|
PANM 23
|
Jako tradičně bude po semináři vydán recenzovaný sborník, kam mohou účastníci semináře zaslat své příspěvky.
Pro dopravu do Hejnic doporučujeme vlak z Liberce; doba jízdy je přibližně půl hodiny a spoje odjíždí každou hodinu. Hejnický klášter se nachází cca 500 metrů od vlakového nádraží. Budete-li potřebovat odvoz od vlaku, kontaktujte nás, prosím.
Pro cestující s vlastním autem je v blízkosti hotelu dostatek stání.
The development of blood contacting devices, such as Ventricular Assist Devices (VAD) became one of the most challenging tasks in modern bioengineering. The models of blood rheology, stress- and strain- based models of hemolysis, models of turbulence need to be solved at high resolution in complex geometries. The talk will present an overview of our current work in design of computational test case for VAD development and some of the recent results on combining the models of viscoelastic rheology with strain based models of hemolysis. This is a part of an ongoing joint work with Anna Lancmanova, Nico Dirkes, Alberto Girelli, Adelia Sequeira and Giulia Giantesio.
V současné době době existuje téměř nepřeberné množství numerických technik pro řešení parciálních diferenciálních rovnic hyperbolického typu a dalších příbuzných problémů (od metody konečných diferencí, metody konečných objemů, přes různé verze metody konečných prvků až po spektrální metody). Jednou z občas diskutovaných otázek je, zda je vhodnější aproximovat řešení spojitými, nebo nespojitými funkcemi. V rámci přehledové přednášky se pokusíme popsat vybrané příklady metod (např. metody Godunovova typu, schémata typu active flux) založených na různých typech aproximace hledané funkce, shrnout výhody, nevýhody a omezení těchto přístupů. Dále popíšeme dopad různých typů aproximace řešení na další části metod. Zmíníme také některé otevřené problémy.
Many real-world optimization problems involve multiple conflicting objectives and rely on numerical simulations, where the high computational cost limits the number of feasible evaluations. In such settings, the direct application of conventional numerical optimization is often computationally prohibitive. This contribution focuses on data-driven approaches designed to address such challenges, with an emphasis on surrogate-assisted strategies that replace the costly objective functions with learned approximations. In particular, we consider probabilistic surrogate models within the Bayesian optimization framework that utilize uncertainty information to guide the optimization process. Further aspects, including probabilistic constraint handling, mixed-integer design spaces, and multi-fidelity modelling, are also discussed to address practical challenges in real-world multi-objective optimization.
The lecture will focus on the mesoscopic lattice Boltzmann method (LBM) and will consist of three parts. First, the basic principles of this method will be explained and illustrated with engaging images and animations. Then, several applications of LBM for solving three-dimensional mathematical modeling tasks will be presented. Finally, we explore what macroscopic partial differential equations we are actually solving using LBM.
Fast Fourier transform (FFT)-based solvers have become a standard tool for homogenization problems in micromechanics [1]. These solvers are particularly effective for elliptic partial differential equations (PDEs) with periodic boundary conditions and data defined on regular grids.
A key component of this class of solvers is the discrete Green's operator preconditioner, applied via the FFT, which substantially accelerates convergence of iterative methods---in our case, the conjugate gradient method (CG).
Inspired by Carson [2], we will examine the convergence of CG-accelerated FFT-based solvers across a series of homogenization problems. We will establish a direct connection between PDE data, spectral properties, and solver performance, providing deeper insight into the mechanisms that govern efficiency and accuracy of these solvers.
References:
[1] M. Schneider, A review of nonlinear FFT-based computational homogenization methods, Acta Mech. 232 (6) (2021) 2051--2100, doi:10.1007/s00707-021-02962-1.
[2] Carson, Erin, Jörg Liesen, and Zdeněk Strakoš. "Towards understanding CG and GMRES through examples." Linear Algebra and its Applications, vol. 692, 2024, pp. 241--291. Elsevier. https://doi.org/10.1016/j.laa.2024.04.003
Fluid flow and rock mechanics are key processes governing the safety assessment of deep geological repositories. Their strong coupling, together with complex fracture networks, poses significant challenges for mathematical modelling and numerical simulation. This lecture discusses key issues in hydro-mechanical modelling of fractured media, including conceptual representation of fractures across multiple scales, formulation and finite-element approximation of coupled nonlinear hydro-mechanical models accounting for fracture-matrix interaction, and efficient numerical solution strategies. Particular attention is paid to monolithic and iterative schemes, solution of contact problems, and parallel domain decomposition methods for large-scale simulations.
References:
I. Berre, F. Doster, E. Keilegavlen. Flow in fractured porous media: A review of conceptual models and discretization approaches. Transport in Porous Media, 130, 215-236 (2019).
J. Stebel J. Kružík, D. Horák, J. Březina, M. Béreš. On the parallel solution of hydro-mechanical problems with fracture networks and contact conditions. Computers & Structures, 298, 107339 (2024).
This lecture deals with the solution of slope stability problems in 3D using the finite element method (FEM) and incremental techniques, including shear strength reduction (SSR) or limit load (LL) methods. The aim is to reliably determine the factor of safety (FoS) of the slope and detect possible slip zones. We build on the Mohr-Coulomb perfectly plastic model, Davis modifications of non-associated plastic flow, and recent mathematical results that connect the determination of FoS with optimization problems. To make the SSR and LL methods more accessible to a mathematical audience, they are also presented in algebraic form and completed with appropriate assumptions enabling numerical analysis. Then, the proposed numerical methods are presented in detail. They are a combination of continuation techniques, Newton-type methods, and deflated Krylov methods with preconditioning. Furthermore, we propose a methodology reducing the overestimation of FoS caused by spatial discretization. It is based on a combination of the continuation techniques with FEM mesh adaptivity. The proposed algorithms are implemented within publicly available Matlab codes and used to investigate the stability of slopes in 3D. Numerical examples also show that the use of deflated Krylov methods leads to a significant acceleration of the calculation. The commercial software Comsol Multiphysics is used to compare the numerical results. The presented results are a joint effort with Michal Béreš, Simona Bérešová, Tomáš Luber, Zdeněk Michalec, Jaroslav Haslinger and Jakub Kružík.
References:
[1] S. Sysala, M. Béreš, S. Bérešová, T. Luber, Z. Michalec: Advanced continuation and iterative methods for slope stability analysis in 3D. Computers & Structures 315, 2025, 107842.
[2] S. Sysala, M. Béreš, S. Bérešová, J. Haslinger, J. Kružík, T. Luber: Convex optimization problems inspired by geotechnical stability analysis. SIAM Journal on Optimization 35(3), 1993-2016, 2025.
Solving linear least-squares (LS) problems and systems of linear algebraic equations (LE) belong to basic tasks of numerical linear algebra. Their solvability is a necessary condition to obtain efficient solution of many higher level tasks of computational science and engineering. Over the years, the size of the problems that have to be solved has constantly increased. But, fortunately, many applications offer them not as tasks formulated in a black-box form. Instead, the LE and LS problems often come with a structure visible either as a limited structural density (sparsity; a substantial part of entries in involved matrices are zeros), data sparsity (existence of low-rank subproblems). Moreover, even algorithmic tools to get the solution may involve well-understood components that involve in a smaller or larger part approximative procedures.
The talk plans to point out basic ways that could be used to solve LS problems, primarily based on considering sparsity to tackle large problems.
| 16. 3. 2026 | registrace účastníků, včetně krátkého abstraktu |
| nejpozději 31. 3. 2026 | organizátoři potvrdí přijetí příspěvku a zašlou údaje pro platbu |
| 17. 4. 2026 | platba účastnického poplatku |
| 14.-19. 6. 2026 | PANM 23 |
| 30. 9. 2026 | zaslání příspěvku do sborníku na mail panm@math.cas.cz |
Účastníci mohou zaslat příspěvky v anglickém jazyce do tradičního recenzovaného sborníku (ukázka PANM 22), který bude vydán elektronicky prostřednictvím České digitální matematické knihovny (články budou volně přístupné, tzv. open access) a rovněž v tištěné podobě. Termín pro podání článků je 30. 9. 2026. U sborníku předpokládáme zařazení do databáze Conference Proceedings Citation Index společnosti Clarivate Analytics, stejně jako u předchozích sborníků PANM.
Pro přípravu příspěvku v systému LaTeX budete potřebovat následující soubory (všechny soubory v jednom archivu lze stáhnout zde):
Organizační výbor:
J. Chleboun, J. Papež, M. Rozložník, K. Segeth, J. Šístek, T. Vejchodský
Sekretariát konference: H. Bílková, J. Papež, J. Šístek
|
|