stacking faults, which increase the magnitude of S while hindering electrical conductivity [9].

References:

[1]. Phys. Rev. Mater. 3, 020301 (2019).
[2]. J. Mater. Chem. C 4, 3905 (2016).
[3]. Phys. Rev. B 86, 195140 (2012).
[4]. Phys. Rev. B 97, 085301 (2018).
[5]. J. Appl. Phys. 115, 153508 (2014).
[6]. Appl. Surf. Sci. Adv. 25, 100674 (2025).
[7]. ACS Appl. Energy Mater., 5 (9), 11025 (2022).
[8]. L. Cigarini et al., in preparation.
[9]. Appl. Surf. Sci. Adv. 31, 100923 (2026).

Speaker: Luigi Cigarini (IT4Innovations, VŠB – Technical University of Ostrava)

37 Machine Learning for Predicting Lignin Solubility and Guiding Solvent Selection

Lignin is a structurally complex heteropolymer that has growing interest as a renewable feedstock for producing aromatic chemicals, biofuels, and bioplastics. One of the key obstacles in lignin valorisation is the compositional and structural variability of lignin obtained from different biomass sources, which makes it difficult to design efficient and generalizable processing strategies. Solvent fractionation is commonly used to narrow this heterogeneity, but choosing the right solvent is not straightforward, especially when lignin composition changes across feedstocks. In this work, we developed a computational approach to identify favourable and unfavourable organic solvents for lignin dissolution as a function of its monolignol composition. A library of 273 lignin polymer models was constructed using LigninBuilder, drawn from three biomass sources (birch, miscanthus, and spruce) to capture a wide and compositionally realistic range of syringyl, guaiacyl, and p-hydroxyphenyl monolignol ratios and interunit linkage distributions. Activity coefficients were calculated for each lignin–solvent combination across a large body of organic solvents using the conductor-like screening model based on segment activity coefficients (COSMO-SAC), which serves as a thermodynamically grounded metric for predicting dissolution tendency. We used this data, together with selected experimental solubilities, to train machine learning model (XGBoost) that allowed us to investigate how solvent properties, monolignol composition, molecular weight and interunit linkage distributions influence predicted lignin solubility. Leveraging the predictive power of this surrogate model alongside Kamlet–Taft solvent parameters, we mapped solubility behaviour across wide ranges of solvent polarity, basicity, and hydrogen bond donating ability, uncovering regions of solvent property space associated with favourable lignin dissolution regardless of feedstock origin. We also identified solvents that show selective dissolving behaviour depending on the monolignol composition of the lignin, suggesting their potential use in isolating fractions enriched in particular subunits. These outcomes provide guidelines for solvent selection that can be used to design fractionation processes tailored to lignin feedstocks of varying composition, as well as to identify solvents that promote lignin solubility regardless of monolignol ratios.
This work was supported from the LUABA24070 project of The Ministry of Education, Youth and Sports of the Czech Republic.

Keywords: Lignin solubility; Solvent selection; Machine learning; Monolignol composition

Speaker: Zaheer Abbas (University of Chemistry and Technology, Prague)

38 Effects of sequence on the conformational ensembles of ribosomal protein fragments

The ribosome is a ribonucleoprotein complex responsible for protein synthesis, with ribosomal proteins surrounding and interacting with the highly conserved RNA catalytic core. Some protein fragments located close to the peptidyl transferase center adopt characteristic backbone conformations in the ribosome. Our previous molecular dynamics simulations showed that isolated ribosomal protein fragments can sample similar conformations in aqueous solution (outside the ribosome), This suggests that part of their ribosome-associated conformational preference may be intrinsic to the peptide sequence.
We investigate whether these preferences are determined by the specific amino-acid order or arise primarily from peptide composition. Short fragments from ribosomal proteins uL2, uL3, uL13 and uL22 were simulated in explicit solvent together with shuffled variants. Multiple independent molecular dynamics simulations were performed for each peptide. Backbone conformational ensembles were characterized using residue-specific Ramachandran φ/ψ torsions and respective free-energy landscapes, allowing direct comparison between native and shuffled sequences.
We found that native and sequence-shuffled peptides showed broadly similar φ/ψ distributions despite differences in amino-acid order. Nevertheless, localized changes were observed at several residue positions, indicating that local sequence context can modulate backbone conformational preferences. These results suggest that the conformational behaviour of evolutionary old ribosomal protein fragments is relatively robust to sequence randomization. The conformations found in the ribosome were readily sampled in the aqueous solution, which implies that the conformational penalties associated with the fragment binding to the RNA were low.

Speaker: Jan Heblt (VŠCHT)

39 Atom-scale contributions to nanofriction response in transition metal dichalcogenides

We investigate the nanoscale friction behaviour of MX$_2$ monolayers (M = Mo, W; X = S, Se) on Au(111) and Ag(111) substrates with a silicon tip using classical molecular dynamics simulations with machine-learning-based force fields. This approach enables an accurate description of tip-surface interactions and friction mechanisms at the atomic scale. We observe a pronounced nonmonotonic dependence of the friction force on the applied normal load, with a corresponding coefficient of friction inversely proportional to the load. Analysis of lateral force signals and their spatial Fourier transforms reveals the coexistence of multiple sliding modes, including longitudinal sliding, lateral slip, and zig-zag motions. We show that the overall friction response is governed by the relative contributions of these motions. While the qualitative features of friction are largely substrate-independent, both the magnitude of friction and the balance between sliding modes depend sensitively on the substrate-monolayer combination. In particular, the Au/MoSe$_2$/Si system exhibits significantly reduced friction due to suppression of lateral slip motion. The analysis method employed in this study is also applicable to other material systems with different chemical compositions or sliding interface geometries. However, the peak positions and intensities are expected to vary with the chemical composition of the material, and consequently, their interpretation may differ across different systems.

Speaker: Mr Suresh Ravisankar (Czech Technical University in Prague)

40 A Quantum Perspective on Interval Power Flow

The integration of wind and solar generation leaves grid operators with input data known only within ranges. Interval power flow (IPF) seeks guaranteed bounds on bus voltages and line flows given interval-valued injections and loads. Classical approaches based on interval and affine arithmetic propagate these intervals through the power flow equations and suffer from the wrapping effect: dependencies between variables are lost, the computed hull is systematically wider than the true solution set, and the resulting operational margins are more restrictive than physics requires.

This work proposes two alternative routes exploiting quantum computers. The first obtains each bound by directly searching for the extremal operating point over discretized input intervals, encoded as a well-understood QUBO with the power flow equations as penalty terms and solved with QAOA on a 24-qubit IQM Star processor. The second casts IPF as a non-diagonal Hamiltonian, opening the problem to sample-based quantum diagonalization methods.

The poster presents both encodings, the qubit cost as a function of bus count and discretization resolution, and the benchmarking protocol for 3-bus to 5-bus test systems against established classical algorithms.

Speaker: Michal Belina (IT4Innovations, VSB – Technical University of Ostrava, 17. listopadu 2172/15, 708 00 Ostrava-Poruba, Czech Republic)

41 Density functional theory simulations of hydrogenated graphene supported by machine learning

Hydrogenation of graphene is a promising route toward tailoring its electronic properties, including opening the band gap required for transistor applications and advancing hydrogen-storage platforms. Industrial processing techniques—such as exposure to atomic beams, plasmas, or chemical functionalization—depend on the consecutive deposition of individual atomic species onto the surface. As local adatom concentrations increase, cooperative electronic interactions, localized lattice strain, and hydrogen-bonding networks dynamically alter the energy landscape for subsequent adsorption events. However, the precise atomistic mechanisms governing this dynamic coverage evolution, local cluster formation, and structural reorganization remain poorly understood.

Addressing this knowledge gap presents severe computational and experimental challenges. Direct real-time experimental observation of individual atomic adsorption events under processing conditions remains unfeasible. On the theoretical side, density functional theory (DFT) is limited to small unit cells or unrealistically ordered periodic configurations that fail to reflect the disordered, low-coverage topologies present at the beginning of a deposition process. Conversely, classical molecular dynamics (MD) lacks the accuracy required to model bond breaking and shifting atomic interactions as local concentrations change.

Integrating molecular dynamics with on-the-fly machine learning from DFT data provides the necessary framework to overcome these limitations. A machine-learned interatomic potential (MLIP) for hydrogenated graphene was developed using on-the-fly learning with Bayesian uncertainty quantification implemented in the Vienna Ab initio Simulation Package (VASP). The training dataset incorporated pristine graphene as well as structures containing Stone–Wales and monovacancy defects. The potential was iteratively refined through active learning and rigorous validation on adsorption, desorption, and cluster-formation processes.

The final MLIP reproduces DFT energies and forces with high accuracy while reducing the computational cost by two to three orders of magnitude. Reliable simulations were achieved for hydrogen coverages up to approximately 10%. By combining quantum-level precision with the efficiency needed for large-scale systems, this approach captures realistic, non-periodic surface arrangements, avoids unphysical static geometries, and maps the true dynamic evolution of low-ordered, low-coverage adsorbate systems under experimental conditions. These results demonstrate that machine-learning interatomic potentials provide an efficient, highly accurate framework for studying adatom dynamics and enable large-scale investigations into adsorption-driven modifications of graphene's electronic properties.

Speaker: Jiří Richter (Brno University of Technology, Faculty of Mechanical Engineering, Institute of Physical Engineering)

42 Many-body errors of quantum chemistry methods

Density functional theory (DFT) methods are widely used to study properties of molecular solids. This is due to the manageable computational cost of the calculations and acceptable accuracy. This is true also for calculating properties of polymorphs. However, it is also known that part of the success of DFT when applied to, e.g., calculation of energy differences between polymorphs comes from error cancellations. In our contribution we will present the results of our calculations in which we aim to obtain reliable energy differences between polymorphs and understand the errors of DFT and other widely used or emerging methods. Specifically, we use many-body expansion (MBE) of the binding energy to obtain binding energy contributions from dimers, trimers, and larger clusters at the reference-quality level. We then use these datasets to assess simpler schemes, including different dispersion-corrected DFT functionals [1]. Because the computational cost of the reference-quality methods quickly grows with the system size these tests are necessarily limited to molecules with up to around ten atoms. To study larger molecules, reduced scaling approaches, so-called local coupled cluster (LCC) methods, have been proposed. These offer a way to obtain reference MBE contributions. We will show that this is not straightforward and we will present how the precision of LCC needs to be controlled to obtain various MBE contributions for different expected levels of precision [2]. Finally, we will show how can numerical settings of the calculations affect the predicted properties. For example, how the choice of projector-augmented wave data sets in plane-wave codes changes the binding energies for different systems and how the choice can affect optimized structures [3].

Acknowledgements: This work is supported by the project Calculations of reference binding energies of polymorphs of molecular crystals (LUC24039) and by the project Advanced Multiscale Materials for Key Enabling Technologies (CZ.02.01.01/00/22_008/0004558), both supported by the Ministry of Education, Youth, and Sports of the Czech Republic (MEYS), co-funded by the European Union. We acknowledge the computational resources provided by e-INFRA CZ project (ID: 90254) supported by MEYS and by the Poznań Supercomputing and Networking Center.

References

• K. N. Pham, M. Modrzejewski, J. Klimes, J. Chem. Phys. 2023, 158, 144119; 2024, 160, 224101; 2025, 163, 224105.

• K. N. Pham, M. Modrzejewski, P. R. Nagy, J. Klimes, submitted

• S. Yourkhani, J. Klimes, J. Chem. Theory Comput. 2023, 19, 8871; in preparation

Speaker: Jiri Klimes (Faculty of Mathematics and Physics, Charles University)

43 Energy-optimized scaling laws for laser-driven electron accelerators

Laser wakefield acceleration promises compact electron accelerators for applications in medicine, industry, and fundamental science. Yet, despite rapid progress, accurately predicting the electron energy attainable in a given experimental configuration and the acceleration length required to reach it remains an open challenge. Here we use Bayesian optimization combined with advanced particle-in-cell simulation techniques to determine the maximum electron energy that a self-guided laser wakefield accelerator driven by a laser of a given energy and wavelength can produce. By systematically optimizing the accelerator performance across a range of laser energies and wavelengths, we derive energy-optimized scaling laws. These scaling laws yield the highest electron energy over the shortest acceleration length possible, are expressed solely in terms of laser energy and wavelength, and are accompanied by the complete set of laser and plasma parameters required to enable the scaling. The resulting scaling laws provide practical guidance for designing state-of-the-art laser wakefield acceleration experiments operating at their fundamental performance limits [1, 2, 3]. This work was supported by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID:90254).

[1] P. Valenta et al., Phys. Rev. Accel. Beams 28, 094601 (2025).
[2] P. Valenta et al., Mach. Learn.: Sci. Technol. 7, 025030 (2026).
[3] P. Valenta et al., arXiv:2608.08903 (2026).

Speaker: Dr Petr Valenta (Beamlines Facility, Extreme Light Infrastructure ERIC)

44 From Batch Jobs to AI Services: Training and Inference Workflows on HPC

AI workloads on typically involve two distinct usage patterns: scheduled training and fine-tuning jobs, and interactive model inference. Both require access to accelerator resources, but differ substantially in how these resources are allocated and used.

The AI Training App, built on the LEXIS Platform, provides a unified interface for both modes. For training, it simplifies data management, environment configuration and execution of AI workloads on the EuroHPC JU supercomputers, with prepared workflows for models such as YOLO and Whisper, and Unsloth. For inference, the integrated EXA4MIND AI Inference Service enables interactive access to models running on HPC resources through an OpenAI-compatible API, while handling the underlying resource and model management.

The poster presents the architecture and practical use of this unified approach, showing how higher-level interfaces can make both batch-oriented AI training and interactive inference available on HPC infrastructure.

Speaker: Tomáš Martinovič (IT4Innovations, VŠB - Technical University of Ostrava)

45 Quantum-Circuit Tensor Networks for Correlated Superconducting Nanostructures

Artificial nanostructures built from magnetic atoms, molecules, or quantum dots coupled to superconducting substrates offer a tunable platform for realizing correlated quantum states and, eventually, designer quantum materials. Extending such few-site structures toward larger artificial quantum materials quickly leads to strongly correlated many-fermion problems that challenge established classical methods. Building on a canonical transformation that maps these superconducting cluster problems onto particle-number-conserving fermionic Hamiltonians, this project investigates which quantum-circuit representations can efficiently capture the resulting correlated physics in one-dimensional clusters, in regimes related to the Hubbard and Heisenberg models.

Our current work focuses on one-dimensional systems using matrix-product-state-based quantum circuits (qMPS). In this representation, each site of the classical MPS is mapped onto a small quantum register whose qubits are split into physical and auxiliary groups, with the auxiliary qubits encoding the bond dimension logarithmically rather than polynomially, as in classical tensor networks. Sites are prepared and optimized through variational routines, including quantum singular value decomposition and reshape operations, that bring each site into canonical form, allowing a variational qMPS (vqMPS) scheme closely analogous to the classical density matrix renormalization group algorithm (DMRG), updating one site at a time via sweeps rather than optimizing the full system simultaneously. Preliminary implementations of this approach have already been tested on quantum simulators with promising accuracy.

We aim to extend this approach to larger and more strongly correlated 1D clusters, characterizing how circuit depth, gate count, and accuracy scale with system size, bond dimension, and correlation strength. Looking further ahead, this framework could in principle be extended toward more complex two-dimensional geometries, where correlations and entanglement structure become substantially richer.

Computational resources were provided by IT4Innovations National Supercomputing Center, the MetaCentrum, and CERIT-SC under the Project e-INFRA CZ (ID:90254) supported by the Ministry of Education, Youth and Sports of the Czech Republic. Access to CESNET storage facilities provided by the project e‑INFRA CZ under the program Projects of Large Research, Development, and Innovations Infrastructures (LM2023054) is appreciated. Financial support from the Czech Academy of Sciences via the Praemium Academiae awarded to Martin Friák, from the Institute of Physics of Materials, Czech Academy of Sciences, Brno, Czech Republic, is gratefully acknowledged.

Speaker: Jan Hojač (Institute of Physics of Materials, Czech Academy of Sciences, Brno, Czech Republic)

46 SELECTING VAN DER WAALS FUNCTIONALS FOR 2D MAGNETS: A DFT BENCHMARK FOR STRUCTURAL, ELECTRONIC AND MAGNETIC PROPERTIES OF CHROMIUM TRIHALIDES (CrX₃)

Two-dimensional van der Waals (vdW) magnets such as the chromium trihalides CrX₃ (X = Cl, Br, I) suit spintronic and magneto-optical heterostructures. First-principles methods struggle to describe them: strong intralayer covalent bonding coexists with weak interlayer dispersion, semi-local functionals overestimate interlayer spacing and underestimate exfoliation energy, and localized Cr 3d states need a Hubbard correction. We benchmark 17 exchange-correlation functionals spanning semi-local GGA (PBE, PBEsol) and their D3(BJ)-corrected variants [1], nonlocal vdW functionals (vdW-DF, vdW-DF2, optPBE-vdW, optB88-vdW, optB86b-vdW, rev-vdW-DF2, vdW-DF-cx [2], rVV10, PBE+rVV10L), and meta-GGA nonlocal functionals (SCAN+rVV10, r²SCAN+rVV10) [3], for bulk, monolayer, and bilayer CrX₃, using DFT+U (Liechtenstein) [4] in VASP [5].

All functionals match the experimental in-plane lattice constant a to within about 5%, so covalent Cr–X bonding is captured even semi-locally. The out-of-plane constant c is far more sensitive: dispersion-free functionals overestimate it by up to 16%, while nonlocal vdW functionals and dispersion corrections cut that to a few percent. Only PBE+D3BJ, optB86b-vdW, and vdW-DF-cx [6] keep errors below 1% for a and below 2% for c across all three compounds, with vdW-DF-cx giving the lowest average error (Figure 1).

Exfoliation energies are near zero or slightly negative for PBE/PBEsol and rise once dispersion is included, following CrCl₃ < CrBr₃ < CrI₃ as halogen polarizability grows. PBE+rVV10L binds noticeably more strongly than other nonlocal functionals, which otherwise resemble the simpler D3BJ correction.

Band gaps (CrCl₃: 2.4–3.7 eV; CrBr₃: 1.8–2.9 eV; CrI₃: 1.0–2.0 eV, bulk to monolayer) are consistently underestimated by semi-local and D3(BJ)-corrected GGA functionals. SCAN+rVV10 overshoots the experimental gaps considerably, while r²SCAN+rVV10 gives a better estimate at lower cost. Both capture the widening from bulk to monolayer and decrease from CrCl₃ to CrI₃.

Except for PBE+rVV10L in CrCl₃, every functional predicts a ferromagnetic monolayer with an out-of-plane easy axis, matching experiment. The MAE is small for CrCl₃ (−0.01 to −0.03 meV/Cr) and grows with halogen spin–orbit coupling, reaching about −0.5 meV/Cr for CrI₃ (Table 1). The exchange constant J is positive for all halides except under PBE+rVV10L and rises from CrCl₃ to CrI₃, pointing to stronger Cr–X–Cr superexchange as halogen polarizability grows. CrCl₃ sits closest to the ferromagnetic–antiferromagnetic crossover, so its J is most sensitive to the functional used.

Exfoliation energy, band gaps, and MAE depend strongly on the functional, but structural and magnetic trends across CrX₃ hold regardless. PBE+D3BJ matches the accuracy of the costlier nonlocal vdW functionals at lower cost, while vdW-DF-cx and optB86b-vdW remain most reliable for higher structural precision.

References
[1] S. Grimme, S. Ehrlich, and L. Goerigk, Journal of Computational Chemistry 32, 1456 (2011).
[2] K. Berland and P. Hyldgaard, Phys. Rev. B - Condens. Matter Mater. Phys. 89, 1 (2014).
[3] J. Ning et al., Phys. Rev. B 075422, 1 (2022).
[4] A. I. Liechtenstein, V. I. Anisimov, and J. Zaanen, Phys. Rev. B 52, 5467 (1995).
[5] G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996).
[6] J. Klimeš, D. R. Bowler, and A. Michaelides, Phys. Rev. B - Condens. Matter Mater. Phys. 83, (2011).

Speaker: SUBHASMITA RAY (POSTDOCTORAL RESEARCHER, CHARLES UNIVERSITY)

47 Kinetic study of shock formation in protostellar jets

Synchrotron radio emission has been detected from jet termination regions of young stars, indicating the presence of relativistic electrons. Non-thermal electrons are most probably accelerated through the Fermi I acceleration. However, thermal electrons in the jet are not energetic enough to traverse the shock. This brings up the need for an electron pre-acceleration mechanism. We perform Particle-in-Cell (PIC) simulations of these astrophysical shocks to study the Shock Surfing Acceleration (SSA) mechanism. We will present preliminary results of 1D simulations of perpendicular shocks in electron-proton weakly magnetised plasma.

Speaker: Jakub Gazdoš (FZU AVČR)

48 Influence of Antisite Defects on the Stability of Martensitic NiTi

The martensitic phase of NiTi has long posed a challenge for first-principles modelling due to the discrepancy between the experimentally observed monoclinic B19′ structure and the B33 structure predicted by density functional theory as the low-temperature ground state. Recent molecular-dynamics simulations have shown that vibrational entropy can stabilize the B19′-like phase at finite temperature, providing a possible explanation for this apparent discrepancy. However, the predicted stabilization occurs at temperatures substantially higher than those observed experimentally.

In this work, we investigate the role of antisite defects in the finite-temperature stability of NiTi martensite using large-scale molecular-dynamics simulations driven by a newly developed machine-learning interatomic potential. We show that antisite disorder markedly shifts the stabilization of the B19′-like structure towards lower temperatures, bringing the predicted structural behaviour significantly closer to the experimentally relevant temperature range.

Speaker: Petr Jaroš (Ústav termomechaniky AVČR v.v.i.)

49 Evaluating Generative Structural Models Against Multi-Microsecond Molecular Dynamics for Intrinsically Disordered Ensembles: Lessons from p53 TAD1

Previously, protein function was traditionally understood through the structure-function relationship and biological activity. However, a substantial portion of the human proteome contain intrinsically disordered proteins and regions that lack stable, globular folds. Instead these proteins function through a combination of “fly-casting” effects and transient binding structures which become ordered upon binding. Their structural heterogeneity make them difficult to characterize experimentally (CryoEM, X-Ray) or computationally (molecular dynamics). These proteins are of particular importance as they are involved in major disease pathways, including cancer and neurodegenerative diseases.

Recent advances in generative artificial intelligence have enabled protein structures and conformational ensembles to be generated at extraordinary speeds compared with conventional molecular simulations, but pharmaceutical drug discovery pipelines are hesitant to implement these techniques due to concerns surrounding their ability to reproduce unbound thermodynamic ensembles of highly flexible proteins. Building on our recently published work in PLOS Computational Biology, we benchmarked six state-of-the-art generative-AI models (AlphaFold3, Boltz-2, AlphaFlow, BioEmu, idpSAM, and IDPFold) using the disordered transactivation domain 1 of the tumor suppressor protein, p53, as a model system. The generative-AI were compared against solvent all-atom MD across four different forcefields, coarse-grained and CALVADOS simulations, temperature replica-exchange molecular dynamics (T-REMD), and multimodal experimental measurements from NMR, SAXS, circular dichroism and smFRET.

Our structural and thermodynamic analyses reveal substantial differences and challenges between the generative-AI models. General-purpose models, e.g. AlphaFold3 and Boltz-2x, strongly restrict the conformational ensemble and favor compact, highly helical states resembling experimentally determined p53-MDM2-bound states. In contrast, dynamic and IDP-oriented models, including BioEmu and idpSAM, more effectively reproduce extended conformations and broad free-energy basins observed in physical simulations. Nevertheless, reduced representation generative approaches remail less successful at recovering the atomic environments required to reproduce experimental NMR chemical shifts. Despite these initial successes, proper modeling of the protein through choice selection of forcefield and water models in an all-atom representation MD still produced the most accurate to experimental conformational ensembles.
Together, these results show that generative-AI can provide rapid exploration of the global conformational landscape of intrinsically disordered proteins, but that physics-based molecular dynamics simulations and high-performance computing remain essential for resolving their thermodynamics, local molecular structure, and therapeutically relevant interaction mechanisms. The best results were obtained from a combination of the generative-AI’s ability to freely jump across free energy barriers, while still restricted by short MD-physics based trajectories.

Speaker: Michael Bakker (Charles University)

50 Laser-driven acceleration of sub-GeV and GeV high-quality electron beams

Laser–wakefield acceleration (LWFA) [1] is a promising route towards compact high-energy electron accelerators. In this scheme, an intense ultrashort laser pulse propagates through a gas target, driving a plasma wave that can trap and accelerate electrons to relativistic energies over only a few centimeters. LWFA therefore offers a potential path towards compact electron sources for next-generation free-electron lasers (FELs). In this work, we investigate controlled electron injection into the plasma wave using two mechanisms showing promising beam quality in the recent studies: density down-ramp injection [2,3] and truncated ionization injection [4], aiming at high-quality laser-driven acceleration of electron beams at sub-GeV and GeV energies. Using quasi-3D particle-in-cell (PIC) simulations [5,6] together with post-processed interactive visualizations [7], we study the laser–target dynamics and optimization of the longitudinal plasma-density profile, including a density up-ramp in the acceleration region and an optimized exit down-ramp. The goal is to identify experimentally feasible regimes enabling efficient production of high-charge, low-divergence electron beams with favourable energy spread and emittance for FEL realization at the ELI Beamlines Facility [8]. Current work further focuses on adapting the laser and target parameters for initial experiments, supported by combined PIC and computational fluid dynamics (CFD) modelling.

References:
[1] T. Tajima et al., Phys. Rev. Lett. 43, 267–270 (1979)
[2] K. Feng et al., arXiv:2501.09916 (2025)
[3] J. Zhan et al., Phys. Rev. Research 8, 013207 (2026)
[4] S. Maity et al., Phys. Rev. Accel. Beams 29, 093601 (2026)
[5] J. Derouillat et al., Comput. Phys. Commun. 222, 351-373 (2018)
[6] R. A. Fonseca et al., LNCS 2331, 342-351 (2002)
[7] M. Matys et al., Photonics 12, 436 (2025); vbl.eli-beams.eu
[8] A. Molodozhentsev et al., in Proc. IPAC’24, MOPG32 (2024)

Acknowledgements

This work received funding from the European Union’s Horizon Europe research and innovation program, under Grant Agreement No.101079773 (EuPRAXIA Preparatory Phase), No.101188004 (PACRI—Plasma Accelerator systems for Compact Research Infrastructures). Portions of this research were carried out at the ELI Beamlines Facility, a European user facility operated by the Extreme Light Infrastructure ERIC. The computational time was provided by the supercomputers Sunrise of ELI Beamlines facility and Karolina of IT4Innovations - supported by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID:90254).

Speaker: Dr Martin Matys (ELI Beamlines Facility, The Extreme Light infrastructure ERIC)

51 Atomistic Insights Into Interaction of Doxorubicin With DNA: From Duplex to Nucleosome

Doxorubicin (DOX) is a widely used chemotherapeutic agent known for intercalating into DNA. However, the exact modes of DOX interactions with various DNA structures remain unclear. Using molecular dynamics (MD) simulations, we explored DOX interactions with DNA duplexes (dsDNA), G-quadruplex, and nucleosome. DOX predominantly stacks on terminal bases of dsDNA and occasionally binds into its minor groove. In the G-quadruplex, DOX stacks on planar tetrads but does not spontaneously intercalate into these structures. Potential of mean force calculations indicate that while intercalation is the most energetically favorable interaction mode for DOX in dsDNA, the process requires overcoming a significant energy barrier. In contrast, DOX spontaneously intercalates into bent nucleosomal DNA, due to the increased torsional stress. This preferential intercalation of DOX into regions with higher torsional stress provides new insights into its mechanism of action and underscores the importance of DNA tertiary and quaternary structures in therapies utilizing DNA intercalation.

Speaker: Andrea Nedělníková

52 Controlled Electron Injection and keV Betatron X-rays from Nanoparticle-Assisted Laser Wakefield Acceleration

Laser wakefield acceleration (LWFA) can generate compact relativistic electron beams, but its performance is often limited by uncontrolled self-injection. In this work, we use quasi-3D Smilei particle-in-cell simulations to study nanoparticle-assisted injection in helium plasma with lithium, aluminium, and copper nanoparticles. These nanoparticles act as localized plasma perturbations that modify the injection dynamics, producing material-dependent regimes ranging from weak perturbation to enhanced injection. We then focus on lithium, distinguishing helium plasma electrons injected by the nanoparticle-triggered wake from electrons born directly from the nanoparticle. Further optimization of the lithium case produces keV-range betatron X-ray emission and, to our knowledge, represents the first study of betatron radiation in nanoparticle-assisted LWFA. These results not only show that nanoparticles can control electron beam injection and quality,but also suggest a promising path toward reliable compact plasma-based X-ray generation for multidisciplinary applications.

Speaker: Alfred Mishi (Czech Technical University in Prague)

53 Magnetic field effect on acoustic wave propagation in cubic ferromagnets

The coupling between spin waves (magnons) and acoustic waves (phonons) is a crucial phenomenon with significant implications for spintronics, magnonics, and quantum materials research [1]. The advance of the computing technology allow us to use numerical methods to investigate magnon-phonon coupling beyond simplified models used so far. We show that a significant role in the magnetic field effect on acoustic wave propagation is played by its dispersion [2]. The k-wave vector dependence in magnetic field has a significant impact on the low-field regime, i.e. close to magnon-phonon resonance, and therefore, in contrary to the high-field regime, it can’t be neglected.

In our work, based on the example of bcc Fe, we perform a detailed analysis of spin waves to lattice vibrations coupled effects using numerical atomistic simulations and also demonstrate various k-wave vector dependences in the low magnetic field using our web-based interactive tool called VelCrys [3]. We analyze the role of the direction of the magnetic field in the change of sound wave propagation speed for the bcc Fe crystal for three types of acoustic waves (one quasi-compressional wave and two quasi-shear ones).

References:
[1] Rostyslav O. Serha, Carsten Dubs, Andrii V. Chumak, APL Mater. 14 (2026), 030901
[2] I. Korniienko, P. Nieves, D. Legut, Results in Physics, 73 (2025) 108264
[3] P. Nieves, I. Korniienko, A. Fraile, J.M. Fernández-Díaz, R. Iglesias, D. Legut, SoftwareX, 33 (2026) 102472

Speaker: Ievgeniia Korniienko (IT4Innovations)

54 Machine-learned interatomic potentials for fracture simulations in hydroxylated silica minerals

Flint, a cryptocrystalline form of silica, is known for its characteristic conchoidal fracture, which produces smooth, curved surfaces and exceptionally sharp edges. Although this behaviour has been exploited for millennia in the production of cutting tools, the atomistic mechanisms governing crack initiation and propagation in flint remain poorly understood. Direct quantum-mechanical simulations can provide the required atomistic accuracy but are limited to relatively small systems and short time scales, making realistic simulations of fracture computationally prohibitive.

To overcome these limitations, we develop a machine-learned Si–O–H interatomic potential trained on high-accuracy quantum-mechanical data, aiming to retain near-ab initio accuracy while enabling simulations at substantially larger length and time scales. Particular attention is given to hydrogen-containing environments, since natural flint contains both molecular water and hydroxyl groups that may modify local bonding and fracture behaviour. The training dataset therefore spans a broad range of relevant structural environments, including hydroxylated and hydrated configurations, different surface terminations, clusters, strained structures, amorphous silica and defects.

The potential will be systematically validated against quantum-mechanical reference data before being employed in large-scale molecular dynamics (MD) simulations. These simulations will be used to investigate crack initiation, bond-breaking processes and crack propagation in chemically heterogeneous silica. The resulting model should provide an atomistic framework for understanding the characteristic fracture behaviour of flint and, more generally, the role of hydrogen-containing environments in brittle fracture of silica-based materials.

Speaker: Valentína Berecová (Institute of Physics of Materials CAS, v. v. i.)

55 Simulation-Driven Optimization of Betatron X-ray Sources

Laser wakefield acceleration (LWFA) [1,2] is a promising approach for producing compact sources of high-energy electron beams and ultrashort X-ray pulses [3], offering a potential alternative to large-scale synchrotron facilities [4,5] for applications in imaging, materials science, and ultrafast science. One promising strategy for increasing the X-ray yield is the accelerator–radiator scheme[6], in which electrons are first accelerated in a low-density plasma stage and subsequently enter higher-density radiator, where enhanced transverse oscillations lead to stronger X-ray emission.
In this work, we investigate the optimization of such accelerator–radiator configurations using particle-in-cell simulations [7]. Due to the highly nonlinear nature of laser–plasma interactions and the large parameter space, we employ Bayesian optimization [8] to efficiently explore the dependence of X-ray generation on the radiator position and plasma density. The optimization targets the laser-to-X-ray conversion efficiency and identifies operating regimes where electron rephasing in the radiator significantly enhances betatron emission, achieving more than an order-of-magnitude increase in X-ray yield compared to the accelerator-only configuration.
These results demonstrate the potential of accelerator–radiator schemes for developing brighter compact X-ray sources while highlighting the role of efficient simulation-driven optimization in studying complex laser–plasma systems. The work contributes to the development of advanced betatron sources for experiments at ELI Beamlines [9] and illustrates the value of high-performance computing in accelerating their design and optimization.

References:
[1] T. Tajima and J. M. Dawson, Laser electron accelerator, Physical review letters 43, 267 (1979).
[2] A. Picksley et al., Matched guiding and controlled injection in dark-current-free, 10-GeV-class, channel-guided laser-plasma accelerators, Physical Review Letters 133 (2024) 255001.
[3] A. Pukhov, et al., Laser wake field acceleration: the highly non-linear broken-wave regime, Applied Physics B 74, 355 (2002).
[4] S. Corde, et al., Femtosecond x rays from laser-plasma accelerators, Reviews of Modern Physics 85, 1 (2013).
[5] F. Albert, et al., Laser wakefield accelerator based light sources: potential applications and requirements, Plasma Physics and Controlled Fusion 56, 084015 (2014).
[6] J. Ferri, et al., High-brilliance betatron g-ray source powered by laser-accelerated electrons, , Physical review letters 120, 254802 (2018)
[7] J. Derouillat, A. Beck, F. Pérez, T. Vinci, M. Chiaramello, A. Grassi, M. Flé, G. Bouchard, I. Plotnikov, N. Aunai, et al., Smilei: A collaborative, open-source, multi-purpose particle-in-cell code for plasma simulation, Computer Physics Communications 222, 351–373 (2018).
[8] A. Döpp, C. Eberle, S. Howard, F. Irshad, J. Lin and M. Streeter, Data-driven science and machine learning methods in laser–plasma physics, High Power Laser Science and Engineering 11, e55 (2023)
[9] U. Chaulagain, et al., Eli gammatron beamline: a dawn of ultrafast hard x-ray science, in Photonics, Vol. 9 (MDPI, 2022) p. 853.

Speaker: Dominik Čáp (Beamlines Facility, Extreme Light Infrastructure ERIC)

56 The stability and material properties of Zn5Sb4In2 computed by quantum mechanical calculations

Intermetallic compounds in the Sb-Zn system exhibit excellent thermoelectric properties, which can be further improved by indium alloying. Accurately describing the underlying In-Sb-Zn ternary phase diagram using phenomenological CALPHAD modeling relies on precise input datasets for all constituent phases. Among these, Zn5Sb4In2 represents the sole known ternary compound in the system. This work presents an ab initio study of the ground state of the ternary phase Zn5Sb4In2. Density functional theory (DFT) was employed to evaluate the equilibrium structural parameters, electronic properties. Additionally, the mechanical stability of the phase was established. Furthermore, dynamical stability was investigated using the Phonopy package. The calculated density of states revealed a narrow energy gap at the Fermi level, indicating the semiconducting character of Zn5Sb4In2. The calculated elastic properties confirm that Zn5Sb4In2 is mechanically stable, but phonon calculations revealed a dynamical instability of this compound at T = 0 K, consistent with its experimental observation only at elevated temperatures.

Acknowledgments
This study is a part of the project No. CZ.02.01.01/00/22_008/0004631 Materials and technologies for sustainable development within the Jan Amos Komensky Operational Program financed by the European Union and from the state budget of the Czech Republic. The authors acknowledge the Ministry of Education, Youth and Sports of the Czech Republic for the financial support of large research infrastructure INFRA CZ (ID:90254) at the IT4Innovations National Supercomputing Center.

Speaker: Martin Friák (Institute of Physics of Materials, Czech Academy of Sciences, Brno, Czech Republic)

57 Implicit solvation models for modern neural network potentials

Modern ML force fields approach the accuracy of Density Functional Theory (DFT) calculations for the properties of organic molecules and their reactions.[1,2] The vast majority of them do not model the effects of the environment, specifically the effect of the solvent.[3] However, correctly modeling interaction of molecules with the solvent is crucial for modeling biological systems and organic reactivity. Therefore, we set out to explore whether we could model these effects with electrostatics-based implicit solvation models, such as the COSMO model.[4] To that end, we reimplemented a linear-scaling variant of this model (ddCOSMO[5]) in the PyTorch framework. Also, to model non-electrostatic effects, we decided to reimplement the OpenCOSMO-RS solvation model[4,6] in a similar fashion. These can be then coupled to models that produce electrostatic potentials, such as MACE-POLAR[7] or MACE-MDP.[8] We will discuss the performance of the resulting hybrid models and outline possible improvements.

References

[1] Yuan, E. C. -.; Liu, Y.; Chen, J.; Zhong, P.; Raja, S.; Kreiman, T.; Vargas, S.; Xu, W.; Head-Gordon, M.; Yang, C.; Blau, S. M.; Cheng, B.; Krishnapriyan, A.; Head-Gordon, T. Foundation Models for Atomistic Simulation of Chemistry and Materials. arXiv 2025. doi:10.48550/arXiv.2503.10538

[2] Kalita, B.; Gokcan, H.; Isayev, O. Machine learning interatomic potentials at the centennial crossroads of quantum mechanics. Nat. Comput. Sci. 2025, 5, 1120-1132. doi:10.1038/s43588-025-00930-6

[3] Zhang, L.; Zavadlav, J. ConSolv: Solvent-Conditional Machine Learning Implicit Solvent Potential. arXiv 2026. doi:10.48550/arXiv.2606.24983

[4] Klamt, A. The COSMO and COSMO‐RS solvation models. WIREs Comput. Mol. Sci. 2018, 8, e1338. doi:10.1002/wcms.1338

[5] Stamm, B.; Lagardère, L.; Scalmani, G.; Gatto, P.; Cancès, E.; Piquemal, J.; Maday, Y.; Mennucci, B.; Lipparini, F. How to make continuum solvation incredibly fast in a few simple steps: A practical guide to the domain decomposition paradigm for the conductor‐like screening model. Int. J. Quantum Chem. 2019, 119, e25669. doi:10.1002/qua.25669

[6] Müller, S.; Nevolianis, T.; Garcia-Ratés, M.; Riplinger, C.; Leonhard, K.; Smirnova, I. Predicting solvation free energies for neutral molecules in any solvent with openCOSMO-RS. Fluid Phase Equilib. 2025, 589, 114250. doi:10.1016/j.fluid.2024.114250

[7] Batatia, I.; Baldwin, W. J.; Kuryla, D.; Hart, J.; Kasoar, E.; Elena, A. M.; Moore, H.; Gawkowski, M. J.; Shi, B. X.; Kapil, V.; Kourtis, P.; Magdău, I.; Csányi, G. MACE-POLAR-1: A Polarisable Electrostatic Foundation Model for Molecular Chemistry. arXiv 2026. doi:10.48550/arXiv.2602.19411

[8] Gönnheimer, N.; Reuter, K.; Kapil, V.; Margraf, J. T. MACE-MDP: A General Dipole and Polarizability Model for Organic Molecules and Materials. ChemRxiv 2026. doi:10.26434/chemrxiv.15000716/v2

Speaker: Erik Andris (IOCB Prague)

58 The Quantum Access Trail: Reach QPUs at EU/National scale with QHEAppE

Current quantum computers (QC) are largely operated as single-user systems serving small research groups. Scaling access to thousands of users across the Europe, like institutions as part of the EuroHPC Joint Undertaking (EuroHPC JU), introduces two major challenges.

The first concerns multi-tenancy and resource governance. Users must be organized into projects with allocated quantum processing unit (QPU) time and policies governing resource consumption, while usage must be accurately accounted for. Although a QPU is inherently a serial resource, the infrastructure must support concurrent job submissions from multiple endpoints, efficient scheduling, and strict isolation of user data and results.

The second challenge is HPC–QC co-scheduling: coordinating workloads across classical HPC and quantum resources while accounting for their fundamentally different execution models and maximizing overall resource utilization.

To address these challenges and provide national and EuroHPC JU quantum computing resources to users, we developed QHEAppE, a Quantum-as-a-Service architecture integrating quantum resources with HEAppE Middleware.

The architecture combines four core components: the qaas Python client, HEAppE Middleware, QScheduler, and the LEXIS Platform. Together, they provide an end-to-end workflow covering user authentication and authorization, quantum workload submission and scheduling, QPU execution, result retrieval, and resource accounting.

Users submit quantum circuits through the QHEAppE client provided by the qaas Python package, supporting Qiskit, OpenQASM 3.0, and pulse-level playlists. Authentication and authorization are provided through the LEXIS Platform using MyAccessID, a federated identity provider proxy. HEAppE Middleware manages asynchronous job submission, lifecycle and state synchronization, and output retrieval, extending its HPC-as-a-Service abstraction to quantum systems.

HEAppE subsequently forwards jobs to QScheduler, a low-latency Rust-based scheduler forming the final scheduling layer between HEAppE and the QPU. QScheduler implements quantum-specific resource allocation and scheduling policies, including circuit grouping, job prioritization, resource reservations, runtime limits, and execution timeouts.

This architecture provides a scalable service layer for controlled, multi-user access to quantum computing resources and establishes a foundation for integrating quantum resources into distributed HPC environments.

Speaker: Jan Swiatkowski (IT4Innovations, VSB - TUO)

59 Developing Machine Learning Interatomic Potential for 2D TMDC High Entropy Alloys

Two-dimensional (2D) transition-metal dichalcogenides (TMDCs) exhibit a broad range of attractive electronic, optical, catalytic, and tribological properties. Incorporating multiple transition-metal species into the metal sublattice offers further opportunities for tuning these properties through the high-entropy alloy (HEA) concept. However, the vast configurational space of multicomponent 2D TMDCs makes their systematic investigation using density functional theory (DFT) alone computationally demanding.
In this work, we develop a machine-learning interatomic potential (MLIP) for the pentanary 2D high-entropy TMDC using the Gaussian Approximation Potential (GAP) framework. The performance of the MLIP is assessed by comparing its predicted energies and forces with DFT results and by evaluating its ability to reproduce DFT-optimized atomic structures.
The developed MLIP provides a computationally efficient framework for exploring the large configurational space of 2D TMDC high-entropy alloys beyond the practical limits of direct DFT calculations. The work establishes a foundation for large-scale atomistic simulations and further investigation of the structural and thermodynamic behavior of these complex 2D materials.

Speaker: Shubham Agarwal

60 Magnetoelastic behavior in tetragonal Pt-base compounds

Energy conversion effects are important in the design of efficient devices; they also involve a direct conversion between magnetic and mechanical energy – the so-called magnetoelasticity, which takes place in various applications such as force sensors, or magneto-acoustically driven low-energy demanding electronics. Mostly, the magnetoelasticity has been thoroughly studied in a high symmetry systems. Here, we focus on magnetoelastic behavior in a Pt-based system, which is able to
provide a stronger magnetoelastic response, focusing on the behavior of the ferromagnetic and antiferromagnetic systems represented, namely by FePt resp. MnPt.
Based on the ab-initio calculations, we discuss the relation of the magnetoelasticity and the magnetic ordering with the aim to make the link of the origin of electronic structure changes in the resulting magnetoelastic behavior. In addition, we discuss the modeling of the magnetoelastic behavior is such type of materials.
This work has been supported by GACR project 24-11388I of the Czech Science Foundation (GAČR).

Speaker: Dr Jakub Šebesta (IT4Innovations, VSB - Technical University of Ostrava, 70800 Ostrava-Poruba, Czech Republic)

61 Interlayer exchange coupling in antiferromagnetic multilayers with magnetically alloyed spacer

Two layers made of ferromagnetic (FM) materials exhibit an antiferromagnetic (AFM) interlayer exchange coupling (IEC) when separated by a nonmagnetic (NM) material [1]. In addition, the IEC oscillates between the AFM and the FM coupling for a variable width of the NM spacer [2]. The oscillating nature of the IEC with the width of the spacer can be explained by the Ruderman-Kittel-Kasuya-Yosida (RKKY) exchange coupling [3]. This phenomenon allows one to tune the interaction from FM to AFM by changing the thickness of NM material between two magnetic layers, and optimize the various magnetic properties. The tunability of magnetic multilayers makes them attractive for magnetic sensors, magnetic random access memory (MRAM), spin-torque oscillator, magnetic racetrack memory, etc. [4]. The IEC can also be modified by chemically altering the composition of nonmagnetic material by introducing magnetic impurities. The presence of magnetic impurities results in the renormalization of IEC by changing the oscillatory period and amplitude via Fermi surface shifts of the nonmagnetic layer. In Co/Ru multilayers, the Ru spacer alloyed with Fe develops a vortex-like spine texture at intermediate Fe concentrations. Fe spins tend to couple antiferromagnetically, resulting in a noncollinear alignment of Co layers [5]. The noncollinear alignment of FM layers is favored by the biquadratic coupling constant. Alloying NM Ru with FM Fe can be used to control the noncollinear alignment of magnetic moments in FM layers, opening the perspective of using Co/Ru multilayers for spin-transfer torque MRAM devices [6].
In our previous work, we had shown that the experimentally measured values of the bilinear coupling for different concentrations x of magnetic impurities Y = (Co,Fe) for Co/Ru(1−x)Y(x) multilayers are well reproduced by Density Functional Theory (DFT) calculations [7]. In this work, we used the same computational framework to quantify the effect of alloying Ru with all known magnetic elements Y=(Cr,Mn,Co,Fe,Ni) on the AFM bilinear exchange coupling. Our analysis reveals a correlation between the distribution of spin moments and the changes in the IEC value for different magnetic impurities. We quantify this analysis within a simple Heisenberg model considering both direct exchange and RKKY-type interactions.

[1] P. Grünberg et al., Phys. Rev. Lett. 57, 2442 (1986), URL https://link.aps.org/doi/10.1103/PhysRevLett.57.2442.
[2] S. S. P. Parkin, Phys. Rev. Lett. 67, 3598 (1991), URL https://link.aps.org/doi/10.1103/PhysRevLett.67.3598.
[3] P. Bruno, Phys. Rev. B 52, 411 (1995), URL https://link.aps.org/doi/10.1103/PhysRevB.52.411.
[4] S. Parkin and S.-H. Yang, Nature Nanotechnology 10, 195 (2015), ISSN 1748-3395, URL https://doi.org/10.1038/nnano.2015.41.
[5] Z. R. Nunn et al., Science Advances 6, eabd8861 (2020), URL https://www.science.org/doi/abs/10.1126/sciadv.abd8861.
[6] A. D. Kent and D. C. Worledge, Nature Nanotechnology 10, 187 (2015), ISSN 1748-3395, URL https://doi.org/10.1038/nnano.2015.24.
[7] K. Winther et al., Phys. Rev. Appl. 22, 024058 (2024), URL https://link.aps.org/doi/10.1103/PhysRevApplied.22.024058.

Speaker: Sergiu Arapan (IT4Innovations VŠB - TU Ostrava)

62 Bridging Phonon Dynamics and Nanotribology in TMD Heterostructures via Machine Learning Force Fields

In this work, we study the tribological response of transition metal dichalcogenide monolayers interacting with a metallic substrate and a silicon atomic force microscopy tip. High-quality machine learning force fields [1,2] are developed and subsequently used to perform classical molecular dynamics (MD) simulations at room temperature, to investigate the nanofriction behavior under varying applied loads and sliding velocities of the Si tip. The resulting MD trajectories are then analysed in terms of renormalized phonon, eigendisplacements, eigenfrequencies and relative atomic participation.

This analysis provides detailed insight into how vibrational modes, such as sliding and breathing modes, determine the friction response and what is the relative role of substrate, monolayer and tip. The results provide information on how to tune the friction response by proper selection of the kind of substrate and monolayer. This analysis is general and can be promptly applied to the study of frictional interfaces irrespective of the chemical composition and atomic topology.

Keywords: machine learning force fields, transition metal dichalcogenides, nanofriction, phonon analysis, molecular dynamics, heterostructure

References
[1] Shaidu et al. npj Computational Materials 2025, 11, 273
[2] Denghui Lu et al. Computer Physics Communications 2021, 259, 107624

Speaker: Ravikant Kumar (Department of Control Engineering - KN:G-204, Faculty of Electrical Engineering, Czech Technical University in Prague, Czech Republic)

63 Feature-Complete Multi-GPU Solver for the k-Wave Ultrasound Simulation Toolbox

Large three-dimensional k-Wave simulations are often limited by the memory capacity of a single GPU, while CPU execution can be prohibitively slow. We present a feature-complete multi-GPU CUDA solver that preserves the global k-space operator, supported physical models, and existing k-Wave workflow while distributing simulation fields across multiple devices. Global three-dimensional FFTs use cuFFT Xt to redistribute slab-partitioned data between spatial and spectral layouts. CPU, single-GPU, and multi-GPU execution are integrated within one unified C++17 and CUDA codebase with shared input, output, logging, and regression infrastructure. The solver is continuously validated against 137 MATLAB-reference simulations, performing 859 output-dataset comparisons per execution mode across multiple HPC systems. On Karolina, eight A100 GPUs achieve up to 29.3 times speedup over the optimized 128-thread OpenMP solver. Distributed GPU memory also enables a measured $1400^3$ simulation containing 2.744 billion grid points.

Speaker: Oliver Kuník

64 Modular multiscale approach to modelling HHG in gases & spectrally selective HHG scheme

We will present the recently released open-source MMA-HHG toolbox (mma-hhg.github.io) for modelling high-harmonic generation (HHG) in gases. The package provides a common framework for coupling the microscopic response of elementary emitters to the local driving field resulting from nonlinear propagation in the macroscopic generating medium, and for coherently combining these contributions to reconstruct the macroscopic XUV field. The current public version focuses on linearly polarised driving fields and cylindrically symmetric propagation, with the aim of enabling detailed comparison with laboratory HHG experiments while preserving stand-alone use of the individual modules. The development of this package is strongly driven by laboratory applications.

This release contains Docker container to learn all the features of the package locally and to test different analyses of the results, there is also a Pythonic interface to microscopic HHG response obtained from numerically solving time dependent Schrödinger equiation. It provides a detailed insight into the process (ionisation, Gabor transform, ...). Then, we show templates for computational pipelines to deploy the code on a supercomputer

One of the ongoing simulation efforts is the application on monochromatising HHG source using Bessel-Gauss beams in density-modulated gas targets. We will present the actual results of full-scale simulations of this
scheme.

Acknowledgements:
This work was supported from from GACR Grant No. 25-17853S.
This work was supported by the Ministry of Education, Youth and Sports of the Czech Republic through e-INFRA CZ (ID:90254).

Speaker: Jan Vábek (ELI Beamlines Facility, The Extreme Light Infrastructure ERIC, Czech Republic; Czech Technical University in Prague, Faculty of Nuclear Sciences and Physical Engineering, Czech Republic)

Excursion - Showroom

19:00 Conference Dinner

Friday 6 November

Keynote III

65 Improving the ground state calculations on NISQ devices

The paradigm of quantum computing offers to solve certain computational problems faster compared to the classical computers. However current so called Noisy Intermediate Scale Quantum (NISQ) devices are too noisy to produce results faster or more accurately compared to classical computers. In our research we focus on combining two types of quantum algorithms Variational Quantum Eigensolver (VQE) with Robust Phase Estimation (RPE) to solve time independent Schroedinger equation which can be interpreted as an eigenvalue problem. RPE is a variant of phase estimation algorithm that can correctly returns the eigenvalue as long the eigenstate is known which is not typically true for an eigenvalue problem. Results from the conducted numerical testing show that only the rough estimate of an eigenstate is sufficient input to the RPE in order to produce accurate eigenvalue. To produce the rough estimate of an eigenstate the VQE algorithms was implemented with generic non expressive ansatz, this approach reduces the QPU time and also the size of the parametrized quantum circuit making it more resilient to noise. The most important part of the RPE is an efficient implementation of the controlled unitary operation in the native gate set of given quantum processor since it can produce circuit too large to be efficiently implemented. To prevent creating circuit with thousands two qubit gates the Transitionally Invariant Compressed Controlled (TICC) time evolution was implemented so the created circuits are executable on NISQ devices. This combination of VQE and RPE was implemented to solve the ground state of the Ising model accurately on the noiseless and noisy simulator of the quantum computer. The simulations were conducted using IT4Innovations based supercomputers and models are also prepared to run on the VLQ quantum computer.

Speaker: Vojtěch Vašina (Institute of Physics of Materials, Czech Academy of Sciences)

Users' Talks III

66 Certifying Quantum Hardware

In this presentation, we propose a quantum certification method whose main goal is to verify if a quantum provider is honest about the capabilities of their quantum machine. To do this, we use machine learning methods to carefully create a quantum circuit to be executed by the provider. By investigating the outputs of the computation, we are able to certify the validity of the quantum machine with high probability. If the machine is not of good quality, or if the provider is dishonest and wishes to simulate the results instead, the test will most likely fail.

In an era where new quantum infrastructure pops up frequently with only API access given, such a method should be of great relevance.

Speaker: Ryszard Kukulski (IT4Innovations National Supercomputing Center)

67 Structural Models for Carbon Nanodots: Graphitic, Nitrogen-doped and Graphyne-Based systems

Graphyne and graphdiyne represent two-dimensional carbon allotropes structurally related to graphene. Unlike graphene, which is composed exclusively of sp²-hybridized carbon atoms arranged in a honeycomb lattice, graphyne and graphdiyne incorporate sp-hybridized carbon atoms through acetylenic linkages. In graphyne, aromatic rings are connected by single –C≡C– units, while in graphdiyne they are linked via diacetylenic –C≡C–C≡C– bridges. These structural motifs introduce intrinsic porosity and significantly modify the electronic properties, leading to tunable band structures, enhanced charge transport characteristics, and well-defined nanoporous architectures suitable for energy conversion, sensing, and photocatalytic applications.
Carbon dots (CDs) are emerging luminescent carbon-based nanomaterials that have attracted significant attention due to their unique photoluminescence, low toxicity, chemical stability, and excellent biocompatibility. Typically considered quasi-zero-dimensional systems, CDs consist of a hybrid sp²/sp³ carbon core functionalized with oxygen- and nitrogen-containing groups, making them highly versatile for applications in bioimaging, optoelectronics, and photocatalysis.
In this work, we develop and investigate atomistic models of carbon dots based on different structural building blocks, including:
• Nitrogen-doped graphitic layers,
• Graphitic carbon dots functionalized with hydroxyl groups on their surface,
• Hybrid assemblies combining circumcoronene units with graphyne and graphdiyne sheets.
For both the graphitic and nitrogen-doped carbon dot models, we also investigate their interactions with fluorophores belonging to the citrazinic acid family.

Classical molecular dynamics simulations are performed to investigate the interactions between these constituents and to elucidate their self-assembly mechanisms leading to carbon dot formation. Prior to the simulations, all structural components are carefully parametrized to ensure a consistent and reliable description of their interatomic interactions.
All simulations are carried out on high-performance computing (HPC) infrastructures, specifically the LUMI and Karolina supercomputers. Molecular dynamics simulations are performed using GROMACS, while the parametrization of all molecular building blocks is carried out using Gaussian. This computational framework enables an extensive exploration of different carbon dot design strategies, offering a systematic understanding of how structural motifs influence assembly behavior and ultimately govern the resulting morphology. The overall goal is to elucidate carbon dot formation mechanisms and identify optimal structural configurations that could lead to improved physical and functional properties.

Speaker: Zaira Carboni (Università degli studi di Cagliari)

68 Controlled Continued Pretraining at Scale: Measuring Entity Learnability in LLMs on LUMI

How much exposure does a language model need before it reliably learns a new entity, and how can this process be measured experimentally on large-scale HPC infrastructure? We present the computational design of a dissertation experiment focused on entity learnability in large language models, using fictitious brands and their attributes as a controlled case. The experiment uses continued pretraining on the EuroHPC LUMI system.

The scientific objective is to measure how information about a previously unseen entity is acquired by a model as a function of exposure volume, source type, and distribution. From an HPC perspective, this also creates an orchestration problem: a large matrix of comparable training runs must be generated, scheduled, monitored, checkpointed, validated, and evaluated while preserving tightly controlled experimental conditions.

For each condition, synthetic corpora are constructed to represent different source types, including entity-owned websites, catalogues, independent editorial content, press releases, user discussions, and texts containing contradictory or denying statements. The text is generated through a frontier LLM API and then passed through a mechanical validation layer. Validated content is inserted into a fixed reference corpus at controlled token positions, allowing the total training volume, cumulative entity exposure, and exposure distribution to be varied independently.

Selected open language models then undergo continued pretraining on LUMI GPU accelerators. A central design issue is checkpoint comparability. Conventional checkpointing by training step ties measurement points to batch configuration and parallelization strategy. Instead, checkpoints are scheduled by cumulative exposure to the target entity. Measurement points are therefore defined by experimental meaning rather than by a particular compute configuration, making runs directly comparable across execution geometries.

The poster focuses on the engineering required to make such an experimental campaign reproducible and operationally manageable on shared HPC infrastructure. Corpus construction is deterministic, with cryptographically fingerprinted shards and explicit manifests describing corpus composition and entity placement. Configurations, datasets, checkpoints, and derived artifacts are fingerprinted in the same way. Saved checkpoints are automatically verified before downstream processing to detect incomplete or corrupted artifacts early.

Training throughput is calibrated empirically across micro-batch and parallelization configurations rather than estimated from nominal accelerator performance. This supports resource planning based on observed performance and enables cost comparisons across experimental cells.

The pipeline spans CUDA and ROCm environments, providing practical insights into portability across containers, libraries, numerical precision, memory behavior, and job configuration. Monitoring uses a pull-based architecture without persistent cluster-side services: jobs emit structured state and metrics that are aggregated externally for progress tracking and failure detection.

The study is preregistered on OSF, and the methodology and code are intended to be public. Beyond the specific application to brand entities, the project serves as a case study in reproducible orchestration of controlled continued-pretraining experiments on shared HPC infrastructure.

Speaker: Mr Michal Kubicek (Silesian University in Opava, School of Business Administration in Karvina)

10:30 Coffee Break

Users' Talks IV

69 Decoding Lipid Nanoparticle Structure: From Lipid Composition to pH through Multiscale Simulations

Lipid nanoparticles (LNPs) have enabled the rapid development of RNA-based medicine in vaccination, cancer treatment and the therapy of rare diseases. These systems encapsulate RNA within assemblies of ionizable lipids (ILs), helper lipids, and a PEG shell, achieving high loading efficiency through electrostatic interactions of ILs with RNA, while maintaining low cytotoxicity due to the neutral charge of ILs under physiological conditions. Despite intense efforts in ILs design aiming to improve the RNA delivery efficiency, progress is hindered by a limited molecular-level understanding of LNP structure, its pH induced changes, and the mechanisms underlying endosomal escape.
Multiscale molecular dynamics (MD) simulations provided an important and detailed insight into the LNP structure. However, the systems of interest contain a mixture of variously charged lipids and nucleic acids in LNP and possible interacting biocompartments, all dynamically diffusing, therefore changing their composition definition. Their dynamical analyses offer new challenges and opportunities in the possible extracted message of the behavior of LNPs.
Here we present the combined all atom and coarse grained simulations decoding the role of LNP composition of the adopted LNP phase, its curvature and mobility. We focus on role of the structure of IL, the ratio of individual lipid components in LNP and of RNA under different environmental conditions. The dynamic decoding of the LNP internal structure offers a detailed insight into the role of its individual components on the resulting particle.

Speaker: Markéta Paloncýová (CATRIN, Palacký University Olomouc)

70 SEANERGYS: From Day-0 Datasets to Predictive AI for HPC Resource Management

As High-Performance Computing (HPC) scales toward Exascale, optimizing energy efficiency and resource utilization requires a unified, data-driven approach. The EuroHPC JU project SEANERGYS [1] addresses this by establishing a telemetry-to-intelligence pipeline built around three pillars: Monitoring (CMI), Data Analytics (AIDAS), and Resource Management (DSRM). The AIDAS transforms the CMI monitoring data into insights, predictions, and actionable feedback that can be used by DSRM. The main AIDAS component is the Model Zoo, a centralized framework containing a curated set of AI/ML models providing system-, node-, scheduler-, and job-centric predictions. To enable model training prior to full CMI deployment, anonymized “Day-0 datasets” were collected from eight supercomputing centers, standardized into the unified SEANERGYS format, and stored as compressed Apache Parquet [2] files in a centralized data repository. Day-0 datasets establish a baseline for pre-training and benchmarking the Model Zoo’s models, enabling their preparation prior to integration with live monitoring data streams. Day-0 datasets may also be made publicly available in the future through a data repository to support research and reproducibility.

Speaker: Petr Ptáček (IT4Innovations)

71 From Methods to Molecules: Four Years of Open-Source Computer-Aided Drug Design

Over the past four years within the INTER-EXCELLENCE II grant project LUAUS23262, we have developed an integrated ecosystem of open-source tools for computer-aided drug design, covering virtual screening, molecular simulation, de novo design, and ligand optimization. EasyDock provides automated and scalable molecular docking workflows, while StreaMD supports standardized molecular dynamics and free-energy calculations. In parallel, we substantially extended the CReM framework for chemically reasonable molecular transformations and developed specialized tools for pharmacophore-guided generation (CReM-pharm), docking-guided de novo design (CReM-dock), and docking-driven ligand optimization (CReM-opt).

A central objective of this work was to create modular, interoperable, and computationally efficient tools that can be used independently or combined into higher-level automated workflows. Their practical utility was evaluated in two prospective drug-discovery studies. For CDK16, de novo generated compounds were used to guide similarity searching in an ultra-large chemical space of Enamine REAL DB (~10B compounds), followed by experimental testing of selected candidates resulted ion discovery of 14 new CDK16 inhibitors. A second prospective study targeting DYRK1B combined pharmacophore-guided generation, docking-guided design, iterative optimization, physicochemical and structural filtering, and free-energy calculations to prioritize novel compounds for synthesis. Top selected compounds are currently undergo experimental evaluation.

These studies illustrate how open-source computational tools can be integrated into reproducible end-to-end workflows that connect large-scale virtual screening with molecular generation and prospective experimental validation.

Speaker: Pavel Polishchuk (IMTM, Palacky University)

Users' Council

13:00 Lunch