8th Users' Conference of IT4Innovations

4 Nov 2024, 10:00 → 5 Nov 2024, 14:00 Europe/Prague

atrium (IT4Innovations)

8^th Users' Conference of IT4Innovations will take place on 4 and 5 November 2024. All of our users, as well as research and project partners from various organisations, research institutions, and industry, are welcome to attend the Conference.

Find the conference programme in PDF format, along with the list of poster presentations, at the bottom left corner of this page.

The submission of contributions deadline is 11 September 2024.

Registration is open until 29 October 2024.

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

Users' talks

Selected talks by our prominent users will be presented at scheduled times during the whole conference. Each talk is expected to take a maximum of 20 minutes (with discussion included):

• IT4I Project(s) (5 minutes): Dedicate approximately 5 minutes to provide insights into your computational approach and methods. Discuss software utilisation, job sizes, parallelisation techniques, and the extent of computational resources employed. Sharing your computational experience will provide valuable context for the audience.
• Research (10 minutes): Allocate around 10 minutes to delve into your research endeavours. Highlight key findings, address challenges encountered, and emphasise the significance of your work. This segment should offer a concise yet comprehensive overview of your research area.
• Discussion (5 minutes): Please reserve 5 minutes to engage with questions and discussions from the audience.

Posters

Please note that the required poster size is A1 portrait orientation. 

If you would like IT4Innovations to print and display your poster at the conference, please ensure you send your printing file to pr@it4i.cz by 23 October.

The poster session will take place on 4 November during lunch and dinner.

8th Users Conference of IT4I Programme

How to get to Ostrava and Accommodation tips

Poster presentations

Presentations

• EuroHPC JU Access Calls
• High Level Support Team
• IT4Innovations News

Monday, 4 November

Registration and networking

Welcome and IT4Innovations news

Keynote I

1 Synergy of semiempirical quantum chemistry and machine learning

Semiempirical quantum-mechanical (SQM) methods combine proper quantum-mechanical description of molecular systems with approximations that offer enormous increase of computational efficiency compared to more complex ab initio quantum mechanical or DFT methods. In the past decade, we have worked on improving the description of non-covalent interactions in SQM methods, and our PM6-D3H4X method is one of the top contenders in the field. Despite these successes, our extensive work with SQM methods also revealed their limitations. The most important unsolved issues are non-covalent interactions at short range, and the description of conformation energies.

To overcome these issues, we built a machine learning (ML) model serving as a correction for SQM method PM6 and trained it on an extensive database computed with quality DFT. The resulting PM6-ML method is able to correct all the remaining problems noted above, and it clearly outperforms both all previous SQM methods and standalone ML models in an extensive set of validation data sets covering different phenomena [ChemRxiv 2024]. In comparison to the previous Δ-ML approaches based on SQM methods, PM6-ML covers larger chemical space what makes it applicable to e.g. computer-aided drug design, and takes an advantage of a linear-scaling implementation of the SQM calculation what allows working with systems with thousands of atoms. An implementation of the method is available at [GitHub].

Speaker: Jan Řezáč (UOCHB AV ČR)

Users' Talks I

2 Machine learning assisted phonon calculations of Al-Ge solid solution

The relative stability of different phases, or their mixtures at finite temperatures, can be estimated by comparing their Gibbs free energies. In the case of disordered phases, such as solid solutions, the primary contribution to the free energy comes from configurational entropy, which arises from the number of ways atoms can be arranged in the system. However, other contributions, such as the vibrational contribution to the free energy, which arises from the system's phonon modes, can significantly influence the stability of a particular solid solution.

As an example of such a system, we examine the Al-Ge binary alloy, which exhibits limited solubility of Ge in Al and almost no solubility of Al in Ge. Since ab initio calculations are typically limited to a relatively small number of atoms, reducing the available configuration space for disordered systems, a machine-learning-based forcefield potential for the Al-Ge alloy was developed. This forcefield was trained on data from smaller-scale ab initio calculations. The trained forcefield potential was then used to calculate the phonon density of states at various temperatures and the corresponding vibrational contributions to the Gibbs free energy across a much broader concentration range. By combining these vibrational contributions with configurational entropy in the total Gibbs free energy, we can estimate the phase stability and solubility limits of the alloy. This approach allows for a more comprehensive estimation of solubility limits, considering not only configurational entropy but also vibrational effects, which are essential for accurately describing the behavior of the Al-Ge system.

Speaker: Martin Zelený (Faculty of Mechanical Engineering, Brno University of Technology)

3 Introducing FireCore: Efficient Modeling Of (Small) Organic Molecules On Ionic Substrates

Understanding the self-assembly and binding dynamics of organic molecules on ionic substrates is critical for advancing molecular nanotechnology. General-purpose molecular dynamics codes (such as LAMMPS[1], AMBER[2] or GROMACS[3]) are written and optimized in order to efficiently simulate large systems (e.g. millions of particles and more) via parallelisms such as spatial domain decomposition. On the other hand, they often lack the flexibility required for small systems (~100 atoms) and tailored intermolecular interactions. FireCore[4] is designed to address these limitations by providing a specialized, multiscale modeling platform optimized for the simulation of small organic molecules on rigid ionic substrates.
FireCore integrates several state-of-the-art techniques to offer a flexible and scalable tool for researchers. The code combines tight-binding density functional theory (TB-DFT) via the Fireball package[5], with classical force fields (UFF[6] with automatic typization for the most common atomic species, and a custom force field with an explicit representation of π-orbitals).
Furthermore, FireCore features GridFF, an effective approach to projecting interactions on a grid, enabling fast and accurate simulation of molecule-substrate interactions. One of the highlights of FireCore is its ability to simulate systems using graphical processing unit (GPU) acceleration with a replica-based parallelism, offering significant speedups for small systems (e.g. thousands of replicas on one GPU). This acceleration allows for rapid exploration of molecular configurations, either for global optimization algorithms or for free energy calculations. It also features a spartan but intuitive graphical user interface (GUI). The possibility of performing force field parameterization/refining can also be of interest for researchers active in model development or for tailored applications. In this regard, we are currently working on the development of hydrogen-bond corrections with explicit charges to model lone pairs for a simple yet accurate description of nonbonded interactions. Moreover, FireCore also implements the probe-particle model[7], enabling atomic force microscopy (AFM) simulations for high resolution imaging.
In this presentation, we will show the current state of development of FireCore, and present results from a recent study[8] on molecular self-assembly on ionic substrates, showcasing its potential for advancing molecular nanotechnology.

References
[1] C. Trott, S.J. Plimpton, Comp. Phys. Comm., 271, 10817 (2022).
[2] D.A. Case et al., J. Comput. Chem., 26, 1668 (2005).
[3] D. van der Spoel et al., J. Comp. Chem., 26, 1701 (2005).
[4] https://github.com/ProkopHapala/FireCore
[5] https://www.fireball-dft.org
[6] A.K. Rappe et al., J. Am. Chem. Soc., 114, 10024 (1992).
[7] P. Hapala et al., Phys. Rev. Lett., 113, 226101 (2014).
[8] M. Manikandan, P. Nicolini, P. Hapala, ACS Nano, 18, 9969 (2024).

Speaker: Dr Paolo Nicolini (FZU - Institute of Physics of the Czech Academy of Sciences)

4 Bottom-up machine learning potentials for peptides

Can larger peptides be described just knowing their smaller constituents? Concretely: can we infer the potential energy surface (PES) of a 20-peptide just from the PESs of single amino acids and dipeptides? To answer these long-standing questions,[1] we trained equivariant neural network potentials[2] on oligopeptides of varying sizes (1-3; taken from PeptideCS[3] and P-CONF_1.6M[4] datasets) and tested the performance of these potentials on larger peptides. The training as well as the evaluation data consisted of structures optimized at GFN-2+ALPB(water) level of theory, some of them with fixed main chain and side chain dihedral angles. The energies were calculated at BP86/D3Rezac, COSMO-RS level described previously.[3] Because the neural network has no built-in inductive biases besides the locality of interactions (5 Å distance cutoff) and equivariance, we can test if any new interactions appear in larger peptides that are not present in the dipeptides used for training. Previous research in our group indicated that this should not be possible, and energy function that would estimate energy of longer chains from shorter ones could not be constructed.[5] Interestingly, a system trained on dipeptides and amino acids only can already predict energy of pentapeptides with 1 kcal mol-1 RMSE and it can also correctly identify the global minimum of a larger protein out of 1000 structures (Figure 1). We believe that resulting potentials can be immediately used to significantly accelerate calculations. In addition, the excellent performance of the ML potentials also indicates that a bottom-up theoretical approach to predicting protein structures from first principles might be possible.

Figure 1. (a) Description of the training process and the training peptide structures. (b) Example of a larger test peptide. (c,d) Actual (DFT) vs predicted (NN trained on mono- and dipeptides) absolute energies of (c) pentapeptides and (d) conformers of (b) (in eV; atomic energies were subtracted).

References
[1] Schweitzer-Stenner, R. The relevance of short peptides for an understanding of unfolded and intrinsically disordered proteins. Phys. Chem. Chem. Phys. 2023, 25, 11908-11933.
[2] Batzner, S.; Musaelian, A.; Sun, L.; Geiger, M.; Mailoa, J. P.; Kornbluth, M.; Molinari, N.; Smidt, T. E.; Kozinsky, B. E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials. Nat. Commun. 2022, 13, 2453.
[3] Kalvoda, T.; Culka, M.; Rulíšek, L.; Andris, E. Exhaustive Mapping of the Conformational Space of Natural Dipeptides by the DFT-D3//COSMO-RS Method. J. Phys. Chem. B 2022, 126, 5949–5958.
[4] Culka, M.; Kalvoda, T.; Gutten, O.; Rulíšek, L. Mapping Conformational Space of All 8000 Tripeptides by Quantum Chemical Methods: What Strain Is Affordable within Folded Protein Chains? J. Phys. Chem. B 2021, 125, 58–69.
[5] Kalvoda, T. Studium konformačního chování krátkých peptidových fragmentů metodami kvantové chemie. Master Thesis [Online], Charles University, Prague, July 2020. http://hdl.handle.net/20.500.11956/122714 (accessed Sep. 4, 2024).

Speaker: Erik Andris (IOCB Prague)

Lunch & Poster Session

Users' Talks II

5 Engineering Defect Clustering in Diamond-based Materials for Technological Applications via Quantum-mechanical Descriptors

Dopant-dopant and dopant-vacancy complexes in diamond can be utilised in quantum computers, single-photon emitters, high-precision magnetic field sensing and nanophotonic devices. While some dopant-vacancy (e.g. N-V centre) complexes are well-studied, research on other dopant/vacancy clusters is focused mainly on defect detection, with minimal investigation into their electronic features or how to tune their electronic and optical properties. Moreover, the formation of cluster defects is often seen as undesirable, and their potential role in technological applications is overlooked.

In this work, we aim to reveal coupled structural electronic features of different dopant/vacancy configurations and their effect on the band gap of diamond-derived materials. We conduct a quantum mechanical investigation of diamond-based structures containing various types of cluster defects and dopant atoms in different concentrations. Moreover, we compare the results with systems containing a single dopant where relevant. Our findings reveal that doping with a p-type (n-type) dopant does not always lead to the creation of p-type (n-type) diamond structures, depending on the kind of cluster defect. We also identify quantum mechanical descriptors (e.g. charge redistribution) which are most suitable to tune the electronic band gap about the Fermi level for each defect type. We suggest how to choose suitable dopant atomic types, concentration and kind of aggregation to achieve the target electrical or optical effect. Finally, we discuss how the different cluster defects can be exploited in several technological applications such as transparent conductive materials, laser diodes, intermediate band photovoltaics and multi-colour emitters, among others.

This work provides a set of guidelines on how to achieve the desired electronic or optical properties of the material. Furthermore, we already present a variety of promising material options for specific applications which can be promptly used in the design of particular devices.

This work was supported by the project “The Energy Conversion and Storage“, funded as project No. CZ.02.01.01/00/22 008/0004617 by Programme Johannes Amos Commenius, call Excellent Research. This work was supported by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID:90254). The access to the computational infrastructure of the OP VVV funded project CZ.02.1.01/0.0/0.0/16 019/0000765 “Research Center for Informatics“ is also gratefully acknowledged.

Speaker: Matúš Kaintz (Czech Technical University in Prague)

6 Efficient and accurate in silico polymorph ranking for molecular pharmaceuticals

Accuracy and sophistication of in silico models of structure, internal dynamics and cohesion of molecular materials at finite-temperatures has evolved over time. It has become possible to perform ab initio predictions of polymorphism of crystals of small molecules with a qualitative accuracy. Obviously, it is essential to properly capture all the non-covalent interactions within the lattice of a molecular crystal to be able to predict its lattice energy, thermodynamic properties, or even relative stabilities of multiple polymorphs. Extending the applicability of such first-principles models to larger systems with a real-life significance, such as pharmaceuticals or semiconductors is now vital for material research.
Efficient and accurate computational models of polymorphism would enable to perform an initial high-throughput screening when designing novel molecular materials. Development of such methods still represents a massive challenge to be tackled by computational chemists. This work presents a novel composite method that combines the computational efficiency of density-functional tight binding (DFTB) methods with the accuracy of density functional theory (DFT). Following the quasi-harmonic approximation, it uses a fast method to perform the otherwise costly scans of how static and dynamic cohesive characteristics of the crystal vary with respect to its volume. Such data are subsequently corrected to agree with results of a higher-level more expensive model, which needs to be evaluated only at a single volume of the crystal. It thus enables predictions of structural, cohesive and thermodynamic properties of complex molecular materials, such as pharmaceuticals or organic semiconductors, at a fraction of the original computational cost.
As the composite model retains the solid physical background, it suffers from a minimum accuracy deterioration when compared to the full treatment with the costly approach. The novel methodology is demonstrated to provide consistent predictions of structural and thermodynamic properties of real-life molecular crystals of selected pharmaceuticals and their polymorph ranking.

Speaker: Ctirad Cervinka

7 The Impact of Atomic Polarizability on the Thermodynamic and Structural Properties of Ionic Liquids

Molecular dynamics (MD) simulations at all-atom resolution provide valuable insights into the relationship between atomic-scale structures and the macroscopic properties of materials. In this study, we performed MD simulations on a broad set of aprotic ionic liquids (ILs) to assess whether incorporating atomic polarizability into the force field enhances the accuracy and reliability of predicted thermodynamic and structural properties of these materials, specifically the melting enthalpy and glass-transition temperature. Three approaches to the polarizability were evaluated: the Drude oscillator model, the self-consistent AMOEBA force field (explicit polarizability treatments), and charge scaling (implicit polarizability). A non-polarizable classical force field served as a reference level of theory. Leveraging IT4I computational resources at Karolina and the LAMMPS simulation software, we found that the Drude oscillator model provided the best overall performance for both properties studied, with average deviations from experimental data of about 30% and 11 K for the fusion enthalpy and glass-transition temperature, respectively. Furthermore, the Drude model demonstrated stronger qualitative correlations with experimental data, highlighting its potential for IL design. Additional analyses, exploring correlations between simulated data and IL-specific descriptors, were also performed to support the understanding of the behavior of ILs in the context of their structure and interactions.

Speaker: Martin Klajmon (University of Chemistry and Technology Prague)

Session: Accessing EuroHPC Systems and Support from the High-Level Support Team (HLST)

8 Access to the EuroHPC JU infrastructure

Speaker: Tomas Karasek (IT4Innovations, VSB-Technical University of Ostrava)

9 Empowering Research through the High-Level Support Team (HLST)

The High-Level Support Team (HLST) at IT4Innovations was established to meet the advanced computational needs of researchers working on high-performance computing (HPC) projects. Created in response to user feedback, HLST provides tailored support to optimize workflows, troubleshoot issues, and efficiently allocate resources. By simplifying HPC access and enabling collaborative problem-solving between project researchers and the IT4i team, HLST empowers users to focus on achieving faster, more impactful research outcomes. This presentation details the unique advantages HLST offers, including guidance on code modernization, data management, and compliance, to enhance productivity and accelerate scientific progress.

Speaker: João Barbosa (IT4Innovations)

Users' Council

Coffee break

Keynote II

10 StreaMD: high-throughput automated molecular dynamics simulations tool

Molecular dynamics (MD) simulations and binding free energy calculations are critical techniques in computational chemistry and molecular biology. This study presents the implementation of an advanced tool designed to fully automate system preparation, start or extend MD simulations, perform trajectory analysis, and compute binding free energies (using gmx_MMPBSA), along with detailed protein-ligand interaction analyses. The pipeline supports diverse systems, including protein-only, protein-cofactors, protein-ligand, and protein-ligand-cofactors complexes in explicit water environments. By requiring only a PDB file for the protein and, where necessary, SDF or MOL files for ligands or cofactors, the tool streamlines the input process.
The tool's efficiency is significantly enhanced through the use of the Dask Python library, which enables parallelization and distributed computing across a network of servers via SSH connections independent from a particular scheduler. It also supports GPU-accelerated computations, offering a substantial reduction in processing time. Advanced features include the Gaussian-based parameterization of non-standard ligands (such as boron-containing molecules) and MCPB.py parameterization for atoms involved in metal coordination, making the tool highly versatile for complex molecular systems.
For validation we run 1 ns simulations and calculated the GBSA energies for 161 molecules of human β-secretase 1 (UniProt ID: P56817), 63 molecules of human α-thrombin (UniProt ID: P00734) and 51 molecules of bovine trypsin (UniProt ID: P00760). The resulting Pearson correlation coefficients between GBSA energies and experimental activity values were found to be -0.67 for the β-secretase 1, -0.53 for the trypsin and -0.72 for the α-thrombin datasets. The StreaMD and MM-GBSA method were applied in advanced high-throughput screening process during the Critical Assessment of Computational Hit-Finding Experiments (CACHE) Challenge #1 competition as a final ranking procedure to select 150 molecules from 900 candidates obtained from the previous ranking stages. The achieved hit rate was ~10% (8 compounds were active out of 82 total studied ones).
The tool is available as an open-source package at https://github.com/ci-lab-cz/streamd/tree/master.

Jurášek, M., et al. (2023). Triazole-based estradiol dimers prepared via CuAAC from 17α-ethinyl estradiol with five-atom linkers causing G2/M arrest and tubulin inhibition. Bioorganic Chemistry, 131, 106334.
Řehulka, J., et al. (2022). Anticancer 5‐arylidene‐2‐(4‐hydroxyphenyl)aminothiazol‐4(5 H )‐ones as tubulin inhibitors. Archiv Der Pharmazie, 355(12), 2200419.

The work was supported by the Ministry of Education, Youth and Sports of the Czech Republic through INTER_EXCELLENCE II grant LUAUS23262, the e-INFRA CZ (ID:90254) and projects ELIXIR-CZ (LM2023055) and CZ-OPENSCREEN (LM2023052).

Speaker: Aleksandra Ivanova

Users' Talks III

11 Structure and Tribological Properties of Mo-S-C Films via Reactive Molecular Dynamics

Molybdenum disulfide (MoS₂) is a layered material from the transition metal dichalcogenide (TMD) family, widely applied in areas ranging from tribological coatings to electronics, optics, and catalysis. However, pure TMD coatings face limitations in tribological applications due to their low hardness and poor oxidation resistance. To address these issues, combining TMDs with carbon has emerged as a promising strategy. Carbon contributes to increased hardness, while the TMD forms tribolayers during use, reducing friction.
The structure and evolution of TMD-C coatings during tribological processes are subjects of great interest, requiring both experimental and computational approaches. These coatings exhibit complex structural and bonding environments that are challenging to capture experimentally, particularly under tribological conditions. Simulations offer the ability to model these interactions in a controlled virtual environment, providing a dynamic view of how the structure evolves over time or in response to external factors.
Using our developed ReaxFF parameter set, we investigated the structure and properties of TMD-C films. First, we studied the bonding in amorphous Mo-S-C and observed a strong tendency for phase separation into MoS₂ and C phases. Our results on the kinetics of this phase separation suggest that it likely occurs during the deposition of Mo-S-C films, which aligns with experimental observations: Mo-S and C-C bonds dominate, with Mo-C bonds forming only when the S-to-Mo ratio significantly deviates from 2:1.
Further structural exploration of Mo-S-C revealed notable variations in the carbon phase's morphology depending on carbon content. Models with 25 atom-% carbon displayed a one-dimensional carbon phase in the form of a tube within the MoS₂ matrix. At 50 atom-% carbon, two-dimensional carbon sheets emerged, while models with 75 atom-% carbon formed a three-dimensional carbon phase with embedded MoS₂.
We also examined sliding behavior and its relationship to structural changes in these models. Friction coefficients were notably low, below 0.02 for "carbon in MoS₂" configurations, and below 0.2 for "MoS₂ in carbon" configurations. The crystallization of MoS₂ plays a significant role in reducing friction.

Speaker: Ilia Ponomarev (Czech Technical University in Prague)

12 Multiscale modeling of magnetoelasticity

Magnetoelastic interactions are responsible for many interesting phenomena such as Joule magnetostriction, the Wiedemann effect, the Villari effect, effects on sound velocity, and many others. In this work[1], we investigate the effect of microstructure on saturation magnetostriction of Terfenol-D (Tb0.27Dy0.73Fe2) by means of Finite Element Method. The model is based on the equilibrium magnetoelastic strain tensor at magnetic saturation, and shows that the crystal orientation jointly with the grain volume fraction play a more significant role on saturation magnetostriction than the morphology of the grains. We also calculate the dependence of saturation magnetostriction on the dispersion angle of the distribution of grains in the oriented growth crystal directions < 011 > and < 111 >. This result evinces the importance of high-quality control of grain orientation in the synthesis of grain-aligned polycrystalline Terfenol-D. The input parameters, i.e. magnetostriction coefficients (100 and 111), could be determined using novel developed method [2,3] based on the first-principles calculations (microscopic scale) combined with classical spin-lattice simulations [4] (mesoscopic scale) within a multiscale approach. The computational aspects will be discussed.
References:
[1] P. Nieves, and D. Legut, Solid State Comm., 352, (2022), 114825
[2] P. Nieves, D. Legut et al, Comput. Phys. Comm., 271, (2022), 108197
[3] P. Nnieves, D. Legut et al, Comput. Mat. Sci. 224, (2023) 112158
[4] P. Nieves, D. Legut et al, Phys. Rev. B, 103 (2021), 094437

Speaker: Dominik Legut (IT4I)

13 Combined theoretical and experimental study of Fe-Sn intermetallic phases

There are conflicting literature reports related to Fe-Sn intermetallic phases, when, for example, the FeSn2 phase is theoretically predicted to be dynamically unstable due to imaginary phonon modes (see, e.g, C.-J. Yu et al., New J. Chem. 44 (2020) 21218, DOI:10.1039/d0nj04537c). We have, therefore, performed a combined theoretical and experimental study of both FeSn2 and FeSn intermetallics. The theoretical part consists of quantum-mechanical calculations of ground-state properties, including structural and magnetic properties. Computing phonon modes tested the dynamic stability, and the thermodynamic properties were subsequently assessed using quasi-harmonic approximation (QHA). The FeSn2 phase is computed stable, i.e., free of imaginary phonon modes. Importantly, vibrational degrees of freedom significantly affected the finite-temperature stability of FeSn2. We have also characterized Fe-Sn phases using our experimental samples, including X-ray analysis of structural aspects and Moessbauer measurements of magnetic properties. Both the lattice parameters and temperature-dependent Moessbauer factor (see also our previous paper M. Friák et al., Comp. Mater. Sci. 215 (2022) 111780, DOI:10.1016/j.commatsci.2022.111780) turned out to be in excellent agreement with our theoretical results. Financial support received under Project No. 22-05801S from the Czech Science Foundation is gratefully acknowledged. We also gratefully acknowledge the financial support from the Czech Academy of Sciences (the Praemium Academiae of M.F.). Computational resources were provided by the e-INFRA CZ project (ID:90254), supported by the Ministry of Education, Youth and Sports of the Czech Republic. These resources were utilized through IT4Innovations National Supercomputing Center.

Speaker: Martin Friák (Institute of Physics of Materials, Czech Academy of Sciences, Brno, Czech Republic)

Conference Dinner and Poster Session

Tuesday, 5 November

Keynote III

14 Utilization of AI in Shape Optimization

In shape optimization, especially for designs like airfoils and car bodies, computational fluid dynamics (CFD) simulations are essential for assessing aerodynamic performance. To speed up the optimization process, surrogate models—such as machine learning techniques can be used. These models approximate CFD outcomes at a fraction of the cost, enabling faster design exploration and optimization while maintaining sufficient accuracy. This makes it possible to perform efficient design iterations, making the process more feasible for real-world engineering.

In our work, we develop a differentiable model that uses a parameterized mesh as input and outputs key performance metrics like drag (CD) and lift (CL) coefficients. This approach allows for automatic shape optimization using gradient-based methods.

We also present preliminary results comparing different methods, including geodesic convolutional networks and traditional convolutional techniques, where shapes are represented as images for optimization.

Speaker: Dr Michal Kravčenko (IT4Innovations, VSB - Technical University of Ostrava)

Users' Talks IV

15 Collimated γ-flash radiation via laser-target interaction

The interaction of high-power lasers with targets provides a way to produce γ-photon beams [1,2]. In this work we demonstrate via three-dimensional particle-in-cell simulations a regime where emission of a collimated γ-photon beam occurs for oblique laser incidence angles on the flat target. The electromagnetic field accelerates electrons to the gigaelectronvolt energy level. Consequently, these electrons emit a collimated γ-photon beam when interacting with the electromagnetic field. The dependencies of γ-photon emission on the incident angle, laser polarization, laser power, laser duration and target thickness are addressed. The beam directionality is important for designing future experiments. We visualize the γ-photon emission temporal evolution via our Virtual Beamline - VBL application [3,4,5], accessible in a regular web browser and in a virtual reality headset.

References:
[1] P. Hadjisolomou et al., Phys. Rev. E 104, 015203 (2021), https://doi.org/10.1103/PhysRevE.104.015203
[2] P. Hadjisolomou et al. Sci Rep 12, 17143 (2022), http://doi.org/10.1038/s41598-022-21352-8
[3] Virtual beamline, https://vbl.eli-beams.eu/
[4] M. Danielova et al., In Proceedings of the EuroVis 2019-Posters (2019), https://doi.org/10.2312/eurp.20191145
[5] M. Matys et al., Photonics, 10, 61 (2023), https://doi.org/10.3390/photonics10010061

Speaker: Dr Martin Matys (Extreme Light Infrastructure ERIC / ELI Beamlines Facility)

16 State-Averaged Orbital-Optimized VQE: Enhancing Quantum Simulations for Ground and Excited States

In this presentation, we delve into the State-Averaged Orbital-Optimized Variational Quantum Eigensolver (SA-OO-VQE), a hybrid quantum-classical algorithm designed to provide a balanced and accurate description of both ground and excited electronic states. SA-OO-VQE leverages state-averaged orbital optimization to effectively handle degenerate and quasi-degenerate states, overcoming common numerical challenges associated with state-specific methods near avoided crossings and conical intersections. Additionally, we introduce a novel diabatization approach that enhances the reliability of electronic state representations in strongly coupled regions, which is crucial for accurately modeling non-adiabatic processes. Our implementation uses Qiskit and allows users to choose between a real quantum computer or simulator, exploiting CPU and GPU efficiently. It also demonstrates significant improvements in computing potential energy surfaces, gradients, and non-adiabatic couplings for complex molecular systems. These advancements pave the way for more precise and scalable quantum chemistry simulations, addressing key computational bottlenecks in modern theoretical chemistry.

Speaker: Silvie Illésová (IT4Innovations)

17 “But it works on my machine”- escaping dependency hell through containerization for faster deployment in high-throughput docking benchmarks

Docking is a computational method which searches through conformational space to find an optimal pose of a molecule in a specified environment within a protein target. It is designed to scan vast libraries of molecules against proteins to recreate binding geometry which is later used in drug design.
There are a lot of docking engines available, and it is of utmost importance to choose the right one for a specified task. It requires many engines to try, many trials and, we all know that, many errors even before we start to use the software. The question is: how to avoid it?
The core of the presentation focuses on escaping dependency hell for installation and configuration of software using the example of docking engines. Traditional software deployment requires one to recreate the development environment which, in general, might be hard to do especially for novel machine learning models used in the scientific process. Containerization encapsulates the entire software environment, ensuring that the software can run consistently regardless of the underlying infrastructure.
This work focuses on cytochromes P450 enzymes, a major drug metabolizing enzyme from the liver. It is reported that 75% of all commercially available drugs is metabolized by only few out of many of those enzymes in humans [1]. The CYPs contains an important structural feature for docking simulations: active site is flexible and therefore presents a challenge for docking with rigid protein approximation which shows limited reliability [2,3].
We will present our simulation workflow, discuss about the containers and how they can improve deployment, provide comments and tips for apptainer containers on IT4I, and finally discuss benchmark results of such high-throughput simulations using significantly differing docking software [4-8].

[1] Brändén, G., Sjögren, T., Schnecke, V., & Xue, Y. (2014). Structure-based ligand design to overcome CYP inhibition in drug discovery projects. Drug Discovery Today, 19(7).
[2] Lokwani, D.K.; Sarkate, A.P.; Karnik, K.S.; Nikalje, A.P.G.; Seijas, J.A. Structure-Based Site of Metabolism (SOM) Prediction of Ligand for CYP3A4 Enzyme: Comparison of Glide XP and Induced Fit Docking (IFD). Molecules 2020, 25, 1622.
[3] Matthew R. Masters, Amr H. Mahmoud, Yao Wei, and Markus A. Lill, Deep Learning Model for Efficient Protein–Ligand Docking with Implicit Side-Chain Flexibility, Journal of Chemical Information and Modeling 2023 63 (6), 1695-1707
[4] C. Lee, J. Yang, S. Kwon, C. Seok. GalaxyDock2-HEME: Protein–ligand docking for heme proteins J. Comput. Chem. 2023, 44(14), 1369
[5] Rohith Krishna et al. Generalized biomolecular modeling and design with RoseTTAFold All-Atom.Science 384,eadl2528(2024).
[6] Eberhardt, J., Santos-Martins, D., Tillack, A.F., Forli, S. (2021). AutoDock Vina 1.2.0: New Docking Methods, Expanded Force Field, and Python Bindings. Journal of Chemical Information and Modeling.
[7] Trott, O., & Olson, A. J. (2010). AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. Journal of computational chemistry, 31(2), 455-461.
[8] McNutt, A.T., Francoeur, P., Aggarwal, R. et al. GNINA 1.0: molecular docking with deep learning. J Cheminform 13, 43 (2021).

Speaker: Dominik Suwala (Faculty of Pharmacy in Hradec Kralove, Charles University)

18 Cratering of Tungsten Walls Under High-Velocity Dust Impacts

In fusion reactors, the durability of plasma-facing components (PFCs) is heavily compromised by the impacts of high-velocity tungsten (W) dust, particularly during extreme events such as runaway electron terminations. These incidents can propel dust particles at velocities reaching several kilometers per second, leading to significant material erosion and structural damage. While existing models primarily focus on low-velocity impacts, there remains a critical knowledge gap concerning high-velocity impacts, which is essential for the development of reactors like ITER and DEMO.

To address this gap, our study utilizes large-scale molecular dynamics simulations to explore the effects of high-velocity W dust impacts on tungsten walls under extreme operational conditions. The simulations cover a wide range of impact velocities (2.5 to 4.5 km/s), angles (0° to 75°), and temperatures (300 to 3000 K). By incorporating up to 300 million atoms, these simulations provide a detailed analysis of how impact angle and temperature influence crater morphology and ejecta distribution.

Our findings reveal that both the angle of impact and the operational temperature significantly affect the resulting crater structures and the dynamics of material ejection. The study delves into sputtering processes, material degradation, and deformation mechanisms at the atomic level, offering critical insights into the behavior of tungsten walls under high-velocity impacts.

This research substantially enhances our understanding of dust-wall interactions in fusion environments, contributing to the development of more resilient materials for future fusion energy systems. The insights gained from this work will aid in improving the durability and operational efficiency of PFCs, which are crucial for the successful deployment of next-generation fusion reactors.

Speaker: Prashant Dwivedi (Czech Technical University)

Coffee break

User's Talks V

19 Quasivalent Carbon: a new face-centred cubic carbon allotrope

Besides the three basic forms of carbon, i.e. diamond, graphite/graphene and fullerene, a fourth carbon allotrope has been identified. We obtained this carbon allotrope in form of epitaxial films on diamond in a quantity sufficient to perform their comprehensive studies, and provided clear evidence for its unique crystal and electronic structure. Its band gap was found to be typical for insulators, whereas the material has a noticeable electrical conductivity, and its temperature dependence is the typical one for semiconductors. In this work, we present the physical insights we obtained on the electronic properties of such carbon allotrope by means of ab initio simulations. The quantum mechanical results provide a possible explanation on the apparent contradiction between the large bandgap and the conductivity features, by pointing at noncovalent electron sharing of p-electrons of neighbouring carbon atoms. The individuated carbon allotrope can create a new pathway to ‘carbon electronics’, as it is the first carbon material having properties of intrinsic semiconductors.

This work was co-funded by the European Union under the project “Robotics and advanced industrial production” (reg. no. CZ.02.01.01/00/22_008/0004590). This work has also been done with the support of the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID:90254) and access to the computational infrastructure of the OP VVV-funded project CZ.02.1.01/0.0/0.0/16_019/0000765 “Research Center for Informatics”.

References
[1] I. Konyashin, R. Muydinov, A. Cammarata et al. “Face-centered cubic carbon as a fourth basic carbon allotrope with properties of intrinsic semiconductors and ultra-wide bandgap”, Communications Materials 5, 115 (2024).

Speaker: Antonio Cammarata (Czech Technical University in Prague)

20 Interaction of Carbon Nanomaterials with Biomaterials

Carbon nanomaterials have revolutionized the field of biomedicine, offering opportunities for their diverse applications. Their unique tunable physicochemical, optical, mechanical and electronic properties make them ideal candidates for a wide range of biomedical applications such as drug delivery, biosensing, and imaging. However, to harness their full potential and ensure their safe and efficient use, it is crucial to gain a comprehensive understanding of their interactions with biomaterials at the molecular level. For this, experimental techniques yet lack the required atomistic resolution and therefore we can rely on molecular dynamics simulations, which can provide valuable insights into the underlying mechanisms governing the biocompatibility of carbon nanomaterials.
Here we focus on the modeling of carbon dots (CDs) and their interactions with biomolecules. Using our dedicated builder (cd-builder.upol.cz), we generated structures and topologies of CDs for simulations. We specifically investigated the interaction of CDs with biomolecules, particularly nucleic acids. Through simulations, we identified preferential modes of interaction between CDs and various nucleic acid shapes, varying their type and complexity. Such simulations require a careful choice of simulation set-up compatible with each system part. The simulations show the effect of CDs on the structure of the nucleic acid, however in a different way than known DNA poisons.
Further, we focused on understanding of the behaviour of graphene derivatives and their interactions with complex lipid membranes in multiscale resolution. Through these simulations, we elucidated the nature of interactions between graphene derivatives and lipid membranes, providing insights into graphene-membrane interactions and the effect of graphene on membrane organization.
We present a complex approach to simulations of the bio-nano interface in multiscale resolution, necessary for capturing the required level on details. These simulations can be a start of in-silico studies on nanotoxicity of the nanomaterials or used for targeted design increasing their biocompatibility.

Speaker: Markéta Paloncýová (Regional Centre of Advanced Technologies and Materials, Czech Advanced Technology and Research Institute (CATRIN), Palacký University Olomouc)

21 Simulated nano-extrusion of graphene hyperbolic pseudosphere surfaces

We produce curved graphene hyperbolic pseudosphere surfaces in molecular dynamics (MD) simulation of a nano-scale extrusion process, in which C atoms are forced down a pseudo two-dimensional volume shaped like a hyperbolic pseudosphere. During the extrusion process the carbon atoms form pentagons, hexagons and heptagons and such a mixture is unrealistically less stable than pure graphite or diamond, by several eV/atom. During relaxation and lengthy high temperature annealing up to 0.1 microsecond (100mil. MD steps) after the extrusion process, the structure fast finds its equilibrium and polycrystalline curved graphene with a limited number of point defects is formed. The point defects cause bending of the graphene and the pseudosphere edges even more. When these free edges are removed from the simulations by attaching periodic flat graphene sheets to the pseudosphere edges, the carbon atoms assume positions with a root mean square deviation of some tenths of Å from the mathematical hyperbolic pseudosphere surface. The hyperbolic pseudospheres proved to be mechanically stable against large shearing and elongation deformations as well as against annealing at 1500 K. Our methodology is easy to use, employing the REBO2 carbon interaction potential [1] within the open source MD code LAMMPS [2]. Systems of less than 10000 atoms only scale efficiently on small numbers of cpu cores, but do require ~one week of runtime to complete 100mil. MD steps. Our method offers a practical way to create simulated stable, curved graphene surfaces with a wide variety of desired shapes. It allows for the testing in advance of the stability of graphene shapes that are to be produced experimentally.

[1] D. W. Brenner, O. A. Shenderova, J. A. Harrison, S. J. Stuart, B. Ni, S. B. Sinnott, “A second-generation reactive empirical bond order (REBO) potential energy expression for hydrocarbons”, J. Phys.: Condens. Matter 14 (2002) 783–802
[2] S. Plimpton, “Fast Parallel Algorithms for Short-Range Molecular Dynamics”, J. Comput. Phys. 117 (1995) 1

Speaker: Peter Klaver (VSB Technical University of Ostrava)

Lunch