Optimisation of design and production processes of vertical pumps using modern technologies

Call: 31st Open Access Grant Competition; OPEN-31-33

Researcher: Tomáš Blejchař

Institution: IT4Innovations

Field: Engineering

The project of Tomáš Blejchař from IT4Innovations focuses on improving the design of vertical pumps using artificial intelligence and machine learning. Pumps are among the most used machines in industry and everyday life. Their efficient design is critical to reducing energy consumption, which is especially important for pumps running continuously, such as water and wastewater treatment plants, heating and air conditioning systems, etc.

Pump design is a complex process influenced by many input parameters and involves manual calculations followed by numerical simulations. This project aims to use neural networks and artificial intelligence to simplify pump design, specifically to automatically correct the initial rough calculations, allowing the optimal pump design to be found in a single step. This will save not only the energy and computational resources required for iterative numerical simulations but also the time required to find the optimal pump design.

This research is closely related to the project funded by the Trend programme of the Technology Agency of the Czech Republic (FW10010202).

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

Call: 31st Open Access Grant Competition; OPEN-31-51

Researcher: Pavel Hobza

Institution: Institute of Organic Chemistry and Biochemistry
of the Czech Academy of Sciences and IT4Innovations

Field: Material sciences

The project of Professor Pavel Hobza from the Institute of Organic Chemistry and Biochemistry of the CAS and IT4Innovations focuses on studying the effect of solvent polarity on the stability of covalent dative and non-covalent complexes.

To analyse these interactions, Professor Hobza's team will use the Karolina, Barbora, and LUMI supercomputers and modern laboratory methods. Solvation, the envelopment of ions by solvent molecules, plays a crucial role in supramolecular chemistry.

The research aims to gain a deeper understanding of the chemical bonds and interactions in different solvent environments, leading to the development of new technologies and materials necessary for solvation processes.

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

Call: 31st Open Access Grant Competition; OPEN-31-8

Researcher: František Karlický

Institution: University of Ostrava

Field: Material sciences

František Karlický and his team of nanostructure physicists from the University of Ostrava (nano.osu.cz) focus, among other things, on two-dimensional (2D) materials that have the potential for flexible and ultrathin functional devices such as future nanoelectronics and solar cells. They also investigate van der Waals (vdW) heterostructures formed by layering 2D materials and have unique electronic and optical properties.

They use computationally intensive multiparticle methods to simulate physical phenomena that traditional methods cannot capture. This research, funded by the Just Transition Operational Programme (LERCO), will contribute to a better understanding of vdW heterostructures and support further experimental research and technological applications.

Foundational Czech Language Model

Call: 31st Open Access Grant Competition; OPEN-31-52

Researcher: Petr Marek

Institution: Czech Technical University in Prague

Field: Informatics

Petr Marek from the Czech Technical University in Prague (CTU) plans to use the computational resources of the Karolina and LUMI supercomputers to develop a modern language model for the Czech language that will be able to understand, generate, and interact with written text.

This project will focus on creating a basic Czech language model with an emphasis on maximum efficiency, scalability, and performance, using state-of-the-art computing techniques and platforms such as CUDA, the open-source Transformers library from Hugging Face, and Hugging Face Accelerate.

The goal is also to eventually create a bilingual Czech-English version of the model, which could be extended to other Central European languages, inspired by advanced Phi language models. The CTU researchers plan to use a large dataset containing 350 billion Czech tokens.

In-silico insight into endosomal escape

Call: 31st Open Access Grant Competition; OPEN-31-26

Researcher: Martin Šrejber

Institution: Palacký University Olomouc

Field: Biosciences

Liposome delivery systems, especially lipid nanoparticles (LNPs), have played a key role in cancer chemotherapy, gene therapy, pharmacotherapy, and vaccine research in recent decades. LNPs play a key role in the targeted delivery of mRNA to vaccines, where they form a lipid carrier for mRNA fragments, thereby ensuring their safe transport into human cells. The project of Martin Šrejber from CATRIN at Palacký University in Olomouc aims to use the Karolina and LUMI supercomputers to investigate the complex processes involved in targeted drug delivery to the body using state-of-the-art computational methods. The research will focus on investigating the interaction of lipid nanoparticles with the plasma membrane, the process of endosome maturation, and the subsequent release of mRNA fragments. Endosomes are membrane "vesicles" formed by endocytosis and are crucial for transporting substances into the cellular environment. In the context of lipid nanoparticle drug delivery, the endosomal release process is critical for successful delivery of mRNA into the cellular environment. This research is part of the MINIGRAPH project funded by the European Union's Horizon Europe programme.

Optimizing ARPES experiments through AI: Achieving precise photon beam polarization with graphene calibration

Call: 31st Open Access Grant Competition; OPEN-31-35

Researcher: Ridha Eddhib and Ján Minár

Institution: University of West Bohemia

Field: Material sciences

Graphene and polarization detection: The figure shows the MAX 4 beamline setup, illustrating the key optical components involved. The probed Brillouin zone is highlighted, with a photon energy of 22 eV. The right side presents experimental circular dichroism patterns for different angles (alpha = 22, 37, and 45 degrees) using right- and left-circularly polarized light (RCP and LCP). These experimental results are compared with theoretical predictions, demonstrating good agreement across the different angles.

The EuSpecLab-ML supercomputing project, led by Prof. Ján Minár and his PhD student Ridha Eddhib from the New Technologies Research Center (NTC), University of West Bohemia, is set to revolutionize Angle-Resolved Photoemission Spectroscopy (ARPES) by harnessing the power of Machine Learning (ML) techniques. ARPES is a critical technique in condensed matter physics, providing deep insights into the electronic structures and quantum states of materials.
However, the precision of ARPES measurements critically depends on the accurate calibration of the photon’s beam polarization – a challenge that has long confronted experimentalists.

In this project, researchers aim to achieve 100% polarization accuracy by developing an advanced Neural Network (NN) model, based on PyTorch library. This model will be meticulously trained on datasets generated by the sophisticated one-step model of the SPRKKR (Spin-Polarized Relativistic Korringa-Kohn-Rostoker) code, renowned for its accurate ARPES simulation capabilities. The ase2sprkkr package will be employed for seamless interfacing with SPRKKR, enriching the AI model’s training data with high-fidelity simulations of graphene – a 2D material celebrated for its exceptional electronic properties and pivotal role in material science. To streamline and enhance the efficiency of the model training process, the project introduces SparkkFLOW, a novel workflow tailored for managing and analyzing ARPES simulation data. Coupled with the JobMaster workflow manager, SparkkFLOW enables the automated generation, organization, and sorting of simulation data, significantly reducing the time and computational resources required for model training. This integration of AI with cutting-edge computational tools allows for real-time prediction of optimal ARPES experimental conditions, ensuring that each measurement is conducted under the most precise polarization conditions.

The research leverages the high-performance computing resources of the supercomputer Karolina, part of the European High-Performance Computing Joint Undertaking (EuroHPC JU). The ultimate goal is to provide experimentalists with a robust tool to dynamically adjust their ARPES setups, thereby optimizing beamline time and enhancing efficiency at synchrotron facilities. Funded by the MSCA Horizon project and OPJAK, forming part of the EuSpecLab Doctoral Training Network, this research not only aims to elevate the quality and accuracy of ARPES experiments but also opens new avenues for exploring the electronic behaviors of advanced materials, including novel 2D systems and quantum materials. project is funded by OPJAK: Researchers are thankful for the support of the QM4ST project funded by Programme Johannes Amos Commenius, call Excellent Research, (Project No. CZ.02.01.01/00/22 008/0004572).