Researcher: Georg Zitzlsberger        


Applications of Deep Neural Network based Urban Change Detection using Remote Sensing (part 2)

Barbora CPU-100; Barbora GPU-100; Karolina CPU-900; Karolina GPU-700, LUMI-C 1000, LUMI-G 1000         

This work is the second part of a continuation of the HS BLENDED [1] project (funded by ESA in 2020-2021) and OPEN-21-31/OPEN-25-24. 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 would like to continue finding further uses and tailor the trained networks and involved methods. To extend the use, we consider a documentation of war related changes in Ukraine over 2022 and 2023, as well as monitoring other areas.


Researcher: Michael Bakker              


Utilizing Machine Learning Algorithms and Fragmentation Techniques to Probe Intrinsically Disordered Proteins Phase Space of NMR Chemical Shifts   

Barbora CPU-2800; Barbora GPU-3000; Karolina CPU-6300; Karolina GPU-3200

In the last 20 years, great interest has emerged for the structural characterization of intrinsically disordered proteins. These proteins have been connected with many diseases and abnormalities and understanding their structural features plays a major role in connecting-the-dots for these disorders. Molecular dynamics (MD) trajectories are necessary in order to properly describe the inherent disorder in these proteins, however current chemical shift prediction software are insufficient as they do not properly describe the electronic distribution within a system. For the purpose of accurately computing chemical shifts, we propose a four part combined approach, using machine learning algorithms known as cluster analysis to sort and distinguish distinct conformations and weight them based on their populations. Next we fragment the protein files using a technique developed by Thomas Exner known as the adjustable-density matric assembler (ADMA). Finally we optimized the structures and computed the chemical shifts using density functional theory. In practice, we have already had significant success in this endeavor, with some preliminary results published recently in Physical Chemistry and Chemical Physics (Physical Chemistry Chemical Physics, 2022, DOI: 10.1039/D2CP01638A). The results collected are promising, showing great alignment with experimentally obtained data at a fraction of the amount of the time expended for computation.


Researcher: Ctirad Červinka              


Fragment-based ab initio refinement of lattice energies of crystalline ionic liquids as archetypal low-volatile molecular materials            

Karolina CPU-82000; Karolina FAT-1300; Karolina VIZ-40, LUMI-C 10000             

Ionic liquids represent archetypal low-volatile materials, the crystals of which are an intermediate between molecular and ionic crystals. Chemical entities constituting these crystals concurrently exhibit a polyatomic nature and a permanent ionicity. Semi-organic character of the typical ions present in the ionic liquids interact via a mixture of electrostatic, induction, and dispersion forces, and optionally also hydrogen bonding. Such a complex interplay of various non-covalent interactions calls for a high-level ab initio treatment to reach a sufficient accuracy of the lattice energies of these crystals. A physically sound way of a fragment-based ab initio refinement, building the total lattice energy as a sum of monomer contributions, pair interaction energies, higher-body interactions, and long-range corrections could potentially yield highly accurate sublimation data for ionic liquids. Reaching a subchemical accuracy, with the computational error ranging to 1–2 kilojoules per mole would challenge even the experimental state of the art in this low-volatile region. This proposal targets tuning of the fragment-based scheme, optimizing the combination of a high-tier method for treatment of interaction of proximate ions and a low-tier method for screening of more distant or higher-body interactions.


Researcher: Martin Hurta    


Evolutionary Design of Energy-Efficient Movement Disorders Classifiers

Karolina CPU-2000        

The evolutionary design (ED) takes inspiration from biology to develop new methods of computer programming that deliver solutions to unusual and challenging problems. This research focuses on the part of the ED called cartesian genetic programming (CGP) and its use in solving real-world challenges connected to the classification of movement abnormalities, particularly Parkinson's disease (PD).

PD's symptoms treatment can cause motor abnormalities called levodopa-induced dyskinesia (LID), which worsen the patient's quality of life. Hardware-oriented LID-classifier could be implemented directly in a home-wearable device and thus help clinicians easily adjust a patient's medication dosage and find a tolerable balance between the benefits and side effects.

This research builds on our previous work in the automated design of hardware-oriented LID-classifier. We aim to further improve the proposed method by employing other advanced techniques (e.g. multi-objective design). According to this case study, we plan to introduce and evaluate a methodology for the automated design of hardware-oriented implementation of feature extractors and classifiers of other movement disorders accompanying neurological conditions.


Researcher: Milan Červenka


Acoustic black holes for absorption of air-borne sound in ducts

Barbora CPU-10000; Barbora VIZ-40     

An acoustic black hole (ABH) was theoretically proposed 20 years ago as a retarding structure consisting of a piece of a fluid-filled tapered duct with a compliant wall, serving as an anechoic termination for a waveguide. The duct tapering, together with its variable wall admittance, can cause the acoustic wave slow-down, theoretically, to zero speed. This way, the wave needs an infinite time to reach the end of the structure, and thus, it cannot reflect. In all practical realizations of ABHs, the zero wave speed is never reached; however, due to the wave slow-down, its amplitude rapidly increases, and a great amount of acoustic energy can be absorbed thanks to weak but omnipresent thermos-viscous dissipation processes. This mechanism also leads to a significant reduction of wave reflection. ABHs seem to represent an attractive counterpart of classical means of reduction of sound propagating in ducts, which usually employ porous/fibrous materials, the use of which is in many cases problematic.

Within this project, we intend to conduct a detailed study of the absorbing properties of ABHs. First, we will employ multi-physical finite element approach to assess the role of individual processes in the acoustic energy absorption in an ABHs. Second, we will propose a simplified, computationally efficient model of an ABH. Third, we will employ techniques of evolutionary computing to optimize the performance of ABHs under given physical constraints to understand their performance limits.


Researcher: David Gregocki


Transition of electron beams between vacuum and plasma in the external injection into a laser wakefield accelerator    

Karolina CPU-8400; Karolina VIZ-40       

The largest accelerating fields that are achievable by conventional particle accelerators are in the order of 100 MV/m. This limit is inherent to the way conventional accelerators work. The state-of-art generation of high-intensity laser pulses enables the construction of a particle accelerator with very high field gradients. Such accelerators are compact and have very efficient acceleration compared to the conventional types of accelerators. Laser wakefield acceleration (LWFA), namely, has proven its potential of accelerating electron bunches, reaching accelerating gradients up to hundreds of GV/m. Nonetheless, current experiments demonstrating LWFA suffer from the lack of control or large shot-to-shot fluctuations. This is partially because the injection of electrons into the accelerating structure and their subsequent acceleration are inherently coupled in all of the traditional schemes. These schemes are based on nonlinear dynamics, such as wavebreaking, and injecting electrons from the plasma background itself. Therefore, there is a growing interest in exploring the mechanism where electron bunches are injected from a different accelerator. Thereby, the injection can be controlled more effectively. To address this problem, several 3D numerical “particle-in-cell” simulations with external injection of an electron beam are aimed to be performed within the frame of this project. In particular, different plasma density profiles used for the acceleration will be analysed.


Researcher: Simona Sajbanova    


Investigation of phases from salt-cocrystal continuum ΔpKa area             

Karolina CPU-3000; Karolina VIZ-40       

Nowadays, the portfolio of solid forms of pharmaceutical molecules is very wide. Pharmaceutical salts are one of the forms that can be used for formulation of active components.  About 50% of drugs used today are salts. Cocrystals are the other most dynamically developing pharmaceutical forms. The difference between salt and cocrystal is given only by the position of single hydrogen. The information about the exact position of this hydrogen is required in registration and patent documentation by regulation authorities. Since this difference between cocrystal and salt is so small, it is an essential task to develop new methods that could determine the exact position of responsible hydrogen. We are developing a computational method for such hydrogen position determination. The method does not require crystallographic data from high-quality monocrystal, and it work even with data from powder samples.


Researcher: Miroslav Medveď


Mechanistic Insights into Molecular Photoswitching via Quantum Chemistry Methods  

Karolina CPU-25390; Karolina VIZ-40    

In daily life, we are used to being able to switch ON/OFF a device depending on our needs. To reach such control at the microscopic scale is one of the holy grails of current material chemistry. Molecular photoswitches are in this respect of particular interest as the light allows for a non-invasive localized activation of, for example, a drug only at the specific site and time. Despite a wide palette of existing switches, most of them are far from ideal for biomedical applications, as they usually require harmful UV light for their operation, do not exclusively switch from one form to the other, and/or do not function under physiological conditions. The optimization of existing switches and the design of new scaffolds exhibiting light-sensitive structural changes are thus highly appealing. Clearly, such efforts require a multidisciplinary approach. Rapid development of ultrafast spectroscopies has enabled investigation of photochemical transformations with unprecedently deep insights into the radiative and non-radiative processes. Simultaneously, new theoretical approaches offer crucial information about the electronic structure and the nature and dynamics of excited states. Combined theoretical and experimental efforts offer not only consistent interpretation of the experimental data but also open a path to rational design of new photoactive materials with attractive properties.

The main objective of the project is to elucidate individual steps of photoswitching mechanisms for experimentally interesting photo-responsive systems using state-of-the-art quantum chemistry methods and -based on the revealed structure-activity relationships- to design new chromophore structures with desirable photoswitching characteristics.


Researcher: Jan Rezac           


Benchmarking non-covalent interactions and SQM method development            

Karolina CPU-15000; Karolina GPU-1600             

Our aim is to develop novel approximate quantum-mechanical computational chemistry methods applicable to very large systems such as biomolecules and complex materials. To improve the accuracy in properties crucial for these applications, special attention has to be paid to non-covalent interactions. This project combines accurate benchmark calculations needed to complete our database of reference data with method development, namely combining semiempirical quantum-mechanical methods with machine learning.

The benchmarking is an extension of my Non-Covaleint Interactions Atlas project (www.nciatlas.org), a large publicly available database of state-of-the-art benchmark data on intermolecular interactions, which will be extended with another dataset of ionic species.

In the method development part, we will re-train the already available modern machine learning potential TorchMD-NET as a correction for semiempirical quantum chemistry.


Researcher: Marek Pecha   


ml4py – distributed machine learning tools (dev/stage) III           

Barbora CPU-1100; Barbora FAT-10; Barbora GPU-200; DGX-2-200; Karolina CPU-3400; Karolina FAT-10; Karolina GPU-200,
LUMI-C 5000, LUMI-G 400       

This project continues with the implementation and benchmarking of emerging tools for distributed machine learning called ml4py. Unlike standardly used machine learning kits, ml4py implements a mechanism for training predictors over cluster in its runtime using Message Passing Interface (MPI). Thus, no other frameworks such as Horovod or orchestration based on running multiple container instances of an application are required. Accelerators such as graphic cards are utilized using the OpenCL and CUDA technologies through the ml4py runtime.

Current research related to machine learning commonly prefers deep learning techniques using (deep) neural networks. An associated training phase could be considered as a stochastic process that converges to an expected model. Our previous research on machine learning shows that determinism allows us to design training pipelines directly and explain the qualities of an attained learning model and numerical behaviour of the underlying solver more straightforwardly. However, these approaches are expensive in the sense of computational cost, and they can efficiently solve up to medium scale tasks by the state-of-the-art. Since ml4py is designed for massively parallel approaches, determinism can be incorporated into training pipelines for big data analysis. To speed the deterministic training phase up, stochastic optimization methods can be exploited. Basically, we find a good candidate for a solution using stochastic approaches expeditiously, and this candidate is used as a hint for deterministic and parallel predictor training subsequently. It helps to attain high-quality explainable models within a reasonable time.

The ml4py framework is tested on datasets ranging from remote sensing, geoscience to computer vision.


Researcher: Pavlína Pokorná        


Characterizing transitory ensembles on the G-quadruplex folding landscape       

Karolina CPU-1600; Karolina GPU-12000             

G-quadruplexes (G4s) are non-canonical nucleic acid structures formed by guanine-rich sequences. Localized dominantly in telomeres, they contribute to gene expression control, genome replication, maintenance, and are implicated in cancer development. Despite extensive research efforts, the G4 folding mechanism is still not well understood since its observation on the atomic level is below the current experimental resolution. Molecular dynamics simulations are a modeling tool complementary to experiments that allow us to study such processes. Due to a large number of possible metastable states on the G4 folding landscape, simulation studies of G4 folding are computationally demanding, necessitating the use of high-performing computers and enhanced sampling methods. We aim to characterize intermediates on G4 folding landscapes and transition pathways among them. This will help us better understand the folding process and its determinants. Besides, we also aim to evaluate performance of the applied enhanced sampling methods in order to improve methodology development.


Researcher: Diana Csontosová    


Magnetic Excitations in Ruthenates with Strong Spin-Orbit Coupling       

Karolina CPU-14100; Karolina VIZ-40     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: Jakub Planer    


Tunable Charge Injection Layers for Organic Semiconductors     

Karolina CPU-8000; Karolina VIZ-40       

In recent time, light weight, flexibility, low-cost and tunability of organic semiconductors (OS) have drawn the attention of a broad scientific community and semiconductor industry, making them usable in many applications such as active-matrix organic light-emitting diodes (OLEDs) in smartphones, organic field-effect transistors (OFETs), solar cells, memories, photoswitches and sensors. Therefore, tremendous effort has been made to improve OS functionality. The development has reached the point where the contact resistance with the metal electrode, especially in OFETs, is becoming the main parameter limiting the performance of OS-based devices. Recent research in the Host Group has revealed that molecular monolayers (MoMs) composed of aromatic carboxylic molecules can be deprotonated on a silver substrate in a controlled manner. This finding opens a way towards developing entirely new highly efficient Charge Injection Layers for organic semiconductors.

In this project I will employ state-of-the-art computational methods to characterize interfaces between Ag substrate and MoMs composed of aromatic carboxylic acid molecules, and design tunable CILs based on the controlled deprotonation. If successful, our collaboration can help the OFETs finally reach their full market potential.


Researcher: Michael Matějka           


Climate variability and atmosphere-glacier interaction on James Ross Island, Antarctica

Karolina CPU-20000; Karolina VIZ-40, LUMI-C 5000    

Our simulations with IT4I will contribute to climate research in the Antarctic Peninsula region where the J. G. Mendel Czech Antarctic Station is located. The evolution of climate in 2000 – 2100 will be simulated using WRF, a state-of-the-art atmospheric model. The results will improve our knowledge about climate change and climate variability in this region known by large temperature sensitivity of glaciers. In the second part of our project, glacier-climate interaction will be studied in a greater detail with very high-resolution simulation of atmospheric and snow cover processes. The output will be validated with in-situ observations and will help us to better understand processes on Antarctic glaciers like snow accumulation and intense melt events.


Researcher: Michal Schmid


Aeroacoustics simulations of a cooling fan and heat exchanger  

Karolina CPU-5500; Karolina VIZ-40       

Modern construction equipment like excavators must meet demanding environmental and safety targets. These targets could be represented for example by emission and noise limits or by electric vehicles’ battery cooling. At the same time, the vehicle must satisfy other functional requirements such as occupants’ comfort, fuel consumption, etc. One of the critical components, even within modern electric vehicles, is a rotating fan and a heat exchanger. It is necessary to produce sufficient air flow via a rotating fan to cool down batteries, engines, or other fluids in heat exchangers (e.g. hydraulic oil, AC refrigerant). However, a rotating fan and airflow generate a noise, which is unfavorable. All these challenging design requirements lead to higher demands for vehicle development, where so-called Computational Aided Engineering (CAE) is an essential part. Hence, accurate and fast mathematical modeling of airflow generated by the rotating fan and flow through a heat exchanger as well as aero-acoustics prediction is crucial. The research project is focused on aeroacoustics theory, mathematical modeling, and industrial application in cooperation with Doosan Bobcat s.r.o.


Researcher: Jemal Yimer Damte      


Triboelectric Nanogenerators for Green Energy Production (TENGs)       

Barbora CPU-9800; Barbora VIZ-40; Karolina CPU-11600; Karolina VIZ-40            

Most of the world’s population uses fossil fuel such as coal, oil and natural gas which cause emission of carbon dioxide and green house gases, and led to the increase of global warming.  In parallel, the energy crisis across the globe is escalating,  since extraction of oil is falling while the rate of fuel consumption is increasing. Accordingly, harvesting energy which is environmental friendly as well as satisfying the energy depletion is the global concern. Due to the advancement of nanotechnology, extensive researches have been studied to find alternative energy sources. Triboelectric nanogenerators (TENGs) have attracted widespread attention for energy harvesting. They convert mechanical energy into electricity, produce high power density and have low cost fabrication.  An ideal solution would be to develop hybrid photo-triboelectric devices, which owns both photon harvesting and triboelectric capabilities for electricity generation. In this respect, two dimensional (2D) transition metal dichalcogenides have received significant attention thanks to their close features with graphene, demonstrating promising electrical, optical and mechanical properties. In this work, we plan to design PV-TENG materials that exhibit both photovoltaic and triboelectric current generation by using density functional theory simulations, with the ultimate goal to suggest possible device setups.


Researcher: Vojtech Kostal 


Aqueous solution/air interfaces approached by the first-principles molecular dynamics

Karolina CPU-48700; Karolina VIZ-40    

Interfaces between aqueous solutions and air are widespread and play an important role in, e.g., uptake and release of gases from the sea water droplets that has implications to atmospheric and environmental chemistry. Number of experimental studies was performed elaborating on structure and dynamics of water and dissolved salts in contact with air. However,  they often lack a detailed information about the underlying molecular structure. In this direction, molecular simulations are of a great advantage thanks to their unprecedented temporal and spatial resolution and thus it is not only a complement to the experiments but also has predictive abilities. It was shown that behaviour of the dissolved ions close to the liquid/air boundary cannot be recovered by simple nonpolarizable models. So far, theoretical studies were limited to polarizable, yet still empirical models due to the computational costs. In this project, we propose to approach the aqueous electrolyte solutions in contact with the air by molecular dynamics based on high-level ab initio potentials including explicit electronic structure. This will provide us with an insight into the structure of these interfaces without relying on empirical parameterizations with all their potential pitfalls.


Researcher: David Zihala     


Biology and resistance of minimal residual disease in multiple myeloma

Karolina CPU-500; Karolina VIZ-40         

Multiple myeloma (MM) is a malignancy of plasma cells, terminally differentiated B-cells, producing aberrant immunoglobulins and causing damage to organs, and ultimately leading to a patient’s death. The malignant phenotype is caused by changes at the genomic level, with many mutations being already described, however the complete picture of the genomic landscape and its relation to treatment is still missing, particularly due to unprecedented heterogeneity of tumor-specific mutations in cohorts of patients and also within tumor cell populations. Many recent genomic and transcriptomic studies in patients before treatment have deeply described the molecular profiles of tumor cells and identified a few important players for the prediction of disease course or sensitivity to treatment. Much less research has been made on patients in relapsed and treatment-resistant disease stages. These patients usually do not benefit from standard care and more efficient and better-tailored treatment is needed. The genomics studies describing the landscape of small mutations, structural variants, and deferentially expressed genes may help to understand drug resistance and predict potential drug targets. In our study, we focus on molecular characteristics of minimally residual disease in MM and consequent relapse samples to shed more light on the biology and clonal evolution of these treatment-resistant stages.


Researcher: Dana Nachtigallova 


Hydrogen production by photocatalytic water splitting over TiO2 surfaces: A Computational study

Barbora CPU-30000; Barbora FAT-500; Barbora GPU-1000; Karolina CPU-120000; Karolina FAT-400; Karolina GPU-20000,
LUMI-C 40000, LUMI-G 5000 

Photocatalytic water splitting is an attractive and sustainable way of harnessing solar energy that avoids intermittency problems. Despite the tremendous efforts, most existing photocatalysts suffer from low activity, narrow absorption range, and low solar energy conversion efficiency. This project aims to improve the understanding of water splitting on pristine or doped TiO2 semiconductors using computational approaches. These studies will be carried out in close collaboration with experimental partners. This collaboration has the potential to contribute to the realization of the hydrogen economy with great applicability in various fields and to identify the challenges that have to be addressed to move the field forward. The DFT calculations, employing selected DFT functionals, will be performed to calculate the electronic, optical, and mechanical properties of TiO2 polymorphs. These studies will be performed using crystal models that mimic experimental conditions. The cluster models generated from crystal structures will be studied with expensive single and multi-reference calculations.


Researcher: Ahmed Alasqalani         


Improving the thermal STAbility of Tungsten via Grain boundaries Egregation for Nuclear application (STATGEN)             

Barbora CPU-10100; Barbora VIZ-40; Karolina CPU-9760; Karolina VIZ-40            

Developing high-performance materials is the primary concern for realizing magnetic confinement fusion reactors. The plasma-facing materials (PFMs) work under extreme conditions, including high-energy neutron (14.1 MeV) bombardments, severe thermal loads (up to 20 MWm−2), and high fluences of high-flux (>1021 m−2 s−1) and low-energy (<100 eV) hydrogen and helium plasma irradiations. Tungsten (W) is a refractory metal with superior properties such as high melting temperature, high thermal conductivity, good erosion resistance, low vapor pressure, low swelling, and low tritium retention. These properties are appealing for applications as PFMs in future fusion facilities. However, high brittleness in several regimes, including low-temperature embrittlement (relatively high ductile-brittle transition temperature (DBTT)), irradiation embrittlement, and recrystallization embrittlement. Different strategies have been proposed to achieve higher strength levels, such as solid solution and grain boundary (GB) solute segregation. Here, the propensity of Zr segregation at different grain boundaries (GBs) will be investigated using DFT. Furthermore, MD cascade simulations will examine the influence of GBs structure on the point defects (radiation damage). 


Researcher: Pavlo Polishchuk           


Molecular modeling and de novo design in drug discovery          

Karolina CPU-4500; Karolina VIZ-40       

Development of new drugs is a multistage process. On early stages researchers try to indentify promising hits which can be further optimized in lead compounds and drug candidate. Chemoinformatics can contribute to these early stages of drug development by design of compounds with new scaffolds or optimize structures of previously established hit molecules. The developed automated tools can substantially simplify these tasks. Within this project we will continue our previous research on development of tubulin inhibitors of various classes, MARK4 inhibitors, inhibitors of zinc metalloprotease Zmp1 of Mycobacterium tuberculosis. Those compounds may be used for treatment of cancers, Alzheimer’s disease, and tuberculosis.

Recently ultra-large libraries became popular in drug discovery applications. These libraries are virtually enumerated from building blocks and reaction rules creating a large number of compounds with high synthetic accessibility. Searching in such databases is challenging due to their size. Within this project we will explore applicability of our developed approach for fast searching of promising molecules in large databases.


Researcher: Jiri Klimes         


Accuracy and precision for extended systems X

Barbora CPU-9300; Barbora VIZ-40; Karolina CPU-23500; Karolina VIZ-40            

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 Friak     


Defect engineering of new eco-friendly magnets based on non-magnetic oxides

Barbora CPU-46080; Karolina CPU-92160           

Many currently used magnets contain so-called rare-earth (RE) elements, such as neodymium (Nd) in the strongest known permanent magnet Nd2Fe14B. The mining and production of RE elements in China is very environmentally unfriendly, hazardous and unreliable due to global restrictions related to current fast changing geo-political situation. Development of new RE-free magnets is, therefore, a very active area of research worldwide. Interestingly and surprisingly, magnetism was experimentally observed even in semiconductors containing elements that are normally non-magnetic, such as thin films of HfO2 (see M. Venkatesan et al., Unexpected magnetism in a dielectric oxide, Nature 430 (2004) 630). These materials would open an unexpected and potentially very promising direction when designing new magnetic systems. The origin of the magnetism in non-magnetic oxides is, unfortunately, not fully understood and explained yet. It has been speculated that this phenomenon is closely related to surfaces defects. The proposed research aims at performing quantum-mechanical calculations of surfaces of TiO2 with defects, in which magnetism was also detected, in order to explain experimental data and shed a new light on this interesting phenomenon that can have numerous future applications.


Researcher: Prashant Dwivedi          


InvesTigation of High-velocity dUst iMpacts on Plasma-facing materials. (THUMP-II)      

Karolina CPU-16200; Karolina VIZ-40    

Future fusion reactors will need to consider the multifaceted safety and operational implications of dust in tokamaks [1-8]. Due to the vast range of dust velocities in fusion devices (from a few hundred m/s to over 1 km/s), energetic dust grains and wall materials may interact in a variety of ways. Low-energy dust particles can alter the characteristics of walls by depositing them. Strong dust-wall interactions can seriously harm Plasma Facing Material (PFMs), through sputtering, erosion, deformation, and other effects. In the proposed study, molecular dynamics (MD) simulations will be carried out to examine the interactions between tungsten (W) plasma-facing materials (PFMs) and (W) dust grain/projectile. The research will focus on W since this is the main candidate in different PFMs and divertor designs. To validate our simulations with in-progress experimental work, a wide range of dust velocities will be taken into consideration, from 0.1 km/s to 1 km/s. The goal of the project is to first investigate and understand the mechanical damage and structural modifications suffered by the wall material after the high-velocity impacts. Secondly, to develop an analytical model to explain the underlying mechanisms, and finally, to compare our model with experimental studies. Our findings will help to understand the experimental results and make predictions about dust-wall interactions in future fusion devices.


Researcher: Michael Komm


Particle-in-cell simulations of Langmuir probes in MAST-U tokamak        

Karolina CPU-18000; Karolina VIZ-40    

Research of plasma properties in candidate concepts of future thermonuclear reactors, such as tokamaks, is vitally dependent on correct interpretation of plasma diagnostics. This is especially challenging in the plasma edge, where the hot plasma particles interact with the plasma-facing components. This interaction has to regulated in order to prevent damage to the first wall of the future reactor. One of the most prolific diagnostics in this area are Langmuir probes, which serve to measure the plasma temperature and density. Within this project, we propose to simulate the particle collection of such probes in the MAST-U tokamak (UK), which is dedicated for studies of advanced strategies of power exhaust.


Researcher: Elliot Perviz      


First-principles Investigation into the Thermodynamic phase Stability of TMD-based heterostructures (FIRST-TMD)               

Barbora CPU-9360; Barbora VIZ-40; Karolina CPU-14300; Karolina VIZ-40

Transition metal dichalcogenides (TMDs) are perhaps the most promising substitute for graphene for building the next generation of low-dimensional functional materials. Due to their naturally layered and anisotropic structures with a vanishing coefficient of friction in High Vacuum, TMDs have found use in nanotribology as thin film lubricants. However, issues such as instability in dynamic and humid conditions create major challenges for their design, control and reliability. Investigations in recent years have successfully explored the doping and multi-layer design of TMDs to mitigate these weaknesses, but a comprehensive study of their thermodynamic phase stability does not yet exist.

To that end, we will perform a computational investigation to study the doping of monolayer and bilayer TMDs, based on MoS2 and WS2, from first principles using the Alloy Theory Automated Toolkit (ATAT) together with the Vienna Ab initio Simulation Package (VASP). We will construct the thermodynamic temperature/pressure-composition phase diagrams for these TMDs with a selection of dopants, in order to provide guidelines for their experimental synthesis. We also seek to propose a mechanism leading to the formation of the doped compounds in terms of the atomic types forming the structure and the electronic environment in which they are embedded. Once complete, the results of the study will be submitted to an impactful scientific journal.


Researcher: Ivan Kološ         


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

Karolina CPU-1100; Karolina VIZ-40       

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


Researcher: Damien Lucien Michael Gagnier            


Post-dynamical in-spiral common envelope evolution   

Karolina CPU-96000; Karolina VIZ-40    

Common envelope is a phase in the life of binary systems when two stars interact inside a single shared envelope. It is thought to be at the origin of a wide variety of observed close binary systems such as cataclysmic variables, X-ray binaries, progenitors of Type Ia supernovae, or planetary nebulae nuclei, and precedes the majority of stellar mergers detected by ground-based gravitational wave detectors. Despite great efforts over the last decade, detailed hydrodynamic calculations have failed to reproduce the orbital properties of the observed post-common envelope binaries. Because of the wide range of temporal and spatial scales that need to be resolved and the associated high numerical cost, such simulations are always halted soon after the end of the initial dynamical in-spiral phase, and a subsequent slow contraction of the orbit on a thermal time scale is assumed. In my previous project (OPEN-24-67), I showed that this assumption is erroneous and that it most likely contributes to the discrepancy between observed and simulated properties of post-common envelope remnants. In this project, I will explore a wide orbital parameter range and devise semi-analytic formulae to predict the long-term evolution of orbital separation and eccentricity. The use of such prescriptions in 1D stellar evolution codes will provide access to invaluable and unique constraints on the properties of post-common envelope binaries, and on the rate predictions of mergers and of their resulting transients.


Researcher: Martin Dračínský           


Investigation of nuclear quantum effects in pharmaceutical solids           

Karolina CPU-40000; Karolina VIZ-40    

Many efforts have been recently devoted to the design and investigation of multicomponent pharmaceutical solids, such as salts and cocrystals. The understanding and correct description of intermolecular hydrogen bonds are crucial in these systems. However, experimental distinction between these solid forms is often challenging and common computational approaches may also fail. We have recently shown that the transformation of a salt into a cocrystal with a short hydrogen bond does not occur as a sharp phase transition but rather a smooth shift of the positional probability of the hydrogen atoms. Nuclear quantum effects, such as hydrogen atom delocalization and tunneling, are very important for accurate predictions of the structure and properties of these pharmaceutical solids. This project aims to develop a reliable computational protocol for accurate predictions of hydrogen-atom positions. The approach will be based on advanced path-integral molecular dynamics simulations that account for the nuclear quantum effects. These computational studies will be complemented with solid-state NMR and X-ray diffraction experiments. The results of this project will help to understand intermolecular interactions in pharmaceutical solids.


Researcher: Fabien Jaulmes


Computational modelling of fast ion orbits and their consequences in tokamak

Barbora CPU-18400; Barbora VIZ-40; Karolina CPU-1000; Karolina VIZ-40            

Nuclear fusion technology might enable us to generate energy without releasing large amounts of greenhouse gases into the atmosphere or leaving behind us long lived radioactive waste. The tokamak concept involves the use of magnetic fields to confine plasma hot enough to sustain fusion within itself. Today, as a part of international project ITER, a new tokamak is built in southern France. If successful, the device would be the first one of its kind to produce net energy. Fusion is now supported as a way to revert climate change and was discussed at the COP26.

COMPASS Upgrade (COMPASS-U) will be a large magnetic field (5T) tokamak that will allow the scientific investigation of various physics issues related to the operation of the future ITER. In particular, an 80keV Neutral Beam Injection (NBI) system is planned to heat up the plasma with 4MW of external power. Such a unit was tested on the COMPASS tokamak before its shut down and our modelling contributed to the interpretation of the results. The study and modelling of NBI-born particle behavior is of great relevance: it might influence future design of the system and its integration in the overall reactor design. We request computational time for the modelling of the interaction of the fast particles with the background plasma. Our code, EBdyna, with its new collisional features, was benchmarked against the NUBEAM code on several test cases.


Researcher: Tomas Martinovic         


Biodiversity Digital Twin development  

Karolina CPU-600; Karolina GPU-200     

Project aims to improve biodiversity simulations by running them on HPC clusters and allowing for better precision of the models, or increased scale. We are focusing on three main models: bee survival simulation, grassland simulation and forest/bird dynamics model. These models are part of the BioDT project, where we are working towards creation of several digital twins in the field of biodiversity research. Running these models on HPC clusters is first part of this effort.


Researcher: Róbert Babjak 


Laser-plasma based accelerators as secondary radiation sources              

Karolina CPU-32500; Karolina VIZ-40    

The interaction of intense lasers with plasmas has led to many technological advances in the last decades. One of the potential applications of such interaction is the efficient acceleration of electrons on distances orders of magnitude shorter compared to conventional accelerators. Due to the novelty of acceleration concepts, many paths are being actively explored in order to produce high-quality electron beams. One of the possible acceleration mechanisms is called direct laser acceleration which has the potential to produce multi-GeV electron beams using currently available laser facilities. Such electron beams can be used as sources of high-energy photons, neutrons, or they can be used to enhance the acceleration of ions from thin dense foils. Our project is dedicated to developing theoretical models capable of describing the acceleration and properties of radiation produced during the interaction. Such understanding of the fundamental processes during the laser-plasma interaction will allow for proposing optimal configuration of future accelerators, x-ray or neutron sources.


Researcher: Sára Simandlová       


Population genomics of gray wolf in Central Europe       

Barbora CPU-2000; Barbora VIZ-40; Karolina CPU-8000; Karolina VIZ-40, LUMI-C 5000

During the 21st century, the grey wolf population in central Europe has been increasing. As time goes on, the ways in which we can obtain genomic data from the animal populations get bigger. Wolf DNA samples are being collected from scat, urine, and even dead individuals (e.g. roadkills) now. Our laboratory is responsible for genetic monitoring of grey wolf in Czech Republic and in Slovakia (https://www.navratvlku.cz/o-vlkovi-genetika/) and we are members of international consortia studying wolf genetics (CEwolf consortium | Senckenberg Society for Nature Research). This gives us a large amount of data that is suitable for in-depth genomic analyses. The data can shed light on many interesting facts about wolf populations. We can read from the data whether wolf populations are merely coexisting or hybridizing. We can also estimate fitness, the effect of landscape fragmentation on the population, or verify possible hybrid status with domesticated dogs. Genomic research is currently using the cutting edge bioinformatics approaches that can cope with data overwhelming, that are well parsable and provide more comprehensive answers to the genomic questions being asked.


Researcher: Amina Gaffour            


Efficiency and Accuracy of Fragment-Based NMR Spin-Spin Coupling Calculations for the Backbone in Structured Proteins              

Barbora CPU-59500; Barbora GPU-6000; Karolina CPU-3200; Karolina GPU-2800             

A protein’s molecular three dimensional structure is linked to its function, thus, understanding the structure of proteins is essential in understanding its function. Enzymatic processes, substrate recognition, and protein interactions all take place at the molecular level. This is a fundamental concept of structural biology, which has successfully supplied us with a molecular-level understanding of numerous cell functions. In fact, proteins are highly dynamic, which is generally a strong characteristic in their function and regulation. Solution NMR spectroscopy is an ideal technique to probe both the protein dynamics and structure. In-silico prediction and modelling of NMR parameters including the Spin-Spin couplings has become indispensable in designing structural ensembles that match the experimental NMR observables. In this project, we aim to validate the accuracy and efficiency of computational protocols for calculating spin-spin couplings using structural ensembles devised from molecular dynamic (MD) as well as using fragment-based calculations and density functional mthods. The objective of this project is to facilitate the structural characterization of proteins and support the analysis of experimental NMR spectroscopy data.


Researcher: Valeria Butera            


Ru@MoS2 as Single-Atom catalyst for the CO2RR           

Barbora CPU-3000; Barbora VIZ-40; Karolina CPU-3000; Karolina GPU-11500    

Single atom catalysts (SACs) have received great attention due to their promising catalytic activity and sustainability. In this regard, the catalytic activity towards CO2 conversion reaction can be efficiently improved by the addition of single transition metal anchored on the 2D MoS2 monolayer. Within this project, we aim to explore the potential utilization of transition metals (TMs) anchored on MoS2, TM@MoS2, as promising SAC for the CO2 reduction reaction (CO2RR) by using first-principles simulations.  We will first evaluate the stability of the so-formed SACs in terms of binding energy. Once that the most stable SACs will be identified, we will investigate the reaction paths leading to the production of CO, CH4 and CH3OH as main value-added products. Our results will contribute to shed light on the potential of TM@MoS2 as selective SACs for the CO2RR and the understanding of the complex reaction mechanisms, aiming at the development of more efficient catalytic systems. 


Researcher: Mireia Diez Sánchez


Robust End-to-End Neural Diarization for Realistic Speech Data 

Karolina GPU-7400        

End-to-End neural diarization (EEND) systems are currently the research wellspring for speaker diarization tasks. EEND systems have reached state-of-the-art performance on two-speaker conversations, but struggle with recordings containing more speakers. This is a problem in realistic datasets such as Ego4D, DIHARD and VoxConverse, which comprise realistic and difficult situations and contain several more speakers in most recordings. This project will explore the usage of new architectures for EEND systems, mostly based on attention schemes, that will handle more speakers effectively.


Researcher: Urszula Wdowik        


Interplay between magnetism and superconductivity in the U-Te system under extreme conditions

Barbora CPU-3700; Barbora GPU-4000; DGX-2-350; Karolina CPU-3700; Karolina GPU-4300, LUMI-C 1250

Strongly correlated-electron systems with unconventional superconductivity such as high-Tc cuprates, heavy-fermion systems, Fe-based pnictides/chalcogenides, and newly-discovered nickelates pose a real challenge for theoretical condensed matter physics. Among them, a novel heavy-fermion spin-triplet superconductor UTe2 with Tc < 1.6 K is a paramagnetic end-member in the family of uranium-based unconventional ferromagnetic superconductors where spin-fluctuations without an ordered magnetic state play a major role in Cooper pairing. Signatures of spin-triplet pairing make this compound a strong candidate for topological superconductivity, and hence a material of interest in quantum computing. Unique properties of UTe2 can be dramatically changed by external magnetic field as well as pressure. To elucidate the interplay between superconductivity and magnetism in the U-Te system we propose extensive ab initio research of diverse properties of UTe2, UTe3 and U2Te5 compounds, especially changes of these properties upon external pressure and magnetic field. Our theoretical research is to support the ongoing experimental studies performed within the Czech Science Foundation (GAČR) project No. 22-22322S entitled Unconventional superconductors under extreme conditions by providing essential information at the microscopic/atomic scale for interpretation of the measured quantities.


Researcher: Eugen Hruska   


Machine learning of pharmaceutical compound binding

Karolina CPU-1000; Karolina GPU-2500

In silico discovery of pharmaceutical inhibitors requires the virtual screening of binding free energies for a large number of compounds. The large number of potential compounds limits the accuracy of these predictions. Higher accuracy methods are limited by restrictively large computational resource requirements, which results in a low number of potential inhibitors screened. Here, the reliability of virtual screening will be improved by combining faster methods with higher accuracy methods.


Researcher: Petra Kuhrova            


Prediction of 3D structure and description of the mechanism of action of selected aptamers

Karolina CPU-5000; Karolina GPU-18100             

Aptamers are short single-stranded oligonucleotides with high affinity towards various targets such as antibiotics, pesticides, inorganic materials, proteins, viruses, and cells. This high affinity of aptamers to recognize target molecules, especially low-molecular-weight molecules, is now increasingly used in the development of biosensors and detectors and in drug delivery. The ability to bind to a target with high affinity and specificity is strongly dependent on functional three-dimensional conformation of aptamer. However, it is very difficult to experimentally obtain structure-based mechanistic information. Moreover, only a limited number of experimentally determined 3D aptamer structures are currently available.

Using in silico tools, it is not only possible to determine the 3D structure of aptamers, but also to design aptamers, improve their affinity to the target, and identify structural patterns responsible for aptamer/target interface. In this proposal, the enhanced sampling simulations together with classical molecular dynamic simulations and docking tools will be used not only to predict the 2D and 3D structure of selected DNA and RNA aptamer-based biosensors, describe the mechanism of their function, but also to improve their efficiency. We plan to propose a workflow scheme including aptamer structure prediction, preparation of biosensor complex for molecular dynamic simulations, and validation of the obtained results with experimental data coming from our laboratory.


Researcher: Petra Čechová            


Interactions of Ionizable and Cell Membrane Lipids        

Karolina CPU-960; Karolina GPU-4600, LUMI-C 15360  

Lipid-based drug delivery systems, such as lipid nanoparticles (LNPs), are a promising branch of current medical research. The most notable recent example of such a LNP-based molecular delivery are the novel COVID-19 vaccines, where a LNP containing engineered ionizable lipids carries an mRNA molecule into the human body. To deliver its cargo into the cell’s cytoplasm, the LNP must first cross the outer cellular membrane. There are several proposed mechanisms of this process, however, the exact nature of the LNP-membrane dynamics has so far gained only limited attention as it is challenging to study experimentally. In this project we use molecular dynamics to visualize and describe the interactions of the ionizable and membrane lipids, to help understand the fine details of LNP entering the cell on a molecular level. This can in turn provide information about the ionizable lipid properties, which might suggest an improvement of the ionizable lipid design, to increase the LNP stability and cellular targeting.


Researcher: Martin Šurkovský          


Deterministic Road Traffic Simulator – III. phase

Barbora CPU-500; Barbora VIZ-40; Karolina CPU-1300; Karolina VIZ-40, LUMI-C 1000 

A usage of historical data observed from traffic combined with high performance computing resources for traffic simulation can optimize traffic flow in a macro scale level. A set of algorithms used for route planning has been developed in ADAS laboratory, including in-house traffic simulator and a scalable, distributed system for serving vehicle routing requests which is able to handle tens of thousands individual cars.

The aim of this effort is to reduce number of non-deterministic events in the traffic simulator to increase the reproducibility of results. In this way we will be able to fine-tune the setting of routing algorithms without need of doing several experiments with a particular setting to get statistically significant results.


Researcher: Andreas Erlebach          


Towards mechanistic understanding of zeolite synthesis using neural network potentials

Karolina CPU-34900; Karolina GPU-700

Sodium aluminosilicates (SAS) are mass-produced materials with a vast range of industrial applications from damage resistant glasses to catalysts for green chemistry. Optimizing their energy-intensive synthesis as well as their properties for new applications requires detailed knowledge of the chemical reactivity of SAS with water. However, modeling of chemical reactions at the SAS-water interface is extremely challenging due to the high structural complexity of such interfaces and the need of simulations with quantum mechanical (QM) accuracy. In this project, we will train Machine Learning (ML) models to enable fast simulations of the SAS-water interface with QM quality. We will demonstrate the performance of our ML models by tackling a long-standing challenge: the realistic modeling of the initial phases of the crystallization process at the external SAS zeolite surface.


Researcher: Jiří Nádvorník  


Hierarchical Semi-sparse cubes (HiSS cubes)      

Karolina CPU-60800; Karolina VIZ-40    

Data volumes in astronomy are often defining what Big Data means. A recent example is the LSST telescope that will produce 20TB of raw data per night, which will result in cca 60 PB for the whole duration of the project. This means that we are already in the age where we are measuring data nobody will look at, as there is just not enough time in our lives for it. Only through machine learning algorithms that filter the interesting data from these huge volumes can we hope to make new discoveries. Our efforts in this project are even one step further – combining such huge amounts of imaging data with spectroscopic observations to gain even better ways how to interpret the data and gain new knowledge about the way universe works.

The Hierarchical Semi-Sparse (HiSS) cube framework aims to combine imaging and spectroscopic Big Data in an efficient and scalable way.


Researcher: Martin Srnec    


Control of C-H cleavage selectivity by design: the joint effects of substrate and oxidant structure, and fine tuning through the reaction environment

Barbora CPU-13000; Karolina CPU-56250, LUMI-C 10967           

The synthesis of new drugs and materials is an expensive task, requiring a large energy investment and producing material waste in the process. A powerful strategy to minimize costs and byproducts is the direct functionalization of molecular leads through C-H activation, as most biologically relevant molecules are organic, and they display many similar but non-identical C-H bonds. By the selective cleavage of a targeted C-H bond, the chemist can produce new and more complex products in a succinct manner. In this project we will investigate the critical factors influencing the selectivity of C-H activation based on a chemical theory recently developed by us, made possible by the IT4I supercomputing facilities. The theory is based on the thermodynamic characterization of the reaction partners (a substrate and an oxidant), which provides not only the well-known effect of reaction energy (ΔG0) on reactivity, but also the nonclassical thermodynamic variables named frustration (σ) and asynchronicity (η) factors. With this toolbox, we will use quantum chemical and hybrid QM/MM calculations to investigate (a) strategies to tune ΔG0, η and σ independently via modular design, (b) the control of reactivity and selectivity through the solvent composition and (c) the effect of additives, such as Lewis acids. With this study, we will seek a unified set of guidelines for selective C-H activation. 


Researcher: Štěpán Marek  


Molecular Electronics in DFT and GW Approximation    

Karolina CPU-2000; Karolina FAT-100; Karolina VIZ-40, LUMI-C 500  

Calculation of conductance of molecular junctions is fundamental for our understanding of complex processes taking place on the nanoscale. The interplay of molecular properties and macroscopic leads makes it a challenging problem where first-principles approaches can prove useful, given that the quantitative description of these systems remains elusive. In this project, we investigate several properties of molecular junctions, both general and specific. Firstly, a way to correct for some quantitative shortcomings of DFT approach could be to carry out GW calculation – a specific approximation to many-body perturbation theory that is known to correct some aspects of electronic structure that DFT determines erronously. We aim to carry out the convergence study for this approximation, specifically how the transmission function changes for different electrode sizes and shapes. Secondly, we are interested in two particular cases of junctions – ferrocene and “Geländer” molecule. Ferrocene molecule exhibits unusually high conductance even at room temperature and Geländer molecule was observed to rotate under the influence of current flow. Explanations of the origin of these experimentally observed effects is missing and this project aims to discover these.


Researcher: Pavel Ondračka              


Machine learning-enabled modelling of amorphous W-B-C         

Karolina CPU-11250; Karolina VIZ-40    

The bench-mark protective coating material utilized in the cutting and forming applications is TiAlN exhibiting high hardness as well as stiffness. Unfortunately, these favorable properties are associated with unwanted brittle deformation behavior resulting upon mechanical loading in the formation of cracks which limits the performance and lifetime of the coated tools. From a materials design point of view a rather unusual combination of properties - high hardness and stiffness together with moderate ductility - is therefore required for the next generation of protective coating materials. This project aims to design coating materials with high hardness and enhanced fracture toughness for cutting and forming applications with extended lifetime and performance by ab initio calculations on a novel class of amorphous tungsten-boron-carbide material.


Researcher: Martin Čadík    


Deep-Learning Approach to Information and Generative Visualisation   

Karolina CPU-2400; Karolina GPU-7300

Automatic label placement is an area of information visualization research focusing on positioning textual annotations (so-called labels) on top of a visualization serving as an additional informational layer. For an experienced cartographer, manual labeling takes up to 50 % of the overall production time of maps. On the other hand, even simplified versions of label placement algorithms are NP-hard. Therefore, it is often necessary to use heuristics to find feasible positions of many labels automatically. The general label placement has many practical aspects. In the second stage of the project, we want to consider a specific problem: visualization of an arbitrary label (usually a number) on the surface of a real object, such as a soccer player’s jersey. Recent advancements in computer vision allow us to use generative models to generate novel images with specific conditions. Our aim is to explore the possibilities of using conditional generative models in this domain. Our research is directly applicable in gaming and broadcast industries having an unprecedented impact on individuals.


Researcher: Jan Brezina       


Deep learning meta-model for numerical homogenization          

Karolina CPU-10000; Karolina GPU-4000             

A deep geological repository (DGR) of radioactive waste for the Czech Republic will be located in a crystalline rock where the dominant water flow is through a fracture network. Despite all protective layers, the long-term barrier is the surrounding rock. The necessity to describe water flow through a wide range of fracture scales has motivated our aim to investigate groundwater processes with discrete fracture-matrix (DFM) models that consist of a continuum and a discrete fracture network. As modeled processes involve uncertainties, we utilize the multilevel Monte Carlo method (MLMC) to estimate the expected value of various safety indicators based on the models. The MLMC concept depends on the ability to construct a sequence of models with decreasing resolution. In order to represent small-scale fractures in a coarse DFM model, numerical homogenization is employed. This procedure is accelerated by a meta-model based on deep learning techniques. Convolutional neural networks, graph convolutional neural networks, and generative adversarial networks (GANs) are considered meta-models of numerical homogenization procedure that determines equivalent continuum properties (permeability tensor, porosity, etc.) of a DFM model.


Researcher: Vojtech Cima   


ViderAI Transfer Learning            

Karolina GPU-100          

The medical data collection, curation, and labeling process are known to be a very expensive and long-term process. In deep learning scenarios where a large number of data samples are required to create well-performing predictive models, this may also result in very high computational costs. Therefore, in production use cases, transfer learning is often used to create or adjust robust predictive deep learning models using a much smaller number of samples. We aim to explore the exploitability of this technique to efficiently adapt existing models used for diabetic retinopathy detection to work with new sampling devices (cameras) and so enable deployment of the new facilities (e.g. hospitals) without a further need for large-scale data collection. This promises to save human and computational resources as well as shorten the time required for deployment in new facilities.


Researcher: Pavel Praks       


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

Barbora CPU-14000; Barbora FAT-1800; Barbora GPU-1000       

Electrical power consumption is gradually increasing over the years. In combination with the ageing of distribution grids and the required integration of new uncontrolled sources (wind and solar systems), a higher emphasis is placed on power flow control and monitoring elements to ensure continued supply and required quality of the provided electrical energy for society. However, the purchase price and the maintenance cost of the switching and monitoring devices are high, therefore discrete optimization must be employed to identify the optimal placement and operating mode of the control devices. The stochastic approach is very robust but extremely time-consuming. Fortunately, the performance of stochastic methods can be accelerated by a parallel implementation. It is the third project for testing of HPC infrastructure of IT4innovations for the modelling and optimisation of Czech distribution networks. The novelty of the project includes testing the Ray Tune Python library for large-scale distributed hyperparameter optimization of energy networks. Moreover, the PySR library will be used as a tool for explainable AI of selected thermochemical processes (for example, gasification).


Researcher: Samuel Lukeš  


Simulation of liquid metal vapor transport and re-deposition on COMPASS Upgrade tokamak

Karolina CPU-8000; Karolina VIZ-40       

With the ever-increasing energy consumption of mankind, the need to use ecologically and politically acceptable, reliable 24/365 and inherently safe sources also increases. One of the very few ways to satisfy these requirements seems to be nuclear fusion. That is why its flagship research project ITER is currently the most expensive science experiment on the planet. However, several unanswered questions still remain, one of them is the long-term reliable heat shield of the reactors. Current solid shields (W, Be) must withstand extreme heat fluxes from thermonuclear plasma in the form of charged energetic particles, which significantly degrade any crystalline lattice or overheat the matter to the point of cracking or melting. Liquid metal technology is supposed to solve this. Compared to solid metals, their liquid state is assumed even from low temperatures (Li, Sn). Capillary forces in a porous mesh (W, Mo) then hold the liquid in the desired shape.

This project simulates the transport of vapor from liquid metal heat shields (the main way how liquid metals are cooled under heavy heat loads) in the COMPASS upgrade tokamak. 3D Monte Carlo code ERO2.0, which solves the transport and plasma-wall interactions of impurities in the region between the confined plasma and the solid wall of current fusion devices, is used for this purpose.


Researcher: Jiri Brabec         


Theoretical study of polycyclic aromatic hydrocarbons at the post-DMRG level using optimized orbitals

Karolina CPU-46200; Karolina VIZ-40    

Recent progress in ultra-high vacuum (UHV) on-surface chemistry and scanning probe microscopy (SPM) with the unprecedented sub-molecular resolution [spm] enabling precise determination of molecular products opened new possibilities to synthetize and characterize molecular species not available via traditional solution chemistry [syn]. For example, a possibility to synthetize polycyclic aromatic hydrocarbons (PAH) possessing open shell electronic structure featuring magnetic properties. The researchers have now succeeded in producing room-temperature stable magnetic carbon nanostructures in practice and showed that the theory does correspond to reality [gob].

Recently, we have studied the electronic structure of several interesting PAH at the DMRG level [app], which allows us to determine electronic structure and spin density. However, some PAHs are very challenging, for example, PAHs with ring topology. In this work, we plan to study the electronic structure of these systems using DMRG with adiabatic connection approach, also we plan to use the mode transformation [modetrafo] to find the optimized orbitals for these structures. These structures have been recently prepared by means of UHV on-surface synthesis by our experimental partners.


Researcher: Jiří Jaroš             


Modeling of Low Intensity Focused Ultrasound Using Convolutional Networks III             

Barbora CPU-5000; Barbora FAT-10; Barbora GPU-200; Karolina CPU-1000; Karolina FAT-10; Karolina GPU-1000,
LUMI-C 500, LUMI-G 100             

Transcranial low-intensity focused ultrasound (LIFU) therapy is increasingly used for the non-invasive treatment of brain disorders. However, conventional numerical wave solvers are currently too computationally expensive to be used online during treatments to predict the acoustic field passing through the skull.

As a step towards real-time predictions, we developed a fast iterative solver for the heterogeneous Helmholtz equation in 3D using a fully-learned optimizer. The lightweight network architecture is based on a modified UNet that includes a learned hidden state. The network is trained using a physics-based loss function and a set of idealized sound speed distributions with fully unsupervised training (no knowledge of the true solution is required). The learned optimizer shows good performance on the test set, and is capable of generalization well outside the training examples, including to much larger computational domains, and more complex source and sound speed distributions, for example, those derived from x-ray computed tomography images of the skull.

The aim of this project is to introduce fully heterogeneous description of the skull into the model couple it with the thermal model and minimize computational requirements which still exceed the real-time requirements for inference and reaches days for training."


Researcher: Martin Žonda  


Complex Magnetic Systems for Neuromorphic Computing          

Barbora CPU-500; Barbora GPU-200; Karolina CPU-3200; Karolina GPU-1000    

Artificial neural networks have been proven as a highly efficient computational model allowing us to solve complex tasks like image classification, face recognition, or spam filtering. Although the architecture of artificial neural networks completely differs from the von Neuman one utilized in our computers, they are usually implemented as a computer code running on standard processing units. Thus, the most remarkable features of neural networks, like decentralized data storage and on-chip memory, cannot be fully used in practice. Therefore, a physical implementation of artificial neural network is desired.

Scientific discipline aiming to develop a computer chip able to mimic functionality of a biological neural network is called neuromorphic computing and engineering. The desired device should response to an external input signal. There are two main features of the response: nonlinearity and memory. The goal of our project is to find a magnetic system obeying these conditions. By means of multilevel hybrid simulations combining quantum transport calculations, atomistic spin dynamics, and micromagnetic simulations, we shall inspect ability of complex magnetic textures featuring nontrivial topology to be used for solving classification and regression tasks. We shall provide appropriate schemes for performing the computation.


Researcher: Paolo Nicolini  


NANOscale study of early-stage abrasive WEAR of molybdenum disulfide in silico (NANOWEAR)

Barbora CPU-30000; Barbora VIZ-40     

The proposed project constitutes part of a joint experimental-computational study aimed to a deeper understanding of abrasive wear processes in

molybdenum disulfide (and more in general on layered materials). Indentation and scratching simulations will be performed in order to haracterize the nanostructures (flakes, chips, clusters) formed in the wear process (a similar analysis with different resolution will be conducted via atomic force microscopy measurements). Quantitative estimations of the amount of debris formed will also be obtained. In addition, in-depth investigation of the phenomena involved (e.g. layer rupture, exfoliation and/or folding/wrinkling processes) will also be performed, ultimately unraveling atomistic mechanisms which are not accessible experimentally. All in all, the project will shed light on fundamental mechanical and tribological properties of 2D materials, which are rarely addressed in the literature. The results can also suggest optimum strategies for applications of layered materials as solid lubricants or as protective layers for machine components.


Researcher: Barnabas Barna              


Turbulent and windy: studying the environment of the galactic circumnuclear disk         

Karolina CPU-2500; Karolina VIZ-40       

At a distance of 25.500 lightyears, the Galactic Center of the Milky Way (GC) is subject of particular interest to both the observations and theoretically oriented astronomical studies. Despite the decade-long intensive observational campaigns, we still know only a little about the exact structure of this region, because observations are limited by the line of sight effects and the extinction from interstellar matter (ISM). Thus, numerical methods are required to simulate the inner ~10 pc where the gas forms the circumnuclear disk (CND) around Sgr A*, the supermassive black hole of the Galaxy (SMBH).

We are planning multiple hydrodynamic simulations with FLASH in 3D to investigate the impact of stellar wind and magnetic field on the CND. The results will be matched with recent observations, e.g. the current distribution, velocity, and turbulence of the ISM inside and around the CND (i.e. in the halo). The most realistic setup will serve as a model environment in future simulations for studying whether supernovae can deliver mass to the SMBH.

This project is the direct continuation of our previous efforts in modeling the GC environment and the feeding of the SMBH by the nearby supernovae .


Researcher: Jiří Chudoba     


Simulations for LHC experiment ATLAS 

Karolina CPU-200000; Karolina VIZ-40  

Experiments measuring particle collisions in Large Hadron Collider at CERN require huge computing capacity for data analysis and Monte Carlo simulations. In cooperation with other projects, they developed and use distributed environment called Worldwide LHC Computing Grid (WLCG). Basic needs for computational and storage resources for Run III starting in July 2022 are covered by WLCG pledge resources. However very significant fraction of resources is used from so called unpledged resources, which are provided by HPC centers, occasional cloud resources and resources from volunteers via BOINC. In 2021, Vega supercomputer (part of PRACE network) almost doubled available computing resources for the ATLAS experiment. The second most significant HPC contributor was Karolina supercomputer.

We already created a well working environment for job submission to IT4I supercomputers. Jobs are sent automatically by a production system and are able to keep busy available resources. Optimalisation of CPU usage was done using HyperQueue.


Researcher: Adam Pecina   


Universal Quantum Mechanical Scoring Function for Virtual-Screening Applications       

Karolina CPU-9000; Karolina VIZ-40       

Drug discovery is a very costly and time demanding process, through which potential new drugs are identified. There is thus an urgent need of more efficient methods that will enhance the process. In a computer-aided drug design, standardly used ultrafast scoring functions fail even in simple tasks like native pose identification and affinity ranking. Due to the enormous advances in computational power, quantum mechanical (QM) methods have been gaining in importance, especially in cases where a reliable description of protein-ligand interactions is of utmost significance. The goal of this project is to develop and test a fast, accurate and system-transferable QM-based scoring protocol for virtual screening applications that will provide an unambiguous advantage over the current scoring methods.


Researcher: Dominik Legut 


Calculations of thermoelectric effects and spin-orbit torque in 2D van der Waals materials

Barbora CPU-40000; Barbora GPU-2000; DGX-2-500; Karolina CPU-25000; Karolina GPU-9800, LUMI-C 25000  

Two-dimensional (2D) van der Waals (vdW) materials assembled in vertical stacks allow for a combination of many quantum features within a single heterostructure via proximity effects. For instance, considering a heterostructure made of a 2D material with strong spin-orbit coupling and a 2D ferromagnet allows us to propose a variety of electronic properties with promising quantum features for future nanoelectronics. Especially, one of the most fascinating fundamental research is spin-orbit torque (SOT) technology which involves the vdW stacks of 2D ferromagnet and the 2D spin-orbit coupling material. The SOT is associated with anomalous spin currents due to strong spin-orbit coupling in one material, which can then purely electrically manipulate or excite magnetization in the ferromagnetic materials. One of the applications is a non-volatile magnetoresistive random access memory device with significantly low-energy cost operation. The efficiency of anomalous spin-polarized current is tightly related to electric conductivity, spin-orbit coupling strength, and thermal properties of the materials. In this project, we will investigate the thermoelectric properties of the vdW heterostructures made of 2D materials. We will study the twist angle between the 2D materials and its effect on electronic structure, and calculate induced SOT and related thermal effects. We will employ ab-initio calculations of electronic structure, determine phonons in harmonic and anharmonic approximation, determine thermal conductivity, and calculate electron-phonon interaction in order to address the thermoelectrical effects allowing us to seek the best figure of merit.


Researcher: Diego Nicolas Calderón Espinoza           


Moving-mesh Radiation-hydrodynamic Simulations of Tidal Disruption Events   

Barbora CPU-61400; Barbora VIZ-40     

Most if not all galaxies in the Universe harbour a super-massive black hole in their centre. Unfortunately, only a small fraction of them accretes at rates that allow generating detectable radiation. However, a black hole can become active over human timescales if a star travels close enough to be disrupted, causing the accretion of the stellar material. As a result, a bright flare is generated known as tidal disruption event (TDE). TDEs present a unique opportunity for studying super-massive black holes, which otherwise would remain undetectable. Here, we propose conducting radiation hydrodynamic simulations of the outcomes of TDEs to synthesise light curves and quantify the effect of different black hole mass and rotation on them. Unlike previous works, we will use the state-of-the-art moving-mesh code JET. This will allow us to model the evolution of the radiation and the wind generated from the TDE while they interact with the surrounding medium. The models will cover length scales from a couple to tens of thousands gravitational radii, and timescales from seconds to decades. As JET has already been tested satisfactorily for running two-dimensional simulations on the Barbora cluster, it will be straightforward to investigate a more realistic three-dimensional scenario. The results will provide for the first time a set of synthetic light curves of TDEs interacting with the circumstellar medium based on self-consistent three-dimensional radiation hydrodynamic models.


Researcher: Jan Martinovic


Large-scale Generation of Pose Selector Training Data for LIGATE            

Karolina CPU-19000; Karolina GPU-5000, LUMI-G 45000             

COVID-19 pandemic demonstrated the need of efficient workflows to rapidly develop new drugs against emerging diseases. A promising approach consists in using HPC resources to virtually screen libraries of drug candidates to find the compounds that binds most tightly to a viral protein. To this end, molecular docking generates structures of the complex formed by the target protein and the candidate drug and the most promising ligands are sent forward towards experimental validation. Unfortunately, promising ligands are selected by approximate scoring functions. Recently, Machine Learning approaches have been demonstrated to outperform legacy ones with respect to docking power, i.e.: model’s ability to predict ligand poses close to experimental ones. ML models’ hyperparameters are usually fine-tuned by supervised training using RMSD labels of individual ligand poses generated in silico wrt experimental crystal poses available in e. g. PDBbind database. However, absolute binding free energies have been demonstrated to be higher quality labels to train pose selector ML models wrt RMSD ones. Unfortunately, full cycle ABFEs can only be obtained by means of long MD simulations and cannot be linked to individual, interconverting poses. We plan to overcome this issue by means of short MD simulations to obtain approximated ABFEs for individual ligand poses with Explicit Solvent and Implicit Solvation models. Our plan is to build a completely new training set made up of 4M in-silico generated crystal near ligand poses individually labelled with slightly approximated but ranking preserving ABFEs. To the best of our knowledge, this is the first attempt to generate ABFEs for such a large dataset to be used for ligand pose prediction in HTVS studies.


Researcher: Petr Macha       


GBS simulation of the COMPASS tokamak          

Karolina CPU-26900; Karolina VIZ-40    

The proposed project focuses on simulations of plasma turbulence of the COMPASS tokamak edge and the divertor region. Plasma turbulence plays an important role in tokamak confinement and has a significant impact on the heat loads on plasma facing components. For the simulation, a 3D flux-driven, global, two-fluid turbulence GBS code is used. The code is capable of both interpretative and predictive modeling due to both the plasma equilibrium and fluctuation dynamics. GBS already demonstrated good agreement on TCV tokamak. In this project, we plan to perform GBS validation against COMPASS experimental data, which is also part of GACR project (22-03950S). COMPASS offers a unique possibility to also validate electron temperature and plasma potential fluctuations, which are routinely measured. The first COMPASS simulation was already started, and the turbulent phase was reached, however, more computational resources are needed in order to capture sufficient turbulence statistics and validate the code against COMPASS experimental data. Furthermore, because recent research shows a significant impact of neutrals on plasma profiles, especially at the divertor, the kinetic neutral model, which is included in the GBS code, is planned to be used during validation. Based on the validation, the GBS code will be used for predictive modeling of the COMPASS Upgrade in the future, which is relevant to ITER.


Researcher: Marek Vaško   


Quality assessment of computer vision data-sets            

Karolina GPU-100, LUMI-G 200          

The area of image recognition and computer vision overall suffers from faulty input data. By assessing the image quality one can possibly eliminate inputs which may cause under performance of networks. The overall problem with data-sets is their shear scale which makes the task nearly impossible to do by hand. Current methods allow automatic assessment of image quality based on multiple different criteria. The main idea in state-of-the-art approaches is to train network with auxiliary information, which may include values of loss function per samples, statistical properties of samples when compared to each other and estimation of sample deviation from ideal examples. These methods are useful in the assessment of quality for specific task such as recognition, where it is possible to reject low quality samples before they are evaluated in recognition system. With computational resources of the project we would like to test out if application of these methods on different data-sets improves  the accuracy of recognition systems based on iris biometry and possibly other types of biometric recognition tasks, making these methods generally applicable to one more different tasks and experiment with different types of neural networks.


Researcher: Petr Šesták       


Grain boundary segregation revised via on-the-fly machine-learned force field  

Barbora FAT-300; Karolina CPU-18300; Karolina FAT-200; Karolina VIZ-40, LUMI-C 5000           

The proposed research is focused on the study of impurity segregation at grain boundaries using the on-the-fly machine-learned force fields as a new promising tool for atomistic simulations. In this approach, reliable force fields (atomic interactions database) are generated using the ab initio molecular dynamics and on-the-fly machine learning. Subsequently, the obtained force fields are used to study selected grain boundary properties. This helps to overcome some limitations of the present atomistic simulations, such as the necessity to use small computational cells and 0 K temperature in ab initio models and insufficient precision of interatomic potentials in classical molecular dynamics. This way, we can study lower and more realistic impurity concentrations under finite temperatures even for grain boundaries requiring large simulation cells. Thus, the atomistic simulations in this project will approach more physically realistic conditions, increase the precision of our predictions and, moreover, save a lot of CPU time as classical molecular dynamics is faster than quantum mechanics approaches.


Researcher: Vladislav Pokorný         


Correlation effects in superconducting quantum dot junctions III             

Barbora CPU-10700; Barbora VIZ-40     

Josephson junctions are miniature devices in which two superconducting electrodes are separated by a weak link, usually a thin insulating barrier. These junctions had become standard building blocks of various devices including rapid single flux quantum (RSFQ) electronics and qubits for quantum computing. If the insulating layer is replaced by a quantum dot, e.g., a short semiconducting nanowire, we obtain a hybrid device in which various quantum-mechanical and relativistic phenomena can be separately tuned. Understanding the complex interplay of these effects is a necessary step in developing a new generation of superconducting devices. Supercomputers are now a necessary tool for simulating such systems, understanding the available experimental results and predicting their behavior.


Researcher: Sergiu Arapan  


A computational study of the magnetic phase transitions in the Fe-Ta Laves phase alloys

Karolina CPU-10000; Karolina GPU-5000, LUMI-C 20000             

Magnetic solids with considerable magnetocaloric (MC) effect (MCE) have recently attracted an increasing research interest due to the potential application for the magnetic refrigeration (MR) technology. MCE is a heating or cooling of the magnetic material under the influence of the applied magnetic field. Rare Earths (RE) Laves phase compounds show MCE at low transition temperatures and were suggested as MC materials for hydrogen liquefaction. However, the high cost and low abundance of RE call for the search and development of RE-free MC materials. Fe-based alloys are one of the potential options due to their excellent magnetic properties and low cost. In particular, the hexagonal Laves phase (Mg2Zn type, C14) in Fe-Ta alloys show transitions from magnetically ordered states (ferromagnetic or antiferromagnetic) to disordered (paramagnetic) one upon Fe concentrations. In this work, we will investigate the magnetic transitions (MT) in the non-stoichiometric Fe-Ta alloys by means of first-principles electronic structure calculations.


Researcher: Oldřich Plchot 


Audio-visual data processing via large-scale pre-trained models for the assessment of psychiatric patients          

Karolina CPU-1500; Karolina GPU-3600

The goal of this project is to use large-scale, pre-trained models for analysis of audio and visual data of psychiatric patients for the prediction of mental health questionnaire scores and psychiatric diagnosis. Both audio and visual data of psychiatric patients, can provide a wealth of information and with further processing can provide digital biomarkers representative of the state of the patient. With the advent of deep learning, there is an opportunity to use large-scale, pre-trained models for speech and visual analysis combining these models into a multimodal network that is able to learn how to extract meaningful audiovisual biomarkers and then better predict the state of psychiatric patients. Ultimately, this work seeks to help push psychiatry to more objective methods of digital sensor based data gathering and create an automatic artificial intelligence pipeline for patient state prediction and mental health diagnosis.


Researcher: Matus Dubecky              


Modeling of Excitons in Sc2C(OH)2 MXene         

Karolina CPU-65600; Karolina VIZ-40    

Fixed-node diffusion quantum Monte Carlo (FNDMC) is a promising many-body method that recently achieved reference accuracy for fundamental gaps in 2D materials, like, e.g., fluorographene, or, MXenes. This project focuses on development of currently unexplored FNDMC-based pproaches for explicit modeling of localized exciton quasi-particles in extended systems using multi-determinant trial wave functions. Application involves the recently studied and promising non-magnetic semiconducting MXene 2D system, Sc2C(OH)2. The numerical results, like optical gaps and exciton binding energies, will be compared to the accurate Bethe-Salpeter (BS) results, available for the selected MXene. The planned FNDMC computations will serve as a proof-of-concept example and have a potential to open new routes toward explicit modeling of excitons in larger systems than possible today by more demanding BS method.


Researcher: Marta Jaroš      


Offloading of Workflows Executions to Remote Computational Resources III.     

Barbora CPU-200; Barbora GPU-100; Karolina CPU-500; Karolina GPU-1000       

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: Martin Zelený 


Ab initio study of the layered van der Waals compound VBr3     

Karolina CPU-56300; Karolina VIZ-40    

VBr3 compound belongs to the family of layered van der Waals magnetic materials. The crystal structure of transition-metal trihalides given by a graphene-like honeycomb network of transition-metal ions and their magnetic ordering at finite temperatures provide promising opportunities for spintronic device applications. This compound exhibits anomalous phase transition at 90.4 K from high-temperature trigonal structure to low-temperature monoclinic structure which exact nature is not fully described yet. In the proposed project, we are going to reveal, how the lattice stability is influenced by van der Waals interaction between layers, magnetic ordering of the compound and presence of lattice defects. By this route, the results will contribute to the improvement of the understanding of complex behavior of layered van der Waals magnetic materials.


Researcher: Oldřich Plchot 


Multichannel Speaker Recognition with Large-scale Pre-trained Models

Karolina CPU-1000; Karolina GPU-7800

The project's main goal is to substantially advance the state-of-the-art in robust multi-channel Speaker Recognition (SR). We will continue our basic research in the domain of trainable multi-channel front-end based on beamforming and interconnect it with state-of-the-art feature and embedding extraction via large-scale transformer-based models. In a later stage of the project, we will attempt to sidestep the beamforming frontend by designing a simpler MISO (Multiple Input Single Output) neural networks and jointly learn temporal-spatial features from multichannel input. The whole architecture will support joint training of the feature extraction and the objective of the target speech application. The  joint training is close to the End-to-End training, which generally provides better performance than what can be achieved with modular systems when individual blocks are trained on separate objectives, such as beamforming tuned only for perceptual speech quality. We will compare our results with the latest state-of-the-art beamforming methods and our previous research where we trained the beamformer w.r.t target application (speaker verification).


Researcher: Jan Zemen        


Modeling Magnetic Structure of Transition-Metal Nitrides with Large Perpendicular Magnetic Anisotropy           

Barbora CPU-25900; Barbora VIZ-40     

Magnetically ordered thin films with large perpendicular magnetic anisotropy (PMA) and low saturation magnetization (MS) enable construction of non-volatile magnetic random access memory (STT-MRAM) with efficient magnetization switching, high retention and external field immunity. Transition-metal nitrides with anti-perovskite structure have been explored in this spintronic context due to their complex magnetic order including collinear and non-collinear ferrimagnetic phases (FIM, ncFIM). We have recently simulated magneto-optical and magneto-transport properties of anti-perovskite nitrides, focusing on Mn3NiN and Mn4N with non-collinear antiferromagnetic (ncAFM) and ferrimagnetic phases, respectively. The films with ncAFM phase can generate spin-polarised currents despite the zero net magnetization so they could be used as electrodes in antiferromagnetic tunnel junctions (AFMTJs). Here we propose to explore related ncFIM, FIM, and ferromagnetic (FM) phases of Mn4-xAxN (A = Co, Fe, Ni). Large PMA, Ku = 3.9 MJ/m3, has been predicted in strained ferrimagnetic Mn4N. However, ferromagnetic Fe4N shows much larger spin-polarization of electrical conductivity and better mechanical properties. Our aim is to simulate the magnetic order and corresponding magneto-optical Kerr effect (MOKE) spectra across a range of compositions where Mn is partially or fully replaced with Fe, Co or Ni. A range of strains induced by lattice mismatch with substrates such as MgO or Si will be considered.


Researcher: Rafael Dolezal 


Generative adversarial networks for biomolecular dynamics discovery of new nonrapalog allosteric mTORC1 kinase inhibitors           

Karolina CPU-12000; Karolina GPU-5900             

Generative adversarial networks (GANs) have become one the most attractive tools from the machine learning spectrum which can even produce artistic objects like a picture of a human face. The powers of GANs will be utilized in this project for learning molecular dynamics trajectories of an enzyme mTORC1 that controls hundreds of biochemical processes in cells, including, for instance, cancer or ageing. Through a receptor-based virtual screening, a set of mTORC1 inhibitors will be selected and investigated by relatively short molecular dynamics (MD) simulations. Next, the MD data will be used for training GANs which after proper optimization will provide considerably longer artificial atomistic trajectories, and consequently a better chance to discover thermodynamically stable mTORC1 inhibitors. The obtained knowledge will be utilized in our projects focused on treatment of neurodegenerative disorders and research of ageing.


Researcher: Martin Friak     


High-throughput analysis of thermal vibrations in magnetic high-entropy alloys

Barbora CPU-88100; Barbora VIZ-40; Karolina CPU-110160; Karolina VIZ-40       

So-called high-entropy alloys are typically metals containing a few different chemical elements with similar (or even equal) concentrations. The atoms are randomly distributed over the crystal lattice sites. Mechanical properties of metallic alloys, such as their strength, can be linked to size difference between the constituting atoms. The high-entropy alloys represent an extreme case when numerous different chemical elements are mixed together and a significant impact on the strengthen of the material is observed. Another consequence of the atomic size differences are the displacements of the atoms from their lattice sites during thermal motion, so-called ”lattice distortions”. These atomic displacements can be quantified by so-called mean square atomic displacement, that can be measured, for example, by X-rays or electron microscopy. Importantly, the connection between the atomic size differences on one hand and different thermal vibrations of atoms (and subsequently the strength of the material) is not fully understood yet. We plan to use theoretical calculations to determine the thermal vibrations of a series of different alloys in order to explore the above mentioned structure-property relation. Our theoretical study will be complemented by experiments by a project partner from Japan.  


Researcher: Michal Cifra      


SubTHz spectra of functional vibration modes of proteins            

Karolina CPU-6000; Karolina GPU-6000

Understanding the interaction of electromagnetic field with the biological matter at the molecular level is crucial both for bionanotechnological applications as well as for setting safety limits for undesired electromagnetic exposure of complex biological systems, including humans. In the current computational project, we aim to explore how the electromagnetic field interacts with tubulin – a protein that serves several important roles in cells, including cell division. Our focus is on frequencies of electromagnetic field which match the frequencies of biologically functional protein vibrations. The frequencies of these collective protein vibration modes lie typically in subTHz - THz frequency range depending on the size and other properties of the protein molecule. The goal of the project is to employ molecular dynamics simulations and modal analyses to calculate the spectra of tubulin vibration modes that are associated with the tubulin conformational transitions. Such knowledge will guide experiments and provides the first step for the development of electromagnetic control of tubulin-based nanomachines.


Researcher: Jan Kuriplach   


Structure, magnetism and defects in inverse Heusler alloys Mn2FeZ (Z = Si, Al, Sn)          

Karolina CPU-3000        

The purpose of the proposed project is to investigate computationally the interplay between point defects and structural and magnetic properties of selected inverse Heusler alloys with the composition Mn2FeZ (Z = Si, Al, Sn). The results of ab initio calculations and simulations will be compared and confronted with experimental data, which enables to check the adequacy of theoretical approaches used and to assist the proper interpretation of measurements. Studying Heusler alloys will bring a deeper understanding of their fundamental physics and will pave and ease the way to their possible applications.


Researcher: Jan Pokorny     


Holistic approach to Rotating Packed Bed (RPB) Carbon Capture process with the use of 3D CFD

Karolina CPU-780           

The project will study the mechanisms that take place in a rotational absorber for CO2 capture (Rotational Packed Bed RPB). The RPB combines hydrodynamic and process mechanisms. The major issues are liquid spray in the rotor eye, in the rotating packing made of knitted wire mesh or a zigzag geometry and in the outer cavity, incl. liquid spray on the outer surface of the packing and formation of a liquid film on casing walls. So far attention has been mostly paid to the effect of rotation or type of the packing on the overall efficiency of gas capture, often supplemented by CFD. However this engineering approach doesn´t elucidate the role of individual elements of the RPB and their contribution to CO2 capture. Moreover there is no relevant literature that illuminates the abovementioned mechanisms in context, and which studies in detail the interaction between liquid distribution and the efficiency of gas capture. Therefore the main goal of this project is to consider the RPB as a whole and using 3D computational modelling with subsequent limited validation to unveil the interconnection and the influence of individual RPB segments on CO2 capture and to optimize their capture efficiency. Due to practically impossibility of experiments inside the packing, 3D CFD simulations using Volume of Fluid technique with a limited validation by own experiments is the only way how to recognize and optimize processes in the entire RPB.