Researcher: Petr Vrchota            


Winglet for Tilt Wing     

Barbora CPU 1810         

The project is focused on the design of a winglet behind a low-loaded propeller at the wing tip to effectively reduce the induced drag and to increase the aerodynamic efficiency of the entire aircraft. In cruise, the slip stream behind the low-loaded propeller is usually slightly twisted and thus may not be sufficient to reduce the intensity of the induced vortex at the wing tip. Airplanes with distributed propulsion usually have a high aspect ratio, but on the other hand a smaller wing area, corresponding to the demands in cruise, and thus higher lift coefficient and therefore higher induced drag. Tilt wing aircraft has a wing with small aspect ratio to be able to transient regime from hover to horizontal flight and vice versa to land vertically. In case of small aspect ratio wing the high induced drag is difficult to reduce effectively. A winglet designed with consideration of the flow field behind the propeller is one option to reduce induced drag, fuel consumption and emissions or to extend the range and endurance.


Researcher: Daniel Langr            


Eigensolver for nuclear structure studies             

Barbora CPU 410;  Karolina CPU 2800   

Nuclear structure calculations aim to obtain wavefunctions of atomic nuclei as eigenvectors of physically-relevant Hamiltonian matrices. The memory requirements for computer representations of these matrices grow rapidly with the increasing number of nucleons. This factor hinders the application of nuclear structure calculations to non-light nuclei, even with the most powerful supercomputers. A possible solution to this problem is recalculating the matrix element “on-the-fly” each time they are needed. The key to this approach is to employ numerical methods that reduce the number of eigensolver iterations as much as possible. The goal of this project is to develop an eigensolver software package based on such methods and verify it on large-scale nuclear structure problems.


Researcher: Ales Vitek 


Big water clusters II        

Karolina CPU 3600         

Water clusters are challenging molecular systems for molecular simulations because the combination of wan der Waals forces and hydrogens bonds create complicated potential energy surfaces of this species. Water clusters have different properties than bulk limit of water molecules, exhibits smooth phase changes and phase coexistence, play important role e.g. in atmospheric chemistry. Their thermodynamics can be also measured experimentally so we can try to develop the best computational methodology for simulations of finite molecular systems to achieve good correspondence with real experiment. Smallest water clusters, which thermodynamic properties can be measured experimentally have size about 50 – 100 water molecules . We would like to access the same size of simulated water clusters to create theoretical study of water clusters thermodynamics comparable with the experiment.


Researcher: Klára Kalousová


Transport processes through the ice layers of Ocean Worlds – Effect of organics

Karolina CPU 1300         

Within the last few decades, the spacecraft-based missions have discovered an increasing number of Ocean Worlds - planets or moons that harbor deep oceans locked beneath an outer shell of ice. Their exploration is driven by the question of the emergence of life in places where liquid water has been present. The interior of an Ocean World is typically formed by a rocky core and a layered hydrosphere. In case of small bodies, the ocean is in direct contact with the core, while in larger bodies, a layer of high-pressure ice forms that may seem to prevent an exchange of material between the core and the ocean. However, our previous work has demonstrated that such an exchange may be possible if melting can occur at the core-hydrosphere interface. Recently, the presence of organic molecules was suggested to explain the low value of core density constrained by the mission data. Organic molecules, some of them like amino acids being the building blocks of life, are ubiquitous in the outer solar system as demonstrated by the results of the Rosetta mission and analysis of carbon-rich meteorites. The interaction of these molecules with water is essential for understanding the evolution of Ocean Worlds and for assessing their habitability potential. The goal of this project is to investigate the transport of the core-originated organic molecules through the high-pressure ice layer of large Ocean Worlds, and to infer the ocean composition and atmosphere evolution.


Researcher: Vladimír Ulman     


Basic and Applied Research on Parallelization of Image Processing           

Barbora CPU 150;  Barbora GPU 50;  Karolina CPU 400;  Karolina GPU  120          

The sheer amounts of image data per usual single biological or medical experiment keeps increasing and is today often beyond what researchers can process and analyse using available software and hardware. Both fields are thus in a constant lack of new computer science solutions. While the hardware solutions available at the IT4Innovations is extremely performant and well accessible to the community, the software stack, that would bring the computational power to the hands of everyday image practitioners in a comfortable and understandable way, is still scarse. Easier said than done, developing such software is still largely unsolved computer science research task.The principal investigator, who is a researcher at IT4Innovations, has recently designed and succesfully implemented several projects, including the projects HPC Workflow Manager for Fiji and the Cell Tracking Challenge silverGT. The former is bringing a graphical user interface to the popular desktop image processing software ImageJ/Fiji, using which users can relatively easily execute basic image processing pipelines on a supercomputer. In the silverGT project, the use of supercomputer enabled to establish and release densely annotated segmentation training data for tens of real microscopy videos. This data are today used in deep learning networks for training to perform more accurate cell segmentations. To continue the research and development in these topics, this grant is needed to secure access to HPC.


Researcher: Jan Šimkanin          


Magnetic field polarity reversals and their properties in hydromagnetic dynamos driven thermally, chemically and thermochemically           

Barbora CPU 9500         

The Earth's magnetic field is one of the most variable

geophysical fields and provides us with effective protection against high-energy, electrically charged particles from the solar wind, and solar flares and is a useful tool for navigation, not only for us, humans but also for animals. The Earth's magnetic field is generated by convective motions of an electrically conductive melt in the Earth's outer core and penetrates the surface of the Earth. These generation processes are called Geodynamo for short. On the Earth’s surface, a large-scale, dipole-dominated field is observed. However, we have no direct information about the magnetic field in the Earth's core. For this reason, we model magnetohydrodynamic processes in the Earth's core numerically. Numerical modelling of Geodynamo has made enormous progress during the last years thanks to the development of computer technology. Numerical models produce magnetic fields that are close to the observed geomagnetic field. We are also able to reproduce its temporal changes, whether short-term or long-term ones (so-called secular variations). However, we cannot use parameters typical of the Earth's core in our models. The real parameter values cannot be used for computational reasons – so far no supercomputer in the world could solve a Geodynamo model with such values. Nevertheless, we are gradually approaching these values as computers become more and more powerful.


Researcher: Uladzislau Yorsh    


Efficient Transformers   

Karolina GPU 300           

Transformer neural networks, while achieving impressive results in natural language processing, computer vision and other domains, require a massive effort for training and have high computational complexity. The latter disables them from being effectively applied on tasks requiring processing of long inputs, such as summarization, long text question answering and others. The issue comes from an attention computation mechanism, which provides the model a great flexibility at the cost of quadratic time and memory requirements with respect to the input length.In the current work, we are going to explore alternative attention mechanisms which can be used to reduce the computational demands without major performance tradeoffs. In particular, we explore mechanisms employing external memory, which not only allows to factorize attention computation, but also provides an additional source for interpretability.


Researcher: Pavlo Polishchuk   


Fragment-based de novo design and structure optimization       

Karolina CPU 8900         

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 of such computational tools based on the previously developed fragment-based structure generation framework CReM which the main advantage is generation of synthetically accessible molecules that will improve outcomes of medicinal chemistry projects.


Researcher: Tahir Wahab            


First-principles study of optoelectronic properties and photocatalytic performance of ZnO-VXY ( X= Cl, Br, Y = Se, Te) vdW heterostructures  

Karolina CPU 9352         

Two-Dimensional materials have received enormous consideration as photocatalysts for hydrogen production to address worldwide energy crises. In particular, van der Waals heterostructures are considered promising materials due to their adjusting nature in optoelectronic properties, such as efficient carrier separation and type-II band alignment. In this respect, we will consider heterostructures based on ZnO and Janus VXY (X= Cl, Br, Y= Se, Te), and investigate their structural, electronic, Spin-Orbit Coupling (SOC), optical, and photocatalytic properties by means of ab initio techniques. The results will constitute a guide to design new van der Waals materials for photocatalytic applications.


Researcher: Mario Vazdar          


Translocation of cell-penetrating peptides across lipid membranes          

Karolina CPU 10000      

Biological membranes are vital for proper functioning of all living systems. They are composed of phospholipid bilayers which are practically impermeable for water and charged species due to strong hydrophobic character in their interior. However, some charged peptides rich in arginine amino acids (Arg-rich peptides), such as nona-arginine (R9) or penetratin, easily translocate and bring drugs across the membranes by a mechanism which is still not understood at the molecular level.The main aim of the proposed project is to study in detail the two steps of the translocation mechanism. In the first step, we will study how Arg-rich peptides adsorb to the membranes in different experimental conditions mimicked in our simulations, using state-of-the-art molecular dynamics simulations at large scales and supported with free energy calculations of peptide binding. In this way, we will identify what are optimal conditions for peptide binding which is the first step of translocation process. In the subsequent second step of the process, we will use free energy calculations of pore-free translocation across the membranes using special pore reaction coordinates, to obtain the translocation energetics so far not reported in detail in the literature.The knowledge of the energetics of the transport mechanism of Arg-rich peptides will provide new knowledge which is fundamental for further development of smart and controlled drug delivery, which is especially needed in medicinal research.


Researcher: Ondřej Maršálek   


Validation of an MP2-based machine learning potential for bulk aqueous systems           

Karolina CPU 8500         

The field of machine learning is currently one of the most dynamic areas in science. Fast progress can also be seen in the area of machine learning potentials, which have become an important tool for the investigation of static and dynamic properties of atomistic systems using molecular dynamics simulations. Due to this progress, models that used to be state of the art not that long ago are inferior to many more recent models. We have developed a committee neural network potential that promises to have higher accuracy for the calculation of energies and forces for water, a crucial system for many applications. However, to prove the higher accuracy, we need to rigorously validate the accuracy of our model on an independent test set evaluated using an expensive high-level electronic structure method.


Researcher: Raman Samusevich             


Enzymes function prediction using machine learning      

Karolina CPU 600;  Karolina GPU 3800  

Terpene synthases (TPSs) are enzymes (catalytic proteins) responsible for the biosynthesis of the largest class of natural products, including widely used flavors, fragrances, and first-line medicines. Although the amount of available TPS protein sequences is increasing exponentially, characterizing the function of each TPS requires challenging and time-consuming experiments as well as significant domain expertise. The objective of this project is to develop predictive models for characterizing the function of TPSs directly from their protein sequences. Such a model will have a multitude of applications in drug discovery and synthetic biology and will provide an important precedent towards computational characterization of the catalytic function of enzymes directly from their protein sequences.


Researcher: Marta Jaroš             


Offloading of Workflows Executions to Remote Computational Resources IV.      

Karolina CPU 2700;  Karolina GPU 500  

In recent years, the therapeutic ultrasound has grown in a number of applications such as tumor ablation, targeted drug delivery or neurostimulation. Precise preoperative treatment planning tailored individually to each patient is crucial for the maximalization of the treatment’s outcome. The fundamental challenge shared by all applications of therapeutic ultrasound is that the ultrasound energy must be delivered accurately, safely, and noninvasively to the target region within the body identified by a medical doctor. The estimation of treatment outcome heavily depends on the computationally very intensive ultrasound, thermal and tissue models which are only realizable with the use of HPC facilities.Since there is a lack in expertise of the clinical end-users to use HPC resources efficiently, we developed the k-Plan software simplifying the everyday use of the HPC resources without a need to specify task execution parameters, dependencies, and their monitoring. k-Plan also reacts to the problem of task execution parameters selection, e.g., number of nodes, estimated execution time, storage space, etc. The end-users have no knowledge about the strong and weak scaling of the software being used, yet these characteristics have great impact on the calculation cost and overall execution time. The goals of this project are to (1) deploy the k-Plan software and offer it to a small set of pilot end users from clinical environment to execute realistic treatment plans, (2) to tailor the task submission planning logic to IT4Innovations clusters, (3) and investigate methods that allow k-Plan to automatically tune execution parameters for individual tasks.


Researcher: Karel Sindelka        


Computer simulations of colloidal solutions and organic crystals

Barbora CPU 15500;  Karolina CPU 4500              

Aqueous solutions are omnipresent in nature, industrial processes, and daily life. Understanding their behaviour in inhomogeneous environments (self-assembled or confined systems) in equilibrium aand non-equilibrium (shear flow) conditions is important in many applications from medicine to environmental protection. The first part of this project focuses on interactions of surfacant monolayers with soft surfaces that play important role in industrial or household products (e.g., fabric softeners, cosmetic products). We use mesoscopic simulations to provide molecular-level insights into chemical and physical behaviour of these systems in and out of equilibrium.The second part of the project focuses on polycyclic aromatic hydrocarbons (PAHs) and their nitrogen derivatives. PAHs are useful functional materials in organic electronic devices (such as OLEDs or solar cells), but they are also of interest to widely different fields, such as astrochemistry (components of interstellar dust), so understanding their properties and behaviour is important. Incorporating nitrogen into these compounds (azaPAHs) allows for tuning their properties to find even more use in modern organic electronics materials. However, the accessibility of these compounds is very limited largely because the research is scarce. We use all-atom simulations to investigate the PAHs and azaPAHS with stress on their structural properties under wide range of temperatures above and below their melting points.


Researcher: Jana Pavlů


Structure defects on phase boundaries in SiC-TiSi2 nanocomposites       

Karolina CPU 13998      

Our modern society based on the utilisation of advanced materials requires the development of new structural materials that could be used at higher operational temperatures or reveal unique properties. Nevertheless, the properties of those materials are significantly affected by structural defects such as interfaces and the processes taking place on them. Hence, the proposed project aims to understand how the atoms and structural defects, such as phase boundaries, antisites or vacancies, behave in the SiC-TiSi2 nanocomposites and affect their properties. Unfortunately, these relations very often touch the experimentally unreachable areas. However, they can be studied through theoretical approaches such as computational modelling; here, the new on-the-fly machine-learned force fields based on the ab initio molecular dynamics. This new approach overcomes some limitations of the present atomistic simulations (small computational cells and 0 K temperature) and classical molecular dynamics (low precision of interatomic potentials). Hence, we will be able to include more realistic conditions in our study and save CPU time.


Researcher: Michael Komm      


Collisional particle-in-cell simulations of inverse sheath

Karolina CPU 15000      

Since the dawn of plasma physics research, it has been assumed that the plasma potential should be decreasing towards a floating wall, either monotonically (Debye sheath) or forming a virtual cathode (space-charge limited sheath). Recently, a fundamentally different regime dubbed  inverse sheath has been proposed in case of strongly emitting walls and presence of charge exchange collisions, featuring increasing potential towards the wall. In this project we aim to investigate this regime by means of particle-in-cell simulations with thermionic emission and charge exchange collisions in conditions relevant to future thermonuclear reactors.


Researcher: Rene Kalus


Ternary recombination processes in cold rare-gas plasmas – phase II      

Karolina 9100  

The present project extends a preceding one (Ternary recombination processes in cold argon plasmas, OPEN-25-37) and aims to provide additional data on the formation of molecular ions in rare-gas plasmas. The molecular ions of rare gases play an important role in cold rare-gas plasmas and crucially influence their interaction with the environment and, as a consequence, potential benefits of their use in applications. In the preceding project, we performed an introductory study of recombination processes in argon, in the present project, we plan a) to amend this study by additional calculations yielding a more complete view of the formation of diatomic argon ions in strong electric fields (present in plasma jets) and b) to use the experience we got during the preceding calculation in subsequent investigations of krypton, a heavy rare gas for which the molecular are even more important than for argon and for which relativistic effects play a major role in the electronic subsystem. Like in the previous project, recombination rates will be calculated as well as properties of nascent populations of molecular ions formed in recombination processes (e.g., distributions of electronic and/or rotational-vibrational excitations in formed ions, both influencing their reactivity with environment species). Like before, the project will be solved in collaboration with several groups at the Université Toulouse III – Paul Sabatier, France, presently formalized by a cotutelle agreement.


Researcher: Martin Pykal        


Initial steps in carbon dots formation: a molecular dynamics study          

Karolina CPU 18000      

Carbon dots (CD) have come to prominence in the last decades mainly due to their excellent fluorescent properties, biocompatibility, and rich surface chemistry that offers versatile application possibilities, including sensing, imaging, and catalysis. Despite the significant progress in those fields, there is still a limited knowledge about the atomistic insight and the related underlying mechanisms governing their exceptional properties and behavior. Understanding the origin of CDs is crucial for developing new synthesis methods and fine tailoring their properties for specific applications. It is generally considered that the carbon dots are composed of carbon (often graphene-like) core with various functional groups (OH-, COOH-, NH2-) on the surface. However, the chemical structure of the CDs and especially the surface functionalization strongly differs from the synthesis method used. A frequently studied reaction is the pyrolysis of the citric acid (CA) and amines which provides nitrogen-rich CDs with a high quantum yield. Here, the focus is set on modelling of the initial steps of the polymerization reaction involving CA as well as on the processes preceding the formation of carbon core itself. In silico studies can provide unprecedented resolution required for prediction of reaction products and deep rationalization of the underlying chemical processes.


Researcher: Diana Csontosová


Magnetic Excitations in Ruthenates with Strong Spin-Orbit Coupling       

Karolina CPU 10600      

Dynamics of magnetic excitations is one the key properties of materials for spintronic applications. While it is routinely studied in experiments, its theoretical prediction for specific materials is notoriously difficult. We have recently developed a method to do so and tested it successfully on simplified models. Now it is time to demonstrate its power for a real material. SrRu2O6 provides an ideal case for such a study. It has a simple enough structure to be tractable, yet the interpretation of its properties remains controversial. Our method  is capable of answering the outstanding open questions concerning the electronic structure of SrRu2O6. The implications of our results will not be limited to this specific material, but will hold for antiferromagnets in general. Moreover, if successful our results will be the first calculations of magnon spectra using dynamical mean-field theory (DMFT) for a specific material. This would be a large methological step expanding the capabilities of DMFT, which is currently the method of choice for investigation of materials with strongly correlated electrons with hundreds of users around the world.


Researcher: Jan Hůla    


Efficient Language Models          

Karolina GPU  1000       

Large language models (such as chatGPT) have enabled many exciting applications, but these mainly benefited English speakers and large tech companies that can internally develop and train models with a budget not accessible to ordinary academic researchers. It is conceivable that this will increase the disbalance in terms of access to information for non-English speakers and, in terms of research, create a gap between academic and industrial labs. This project is focused on developing efficient language modeling techniques that should democratize access to these technologies. We will explore several ways how to automatically obtain much smaller/cheaper models from the larger/expensive ones. Our work will utilize ideas from model fine-tuning, a distillation of an expensive “teacher” model to a cheaper “student” model, and model ensembling. We will also develop a metric that will help us to estimate which student model is suitable for finetuning on a given task. This will allow researchers/developers to quickly pick a suitable model from a large pool of pre-trained models. Using the results from many computational experiments, we will produce training pipelines and models with a different tradeoff between accuracy and computational requirements.


Researcher: Jan Zemen


Probing magnetic structure of Heusler alloys with martensitic transformation    

Barbora CPU 38000       

A large subgroup of Heusler alloys has been investigated systematically  for their elastomagnetic multiferroic behaviour, so-called magnetic shape memory (MSM) phenomena. This behaviour is underlined by structural diffusionless and displacive transitions to low symmetry phase called martensitic transformation (MT). The occurrence of MT shows extreme sensitivity to chemical composition of the alloy which constitutes a great challenge to materials science both theoretical and experimental. The sensitivity may originate in the interplay of chemical (dis)order with specific magnetic order which can be investigated in detail by local methods – Mössbauer spectroscopy (MS) and nuclear magnetic resonance (NMR) making use of the hyperfine interactions. Disorder can also affect the hysteresis of MT which is a crucial parameter for the efficiency of magnetocaloric materials and other energy-conversion devices. Here we intend to study Heusler alloys with general formula X2MnY where X = Fe, Ni and Co and Y = Ga, Sn and their off-stoichiometric derivatives. We will use density functional theory (DFT) to relax large supercells incorporating the chemical disorder to find the local magnetic structure and determine the likelihood of the sought-after MT by comparing the density of states of the austenite and martensite structure. The results can be used in future as an input of a full-potential DFT study of the hyperfine parameters to be compared to values observed in MS and NMR experiments.


Researcher: Jan Navrátil             


Enhancing photocatalytic performance of modified TiO2(001) anatase in water splitting for green hydrogen production

Karolina CPU 23000      

The Green Deal’s ambitious goals require a significant improvement in the current technologies used to produce green fuels, such as hydrogen. Solar energy is a widely accepted and sustainable way to produce hydrogen from water through a process called \photocatalytic water splitting.\" However, there are still limitations and challenges to be addressed to make the process more competitive with other hydrogen production methods. This is because the currently used materials are far from achieving the theoretical maximal efficiency. Titanium dioxide (TiO2), one of the most prominent materials used for photocatalytic water splitting, has a wide bandgap, which limits its absorption of light to only the UV region, resulting in wasted incoming light. We have shown within IT4Innovations OPEN-24-57 that the TiO2 bandgap can be tuned allowing to utilize a broader range of light spectrum, by creating surface defects, such as atomic vacancies, doping, and adsorbed single-atoms (SA) of Pd or Pt co-catalyst. Our aim is to extend the above-mentioned research by exploring the formation and desorption of H2 and O2 molecules from water, comparing the processes on various modified TiO2 surface(s) and identifying the catalytic role of SA-co-catalysts. These results will aid in engineering materials with the best catalytic performance for water splitting, ultimately contributing to the development of more efficient and sustainable hydrogen production."


Researcher: Frantisek Mihok    


Simulated thermoelectric properties of alloys    

Karolina CPU  1400        

Energy consumption is steadily growing with a growing population. Energy needs of modern societies are on the rise as well. Renewable resources and energy management efficiency are proving essential in maintaining energy suficiency while not destroying the planet environment further. Thermoelectric generator provide unique opportunity to improve energy management. These modules made from specialized materials with suitable thermoelectric properties offer cost effective, simple, maintenance free and reversible solution for managing energy from waste heat. Modules consist of multiple N and P semiconductor pairs connected in series. These devices offer flexibility in their use as heating or cooling device or electricity generating device because of Peltier and Seebeck effects respectively. The main downside of current thermoelectric modules is their low efficiency but new materials with marginally improved conversion efficiency sparked renewed scientific and even commercial interest in thermoelectric modules.This project looks to find and evaulate new materials which exhibit significant thermoelectric properties. Specific focus will be placed on promising SnSe alloys. Materials will be modeled and optimized using DFT methods of NWChem. Afterwards their theoretical thermoelectric properties will be determined using Quantum Espresso and LAMMPS molecular dynamics. Lastly, interaction of materials with their surroundings and their properties under different temperature gradients will be investigated using molecular dynamics methods as well. 


Researcher: Marco Vitek            


Investigating the Metal-Insulator Transition in Metal-Ammonia Solutions Using the GW Method

LUMI-C  6500   

The Metal-insulator transition (MIT) is a phase transition that arises from nonclassical properties in condensed matter systems. During MIT, transport properties such as the electrical conductivity or optical reflectivity undergo significant changes, shifting by orders of magnitude between values typical for metals or insulators. MIT is well-understood in solid-state materials, which has allowed for the artificial tailoring of semiconductor conductivity through impurity atom doping, a technique used in the production of microchips and micro-electronics. However, the liquid analogue of MIT, the electrolyte-to-liquid-metal transition, remains poorly understood despite more than a century of investigation. With recent advancements in liquid-jet photo-electron spectroscopy (PES) combined with quantum chemical calculations, we can now conduct detailed microscopic studies. The broad aim of this project is to use theoretical approach to uncover the underlying mechanisms driving MIT in liquid alkali-metal solutions, thus complementing our experimental efforts. In the early 19th century, it was observed that upon dissolving alkali metals in liquid ammonia, the solutions remained stable for extended periods of time despite a high concentration of electrons. At low concentrations, alkali metal cations and solvated electrons are formed, imparting a distinct blue coloration to the solution. Upon increasing the concentration of alkali metals, a distinct change in color from blue electrolyte solutions to bronze or gold-colored metallic solutions is observed. This change in color is an indication of Metal-Insulator Transition (MIT) occurring in the alkali-metal liquid ammonia system, which is directly correlated with a significant increase in electrical conductivity by several orders-of-magnitude. The present project focuses on the direct investigation of solvated electrons originating from dissolving increasing amounts of alkali metals in bulk liquid solvents through electronic structure calculations and from previously calculated ab initio molecular dynamics (AIMD) simulations. The goal is to provide insight into the transition from individual solvated electrons through dielectrons to the onset of more delocalized states.


Researcher: Zdeněk Futera        


Electron transport on bio/metallic interfaces      

Barbora CPU 30720;  Karolina CPU 5760              

Biomolecular electronics is a growing field of technology utilizing biomolecules such as peptides and proteins to form electric contacts between other metallic components. This is motivated by extraordinarily high temperature-independent currents detected on single-protein junctions by STM-like measurements. Although the redox-active proteins are known to transfer charge by the so-called hopping mechanism in their native environment, the transport through protein junctions proceeds via efficient, coherent tunneling. Recently, the bio/metal contact region was identified to control the transport mechanism by the potential drop emerging from the misalignment of the metallic and biomolecular electronic states. Here, we investigate how the surface dipole affects the interface band alignment, particularly surface material, coating, and linker effects. Detail knowledge of these relations between the atomic and electronic structure of the surface and the biomolecular states is key for the further development of biomolecular electronics.


Researcher: Jiří Pittner


Benchmarks and training sets for machine-learning accelerated excited state molecular dynamics

Karolina CPU  Alloc=25000         

Molecular dynamics in excited states including the non-adiabatic and spin-orbit effects is an important theoretical tool for the simulation of photochemical processes which play an important role in nature and technology, like e.g. photosynthesis, phototherapy, photovoltaics, etc. Its computational cost when ab-initio or DFT methods are employed is very limiting in both size of the molecules treated and the length and number of computed trajectories. Machine learning (ML) has recently become very popular thanks to its widespread applications in many areas of science, industry, and commerce. Recently the machine learning methods have been successfully employed to speed up the molecular dynamics in the ground state, and some progress was done also on the ML of excited states and non-adiabatic effects. The aim of this project is to implement ML methods for MD with nonadibatic and spin-orbit couplings and perform ML-accelerated molecular dynamics in excited states undergoing both internal conversion and intersystem crossing processes, with application to (modified) DNA bases.


Researcher: Taoufik Sakhraoui 


Electronic structure and magnetic properties of MXene 2D material        

Barbora CPU 20000;  Karolina CPU 15000           

Magnetic 2D materials have gained great attention in recent years due to their promising applications in electronic and spintronic devices. As a new family of 2D materials, MXene materials Mn+1XnTx (where M = a transition metal, X = carbon or nitrogen, and T = surface terminations such as O, F and OH) may have unusual magnetic properties. In some cases, different transition metals (M = Ti, Sc, Cr, Mn, …) may be well combined and further extending application potential of these promising materials.On the other hand, accurate theoretical description of MXenes remains challenging. Due to the presence of multiple configurations and magnetic alignment for each of them, and the need for accurate predictions of MXene properties, supercells with hundreds of atoms are required. The project will contribute to the fundamental understanding of magnetism on MXene which should stimulate interest in their experimental research as well as technological applications. In this project, the structural stabilities, electronic structure and magnetic properties of MXenes with different magnetic configurations will be calculated by using DFT and DFTB methods.


Researcher: Mikulas Matousek


High level quantum chemistry for potential energy surfaces.      

Karolina CPU 24700;  Karolina FAT 100;  LUMI-C 1400    

Even though the majority of quantum chemical calculations are carried out using density functional theory, which is suitable for high throughput calculations, for more challenging molecules (like transition metal complexes) high level multireference methods are required. As this type of calculations is very demanding, the use of these methods is limited to a few geometries. Unfortunately, this way we lose a lot of information about the potential energy surface (PES). We aim to change that by performing PES scans for several challenging molecules using multireference computational methods we are developing in our group.


Researcher: Athanasios Koliogiorgos    


Study of Phthalocyanine-metal complexes on a superconductor (SUPERPC)        

Karolina CPU 19700      

Phthalocyanine-metal (PcM) complexes, consisting of a phthalocyanine (Pc) anion and a metal or metalloid (M) cation, exhibit notable electronic, optical and vibrational properties. Among their applications are solar cells, nanoelectronics and molecular electronics and STM microscopy. Lesser studied, especially in a consistent computational way, are their vibrational properties and their electronic-vibrational (EV) coupling. Knowledge of these properties can serve many applications, such as enhancement of STM probe analysis ability and charge transport in molecular junctions. In this project, we will study an array of PcM conjugates with emphasis on their electron-vibration coupling. We will also study their interaction with monolayers of superconducting Pb, in order to elucidate the enhancement of their EV coupling and the possibilities that this opens for applications in STM microscopy and molecular electronics.


Researcher: Jiří Jaroš    


Closed-loop individualized image-guided transcranial ultrasonic stimulation        

Barbora CPU 5000;  Barbora FAT 50;  Barbora GPU 100;  DGX-2 10;  Karolina CPU 1000;  Karolina FAT 50;  Karolina GPU 1000;  LUMI-C 300;  LUMI-G 100             

Disorders of the brain, including neurological and psychiatric diseases, affect one in four people. New treatment options are needed with enhanced efficacy and reduced side-effects, costs, and invasiveness. Neurostimulation techniques that modulate the electrical activity of the brain have evolved as an important class of second-line treatments for pharmacoresistant cases. What is needed is a non-invasive brain stimulation technique that can stimulate brain targets with high anatomical precision, unlimited penetration depth, full reversibility, and low risk-profile. This can be achieved using the newly emerging technique of low-intensity focused transcranial ultrasonic stimulation (TUS)  for neuromodulation.This project focuses on the development of closed-loop individualized image-guided transcranial ultrasonic stimulation, under Horizon Europe CITRUS project. The ultimate goal of the CITRUS project is to develop a fully functional prototype of a medical device that integrates an ultrasound transducer system possessing advanced 3D steering capabilities with a custom-built magnetic resonance receiver, enabling high-resolution transcranial neuromodulation with unprecedented flexibility and sensitivity. The computational resources will be used for preoperative MR-based brain imaging, personalized ultrasound treatment planning including temperature mapping, and validation of fast real time re-planning software based on advanced mathematical models and artificial neural networks.


Researcher: Tadeáš Kalvoda      


Mapping the Conformational Space of Non-Canonical Amino Acids and Dipeptides: Insights into the Origin of the Natural Amino Acid Set by Quantum Chemical Calculations         

Karolina CPU 26300      

Why was amino acid set formed into its current shape? Natural proteins are composed of 20 amino acids, which serve as building blocks for a vast array of biological functions. Despite the decades of extensive knowledge of protein structure and function, the fundamental question of the origin of the set of 20 canonical amino acids remains unanswered: why are these 20 natural amino acids utilized by nature, and not others? Moreover, is it possible to create existing secondary structure elements with non-canonical amino acids? Gaining insight into this question can reshape our understanding of protein structure, design, and function and may open new avenues for protein engineering and biotechnology.


Researcher: Masao Obata          


Electronic structure investigation on Ni-Mn-Ga alloy via quasiparticle self-consistent GW approach

Karolina CPU 16000;  Karolina GPU 800

Ni-Mn-Ga alloys possess magnetic and shape memory properties. While it is commonly known that temperature change plays a part in shape distortions, it is later found that external magnetic fields also induce an enormous one that far exceeds the typical magnetostriction. This property comes from transitions among energetically close structures involving nanoscale-modulated structures attributed to magnetic and electronic states. Investigation of electronic structure based on density function theory indicated an incomplete description of the electron correlation effect due to electron localization. This project reveals how electron correlation influences the electronic structure of Ni-Mn-Ga alloys by means of the quasiparticle self-consistent GW approach. The results will contribute to improving and reinforcing the understanding of the electronic structure of Ni-Mn-Ga alloy.


Researcher: Bedřich Smetana   


Training an electrochemical surrogate model     

Karolina CPU 12000;  Karolina GPU 1000             

The applicants plan to build a computationally efficient surrogate model from computationally expensive finite-difference (FD) simulation that models the electrochemical experiment of voltammetry, which is the underlying methodology in glucose sensors, and gas sensors (including carbon monoxide, hydrogen and hydrogen sulphide etc.). The Oxford group has authored the leading textbook on electrochemical simulation Understanding Voltammetry: Simulation of Electrode Processes1 and hundreds of peer-reviewed electrochemistry simulation papers,2-7. Currently they are leading the application of AI in electrochemistry.8-11Typically, experimental results are compared with simulated results. Similar results then indicate that the assumed physics in the simulation may describe the electrochemical system. These simulations typically solve coupled mass-transport PDEs subject to electrode reaction models, which result in so-called voltammograms that take the form of a 2D graph and model the data an experimentalist would measure. Importantly, there is no simple inverse function that could extract the parameters of the mass transport or the reaction from the measured voltammogram.Hence, we plan to train an electrochemical surrogate model, by learning from millions of simulations to:1. Predict voltammograms using neural network in less than 1/10 of a second.2. Extract experimental parameters from voltammograms, a functionality dreamed by scientists for decades!3.  Incorporate the surrogate model to create a new open-source electrochemical GUI software based on the proof-of-concept called “FreeSim”  allowing “click-and-run” AI inference to maximize global impact. However, simulating millions of experiments can be very expensive which is the reason no one has done this before.  IT4Innovations can step in to offer a solution.


Researcher: Ales Prachar            



Barbora CPU 820;  Karolina CPU 800      

The project is focused on the study of the flow field behind the propeller and its impact on the interaction with the wing by advanced CFD methods. The propellers in the UAV (unmanned aerial vehicle) has very often several very different working points. It is highly loaded during the take-off and landing and fulfills very different expectations in cruise. Especially in the case of UAV with tilt wing, it is crucial to understand the impact of the propeller flow on the wing. This is affected by mutual position of the wing and the propeller, which manifests itself very differently for each of the flight phase. This knowledge is useful within the design process of the UAV’s, it seems to be a very important parameter that will allow aerodynamic design of aircraft with higher efficiency and lower emissions, which is fully in line with the current trends.


Researcher: Pavel Jungwirth     


Ab-initio molecular dynamics simulation of an excess electron in small ammonia clusters

Barbora CPU 54800       

The dissolution of alkali metals in liquid ammonia produces blue or bronze solutions that contain alkali metal cations and electrons localized in cavities between solvent molecules. These solutions find extensive applications in organic chemistry, such as the Birch reduction, due to their strong reducing properties.The stabilization of the excess electron in the solution is ascribed to the presence of solvent molecules. Therefore, it is of great interest to investigate the stabilization mechanism in various solvents and determine the minimum amount of molecules required to stabilize these species. Experimental studies have demonstrated that the smallest stable cluster containing a solvated electron consists of 13 ammonia molecules [1]. In contrast, theoretical calculations on linear clusters at 100 K suggest that a minimum of 8 ammonia molecules is required [2]. Nevertheless, the linear structures fail to capture the actual experimental structures observed at finite temperatures and thus do not offer insights into their electronic stability.To address this issue, we will perform ab-initio molecular dynamics (AIMD) simulations to obtain the thermal structures of small experimental clusters containing 2 to 48 ammonia molecules with an excess electron. By calculating their vertical binding energy (VBE), we will investigate the electronic stability (bond strength with solvent) of the finite temperature clusters. This study will provide precious information on the solvation and electronic properties of ammonia solutions, with potential applications in various areas of chemistry.


Researcher: Jiri Klimes


Accuracy and precision for extended systems XI

Barbora CPU 30000;  Karolina CPU 14000;  LUMI-C 2700             

Computer simulations have become indispensable for understanding the majority of experimental observations and thus understanding our world. This is especially true at the atomic and molecular level where experimental observations can become very difficult to interpret. However, if simulations are to be useful, they need to model the system reliably. For example, if there are different pathways for a chemical reaction, the simulation needs to be able to correctly predict their relative importance so as to increase our understanding of the system. The world at the atomic level is governed by the complex laws of quantum mechanics. In practice, this means that increasing the reliability of simulations, i.e., using less approximations, increases the computational cost substantially. While we understand well the impact of some of the approximations, there are some for which our knowledge is limited. In some cases, these can lead to unexpectedly large loss of the reliability of calculations. In our project we analyse some of the problematic cases occurring for calculations of intermolecular interactions, either in small clusters, or in molecular crystals.


Researcher: Martin Srejber       


Graphene-based materials in biomedicine: a molecular approach            

Karolina CPU 40000      

In recent years, there has been a significant increase in the utilization of graphene and graphene-based materials in various biomedical fields, such as biomedical sensing of small molecules, drug delivery, imaging techniques or photothermal therapy applications. This is attributed to the exceptional properties exhibited by graphene, such as its high surface area, thermal and electrical conductivity, and remarkable mechanical strength. These properties arise from its single-layered, hexagonal \honeycomb\" lattice nanostructure consisting of sp2 hybridized carbon atoms that can be functionalized using covalent or non-covalent means. This exceptional combination of characteristics makes graphene a promising candidate for various biomedical research applications.The understanding of the effects of graphene and graphene-based materials on biological membranes is critical to comprehending their potential for use in various biomedical disciplines since biological membranes are typically the initial barriers encountered by foreign substances upon administration. Therefore, the aim of this project is to investigate the interaction between graphene and biological membranes using molecular dynamics simulations at both atomistic (AA) and coarse-grained (CG) resolutions. These simulations will provide insights into the behavior of graphene and biological membranes at the atomic level, allowing for a better understanding of the underlying mechanisms and interactions that occur at the interface.The use of both atomistic and coarse-grained simulations will enable the study of systems with varying levels of complexity, thus providing a more comprehensive understanding of the behavior of graphene-based materials in biological environments. Ultimately, the findings from this project will contribute to the development of graphene-based materials with optimized properties for use in various biomedical applications.


Researcher: Lukas Neuman       


Inductive Bias of Deep Neural Networks for Computer Vision     

Karolina GPU 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.


Researcher: Alžběta Špádová 


Laser-driven electron acceleration reaching GeV energies            

Karolina CPU 20000;  Karolina FAT 100;  Karolina GPU 1000        

Since Tajima and Dawson's proposal in 1979, the laser wakefield acceleration (LWFA) held the promise of shrinking km-scale conventional accelerators and radiation sources down to room-size machines.  This mechanism uses plasma waves generated via the interaction of intense ultrashort laser pulses with gas targets to capture and accelerate electrons to relativistic velocities. One of the main advantages of this concept lies in the fact that the accelerating field created by the plasma wave can be up to three orders of magnitude higher compared to the fields in conventional accelerators based on the radio-frequency technology. This enables obtaining electrons with energies up to several GeV with acceleration length only several centimeters long. Although the LWFA research has made significant progress in the last 40 years, generation of stable electron beams with high energy and charge is still a great challenge. One of the key moments in the acceleration process is the electron injection for which several injection schemes were proposed over the past several years. Within the scope of this project, we plan to use computer simulations to study the nonlinear physical processes present during the acceleration process as well as to determine the optimal injection method and parameters to obtain stable electron bunches reaching GeV energies in ELBA experiment at ELI Beamlines.


Researcher: Jun Terasaki            


Nuclear calculation for neutrinoless double-β decay       

Karolina CPU 42500      

The subject of this project is to establish the reliable prediction of nuclear matrix elements of the neutrinoless double-β decay, which is a hypothetical decay of a nucleus caused by a new type of neutrino. If this decay is observed, it proves the fundamental assumption of the theory beyond the standard model to explain the matter-antimatter imbalance in the universe. This explanation will be a historical success of science. Currently a few tens of experimental projects continue their operations day and night around the world to observe this extremely rare decay; the half-life is longer than 1026 years. The nuclear matrix element is indispensable because nuclei are used in these observation attempts.  The nuclear matrix element is also a key physical quantity to determine the mighty transformation matrix enabling calculations of transition rates of any reactions involving the neutrino. These calculations may give a tool to investigate inner part of stars in the future. It could contribute to the studies on how the materials necessary for the birth of lifeform were created in the universe.  The nuclear matrix element cannot be measured experimentally. Further, the accurate calculation of the nuclear matrix element is currently difficult because the decay probability is extremely small. I will break this difficulty by the state-of-the-art theoretical method, high-performance supercomputers, broad range of knowledge and innovative ideas.


Researcher: Pavel Jungwirth     


Ab-initio molecular dynamics simulation of an excess electron with alkali-metal counter-ion in liquid ammonia  

Barbora CPU 70600       

Alkali metals readily dissolve in anhydrous liquid ammonia, creating intensely colored solutions that are widely used as reducing agents in organic chemistry, for example in Birch reduction. These solutions are characterized by a fine blue color for dilute solutions and a copper/bronze metallic color for concentrated solutions. The process results from the alkali metal valence electron being detached from the metal and being dissolved by the solvent molecules.Using ab initio molecular dynamics simulations (AIMD), we can investigate the properties of these solutions at the molecular level. AIMD has been successful in describing the spatial structure and dynamics of excess electrons in ammonia solutions, but previous simulations have not included the alkali metal counter-cation.The proposed research aims to perform AIMD simulations of bulk systems containing one excess electron, one alkali metal cation, and solvent molecules to obtain a more accurate description of the structural and dynamical properties of low-concentration metal-ammonia solutions. We will use an improved methodology to accurately simulate these systems. This research is the first step towards a complete theoretical understanding of more concentrated metal-ammonia solutions.


Researcher: Erik Andris



Barbora CPU 14834;  Karolina CPU 37947           

Metal binding sites are ubiquitous feature of proteins and they often play crucial role in mediation of enzymatic chemical reactivity. The structures of these binding sites can be obtained from structural databases such as Protein Data Bank, or more specialized databases such as MetalPDB. However, these databases include only metal sites with experimentally determined structure. On the other hand, systematic screening of all possible binding sites can provide a way to remove the biological constraints from the equation and find a complementary way to design an “ideal” site. To that end, we plan to employ accurate Density Functional Theory (DFT) calculations to extensively screen “all possible” metal binding sites and calculate their properties. The resulting dataset will aid the design of selectively metal binding or functional metal-binding peptides, with applications in catalysis or bioremediation.


Researcher: Václav Bazgier        


Development of Disease-Modifying Therapy Agents for Sporadic Parkinsons Disease via Molecular Docking and Ligand Selection            

Karolina CPU 28300      

Parkinson's disease is a debilitating neurodegenerative disorder with no known cure. In this project, we aim to develop disease-modifying therapy agents for sporadic Parkinson's disease using a molecular docking approach. We will use ligands from various drug databases, including DrugBank and the Human Metabolome Database, and convert them into 3D structures for molecular docking. Multiple docking methods will be employed to select potential small molecules that target human cannabinoid receptors and butyrylcholinesterase. Our molecular docking workflow will help pre-select ligands and reduce the number of compounds that need to be tested, with the ultimate goal of identifying novel drugs for Parkinson's disease treatment.


Researcher: Anna Špačková      


Biologically significant protein pathways             

Karolina CPU 55405      

Protein tunnels are pathways within a protein structure that allow small molecules, such as substrates, products, and ligands, to access the protein's active site or to move between different regions of the protein. These tunnels are essential for protein function, as they regulate the flow of molecules into and out of the protein's interior. [1] The geometry, size, and physico-chemical properties of the tunnel determine which molecules can pass through it, and the tunnel's flexibility can influence the protein's conformational changes and allosteric regulation. [2]Recent advances in computational and experimental techniques have enabled the identification (MOLE[3, 4, 5], Caver[6]), characterization, and manipulation of protein tunnels, opening up new opportunities for drug discovery and enzyme engineering. Understanding the structure and dynamics of protein tunnels is a key challenge in structural biology and biochemistry. [7]Initial values of tunnels properties affect amount of detected tunnels by algorithm, and which tunnels will be found. In some cases algorithms can find out huge amount of tunnels in one protein structure, most of them are false positive. We want to find out which of tunnels that can be detected by MOLE algorithm[8] are biologically relevant, which of them are crucial in developing new drugs. Our hypothesis is create machine learning model base on known biological tunnels, that will be able to suggest about biologically importance of tunnels.


Researcher: Karel Tůma              


Numerical study of contact-less rebound of elastic solid in viscous incompressible fluid 

Karolina CPU 1700         

The interaction between fluids and solids is a fascinating topic that has important implications for a wide range of fields such as engineering, medicine, and geology, among others.One particularly interesting phenomenon is the rebound of elastic bodies in a viscous incompressible fluid without any contact. This is completely counterintuitive as one would expect that when an elastic ball is thrown in the swimming pool against the wall, the ball will touch the wall. However, if the perfectly incompressible fluid is assumed, it can be proven that the ball does not touch the wall, while it still bounces off due to the pressure singularity between the ball and the wall.This unexpected non-contact phenomenon is used to numerically implement fluid-structure interaction where a change in topology is undesirable.


Researcher: Libor Veis 


Organic molecules with inverted singlet-triplet gaps       

Karolina CPU 64000      

One of the recent proposals for design of new organic light-emitting diodes (OLEDs) is the principle of thermally activated delayed fluorescence (TADF), in which small energy difference between first excited singlet and triplet states allows for thermal upconversion of excited-state triplets to excited-state singlets. This way, the internal quantum efficiency of OLEDs can be improved from 25 % up to 100 %, since there is no emission from the excited triplets. Recently, some “exotic” molecules, which violates the Hund’s first rule and exhibit even the inverted singlet-triplet (INVEST) gap appeared. It was soon realized by computational chemists, that doubly excited configurations are primarily responsible for this effect and thus computationally cheap TDDFT methods are not able to capture it. In this project, we will benchmark the newly developed computational tools which are based on the adiabatic connection approach against the high-level coupled cluster methods with the aim of finding a cheap and reliable methodology for future high-throughput design of the INVEST molecules.


Researcher: Georg Zitzlsberger


Applications of Deep Neural Network based Urban Change Detection using Remote Sensing (part 3)

Karolina CPU 800;  Karolina GPU 600;  LUMI-C 400;  LUMI-G 900              

This work is the third part of a continuation of the HS BLENDED project (funded by ESA in 2020-2021) and OPEN-21-31/OPEN-25-24/OPEN-27-1 . We developed and trained a set of neural networks to monitor urban changes using a combination of SAR and optical observations with mission pairs ERS-1/2 & Landsat 5 TM (1991-2011), and Sentinel 1 & 2 (2017-now). Originally, the three sites Rotterdam, Liege and Limassol were used. We also applied transfer learning to help monitor Mariupol (Ukraine) during the years 2022/23 (and ongoing), and demonstrated its utility. In the third part, we would like to further optimize the trained networks and involved methods.


Researcher: Petr Hellinger         


Anisotropy of plasma turbulence at ion scales   

Karolina CPU 80000      

Turbulence is a ubiquitous phenomenon in space a laboratory plasmas. The nature does not like a concentration of energy on a narrow range of scales. The nonlinear coupling between scales leads to a spread of energy over a wide range of scales, the spectral energy densities exhibit a power-law behaviour, and the energy flows/cascades from large to small scales. On small scales the energy is dissipated and particle are heated. In usual fluids the heating is connected with the irreversible particle-particle collisions, however, in many space and astrophysical plasmas the collisions are too rare and the heating/energization proceeds via reversible channels. This phenomenon is not well understood: a wide range of turbulent fluctuating fields and weak collisions lead to complex properties of particle distributions and their energization. Moreover, space and astrophysical plasmas are magnetized and the magnetic field introduces an anisotropy, the system behaves very differently along and across the background magnetic field. To have a better insight to this problem numerical simulation are necessary. We propose to study plasma turbulence in the context of one well known example of weakly collisional turbulent system, in the solar wind (a flow of magnetized plasma from the Sun). We will use a three dimensional hybrid code (where electrons are assumed as a fluid but ions are treated as particles) to study properties of turbulent cascade towards ion scales and their energization.


Researcher: Tomas Jenicek        


Day-night image retrieval            

DGX-2 200;  Karolina GPU 3800

Image retrieval is an important and active area in computer vision. The task is to query-by-image in a large indexed collection of images, where the search is based purely on the image content. Applications include content- based browsing and search in large image collections, visual localization, image annotation, data collection for 3D reconstruction, and many others. The current state-of-the-art retrieval methods are based on Deep Learning Models which are trained on a GPU.In our project, we address image retrieval under significant illumination changes, such as between day and night images, where the appearance changes dramatically. This is currently an active field of research because of its applicability for real-world tasks such as autonomous driving.


Researcher: Jana Precechtelova             


High Flexibility to High Fidelity: Exploring the Complexity of Intrinsically Disordered Proteins with Machine Learning       

Barbora CPU 9500;  Karolina CPU 2900;  Karolina GPU 1600       

It has been well established that highly flexible systems in chemistry require special focus to account for the engorged conformation landscape. This investigation attempts to explore the complexity of intrinsically disordered proteins (IDPs) through established computational techniques. We will use molecular dynamics (MD) to simulate the conformational landscape of small-chain IDPs, and then implement replica-exchange molecular dynamics (REMD) to better sample the landscape. From there, we plan on implementing machine learning algorithms such as PCA, tSNE, and UMAP to reduce the dimensionality of the system and uncover key features of the molecule's behavior, then apply clustering algorithms to generate smaller, high-fidelity ensembles. These ensembles will be evaluated by comparing their radii of gyration and through chemical shift predictions to assess their ability to describe the conformational space enclosed in the trajectory. To further our research, we plan on using fragmentation, optimization, and ab-inito NMR calculations using ADMA fragmentation and DFT using a multiscale procedure established and refined in a previous publication. Our investigation intends to give a robust procedure for utilizing machine learning on highly flexible systems, parameterized for NMR chemical shifts. Additionally, the procedure can be implemented on other systems with a refined focus. We hope to gain insight into the complexity of IDPs and offer new strategies.


Researcher: Michal Hradiš         


The Next Generation of large scale Czech language models         

Karolina GPU  4800       

Large Language Models (LLM) pre-trained on vast amounts of text data have recently re-defined the State-of-the-Art in many tasks of natural language processing (NLP), information retrieval and human-computer interaction. LLMs are the foundations needed to build a wide range of applications and to keep up with advanced AI research. However, there are no truly large pretrained models available for Czech and other Central European languages. In this project, we plan to remedy the situation by training large models on a mixture of data from internet sources as well as, uniquely, on hundreds of thousands of printed sources in cooperation with main Czech libraries. Our primary aim is to create a range of models starting at hundreds of million parameters up to several billion parameters, and to release them to the general public for unrestricted use and for further development. We believe that these models will boost research in AI, NLP and digital humanities, and, perhaps more importantly, the models will allow Czech companies to build innovative language products and stay competitive in the age of ChatGPT.


Researcher: Rostislav Langer              


Investigation of stability-activity relationship of transition metal-doped defective graphene for various catalytic reactions            

Karolina CPU 52200;  LUMI-C 24900      

Single atom catalysis (SAC) is a type of heterogeneous catalysis where metal atoms are dispersed as individual atoms on the surface of a support material, rather than being clustered together as nanoparticles. SAC has gained significant attention in recent years due to its potential advantages over traditional nanoparticle catalysts, including increased catalytic activity and selectivity, improved stability and durability, reduced material usage and material costs, reduced environmental impact and/or enhanced control of catalytic processes. This research project aims to investigate the stability and efficiency of single-atom catalysts (SACs) anchored on two-dimensional (2D) materials. The goal is to understand the relationship between the stability of the active site and the effectiveness of catalytic reactions, specifically the hydrogen evolution reaction (HER) and the oxygen evolution/reduction reaction (OER/ORR). In this project, the density functional theory (DFT) will be used to calculate properties of 2D materials, using both finite models and infinite models with periodic boundary conditions. The Vienna Ab initio Simulation Package (VASP) and Gaussian software will be used for the calculations. The findings of this research will have significant implications for the development of more efficient and cost-effective catalysts for various industrial applications.


Researcher: Igor Szőke


Large pre-trained multilingual acoustic models for streaming speech recognition              

Karolina CPU 640;  Karolina GPU 5800  

The project's main goal is to significantly advance state-of-the-art in large pre-trained acoustic models suitable for streaming Automatic Speech Recognition (ASR).  We will perform advanced research in the domain of transformer models architecture adaptation to streaming scenarios (on-line, low-latency transcription). The actual well recognized models  (Transformers pre-trained with Wav2Vec 2.0, HuBERT, or the Whisper model) are trained to see the whole utterance (and they are looking into the future). The “streaming” application of such models is very computationally inefficient, as the input is incrementally extended and the “prefix” part of it is processed several times.  Our goal is to come up with an adapted architecture which will be tailored to the streaming scenario - computationally effective. As a by-product, we expect using acoustic data with open licensing so the model will be traceable and fulfill upcoming EU regulations. The trained models including the recipe will be published and freely accessible.


Researcher: Oldřich Plchot        


Automatic design of conversational models from observation of human-to-human conversation

Karolina CPU 1000;  Karolina GPU 5300

This work investigates approaches for the design of conversational models by observing human-to-human conversation with little supervision. The models will be in a human auditable form, possibly finite state graphs, to bring trust to such models from their users and ensure their faster adoption by the industry. These models can make conversational agents more fluent, speed up their development, and reduce costs. The models can also be used in new user interfaces and for speech analysis with a deeper understanding.  Also, we believe these models can enable much faster parallel development of conversational agents for multiple languages due to shared “language independent” semantic representation in a vector space. The work will use spoken and textual conversational data like MultiWOZ, DSTC11 Speech Aware Dialog System Technology Challenge data, newly collected spoken data, and large spoken and textual collections for pre-training. We want to study what the best content embeddings (for audio and text) are, how to convert content embeddings to some dialog states through autoencoders or clustering, and how to build and prune a dialog graph. Then how to automatically annotate the dialog graph by selecting characteristic sentences from the training data or using some natural language generation approach.


Researcher: Vojtěch Mrázek     


Optimization of hardware accelerators for machine learning       

Barbora GPU 400;  Karolina CPU 50;  Karolina GPU 1400

Machine learning models such as regression algorithms, decision trees, or neural networks have found applications not only in high-performance systems but also in low-power embedded devices. They are typically accelerated by specialized hardware accelerators implemented on these devices. The goal of this project is to optimize the machine learning models to be more efficient in inference. This optimization consists of the following steps: optimizing the representation of the weights to achieve a better throughput from memory, introducing approximate components to the computation, and modifying the architecture of the neural networks.


Researcher: Vitezslav Hanzal    



Karolina CPU 1400         

The L-39 is a well-known jet trainer that has been in service for decades. However, as technology advances and new materials become available, there is always room for improvement. This project aims to contribute to development of a new generation of the L-39, with a focus on incorporating modern manufacturing processes to the airplane structure.Manufacturer is exploring several different directions for further improvement. One of these is the greater use of composite materials, which offer numerous benefits over traditional metal alloys. Composite materials have a high strength-to-weight ratio, are corrosion-resistant, and can be molded into a wide range of shapes and sizes, which allows for greater design flexibility.Specific area of focus for our project is the redesign of the rear fuselage around the exit nozzle. This area was identified as not being optimally designed, and it's believed that a redesign using composite materials could offer significant improvements in terms of weight, noise damping, vibration resistance, and manufacturing costs.It is believed that by replacing the rear end with composite materials and optimizing the design, new levels of performance and efficiency can be achieved. The redesigned rear fuselage can create room for new features such as sensors, and will improve aerodynamics, reducing drag and lowering engine exit losses. Additionally, the redesigned jet exhaust mixing could lead to lower noise generation.


Researcher: Assia Benbihi         


The Prague Benchmark for Day-Night Visual Localisation              

Karolina CPU 50000;  Karolina GPU 3800             

This project aims at improving the benchmark for day-night visual localisation with an evaluation dataset that exhibits a new set of challenges. Research on visual localisation thrives with the availability of benchmark datasets that emulate the real-world challenges yet to be tackled. Such benchmarks allow researchers to investigate and evaluate novel methods in controlled environments and collect valuable experimental feedback. The benchmark dataset developed in this project completes the existing ones with novel challenges yet to be tackled such as the handling of dynamic objects (e.g. car, pedestrians), the robustness to occlusions, and the stability with respect to several artifacts induced by the “walking pedestrian scenario” such as motion blur, low-quality images, and light artifacts. The data collection, which is already completed, is taken from the viewpoint of a pedestrian over 14km in famous Prague districts among which Old Town Square, Charles Bridge, the Prague Castle, Vyšehrad, and this project aims at unifying the data under one global map. By providing geo-located images and 3D maps under such a novel scenario, this project spurs the research on day-night visual localisation toward new directions.


Researcher: Miroslav Kolos       


Calculations of exciton-phonon driven optical luminescence and nonlinear responses in two-dimensional semiconductors              

Barbora FAT 1778;  Karolina CPU 32000

An exciton is a bound electron-hole pair in a solid due to an attractive Coulombic force generated by, for example, photons. The binding energy determines the lifetime of the exciton, which is responsible for photo-luminescence in a material. Understanding the dynamics of excitonic motion has been a challenge for scientists for a long time, but the development of quantum field theory and the use of next-generation computer architectures have allowed for the accurate calculation of excitonic binding energies using the Bethe-Salpeter equation (BSE). The BSE provides direct access to a material's linear optical responses against all frequencies, and the resulting absorption and photoluminescence spectra contain valuable information about the fundamental optical gap, bright and dark excitons, excitonic binding energies, and interaction strength. These calculations are essential for optimizing and fine-tuning the features of various materials to enhance their luminosity, optical frequency conversion efficiency, and nonlinear optical mixing. Experimental conditions however include temperature effects. Including such effects in calculations is possible through exciton(electron)-phonon coupling which needs advanced techniques. In this work, we are using such techniques to predict accurate optical properties of a set of 2D materials.


Researcher: Jakub Šístek            


Vortical structures: efficient numerical simulation and advanced identification   

Karolina CPU 3800;  Karolina GPU 1300

The main aim of the project is performing high-resolution computational fluid dynamics simulations of prototype problems of incompressible viscous flows. The primary goal of these simulations is to generate high-resolution 3D data with vortical structures, which will be subsequently used for development of new methods for flow-field analysis and vortex identification and visualization. Unsteady flows considering very fine computational meshes are required for this purpose. The computations will contribute significantly to the project GAČR 23-06159S \Vortical structures: advanced identification and efficient numerical simulation\" funded by the Czech Science Foundation during 2023-2025.The computations will be performed using several solvers. First, an in-house parallel finite element solver based on multilevel domain decomposition will be used for low to moderate Reynolds number flows. Large Reynolds number turbulent flows will be solved by the Ansys Fluent/CFX software.A subsequent goal of the project is further development of the computational method and optimization of the open-source BDDCML library for large numbers of computer cores combined with GPU accelerators by NVIDIA and AMD.


Researcher: Dominik Legut       


Optimizing electron and heat transport properties of early-transition-metal nitrides for thermoelectric applications        

Barbora CPU 18000;  Barbora FAT 500;  Barbora GPU 2000;  DGX-2 1000;  Karolina CPU 30000;  Karolina FAT 500;  Karolina GPU 5500         

Although the early-transition-metal nitrides are well-known for their outstanding physical properties including high hardness, mechanical strength, high melting point, and electrical conductivity varying from metallic to semiconducting, they have attained renewed experimental and theoretical interest in recent years due to their potential thermoelectricity. Tailoring the electronic and phonon properties of these nitrides via doping with transition-metal and non-transition-metal impurities may open a route to develop high-efficiency thermoelectric devices. The main goal of the proposed theoretical research is to examine the effect of doping on the electronic, dynamic, and thermoelectric properties of pristine as well as oxygen-contaminated ScN and CrN. Our theoretical study is intended to support and explain the results of the ongoing experimental investigations performed within the Czech Science Foundation (GAČR) project No. 23-07228S, Novel thermoelectric, thermovoltaic, and phonoelectric heat conversion systems based on nitride semiconductors.


Researcher: Petr Strakos             


Research and Development of Libraries and Tools in the INFRA Lab III    

Barbora CPU 3500;  Barbora FAT 100;  Barbora GPU 500;  Barbora VIZ 200;  DGX-2 1000;  Karolina CPU  40000;  Karolina FAT 100;  Karolina GPU 6400;  Karolina VIZ 200;  LUMI-C 20400;  LUMI-G 6700

As members of the infrastructure research laboratory, our goal is to bring improvements and extensions of available tools that support the users of the IT4I clusters and their research. The key topics of our research are Energy efficiency, the development of scalable libraries for engineering applications MESIO and ESPRESO, and the development of Visualization tools. Developments in these areas will also serve to meet the objectives of the individual projects in which members of the laboratory are involved. The energy efficiency topic focuses on the measurement and tuning of the HPC applications in terms of possible energy savings. The team developing engineering applications will continue improving the capabilities of the ESPRESO library to solve a wider set of engineering applications as well as the capabilities of the MESIO library to pre-post processes of large-scale simulations. The visualization team will extend the possibilities of cluster rendering and further research the areas of image processing and visualization with help of a cluster.


Researcher: Jakub Klinkovský  


Development of LBM solver for enhancement of 4D PC-MRI data            

Karolina CPU 2100;  Karolina GPU 2700;  LUMI-C ;  LUMI-G 233 

The lattice Boltzmann method (LBM) is an efficient approach for simulating fluid flow, suitable for modern parallel architectures such as multicore CPUs and GPUs. It is based on solving the Boltzmann equation derived from statistical physics. The main advantage of LBM is its higher efficiency compared to other methods such as finite volume or finite element methods, enabling simulations to be performed at much higher resolutions to capture turbulent flow in more detail. 4D PC-MRI is a modern technique from magnetic resonance imaging (MRI) used to acquire 3D MRI data that evolves over time, particularly to measure the velocity of blood flow in the heart and large vessels. However, its application in clinical practice is challenging due to the need for advanced post-processing techniques to enhance the limited quality given by low spatial and temporal resolution, signal-to-noise ratio. The flow acquired by 4D PC-MRI can be reconstructed by an appropriate LBM simulation. This requires minimizing the difference between the high-resolution, noise-free velocity field obtained by LBM simulation and the measured 4D PC-MRI flow, a problem known as PDE-constrained minimization, usually solved by the gradient descent method. Solving this problem involves many LBM simulations, making it highly computationally demanding.This project aims to develop new methods for enhancing 4D PC-MRI using the in-house implementation of LBM based on the open source library TNL (https://tnl-project.org/). The implementation will be optimized to run efficiently on multicore CPUs and GPUs by NVIDIA and AMD.


Researcher: David Antoš            


High Performance Language Technologies (HPLT)            

Karolina CPU 100000;  Karolina GPU 8000;  LUMI-G 111000       

Large language models (LLMs) are at the forefront of enormous progress in natural language processing and AI. Learning from vast amounts of unlabeled text enables them to generalize in a data efficient way to support various operations with natural language.Pre-trained language models and machine translation models are in regular use by search engines, making recommendations, classifying speech and documents, and many similar applications. These models are deployed by a variety of enterprises. On the other hand, training very large language models is in hands of several large American and Chinese companies. This gives their modeling decisions exceptional power. Those models do not encode values like multilinguality, FAIR principles, minimizing bias, and energy efficiency.The goal of the project is to overcome those limitations, training open language models for over 50 languages. The project will ingest over 7 PB of archived web pages, parallel corpora, and other sources. The data will be cleaned and models trained, thus building the largest collection of reproducible language and translation models, keeping track of how the models have been built, ensuring highest standards of open science, reproducibility and transparency.


Researcher: Sergiu Arapan         


A spin-spiral study of the dependence of magnetic interaction on layers width at the Co/Pt interface

Karolina CPU 9200;  Karolina FAT 400;  Karolina GPU 3900;  LUMI-C 8800             

The occurrence of a specific spin textures is determined by the competition of various energy contributions that describe the spin interaction. In magnetic thin films with broken inversion symmetry the interplay of the symmetric Heisenberg exchange interaction and the asymmetric Dzyaloshinskii–Moriya interaction leads to the formation of spin spirals and skyrmions. This interplay is determined by the nature of the interface and can be tuned by the width of the films. In this project we will perform a computational study of the dependence of the parameters of magnetic interaction at the Co/Pt interface on the thickness of films using a spin-spiral method approach.


Researcher: Jindřich Libovický 


Pixel-token Hybrid Language Models     

Karolina GPU 18500      

Recent advances in natural language processing (NLP) have significantly impacted our digital lives, yet their benefits remain predominantly accessible to speakers of a select few high-resource languages and primarily those using the Latin script. Lower-resource and endangered languages, which often use non-Latin scripts, are left behind in this development. Representing text as pixels, where the alphabet is no longer a bottleneck, enables a pretrained language model to share information in the form of visual similarity across all languages [1]. While pixel-based language modelling is a promising research direction, especially for lower-resource NLP, the pioneering work is currently limited by the model’s inherent lower sample efficiency compared to tokenizer-based pretrained language models of the same size. To address this shortcoming, this project will explore additional pretraining objectives along with architectural choices that allow fusing the idea of true open vocabulary when operating over a visual representation of text with well-researched subword-based language models [2].The goal of this project is to train a hybrid language model that can perform equally well for high- and lower-resource languages. The resulting model will still support any written language that can be typeset on a modern computer without loss of information, but with a reinforced understanding of sentence structure and semantics.


Researcher: Ondřej Vysocký     


Energy efficiency audits of the community HPC codes in material science within the EuroHPC project Centre of Excellence MaX phase 3               

Barbora CPU 20000;  Karolina CPU 20000;  Karolina GPU 6900  

One of the main goals of the EuroHPC Center of Excellence MAX (phase 3) is continuously evaluating and optimizing performance and energy consumption of parallel applications developed under the MAX project – BigDFT, FLEUR, Quantum Espresso, SIESTA, YAMBO. These codes are the most successful and most widely used open-source, community codes in quantum simulations of materials, that should be prepared for the exascale and beyond. The goal of this computational resources project is to deliver the audits.


Researcher: Pierre Koleják        


Enhancement of terahertz spintronic emitters   

Karolina CPU 800;  Karolina FAT 40;  Karolina GPU 100;  LUMI-C 100       

Terahertz technologies find many applications across medical imaging, security revealing of drugs, explosives and weapons, ultrafast telecommunications, and quality diagnostics of integrated micro-circuits, food and manufacturing. Nowadays, the main barrier is the acquisition cost of such technologies for widespread application. Although many terahertz sources were already developed, the recent progress in terahertz spintronics changes the play rules due to high versatility and low-cost fabrication.  However, the efficiency of the spintronic device is lower than standard sources.  We simulate spintronic devices integrated with structures, which can drastically enhance their performance. Here, we apply the photonic and plasmonic structures, which are designed to maximize the optical-to-terahertz conversation by modifying field distribution within the spintronic layers.


Researcher: Luboš Šmídl            


Acoustic and multimodal transformers for speech recognition and denoising      

Karolina GPU 7400         

The goal of the project is to pre-train a large SpeechT5 model and perform fine-tuning for several tasks, speech recognition and speech signal denoising. Unified-modal SpeechT5 framework is based on the encoder-decoder pre-training for self-supervised speech/text representation learning. The SpeechT5 framework consists of a shared encoder-decoder network and six modal-specific (speech/text) pre/post-nets. The training of such large neural networks is based on and is possible only due to a huge amount of unlabeled data together with the stunning computing power of modern GPU clusters. The aim of the project is to use the power of a node with 8 A100 GPU for training acoustic transformers for speech recognition and speech enhancement (denoising). The results will be compared with current models based on wav2vec representation and publicly available models such as Whisper or XLSR.


Researcher: Petr Slavicek           


Time-dependent simulations for time-resolved electronic spectroscopies             

Karolina CPU 3500;  Karolina GPU 2300

This project aims at developing computational tools for the interpretation of dynamical processes revealed by time-resolved photoemission experiments. Present-day laser technologies allow for real-time probing of molecular motion on a femtosecond. This has transformed our understanding of bond formation and breaking, chemical reactions, charge migration, and ionization. The interpretation of the measured data, however, heavily depends on the direct quantum simulations. The extension of the time-resolved spectroscopies to the VUV/XUV domain brings in new advantages such as site and element specificity.  We aim at developing new unorthodox tools for the ab initio simulations of time-resolved spectroscopies. The effort will include developing and testing new approaches for non-adiabatic dynamics of excited and ionized systems, their simulation in the liquid phase, and exploring the solvent effects on the transient structures. We will further focus on the time-resolved processes in the core electron region and time-resolved Auger spectroscopy.


Researcher: Martin Dočekal      


Multi-Document Summarization From Scientific Literature          

Karolina GPU 4000;  LUMI-G 4200          

Never in human history were so many scientific papers published as in our time. This fact brings pressure on scholars to stay up to date in their own highly specific field and makes it really difficult to follow other relevant works.In the scientific community, it is a common practice to write overview articles called reviews or surveys that summarizes the literature on a specific topic and helps scholars to catch up on new knowledge in the field. A special type of overview could also be written as a part of a paper, in a dedicated section, that summarizes related work to the given article. Such a section will help the reader to comprehend the paper in a broader context.Both of these types of summaries require a significant amount of human labor. However, thanks to advances in machine learning, it is possible to develop models that will help scholars with that work.Our project will focus on the challenging task of multi-document summarization from very long texts. We will develop models that can select relevant pieces of information from multiple scientific papers to create a summary on demand. A significant amount of data is needed to train such models. For this reason, we will create datasets that we will make public. We believe that this kind of tool would be beneficial for scholars in their everyday work.


Researcher: Jonas Kulhanek     


Unified 3D Scene Representation            

Karolina GPU 5500         

3D scene representations, or 3D maps, are an essential component of a wide range of intelligent systems, such as self-driving cars, robots, or virtual reality. A fundamental limitation of the current approaches is, however, that they are designed for a specific sensor setup which makes them difficult to share between applications. The goal of this thesis project is thus to develop a unified map representation. One possibility is to represent the scene via an implicit neural function, i.e., a function that takes a 3D point as input and outputs density and colour value. This type of data structure has been shown to be able to model detailed scene geometry with high fidelity and can be trained from any 3D data, as well as raw sensor measurements such as images. However, unlike current neural radiance field approaches, which are optimised per scene, we aim to be able to estimate or initialise such a data structure quickly for new scenes. This could be enabled by building a database of 3D geometry parts which can be queried efficiently. We hope that our 3D scene representation will bridge the barriers between different modalities and will enable large-scale applications of systems which would otherwise require difficult-to-obtain data.


Researcher: Pavel Hobza            


Covalent Dative Bonding, Ionic, H-Bonding, and Charge Transfer Complexes:       

Barbora CPU 30000;  Barbora FAT 200;  Barbora GPU 1000;  Karolina CPU 70100;  Karolina FAT 800;  Karolina GPU 29700;  LUMI-C 16700;  LUMI-G  31200

The objective of the project is to enhance comprehension of the impact of solvent polarity on the stability of both covalent dative, ionic and non-covalent complexes via computational methods. The computational studies will be closely coordinated with experimental partners who employ cutting-edge experimental techniques. This collaborative effort has significant potential to advance understanding of the solvent's effect on complex stability, which could have broad practical applications across multiple fields. Using selected DFT functionals, the project will conduct to evaluate the electronic and optical properties of various complexes. The studies will employ both implicit and explicit solvent environments.


Researcher: Martin Zelený        


Ab initio design of compositionally complex ceramic oxides         

Barbora CPU 10000;  Karolina CPU 61440;  LUMI-C 33900           

Compositionally-complex ceramics (CCCs) are believed to be one level up from high-entropy ceramics (HECs) in their material properties due to their non-equimolar metal ions ratio, compared to equimolar ratio dedicated to HECs. The HECs exhibits the unique Li+ super ionic conductivity, enhanced dielectric behavior as well as other excellent physical properties. We propose that their CCC oxides should have unprecedently different material properties than nowadays materials including HEC oxides. In the proposed project, we employ ab initio calculation based on density functional theory to study material properties like structural stability, ions mobility and dielectric properties of selected CCCs derived from HEC oxides with rock salt (Me0.2Co0.2Mg0.2Ni0.2Zn0.2)O and perovskite BaxSr1-x((Zr0.94Y0.06) 0.2Sn0.2Ti0.2Hf0.2Me0.2)O3−y structures.


Researcher: Štěpán Sklenák      


Splitting dinitrogen        

Karolina CPU 29000;  LUMI-C 6500        

Zeolite based catalysts are the most important industrial catalysts. Zeolites are crystalline microporous aluminosilicates with a unique microporous nature, where the shape and size of a particular pore system exerts a steric influence on the reaction, controlling the access of reactants and products. Periodic DFT methods permit investigations of properties of zeolite-based catalysts which are needed for their fine-tuning. DFT calculations are complementary to experimental examinations and together they can provide more complex knowledge of the properties of the studied catalysts and the reactions they catalyze.


Researcher: David Barina           


Computational Verification of the Collatz problem          

Karolina CPU 260100;  LUMI-G 80000   

One of the most famous problems in mathematics that remains unsolved is the Collatz conjecture, which asserts that, for arbitrary positive integer n, a sequence defined by repeatedly applying the function f(n) = 3n+1 if n is odd, or f(n) = n/2 if n is even will always converge to the cycle passing through the number 1. The terms of such sequence typically rise and fall repeatedly, oscillate wildly, and grow at a dizzying pace. The conjecture has never been proven. There are however experimental evidence and heuristic arguments that support it. As of 2023, the conjecture has been checked by computer for all starting values up to 2^68 [Barina2021]. Our project aims to extend the computational records to higher values, and possibly to find some interesting results (numbers with extraordinary expansion factors, climbing to extremely high values, an iteration length that is strongly outside the expected distribution, a counterexample, etc.)


Researcher: Radim Špetlík         


Temporally-Consistent Style Transfer with Denoising Diffusion Probabilistic Models         

Karolina GPU 7300         

Denoising Diffusion Probabilistic Models (DDPM) are generally regarded as the major research direction in the generative image modelling. However, image-guided style transfer with DDPM remains an unexplored research area due to its vast computational requirements and severe ill-posedness. Given a photograph and stylised artwork, the task of image-guided style transfer is to mimic distinctive visual features of the style exemplar while producing an image with content similar to the input photo. Furthermore, while existing methods are commonly designed to work only with a single image, a handful of style-transfer approaches tackle the problem of video-sequence style transfer. Here, the task is even more challenging. Given a single stylised image, the style of the image is transferred to the whole image sequence.


Researcher: Ladislav Bartoš       


Mechanism and energetics of peptide translocation across phospholipid membranes    

LUMI-C 10200;  LUMI-G 19000 

In this project, we aim to address the growing threat of antibiotic-resistant bacteria and the need for new treatments. Potential treatment options include utilizing antimicrobial or cell-penetrating peptides that possess the capabilities to permeate bacterial membranes and can either directly kill bacteria or enhance the transport of other molecules, such as drugs, into bacteria. By studying the properties of amphiphilic peptides and their ability to penetrate cell membranes, we hope to better understand the biophysical and biochemical principles that govern their activity. Using detailed atomistic molecular dynamics simulations with free energy calculations, we will investigate the ability of specific peptide sequences to translocate through membranes. Furthermore, we will study the effect of transmembrane proteins on the translocation process. The computational results will be subsequently validated by experiments. Through this research, we hope to pave the way for the development of more potent and effective treatments for antibiotic-resistant bacteria.


Researcher: Tomáš Karásek       


Reliability-based approach to drive inspections of steel bridges as a way to ensure their safety and lifespan         

Karolina CPU 2200         

Successful solution of the project would lead to an improvement of one essential part of ensuring safety and longevity of such critical parts of transport infrastructure such the bridges are. It is expected that the proposed approach, with a necessary adjustment, will be usable for bridges not only in the European union but also in other countries.


Researcher: Peter Huszár           


Impact of urban surfaces on regional scale rain and cloud patterns          

Karolina CPU 5000         

Urban environments not only affect the warming rate over cities through the so-called urban heat island (UHI) but also induce changes in other relevant meteorological variables. This study aims to evaluate the impact that different combinations of urban, microphysics, and convective parameterizations have on a number of meteorological variables, including temperature, wind and those related to cloud/rain microphysics. Simulations are performed using the WRF model with a domain at 9km horizontal resolution centered over Prague covering central Europe for the 2008-2017 period. The length of the period ensures statistically robust results. The urban schemes include bulk, the single-layer urban canopy model (SLUCM), and the multilayer urban models (BEP-BEM) with a building energy model including anthropogenic heat due to air conditioning. We further consider another scenario in which the urban land use is fully replaced by a rural one (NOURBAN). Besides, we also used three different options for the microphysics and convective schemes. We expect to show that the inclusion of urban canopy schemes leads to an increase in temperature and a decrease in wind speed as well as to changes in other relevant meteorological values such as cloudiness and precipitation, which also depend on the microphysics and convective scheme used.