VÝSLEDKY 37. KOLA VEŘEJNÉ GRANTOVÉ SOUTĚŽE

ALOKAČNÍ KOMISE VE 37. KOLE VEŘEJNÉ GRANTOVÉ SOUTĚŽE IT4INNOVATIONS ROZDĚLILA VÝPOČETNÍ ZDROJE TAKTO:

Řešitel: Van Binh Henri VU        

OPEN-37-1                    

Quantum-Enhanced Anti-Money Laundering (QE-AML) using Dynamic Quantum Reservoir Computing  

Barbora NG  Alloc=1200;  Karolina CPU Alloc=4500;  Karolina FAT  Alloc=100;  Karolina GPU Alloc=2900;  LUMI-C  Alloc=7600;  VLQ  Alloc=350000

Money laundering is a trillion-dollar global problem that hides criminal activities behind complex webs of transactions. While banks use artificial intelligence to spot these patterns, today’s \classical\" computers often struggle to keep up with the sophisticated, fast-changing tactics of modern  fraudsters. Our project, Quantum-Enhanced Anti-Money Laundering (QE-AML), aims to solve this by using the next generation of technology: Quantum Computing with Artificial Intelligence.  We are developing a specialized \"Quantum Reservoir\"—a type of brain-inspired quantum system—to process transaction data from a world-leading research dataset (Kaggle). Unlike traditional systems that look at single snapshots of data, our quantum model uses a \"Rewinding Protocol\" to maintain a memory of transaction history, allowing it to \"see\" suspicious patterns unfolding over time. By running these simulations on the Karolina supercomputer and the new VLQ quantum computer at IT4Innovations, we will compare this quantum approach against industry-standard AI like LSTMs, RNNs and GRUs. Our goal is to prove that quantum technology can detect fraud more accurately and  with fewer \"false alarms.\" This research not only pushes the boundaries of Czech science but also provides a roadmap for banks to create more secure financial systems, protecting the economy and society from the impact of organized crime.

Řešitel: Ales Vitek   

OPEN-37-10                 

From Structure Factors to Hydrogen-Bond Networks: Decoding Nanoscale Water   

Karolina CPU  Alloc=300             

Understanding the structural organization of water at the nanoscale remains a key challenge in molecular physics, with relevance to atmospheric processes, chemistry, and biology. We propose a computational study of water clusters containing 50 molecules using classical Monte Carlo simulations in the NPT ensemble combined with parallel tempering, enabling efficient sampling of complex hydrogen-bonded configurations across a range of conditions.  The main goal is to connect experimentally accessible observables with microscopic structural descriptors. We will compute the static structure factor (S(k)), directly comparable to X-ray and neutron scattering experiments, and complement it with real-space analysis of local and network structure. In particular, the tetrahedral order parameter [1] will be used to quantify local “ice-like” ordering, while the hydrogen-bond network will be analysed using graph-theoretical methods.  This network-based approach allows us to determine the frequency of four-, five-, and six-membered rings, which distinguish disordered, liquid-like structures from more ordered, ice-like arrangements. By combining reciprocal-space and graph-based analyses, we aim to disentangle local geometric order from global network topology.  The project builds on established methodologies for water structure characterization [1,2] and extends them to finite clusters, where surface effects and finite-size constraints play a crucial role. The results will provide experimentally relevant structure factors together with a detailed understanding of the relationship between tetrahedrality and hydrogen-bond network topology in nanoscale water systems. 

Řešitel: Jakub Kopko                     

OPEN-37-11                 

Advancing QM-Enriched Machine Learning Scoring Functions for High-Throughput Discovery of Next-Generation Therapeutics

Karolina CPU  Alloc=21300;  Karolina GPU  Alloc=1700;  LUMI-G  Alloc=8000              

In computational drug discovery, scoring is the estimation of the binding affinity between a small molecule and a protein target, which is essential for identifying effective drugs. Modern structure-based drug discovery is frequently bottlenecked by the trade-off between computational speed and accuracy during this scoring phase. In our previous IT4Innovations project (OPEN-34-5: Quantum Mechanics-Enhanced Structural Data Generation: Building Foundations for Advanced ML Models in Drug Discovery), we successfully developed a proof-of-concept Machine Learning (ML) scoring function trained exclusively on high-quality, Quantum Mechanically (QM) scored protein-ligand conformations. This approach allowed for training models useful even in the absence of experimental training data for medically important targets.  We are now scaling our proof-of-concept model into a comprehensive, real-world pipeline. In the first phase, we will allocate significant computational resources to expand our QM-enriched data generation. We will specifically tackle highly complex targets, such as metalloproteins, and incorporate induced-fit docking to accurately account for natural protein flexibility. In the second phase, we will leverage this newly generated, highly complex data to conduct rigorous ML ablation studies. Ultimately, this pipeline will drastically reduce computational costs associated with the discovery of novel therapeutic candidates across a broad spectrum of challenging diseases.

Řešitel: Martin Friak

OPEN-37-12                 

The impact of noise on quantum computing of the electronic structure of solids      

Barbora NG  Alloc=700;  Karolina CPU  Alloc=600;  Karolina GPU  Alloc=300;  VLQ  Alloc=360000                

Quantum computing is currently emerging as a very useful novel paradigm for solving complex computational tasks. We will apply it to the calculations of the electronic structure of crystals. The electronic structure is critically important for accurately predicting material properties. We are motivated by the fact that calculations of the electronic structure by classical computers are not effectively scalable and lack accuracy when applied to larger systems. Quantum computing offers a solution to this problem but the influence of the noise in the current superconducting quantum computers needs to be studied. We aim at filling this gap in knowledge by computing and analyzing the electronic structure using a representative of current superconducting quantum computers, the VLQ at the IT4Innovation supercomputing center.

Řešitel: Michal Belina                  

OPEN-37-13                 

Quantum Interval Power Flow 

Karolina CPU  Alloc=100;  VLQ  Alloc=180000             

Modern power grids face growing uncertainty from renewable sources like wind and solar, whose output fluctuates unpredictably. Engineers must ensure safe grid operation not under one set of conditions, but across all possible ranges of these fluctuations. This challenge, known as interval power flow analysis, is computationally demanding, and conventional methods often produce overly conservative safety estimates.     This exploratory study will investigate whether quantum computing methods can be meaningfully applied to this problem. Using a 24-qubit superconducting quantum processor by IQM, the interval power flow will be reformulated as an optimization task and tested on small power networks. Instead of propagating uncertainty through equations, which inevitably inflates the results, the quantum approach directly searches for worst-case operating conditions, which may yield tighter bounds.     The study will focus on testing the feasibility of the formulation, identifying practical hardware limitations, and evaluating how the approach behaves as problem size grows. In a second stage, alternative problem encodings that could exploit quantum effects beyond classical combinatorial search will be explored. Results will be systematically benchmarked against established classical methods. The work aims to establish methodological groundwork designed to scale with future generations of quantum hardware.

Řešitel: Sergiu Arapan                  

OPEN-37-14                 

The study of magnetocaloric properties in Fe2X Laves phases: calculating finite temperature magnetic properties from ab-initio               

Barbora NG  Req=5500;  Karolina CPU  Alloc=3400;  Karolina FAT  Alloc=50;  Karolina GPU  Alloc=400;  LUMI-C  Alloc=2000    

Within the paradigm of green energy, magnetic refrigeration is a promising environmentally friendly cooling technology.  It is designed based on the magnetocaloric effect, which is a heating or cooling of the magnetic material under the influence of the applied magnetic field. The magnetocaloric effect is large at temperatures where phase transitions take place. Fe2X (X=Ti, Ta, Sc, Nb, Hf) Laves phases show various competing magnetic configurations, with magnetic transitions in a large range of temperatures. Furthermore, the transition temperature and the magnetic entropy change can be tuned by transition metal substitutions (chemical pressure) or applied external pressure (the barocaloric effect). For the large configurational space various compositions, magnetic arrangements, pressures, and temperatures, we can use the computational power to determine optimal set of parameters. The numerical modeling of magnetic properties at finite temperature is based on the knowledge of Heisenberg exchange interactions between magnetic moments. In this study we will determine the Heisenberg exchange parameters for various Fe2X Laves phases by using quantum mechanical calculations based on the density functional theory.

Řešitel: Maximilian Lamanec  

OPEN-37-15                 

Quantum and Classical Computing for Auger Spectroscopy of Molecules                     

Barbora NG   Alloc=8000;  Karolina CPU  Alloc=24900;  Karolina FAT  Alloc=100;  Karolina GPU  Alloc=1400;  LUMI-C  Alloc=18700;  LUMI-G  Alloc=13600;  VLQ  Alloc=362200

This project aims to develop and apply an innovative and scalable theoretical framework for the modeling of Auger spectroscopy in molecules. The main goal is to improve the interpretation of modern X-ray spectroscopic experiments and to increase the chemical insight that can be extracted from Auger electron spectra. The proposed framework will be used to address fundamental questions concerning the electronic structure of molecular core-excited and core-ionized states and the mechanisms governing Auger decay. Particular emphasis will be placed on resonant Auger electron spectroscopy, which is site- and element-specific and provides exceptionally rich information on the structural and electronic properties of molecules. By combining advanced electronic-structure methods with scalable computational workflows, the project will establish reliable theoretical tools for the accurate simulation and analysis of Auger spectra in molecular systems.

Řešitel: Alžběta Špádová           

OPEN-37-16                 

Nanoparticle-Assisted Electron Injection for Guided Multi-GeV Laser Wakefield Acceleration

Karolina CPU  Alloc=14200;  Karolina GPU  Alloc=1200                

Particle accelerators are essential tools in modern science and medicine, but conventional facilities are large, expensive, and accessible only to a few research institutions. Laser wakefield acceleration (LWFA) offers a path toward compact, affordable alternatives by using intense laser pulses to drive plasma waves that accelerate electrons to near-light speeds over just a few centimeters.   Despite impressive progress, producing stable, high-energy electron beams remains challenging. Reaching energies above 10 GeV requires guiding the laser pulse over extended distances, and while plasma guiding structures have been demonstrated, controlling electron injection into the accelerating wave at the required low plasma densities remains difficult. We investigate nanoparticle-assisted injection as a solution, where an ionized nanoparticle triggers controlled electron injection, enabling stable and reproducible beams even under demanding long-distance acceleration conditions.  In this project, we will use particle-in-cell simulations to study the underlying physics and optimize injection parameters, with the goal of producing stable, GeV-scale electron beams in the ELBA experiment at ELI Beamlines. Success here would mark a key step toward compact accelerators that could one day fit inside a hospital or industrial facility — bringing the capabilities of national laboratories within reach of researchers and medical professionals worldwide.

Řešitel: Aleš Horák 

OPEN-37-17                 

Slama - Slavonic Large Foundational Language Model for AI, pretraining larger multilingual variant            

LUMI-G  Alloc=8700

The proposed Slama project focuses on building a new foundational language model concentrated on main Slavonic languages (Czech, Slovak, Polish, ...). The project’s primary goal is to explore the performance differences between state-of-the-art pre-trained multilingual models (where English texts represent the majority of training data) and a model tailored specifically to the Slavonic language group. The research focuses on developing generative models whose training data are more balanced in favor of the Slavonic language group rather than English. Therefore it should provide better results when used in AI tools processing mainly Slavonic languages. The resulting foundational model can then be easily applied in a range of AI tasks.  To achieve this, we will continue the Slama models development with training a new larger 14B-parameter generative language model on a large-scale, carefully balanced multilingual corpus (approx. 367B tokens), with significantly increased representation of Slavonic languages. The model size and training token budget are designed in accordance with Chinchilla scaling laws, ensuring a compute-optimal balance between model capacity and data. Training will be conducted on the LUMI-G supercomputing infrastructure, leveraging empirically validated scaling from prior 7B model experiments.

Řešitel: Antonín Jarolím              

OPEN-37-18                 

Scalable Fine-Grained Evidence Retrieval for Explainable Fact-Checking

Karolina GPU  Alloc=3000          

Online discussions and reviews significantly influence public opinion, yet they are increasingly affected by misinformation. Detecting misleading claims and verifying them reliably remains a challenging task, especially when relevant evidence is hidden within long and complex documents.  This project focuses on developing scalable methods for evidence-based fact-checking from real-world sources such as news comment sections. We aim to automatically identify factual claims and retrieve both supporting and refuting evidence from large document collections. Unlike current approaches that rely on generating unfaithful explanations, we propose to directly locate and highlight precise evidence within documents.  The proposed approach builds on recent advances in large language models, which are capable of understanding long texts and identifying fine-grained evidence spans. To transfer these capabilities into efficient retrieval systems, training must explicitly incorporate long-context understanding. Training these models requires processing large-scale datasets and handling long-context inputs.  The expected outcome is an efficient and interpretable fact-checking retrieval system that improves transparency and reliability of automated verification. The results will support journalists, analysts, and the public in making better-informed decisions and contribute to reducing the spread of misinformation.

Řešitel: Jozef Hritz  

OPEN-37-19                 

Biomolecular interactions relevant in neurodegenerative diseases             

Karolina CPU  Alloc=20800;  Karolina GPU  Alloc=2500;  LUMI-C  Alloc=3500;  LUMI-G  Alloc=13400          

Understanding the molecular-level mechanisms behind diseases is a key step in developing successful treatments.  This is especially the case for complex neurodegenerative diseases such as Alzheimer’s. In our research group, we  use computational approaches to supplement experimental data to determine the precise interactions and  dynamics that lead to such pathologies, as well as both natural and artificial mechanisms that help mitigate  those. We focus on protein-protein and protein-ligand binding, and how it is influenced by the wider environment.  In doing so, we help rationalize biological observations and discover potential targets for drugs and antibodies  alike

Řešitel: Martin Matys                    

OPEN-37-2                   

Laser Wakefield Acceleration for Generating High-Quality Electron Beams                   

Karolina CPU Alloc=15100;  Karolina GPU Alloc=200                     

Laser–wakefield acceleration (LWFA) is a promising approach for developing compact high-energy electron accelerators. In this scheme, an intense ultrashort laser pulse propagating through a gas creates a plasma wave that can trap and accelerate electrons to relativistic energies over only a few centimetres. Because the accelerating fields in plasma are orders of magnitude higher than in conventional radio-frequency structures, LWFA offers a path toward smaller and more cost-effective electron sources for applications such as free-electron lasers, compact radiation sources, and advanced accelerator concepts. A key challenge is to generate electron beams with the quality required by these applications, in particular high charge, low energy spread, low emittance, and low divergence. In this project, we will investigate two promising schemes for injecting electrons into the plasma wave: density downramp injection and ionization-induced injection. Using large-scale Particle-In-Cell simulations on high-performance computing systems, we will study the influence of laser evolution and plasma-density shaping on electron trapping and acceleration, and identify conditions leading to efficient generation of high-quality electron beams.

Řešitel: Pavel Novosad                

OPEN-37-20                

Primary breakup liquid spray in internal mixing twin-fluid atomizers           

Karolina CPU  Alloc=9400;  Karolina GPU  Alloc=1500                   

Spraying highly viscous liquids efficiently is a challenging engineering problem with applications ranging from agricultural fertilizer distribution to heavy fuel combustion. Internal mixing twin-fluid atomizers offer a promising solution by mixing a small amount of compressed air directly with the liquid inside the nozzle, using the expanding gas to drive droplet formation. Compared to conventional nozzles, this design requires less energy and is resistant to clogging, but suffers from spray instability at low air flow rates.  The instability originates at the discharge orifice, where the air-liquid mixture exits the nozzle. Depending on operating conditions, the air passes through either as a stream of small bubbles or as a single large gas mass, each producing a fundamentally different breakup mechanism. This region remains poorly understood — the processes occur on a very small scale and change too rapidly for experiments to resolve reliably, and existing numerical studies have relied on overly simplified models.  This project combines detailed CFD simulations with high-speed imaging to study the breakup at the discharge orifice systematically. Simulations resolving the full three-dimensional interface dynamics will investigate how bubble size, position and number affect spray formation, with atomization efficiency tracked through interface surface area growth. The results will provide a foundation for designing more stable and energy-efficient atomizers.

Řešitel: Alexander Vikman        

OPEN-37-21                

Gravitational Waves from the Early Universe: Axion Scenario  

Karolina CPU  Alloc=20600        

The early universe between inflation and big bang nucleosynthesis is a largely unexplored epoch of cosmic history. The current paradigm to explain the properties of the universe at large scales posits a phase of exponential expansion (inflation), an epoch of energy transfer (reheating), and finally the onset of the thermal era. The range of processes that could take place during this period is vast, and most probably involves high-energy physics beyond the Standard Model. For instance, some phenomenologically interesting realizations of inflation consider the inflaton to be an axion-like particle (ALP). On the other hand, ALPs are also interesting for processes after inflation.  In particular, they can form domain walls (DW) and other topological defects. Both phenomena are known to result in effective emission of gravitational waves (GW).   Currently, the unique opportunity to obtain direct information about these very first moments of the universe is to observe GW free-streaming to us. This is an active and promising direction of research given the strong evidence of a nHz stochastic GW background (GWB) reported by NANOGrav collaboration in 2023 and current efforts to observe GW in more than million times higher frequencies in the next decade. These observational programs include space mission LISA and the ground based Einstein Telescope. This proposal aims to numerically determine the dynamics of axion inflation and DW and their signatures in GW using publicly available CosmoLattice code.

Řešitel: Libor Veis                  

OPEN-37-22                

Molecular Electronic Structure Calculations on a VLQ Quantum Computer                 

VLQ  Alloc=72000    

Řešitel: Rachana Yogi         

OPEN-37-23                

Doped Penta-2D Materials for Emerging Electronic Technologies: First-principles Engineering

Barbora NG  Alloc=2500;  Karolina CPU  Alloc=9400;  Karolina FAT  Alloc=150;  Karolina GPU  Alloc=300;  LUMI-C  Alloc=2100

This work aims to investigate the structural stability, electronic, and optical properties of emerging two-dimensional (2D) materials, specifically Ge-based penta-nanosheets, using first-principles computational methods. We focus on the design and characterization of novel 2D systems with tunable properties that are suitable for applications in optoelectronics, including light-harvesting devices, transistors, and photocatalysis. The structural stability of the proposed materials is systematically evaluated through ab initio molecular dynamics and phonon dispersion calculations, ensuring both thermal and dynamical stability. To accurately describe the electronic properties, we employ density functional theory (DFT) using both the generalized gradient approximation in the PBE functional and the hybrid HSE06 functional, which enables the reliable prediction of band structures. Furthermore, we investigate the role of  defects and substitutional doping in modulating the electronic properties of these materials. Our results reveal that defect engineering and doping can induce significant changes in the electronic behavior, including transitions from semiconducting to half-metallic and ultimately metallic states. Such tunability highlights the potential of these materials for a wide range of advanced technological applications. Overall, this work provides a comprehensive understanding of the computational physics underlying Ge-based 2D penta-nanosheets are good for next-generation energy and electronic applications.

Řešitel: Ankit Gupta            

OPEN-37-24                

Efficient and Scalable AI Methods for Radar-Based Non-Contact Biosignal Processing                

Karolina CPU   Alloc=200;  Karolina GPU  Alloc=400

This project focuses on artificial intelligence-based methods for non-contact pulse wave reconstruction and cardiopulmonary vital sign estimation, including heart rate, respiration rate, and blood pressure, using radar-based chest displacement measurements. The proposed approach focuses on extracting physiological information from high-dimensional radar data in a non-invasive and unobtrusive manner. To achieve this, the project will investigate advanced machine learning techniques, including transformer-based architectures, generative models, and self-supervised learning strategies, for robust biosignal reconstruction and representation learning under challenging conditions such as low signal-to-noise ratio and motion artefacts. These methods will be designed to improve generalisation and reliability in real-world environments. In addition, the project will develop computationally efficient models suitable for deployment on embedded and edge devices, enabling real-time, non-contact monitoring in both clinical and home-care settings. This supports the development of scalable and patient-friendly healthcare solutions with minimal discomfort.  Overall, the project combines high-performance computing and advanced AI methodologies to enable robust, real-time radar-based vital sign monitoring, contributing to next-generation digital health technologies.

Řešitel: Marta Jaroš             

OPEN-37-25                

Automated Tuning of Workflows Executions on Remote Computational Resources

Karolina CPU  Alloc=1000;  Karolina GPU  Alloc=300

In recent years, therapeutic ultrasound has diverse applications like tumor ablation and targeted drug delivery. Optimal outcomes require precise, customized preoperative planning. A challenge is accurate, safe, and noninvasive ultrasound energy delivery to the target region. Computationintensive models for treatment estimation use high-performance computing (HPC). Despite the significance of HPC, clinical end-users lack efficient utilization expertise. The k-Plan software simplifies HPC use without specifying parameters, dependencies, or monitoring. It addresses parameter selection challenges and scaling issues, critical for calculation cost and execution time. Having deployed k-Plan for initial workflows, this project aims to (1) develop and test a new GPU code for thermal simulation, (2) apply real clinical and biomedical workflows, (3) customize task submission planning logic for IT4Innovations clusters with machine learning, and (4) explore methods for kPlan to auto-tune execution parameters for tasks. Creating a publication covering experiments is a key goal.

Řešitel: Frantisek Karlicky

OPEN-37-26                

Many-body physics of van der Waals heterostructures from 2D materials III                 

Barbora NG  Alloc=8600;  Karolina CPU  Alloc=14700;  Karolina FAT  Alloc=150;  Karolina GPU  Alloc=500;  LUMI-C  Alloc=8000                    

Current needs of applied research on two-dimensional (2D) materials for flexible and ultrathin functional devices require computational predictions and computer-aided design. Stacking of two or more 2D materials leads to van der Waals (vdW) heterostructures, which are promising for engineering of its electronic and optical properties. Delicate physical effects in heterostructures will be studied by costly many-body methods because the usual density functional theory cannot describe the corresponding physics correctly. The project will contribute to the fundamental understanding of vdW heterostructures and boost their experimental research as well as technological applications.

Řešitel: Dmytro Khursenko                     

OPEN-37-27                

Visual Question Answering for Autonomous Driving

Karolina GPU  Alloc=2900          

Improving autonomous driving safety requires vehicles to fully understand their environment, not just detect objects. This project investigates Vision-Language Models (VLMs) to enable self-driving systems to reason about and explain complex traffic scenarios in plain language. The research utilizes the open-source nuScenes dataset, which provides 360-degree surroundings through camera and LiDAR sensors, alongside the DriveLM dataset for structured questions and answers (QAs) on selected driving frames. To enhance these datasets, a Large Language Model acts as an orchestrator, combining extracted 2D and 3D object data to generate new information-rich QAs and scene captions. By training and evaluating open-source VLMs on this comprehensive language data and images, the model learns to better perceive risks, predict the movement of others, and plan safer maneuvers. Finally, the project aims to ensure AI decision-making is interpretable through clear, natural language explanations. Such transparency not only builds public trust in autonomous systems but also significantly contributes to safe autonomous vehicles by preventing accidents and protecting all road users.

Řešitel: Renáta Praksová 

OPEN-37-28                

Multi-Objective Optimization Techniques for Magneto-Biplasmonic Photonic Devices                 

Karolina CPU  Alloc=700;  Karolina GPU  Alloc=600 

The CIRCULIGHT project (EIC Horizon Europe Grant No. 101129645) is pushing the boundaries of photonics by developing the first fully integrated magneto‑biplasmonic optical circulator operating at the key telecom wavelength. At the heart of this effort is a challenging hybrid structure that must guide light smoothly between a conventional indium‑phosphide waveguide and a nanoscale plasmonic slot. Achieving this requires extremely precise design and optimization. After an initial optimization phase, based on 336 simulations, that already boosted device transmission by nearly 6% and cut unwanted reflections in half, the project is now scaling up [1]. Using the latest CST Studio Suite 2026 and advanced GPU acceleration on the Karolina supercomputer, the team will run 10,000 high‑fidelity simulations to refine the device further. This GPU‑based workflow delivers significant performance gains which showed the benchmark performed by e-INFRA project FTA-26-36: each simulation runs about 1.6× faster than on a CPU, and overall throughput is more than 12× higher.   To drive this large‑scale search efficiently, Python‑based multi‑objective optimization tools automatically generate new device geometries and submit them as parallel jobs through the cluster’s SLURM scheduler. By combining high‑performance computing with intelligent optimization, CIRCULIGHT aims to deliver a breakthrough optical component with major implications for next‑generation communication and sensing technologies.

Řešitel: Georgios Tolias     

OPEN-37-29                

Instance-level Visual Recognition and Generation   

LUMI-G  Alloc=13600                     

Recent advances in foundation models have greatly improved visual recognition and image generation, yet they still struggle at the instance level, where the goal is to distinguish, retrieve, localize, or generate a particular object rather than a broad semantic category. This project will study instance-level visual understanding in open-world settings, where new objects, domains, and visual conditions appear beyond the training data. We will develop and evaluate models for image retrieval, object localization, and image generation guided by text or visual exemplars, with a particular focus on fine-grained visual details and identity preservation. A central part of the project is the synergy between recognition and generation: generative models will provide diverse synthetic data and hard visual variations for training recognition systems, while recognition models will guide generation toward higher fidelity and better instance preservation. Large-scale GPU resources are essential for adapting foundation models, extracting and indexing descriptors over massive image collections, and generating synthetic training data at scale. The project aligns with the broader IVee research program (GACR) on instance-level visual recognition and generation.

Řešitel: Elisabet Maňásková                 

OPEN-37-3                   

Pairwise ranking for ML-guided directed evolution: benchmarking and therapeutic application                     

Karolina GPU  Alloc=1500          

Designing improved therapeutic proteins involves exploring a large number of possible mutations while relying on limited experimental data. We developed EVOLVEpro-Rank, a pairwise ranking model for data-efficient protein engineering that outperforms EVOLVEpro [1]. We applied our method to engineer staphylokinase, a low-cost thrombolytic drug candidate: in the first round of directed evolution, three single mutants with activity exceeding wild type were identified and experimentally confirmed at Loschmidt Laboratories, with the second round currently undergoing validation. Based on these results, the project focuses on three main tasks: (1) accelerating EVOLVEpro-Rank to enable large-scale variant screening using the established EVOLVEpro benchmark of 12 DMS datasets; (2) benchmarking the improved method against several baseline and state-of-the-art methods on a new large-scale benchmark derived from MaveDB — comprising 334 DMS datasets, surpassing the scale of ProteinGym [5], and filling a critical gap in current evaluation practices for ML-guided directed evolution; and (3) extending the staphylokinase engineering campaign to multi-mutant variants.

Řešitel: Linda Kučerová    

OPEN-37-30                

Adaptive genomics of grey wolf in Central Europe    

Barbora NG  Alloc=1000;  Karolina CPU  Alloc=5200;  LUMI-C  Alloc=300

The project is related to analysing the processes in re-expanding European wolf populations using a genomic approach. To study differences across particular populations and identify their environmental demands, we aim to identify islands of differentiation - regions where the fixation index (FST) is high. The fixation index is a measure of population differentiation resulting from genetic structure and is typically estimated using genetic polymorphism data such as single-nucleotide polymorphisms (SNPs) or microsatellites. A  useful tool for functional analysis is the MAVEN server (Narayanan et al., 2010). To monitor adaptive processes in wolves, we will examine candidate genes associated with phenotypic changes that facilitate adaptation to specific environments. Gene ontology plays an important role in this context, and tools such as the LiftOver utility in the UCSC Genome Browser (https://genome.ucsc.edu/cgi-bin/hgLiftOver) can be employed for this purpose (Caniglia et al., 2018). To investigate potential admixture among particular populations and hybridization events with dogs, we plan to apply the ABBA–BABA method (Green et al., 2010). This approach can help distinguish between ancestral polymorphism and genuine introgression (horizontal gene flow).   

Řešitel: Sofia Canola          

OPEN-37-31                

Photochemical formation of arynes: mechanisms and molecular design

Karolina CPU  Alloc=10100;  Karolina GPU  Alloc=1000                

The synthesis of large aromatic molecules (PAHs) with atomically precise structure is a complex task. Many synthetic strategies pass through specific molecular intermediates, arynes, which are highly reactive and capable of forming extended aromatic structures. Recently, their generation via light irradiation (photolysis) has been experimentally applied, however the underlying mechanism and the molecular structure-reactivity relation remain unexplored. For example, experiments show that the photolysis of precursors with anhydride functional group produces arynes, with an efficiency that strongly depends on the molecular structure. In this project, we propose to study the photochemical mechanism that forms molecular aryne intermediates from selected precursor molecules with anhydrides functional groups. With computational tools, we plan to look for systematic relations  between the molecular structure and the photoreactivity, particularly for molecules with extended aromatic backbone. Our objectives are to clarify the photoreaction mechanism of known compounds, derive predictive rules for larger derivatives, and ultimately propose molecular design guidelines to support experimental efforts.

Řešitel: Barbora Venosova                     

OPEN-37-32                

Impact of Vacancy Defects on the Structure and Properties of MXenes and MXene Quantum Dots             

Barbora NG  Alloc=7800;  Karolina CPU  Alloc=20700;  Karolina FAT  Alloc=100;  Karolina GPU  Alloc=2700                  

MXenes represent a rapidly expanding class of two-dimensional (2D) materials characterized by high  electrical conductivity, chemical stability, and tunable surface properties. These features make them  promising candidates for applications in nanoelectronics, spintronics, catalysis, and energy storage.  MXene quantum dots (MXQDs) have recently attracted significant attention due to quantum  confinement effects, which give rise to distinct electronic and optical properties compared to bulk  systems. Despite this growing interest, the role of structural defects, particularly vacancies, in MXQDs  remains insufficiently understood. Vacancy defects are expected to strongly influence the local  electronic structure, stability, and reactivity, making the identification of the most stable vacancy types  and configurations essential. Moreover, vacancy sites may act as active centers for the adsorption and  activation of small molecules relevant to environmental and energy applications, such as CO₂, hydrogen  (H₂), or toxic water contaminants like trichloroethylene (TCE). This project will systematically investigate  vacancy defects in MXQDs, with a focus on their stability and their impact on electronic properties. The  results will provide a foundation for future studies of catalytic processes, including gas adsorption,  hydrogen storage, and environmental remediation.

Řešitel: Nikola Machacova                     

OPEN-37-33                

Defects and strain engineering in non-collinear antiferromagnets and ferrimagnets

Karolina CPU   Alloc=6800;  Karolina FAT  Alloc=100

Antiferromagnetic spintronics is gaining interest due to zero stray fields and faster magnetization dynamics compared to ferromagnets. In certain antiperovskite compounds, such as Mn3NiN, magnetic atoms form a Kagome lattice. Combined with antiferromagnetic exchange, this geometry results in a complex non-collinear magnetic order. Since substrate lattice mismatch introduces epitaxial strain and defects, it is necessary to explore their effect. The project will first focus on the evolution of exchange interactions in strained systems. Subsequently, we will describe changes in electronic and magnetic properties in the vicinity of edge defect by the means of density functional theory. Finally, the data will be used in spin dynamics simulations to bridge various time and length scales. This research aims to deepen our understanding of spin dynamics in realistic non-collinear antiferromagnets, possibly facilitating the design of spintronic devices.

Řešitel: Lukas Neuman     

OPEN-37-34                

Inductive Bias of Deep Neural Networks for Computer Vision 

Karolina GPU   Alloc=1900         

Deep Neural Networks models have in the recent years dominated virtually all areas of Artificial Intelligence and Computer Vision. Their main advantage is that, given enough training samples, a training algorithm can automatically update network parameters to directly maximise given objective, such as image classification accuracy. Despite the recent success, the models are easily confused by trivial samples not present in the training set and even the largest models lack basic generalisation and reasoning abilities despite having hundreds of millions of parameters and despite being trained on millions of very diverse data samples -- suggesting that a fundamental piece of understanding is still missing.    We propose that one of the missing pieces in current models compared to humans is an appropriate inductive bias -- the set of prior assumptions used to generalise and make a prediction based on a finite set of training samples. In this project, we want to exploit this observation and search for new inductive biases to incorporate them into modern Deep Neural Networks used in common Computer Vision tasks. This will result in Deep Neural Network models which require less parameters, which are more efficient, which are less confused by out-of-distribution data samples and which require less training data, as using an appropriate inductive bias is likely equivalent to even exponentially less training data.

Řešitel: Jai Bardhan             

OPEN-37-35                

World Models for Robotic Applications    

LUMI-G  Alloc=600  

Robot World Models aim to close the performance gap between humans and robots on manipulation tasks by enabling robots to imagine how their actions will affect the scene before executing them. Trained on large-scale interaction data, these models learn to simulate the physical world, predicting how objects move, deform, and interact in response to a robot's actions. By \thinking ahead\" in this way, robots can plan more effectively, recover from mistakes, and generalise to novel objects and scenes — capabilities that are essential for deploying manipulation systems outside of controlled laboratory settings."

Řešitel: Martin Sedláček  

OPEN-37-36                

Video Action Models for Generalist Robotic Manipulation          

LUMI-G  Alloc=13400                     

Recent advances in learning generalist policies for robotic manipulation [1,2,3,12] from large-scale datasets [4,5] have demonstrated remarkable success in learning diverse tasks including cloth folding, opening doors, drawers, and sorting various objects in an open-vocabulary setting. Specifically, a new paradigm of Video-Action (or World-Action) models [6,7,8] stands out in their remarkable adaptability across objects, scenes, and other factors; achieving top-most rankings in various challenging benchmarks - both in simulation and the real-world [9].   Such models can “imagine\ the physical consequences of their movements before executing them while predicting the causal action corresponding to the imagined future (such as the target joint angles of a robot arm or pose of the end-effector in 3D space). While this approach works remarkably well for common manipulation tasks, these models still struggle to match human dexterity and precision on tasks such as assembling small parts, turning a key, and long horizon manipulation that requires memory.   With the use of our own cutting-edge tools for evaluating generalist policies - including a highly realistic simulation [10] and a World Model [11] - we will stress-test and understand these new models, uncovering their limitations lying beyond the demos, and develop new methods to increase robustness and acquire new skills without forgetting.  "

Řešitel: Klára Mitošinková

OPEN-37-37                

Full-Orbit Modeling of Fast and Thermal Ions in Tokamak Environment    

Karolina CPU  Alloc=5000          

Fusion energy offers a clean and sustainable power source, but controlling hot plasma remains a major challenge. This project studies the behavior of hydrogen plasma ions with focus on the plasma edge, a region that strongly influences overall performance.  Fast ions, generated by heating systems, can affect plasma stability and may help create an electric field needed to access high-confinement operation. Using advanced simulations, this work investigates their role and predicts their behavior in the upcoming COMPASS Upgrade tokamak in Prague as well as DIII-D tokamak operated in the USA.  The project also examines thermal ions, which may deviate from commonly assumed maxwell distribution due to collisions and losses at the plasma edge. These effects can influence electric fields and plasma rotation.  By improving our understanding of main ion dynamics, this research supports the development of more efficient and reliable fusion devices.

Řešitel: Pavel Baláž             

OPEN-37-38                

Quantum State Tomography from Incomplete set of Measurements          

Karolina CPU  Alloc=600;  Karolina GPU  Alloc=200;  VLQ  Alloc=43200

Quantum computers and quantum technologies rely on our ability to precisely characterize quantum states — the fundamental carriers of quantum information. This characterization, known as quantum state tomography, requires reconstructing the density matrix of a system. For a system of N qubits, this demands at least 4N-1 independent measurements, each repeated thousands of times to suppress statistical noise. The resulting financial and time costs grow exponentially with system size, making full tomography of larger quantum systems practically unfeasible.    Two complementary strategies have emerged to tackle this challenge: working with an incomplete set of measurements, or accepting noisy measurements with fewer repetitions. However, a unified framework capable of handling both limitations simultaneously has remained elusive — until now.    We propose to develop a diffusive transformer model that bridges these two regimes. Thanks to its stochastic nature, the diffusion model is a natural fit for handling measurement noise and uncertainty. As a generative model, it can produce entire ensembles of density matrices consistent with a given incomplete dataset, enabling principled uncertainty quantification. Crucially, the transformer architecture brings exceptional scalability, opening the door to quantum state tomography well beyond the two-qubit systems accessible today.    High-performance computing resources are essential to train and benchmark these models at scale — across system sizes, noise levels, and competing architectures — and to make this approach viable for real-world quantum devices. 

Řešitel: William Shakespeare Morton                     

OPEN-37-39                

Designer Disordered Proteins  

Karolina GPU  Alloc=1300          

Disordered proteins don't fold into a single shape, but constantly shift between conformations. Despite this, they carry out important jobs in the cell, from regulating transcription to organizing proteins into condensates. The problem is that our usual tools for comparing proteins rely on structure and sequence alignment, and neither works for disordered regions. In this work we are building a computational space where any disordered protein can be positioned based on both its physical properties (how compact it is, whether it phase separates) and its biological function. This lets us design changes to a disordered region, for instance stopping it from phase separating, without disrupting what it does in the cell. We will validate our designs using coarse grained molecular dynamics simulations, confirming that designed sequences behave as predicted. This matters because many diseases, such as cancers and alzheimer's, involve disordered regions with altered physical properties. If we can correct those alterations while keeping biological function intact, that opens the door to new ways of tracking down disease prone proteins, and developing targeted therapies to correct their function.

Řešitel: Jan Zemen               

OPEN-37-4                   

Magnetic anisotropy in Heusler superlattices             

Barbora NG  Alloc=10000;  Karolina CPU  Alloc=16000                 

We propose first-principles simulations of interface-induced magnetic anisotropy in multilayers based on Heusler alloys, with primary focus on Ni2MnX/A superlattices (A = Cr, Ni, Mn, Tb, Dy; X = Ga, In, Sb). The aim is to determine how the magnetic anisotropy depends on the interface composition, layer thickness, and in-plane biaxial strain imposed by the substrate. The calculations are motivated by the discrepancy between the large magnetic anisotropy predicted for bulk martensitic Ni2MnGa and the substantially smaller values observed experimentally, which suggests an important role of interfaces and microstructure. Using DFT calculations, we will first relax the multilayer structures and then evaluate their magnetic anisotropy energy. The results will identify the most promising material combinations for interface engineering of magnetic properties. More broadly, the project will contribute to the design of functional intermetallic multilayers for applications in sensing, energy-conversion, information storage and processing.

Řešitel: Ondřej Vysocký    

OPEN-37-40                

Development and deployment of SEANERGYS solution               

Barbora NG    Alloc=300;  Karolina CPU  Alloc=400;  Karolina GPU  Alloc=100;  LUMI-G  Alloc=2700            

SEANERGYS is a EuroHPC project that creates an integrated European software solution that optimises resource utilisation and reduces the energy used for real-world workload mixes. It therefore improves the throughput of HPC systems, generating more R&D results for a given energy budget. The solution consists of a comprehensive monitoring infrastructure (CMI), an Artificial Intelligence data analytics system (AIDAS), and a dynamic scheduling and resource management system (DSRM).      The SEANERGYS solution will also be deployed at the IT4Innovations Karolina system. To ensure production-quality, the project builds on results from European projects, the competency of well-established research groups and companies, and widely used open-source codes. These are inputs for an integrated software system that achieves the functionality, performance and stability needed by European HPC centres. The IT4Innovations team significantly contributes to the SW stack from their MERIC Energy Aware suite, the flagship project of the IT4Innovations [3].      An agile, professional software development method based on Continuous Integration/Continuous Deployment (CI/CD) is complemented by a staged testing and validation process, with functionality tests on single-nodes, scaling tests on mid-sized platforms, and finally acceptance tests on production supercomputers.

Řešitel: Ahmed Alasqalani                      

OPEN-37-41                

Data-driven Approach to Design W-based High Entropy Alloys with Excellent Radiation Resistance (ADHEARR)       

Barbora NG  Alloc=4100;  Karolina CPU  Alloc=13000                    

As the demand for ever more advanced nuclear reactors pushes the capacity of the current energy system, advanced nuclear systems are seeking materials with superior high-temperature properties, as well as high corrosion and radiation resistance, among others. Recent advances in the field of alloys have led to the discovery that some selected high-entropy alloys (HEAs) possess excellent properties that address the increasing demand for the previously mentioned characteristics, in addition to no grain coarsening and self-healing abilities under irradiation. In this way, HEAs are considered to be potentially the next generation of nuclear materials for future fission or fusion reactors, designed to operate at higher temperatures and radiation doses than previously known materials. Unlike conventional alloys, which contain one or two base alloys with possible multiple minor additions of other elements, HEAs are characterized by being composed of multiple principal elements. This considerably extends the range of possible HEA compositions, and HEAs are at present of great research interest in materials science and engineering. Additionally, new theories, models, and methods will be developed for their study. To address this, we will utilize high-throughput computational methods, including density functional theory and molecular dynamics simulations, to explore the composition space of MoNbTaW HEAs and predict their phase stability and mechanical properties.

Řešitel: David Adamczyk  

OPEN-37-42                

Exploiting Forking Tokens in Reinforcement Learning for LLMs

Karolina GPU   Alloc=1000;  LUMI-G  Alloc=1700       

When large language models are trained to reason using reinforcement learning, they must explore, trying different reasoning paths and learn from feedback. Recent findings show that only a small fraction of the tokens a model generates carry real uncertainty, while the rest are predictable continuations. These rare high-uncertainty tokens act as decision points that steer the reasoning process, yet current training methods treat all tokens the same.  We plan to study these decision points and develop training methods that focus exploration where it matters. We will first work with compositional reasoning tasks over symbolic inputs, where the location and structure of every decision point is known by design. This lets us precisely measure how uncertainty evolves during training and how it affects the model's ability to generalize to new problem combinations. We will then apply the resulting techniques to train open-weight language models on mathematical reasoning benchmarks using the Karolina and LUMI-G clusters. By directing exploration to the positions where it is most productive, these methods have the potential to make reinforcement learning for language models more efficient and to produce models that generalize more reliably to unseen problems.

Řešitel: Oldřich Plchot       

OPEN-37-43                

Joint Speech And Text Training For LLM-based Components of Conversational Dualogue Systems              

LUMI-C  Alloc=1400;  LUMI-G  Alloc=13200                   

This work is linked to the  H2020 ELOQUENCE project, which develops multilingual conversational systems and cross-lingual Factual Information Retrieval (FIR) capabilities to reduce reliance on handcrafted rules. As a core objective of ELOQUENCE is to build upon prior achievements in speech and language foundation models, we propose an architecture and methodology to effectively interconnect these models while leveraging limited open-source resources within individual EU languages. Specifically, we introduce a joint speech and text training framework for end-to-end spoken dialogue state tracking (DST) that addresses the scarcity of annotated audio data. By training on speech data from one domain alongside text from others, the model learns to extract structured information from the target domains. Our experiments show that this approach recovers up to 79% of the performance gap compared to models trained on target-domain speech, without adding computational latency during inference. By taking advantage of text augmentation techniques, we want to scale textual data to train a “Universal conversational model” that not only extracts dialogue states but also generates responses to the users.

Řešitel: Martin Žonda         

OPEN-37-44                

Phase dynamics in complex Josephson junctions    

Barbora NG  Alloc=2100;  Karolina CPU  Alloc=1900;  Karolina GPU  Alloc=100;  LUMI-C  Alloc=700;  LUMI-G  Alloc=200            

Josephson junctions exhibit rich phase dynamics central to superconducting electronics and quantum technologies. While the standard resistively capacitance shunted junction (RCSJ) model captures basic behavior, many experimentally relevant systems involve greater complexity, including multi-junction configurations, strong environmental coupling, and colored noise, leading to non-Markovian dynamics beyond conventional descriptions.    We propose to develop and apply Neural Ordinary Differential Equations (Neural ODEs) and Neural Stochastic Differential Equations (Neural SDEs) to model the phase dynamics of such complex Josephson systems. These approaches enable data-driven reconstruction of effective dynamics, including memory effects and structured noise, directly from simulations and experimental data.    Our focus is on escape and switching dynamics from metastable states under dc and ac bias, extending to coupled junctions and environments with nontrivial noise spectra. This work is motivated by recent experiments showing clear deviations from RCSJ predictions. The proposed framework will provide efficient and flexible tools for simulating and interpreting non-Markovian Josephson dynamics in realistic settings.

Řešitel: Marie Behounkova                     

OPEN-37-45                

Modeling Evolutionary Pathways of Moons and Exoplanetary Systems     

Karolina CPU  Alloc=2000          

Moons and planets are not static worlds, but dynamic systems shaped by internal heating, gravitational forces, and orbital motion. In many cases, these processes generate melting inside planetary interiors, which can drive geological activity and even create conditions suitable for life. However, much of what happens beneath the surface remains hidden, and current models often fail to capture how these processes interact. This project will use advanced computer simulations to study how (1) internal evolution and presence of melting, (2) tidal forces, and (3) orbital dynamics work together to shape the evolution of moons and exoplanets. By focusing on well-known bodies such as Jupiter’s moon Europa and planets in the TRAPPIST-1 system, we aim to understand how internal processes influence observable features such as surface deformation and heat release. The results will be relevant for interpreting data from ongoing and future space missions, improving our understanding of potentially habitable environments beyond Earth, and could contribute to the broader exploration of our Solar System and other planetary systems.

Řešitel: Jindřich LNĚNIČKA                    

OPEN-37-46                

Investigating behaviour and dynamics of glucose transporters and possible small-molecule interactions                     

Barbora NG  Alloc=300;  Karolina CPU  Alloc=400;  Karolina GPU  Alloc=100;  LUMI-C  Alloc=400;  LUMI-G  Alloc=3200               

Glucose transporters are a wide group of membrane-embedded proteins, that facilitates glucose transport through membrane. We are planning to focus on the GLUT family found in humans, expressed primarily in the brain and erythrocytes (red blood cells). Therefore we will be focusing on their effect on the epilepsy and other related diseases. We will study the various possible forms and mutations of the protein as well as interactions with potentially efficient drugs.

Řešitel: David Zihala           

OPEN-37-47                

Circulating Tumor Cells in Multiple Myeloma: From Biology to Blood-Based Genomic Diagnostics             

Karolina CPU   Alloc=10600       

Multiple myeloma is a type of blood cancer where abnormal plasma cells multiply uncontrollably within the bone marrow. While these cells typically remain in the marrow, some escape into the bloodstream. These \circulating tumor cells\" (CTCs) are an important indicator of how the disease might progress, with higher numbers generally pointing to worse outcomes for patients. Currently, doctors rely on bone marrow biopsies—an invasive and uncomfortable procedure—to study the disease and determine the best course of treatment. Our project aims to show that we can gather this crucial genetic information using a standard, much less invasive blood draw. By analyzing the complete DNA of the tumor cells found in the blood, we hope to prove that these routine \"liquid biopsies\" can effectively assess a patient's risk level just as accurately as traditional methods. Additionally, we want to understand the biological factors that cause these cells to enter the bloodstream. Previous research has only focused on patients with very high numbers of CTCs. By comparing the genetic data of patients across the entire spectrum—including those with zero CTCs—we aim to uncover what drives the cancer to spread. Ultimately, this research strives to replace difficult procedures with a simple blood test, leading to more comfortable and personalized management of multiple myeloma."

Řešitel: Martin Zelený         

OPEN-37-48                

First-principles description of the hypothetical fcc-Ge phase for Al–Ge phase-diagram prediction              

Barbora NG  Alloc=2400;  Karolina CPU  Alloc=12600                    

Understanding phase diagrams is essential for designing new alloys with desired mechanical, physical, and chemical properties. Reliable prediction of phase stability requires thermodynamic data not only for stable phases, but also for hypothetical reference phases that may not exist under normal conditions. In the Al–Ge system, such a phase is fcc Ge. While stable fcc Al and diamond Ge can be described using standard first-principles methods, hypothetical fcc Ge is much more difficult because it is dynamically unstable and cannot be treated by conventional phonon calculations. The aim of the proposed project is to develop a computational methodology for obtaining thermodynamic properties of hypothetical fcc Ge from first principles. The project will combine quantum-mechanical calculations with constrained atomistic simulations that keep the system in an fcc-like state and then reconstruct the corresponding unbiased free-energy branch. In this way, temperature-dependent Helmholtz free energy, Gibbs free energy, and heat capacity of hypothetical fcc Ge will be obtained. These data will provide the missing thermodynamic input needed for more reliable prediction of phase stability and solubility in the Al–Ge system. More generally, the project will establish a transferable methodology for describing unstable hypothetical phases needed in future computational prediction of complex phase diagrams.

Řešitel: Paulina Lewandowska            

OPEN-37-49                

Benchmarking of hybrid HPC+QC architectures based on Heavy Output Generation problem

Karolina CPU  Alloc=3;  Karolina GPU  Alloc=36;  VLQ  Alloc=36000            

As high-performance computing (HPC) converges with emerging quantum computing (QC) technologies, there is a growing need for reliable methods to evaluate and compare hybrid systems. This project proposes a benchmarking approach based on the “heavy output generation” (HOG) [1,2,3] problem, a task widely used to assess quantum computational advantage [4,6].  The research will design and implement benchmark protocols that measure how efficiently HPC+QC systems generate statistically significant heavy outputs-results that occur with higher-than-median probability in quantum circuits. In addition, we will perform standardized Quantum Volume [2,3] experiments on a VLQ quantum computer as well as on hybrid HPC+QC infrastructures, enabling direct comparison between standalone quantum hardware and integrated hybrid workflows.  For the hybrid approach, we will employ circuit cutting techniques [5], which allow large quantum circuits to be partitioned into smaller subcircuits that can be executed across quantum and classical resources. This enables scaling beyond current hardware limits while maintaining meaningful benchmarking capabilities.  Methods include developing standardized workloads, executing experiments across diverse platforms, and analyzing key performance metrics such as fidelity, runtime, scalability, and resource efficiency. The project will also explore optimization strategies that improve coordination between classical and quantum components.  The expected outcome is a practical and widely applicable benchmarking framework that enables fair and transparent evaluation of hybrid computing systems, supporting the advancement of next-generation computing technologies.

Řešitel: Tadeáš Kalvoda    

OPEN-37-5                   

Solvation structure of selected ions in aqueous solution determined via density functional theory and ab initio molecular dynamics

Karolina CPU  Alloc=4800          

The interactions of various transition, rare metal, or lanthanide ions with water molecules in aqueous solution, resulting in the formation of solvation shells, play an important role in various aspects, desired or undesired, of (bio)chemical reactions involving these ions, like binding to enzymes, luminiscence, or redox. Therefore, an accurate description of solvation effects of such ions is desirable, potentially improving the modelling of thermodynamic properties of metal-ligand reactions, as well as drug design involving these ions. We will therefore use density functional theory to determine the structure of selected metal ions in water.

Řešitel: Ryszard Stefan Kukulski         

OPEN-37-50                

Probabilistic quantum error correction for randomized Ising problem optimization 

VLQ  Alloc=14400    

Quantum computers promise advances in solving complex optimization problems, many of which can be expressed using the Ising model. However, their practical use is limited by noise, which corrupts fragile quantum information. This project explores a novel approach to quantum error correction dedicated for Ising-based optimization called probabilistic quantum error correction. This method uses selective acceptance to determine whether errors have been successfully corrected. Probabilistic methods can outperform traditional deterministic approaches by enabling correction under a wider range of noise conditions. In particular, when errors arise from interactions with small auxiliary quantum systems, our framework allows for efficient recovery using minimal additional resources. This research paves the way for more resilient quantum computations for limited size quantum computers.

Řešitel: Ryszard Stefan Kukulski         

OPEN-37-51                

Quantum error correction for Ising model solutions enhanced through reinforcement learning agents     

VLQ  Alloc=36000    

Mathematical optimization lies at the heart of modern science and industry, powering applications from internet routing and drug discovery to financial planning. Many of these complex problems can be reformulated using the Ising model—a mathematical framework that is particularly well-suited for emerging quantum computing technologies. However, today’s quantum devices, known as noisy intermediate-scale quantum (NISQ) computers, are highly sensitive to errors caused by environmental noise, limiting their practical performance. This project aims to develop novel quantum error correction strategies tailored specifically for Ising model-based optimization. By integrating reinforcement learning, we propose an adaptive approach in which an intelligent agent learns to design and refine quantum error correction codes through repeated interaction with noisy quantum systems. Over time, the agent identifies patterns in the noise and optimizes the codes to improve computational reliability and efficiency.

Řešitel: Ondřej Kobza         

OPEN-37-52                

Trustworthy Agentic Coding      

Karolina CPU  Alloc=300;  Karolina GPU  Alloc=2200;  LUMI-C  Alloc=300;  LUMI-G  Alloc=4000                     

Recent advances in generative AI have made it possible for language models to assist with increasingly complex programming tasks, including work across entire software repositories. However, these systems can still produce insecure, unreliable, or even harmful code, which creates serious risks for software development and cybersecurity. In this project, we propose a secure agentic coding assistant designed to generate, test, and refine code while actively checking for vulnerabilities and unsafe behavior. The system combines a central orchestration module with specialized tools for code generation, automated testing, static analysis, and LLM-based security evaluation. We will also develop new training methods to improve the model’s reasoning, safety, and robustness, including adversarial reinforcement learning and a novel evolutionary extension of policy optimization. To support both training and evaluation, we will build a synthetic data generation pipeline and assess the system on public benchmarks and security-focused datasets. The result will be a practical and open framework for safer autonomous code generation.

Řešitel: Ales Prachar           

OPEN-37-53                

DeltaWT  

Barbora NG   Alloc=700;  Karolina CPU   Alloc=800  

Delta-wing aircraft perform especially well during high-speed flight and at high angles of attack—conditions under which conventional wings would typically stall. Instead of experiencing abrupt flow separation, delta wings generate what is known as vortex lift. The sharp leading edges of the wing create controlled, spinning vortices over the upper surface. This characteristic makes delta-wing configurations particularly well-suited for agile, high-performance aircraft.    In wind tunnel tests of the aircraft, there are several simplifications to the real operation. First, there is an effect of wind tunnel walls and model supports, which interfere with the aerodynamic properties. The effect of propulsion is also missing for most wind tunnel models.    The role of CFD is to provide corrections with respect to the mentioned effects, which can be incorporated into the results, giving more accurate representation of the reality. This is important for the flight control algorithms and for correct estimation of the flight performance. 

Řešitel: Filip Jozefov            

OPEN-37-54                

MassSpecGym v2: A trustworthy benchmark for molecular discovery from mass spectrometry                    

Karolina CPU  Alloc=100;  Karolina GPU  Alloc=1100;  LUMI-C  Alloc=200;  LUMI-G  Alloc=7300                     

Mass spectrometry is a key technology for identifying small molecules in biology, medicine, and  environmental science, but most detected signals still cannot be linked to exact chemical structures.  This limits progress in metabolomics, biomarker discovery, drug development, and environmental  monitoring [1].  Our team recently introduced MassSpecGym, the first large public benchmark for AI methods that  predict molecular structure from tandem mass spectra [1]. Since its release, new spectral libraries  have become available and benchmark use has revealed artifacts that should be corrected for fairer  and more realistic evaluation.  We will develop MassSpecGym v2, a larger and more reliable benchmark for computational mass  spectrometry. We will expand the dataset with new LC-MS/MS and curated GC-MS spectra, improve  metadata and structure harmonization, correct evaluation artifacts, add new tasks including novelty  detection and adduct-aware search, and retrain leading models on the corrected benchmark. The  result will be open data, reproducible evaluation, and stronger baselines for trustworthy molecular  discovery.

Řešitel: Fabien Jaulmes    

OPEN-37-55                

Computational modelling of fast ion orbits and their consequences in tokamak       

Barbora NG  Alloc=8300;  Karolina CPU  Alloc=100  

Nuclear fusion will enable us to generate energy without releasing large amounts of greenhouse gases into the atmosphere or leaving behind us long lived radioactive waste. The tokamak concept involves the use of magnetic fields to confine plasma hot enough to sustain fusion within itself.   COMPASS Upgrade (COMPASS-U) will be a large magnetic field (5T) tokamak that will allow the scientific investigation of various physics issues related to the operation of the future ITER. In particular, an 80keV Neutral Beam Injection (NBI) system is planned to heat up the plasma with 4MW of external power. Such a unit was tested on the COMPASS tokamak before its shut down and our modelling contributed to the interpretation of the results. The study and modelling of NBI-born particle behavior is of great relevance: it might influence future design of the system and its integration in the overall reactor design. We request computational time for the modelling of the interaction of the fast particles and the design of the fast ions related diagnostics. Our code, EBdyna, with its new collisional features, was benchmarked against the NUBEAM code on several test cases. A publication in the Nuclear Fusion journal summarizes the results of our initial modelling effort [ F. Jaulmes, et al., 2021 Nucl. Fusion 61 046012, Modelling of charge-exchange induced NBI losses in the COMPASS Upgrade tokamak, Open Access: https://iopscience.iop.org/article/10.1088/1741-4326/abd41b/pdf]. The simulations allow for a thorough investigation of the interaction of the fast particles with the edge of the confined plasma. 

Řešitel: Václav Bazgier      

OPEN-37-56                

Integrated geometric and deep learning-based discovery of novel cannabinoid receptor modulators from biological assemblies

Karolina CPU  Alloc=900             

The human cannabinoid system is a key therapeutic target for treating chronic pain, inflammation, and neurodegenerative disorders such as Parkinson’s disease. This project aims to discover novel natural modulators of cannabinoid receptors by leveraging the supercomputing power of IT4Innovations. The research focus is on the large-scale analysis of over 350,000 Biological Assemblies from the Protein Data Bank.    By utilizing the Karolina CPU partition, we will perform a high-throughput geometric screening using the MOLE 2.0 software to identify hidden access tunnels and binding cavities within these complex protein structures. This structural information will then be used for extensive molecular docking simulations (AutoDock Vina) to evaluate the binding potential of compounds from natural product databases, such as SANCDB and DrugBank. This integrated \digital-first\" approach significantly accelerates the drug discovery process and provides a cost-effective path toward developing safer, non-synthetic therapeutic alternatives. The results of this pilot study will serve as a foundational dataset for future large-scale pharmacological research and national grant applications."

Řešitel: Ivan Zelinka            

OPEN-37-57                

QUASION - Quantum algorithms sythesis and verification         

VLQ  Alloc=50000    

The QUASION project (Quantum Algorithms Synthesis and Verification) focuses on the AI-assisted design, optimization, and validation of quantum algorithms for near-term and future quantum computing platforms. Its goal is to develop computational workflows that automatically propose candidate quantum circuits, evaluate their correctness and robustness, and verify their behavior under realistic simulation and hardware constraints.  The project will use HPC, GPU, and quantum simulation resources to support large-scale experimentation with quantum circuit synthesis, hybrid quantum-classical optimization, formal and empirical verification, and benchmarking across different software and hardware backends. Special attention will be given to the use of artificial intelligence methods for guided search in the space of quantum circuits, automated tuning of algorithmic parameters, detection of failure modes, and improvement of circuit quality with respect to depth, gate count, fidelity, and hardware compatibility.  These resources will primarily support PhD education and advanced student training by enabling work on reproducible, research-oriented tasks in quantum programming, algorithm design, and verification. At the same time, the project will provide demonstrators for teaching quantum informatics, allowing students to observe how quantum algorithms are synthesized, tested, and compared in practice on simulators and selected real quantum platforms. The expected impact includes new methods for AI-driven quantum algorithm engineering, stronger expertise in trustworthy and scalable quantum software development, and the creation of reusable experimental pipelines for research, doctoral education, and modern teaching in quantum computing.

Řešitel: Rafal Porowski      

OPEN-37-58                

FD-Based Optimization of Hydrogen–Methane Combustion in Industrial Burners     

Karolina CPU  Alloc=43400        

At present, there is strong pressure to decarbonise industry and other energy-intensive sectors, as well as to move away from the combustion of fuels. Unfortunately, many processes, whether in metallurgy or the chemical industry, still cannot operate without heat sources based on combustion. Therefore, hydrogen is one of the progressive alternative fuels, as it is carbon-neutral when produced by water electrolysis. Currently, it is not possible to immediately convert all infrastructure to 100% hydrogen, but hydrogen can be blended with methane to create a mixture that reduces methane consumption while maintaining properties similar to pure methane from the user’s perspective. At present, the main objective is to determine the maximum amount of hydrogen that can be added to methane without affecting flame characteristics or other combustion equipment parameters. Outcomes of this project will also contribute to the hydrogen safety knowledge base, building on the PI's ongoing work within the CESAR (Centre of Excellence for Safety Research) Horizon Europe project (project No.101186946) on self-ignition and accidental ignition mechanisms.

Řešitel: Roman Bushuiev 

OPEN-37-59                

DreaMS-Mol: Contrastive deep learning for molecular structure identification from mass spectra              

LUMI-G  Alloc=6800

Advancing scientific knowledge in life sciences and drug discovery relies on identifying novel molecules. However, less than 10% of the chemicals present in the human body and plant kingdom have been characterized to date. Mass spectrometry is the primary experimental technique for molecular identification, yet interpreting its data remains a major bottleneck due to the scarcity of annotated spectral libraries. In our previous projects (OPEN-26, OPEN-29), enabled by IT4Innovations high-performance computing, we developed DreaMS, a self-supervised Transformer model trained on 700 million unannotated mass spectra, published in Nature Biotechnology. We also established MassSpecGym, the first comprehensive benchmark for molecular structure identification, published at NeurIPS 2024 as a Spotlight paper. Building on these foundations, we propose DreaMS-Mol, a contrastive learning model combining DreaMS with a pretrained molecular encoder to retrieve and score candidate molecular structures for unknown spectra. Preliminary results show up to 140% improvement over the state-of-the-art. LUMI GPU resources are essential to scale DreaMS-Mol training and enable generative screening of novel molecular candidates.

Řešitel: Jana Pavlů                

OPEN-37-6                   

Theoretical investigation of the nitrogen-doped 4H-SiC/δ-Ni₂Si system   

Barbora NG  Alloc=7700;  Karolina CPU  Alloc=14600;  Karolina GPU  Alloc=100        

The growing demand for high-power, high-frequency, and high-temperature electronics has highlighted silicon carbide (SiC) as a key material due to its excellent electrical properties. Among its polytypes, 4H SiC is particularly suitable for power devices due to its wide band gap and high carrier mobility. Achieving reliable, low-resistance ohmic contacts is essential for device performance. For nitrogen-doped SiC, nickel-based metallization is commonly used, where high-temperature annealing forms primarily δ-Ni₂Si silicide. This phase plays a crucial role in reducing the electron injection barrier and enabling low-resistance contact behavior. Despite extensive experimental investigations, the microscopic origin of this barrier reduction and the role of the preferentially oriented δ-Ni₂Si remain insufficiently understood. In particular, a quantum-mechanical description is required to capture the atomic-scale mechanisms governing contact formation. Our calculations provide direct insight into the electronic structure of the nitrogen-doped 4H-SiC/δ-Ni₂Si interface, including charge redistribution, interface-induced states, band alignment, and electrostatic potential profile across the interface. Such an analysis is crucial for determining key quantities such as the work function of the δ-Ni₂Si, the electron affinity of 4H-SiC and the resulting barrier height. Our approach represents an essential step toward a fundamental understanding of 4H-SiC/ δ-Ni₂Si ohmic contact formation.

Řešitel: Leos Pohl                  

OPEN-37-60                

Scale dependence of pressure–strain decompositions in 2D plasma turbulence      

Karolina CPU  Alloc=5200          

Turbulence is a ubiquitous phenomenon in space and laboratory plasmas. The solar wind, a flow of magnetised plasma from the Sun, is an example of a turbulent system with a low collision rate between its constituent particles. In turbulent systems, energy is distributed across a wide range of spatial scales that are coupled through nonlinear interactions. The energy flows, or cascades, from large scales to small ones. At small scales, energy is transferred to particles and can eventually be converted into heat. In ordinary fluids, heating is mainly associated with irreversible particle-particle collisions. However, when collisions are rare, as in the solar wind, particle energisation can proceed through reversible channels. This process is still not well understood: the combination of turbulent fluctuations and weak collisions leads to complex particle distribution functions and their energisation. Moreover, space and astrophysical plasmas are magnetised, and the background magnetic field affects the turbulent dynamics and the associated energy-transfer processes. To gain better insight into this problem, numerical simulations are necessary. We propose to study plasma turbulence in the context of the solar wind, a weakly collisional turbulent system. We will use a two-dimensional hybrid code, in which electrons are treated as a fluid and ions as particles, to study the properties of the turbulent cascade toward ion scales, the transfer of energy across spatial scales, and the role of reversibility in these processes.

Řešitel: Martin Elias             

OPEN-37-61                

Augmenting NuScenes with Spatial Reasoning          

Karolina CPU  Alloc=100;  Karolina GPU  Alloc=2900

This project addresses the critical need for advanced reasoning capabilities in autonomous driving models. Current models often struggle to interpret complex, dynamic scenes involving movement and spatial dependencies. We aim to augment the nuScenes dataset—prominently used in the DriveLM competition —with multi-frame spatial reasoning and LiDAR sensor data. By leveraging video language and spatial models, we extract scene descriptions and object detections from videos. We then feed this information to large language models to analyze sequential frames, reason over objects and their trajectories to generate a robust dataset of Question-Answer pairs. We will initially match the original 300k training QA pairs for direct comparison with the original DriveLM dataset, but ultimately aim to exceed this by generating around 750k pairs per dataset variant. Additionally, we will conduct smaller-scale testing of state-of-the-art LLMs to evaluate their inherent capabilities in reasoning about spatial information like distance and speed. Our initial tests showed performance increase with only a 30k sample of this augmented data. The final dataset will be publicly available benefiting the autonomous driving research community.

Řešitel: Pavlo Polishchuk 

OPEN-37-62                

Fragment-based de novo design and structure optimization    

Karolina CPU  Alloc=1100;  Karolina GPU  Alloc=900

Fragment-based approaches have become an important part of modern chemoinformatics and drug discovery. Instead of screening only large, complex molecules, they start from small chemical fragments that bind efficiently to a biological target and can then be expanded or combined into more potent compounds. This strategy helps explore chemical space more effectively, supports rational design, and is well suited for computational methods that predict binding, optimize structures, and guide synthesis.  Within this project, we develop computational tools for fragment-based drug discovery and validate them in a real medicinal-chemistry application focused on the kinase DYRK1B. DYRK1B is involved in the regulation of cell growth, metabolism, and survival, and its abnormal activity has been linked to cancer and other diseases. The development of DYRK1B inhibitors may therefore open new opportunities for targeted therapy, while also providing an excellent test case for advanced computational methods in fragment-based drug design. 

Řešitel: Jiri Tomcala             

OPEN-37-63                

Quantum Computing for ZEUS project      

Barbora NG  Alloc=1000;  Karolina CPU  Alloc=3300;  Karolina FAT   Alloc=10;  Karolina GPU  Alloc=300;  VLQ  Alloc=360000  

The ZEUS project focuses on research and development of technologies for the production, transmission and storage of energy. The purpose of this subproject is to use quantum computing for power flow analysis and also for the optimization of community power networks. Work on this subproject will therefore develop in two directions.   In one direction, it will be the first practical implementation of a quantum algorithm for calculating state values of a transmission energy network, the so-called Quantum Power Flow (QPF). So far, this method has only been proposed in theory and verified on a simulator of a gate-based quantum computer. A practical implementation on the VLQ quantum computer will therefore be of potentially great value to the expert community in this field.  The second direction will involve the quantum optimization of community energy network interconnections with a view to maximizing the use of local energy sources. This is essentially an optimization task of minimizing energy overflows from the community energy network to the public distribution network. Quantum optimization can be performed using quantum annealing using the Optimal Power Flow (OPF) method or on gate-based quantum computers, such as VLQ, using Quantum Approximate Optimization Algorithm (QAOA). 

Řešitel: Anastasia Ostapenko              

OPEN-37-64                

Fine-tuning video generation models for driving scenario synthesis and editing         

Karolina CPU  Alloc=100;  Karolina GPU  Alloc=2500

Autonomous driving systems require large amounts of diverse visual data, but collecting real driving   videos for all weather conditions, road types, and rare situations such as traffic collisions is expensive and   time-consuming, as it depends on human annotators, drivers and domain experts. This project that is   carried out as a Master’s thesis investigates how modern AI video generation models can be used to   create and edit realistic driving videos for research, development, and testing.   In particular, we plan to utilize the NVIDIA Cosmos platform and related generative models such as   MAD and LTX to produce driving scenarios guided by multiple inputs, such as text descriptions,   reference images and edge images. The research will also investigate whether existing driving videos can   be meaningfully edited, for example by changing weather conditions, modifying the scene type, or   inserting new objects into the environment.   If successful, the generated scenarios could support research on autonomous driving and improve testing   workflows for driving systems. The project will evaluate not only the visual quality of the generated   videos, but also the practical usability of the selected models, including ease of fine-tuning and suitability   for real development pipelines. In the long term, this approach may help reduce the cost of preparing   diverse test data and support safer development of automated driving technologies.

Řešitel: Dalibor Javůrek    

OPEN-37-65                

Shape Optimization of Asymmetrizer for Integrated Magneto-Biplasmonic Circulator                   

Karolina CPU  Alloc=1900;  Karolina GPU  Alloc=800

Photonic chips use light instead of electrical signals to process data at extreme speeds with low energy consumption. One critical component is still missing: a micrometer-sized circulator — a device that steers light between ports in a fixed rotational order, like a roundabout directs traffic. Without it, reflected signals destabilize on-chip lasers and prevent integration of multiple key functions. This gap has persisted for over 30 years of worldwide research with no chip-integrated solution achieved.    The European CIRCULIGHT project (EIC Pathfinder Open, Horizon Europe, 9 partners, 6 countries) aims to create the world's first on-chip circulator using a novel principle — magneto-biplasmonics. Our team at VŠB–TUO optimizes the device structure via large-scale simulations on IT4Innovations supercomputers. The key bottleneck is the asymmetrizer — the section where the optical field reshapes so that light is routed to the correct output. It is currently too long, causing unacceptable loss in metallic parts. Our goal is to shrink it by over 75%, bringing losses toward the industry threshold of 0.5 dB.    Timing is urgent — international roadmaps rank on-chip circulators among the top missing building blocks. Success would unlock mass-producible integrated circulators, giving European foundries a decisive edge in telecom, data centers, quantum technologies, healthcare, and AI-driven optical computing.

Řešitel: Vijay Miriyala         

OPEN-37-66                

Computational Investigation of Atropisomeric Systems and Reaction Mechanisms

Barbora NG  Alloc=3400;  Karolina CPU  Alloc=12600;  Karolina GPU  Alloc=300        

This project focuses on the computational investigation of atropisomeric systems, their conformational dynamics, and catalytic reaction mechanisms relevant to modern synthetic chemistry. Atropisomerism, arising from restricted bond rotation, leads to stereoisomers with distinct chemical and biological properties, making it particularly important in medicinal chemistry and asymmetric catalysis. Using advanced quantum chemical methods, including density funtional theory (DFT) and correlated wavefunction approaches, we will analyze rotational barriers, stereochemical stbility, and key electronic and non-covalent interactions governing these systerms. In parallel, the project will explore palladium and nickel catalyzed reaction mechanisms through detailed free energy profiling and transition state characterization. By correlating computational results with experimental observations, the study aims to uncover the origins of reactivity and selectivity. The outcomes will provide fundamental mechanistic insights and predictive guidelines to support experimental research and enable the rational design of molecules and catalysts.

Řešitel: Jakub Šebera          

OPEN-37-67                

The molecular docking and molecular dynamics simulations of methyltransferases; the way to the discover of new inhibitors.

Karolina CPU  Alloc=28300;  Karolina GPU  Alloc=2900                

We will study several methyltransferases (METTL) to guide large scale in silico discovery of small molecule inhibitors targeting this enzyme family. We will perform high throughput virtual molecular docking (HTVS) of several millions commercially accessible compounds (Enamine database). Docking will be followed by molecular dynamics simulations on top ranked complexes to validate binding modes, assess pocket plasticity and quantify binding stability in atomistic detail. The project will systematically compare active sites across methyltransferases to identify opportunities for pan family versus selective inhibition, with particular emphasis on enzymes implicated in cancer relevant pathways such as translational control and epitranscriptomic regulation. Integration of docking scores, MD derived descriptors and basic cheminformatics will enable prioritization of chemically diverse lead candidates for experimental testing, and generate structure–function hypotheses about how small molecules modulate methyltransferase activity.

Řešitel: Miroslav Kolos      

OPEN-37-68                

Reconstruction-Induced Reactivity in Low-Dimensional Nanostructures

Karolina CPU  Alloc=14300;  Karolina GPU  Alloc=2900                

Understanding how materials activate chemical reactions is essential for the development of new technologies in energy and environmental protection. In this project, we study a new mechanism of chemical reactivity in nanoscale materials, where small structural changes can dramatically enhance catalytic activity.  Our recent results have shown that in nanoscale systems, atoms can rearrange into new configurations that create electronic states capable of transferring charge to reacting molecules. This can significantly lower reaction barriers and enable the breakdown of otherwise stable compounds, such as chlorinated pollutants.  Using advanced quantum-mechanical simulations, we will investigate how structural distortions in nanostructures influence electronic properties and chemical reactions. The project will focus on representative processes such as bond breaking in chlorinated molecules, water dissociation, and carbon dioxide activation.  The results will contribute to a deeper understanding of how nanoscale materials function and may guide the future design of efficient catalysts for environmental and energy applications.

Řešitel: Libor Dostál            

OPEN-37-69                

Activation of small molecules by P,C,P-pincer p-block element compounds – A theoretical insight.           

Barbora NG  Alloc=7800;  Karolina CPU  Alloc=38700                    

The replacement of transition metals in the activation of small molecules and catalysis represents one of the main targets for modern main-group organometallic chemistry. In this respect, various types of pincer complexes have gained considerable attention. Their coordination pocket enables stabilization of complexes bearing the central atom in unusual oxidation states or unique geometries, which is necessary to reach the intended unprecedented and value-added reactivity. This realm has been for a long time governed by ligands armed with N-donors, where previous works on bismuth are glancing examples. We have recently introduced P,C,P pincers as promising platforms for p-block elements also with the help of our previous IT4Innovations project (OPEN-34-62). Now, we are able to synthesize various low-valent compounds of Group 14 and 15 and reactive hydrides from Group 13, all systems are particularly suitable for the activation of various substrates. This project is intended to support our synthetic effort by theoretical studies, which is nowadays integral part of such type of research. Three major topics will be touched: i) Support for a reasonable development of properly substituted pincer compounds by pre-synthetic screening using theoretical approach. ii) Explaining interaction and activation mechanism for target substrates using promising pincer compounds (gained from the point i)). iii) Detailed description of catalytic cycles for the most successful systems.

Řešitel: Jan Heyda                

OPEN-37-7                    

Salt-specific phase behavior, condensation, and membrane interactions of intrinsically disordered elastomeric polypeptides

Barbora NG   Alloc=2600;  Karolina CPU  Alloc=2400;  Karolina GPU  Alloc=1100;  LUMI-G  Alloc=3400     

Intrinsically disordered proteins and protein-inspired polypeptides can reversibly change their structure, solubility, and phase behavior depending on temperature, salt composition, and nearby biological interfaces. These processes are important both for biology, where related proteins participate in biomolecular condensates, and for materials science, where elastomeric polypeptides are promising building blocks of responsive biomaterials. In this project, we will use large-scale molecular dynamics simulations to uncover how different salts, sequence mutations, and lipid membranes control the behavior of resilin-like polypeptides, elastin-like polypeptides, and the biologically important FUS protein. We will study isolated chains, protein-rich condensed phases, membrane-contact systems, and selected machine-learned interatomic potential simulations. The simulations will reveal which interactions stabilize expanded or compact states, which salts promote or suppress condensation, and how membrane composition modulates adsorption and phase behavior. The results will provide molecular-level design rules connecting sequence, environment, and responsiveness of disordered proteins, with direct relevance for biomolecular condensates and smart peptide-based materials.

Řešitel: Dominik Legut      

OPEN-37-70                 

Electron-phonon coupling in transition metal nitrides for energy conversion                

Barbora NG  Alloc=7400;  Karolina CPU  Alloc=12200;  Karolina FAT  Alloc=100;  Karolina GPU  Alloc=1200;  LUMI-C  Alloc=19400;  VLQ  Alloc=200000              

Modern technologies waste a large fraction of energy as heat. New-generation materials could efficiently convert this waste heat into reusable energy. Transition metal nitrides are a class of materials with promising properties for this purpose; however, further research is needed to improve their energy conversion performance and make their large-scale use economically feasible. By studying how electrons and atomic vibrations interact at the microscopic level, we aim to devise new routes for enhancing energy conversion efficiency beyond current limits. In particular, we investigate both thermoelectric and emerging phonoelectric effects, which could enable energy harvesting even at very small scales where traditional approaches fail. The results will support the design of next-generation materials for more efficient and sustainable energy technologies.

Řešitel: Yevgen Yurenko    

OPEN-37-71                 

GPU-accelerated molecular dynamics of protonation effects in carbonic anhydrase II                 

Karolina CPU   Alloc=300;  Karolina GPU  Alloc=2700                     

Accurate prediction of protein–ligand binding remains a key challenge in computational chemistry. In many systems, binding strongly depends on protonation states, which are often fixed in standard simulation protocols.   This project focuses on carbonic anhydrase II (CAII), a well-characterized enzyme where ligand binding is directly linked to protonation and metal coordination. Using GPU-accelerated molecular dynamics simulations, we will investigate how different protonation states influence structural stability and electrostatic properties of protein–ligand complexes.  Multiple protonation variants will be simulated for a series of CAII inhibitors. The simulations will be analyzed in terms of structural stability, hydrogen bonding, and electric fields in the binding site. A limited number of alchemical calculations will be performed for selected cases to estimate energetic differences.  The project aims to clarify when standard modelling assumptions become unreliable and to improve practical simulation strategies in drug discovery. 

Řešitel: Miroslav Rubes     

OPEN-37-72                 

GPU-Accelerated Modeling of Cu-Based Catalysts: Extension of Quantum-Chemical Simulations             

Karolina GPU  Alloc=1300;  LUMI-G  Alloc=7500        

This project focuses on computational modeling of catalytic materials based on copper clusters and small nanoparticles, which are widely used in hydrogenation and dehydrogenation reactions. Understanding how the atomic structure of these catalysts affects their performance is essential for improving their efficiency and stability. Building on previous computational work, this project introduces modern machine learning (AI/ML) approaches to accelerate simulations. We will test advanced AI/ML models, such as MACE and UMA, on GPU-based infrastructure to evaluate how efficiently they can reproduce accurate quantum-chemical results. The project is designed as a testing platform for GPU-accelerated materials modeling, comparing traditional simulation methods with AI/ML-based approaches. The goal is to determine how much these new methods can speed up research while maintaining accuracy. The results will support faster catalyst design and contribute to more sustainable chemical technologies.

Řešitel: Jiri Tomcala             

OPEN-37-73                 

Gaussian Boson Sampling for Earth Observation (GBS4EO) on VLQ           

VLQ  Alloc=360000  

The purpose of this project is to test the developed GBS4EO libraries on a quantum computer. In the future, these libraries, running on HPC and quantum computer will be integrated into the overall hybrid algorithm of GBS4EO project.  The primary objective of GBS4EO project is to build end-to-end solutions that benefit from the Gaussian Boson Sampling (GBS) for satellite data analysis. GBS outputs cannot be simulated by classical computers in polynomial time, but recent experiments with programmable optical circuits have demonstrated GBS's advantages. Consequently, GBS presents a validated method for surpassing classic capabilities, with tangible implementations. It is distinct amidst ongoing efforts to realize practical quantum computers. While GBS does not provide a foundation for universal QC, it excels in specific tasks (e.g., graph problems), which we build upon in GBS4EO. 

Řešitel: Michal OTYEPKA  

OPEN-37-74                 

ADAMS4SIMS              

Karolina CPU  Alloc=500             

EXA4MIND (HE project N° 101092944) is developing a new European platform for handling extremely large scientific datasets across the full digital continuum, from data generation to analysis on supercomputers, and within this effort ADAMS4SIMS focuses on molecular simulations by creating tools to store, standardize, and analyse massive amounts of simulation and experimental data. The platform helps researchers compare simulation results with experiments and improve molecular force fields in a faster and more systematic way, which is especially important for RNA and other biomolecules, where accurate simulations can support advances in biology, biotechnology, and medicine. By combining high-performance computing, automated workflows, and FAIR data management, the project makes complex analyses more efficient and reproducible, helping turn large-scale simulation data into practical scientific knowledge with broad value for European research and innovation.

Řešitel: Pavel Hobza           

OPEN-37-75                 

Solvent Control of Stability, Charge Transfer, and Spin States in Donor–Acceptor Complexes  

Barbora NG  Alloc=9200;  Karolina CPU  Alloc=20400;  Karolina FAT  Alloc=100;  Karolina GPU  Alloc=1800;  LUMI-C  Alloc=30600;  LUMI-G  Alloc=15600;  VLQ  Alloc=362200

This project aims to elucidate how solvent polarity influences the stability of covalent, dative, and non-covalent complexes using advanced quantum-chemical calculations. The computational study will be performed in close connection with ongoing experimental investigations employing state-of-the-art techniques, allowing for direct validation of theoretical predictions. The results will deepen our understanding of solvent-controlled complex stability and provide knowledge relevant to a broad range of chemical and materials systems. To achieve this, we will employ selected DFT functionals to evaluate the electronic and optical properties of various complexes in both implicit and explicit solvent models, an approach that requires substantial computational resources and high-performance computing infrastructure.

Řešitel: Theodorus Petrus Cornelis Klaver          

OPEN-37-76                 

Tight Binding calculations of bcc Fe           

Barbora NG  Alloc=2100;  Karolina CPU  Alloc=1900;  Karolina FAT  Alloc=100;  Karolina GPU  Alloc=50   

Iron is the most widely used engineering metal, forming the main constituent of tens of thousands of steel alloys that are used in the vast majority of all technical applications. While iron making is an old craft, it continues to be an active area of research, with improvements achieved through both experimental and computational research. In the latter category, a type of quantum mechanical calculations, known as Density Functional Theory, provides accurate results, but at a high computational cost. Another type of calculations, known as Tight Binding, delivers only slightly less accurate results but at hundreds of times lower computational cost. Tight Binding calculations can therefore handle systems with many more atoms, enough to include crystal defects called self-interstitial clusters and dislocations. The latter govern how ductile or brittle steel alloys are. The output of Tight Binding calculations enables yet another type of calculation method, known as machine learning interatomic potentials, that are again computationally much cheaper and allow further scaling up of the system sizes in computational research on iron alloys. Such research includes looking at how to understand and limit radiation damage in steels used in generating nuclear energy.

Řešitel: Przemyslaw Karol Grenda    

OPEN-37-77                 

High accuracy machine learning force fields for pharmaceutical research                    

Karolina CPU  Alloc=3300;  Karolina GPU  Alloc=700;  LUMI-C  Alloc=75;  LUMI-G  Alloc=100;  VLQ Alloc=35700       

Most modern medicines are processed in the human body by Cytochrome P450 family enzymes. Due to the high impact of these enzymes on drug metabolism it is necessary to take them into consideration during the process of drug development. However, simulating these enzymes is difficult due to the cofactor – additional molecule that enables the protein’s function – which contains iron, making it difficult to simulate using fast and standard methods. Our project aims to bridge the gap between fast simulations and high accuracy by applying Machine Learning Interaction Potentials – AI models trained to mimic high accuracy quantum mechanical simulations at a small fraction of original calculations’ cost. Finding an accurate model that would enable fast and accurate estimation of drug interactions with the enzyme will decrease both time and financial cost of developing new drugs. 

Řešitel: Saurabh Kumar Pandey          

OPEN-37-78                 

Electric-Field Effects on Post-Translationally Modified Tubulin: All-Atom Molecular Dynamics of PTM-Specific Response         

LUMI-G  Alloc=6200

Microtubules are protein filaments composed of α- and β-tubulin heterodimer units, that organize the interior of cells and are essential for cell division, intracellular transport and mechanical stability. Their behavior is often regulated by post-translational modifications (PTMs), chemical tags that form a molecular “tubulin code” and influence stability, dynamics and interaction with other proteins and intracellular trafficking. We propose to use large-scale molecular dynamics simulations on the LUMI-G supercomputer to determine how selected PTMs specifically polyglutamylation at C-termini tails of tubulin change the response of tubulin to external electric fields. The project will compare unmodified and several variants of post-translationally modified tubulin dimers under electric fields of different intensities and will quantify changes in structure, flexibility, charge distribution, hydration and dipole response. The expected outcome is a mechanistic picture of how chemical modifications and electric perturbations jointly regulate the basic building blocks of microtubules. This knowledge will advance fundamental biophysics and support future biomedical and nanobiotechnology research that aims to influence protein function in a controlled and non-invasive way.

Řešitel: Alfred Haavaan Mishi              

OPEN-37-79                 

Modeling of Betatron Radiation in Nanoparticle-Assisted Laser Wakefield Acceleration              

Karolina CPU  Alloc=38000;  Karolina GPU  Alloc=2800                

Nanoparticle-assisted laser–plasma wakefield acceleration (LWFA) offers a transformative pathway toward compact, high-brightness X-ray sources by enabling deterministic control over electron injection. In contrast to conventional self-injection, which is inherently stochastic, nanoparticles generate localized electrostatic fields that trigger controlled electron injection with precise timing and phase-space properties. Building on results obtained using IT4Innovations resources under project FTA-25-77, we have demonstrated controlled nanoparticle-triggered injection and enhanced beam charge. This project will perform full three-dimensional (3D) Particle-In-Cell simulations using Smilei to investigate how nanoparticle parameters influence both electron beam properties and the resulting X-ray emission. A central objective is to compute betatron radiation using Smilei’s inbuilt radiation diagnostics in full 3D geometry, enabling, for the first time, a comprehensive study of radiation from nanoparticle-assisted LWFA.

Řešitel: Michal Kolar           

OPEN-37-8                   

Hydration shells of ribosomes across domains of life                    

LUMI-G  Alloc=16200                     

From bacteria in a kitchen sponge to the neurons in the human brain, all cells produce proteins using ribosomes. Ribosomes are large biomolecular complexes composed of several strands of RNA and a few dozen proteins. Their structure is permeated by numerous channels, the largest of which is the polypeptide exit tunnel. In fact, the ribosome resembles a microscopic sponge: roughly one third of its mass is water. Although the main principles of protein synthesis are the same across all known life, ribosomes differ notably. A so-far underappreciated feature is their water content. This project focuses on the water inside the ribosome, which behaves differently from bulk water in a test tube. Using unique atomistic models, we will simulate ribosome-water dynamics at the microsecond timescale.

Řešitel: Jakub Podgorny    

OPEN-37-80                

Filling the gaps in our knowledge of black holes II.

Barbora NG  Alloc=9000;  Karolina CPU  Alloc=39100;  LUMI-C  Alloc=8000                  

Black holes (BH) are undoubtedly part of contemporary physics. An accretion disc of orbiting matter around BHs is often formed, which is illuminated by a hot X-ray-emitting corona. The location of the corona, the structure of the disc and its detailed high-energy micro-physics are for decades one of the main unsolved mysteries in astrophysics. With this project we aim to follow-up on a previous project at IT4Innovations in order to calculate the interaction of light and matter in the accretion disc's atmosphere, including known physical effects that are largely neglected in the solutions currently existing in literature. Using state-of-the-art Monte Carlo code STOKES, we aim to provide unique numerical estimates of polarization of X-rays reflected from and transmitted through the disc's stratified atmosphere and its warm corona, constructing a table model to be compared with the latest observations of BHs by the IXPE mission.

Řešitel: Oyvind Christiansen                 

OPEN-37-81                

BeStMIC: Beyond Standard Model Interactions in Cosmology 

Barbora NG  Alloc=8500;  Karolina CPU  Alloc=10800;  Karolina GPU  Alloc=400        

The dark sector of cosmology remains poorly understood and may exhibit behaviour that could explain observed discrepancies within the standard cosmological model, including tensions in the present-day expansion rate of the Universe, large-scale inhomogeneities, and the expansion history. There are also possible indications of structure in the dark sector on astrophysical scales, for example in the radial acceleration relation. Assessing the significance of these empirical tensions requires deriving predictions from alternative models that can be tested against observations. In this project, we perform cosmological simulations using a newly developed code based on the relativistic particle-mesh code gevolution. The code incorporates generalized dark matter with momentum-dependent interactions and interactions with the neutrino sector. In particular, we have implemented simulations of aether scalar–tensor dark matter and fuzzy, symmetron-like dark matter. We use these simulations to derive predictions for relativistic observables, such as weak-lensing foregrounds of the cosmic microwave background. Statistics of nonlinear structures, including baryonic and dark matter halos, can be compared with galaxy surveys. The completion of this project may rule out large regions of the model parameter space, improve our understanding of the required dark sector physics, and identify distinctive observational signatures that can be targeted in future surveys.

Řešitel: Šimon Vrba              

OPEN-37-82                

Modelling of SOL in the EAST tokamak with and without RMPs

Barbora NG  Alloc=9300;  Karolina CPU  Alloc=36500                    

One of the key challenges in magnetic confinement fusion is managing extreme power loads to the plasma-facing components and the related erosion of these elements. One of the most critical threads are the Edge-Localized Modes (ELMs) in tokamaks. They cause sudden bursts of hot plasma to escape the confined region into the scrape-off layer (SOL), where they propagate and then strike the divertor plates. This can lead to surface cracking, melting, material erosion and subsequent contamination of the core plasma terminating the discharge. In future large devices like ITER and DEMO, these loads scale unfavorably with machine size, threatening component lifetime and discharge stability. One of the mitigation techniques are the so-called resonant magnetic perturbations (RMPs). Making the edge magnetic field lines chaotic causes a spread of the heat loads over larger area and a decrease in the peak power. RMPs have been used in tokamaks such as EAST, DIII-D, JET, KSTAR and are a baseline tool for ITER. This project will perform the first fully kinetic BIT1 simulation of the EAST tokamak regular and RMP-affected SOL, directly benchmarking our model against experimental measurements from EAST. Validating this advanced kinetic approach is essential for subsequent predictive simulations of the ITER baseline scenarios.

Řešitel: Karel Sindelka       

OPEN-37-83                

Pectin gelation: Computer simulations of polysaccharide-based annealed polyelectrolytes    

Barbora NG  Alloc=6900;  Karolina CPU Alloc=18900                     

Pectin, a natural polysaccharide, plays an essential role in food processing as a gelling agent and is increasingly used in biomedical applications, such as drug delivery. This project investigates pectin gelation mechanisms via computer simulations, addressing a significant gap in understanding the behaviour of pectin-based materials. Pectin gelation reles on electrostatic crosslinking with calcium ions, resulting in stable “egg-box” structures. We therefore use mesoscopic simulations with explicit electrostatics to study interactions between many pectin chains of realistic lengths.

Řešitel: Ievgeniia Korniienko                 

OPEN-37-84                

Magnetic contribution to acoustic wave propagations in antiferromagnets                   

Barbora NG  Alloc=5800;  Karolina CPU  Alloc=7900;  Karolina FAT  Alloc=150;  Karolina GPU  Alloc=1200;  LUMI-C  Alloc=13800;  LUMI-G  Alloc=6100               

The coupling between spin waves (magnons) and acoustic waves (phonons) is a crucial phenomenon with significant implications for spintronics, magnonics, and quantum materials research. The development of computing technology and approaches based on numerical methods has opened up the possibility of both a more qualitative, accurate study of those phenomena and overcoming the limitations associated with simplifications in analytical approaches based on the linear theory of magnetoelasticity. In our study we will develop spin-lattice models for selected cubic antiferromagnets and use them for investigation of coupled magnon-phonon effects, in particular, for specifying the magnetic contribution to acoustic wave propagation in those crystals.

Řešitel: Anton Firc                

OPEN-37-85                

Hybrid quantum-classical methods for robust audio deepfake detection

Karolina GPU  Alloc=500;  LUMI-G  Alloc=1800;  VLQ  Alloc=36000              

AI-generated speech is becoming harder to distinguish from human speech, posing risks to media trust, digital communication, and security. This project will develop new hybrid quantum-classical methods for detecting audio deepfakes. Our preliminary results show that small quantum components can improve the reliability and generalization of modern detection systems, while our optimized implementation enables large-scale simulation and testing to run much faster. Building on this, we will design improved quantum modules, explore more effective ways to represent speech for hybrid models, and optimize execution across HPC and quantum platforms. The expected outcome is a new class of efficient and trustworthy detection methods that can better handle changing real-world conditions and help strengthen practical audio-forensic tools.

Řešitel: Jan Kuneš                 

OPEN-37-86                

Optical Kerr effect in altermagnets              

Karolina CPU  Alloc=22600        

Altermagnetism is a form of magnetic order that combines features of ferromagnets and antiferromagnets, enabling spin-polarized electronic states without net magnetization. A prototypical material is manganese telluride (MnTe), whose symmetry leads to a momentum-dependent spin splitting with promising implications for spintronics. X-ray circular dichroism has proven its utility as a tool to investigate altermagnetic order, however, large facilities (synchrotron) are required. Its optical analogy, the magneto-optical Kerr effect, which can be run on table top devices is less investigated and understood.  We will combine the density functional theory with the dynamical mean-field theory to capture both the realistic band structure and the effects of electronic correlations in MnTe. In particular, this approach allows us to study the evolution of optical response in the paramagnetic as well as altermagnetic phases.  The results will establish theoretical interpretation of experiments run by our collaborators.

Řešitel: Martin Hurta           

OPEN-37-87                

Synergy of Evolutionary Algorithms and Advanced Machine Learning Algorithms for Digital Circuit Design                    

Barbora NG  Alloc=700;  Karolina CPU  Alloc=800;  Karolina GPU  Alloc=300                

Designing digital circuits is a very complex engineering task. Arranging individual components to achieve optimal performance, efficiency, and reliability demands significant expertise, time, and computational resources. Two modern automated approaches each carry notable limitations. Evolutionary Algorithms (EAs) draw inspiration from natural evolution by iteratively generating increasingly functional candidate circuits. However, EA heavily relies on trial-and-error search, making it inefficient. Machine Learning (ML) algorithms excel at pattern recognition, yet they lack sufficient domain-specific training data to perform reliably in hardware design contexts. This project proposes a hybrid framework that combines the strengths of both methods. ML models guide the evolutionary search process, directing it toward promising design regions and eliminating unproductive exploration. Simultaneously, the evolutionary component continues to investigate unconventional solutions that ML alone would not be able to find. The framework will be validated across several real-world scenarios, including the design of energy-efficient accelerators, automated design verification, and medical signal classifiers. The overarching objective is to significantly accelerate and reduce the cost of chip design, supporting both current CMOS (Complementary Metal–Oxide–Semiconductor) and future post-CMOS technologies.

Řešitel: Anton Bushuiev    

OPEN-37-88                

Expanding our understanding of viral proteins through protein language model customization

Karolina GPU  Alloc=2400          

Viral proteins are critical targets for vaccine development, antiviral drug design, and understanding infection mechanisms. However, predicting the structures of viral proteins remains challenging due to their high mutation rates and limited availability of homologous sequences. This project aims to substantially expand the Big Fantastic Virus Database (BFVD), a comprehensive repository of viral protein structures, by applying our recently developed ProteinTTT method enhanced with multiple sequence alignment (MSA) information. ProteinTTT customizes protein language models to individual target proteins at test time, substantially improving structure predictions for challenging, out-of-distribution proteins. By combining ProteinTTT with MSA-based customization and ESMFold, we will re-predict structures for the hundreds of thousands of viral proteins in BFVD where current AlphaFold and ESMFold predictions remain low-confidence. Our preliminary results show that ESMFold + ProteinTTT already improves 19% of BFVD entries.

Řešitel: Vojtech Horny       

OPEN-37-89                

Particle beam acceleration and gamma-ray generation with multi-PW laser pulses: particle-in-cell simulations      

Karolina CPU  Alloc=18600;  LUMI-C  Alloc=7000;  LUMI-G  Alloc=4400    

Ultra-intense laser pulses can generate beams of charged particles, including protons, electrons, and exotic particles such as positrons and fast neutrons [2]. These beams have promising applications ranging from cancer treatment [1] to nuclear waste transmutation and medical isotope production. The commissioning of multi-petawatt laser systems such as ELI Beamlines (Czechia), ELI NP (Romania), and Apollon (France) significantly expands the capabilities of laser-driven particle acceleration.  In this project, we will use large-scale simulations to study particle acceleration and gamma-ray generation in interactions of multi-PW laser pulses with matter. Using high-performance computing and methods developed in our previous allocation, we will identify conditions that maximize particle and radiation output and support the design and interpretation of experiments at these facilities. Ultimately, this work contributes to the development of compact radiation and particle sources with applications in medicine, industry, and nuclear science.

Řešitel: Vasileios Psomas OPEN-37-9                   

Retrieval-Augmented Vision-Language Models for Open-Vocabulary Segmentation v2 (RAVLOSv2)            

LUMI-G  Alloc=15600                     

How do self-driving cars safely navigate busy streets? How do satellites monitor environmental changes or assist in disaster relief? At the heart of these innovations lies visual scene understanding — the ability of technology to interpret complex images and make sense of the world. However, today's systems often fail when encountering unfamiliar objects or scenarios, limiting their potential in dynamic, real-world environments.    RAVLOSv2 (Retrieval-Augmented Vision-Language Models for Open Vocabulary Segmentation v2) aims to address this challenge. Inspired by the way humans recall memories to better understand new situations, RAVLOSv2 combines advanced artificial intelligence models with a powerful memory system that stores examples of objects and scenes. By using this memory, the system can adapt on the fly, recognising not only generic objects like \cat\" or \"vehicle\" but also highly specific ones like \"Egyptian cat\" or \"electric scooter\". This breakthrough enables more accurate and detailed visual understanding, whether in autonomous vehicles, medical diagnostics, or environmental monitoring.    The impact of RAVLOSv2 could transform industries and improve lives — from safer autonomous vehicles and smarter assistive technologies for people with disabilities to better tools for protecting our planet. By teaching machines to see and understand the world as we do, RAVLOSv2 builds on the internationally recognised success of RAVLOSv1, bringing us closer to a future where technology seamlessly enhances our daily lives.

Řešitel: Michael Komm     

OPEN-37-90                

3D PIC simulations of tungsten scarification               

Karolina CPU  Alloc=3900          

ITER tokamak, currently under construction in Cadarache (France), is aimed to demonstrate the feasibility of power generation by fusion of deuterium and tritium, targeting Q=10. Recently, the ITER Organisation decided to change the material of the first wall from beryllium to tungsten. This brought to the attention of the fusion community the problem of tungsten sputtering from the first wall and proliferation of tungsten impurity into the confined plasma. Tungsten can radiate at high plasma temperatures and as such compromise the goals of ITER. Our study is aimed on particle-in-cell simulations of specially designed tungsten surfaces, which should feature significantly lower effective sputtering yield than a flat surface.

Řešitel: Jiří Pittner                 

OPEN-37-91                

Excited state molecular dynamics of radicals with non-adiabatic and spin--orbit effects            

Karolina CPU  Alloc=18900        

Photochemistry is the study of chemical reactions initiated or influenced by light, and it plays a critical role in both natural processes and technological advancements. Its importance spans multiple domains, making it a cornerstone of science and innovation. Molecular dynamics in excited states including the non-adiabatic and spin-orbit effects is an important theoretical tool for the simulation of the photochemical processes. Many of the compounds employed in current applications contain expensive rare elements like e.g. ruthenium, iridium, or germanium. Finding their replacement involving more abundant elements is thus an important task. In the last decade, a novel class of photoactive octahedral complexes of chromium(III) called “molecular rubies'' with ligands forming six-membered chelate rings and inducing very strong ligand field have been discovered and studied. Thanks to their photochemical properties and high abundance of chromium, these complexes have many potential applications. However, their theoretical study is complicated by their radical nature with odd number of electrons leading to doublet and quartet states. We aim to develop methods suitable for the excited state MD simulations of radicals and apply it to these species.