6th Users' Conference of IT4Innovations will take place on 3 and 4 November 2022. All of our users, as well as research and project partners from various organizations, research institutions, and industry, are welcome to attend the Conference.
The submission of contributions has been closed.
Registration is open until 25 October 2022.
Attendees will discover more about our future upgrade plans, listen to talks given by our prominent users, and can engage in discussions during the Users' Council meeting and a poster session.
Contribution types: Users' talks/Posters
Selected talks by our prominent users will be presented at scheduled times during the whole conference. Each talk is expected to take max. 20 mins (with discussion included).
Please note that the required poster size is A1 portrait orientation.
The poster session will take place on 3 November during lunch and dinner.
The extracellular space is a very dynamic and complex environment, especially above the plasma membrane. For decades, theoreticians and experimentalists have been trying to describe it piece by piece, i.e., grasp the idea of how cell membranes, proteins, and polysaccharides interact with each other. However, the complexity and functional length scales have made until now unfeasible to model a significant portion of the extracellular interaction network. The main limitations were the lack of computational power required for such simulations, the absence of uniform and accurate simulation models, and the large uncertainty of which molecular motifs should be modeled.
In our work, using the computational power provided by LUMI supercomputer via IT4Innovations together with recently improved all-atom molecular models of major classes of biomolecules, we modeled the outer-cell network on an unprecedented scale, combining in a single set of simulations all key components simultaneously (i.e., lipids, proteins, sugars, ions, and water) modeled at full atomic resolution. Our model but realistic systems, ranging from the lipid membranes with thoroughly detailed lipid composition to huge lipid membranes/glycocalyx complexes containing a wide variety of biologically relevant molecules, account for millions of atoms simulated at the microsecond timescale. The preliminary results include structural and interaction patterns present in such systems providing unique molecular insights into the organization and communication of biomacromolecules in realistic environments. Moreover, our large-scale molecular dynamics simulations offer the first of its kind opportunity to visualize in silico numerous biologically relevant processes at the atomic level.
RNA-based therapy covers a wide range of applications, from cancer therapy, treatment of inherited diseases up to vaccination. The encapsulation of RNAs into ionizable lipids (ILs) containing lipid nanoparticles (LNPs) enabled their safe targeted delivery, but for design of more efficient RNA-LNPs, molecular understanding of LNP-structure and IL-RNA interactions is needed.
Here we simulated LNPs with composition corresponding to currently developed covid-19 vaccines. We observed, that in conditions acidobasically representing LNP preparation (acidic pH), lipids assembled around RNA into a hexagonal phase and formed a rough, non-spherical hydrated LNP.
The change of pH to neutral conditions, representing the environment after LNP administration into human tissues drastically affected the morphology of LNPs. Immediately after IL deprotonation, LNP became spherical and in microseconds time scale, they got gradually dehydrated. Further, we observed lipid separation and creation of an IL-rich phase and a phase rich in ordered saturated phospholipids, that organized to be in contact with water or RNA. During the simulations in neutral conditions, majority of RNA was expelled from LNP, as it lost the favorable electrostatic interactions by neutralization of ILs and the RNA remaining inside LNP was surrounded by helper phospholipid.
We present here the first ever simulation of self-assembly of LNPs containing ILs and their internal morphology. The simulations were performed combining coarse-grained and atomistic resolutions, capturing the dynamics of systems with up to ~3 million of atoms. The simulations support the hypothesis of a solid electron-rich hexagonal phase in the LNP which hosts the RNA. We also give evidence for the fragility of LNPs in neutral conditions, that can explain still very limited efficiency in RNA delivery.
Computational treatment of ligand:protein interactions represents an old, but yet unsolved challenge in theoretical chemistry. The hybrid quantum mechanical and molecular mechanical (QM/MM) approach, combined with X-ray crystallography, molecular docking and conformational sampling, seems to be the most promising approach for the computational ligand (drug) design. The accuracy of the QM/MM approaches, though, comes at a price of huge computational resources requested by the calculations which are available in supercomputer centres.
Specifically, within the OPEN-15-18 IT4I project, we focused on the stimulator of interferon genes (STING), a protein that is central to the immune system and whose activation and inhibition represents an underexplored therapeutic target in medicinal chemistry. The natural ligand of the STING is the cyclic dinucleotide 2′,3′-cGAMP (cyclic [G(2′,5′)pA(3′,5′)p]), which is synthetized by the cyclic- GMP-AMP synthase, upon detection of the double-stranded DNA in the cytosol which then initiates the immune response.
Our computational efforts, carried out at the IT4I supercomputer center, aimed at (1) the quantitative evaluation and interpretation of the observed STING:CDN binding constants (CDN= cyclic dinucleotide), (2) elucidation of the structural and energetic details accompanying the STING ligand binding and ultimately to (3) predictions of new molecular scaffolds to synthesized and tested experimentally. [1-7]
 Gutten, O. et al. Phys. Chem. Chem. Phys. 2021, 23, 7280-7294.
 Novotná, B. et al.: Enzymatic Preparation of 2‘-5‘, 3‘-5‘Cyclic Dinucleotides, Their Binding Properties to Stimulator of Interferon Genes Adaptor Protein, and Structure/Activity Correlations. J. Med. Chem. 2019, 62, 10676-10690.
 Pimková Polidarová, M.; Břehová, P.; Kaiser, M. M.; Smola, M.; Dračínský, M.; Smith, J.; Marek, A.; Dejmek, M.; Šála, M.; Gutten, O.; Rulíšek, L.; Novotná, B.; Brázdová, A.; Janeba, Z.; Nencka, R.; Boura, E.; Páv, O.; Birkuš, G.*: Synthesis and Biological Evaluation of Phosphoester and Phosphorothioate Prodrugs of STING Agonist 3′,3′-c-Di(2′F,2′dAMP). J. Med. Chem. 2021, 64,7596–7616
 Smola, M.; Gutten, O.; Dejmek, M.; Kožíšek, M.; Evangelidis, T.; Aliakbar Tehrani, Z.; Novotná, B.; Nencka, R.; Birkuš, G.; Rulíšek, L.; Boura, E.: Ligand Strain and Its Conformational Complexity Is a Major Factor in the Binding of Cyclic Dinucleotides to STING Protein. Angew. Chem. Int. Ed. 2021, 60, 10172–10178.
 Vavřina, Z.; Gutten, O.; Smola, M.; Zavřel, M.; Aliakbar Tehrani, Z.; Charvát, V.; Kožíšek, M.; Boura, E.; Birkuš, G.; Rulíšek, L.*: Protein-Ligand Interactions in the STING Binding Site Probed by Rationally Designed Single-Point Mutations: Experiment and Theory Biochemistry 2021, 60, 607 620.
 Aliakbar Tehrani, Z.;* Rulíšek, L.; Černý, J.: Molecular dynamics simulations provide structural insight into binding of cyclic dinucleotides to human STING protein. J. Biomol. Struct. Dyn. 2021
 Novotná, B.; Holá, L.; Staś, M.; Gutten, O.; Smola, M.; Zavřel, M.; Vavřina, Z.; Buděšínský, M.; Liboska, R.; Chevrier, F.; Dobiaš, J.; Boura, E.; Rulíšek, L.; Birkuš, G.*: Enzymatic Synthesis of 3′–5′, 3′–5′ Cyclic Dinucleotides, Their Binding Properties to the Stimulator of Interferon Genes Adaptor Protein, and Structure/Activity Correlations. Biochemistry 2021, 60, 3714–3727.
Molecular crystals are known for the delicate interplay of non-covalent interactions governing the molecular packing and cohesion. This work exploits first-principles methods to predict, quantify and interpret structural and thermodynamic phenomena related to crystalline biogenic carboxylic acids. Going beyond the established methods for treatment of molecular crystals focusing on sublimation or polymorphism, this works focuses also on phenomena such as local disorder and related configurational entropy, and local anisotropy.
Crystalline carboxylic acids are known to exhibit a considerable anisotropy of their thermal expansion, which is interpreted in this work using fragment-based calculations of the cohesion of these crystals. SAPT calculations are then used to decompose the pair interactions, and to identify the directions of the strongest cohesion, imposed by the site-specific and spatially-oriented hydrogen bonding, restricting the thermal expansion in particular directions.
In addition, two polymorphs are currently known for each of succinic acid, fumaric acid and malic acid. Various ambiguities have prevailed in the literature on their experimental ranking of the relative thermodynamic stability. Quasi-harmonic density functional theory calculations, refined with the ab initio fragment-based cohesive energies and a first-principles model of the configurational entropy, related to the static disorder of carboxyl hydrogen atoms is used to rank these polymorphs in silico. The configurational entropy is demonstrated to shift the polymorph ranking in the correct direction, mostly yielding qualitatively correct ranking. Still, the tiny sub-kilojoule per mole differences of the chemical potentials of the polymorph pairs render reaching a semi-quantitative agreement of theory and experiment in this field extremely challenging.
Ribosome is a huge biomolecular complex responsible for protein synthesis in all living organisms. As a multi-component system consisting of several strands of RNA and a few dozen proteins, the ribosome is particularly challenging for current molecular dynamics simulations . In the past few years, we carried out many atomistic computer simulations of the entire ribosome in explicit aqueous environment to better understand how nascent proteins emerge through the ribosome exit tunnel to the cytosol. We have learned that the tunnel offers a unique conditions to form a secondary structure of certain nascent protein sequences causing translational arrest . We studied how the information about ribosome surface is transferred to its interior . Most recently we have described the structure and dynamics of the narrowest part of the ribosome exit tunnel, the function of which has remained under debate for decades. The talk will summarize our simulation efforts to understand the ribosome as a crucial player in the life as we know it.
Phylogenomics has proven to be indispensable for inferring the eukaryotic tree of life. During the ongoing project, we are using Phylofisher software for proper ortholog detection, contamination removal, and reconstruction of phylogenetic relationships within the neglected eukaryotic major group Heterolobosea. This group contains free-living as well as facultatively parasitic microbial eukaryotes including genus Naegleria; Naegleria gruberi is an important cell biology model and N. fowleri is a facultative human pathogen responsible for fatal primary amoebic meningoencephalitis (PAM). During the project, we also aim to trace origin of anaerobic enzymes present in Naegleria and their possible homologs in anaerobic lineages of Heterolobosea and to reconstruct energy metabolism of anaerobic lineages, to understand evolutionary changes that these linaeges underwent during transition to anaerobiosis.
Computing for Large Hadron Collider experiments is based on the idea of distributed computing in centers located all around the world. Supporting computing centers were mostly adjusted for needs of the Worldwide LHC Computing Grid (WLCG). The existing capacities were extended by large supercomputing centers in the USA and Europe. We will describe different obstacles which we had to resolve to be able to use these HPC centers effectively. The examples will be based on the experience of the ATLAS experiment, which is the most CPU consuming project in the WLCG. We will discuss software installation in the relatively close environment of HPC centers, experience with different CPU architectures and procedures to effectively run jobs on systems with many cores.
There is overwhelming evidence that new power sources and more complex geometries are strictly necessary for explaining the observed light curves of certain astrophysical transients, e.g. supernovae in the presence of surrounding medium. Numerical models are challenging due the need of performing radiation-hydrodynamic simulations across a wide dynamic range of both time and space, which typically implies an elevated computational cost. In this context, we have developed a new tool that solves the multi-dimensional radiation-hydrodynamic equations in spherical coordinates over a domain that is allowed to expand radially. The equations are solved using the operator-splitting technique in three steps: an explicit update using the Godunov method for the hydrodynamic hyperbolic equations, an explicit update to account for radiation sink/source terms, and an implicit update for radiation diffusion, emission, and absorption processes. We have shown that this approach is ideal for simulating astrophysical transients over ten orders of magnitude spatially and temporarily even if the sources lack of spherical symmetry . We present the implementation, results of our first application: wind-reprocessed transients, and also ongoing work simulating supernovae and tidal-disruption events. Part of the content is based on our published peer-reviewed article: Calderón D., Pejcha O., Duffell P. C., 2021, Monthly Notices of the Royal Astronomical Society, 507, 1092. doi:10.1093/mnras/stab2219
This talk/poster investigates the performance of the distributed k-Wave toolbox on Barbora, Karolina and LUMI clusters. The solver uses 1D decomposition of the k-space corrected pseduospectral method heavily employing the distributed FFTW library. The goal is to show the strong and weak scaling on domains up to 3072^3 grid points (3.6TB of RAM) using up to 3072 compute cores. The comparison also investigates the computation/communication ratio and several low level performance statistics.
Thibault J.-Y. Derrien, Nadezhda M. Bulgakova
HiLASE Centre, Institute of Physics (AS CR), Za Radnici 828, 252 41 Dolni Brezany, Czech Republic
Ultrashort pulse laser processing of dielectric materials by ultrashort laser pulses has led to a number of valuable commercial applications such as nanograting formation (1) employed for fabrication of polarizers, waveguide direct writing (2), and for making quantum photonic devices for cryptography (3).
In these examples, the excitation of electrons by a linearly polarized field within the bulk of a wide band gap solid plays a key role in the control of modified material properties. While experimental aspects have been already well explored, the mechanisms of material modification, electron excitation routes, and the effect of light polarization remain not well quantified, in particular for different types of polarization. In this context, numerical modeling is an important tool to describe the deposition of energy in a quantitative
manner (4, 5).
So far, the case of linear polarization has been widely treated from the pioneering development of a two-bands model (6) and its improvements (7) to their comprehensive comparison with realistic multi-band approaches (8, 9) that provide first-principle excitation rates in the strong field regime (9, 10), the consequences of other polarization states have been less explored in the context of laser processing and still require an assessment of new models (11) in comparison to more advanced simulations taking into account a full three-dimensional band structure (12). In this work, we use the density functional
theory (DFT) and its time-dependent implementation (TDDFT) (13) to explore the mechanisms of excitation of wide band gap materials on the example of quartz crystal (14) illuminated by an intense beam with a nonlinear polarization in comparison with linearly polarized light.
The PALM model system  is an open-source, HPC-enabled, extensible large-eddy simulation (LES) modelling system which includes components for many related processes that are necessary for a complex environmental modelling in micro-scale, such as resolved urban climate simulations in the street level . These simulations are vital for studying the adverse effects of urbanization and climate change such as air pollution and heat stress and they enable the urban planners to take into account and mitigate these effects . One of the most important processes in micro-scale atmospheric modelling is radiation and its interaction with surfaces, plant canopy etc. which provides thermal energy that powers most atmospheric processes, especially in the boundary layer.
The radiative transfer model (RTM, ) in PALM provides fully resolved, explicit 3-D simulation of short-wave and long-wave radiation that includes multiple reflections, shading, absorption and emission, interaction with semi-transparent plant canopy and other processes, and it is fully integrated in the model, working on a common 3-D grid, time-steps, data structures and parallelization via horizontal domain decomposition. Such simulation is computationally expensive and it presents challenges to parallelization, because the immediate interactions are not limited to neighbouring sub-domains.
The RTM is centred around the radiosity approach and it uses pre-calculated view factors as the basis for interaction among discretized surface elements, plant canopy grid cells and other objects of radiation. The most computationally intensive task is the calculation of the view factors, which is performed as part of the pre-processing step and it uses a custom column-integrated raytracing algorithm, which is optimized with respect to the discretization scheme and the horizontal domain decomposition.
Over the last few years, the RTM has received numerous improvements which greatly enhanced its applicability by adding new simulated processes and subjects of radiation. As a result of substantial improvements of algorithms and representation schemes, it features excellent scalability and it has been successfully used for extremely large simulations with tens of thousands of parallel processes. Many of these improvements have been tested and validated on the IT4I supercomputers Salomon and Karolina . An important and significant improvement is the recently developed new parallelization scheme for the raytracing algorithm. By reorganizing the parallel computation and using new patterns for data exchange among the MPI processes, it allows to avoid certain problematic MPI calls as well as some large arrays, thus greatly enhacing scalability. It is currently being tested on several HPC systems including Karolina. First tests show that it can speed up raytracing by an order of magnitude in certain scenarios. This presentation shows the new paralelization scheme and other new computational improvements in the RTM.
Recently, self-supervised Transformer-based models have become an integral part of state-of-the-art speech modeling and are being integrated into many speech applications such as Automatic Speech Recognition (ASR), Speaker Verification (SV), Language Identification (LID), emotion detection, etc. These models are trained on datasets comprising tens or even hundreds of thousands of speech and can reach several hundreds of millions of parameters. In my talk, I will briefly overview their architecture and a self-supervised training paradigm based on masked speech prediction. Later on, I will describe a use case in speaker verification where we use these already pre-trained models, which we subsequently fine-tune to serve as powerful feature extractors for speaker embedding extraction. I will also discuss methods that can be employed for fine-tuning such large models when there is only a relatively small amount of target and labeled data available.
Uranium hydrides are not only materials necessary for an understanding of fundamental aspects of actinides. They are also relevant for nuclear technologies as well as for specific hydrogen storage tasks, e.g. storing tritium in nuclear fusion devices. The electronic structure of uranium hydrides (α- and β-UH3, UH2) reflects two contradictory tendencies. One is a charge transfer from U towards H, the other is the stability of the f shell. Here, based on the first-principle calculations we reveal the phase stability of three of them: UH2, Alfa-UH3, and Beta-UH3 with respect to the relative occupancy of the majority f-states. The transfer is thus realized in U by the 6d and 7s electrons, which become noticeably depleted, but the 5f occupancies remain high, together with the volume expansion, to pronounced ferromagnetism of the U hydrides with Curie temperatures far above 100 K. We found a relatively simple model of the electron structure that is able to correctly and still accurately model the elastic, magnetic (spin, orbital moments) properties as well, as reveal the lattice dynamics properties for further spectroscopic measurements. Theoretical calculations reveal for the first time the non-collinear ferromagnetic order of the one of Uranium sites in Beta-UH3 and allow the estimate of the Curie temperatures in reasonable agreement with the experiment.
1. L. Kyvala, L. Havela, A. Kadzielawa, D. Legut, J. Nucl. Mater. 567, 153817 (2022), doi.org://10.1016/j.jnucmat.2022.153817
The Non-Covalent Interactions Atlas is a database gathering next-generation computational chemistry benchmark data sets covering wide range of non-covalent interactions. The data sets are openly available in an on-line repository at www.nciatlas.org.
The aim of the project is to map all the important classes of non-covalent interactions with high-quality data sets significantly exceeding the previous state of the art in multiple aspects: 1) Size, with hundreds of systems for each class of interactions; 2) Accuracy, with true gold-standard CCSD(T)/CBS interaction energies extrapolated from large basis sets; 3) Wider coverage of chemical space; and 4) Quality, availability and usability of the data.
The database now contains seven data sets covering all important classes of non-covalent interactions in extended organic chemical space with almost 20k data points.
These data sets have been used to validate the accuracy of wide range of approximate computational methods, but their main purpose is the development of next-generation methods bringing better accuracy to broader chemical space.
Activation of dioxygen attracts enormous attention due to its potential for utilization of methane and applications in other selective oxidation reactions. Employing periodic DFT calculations we discovered a cleavage of dioxygen over distant binuclear Fe(II). Our experiments confirmed splitting dioxygen at room temperature and showed that pairs of the formed distant Alpha-oxygen species [i.e., (Fe(IV)═O)2+] exhibited unique oxidation properties reflected in an outstanding activity in the oxidation of methane to methanol at room temperature. Designing a man-made system that mimicks the enzyme functionality in the dioxygen activation using both a different mechanism and structure of the active site represents a breakthrough in catalysis.
Professor Yuriy Roman-Leshkov of MIT about our discovery:
"Recently, Tabor et al. have observed the formation of Fe(IV)=O from O2 at room temperature by a pair of distant Fe(II) centers stabilized in the matrix of a zeolite. This delicately designed binuclear Fe(II) site converts methane into methanol at room temperature using O2 as oxidant, representing a breakthrough in methane oxidation catalysis."
"The direct methane-to-methanol conversion has been considered as one of the "holy grail" reactions in the field of catalysis."
The authors of the discovery were awarded the Czech science prize "Česká hlava Cena Invence“ in 2020.[3, 4]
In our contribution, we present the results of follow-up periodic DFT calculations used to investigate the effect of (i) the Al siting in the rings forming the cationic sites, (ii) the distance, (iii) the mutual geometrical position of the rings accommodating Fe(II), and (iv) the type of transition metal cation (Me(II) = Co(II), Mn(II), and Fe(II)) accommodated in the active centers on the activity of the distant binuclear Me(II) sites in splitting dioxygen. The outcome of our work reveals that employing periodic DFT and supercomputers we can suggest new materials which when experimentally prepared and tested have a high chance of being successful catalysts.[5-7]
(1) Tabor, E.; Dedecek, J.; Mlekodaj, K.; Sobalik, Z.; Andrikopoulos, P. C.; Sklenak, S. Dioxygen Dissociation over Man-Made System at Room Temperature to Form the Active Alpha-Oxygen for Methane Oxidation. Science Advances 2020, 6, eaaz9776.
(2) Yuan, S.; Li, Y. D.; Peng, J. Y.; Questell-Santiago, Y. M.; Akkiraju, K.; Giordano, L.; Zheng, D. J.; Bagi, S.; Roman-Leshkov, Y.; Shao-Horn, Y. Conversion of Methane into Liquid Fuels-Bridging Thermal Catalysis with Electrocatalysis. Advanced Energy Materials 2020, 10, 2002154.
(3) https://www.ceskahlava.cz/ceska-hlava/vitezove/tym-mgr-jiriho-dedecka-csc-dsc/ (accessed 04-25-2022).
(4) https://www.it4i.cz/o-it4i/infoservis/novinky/vyzkum-s-prispenim-superpocitacu-it4innovations-ziskal-cenu-ceska-hlava (accessed 04-25-2022).
(5) Dedecek, J.; Tabor, E.; Andrikopoulos, P. C.; Sklenak, S. Splitting Dioxygen over Distant Binuclear Transition Metal Cationic Sites in Zeolites. Effect of the Transition Metal Cation. International Journal of Quantum Chemistry 2021, 121, e26611.
(6) Mlekodaj, K.; Lemishka, M.; Sklenak, S.; Dedecek, J.; Tabor, E. Dioxygen Splitting at Room Temperature over Distant Binuclear Transition Metal Centers in Zeolites for Direct Oxidation of Methane to Methanol. Chemical Communications 2021, 57, 3472-3475.
(7) Tabor, E.; Lemishka, M.; Olszowka, J. E.; Mlekodaj, K.; Dedecek, J.; Andrikopoulos, P. C.; Sklenak, S. Splitting Dioxygen over Distant Binuclear Fe Sites in Zeolites. Effect of the Local Arrangement and Framework Topology. ACS Catalysis 2021, 11, 2340-2355.
One of the main difficulties in the control of the nanoscale friction is represented by the complexity of non-equilibrium processes occurring in tribological contacts.
To understand the origin of the nanoscale friction and to design new tribological materials, we conduct quantum mechanical studies on Transition Metal Dichalcogenides and TiO2 surfaces, chosen as prototypical materials. We combine structural and dynamic information from group theory and phonon calculations with the electronic density characterization from orbital polarization, bond covalency and cophonicity analyses. We define a phonon-based method to identify and tune possible sliding paths, corresponding energy barriers and dissipation channels. The method allows to extract energy information from atomic force microscopy signals. We identify electrostatic and electromagnetic fields as possible external knobs which allow the fine tuning of the nanofrictional behaviour; the latter case corresponds to a new field of research named "photofriction". We finally formulate guidelines on how to engineer the intrinsic friction at the nanoscale in order to design novel materials with controlled tribological properties.
Thanks to the general formulation of the proposed analysis, the present outcomes can be promptly extended to the design of new materials with diverse applications beyond tribology.
Martin Friák1, Martin Zelený2,3, Martina Mazalová1, Ivana Miháliková1, Jiří Kamarád4, Jiří Kaštil4, Ilja Turek1, Oldřich Schneeweiss1, Mojmír Šob5,1
1 Institute of Physics of Materials, v.v.i., Czech Academy of Sciences, Žižkova 22, CZ-616 62 Brno, Czech Republic;
2 Institute of Materials Science and Engineering, Faculty of Mechanical Engineering, Brno University of Technology, Technická 2896/2, Brno, 61669, Czech Republic;
3 Faculty of Mathematics and Physics, Charles University, Ke Karlovu 5, Prague, 12116, Czech Republic;
4 Institute of Physics, v.v.i., Czech Academy of Sciences, Na Slovance 2, Prague, 18221, Czech Republic
5 Department of Chemistry, Faculty of Science, Masaryk University, Kotlářská 2, CZ-611 37 Brno,Czech Republic;
We have performed a quantum-mechanical study of thermodynamic, elastic, magnetic and structural properties of four different ferrimagnetic states in Ni1.9375Mn1.5625Sn0.5 martensite. They are modeled by the four-layer modulated 4O structure with Mn-excess atoms randomly distributed in Ni and Sn sublattices. The Mn atoms at the Ni sublattice turn out to be decisive for both thermodynamic and magnetic properties. A reversal of the orientation of their local magnetic moments has a huge impact on the properties of the whole system. The lowest-energy configuration exhibits anti-parallel local magnetic moments of these Mn atoms with respect to the orientation of the total magnetic moment. By testing both elastic properties and phonon modes we conclude that the lowest-energy state is mechanically stable. Magnetic as well as vibrational properties of individual atoms are found to be very sensitive to the chemical disorder. For details see M. Friák et al., Intermetallics (2022) accepted for publication.
The authors acknowledge the Czech Science Foundation for the financial support received under the Project No. 20-16130S entitled “Multifunctional properties of powdered Ni-Mn-Sn Intermetallics”. Computational resources were made available by the Ministry of Education, Youth and Sports of the Czech Republic under the Project of the IT4Innovations National Supercomputer Center (project “e-Infrastructure CZ-LM2018140”) within the program Projects of Large Research, Development and Innovations Infrastructures and via the CESNET (Project No. LM2015042) and CERIT-Scientific Cloud (Project No. LM2015085).
Neural network potentials (NNPs) are becoming increasingly popular in multiple areas of material science and chemistry thanks to their ability to keep ab initio accuracy at the cost of the standard reactive force fields such as ReaxFF [1,2]. However, the vast majority of the studies focused on systems with rather low-dimensional configurational space. This apparent curse of dimensionality could be behind the fact that NNPs for microporous solids such as zeolites are, to the best of our knowledge, non-existent [1,2].
In this work, we present the development of linear-scaling, reactive NNPs using the SchNet architecture  for various zeolite classes by using robust training and data curation procedures. The resulting NNPs retain DFT accuracy across the complex zeolitic classes considered, outperforming specialized ReaxFF by order(s) of magnitude in accuracy, while speeding-up calculations by at least three orders of magnitude compared to DFT. Using the developed NNPs we have been witnessing intriguing results such as: i) large-scale simulations of zeolite databases (>330k hypothetical zeolites) revealing more than 20k additional hypothetical frameworks in the thermodynamically accessible range for zeolite synthesis ; ii) multi-nanosecond molecular dynamics (MD) simulations of reactive diffusion of sub-nanometer Pt clusters in zeolites showing intermittent bond breaking of the zeolite framework ; iii) effects of minor topology variations on the germanium distribution in germanosilicate zeolites with profound effects on their delamination propensity [6,7]; iv) MD simulations of aluminosilicate zeolites quantifying the effects of water loadings and Si/Al ratios on proton solvation and water diffusivity . Lastly, our recent work  suggests that the learned NNP representations of atomic environments, a by-product of our NNP generation, can be reused to construct (using variational autoencoders) robust collective variables (CVs) that are "aware" of the underlying potential energy surfaces - such CVs will facilitate any subsequent biased MD simulations that generate realistic free energy surfaces of reactive processes in zeolitic systems. Therefore, the herein developed, reactive zeolite NNPs enable calculations of complex zeolitic frameworks under experimentally relevant - operando - conditions making them a new standard in the field.
 N. Artrith, J. Phys. Energy 2019, 1, 032002.
 O. T. Unke, S. Chmiela, H. E. Sauceda, M. Gastegger, I. Poltavsky, K. T. Schütt, A. Tkatchenko, K.-R. Müller, Chem. Rev. 2021, 121, 10142–10186.
 K. T. Schütt, H. E. Sauceda, P.-J. Kindermans, A. Tkatchenko, K.-R. Müller, J. Chem. Phys. 2018, 148, 241722.
 A. Erlebach, P. Nachtigall, L. Grajciar, npj Comput. Mater. 2022, 8, 1–12.
 D. Hou, L. Grajciar, P. Nachtigall, C. J. Heard, ACS Catal. 2020, 10, 11057–11068.
 M. Opanasenko, M. Shamzhy, Y. Wang, W. Yan, P. Nachtigall, J. Čejka, Angew. Chem. 2020, 132, 19548–19557.
 M. Jin, O. Veselý, C. J. Heard, M. Kubů, P. Nachtigall, J. Čejka, L. Grajciar, J. Phys. Chem. C 2021, 125, 23744–23757.
 I. Saha, A. Erlebach, P. Nachtigall, C. J. Heard, L. Grajciar, preprint: ChemRxiv. Cambridge: Cambridge Open Engage 2022, https://doi.org/10.26434/chemrxiv-2022-d1sj9.
 M. Šípka, A. Erlebach, L. Grajciar, J. Chem. Theory Comput. (under revision) 2022, preprint: http://arxiv.org/abs/2203.08097.
Atomistic simulations are the most used theoretical methods to study mechanical properties at nano-scale level. However, the most frequently used molecular dynamics (based on Newton mechanics) is limited by the reliability of available interatomic potentials while the much more precise ab initio methods (based on quantum mechanics) can be performed only for relatively small computational cells and at absolute zero temperature. To overcome these limitations, one can use the ab initio molecular dynamics with on-the-fly machine learning as it is implemented in the current version of the Vienna Ab initio Simulation Package (VASP). This allows to perform molecular dynamics simulations with almost ab initio precision and overcome the limitations of both mentioned approaches.
In this work, we demonstrate how the aforementioned tool is used for segregation study of Sn, P and Ge atoms at selected GBs in bcc iron. In the first stage, segregation characteristics were obtained using small supercells containing up to 100 atoms or medium cells containing approximately 400 atoms and compared with the data obtained from the ab initio simulations. This comparison served as a benchmark of the machine learned force fields (MLFFs) obtained from the VASP on-the-fly machine learning. The comparison shows that the computed segregation energies and other GB characteristics determined via MLFF are very consistent with those obtained from the ab initio simulations. This provides a proof of reliability of the obtained MLFF. After these benchmarks, we employed significantly larger atomistic models capable of describing low impurity concentration at finite temperatures. The received results clarify how simulation cell size might affect the GB characteristics, shed light into temperature influence and provide information about the impurity segregation for many GB types.