2025 Meeting

The participants of the Dutch-Flemish Scientific Computing Societies (SCS) Springmeeting at the Rijksuniversiteit Groningen in 2024.

This years Springmeeting will be held on Friday 13 June 2025, at the Hasselt University.

The Springmeeting is free of charge, but you will need to register. Fill in the registration form here.
If you registered but can't attend, please notify me in time via numwisk@cwi.nl.

Location:

Hasselt University, Campus Diepenbeek

Agoralaan, Building D

3590 Diepenbeek

The presentations will be given in room H2, Building D

How to reach Campus Diepenbeek

https://www.uhasselt.be/en/about-hasselt-university/contact/campus-diepenbeek

More about Hasselt, including hotels and other accommodations:

https://www.visithasselt.be/en

Organizing committee:

Sorin Pop (Hasselt University), Jason Frank (Utrecht University) and Martine Anholt (CWI, Secretary SCS).

Program

09:30-10:00	Registration, coffee and tea
10:00-10:30	Jochen Schuetz: Convergence acceleration for multiderivative SDC schemes
10:30-11:00	Matthias Schlottbom: On accelerated iterative schemes for neutron transport using residual minimization
11:00-11:25	Wouter van Harten: Distributed preconditioning for the Helmholtz equation
11:25-12:00	Coffee and tea break
12:00-12:25	Koondi Mitra: Robust & adaptive iterative linearization methods for nonlinear elliptic problems
12:25-12:50	Sanjana Verma: Design of Freeform Imaging Systems: Mathematical Model and Numerics
12:50-13:00	Group picture
13:00-14:00	Lunch
14:00-14:25	Karel van Bockstal: Numerical algorithms for the reconstruction of space-dependent sources in thermoelasticity
14:25-14:50	Syver Agdestein: Learning model-data consistent closure models in large-eddy simulation
14:50-15:20	Coffee and tea break
15:20-15:45	Florian Feppon: Stokes flows in semi-infinite periodic strip
15:45-16:25	Barry Koren: Machine Learning and Reduced Order Modeling for Uncertainty Quantification in Nonlinear PDE Problems
16:25-16:30	Closing

Speakers:

	Barry Koren, Technical University Eindhoven Barry Koren received his MSc degree in aerospace engineering and his PhD degree in numerical mathematics, both from TU Delft. Prior to his appointment at TU Eindhoven (TU/e), he was part-time full professor Numerical Analysis at Leiden University, part-time full professor Computational Fluid Dynamics at TU Delft, and leader of the research group Modelling, Analysis and Simulation and member of the management team of the Centrum Wiskunde & Informatica (CWI) in Amsterdam. Barry serves as an associate editor of the Journal of Computational Physics and as a member of several national and international scientific boards, committees and research programs. He was president of the Royal Dutch Mathematical Society (KWG) and chair of the Dutch-Flemish Scientific Computing Society. At the TU/e Department of Mathematics and Computer Science, he was vice-dean for four years and acting dean twice. He is a scientific adviser at CWI and at the Royal Netherlands Aerospace Centre (NLR) in Amsterdam. So far, he has (co-)authored two books and about 200 scientific papers, and (co-)edited four books and five special journal issues. Barry has supervised about 25 PhD, 65 MSc and 15 BSc students.
	Matthias Schlottbom, University of Twente MS research is driven by mathematically tackling real-world challenges with tools from inverse problems, numerical analysis, numerical simulation and optimization. The kind of applications range from optical tomography, over simulation and metrology for (photonic) nanostructures, to biological and socio-economic inverse problems, like formation of biological transportation networks or population dynamics. Within our research group (Mathematics of Computational Science - MACS), we have a strong team of researchers with complementary knowledge in designing, analyzing and implementing advanced finite element and boundary element solvers. Members of MACS (co)-develop for example the community finite element software packages HPGEM and NGSOLVE or integral equation solvers, like INTI, that offer state-of-the-art codes for solving partial differential equations (PDEs) and high-level optimization using PDE models.The expertise of the group is further complemented by members of our cluster (Systems, Analysis, and Computational Science) with expertise for instance in model order reduction, time series and statistical analysis, or control theory. Next to research in mathematics, we have active collaboration with groups in physics and geosciences.
	Jochen Schuetz, Hasselt University Jochen Schütz is associate professor (hoofddocent) at UHasselt. He started as an assistant professor (tenure track) at UHasselt back in 2016, after having obtained a PhD (2011) and done a postdoc (2012-2015) at RWTH Aachen University. His research expertise is in high-order schemes for ordinary and partial differential equations. He has contributed to the (hybridized) discontinuous Galerkin methods for compressible Navier-Stokes equations, asymptotically preserving schemes, multiderivative time integration schemes, and many more. Amongst other things, he is co-organizer of the Belgian “Mathematics for industry week” (https://be-maths-in.be/mfi25/).
	Karel van Bockstal, Gent University Karel Van Bockstal obtained his PhD (in mathematical engineering) in 2015 at Ghent University, Belgium, and is currently working as a postdoctoral researcher at the Ghent Analysis & PDE Center of Ghent University (Department of Mathematics: Analysis, Logic and Discrete Mathematics). His area of specialisation is related to mathematical analysis, evolutionary partial differential equations, and the development and implementation of numerical algorithms. This research focus concerns direct and inverse problems finding applications in diverse fields such as heat transfer, elasticity, electromagnetism and thermoelasticity. He (co-)authored 39 publications, co-supervised 2 PhD students (1 finished, 1 in progress) and was awarded the EAIP Young Scientist Award in May 2016.
	Florian Feppon, KU Leuven Florian Feppon is a tenure-track assistant professor at the Department of Computer Science of KU Leuven since October 2022. He is a member of the Numerical and Applied Mathematics research Unit (NUMA). He teaches courses on numerical simulation of differential equations and nonlinear systems in the Master of Mathematical Engineering program. Before joining KU Leuven, he was a Hermann–Weyl postdoctoral instructor at the Seminar for Applied Mathematics (SAM) at ETH Zürich, Switzerland, from September 2020 to September 2022, in the group of Prof. Habib Ammari. From January to August 2020, he held a postdoctoral researcher position at the Centre de Mathématiques Appliquées (CMAP) at École polytechnique, Palaiseau, France. Florian Feppon completed his PhD at CMAP, École polytechnique, from April 2017 to December 2019, supported by CIFRE industrial research funding from SAFRAN. His doctoral research, supervised by Prof. Grégoire Allaire, focused on the shape and topology optimization of multiphysics systems. He currently supervises 3 PhD students, including two as a promotor and one as a co-promotor.
	Koondi Mitra, Eindhoven University of Technology Kondanibha (a.k.a. Koondi) Mitra is an Assistant Professor at Eindhoven University of Technology (TU/e), specializing in Computational Illumination Optics within the Department of Mathematics and Computer Science. Koondi completed a dual degree (B.Tech and M.Tech) in Mechanical Engineering at the Indian Institute of Technology (IIT) Kharagpur, graduating in 2015. He earned his doctorate jointly from TU/e and Hasselt University in Belgium with a cum laude distinction. Then he completed successive post-docs in TU Dortmund, INRIA Paris, Radboud University Nijmegen, and Hasselt University, before joining Eindhoven as a faculty in Nov 2023. Koondi's research encompasses nonlinear partial differential equations, mathematical modelling, applied and numerical analysis. It is focused on optics, porous media flow problems and mathematical biology.
	Syver Agdestein, CWI I am currently doing a PhD in the Scientific Computing Group at Centrum Wiskunde& Informatica in Amsterdam. My field of research is discretization and machine learning for turbulence modeling and large-eddy simulation. I enjoy bouldering,playing the piano, writing differentiable solvers in Julia, and running largefluid simulations on GPUs.
	Sanjana Verma, Technical University Eindhoven Sanjana Verma obtained a Master’s degree in Mathematics from Indian Institute of Technology Bhubaneswar. Since September 2021, she is a Ph.D. student in the Computational Illumination Optics group at the Department of Mathematics and Computer Science, Eindhoven University of Technology. Her research focuses on developing inverse methods for the design of imaging optical systems.
	Wouter van Harten, Radboud University After obtaining a MSc degree in Applied Mathematics at the University of Twente, Wouter van Harten started his PhD research project at the Radboud University in late 2021 under the supervision of Laura Scarabosio. His research is focussed on high dimensional uncertainty quantification, with a specific interest in uncertainty in the domain of a Partial differential equation and computational approaches.

Abstracts:

Barry Koren TU Eindhoven	Machine Learning and Reduced Order Modeling for Uncertainty Quantification in Nonlinear PDE Problems A multigrid method for uncertainty quantification in nonlinear PDE problems is proposed. The principle behind the method is that the relative solution error between grid levels has a spatial structure that is by good approximation independent of the actual grid level. The method learns this structure by employing a sequence of convolutional neural networks, that are well-suited to automatically detect local truncation errors as latent quantities of the solution. By using transfer learning, the information of coarse grid levels is reused on fine grid levels to minimize the required number of samples on fine grid levels. The method outperforms an existing multi-level method for uncertainty quantification, for a relevant test case. As an alternative for machine learning in uncertainty quantification, the potentials of the use of a reduced order model approach in uncertainty quantification are quickly explored. Advantages and challenges of the machine learning approach and the reduced order modeling approach are compared. Combination of both approaches in uncertainty quantification may be an interesting future research topic.
Matthias Schlottbom University of Twente	On accelerated iterative schemes for neutron transport using residual minimization The numerical solution of neutron transport problems requires the solution of very large linear systems. In the past decades a vast amount of iterative schemes has been devised to solve this task. To accelerate convergence of such iterative methods, preconditioners have been developed that solve a diffusion problem, which can be well motivated using arguments from asymptotic analysis. It has been observed that special care needs to be taken in the discretization of such diffusion problems to preserve convergence, leading to so-called consistent schemes. In this talk, we take a slightly different point of view and use preconditioners that are based on residual minimization over suitable subspaces. We prove convergence of the resulting iteration using Hilbert space norms, which allows us to obtain algorithms that converge robustly with respect to finite dimensional realizations via Galerkin projections. We investigate in particular the behavior of the iterative scheme for discontinuous Galerkin discretizations in the angular variable in combination with subspaces that are derived from related diffusion problems. The performance of the resulting schemes is investigated in numerical examples for highly anisotropic scattering problems with heterogeneous parameters.
Jochen Schuetz Hasselt University	Convergence acceleration for multiderivative SDC schemes In this talk, we consider a class of multiderivative SDC (spectrally deferred correction) schemes for the approximate solution of initial value problems. SDC schemes approximate solutions through an iterative procedure, where the iteration is usually towards a high-order fully implicit collocation (multiderivative) Runge-Kutta scheme, called the background scheme. This iteration allows for methods that can be parallelized in time. SDC methods have the distinct feature that in each iteration, the formal order of consistency is increased by one, until the order of the background scheme is reached. So even if convergence towards the background scheme is not obtained, the scheme is of high-order. However, in many practical cases, e.g., for stiff problems, it is desirable to converge to the background scheme to mitigate order reduction. In this talk, we therefore discuss possible convergence acceleration of the iteration towards the background scheme.
Karel van Bockstal Gent University	Numerical algorithms for the reconstruction of space-dependent sources in thermoelasticity In this talk, I will discuss inverse problems of determining space-dependent sources in thermoelasticity from time-averaged and final-in-time measurements. I will briefly discuss the uniqueness of a solution to the inverse problems under appropriate conditions on the given time-dependent part of the sought source. Afterwards, I will apply several numerical methods to reconstruct the sources, including a Landweber scheme and minimisation methods for the corresponding cost functionals. A shortcoming of these methods is that the initial guess fixes the values of the sought source at the boundary. I will explain how to overcome this by applying the Sobolev gradient method. Numerical examples (implemented on the FEniCSx platform) will be presented to discuss the different approaches and support the findings.
Florian Feppon KU Leuven	Stokes flows in semi-infinite periodic strip It is well-known in the heat transfer engineering community that a laminar flow entering a periodic duct becomes periodically developed after a short distance to the inlet. This enables the efficient design of periodic channels by optimizing the flow profile in a single unit periodicity cell. However, the performance of the obtained design is usually worse than predicted, because this procedure does not take into account the boundary layers at the entrance and outlet flow regions. In this talk, I will present a numerical method for effectively computing such boundary layers and obtaining quantitative descriptions of their exponential decay rates. This is a joint work with S. Fliss from POEMS, ENSTA Paris.
Koondi Mitra TU Eindhoven	Robust & adaptive iterative linearization methods for nonlinear elliptic problems In this work, we consider iterative linearization schemes for nonlinear elliptic equations. We show that the iterative process can be represented by computing updates from the residual of the problem evaluated at the current iteration. For a large class of problems (e.g., porous media flow, mean curvature flow, biological flows, mixed dimensional equations, optimal transport problems, and design of optical systems) and for almost all standard linearization schemes (Newton scheme, Picard scheme, L/M-schemes) this structure is shown to hold. Moreover, by considering an update equation which does not depend on previous iterates, we get schemes that are, under some minor assumptions on coefficients, more robust in terms of discretization, nonlinearities, and degeneracies (the ‘so called’ Zarantonello/L-scheme). It also results in solving the same problem in every iteration implying that preconditioners, or Green’s functions, or neural networks can be used to solve each iteration faster. However, the convergence is at most linear, and consequently, more iterations are taken compared to quadratic schemes such as the Newton method unless optimal coefficient values are chosen. Based on a posteriori error estimates, we devise an adaptive scheme which automatically chooses such (quasi-)optimal parameters for convergence. Numerical results are presented for a wide variety of problems demonstrating the stability and efficiency of this class of schemes, and their usage.
Syver Agdestein CWI	Learning model-data consistent closure models in large-eddy simulation Large-eddy simulation (LES) aims to compute large-scale motions of turbulent fluid flows at lower computational costs than direct numerical simulation of all scales. LES requires closure models to account for the effects of unresolved small-scale fluctuations. Neural network closure models have been used to achieve high accuracy, but instabilities have been observed when neural closure models are inserted into LES environments. These instabilities can be attributed to inconsistencies between the closure model and training data. We show how to diminish or eliminate these inconsistencies altogether, leading to both stable and accurate closure models. While a-posteriori training of the neural networks embedded in the LES solvers is considered necessary to stabilize the models, we found that simple a-priori training is sufficient when the model-data inconsistencies are properly addressed, reducing the cost and complexity of training the closure models.
Sanjana Verma TU Eindhoven	Design of Freeform Imaging Systems: Mathematical Model and Numerics Freeform optical systems are gaining significance in imaging applications due to enhanced design flexibility and superior performance over traditional designs. I present an inverse model for the design of a parallel-to-point freeform imaging system with two reflectors and describe its connection with optimal transport theory. In nonimaging optics, inverse methods compute the shapes of freeform surfaces that convert a given source distribution to a desired target distribution. The mathematical model underlying inverse methods consist of an optical map connecting source and target coordinates, and the law of conservation of energy. The goal of nonimaging systems is the transfer of energy, whereas imaging systems aim to minimize aberrations, i.e., deviations from a linear optical map in phase space. The propagation of light in an optical system is governed by a Hamiltonian system, which leads to the optical map being symplectic. From global energy conservation and the symplectic nature of the linear optical map in phase space, we conclude that the ratio of the energy distributions at the source and the target must be constant for designing an imaging system. Subsequently, inverse methods from nonimaging optics are utilized to compute freeform imaging reflectors. The design is verified with a highly accurate raytracer based on quasi-interpolation, a local approximation method. The inverse imaging design outperforms the traditional design.
Wouter van Harten Radboud University	Distributed preconditioning for the Helmholtz equation tba