BEGIN:VCALENDAR
PRODID:calendar.unige.ch
VERSION:2.0
CALSCALE:GREGORIAN
X-WR-TIMEZONE:Europe/Zurich
BEGIN:VEVENT
DTSTART:20230830T080000Z
DTEND:20230830T080000Z
DTSTAMP:20240804T175332Z
UID:37219@calendar.unige.ch
CREATED:20230823T100049Z
LAST-MODIFIED:20230823T100049Z
SUMMARY:Machine Learning and Science: It's All Fun and Games Until CUDA Crashes (Constructor Technology)
DESCRIPTION:Recently we published a paper where we developed a machine learning algorithm for predicting properties of 2D materials. The core idea of the paper was invented and tested in about a week. All the remaining 1.5 years went into writing the code to generate the data, writing the code to process it, writing the code to implement baselines, writing the code to do hyperparameter search (which didn’t change the conclusions in the slightest, we still won by a large margin). And then running, debugging and re-running on several different HPC platforms.\n\nWe'll discuss the technical challenges of conducting computational research labour-efficiently, and the ways to address them, followed by the demonstration of our project on the Constructor Research Platform.
LOCATION:Room 1-15, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37219
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20230919T120000Z
DTEND:20230919T120000Z
DTSTAMP:20240804T175332Z
UID:37163@calendar.unige.ch
CREATED:20230816T081013Z
LAST-MODIFIED:20230908T150831Z
SUMMARY:The solution path of the Sorted L One Penalized Estimation (SLOPE) (Patrick Tardivel, Université de Dijon)
DESCRIPTION:The SLOPE estimator has the particularity of having null components (sparsity) and components that are equal in absolute value (clustering).\nThe number of clusters depends on the regularization parameter of the estimator. This parameter can be chosen as a trade-off between interpretability (with a small number of clusters) and accuracy (with a small mean squared error or a small prediction error). Finding such a compromise requires to compute the solution path, that is the function mapping the regularization parameter to the estimator.\nDuring this presentation I will provide some theoretical results on the solution path of SLOPE as well as a numerical scheme to solve this path.
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37163
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20230926T120000Z
DTEND:20230926T130000Z
DTSTAMP:20240804T175332Z
UID:37487@calendar.unige.ch
CREATED:20230920T150425Z
LAST-MODIFIED:20231009T142550Z
SUMMARY:Sur les potentielles limitations du Deep Learning et ses utilisations pour la résolution d'EDPs (Teilo Wahl, UNIGE)
DESCRIPTION:La première partie de cette présentation est consacrée aux résultats de l'article "The troublesome kernel", qui montrent l'instabilité des réseaux Deep Learning comparés aux méthodes classiques dans le cadre des problèmes de reconstruction d'images à partir de données limitées. La seconde partie est centrée autour de l'élaboration de réseaux neuronaux pour la résolution d'EDPs, avec essais numériques effectués sur l'équation de Poisson.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37487
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231003T120000Z
DTEND:20231003T120000Z
DTSTAMP:20240804T175332Z
UID:37425@calendar.unige.ch
CREATED:20230918T143652Z
LAST-MODIFIED:20230919T150453Z
SUMMARY:A Comparison of IMEX- Spectral Deferred Correction, Multistep and Runge-Kutta Methods for the rotating Shallow Water Equations (Joscha Fregin, TU Hamburg)
DESCRIPTION:Atmospheric motion covers a broad range of time- and spatial scales. Low and high pressure systems can influence us for days or even weeks and they extend up to hundreds of kilometers. In contrast, sound waves pass by in seconds with wavelengths of centimeters to meters. Implicit-explicit (IMEX) time stepping methods can help to avoid drastic limitations on the time step induced by the variety of scales without requiring computationally expensive fully nonlinear implicit solves.\nIn our talk, we compare IMEX- Spectral Deferred Correction (SDC), multistep and Runge-Kutta time integrators for the Galewsky test case using the Python spectral method framework Dedalus. We demonstrate that SDC methods have superior stability properties and can provide shorter time to solution for comparable errors. In addition, we outline strategies that could further reduce simulation times by using the SDC residual to minimise the computational effort and by parallelising SDC "across the method". Finally, we discuss the interpretation of SDC as a Runge-Kutta method.
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37425
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231031T130000Z
DTEND:20231031T130000Z
DTSTAMP:20240804T175332Z
UID:37424@calendar.unige.ch
CREATED:20230918T143205Z
LAST-MODIFIED:20231020T151310Z
SUMMARY:High-order averaging and machine learning (Léopold Trémant, Université de Strasbourg)
DESCRIPTION:Highly-oscillatory phenomena combine the well-known numerical challenges of stiff equations and geometric problems, such as order reduction and energy preservation. These problems involve both fast oscillatory dynamics and slow drift dynamics, which interact to create complex dynamics. High-order averaging allows to asymptotically decouple these dynamics using formal calculations, which generates modified problems that can be solved with better numerical accuracy.\n\nIn this talk, I will introduce high-order averaging using a (somewhat recent) closed form approach, and exhibit the improvement in numerical accuracy it allows. I will then present the geometric properties of the method, notably regarding the preservation of a Hamiltonian structure. The talk will end with some preliminary results regarding geometric neural networks, which can replace the offline symbolic computations of averaging with an offline training procedure.
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37424
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231114T130000Z
DTEND:20231114T130000Z
DTSTAMP:20240804T175332Z
UID:37423@calendar.unige.ch
CREATED:20230918T142841Z
LAST-MODIFIED:20231018T154840Z
SUMMARY:Numerics with Coordinate Transforms for Efficient Brownian Dynamics Simulations (Dominic Phillips, University of Edinburgh)
DESCRIPTION: Many stochastic processes in the physical and biological sciences can be modelled using Brownian dynamics with multiplicative noise. However, numerical integrators for these processes can lose accuracy or even fail to converge when the diffusion term is configuration-dependent. One remedy is to construct a transform to a constant-diffusion process and sample the transformed process instead. In this work, we explain how coordinate-based and time-rescaling-based transforms can be used either individually or in combination to map a general class of variable-diffusion Brownian motion processes into constant-diffusion ones. The transforms are invertible, thus allowing recovery of the original dynamics. We motivate our methodology using examples in one dimension before then considering multivariate diffusion processes. We illustrate the benefits of the transforms through numerical simulations, demonstrating how the right combination of integrator and transform can improve computational efficiency and the order of convergence to the invariant distribution. Notably, the transforms that we derive are applicable to a class of multibody, anisotropic Stokes-Einstein diffusion that has applications in biophysical modelling.
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37423
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231121T130000Z
DTEND:20231121T140000Z
DTSTAMP:20240804T175332Z
UID:37473@calendar.unige.ch
CREATED:20230920T055749Z
LAST-MODIFIED:20231111T083747Z
SUMMARY:Sketched and truncated Krylov subspace methods for matrix equations
DESCRIPTION:Sketching can be seen as a randomized dimensionality reduction technique able to preserve the main features of the original problem with probabilistic confidence.\nThis kind of technique is emerging as one of the most promising tools to boost numerical computations and it is quite well-known by theoretical computer scientists.\nNowadays, sketching is gaining popularity also in the numerical linear algebra community even though its use and understanding are still limited.\n\nIn this talk we will present cutting-edge results about the use of sketching in numerical linear algebra. In particular, we will focus on showing how sketching can be successfully combined with Krylov subspace methods. We will specialize our results to the solution of large-scale matrix equations, but similar techniques can be applied to a variety of important algebraic problems, including the solution of linear systems, eigenvalue problems, and the numerical evaluation of matrix functions.\n\nThis talk is based on joint work with Valeria Simoncini and Marcel Schweitzer.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37473
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231128T130000Z
DTEND:20231128T130000Z
DTSTAMP:20240804T175332Z
UID:37628@calendar.unige.ch
CREATED:20231003T115747Z
LAST-MODIFIED:20231106T095626Z
SUMMARY:Solving numerically differential equations with distributed delays (Nicola Guglielmi, GSSI, L'Aquilla)
DESCRIPTION:Authors: Nicola Guglielmi and Ernst Hairer\n\nSinopsis:\n\nIn the early Nineties Ernst Hairer developed a famous code for the numerical approximation of stiff and differential algebraic problems, Radau5, which is based on the $3$-stage Radau IIa Runge Kutta method. The code became very popular and is nowadays one of the most important for the numerical approximation of implicit and stiff ODEs.\n\nLater, between 2000 and 2005, Ernst and I developed a code, Radar5, which extended Radau5 to delay differential equations with discrete delays (in short DDEs).\n\nDue to the increasing importance of models with distributed delays, for example in pharmacodynamics and pharmacokinetics (e.g Shuhua Hu et al., 2018), we have been recently addressed to certain kind of differential equations with distributed delays (in short DDDEs), where the dependence on the solution in the past appears through an integral term.\n\nIn this talk I will explain how these problems can indeed be approximated by suitable systems of ODEs or DDEs, according to the kind of distributed term.\n \nThe main ideas here are two:\n(i) replacing the kernels by suitable quasi-polynomial expansions, that is sums of polynomials multiplied by exponentials, and then transforming the distributed (integral) delay term into a set of ODEs (or DDEs);\n(ii) making the method efficient by exploiting the structure at the level of the linear algebra.\n\nNumerical evidence of the effectiveness of the proposed approach is illustrated on a few illustrative examples.
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37628
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231129T093000Z
DTEND:20231129T093000Z
DTSTAMP:20240804T175332Z
UID:37771@calendar.unige.ch
CREATED:20231019T082746Z
LAST-MODIFIED:20231115T093543Z
SUMMARY: Metrics and a cut-off phenomenon in deep learning (Anders Karlsson, UNIGE)
DESCRIPTION:Today’s AI is based on neural networks and deep learning. These concepts are easily explained to a mathematician. Why and exactly how it works are however much more unclear, and the problems and risks associated with this are widely discussed. Products of random transformations occur in practice in several ways in deep learning (initialization, stochastic gradient descent, drop-out, etc). There is a noncommutative ergodic theorem that governs such products asymptotically. For less large neural networks we also found empirical evidence of a cut-off phenomenon for random initialization. Ongoing work with B. Avelin (Uppsala) and B. Dherin (Google).
LOCATION:Room 1-07, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37771
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231212T130000Z
DTEND:20231212T140000Z
DTSTAMP:20240804T175332Z
UID:37467@calendar.unige.ch
CREATED:20230919T153345Z
LAST-MODIFIED:20231208T105630Z
SUMMARY:Modal analysis of the Robin-Schwarz operator (Laurence Halpern, Sorbonne Paris Nord)
DESCRIPTION:The Robin-Schwarz operator describes one iteration of the Schwarz domain decomposition algorithm with Robin transmission conditions. Little is known on the spectral properties of this operator which is neither self-adjoint nor compact. We give here for the first time a method to compute the eigenmodes with separation of variables, and prove that they are a complete family. These eigenmodes are of paramount importance for the design and understanding of multi-level domain decomposition methods. The problem inherently couples oscillatory and exponential series, but no result in approximation or spectral theory existing so far can be applied. The analysis calls for multiple tools: regularity results, complex analysis, including non harmonic Fourier series and almost periodic functions. We focus on the special case of 2 by 2 subdomains, but we have little doubts that it opens the door\nfor more general geometries.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37467
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20231219T130000Z
DTEND:20231219T130000Z
DTSTAMP:20240804T175332Z
UID:37768@calendar.unige.ch
CREATED:20231018T155237Z
LAST-MODIFIED:20231211T130928Z
SUMMARY: Numerical integration of the nonlinear Klein-Gordon equations at low regularity and conservation properties (Joackim Bernier (Nantes Université)
DESCRIPTION:Close to the origin, the nonlinear Klein--Gordon equations on the circle are nearly integrable Hamiltonian systems which have infinitely many almost conserved quantities called "harmonic energies" or "super-actions". In a series of works in 2008, D. Cohen, E. Hairer and C. Lubich proved that, at high regularity, classical symplectic numerical integrators preserve this qualitative property (even if the CFL number is not small). I will present a joint work with C. Abou Khalil (in progress), in which we extend this result for non-smooth solutions.
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/37768
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240123T130000Z
DTEND:20240123T130000Z
DTSTAMP:20240804T175332Z
UID:38249@calendar.unige.ch
CREATED:20231201T172438Z
LAST-MODIFIED:20231222T230321Z
SUMMARY:An introduction to fully homomorphic encryption (Philippe Chartier, Ravel Technologies, Paris)
DESCRIPTION:In this talk,I will give a brief overview of fully homomorphic encryption (FHE) and present one of the most promising method, the so-called TFHE. I will in particular detail the central “bootstrapping” technique.
LOCATION:Conference room on 8th floor, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/38249
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240220T130000Z
DTEND:20240220T140000Z
DTSTAMP:20240804T175332Z
UID:38234@calendar.unige.ch
CREATED:20231130T134134Z
LAST-MODIFIED:20240220T123243Z
SUMMARY:Domain decomposition for physics-informed neural networks (Alexandre Heinlein, Delft University)
DESCRIPTION:Physics-informed neural networks (PINNs) are a class of methods for\nsolving differential equation-based problems using a neural network as\nthe discretization. They have been introduced by Raissi et al. and\ncombine the pioneering collocation approach for neural network functions\nintroduced by Lagaris et al. with the incorporation of data via an\nadditional loss term. PINNs are very versatile as they do not require an\nexplicit mesh, allow for the solution of parameter identification\nproblems, and are well-suited for high-dimensional problems. However,\nthe training of a PINN model is generally not very robust and may\nrequire a lot of hyper parameter tuning. In particular, due to the\nso-called spectral bias, the training of PINN models is notoriously\ndifficult when scaling up to large computational domains as well as for\nmultiscale problems.\n\nIn this talk, overlapping domain decomposition-based techniques for\nPINNs are being discussed. Compared with other domain decomposition\ntechniques for PINNs, in the finite basis physics-informed neural\nnetworks (FBPINNs) approach, the coupling is done implicitly via the\noverlapping regions and does not require additional loss terms. Using\nthe classical Schwarz domain decomposition framework, a very general\nframework, that also allows for mult-level extensions, can be\nintroduced. The method outperforms classical PINNs on several types of\nproblems, including multiscale problems, both in terms of accuracy and\nefficiency. Furthermore, the combination of the multi-level domain\ndecomposition strategy with multifidelity stacking PINNs for\ntime-dependent problems will be discussed. It can be observed that the\ncombination of multifidelity stacking PINNs with a domain decomposition\nin time clearly improves the reference results without a domain\ndecomposition.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/38234
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240227T130000Z
DTEND:20240227T130000Z
DTSTAMP:20240804T175332Z
UID:38243@calendar.unige.ch
CREATED:20231201T070143Z
LAST-MODIFIED:20240223T145328Z
SUMMARY:Time adaptive waveform methods for dynamic multiphysics problems (Philipp Birken, University of Lund)
DESCRIPTION:We discuss partitioned time-integration of dynamic multiphysics problems, which commonly exhibit a multiscale behavior, requiring independent time-grids. The ideal method for solving these problems allows independent and adaptive time-grids, higher order time-discretizations, is fast and robust, and allows the coupling of existing subsolvers, executed in parallel.\n\nWaveform relaxation (WR) methods can potentially have all of these properties. These iterate on continuous-in-time interface functions, obtained via suitable interpolation. This way, one can allow for time adaptivity in the subsolvers. We present a fast and highly robust, second order in time, adaptive WR method. Basis is a Dirichlet-Neumann coupling at the interface, combined with an acceleration technique based on Quasi-Newton methods. As a leading example, we use coupled heat equations and discuss existing convergence results.\n\nNumerical results demonstrate the robustness and efficiency of the method for different multiphysics problems.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/38243
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240305T130000Z
DTEND:20240305T140000Z
DTSTAMP:20240804T175332Z
UID:38233@calendar.unige.ch
CREATED:20231130T133113Z
LAST-MODIFIED:20240207T111024Z
SUMMARY:Scalable algorithms for time-parallel optimal control based on diagonalisation (Arne Bouillon, KU Leuven)
DESCRIPTION:Optimal-control problems can give rise to boundary value problems (BVPs) whose solution contains the required control term. ParaDiag [1,2,3] and ParaOpt [4] are two algorithms to solve these optimality BVPs in a time-parallel way. The former is limited to linear problems with a symmetric system matrix, but generally scales well; the latter applies to non-linear problems, but includes a linear-system solving step that does not scale well when increasing time-parallelism.\n\nThe presentation consists of two main parts. In the first, we propose algorithmic improvements and extensions to these methods that make them more generally applicable and that improve their scaling. More specifically, we extend ParaDiag's applicability to non-symmetric linear problems and more general control objectives [5]. We then apply these new ParaDiag preconditioners to the linear system in ParaOpt, which improves its scaling [6].\n\nThe second part of the presentation discusses theoretical results about ParaDiag and ParaOpt applied to linear diffusive problems. We show how convergence for both algorithms can be studied by finding eigenvalues of related matrices and present convergence bounds that are parameterised by the time discretisation scheme(s) used. This generic approach allows us to improve the state-of-the-art coarse propagator for ParaOpt and to directly apply ParaDiag's analysis to the ParaOpt preconditioners we developed.\n\n[1]: S.-L. Wu and T. Zhou. Diagonalization-based parallel-in-time algorithms for parabolic PDE-constrained optimization problems. ESAIM: Control, Optimisa- tion and Calculus of Variations, 26, 2020.\n[2]: S.-L. Wu and J. Liu. A parallel-in-time block-circulant preconditioner for optimal control of wave equations. SIAM Journal on Scientific Computing, 42(3):A1510–A1540, 2020.\n[3]: M. J. Gander, J. Liu, S.-L. Wu, X. Yue, and T. Zhou. ParaDiag: parallel-in-time algorithms based on the diagonalization technique. arXiv:2005.09158 [cs, math], Apr 2021.\n[4]: M. J. Gander, F. Kwok, and J. Salomon. PARAOPT: A Parareal algorithm for optimality systems. SIAM Journal on Scientific Computing, 42(5):A2773–A2802, Jan 2020.\n[5]: A. Bouillon, G. Samaey, and K. Meerbergen. On generalized preconditioners for time-parallel parabolic optimal control. arxiv:2302.06406 [cs, math], Feb 2023.\n[6]: A. Bouillon, G. Samaey, and K. Meerbergen. Diagonalization-based preconditioners and generalized convergence bounds for ParaOpt. arxiv:2304.09235 [cs, math], Apr 2023.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/38233
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240312T130000Z
DTEND:20240312T140000Z
DTSTAMP:20240804T175332Z
UID:39020@calendar.unige.ch
CREATED:20240222T102357Z
LAST-MODIFIED:20240302T150413Z
SUMMARY:Space-Time discretization of traffic flow Mean-field game model (Amal Machtalay, EPFL)
DESCRIPTION:We explore a multi-class traffic model and examine the computational feasibility of mean-field games (MFG) in obtaining approximate Nash equilibria for traffic flow games involving a large number of vehicles. The obtained MFG system is challenging for different reasons, among them, the non-smoothness of the solutions, the high nonlinearity of the MFG system in addition to the large-scale computations.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39020
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240319T130000Z
DTEND:20240319T130000Z
DTSTAMP:20240804T175332Z
UID:38830@calendar.unige.ch
CREATED:20240206T134848Z
LAST-MODIFIED:20240223T140229Z
SUMMARY:On a selection problem for small noise perturbation of unstable dynamical systems (Andrey Pilipenko,Kyiv)
DESCRIPTION:The well-known Peano existence theorem states that an ordinary differential equation (ODE) with continuous coefficients has a local solution, which, however, may not be unique. On the contrary, an addition of a non-degenerate noise term usually yields the existence of a unique solution to the corresponding stochastic differential equation, even if the drift term is discontinuous. We study the limit of the stochastic equations as the noise intensity converges to zero. This limit can be interpreted as a natural selection of a solution for the initial ODE.\nThe identification of the limit is closely related to a study of the exact growth rate of certain stochastic equations, as well as to an averaging principle.\n\n
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/38830
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240326T130000Z
DTEND:20240326T140000Z
DTSTAMP:20240804T175332Z
UID:38590@calendar.unige.ch
CREATED:20240119T164635Z
LAST-MODIFIED:20240325T130033Z
SUMMARY:Solving convex optimization problems in function spaces (Sebastien Loisel, Heriot-Watt University)
DESCRIPTION:Many stationary partial differential equations and boundary value problems can be rephrased in terms of the minimization of a functional, sometimes called the energy, the Lagrangian, or the action. In the usual case (e.g. the usual linear Laplacian), the Lagrangian is a convex quadratic functional of its argument. We consider the generalization to nonquadratic but convex Lagrangians (e.g. the p-Laplacian, a highly nonlinear PDE operator). One may solve such problems by the barrier method on a finite element grid with $n$ vertices, but usually the performance is no better than $O(n^{1.5})$ FLOPS. We will discuss new algorithms that converge in $O^*(n)$ FLOPS, where the star indicates that certain polylogarithms are neglected.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/38590
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240416T120000Z
DTEND:20240416T130000Z
DTSTAMP:20240804T175332Z
UID:38799@calendar.unige.ch
CREATED:20240205T121759Z
LAST-MODIFIED:20240321T165624Z
SUMMARY:The geometric properties of exotic aromatic Butcher-series in stochastic backward error analysis (Adrien Laurent, INRIA Rennes)
DESCRIPTION:The Butcher series are a strong tool for the backward error analysis of ordinary and stochastic differential equations. After introducing the standard B-series, their exotic and aromatic extensions, and some of their numerical applications, we will ask ourselves the question: are these B-series just a tool for tedious calculations or are they interesting beyond their numerical applications? We will show that exotic aromatic B-series satisfy a universal geometric property of equivariance, which confirms that they are a natural geometric object that is interesting not only from the numerical point of view. If time allows, we shall quickly discuss the extension of exotic aromatic B-series to the creation of methods on manifolds.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/38799
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240423T120000Z
DTEND:20240423T130000Z
DTSTAMP:20240804T175332Z
UID:39034@calendar.unige.ch
CREATED:20240223T084647Z
LAST-MODIFIED:20240422T102010Z
SUMMARY:Discrete Transparent Boundary Conditions for 1D linear equations (Antoine Tonnoir, University of Rouen)
DESCRIPTION:In this work, we consider the problem of constructing Transparent Boundary Conditions (TBCs) for discretised 1D linear evolution equations (such as the heat equation, wave equation or Klein-Gordon equation). TBCs are used to reformulate a problem posed in an unbounded domain into a problem posed in a bounded domain in a way that the solutions of the two problems (in the bounded and unbounded domain) are the same in the bounded domain. At the continuous level, many results are already known and efficient approximations of the TBCs have been developed (see for instance the absorbing boundary conditions for the wave equation). Yet, for the discretised PDE, the construction of TBCs (called D-TBCs for Discrete-TBCs) is less obvious and fewer results exist (see for instance the work of C. Besse, M. Ehrhardt, X. Antoine). In this talk, we will present an original way to compute these D-TBCs for a class of 1D linear PDEs and arbitrary discretisation schemes. Several numerical examples will be shown to illustrate the results. \n\n
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39034
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240430T120000Z
DTEND:20240430T130000Z
DTSTAMP:20240804T175332Z
UID:39019@calendar.unige.ch
CREATED:20240222T102132Z
LAST-MODIFIED:20240429T145642Z
SUMMARY:Origin of the Fourier Series (Gerhard Wanner, UNIGE)
DESCRIPTION:Joint work with Philippe Henry.\n\nTaylor 1715\nHuygens 1673\nJoh.Bernoulli 1698,1728\nEuler 1729,1754\nd'Alembert 1746/49\nEuler 1746/49\nLagrange 1759\nEuler 1760,1777\nFourier 1807,1811,1822\nKahane 2008\nDirichlet 1829\nRiemann 1854\nArago 1833
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39019
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240507T120000Z
DTEND:20240507T130000Z
DTSTAMP:20240804T175332Z
UID:39127@calendar.unige.ch
CREATED:20240304T142619Z
LAST-MODIFIED:20240508T135107Z
SUMMARY:Long-time behaviour of numerical integrators for charged particle dynamics (Ernst Hairer, UNIGE)
DESCRIPTION:The Boris algorithm is the most popular time integrator for charged particle motion in electric and magnetic force fields. It is a symmetric one-step method, and it preserves the phase volume exactly. However, it is not symplectic. In this talk we prove near-conservation of energy over very long times in the special cases where either the magnetic field is constant or the electric potential is quadratic. When none of these assumptions is satisfied, it is illustrated by numerical examples that the numerical energy can have a linear drift or its error can behave like a random walk.\n\nWe thank Martin Gander for drawing our attention to this problem.\nThe presented results have been obtained in collaboration with Christian Lubich.\n
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39127
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240521T120000Z
DTEND:20240521T120000Z
DTSTAMP:20240804T175332Z
UID:39573@calendar.unige.ch
CREATED:20240418T125249Z
LAST-MODIFIED:20240428T200402Z
SUMMARY:On the contractivity of neural differential equations (Nicola Guglielmi, GSSI, L'Aquilla)
DESCRIPTION:abstract: see attached PDF file.
LOCATION:Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39573
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240618T120000Z
DTEND:20240618T120000Z
DTSTAMP:20240804T175332Z
UID:39035@calendar.unige.ch
CREATED:20240223T135159Z
LAST-MODIFIED:20240605T083629Z
SUMMARY:Advanced Time-Adaptive PIROCK Method with Error Control for Magnetic Reconnection Simulations in Chromospheric Environments (Q.M. Wargnier, Lockheed Martin Solar and Astrophysics Laboratory and Baeri)
DESCRIPTION:Understanding the heating mechanism of the Sun's chromosphere remains a central challenge in solar physics, with magnetic reconnection (MR) playing a pivotal role in its heating and dynamics. Recent observations have highlighted the significance of ion-neutral interactions, which current models struggle to fully incorporate. In this presentation, we will introduce a numerical simulation with a Multi-Fluid Multi-Species (MFMS) model to investigate MR in upper chromospheric environments, accounting for multi-fluid/multi-species interactions, including helium and hydrogen species.\n\nThe MFMS model is a multi-scale temporal model presenting challenges for temporal integration with classical integration methods. The model comprises stiff source terms from collisional or reactive effects, coupled with diffusion terms, and an increased number of equations due to the duplication for each fluid considered, complicating the time integration process. Therefore, our approach employs the second-order Partitioned Implicit-Explicit Runge-Kutta (PIROCK) method, renowned for its ability to handle complex convective-diffusive-reactive systems of equations like the MFMS model. Comparative analyses with alternative integration methods, including third-order explicit Runge-Kutta time integration and another method based on a first-order Lie splitting approach, underscore the superior computational efficiency and accuracy of PIROCK. Our findings underscore the vital role of particle decoupling in facilitating efficient heating mechanisms within the chromosphere, with implications for phenomena such as helium enrichment observed in switchbacks and coronal mass ejections.\n
LOCATION:conference room, 8th floor (unusual room!), Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39035
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240625T120000Z
DTEND:20240625T120000Z
DTSTAMP:20240804T175332Z
UID:39036@calendar.unige.ch
CREATED:20240223T135206Z
LAST-MODIFIED:20240605T083646Z
SUMMARY:Approximate Matrix Factorization and W-methods for the time integration of multidimensional parabolic problems (Domingo Hernández Abreu, Universidad de La Laguna)
DESCRIPTION:This talk deals with the time-integration of space-discretised parabolic problems subject to Dirichlet boundary conditions on a rectangular m-dimensional domain.\nWe consider the combination of linearly implicit methods (W-methods) along with Approximate Matrix Factorization based on an alternating direction implicit approach, which allows to reduce the algebra cost to the level of 1D problems. Optimal results on PDE-convergence will be presented for linear problems, the Euclidean norm and arbitrary spatial dimensions m ≥ 2. Some techniques aimed at mitigating the order reduction phenomenon due to time-dependent boundary conditions are also presented and numerically illustrated in two and three spatial dimensions.\n\nThis talk is based on joint work with S. González-Pinto, E. Hairer and S. Pérez-Rodríguez.
LOCATION:Conseil Général 7-9, Room 1-15 (unusual room!), Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39036
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240702T120000Z
DTEND:20240702T130000Z
DTSTAMP:20240804T175332Z
UID:39459@calendar.unige.ch
CREATED:20240409T093245Z
LAST-MODIFIED:20240418T125947Z
SUMMARY:Adaptively optimized Schwarz methods (Conor McCoid, Laval Universitiy, Quebec)
DESCRIPTION:Optimized Schwarz methods use Fourier analysis or similar to find transmission conditions between subdomains that provide faster convergence over standard Schwarz methods. However, this requires significant upfront analysis of the operator, and may not be straightforward for all problems. This work presents black box methods for adaptively optimizing the transmission conditions, which is equivalent to an optimal Krylov subspace method.
LOCATION:Conseil Général 7-9, Room 1-05, Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/39459
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240909T220000Z
DTEND:20240910T220000Z
DTSTAMP:20240804T175332Z
UID:37596@calendar.unige.ch
CREATED:20230928T152151Z
LAST-MODIFIED:20230928T152407Z
SUMMARY:Swiss Numerical Analysis Day 2024
DESCRIPTION:The Swiss Numerical Analysis Day is an annual informal meeting following the tradition of the "Journées Suisses d'Analyse Numérique" launched in 1976. Its intention is to bring together researchers in Switzerland who are active in the fields of\n-scientific computing\n-numerical analysis\n-high-performance computing\n-visualization\n-numerical analysis in engineering\n-computational chemistry, biology, etc.\n\nThe program consists of two plenary lectures, selected contributed talks, and a poster session.
LOCATION:A150
TRANSP:TRANSPARENT
CATEGORIES:Conférence
CONTACT:
URL:http://agenda.unige.ch/events/view/37596
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240911T220000Z
DTEND:20240913T220000Z
DTSTAMP:20240804T175332Z
UID:39990@calendar.unige.ch
CREATED:20240612T054436Z
LAST-MODIFIED:20240612T054436Z
SUMMARY:HealthCH : Symposium en management de la santé
DESCRIPTION:Les formations en management des organisations de santé présentent le Symposium HealthCH 2024, première édition organisée par la faculté de médecine de l’UNIGE les 12 et 13 septembre 2024 au Campus Biotech. \n\nLe symposium vise à discuter et co-élaborer des solutions et des nouveaux partenariats pour répondre aux défis qui se présentent aux organisations de santé suisses. HealthCH est destiné aux managers de la santé, aux cliniciens, aux étudiants en management et aux chercheur.e.s en management des organisations de santé.\n\nL’évènement alternera des tables rondes sur les grands défis de l’évolution du système de santé suisse (santé numérique, financement des soins intégrés, éco-responsabilité des organisations de santé, continuité des soins…), des Masterclass sur des sujets pointus de management opérationnel et des analyses d’experts internationaux.
LOCATION:Campus Biotech, Chem. des Mines 9, 1202 Genève
TRANSP:TRANSPARENT
CATEGORIES:Symposium
CONTACT:
URL:http://agenda.unige.ch/events/view/39990
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240911T220000Z
DTEND:20240913T220000Z
DTSTAMP:20240804T175332Z
UID:39990@calendar.unige.ch
CREATED:20240612T054436Z
LAST-MODIFIED:20240612T054436Z
SUMMARY:HealthCH : Symposium en management de la santé
DESCRIPTION:Les formations en management des organisations de santé présentent le Symposium HealthCH 2024, première édition organisée par la faculté de médecine de l’UNIGE les 12 et 13 septembre 2024 au Campus Biotech. \n\nLe symposium vise à discuter et co-élaborer des solutions et des nouveaux partenariats pour répondre aux défis qui se présentent aux organisations de santé suisses. HealthCH est destiné aux managers de la santé, aux cliniciens, aux étudiants en management et aux chercheur.e.s en management des organisations de santé.\n\nL’évènement alternera des tables rondes sur les grands défis de l’évolution du système de santé suisse (santé numérique, financement des soins intégrés, éco-responsabilité des organisations de santé, continuité des soins…), des Masterclass sur des sujets pointus de management opérationnel et des analyses d’experts internationaux.
LOCATION:Campus Biotech, Chem. des Mines 9, 1202 Genève
TRANSP:TRANSPARENT
CATEGORIES:Symposium
CONTACT:
URL:http://agenda.unige.ch/events/view/39990
FREEBUSY;FBTYPE=FREE:
END:VEVENT
BEGIN:VEVENT
DTSTART:20240924T120000Z
DTEND:20240924T120000Z
DTSTAMP:20240804T175332Z
UID:40063@calendar.unige.ch
CREATED:20240705T075644Z
LAST-MODIFIED:20240705T075644Z
SUMMARY:TBA (Daniel Ruprecht, Technische Universität Hamburg)
DESCRIPTION:TBA
LOCATION:Conseil Général 7-9, Room 1-07 (unusual room!), Séminaire d'analyse numérique
TRANSP:TRANSPARENT
CATEGORIES:Séminaire
CONTACT:
URL:http://agenda.unige.ch/events/view/40063
FREEBUSY;FBTYPE=FREE:
END:VEVENT
END:VCALENDAR