SCIENCE BRINGS NATIONS TOGETHER
Mathematical Modeling and Computational Physics - MMCP’2026

→ Europe/Moscow
Description

MMCP’2026 will take place on 22-26 June 2026

The International Conference "Mathematical Modeling and Computational Physics” (MMCP’2026) will be held at the Meshcheryakov Laboratory of Information Technologies (MLIT) of the Joint Institute for Nuclear Research (JINR) on 22 – 26 June 2026 in Dubna. The conference MMCP’2026 is dedicated to the 70th anniversary of the Joint Institute for Nuclear Research and to the 60th anniversary of the Meshcheryakov Laboratory of Information Technologies. MMCP’2026 promises to be a stimulating gathering for researchers and practitioners in mathematical modeling and computational physics.


Conference topics:

  • Mathematical methods and tools for modeling complex physical systems;
  • Mathematical methods in life sciences;
  • Modern methods for data processing and analysis in Mega-science projects;
  • Machine learning and big data analytics;
  • Methods of quantum computing and quantum information processing;
  • Numerical and analytical calculations in modern mathematical physics;
  • Methods and numerical algorithms in high-energy physics.

The program of the Conference comprises plenary reports in English (30 min), sessional reports (15 min).

Conference languages – English and Russian.

Contacts: 

Address: 141980, Russia, Moscow region, Dubna, Joliot Curie Street, 6
Phone: (7 496 21) 64019, 64826
E-mail: mmcp2026@jinr.int
URL: http://mmcp2026.jinr.ru/
Proceedings
Programm
Participants
    • 09:00 → 10:00
      Registration Conference Hall, 6th floor (MLIT)

      Conference Hall, 6th floor

      MLIT

    • 10:00 → 13:00
      Plenary Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 10:00
        Opening 30m
      • 10:30
        JINR scientific program 30m
        Speaker: Elena Kolganova (BLTP JINR)
      • 11:00
        Methods, algorithms and software for processing experimental data at JINR 30m
        Speaker: Dr Nikolay Voytishin (JINR)
      • 11:30
        On the сompetition of alternative Deep Learning algorithms for processing data from SPD and MPD experiments at NICA 30m

        In my overview talk at the previous MMSR-2024 conference, I explained how the "Tracking Crisis" caused by the advent of new high-luminosity collider experiments has led to the development of new deep learning methods for tracking problems based on the use of recurrent, transform, and graph neural networks, as well as quantum annealing methods for Hopfield networks.
        Over the past two years, these new approaches have been rigorously tested on simulated datasets for SPD and MPD experiments, and other alternative algorithms have also emerged.
        This talk is devoted to a comparative discussion of the latest results of various algorithms for processing SPD and MPD data.

        Speaker: Prof. Gennady Ososkov (Joint Institute for Nuclear Research)
      • 12:00
        Status of ARIADNA collaboration for applied research at NICA facility and its requests for advanced computation techniques 30m

        The talk covers the status of ARIADNA collaboration for applied research at NICA facility and its topics connected with computing, data processing and mathematical modelling. The overall scope of applied research performed with NICA beams includes, but is not limited to, radiation protection in space, life sciences, radiation testing of microelectronics, ion-beam-based materials research, and novel approaches to accelerator-driven systems. The experiments being performed in these directions of research require advanced simulation techniques to reconstruct physical processes of radiation interaction with matter as well as reliable data storage to keep the results for their further exploitation. The talk focuses on the first results obtained during the test operation of the ARIADNA beamlines and discusses their potential transfer to various industrial sectors, including the space industry, materials engineering, radiation therapy technologies, etc. Special emphasis is placed on ARIADNA’s mission to disseminate and deepen expertise in heavy-ion interactions with matter as well as potential application of various computation techniques.

        Speaker: Олег Белов (VBLHEP JINR)
      • 12:30
        AI-Driven Systematic Research on Oceanic Internal Solitary Waves: From SAR Image Interpretation to Governing Equation Discovery 30m

        Oceanic internal solitary waves (ISWs) are large-amplitude, nonlinear internal waves propagating within density-stratified ocean layers. Their accurate observation and modeling are critical for understanding ocean energy transport, water mass mixing, and marine navigation safety. Spaceborne synthetic aperture radar (SAR) has become the primary instrument for ISW remote sensing due to its all-weather, day-and-night imaging capability. However, ISW extraction from SAR imagery faces three fundamental challenges: (i) severe speckle noise corrupts wave signature fidelity; (ii) the scarcity of annotated training samples limits the applicability of deep learning models; and (iii) satellite revisit cycles restrict continuous temporal observation of ISW propagation dynamics. Furthermore, conventional physics-based models such as the Korteweg–de Vries (KdV) equation exhibit limited validity under complex oceanographic conditions, necessitating a paradigm shift toward artificial intelligence (AI)-driven systematic methodologies.
        This study establishes an end-to-end technical pipeline spanning five stages: image preprocessing, feature extraction, segmentation and extraction, process reconstruction, and physical modeling. For image preprocessing, a multichannel structure-aware speckle (MCS) filter framework has been proposed to address speckle suppression. For intelligent segmentation, several architectures are developed: MTU2-Net, PCGAN, GAN-MSDU, etc. For process reconstruction, a collaborative diffusion framework is introduced to generate continuous visual sequences from sparse satellite observations. For physical modeling, sparse regression algorithms coupled with Levenberg–Marquardt optimization are employed to discover data-driven governing equations directly from observational data.
        These advances collectively demonstrate the transformative potential of artificial intelligence in oceanographic remote sensing and geophysical fluid dynamics research.

        Speaker: Prof. Saheya Barintag (Inner Mongolia Normal University, China)
    • 13:00 → 13:10
      Group Photo Conference Hall (MLIT)

      Conference Hall

      MLIT

    • 13:10 → 14:00
      Lunch
    • 14:00 → 16:00
      Mathematical methods and tools for modeling complex physical systems Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 14:00
        Modeling of thermophysical processes by multithreaded parallel computing algorithms 15m

        This study addresses the implementation of nonlinear mathematical models of fluid flow and heat transfer using parallel multi-threaded algorithms. The rationale for the proposed approach lies in constructing linearized algebraic models that approximate the fundamental laws of hydrodynamics and thermodynamics. Methods for formulating three-dimensional unsteady conservation equations of mass, momentum, and energy for liquid media, specifically tailored to parallel computing architectures, are proposed. Finite volume and finite element methods have been optimized for graphics processing units (GPUs), enabling high-speed and scalable simulation of complex thermophysical processes. The developed algorithms facilitate the efficient solution of computational fluid dynamics (CFD) and heat transfer problems in multi-threaded computing environments.

        Speaker: Nikolay Stepenko (St. Petersburg State University)
      • 14:15
        Modelling cosmic-ray propagation in electromagnetic fields and matter interactions with a Python package GT simulation 15m

        The standard method for analysing cosmic-ray propagation relies on a statistical description based on numerically solving the transport equation. It describes the evolution of the cosmic-ray distribution function through a diffusion process, accounting for interactions with matter and energy losses. However, a limitation of this approximation is its reliance on spatially averaged quantities. Consequently, the method suffers from the loss of kinetic information and fails to resolve the motion of individual particles, which might be critical for a complete understanding of cosmic-ray transport in various astrophysical systems.

        This paper presents an overview of the open-source GT simulation package, which is designed to be freely available to the community. In contrast, GT adopts a more fundamental approach by modelling the motion of individual particles. It employs particle tracing method to integrate the equations of motion, incorporating numerous electromagnetic field models, including those for the Earth's magnetosphere, the heliosphere, and the Galactic magnetic field, as well as various medium models, such as the atmosphere or the interstellar medium. Additionally, the package provides an interface to the Geant4 platform, enabling simulations of nuclear interactions, decays, and energy losses. This approach offers an alternative way of modelling cosmic-ray physics. The package's structure allows it to be applied to a variety of astrophysical environments and easily extended to incorporate new physical models.

        Speaker: Andrey Dyagilev (NRNU MEPhI)
      • 14:30
        Comparative assessment of quadratic programming solvers for neutron spectrum unfolding 15m

        This paper presents an approach for neutron spectrum unfolding formulated as a quadratic programming problem. The relationship between measurements obtained with a Bonner multisphere spectrometer (BSS) and the neutron spectrum is described by a system of Fredholm integral equations of the first kind, the discretization of which yields an ill-posed system of linear algebraic equations. To obtain a stable solution, Tikhonov regularization is used. In our work the inverse problem is transformed into a constrained optimization problem with a non-negativity constraint on the spectrum. A comparative analysis is conducted on solvers integrated into the Python environment via the qpsolvers and cvxpy packages: ECOS, Clarabel, SCIP, SCS, SDPA, CPLEX, HiGHS, PROXQP, OSQP, and QUADPROD. An interface to the Cosmo solver from Julia language was also implemented. The optimal regularization parameter was selected using generalized cross-validation as implemented in the pytikhonov package. The methods are validated using an ensemble of 20 spectra from the IAEA Compendium, simulated via Monte Carlo transport code, including reference spectra such as 252Cf and 241AmBe, as well as realistic workplace neutron fields. Unfolding quality is quantitatively assessed using a suite of metrics: mean square error (MSE), Wasserstein distance, coefficient of determination (R2), cosine similarity, and an integral metric—the relative uncertainty in effective dose rate estimation. The study was done for a BSS with polyethylene spheres with diameters of 3, 4, 5, 6, 8, 9, 12, and 18 inches, supplemented by spheres with lead inserts of 9, 12, and 18 inches in diameter to enhance sensitivity above 20 MeV. The results demonstrate that solver selection significantly influences solution quality. Based on a composite evaluation of the metrics, the SCS solver yields the best overall performance. The relative error in effective dose rate estimation across solvers ranges from 0.7 to 1%, with Cosmo achieving the highest accuracy. All methods exhibit comparable performance in terms of R2 (0.69–0.71) and cosine similarity (0.90–0.91). Notably, unfolding methods are shown to introduce artifacts manifesting as spurious peaks in the energy range above 20 MeV. In the context of radiation safety, such artifacts risk being misinterpreted as indicative of shielding deficiencies or the presence of non-existent high-energy radiation. Furthermore, configurations relying on BSS spheres with reduced sensitivity in the high-energy region (i.e., without lead inserts) exacerbate the prevalence of such spurious peaks. The proposed algorithms are implemented in bssunfold python package. It can be used to improve dose assessment in high-energy neutron fields.

        Speaker: Константин Чижов (Radiation safety department)
      • 14:45
        Analysis of traffic flows in a roundabout system based on cellular automata models 15m

        Software for traffic flow modeling is currently in high demand to conduct experiments with digital models of road networks for sensible design and optimal traffic management One promising type of road junction is the roundabout: it is considered more efficient than conventional intersections (with or without traffic lights), especially in terms of safety and capacity under moderate road congestion. Describing traffic at roundabouts is a challenging problem. The original software TrafficFlowViz3D was supplemented with a model, related algorithms and visualization tools for multi-lane roundabout simulation [1].
        The new model is an extension of the Nagel-Schreckenberg cellular automata model [2]. Traffic cellular automata are discrete in space and time. The road is represented as a lattice. The cells' state is updated at each time step, which is interpreted as the movement of vehicles. In case of heterogeneous traffic, vehicles of different types occupy different numbers of cells [3]; the use of not only traditional rectangular grids, but also grids with hexagonal cells allows approximating road networks with complex geometry [3]; the developed routing algorithms ensure movement to a global target within the entire computational domain [3].
        The roundabout routing algorithm uses the Manhattan distance as the path cost to the desired exit [1]. The present study proposes an improved algorithm that relaxes strict entry/exit rules and safety rules to enable more intensive lane changes on the roundabout, thereby improving the roundabout capacity.
        For verification, traffic was simulated in a closed system of four two-lane roundabouts connected by two-way one-lane straight road sections. At the initial moment all road cells are free. Vehicles appear in the system at the predetermined generation points until the total number of vehicles reaches the specified value. Periodic boundary conditions are implemented. Calculations were carried out for different lengths of straight sections. The problem statement and preliminary results can be found in [1].
        In the environment TrafficFlowViz3D the simulation is inseparable from dynamic visualization. In addition to visual observation, the program allows for the evaluation of integral numerical characteristics for system analysis. At any time moment, we have a certain state of the system: some cells are occupied by vehicles at corresponding discrete speeds. For each traffic lane the density is determined based on the number of occupied cells, the speed is calculated as the arithmetic mean, and the flow is obtained as their product using the hydrodynamic relation. The roundabouts have two lanes, and the lane-averaged speed is the arithmetic mean weighted by the lane densities, the total flow is the product of the total density and the lane-averaged speed. The data are further aggregated over 100-second intervals, thus discrete simulation results on all straight sections and roundabouts are averaged over space and time.
        The so-called macroscopic fundamental diagram (MFD) is of great interest [4]: the reference dependence of traffic flow on density is true not only for a single road, but also for a road network. The existence of MFD at the level of an area allows monitoring and refining control strategies in real urban transport networks. For the roundabout system under consideration the dependence of the weighted average total flow on the weighted average total density is obtained at different lengths of straight links and at different accumulation of vehicles. The weight is the proportion of the link length in the total length of the network roads.
        The developed cellular automata model is stochastic: it reflects random braking while driving along a lane, random failure to change lanes, and randomness when choosing a roundabout exit. From a human factor perspective, random braking simulates driver imperfections, which leads to a decrease in average speed. Computational experiments with MFD show that the roundabout system capacity, defined as the maximum flow, increases as the probability of random perturbations decreases. However, a deterministic model cannot reproduce spontaneous processes in the system. Also, the capacity is higher if straight sections are longer. Traffic breakdown occurs when the system becomes fully saturated with vehicles.
        In the future, the problems of establishing a steady-state mode and self-organization of the system [5] will be studied.

        References

        1. M.A. Trapeznikova, M.V. Yashina, A.G. Garibyan. Modeling and visualization of traffic flow at roundabouts using cellular automata // Mathematical Models and Computer Simulations, 2025, V. 17, Suppl. 1, P. S85-S97. DOI: 10.1134/S2070048225700693
        2. K. Nagel, M. Schreckenberg. Cellular automaton model for freeway traffic // J. Phys. I France, 1992, V. 2, P. 2221-2229.
        3. M.A. Trapeznikova, A.G. Garibyan. Development and visualization of cellular automata models for predicting heterogeneous traffic flows // Physics of Particles and Nuclei, 2025, V. 56, No. 6, P. 1638-1642. DOI: 10.1134/S1063779625701011
        4. N. Geroliminis, C.F. Daganzo. Existence of urban-scale macroscopic fundamental diagrams: some experimental findings // Transportation Research Part B: Methodological, 2008, V. 42, No. 9, P. 759-770.
        5. M.V. Yashina, A.G. Tatashev. Loading conditions for self-organization in the BML model with stochastic direction choice // Mathematical Methods in the Applied Sciences, 2024, V. 47, No. 18, P. 14341-14352.
        Speaker: Dr Marina Trapeznikova (Keldysh Institute of Applied Mathematics RAS)
      • 15:00
        NuSt-Vis, Nuclear Structure Visualization Software 15m

        We present software designed for visualization of nuclear structure. The underlying model based on the quark model of nucleon structure leads to the arrangement of nucleons in a face-centered cubic (FCC) lattice [1]. Building on the algorithm proposed by N. Cook and co-authors [2], we developed the software NuSt-Vis, which enables the construction of any nucleus from the periodic table. The Qt-based graphical interface provides intuitive, interactive construction of nuclei and renders them as three-dimensional models [3]. The software runs under both Linux and Windows, owing to the cross-platform Qt toolkit and the C programming language.
        The Gnuplot package [4] was used to verify the visual representation of the structures.

        1. G. Musulmanbekov, Adv. Nucl. Sci. Appl., v1, № 4, p. 250 (2025);
          https://doi.org/10.63907/ansa. v1i4.53.
        2. N. D. Cook, T. Hayashi and N. Yoshida, IEEE Comp. Graph. Appl. 19, 54-60, 1999.
        3. https://www.qt.io/development/qt-framework
        4. A.G. Soloviev, The Gluplot Plotting Package [Electronic resource]: http://wwwinfo.jinr.ru/programs/jinrlib/gluplot/indexe.html
        Speaker: Женис Мусульманбеков (JINR, LIT)
      • 15:15
        Solvable models for interacting particles with spin and fundamental constants 15m

        Solvable models of interacting particles with spin are constructed. Systems of units based on the fundamental constants of physics are considered.

        Speaker: Dr Nugzar Makhaldiani (JINR)
      • 15:30
        Maxwell-Fourier regularisation of MSC-230 cyclotron magnetic field map 15m

        Magnetic maps for superconducting medical cyclotrons are design-critical elements for the performance of the machine. In this respect interpolation errors from the magnetic field simulations must be regularised for strict observance of Maxwell's equations, to ensure stable beam confinement and focusing. We present a robust Maxwell–Fourier filter that enforces exact Maxwell compliance via spectral Laplacian projection in Fourier space, guaranteeing that the corrected field derives from a harmonic scalar potential. This approach enables generation of accurate, divergence-free magnetic maps essential for compact, high-performance cyclotrons, eliminating simulation-induced errors. The results we present show a significant reduction of the non-physical artifacts in the high-gradient section and extraction radius region, ensuring reliable beam dynamics.

        Speaker: Svetlana Dima
      • 15:45
        Simulation of gasdynamic flows in regions with complex geometries using the fictitious domain method 15m

        Many applied problems deal with simulation gas or liquid flows in complexly shaped domains. This requires setting boundary conditions on curved external and, possibly, internal boundaries. For example, one of the current trends is direct (pore-scale) numerical simulation of gas- or hydrodynamic flows in porous media, which are considered as a system of channels with complex geometries. In this case, it is necessary to correctly impose boundary conditions on the set of internal interfaces between a free fluid and solid matrix. For these situations, a method of fictitious domains was proposed in [1], which allows to construct a unified numerical method for the entire computational domain using difference schemes constructed in a special way. This algorithm automatically ensures no-flow conditions where necessary. In the present work, a version of the fictitious domain algorithm based on quasi-gasdynamic (QGD) equations [2] is applied to simulate flows at low Reynolds numbers in a channel with a backward-facing step as well as in a channel with cylinder arrays of square or staggered arrangement, which can be treated as a porous medium.
        The algorithm can also be successfully applied to calculations involving objects with complex geometric shapes. To ensure end-to-end computation, such a calculation domain can be placed, for example, within a rectangle. Calculations are performed within the rectangular domainin a unified way, and the fictitious domain method ensures correct boundary conditions at the boundaries of the complex domain.
        This paper presents calculations of gas-dynamic flows in a catalytic reactor, where there are both a free flow and filtration flow through a porous filter. In this case, the calculation is performed using a generalized quasi-gasdynamic system of equations [3], similar to the Darcy-Forchheimer equations, allowing calculations to be performed across the entire computational domain without distinguishing interfaces between the porous body and free flow.

        1. M.A. Trapeznikova, N.G. Churbanova, G.U. Juraev, Application of the fictitious domains method for the simulation of gas flows at low Mach numbers // Preprints of the Keldysh Institute of Applied Mathematics. 2019, № 106, 12p.
        2. B.N. Chetverushkin, Kinetic Schemes and Quasi-Gas Dynamic System of Equations, Barcelona: CIMNE, 2008.
        3. N. Churbanova, V. Emets, Predictions of Flows with Porous Inclusions Using the Generalized QGD-System of Equations // Physics of Particles and Nuclei, 2025, V. 56, N 6, pp. 1413-1417.
          DOI: 10.1134/S1063779625700595.
        Speaker: Dr Natalia Churbanova (Keldysh Institute of Applied Mathematics of Russian Academy of Sciences (KIAM RAS))
    • 14:00 → 16:00
      Modern methods for data processing and analysis in Mega-science projects in HEP Room 310 (MLIT)

      Room 310

      MLIT

      • 14:00
        Processes of gravitino production as a dark matter particle in stealth supersymmetry 15m

        The report presents the results of a study of the stealth supersymmetry model, which contains a hidden sector additional to the MSSM with nearly mass-degenerate states interacting with the MSSM fields through a portal. In this model, the lightest superpartner and a dark matter particle candidate is the gravitino, while the decays of the other superpartners proceed through the hidden sector, allowing one to circumvent existing constraints on the observation of supersymmetry signals at the LHC. An analytical calculation of the decay process of the hidden sector singlet via the portal into Standard Model particles is shown, which is part of the cascade production process of the gravitino. The results of the simulation of this process for the LHC case are presented.

        Speaker: Alina Nurysheva (MIPT, JINR)
      • 14:15
        Gravitino pair production at the LHC 15m

        The report presents the results of a study of the processes of paired gravitino production at the Large Hadron Collider (LHC). The observability of such processes under LHC conditions is assessed. In the model under study, the gravitino is the LSP and is also considered a candidate for a dark matter particle.

        Speaker: София Франк-Каменецкая
      • 14:30
        Dark matter in the two-doublet Higgs model with b-quarks in the final state 15m

        A search for dark matter particles produced in association with a Higgs boson decaying to a bottom quark-antiquark pair in proton-proton collisions at $\sqrt{s}= 13$ TeV is presented. An analysis of the predictions of the extended two-doublet Higgs scenario is performed. The data, collected with the CMS detector at the LHC, correspond to an integrated luminosity of 101 fb$^{−1}$. The analysis is performed in exclusive categories targeting both Lorentz-boosted (merged) and resolved b jet pair topologies, covering a wide range of Higgs boson transverse momentum. A statistical combination is made with a previous search using data collected in 2016 and corresponding to an integrated luminosity of 35.9 fb$^{−1}$. The observed data is in agreement with the standard model background predictions. Constraints are placed on a type-II two-Higgs-doublet model featuring a heavy pseudoscalar with an additional light pseudoscalar (2HDM+a).

        Speaker: Zaurbek Khabaev
      • 14:45
        Application of deep neural network methods for the identification of Higgs boson production in association with a single top quark in proton–proton collisions at $\sqrt{s}=13.6$ TeV 15m

        Modern methods for the analysis of large volumes of data play a key role in the study of complex physical systems characterized by high-dimensional feature spaces, significant class imbalance, and the presence of statistically rare processes. In high-energy physics experiments, such challenges arise, in particular, in searches for processes with small cross sections in the presence of dominant background contributions with similar final-state topologies. In this work, a method for increasing sensitivity to the rare process of Higgs boson production in association with a single top quark ($tH$, hereafter referred to as the signal process), based on the application of deep learning techniques, is presented.

        The analysis is performed using Monte Carlo simulated proton–proton collision events at a center-of-mass energy of $\sqrt{s}=13.6~\text{TeV}$. The dataset includes the signal process ($tH$), simulated at next-to-leading order (NLO) in perturbation theory. Two main background processes, $t\bar{t}$ and $t\bar{t}H$, are also simulated at NLO. The remaining background processes with smaller contributions — $t\bar{t}b\bar{b}$, $tZbq$, $t\bar{t}Z$, and $t\bar{t}W$ — are generated at leading order (LO). The final training dataset is constructed taking into account the characteristic kinematics of final states containing a charged lepton, three or four $b$-jets, and a forward jet.

        Through hyperparameter optimization, including the number of hidden layers, the dimensionality of hidden representations, activation functions, and regularization parameters, the optimal neural network architecture is determined. The model is applied to a set of weakly correlated kinematic variables to ensure training stability. During training, differences in the expected event yields of signal and background processes are taken into account by introducing event weights. Additional improvement in classification performance is achieved through preprocessing of the input variables supplied to the neural network.

        To select a neural network architecture with the most reliable generalization capability, the stability of the training procedure is evaluated for several architectures with the highest classification accuracy. The obtained results demonstrate a significant improvement in signal–background separation compared to the traditional method based on imposing threshold cuts on kinematic variables and indicate the possibility of achieving a higher expected statistical significance of the signal process. The current classification performance reaches $\mathrm{ROC\text{-}AUC} = 0.87$, where ROC–AUC (area under the receiver operating characteristic curve) quantifies the probability that the classifier will assign a higher score to signal than to background events.

        Speaker: Mr Rustam Azimzade (Joint Institute for Nuclear Research)
      • 15:00
        Vectorized graph computations in Python with dag-modelling 15m

        High Energy Physics (HEP) analyses, particularly in neutrino experiments, demand computationally intensive simulations and iterative statistical evaluations. Rapid analysis pipelines are essential for pre-experiment parameter optimization, real-time hypothesis testing, and efficient resource allocation. However, conventional modeling frameworks often suffer from rigid architectures, redundant computations, and limited scalability, which slow down scientific iteration.

        We present the dag-modeling framework designed for efficient and transparent neutrino experiment simulation and analysis. By representing analytical workflows as directed acyclic graphs—where nodes encapsulate data and transformations, and edges define computational dependencies—the framework enables modular, reproducible, and easily interpretable pipeline construction. The architecture integrates lazy evaluation with just-in-time (JIT) compilation, automatically optimizing the computational graph at runtime to eliminate redundant operations and accelerate execution.

        We describe the framework’s internal structure, application scope, and implementation details, and evaluate its computational performance using open data from the Daya Bay Reactor Neutrino Experiment.
        This tool provides a scalable, efficient foundation for modern HEP analysis workflows, with direct applicability to neutrino physics and broader experimental domains.

        Speaker: Vitalii Zavadskyi (JINR)
      • 15:15
        Neutrino Trapping Effects in Hybrid and Strange Stars Revealed Through Gravitational Waves from Merger Events Using AI Analysis 15m

        Binary neutron star mergers create hot, lepton‑rich conditions where dense matter may transition from hadronic to deconfined quark phases. Using a covariant density functional for hadronic matter and a two‑flavor color‑superconducting (2SC) to CFL description for quark matter, we construct mixed phases that conserve baryon and lepton number and track thermodynamic trajectories at fixed entropy and lepton fraction. Neutrino trapping is found to significantly alter the composition and delay the onset of deconfinement, producing a density‑dependent pressure in the mixed phase.

        The resulting finite‑temperature equations of state are used to compute hybrid‑star configurations, showing that hot, neutrino‑rich stars have larger radii and slightly higher maximum masses than cold, deleptonized counterparts. We further examine how these microphysical effects imprint themselves on gravitational‑wave observables in merger events, including tidal deformabilities and post‑merger oscillation spectra. Finally, we demonstrate that AI‑assisted gravitational‑wave analysis can enhance the detection of subtle signatures of deconfinement and neutrino trapping, offering a promising pathway to constraining the phase structure of dense matter.

        Speaker: Dr Chitta Ranjan Das (BLTP, JINR)
      • 15:30
        Some algorithms for evaluation of Goncharov multilogarithms and related functions 15m

        Goncharov multilogarithms (MPLs) serve as a fundamental functional basis for expressing multiloop scattering amplitudes in high-energy physics. Certain classes of multilogarithms are deeply intertwined with the broader class of generalized hypergeometric functions with good convergence near singular points.

        Speaker: Егор Дыдышко (JINR)
      • 15:45
        Event-by-Event analysis in the multiparticle events 15m

        Event-by-event analysis of heavy-ion collision events is an important tool for the study of new phenomena in formation of a quark-gluon plasma. Such
        analysis may provide useful information about the properties of the matter
        created in these collisions. In this work EbyE analysis is used for p_T^{max} global variable, which is calculated for each event in the dataset. It is shown that usage of this variable give a new sight useful for events taxonomy.

        Speaker: Олег Рогачевский (JINR)
    • 14:00 → 16:00
      Numerical and analytical calculations in modern mathematical physics Room 406 (MLIT)

      Room 406

      MLIT

      • 14:00
        Modelling S-parameters in PCBs with the method of moments and a scattering algorithm 15m

        Printed circuit boards (PCB) is a widely used technology in radio and microelectronics. PCB modelling is an important part of the PCB manufacturing process, since it allows the designer to optimize the PCB output characteristics prior to investing cost and effort into physical PCB production. A PCB is essentially a 2.5D structure, since it consists of a layered dielectric host containing thin and highly conductive metal components; therefore, PCB modelling problem is amenable to the solution by the method of moments (MoM) [1]. MoM is a projection method applied to Maxwell's equations in the integral equation form, where the kernel of the integral operator is the Green's function of the host medium, and the known source fields and the unknown currents are expanded in terms of a simple basis function set defined on the conductors. An essential part of the modelling process is the calculation of the host Green's function and its overlap integrals with the MoM basis functions. A large part of this work is performed analytically.

        In this contribution, we consider MoM applied to multilayer shielded PCB modelling, where the Green's dyadic in a rectangular waveguide is calculated using the scattering matrix (S-matrix) formalism. Instead of using transmission line theory, which is a commonly used approach in the community, we show how the complete Green’s dyadic including both transverse and longitudinal components, can be calculated from scattering matrices, rigorously treating all the multiple reflections and transmissions of the waveguide modes. We express the Green's function components in terms of the waveguide modes and the aggregated scattering matrices of the wavegiude sections above, below and between the layers, in which the electric current source and the field observation point are located. We formulate the algorithm to construct the necessary aggregated S-matrix components for an arbitrary layered host medium from elementary layer and interface S-matrices. Using the aggregated S-matrices, the general form of the Green's function for any two points can be expressed. Full Green's dyadic can be used to model both transverse and longitudinal currents, which are utilized for modelling flat metallizations and interconnecting structures. The method is implemented with the help of the surface and volume piecewise linear functions (pulses, rooftops and half rooftops), for which overlap integrals are evaluated analytically [2]. The method is implemented with FFT-acceleration of the MoM matrix construction and iterative MoM linear system solver on a regular mesh in the transverse directions. The resulting method is numerically stable. Performance is validated on a set of numerical examples. The method can be extended to other basis function sets and applied to modelling other metal shapes in a shielded layered waveguide.

        Acknowledgement: This work was supported by the Russian Science Foundation grant no. 22-11-00153-$\Pi$.

        References:
        [1] V.Okhmatovski and S.Zheng, Theory and Computation of Electromagnetic Fields in Layered Media, Wiley-IEEE Press, 2024.
        [2] J.C.Rautio, Method of Moments Analysis of Structures Embedded in Shielded Layered Media, USA, 2020.

        Speaker: Polina V. Zheglova (ITMO University)
      • 14:15
        Relaxation Method Based on Energy Landscape Information in the Vicinity of the Global Minimum of the Classical Thomson Problem on a Disk 15m

        The search for equilibrium configurations of classical Coulomb charges confined to a circular domain with an infinite confining potential at the boundary (the Thomson problem on a disk [1]) remains a fundamental problem with direct applications in the physics of mesoscopic systems, nanoelectronics, and plasma physics [2]. In contrast to the problem on a sphere (S. Smale's 7th problem [3]), the two-dimensional case on a disk is characterized by a strong boundary effect, leading to the formation of ring-like structures, topological defects, and an inhomogeneous density that increases toward the boundary [2,4]. The total ground-state energy $E_{GS}$ of such systems comprises the following contributions: the electrostatic energy of the continuum limit $(∝N^{2}⁄R)$, the correlation energy of the discrete lattice $(∝N^{3⁄2}⁄R)$, the surface energy of boundary charges $(∝N⁄R)$, and the elastic energy of topological defects $(∝N^{1⁄2}⁄R)$, where $R$ is the disk radius [4].
        Finding the global minimum for this problem is considered to be an $NP$-hard problem. Standard approaches, such as quenched molecular dynamics [2] and simulated annealing [4], face an exponential growth in the number of metastable states and the inability to reach the ground state. Significant progress was achieved by the authors of Ref. [5] using a more advanced hybrid Basin-Hopping / Divide & Conquer algorithm and fixing the number of boundary particles $N_{b}$, which allowed them to find minima believed to be global. However, in terms of computation time, the result is achieved far beyond polynomial scaling.
        In the present work, we propose a physically motivated approach that enables computations for large $N$ (up to tens of thousands) in linear time without the use of high-performance computing systems. The key idea of the method consists in introducing a feedback loop based on the analysis of the local stress field. In each cycle, the algorithm performs the following steps:
        1. Diagnostics of the defect structure: the local energy, serving as a measure of mechanical stress, is calculated for each particle. Statistical analysis of the distribution of these energies allows for the identification of energetically "hot" spots — dislocations, disclinations, and elastic deformation zones, which, according to the theory [4], are responsible for lattice frustration and the formation of extended linear defects with a non-zero net charge.
        2. Targeted stress relaxation: the proposed approach focuses computational resources exclusively on the identified "hot" regions. The adaptive radius of the optimization zone is cyclically varied in order to eliminate both point defects and extended dislocation chains. Additionally, following the methodology of Ref. [5], the optimal number of boundary particles $N_{b}$ is used, which exponentially accelerates the search.
        Previously known results have been confirmed numerically with high accuracy, and for $N=61$ and $N=64$, configurations with an energy lower than that published in Ref. [5] have been found. The maximum performance of the method is demonstrated for the record cluster of $N=40886$: using a physically justified number of boundary particles, the algorithm achieved an energy of $E_{min}=1300045675.26$ in just $18138$ seconds of machine time on a standard PC; the relative difference from the best-known result [5] is $∼1.46×10^{-6}$.
        The obtained results demonstrate that integrating physical insights into the energy landscape structure of the defect subsystem directly into the computational algorithm allows for a dramatic reduction in the cost of searching for the global minimum in classical Coulomb clusters.

        References
        1. J. J. Thomson, Philosophical Magazine, Series 6, vol. 7, no. 39, pp. 237–265 (1904)
        2. R. G. Nazmitdinov, A. Puente, M. Cerkaski and M. Pons, Phys. Rev. E 95, 042603 (2017)
        3. S. Smale, Math Track, School of Mathematical Sciences, vol. 1, no. 1, pp. 5-30 (2005)
        4. A. Mughal, M. A. Moore, Phys. Rev. E 76, 011606 (2007)
        5. P. Amore, U. Zarate, Phys. Rev. E 108, 055302 (2023)

        Speaker: Georgiy Lavrov (DLNP JINR)
      • 14:45
        A numerical study of the free-fall three-body problem with Pythagorean masses 15m

        This work is motivated by the famous Pythagorean three-body problem (Burrau's problem of three bodies). The problem is to find the trajectories of three bodies with masses of 3,4,5 that are gravitationally attracted to each other. Аt t=0 the bodies have zero velocities (have free-fall initial conditions) and are located at the apexes of a right triangle with sides 3,4,5, so that the corresponding masses and sides are opposite one another. The problem has exponential sensitivity to small changes in the initial conditions (a property of chaotic systems), hence being difficult to calculate. This sensitivity can be overcome by using a high-precision floating-point arithmetic.
        In the current work, we consider the 2D Agekyan-Anosova's domain corresponding to all possible free-fall initial conditions and for all possible permutations of the masses of 3,4,5. Burrau's problem corresponds to a point in this domain. This means that only the masses remain Pythagorean, and not the geometry of the initial conditions. The initial conditions considered are not fixed, but arbitrary free-fall ones.
        High-precision results for the decay times of the system in Agekyan-Anosova's domain are presented. A search for collisionless periodic free-fall orbits based on Newton's method is also conducted. The database of the known periodic orbits of the type considered is expanded significantly. As an ODE-integrator, the high-order and high-precision Taylor series method is used. The intensive high-precision computations are performed on the ''Govorun'' supercomputer and using the resources of the JINR Member States distributed information and computing environment (JINR DICE), Dubna, Russia.

        Speaker: Mr Zarif Sharipov (JINR)
      • 15:00
        Pion off-shell form factors 15m

        In the report, the off-shell electromagnetic pion form factors in the Bethe-Salpeter formalism are considered. The one-rank separable separable kernel of the quark-antiquark interaction is used to solve the equation analytically. The expressions for the form factors are four-dimensional integrals and calculated using the Feynman parameterization method. The obtained integrals are calculated in the C++ programming language using the VEGAS library. The Ward-Takahashi identities are proved numerically and importance of the interaction part of the electromagnetic current is demonstrated. The obtained results are compared with experimental data.

        Speaker: Mikhail Slautin (Dubna University/BLTP JINR)
      • 15:15
        A Lower Bound on the Rank of Integer Matrices with Peculiar Leading Diagonal 15m

        Some lower bound on the matrix rank is found for square integer matrices in which all but diagonal entries are divisible by a given integer number. The bound is equal to the ratio of the order to the number of prime divisors of the common factor of the off-diagonal entries. Infinite series of examples are found where the lower bound is optimal. In particular, we prove a new analogue of the well-known theorem on strictly diagonally dominant matrices. The transfer of the obtained results to rings of algebraic integers is also considered. In the same way, one can prove nondegeneracy of a perturbed matrix over the ring of polynomials. We demonstrate the possibility of applying our approach to solve systems of ordinary differential equations.

        Speakers: Alexandr Seliverstov (Institute for Information Transmission Problems of the Russian Academy of Sciences (Kharkevich Institute)), Ksaverii Malyshev (Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University; Peoples' Friendship University of Russia named after Patrice Lumumba)
      • 15:30
        Spinor Field Dark Energy in FLRW Space-time: Theory and Observation 15m

        This work investigates a nonlinear spinor field unified dark energy model in a spherically symmetric Friedmann--Lema\^{\i}tre--Robertson--Walker (FLRW) spacetime, where the spinor self-interaction naturally reproduces various types of dark energy such as quintessence, generalized Chaplygin gas (GCG) and a Modified Chaplygin Gas (MCG). Unlike scalar-field or $\Lambda$CDM models, the spinor field generates non-diagonal components of the energy--momentum tensor as a consequence of spacetime geometry, providing new insight into the coupling between spinor dynamics and cosmic expansion. We confront the model with recent late-time observations, including cosmic chronometer measurements of $H(z)$, Type~Ia supernovae from the Pantheon+ compilation, and baryon acoustic oscillations (DESI DR2). A Markov Chain Monte Carlo analysis yields tight constraints on the model parameters, with a best-fit Hubble constant $H_{0}\simeq 67.78\,\mathrm{km\,s^{-1}\,Mpc^{-1}}$. The model reproduces late-time cosmic acceleration with a present-day deceleration parameter $q(0)\simeq -0.493$ and yields information criteria values that are competitive with $\Lambda$CDM. These results demonstrate that the spinor-field realization of MCG provides a viable and geometrically enriched alternative to standard dark energy scenarios.

            It was found that 
            Spinor field provides a viable model for dark energy;
            Fits observational H(z) and SN data well;
            Future work: Include CMB constraints, explore non-minimal couplings.
        
        Speaker: Prof. Bijan Saha (Laboratory of Information Technologies, Joint Institute for Nuclear Research)
      • 15:45
        Inelastic backscattering of high-energy electrons from He: Calculational details and approaches 15m

        See content in Abstract_Popov.zip

        Speaker: Yuri Popov (Nuclear Physics Inst., Moscow State Univ., BLTP JINR)
    • 16:00 → 16:20
      Coffee break 20m
    • 16:20 → 18:20
      Mathematical methods and tools for modeling complex physical systems Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 16:20
        Breaking of Structural Symmetry and Topological Defects in the Classical Thomson Problem on a Disk at Zero Temperature 15m

        The classical Thomson problem on a plane [1] serves as a canonical model for studying Wigner crystallization and spontaneous symmetry breaking. At absolute zero temperature, the particle configuration is determined solely by the minimum of the Coulomb energy. The key physical aspect of the problem is to find a compromise between the isotropic geometry of the disk boundary and the intrinsic tendency of the system to form an anisotropic hexagonal lattice.
        Based on the analysis of computer simulation results for a wide range of $N$, as well as data from [2,3] confirmed by our calculations, it is established that the transition from a system with a small number of particles to the thermodynamic limit $(N \rightarrow \infty)$ is accompanied by a change in structure and the appearance of topological defects. Depending on the value of $N$, different types of ordering and symmetry are formed in the system.
        For $N \lesssim 100$, the boundary symmetry dominates. The system minimizes the boundary energy, and particles arrange themselves into concentric rings [2]. Each ring has a fixed capacity; when this capacity is exceeded, the structure rearranges (a particle moves inward). The competition between the boundary geometry and the interparticle interaction generates multiple defects with coordination numbers 5, 7, 8, 9 in the disk interior. For a number of values $N = \left\{ 62,\ 63,\ 71,\ 72,\ 77,\ 78,\ 79,\ 83,\ 84,\ 89,\ 95,\ 96,\ 97 \right\}$ this leads to the formation of isolated disclination in the center with coordination number 7, and in the case $N = 68$ with coordination number 8, forming a topological pattern $[1,\ 8,\ 16,\ 43]$.
        For $100 \lesssim N \lesssim 300$, a “plateau” appears in the configurational entropy dependence, which is characterized by strongly varying defect configurations in the ground state as $N$ increases. In the composition of defects, a relatively large number of disclinations with coordination numbers 5, 7 and 8 are observed, and isolated disclinations with coordination number 7 in the disk interior become stable.
        For $300 < N < 1000$, a conflict arises between the center and the periphery. A hexagonal domain forms in the central part of the disk, while the ring structure is retained near the boundary. Individual topological defects in the form of dislocations (pairs of pentagons and heptagons) appear, necessary to reconcile the curved geometry with the flat lattice. The relative number of octagons in the Voronoi decomposition begins to steadily decrease, and their formation is possible only closer to the disk edge.
        For $N \gtrsim 1000$, defects become linear. Individual dislocation pairs merge into chains – so-called “scars”, if the total charge of such a defect is equal to $\pm 1$ and "pleats" if it is equal to 0 [3]. Linear defects can start either at the boundary or in the bulk of the system. Scars separate perfectly ordered hexagonal regions slightly rotated with respect to each other. The system resembles a mosaic of almost perfect crystals separated by defective lines (“cracks”).
        Numerical calculations show that the density of local energy minima grows exponentially with $N$. In the thermodynamic limit, precise determination of the global minimum becomes practically impossible, and the system behaves like a Coulomb glass: metastable states dominate the dynamics.
        For $N \lesssim 400$, the ground state energy minimum is achieved through the deformation and subsequent destruction of the hexagonal lattice in the central region of the disk, giving rise to dislocations and disclinations. For certain $N$, the energy minimum is realized when defects are symmetrically distributed throughout the entire system, forming a rotational $(C_{2},C_{3},C_{4},C_{6})$ or axial $(D_{1})$ symmetry. Breaking the symmetry in the defect arrangement increases the configurational energy, making the global minimum unattainable. For $N \gtrsim 400$, the relative number of defects decreases, and a domain structure becomes energetically more favorable: Wigner lattices with identically oriented hexagonal cells separated by dislocation chains.
        Thus, the transition from a ring-like to a domain structure of the Wigner crystal is governed by fundamental topological laws and is accompanied by a successive change of structural regimes and increasing complexity of defect topology, which leads to the emergence of geometric frustration as the system strives for the ground state energy minimum. Understanding these regularities is critically important for the design of mesoscopic electronic and colloidal systems.

        References
        1. J. J. Thomson, Philosophical Magazine, Series 6, vol. 7, no. 39, pp. 237–265 (1904)
        2. R. G. Nazmitdinov, A. Puente, M. Cerkaski and M. Pons, Phys. Rev. E 95, 042603 (2017)
        3. P. Amore, U. Zarate, Phys. Rev. E 108, 055302 (2023)

        Speaker: Eduard Nikonov (LIT JINR)
      • 16:35
        Regularized Equations for Dynamics of Heterogeneous Binary Mixtures and Water–Vapor Phase Transitions 15m

        We apply a system of four balance PDEs for dynamics of heterogeneous compressible binary mixtures with common velocity, temperature and pressure for gas or liquid components which are described by the stiffened gas equations of state. Elimination of volume fractions of components from the unknown functions is possible leading to the quasi-homogeneous form based on an explicit quadratic equation for the common pressure. We further apply to this form a regularization of the quasi-gasdynamic type. We also exploit a description of the phase transition in the limit of instantaneous relaxation to thermodynamic equilibrium between components that is reduced to nonlinear equations for the saturation temperature and pressure. This approach is implemented numerically by an explicit conservative two-level in time and symmetric three-point in space finite-difference scheme without limiters and with splitting with respect to physical processes (in the 1D case). Examples of computations of the water–vapor phase transitions are presented.

        Details can be partially found in: A. Zlotnik, T. Lomonosov, Application of Regularized Equations for Dynamics of Heterogeneous Binary Mixtures for Modeling Water–Vapor Phase Transitions, Mathematical Models and Computer Simulations, 2025. Vol. 17. No. 3. P. 352–364.

        Speaker: Alexander Zlotnik (Higher School of Economics University)
      • 16:50
        Electrical conductivity of transparent conducting films based on crack templates 15m

        Transparent conducting films based on crack templates are often modeled using random resistor networks. In our work, such networks are constructed as the edges of a two-dimensional Poisson--Voronoi diagram. Two types of networks are considered: the original one, in which the conductivity of each edge is inversely proportional to its length, and the effective one, where all edges have the same conductivity obtained from effective medium theory. For analytical estimation of electrical conductivity, the mean-field approximation is used, based on the assumption that potential fluctuations at the nodes are small. Direct numerical calculations for the Poisson--Voronoi diagram show that the mean-field approximation without taking into account potential fluctuations overestimates the conductivity of the original network by about 13\%, and for the effective network by 79\%. Taking into account potential fluctuations at network nodes does not improve these estimates. The obtained results show that when modeling transparent conducting films based on crack templates, especially in the case of hierarchical crack networks with variable width of cracks (where resistance is not simply proportional to length), the direct application of the mean-field approximation can lead to significant errors.
        Additionally, a hexagonal network with a distribution of edge conductivities corresponding to the Poisson--Voronoi diagram is investigated: for it, the predictions of effective medium theory are more accurate than for the Poisson--Voronoi diagram, which is explained by the greater structural homogeneity of the periodic hexagonal lattice.
        In our opinion, this result for the hexagonal network with random edge conductivities is interesting. Although Foster's theorem relies only on the fact that the degrees of all graph vertices are equal, the EMT estimate for the hexagonal network turned out to be better than for the regular graph corresponding to the edges of a PVT. We attribute this to the fact that the PVT is less homogeneous compared to the hexagonal graph.
        A.V.E. acknowledges funding from the Russian Science Foundation, grant No.~25-21-00460.

        Speaker: Mr Andrei Eserkepov (Astrakhan Tatishchev State University)
      • 17:05
        Methods and algorithms for parallel modeling magnetostatic fields of a compact electromagnetic device of experimental setup to study rot effects in fission of heavy nuclei 15m

        The report is focused on more detailed presentation of mathods and algorithms that were used for parallel modeling magnetic firld of a compact electromagnetic device. It has a number of specific features [1], therefore, for simulation based on formulation of the magnetostatic problem for two scalar potentials, the finite element method was applied. Usually, the error of the approximate solution in this method is estimated as O(h^{p+1}), where h is the diameter of the mesh cell and p is the order of the approximating polynomials (1< p < 5). The error can also be estimated as O(N^{-(p+1)/3}), where N is the number of unknowns in the finite element discretization. A block structured conforming mesh with cells in the form od parallelepipeds is used. This allows us to reduce the number of data excanges between the cores memory and use an economical variant of the matrix-free method because the Jacobian is the same within each block when calculating the element matrices. Note that at the same time, each block can contain cells of different diameters, and also each block can have its own approximation order. To calculate the elemnt matrices from standard elements, we use the tensor product of the one-dimensional Lagrangian basis functions of different order in each direction with the special rules for parallel computation of cubature formulas. The finite element system of equations is solved by an economical variant of the matrix-free method with the diagonal preconditioned conjugate gradient (DPCG) method in version of merged vector operations. This approach also significantly reduces the number of data exchanges between the cores, and the method is more fast than the standard DPCG. As calculations show [1], the experimental requirements for the homogeneity of the magnetic field inside the electromagnetic device are satisfied already at N=1.7 *10^{7} and p=2.

        [1] M.B. Yuldasheva, O.I. Yuldashev, V.V. Novitsky. Modelling magnetic field of a compact electromagnetic device of experimental setup to study ROT effects in fission of heavy nuclei. Russ. J. Numer. Anal. Math. Modelling 2025, 40(2): 153-167.

        Speaker: Yuldashev Oleg
      • 17:20
        Numerical solution of nonstationary problems with memory 15m

        This talk is dedicated to the numerical solution of evolutionary integrodifferential equations with memory, specifically those involving difference kernels of convolution type. Such models generalize classical parabolic and hyperbolic partial differential equations and are fundamental in various applications, including heat conduction in media with memory and viscoelasticity. The central challenge is the nonlocality in time, where the solution’s evolution depends on its entire history. To address this, a systematic computational approach is outlined that reduces the nonlocal problem to a local system of ordinary or partial differential equations. The method begins by approximating the positive-definite difference kernel using a Prony series, which is a sum of exponentials. By introducing auxiliary variables defined via convolution integrals with these exponential kernels, the original integrodifferential equation is transformed into an equivalent system of local evolution equations without memory. Spatial approximation using finite elements or finite volumes results in a Cauchy problem for a system of first- or second-order ordinary differential equations in a finite-dimensional Hilbert space, where stability estimates are derived based on the properties of the operators and the kernel. Time discretization is performed using a two-level scheme with a weight, including backward Euler and Crank–Nicolson schemes, and unconditional stability is proven for a range of the weight parameter. The key result is a computationally efficient algorithm where the complexity of solving the nonlocal problem is only marginally greater than that of solving a standard local evolution problem with the same spatial operator. The talk concludes by extending the methodology to more general equations, including those with memory in the solution’s derivative, second-order problems, and the heat conduction equation with memory from Gurtin–Pipkin theory.

        Speaker: Petr Vabishchevich (Lomonosov Moscow State University)
      • 17:35
        Symbolic recursion method for strongly correlated quantum dynamics 15m

        Dynamics of strongly correlated quantum matter have long been considered to be quantitatively intractable beyond phenomenological or perturbative approaches. However, currently the field is being transformed by new ideas revolving around the phenomenon of operator growth. I will report an advanced implementation of the recursion method – a technique to solve coupled Heisenberg equations in the tridiagonal operator basis – and its application to near- and far-from-equilibrium dynamics of strongly correlated spin-1/2 and fermionic systems in one, two and three dimensions [1-5]. From the computational point of view, the core routine of the method is the symbolic computation of nested commutators between the system Hamiltonian and the observable of interest [1-3]. The result of this computation is a colossal computer formula containing Pauli matrices/fermionic creation and annihilation operators along with symbolic parameters entering the Hamiltonian. After this formula is obtained and stored in computer memory, various post-processing techniques are applied to convert it to correlation functions, transport coefficients or time dependence of the observable after a quantum quench [2-5]. Thanks to the symbolic nature of computation, a single run of the core routine covers the whole parameter space of the microscopic model, as well as the whole space of initial states for the quench problem, resulting in a drastic gain in the efficiency of the method.

        [1] O. Lychkovskiy, V. Khrushchev, I. Shirokov. QCommute: a tool for symbolic computation of nested commutators in quantum many-body spin-1/2 systems // arXiv: 2604.04778.
        https://github.com/QCommute
        [2] F. Uskov, O. Lychkovskiy. Quantum dynamics in one and two dimensions via the recursion method // Phys. Rev. B 109, L140301 (2024).
        [3] I. Ermakov, O. Lychkovskiy. Symbolic recursion method for strongly correlated fermions in two and three dimensions // arXiv: 2512.23678.
        [4] A. Teretenkov, F. Uskov, O. Lychkovskiy. Pseudomode expansion of many-body correlation functions // Phys. Rev. B 111, 174308 (2025).
        [5] I. Shirokov, V. Khrushchev, F. Uskov, I. Dudinets, I. Ermakov, O. Lychkovskiy. Quench dynamics via recursion method and dynamical quantum phase transitions // arXiv: 2503.24362.

        Speaker: Oleg Lychkovskiy (Skolkovo Institute of Science and Technology)
      • 17:50
        Analytical and Numerical Calculations within the Framework of the Basic Element Met 15m

        A new approach is presented to analytical and numerical solving
        polynomial approximation problems on an interval of limited length. A new kind
        of polynomials has been developed within the frame of the basic element method
        (BEM, BEM-polynomials) [1]. On the base of BEM the Taylor polynomial is
        transformed into the hierarchical BEM-polynomial, which fuses the properties of
        n-degree Taylor polynomial and a second-degree Lagrange polynomial on a three-
        point grid. Maximal order of derivatives at formulas for calculating coefficients of
        the BEM polynomial is smaller than a degree of the approximating polynomial
        more than three times. This significantly reduces the computational complexity of
        the algorithms while preserving calculation precision. The BEM-polynomials were
        used for effective solution several current tasks. Among them are the method for
        the segmentation of 2D-contours with a complex geometrical form [2], which is
        relevant for robotics and for AI; a fundamentally new explicit method for
        numerical solving of the stiff Cauchy problem based on the predictor-corrector
        scheme [3]; neutron noise analysis has been using the fifth-degree BEM-polyno-
        mials [4]. The higher-degree BEM-polynomials are used to normalize the spectra
        during investigation of pulse energy fluctuations of the IBR-2M reactor [5] etc.
        References
        [1] N.D. Dikusar. The Basic Element Method, Math. Models and Comp. Simulat.,
        2011, Vol. 3, No.4, pp. 492–507. https://doi.org/10.1134/S2070048211040053
        [2] Nikolay Dikusar. Shape Approximation Based on Higher-Degree Polynomials,
        Proceedings of MMCP2017, The European Physical Journal, EPJ Web of
        Conferences “Mathematical Modeling and Computational Physics”,
        DOI:10.1051/epjconf/201817301003, www.epj-conferences.org, 173(4):01003
        [3] N.D. Dikusar, Numerical Solution of the Cauchy Problem Based on the Basic
        Element Method. ISSN 2070-0482, Mathematical Models and Computer
        Simulations, 2023, Vol. 15, No. 6, pp. 1024–1036. DOI: 10.20948/mm-2023-05-06
        [4] N.V. Korepanova, N.D. Dikusar, Yu.N. Pepelyshev, M. Dima. Neutron Noise
        Analysis using the Basic Element Method . Annals of Nuclear Energy, 131,
        (2019). 475-482. https://doi.org/10.1016/j.anucene.2019.04.008
        [5] Pepelyshev Yu.N., Dikusar N.D., Tsogtsaikhan Ts., Sumkhuu D. Investigation
        of Pulse Energy Fluctuations of the IBR-2M Reactor Using the Basic Element
        Method. JINR preprint R13-2025-49, accepted for publication in the journal
        "Problems of Atomic Science and Technology. Series: Physics of Nuclear
        Reactors" (in Russian)

        Speaker: Nikolay Dikusar (JINR)
      • 18:05
        Analysis of the resonant properties of the IBR-2M reactor during a cycle using the basic elements method 15m

        The fluctuations of the IBR-2M reactor pulse energy during the third cycle of 2025 were studied using the basic elements method. The spectrum was divided into a resonant component and a line component. It was shown that the resonant noise component (peak at ~0.1 Hz) coincides completely with the amplitude-frequency characteristic derived from the reactor dynamics model. The amplitude-frequency characteristic was analyzed using estimated power feedback parameters obtained experimentally at the beginning, middle, and end of the same cycle. These results confirm the accuracy of the dynamics model and the applicability of the basic elements method for reactor condition monitoring.

        Speaker: Sumkhuu Davaasuren (Mongolia)
    • 16:20 → 18:20
      Modern methods for data processing and analysis in Mega-science projects in HEP room 310 (MLIT)

      room 310

      MLIT

      • 16:20
        Time-reversed anisotropic diffusion approach for probing the Galactic contribution to the TeV–PeV astrophysical neutrino flux 15m

        A new quantitative likelihood measure for testing the Galactic vs. extragalactic origin of high-energy astrophysical neutrinos is proposed. The method involves two steps: (i) reconstruction of the neutrino production point and of the parent cosmic-ray particle parameters and (ii) backtracking the parent cosmic-ray particle in the stochastic magnetic field via a three-dimensional time-reversed anisotropic diffusion equation. The method is applied to the IceCube Neutrino Observatory TeV–PeV events. The Galactic contribution to IceCube’s total flux is discussed.

        This work is supported in the framework of the State project “Science” by the Ministry of Science and Higher Education of the Russian Federation under the contract 075-15-2024-541.

        Speaker: Nickolay Martynenko (MSU & INR RAS)
      • 16:35
        Overlapped signals delimitation for HEP experiments 15m

        The third phase of the Large Hadron Collider's operation continues, characterized by higher luminosity and an increased multiplicity of overlapping signals.
        In the cathode-strip chambers of the CMS experiment, when charged particles pass through, electron avalanches are formed, which are registered as clusters (groups of signals from adjacent strips) with charges whose distribution in space can be approximately described by Gauss functions. Often there is an overlap of clusters from closely passing particles, which can lead to significant losses in accuracy in determining their coordinates.
        The currently used simple center of gravity algorithm does not provide the required accuracy for overlapped signals (the error is up to 40% of the strip width with the required accuracy of up to 5%).
        To solve this problem, a sequential elimination algorithm, wavelet analysis and the use of Kolmogorov-Arnold neural networks are considered. The results of overlapped signals delimitation for all algorithms are presented and discussed.

        Speaker: Dmitry Fomenok (JINR)
      • 16:50
        3D hit reconstruction in End-Cap straw trackers in High Energy Physics 15m

        Precise 3D hit reconstruction in the end-cap regions of tracking detectors presents unique challenges due to non-uniform magnetic fields, high particle flux, and complex geometry.
        This work presents a novel algorithm for reconstructing three-dimensional space points from crossed straw tubes in the end-cap sub-system. The method relies on hit recovery through tangent planes. Performance is evaluated using SPDroot simulations for the SPD detector at NICA.
        The results demonstrate a 3D resolution of ca. 160 microns, significantly improving track seeding efficiency in the high rapidity regions compared to conventional 2D projection methods.
        The reconstruction is computationally efficient, ideal for real-time processing on
        CPU based architectures. These developments are critical for the continuous DAQ system of SPD,
        providing a robust foundation for event tagging and efficient data to disk acquisition

        Speaker: Maria Dima (JINR - DLNP)
      • 17:05
        Geant4 hadronic model developments for HEP experiments 15m

        In this work, we present the status of several models in the Geant4 software package for modeling hadronic processes in high-energy physics. We focus on models, which are important for Monte Carlo simulation of experiments at the collider NICA. The latest updates are discussed in the de-excitation module, the pre-compound model, and in the modeling of elastic interactions. Validation of hadronic models is performed versus experimental data, which will be discussed in detail. Additionally, a new charge exchange model for pions and kaons is presented.

        Speaker: Nikita Chalyi (Tomst State University (TSU))
      • 17:20
        Studies of Quark- and Gluon-Jet Properties at CMS: Methods, Challenges,and Solutions 15m

        The talk is devoted to methods for the experimental study of quark (q)
        and gluon (g) jet properties at hadron colliders. Measurements of
        q/g-jet properties begin with the determination of the q- and g-jet
        fractions in the investigated jet sample. This is achieved using model
        q/g templates, i.e. distributions of observable jet characteristics
        obtained within a given model. Consequently, measurements of q/g-jet
        properties are intrinsically model-dependent. The talk discusses the
        problem of defining model uncertainty, the main types of model
        uncertainties, and methods for their quantitative evaluation and
        treatment in the analysis of experimental data at the LHC.

        Speaker: Сергей Шульга (JINR)
      • 17:35
        Polarized gamma-gamma collisions in the ReneSANCe event generator 15m

        We give a report on new developments in the gamma-gamma collision mode of the ReneSANCe Monte Carlo event generator. Results for a number of polarized Standard Model processes at low and high energy will be presented.

        Speaker: Renat Sadykov (JINR)
    • 16:20 → 18:20
      Numerical and analytical calculations in modern mathematical physics Room 406 (MLIT)

      Room 406

      MLIT

      • 16:20
        High-performance computer simulation of complexation of fullerenes and fullerenols with the peptide dendrimer Lys2Gly 15m

        Using high-performance computer simulations, we studied the complexation of fullerenes and fullerenols with lysine-based dendrimers. Fullerenes have potential for biomedical applications. However, these molecules are poorly soluble in water. Fullerenols are more soluble in water due to the attached OH groups. Their hydrophilicity can be increased by encapsulating fullerenols in dendrimer-based nanocontainers. We chose a peptide dendrimer because it consists of amino acids, which are highly compatible with the human body.
        Computer simulations were performed using molecular dynamics (MD) in the Gromacs package with the AMBER-99SB-ILDN force field using high-performance computing. The study examined fullerenes and C60(OH)n fullerenols, with n = 12, 24, and 36 is the number of OH groups, and the peptide dendrimer Lys2Gly. Each system contained of a single fullerene/fullerenol, a dendrimer, and counterions immersed in an aqueous solution.
        We demonstrated the formation of a stable complex within 10-100 ns for all studied systems. The dendrimer size remained practically the same in all cases. However, there are differences in the internal structure of the complexes. The fullerene is positioned almost at the center of mass of the dendrimer and remains firmly there. Fullerenols are often located near the center of mass of the dendrimer, although conformations deviating from this position exist. The fullerene and fullerenols displace part of the dendrimer atoms and part of the counterions from the dendrimer center of mass. This increases the zeta potential of the dendrimer in the complex by 50% compared to the dendrimer outside the complex. The increase in the number n of OH groups in the fullerenols practically does not affect their complexation with dendrimer but changes its mechanism: the number of hydrophobic contacts decreases while the number of hydrogen bonds between the dendrimer and the fullerenol increases compensating the loss of hydrophobic contacts with n.
        MD simulations demonstrate that peptide dendrimer Lys2Gly performs well as a nanocontainer both for fullerenes and fullerenols, ensuring solubility and maintaining their high biological activity. This simulation result is in good agreement with the experimental data.

        Speaker: Oleg Shavykin (St. Petersburg State University, 7/9 Universitetskaya nab., 199034 St. Petersburg, Russia)
      • 16:50
        Equation of State in the Directed Percolation Universality Class: Numerical Three-Loop Analysis 15m

        We investigate the critical behavior of directed percolation within the framework of the field-theoretic renormalization group. In particular, we focus on the perturbative evaluation of the equation of state and the corresponding universal amplitude ratio in the $\varepsilon = 4 - d$ expansion. Our work extends previous analyses toward three-loop order and develops a numerical technique for carrying out the required higher-order calculations. We reproduce the known two-loop results as a consistency check and present the complete set of Feynman diagrams contributing to the three-loop approximation. These results constitute a step toward a systematic high-order determination of universal amplitude ratio in the directed percolation universality class.

        Speaker: Lukas Mizisin (BLTP, JINR)
      • 17:05
        Multiple Right-Hand Sides and Multi-GPU CUDA Implementations of the Tridiagonal Partition Method 15m

        The parallel partition algorithm for solving systems of linear algebraic equations (SLAEs) with tridiagonal coefficient matrices proposed in [1] represents a conceptually simple yet highly efficient numerical method. It operates by decomposing the original matrix into smaller sub-matrices, thereby enabling the concurrent solution of smaller SLAEs. The algorithm was originally designed for execution on systems with a large number of processors and accordingly, the computational experiments reported in [1] use up to 250 processors.

        There exists a class of problems which requires solving a series of tridiagonal SLAEs with the same coefficient matrix, but different right-hand sides (RHSs). For instance, problems discretized by the finite-difference methods, and computations of cubic and B-splines, where the RHS changes for each fit. These systems are normally solved in a batch fashion. The big computational intensity of the task justifies the usage of GPUs for its solution. Here, we are going to present our CUDA multiple RHSs implementation of the partition method. A discussion for the difference between an aggregate and an interleaved data storage format for the RHSs is made. Computational experiments for different SLAE sizes with different number of RHSs are conducted. The obtained results are compared with the cuSPARSE multiple right-hand sides solver gtsv2_nopivot.

        The tridiagonal partition algorithm contains sufficient parallelism to justify the usage of multiple GPUs to improve performance, and/or to help handling SLAEs which are too big to be solved on a single GPU due to global memory limitations. A multi-GPU CUDA implementation and a multi-GPU recursive CUDA implementation of the partition method are made. The usage of pageable vs. pinned memory is investigated. Computational experiments on one and two GPU cards, for different SLAE sizes and with different number of recursive steps are conducted. The findings regarding the optimal number of recursions for single-GPU execution, as discussed in [2], remain valid in the context of dual-GPU execution. The efficiency of the recursive partition method when moving from one to two devices is computed and discussed.

        [1] T. M. Austin, M. Berndt, and J. D. Moulton, A Memory Efficient Parallel Tridiagonal Solver, Preprint LA-VR-03-4149 (2004) 13 p.

        [2] M. Veneva, ML-Based Optimum Sub-system Size for the GPU Implementation of the Tridiagonal Partition Method, Moscow University Physics Bulletin. Machine Learning in Fundamental Physics, 80 (3, Suppl.), pp. S1085--S1095, doi: 10.3103/S0027134925703138, arxiv: 2510.27351 [cs.DC] (2025).

        Speaker: Ms Milena Veneva (RIKEN)
      • 17:20
        A Modified Chebyshev Collocation Method for Solving Two-Point Problems for General Second-Order Linear ODEs 15m

        Chebyshev collocation methods have earned a well-deserved place among the many approximate numerical methods for solving linear and nonlinear ODEs of both first (single-point problems) and second order (two-point problems). A version of the modified Chebyshev Collocation Method for single-point problems for first-order ODEs was proposed and implemented in [1-2].

        In [3], sufficient conditions for the reducibility of a (homogeneous or non-homogeneous) second-order ODE to the form of a total derivative were established for a second-order non-homogeneous ODE. In this case, solving the original problem reduces to solving its "first" integral—the first-order ODE for which the MCCM was developed. If the sufficient conditions are not met, it is necessary to present (develop) a method for finding a "first" solution to a two-point problem for a second order (homogeneous or) non-homogeneous LODE using a modified Chebyshev Collocation Method. If one particular solution is known, the second is found by reducing the problem to an auxiliary first-order LODE. Given the general solution to the homogeneous LODE, we find the complete solution to the non-homogeneous LODE. Considering the general conditions in the "basic" formulation of the two-point problem using a modified Chebyshev Collocation Method reduces solving a two-dimensional SLAE. Its solvability or non-solvability is equivalent to the solvability or non-solvability of the original problem, which is an advantage of the Chebyshev Collocation Method over traditional Collocation Method.

        This paper proposes a stable Newton method based on a modified Chebyshev Collocation Method for finding the "first" solution to a second order homogeneous ODE.

        References
        1. L. A. Sevastianov, K. P. Lovetskiy, and D. S. Kulyabov, “Multistage collocation pseudo-spectral method for the solution of the first order linear ODE,” in 2022 VIII International Conference on Information Technology and Nanotechnology (ITNT), May 2022, pp. 1–6.
        2. Lovetskiy, K.P.; Sevastianov, L.A.; Hnatiˇc, M.; Kulyabov, D.S. Numerical Integration of Highly Oscillatory Functions with and without Stationary Points. Mathematics 2024, 12, 307.
        3. Lovetskiy K. P., Malykh M. D., Sevastianov L. A., Sergeev S. V., Solving a two-point second-order LODE problem by constructing a complete system of solutions using a modified Chebyshev collocation method. Discrete and Continuous Models and Applied Computational Science. 2024. Т. 32. № 4. С. 414–424.

        Speaker: Константин Ловецкий (RUDN University)
      • 17:35
        Continuous-corpuscular schemes of computational experiment 15m

        Tensor mathematics appears to be an adequate tool for the coherent description of inseparable spatial phenomena using linear approximations of smoothly changing properties in continuum mechanics. Spatial tensor arithmetic was historically developed to generalize and synthesize algorithms for direct computational simulation. This allowed, for example, to ensure algorithmic uniformity in cumbersome numerical schemes for aerohydromechanics. Key formalizations were synthesized in algorithms for separating modeling stages into independent physical processes of time integration.
        In numerical algorithms, formalizations of the laws of continuum mechanics have naturally developed using arithmetic and geometric formalizations in tensor notation. In conditionally independent or local bases, the laws of mechanics acquire extremely clear or canonical notations. In the present research, similar tensor approaches are developed for solving problems of thermodynamics and impurity diffusion, where the continuum stage involves constructing gradient surfaces followed by a corpuscular stage of modeling in accordance with kinetic theory. The resulting products or transformations in global reference frames in traditional analytical notation are extremely cumbersome and difficult to interpret for extremely complex convective dependencies. In computer notation, within local bases, physical laws acquire arithmetic form with independent tensor components surrounded by polarized corpuscles linked into a single system.
        In conclusion of the current work, the second corpuscular stage of the method of fixed cells and free but interdependent particles is highlighted.
        Recalling Isaac Newton's corpuscular interpretations of continuous medium dynamics, interesting results are obtained with sets of polarized corpuscles, which, depending on the laws of spatial interaction, form adequate forms of spatial arrangement. Previously, dense filling of a sphere was observed under the influence of conventionally "nuclear" potentials for attractive forces. In cases of particle repulsion in conventionally molecular interactions, all corpuscles are arranged on a sphere, illustrating surface tension. The dipole polarization of such particles implies a predetermined velocity of their free straight linear motion. The spatial arrangement of large particle conglomerates is easily disrupted by the ultra-low digital noise of computer approximations, after which they separate into small groups of mobile corpuscles, the multi-dipole bonds between which are significantly weakened by their relative orientation.

        Speaker: Vasily Khramushin (Saint-Petersburg State University)
      • 17:50
        Compensation constant of magnetocrystalline anisotropy K0 as a marker of spin-reorientation phase transition in a tetragonal ferromagnetic crystal. 15m

        In [1], a method is proposed for calculating the volumetric energy density of magnetocrystalline anisotropy (MCA) Ea(K0(T), Ki(T)) (i ≤ N) of a tetragonal ferromagnetic crystal with spin-reorientation transition (SRT), using N experimentally determined MCA constants Ki(T) , depending on the temperature T, and the calculated compensation constant K0(T) from the condition Ea ≥ 0. The report shows that the temperature of the SRT TSRT coincides with the zero values of the functions K0(T) = 0 and K1(T) = 0 at the SRT point. The condition K0(T)=0 also holds for easy-axis MCA type. This implies a close relationship between the compensation constant of the MCA and the SRT. The temperature-dependent value of this constant determines the degree of rearrangement of the analytical structure Ea during the SRT and can serve as a marker of this phase transition.
        An analysis of the structure of Ea is carried out as a function of the angles  and , which determine the orientation of the magnetization vector in a spherical coordinate system with the z- axis located along the axis of high symmetry of the ferromagnetic crystal under study. Here, we considered the most interesting case of the value φ = φm at the minimum point of the function Ea. In particular, the dependence of the set of critical points of Ea(θ,φm) is considered within the framework of catastrophe theory. It is shown that, from a mathematical point of view, the SRT represents a type A3 catastrophe.

        1. Pastushenkov Yu . G ., Tsvetkov V . P ., Tsvetkov A . I ., Mikheev S . A ., Tsvetkov I. V. Method for calculating the surface energy density of 180 0 Bloch domain walls in uniaxial ferromagnetic crystals. Phys. Rev. B. 2026 v.113. Iss.2. 024410. https://doi.org/10.1103/9f6n-q4gx.
        Speaker: Anton Tsvetkov
    • 18:30 → 19:30
      Welcome party
    • 09:25 → 11:00
      Plenary Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 09:30
        Schemes of Finite Element Method on Hyperparallelepiped Meshes 30m

        New finite element method schemes have been developed for solving d-dimensional self-adjoint boundary value problems (BVPs). Basis functions are constructed using multidimensional Hermite interpolation polynomials on hyperparallelepiped meshes. As test examples up to 6 dimensional BVPs for hyperellipsoids and Moshinsky atom is solved using additional nonorthogonal coordinate transformations.

        Speaker: Александр Гусев (Laboratory of Information Technologies Joint Institute for Nuclear Research)
      • 10:00
        On a Generic 2-Qubit Entangler 30m

        The report aims to present a unified landscape of the entanglement characteristics of 2-qubit states and their associated gates, framed within classification schemes that rely on the principle of local unitary symmetry in composite quantum systems. I will introduce the concept of a 2-qubit generic entangler, namely a 6-parameter SU(4) gate capable of generating any state within the nine entanglement classes of 2-qubit mixed states. The connection between such generic entanglers and the commonly used 3-parameter entanglers will be examined. Furthermore, the implementation of generic entanglers in parameterized quantum circuits will be explored.

        Speaker: Arsen Khvedelidze (Joint Institute for Nuclear Research)
      • 10:30
        Spatial confinement-deconfinement transition in accelerated gluodynamics within lattice simulation 30m

        In this work we investigate the influence of weak acceleration on the confinement-deconfinement phase transition in gluodynamics. Our study is carried out within lattice simulation in the comoving reference frame of accelerated observer which is parameterized by the Rindler coordinates. We find that finite temperature confinement-deconfinement phase transition turns into spatial crossover in the Rindler spacetime. In other words, spatially separated confinement and deconfinement phases can coexist in the Rindler spacetime within certain intervals of temperature and acceleration. We determine the position of the boundary between the phases as a function of temperature for several accelerations and find that it can be described by the Tolman-Ehrenfest law with rather good accuracy although a minor deviation takes place. Moreover, the critical temperature of the system in the weak acceleration regime is found to remain unchanged as that of the standard homogeneous gluodynamics. Our results imply that the spatial confinement-deconfinement transition might take place in the vicinity of the Schwarzschild black hole horizon.

        Speaker: Prof. Victor Braguta (JINR)
    • 11:00 → 11:20
      Coffee break
    • 11:20 → 13:00
      Plenary Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 11:20
        A digital platform for supercomputer modeling of complex technical processes 30m

        This report focuses on the development and validation of a digital platform designed
        for supercomputer modeling of complex technical processes.
        The platform is based on several classes of mathematical models used
        in the numerical solution of problems in gas dynamics, hydrodynamics,
        filtration, electrodynamics, and micro- and nanoelectronics.
        Numerical implementation of the models utilizes splitting by physical processes,
        the finite volume grid method on regular and irregular grids, particle methods,
        and multiscale approaches. Most of the developed numerical algorithms are adapted
        for parallel computing on modern supercomputer systems.
        The report provides examples of solutions to relevant applied problems.

        Speaker: Sergey Polyakov (Keldysh Institute of Applied Mathematics)
      • 11:50
        Numerical solution of the time-dependent Schrödinger equation and modeling of nuclear transfer reactions 30m

        The numerical solving of the time-dependent Schrödinger equation taking into account the spin-orbit interaction [1] is used to describe nucleon transfers reactions. A quantum description of the rearrangement of outer neutrons in collisions of nuclei makes it possible to study processes proceeding within various characteristic times [2]. In the case of a slow (adiabatic) relative motion of colliding nuclei, outer nucleons of colliding nuclei penetrate from one into another nucleus and populate slowly changing two-center (“molecular”) states. In the case of a fast (diabatic) relative motion the nucleons probability density does not have time to occupy the whole volume of the another nucleus, so that the change in it has a more local character [2]. The time-dependent model has shown that the rearrangement of loosely bound outer neutrons of $^{6}$He and $^{9}$Li nuclei in the collision process changes the real and imaginary parts of the interaction potential, and this may be a reason for the local enhancement of the total reaction cross section of the reactions $^{6}$He and $^{9}$Li with $^{28}$Si [2]. This time-dependent quantum-mechanical approach was successfully applied to describing a number of important processes in low-energy nucleus–nucleus collisions [3]. This approach makes it possible to study both multinucleon-transfer reactions involving heavy and superheavy nuclei and alpha-cluster transfers in reactions involving relatively light nuclei such as those of $^{12}$C and $^{16}$O [3]. The experimental cross sections for the production of the isotopes Au in $^{4,6,8}$He+Au reactions are explained based on numerical solution of the time-dependent Schrödinger equation for outer neutrons of He nuclei [4].
        1. V.V. Samarin, Phys. At. Nucl. 78, 128 (2015).
        2. Yu. E. Penionzhkevich, Yu.G.Sobolev, V. V. Samarin, and M. A.Naumenko, Phys. At. Nucl. 80, 928 (2017).
        3. V.V. Samarin, Phys. At. Nucl. 81, 486 (2018).
        4. V.V. Samarn, and M.A. Naumenko, Phys. At. Nucl. 85, 880 (2022).

        Speaker: Viacheslav Samarin (Joint Institute for Nuclear Research, Flerov Laboratory of Nuclear Reactions)
      • 12:20
        A Localized Kochen-Specker Theorem for Unsharp Measurements 30m

        Jose Manuel Ag\"{u}ero Trejo, Cristian S. Calude, Ovidiu C. Stoica

        Kochen-Specker Theorem is the non-probabilistic no-go theorem in quantum physics. The located Kochen-Specker Theorem shows that
        if we choose a value-definite vector in C^n, n>3, then every vector that is neither perpendicular nor parallel to it is value indefinite.
        3D QRNG generators based on measuring value-indefinite vectors have been studied theoretically (they do not require entanglement, generate quantum random strings that are maximally unpredictable, hence provably better than any pseudorandom strings) and experimentally. Robust experimental results probing the maximal unpredictability of quantum random strings generated by a 3D QRNG implemented with a superconducting transmon confirmed the theoretical results.

        A simpler implementation of a 3D QRNG based on the located Kochen-Specker Theorem uses photons. As a 3D QRNG works in a noisy environment due to various errors, we can no longer assume that the operators used to model the photonic embodiment are unitary. Hence, we model the 3D QRNG as an open quantum system, constructed as an arrangement of imperfect beam splitters with losses determined by the fidelity of the actual components. In this framework, we show that the process remains within the scope of the Kochen-Specker Theorem, thereby guaranteeing the maximal unpredictability of quantum random strings.

        Speaker: Cristian S. Calude
    • 13:00 → 14:00
      Lunch
    • 14:00 → 16:00
      Mathematical methods and tools for modeling complex physical systems Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 14:00
        Numerical simulation of the orientation symmetric top molecule by shaped laser pulse. 15m

        We numerically investigate the alignment and orientation of symmetric-top molecules driven by two-color laser pulses by solving the time-dependent Schrödinger equation. For fixed pulse energy and duration, shaped pulse envelopes yield a higher degree of alignment compared to Gaussian pulses. We analyzed how pulse duration affects the maximum achievable alignment and orientation, and showed decrease of both effects with increasing rotational temperature.

        Speaker: Eugene Koval (BLTP, JINR)
      • 14:15
        An investigation of the controllability and optimal control problems for the fractional dynamical systems of variable order 15m

        In this work the following system considered:
        $$ {}_0D^{\alpha(t)}_t q(t)=f(t,q(t),u(t)),\quad t\in[0,T], $$ with initial and final conditions $q(0)=q^0$, $q(T)=q^T$ correspondingly. Here ${}_0D^{\alpha(t)}_t$ --- fractional variable-order derivative, defined in [1], $u(t)$ --- control, $q(t)$ --- system state, $f(t,q(t),u(t))$ --- some known function. The minimal requirments to functions $u(t), q(t), f(t,q(t))$ consists in summability. For such system the two problems analyzed: the controllability and optimal control exisyence problem in case of $\alpha(t)=u(t)$ ($u(t)$ is unknown) and $f(t,q(t),u(t))=f(t,q(t))$ and the optimal control problem in case of $\alpha(t)$ is known function and $f(t,q(t),u(t))=f(t,q(t))+u(t)$. These problems investigateded using a moment method. Some conditions of controllability and optimal control exisyence derived. It's shown that optimal control problem with minimal control norm can be reduced to the generalized problem of moments.
        Obtained results can be useful in mathematical modeling of complex dynamical systems with non-stationary parameters, like the semiconductors, electrochemical and biological systems.

        [1] Garrappa R., Giusti A., Mainardi F.,Variable-order fractional calculus: A change of perspective, Commun. Nonlinear Sci. Numer. Simulat., \textbf{102}, 105904 (2021)

        Speaker: Sergey Postnov (V.A. Trapeznikov Institute of Control Sciences, MIREA - Russian Technological University, Moscow, Russia)
      • 14:30
        Application of the grid-characteristic method to the numerical study of anisotropy in fractured media 15m

        Numerical modeling of wave processes is relevant to geophysics, seismic exploration, nondestructive testing, medical diagnostics, and the study of metamaterials. This work focuses primarily on seismic exploration. Oil and gas fields are typically associated with porous and fractured geological structures. The presence of fractures leads to anisotropy, that is, a dependence of wave velocity on direction. Considering this effect improves the accuracy of seismic data interpretation.
        The present work is dedicated to the numerical evaluation of the anisotropic properties of fractured media. The equations of linear elasticity are used to describe an elastic solid, while the grid-characteristic method is applied for their numerical solution. To describe fractures, a special grid model is developed: the fracture region is represented by a separate grid with fluid parameters, surrounded by grids corresponding to the host elastic medium. This approach makes it possible to consider both the geometry of fractures and the influence of their filling material on the propagation of elastic waves.
        To demonstrate the method’s capabilities, a series of numerical experiments was conducted. In the first problem, the conditions of a laboratory experiment measuring anisotropy in concrete specimens containing water-saturated fractures were reproduced. Chimera meshes were used to accurately describe the boundaries of inhomogeneities, and hierarchically refined meshes were applied to improve computational efficiency. The obtained results were compared with experimental data and theoretical estimates. In addition, the dependence of the wave propagation velocity on the angle of inclination of a system of parallel fractures was investigated. The comparison of the obtained results demonstrated both qualitative and quantitative agreement with theoretical models. The effect of wavelength on the effective longitudinal wave velocity for various fracture orientations was also investigated. The results showed that the wave characteristics of the medium depend significantly not only on the orientation of the fractures but also on the ratio between the wavelength and the characteristic size of the inhomogeneities.
        Thus, the proposed numerical approach makes it possible to model the propagation of elastic waves in fractured media while accounting for the geometry of heterogeneities and fluid saturation of fractures. It is shown that the grid-characteristic method, combined with Chimera meshes and hierarchically refined grids, provides an adequate description of anisotropy and can be used to study a wide range of problems in seismic exploration of heterogeneous media. The study was supported by the Russian Science Foundation, project 23-11-00035-П.

        Speaker: Ilia Agrelov (MIPT)
      • 14:45
        Multi-agent approach to nuclear many-body system modelling 15m

        The large-amplitude collective motion of nucleons in fusion-fission and quasi-fission processes has presented a major challenge for theoretical and numerical modeling for many decades.
        For example, in calculations performed by V. Pashkevich in Dubna, high-dimensional minimization algorithms were applied in a deformation space with 10-20 parameters, and a catastrophe-based explanation was given for the discontinuities and fluctuations arising in this space [1,2].
        In this paper, a new multiagent algorithm for numerical simulation of the nuclear collective motion is proposed. The modeling is performed within the framework of the modified Vicsek Model, where the interaction of particles is determined by the nearest neighbor rule [3]. The Vicsek Model, introduced in 1995, is a minimalist agent-based model used to study the emergence of collective movement in various areas of modern complex systems science. In our approach, the Vicsek Model is modified by introducing realistic nuclear dynamics through an improved neighbor selection strategy.
        The developed algorithm is designed to simulate the occurrence of collective aggregation of nucleons due to short-range mutual interaction. It allows modeling the liquid drop component of the Strutinsky shell correction method, which combines the macroscopic smooth component with microscopic shell corrections.
        Our approach has potential for application in other areas related to the nuclear many-body problem and complex multiscale processes in nuclear physics [4,5,6].

        1. V. V. Pashkevich, Proceedings of the XV EPS Conference on Low Energy Nuclear Dynamics, St. Petersburg, Russia, (1995), p. 161.
        2. V. V. Pashkevich, A. V. Unzhakova, Int. Conf. «NUCLEUS 2014»: book of abstracts, Minsk, (2014), p. 179.
        3. A. S. Morse, A. Jadbabaie, J. Lin, IEEE Trans. Autom. Control, 48. (2003) p. 988.
        4. A. N. Sissakian, A. S. Sorin, M. K. Suleymanov et al., Physics of Particles and Nuclei Letters 5(1), (2005), p. 1.
        5. V. Vechernin, S. Belokurova, S. Yurchenko, Symmetry, 16 (1), (2014), p. 79.
        6. S. N. Mamaeva, A. N. Pavlov, N. A. Nikolaeva and G. V. Maksimov, Technical Physics, 69 (7), (2024), p. 2045.
        Speaker: Yury Ivanskiy (Saint Petersburg State University)
      • 15:00
        Analytical solutions of the Kuramoto–Sivashinsky equation with nonlinear convection 15m

        This report is devoted to the study a generalization of the Kuramoto-Sivashinsky equation taking into account nonlinear convection \cite{Kudryashov:Article_1}
        [u_t+u_{xx}+\sigma\,u_{xxx}+u_{xxxx}+\beta\,u\,u_x+\chi\,(u^2)_{xx}=0.]

        This equation is used to model processes in combustion physics, plasma physics, hydrodynamics, and other fields. Since the Cauchy problem for this equation cannot be solved by the inverse scattering transform, a traveling wave reduction ($u(x,t)=y(z)$ and $z=x-C_0\,t$) is employed to seek solutions.
        It is shown that the equation generally fails the Painlev\'e test. However, under additional constraints on the model parameters, the corresponding nonlinear ordinary differential equation (ODE) \cite{Kudryashov:Article_2}
        [C_1-C_0\,y+y_{z}+\sigma\,y_{zz}+y_{zzz}+\frac{1}{2}\beta\,y^2+\chi\,(y^2)_{z}=0, \quad (C_1 = const)]
        passes the Painlevé test. Using the results of the Painlev\'e analysis, first integrals for the nonlinear ODEs are derived. The general solutions for two specific cases, each involving four arbitrary constants, are presented in terms of the Weierstrass elliptic function and the transcendental solutions of the first Painlev\'e equation. Furthermore, exact solutions with one and two arbitrary constants are obtained using the simplest equation method \cite{Kudryashov:Article_3}.

        This research was supported by the Ministry of Science and Higher Education of the Russian Federation according to the state task project No. FSWU-2026-0006.

        \begin{thebibliography}{9}

        \bibitem{Kudryashov:Article_1}
        Aspe H., Depassier M. C. \textit{Evolution Equation of Surface Waves in a Convecting Fluid.} Physical Review A; 1990; 22(19): 4135–4142.

        \bibitem{Kudryashov:Article_2}
        Kudryashov N. A. \textit{Painleve Test, First Integrals and Exact Solutions of Nonlinear Dissipative Differential Equations.} Regular and Chaotic Dynamics; 2025; 30(5): 819–836. DOI: 10.1134/S1560354725050041.

        \bibitem{Kudryashov:Article_3}
        Kudryashov N. A. \textit{Traveling Wave Solutions of the Kuramoto-Sivashinsky Equation with Nonlinear Convection.} Physics Letters A; 2026;585:131644. DOI: 10.1016/j.physleta.2026.131644.

        \end{thebibliography}

        Speaker: Prof. Nikolay Kudryashov (National Research Nuclear University MEPhI)
      • 15:15
        APPLICATION OF THE CABARET SCHEME FOR MODELING TWOPHASE FLOWS IN PIPELINES 15m

        The paper investigates the application of the CABARET balance-characteristic scheme for the numerical simulation of gas-liquid flows in a one-dimensional approximation. The mathematical model is based on the Saurel–Abgrall hyperbolic system of equations (a two-phases seven-equation model) with separate pressures, velocities, and temperatures for each phase. The model is supplemented by accounting for interfacial friction, gravity, and relaxation processes.
        To implement the algorithm, the characteristic form of the system and the analytical form of Riemann quasi-invariants were derived. The numerical scheme is extended with procedures for the instantaneous relaxation of parameters (pressure, velocity,and temperature) based on the principle of splitting by physical processes.
        The verification of the method was carried out on a series of test problems. The simulation results of the gas bubble transport problem demonstrate the high accuracy of the scheme in describing contact discontinuities. A comparative analysis of the CABARET scheme and the second-order Godunov-type scheme (HLLC + MUSCL) in the double expansion problem (Einfeldt test) confirmed the advantage of the proposed approach in minimizing numerical dissipation.

        Speaker: Mr Mikhail Turbylev (Lomonosov Moscow State University (MSU))
      • 15:30
        Conservation laws for a family of generalized nonlinear Schrödinger equations 15m

        The report examines a family of generalized nonlinear Schrödinger equations \cite{Nifontov:Article_1}
        \begin{equation}\begin{gathered}\label{Nifontov:equation1}
        iq_t + \left(F_1(|q|^2)q\right){xx} + \left(F_2(|q|^2)q\right){xt} + H(|q|^2)q + iG_1(|q|^2)q_x + iG_2(|q|^2)q^2q^*_x = 0,
        \end{gathered}\end{equation}
        where $q(x,t)$ is a complex-valued function, $x$ and $t$ are independent variables, $i^2 = -1$, $F_1(|q|^2),$ $F_2(|q|^2),$ $H(|q|^2),$ $G_1(|q|^2)$ and $G_2(|q|^2)$ are real functions.
        Considered nonlinear partial differential equations depend on arbitrary functions, and the Cauchy problem for them cannot generally be solved by the inverse scattering transform. The functions presented in Eq. \eqref{Nifontov:equation1} have a physical meaning and are associated with processes occurring during the propagation of an optical pulse in a nonlinear medium. The function $F_1(|q|^2)$ corresponds to second-order dispersion effects, while $F_2(|q|^2)$ is related to an additional frequency shift. The function $H(|q|^2)$ characterizes the nonlinear dependence of the refractive index of a light pulse in an optical medium. The dependencies $G_1(|q|^2)$ and $G_2(|q|^2)$ make it possible to account for the effects of self-steepening of the pulse front and stimulated Raman scattering.

        We construct conservation laws for Eq. \eqref{Nifontov:equation1}. To achieve this, a modified multiplier method based on direct transformations of the equation is used \cite{Nifontov:Article_2, Nifontov:Article_3, Nifontov:Article_4}. We demonstrate that under certain constraints on the functions, the equation under study possesses several conservation laws. The paper presents a series of theorems on the construction of conservation laws with the obtained densities and fluxes.

        This research was supported by the Ministry of Science and Higher Education of the Russian Federation according to the state task project No. FSWU-2026-0006.

        \begin{thebibliography}{9}

        \bibitem{Nifontov:Article_1}
        Kudryashov N. A. \textit{Families of Nonlinear Schrödinger Equations in General Form with Exact Solutions.} Physics Letters A; 2025; 552: 130648.

        \bibitem{Nifontov:Article_2}
        Anco S. C., Bluman G. \textit{Direct Construction Method for Conservation Laws of Partial Differential Equations Part I: Examples of Conservation Law Classifications.} European Journal of Applied Mathematics; 2002; 13(5): 545–566.

        \bibitem{Nifontov:Article_3}
        Anco S. C., Bluman G. \textit{Direct Construction Method for Conservation Laws of Partial Differential Equations Part II: General Treatment.} European Journal of Applied Mathematics; 2002; 13(5): 567–585.

        \bibitem{Nifontov:Article_4}
        Kudryashov N. A., Nifontov D. R. \textit{Conservation Laws and Hamiltonians of the Mathematical Model with Unrestricted Dispersion and Polynomial Nonlinearity.} Chaos, Solitons & Fractals; 2023; 175: 114076.

        \end{thebibliography}

        Speaker: Daniil Nifontov (National Research Nuclear University MEPhI (Moscow Engineering Physics Institute))
      • 15:45
        About the hysteresis model in biological systems 15m

        The effective use of fractional differential models to describe the properties of biological materials is confirmed by numerous experimental results. In this paper a fractional differential model uses to solve the following problem: to determine the dependence of relative deformatioin on tense if the external action is a sequence of triangular pulses; to investigate the hysteresis. Among many models, the Caputo-Mainardi model was chosen for the numerical solution of the problem, since it more accurately reproduces experimental data describing the viscoelastic behavior of materials. The algorithm for numerical simulation of the solution was written in the Matlab programming environment version R2016a. The order of the Caputo fractional derivative was set by the user when running the program code. For a visual representation of the results, four dependencies were obtained: 1) stress versus time; 2) fractional derivative of Caputo; 3) dependence of tense on time; 4) dependence of tense on deformation. It is shown that the dependence of tense on deformation has a «loop-like» character, which indicates the presence of hysteresis, i.e. there is a delay in deformation relative to the applied stress and a mismatch between the loading and unloading paths of the material. The obtained simulation results clearly demonstrate that the model makes it possible to take into account not only the prehistory of material loading under pulsed load, but also to give numerical characteristics of its viscoelastic properties. Therefore, it will be useful in studying, for example, the strength properties of blood vessels or their pathologies.

        Speaker: Elena Postnova (Russian State Medical University)
    • 14:00 → 16:00
      Methods of quantum computing and quantum information processing Room 310 (MLIT)

      Room 310

      MLIT

      • 14:00
        Quantum reservoir classifiers based on decoherence-free subspaces 15m

        Quantum reservoir networks [ S. Ghosh et al., Quantum Information 5 (2019)35] make up an efficient class of quantum neural networks [M.Altaisky, N.Kaputkina, V.Krylov, PEPAN 45 (2014) 1013], where the whole learning process is concentrated in the output layer, whereas the random bulk connections provide high dimensionality of classification. Following our previous papers [M.Altaisky,N.Kaputkina, J.Innov. Sci. Sust. Tech. 5 (2025) 83; pp.6-12 in Proc. Days on Diffraction 2025, IEEE], we implement quantum reservoir network with an output fed to a basis of decoherence-free subspaces (DFS), which makes the information processing stable under environmental fluctuations [P.Zanardi, M. Rasetti, Phys. Rev. Lett. 79 (1997)3306]. We present results of numeric simulation of 4-qubit and 6-qubit reservoir networks harnessed for classification between separable and entangled 2-qubit input states. If a 4-qubit reservoir network needs a sampled trajectories for reliable classification, for a 6-qubit network a single-shot measurement of the output state turns to be sufficient. The simulation of reservoir dynamics was performed in Gorini-Kossakowski-Sudarshan-Lindblad approximation using standard qutip Python library [arXiv.org:2412.04705].

        Speaker: Dr Mikhail Altaisky (Space Research Institute RAS)
      • 14:15
        Quantum Version of Decision Tree Construction Algorithm Subroutine 15m

        \documentclass{article}
        \usepackage[english]{babel}
        \usepackage[utf8]{inputenc}
        \usepackage{johd}
        \usepackage{amsmath}

        \title{Quantum Version of Decision Tree Construction Algorithm Subroutine}

        \author{Ilnaz Mannapov \
        \small Kazan Federal University, Kazan, Russia \
        \small \tt{ilnaztatar5@gmail.com} \
        }
        \date{}

        \begin{document}
        \maketitle

        % \noindent{2026}

        \begin{abstract}
        \noindent
        Machine Learning is one of the most famous and promising areas of applied science. The classification problem is a widely-used approach.
        Decision Trees are one of the tools for solving classification problems.
        Many algorithms for constructing decision trees have been developed.

        As quantum computation is a very fast-growing field, it is used to solve such problems [\citep{PhysRevA.58.915}, \citep{lu2014quantum}].

        This paper investigates and constructs the main part of a decision tree construction algorithm: the subroutine for calculating the best split of the data.

        In our previous works [\citep{Mannapov2019}, \citep{kms2021_generic}, \citep{Mannapov2023}], we developed a quantum version of the decision tree construction algorithm based on classical algorithms such as C4.5, C5.0, and CART.

        Let us to give some preliminaries. Let ${\cal X}=\{X^1, X^2, ..., X^N\}$ is a training data set and ${\cal Y}=\{y_1, y_2, ..., y_N\}$ is a set of corresponding classes. $\sigma_{y=c_{i}} {\cal X}$ is a subset from training set which elements are related to class with number $i$.
        Here $X^i=\{x^i_1, x^i_2, ..., x^i_d\}$ is a vector of attributes, where $i\in\{1,\dots,N\}$, $d$ is a number of attributes, $A = \{a_1, a_2, ..., a_d\}$ is the set of all attributes,
        $N$ is a number of vectors in the training data set, M is a number of classes, $y_i \in C=\{1,\dots,M\}$ is an index of class of $X^i$ vector.
        The problem is to construct a function $F:DOM_1\times\ldots\times DOM_d\to C$ that is called classifier. The function classifies a new vector $X=(x_1,\dots,x_d)$ that is not from ${\cal X}$.

        Decision tree construction algorithms use a "divide and conquer" method to build a suitable tree.
        The main part of the algorithm is to find the best split of the data, where some attribute is used for splitting.
        The best split is determined based on a criterion such as Gini impurity, information gain, or mean squared error.
        It is the split that maximizes the gain, which is calculated from the impurity of the data before and after the split.
        The construction process stops when all data points belong to one class or when a maximum tree depth is reached.

        In this paper, we investigate the main part of this algorithm—the subroutine for calculating the best split of the data. We use the Gini impurity as the split criterion.

        The best split based on the Gini index is calculated using the following formula:
        $$ \text{BestSplit} = \arg\max_{s \in S} \text{Gain}(s) $$ where $S$ is the set of all possible splits and $\text{Gain}(s)$ is the gain of split $s$. The gain is calculated using the following formula: $$ \text{Gain}(s) = \text{Impurity}({\cal X}) - \sum _{i=1}^{k} \frac{|{\cal X}_i|}{|{\cal X}|} \text{Impurity}({\cal X}_i) $$ where ${\cal X}$ is the set of all data points, $k$ is the branch count of the split $s$, ${\cal X}_i$ is the set of data points that are in the $i$-th partition of the split $s$, and $\text{Impurity}({\cal X})$ is the impurity of the set ${\cal X}$. The impurity is calculated using the following formula: $$ \text{Impurity}({\cal X}) = 1 - \sum_{c=1}^{M} \left( \frac{|{\cal X}_c|}{|{\cal X}|} \right)^2 $$ where $M$ is the number of classes and ${\cal X}_c$ is the set of data points that are in class $c$. In this paper, we investigate the quantum version of the algorithm for calculating the best split of the data based on Gini impurity. We use the quantum version of the algorithm to calculate the gain of a split and the impurity of a set. We also investigate the complexity of the quantum version of the algorithm, using Circuit Depth as a complexity measure. We consider different quantum approaches for calculating the best split of the data based on Gini impurity: \begin{enumerate} \item QFT approach [\citep{Ramezani_2023}]. \item Approximated QFT approach [\citep{Nam_2020}]. \item VBERippleCarryAdder approach [\citep{Vedral_1996}]. \item VBERippleCarryAdder + RCCX approach. \end{enumerate} All of these approaches are implemented in Qiskit and tested on a quantum computer simulator. All of them compute the correct answer. The VBERippleCarryAdder approach is more effective in terms of circuit depth, but VBERippleCarryAdder–based approaches use more ancilla qubits than QFT-based approaches. Moreover, two different variants of the algorithm were implemented: a quantum subroutine that works with precalculated values and a fully quantum subroutine. At the end of the paper, we compare the complexity of the quantum version of the algorithm with that of the classical version for calculating the best split of the data based on Gini impurity. We also compare the circuit complexity of each developed algorithm. We obtained the following complexity (Circuit Depth) for our investigated methods. Let $n = |S|$, $m$ is total qubits size for oracle. \begin{tabular}{|c|c|c|c|} \hline QFT & A-QFT & VBERippleCarryAdder & VBERippleCarryAdder + RCCX \\ \hline $O(n^{1.5} \cdot m^2)$ & $O(n^{1.5} \cdot m \cdot \log(m))$ & $O(n^{1.5} \cdot m)$ & $O(n^{1.5} \cdot m)$ \
        \hline
        \end{tabular}

        \end{abstract}

        \noindent

        \bibliographystyle{johd}
        \bibliography{bib}

        \end{document}

        Speaker: Ilnaz Mannapov (Kazan Federal University)
      • 14:30
        Возможности и перспективы применения квантовых и квантово-инспирированных отжигателей для решения задач управления сложными технологическими системами на примере задачи управления энерго-электро-сетями 15m

        Многие задачи управления сложными технологическими системами математически сводятся к задачам решения систем нелинейных уравнений и неравенств, которые, в свою очередь, могут быть сведены к решению задач оптимизации нелинейных полиномиальных функций с ограничениями на диапазоны возможных значений переменных. При большом количестве переменных такие функции могут иметь очень сложный ландшафт, и поиск их глубоких минимумов часто оказывается весьма ресурсоемкой задачей, включающей в себя большую переборную составляющую. Поэтому перспективным представляется использовать для решение таких задач квантовых и квантово-инспирированных отжигателей – специализированных вычислителей, предназначенных для решения задач оптимизации.

        В первой части доклада будут даны: 1) краткие описания принципов работы отжигателей; 2)обзор современного состояния развития отжигателей и примеров их использования сегодня.
        Во второй части доклада будут представленыт некоторые результаты анализа возможности использования отжигателей для решения задач управления энерго-электро-сетями.

        Speaker: Николай Владимирович Малетин (Центр квантовых технологий МГУ, Центр инжиниринга перспективных технологий МГУ)
      • 14:45
        Neural network-based optimization of parameter search in quantum hashing 15m

        NEURAL NETWORK-BASED OPTIMIZATION OF PARAMETER SEARCH IN QUANTUM HASHING
        I. A. Shaimardanov
        Kazan (Volga region) Federal Uviversity, Kazan, Russia
        uvixmi@mail.ru
        Quantum hashing is considered one of the promising approaches in post-quantum cryptography [1]. However, the construction of optimal parameter sets for quantum hash functions remains a computationally difficult optimization problem due to the exponential complexity of the target function and the non-differentiable nature of the search space [2]. Classical optimization algorithms, such as brute-force search, random search, simulated annealing, and particle swarm optimization, are capable of finding high-quality solutions, but they require substantial computational resources and repeated evaluations of the objective function.
        The purpose of this work is to develop a neural network model capable of approximating the mapping between the input parameters of a quantum hashing problem and the optimal parameters obtained by classical optimization algorithms. The proposed approach allows replacing iterative optimization procedures with direct prediction of near-optimal solutions.
        (bits,"  " qubits)→parameters[], (1)

        where parameters represent the optimal parameter values used in the computational function defined by the model in the C# programming language.
        The developed model is based on a multilayer perceptron architecture [3]. The network contains an input layer of two neurons representing the normalized values of the parameters bits and qubits, three hidden layers with ReLU activation functions, and an output layer whose size dynamically depends on the number of qubits. The model was trained using datasets generated by several optimization techniques, including brute-force search, adaptive random search, repeated random search, greedy adaptive search, simulated annealing, and particle swarm optimization.
        Approximately 2000 samples were generated for combinations of input parameters with values bits=3…6and qubits=1…2. The dataset was split into training, validation, and testing subsets with proportions of 70%, 15%, and 15%, respectively. Training was performed using the Adam optimization algorithm [4] with a learning rate of 0.001, batch size of 32, and 50 training epochs. The loss function used during training was the mean squared error (MSE), defined as the average squared difference between predicted and target parameter values. Adam was selected because it combines momentum and adaptive learning rate techniques, providing stable and efficient convergence during training. A batch size of 32 was chosen to balance computational efficiency and generalization capability, while 50 training epochs allowed the network to iteratively refine its weights over multiple passes through the dataset.
        Experimental results demonstrate that the neural network achieves prediction accuracy comparable to fast adaptive optimization algorithms while significantly reducing computation time. The average prediction error on the test dataset corresponds to an MSE value of approximately 0.5–2.0, while the average deviation of predicted parameters remains within ±1–3 integer values. At the same time, the inference time of the neural network is approximately 0.001 ms, which is substantially faster than classical optimization methods.
        An important property of the proposed approach is its ability to generalize optimization patterns learned from multiple algorithms. Unlike stochastic optimization methods, which may return different solutions for repeated runs, the neural network produces stable predictions for known data and provides high-quality approximate solutions even for previously unseen parameter combinations.
        Thus, the proposed neural network model can be considered an efficient alternative to classical optimization methods in problems related to quantum hashing parameter selection. The combination of high computational speed, acceptable accuracy, and the ability to generalize makes the approach promising for intelligent optimization systems in quantum cryptography.
        References
        Ablayev F. M., Ablayev M. F., Vasiliev A. V. / Universal Quantum Hashing // Ученые записки Казанского университета. Серия Физико-математические науки. – 2014. – Vol. 156, No. 3. P. 7–18.
        Vasiliev A. V., Vasilov A. R., Latypov M. A. / Analysis of the Properties of Quantum Hashing // Results in Science and Technology. Series “Modern Mathematics and Its Applications”. 2017. Vol. 138. No. 0. P. 11–18.
        Goodfellow I., Bengio Y., Courville A. Deep Learning. MIT Press. 2016.
        Kingma D. P., Ba J. / Adam: A Method for Stochastic Optimization// International Conference on Learning Representations (ICLR). – 2015.
        Vasiliev A. V / Dense Quantum Hashing// Phys. Part. Nuclei. – 2025. – Vol. 56. – P. 974–977.

        Speaker: Mr Ivan Shaimardanov
      • 15:00
        Quantum Hashing with QKD States 15m

        We propose a new quantum hash function that encodes information into the sequence of quantum states that are usually employed in the quantum key distribution protocols. Here we also choose a basis and a state in that basis. However, the choise of the basis is not random, but is determined by the input being encoded. The cryptographic properties of the quantum hash function are guaranteed by both indistinguishability of the nonorthogonal states and the usage of the error-correcting codes to avoid quantum collisions.

        Speaker: Alexander Vasiliev (Kazan Federal University)
      • 15:15
        Implementation of the Quantum Hashing Algorithm for Real Devices 15m

        One of the techniques that allows us to develop space-efficient quantum algorithms is quantum fingerprinting or quantum hashing. This technique is well known, and it allows us to compute a short hash or fingerprint that identifies a (potentially long) string of data with high probability. In this paper, we are interested in minimizing the number of CNOT gates that we call the CNOT cost of the quantum circuit.

        In this paper we present an algorithm for optimizing a shallow circuit for the algorithm in amplitude form. The main idea of circuit's optimization is to move the target qubit by the path $P$ and application of the control rotation with all neighbor vertices. he method has polynomial time complexity that is an important adventure comparing to existing methods that has exponential time complexity $O((n+m)2^n)$. So, our new result allows us to use the method for practical important qubit connectivity graphs with $n\geq 40$.

        We applied the method for seven architectures of quantum devices that are IBMQ Fez, Rigetti Aspen, Rigetti Ankaa, Google, Rigetti Cepheus.

        Our algorithm generates a circuit that is better than all of presented strategies for all architectures. Note that we cannot use existing algorithm for this architectures because of two reasons: (i) it always uses all the vertices, but not only required $k$ (ii) it has exponential time complexity for exact algorithm and big polynomial time complexity for approximate algorithm that cannot be used for graphs with more than 100 vertices.

        Additionally, us discuss the phase form of the quantum hashing algorithm. Vasiliev in 2023 presented the shallow circuit for the algorithm with no two-qubit operations. Effectiveness of the method was shown in computational experiments by Khadiev et al. in 2025.

        In this paper, we implement the algorithm for checking equality of two binary strings for 4 and 8 qubit Bauman octillion quantum devices.

        Speaker: Kamil Khadiev (Kazan Federal University)
      • 15:30
        Probabilistic perfect transfer of arbitrary multiqubit pure state 15m

        Quantum information transfer is a widely acknowledged branch of quantum informatics.
        Although the basic information courier over long distance is a photon, the solid-state architecture
        of a transmission line is of interest in small-scale devises placed on a particular quantum
        platform. In this case, implementing optical environment is associated with additional technical
        complications. Such transmission line can be represented by a spin-chain. There are many
        algorithms of perfect state transfer based on graph theory. All of them require a highly engineered
        spin-chain and they are very sensitive to perturbations of coupling constants. Alternatively, our
        algorithm is applicable to any spin-chain governed by the Hamiltonian preserving the excitation
        number during evolution.
        We propose the probabilistic protocol for perfect state transfer of an arbitrary multiqubit
        pure quantum state along the spin-1/2 chain governed by the XX-Hamiltonian [1]. We show
        that the k-excitation pure sender's state can be restored at the receiver at some time instant
        t0 using only the local unitary transformation U over the qubits of the extended receiver. The
        restored state appears in the superposition with other states which form garbage. This garbage
        can be easily removed by including the 1-qubit ancilla, whose state labels the garbage, and then
        measuring the ancilla state with desired output. The resulting state of the receiver coincides
        with the initial sender's state. The probability of the desired output of ancilla measurement
        (success probability) can be maximized using parameters of the above unitary transformation
        U . It is important that this optimization can be done analytically by solving certain polynomial
        equation. An arbitrary pure state can be perfectly transferred by encoding it into the k-
        excitation state of the sender and then decoding the perfectly transferred k-excitation state
        at the receiver side at the time instant t0 selected for state registration.
        The work is funded as a state task of Russian Fundamental Investigations (State Registration No. 124013000760-0).

        [1] E. B. Fel'dman, J. Wu, and A. I. Zenchuk, Perfect pure quantum state transfer via state restoring and ancilla measurement, arXive:2509.10100[quant-ph]

        Speaker: Dr Alexander Zenchuk (Institute of Problems of Chemical Physics of RAS)
      • 15:45
        Towards description of the orbifold $\mathrm{SU(2)\times SU(2)}\backslash \mathrm{SU(4)}/\mathbb{T}^3$ as an entanglement space of two qubits 15m

        The $\mathrm{KAT}$ decomposition of the $\mathrm{SU(4)}$ group [1],
        \begin{equation}
        \nonumber
        g = \mathrm{K}\mathcal{A} \mathbb{T}^3\,, \qquad \ g\in \mathrm{SU(4)}\,,
        \end{equation}
        is adapted to describe the double coset $\mathbb{B} = \mathrm{K}\backslash \mathrm{SU(4)}/\mathbb{T}^3$ with the subgroup $\mathrm{K} = \mathrm{SU(2)\times SU(2)}$ and the maximal torus $\mathbb{T}^3$. The double coset $\mathbb{B}$ is shown to be a 6D orbifod which admits several differential-geometric structures. In the talk a few of them are discussed and the results of the computation of the following issues:

        -- The left-invariant forms and the bi-invariant metric on the SU(4) group,
        -- The induced metric on the principle orbit of the orbifold $\mathbb{B}$,
        -- The Haar measure on the SU(4) group and on the orbifold $\mathbb{B}$

        are presented.

        The impact of these differential-geometrical structures on the entanglement phenomena in the system of two qubits is discussed.

        Speaker: Dimitar Mladenov (Sofia University "St. Kliment Ohridski", Sofia, Bulgaria; Institute of Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, Sofia, Bulgaria)
    • 14:00 → 16:00
      Numerical and analytical calculations in modern mathematical physics Room 406 (MLIT)

      Room 406

      MLIT

      • 14:00
        Numerical method with error control for designing thin-film waveguide Luneburg lens 15m

        The work is dedicated to the design of a thin-film Luneburg waveguide lens in the zeroth approximation of the asymptotic model of electromagnetic radiation under approximate "horizontal" boundary conditions.

        The lens is designed based on the varying effective refractive index, which is a solution to the Morgan equation. An algorithm for numerically solving the Morgan equation with error control is proposed and implemented.

        The varying thickness of the waveguide lens is also calculated with error control, considering the error in the numerical solution of the Morgan equation.

        Speaker: Данила Стариков
      • 14:15
        Analytical solutions of the nonlinear fifth-order equation for waves in rotating fluid 15m

        This research was supported by the Ministry
        of Science and Higher Education of Russian Federation according to the state task project No. FSWU-2026-0006.

        \bigskip

        The nonlinear fifth-order equation for waves in rotating fluid
        \begin{equation}
        \label{oks}
        (u_t+\alpha (u^2)x+\chi (u^2){xx}+\mu u_{xx}+\sigma u_{xxx}+\beta u_{xxxx})x=\delta u+\epsilon u^2
        \end{equation}
        for the real-valued fucntion $u=u(x,t)$ is considered. In the case of $\delta=\epsilon=0$ eq. \eqref{oks} transforms to the Kuramoto-Sivashinsky equation with nonlinear convection \cite{kudr2026}, the form having the general solution in terms of traveling wave variables. When $\beta=\chi=0$, eq. \eqref{oks} represents the Burgers-Ostrovsky equation \cite{Zuev, MITRA2017639}. This equation describes, for example, the wave propagation in a viscous self-gravitating quantum fluid, modelling the star formation process in astrophysics. It does not pass the Painlevé test and has exact solutions in the kink form. For $\beta=\chi=\mu=0$ exact solutions to eq. \eqref{oks} in the form of solitary and periodic waves are shown in \cite{tian2008, Gandarias2009}. In this paper, we assume that $\beta=1$ and all other parameters in eq. \eqref{oks} are real nonzero numbers. Thus, eq. \eqref{oks} is reduced to the nonlinear ordinary differential equation of the fifth order
        \begin{equation}
        \label{oks_trw}
        y
        {zzzzz}+\sigma y_{zzzz}+2 \chi yy_{zzz}+\mu y_{zzz}+6\chi y_{z}y_{zz} +2 \alpha y y_{zz} -C_{0}y_{zz}+2 \alpha y_{z}^{2}-\delta y -\epsilon y^{2}
        = 0
        \end{equation}
        via traveling wave variables $u(x,t)=y(z),\ z=x-C_0t$, where $C_0$ is the wave velocity. An integrability of the obtained equation is studied using the Painlevé test in the form of the Kovalevskaya algorithm. It is shown that eq. \eqref{oks_trw} does not possess the Painlevé property, hence the initial value problem for eq. \eqref{oks} cannot be solved by the inverse scattring transform. However, a possibility of the exact solutions to eq. \eqref{oks_trw} existence is shown by obtaining Laurent series of its solution with the number of arbitrary constants less than its order. By means of the simplest equations method \cite{KUDRYASHOV20051217}, solitary and kink wave solutions to the eq. \eqref{oks} expressed via the solution of the Riccati equation with constant coefficients are found. Furthermore, using the equation for Jacobi elliptic sine as the simplest equation, exact solutions to the eq. \eqref{oks} in the form of periodic waves are obtained.

        \begin{thebibliography}{9}

        \bibitem{kudr2026} Kudryashov N.A. \textit{Traveling wave solutions of the Kuramoto-Sivashinsky equation with nonlinear convection}. Physics Letters A, V. 585, 2026.

        \bibitem{Zuev} Zuev K.M., Kudryashov N.A. \textit{Painlevé analysis and exact solutions of generalized Ostrovsky – Burgers equations}. Vestnik natsional'nogo issledovatel'skogo yadernogo universiteta "MIFI". 2026; 15(1): 37-50. (In Russ.)

        \bibitem{MITRA2017639} Aniruddha Mitra, Rajkumar Roychoudhury, Radhaballav Bhar, Manoranjan Khan \textit{Nonlinear wave breaking in self-gravitating viscoelastic quantum fluid}. Physics Letters A; 2017; V. 381 No. 6: P. 639-645.

        \bibitem{tian2008} Lixin Tian, Jiuli Yin \textit{Multi-compacton and double symmetric peakon for generalized Ostrovsky equation}. Chaos, Solitons \& Fractals; 2008; V. 35 No. 5: P. 991-995.

        \bibitem{Gandarias2009} Gandarias M.L., Bruzon M.S. \textit{Exact Solutions for a Generalized Ostrovsky Equation}. AIP Conf. Proc.; 2009; 1168: 406–409.

        \bibitem{KUDRYASHOV20051217} Kudryashov N.A. \textit{Simplest equation method to look for exact solutions of nonlinear differential equations}. Chaos, Solitons \& Fractals; 2005; V. 24: P. 1217-1231.

        \end{thebibliography}

        Speaker: Kirill Zuev (National Research Nuclear University MEPhI)
      • 14:30
        Trajectory analysis for the Moyal equation 15m

        The research investigates whether the Moyal equation, governing the evolution of the Wigner function in phase space, serves as the quantum counterpart to the classical Liouville equation for statistical distributions. It is frequently noted that the Moyal equation contains an infinite series with coefficients proportional to powers of the Planck constant on the right-hand side, and therefore it could easily be reduced to the well-known Liouville equation in the limit as the Planck constant approaches zero. In this limit, the right-hand side of the equation vanishes, and the Moyal equation becomes the classical Liouville equation for the distribution function. However, this process is seldom examined critically, and it remains unclear whether any significant terms are being excluded.
        Using the Helmholtz theorem, as well as the Vlasov-Moyal approximation for the mean acceleration, this study shows that the right-hand side of the Moyal equation does not explicitly depend on the Planck constant, and all terms in the series can make significant contributions. In the classical limit (ℏ → 0), the Moyal equation reduces to the Liouville equation, demonstrating that classical mechanics emerges naturally from the quantum description. However, the transition between classical and quantum descriptions is related not to the Planck constant, but to the spatial scale.
        For a model quantum system with a potential in the form of a “quadratic funnel,” an exact 3D solution of the Schrödinger equation is obtained, and the corresponding Wigner function is constructed. Using trajectory analysis in phase space, based on the representation of the right-hand side of the Moyal equation, it is shown that at the spatial microscale there exists an infinite number of “trajectories” of particle motion (thus the concept of a trajectory is indefinite). When transitioning to the macroscale, all “trajectories” converge around the classical trajectory. Properly understanding the analogy between the Moyal and Liouville equations facilitates modeling quantum dynamics using phase-space methods, which can be useful in quantum optics, semiclassical approximations, and quantum statistical mechanics.

        Speaker: Pavel Afonin (Moscow State University)
      • 14:45
        On invariant manifolds of reversible difference schemes 15m

        This work investigates invariant manifolds onto which solution trajectories of ordinary differential equation systems are wound, as obtained via reversible difference schemes. Particular emphasis is placed on the dimensionality of these manifolds, and methods for the numerical computation of dimensionality characteristics are presented. The correlation dimension method is applied to solutions of the equations governing the motion of a spinning top in the Kovalevskaya case; the dimension of the system's manifold is computed for various time-step parameters. The stability of the computed dimension with respect to enlargement of the integration interval is examined. The method is further extended to cases involving small deviations from the Kovalevskaya case, revealing a network of "resonances" wherein a reduction in dimensionality is observed. Computational experiments conducted over a parameter grid have enabled the identification of a critical parameter boundary defining the transition between finite and unbounded dynamical regimes of the system.

        Speaker: Viktor Kadrov
      • 15:00
        Two-dimensional shallow water flows on adaptive moving triangular meshes 15m

        Numerical solution of the shallow water equations over long times is prone to numerical dissipation on fixed Eulerian meshes. Continuing earlier moving-mesh work of the authors on the same equations in one [1] and two [2] dimensions, we describe a two-dimensional finite-volume scheme on adaptive moving triangular meshes.

        The method is formulated in an arbitrary Lagrangian-Eulerian framework. Cell-centered states are advanced by a Godunov-type update with fluxes from an exact one-dimensional Riemann solver [3] evaluated along each edge normal. The tangential velocity is transported by the contact discontinuity. Node velocities are recovered from edge-normal velocities by a local constrained least-squares problem whose admissible motion basis distinguishes interior, sliding-on-boundary, and fixed-corner nodes. The physical domain boundary is described as a closed contour assembled from straight segments and circular arcs, so curved boundaries are reproduced exactly throughout the simulation. Initial discontinuity surfaces and other prescribed alignment lines enter the initial triangulation as constrained edges. After each time step, locally over-stretched edges are refined by longest-edge bisection, with the conservative state copied to the resulting child triangles without interpolation.

        The scheme is implemented as a stand-alone Python package, with ongoing testing on two-dimensional Riemann problems and dam-break configurations.

        This work was supported by the Russian Science Foundation, project No. 24-71-00113.

        References

        [1] K. E. Shilnikov. Numerical modelling of one dimensional problems for the shallow water equations on adaptive moving meshes. J. Comput. Appl. Math. 476 (2026) 117134.

        [2] S. A. Ladygin, K. E. Shilnikov. Shock-ALE-SW: shallow-water shock-wave simulations on moving triangular meshes. Software, 2026. doi:10.5281/zenodo.19570235.

        [3] E. F. Toro. Riemann Solvers and Numerical Methods for Fluid Dynamics. Springer, 3rd ed., 2009.

        Speaker: Stanislav Ladygin (National Research Nuclear University MEPhI, Moscow, Russia)
      • 15:15
        Numerical modeling of planetary wave propagation using overset and multiblock grids 15m

        The paper presents a numerical study of wave propagation in complex planetary-scale domains. A two-dimensional model of Mars with internal structure and surface topography based on the Duran22~\cite{Duran2022Mars} and MEGDR~\cite{Smith2001} datasets is implemented. Governing equations are solved for heterogeneous media with a Ricker wavelet source using the grid-characteristic method. A novel approach based on combination of overset and multi-block mesh structures for planetary domain modeling if performed. The comparison with the solution obtained by the spectral-element method using the SPECFEM2D~\cite{Komatitsch2005} software package is provided. The considered problem for numerical analysis is analogous to that described in~\cite{Qiu2024}. The study highlights the differences between multi-block and overset grid structures for planetary wave simulations. The study was funded by grant No. 26-11-00143 from the Russian Science Foundation, https://rscf.ru/project/26-11-00143/

        Speaker: Ivan Yakovlev (Moscow Institute of Physics and Technology)
      • 15:30
        Monte Carlo study of threshold effects on cross-talk in close-packed multi-detector scintillator arrays 15m

        Multi-detector arrays based on liquid organic scintillators are widely employed for prompt fission neutron detection. The close packing of detector modules, required to achieve high geometric efficiency, gives rise to cross-talk—the detection of a single gamma ray in multiple detectors. Such events cause spurious coincidences and distort measured correlations. Monte Carlo simulation enables systematic study of cross-talk over a continuous energy range, which is difficult to achieve experimentally.
        A Monte Carlo model of photon transport has been developed for a multi-detector system consisting of 32 cylindrical liquid-scintillator modules arranged on a spherical surface. The model includes Compton scattering and photoelectric absorption and tracks all generations of secondary photons. The model has been validated against GEANT4 simulations.
        This work investigates the influence of pulse-height discrimination thresholds on detection efficiency and cross-talk characteristics. Systematic simulations are carried out for photon energies in the range 100–1500 keV. The introduction of realistic thresholds enables separation of noise-equivalent events from genuine energy depositions and provides practical estimates of single-detector efficiency and cross-talk probabilities under experiment-like conditions.
        Special emphasis is placed on two-dimensional energy deposition maps (E₀ vs. E_t), where E₀ and E_t denote the energies deposited in the target and neighbouring detectors, respectively. These maps reveal characteristic structures—full-energy line and photoelectric depletion zone—whose visibility and extent depend on the applied threshold. The results serve as a reference database for cross-talk correction and can guide the design and optimization of close-packed scintillator arrays for neutron spectrometers.

        Speaker: Olga Sidorova (JINR)
      • 15:45
        The a priori $\tan\Theta$ theorem and other estimates in the subspace perturbation theory 15m

        We start with reviewing known bounds on variation of the spectral
        subspace of a selfadjoint operator (say, hermitian Hamiltonian)
        under an additive symmetric perturbation. To this end we first
        recall the concept of operator angle between two subspaces of a
        Hilbert space $\mathfrak{H}$ as well as the concept of symmetric
        norm on a two-sided ideal in the algebra of bounded linear operators
        on $\mathfrak{H}$. Then we recollect the spectral dispositions for
        which sharp norm bounds on the variation of the spectral subspace
        associated with an isolated spectral set have already been
        established. Starting from the celebrated Davis-Kahan trigonometric
        theorems in the subspace perturbation theory, all of these bounds
        have the form of an estimate of a norm of some trigonometric
        function of the operator angle $\Theta$ between the unperturbed and
        perturbed spectral subspaces through the same norm of the
        perturbation operator. The latest known sharp norm estimate called
        the a priori $\tan\Theta$ theorem serves for the case where the
        unperturbed spectral set lies in a finite gap of the remainder of
        the spectrum. Unlike the previous trigonometric estimates that were
        proven to hold for any symmetric norm, by now the $\tan\Theta$
        theorem was only established for the standard operator norm. In the
        present research, we prove estimates for the operators $\tan\Theta$
        and $(\tan\Theta)^2$ that already involve their symmetric norms. We
        conclude the presentation by explaining how the established bounds
        on variation of a spectral subspace help to control the numerical
        quality of linear-algebraic calculations and, in some cases, even
        provide sharp estimates for numerical errors.

        Speaker: Alexander K. Motovilov (Joint Institute for Nuclear Research)
    • 16:00 → 16:20
      Coffee break
    • 16:20 → 18:35
      Mathematical methods and tools for modeling complex physical systems Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 16:20
        Generalized ordinary differential equations of Chazy class VII: first integrals and general solutions 15m

        We consider a generalized nonlinear third‑order ordinary differential equation
        $$y_{zzz} - y\, y_{zz} - 2\,y_z^2 - 2\,y^2 y_z + \alpha y + \beta y^2 + \gamma y^3 + \delta y\, y_z + \sigma y_{zz} + \mu y_z = 0.$$
        The leading terms of this equation coincide with the equation of Chazy class VII.
        The Painlevé test is performed, yielding three sets of parameter restrictions for which the test is satisfied. For each of the three integrable cases, first integrals are constructed using the method of the integrating factor. By applying Lie symmetry analysis to the obtained first integrals, the order is reduced, and the problem is reduced to solving first‑order equations. The general solutions for all three cases are obtained in quadratures.

        Speaker: Aleksandr Kutukov (National Research Nuclear University MEPhI)
      • 16:35
        Self-Consistent Modeling of Oxygen Atom Transport in a Remote Plasma Reactor for Photoresist Ashing 15m

        In this work, a two-dimensional self-consistent model of oxygen-containing inductively coupled plasma in a remote-type photoresist ashing reactor for 300 mm wafers was developed and verified. The model combines the description of plasma generation, gas heating, and transport of chemically active oxygen atoms responsible for photoresist removal. Numerical simulations were carried out in COMSOL Multiphysics 6.2 using a kinetic scheme including 41 plasma-chemical reactions. The verification was performed by comparison with Langmuir probe diagnostics and ellipsometry measurements of photoresist etching rate. The obtained results demonstrate that the radial non-uniformity of etching is strongly affected by gas pressure, RF input power, and separation grid geometry. The simulations reveal the formation of enhanced oxygen atom fluxes near the wafer edge caused by intensified peripheral gas flow and redistribution of plasma density. It was also shown that the separation grid configuration significantly influences both ultraviolet shielding efficiency and oxygen atom transport. Among the investigated configurations, the single grid with angled openings provided the best compromise between etching rate, process uniformity, and UV protection.

        Speaker: Yury Zakharov (JSC «Molecular Electronics Research Institute» (MERI))
      • 17:05
        Reaction-diffusion processes: Non-perturbative renormalization group treatment of advection effects 15m

        In this work we study the influence of turbulent mixing on
        reaction-diffusion systems undergoing active-to-absorbing phase transitions by means of non-perturbative renormalization group formalism (NPRG).
        We mainly focus on physics of branching and annihilating random walks advected by incompressible velocity field. Within the used NPRG framework
        we obtain a phase diagram for the model and proceed to study both its universal and non-universal properties in non-trivial regimes.

        Speaker: M. Kecer (BLTP JINR)
      • 17:20
        Real-time identification and classification of defects in radiation damage simulations 15m

        Large-scale molecular dynamics (MD) simulations are standard for modeling primary radiation damage, often requiring millions to billions of atoms to mitigate periodic boundary artifacts and prevent the self-interaction of supersonic shockwaves and channeled atoms, which may lead to the creation of artificial defects [1]. However, extracting structural data from such massive systems presents a severe I/O bottleneck. Dumping full atomic trajectories for on-the-fly defect tracking demands massive data storage and drastically degrades the integration performance. Consequently, the common practice in cascade simulations is post-processing discrete, widely spaced trajectory frames, thereby losing critical continuous defect dynamic resolution. While existing on-the-fly topological tools, such as the native Voronoi package in LAMMPS, offer geometric cell evaluation, they demand significant memory overhead for global atom mapping and lack the architectural capability to natively output coordinates for empty lattice sites (vacancies), antisites, and replacement atoms [2].

        To bypass these limitations, we developed a custom Wigner-Seitz compute module for LAMMPS, powered by the Voro++ library [3], to track defect dynamics in real time. Instead of relying on global ID lookups, the algorithm maps current atomic coordinates against a static reference lattice using pure spatial intersections. We implemented dynamic grid scaling to maintain constant O(1) lookup times even under extreme local density fluctuations, while incorporating ghost-atom halos to preserve domain boundary integrity across MPI partitions. The module outputs three concurrent data streams: displaced atom coordinates for interstitials and mixing, exact spatial locations of vacancies in the reference lattice, and macroscopic Frenkel pair tallies for thermodynamic logging. This method effectively eliminates the need for massive trajectory dumps, allowing for continuous, high-fidelity tracking of defect clustering, replacement, and recombination kinetics without bottlenecking computational performance.

        Acknowledgments: We thank the HybriLIT heterogeneous computing platform at the Meshcheryakov Laboratory of Information Technologies, Joint Institute for Nuclear Research for providing computational resources.

        References
        [1] K. Nordlund, S. J. Zinkle, A. E. Sand, F. Granberg, R. S. Averback, R. E. Stoller, T. Suzudo, L. Malerba, F. Banhart, W. J. Weber, F. Willaime, S. L. Dudarev, and D. Simeone. "Primary radiation damage: A review of current understanding and models." Journal of Nuclear Materials, 512 (2018): 450-479.

        [2] A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in 't Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimpton. "LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales." Computer Physics Communications, 271 (2022): 108171.

        [3] J. Lu, E. A. Lazar, and C. H. Rycroft. "An extension to Voro++ for multithreaded computation of Voronoi cells." Computer Physics Communications, 291 (2023): 108832.

        Speaker: Dr Trong Hoang Trung Nguyen (1. DNRI, VINATOM 2. FLNP, JINR)
      • 17:35
        Stability Analysis of the Steady Solution for Incompressible Lid-Driven Cavity Flow 15m

        The linear stability of steady incompressible flows is investigated through the linearization of the Navier-Stokes equations. In the case of a semi-elliptical cavity, the non-linear convective terms lead to the existence of multiple steady-state solution branches for the same Reynolds number [1,2]. These flow solutions are obtained using an iterative linear method, to ensure convergence at high Reynolds numbers.

        The stability of the obtained solutions is verified by formulating a generalized algebraic eigenvalue problem over the computational domain. The spatial discretization is performed using the finite element method on a sequence of triangular grids. For this purpose, the Taylor-Hood finite element pair is utilized, employing continuous quadratic polynomials for the velocity components and linear polynomials for the pressure. To solve the resulting spectral problem, the shift-and-invert transformation is applied [3], the leading eigenvalues are computed using the Krylov-Schur algorithm.

        Numerical results are obtained for the square lid-driven cavity to validate the computational framework against established benchmarks [4,5]. Also, the stability is investigated for multiple steady-state solutions in the semi-elliptical cavity, with various computational grids and a Reynolds numbers up to 10\,000.

        1.E. Erturk and T. Allahviranloo, Bifurcation and Multiplicity of Solutions of the Navier--Stokes Equations in Driven Semi-Elliptical Cavity Flow, \emph{Mathematics}, 10 (2022) 4242.
        2.Lomasov D. V., Vabishchevich P. N. Numerical solution of BVP for the incompressible Navier–Stokes equations at large Reynolds numbers //Russian Journal of Numerical Analysis and Mathematical Modelling. – 2025. – Т. 40. – №. 4. – С. 281-295.
        3.C. Campos, J. E.Roman, E. Romero, and A. Tomas, SLEPc Users Manual: Scalable Library for Eigenvalue Problem Computations, \emph{Technical Report DSIC-II/24/02}, Universitat Polit`ecnica de Val`encia, (2013).
        4.J. J. Gervais, D. Lemelin, and R. Pierre, Some experiments with stability analysis of discrete incompressible flows in the lid-driven cavity, \emph{International Journal for Numerical Methods in Fluids}, 24 (1997) 477--492.
        5.H. C. Elman, K. Meerbergen, A. Spence, and M. Wu, Lyapunov inverse iteration for identifying Hopf bifurcations in models of incompressible flow, \emph{SIAM Journal on Scientific Computing}, 34 (2012) A1584--A1606.

        Speaker: Dmitrii Lomasov (Lomonosov Moscow State University)
      • 17:50
        Multiscale mathematical modeling the cold spray of metallic nanoparticles on surfaces 15m

        The work presents a multiscale approach for studying the processes of creating the required ultra-thin coating on a substrate with specified geometric characteristics, used in micro- and nanoelectronics. Micro-cold spray (MCS) is considered as a technology for its creation. MCS is a modification of supersonic cold gas-dynamic spraying (SCGS) and can be used on temperature-sensitive materials and parts where the use of high-temperature processes is impossible. Experimental research of high-resolution MCS/SCGS technologies is a complex engineering task and requires significant time and financial expenditures, which leads to the need for preliminary mathematical calculations and the development of appropriate mathematical apparatus and program tools. This work proposes a multiscale approach that combines three scale levels of modeling (macro-, meso-, micro-), conforming the concepts of continuum mechanics accounting the nanopowder concentrations, particle and molecular dynamics methods, and also taking into account the possible multicomponent nature of the powder for the creation of composite materials. The numerical algorithm is based on splitting by physical processes and combining the finite-volume grid method, the symmetric Adams scheme, and the velocity Verlet integration. The general computational approach assumes that a three-stage algorithm is implemented in a time-dependent cycle, corresponding to the scale levels. The program implementation of the developed multiscale approach is based on MPI and OpenMP technologies and is designed for supercomputer use. A number of important numerical experiments are conducted, confirming the correctness of the approach and enabling the selection of modeling parameters for nanoparticle spraying on a surface.
        This work is supported by the Russian Science Foundation, project No. 25-71-20016.

        Speaker: Prof. Viktoriia Podryga (Keldysh Institute of Applied Mathematics of RAS)
      • 18:05
        Differentiable Fixed-Topology Mesh Adaptation 15m

        This work presents an optimization-based $r$-adaptation method for computational
        meshes with fixed topology. The cell connectivity is preserved, while the
        spatial resolution is redistributed by changing the nodal coordinates. The
        method is designed to move the existing mesh resolution toward regions
        identified by a monitor function, without increasing the number of cells and without remeshing.

        The adaptation is formulated as a minimization problem with respect to the mesh
        node coordinates:
        $$ L(x) = L_{\mathrm{mon}}(x) + L_{\mathrm{reg}}(x), $$ where $x$ denotes the nodal coordinates. The monitor term penalizes deviations from the equidistribution of the monitor-weighted cell measure. For a cell $K$, let $|K(x)|$ denote its area or volume, and let $M_K$ denote the cellwise monitor value used during a mesh-optimization step. We use the functional $$ L_{\mathrm{mon}}(x) = \frac{1}{N_c} \sum_K \left( \frac{M_K |K(x)|} {\frac{1}{N_c}\sum_J M_J |J(x)|} - 1 \right)^2 . $$ Minimization of this functional corresponds to the condition $$ M_K |K| \approx \mathrm{const}, $$ so that cells become smaller in regions where the monitor function is large. The regularization term $L_{\mathrm{reg}}$ penalizes excessive nodal
        displacement, non-smooth deformation, poor element shapes, and near-degenerate
        cells.

        The gradients of the functional with respect to the nodal coordinates are
        computed using automatic differentiation. The optimization is performed by an

        iterative gradient-based method.

        The method is tested on two-dimensional triangular and three-dimensional
        tetrahedral meshes. Numerical results demonstrate that the adaptation moves
        cells toward regions of large monitor values while preserving valid mesh
        elements. In unsteady test problems, this redistribution improves the resolution
        of localized solution features compared with a fixed mesh with the same number
        of cells. The results indicate that the proposed method can improve solution
        accuracy at a moderate additional optimization cost.

        Speaker: Sergei Voitov (Skoltech, AI Center)
    • 16:20 → 18:20
      Methods of quantum computing and quantum information processing Room 310 (MLIT)

      Room 310

      MLIT

      • 16:20
        On randomization in variational quantum algorithms 15m

        Currently, the only practically significant quantum computing algorithms with hardware implementation are variational quantum algorithms. At the same time, it is well known that these algorithms encounter difficulties when applied to high-dimensional data. We analyze the potential of Monte-Carlo methods to efficiently overcome the problems of barren plateau and multiple local minima that arise in variational quantum algorithms using gradient methods to select optimization parameters. In particular, we consider approaches to the randomization of unitary operators in ansatz layers, random initialization of optimization parameters in the framework of annealing method, and classical simulations of variational quantum algorithms. The results are illustrated by two model examples.

        Speaker: Alexander Tsirulev (Tver State University)
      • 16:35
        Connection of dipolar spin dynamics and multiple quantum dynamics of two-spin system with dephasing relaxation 15m

        We consider the spin dynamics of a two-spin system described by the secular part of the dipole-dipole interaction (due to being in a strong magnetic field) [1] and an analogous system governed by the multiple-quantum Hamiltonian [2], both systems being subject to dephasing relaxation [3]. These methods are widely used for solving problems of quantum correlations (entanglement and the quantum discord [4]). In both cases we obtained the exact solution of the Lindblad equation [5] with the initial state obtained by applying a π/2 resonance $y$-pulse to the thermodynamic equilibrium state. We showed that the obtained solutions are closely related: the system’s density matrix for the dipolar Hamiltonian at time $t$ coincides with the density matrix for the multiple quantum Hamiltonian at time $3t$ if the dephasing relaxation rate for the system with the multiple quantum Hamiltonian in three times less than for the system with the dipolar Hamiltonian.

        The experimental verification of obtained results in NMR were also performed. The main results are published in Journal of Magnetic Resonance [6].

        The work was performed as part of the state task, state registration number No. 124013000760-0.

        [1] A.Abragam. The Principles of Nuclear Magnetism. Oxford, Clarendon Press. 1961
        [2] J. Baum, M. Munowitz, A. N. Garroway, and A. Pines, J.Chem. Phys. 83 (1985) 2015
        [3] E.B.Fel’dman, E.I.Kuznetsova, E.I.Shipulya, S.G.Vasil’ev, A.I.Zenchuk, Appl. Magn. Res. 56 (2025), 1053
        [4] S.M.Aldoshin, E.B.Fel’dman, M.A.Yurischev. Low Temp. Phys. 40 (2014), 3
        [5] G.Lindblad, Commun. Math. Phys. 48 (1976) 119 – 130
        [6] G.A.Bochkin, E.B.Fel’dman, S.G.Vasil’ev. Journal of Magn. Res. 389 (2026), 108078

        Speaker: Dr Georgii Bochkin (Federal Research Center of Problems of Chemical Physics and Medicinal Chemistry RAS)
      • 16:50
        Spin Entanglement in the Deuteron within a Simple Algebraic Model 15m

        This paper further develops an algebraic model, previously proposed for studying entanglement in bipartite fermionic systems within the framework of Canonical Anticommutation Relations ($\mathcal{CAR}$). We explore the embedding of the parafermion algebra of order $p$ into the Cuntz algebra $\mathcal{O}_{2p}$ to analyze spin entanglement in the deuteron. For $p=2$, the Green components are defined as proton and neutron creation-annihilation operators; these operators anticommute within their respective components and commute between different ones. It is established that the deuteron’s total angular momentum is $J=1$, with possible projections $M=0, \pm1$. Due to parity conservation, the ground state is a superposition of the $\mathbf{S}$-state ($L=0$) and the $\mathbf{D}$-state $(L=2)$, with the $\mathbf{D}$-state admixture (approx. 4–5 percent) resulting from the deuteron's quadrupole moment. Our model indicates that maximum entanglement occurs in the $\mathbf{S}$-wave. While entanglement is absent in the $\mathbf{S}$-wave for $M=\pm 1$, a small degree of entanglement emerges from tensor forces that facilitate $\mathbf{S}-\mathbf{D}$ wave mixing.

        References

        [1] A.S. Sitdikov, A.S. Nikitin, D.V. Bushtets// Lobachevskii Journal of Mathematics. 2026. Vol. 47. No4. pp. 1831–1843.

        [2] M. Abe and K. Kawamura// Commun. Math. Phys. 2002. V.228. pp.85-101.

        Speaker: Airat Sitdikov (Kazan State Power Engineering University)
      • 17:05
        Two-parameter quantum hashing based on difference sets 15m

        Two-parameter quantum hashing based on difference sets

        Kirill Stepanenko

        Kazan (Volga Region) Federal University, Kazan, Russia

        Zavoisky Physical-Technical Institute, FRC Kazan Scientific Center of RAS

        In classical cryptography, hashing has a significant importance. For example, this primitive is used when constructing a digital signature or when calculating a certain checksum [1], which serves as a public verifier of the information source.

        In quantum cryptography, there is also a concept that provides the construction for quantum hash images. This concept uses information loss as a result of measuring quantum states to perform the necessary cryptographic properties. For example, in [2] a one-parameter construction of the quantum hash function for finite Abelian groups was proposed. A significant feature of this approach is the use of so-called $\varepsilon-$biased sets, which provide the required collision-resistance property for such a design of quantum hash function.

        Earlier [3], we've discussed that if we consider a difference set as an $\varepsilon-$biased set, this will give a slightly different quantum hash function, for which we can specify an explicit construction with deterministic collision-resistance and one-way properties. In this report, we develop the proposed concept of quantum hash functions based on difference sets and present a two-parameter quantum hash function of the form

        \begin{equation}
        \left|\psi(a,b)\right\rangle=\frac{1}{|\sqrt{D}|}\sum_{x \in {D+b}}e^{i\frac{2\pi}{q}x\cdot a}\left|x\right\rangle
        \end{equation}

        In this formula $D$ is a difference set for some finite Abelian group $\left\langle G,+\right\rangle$, $a\in G, b\in G, q=|G|$. The report will also consider the cryptographic properties of this function and the possibility of using such a construction as a quantum message authentication code (QMAC).

        Acknowledgement: The research was supported by the government assignment for FRC Kazan Scientific Center of the Russian Academy of Sciences.

        References
        [1] J. Talbot and D. Welsh, Complexity and cryptography: an introduction. (Cambridge University Press, 2006).
        [2] A. Vasiliev, Quantum hashing for finite abelian groups, Lobachevskii Journal of Mathematics 37 (6), 753-757 (2016).
        [3] K.A. Stepanenko, Quantum hashing based on difference sets in finite abelian groups, XII International Scientific Chetaev Conference "Analytical Mechanics, Stability and Control": Proceedings of the Conference. (2025). Vol. 2, pp. 328-332, Kazan: KNRTU-KAI pub. (in Russian)

        Speaker: Kirill Stepanenko (Kazan (Volga Region) Federal University; Zavoisky Physical-Technical Institute, FRC Kazan Scientific Center of RAS)
      • 17:20
        DNA molecule as a quantum channel for the effective transfer of a photoinduced charge carrier 15m

        A computer simulation of a photoinduced-charge propagation in a fragment of an artificial DNA molecule has been developed. The tensor network formalism is first used to represent a simplified tight-binding model for DNA and then to describe the quantum dynamics of charge propagation in this molecule by solving Lindblad's master equation. Calculations are performed using the Lindblad MPO Solver program, and the results are interpreted from two perspectives.
        The first concept suggests to consider the DNA as a one-dimensional quantum channel enabling to transfer with a high efficiency a quantum state (a single fermionic charge carrier) between two ends of this channel. The second concerns a prototypical effect known as an environment-assisted quantum transport (ENAQT) explaining the enhancement of the transfer efficiency by the interaction of the molecule with an environment.

        Speaker: Виктор Юшанхай (JINR, BLTP)
      • 17:35
        Transition to a many-body localization state in the disordered spin-1/2 Ising chain in the tensor network approach 15m

        A fundamental property known as ergodicity is violated if a quantum many-particle system, such as a chain of spins s=1/2 coupled by an exchange interaction, fails to thermalize. This phenomenon, called multiparticle localization (MPL), which occurs in the presence of some disorder in the system, is studied by us within the framework of the disordered chain model TFIM, that is, a one-dimensional Ising model with a long-range exchange interaction between spins in a disordered transverse magnetic field.
        Based on the disordered TFIM model, nonequilibrium quantum dynamics is simulated on a classical computer using the tensor network formalism and the open-source TenPy library. In this approach, we observe changes in quantum dynamic behavior, indicating a transition from a thermalized state to an MBL state, over a wide range of model parameters.
        In the more general context of quantum information processing, MBL systems retain memory of their initial conditions and are therefore good candidates for storing quantum information. Another possibility is to exploit the property of MBL states to maintain topological order, which is effectively protected from disorder-induced decoherence.

        Speaker: Людмила Сюракшина (MLIT JINR)
    • 16:20 → 18:35
      Numerical and analytical calculations in modern mathematical physics Room 406 (MLIT)

      Room 406

      MLIT

      • 16:20
        Analytic continuation of the T- and S-matrices to unphysical sheets in a multichannel extension of the Friedrichs-Faddeev model 15m

        In this work, we consider a matrix multichannel extension of the
        Friedrichs-Faddeev model. First, we formulate conditions on the
        kernels of the potential matrix that guarantee for the $T$-matrix
        (and then for $S$-matrix) the existence of analytical continuation
        to unphysical energy sheets. We also obtain explicit representations
        for the analytically continued $T$- and $S$-matrices on unphysical
        sheets in terms of these matrices themselves but only taken in the
        physical sheet. From the explicit representations obtained,
        it follows that the resonances on a certain unphysical sheet are nothing but
        the points at which the corresponding truncation of the scattering
        scattering matrix in the physical sheet has an eigenvalue of 0.

        This is a joint work with A.K.Motovilov.

        Speaker: Михаил Кременецкий (Dubna State University)
      • 16:35
        Exact solutions of reaction–diffusion systems with polynomial nonlinearities via the logistic function method 15m

        Reaction–diffusion systems model various spatiotemporal phenomena in chemistry, biology, and physics. Consider the system

        $$ u_t = D_1 u_{xx} + F(u,v), \qquad v_t = D_2 v_{xx} + G(u,v), $$ where $F$ and $G$ are third-degree polynomials. The aim is to determine which forms of $F$ and $G$ admit exact traveling wave solutions via the logistic function method. This method has been successfully applied to many individual equations [1] (Fisher, Kuramoto–Sivashinsky, KdV, etc.), but its potential for systems remains underexplored. The logistic function satisfies the Riccati equation and admits various generalizations [2]. Using $z = x - C_0 t$, we seek $$ u(z) = A_2 Q^2 + A_1 Q + A_0, \qquad v(z) = B_2 P^2 + B_1 P + B_0, $$ with $Q(z) = 1/(1+e^{kz})$, $P(z) = 1/(1+e^{mz})$. Substitution into the reduced ODE system yields algebraic conditions on the coefficients. Dependence on $k$ and $m$ is analyzed. Both generic ($k \neq m$) and degenerate ($k = m$) cases are considered.

        We provide examples of exact solutions for classical models such as the FitzHugh–Nagumo system, the Belousov–Zhabotinsky reaction, the Lotka–Volterra system with diffusion, and the Turing activator–inhibitor model. Explicit wave profiles (kinks, fronts, pulses) are constructed. The main value is extending the method to systems of ODEs and determining all ansatz coefficients. The results are directly relevant to biological pattern formation, where reaction–diffusion mechanisms are central [3].

        Acknowledgement: This work was supported by the Ministry of Science and Higher Education of the Russian Federation (state assignment project No. FSWU-2026-0006).

        References

        [1] N.A. Kudryashov, Analytical solutions of the modified Korteweg–de Vries–Burgers equation with a nonlinear source, Vestnik natsional'nogo issledovatel'skogo yadernogo universiteta "MIFI" 15 (2026) 21–36. (In Russ.)

        [2] G. Dattoli and R. Garra, A note on differential equations of logistic type, Reports on Mathematical Physics 93 (2024) 301–312.

        [3] M.C. Milinkovitch, E. Jahanbakhsh, and S. Zakany, The unreasonable effectiveness of reaction diffusion in vertebrate skin color patterning, Annual Review of Cell and Developmental Biology 39 (2023) 145–174.

        Speaker: Agniia Borodina (NRNU MEPhI)
      • 16:50
        Feshbach resonances of the second order in a Friedrichs model 15m

        We consider an extended version of the celebrated matrix Friedrichs model that may produce Feshbach resonances of the second order. In this model, under enlarging of the coupling constant, a pair of eigenvalues of the unperturbed Hamiltonian embedded into the continuous spectrum first transforms in resonances of the first order. At some moment later on, these two resonances merge into a single resonance but of the second order, and we fix conditions for this event. Being based on the operator interpretation of resonances [1], we found that, in the case under consideration, the operator roots of the transfer function have a very particular structure. Finally, we establish temporal behavioral consequences for the decay of the system.

        References
        [1] R. Mennicken and A. K. Motovilov, Operator interpretation of resonances arising in spectral problems for 2$\times$2 operator matrices, Math. Nachr. 201 (1999), 117–181.

        Speaker: Dmitry S. Grudko (BLTP, JINR)
      • 17:05
        Tensor Expansions in the Hermite Interpolation Polynomial Basis and Algorithmic Basis Reduction in Multidimensional Finite Element Method Schemes 15m

        The report investigates tensor expansions of the solution in the basis of multivariate Hermite interpolation polynomials, which form the foundation of high-order finite element method (FEM) schemes for boundary value problems involving systems of elliptic-type differential equations. Particular emphasis is placed on algorithmic basis reduction—a deterministic exclusion of FEM basis functions with high-order mixed partial derivatives—which effectively suppresses the exponential growth of problem dimensionality. Geometric domain decomposition, including smooth mapping of hyperellipsoids onto hyperparallelepipedal meshes, ensures the correct construction of discrete analogs of elliptic operators. Numerical results demonstrate a reduction in the dimension of the algebraic system by a factor of 1.5–3 while preserving the theoretical order of accuracy of the FEM scheme. The developed algorithms are verified on benchmark problems (Dirac oscillator, the Helmholtz equation) for dimensions $d=5$ and $d=6$, achieving an accuracy of 3 and 2 significant digits.

        Speaker: Oleg Kovalev (MLIT JINR)
      • 17:20
        Minimization based approach and finite element method for solving the two-center Dirac equation 15m

        The relativistic Dirac equation for a bound electron in the field of two fixed positive charges is revisited. In contrast to the one center case this three dimensional equation is separable only partially around the azimuthal angle $\varphi$, because of the commutation of the Dirac Hamiltonian only with the $z$ component of the total angular momentum $J_z$.

        In this report, we consider the two-center Dirac equation in the prolate spheroidal coordinates.
        After eliminating the azimuthal angle $\varphi$, we used a variational minimax approach, which involves directly expressing small components in terms of large components. As a result, we obtain a nonlinear generalized eigenvalue problem for a system of coupled two-dimensional two-variable elliptic equations.
        This problem is solved numerically using the high-order finite element and continuous analogue of Newton's methods.
        Very-high-precision energies of the relativistic $1s\sigma_g$ and $2p\sigma_u$ states for H$_2^{+}$ and Th$_2^{179+}$ are presented. A comparison with known results obtained by other authors is also

        Speaker: Ochbadrakh Chuluunbaatar (JINR)
      • 17:35
        Fourth-order schemes for solving the time-dependent Schrödinger equation based on decomposition of the evolution operator 15m

        Implicit operator-difference schemes based on decomposition of the evolution operator are constructed. The forth-order decomposition by time variable is obtained via the explicit truncated Magnus expansion and Padé approximations. The decomposition by spatial variable is performed by the finite-element method with interpolation Hermite polynomials. The convergence of the proposed schemes is demonstrated.

        Speaker: Диана Волобуева (Московский Государственный Университет им. Ломоносова, Объединенный Институт Ядерных Исследований)
      • 17:50
        Critical distances in some matrix robustness problems 15m

        Several matrix nearness problems are treated in the Frobenius normed space of real matrices. These are, distance computation from a given square matrix to the manifold containing either matrices with pure imaginary eigenvalues (stability radius problem), or matrices with eigenvalues in the unit circle in the complex plane (robust Schur stability problem), or matrices with multiple eigenvalues (Wilkinson's problem). The problem is crucial for parametric synthesis and numerical algorithms robustness. We develop a universal approach for its solution in the framework of symbolic computations ideology. It consists of elimination of variables procedure applied for the system of equations providing the stationary points of the distance function corresponding to the stated problem. The resulting univariate distance equation possesses the zero set containing the critical values of the distance functions. This allows one to solve the problem with absolute reliability of the obtained result. The matrix nearest to the treated one in any of the manifolds in discussion is also provided.

        Speaker: Alexei Uteshev (St.Petersburg State University)
      • 18:05
        The cubic spline method for solving two- and three-body Schrodinger Equation 15m

        The cubic spline interpolation (CSI) is applied to two- and three-body Schrodinger equation. The problem is reduced to eigenvalue and eigenvectors problem of matrix. The method usues picewise smooth solution of wave functions without using any additional methods, in particular, finite-difference methods. The method reduces the number of mesh points in exchange for more computational memory. This method primarily used to investigate spatial sctructures of alpha-cluster nuclei.
        CSI is applied to develop cross-platform (Windows, Linux, MacOS) program of two-body Schrodinger equation using modern libraries (Qt, Eigen, exprtk). The source code of this program is also available on GitHub.

        Speaker: Anton Bazhin (JINR BLTP)
    • 09:30 → 11:00
      Plenary Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 09:30
        Analytical and Numerical Treatment of Molecular Integrals – The Case of the Three-Center Nuclear Attraction Integral 30m

        It is well known that in any ab initio molecular orbital (MO) calculation, the major task involves the computation of molecular multi-center integrals, which are needed for the computation of different energy components, such as kinetic energy, nuclear attraction energy and electron-electron repulsion energy. A large number of these integrals is required for molecular calculations even for small molecules. As the molecular system gets larger, computation of these integrals becomes one of the most laborious and time consuming steps in molecular structure calculation.

        The most popular functions used in ab initio calculations are Gaussian type functions (GTFs). Unfortunately, these GTFs fail to satisfy the mathematical conditions for atomic electronic distributions. A large number of GTFs have to be used in order to achieve acceptable accuracy and this increases the computational cost. Exponential type functions~(ETFs) are more suitable than GTFs to represent electron wave functions near the nucleus and at long range; this implies that a smaller number of ETFs is needed for comparable accuracy. Unfortunately, molecular ETF-integrals are difficult to evaluate rapidly and reliably to a chemically significant accuracy. Despite the efforts of many scientists, no efficient algorithms were proposed for their numerical evaluation. However, with the recent advances in computer technology and the development of more sophisticated numerical methods, many researchers hope that the next generation of ab initio programs will be based on ETFs, such as ADF.

        Speaker: Hassan Safouhi (Campus Saint-Jean, University of Alberta)
      • 10:00
        An Overview of the Numerical Solution of Gauge Integrals 30m

        The pervasiveness of the integrals in the present day scientific research, high-tech engineering, economics, etc., is the result of a multidisciplinary process, started by Newton and Leibniz, three and a half centuries ago, with the foundation of the Calculus.
        Three types of developments deserve considerations:
        (i) The derivation of a rigorous logical framework aimed at clarifying the very concept of the integral. The today’s most comprehensive definition (proposed seven decades ago), called the Gauge integral, encompasses all the previous ones (Riemann, Lebesgue, Improper Riemann, etc.) as subsets.
        (ii) For the wide community of the physicists, engineers, applied scientists, the development of methods, algorithms, and software are of primary interest.
        The elementary algorithms for numerical quadrature may put emphasis either on simplicity (e.g., Newton-Cotes, equally spaced knots), or on efficiency and accuracy (e.g., Gauss, Gauss-Kronrod, via optimal knot selection; Clenshaw-Curtis: n-level knot interlacing, FFT use).
        The complex algorithms involve inner decision tools enabling flexible change of the resolving path: (a) The trapezoidal rule (limited to analytic functions only) may show exponential convergence. (b) The automatic adaptive quadrature (AAQ) shows the highest decision flexibility.
        (iii) Most of the existing quadrature algorithm implementations in computer codes software use floating point computations, adjusted to the available hardware resources.
        We have to emphasize the existence of both blessing and hardship, originating in the floating point computation of the integrals, due to the occurrence of round-off errors. Ways for their obviation or limitation are discussed.
        The most unexpected, pleasant and important result of the decision tree based AAQ solutions of Gauge integrals is the appropriateness of the δ-fine tagging for generating reliable decision paths to the numerical solution, a feature that was missing in the previous AAQ methods.

        Speaker: Gheorghe Adam (MLIT-JINR and IFIN-HH Bucharest)
      • 10:30
        Precision Physics with the Hydrogen Molecular Ion 30m

        Precision physics with hydrogen molecular ions, including H$_2^+$, HD$^+$, will be reviewed. We examine key experiments, the current state of the field, and future perspectives, alongside a theoretical framework being developed by our group for the high-precision study of molecular ions. Specific focus is given to methods of Quantum Logic Spectroscopy (QLS), experiments in ion traps, and tests of CPT invariance. Furthermore, we evaluate the potential of the H$_2^+$ ion as a candidate for ultra-precise molecular clocks, providing a comparison with strontium and ytterbium atomic standards. Finally, we present significant recent results [1,2], highlighting new achievements in quantum three-body calculations that further extend the horizons of precision.

        [1] S. Alighanbari, M.R. Schenkel, V.I. Korobov, S. Schiller, High-accuracy laser spectroscopy of H$_2^+$ and the proton–electron mass ratio. Nature 644, 69-75 (2025).

        [2] Vladimir I. Korobov, J. Buša, SWEXPHC: Variational bound state solution for the three-body nonrelativistic Schrödinger equation. Comp. Phys. Communications 319, 109920 (2026).

        Speaker: Dr Vladimir Korobov (BLTP, JINR, Dubna, Russia)
    • 11:00 → 11:20
      Coffee break
    • 11:20 → 13:20
      Plenary Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 11:20
        Discrete-variable representation method in quantum physics 30m

        The discrete-variable representation (DVR) [1-3] or the Lagrange-mesh method [4,5] is pawerful and flexible method for numerical integration of the few-dimensional Schrödinger equation in the discrete and continuum spectrums. It is widely used to solve various problems in quantum physics and quantum chemistry abroad, but in our country, DVR is less much popular and undeservedly deprived of attention. This report, to some extent, serves to fill this gap. We discuss the key aspects of the DVR, its features and its computational advantages in conjunction with advanced numerical methods used for integrating the few-dimensional Schrödinger equations: the finite difference method, the Kantorovich method, the trial function method in variational calculus etc.. We also discuss the author's recent results obtained with using a nondirect product 2D DVR for integration of 3D and 4D time-dependent Schrdöinger equations in certain problems of atomic interaction with strong laser radiation and ultracold collisions in confined geometry of atomic traps, which demonstrate the method advantages and perspectives.

        1. J.V. Lill, G.A. Parker, and J.C. Light, Chem. Phys. Lett. 89, 483-489 (1982).
        2. J.C. Light and T. Carrington Jr., Adv. Chem. Phys. 19, 263-310 (2000).
        3. V.S. Melezhik, J. Comp. Phys. 92, 67-81 (1991).
        4. D. Baye and P.-H. Heenen, J. Phys. A19, 2041-2059 (1986).
        5. D. Baye, Phys. Stat. Sol. (b) 243, 1095-1109 (2008).
        Speaker: Prof. Vladimir Melezhik (BLTP JINR Dubna)
      • 11:50
        Применение компьютерных вычислений для исследования универсальных свойств (супер)алгебр Ли 30m

        В конце 20 века французским математиком П. Вожелем было введено понятие универсальной алгебры Ли, являющейся, по предположению, моделью всех простых алгебр и некоторых супералгебр Ли. Универсальность заключается в том, что определенные характеристики представлений всех простых алгебр Ли, возникающих в тензорных степенях присоединенного представления, выражаются в виде однородных рациональных функций трех параметров Вожеля, принимающих фиксированные значения для каждой конкретной алгебры (и супералгебры) Ли.
        В оригинальной работе Вожеля понятие универсальной алгебры было получено с помощью рассуждений, берущих свое начало в теории узлов. Мы развиваем альтернативный подход к изучению универсальной алгебры, основанный на использовании т.н. расщепленного оператора Казимира и его реализации в терминах диаграммной алгебры Брауэра. Данный подход приводит к задачам высокой вычислительной сложности: построению характеристических тождеств, вычислению спектров операторов в больших тензорных степенях присоединенного представления и явному построению проекторов на инвариантные подпространства. Для решения этих задач разработана реализация соответствующих алгебраических конструкций в системе символьных вычислений Wolfram Mathematica. Это позволяет автоматизировать алгебраические преобразования и получать универсальные формулы для широкого набора алгебр Ли.
        В докладе будет представлен используемый вычислительный подход, обсуждены особенности реализации алгебры Брауэра и расщеплённого оператора Казимира, а также приведены результаты, полученные с помощью компьютерных вычислений.

        Speaker: Alexander Provorov
      • 12:20
        Neuroscientific principles in designing hybrid systems 30m

        Studying the neurobiological foundations of cognition not only allows for a deeper understanding of the nature of human intelligence but also the creation of more effective artificial intelligence systems. The human brain exhibits a unique ability for hybrid information processing, enabling it to solve complex problems by incorporating both abstract rules and logical structures, as well as concrete experience and sensory data. Understanding the mechanisms underlying this ability opens up fundamentally new possibilities for designing artificial systems capable of more flexible and efficient problem solving. The results of brain function studies in health and neuropsychiatric disorders (Shoshina et al., 2014-2025) indicate that the traditional division into purely logical (deductive) and purely empirical (inductive) methods of cognition is insufficient. Logical and empirical approaches to cognition represent complementary pathways of describing reality, the interaction of which ensures the integrity of perception and the brain's construction of an internal picture of the external world.
        Large-scale neural networks that support the empirical (heuristic) and logical approaches to cognition originate from retinal ganglion cells. At the subcortical level, all information converges on a multinuclear structure — the thalamus (LGN), which then distributes it to various regions of the neocortex. It is separates information into a specific pathway, preserving modality, and a nonspecific pathway, disregarding information about the modality of the object. The logical (rational) approach to analysis is implemented through a specific information transfer circuit, while the heuristic (empirical) approach is implemented through both specific and non-specific circuits. At the level of the posterior cerebral cortex, two information pathways — the temporal and parietal — are formed to the frontal regions of the brain, which support decision-making and action programming. Moreover, the large-scale neural networks of the parietal and temporal pathways are not isolated from each other but interact at all levels of information processing to ensure the integrity of perception. The parietal pathway analyzes the statistical characteristics of an image, perceives contours, movement, and direction of movement, transmits information quickly (up to 100 ms) to the frontal cortex, and constructs models of the visible world. Perception of large amounts of information is ensured by a mechanism for parallel processing of various characteristics of the visible world — a global mechanism for general-to-specific analysis, or an empirical cognitive approach. The mechanism for parallel information processing in the neural network of global heuristic analysis enables multiple calculations to be performed simultaneously, which is critical for achieving high processing speed. This organization of information perception and processing is one of the mechanisms for speed optimization, along with hierarchical processing and adaptive attentional resource management. Hierarchical processing ensures the compression and filtering of irrelevant information.
        The parietal pathway system allows the brain to make quick predictions without the need for a full analysis of the situation — a mechanism known as a "fast and economical" approach to decision making. This mechanism is most relevant under conditions of high uncertainty and time constraints for decision making. Under time pressure or high workload, the brain focuses on the most critical aspects of the problem, sacrificing a complete analysis for the sake of speed. Decisions made under such conditions are largely intuitive, not reflecting cause-and-effect relationships.
        The temporal pathway, responsible for detailed analysis, is characterized by a more time-consuming information processing time, since the processing of visual characteristics occurs sequentially rather than in parallel, as in the large-scale neural network of the parietal pathway. This strategy of perceiving and analyzing information is referred to as a local analysis mechanism, a deductive analysis of information from the specific to the general, or a logistic approach. This strategy of perceiving and analyzing information is a kind of quality optimization mechanism, ensuring the accuracy of decisions when establishing cause-and-effect relationships.
        Solving complex problems requires the integration of empirical (global analysis) and logical (local analysis) approaches, as logic provides structure and consistency in reasoning, while empiricism ensures connection with reality and adaptation to constantly changing environmental conditions. Depending on the context and current tasks, the brain switches between different strategies for perceiving and analyzing information, and, accordingly, between different information processing mechanisms. The prefrontal cortex plays a central role in this adaptive neurocognitive dynamic. Neurons in the prefrontal cortex maintain activity consistent with current goals and suppress activity associated with irrelevant stimuli. This "cognitive control" mechanism enables dynamic switching between processing modes: from fast, automatic, and heuristic-based to slow, analytical, detailed, and rule-based.
        The dynamic switching between empirical and logical processing modes is mediated by task complexity but also by the amount of time available to make a decision and relevance to experience. Depending on how closely the current situation matches previous experience, the brain chooses between rapid recognition (the empirical mode, given similarities) and analytical reasoning (the logical mode). Under conditions of high cognitive load on working memory or limited attentional resources, the brain tends to use simplified strategies, global information analysis, and the heuristic mode, dominated by the parietal attention system.
        A hybrid system must be able to automatically determine which processing mode is most appropriate for the current task and reallocate computational resources accordingly. This requires developing metrics that evaluate the characteristics of the input data (degree of structure, presence of precedents, level of uncertainty) and the characteristics of the task itself (accuracy requirements, time constraints, and the cost of various types of errors).
        The switching decision mechanism can be implemented through a meta-system that analyzes the task context and the current state of the main components. Such a meta-system can use rules, trained models, or a combination of both. It is important that the switching decision process itself be computationally efficient; otherwise, the overhead of determining the operating mode may negate the benefits of adaptive processing. When creating multi-level architectures, it is necessary to ensure interaction between logical and empirical processing components at different levels of the hierarchy.
        Conclusion. The most effective hybrid systems should result from the creation of an integrated architecture in which the empirical and logical components form a single whole capable of mutual enrichment. Practical implementation of these principles in artificial systems requires the creation of architectures with multiple routing, adaptive resource management, integrated feedback mechanisms, and the ability to symbiotically interact between the empirical and logical modes of information processing. This approach promises to significantly improve the efficiency of artificial intelligence systems and other hybrid systems for solving complex problems.

        Speaker: Irina Shoshina (National University of Science and Technology MISIS)
    • 15:00 → 17:30
      Boat trip
    • 17:30 → 19:30
      picnic in Ratmino
    • 09:30 → 11:00
      Plenary Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 09:30
        Quantum Informatics, Nanobioelectronics and Artificial Intellect 30m

        The accumulation of information generated by humanity is currently growing exponentially, leading to the problem of storing this information. Managing such volumes of information and the development of artificial intelligence, in turn, leads to an exponential increase in energy costs. Conventional silicon-based electronics are unable to solve these emerging problems. This report discusses new approaches based on the development of nanobioelectronics, which are orders of magnitude less energy-consuming than silicon-based technologies and orders of magnitude superior in their capacity to store and record information. This report examines the theoretical aspects of DNA-based nanobioelectronics.
        Such processes as DNA quantum cellular automata dynamics , DNA charge transport , dynamic regular structures, soliton evolution, polaron dynamics, breather creation and breather inspired charge transfer are discussed. The supercomputer simulation of charge dynamics at finite temperatures is presented. Different molecular devices based on DNA are considered. These make the solution of quantum bioinformatics problems based on DNA technologies.
        It has been hypothesized that the cognitive functioning of living organisms is determined by quadruplex motifs in their genome.

        Speaker: Prof. Виктор Лахно (ИМПБ РАН - филиал ИПМ им. М.В.Келдыша РАН)
      • 10:00
        Software Innovations in MPD: Advanced Modeling, Algorithmic Developments, and Ecosystem Modernization in MpdRoot 30m

        The Multi-Purpose Detector (MPD) experiment requires a robust, high-performance software framework for accurate data processing and analysis. This talk highlights recent, significant advancements within the MpdRoot software framework. A central focus is the MPD-specific realization of the A Common Tracking Software (ACTS) package, a modern and highly efficient track reconstruction toolkit planned for deployment in major future High Energy Physics (HEP) experiments worldwide. We detail novel algorithms developed for clusterization and hit extraction during both online and offline data processing. Furthermore, we introduce the machine learning models that have been successfully implemented for advanced particle identification. The presentation also showcases newly developed online visualization tools, discusses crucial tasks for detector calibration, and highlights the transition toward run-time Quality Assurance (QA) and online–offline coherence. Beyond algorithmic developments, the talk addresses the challenges associated with large-scale data handling, storage optimization, and the deployment of distributed computing solutions within the MPD offline ecosystem. Finally, we examine new methodologies for the structural refactoring and continuous deployment of the framework, outlining the ongoing developments and planned upgrades required to ensure full software readiness ahead of the first physics runs.

        Speaker: Dr Slavomir Hnatic (JINR)
      • 10:30
        Theory challenges for present and future colliders 30m

        Precision electroweak physics is entering a new era, where sub-percent measurements at the HL-LHC and future lepton colliders demand theoretical predictions incorporating higher-order QED and weak corrections. This talk focuses on the specific challenges associated with computing and implementing these electroweak corrections for present and future collider phenomenology.

        Speaker: Vitaly Yermolchyk (DLNP JINR)
    • 11:00 → 11:20
      Coffee break
    • 11:20 → 13:00
      Plenary Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 11:20
        Supercomputing without supercomputers 30m

        Any computational problem that can be solved by a classical computer can also be solved by a quantum computer. Conversely, any problem that can be solved by a quantum computer can also be solved by a classical computer, at least in principle, given enough time. But none of them can solve Grand Challenges yet. We argue that now only a combination of quantum algorithms and hybrid system emulators can provide a clue. We discuss the following items: 1. The problems of classical architectures; 2. Quantum systems hopes lost?!; 3. Why quantum computing anyway? 4. What about the underlying mathematics? 5. How to improve it? 6. Can we use quantum math in practice? 7. Parallel algorithms; 8. Qualitative Analysis of PDEs.

        Суперкомпьютерные вычисления без суперкомпьютеров

        Любая вычислительная задача, которую может решить классический компьютер, может быть решена и квантовым компьютером. И наоборот, любая задача, которую может решить квантовый компьютер, может быть решена и классическим компьютером, по крайней мере, в принципе, при наличии достаточного времени. Но ни один из них пока не может решить задачи Большого вызова. Мы утверждаем, что в настоящее время только комбинация квантовых алгоритмов и эмуляторов гибридных систем может дать ключ к разгадке. Мы обсуждаем следующие вопросы: 1. Проблемы классических архитектур; 2. Утрата надежды на квантовые системы или пока нет?!; 3. Зачем вообще нужны квантовые вычисления? 4. Как насчет лежащей в их основе математики? 5. Как это улучшить? 6. Можем ли мы использовать квантовую математику на практике? 7. Параллельные алгоритмы; 8. Можно ли построить качественную теорию уравнений в частных производных?

        Speaker: Alexander Bogdanov (St.Petersburg State University)
      • 11:50
        Mathematical Modelling in Radiation Biology 30m

        A review on current development of mathematical modelling in radiation biology is presented. Special focus is given on applications in the radiation medicine. The development of the hierarchy of models that would allow systematizing experimental data and studying the pathways of how radiation-induced pathologies develop at different organization levels (from a molecule to cell and tissue) and time scales (acute and long-term radiation effects) is discussed.

        Speaker: Prof. Aleksandr Bugay (Joint Institute for Nuclear Research)
      • 12:20
        Progress of the Geant4 toolkit for LHC and NICA experiments 30m

        Currently the Large Hadron Collider (LHC) collect data with Run3, which will be completed soon. The Monte Carlo simulation will be needed for analysis of the data for many years after Run3 end. The Geant4 toolkit is the main Monte Carlo simulation engine for the Large Hadron Collider (LHC) experiments, with a total number of events generated over 1010. New requirements for high luminosity Run4 are already established for detector simulation, be which will generate substantially more data, becomes the main task for all LHC experiments. Experiments at NICA require as high productivity and improved accuracy of Geant4 physics models. So, Geant4 simulations must be significantly updated in terms of physics model accuracy and CPU performance to meet the demands of the new experiment requirements.
        We present a review of the status of the Geant4 toolkit, which includes updates in geometry, physics models, and global code optimisation in recent versions, and the upcoming public release 11.5. Comparisons between Geant4 predictions with published data and the progress of the R&D projects toward LHC Run4 data will be discussed. Upgrades are expected both for CPU and GPU modes of Geant4 runs. We also review the broader application of Geant4 for other domains, including space science, medical applications, nuclear physics, and beam transport simulation.

        Speaker: Prof. Vladimir Ivanchenko (Tomsk State University)
    • 13:00 → 14:00
      Lunch
    • 14:00 → 16:00
      Machine learning and big data analytics Room 420 (MLIT)

      Room 420

      MLIT

      • 14:15
        Application of machine learning to the problem of toroidal plasma control 15m

        Tokamaks are considered a promising approach for future energy production based on controlled thermonuclear fusion. One of the key challenges in tokamak operation is the determination of control actions for magnetic coils given a target plasma current trajectory and a reference plasma boundary. This problem represents a nonlinear inverse control task with complicated coupling between plasma dynamics and actuator responses.

        In this work, a tokamak control simulation application is described. The application allows modeling of plasma current and boundary control in a tokamak device within a closed-loop setting. The framework provides a computational environment for training and evaluation of control algorithms. A classical control approach based on model-driven controllers is considered, along with machine learning-based methods including regression models and LSTM networks.

        Speaker: Mikhail Saakyan (Moscow State University)
      • 14:30
        Interpretability Study of Transformer-Based PID Models in the T2K Super Fine-Grained Detector 15m

        Transformer-based models are a promising approach to particle identification (PID) in the T2K Super Fine-Grained Detector (SFGD), but their decision-making remains difficult to interpret. This work presents an interpretability study of a Transformer-based PID model in a particle-gun setup, focusing on the roles of scalar and sequential features, class-dependent patterns in feature usage, and the robustness of the resulting explanations. The aim of this study is to establish a systematic interpretability framework and to formulate preliminary conclusions for further validation in more realistic detector conditions.

        Speaker: Виктория Кисеева
      • 14:45
        Machine Learning Approaches for Calibration of Low-Cost PM2.5 Sensors: From Linear Regression to Quantum Models 15m

        Accurate air quality monitoring remains a critical challenge for environmental protection and public health assessment. Low-cost particulate matter sensors provide an attractive solution for building dense monitoring networks due to their affordability and ease of deployment. However, their measurements are strongly affected by environmental conditions such as temperature and relative humidity, resulting in systematic bias and reduced reliability under real outdoor conditions.
        This study presents a comparative investigation of machine learning approaches for in-field calibration of low-cost PM2.5 sensors using reference measurements from the Serbian Environmental Protection Agency (SEPA) monitoring stations. Three low-cost sensors (M1, M3, and N48) were analyzed under real operating conditions using synchronized measurements of PM2.5, temperature, and relative humidity. Two stable observation periods were selected to ensure reliable comparison.
        Three groups of models were evaluated: classical regression models (Linear Regression, Random Forest, XGBoost), temporal deep learning models (GRU, LSTM, Transformer), and quantum machine learning approaches based on Quantum Linear Regression (QLR). For temporal models, sequential windows of 6 and 8 hours were used, while the quantum model employed a compact window-1 representation with three qubits to ensure stable optimization.
        Results demonstrate that temporal models outperform classical baselines, with LSTM providing the best overall performance for the M3 sensor (RMSE = 2.96 µg/m3, R2 = 0.62). Random Forest remained the strongest classical baseline, while QLR showed highly competitive performance, particularly for the M1 sensor (RMSE = 4.83 µg/m3), outperforming both XGBoost and Transformer models in this case. The M3 sensor exhibited the highest calibration stability across all model families, while M1 showed strong non-stationary behavior during the first observation period.
        The results indicate that the optimal calibration strategy strongly depends on sensor-specific behavior and the structure of measurement errors. Combining classical, temporal, and quantum approaches provides a robust methodology for improving the reliability of low-cost PM sensors and supports the development of scalable next-generation air quality monitoring systems.

        Speaker: Mariia Khvostova (MSU Dubna Branch)
      • 15:00
        Interpretable Kolmogorov-Arnold Network (KAN) Architecture for Real-Time DDoS Detection in IoT Environments 15m

        Due to resource limitations and device heterogeneity, Distributed Denial of Service (DDoS) attacks present serious security risks to Internet of Things (IoT) networks. This study presents a unique interpretable machine learning framework for real-time DDoS detection in Internet of Things scenarios utilizing Kolmogorov-Arnold Networks (KAN). Leveraging the IoT-23 dataset containing 1,191,264 network instances with 47 characteristics across seven attack categories, we create a KAN-based binary classifier identifying DDoS assaults from regular traffic. Learnable spline-activated layers are used in the suggested architecture to record intricate feature interactions while preserving interpretability through activation functions that may be seen. Stratified sampling, feature standardization, weighted loss functions to address class imbalance, and gradient clipping for training stability are all components of our approach. In terms of interpretability, experimental results show 83.7% accuracy with 0.206 validation loss, exceeding conventional multilayer perceptrons while retaining competitive detection performance. Comprehensive measures such as ROC-AUC (0.804) and F1-score (0.783) support the model's 80.65% recall for DDoS assaults with few false positives. With probability levels that may be changed to meet security policy requirements, the deployed system has real-time detection capabilities. By offering security analysts interpretable decision bounds and preserving operational efficiency for environments with limited resources, this work adds an explainable AI approach to IoT security. With model checkpointing, inference pipelines, and deployment options appropriate for IoT gateway integration, the entire implementation is packaged for reproducibility.

        Speaker: Hakizimana Evelyne (Eve@20155!)
      • 15:15
        Glia-to-neuron ratio estimation from histological images using machine learning 15m

        The Sector of Radiation Physiology of LRB JINR investigates radiobiological effects of ionizing radiation on the central nervous system (CNS) of small laboratory animals [1, 2]. One of the key methods for assessing radiation-induced CNS lesions is morphological analysis of brain tissue sections, which traditionally requires manual annotation and classification of all cellular objects by a trained expert — a process that is both time-consuming and subject to inter-observer variability. Machine learning approaches offer a promising avenue for automating this analysis.

        Building upon our previously developed machine learning framework for histological image analysis [3], the work presents an automated solution for neuroglia index computation on histological brain tissue images. The pipeline comprises sequential stages of instance segmentation and cell classification. Instance segmentation is performed by a random forest classifier trained on handcrafted multiscale morphological and textural features. Cell classification is carried out by a separate random forest model. The proportion of glial to neuronal cells — defined as the ratio of glia to neurons (GNR) — is computed automatically from the predicted cell classes, providing a reproducible quantitative measure of experimental results.

        The proposed approach substantially reduces the time required for morphological analysis and minimizes subjectivity inherent to manual cell enumeration, demonstrating clear prospects for integration in radiobiological research.

        Acknowledgement:
        Data annotation and algorithm development were carried out based on the ML/DL/HCP ecosystem of the HybriLIT Heterogeneous Platform (MLIT JINR).

        References:
        [1] A. Stadnik, O. Streltsova,D. Podgainy, I. Kolesnikva, Yu .Butenko, A. Nechaevskiy,A. Anikina, T. Gudiev, and A. Streltsov, 9th International Conference "Distributed Computing and Grid Tech-nologies in Science and Education", 2021, pp. 316-320.
        [2] I. A. Kolesnikova, N. N. Budennaya, Yu. S. Severyukhin and M. G. LalkovicovaProceedings of the Workshop on Information Systems for the Radiation Biology Tasks, Dubna, 2020 pp. 7-16.
        [3] E. Arrokho, D. Divakov, I. Kolesnikova and A. Khokhlov Cells Recognition of the Sensomotor Cortex on Histological Images, Physics of Particles and Nuclei, vol. 56, no. 6, pp. 1348–1352, 2025.

        Speakers: Dmitriy Divakov (Peoples' Friendship University of Russia), Enoel Arrokho
      • 15:30
        Machine Learning to detect collisions leading to the formation of a diatomic pair 15m

        In this abstract, we aim to provide a point-of-interests in the process of diatomic pair formation based on the spin-exchange interaction. The essence of this interaction is that certain atoms, such as noble gas atoms, are "able to transfer" their spin to a close-by atom. However, the spin transfer requires a long-term interaction between the atoms.

        This challenge can be considered in the context of classical mechanics, provided that the gas is hot enough to ignore the inaccuracies caused by quantum effects. Classical mechanics demonstrates that chaotic collisions occur over a short time interval insufficient for spin-exchange interactions to take place. Therefore, the common mechanism for the transfer of spin-exchange interaction is the atomic trapping effect, in which a multi-atom collision leads to the formation of a diatomic pair, which exists for a sufficiently long time for spin-exchange interaction to occur with high probability.

        The issue with this method is that there are so many collisions that it's hard to estimate how often diatomic pairs form. To determine this, we need to simulate every potential collision to check if a diatomic pair forms in each case.

        To address this problem, we propose using machine learning, in which a dataset of kinematic characteristics of atoms is fed as input, and the output provides a prediction regarding the formation of an atomic pair. Studies were also conducted to determine how versatile this approach is for systems with various temperatures or other types of atoms.

        As a result of the research, it was concluded that machine learning is effective at filtering out cases that do not lead to the formation of a diatomic pair. When the system temperature changes, the model demonstrates stable filtering, with only a small part of the cases in which a diatomic pair is formed being lost.

        Speaker: Stanislav Kim (ITMO University)
      • 15:45
        Improving an AI Assistant in a Web Application for Fitting Experimental Data Using Retrieval-Augmented Generation and Hyperparameter Optimization 15m

        This paper investigates the application of Retrieval-Augmented Generation (RAG) to improve the quality of C++ code generation by Large Language Models (LLMs) in the FITTER_WEB application. It was designed for fitting experimental physical data using the ROOT package. To achieve this goal, FITTER_WEB requires accurate and reliable C++ code, which the AI assistant is expected to provide. The main challenge lies in "hallucinations", which lead to uncompilability and functional incorrectness of the code when converting LaTeX formulas into software implementation. The use of a RAG collection containing "LaTeX formula — reference C++ code" pairs allowed a significant increase in the compilability level of the generated code. For example, for qwen3-coder:30b model it raised from 58% to 93%. The functional correctness of results was also improved.

        A comparison of the effectiveness of the proposed method across a range of other LLM models is provided. A significant improvement in the quality of the generated code has been demonstrated with an increase in the number of LLM model parameters. Furthermore, the correlation between the value of the language model temperature parameter and the levels of compilability and functional success was approved. Despite the complexity of the task due to the specifics of the ROOT library, the proposed approach demonstrates potential for scalability and can serve as a basis for expanding the AI assistant's functionality without the need for model fine-tuning.

        Speaker: Konstantin Lukyanov (MLIT, JINR)
    • 14:00 → 16:00
      Mathematical methods and tools for modeling complex physical systems Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 14:00
        Precision laser inclinometer. Method for determining vertical displacement of a surface from its tilt. 15m

        The paper proposes and investigates an algorithm for reconstructing the magnitude of vertical surface displacement, for example, of the Earth’s surface, as a function of the measured angle of its inclination. The role of the dominant oscillation, accompanying oscillations such as the microseismic peak and industrial background, as well as the role of an irreducible noise, is demonstrated. Variants of the algorithm operation are studied both under clean-signal conditions, and the permissible level of noise that still allows the main task of calculating vertical surface oscillations to be solved is determined. Based on the numerical analysis and modeling performed, proposals are made for further work in this direction and for possible areas of application this algorithm.

        Speaker: Иван Бедняков
      • 14:15
        Floquet multipliers of $\phi^4$ standing waves in a ball: comparison of JULIA and MATLAB implementations 15m

        Temporally periodic $\phi^4$ standing waves in a ball are considered as approximation of weakly radiating spherically symmetric oscillons in the $\phi^4$ theory. The standing waves are determined, by means of Newtonian iteration, as solutions of a nonlinear boundary-value problem on the two-dimensional domain $[0,T]\times[0,R]$ where $T$ is the period of oscillations and $R$ radius of the ball. The stability of standing waves is classified by evaluating the associated Floquet multipliers. This computational procedure takes sufficiently long computer time while the numerical investigation requires massive calculations in a wide range of $R$ and $T$. The recently developed JULIA implementation for the Floquet multipliers calculations is presented and compared with the earlier MATLAB version from the point of accuracy and the execution time in both serial and parallel mode.

        Speaker: Maxim Bashashin (MLIT, Joint Institute for Nuclear Research)
      • 14:30
        Derivation Weierstrass elliptic equation within Appell sequences of mass-shell polynomials and its application in relativistic kinematics 15m

        We suggest a new way of derivation of the Weierstrass elliptic equation within Appell sequences and exponential of endomorphism
        of polynomials. In this way the Weierstrass elliptic equation is treated as an principal equation in
        the extended relativistic kinematics .
        Two kinds of evolution parameters can be introduced , the first one has a nature of the hyperbolic
        angle and its translations strongly are correlated with translations of the energy. The second type
        is defined by the periodic angle, the translations of this parameter is correlated by translations
        of the momentum. The formulation of the relativistic motion within Weierstrass elliptic equation
        generalized two kinds of motion into unique scheme by taking advantage from the tool of the theory.
        of elliptic functions.

        Speaker: Dr Robert Yamaleev (JINR, LIT)
      • 15:00
        The application of the computational domain symmetry in the magnetostatics problem integral approach based on a linear approximation of magnetization. 15m

        3D volume integral equations method is used to solve the magnetostatics problem.
        The piecewise linear approximation of magnetization is used for description of magnetic
        field. There exists the significant complexity of usage of integral equations due
        to singularity of the kernel. The collocation method was previously applied
        to avoid such a problem [1]. Additional integration over the partition elements of
        the computational domain is used to eliminate problems with the kernel of integral
        equations singularity [2]. Taking in account of computational domain symmetry
        the modification of the 3D volume integral equations is proposed.
        The matrix elements calculation problems and the methods for solving nonlinear
        systems of discretized equations are discussed.
        References
        1.A. G. Armstrong, GFUN3D User Guide. RL-76-029/A, CERN (1976).
        2.The volume integral equation method in magnetostatic problem.
        Akishin P., Sapozhnikov A., Peoples’ Friendship University of Russia, Discrete
        and continuous models and applied computational science, 2019;27(1):60-69.

        Speaker: Павел Акишин (LIT, JINR)
      • 15:15
        Gauge Averaging and Quantum Indeterminism: A Finite‑Field Model 15m

        In standard quantum mechanics, pure states are identified with rays in a complex Hilbert space --- equivalence classes of vectors under multiplication by non-zero complex scalars. Projective invariance is built into the Born rule, ensuring that all physical predictions are independent of the chosen representative vector.
        This local gauge freedom is postulated, not derived.

        We propose a hypothesis that the probabilistic nature of quantum events originates from a breakdown of projective invariance at the microscopic level: individual vector-vector interactions depend explicitly on the chosen representatives, while observable probabilities emerge after averaging over all admissible gauge parameters.
        This transition from local non-invariance to global invariance mimics a path integral over the gauge group.

        To explore this idea in a mathematically tractable setting, we replace the complex field by a finite Galois field $\mathbb{F}_q$. The discreteness of $\mathbb{F}_q$ makes explicit averaging over the multiplicative group $\mathbb{F}_q^{\times}$ computationally feasible and eliminates analytical ambiguities. Using a non-degenerate quadratic form $Q$ on $\mathbb{F}_q^n$ and an additive character $\chi$, we define a candidate transition probability between projective points $[v],[w]$ as

        $$ P([v],[w]) = \frac{1}{q^2}\Bigl|\sum_{\lambda,\mu\in\mathbb{F}_q}\chi\bigl(Q(\lambda v+\mu w)\bigr)\Bigr|^2. $$ The double Gauss sum averages over all lifts of the two projective points to vectors. We compute these probabilities for all dimensions $n \in \{2,\dots,7\}$ with $q=5$, along with several other representative combinations of $n$ and $q = p^m$ that constitute moderately sized problems. The results show that $P([v],[w])$ takes only three distinct values: $1$, $1/q$, and $1/q^2$. The precise geometric type of the pair (whether the span is totally isotropic, degenerate, or non‑degenerate, and the square class of the discriminant) determines which value occurs. These probabilities are projectively invariant by construction and exhibit a discrete, rational spectrum that parallels the continuous Born rule in the limit $q\to\infty$.

        Our approach demonstrates that projective invariance can be an emergent property arising from gauge averaging, and that finite fields provide a rich laboratory for constructing discrete analogues of quantum probability without the analytic complexities of the continuum.

        Speaker: Владимир Корняк
      • 15:30
        On the Calculation of Feynman Integrals via Neural Network Approximation of Antiderivatives 15m

        High-dimensional numerical integration is a critical computational task in particle physics, particularly in modeling meson properties where complex interaction kernels require repeated evaluations over varying parameter spaces. This work presents an advanced neural network-based approach for the numerical integration of multi-dimensional functions, implemented within the updated Python library "Skuld".

        The proposed [1] methodology shifts from direct integrand approximation to learning the antiderivative of a given integrand function. By training models to represent the function’s antiderivative, the definite integral over a unit hypercube can be calculated effectively. A significant feature of this approach is the inclusion of the integrand’s parameters, which are not used for integration, as model inputs, enabling a single trained model to be reused across different sets of parameters without retraining.

        This work provides further development of the stated approach, where we imply the usage of other architectures aside Multi-Layer Perceptrons (MLP), and a comparative analysis of several neural network architectures for antiderivative approximation, including: Sinusoidal Representation Networks (SIREN), Wavelet Implicit Representation Networks (WIRE), and Kolmogorov-Arnold Networks (KAN). These architectures were validated using the Genz test function package and applied to a practical physics problem: modeling of meson properties within the QCD-motivated model with the separable interaction kernel for the NICA experiment. This modeling requires calculation of 2-d and 3-d Feynman integrals with 4 non-integratable parameters.

        Results indicate that the newer architectures — SIREN, WIRE, and KAN — outperform the previous integrand-based models in terms of precision and generalization.

        Keywords: Numerical Integration, Neural Networks, SIREN, WIRE, KAN, QCD, Mesons, Skuld Library.

        [1] Maître, Daniel, and Roi Santos-Mateos. "Multi-variable integration with a neural network." Journal of High Energy Physics 2023.3 (2023): 221.

        Speaker: Gregory Shipunov (Meshcheryakov Laboratory of Information Technologies, JINR)
      • 15:45
        Python driven Atomistic Simulation Environment modeling of adsorption based on interatomic potentials 15m

        We present atomistic modeling of adsorption using the open-source soft-
        ware Atomistic Simulation Environment, ASE [1]. ASE is a versatile Python library suitable for automating the construction and structural relaxation of adsorption systems. It has functions for creating surface slabs, placing adsorbates, applying constraints, selecting calculators, and so on. Various interatomic potentials (IPs), both simple and complex, as well as machine learned (MLIPs), are utilized for fast, computer little-demanding atomistic simulations of the adsorption.

        In the talk, a couple of atomic and molecular systems adsorbed on surfaces such as gold, quartz, and selenium are computationally studied with ASE and discussed. Selected cases are taken from published works; as examples, Nh and Tl atoms on quartz from Ref. [2].

        Overall, the ASE ecosystem serves as a fast and flexible tool that, in
        some cases — especially with employed MLIPs — yields results in reasonable agreement with available reference data. Furthermore, the ASE-optimized structures serve well as input for subsequent, more rigorous first-principles quantum-chemical calculations.

        Our theoretical approach is of interest for providing assistance in the
        synthesis of superheavy elements, where created species are adsorbing on the detector’s surface, and the adsorption enthalpy is estimated.

        REFERENCES

        1. Hjorth Larsen et al: The atomic simulation environment-a Python library for working with atoms. J Phys Condens Matter. 2017 Jul 12;29(27):273002. doi: 10.1088/1361-648X/aa680e.

        2. Valeria Pershina : Theoretical Study on the Adsorption Behavior of Element 113 and Its Homologue Tl on a Quartz Surface: Relativistic Periodic DFT Calculations ; J.Phys.Chem. C 2016 120 (36), 20232-20238
          DOI: 10.1021/acs.jpcc.6b07834

        Speaker: Dr Мирослав Ильяш (BLTP JINR Dubna, Russia)
    • 14:00 → 16:00
      Mathematical methods in life sciences Room 406 (MLIT)

      Room 406

      MLIT

      • 14:00
        Computation of Structural Chromosome Aberrations in Human Blood after Irradiation with Therapeutic Proton Beams 15m

        Chromosome aberrations (CAs) are produced in the genetic structure of human cells after exposure to ionizing radiation, such as photons and protons. Peripheral blood lymphocytes (PBLs) are among the most accessible human tissue samples and are widely used in biodosimetry and health risk assessment.
        The study of cytogenetic effects induced by accelerated proton beams is of particular interest in medicine because of the increasing use of proton therapy in cancer treatment. In this case, CAs in human PBLs are a useful biomarker for assessing carcinogenic risk and evaluating patient sensitivity to therapeutic proton beams. Unlike photons (X-rays and gamma rays), proton beams deliver dose along their path with a relatively low entrance dose to healthy tissue and a maximum at the end of their path, known as the Bragg peak, where the tumor is targeted. This depth-dose distribution results in higher biological effects compared with photons. However, accurate assessment of radiation-induced CAs remains challenging due to large uncertainties in data from differences in experimental conditions between laboratories.
        In this work, we developed a specialized radiobiological model to predict and explain experimental results obtained under the same conditions as in our laboratory. The depth-dose and linear energy transfer (LET) distributions of 150 MeV and 10 mm wide spread-out Bragg peak (SOBP) proton beams at the JINR Phasotron medical complex are primarily simulated using the Geant4 Monte Carlo radiation transport code. A detailed G0-phase PBL nucleus model, incorporating atomistic descriptions of chromosomes based on polymer physics, full track-structure physics based on the Geant4-DNA interaction cross-sections, and DNA sequence-based algorithms are integrated into a single modeling system.
        Different types of CAs at various depths along a SOBP dose profile of accelerated protons with LET of 0.5 - 20 keV/μm were calculated based on clustered DNA double-strand breaks. The results show that the CAs spectrum depends strongly on the position relative to the Bragg curve. The best agreement between calculations and experiments was observed at the entrance region of the proton beam. Furthermore, the modeling technique allows not only high-resolution studies of the radiobiological mechanisms of accelerated particles, but also helps overcome experimental limitations and increase efficiency of accelerator-based research related to the development of radiation medicine and space exploration.

        Speaker: Dr Batmunkh Munkhbaatar (LRB, JINR, Dubna)
      • 14:15
        Holstein model of charge transfer taking into account the transition between strands of DNA duplex 15m

        The interest to study of charge transfer in DNA is determined by the broad prospects for using DNA fragments of various lengths and compositions as blocks for electronic nanostructures [1]. Currently, many theoretical works have been published devoted to models of charge transfer in DNA. Most models assume that charge transfer occurs only along one DNA strand, without transfer to the complementary strand.
        Previously, a model of hole transfer in the DNA double helix, taking into account all nearest neighbors, was developed and theoretically investigated [2,3] for the case of a rigid lattice. We now combine this description with Holstein polaron model. Here, the lattice site is a complementary nucleotide pair (A/T) or (G/C), and there exist two charge traps per site, which correspond to the two nucleotides in the base pair. This combination allows for a more detailed consideration of the physicochemical properties of the nucleotide chain, which are essential for charge transfer processes.
        Numerical simulations demonstrate that under the influence of an external field with constant intensity, in a DNA duplex the charge exhibits oscillations with the Bloch frequency, as in the single-strand model [4,5].

        1. Lakhno V.D., Vinnikov A.V. Molecular devices based on DNA. Маthematical biology and bioinformatics 16 (1), 115-135 (2021). doi: 10.17537/2021.16.115
        2. V.D. Lakhno, V.B. Sultanov. On the Possibility of Electronic DNA Nanobiochips // J. Chem. Theory Comput. 3, 703-705 (2007). doi: 10.1021/ct6003438
        3. V. D. Lakhno, V. B. Sultanov. Band structure of the spectra of Hamiltonians of regular polynucleotide duplexes // Theoretical and Mathematical Physics 176 (3), 1194–1206 (2013). DOI: https://doi.org/10.1007/s11232-013-0100-7
        4. Lakhno, V.D., Fialko, N.S. Bloch oscillations in a homogeneous nucleotide chain // Jetp Lett. 79, 464–467 (2004). https://doi.org/10.1134/1.1780553
        5. Fialko N.S. Modeling Of Charge Dynamics in Synthetic DNA under the Influence of an External Electric Field // Маthematical biology and bioinformatics 19 (2), 439-452 (2024). doi: 10.17537/2024.19.439
        Speaker: Nadezhda Fialko (Institute of Mathematical Problems of Biology RAS - the Branch of Keldysh Institute of Applied Mathematics of Russian Academy of Sciences (IMPB RAS- Branch of KIAM RAS))
      • 14:30
        Quantum entanglement processes in the DNA - mRNA system 15m

        The work investigates the processes of quantum entanglement formation between DNA and mRNA chains during transcription. The entanglement processes were studied in the TP53 - mRNA system. As a result of the model calculation, we have shown that quantum entanglement in the system under consideration has a dual nature. On the one hand entanglement has dynamic and movable nature. On the other hand the entanglement binds the two DNA and mRNA subsystems together into a single entangled composition. It was shown that in the both cases entanglement forms a quantum communication channel between DNA and mRNA subsystems where information can be transmitted without any chemical processes. In the work we study the quantum entanglement using the von Neumann entropy as a quantitative characteristic of the entanglement degree.

        References
        1. S.Eh. Shirmovsky, A.V. Chizhov. // BioSystems. 2025. V. 256, P. 105566.
        2. S.Eh. Shirmovsky, A.V. Chizhov. // BioSystems. 2023. V. 231, P. 104967.

        Speaker: Sergey Shirmovsky (Far Eastern Federal University, Institute of High Technologies and Advanced Materials, Department of General and Experimental Physics)
      • 14:45
        Кинетическая модель механизмов НГСК и МГСК в клетках млекопитающих и человека 15m

        Представлены результаты кинетического моделирования механизмов негомологичного (НГСК) и микрогомологичного соединения концов (МГСК), ответственных за восстановление радиационно-индуцированных двунитевых разрывов ДНК (ДР ДНК) в клетках млекопитающих и человека, находящихся в начальных фазах клеточного цикла. Модель описывает взаимодействие ключевых репаративных белков НГСК и МГСК (Ku70/80, ДНК-ПКкс, Artemis, ATM, XLF/XRCC4/LigIV, MRN, PARP, Fen1, Polβ) с участками поврежденной ДНК на основе кинетики Михаэлиса-Ментен [1-3].

        Выбор определяющего пути репарации осуществляется в зависимости от сложности первичных ДР ДНК. В работе рассматриваются повреждения ДНК 5 типов: простые ДР и кластерные ДР (кДР) ДНК, содержащие поврежденное основание, однонитевой разрыв ДНК и 1-2 ДР ДНК, находящиеся в пределах 10-15 пар оснований от ДР ДНК. Их количество определяется зарядом и линейной передачей энергии частицы. Вычисление первичных повреждений ДНК произведено с использованием данных [4], полученных с помощью Монте-Карло расчетов. В работе произведен приближенный аналитический расчет выхода ДР ДНК для различных экспериментальных методов, таких как метод ДНК-комет, метод ДНК-фокусов и метод преждевременной конденсации хроматина (PCC-метод). Это позволяет определить вклад механизмов НГСК и МГСК в репарацию ДР ДНК, а также провести количественную оценку каждого пути репарации.

        Speaker: Эрмухаммад Душанов (laboratory of radiation biology, Кафедра биофизики, Государственный университет Дубна)
      • 15:00
        Comparative Analysis of Ionisation Cluster Size Distributions in Sub-cellular Structures using Geant4-DNA 15m

        Deep space exploration and long-term lunar missions pose significant challenges due to the complex field of cosmic radiation. Accurate estimation of biological risks requires sophisticated Monte Carlo simulations capable of tracking particle interactions at the nanometer scale. This study focuses on calculating the Ionisation Cluster Size Distribution (ICSD), a key nanodosimetric quantity that correlates with the induction of complex DNA damage.

        Using the Geant4-DNA toolkit, we simulated the passage of various charged particles, including protons, deuterons, alpha particles, lithium ions (Li-6 & Li-7), and carbon-12, through biological targets using G4EmDNAPhysics_option2/4/6/8. The analysis was extended beyond simple water volumes to realistic sub-cellular geometries: histones, ribosomes, the NMDA receptor, and elements of the cytoskeleton.

        The results demonstrate the sensitivity of ICSD to particle type and Linear Energy Transfer (LET) at an energy of 1 MeV/u. The shift in distribution peaks for heavier ions highlights the increasing probability of high-density ionization events within critical target volumes. These data provide a physical basis for predicting direct radiation damage to biomolecules. The obtained ICSD data serves as a reference point for validating future implementation of BEB-based ionization cross-sections for biological media within the Geant4-DNA framework. While focused on sub-cellular scales, this data contributes to the refinement of risk assessment models used in space radiobiology and hadron therapy research.

        Speaker: Matvey Vologzhin (Tomsk State University)
      • 15:15
        Mathematical Modeling of Albumin Filtration in the Glomerular Capillary of the Human Nephron 15m

        Numerical modeling of filtration in the glomerular capillary of the human nephron is a complex biophysical problem. Solving this problem allows us to understand the mechanisms that prevent large proteins (in particular, albumin) from passing from the blood into the primary urine. The human glomerular filter includes the glycocalyx, endothelium, glomerular basement membrane (GBM), and podocytes. These structures form a sequence of filtering layers of various thicknesses with different permeability coefficients. Filtration in the glomerulus is passive. Its driving force is the high hydrostatic pressure in the capillaries, created by the fact that the diameter of the afferent arteriole exceeds that of the efferent arteriole. This pressure forces the liquid part of the blood through the capillary walls. Blood cells (erythrocytes, leukocytes) and large plasma proteins (albumin) are retained by the multilayered filtration barrier. As a result, primary urine is formed, which enters the nephron capsule. This work presents a three-dimensional numerical simulation of several coupled processes. Inside the capillary, Stokes flow is described. Passive transport of albumin is modeled along the capillary. It is described by the advection–diffusion equation with an advection-dominated regime. The filtration velocity of plasma through the multilayered porous capillary wall is calculated according to Darcy's law. Diffusive transport of filtered albumin is taken into account in the porous medium. The computational domain includes a central subdomain with Stokes flow. It is surrounded by four porous subdomains of different permeabilities. At the capillary–glycocalyx interface, the Beavers–Joseph condition are imposed, along with the Starling and Kedem–Katchalsky conditions. At the interfaces between the porous layers, continuity of pressure and normal velocity components is imposed. A tetrahedral mesh is applied to the computational domain. It is conforming at the internal subdomain interfaces. The solution is obtained using the finite element method. To model the coupled processes in the capillary–porous medium system, a tailored selection of basis functions is applied. It is adapted to the features of each subdomain. The velocity space in the capillary lumen is approximated by quadratic basis functions P2, and the pressure by linear functions P1. This choice of basis guarantees satisfaction of the Ladyzhenskaya–Babuška–Brezzi (LBB) or “inf-sup” condition. For the Darcy flow subdomain, a continuous scalar basis of second order P2\ is used to compute the pressure field, while the vector velocity field in the porous medium is calculated during pos-tprocessing by taking the gradient of the obtained pressure and subsequently applying Darcy's law. To solve the albumin transport equation, a first-order Lagrange basis P1 is employed. In regions with advection dominance, this approximation tends to produce numerical oscillations. To suppress them, the scheme is supplemented with SUPG (Streamline Upwind/Petrov–Galerkin) stabilization.
        Acknowledgments The results were obtained with the financial support of the project within the framework of the state program of the federal territory «Sirius» «Scientific and technological development of the Federal Territory «Sirius» (Agreement No. 18-03 of 10.09.2024).

        Speaker: Ms Anastasiya Gutsu (Sirius University of Science and Technology)
      • 15:45
        Stratification of thyroid nodules by Eu-TIRADS categories using transfer learning of convolutional neural networks 15m

        We performed a pilot study to evaluate a method for assessing malignancy potential in thyroid nodules and their stratification according to the European Thyroid Imaging And Reporting Data System (Eu-TIRADS) scale based on ultrasound diagnostic images using an artificial intelligence system. The method is based on the use of transfer learning technology for multi-parameter models of convolutional neural networks and their subsequent fine tuning. It was shown that even on a small dataset consisting of 1129 thyroid ultrasound images classified by 5 Eu-TIRADS categories, the application of the method provides high training accuracy (Accuracy: 0.8, AUC: 0.92). We have demonstrated, that this technology can be potentially introduced in clinical practice as an additional tool (‘second opinion’) for an objective assessment of the risk of malignancy in thyroid nodules for the purpose of their further selection for fine needle biopsy.

        Speaker: Yulia P. Sych (I.M. Sechenov First Moscow State Medical University of the Ministry of Health of the Russian Federation (Sechenov University))
    • 14:00 → 16:00
      Modern methods for data processing and analysis in Mega-science projects in HEP Room 310 (MLIT)

      Room 310

      MLIT

      • 14:00
        Kubernetes-based online TPC data processing and monitoring for NICA/MPD 15m

        Modern HEP experiments produce high-rate data streams that require fast online event processing for quality control, and visualization during data taking. This paper presents an implementation and resource assessment of a distributed online processing and monitoring system for the MPD experiment at the NICA collider. The proposed system processes raw data from TPC detector through a distributed microservice pipeline based on clustering algorithm and Docker containers with Kubernetes orchestration. The system satisfies observability and automatic recovery requirements that provide a scalable basis for online detector data analysis.

        Speaker: Andrey Budnikov (Voronezh State University)
      • 14:15
        Track Reconstruction in SPD Timeslices Using Graph Neural Networks with Contrastive Learning 15m

        The problem of timeslice decomposition in the SPD experiment at the NICA collider consists of two stages: reconstruction of charged particle tracks with noise hit removal, and subsequent event-by-event separation of the reconstructed tracks. High efficiency of track reconstruction at the first stage is crucial, since an SPD timeslice is designed to contain from 10 to 40 overlapping events, making their separation challenging.
        We propose a track reconstruction method based on graph neural networks with contrastive learning. A k-nearest-neighbor graph is built in the hit coordinate space, after which a neural network based on EdgeConv layers trained with Contrastive Loss produces an embedding space where hits belonging to the same track are compactly clustered. Final clustering is performed using the DBSCAN algorithm. The data are obtained from the official SPD detector simulation using the SpdRoot software package. The proposed method demonstrates high track reconstruction efficiency and is promising as the primary tracking algorithm for the first stage of timeslice processing in the SPD experiment.

        Speaker: Aleksei Didorenko (Meshcheryakov Laboratory of Information Technologies)
      • 14:30
        Graph Attention Neural Network for Clustering Particle Tracks by Events in the SPD experiment 15m

        A Graph Attention Neural Network (GANN) for particle track clustering by events in time slices obtained at the Spin Physics Detector (SPD) of the Nuclotron-based Ion Collider fAcility (NICA) is presented. A novel approach is applied to time slice processing at high particle multiplicity and pile-up. The model is trained using a hierarchical approach and evaluated on time slices generated by the Monte Carlo technique according to the SPD experiment configuration as close as possible at the moment. The model architecture includes a graph encoder and an edge classifier, both of which use the message-passing principle and iteratively structured approach. The GANN exhibits high performance in track clustering by events for the SPD simulation, achieving 95% accuracy and 90% precision and recall.

        Speaker: Savelii Omelianchuk (MLIT JINR)
      • 14:45
        PIdML: ML-Based Particle Identification in MpdRoot 15m

        Particle identification (PID) is a key component of physics analyses in modern collider experiments. In the MpdRoot, PID is traditionally based on detector-specific observables such as energy loss in the Time Projection Chamber (TPC) and time-of-flight (TOF) measurements. Recent advances in machine learning (ML) offer significant improvements in PID performance. In this work, we present PIdML, a universal module for ML-based PID in MpdRoot. The key feature of PIdML is the use of the Open Neural Network Exchange (ONNX) format, which enables seamless deployment of models trained with a wide range of ML algorithms, including gradient boosting methods and deep neural networks. This model-agnostic approach decouples the training procedure from the inference stage within the reconstruction chain.

        Speaker: Vladimir Papoyan (JINR & AANL)
      • 15:00
        Tracklet aided determination of t0 in Straw Trackers 15m

        For both triggered and free-running DAQ, t0 determination independent of clock referencing is of interest: for clock-distribution checks (for the former) and actual tracklet reconstruction (for the latter). The method presented uses solely positions of the hit-straws in tracklets - from which it subsequently derives successive sets of "virtual hits", that are used to determine the distances of closest approach to the axes of the straws. With these it then infers the drift times, via the calibrated inverse drift-to-distance function - thereby obtaining t0, using the hit times. The results will show the t0 error and discuss its feasibility as clock-distribution test and t0 determination method in triggerless DAQ.

        Speaker: Mihai-Octavian Dima (JINR - LYAP)
      • 15:15
        Generative approach to charge cluster simulation for the GEM detector in the BM@N experiment 15m

        The BM@N experiment at the NICA accelerator complex studies the properties of dense baryonic matter in collisions of heavy-ion beams with a fixed target. The core of the central tracker in the experimental setup is a multi-chamber detector with microstrip readout. Detailed simulation at the level of charged clusters is an essential part of physics data analysis. The formation of charged clusters on the readout plane involves the following sequential processes: primary ionization, electron drift and diffusion, avalanche multiplication within the GEM microstructures, and signal induction on the readout strips. Conventional simulation based on the Garfield++ package accurately reproduces these processes at the microscopic level. However, this approach is computationally expensive and too slow for generating large statistics. To overcome this problem, this work proposes a generative approach based on deep learning. A Generative Convolutional Deep Neural Network is used as the model, which is trained on detailed datasets from conventional simulations. Such a network architecture enables the generation of realistic clusters driven by specified parameters. The results presented in this work demonstrate the readiness of the method for integration into the BM@N data processing workflow, thereby opening up prospects for scalable fast simulation of the GEM detector.

        Speaker: Mr Dmitry Baranov (JINR)
      • 15:30
        Full-scale simulation of the MPD experimental setup and data analysis techniques 15m

        The Multi-Purpose Detector (MPD) is one of the three experiments of the Nuclotron Ion Collider-fAcility (NICA) complex, which is currently under construction at the Joint Institute for Nuclear Research in Dubna. With collisions of heavy ions in the collider and fixed target modes, the MPD will cover the energy range 2.5-11 GeV to scan the high baryon-density region of the QCD phase
        diagram. With expected statistics of 50-100 million events collected during the first run, MPD will be able to study a number of observables, including measurements of light hadrons and hypernuclei production, particle flow, correlations and fluctuations.
        We will present selected results of the full-scale simulation of the MPD-NICA experimental setup for and discuss the data analysis techniques.

        Speaker: Аркадий Тараненко (VBLHEP JINR)
      • 15:45
        Fast Weighted-DOCA Seeding for Primary Vertex Linearization in SPD 15m

        We present a fast, non-iterative method for obtaining a primary vertex seed in the transverse plane for the SPD experiment at NICA. The algorithm operates directly on track pairs at their mutual distance of closest approach (DOCA). An initial 2D seed is computed via a weighted average of these DOCA points, where the weight assigned to each pair is given by: $w_{ij} = \left[|\vec{d_i}−\vec{d_j}|^2 + µ^2\right]^{−k}$.

        The parameter µ regularizes the singularity for nearly parallel tracks, while the exponent k controls the degree of outlier suppression. A radial acceptance cut is applied to exclude DOCA points far from the local density maximum, and the remaining points are averaged without further weighting to fix the seed coordinates. All tracks are then analytically swum to the point of closest approach relative to this 2D seed and re-parameterized locally. By expanding around this estimate, the non-linear vertex constraint reduces to a strictly linear problem, permitting a fast analytical vertex fit. This hybrid approach isolates the non-linear gross localization from the precision fit, maintaining the speed required for online flash reconstruction. The parameters k, µ, and the acceptance radius are tuned to the expected multiplicity and background conditions of SPD.

        Speaker: Maria Dima (JINR - DLNP)
    • 16:00 → 16:20
      Coffee break
    • 16:20 → 18:35
      Machine learning and big data analytics Room 420 (MLIT)

      Room 420

      MLIT

      • 16:20
        Methods of orchestration of non-deterministic tasks in analytical workflows: a hybrid of static graphs and agent subgraphs, stop policies and observability 15m

        The orchestration of analytical pipelines is complicated in applications aligned with big data processing and analytics in scientific and enterprise environments by growing data volumes and the increasing adoption of large language model based components. For nondeterministic workloads (external APIs, web integrations, and agentic ReAct loops), a single static execution plan is insufficient: cost drift, output variability, and reproducibility risks arise. Based on an intelligent scientific and technical data analysis system as a case study, a methodological framework is presented and grounded for orchestrating such tasks within a hybrid architecture: a batch orchestrator directed acyclic graph as the scaffold for scalable stages, complemented by agent subgraphs in regions that require adaptation. Minimal requirements are formulated for state logging, together with a catalog of stopping and escalation policies and observability principles, namely correlating run identifiers, retaining orchestrator task logs, and linking those logs to data artifacts, with traceability treated as an integral part of observability.

      • 16:35
        Выявление устойчивых трендов и редких тем в патентных данных на основе консенсуса трех принципиально разных методов и семантической унификации 15m

        Традиционные методы тематического моделирования часто страдают от неустойчивости результатов, зависимости от гиперпараметров и неспособности выявлять редкие, но значимые сигналы в зашумленных патентных корпусах. Это ограничивает их применимость для стратегического анализа технологических трендов и поиска инноваций. В работе предложена методика, основанная на "треугольном" консенсусе принципиально разных алгоритмов: геометрического (K-Means), вероятностного (LDA) и алгебраического (LSA). Методика включает расчет интегральной метрики согласованности для валидации тем, стратегию разделения на устойчивые ядра и редкие ниши, а также семантическую унификацию тем в супер-темы с использованием LLM-нейминга и иерархической кластеризации.
        На основе анализа корпуса из около 45 тыс. патентов по робототехнике выявлены устойчивые базисные уровни гранулярности (K=4–6, 8) и сформировано 7 супер-тем, агрегирующих дублирующиеся кластеры. Обнаружены редкие инновационные ниши. Валидация показала полное соответствие таксономии запросу и корректное игнорирование статистически незначимых сигналов.

        Список литературы

        1. Monti S., Tamayo P., Mesirov J., Golub T. Consensus clustering: a resampling-based method for class discovery and visualization of gene expression microarray data // Machine Learning. 2003. Vol. 52. P. 91–118.
        2. Grootendorst M. BERTopic: neural topic modeling with a class-based TF-IDF procedure [Electronic resource] // arXiv preprint. 2022. arXiv:2203.05794v1 [cs.CL]. URL: https://arxiv.org/abs/2203.05794 (дата обращения: 08.05.2026).
        3. Blei D.M., Ng A.Y., Jordan M.I. Latent Dirichlet allocation // Journal of Machine Learning Research. 2003. Vol. 3. P. 993–1022.
        4. Wang X., McCallum A., Wei X. Ensemble topic modeling // Proceedings of the IEEE International Conference on Data Mining (ICDM). IEEE, 2011. P. 1098–1103.
        Speaker: Дарья Зрелова (JINR MLIT)
      • 16:50
        Machine learning approaches to quality assessment in automated recruitment systems 15m

        Automated tracking systems (ATS) have become a cornerstone of modern recruitment; however, their quality assessment remains an underexplored challenge. This paper proposes a methodology for evaluating the effectiveness of automated recruitment systems through machine learning techniques. We examine candidate clustering algorithms to identify patterns in applicant profiles, as well as predictive models for estimating hiring success probability. A comparative analysis of supervised and unsupervised learning algorithms — including k-means clustering, random forest, and gradient boosting — is conducted to determine the most effective approaches for recruitment quality evaluation. The results demonstrate that ML-based assessment frameworks substantially improve the accuracy and transparency of hiring decisions within organizational-technical systems

        Speaker: Anastasia Novitskaya
      • 17:05
        DU-net architecture for geological inversion with variability in data recording system 15m

        Currently, the approaches that prevail in the scientific community for solving geological inversion problems involve creating additional synthetic data of the required configuration [1] or using stochastic gradient descent methods for a neural network specialized on a specific set of seismograms [2]. This situation in the field is due to the high variability of probing pulse recording systems, which prevents the formation of a sufficient training sample. However, these approaches have significant drawbacks. The method based on additional synthetic data is characterized by a high dependence on the built-in physical models. Meanwhile, set specialized neural networks require computational time comparable to classical FWI methods.

        This work aims to examine the possibility of reducing the impact of data recording system variability for neural networks of the U-net architecture family [3] by modifying the method used for applying the neural network block. These modifications are achieved by changing both the architecture and the approach to calculating the loss function. Validation was conducted on a set of synthetic geological media from the paper [4], for each media seismograms were generated by using the acoustic approximation.

        Based on the results of numerical experiments, it can be concluded that the dependence of U-net-like architectures on data recording system variability was significantly reduced, while the prediction accuracy of convolutional neural networks for geological inversion on unvaried data was improved. Additionally, we show the compatibility of the proposed method with architectures specialized for geological inversion. It was demonstrated using the W-net [5] architecture as an example.

        Speaker: Maxim Nikishin
      • 17:20
        Application of AI agents for particle accelerator modeling 15m

        The development of particle accelerators and large‑scale experiments has led to increasing difficulties in manual control, parameter tuning, and experiment planning. In recent years, AI technologies — including agent‑based systems and machine learning (ML) methods — have been actively integrated into accelerator control processes, beam optimization, and the automation of experimental cycles. At the same time, two trends are emerging: agent‑based AI powered by large language models (LLM) and classical optimization based on reinforcement learning (RL).
        A prominent demonstration of the agent‑based approach is provided by the work of Lawrence Berkeley National Laboratory research group in paper [1] and in the subsequent practical implementation [2]. These publications describe a system in which LLMs control a multi‑agent architecture at the Advanced Light Source (ALS) synchrotron. Such agent‑based AI acts as a “scientific partner” to the operator, drastically reducing the time required to prepare multi‑component experiments and decreasing the need for routine manual programming.
        At the same time, there is another path to automation: the control and optimization of accelerator systems based on RL and ML, as presented in works [3,4]. The described algorithms made it possible to achieve results comparable to manual optimization performed by highly qualified operators — but in significantly less time. This approach creates an “intelligent assistant” for the physics operator, which automatically suggests optimal settings rather than fully autonomously planning the research program, as in agent‑based system in [2]. The proposed approach on practice is able to outperform classical mathematical approach used for control of complex equipment [5]. Complicated systems constructed of many components could be efficiently modeled and controlled using multi-agent system approach [6]. An application of AI agent for the supplementary task of automatic assessment of the research metrics was considered in [7].
        Both approaches demonstrate that AI agents are already moving from the prototype stage into real operational conditions, opening up prospects for automating not only routine tasks but also a significant portion of the research cycle at large‑scale accelerator facilities.

        1. A. Sulc, T. Hellert, R. Kammering, H. Hoschouer and J. S. John, Towards agentic AI on particle accelerators, Machine Learning and the Physical Sciences Workshop, NeurIPS 2024, (2024).
        2. T. Hellert, D. Bertwistle, S. C. Leemann, A. Sulc and M. Venturini, Agentic AI for multi-stage physics experiments at a large-scale user facility particle accelerator, Physical Review Research, (2025), 8(1), p. L012017.
        3. A. Ibrahim, A. Petrenko, M. Kaledin, E. Suleiman, F. Ratnikov and D. Derkach, Reinforcement learning for accelerator beamline control: A simulation-based approach, International Journal of Modern Physics E, (2026), p. 2641027.
        4. J. Kaiser, C. Xu, A. Eichler and A. Santamaria Garcia, Bridging the gap between machine learning and particle accelerator physics with high-speed, differentiable simulations, Physical review accelerators and beams, (2024), 27(5), p. 054601.
        5. S. N. Mamaeva, A. N. Pavlov, N. A. Nikolaeva and G. V. Maksimov, Mathematical Modeling of the Main Characteristics of Cold Field and Thermal Field Electron Cathodes of Scanning Electron Microscopes in the Study of Biological Samples, Technical Physics, 69 (7), (2024), p. 2045.
        6. E. Borisoglebskaya., V. Erofeeva, O. Granichin, Y. Ivanskiy, and N. Kizhaeva, Distributed load balancing based on modified local voting protocol, Proceedings of 2023 9th International Conference on Control, Decision and Information Technologies (CoDIT), IEEE, (2023, July), pp. 2032-2036.
        7. D. I. Koshlan, E. S. Tretyakov, V. V. Korenkov, B. N. Onykij, A. A. Artamonov, Agent technology situational express-analysis in assessment of technological development level of the BRICS countries, Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and Education", (2018), p. 10-14.
        Speaker: Natalia Kizhaeva (Saint Petersburg State University)
      • 17:35
        Сравнительный анализ методов объединения семантических моделей в задаче сопоставления вакансий рынка труда и образовательных компетенций 15m

        Задача оценки семантического сходства текстов является одной из ключевых задач обработки естественного языка. Современные подходы к её решению основаны на построении векторных представлений текстов с последующим вычислением степени их близости с использованием различных метрик. Данная работа продолжает цикл исследований авторов, посвящённых анализу соответствия профессиональных компетенций образовательных программ актуальным требованиям рынка труда на основе технологий обработки естественного языка и методов машинного обучения.
        В настоящей работе проводится верификация устойчивости методов оценки семантического сходства при использовании ансамблей нейросетевых языковых моделей на наборе данных профессиональных компетенций образовательных программ трёх университетов и описания вакансий, размещенных на платформе «HeadHunter» в текущем году.
        Показано, что метод усреднения косинусных расстояний сохраняет наибольшую устойчивость и демонстрирует наиболее высокую степень согласованности при увеличении объёма и разнообразия данных. Полученные результаты могут быть использованы для дальнейшего развития ансамблевых методов оценки семантической близости текстов.

        Speaker: Анна Ильина (MLIT JINR)
      • 17:50
        Multi-Agent Architecture for Intelligent Search in a Scientific and Technical Knowledge Base 15m

        The rapid growth of scientific and technical information makes traditional search systems insufficient for processing complex research queries related not only to finding specific publications, but also to topic exploration, comparison of scientific approaches, trend analysis, and analytical profiling. Studies on academic searching distinguish several types of search activities, including lookup, exploratory, and systematic search, each requiring different search and information processing strategies [1].
        The aim of this work is to develop a multi-agent architecture for intelligent search and analysis in a scientific and technical knowledge base. The proposed approach is based on specialized agents responsible for different stages of query processing. The Planning Agent analyzes the user query, determines the research task type, and constructs a symbolic execution plan. The Retrieval Agent searches for relevant publications, the Aggregation Agent performs data aggregation and structuring, the Analysis Agent carries out analytical processing in overview, comparison, and trend analysis modes, while the Validation Agent controls the correctness of the execution plan and the completeness of the obtained results.
        The architecture dynamically adapts the execution scenario depending on the query type. Simple navigational queries involve only the retrieval stage, whereas analytical tasks may require multiple retrieval calls, data aggregation, and content analysis. This approach is consistent with recent studies on LLM-based multi-agent systems, where large language models are used for coordination, planning, and orchestration of specialized agents [2, 3].
        As a result, a preliminary architecture of a multi-agent system for scientific and technical information retrieval and analysis is proposed. The developed approach can serve as a basis for intelligent search and analytical services for scientific organizations and research platforms.

        References

        1. Gusenbauer M., Haddaway N. R. What Every Researcher Should Know About Searching – Clarified Concepts, Search Advice, and an Agenda to Improve Finding in Academia. Research Synthesis Methods, 2021.
        2. Guo T. et al. Large Language Model Based Multi-Agents: A Survey of Progress and Challenges. IJCAI, 2024.
        3. Wu Q. et al. AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation, 2023.
        Speaker: Tatiana Korenkova
      • 18:05
        Development of a software module for identifying the equations of motion of agents in a multi-agent system using physically-based AI tools 15m

        Abstract:
        This paper examines the Swarmodroid platform — a swarm of 30 self-rotating robots in a circular arena. By interacting with one another, they form collective structures and can be used to simulate pedestrian movement. However, it is first necessary to study the dynamics of the robots in the absence of prior physical knowledge about them. To this end, a tool for the automatic identification of mathematical models in the form of ODE systems based on kinematic data is employed.

        Consequently, the effectiveness of using the Optuna library to tune the tool’s hyperparameters, a model of charged particle motion in an electromagnetic field as prior physical knowledge, and the clustering of robots to obtain mathematical models accounting for their interactions is demonstrated.

        Results:
        This work has resulted in the development of a software module for identifying ordinary differential equation (ODE) systems that describe the motion of robots based on kinematic data. The implementation is available in the repository at the following link: https://github.com/20saaa02/Active_Matter . The software module supports the loading and pre-processing of experimental data, automatic identification of ODE systems using the EPDE tool, and trajectory reconstruction using TEDEouS.

        The Optuna library has been integrated for the automatic selection of EPDE and TEDEouS hyperparameters. For circular robots, using a single example, it has been shown that this reduced the time taken to obtain a solution by a factor of 10 whilst increasing accuracy by a factor of 4 compared to random parameter selection.

        For circular robots, which have the fewest points of contact with others, the use of a charged particle motion model in an electromagnetic field as prior physical knowledge provided, on average, a nearly threefold advantage in accuracy, a 10,000-fold advantage in stability, and a significant simplification of the ODE systems.

        The individual, collective-global and collective-cluster approaches to the identification of robotic motion systems were compared. As a result, the individual approach yielded the highest average accuracy of the differential operators with the shortest identification time. The collective-general approach allowed for the interaction of all robots to be taken into account, but led to an increase in the dimensionality of the models (60 equations), a more than six-fold increase in computation time, and a 2.3-fold reduction in accuracy. The collective-cluster approach provided a compromise between accounting for collective dynamics, model quality stability and computational costs: whilst the identification time increased by less than 1.3 times compared to the individual approach, the accuracy decreased by a factor of 1.8.

        Speaker: Alena Gushchina (ITMO University Saint Petersburg 197101, Russia)
    • 16:20 → 18:35
      Mathematical methods and tools for modeling complex physical systems Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 16:20
        Numerical study of the flow around a descent spacecraft taking into account radiative heat transfer 15m

        Over the past decade, there has been a renewed interest in space exploration. This is due to the high practical utility of near-Earth orbit (navigation, internet), as well as the scientific significance of the deep space explorations. Therefore, the development of new spacecraft capable of withstanding loads over a wide range of parameters is required, including those encountered in near-atmospheric environments. This paper examines the flow around a reentry vehicle. The model is based on the system of quasi-gasdynamic equations for calculating the flow. Equations for the diffuse multigroup approximation of radiative heat transfer are also used. The computational algorithm is constructed using the finite volume method. Its parallel program realization is based on the mesh-operator approach. The program was used to calculate the flow around a three-dimensional vehicle with complex geometry. The calculation results demonstrated the correctness of the numerical algorithm and the program realization.

        The work was supported by the Russian Science Foundation (project No. 25-11-00099).

        Speaker: Nikita Tarasov (KIAM RAS)
      • 16:35
        Numerical study of purging modes of the drift chamber of the TREK detector 15m

        This work presents a numerical study of the purging of a drift chamber manufactured at IHEP for the U-70 accelerator and currently used in the TREK detector (NRNU MEPhI). To ensure the chamber's working condition, the oxygen concentration inside it is reduced to the threshold of 0.1% by filling it with an Ar + CO2 gas mixture. The search for an optimal purging method and gas mixture flow rates remains relevant. Although the problem addressed is similar to known studies on tank purging, the significant difference in the geometry of the drift chamber means that existing approaches cannot be directly applied. The intermediate result of the study is the cross-verification of the LOGOS and OpenFOAM software packages for the simulation of low-speed multicomponent mixture flow, taking into account gravity and diffusion. The difference between the results obtained in the two software packages for all considered purging methods does not exceed 9% in terms of the maximum mass fraction of O2 in the chamber and 2% in terms of the time to reach the threshold value. The predicted purge time is compared with two available experimental values. The study provides the results for a variety of purge modes that have not been documented experimentally. The optimal flow rate and gas supply method have been determined.

        Speaker: Mr Georgii Ageev (NRNU MEPhI)
      • 16:50
        Analytical solution of a fourth-order algebraic equation obtained within the framework of modeling the Brownian motion of particles 15m

        A charged Brownian particle (BP) immersed in a medium of neutral particles under the action of an external magnetic field is considered. The chaotic motion of the BP is described by the generalized Langevin equation with the magnetic Lorentz force, taking into account memory effects in the dynamics of the system. Using its solutions for displacement and velocity in Laplace space, we obtained formulas for the mean angular momentum and mean kinetic energy of BP for Ornstein-Uhlenbeck thermal noise by analytical and exact numerical calculations.

        To find such formulas, it is necessary to determine the inverse Laplace transforms of the key functions determining the solution in the time domain, and also to calculate the integrals that include these functions, their derivatives, and the time correlation function of the thermal noise force.

        This report describes in detail the partial fraction decomposition: finding of analytical formulas for the roots of a given part of the denominator of the Laplace transforms, a fourth-order polynomial, and derivation of corresponding decomposition coefficients.

        Speaker: Dr Ján Buša (MLIT, JINR)
      • 17:05
        Modeling granule processing in a rotary tumbling machine 15m

        Studying the processing and cleaning of various types of granules is a pressing issue today. For example, when filtering an aqueous medium with a granular sorbent, a surface layer of calcium carbonate can form on it. This will significantly reduce cleaning efficiency, as the contact surface between the sorbent and the medium will decrease. Tumbling can be one of the most effective methods for removing deposits. This study examines the dynamics of a granule system in a rotary tumbling machine. The mathematical analysis is based on the discrete element method. The computational algorithm is based on the numerical integration over time of the Newtonian dynamics equations for each granule using the velocity Verlet scheme. A software implementation was also developed. Numerical experiments demonstrated the correctness of the computational algorithm and its software implementation.

        Speaker: Mr Nikita Kalinichev (Keldysh Institute of Applied Mathematics of Russian Academy of Sciences)
      • 17:20
        Numerical solution of elliptic boundary value problems using edge-based approximation 15m

        Elliptic boundary value problems are considered. Two approaches are used for their numerical solution: the finite element method and the finite volume method, both employing a nonstandard arrangement of approximation nodes — on the edges of triangles.

        Within the finite element method, nonconforming Crouzeix–Raviart elements are used. Unlike standard Lagrangian elements, the degrees of freedom here are associated not with vertices but with the midpoints of triangle edges.

        In the finite volume method, the control volumes are quadrilaterals whose diagonals consist of an edge and a segment connecting the centers of the two triangles sharing that edge. The triangle centers considered are the centroid (center of mass) and the circumcenter.

        Using a test elliptic problem with a known analytical solution, the accuracy of both methods is compared on a sequence of successively refined triangular meshes. The convergence rate of the numerical solution under mesh refinement is evaluated.

        Speaker: Mikhail Chernyshov
      • 17:35
        Molecular Dynamics Modeling of Ni and Cu Nanocluster Interaction with a Metal Substrate 15m

        This study investigates the interaction mechanisms of copper and nickel nanoclusters with a metal substrate within the framework of Supersonic Cold Gas Dynamic Spraying (SCGDS). Given the technological significance of SCGDS in modern nanotechnology manufacturing, further clarification is required regarding how coating properties depend on factors such as the critical velocity of nanoclusters, their size, thermal regimes, and the characteristics of the coating itself. Particular attention is paid to the analysis of residual stresses arising within the material structure. Various particle collision scenarios were modeled using the molecular dynamics method. The results were verified by comparison with existing literature data. It was established that the tangential velocity and angular momentum of the particles play a critical role in determining the quality of the resulting composite coating. The findings can be utilized to optimize the technological parameters of the SCGDS process.

        This work was supported by the Russian Science Foundation, project No. 25-71-20016.

        Speaker: IVAN SAFONOV (Keldysh Institute of Applied Mathematics)
      • 17:50
        Calculation of effective response of ring metamaterials in a rectangular waveguide 15m

        Exact analysis of finite metamaterial structures is complex in electromagnetic
        problems and usually requires computationally expensive numerical modeling or
        derivation of effective medium theory. Especially for ring resonators,
        effective magnetic permeability $\mu_{eff}$ based on effective theory for
        infinite homogeneous medium can be easily computed. However, there is still a
        large number of limitations for this theory:
        this theory does not allow
        calculation dielectric permittivity for given structures,
        there is a huge
        divergence of calculated parameters from those obtained for applicable finite
        structures in real experiments
        and slow convergence with increasing number of metacells.
        Moreover, for ring resonators, magnetic permeability is calculated only for
        free space metaatoms configuration assuming a deep quasi-static regime:
        not only the size of the ring is much smaller than the wavelength of
        excitation, but the whole metastructure should be also much smaller.

        This work investigates an optimized numerical approach to analyze
        parameters for ring resonators in a rectangular waveguide with any
        number of dielectric layers. Green’s function expansion in waveguide modes is
        used in Method of Moments formulation with the integral equation for electric
        field (EFIE). Then, the Galerkin method with weighting functions is used for
        approximate computation of current distribution along each metaatom.
        Currents on each ring are represented as uniform along the radial
        direction and expanded into a finite number of Fourier harmonics.

        For calculating dielectric permittivity and magnetic permeability
        we use stepwise Nicolson–Ross– Weir-based material parameter
        extraction method (NRW method). Based on numerical extraction of S-parameters,
        we obrain frequency dependency of material parameters. These results are
        compared with effective medium theory and studied for different lattices of
        given anisotropic metastructures.
        Acknowledgment
        The work was supported by the Russian Science Foundation, grant no. 22-11-00153-$\Pi$.

        Speaker: Aleksandr Makarenko (ITMO University)
      • 18:05
        GPU-Accelerated Particle-in-Cell Simulation of Thermo-Penning Discharges 15m

        Plasma-based separation is a promising approach for the treatment and fractionation of nuclear waste, where selective ion dynamics in crossed electric and magnetic fields may enable efficient separation of heavy-ion species. A key challenge for the design of such systems is the need for predictive numerical tools that can resolve the kinetic behavior of charged particles in a complex discharge while remaining fast enough for parameter scans and experimental optimization.

        In this work, we present a GPU-oriented numerical framework for modeling a thermo-Penning discharge relevant to plasma separation devices. The method is based on a two-dimensional electrostatic Particle-in-Cell formulation with gyrokinetic electrons, kinetic ions, self-consistent electric fields, particle sources and sinks at electrodes, and collision processes including electron-neutral elastic scattering, ion-neutral elastic scattering and charge exchange, and electron-impact ionization. The implementation is developed in JAX with a functional, accelerator-friendly architecture: fixed-capacity particle arrays, mask-based particle life-cycle management, linear charge deposition and field gathering, and an iterative Poisson solver on a circular plasma domain.

        The proposed approach is compared with an existing reference implementation of the thermo-Penning discharge model, focusing on both physical diagnostics and computational performance. Validation is performed using time-dependent integral quantities such as electron and ion populations, total ionization yield, particle losses to electrodes, and spatial distributions of charge density and electrostatic potential. Particular attention is paid to preserving the essential physics of the original model while restructuring the algorithm for efficient just-in-time compilation and execution on modern GPUs.

        The expected performance gain enables repeated simulations over discharge parameters, electrode conditions, magnetic field strength, neutral gas pressure, and source characteristics. This makes the method suitable not only for direct discharge modeling, but also for optimization workflows aimed at selecting operating regimes of experimental plasma separation facilities. The developed JAX/GPU framework therefore provides a bridge between detailed kinetic modeling and practical design-space exploration for plasma-based nuclear waste separation.

        The study was supported by grant No. 25-79-30008 from the Russian Science Foundation, https://rscf.ru/project/25-79-30008/

        Speaker: Daniil Kolotinskii (Joint Institute for High Temperatures of the Russian Academy of Sciences, Izhorskaya 13 Bldg 2, Moscow 125412, Russia)
    • 16:20 → 18:20
      Mathematical methods in life sciences Room 406 (MLIT)

      Room 406

      MLIT

      • 16:20
        Multifactorial mathematical model for selecting optimal ventilation flow values when connecting a patient to an artificial lung ventilation device / Многофакторная математическая модель выбора оптимальных значений вентиляционного потока при подключении пациента к аппарату искусственной вентиляции легких 15m

        The report presents a decision support system for selecting optimal ventilation flow values when connecting a patient to a ventilator. The presented system solves the problem of selecting the optimal option for a set of ventilation flow values depending on the values of the patient's current state indicators. To solve this problem, tables of normalized already implemented optimal values of the initial data are formed - ventilation flow indicators for each already implemented option for their selection for this patient. Based on the calculated values of the distances of the weighted metric, taking into account the priorities of the patient's current state indicators, the average weighted optimal value is calculated for each ventilation flow indicator.

        В докладе представлена система поддержки принятия решений по выбору оптимальных значений вентиляционного потока при подключении пациента к аппарату искусственной вентиляции легких. В представленной системе решается задача выбора оптимального варианта набора значений вентиляционного потока в зависимости от значений показателей текущего состояния пациента. Для решения этой задачи формируются таблицы нормированных уже реализованных оптимальных значений исходных данных – показателей вентиляционного потока для каждого уже реализованного варианта их выбора для этого пациента. На основании рассчитанных значений расстояний взвешенной метрики с учетом приоритетов показателей текущего состояния пациента производится расчет средне взвешенного оптимального значения для каждого показателя вентиляционного потока.

        Speaker: Alexander Kryanev (professor)
      • 16:35
        Neural activity and pattern recognition modeling based on the electrical activity of neurons in the basal ganglia of Parkinson's disease patients 15m

        In most neuronal recordings from primates, burst, pause, and tonic are widely studied neuronal electrical activity patterns. In this work, we utilize clinical data of neuronal electrical activity (pulse sequences) recorded from the basal ganglia of Parkinson's disease patients to construct models and corresponding detection methods that primarily contain these neuronal electrical activity patterns. We develop a method for detecting pulse sequence patterns (burst, pause, and tonic) and apply it to engineering practice. This work is expected to provide new insights for the development of spiking neural networks.

        Speaker: Houchn Zhou (MIPT)
      • 16:50
        Photochemical bleaching kinetics of archaerhodopsin-3 reveal bacterioruberin-dependent stabilization 15m

        Microbial rhodopsins are retinal-binding membrane proteins in which light absorption by the retinal chromophore drives ion transport, sensory, or signaling processes. In some organisms, carotenoids form native complexes with rhodopsins and may act as light-harvesting antennae, structural cofactors, or photoprotective pigments [1,2]. According to the low resolution structural studies of Halorubrum Sodomense claret membranes, trimers of archaerhodopsin-3 (AR-3) coexists with the C50 carotenoid bacterioruberin [3]. AR-3 is also of particular importance for optogenetics, where it has served for proton pumping as well as basis for light-driven neuronal silencers and voltage-sensitive fluorescent reporters [4,5].
        Here, we analyzed hydroxylamine hydrochloride-induced retinal bleaching of AR-3 in native claret membranes. Hydroxylamine reacts with the retinal Schiff base of microbial rhodopsins, resulting in retinal oxime formation and loss of the characteristic retinal absorption band; this reaction is strongly accelerated by illumination [6]. Two membrane states were compared: holo-AR-3, containing both retinal and bacterioruberin, and bacterioruberin-depleted apo-AR-3, where the carotenoid was oxidatively removed by ammonium persulfate treatment. The time-dependent decrease of the retinal-band absorbance was treated as a first-order photochemical bleaching process and fitted with a monoexponential model:

        $$ A(t)=A_{\infty}+\bigl(A_{0}-A_{\infty}\bigr)e^{-k_{\mathrm{obs}}t}, \qquad t_{1/2}=\frac{\ln 2}{k_{\mathrm{obs}}} $$ where $A(t)$ is the retinal-band absorbance at time $t$, $k_{\mathrm{obs}}$ is the apparent bleaching rate constant, and $t_{1/2}$ is the bleaching half-life.
        Under identical high-stress bleaching conditions, including 600 mM hydroxylamine hydrochloride and intense illumination, apo-AR-3 bleached substantially faster than holo-AR-3. Exponential fitting showed that the extracted bleaching half-life differed by approximately a factor of two, corresponding to an approximately twofold increase in the apparent bleaching rate constant after bacterioruberin removal. Our result serves as the evidence that bound bacterioruberin enhances the photochemical stability of the AR-3 membrane complex under photochemical stress. Thus, bacterioruberin appears to act not only as a native membrane pigment but also as a stabilizing and protective cofactor. The proposed kinetic model provides a compact quantitative descriptor of carotenoid-dependent rhodopsin stability.
        This work was supported by the Russian Science Foundation, grant No. 25-24-00851.

        References:
        [1] Balashov, S. P., et al. Xanthorhodopsin: a proton pump with a light-harvesting carotenoid antenna. Science 309, 2061-2064 (2005). doi:10.1126/science.1118046.
        [2] Yoshimura, K., et al. Structural role of bacterioruberin in the trimeric structure of archaerhodopsin-2. Journal of Molecular Biology 375(5), 1267-1281 (2008). doi:10.1016/j.jmb.2007.11.039.
        [3] Bada Juarez, et al. Structures of the archaerhodopsin-3 transporter reveal that disordering of internal water networks underpins receptor sensitization. Nature Communications 12, 629 (2021). doi:10.1038/s41467-020-20596-0.
        [4] Flytzanis, N. C., et al. Archaerhodopsin variants with enhanced voltage-sensitive fluorescence in mammalian and Caenorhabditis elegans neurons. Nature Communications 5, 4894 (2014). doi:10.1038/ncomms5894.
        [5] Vlasova, A. D., et al. Intracellular microbial rhodopsin-based optogenetics to control metabolism and cell signaling. Chem. Soc. Rev. 53(7), 3327-3349 (2024). doi:10.1039/D3CS00699A.
        [6] Zadok, U., et al. The hydroxylamine reaction of sensory rhodopsin II. Biophysical Journal 89(4), 2610-2617 (2005). doi:10.1529/biophysj.105.065631.

        Speaker: Elizaveta Dronova (Moscow Institute of Physics and Technology)
      • 17:05
        Multiscale modeling of inactivated actin: from apo-monomer dynamics to solution-dependent oligomerization 15m

        Multiscale modeling of inactivated actin: from apo-monomer dynamics to solution-dependent oligomerization

        Daria D. Kuklina$^{1,2}$, Ekaterina V. Laptenkova$^{1}$, Ivan O. Bezruchko$^{1,2}$, Elizaveta A. Dronova$^{1,2}$, Egor V. Zinovev$^{1}$, Yury L. Ryzhykau$^{1,3}$

        $^{1}$Research Center for Molecular Mechanisms of Aging and Age-Related Diseases, Moscow Institute of Physics and Technology, MIPT, Dolgoprudny, 141701, Russia

        $^{2}$Laboratory of Molecular Basis of Cell Motility, Institute of Cytology, Russian Academy of Sciences, St. Petersburg, 194064, Russian Federation

        $^{3}$Frank Laboratory of Neutron Physics, Joint Institute for Nuclear Research, Dubna, 141980, Russia

        E-mail: kuklina.dd@phystech.edu, rizhikov@phystech.edu

        Actin is an essential cytoskeletal and nuclear protein whose functional behavior is usually discussed in terms of two canonical forms: monomeric G-actin and filamentous F-actin [1]. However, actin can also form an alternative inactivated state, I-actin, after depletion of bound nucleotide and coordinating divalent cation or under mild stress conditions [2–4]. This state is especially intriguing because I-actin forms non-canonical oligomers that are structurally distinct from F-actin [5,6] and may provide an in vitro model for poorly characterized short actin assemblies observed in the nucleus [6].

        Here we combine atomistic molecular dynamics, geometric conformational analysis and analysis of SAXS-derived structural parameters to investigate early structural responses of actin to nucleotide and divalent-cation depletion and their possible relation to I-actin formation. At the monomer level, molecular dynamics simulations were performed for apo-actin and ligand-bound actin states, including ATP- and ATP/Mg-bound forms. The trajectories were analyzed using conventional structural descriptors and an interdomain twist angle that distinguishes twisted G-like conformations from flattened F-like conformations [7]. Within the simulated nanosecond-scale time window, no unidirectional transition that could be directly assigned to I-actin formation was observed. Instead, apo-actin displayed larger-amplitude interdomain fluctuations than the ligand-bound forms, suggesting increased conformational lability after nucleotide and cation removal. Thus, the simulations are used here mainly to characterize early monomer-level destabilization, whereas the experimentally observed formation of I-actin oligomers likely occurs on much longer timescales.
        In parallel, we analyzed potential aggregation-prone regions and their possible exposure upon destabilization of the native monomer. These calculations were used not as direct evidence for amyloid-like aggregation, but as a test of whether I-actin assembly can be reduced to nonspecific exposure of amyloidogenic fragments. The available data rather support a more specific multistage pathway.

        At the supramolecular level, SAXS-derived structural parameters were used to com-pare different I-actin oligomeric states and to formulate a phenomenological interpretation of solution-dependent oligomerization. Previously observed time-inactivated species are considered as intermediate oligomers [5], while heat-inactivated actin forms flat bead-like particles that can further associate into one-dimensional bead-on-a-string chains [6]. We ad-ditionally discuss whether changes in ionic strength, including those caused by different EDTA concentrations, may shift the distribution of oligomeric states through electrostatic screening and altered association of charged intermediates. This framework relates solution conditions to the experimentally observed oligomer distributions and identifies electrostatic screening as a plausible physical factor controlling transitions between intermediate I-actin assemblies.

        This work was supported by the Russian Science Foundation, grant No. 23-15-00494-P.

        References
        [1] T.D. Pollard, J.A. Cooper, Actin, a Central Player in Cell Shape and Movement, Science (1979). 326 (2009) 1208–1212. https://doi.org/10.1126/science.1175862.
        [2] S.S. Lehrer, G. Kerwar, Intrinsic Fluorescence of Actin, Biochemistry 11 (1972) 1211–1217. https://doi.org/10.1021/bi00757a015.
        [3] K.K. Turoverov, A.G. Biktashev, S.Y. Khaitlina, I.M. Kuznetsova, The structure and dynamics of partially folded actin, Biochemistry 38 (1999) 6261–6269. https://doi.org/10.1021/bi9900976.
        [4] I.M. Kuznetsova, A.G. Biktashev, S.Yu. Khaitlina, K.S. Vassilenko, K.K. Turoverov, V.N. Uversky, Effect of Self-Association on the Structural Organization of Partially Folded Proteins: Inactivated Actin, Biophys. J. 77 (1999) 2788–2800. https://doi.org/10.1016/S0006-3495(99)77111-0.
        [5] Y.L. Ryzhykau, O.I. Povarova, E.A. Dronova, D.D. Kuklina, I.A. Antifeeva, N.S. Ilyinsky, I.S. Okhrimenko, Y.S. Semenov, A.I. Kuklin, V. Ivanovich, A. V Fonin, V.N. Uversky, K.K. Turoverov, I.M. Kuznetsova, Small-angle X-ray scattering structural insights into alternative pathway of actin oligomerization associated with inactivated state, Biochem. Biophys. Res. Commun. (2024) 149340. https://doi.org/10.1016/j.bbrc.2023.149340.
        [6] Y.L. Ryzhykau, D.D. Kuklina, I.O. Bezruchko, M. V Tishkova, T.A. Golozubova, E. V Laptenkova, E.A. Dronova, E. V Zinovev, T.S. Kurkin, A. V Rogachev, A.I. Kuklin, V.N. Uversky, K.K. Turoverov, A. V Fonin, I.M. Kuznetsova, Chain-like supramolecular assemblies of inactivated actin oligomers reveal a multistage assembly pathway, Biochem. Biophys. Res. Commun. 796 (2026) 153122. https://doi.org/10.1016/j.bbrc.2025.153122.
        [7] T. Oda, Y. Maéda, Multiple Conformations of F-actin, Structure 18 (2010) 761–767. https://doi.org/10.1016/j.str.2010.05.009.

        Speakers: Daria Kuklina (Laboratory for Advanced studies of membrane proteins, Research Center for molecular mechanisms of aging and aging-related diseases, Moscow Institute of Physics and Technology (National Research University)), Dr Yury Ryzhykau (MIPT, Phystech; JINR)
      • 17:20
        Using Path Sampling to Model Binding Kinetics in the Barnase-Barstar System 15m

        In this work, the kinetics of association for the Barnase-Barstar complex are investigated using Multiple State Transition Interface Sampling (MSTIS), a path sampling method designed for the study of rare transitions. Publicly available molecular dynamics (MD) trajectories of the Barnase-Barstar system, including distributed and adaptive simulations, are used to construct ensembles of reactive binding pathways between unbound and bound states. The generated transition paths are analyzed to characterize encounter complexes, transition mechanisms, and kinetic properties of the association process. The kinetic estimates obtained from MSTIS are compared with results from direct MD simulations and state-based approaches. The study demonstrates that MSTIS provides an efficient way to reconstruct reactive association pathways and to estimate protein-protein binding kinetics in systems where direct simulation of rare binding events is computationally demanding.

        Speaker: Sergey Poluyan (Dubna State University)
      • 17:35
        Algorithm for detecting palindromes in RNA 15m

        We consider a modification of an algorithm for calculating the distance between a nucleotide sequence and the closest perfect palindrome. The aim of the work is to adapt existing algorithms for the analysis of RNA sequences. To achieve this, the edit distance is modified to account for $\texttt{GU}$ base pairs, which commonly occur in RNA structures. The resulting algorithm can be used to compute a modified version of the $\mathrm{imp}$ function for evaluating palindrome imperfection. The algorithm was evaluated on a set of near-palindromic RNA sequences using trimming algorithms. The results show that the proposed modification of the distance function increases the number of detected palindromes.

        Speakers: Georgii Khaziev (Institute for Information Transmission Problems (Kharkevich Institute) Russian Academy of Sciences), Ms Sofia Mankevich (Russian Technological University MIREA)
      • 17:50
        Automated system for monitoring the phytosanitary state of vineyards using UAVs and computer vision algorithms 15m

        The report presents an end-to-end methodology for automated phytosanitary monitoring of vineyards, integrating three logically connected stages: formalization of UAV imaging requirements, grape leaf classification, and lesion detection with damaged area estimation. At the first stage, UAV flight parameters and RGB image acquisition conditions are defined to ensure reproducible color characteristics of grape leaves, and the feasibility of primary binary classification based on RGB histogram statistics with 95% accuracy is demonstrated. At the second stage, multiclass classification of grape leaf images is performed using Gray World color normalization, transformation into HSV, YUV, and CIE LAB color spaces, and Sobel gradient features; the combination of HSV + LAB + Sobel is shown to be the most informative, achieving a macro F1 score of about 0.96. At the third stage, an algorithm for lesion localization is proposed based on Canny edge detection and threshold filtering in the HLS color space using the Hue component, which makes it possible not only to detect pathological regions but also to calculate the percentage of the affected leaf area. The experimental results show that the proposed approach combines interpretability, moderate computational complexity, and practical suitability for early grape disease diagnosis under real field conditions.

        Speaker: Sergey Samarin (MIPT)
    • 16:20 → 18:35
      Modern methods for data processing and analysis in Mega-science projects in HEP Room 310 (MLIT)

      Room 310

      MLIT

      • 16:20
        $D$--, $D_{S}$--resonances electroweak characteristics in a relativistic quark model employing analytical computing systems calculation. 15m

        The work presents a technique of applying analytical computing systems for the model based on the point form of Poincaré-invariant quantum mechanics parameters calculation. It has been shown that the usage of quark-antiquark potential with modelled effective strong coupling $\alpha_{s}(q^2)$ leads to values of the $V\rightarrow P\gamma$ radiative decays, as well as the magnetic moments of the $u$-, $d$-, $s$- and $c$-quarks, that are consistent with experimental data.

        As a generalization authors consider a variational eigenvalue problem for the Hamiltonian in the $3\times3$ case with a trial oscillator wave function. It has been shown that the usage of computer algebra systems yields the $D$- and $D_{S}$-resonances masses values that are consistent with experimental data.

        Speaker: Dr Vadzim Haurysh (Sukhoi State Technical University of Gomel)
      • 16:35
        On the Log-Normal Law of Ionization Energy Loss in the TPC Detector of the MPD/NICA Experiment / О логнормальном законе распределения ионизационных потерь в детекторе TPC эксперимента MPD/NICA 15m

        In this work, based on Monte Carlo simulation data of heavy-ion collisions, it is shown for the first time for the Time Projection Chamber (TPC) of the MPD/NICA experiment that the distribution of ionization energy loss (signal) follows a log-normal law, both at the population level and at the level of individual tracks. This observed property has a physical justification related to the multiplicative nature of fluctuation accumulation in the process of charged particle registration. Based on this observation, a new approach to particle identification is proposed, utilizing a set of statistical parameters of the track signal distribution instead of the traditional mean value. Using machine learning methods, an optimal set of track parameters is obtained, providing a 10% improvement in identification quality compared to the traditional method. The results obtained can be used to improve track reconstruction and calibration algorithms for the TPC detector in the MPD experiment.

        В настоящей работе на основе данных математического моделирования (Монте-Карло) столкновений тяжёлых ионов впервые для время-проекционной камеры (TPC) эксперимента MPD/NICA установлено, что распределение ионизационных потерь (сигнала) как на уровне генеральной совокупности, так и на уровне отдельного трека подчиняется логнормальному закону. Обнаруженное свойство имеет физическое обоснование, связанное с мультипликативным характером накопления флуктуаций в процессе регистрации заряженных частиц. На основе этого наблюдения предложен новый подход к идентификации частиц, использующий набор статистических параметров трекового распределения сигнала вместо традиционного среднего значения. С помощью методов машинного обучения получен оптимальный набор параметров трека, обеспечивающий улучшение качества идентификации на 10% по сравнению с традиционным методом. Результаты работы могут быть использованы для усовершенствования алгоритмов реконструкции треков и калибровки детектора TPC в эксперименте MPD.

        Speaker: Евгений Осетров (JINR, MLIT)
      • 16:50
        Energy Loss Reconstruction Analysis in the BM@N Experiment 15m

        In 2022–2023, the BM@N experiment recorded 500 million events of xenon beam collisions with a CsI target. The collected data are widely used for various physics analyses. One of the key tasks is the separation of singly charged particles from doubly charged ions. This is important both for studying the production of helium isotopes and for their identification in hypernuclei production studies.

        The BM@N experiment was not originally equipped with a dedicated particle identification detector based on energy loss. However, the main tracking system was repurposed for this task, as the measured signal in the tracking detectors is proportional to the particle charge.

        In this work, we present an algorithm for particle identification using energy loss, provide detector calibration procedures for this observable, and demonstrate how the application of energy loss improves the selection of the hypertriton signal.

        Speaker: Sergei Merts (JINR)
      • 17:05
        Algorithm of geometry alignment of the BM@N detectors 15m

        Geometry alignment is a mandatory ingredient of high-precision detector calibration in high-energy physics: even careful mechanical placement and direct geometric surveys leave residual misalignments that must be estimated numerically. We report on the development of an alignment procedure for the BM@N experiment (NICA, JINR), formulated as a weighted mean-square-error minimization over measured hit residuals with respect to straight-line track models, jointly over local (per-track) and global (per-detector-element) parameters.

        The high dimensionality of the resulting system (∼10⁷ unknowns for a typical BM@N alignment dataset) is handled by the block-structure inversion scheme of Blobel and Kleinwort (the "Millepede trick"). The algorithm is implemented natively in the CERN ROOT framework as part of the BmnRoot software stack, and additionally features an iterative scheme for the non-linear measurement model, Levenberg–Marquardt regularization, and support for arbitrary linear equality constraints via Lagrange multipliers. Companion tools for synthetic test-task generation from real-track data and for quantitative alignment-quality evaluation have been developed.

        We discuss the mathematical formulation, implementation details, and preliminary computational results obtained on the Run-8 dataset. Particular attention is given to the principal numerical difficulties encountered — task underdetermination, the high effective measurement error level, susceptibility to local minima — and to the strategies under investigation to address them.

        Speaker: Игорь Полев (JINR)
      • 17:20
        Framework for realistic flow measurements and centrality determination in MPD at NICA 15m

        Relativistic heavy-ion collisions at the NICA accelerator complex provide an opportunity to study the properties of strongly interacting nuclear matter at high baryon densities. Collective anisotropic flow is a sensitive probe of the equation of state (EOS) of nuclear matter, as the pressure gradients that develop in the collision region are encoded in the flow harmonics. We report on the development of a software framework for realistic measurements of anisotropic flow of identified hadrons and centrality determination in the MPD experiment at NICA. Combined with a systematic comparative analysis of predictions from the modern theoretical models of nucleus-nucleus collisions against existing experimental data, this framework provides a solid basis for the future physics program of the MPD experiment. The framework integrates two methods for centrality determination using charged-particle multiplicity, realistic particle identification via the TPC and TOF systems, and the FHCal detectors for reaction plane estimation. The framework was tested on realistically reconstructed data for two modes of the MPD operation: collider and fixed-target regime.

        Speaker: Peter Parfenov (JINR, NRNU MEPhI)
      • 17:35
        Simulation of elastic and inelastic deuteron-deuteron interactions 15m

        Polarized and un-polarized deuteron-deuteron (dd) interactions will be studied at future NICA/SPD experiment. Thus, Monte Carlo simulations of the dd collisions are needed for design and development of the experiment.
        For estimations of produced particle’s characteristics in the dd interactions we use Geant4/FTF and UrQMD+SMM models. We have obtained that the Geant4/FTF model is applicable for calculations of the inelastic dd collisions. Our calculations show that the models do not describe quasi-elastic processes well due to an underestimation of the process cross-sections.
        At a comparison with corresponding experimental data on elastic dd interactions it was observed that Geant4 models – DIFF and CHIPS, describe the elastic interactions at low four-momentum transfer (t). But, they do not describe the elastic dd interactions at higher t.
        To simulate the elastic dd processes, we apply the Glauber theoretical model. Some promising results are obtained for high energy dd elastic interactions. We have created a new code on C++ based on the Glauber approach for simulations of the elastic proton+proton, deuteron+deuteron and also alpha+alpha scatterings. The program is running under ROOT framework. Some results of the code operation will be presented in our report.

        Speaker: Аида Галоян (Veksler and Baldin Laboratory of High Energy Physics)
      • 17:50
        HEPWEB2: A Web Platform for Cross-Section Calculations and Testing of Monte Carlo Generators in High-Energy Physics 15m

        In high-energy physics, reliable estimation of hadronic cross sections and fast simulation of interaction properties are essential for experiment design, detector studies, and preliminary data analysis. The Glauber approximation remains a standard approach for calculating total, elastic, quasi-elastic, and inelastic cross sections in hadron–nucleus and nucleus–nucleus interactions using nucleon–nucleon cross sections as input. Impact-parameter distributions for hadronic and nuclear projectiles can also be obtained within this framework. However, despite the wide use of such calculations, no widely accessible standard software platform exists for routine interactive use.

        We present HEPWEB2, a web platform developed at the Laboratory of Information Technologies of the Joint Institute for Nuclear Research (JINR), which provides convenient browser-based access to computational tools for hadronic cross-section calculations and event generator studies.

        In addition to Glauber-model calculations, the platform supports fast testing of several well-established Monte Carlo models, including the Dubna Cascade–Evaporation Model, FRITIOF, and HIJING, allowing users to estimate particle production characteristics, kinematic distributions, and global event variables. While modern large-scale generators provide higher precision, their deployment often requires significant computational and technical effort. For preliminary experiment planning and fast feasibility studies, lightweight and robust models remain highly useful.

        The report will present the platform architecture, implementation details, practical examples, and typical use cases. HEPWEB2 can also serve as an effective tool for educational and pedagogical applications in high-energy physics.

        Speaker: Владимир Ужинский (LIT, JINR)
      • 18:05
        Modelling of light nuclei and hypernuclei production in heavy-ion collisions at NICA energies with use of generator THESEUS based on 3-fluid dynamics 15m

        We present a new thermodynamical approach to model hadrons, light nuclei and hypernuclei and their antiparticles production within the updated Three-fluid Hydrodynamics-based Event Simulator Extended by UrQMD (ultrarelativistic quantum molecular dynamics) final State interactions (THESEUS). The only additional parameter related to the light nuclei is the energy density of late freeze-out that imitates the afterburner stage of the collision. The validity of the chosen approach is confirmed by results of simulations of hadronic and light-nuclei production at NICA energies.

        Speaker: Marina Kozhevnikova (JINR)
      • 18:20
        New Geant4 internal cross section data base 15m

        \title{New Geant4 internal cross section data base}
        \author[a]{Berngardt Andrei}
        \affil[a]{The National Research Tomsk State University, Tomsk, Russia}

        In Geant4 there are two approaches for neutron transport simulations: working with high-precision (HP) cross-sections, which are derived from evaluated nuclear data, and working with XS, which are generated by processing HP data into a logarithmic energy grid, XS also uses data form TENDL database. The HP approach provides high precision but requires longer simulation times, whereas the XS approach runs much faster but has limited accuracy in the resonance region due to the smoothing of cross-section.

        To address this limitation, a modified method for generating neutron cross sections for the G4PARTICLEXSDATA database is introduced. The approach incorporates additional resonance (XSR) files with logarithmic energy grid and higher number of bins per decade. These files are generated for elements that have a resonance region, covering energies from 1 eV to the upper resonance boundary, and transition smoothly to standard XS data beyond this range. Element selection was based on accuracy analysis.

        Validation was performed using a detailed Geant4 model of the nELBE time-of-flight facility, evaluating neutron transmission through targets with atomic numbers from 3 to 92. Results were compared to the reference HP model (FTFP_BERT_HP) using mean and maximum absolute deviation metrics. Computational performance was assessed via CPU time and memory usage.

        Optimal number of bins per decade was determined for each element to minimize deviations. Cross-section averaging within XSR-files was also explored, yielding further accuracy improvements for several elements.

        The method has been implemented in the G4PARTICLEXS4.3 database and will be included in Geant4 version 11.5beta. Extensions include forced fission processes for QBBC and FTFP_BERT configurations and application to light ions (protons, deuterons, and alpha particles) and gammas. The approach preserves computational efficiency, with performance differences of approximately 1\%, while improving accuracy in the resonance region.

        Speaker: Андрей Бернгардт (National Research Tomsk State University)
    • 09:00 → 11:00
      Mathematical methods and tools for modeling complex physical systems Conference Hall (MLIT)

      Conference Hall

      MLIT

      • 09:00
        Theoretical modelling of resonance and locking in anomalous Josephson junctions 15m

        We presents theoretical modelling approaches to simulate the resonance and locking behaviour in Josephson junction coupled to ferromagnet under various biasing conditions. In particular, we investigate the influence of external radiation on the dynamical precession of the magnetization in the presence of the coupling with Josephson junction. Our results reveal the emergence of distinct magnetization dynamical states on synchronized steps of the current-voltage characteristics , which are controlled by both the junction parameters and the external field.

        The authors acknowledges the financial support of the Russian Science Foundation in the framework of project 22-71-10022.

        Speaker: Dr Majed Nashaat Abdel Ghani (Senior Researcher at BLTP (JINR))
      • 09:15
        Modeling of a Josephson junction with a second harmonic in the current-phase relation 15m

        In Josephson junctions with complex Cooper pair tunneling mechanisms associated with non-trivial symmetry of the superconducting order parameter or the barrier structure, a second harmonic appears in the current-phase relation. Under external radiation, the current-voltage characteristics (CVC) of such junctions exhibit a multitude of Shapiro step subharmonics in addition to integer harmonics.

        Determining the behavior of Shapiro steps as a function of model parameters, as well as identifying the positions of possible harmonics and subharmonics on the CVC, is a complex task that requires the development of specialized algorithms. Such algorithms must provide automated analysis of the Josephson junction's current-voltage characteristics, identification of the emerging Shapiro steps, and modeling of the step width dependence on model parameters.

        In this work, numerical modeling of the dynamics of a Josephson junction with a second harmonic in the current-phase relation under external radiation was performed. An algorithm for the search and identification of Shapiro steps, including their harmonics and subharmonics arising under external radiation, has been developed. The detection of steps is implemented through the analysis of the derivative of the averaged voltage with respect to current. Additionally, an algorithm for calculating the dependence of the step width and its harmonics on the model and radiation parameters has been implemented. The developed algorithms allow for investigating Shapiro step behavior under varying model parameters and analyzing experimental current-voltage characteristics.

        Speaker: Ирина Брусилова (Филиал Московского государственного университета имени М.В. Ломоносова в городе Дубне)
      • 09:30
        FRG study of the Aharony fixed point in dipolar magnets 15m

        I will discuss a functional renormalization group (FRG) analysis of the Aharony universality class, which describes isotropic systems with dominant dipole-dipole interactions. Within the LPA′ approximation, we compute the critical exponents and compare them with those of the Heisenberg O(3) universality class, showing that the two fixed points are distinct despite having numerically close exponent values. [Based on the paper by G. Kalagov and N. Lebedev, Nuclear Physics B 1027, 117474 (2026).]

        Speaker: Georgii Kalagov (BLTP JINR)
      • 09:45
        Field-Theoretic Approach to Helical MHD Turbulence and Large-Scale Dynamo Formati 15m

        We present a field-theoretic study of solenoidal helical magnetohydrodynamic turbu-
        lence in a statistically stationary regime. The analysis is based on the renormalization-
        group framework, which provides a systematic way to describe the influence of nonlinear
        interactions and fluctuations across scales. In helical media, the turbulent system de-
        velops infrared-relevant contributions that indicate an instability of the initial isotropic
        state. This instability may be interpreted as the onset of a large-scale dynamo mechanism
        and requires a consistent reformulation of the perturbative expansion around a new stable
        state.
        Within this approach, the generation of a homogeneous mean magnetic field is de-
        scribed as a transition to a regime with broken rotational symmetry. The resulting state
        modifies the structure of the propagating modes and leads to anisotropic corrections to the
        turbulent energy transfer. Particular attention is paid to the role of higher-order loop con-
        tributions, which reveal how helicity affects scaling behaviour and magnetic-energy spec-
        tra beyond the leading approximation. The results demonstrate that field-theoretic and
        renormalization-group methods provide an effective mathematical framework for analysing
        dynamo formation, stability, and spectral properties in turbulent magnetohydrodynamic
        systems.

        Speaker: Dr Andrei Ovsiannikov (Joint Institute for Nuclear Research (JINR))
      • 10:00
        Isotopic scaling as a probe of temperature in the heavy-ion fragmentation reaction 15m

        To predict the isotope distributions of the fragments produced in heavy-ion collisions different models can be used, they can be empirical, phenomenological or microscopic ones. In this report the results of calculations of isotopic scaling coefficients in the reactions with Ni and Ca projectiles at energies 140 A MeV in the modified transport-statistical approach are presented and compared with the experimental data and calculations in the EPAX, the FRACS and the Abrasion-Ablation models. It is shown that isoscaling coefficients not only give us the knowledge about the symmetry energy of the nuclear matter but also make it possible to derive the value of the temperature achieved in these processes, which is important to reliable prediction of isotope distributions produced in the collisions of heavy ions.

        Speaker: Татьяна Михайлова (LIT, JINR)
      • 10:15
        Methods of Non-integral Geometry for the Image Reconstruction Problem 15m

        Using the non-integral geometry methods, we study the different
        aspects of the image reconstruction problem.

        Speaker: Prof. Igor V. Anikin (Bogoliubov Lab of Theor Phys JINR)
      • 10:30
        On numerical singularities of computational scheme in the reconstruction problem 15m

        Within the joint project of MLIT, JINR BLTP and the Federal Biomedical Agency (FMBA), the software is being developed for the data restoration process after a computed tomography (CT) procedure with the help of new and improved mathematical tools for solving the reconstruction problem.

        Based on the new non-integral geometry, we consider the different aspects of computational scheme applied to the reconstruction problem. The main attention has been paid to the treatment of different types of numerical singularities.

        Speaker: Anastasia Anikina (MLIT, JINR)
      • 10:45
        Simulation of Photon Detector Based on a Long Josephson Junction 15m

        Interest in single photon detectors has increased significantly in recent years due to the development of quantum computing and quantum communications. However, the registration of photons in the microwave and infrared ranges requires fundamentally new approaches capable of ensuring high detection efficiency. Promising approach is the use of long Josephson structures, in which vortices known as fluxons can appear. This fluxons lead to the formation of branches on the current-voltage characteristic of a long Josephson junction, corresponding to states with a certain number of fluxons.
        In this work, we simulated a photon detector based on a long Josephson junction. The detected photon is represented as a current signal in the form of a wave packet. This wave packet causes oscillations in the spatiotemporal current distribution, which can lead to the formation or destruction of fluxon states. Thus, an incoming signal can induce switching between different fluxon states with different voltages on the I–V characteristic. We have demonstrated the possibility of controlled switching between such states. The dynamics of the considered long Josephson junction is described by a partial differential equation (the sine-Gordon equation). A software toolkit in Python was developed to simulate the detector based on the long Josephson junction. A parallel algorithm using the Numba library was also developed to study the influence of model parameters on switching between fluxon states.
        The work was supported by the Russian Science Foundation within the framework of project No. 22-71-10022.

        Speaker: Adiba Rahmonova (Joint Institute for Nuclear Research)
    • 09:00 → 11:00
      Mathematical methods in life sciences Room 406 (MLIT)

      Room 406

      MLIT

      • 09:00
        Stoichiometric constraints for reducing non-uniqueness in SAXS modeling of detergent-solubilized membrane proteins 15m

        Stoichiometric constraints for reducing non-uniqueness in SAXS modeling of detergent-solubilized membrane proteins

        Yury L. Ryzhykau$^{1,2}$ , Daria D. Kuklina$^{1}$, Ivan O. Bezruchko$^{1}$, Daniil V. Sidorov$^{1}$, Aleksei Yu. Shishkin$^{1}$, Siarhei M. Bukhalovich$^{1}$, Lev M. Vasilenko$^{1}$, Elizaveta A. Dronova$^{1}$, Anatolii E. Mikhailov$^{1}$, Yury S. Semenov$^{1}$, Oleksandr I. Ivankov$^{2}$, Alexander I. Kuklin$^{1,2}$

        $^{1}$Research Center for Molecular Mechanisms of Aging and Age-Related Diseases, Moscow Institute of Physics and Technology, MIPT, Dolgoprudny, 141701, Russia

        $^{2}$Frank Laboratory of Neutron Physics, Joint Institute for Nuclear Research, Dubna, 141980, Russia

        E-mail: rizhikov@phystech.edu, kuklina.dd@phystech.edu

        Membrane proteins are central to signal transduction, transport and bioenergetics, but their quaternary structure in solution is often difficult to determine by high-resolution methods alone. Crystallography and cryo-EM provide indispensable atomic detail, whereas the oli-gomeric organization relevant in detergent or membrane-mimetic environments may remain ambiguous. Small-angle X-ray scattering (SAXS) is therefore an important complementary approach, since it probes membrane-protein assemblies directly in solution and is sensitive to their oligomeric state and overall architecture. For example, cyanobacterial sensory rho-dopsin ASR forms dimers in crystals, whereas SAXS and NMR studies indicated predomi-nantly trimeric species in a lipid-bilayer environment [1]. Thus, for oligomeric microbial rhodopsins, solution scattering addresses not only a structural-biological question, but also a mathematical problem of model selection from indirect low-resolution data.

        The analysis of detergent-solubilized membrane proteins by SAXS represents an intrinsical-ly non-unique inverse problem. The measured scattering curve contains contributions from the protein, the detergent corona and, in many cases, a distribution of oligomeric states. Our previous study showed that these factors can strongly bias structural interpretation and may produce misleading solution models if the detergent belt and polydispersity are not treated explicitly [2]. Hybrid approaches such as MEMPROT improve the situation by representing the detergent corona with pseudo-atoms corresponding to hydrophobic and hydrophilic re-gions and fitting protein-detergent models against the SAXS curve [3,4]. However, the re-sulting optimization problem remains underdetermined: several detergent-corona geome-tries may describe the same scattering data with comparable fit quality, and even an incor-rect oligomeric hypothesis can sometimes be compensated by an alternative detergent-belt configuration.

        Here we consider stoichiometric information as an additional constraint that can regularize SAXS-based modeling of protein-detergent complexes. The key parameter is the detergent-to-protein ratio in the complex. It can be estimated by orthogonal biophysical methods, in-cluding UV/differential refractive index measurements or infrared spectroscopy, but these approaches depend on external calibration parameters such as extinction coefficients and dn/dc values. A complementary route is small-angle neutron scattering (SANS) with con-trast variation. By comparing the contrast-matching point of the whole protein-detergent complex with those of the isolated components, SANS provides access to their relative vol-ume fractions in scattering terms. This stoichiometric information can then be introduced as an a priori restriction in SAXS model construction or used in subsequent joint SAXS/SANS refinement.

        We are testing this approach on DDM-solubilized microbial rhodopsins. In particular, SANS contrast-variation experiments have been performed for KR2, a sodium-pumping rhodopsin whose functional form is pentameric and whose oligomerization is directly related to activi-ty [5]. The ongoing analysis is aimed at evaluating whether SANS-derived stoichiometric constraints reduce the effective parameter space of detergent-corona models and improve the uniqueness and physical plausibility of SAXS-based structural interpretation. In this formulation, the problem is treated as a constrained modeling task in which independent scattering-derived stoichiometry is used to suppress mathematically admissible but physi-cally implausible solutions.

        Yu.R. Ryzhykau and D.D. Kuklina acknowledges the Ministry of Science and Higher Education of the Russian Federation (agreement 075-03-2026-305, project FSMG-2025-0003).

        References
        [1] S. Milikisiyants , S. Wang , R. A. Munro , M. Donohue, M. E. Ward, D. Bolton , L. S. Brown , T. I. Smirnova , V. Ladizhansky , A. I. Smirnov, Oligomeric structure of Anabaena sensory rhodopsin in a lipid bilayer environment by combining solid-state NMR and long-range DEER constraints, J. Mol. Biol. 429 (2017) 1903–1920. https://doi.org/10.1016/j.jmb.2017.05.005.
        [2] Y.L. Ryzhykau, A. V. Vlasov, P.S. Orekhov, M.I. Rulev, A. V. Rogachev, A.D. Vlasova, A.S. Kazantsev, D.P. Verteletskiy, V. V. Skoi, M.E. Brennich, P. Pernot, T.N. Murugova, V.I. Gordeliy, A.I. Kuklin, Ambiguities in and completeness of SAS data analysis of membrane proteins: the case of the sensory rhodopsin II–transducer complex, Acta Crystallogr. D Struct. Biol. 77 (2021) 1386–1400. https://doi.org/10.1107/s2059798321009542.
        [3] J. Pérez, A. Koutsioubas, Memprot: a program to model the detergent corona around a membrane protein based on SEC–SAXS data, Acta Crystallographica Section D 71 (2015) 86–93. https://doi.org/10.1107/S1399004714016678.
        [4] F. De Pol, M. Baranowski, C. Neumann, S. Trampari, P. Nissen, J. Pérez, Simulation and modelling of the detergent corona around the membrane protein MhsT based on SAXS data, J. Appl. Crystallogr. 57 (2024) 1415–1425. https://doi.org/10.1107/S1600576724006721.
        [5] K. Kovalev, V. Polovinkin, I. Gushchin, A. Alekseev, V. Shevchenko, V. Borshchevskiy, R. Astashkin, T. Balandin, D. Bratanov, S. Vaganova, A. Popov, V. Chupin, G. Büldt, E. Bamberg, V. Gordeliy, Structure and mechanisms of sodium-pumping KR2 rhodopsin, Sci. Adv. 5 (2019) eaav2671. https://doi.org/10.1126/sciadv.aav2671.

        Speakers: Daria Kuklina (Laboratory for Advanced studies of membrane proteins, Research Center for molecular mechanisms of aging and aging-related diseases, Moscow Institute of Physics and Technology (National Research University)), Yury Ryzhykau (Moscow Institute of Physics and Technology)
      • 09:15
        Development of the "MOSTLIT" Web Service for Radiation-Induced Foci Analysis 15m

        This report presents the performance results and key modification stages of the "MOSTLIT" web service, which utilizes deep learning and computer vision algorithms for the automated analysis of radiation-induced foci. The service was validated using a dataset of fluorescence microscopy images of human dermal fibroblasts exposed to gamma irradiation (1.25 Gy), with a dynamic analysis of the cellular response from 30 minutes to 24 hours post-exposure. Particular focus is placed on a comparative study conducted by independent expert groups from the Laboratory of Radiation Biology (JINR) and the Burnasyan Federal Medical Biophysical Center (FMBA). This comparison enabled an assessment of the reproducibility, robustness, and accuracy of the automated image processing implemented in the service. The findings confirm the effectiveness of the developed approach and demonstrate the platform's potential for inter-laboratory research and the standardization of radiation-induced damage analysis.
        This work was carried out within the framework of the state assignment of Ministry of Science and Higher Education of the Russian Federation (theme No. 124112200072-2).

        Speaker: Татевик Бежанян (LIT JINR)
      • 09:30
        Investigation of the radiation-induced effects on melanoma growth using the spatially distributed phenomenological mathematical model with immune control 15m

        A spatially distributed phenomenological model of the tumor growth is developed. This model takes into account the cell proliferation, spatial distribution of the cells and their interaction with the immune system. The model has two stable stationary states corresponding to complete tumor elimination and uncontrolled growth, which allows to describe both the control of disease and its progression.

        For quantitative comparison of the model with experimental data on the tumor volume growth, an approach is proposed for restoring biologically interpreted parameters: the effective proliferation rate, the immune control threshold and the invasion coefficient, as well as the parameters of the initial condition. The approach includes several stages: an extraction of initial approximations for the parameters from the two-phase kinetics of the tumor volume growth curve (exponential phase turns into power-law phase), numerical optimization of the parameters using an approximate solution in the form of a moving front (which reduces the initial partial differential equation to a system of ordinary differential equations) and subsequent verification of the values obtained by the direct numerical solution of the initial equation.

        The method is applied to the description of data on the growth of B16 melanoma in vivo in the control group and in the groups after the proton irradiation and combined exposure of protons with the AraC radiosensitizer. It was shown that the irradiation leads to a decrease of the proliferative potential and to an increase of the threshold of the immune control; the AraC addition further suppresses growth and alters the immune response. The proposed approach makes it possible to quantify the radiation-induced changes of the key parameters of the tumor growth dynamics.

        Speaker: Dr Nikolai Ustinov (Лаборатория радиационной биологии)
      • 09:45
        Unsupervised Behavioral Clustering of Rodent Trajectories in the Morris Water Maze 15m

        One of the primary tools for studying central nervous system (CNS) disorders in mammals susceptible to various factors and diseases is the Morris Water Maze behavioral test. Qualitative analysis of movement trajectories allows us to distinguish memory deficits from motor, motivational, or emotional (anxiety) disorders, and also identifies the underlying cognitive process. The development of algorithms and methods for analyzing platform search strategies in the Morris Water Maze is critical for neurobiology, neuropharmacology, and translational medicine.
        In this work, we propose an unsupervised framework for clustering behavioral patterns directly from the rodent trajectories which were constructed by analyzing video data using computer vision methods.
        In the image-based embeddings obtained by a neural network model for clustering the trajectory images, the temporal and behavioral structure are ignored. Here, as an alternative, we extract a set of interpretable, platform-aware features from the trajectories that capture spatial preference, navigation efficiency, and directional behavior. These include measures such as directness, radial distribution, wall preference, platform proximity, sector occupancy, and latency-based descriptors. The resulting feature representations are used for clustering via density-based models, enabling automatic identification of behavioral groups without prior labeling. We show that incorporating target-oriented and temporal features significantly improves cluster separability compared to traditional geometric descriptors. We further evaluate clustering quality by comparison with known strategy labels.
        This work highlights the importance of domain-specific feature design for behavioral trajectory analysis and offers a practical tool for unsupervised behavioral clustering.
        The study was carried out within the framework of the state assignment of Ministry of Science and Higher Education of the Russian Federation (theme No. 124112200072-2).

        Speaker: Сара Шадмехри
      • 10:00
        Computer Vision (CV) Algorithms and Deep Learning Models for Radiobiological Research Experimental Data Analysis Using the «Open Field» Behavioral Test System 15m

        he work is devoted to the application of computer vision (CV) algorithms and deep learning models for the automatic analysis of radiobiological experimental data obtained through using the «Open Field» behavioral test system.

        A comparative analysis of classical CV algorithms and the YOLOv11 architecture was conducted to solve the frame-by-frame object detection problem (tracking problem). The resulting movement trajectory was linked to the marked arena of the «Open Field» behavioral test system to create a heat map of the total time animals spent in different sectors of the laboratory setup. Time intervals were identified in the video data for analyzing behavior patterns, and a framework was developed to implement and demonstrate this functionality.

        Represented approach solves the detection problem with a high degree of accuracy, which makes it possible to create a comprehensive tracking system that determines the current sector-based location of the animal and time intervals that set the boundaries for observing different behavior patterns.

        Literature.

        1. Redmon J., Divvala S., Girshick R., Farhadi A. You Only Look Once: Unified, Real-Time Object Detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 779–788.
        Speaker: Alexey Chubov (Moscow State University branch in Dubna)
    • 09:00 → 11:00
      Methods of quantum computing and quantum information processing Room 310 (MLIT)

      Room 310

      MLIT

      • 09:00
        Dissipative dynamics of multiple-quantum NMR coherences in an ensemble of hundreds of equivalent spin-1/2 nuclei 15m

        Delocalization of quantum information is processed in quantum circuits, or more generally, quantum channels over the entire system is known as scrambling. Scrambling implies that information about the input cannot be deduced by any local measurement of the output. The scrambling rate can be quantified through out-of-time-ordered correlators (OTOCs) [1]. А specific family of OTOCs, first developed in NMR under the name of the multiple-quantum coherence (MQC) spectra, allow us to quantify scrambling in many-body systems experimentally [2]. In fact, the generic decay of such correlators of is a distinguishing feature of quantum chaos. In this work, we investigate the behavior of such a correlator under dissipative dynamics.
        We consider multiple-quantum (MQ) dynamics of a gas composed of spin-carrying molecules with s = 1/2 under nuclear magnetic resonance (NMR) conditions. Initially, the gas confined in nanopores is thermalized at low temperature. The dynamics of the spin system is described by the Lindblad master equation, which accounts for spin dephasing induced by interactions with the pore walls. We present the distribution of MQ NMR coherence intensities at high and low temperatures for various decay rates in a system of 101 spins. For a three-spin system, we provide an analytical solution. Based on the second moment of the MQ NMR coherences, we examine the dependence of the lower bound on the number of entangled spins on temperature and decay rate.
        This work was supported by the Federal Research Center of Problems of Chemical Physics and Medicinal Chemistry, Russian Academy of Sciences, within the framework of the state assignment (registration number 124013000760-0).

        [1] P. Hosur, X.-L. Qi, D. A. Roberts, and B. Yoshida, J. High Energy Phys. 2016, 1 (2016)
        [2] Martin Gärttner, Philipp Hauke, and Ana Maria Rey, Phys. Rev. Lett. 120, 040402 (2018)

        Speaker: Ilia Lazarev (RAS Federal Research Center of Problems of Chemical Physics and Medicinal Chemistry)
      • 09:15
        On Implementation of Variational Quantum algorithm on the HybriLIT Quantum Testbed applied to the Ising Model with a Transverse Field 15m

        We simulate the operation of a noiseless quantum computer using the Cirq library and the cuStateVec package. Average energy and magnetization values are calculated using sampling for various transverse magnetic field values. A distinctive feature of our algorithm is the use of an entangling 6-parameter two-qubit gate to generate the state vector ansatz. Two-dimensional and three-dimensional spin lattices with varying numbers of sites are considered. The cuStateVec package is optimized for computing quantum circuits using graphics cards and yields results for lattices with approximately 30 sites.

        Speaker: Ms Ksenia Verkhovtseva
      • 09:30
        QTensorNet: A High-Performance Tensor-Network-Based Toolkit for Many-Body Quantum Simulations 15m

        As of 2026, quantum computing has reached a stage where it is approaching "quantum supremacy" for specific practical tasks. However, direct quantum computations are currently limited by technological issues, including decoherence and high noise levels. These issues strictly constrain the architecture of modern quantum computers. Consequently, efficient classical simulation of many-body quantum systems has become crucial for the development, validation, and benchmarking of quantum algorithms and hardware. The primary challenge in classical simulation is the so-called "Simulation Bottleneck": the dimension of the Hilbert space for an $N$-particle system grows exponentially as $d^N$. This scaling makes exact diagonalization and unitary dynamics simulation computationally infeasible for realistic system sizes ($N > 50$). To circumvent this, Tensor Networks (TNs) have emerged as a powerful framework. By exploiting the area law for entanglement entropy and the locality of interactions, TN methods reduce the simulation complexity from exponential to polynomial scaling ($O(N^m)$). This enables the efficient approximation of low-energy states in systems with local interactions and the simulation of quantum algorithms with limited entanglement depth.

        We introduce QTensorNet, a high-performance toolkit based on the NVIDIA cuQuantum SDK, specifically designed for many-body problem simulations using the tensor network approach. QTensorNet provides a flexible environment that supports any single-hierarchy-level tensor network architecture. To handle the computational demands of large-scale contractions, the toolkit utilizes one or more NVIDIA GPUs via MPI. The current version of QTensorNet includes implementations of Time-Evolving Block Decimation (TEBD), Density Matrix Renormalization Group (DMRG), and Time-Dependent Variational Principle (TDVP) methods. We demonstrate the performance and scalability of QTensorNet using a computationally challenging system: a 112-spin network on a square-kagome lattice with anisotropic Heisenberg interaction, evaluating its ground-state properties.

        Speaker: Dr Yauheni Talochka (MLIT JINR)
      • 09:45
        Supercomputer simulation of quantum walks to solve complex computational problems 15m

        The development of new quantum algorithms is currently attracting considerable attention due to the significant progress achieved in the practical application of quantum computing. In this paper, we develop a software complex for simulating quantum walks on graphs of various topologies, designed for use on high-performance computing systems.

        The quantum walk model on graphs is a universal computational model equivalent to well-known approaches based on quantum circuits and adiabatic computation. Similar to classical random walks, a quantum system in this model obeys a specific law of state evolution over time, with the set of states identified with the set of vertices of a graph, finite or infinite.

        Quantum walks exhibit properties that sharply distinguish it from a similar classical problem. In particular, due to the superposition of states, the time to reach a specific graph vertex can be significantly shorter than in the case of a classical particle's random walk. This ensures a polynomial speedup of quantum algorithms compared to classical ones in such practically important problems as searching on directed and undirected graphs, finding a triangle in a graph, optimization problems, and many others.

        This paper considers particle motion along a line, a cycle, and a hypercube. The quantum walk algorithm is implemented using parallel computing within the MPI standard. The libraries used were mpi4py, NumPy for high-performance operations on complex arrays, Matplotlib for data visualization, as well as standard Python libraries for working with the file system, saving results in CSV format, and measuring execution time.

        After initializing the MPI environment, each process receives an identifier (rank) and the total number of processes (size). If the platform supports it, the program automatically uses extended precision (complex256, float128) to reduce error accumulation; otherwise, it uses standard double precision (complex128, float64). The graph is then partitioned into tasks assigned to individual processes, which reduces memory usage and improves scalability.

        Two arrays of complex amplitudes corresponding to the internal states of a quantum particle are used for the simulation. The initial state is localized at the selected vertex. The MPI.Barrier function then ensures the simultaneous start of computations. At each step of the main loop, the following are performed: exchanging boundary values between adjacent processes via the non-blocking MPI.Sendrecv function; applying the internal state change operator; and applying a spatial shift. The calculated probabilities are stored in an array to obtain a probability distribution. To compensate for the accumulation of computational error over a specified number of steps, global renormalization is performed: the state norm is calculated using the collective MPI.Allreduce operation, and then the amplitudes are scaled.

        Upon completion, local distributions are collected on the master process (rank = 0) using the MPI.Gatherv operation. The probabilities for all nodes and the analytical distribution are saved to a CSV file; a log of the simulation parameters and execution time is also maintained. Plots are automatically generated for visual analysis.

        The program is built on a modular principle: separate modules for parallel data distribution, quantum evolution, renormalization, results storage, and visualization are implemented. This modular architecture allows for expansion of the system without changing the computing core—implementing other types of internal states, simulating random walks on arbitrary graphs, adding GPU acceleration, and porting the program to hybrid high-performance systems.

        The quantum walk model has previously been successfully applied in a number of works in physical chemistry, yielding results in good agreement with experimental data. In particular, the model makes it possible to study chemical reactions proceeding via a radical mechanism or to predict the outcome by analyzing a graph constructed from individual stages of a physicochemical process. Separately worth noting is the ability to analyze chemical reactions based on the potential energy surface landscape, where particles move from one minimum to another, following the relief of the system's potential energy.

        As an area of application of the mathematical modeling results, one should note such a problem as modeling diffusion processes on the surface and in the bulk of a solid phase. Traditional modeling using classical molecular dynamics methods requires significant computational resources and time in this case. Another serious limitation is interatomic potentials, the construction of which, even when using modern machine learning tools, remains a very labor-intensive process and significantly hinders research. In the case of quantum walks, one can, for example, rely on results obtained from scanning potential energy surfaces, which provide a comprehensive picture of the process in a local region limited to several tens or hundreds of atoms. For simple adsorbates, such as hydrogen, nitrogen, or oxygen atoms, this will significantly accelerate the modeling of diffusion processes, which are an important stage of catalytic reactions in industrial chemical synthesis and also play a significant role in the development of materials for future energy applications.

        The results of the numerical experiments demonstrate a significant speedup compared to direct computations in single-processor mode. This work continues a series of studies of physical and physicochemical processes conducted using supercomputer modeling at the Voronezh State University Supercomputer Center.

        Speaker: Константин Петрищев
      • 10:00
        Propagation of Two-Photon Zernike States in Atmospheric Turbulence 15m

        We analyze propagation and detection of two-photon states expanded in Zernike modes through atmospheric turbulence using the extended Huygens–Fresnel formalism. For SPDC states prepared with a single Zernike pump mode, we analytically reduce the 8-dimensional continuous propagation integrals to an exact, discrete modal expansion. In the absence of turbulence, Zernike addition enforces conservation of azimuthal index and a strict radial-order bound. Turbulence relaxes these constraints, driving structured azimuthal and radial crosstalk dominated by low-order aberration modes. By explicitly removing the lowest-order terms from the discrete turbulence sum, we demonstrate that partial adaptive optics correcting only up to the sixth radial order is sufficient to heavily suppress this crosstalk and restore near-ideal spatial correlations.

        Speaker: Vahagn Abgaryan (JINR MLIT)
      • 10:15
        Implementing 6D Parameterized 2-Qubit Gates on Bauman Octillion Platform 15m

        In this talk, we present an overview of our results on implementing a 6-parameter family of entanglers on the Snowdrop 4Q and Snowdrop 8Q coprocessors developed by Bauman Moscow State Technical University and the All-Russian Research Institute of Automation. We will introduce quantum parameterized circuits that enable testing the entangling capabilities of these gates and discuss the optimization of circuits capable of generating a generic entangled mixed state of two qubits. This gate is used to construct the ansatz in the variational quantum algorithm and is applied to find the minimum energy in the Ising model with external magnetic fields.

        Speaker: Dr Alla Bogolubskaya (MLIT JINR)
    • 09:00 → 11:00
      Modern methods for data processing and analysis in Mega-science projects in HEP Room 420 (MLIT)

      Room 420

      MLIT

      • 09:00
        A Reinterpretation of the LHC Results on the Search for Dark Matter in Various Models with an Extended Higgs Sector 15m

        Title: A Reinterpretation of the LHC Results on the Search for Dark Matter in Various Models with an Extended Higgs Sector

        Abstract
        This work presents an overview of searches for dark matter particles performed by the CMS experiment in models with an extended Higgs sector. The analyses consider events with missing transverse momentum (MET) produced in association with a Higgs boson, as well as events containing a displaced nonresonant muon pair and MET in the final state. The studies are based on proton-proton collision data collected at a center-of-mass energy of 13 TeV, corresponding to integrated luminosities of 35.9 and 138 fb⁻¹.
        The analysis methods are discusse for result interpretation in the framework of simplified 2HDM+a and Z′ mediator models. In addition, a dedicated search for inelastic dark matter is presented using events with a displaced nonresonant muon pair and MET. The interpretation is performed in the context of an inelastic DM (IDM) scenario involving a dark photon A′ and nearly mass-degenerate dark sector states χ₁ and χ₂.

        Speaker: Юрий Корсаков
      • 09:15
        Capabilities of measuring electron drift velocity for MPD TPC 15m

        This presentation are about Laser Calibration System of MPD TPC detector and electron drift velocity reconstruction. The calibration system general construction with current status are given with velocity calculation method. Possible future challenges associated with the detector assembly are also mentioned.

        Speaker: Alexander Bychkov (LHEP)
      • 09:30
        Distributed data Analysis in High Energy Physics: integration of ROOT RDataFrame and Apache Spark 15m

        Modern experiments generate data streams with peak intensity up to 1 PB/s, which requires the creation of new efficient solutions for processing big data in high-energy physics (HEP). Recent papers [1-3] present the results of extending the functionality of the high-level RDataFrame ROOT interface, including improving user functions and reducing memory consumption. Unified methods have been developed for including C++ code both into the local RDataFrame application and the distributed one. Continuing this approach, the present paper describes the architecture of a hybrid data analysis solution in HEP, combining RDataFrame and the Apache Spark framework, a recognized standard for distributed big data processing (Big Data). RDataFrame provides a modern approach to scientific data analysis that describes the logic of computation without clearly specifying the sequence of instructions, optimization of the operation graph, and thread-level parallelism, automatically implemented by the compiler. Apache Spark [4] complements this ecosystem with large-scale distributed processing capabilities: high computing speed, automatic load distribution between cluster nodes, fault tolerance and scalability. Python serves as the connecting link between RDataFrame and Spark, allowing HEP researchers to retain the familiar syntax of data analysis in the ROOT environment, delegating the tasks of scaling, load balancing and resiliency to the Big Data ecosystem.
        The possibility of joint operation of the ROOT RDataFrame and the HDFS distributed file system was investigated. Data integrity has been shown to be preserved during parallel processing. The calculations performed confirmed the linear scalability of the proposed architecture with an increase in the number of computing nodes.Bechmark results demonstrate a significant reduction in data processing and analysis time compared to traditional approaches. Thus, the Spark + RDataFrame combination can be considered as a promising tool for solving computational problems in modern and future experiments in high-energy physics.

        Literature
        1) Padulano V.E. and others. Analysis of distributed data using ROOT RDataFrame. Web Conference EPJ 245, 03009 (2020).
        2) Tededchi T. et al. Prototyping a distributed ROOT-based analysis stream for HL-LHC: an example of using a CMS. Archive: 2307.12579v1. (cs.DC) July 24, 2023
        3) Padulano V.E. and others. Distributed analysis in production using RDataFrame. EPJ 337 Web Conference, 01007 (2025)
        4) Apache Spark Software Site: https://spark.apache.org

        Speaker: Татьяна Соловьева (Jinr)
      • 09:45
        Geometric Alignment Method for GEM Detectors of the BM@N Experiment 15m

        This paper considers a method for the geometric alignment of detector systems, using the GEM detectors of the BM@N experiment at the NICA accelerator complex as an example. High-precision spatial reconstruction of charged-particle tracks in multilayer tracking systems requires accurate determination of the relative positions of detector modules. Even small displacements, angular misalignments, or deformations thereof can lead to systematic errors in track coordinate reconstruction and degradation of the system’s spatial resolution. The proposed alignment method is based on the analysis of residuals between the measured hit coordinates and the parameters of reconstructed tracks. The study uses experimental event data containing charged particles passing through the GEM detector system. To determine the displacement parameters, iterative minimization procedures of the residual functional are applied, allowing for the estimation of translational and angular deviations of individual detector planes. The possibility of applying global alignment algorithms is considered, including methods based on χ² minimization and matrix approaches such as Millepede. It is shown that the application of the geometric alignment procedure reduces systematic deviations in the measured coordinates and improves the quality of trajectory reconstruction. The performed analysis confirms the robustness of the proposed approach with respect to statistical noise and demonstrates its effectiveness for GEM detectors. The developed method can be used to create of software tools for calibration and monitoring of the geometric stability of detector systems in high-energy physics experiments.

        Speaker: Зариф Шарипов (JINR)
      • 10:00
        A method for multimodal analysis of TAIGA experiment data using essential features extracted by autoencoders 15m

        The TAIGA experiment maintains several experimental facilities, including several Atmospheric Cherenkov Telescopes (IACT) and the HiSCORE Cherenkov Array. The Cherenkov telescopes record the spatial distribution of Cherenkov light from secondary particles from extensive air showers (EAS) as images. The HiSCORE facility records the spatiotemporal distribution of Cherenkov light from EAS on the Earth's surface. Currently, the analysis of these multimodal data (MMD) is carried out independently. Therefore, joint analysis of MMD from the TAIGA facilities is an actual problem in cosmic ray physics and gamma-ray astronomy. We present a method for MMD analysis using deep neural networks. The central idea of ​​the proposed approach is to combine data using essential features extracted using autoencoders. Using Monte Carlo simulation data, it is demonstrated that the proposed method allows for efficient MMD analysis. This approach can be used for joint analysis of MMD from other complex experimental facilities.

        Speaker: Alexander Kryukov (SINP MSU)
      • 10:15
        Evaluation of Multi Deep Neural Network Architectures for Neutron Spectrum Unfolding Approaches from Bonner Sphere Systems 15m

        Neutron spectrum unfolding from Bonner sphere spectrometer (BSS) readings is an ill-posed inverse problem traditionally solved through iterative algorithms. This work presents and benchmarks five deep neural network architectures: a simple baseline multilayer perceptron, a deep feed-forward network, a residual network with skip connections, a self-attention-based network, and a wide & deep architecture for direct spectrum reconstruction from BSS detector responses. Models were trained on a synthetic dataset of 500,000 simulated neutron spectra, with input features comprising the detector reading vector (Q values) and output targets representing the full neutron influence spectrum across energy bins. Each model was tested on a separate external test set of validation experimentally relevant spectra for the performance evaluations. The main evaluation metrics are r-squared (R²), mean absolute error (MAE), root mean squared error (RMSE), and per-bin correlation. With excellent per-bin correlation across all energy groups and r-squared greater than 0.93 on the external test set, the residual and attention-based architectures showed better generalization. These findings assure that data-driven unfolding techniques can offer quick, dependable substitutes for algorithms like landweber, maximum likelihood expectation maximization (MLEM) or those based on tikhonov regularization. Machine learning results can also be used as an initial sample for these algorithms, providing additional information about the shape of the true spectrum.

        Speaker: Dr Mohamed Gaber (Meshcheryakov Laboratory of Information Technologies, JINR)
    • 11:00 → 11:30
      Coffee break
    • 11:30 → 12:30
      Closing Conference Hall (MLIT)

      Conference Hall

      MLIT