35^th open access competition RESULTS

We would like to express our sincere thanks to all applicants for their active participation in the 35th round of the IT4Innovations Open Access Grant Competition.

THE ALLOCATION COMMISSION DECIDED ON THE ALLOCATIONS WITHIN THE 34^th OPEN ACCESS GRANT COMPETITION AS FOLLOWS:

Researcher: Pavel Hobza       

OPEN-35-1     

Covalent Dative Bonding, Ionic, H-Bonding, and Charge Transfer Complexes: Surprising Stability/Instability Trends with Increasing Solvent Polarity       

Barbora CPU  Alloc=40000;  Barbora FAT  Alloc=310;  Barbora NG  Alloc=15900;  Karolina CPU  Alloc=85700;  Karolina FAT Alloc=110;  Karolina GPU  Alloc=7900;  LUMI-C  Alloc=8400;  LUMI-G  Alloc=19000

This project seeks to elucidate how solvent polarity modulates the thermodynamic stability of covalent dative and other non covalent assemblies through high level quantum chemical calculations. Theoretical results will be integrated with cutting edge spectroscopic and structural data obtained by experimental collaborators, yielding a cohesive, multidimensional view of solvent mediated effects. State of the art density functional theory—augmented, where required, by higher level wave function benchmarks—will be employed to probe electronic structures and photophysical responses in both continuum (implicit) and explicit solvation regimes. The insights derived will refine our fundamental understanding of solvent stabilization mechanisms and inform applications across diverse areas of chemistry, materials science, and molecular engineering.

Researcher: Karel Carva       

OPEN-35-10   

Quasiparticles in layered van der Waals magnets   

Karolina CPU  Alloc=36800;  Karolina FAT  Alloc=110;  LUMI-C  Alloc=4060

Quasiparticles of spin excitation – magnons - in van der Waals (vdW) materials represent a perfect bosonic reservoir for coherent information processing in topological magnonics. Genuinely weak vdW interlayer interactions make accessible the low-dimensional dynamics of interacting spins. Here, the behavior of magnons is dramatically affected by the spin-orbit interaction. The project focuses on a precise understanding of this system, which should enable the demonstration of tunable coherent, strong coupling between two magnons induced by a magnetic field in a single vdW material, or magnon-phonon coupling. Secondly, we intend to probe magnon band topology by utilizing the thermal Hall effect, which corresponds to the magnon Berry curvature contribution to the transverse magnon current. Spin dispersion simulations based on ab initio calculations will be compared to experimental data in order to provide microscopic interpretations and quantization of magnon Berry curvature. We will focus our research on a class of Mott-insulating vdW halides with a honeycomb lattice with a potentially high orbital magnetic moment as are vanadium trihalides.

Researcher: Sergiu Arapan    

OPEN-35-11   

The effect of magnetic impurities on the antiferromagnetic exchange coupling in Co|Ru|Co multilayers          

Barbora CPU  Alloc=11600;  Barbora NG  Alloc=5600;  Karolina CPU  Alloc=6700;  Karolina GPU  Alloc=2200;  LUMI-C  Alloc=3500   

Interlayer exchange coupling in metallic multilayers between two ferromagnetic layers across a nonmagnetic spacer layer forms the basis of a new field of synthetic antiferromagnetic spintronics. The tunability of magnetic multilayers allows for optimization of properties that are desirable for applications including magnetic field sensing and magnetic random-access memory. The interlayer exchange coupling may be altered by the presence of the ferromagnetic elements in the nonmagnetic spacer. The value of this exchange coupling influences the magnetization dynamics and may decrease, for example, the switching time of the memory cells. Co|Ru|Co multilayers could be important for applications since it has one of the largest reported antiferromagnetic couplings. Here we investigate the role of magnetic elements Cr, Mn, Fe, Co, and Ni in controlling the interlayer exchange coupling by electronic structure calculations.

Researcher: Filip Šebesta      

OPEN-35-12   

QM/MM modelling of electron transfer in copper proteins converted to photoenzymes 

Karolina CPU  Alloc=7300;  Karolina GPU  Alloc=500           

The project is focused on  conversion of multicopper oxidases (MCO) to photoenzymes.  Organometallic photosensitizers will be covalently attached to MCO proteins at specific surface sites. Electron transfer (ET) reactions between the sensitizer and MCO active sites will be triggered by optical excitation and used to investigate the mechanisms of MCO-catalyzed reactions as well as to catalyze new processes, beyond the natural range of MCO enzymatic reactivity. Mechanistic studies will focus on roles tryptophan and tyrosine, testing our hypotheses that they protect enzyme molecules from reactive oxygen species and also actively participate in catalysis. These goals will be accomplished by time-resolved spectroscopic studies of photogenerated intermediates combined with theoretical calculations of electronic interactions between MCO active sites (Cu atoms, tryptophan and tyrosine residues) and simulations of electron-transfer steps using advanced methods of combined quantum and classical molecular dynamics (QM/MM/MD simulations). New mechanistic understanding will be used to accelerate MCO-catalyzed oxidations of organic substrates by sunlight, potentially making biofuel production more effective. We also propose employing photosensitized MCOs as photocatalysts of solar energy conversion. Ultimately, we aim at developing MCO-based photocatalysts for water oxidation by reversing the direction of the natural MCO catalysis with strongly oxidizing photosensitizers.

Researcher: Petr Touš

OPEN-35-13   

Towards the development of an ab initio composite method for predicting active pharmaceutical ingredient co-crystal systems in silico    

Karolina CPU  Alloc=17700    

Within the pharmaceutical industry, all care is applied to efficient, sufficiently precise research and development techniques. In recent years, pharmaceutical companies have started to adopt computational approaches within their product development workflow. These approaches promise to eliminate some of the tedious and expensive trial-and-error work commonly associated with drug development, not to mention reduce its cost and time. Finding a complementary molecule for an active pharmaceutical ingredient (API) to form a co-crystal with, or the so-called “co-formers”, certainly presents one such tedious task. It often involves experimentally investigating hundreds or more binary systems at once with no guarantees of success. There is a myriad of approaches which try to screen for these co-formers based on various criteria, with the output usually being a list of molecules sorted by the probability to form a crystal – the actual packing of the prospect crystal is often neglected. Moreover, pharmaceutically important properties such as solubility depend primarily on the crystal structure rather than the properties of the molecules forming the crystal: Knowledge of these properties for a given prospect system could be of enormous value to the industry. This project aims to combine approaches of co-former prediction, crystal structure prediction and computational thermodynamics in a single workflow to allow for basic materials research at an unprecedented level.

Researcher: Jaroslav Resler  

OPEN-35-14    Microscale urban simulations for project Geohazards         

Karolina CPU  Alloc=82400    

The PALM model system represents a modern tool that allows detailed simulations of conditions in urban areas. These simulations typically concern phenomena of urban heat island and thermal comfort, but still often air quality issues. ICS team significantly contributed to the model development and validation, moreover we are expertized in evaluation of urban climate adaptation measures. For the OP JAK project Geohazards (“Natural and anthropogenic hazards”, CZ.02.01.01/00/22_008/0004605), we prepared several large domains within Prague for complex simulations including thermal comfort and air quality. The simulations represent parts of the Prague metropolitan area in units of meters resolution and will simulate important effects presenting potential hazards for dwellers and test potential measures to forestall them.

Researcher: Artem Moroz     

OPEN-35-15   

AI-powered building management system using image analysis (Systém řízení budovy založený na analýze obrazu využívající umělou inteligenci)         

Karolina CPU  Alloc=1000;  Karolina GPU  Alloc=4500         

Current building management systems depend on a fragmented network of single-purpose sensors for motion, security, or fire detection. This approach lacks the intelligence to grasp the context of a situation, failing to provide a complete picture of events and leading to limited functionality, frequent false alarms, and the risk of overlooking dangerous incidents. Our research addresses a critical challenge in building automation by aiming to create truly intelligent environments that can interpret and react to complex human activities. We will develop a building management system that leverages advanced computer vision and artificial intelligence to analyze a live feed from vision sensors. To ensure speed and privacy, a key requirement is that the system must operate efficiently on a local edge device, processing all data on-site. This project focuses on developing a system that progresses from simple detection to genuine understanding, capable of recognizing specific individuals, identifying their activities, and detecting critical situations in real time to enable a suitable, automated response.

Researcher: Anahí Villalba Pradas    

OPEN-35-16   

Impact of non-CO₂ forcers on climate, weather and air quality using WRF-Chem 

Karolina CPU  Alloc=54000    

While the impact of CO₂ on the climate, weather, air quality and health is very well documented, the effect of the so called non-CO₂ forcers (greenhouse gases and other substances other than CO₂) is less well understood. The FOCI project (“Non-CO₂ Forcers and their Climate, Weather, Air Quality and Health Impacts”, https://www.project-foci.eu/wp/) aims to fill this void and to help policymakers to adopt appropriate mitigation and adaptation strategies. To do so, simulations are conducted using the WRF and WRF-Chem models with a domain at 27 km horizontal resolution covering Europe for the period 2045-2055. The length of the period ensures statistically robust results. Different emission scenarios are used to account for the effect under different climate conditions. We expect to show that the non-CO2 forcers  leads to an increase in temperature as well as to changes in other relevant meteorological values such as humidity and precipitation.

Researcher: Subhasmita Ray

OPEN-35-17   

Interplay of Lattice Dynamics and Magnetism in CrX₃:

Towards Functional 2D Spintronic Materials

Karolina CPU  Alloc=53900;  LUMI-C  Alloc=8400    

The intrinsic ferromagnetism of two-dimensional (2D) chromium trihalides (CrX₃, X = Cl, Br, I) and their potential uses in low-dimensional spintronic devices have garnered a lot of interest. In this work, we use first-principles density functional theory (DFT) coupled with phonon calculations to examine the interaction between lattice dynamics and magnetic properties in CrX₃ monolayers. We investigate the effects of certain vibrational modes on magnetic anisotropy and magnetic exchange interactions. We look at phonon-induced structural changes to find strategies to change magnetism without directly modeling dynamic spin-phonon interaction. This approach sheds light on the vibrational modulation of 2D magnetism and opens the door to phonon-engineered spintronic functionalities.

Researcher: Ján Minár          

OPEN-35-18   

QMSpec         

Barbora CPU Alloc=40700;  Karolina CPU  Alloc=88100;  Karolina GPU  Alloc=5000;  LUMI-C  Alloc=8400          

Understanding how electrons move and interact inside materials is key to designing tomorrow’s technologies, from ultrafast electronics to quantum devices. Spectroscopic techniques such as spin-resolved Angle-Resolved Photoemission Spectroscopy (SARPES), X-ray magnetic circular dichroism (XMCD), X-Ray absorption near edge spectroscopy (XANES), Spin polarized low-energy electron diffraction (SP-LEED), Auger electron spectroscopy (AES), and other transport and magnetic scattering calculations provide deep insights into electronic structure. Leveraging the IT4I supercomputer, we will be investigating different kinds of quantum materials and running high-throughput simulations across a wide range using the SPR-KKR code, a unified first-principles multiple scattering framework capable of simulating all of these spectroscopies in a single DFT-Green’s function formalism.. We are developing robust workflows, based on ASE2SPRKKR for structure conversion and SPARKFlow for scalable job management. This, along with other well-established ab-initio codes for ground state calculation like VASP, ABINIT, and ELK. Data obtained will be  made open access through repositories like NOMAD, Material Project[8], and Lexis platform, for an ultimate goal of contributing to a worldwide database to accelerate spectral prediction through machine learning models. This work is supported by the Czech project Quantum materials for applications in sustainable technologies (QM4ST).

Researcher: Pavlo Polishchuk           

OPEN-35-19   

Fragment-based de novo design and structure optimization          

Karolina CPU  Alloc=5300;  Karolina GPU  Alloc=600           

Exploration of chemical space is chemical space is extremely difficult due to a vast number of potential drug-like molecules – 1036. Approaches which can generate compounds on-the-fly and adaptively explore and navigate in this space will substantially increase search speed of potentially active molecules, improve their novelty and greatly expand the knowledge about favorable and unfavorable regions of chemical space. All this will increase efficacy of searching of new hits, leads and drug candidates. This project is focused on development and application of computational tools based on the previously developed fragment-based structure generation framework CReM which main advantage is generation of synthetically accessible molecules that will improve outcomes of medicinal chemistry projects.

Researcher: Oldřich Plchot   

OPEN-35-2     

Task-Adaptive Hybrid Pruning for Efficient Deployment of Large-Scale Speech Foundation Models        

LUMI-G  Alloc=19000 

Large-scale self-supervised learning (SSL) models like WavLM have achieved state-of-the-art performance in speech processing, but their significant size impedes deployment on resource-constrained devices like mobile phones or smart speakers. Enabling these powerful speech models to run efficiently on devices is a critical challenge for the wider adoption of speech technologies. To address this, we propose to develop a novel \Hybrid Pruning (HP)\" framework that integrates structured pruning directly into the task-specific fine-tuning process. In a single training stage, it jointly optimizes for both task performance and model sparsity, allowing the model to learn a compressed architecture specifically tailored for end tasks like speaker verification and anti-spoofing, while eliminating the need for complex, multi-stage pipelines. Our preliminary experiments show that this method can remove up to 70% of an SSL model's parameters with negligible performance degradation on large-scale benchmarks. Furthermore, this project will allow for a comprehensive, large-scale validation and analysis of this novel technique."

Researcher: Rostislav Langer

OPEN-35-20   

Sustainable Chemistry through Single-Atom Precision: CO₂ Reduction, Nitrate-to-Ammonia Conversion, and Coupling of Aromatic Amines

Barbora NG  Alloc=15900;  Karolina CPU  Alloc=73300       

The transition to sustainable chemical technologies depends on our ability to design catalysts that are both efficient and environmentally benign. Achieving this requires a detailed, atomistic understanding of reaction mechanisms under realistic conditions. In this project, we will employ advanced quantum mechanical simulations to study the reactivity, selectivity, and stability of single-atom catalysts (SACs) across three transformations of environmental and industrial relevance: (1) CO₂ electroreduction to C1 products on SACs, using grand canonical DFT to capture potential-dependent behavior and the balance between stability and activity; (2) nitrate-to-ammonia electroreduction on bioinspired, atomically ordered Cu/Co catalysts, focusing on cooperative site effects; and (3) reductive coupling of nitroarenes with alcohols catalyzed by Fe-based SACs under solvent-free conditions. These subprojects integrate DFT methods, grand canonical potential simulations, and mechanistic modeling. The insights will support experimental efforts and guide the rational design of future catalysts.

Researcher: Michal Langer   

OPEN-35-21   

Unraveling Photocatalysis and Electrocatalysis in Carbon Dots and Low-Dimensional Materials: A Theoretical Perspective    

Barbora NG  Alloc=15900;  Karolina CPU  Alloc=9800;  LUMI-C  Alloc=8400          

Carbon dots (CDs) are a rising star in the world of nanomaterials, showing tremendous promise in clean energy applications and environmental solutions. What sets them apart is their tunable optical and electronic properties, making them ideal for applications like sensing, bioimaging and light driven hydrogen evolution. Still, unlocking their full catalytic potential is no simple task—largely due to the complex interplay between their structure and function. That is where computational modelling steps in. Density functional theory (DFT) and related techniques are key to decoding the electronic landscape, charge flow, and active sites of CDs. Calculations focused on doping, precise size tuning, and surface chemical modification enable fine control over their catalytic behavior. At the same time, molecular dynamics and machine learning potentials are speeding up the search for the structural fingerprints behind high performance, while advanced quantum methods delve into the core mechanisms behind catalytic action. Together, these tools form a powerful theoretical framework, deepening our understanding of CDs and paving the way toward smarter, greener catalysts for a sustainable future.

Researcher: Michal Svatos    

OPEN-35-22    CERN   Barbora CPU  Alloc=11700;  Barbora NG  Alloc=3400;  Karolina CPU  Alloc=15100;  LUMI-C  Alloc=4000   

Experiments measuring particle collisions in Large Hadron Collider at CERN require huge computing capacity for data analysis and Monte Carlo simulations. In cooperation with other projects, they developed and use distributed environment called Worldwide LHC Computing Grid (WLCG). Basic needs for computational and storage resources for Run III which started in July 2022 are covered by WLCG pledge resources. However very significant fraction of resources is used from so called unpledged resources, which are provided by HPC centers, occasional cloud resources and resources from volunteers via BOINC. In 2021, Vega supercomputer (part of PRACE network) almost doubled available computing resources for the ATLAS experiment. The second most significant HPC contributor was Karolina supercomputer.  We already created a well working environment for job submission to IT4I supercomputers. Jobs are sent automatically by a production system and are able to keep busy available resources. Optimalisation of CPU usage was done using HyperQueue.

Researcher: Frantisek Prinz  

OPEN-35-23   

Turbulent Fiberous Particle Motion in Human Respiratory Tract using the Lattice Boltzmann Method    

Karolina CPU  Alloc=24500;  Karolina GPU  Alloc=800         

Computational Fluid Dynamics (CFD) is vital for understanding how particles move and deposit in the respiratory tract, though accurately tracking non-spherical particles experimentally remains difficult. Particle orientation significantly affects their trajectory and deposition. Existing CFD simulations struggle to precisely model this due to the turbulent, non-creeping flow conditions in the respiratory tract.  This project aims to enhance the accuracy of the Lattice Boltzmann Method (LBM) for simulating fibrous particle transport and deposition. We will use a modified Euler-Lagrange Euler-Rotation approach within a realistic female respiratory tract model. A new framework incorporating governing forces and torques for non-creeping flow will be implemented. Results will be validated against traditional CFD and experimental data, providing valuable insights for medical applications.

Researcher: Ivo Oprsal          

OPEN-35-24   

Follow-Up 3D Seismic Modeling and Risk Assessment for Nuclear Facility Infrastructure 

Barbora CPU  Alloc=500;  Karolina CPU  Alloc=1300

Earthquakes have major social and economic impacts, causing casualties and infrastructure damage worldwide. This includes urban areas and critical urban-service infrastructure, such as dams, power lines, pipelines, and, notably, nuclear waste disposal sites. This project is directly related to the proposed expansion of nuclear energy sites into the basin areas of the Czech Republic, highlighting the   significance of local seismic amplification as a key factor influencing seismic hazard. The research aims to enhance planned approaches by focusing on numerical modeling of seismic responses across a range of selected sites with varying geological structures. The results will integrate with geological structure information and previous study outcomes into a regional-scale model, creating optimal methodologies for assessing amplification and screening all external seismotectonic hazards at the site. Advanced computational resources (IT4I) will be utilized for detailed seismic modeling, enabling more precise predictions and analyses of potential seismic events. Numerical modeling using computer resources will play a crucial role in simulating different seismic scenarios and understanding the impact of various geological conditions on seismic amplification.  

Researcher: Klára Kalousová 

OPEN-35-25   

Transport processes through the deep ice layers of water-rich exoplanets

Barbora CPU  Alloc=4300      

Within the last two decades, the observations by the Kepler telescope and the TESS mission led to the discovery of more than five thousands of confirmed exoplanets orbiting stars beyond our Solar System. Some of these planets may have hydrosphere atop the rocky core. It was suggested that these planets, that can only form far from their star beyond the snow line, could migrate closer to their star to end in the habitable zone. Depending on their mass and the pressure-temperature conditions, they can be large rocky planets covered by thin global oceans or planets with thick hydrosphere, where the liquid oceans are sandwiched between the outer ice crust and a deep, high-pressure (HP) ice layer. Such planets (e.g., some of the Trappist planets) may represent an opportunity to explore new classes of potentially habitable planets.    The presence of the HP ice layer is often seen as a barrier for material exchange between the core and the ocean. In our previous work tailored to Solar System icy moons, we have demonstrated that such an exchange may be possible if melting occurs at the core-hydrosphere interface. Furthermore, organics may be present in the rocky cores, some of them being the building blocks of life. The interaction of organics with liquid water is essential for assessing the exoplanets habitability potential. The goal of this project is to study the transport of organic-rich liquids through the HP ice layer of water-rich exoplanets.

Researcher: Masoud Shahrokhikhorneh      

OPEN-35-26   

Metal Single Atoms Anchored on N- and P-Doped Triphenylene-Graphdiyne as High-Efficiency Electrocatalysts for Water Splitting and Oxygen Reduction   

Barbora CPU  Alloc=25800;  Karolina CPU  Alloc=19700      

Single-atom catalysts (SACs) with high efficiency and optimal utilization of metal atoms show great potential for application in the field of renewable energy. Support materials that can stabilize the active metal atoms and tune their catalytic performance are crucial for the effectiveness of SACs. In this project, dispersion-corrected density functional theory (DFT) calculations will be carried out to systematically investigate the stability, activity, and selectivity of a series of potentially efficient and stable single transition and post-transition metal atoms supported on nitrogen- and phosphorus-doped triphenylene-graphdiyne monolayer (N-TpG and P-TpG) for the oxygen evolution reaction (OER), hydrogen evolution reaction (HER), and oxygen reduction reaction (ORR). Hydrogen binding energy (ΔGH*) is employed as a primary descriptor for evaluating HER activity. In contrast, the binding energy of the OH intermediate (ΔGOH*) can act as a single descriptor for predicting the thermodynamic limiting potential of OER and ORR. Among all the studied transitional atoms, we will identify those that exhibit the lowest overpotential for HER, OER, and ORR.

Researcher: Ondrej Olsak     

OPEN-35-27   

Acceleration of Wave Propagation through Spectrum Pruning       

Barbora CPU  Alloc=1000;  Barbora GPU  Alloc=1000;  Barbora NG  Alloc=100;  Karolina CPU  Alloc=2300;  Karolina GPU  Alloc=1700     

Focused ultrasound enables precise, non-invasive treatment for brain disorders using low-intensity ultrasound. However, high-resolution ultrasound simulations required for planning such treatments may not only lead to significant computational resource consumption but also cause delays in time-critical medical procedures. This project advances our research on accelerating spectral method-based ultrasound wave propagation simulations by utilising Fourier basis functions, replacing standard Fast Fourier Transform computations with pruned Fast Fourier Transform algorithms. This optimisation reduces wave propagation simulation times, directly accelerating the treatment planning phase of medical procedures. The results of this project will enable the acceleration of clinical workflows used for planning medical treatment utilising focused ultrasound, especially in cases involving high-resolution ultrasound simulations. Computational resources will be used for developing and evaluating wave propagation simulations involving pruned Fast Fourier Transform and their integration into clinical workflows.

Researcher: Peter Huszár     

OPEN-35-28   

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

Karolina CPU  Alloc=30200    

This project supports the Horizon Europe FOCI initiative, focusing on the climate and air quality impacts of non-CO₂ radiative forcers such as aerosols, methane, and tropospheric ozone. Despite well-established knowledge of CO₂, uncertainties remain around non-CO₂ forcers due to complex atmospheric interactions and regional variability. To address this, the project employs high-resolution regional climate simulations using the RegCM5-Chem model, which integrates atmospheric chemistry and their feedback into the climate system. Simulations will be conducted for two future climate scenarios for the middle of the century, driven by the general climate model EC-Earth3-AerChem. The outcomes aim to improve understanding of non-CO₂ impacts on climate, weather, air quality, and public health, contributing to more effective mitigation strategies at a regional level.

Researcher: Petr Strakos       

OPEN-35-29   

Synthetic Environments for Next-gen Safety and Efficiency in Cities          

Barbora NG  Alloc=2300;  Karolina CPU  Alloc=2200;  Karolina GPU  Alloc=3400;  LUMI-C  Alloc=2300;  LUMI-G  Alloc=2400     

This project aims to advance the frontiers of AI-driven video intelligence by integrating local data processing with cutting-edge machine learning algorithms to enable real-time, energy-efficient, and privacy-preserving solutions. The project targets two key application domains: smart buildings and traffic safety. Core technologies will include object detection, human activity recognition, and anomaly detection, all designed to increase situational awareness and safety in dynamic environments. A central innovation lies in the use of synthetically generated visual data from virtual 3D scenes, enabling the training and evaluation of algorithms in complex or rare scenarios that are difficult or impossible to reproduce in the real world. To support this data-intensive approach, the project will leverage the computational power of a supercomputing infrastructure, ensuring scalable and efficient generation, simulation, and model training workflows. The outcomes will foster next-generation AI systems capable of responding faster and more reliably in safety-critical applications. The project is conducted in collaboration with international partners as a specific work package of the InnovAIte international research project.

Researcher: Denys Biriukov  

OPEN-35-3     

Design of polysaccharide-binding peptides using molecular simulations and evolutionary algorithms   

LUMI-G  Alloc=18150 

This project aims to design novel short peptides with high binding affinity for polysaccharide structures present on the surfaces of mammalian and bacterial cells. These saccharide motifs act as unique biological markers and represent promising targets for biomedical and bioengineering applications, such as targeted drug delivery and biosensing. However, discovering new peptides with high binding specificity remains a major challenge due to the vast combinatorial space of peptide sequences and the intricate nature of saccharide–peptide interactions. To overcome this challenge, we will apply a novel methodological framework that integrates evolutionary algorithms with atomistic molecular simulations and free energy calculations. This framework enables efficient identification of new peptide patterns with high affinity for selected oligosaccharide targets, specifically those within the mammalian glycocalyx and on the surfaces of Gram-negative and Gram-positive bacteria. The project will generate atomic-level insights into peptide–saccharide interactions and deliver a curated library of high-affinity peptide motifs. These outputs will support the development of next-generation molecular tools for precise cell-type recognition, biosensing platforms, and targeted therapeutic delivery systems.

Researcher: Carlos Manuel Pereira Bornes  

OPEN-35-30   

Machine learning potentials for predicting NMR tensors of zeolites under operando conditions 

Karolina CPU  Alloc=48700;  LUMI-G  Alloc=3600    

Zeolites are porous materials widely used in countless industrial processes, from oil refining to water purification. Despite their widespread use, understanding their behaviour under realist conditions remains a major challenge hindering the development of next-generation sustainable technologies, such as green fuels. Current simulations are often too costly to thoroughly explore the diversity and intricate dynamics of zeolites of zeolites. In this project, we will develop a machine learning interatomic potential that can simulate how zeolites behave under operando conditions by predicting their nuclear magnetic resonance spectra. Our models will enable long equilibration simulations while retaining the accuracy of meta(GGA) ab initio calculations, with chemical compositions and hydration levels mimicking those used experimentally. This will allow us to predict time-averaged NMR tensors, enabling direct one-to-one comparison with experimental NMR data and contributing to the atomistic understanding of zeolites under operando conditions.

Researcher: Diana Csontosová         

OPEN-35-31   

Spin Excitations in Half-Metallic CrO₂          

Karolina CPU  Alloc=5700;  LUMI-C Alloc=8400       

CrO₂, well known for its use in magnetic recording, is a half-metallic ferromagnet (FM). Its half-metallicity combined with FM order makes it attractive for industrial applications. Majority-spin electrons are conductive, while minority-spin electrons are semiconductive, resulting in strong spin polarization and a large magnetic moment—key properties for spintronics. Understanding the microscopic origins of FM order and the mechanism of rapid spin depolarization with temperature is essential.  In our project, we will use Dynamical Mean Field Theory based on a realistic band structure from Density Functional Theory to capture local electronic correlations, focusing on both t₂g and eg orbitals. By moving beyond simplified models, we aim to clarify the mechanisms stabilizing FM order. Additionally, inverting the Bethe–Salpeter equation will allow us to calculate dynamic susceptibility and study magnon excitations linked to temperature-driven spin depolarization.  Our work will deepen the understanding of magnetic ordering in CrO₂ and support the design of materials with improved spintronic performance.

Researcher: Paolo Nicolini    

OPEN-35-32   

Early-stage abrasive WEAR of MoS₂ across a STEP edge (WEARSTEP)        

Barbora CPU  Alloc=27000    

Molybdenum disulfide is one of the most widely used solid lubricants. While the nanotribological properties of MoS2 have been intensively investigated for what concerns their frictional behavior, the understanding of abrasive wear processes on these materials is in a very early stage. The proposed project constitutes part of a joint experimental-computational study aimed to a deeper understanding of the atomic-scale wear mechanisms. By sliding a rigid diamond tip on the layered material, we will characterize the wear response and the nanostructures (flakes, chips, clusters) formed in the process. A similar study (at different scales and with different resolution) will be conducted by foreign collaborators by means of atomic force microscopy experiments. In particular, we will focus on the effect that step edges may have on the overall process. We plan to quantitative estimate the friction force generated by the scratching process, together with quantitative estimations of the amount of debris formed. All in all, the project will shed light on fundamental mechanical and tribological properties of 2D materials, which are rarely addressed in the literature. These results offer new insight into the physical mechanisms governing friction and wear in layered solids and provide a framework for precision cutting and nanomachining in van der Waals materials, relevant to next-generation devices at sub-micrometer scales.

Researcher: Martin PYKAL    

OPEN-35-33   

Structure and Activity of Carbonic Anhydrase Immobilized on Amine-propyl-tethered Mesoporous Silica SBA-15 for Enhanced Capture and Sequestration of CO₂

Karolina CPU  Alloc=19600

The active removal of CO₂ remains a key challenge for long-term climate sustainability. A promising large-scale solution is its fixation as carbonate minerals on the ocean floor, though this process occurs too slowly under natural conditions to address the current atmospheric excess. To accelerate the initial rate-limiting step, conversion of CO₂  to bicarbonate in seawater, carbonic anhydrase (CA) has been investigated as a biocatalyst. While free CA increases this reaction rate by up to 107-fold, immobilization required for enhanced stability often reduces its activity from structural changes. To address this, CA is being embedded within mesoporous SBA15 silica, and its structural and functional properties are being studied. A theoretical framework is applied to investigate enzyme-surface interactions at the molecular level using homology modeling, molecular dynamics simulations, and free energy calculations. The immobilization process is modeled in silico to reveal preferred binding orientations, conformational effects on the enzyme, and the impact on active site accessibility. These simulations are combined with solid-state NMR data to assess structural integrity and identify molecular determinants of catalytic activity. Through this integrated approach, SBA15 is being evaluated and optimized as a highly compatible matrix for CA immobilization under seawater-like conditions, supporting future applications in ocean-based CO₂ sequestration technologies.

Researcher: Marie Behounkova       

OPEN-35-34   

Multiscale Deformation of Enceladus`s Ice Shell     

Karolina CPU  Alloc=1500      

Enceladus, one of Saturn’s small icy moons, displays a range of unique features, including water-rich plumes erupting from its South Polar Region (SPR), elevated surface temperatures in that area, and a varied, geologically complex surface. These observations point to a dynamic history shaped by multiple processes acting on different time scales. In this study, we investigate how these processes influence the stress in Enceladus’s ice shell and potentially lead to the formation of the surface features observed today. We use numerical simulations to explore how different physical mechanisms—both internal, such as ice shell flow, and external, such as gravitational interactions with Saturn—affect the moon over time. On short time scales (1.37 day), we examine the role of tidal forces resulting from Enceladus’s slightly elliptical orbit around Saturn. On longer time scales (0.01 to 1 million years), we study the effects of non-synchronous rotation, where the moon’s rotation rate differs slightly from its orbital period. Over geological time scales, we investigate how slow, viscous flow within the ice shell, driven by variations in its thickness, contributes to surface deformation. Our simulations use the most recent data on Enceladus’s ice shell geometry to better understand the link between internal stresses and the moon’s striking surface features.

Researcher: Jan Kuneš          

OPEN-35-35   

Circular dichroism in resonant inelastic x-ray scattering     

Karolina CPU  Alloc=19600    

Resonant inelastic x-ray scattering (RIXS) is a rapidly developing spectroscopic technique for investigation of excitations in solids. Fueled by the advance in energy resolution RIXS has acquired a exceptional status due to its ability 'to see' otherwise invisible excitations and, at the same time, provide momentum resolved information. The number of control parameters, which provides for the versatility of the method, makes the interpretation of the raw data difficult. Therefore the theoretical calculations are typically necessary.  In this project we aim to use the polarization of the incoming photons to extract information.  Similar techniques in absorption, e.g., x-ray magnetic circular dichroism (XMCD), has been used to study ferromagnets and more recently also antiferromagnets. Our theoretical analysis indicated that XMCD and RIXS-CD behave differently when it comes to time-reversal symmetry. This has a important implications on how the RIXS data are to be interpreted, in particular when obtained on altermagnets.  We will simulate RIXS-CD spectra in CrSb, NiF₂ and MnO representing several typical crystallographic structures.

Researcher: Marco Vitek      

OPEN-35-36   

Coherence length in cyclocarbons    

Barbora NG  Alloc=15900;  Karolina CPU  Alloc=9800

Cyclocarbons are molecular carbon rings, a special type of all-carbon clusters. While their spectral signatures were first detected by astrochemists nearly a century ago, they arrived at the forefront of chemistry only in 2019 with the on-surface synthesis of cyclo[18]carbon. Since then, many cyclocarbons have been characterised using scanning probe microscopy, exciting a broad scientific community. A central property is their (anti-)aromaticity, associated with the coherent delocalisation of the electronic wavefunction around the entire molecular ring. The coherent delocalisation results in quantum interference and length-independent conductance, which are critical for designing next-generation molecular electronics devices. Understanding and accurately modelling the electronic structure of cyclocarbons is therefore not only of fundamental interest, but also a key step toward realising quantum-enabled technologies. In contrast to many experimental advances, an accurate theoretical description of cyclocarbons still remains a major challenge. Modelling cyclocarbons is challenging due to their high symmetry, making the computational results very sensitive to the employed level of theory. In this project, we will develop a computational method specifically tailored for all cyclocarbons systems by benchmarking against high-level reference calculations on small cyclocarbons. The resulting method will provide insights into the size limit of quantum behaviour of cyclocarbons and related systems, guiding the design of molecular electronics based on quantum technologies. 

Researcher: Matus Dubecky 

OPEN-35-37   

Benchmarking Iron-Based Single-Atom Catalysts    

Barbora FAT  Alloc=310;  Barbora NG  Alloc=15900;  Karolina CPU  Alloc=45400;  Karolina FAT  Alloc=110         

New materials are crucial for green energy technologies like fuel cells. A key challenge  is replacing costly platinum with catalysts made from abundant elements like iron.  Current computer models struggle to predict the behavior of these promising new  materials, hindering innovation. This project will use the Karolina and Barbora  supercomputers to create ultra-accurate, atomic-level models of such iron-based  catalysts. By establishing a definitive \gold standard\" for how these materials function,  our research will provide the foundational knowledge needed to design the next  generation of efficient, low-cost catalysts, accelerating green economy transition.

Researcher: Ievgeniia Korniienko     

OPEN-35-38   

Spin-lattice modeling of magnetoelastic effects in cubic crystals with antiferromagnetic ordering          

Barbora CPU  Alloc=30000;  Barbora FAT  Alloc=100;  Barbora GPU  Alloc=1424;  Barbora NG  Alloc=9000;  Karolina CPU  Alloc=21700;  Karolina FAT  Alloc=100;  Karolina GPU  Alloc=4000;  LUMI-C  Alloc=2000  

Magnetoelastic interactions couple the motion of atoms in a magnetic material with atomic magnetic moments and are interesting from both a scientific and applied perspective. The development of computing technology and approaches based on numerical methods has opened up the possibility of a more qualitative and accurate study of magnetoelastic phenomena. Only the analytically derived linear theory of magnetoelasticity (which contains number of simplifications and as a result it only works in the region where the neglected higher-order terms do not have a significant effect) were used so far. One such computational approach is to create spin-lattice models that simulate the simultaneous dynamics of atoms and magnetic moments. Recently, a number of successful spin-lattice models have been created for accurate simulating magnetoelastic effects in cubic ferromagnetic crystals. However, for more complex structures and magnetic orderings, such as antiferromagnets, this issue remains open.  In particular, this is due to the lack of effective potentials for antiferromagnets, but they can be created from the analysis of data obtained from Density Functional Theory (DFT). Our project is dedicated to solve this problem for cubic antiferromagnets by developing suitable spin-lattice models.

Researcher: Lukas Neuman  

OPEN-35-39   

Inductive Bias of Deep Neural Networks for Computer Vision        

Karolina GPU  Alloc=5000     

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. 

Researcher: Pavel Šuma        

OPEN-35-4     

Improving Instance-Level Retrieval Generalization with a Similarity-based Transformer

LUMI-G  Alloc=5000   

We are developing a novel image-to-image similarity model for image search (finding photos of the same object), with a primary focus on achieving robust generalization to unseen domains. Current state-of-the-art methods often excel within the domain they were trained on (e.g., landmarks), but their performance on entirely new categories (e.g., products or artworks) remains a significant challenge. This limits their applicability in broad, real-world scenarios.  We are exploring a small and efficient neural network that builds upon powerful, pre-trained foundation models for visual keypoint extraction. The core idea is that our model does not operate on the raw visual descriptors themselves; instead, its primary input is a matrix representing the pairwise similarities between the descriptor sets of two images. By learning from these abstract similarity patterns rather than domain-specific appearances, the model can develop a more transferable and universal measure of image similarity. Furthermore, operating on similarity scalars instead of high-dimensional vectors provides a significant speed-up in image search time.  To validate our model, we establish a comprehensive benchmark of eight datasets spanning diverse domains, such as landmarks, art, products, toys, or household items. 

Researcher: Marek Gebauer 

OPEN-35-40   

High-Fidelity CFD Dataset Generation for AI-Based Surrogate Modelling of AEM Electrolysers    

Barbora NG  Alloc=15900;  Karolina CPU  Alloc=65500

Hydrogen is a key energy carrier for a sustainable future, and AEM (Anion Exchange Membrane) electrolysers offer a promising route to its efficient and affordable production. This project uses advanced computer simulations (CFD – Computational Fluid Dynamics) to analyse the inner workings of AEM electrolysers, with a special focus on determining material parameters of the membrane that are extremely difficult to measure experimentally. Using ANSYS Fluent, we will simulate various operating conditions to generate a high-quality dataset. This dataset will be used to train an AI-based surrogate model, which will later serve as the foundation of a digital twin for the electrolyser. The results will help accelerate design, reduce testing costs, and improve efficiency in hydrogen production systems.

Researcher: Ales Prachar      

OPEN-35-41   

DeltaWing     

Barbora CPU  Alloc=2200;  Karolina CPU  Alloc=1700          

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. These vortices lower the pressure above the wing, enhancing lift and enabling the aircraft to maintain stable, controlled flight at steep angles. This characteristic makes delta-wing configurations particularly well-suited for agile, high-performance aircraft.  However, accurately capturing this behaviour in CFD simulations is challenging. Resolving the vortices requires unsteady simulations, advanced turbulence modelling, and sufficiently refined computational grids. In this project, we investigate the aerodynamic characteristics of delta-wing designs and assess the capabilities of current simulation tools to model them effectively. 

Researcher: Christian Sippl   

OPEN-35-42   

From Deep Learning Association to Real-Time Earthquake Early Warning: Improving TEAM for Operational Deployment   

Karolina CPU  Alloc=200;  Karolina GPU  Alloc=1000           

The ability to issue warnings seconds before strong shaking from an earthquake arrives can drastically reduce loss of life and economic damage. Building on insights acquired under the ERC-funded MILESTONE project (StG-2020-947856), and as follow up of projects OPEN-24-76 and project OPEN-26-52, we will take the state-of-the-art TEAM transformer model, originally developed in TensorFlow, and reimplement it in a clean, modular PyTorch framework (Paszke et al., 2019), laying the groundwork for rapid experimentation and operational deployment. We will introduce targeted architectural enhancements, drawing on lessons from custom TEAM variants (Huang et al., 2024), such as a sliding-window inference scheme for smoother predictions, data augmentation via waveform and station-coordinate rotations, spatially normalized embeddings to improve generalization across network geometries, and advanced loss functions to balance detection speed against false-alert risk. Training and testing on diverse, real-world waveform archives from Japan and Italy,  our system will produce end-to-end ground-motion estimates and ShakeMap predictions directly from raw multi-station data. GPUs are essential to process the vast volumes of continuous seismic data at high throughput, accelerate large-batch model training, and enable real-time inference for timely alerts. The outcome will be an open-source, deployment-ready EEW system powered by cutting-edge transformer architectures and deep learning. With concise documentation and reference implementations, this toolkit will empower seismic agencies worldwide to issue faster, more reliable alerts, ultimately saving lives and strengthening global earthquake resilience.

Researcher: Alexander Molodozhentsev      

OPEN-35-43   

Particle-In-Cell Simulation Study of Laser Wakefield Acceleration for Generating High-Quality GeV-Class Electron Beams         

Karolina CPU  Alloc=24500    

A plasma-based acceleration scheme for particle acceleration by space charge wave was proposed by Y. Fainberg in 1956. This acceleration approach allows one to overcome one of the most significant limitations in conventional accelerators - limited electric field gradient in radio frequency accelerating structures. Extreme laser-plasma accelerating gradients, demonstrated experimentally by different teams, offer a path towards a compact laser-plasma accelerator (LPA). Such a compact accelerator can be used as an electron beam driver needed in a broad variety of applications, including free electron lasers (FEL), Thomson sources and even electron-positron colliders with TeV energy.   Laser-plasma acceleration has been the subject of active research overdecades to generate high-quality electron beams with up to GeV energies. A significant effort is being made to improve the quality, e.g., high charge, low energy spread, small beam emittance, and low divergence of the accelerated electron beam. The electron beam quality in a laser-plasma accelerator (LPA) strongly depends on the injection mechanism. In this study, we will investigate different injection mechanisms and their influence on the accelerated electron beam quality. In particular, we will explore the laser wakefield accelartion process using self-injection scheme, e.g., density downramp injection, injection in a pre-formed plasma channel, and density tailoring in a capilary setup and ionization induced injection mechanism e.g., injection from neutral gas in a single-stage configuration. Additionally, we will investigate the multi-stage acceleration process using Particle-In-Cell simulations. Specifically, we examine the charge coupling and acceleration of externally injected electron beams in the booster stage. We aim to uncover the underlying physical mechanisms and identify critical parameters that affect the trapping of the injected beam into the wakefield. 

Researcher: Ondřej Kobza    

OPEN-35-44   

Trustworthy and Resource-Efficient AI: Advancing Secure Solutions with Small Language Models          

Barbora CPU  Alloc=400;  Barbora GPU  Alloc=600;  Karolina CPU  Alloc=400;  Karolina GPU  Alloc=3000;  LUMI-G  Alloc=1000   

We present a novel architecture designed to improve the reliability, trustworthiness, and security of AI systems, being developed within the Central European Digital Media Observatory (CEDMO) project, whose mission is to combat information disorders. At the core of our approach is a code-based Small Language Model (SLM), deliberately chosen for its fast inference and low computational requirements. This choice allows for rapid, cost-effective deployment in real-world scenarios, enabling efficient scaling and accessibility even in environments with limited resources—an important consideration for actionable, on-the-ground applications in combating information disorders. Our system specializes in coding and semi-logical reasoning use cases. The SLM is strictly optimized to generate safe and secure code both with and without the support of Retrieval-Augmented Generation (RAG), with RAG serving exclusively to enhance factual accuracy and mitigate hallucinations through real-time contextual retrieval, rather than as a primary safety layer. Safety and robustness are embedded within the SLM itself through prompt optimization, curated dataset construction, and reinforcement learning, ensuring it consistently prevents unsafe code generation and demonstrates enhanced reasoning capabilities—addressing a core secondary goal of the CEDMO project. For further reliability, the system incorporates specialized modules: a Planning Tool for workflow management, an Intention Recognition Tool for the proactive detection and mitigation of unsafe or malicious prompts, and comprehensive code analysis tools to verify output safety and functionality. We will evaluate our architecture through a multi-faceted approach: the development of a new security-oriented dataset, systematic expert assessment, reinforcement learning from human feedback (RLHF), and rigorous adversarial testing, such as internal red-teaming, to ensure resilience to emerging threats and maintain reliable system outputs. By leveraging the cost-effectiveness and speed of SLMs within a robust, security-focused architecture, our solution directly advances the reliability of AI systems for coding applications, supporting CEDMO’s broader mission to build societal resilience against mis- and disinformation.

Researcher: Rene Kalus

OPEN-35-45   

Heterogeneous ionic clusters of rare gases I. Interaction model.   

Karolina CPU  Alloc=1600      

A broader project aim of building up a diatomics-in-molecules model of heterogeneous (mixed) rare-gas cluster cations is intended. The project consists of three phases: i) intracluster interactions, ii) interaction with electromagnetic radiation, and iii) nonadiabatic dynamics, the former phase being mainly focused on the development and implementation of a computationally efficient model of the electronic structure of the heterogeneous ionic clusters while the two latter phases mainly comprise applications of the developed software solutions in production calculations of photoabsorption and photodissociation,of the clusters, and simulations of collision dynamics of elementary processes in cold plasmas of mixed rare gases. The present project proposal specifically covers phase (i).

Researcher: Pavel Praks        

OPEN-35-46   

Stochastic and deterministic methods for optimisation of distribution networks in the energy sector VII          

Barbora CPU  Alloc=500;  Barbora GPU  Alloc=500;  Karolina CPU  Alloc=900;  Karolina GPU  Alloc=1500          

Power systems are part of a critical infrastructure. Their operations are affected by various sources which increase demands on the complexity of mathematical models used for simulating their behavior. Many such simulations need to be run by computer control systems to assess consequences of control actions to select the optimal one. The problem is even more demanding in the presence of uncertainties, where multiple possible evolutions need to be considered.   To use computational resources more efficiently, the analytical methods themselves need to be optimized. This includes comparing performance of variants of analytical pipelines, its dependence on power system characteristics and the setup of software control parameters.    This project aims to develop, test and benchmark specialized analytical tools adapted for high-performance computing infrastructure, providing recommendations for the configuration and control of the investigated energy systems.  

Researcher: Ivan Kološ          

OPEN-35-47   

Numerical modeling of load of structures in quasi-static effect of wind    

Karolina CPU  Alloc=900        

The project is focused on numerical modeling of flow around objects in the atmospheric boundary layer. This issue is complicated mainly due to the atmospheric turbulence, which requires the use of advanced numerical models of the flow coupled with detailed computational mesh of the domain. This research will contribute to bigger efficiency in design of building structures.

Researcher: Petr Strakos       

OPEN-35-48   

Multimodal Deep Learning for Clinical Image Segmentation and Classification in Neonatal, Oncologic, and Spinal Diagnostics     

Karolina CPU  Alloc=300;  Karolina GPU  Alloc=1700;  LUMI-G  Alloc=1500

This project aims to develop advanced artificial intelligence tools to support medical image analysis in three critical areas: neonatal eye screening, colorectal cancer surgery planning, and spinal diagnostics. In the case of Retinopathy of Prematurity (ROP), which can lead to blindness in premature infants, we will use deep learning to automatically detect and classify disease symptoms from retinal images. For colorectal cancer patients, we are creating 3D models of abdominal blood vessels from CT scans, helping surgeons plan safer and more effective operations. Finally, for spine X-rays, we are designing real-time algorithms to identify and segment vertebrae, which supports diagnosis of scoliosis and other spinal disorders. All methods rely on powerful deep learning models trained on medical data and run on national supercomputing infrastructure. The project is conducted in collaboration with clinicians from the University Hospital Ostrava and supports the mission of the LERCO research project. Our goal is to bring practical AI innovations to real-world medical care and improve diagnostics and outcomes for patients across multiple fields.

Researcher: Marta Jaroš       

OPEN-35-49   

Automated Tuning of Workflows Executions on Remote Computational Resources III.

Karolina CPU  Alloc=1900;  Karolina GPU  Alloc=700           

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.

Researcher: Dominik Legut   

OPEN-35-5     

Heat transport in novel nuclear fuels           

Barbora CPU  Alloc=40080;  Barbora FAT  Alloc=100;  Barbora GPU  Alloc=1080;  Barbora NG  Alloc=10432;  DGX-2  Alloc=300;  Karolina CPU  Alloc=48200;  Karolina FAT  Alloc=100;  Karolina GPU  Alloc=6000;  LUMI-C  Alloc=8000     

Metallic nuclear fuels are advantegous over the insulating-current used mixed-oxide (primarily based on UOx) fuels as they can operate at higher temperatures and normal pressures, i.e. with higher efficiency. In addition, the desing of novel nuclear reactors of IVth generation[1] assumes that they are immersed into the liquid molten salts and therefor the transfer heat efficiency is even higher. In this project we investigate the electronic and heat transport in uranium(U) and thorium(Th) carbides as these systems are metallic and exhibiting by factor 10 higher thermal conductivity than UOx based fuels. However, U and Th differ substatially by the degree of localized vs. itinerant behavior of 5f-electrons[2]. We  determine the leading quantities for the electron and heat transport using mainly the HPC for solving the density functional theory calculations in order to get the proper model of the electronic structure of thorium-carbon and uranium-carbon systems. We analyze to which temperature the phonons(electrons) govern the heat transport as well as the effects of spin-orbit coupling and strong electron correlations on the electron-phonon coupling strenght that is assumed to be the leading quantity for the heat transfer in novel nuclear fuels for operating temperatures or in general in f-electron systems. In addition, the same quantity (electron and lattice thermal conductivity) are exploited in a design of a highly  efficient thermoelectric materials. In contrast to nuclear fuels one aims to limit them as they are in denominator of so-called figure of merit for thermoelectric materials performance.

Researcher: Jindřich LNĚNIČKA        

OPEN-35-50   

Permeation and interactions of small molecules on membranes   

Karolina CPU  Alloc=1600;  LUMI-C  Alloc=2000;  LUMI-G  Alloc=4500       

To interact with the internal space of a cell, small molecules must first permeate the membrane or bind to proteins embedded within it. Consequently, a vast number of potential interactions can be observed and studied across highly variable membrane systems, especially since membranes themselves differ significantly in composition. This complexity presents a major challenge for drug researchers: they must not only quantify permeability as a numerical value but also investigate the underlying mechanisms of permeation. 

Researcher: Jana Pavlů         

OPEN-35-6     

Machine-learned potentials for atomistic simulations of fracture in SiO₂ flint mineral including hydrogen-related interactions   

Barbora CPU  Alloc=33200;  Karolina CPU  Alloc=36900;  Karolina GPU  Alloc=1500        

For thousands of years, humans have shaped flint into tools of remarkable sharpness and precision. What makes flint exceptional is its ability to fracture along smooth, curved surfaces – known as conchoidal flaking – producing edges so fine, they can surpass even surgical steel. Some flint and obsidian blades are just a few atoms wide and cause less tissue damage than modern scalpels. However, despite its long history, the fundamental question – How does flint break in such a controlled and effective way at the atomic scale? – remains. This project serves as an initial step toward uncovering the underlying atomistic mechanisms behind flint's extraordinary fracture behaviour, including the analysis of how hydrogen-related interactions influence the crack initiation and propagation. We will develop a machine-learned database trained on quantum-mechanical data and then use it in large-scale molecular dynamics simulations of fracture processes. By understanding how such sharpness though fracture is achieved, we seek to guide the design of next-generation brittle materials – ones that break along non-cleavage paths and produce cutting edges sharp through repeated use (compared to flint, which becomes dull after repeated use). From precision tools to medical blades and smart fracture-enabled components, flint may be key to future materials engineered with ancient insight.

Researcher: Vladan Stojnić   

OPEN-35-7      ILTIR: Instance-level text-to-image retrieval 

LUMI-G  Alloc=15500 

Imagine trying to find a photo of your childhood stuffed toy monkey in a huge collection of images. Not just any toy monkey, but your orangutan-like one with shaggy reddish-brown fur, big round eyes, a light beige face, velcro on its hands, and a stitched smile. Most image search tools today would simply return generic pictures of toy monkeys. They often miss the specific details that matter most.  Our aim is to develop a method that can find images based on detailed, object-level descriptions. Instead of searching for broad categories, like “stuffed toy,” our goal is to help computers recognize and retrieve images of specific items that match a unique description. To achieve this, we plan to build on top of recent models that connect images and text and train them with data designed for this task. We’ll select images that show individual, recognizable objects and pair them with rich, descriptive captions that highlight the object's unique look, texture, and features. We’ll also adjust how these models learn, so they better understand the fine details that set one object apart from another.

Researcher: Anton Bushuiev 

OPEN-35-8     

Customizing protein language models for improved therapeutic design   

Karolina GPU Alloc=7900      

Minor structural changes can enable a protein to outperform its native form when repurposed for a specific function. Given the vast computational space of possible variants and limited capacity for experimental validation, efficient AI-driven strategies are essential for guiding protein design. This project aims to customize protein language models for therapeutic protein design. We will develop ProteinTTT to detect subtle local structural changes and extend EVOLVEpro, an AI-guided platform for sequence optimization. Our approach will be applied to enhance the activity of staphylokinase, a low-cost thrombolytic drug candidate with the potential to outperform current clinical standards. Enhanced variants will be experimentally validated in collaboration with Loschmidt Laboratories.

Researcher: Pavlo Golub       

OPEN-35-9     

Machine-Learning Accelerated DMRG for Accurate Excited-State Simulations

Barbora CPU Alloc=5000;  Barbora NG  Alloc=4000;  Karolina CPU  Alloc=34800;  Karolina GPU  Alloc=1000     

Accurately predicting molecular excited states remains one of the central challenges in quantum chemistry, as these states often display strong multireference character and significant electron correlation. The Density Matrix Renormalization Group (DMRG) method is particularly well-suited for such problems, offering high precision in treating dozens of strongly correlated electrons and orbitals. In recent work, we showed that machine learning can significantly accelerate DMRG ground-state calculations, using polycyclic aromatic hydrocarbons (PAHs) as benchmarks. Building on this success, the current project aims to extend the ML-accelerated DMRG framework to excited-state computations. The goal is to develop a highly accurate and efficient tool for modeling molecular excitations, optimized for high-performance computing platforms.