The DNA Seminar (Spring 2024)

The DNA seminar serves as the seminar series for the DNA group (Differential Equations and Numerical Analysis) at NTNU. Our sessions take place on Fridays from 10:15 to 11:00 in Simastuen 656, SBII, located on Campus Gløshaugen.

If you're interested in delivering a presentation or hosting a guest speaker for our seminar, kindly get in touch with me Kurusch Ebrahimi-Fard.

Upcoming seminars:

Time/place: Friday, June 21, 2024, 10:15-11:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Swati Yadav (NTNU, Trondheim)
Title: TBA
Abstract: TBA \( \phantom{mmmmm}\hspace{22cm}\)
Time/place: Monday, June 17, 2024, 11:15-12:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Leonardo Colombo (Centre for Automation and Robotics, CSIC-UPM, Spain)
Title: Machine learning methods for the control and teledetection with multi-rotors UAVs
Abstract: Data-driven modeling is a technique, where the parameters and components of a model are determined based on the characteristics of some set of observations of an unknown system, such that the model best describes the observations. The major drawback of these methods frequently manifests as unpredictable outcomes and the absence of guarantees about the stability of the control loop and performance limitations, which is translated as compromised safety in control systems. Therefore, the current application of data-driven approaches in control is often limited to non-critical and/or low-performance systems. In this seminar, first I will introduce Gaussian Process Regression as a machine learning method to estimate certain uncertainties that appear in the control of an hexacopter vehicle with safety guarantees in different control tasks, as for instance, trajectory tracking problems, or the control allocation problem in fault-tolerant control.
The second part of the talk will be devoted to the use of AI techniques in multi-rotors UAVS for fauna detection with the aim of wildlife management and conservation in the context of the research project “Wildlife Monitoring Bots” (https://wimoboproject.com). In particular, we will explore how computer vision and deep learning techniques can be used to identify deers in their natural environments. The use of high-resolution aerial data captured by unmanned aerial vehicles as part of this detection strategy is discussed.
Time/place: Friday, June 14, 2024, 10:15-11:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Jacob Goodman (University of Nebrija, Spain)
Title: Modified Riemannian Cubics in Obstacle Avoidance Problems with Applications to Quadrotor Load Transportation
Abstract: Riemannian cubic polynomials have a rich history, and have received increased attention in recent years due to their connection to optimal control problems on robotic systems. In this talk, I discuss an extension of Riemannian cubics polynomials which can be used for completing additional tasks by way of artificial potential functions. In particular, I discuss the design of artificial potentials for the purposes of obstacle avoidance, and provide conditions under which safety can be guaranteed. Additionally, I derive reduced necessary conditions for optimality in the case that the underlying manifold is a Lie group equipped with a left-invariant Riemannian metric, or a Riemannian homogeneous space of such a Lie group, and discuss how artificial potential design can be simplified in such cases. Finally, I apply these principles to the application of quadrotor load transportation.
Time/place: Monday, June 10, 2024, 11:15-12:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Arturo De Marinis (GSSI, Italy)
Title: TBA
Abstract: TBA \( \phantom{mmmmm}\hspace{22cm}\)
Time/place: Friday, May 31, 2024, 10:15-11:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Håkon Noren (NTNU, Trondheim)
Title: TBA
Abstract: TBA \( \phantom{mmmmm}\hspace{22cm}\)
Time/place: Friday, May 24, 2024, 10:15-11:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Thomas Christiansen (NTNU, Trondheim)
Title: TBA
Abstract: TBA \( \phantom{mmmmm}\hspace{22cm}\)
Time/place: Wednesday, May 8, 2024, Simastuen 656, SBII, Campus Gløshaugen \( \phantom{mmmmm}\hspace{10.8cm}\)
14:15-15:00
Speaker: Takaharu Yaguchi (Kobe University, Japan)
Title: TBA
Abstract: TBA \( \phantom{mmmmm}\hspace{22cm}\)
15:15-16:00
Speaker: Baige Xu (Kobe University, Japan)
Title: TBA
Abstract: TBA \( \phantom{mmmmm}\hspace{22cm}\)
Time/place: Friday, May 3, 2024, 10:15-11:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Matej Benko (Brno University of Technology)
Title: Discretization of Wasserstein Gradient Flow
Abstract: We consider the nonlocal model with interaction term and linear diffusion as equivalent problem to the gradient flow along a convex functional. In the first part, we use the propagation of chaos technique. It means the particle approximation of the nonlocal model with a system of local models. Then, in the second part we propose the operator splitting technique of local model into the proximal (implicit) and adding Gaussian (diffusion) steps. Also, in the second step the problem is considered as a gradient flow (proximal step along the confinement potential functional and diffusion step along the internal energy functional). Uniform in time bounds for both parts are proven in terms of Wasserstein distance. The talk is based on joint work with Iwona Chlebicka, Jørgen Endal and Błażej Miasojedow.
Time/place: Monday, April 29, 2024, 11:15-12:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Sondre Tesdal Galtung (NTNU, Trondheim)
Title: Equivalence of entropy solutions and gradient flows for the 1D pressureless Euler–Poisson system
Abstract: The thoroughly studied Euler–Poisson (EP) system may serve as a model for, e.g., self-gravitating matter or plasma. Here we show how two apparently distinct notions of solution, introduced for the one-dimensional EP system, turn out to be the same. Each notion is based on an associated equation which admits unique solutions, and which provides distributional solutions of the original EP system. One of these equations is a differential inclusion coming from an \(L^2\)-gradient flow with a minimal selection principle, while the other is a conservation law. It turns out that the minimal selection principle coincides with the Oleinik E condition for entropy solutions of the conservation law. Some examples are provided to illustrate the ideas. This is joint work with José A. Carrillo (University of Oxford).
Time/place: Friday, March 1, 2024, postponed, 10:15-11:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Katerina Nik (Univ. of Vienna, Austria)
Title: Accretive growth in elastic solids
Abstract: I will talk about accretive growth in elastic bodies subject to a prescribed docking condition. The growth process is encoded by viscosity solutions to a suitable generalized eikonal equation describing, roughly speaking, at which time each point of the body is added to the solid. After discussing the global-in-time well-posedness of the equilibrium problem when the growth of the body is prescribed, I will then show existence of a weak local solution for the coupled equilibrium-growth setting, namely when both mechanical displacement and time-evolving set are unknowns of the problem. A key challenge of the latter setting is the low regularity of the growing body. This is overcome by means of a level-set approach allowing to deal with possible singularities originating from self-touching of the growing body. This is joint work with Elisa Davoli (TU Wien), Ulisse Stefanelli (University of Vienna) and Giuseppe Tomassetti (University of Rome III).

Earlier this term:

Time/place: Friday, April 19, 2024, 10:15-11:00, Simastuen 656, SBII, Campus Gløshaugen
Speaker: Hajg Jasa (NTNU, Trondheim)
Title: The Riemannian Convex Bundle Method
Abstract: Within the context of optimization on manifolds, a research direction of particular interest is the investigation of algorithms fit to optimize non-smooth objectives. This research area is relevant since the need for optimizing non-smooth objective functions arises in many real-world problems and applications such as image and signal restoration, denoising, inpainting, etc. In this talk, we introduce the convex bundle method to solve convex, nonsmooth optimization problems on Riemannian manifolds. Each step of our method is based on a model that involves the convex hull of previously collected subgradients, parallely transported into the current serious iterate. This approach generalizes the dual form of classical bundle subproblems in Euclidean space. Several numerical examples implemented using the Julia package Manopt.jl illustrate the performance of the proposed method and compare it to other non-smooth optimization algorithms.
Time/place: Monday, February 19, 2024, 14:15-15:00, Simastue 656, SBII, Campus Gløshaugen
Speaker: Max Pfeffer (Uni Göttingen, Germany)
Title: Manifold Optimization in Data Science
Abstract: Matrix and tensor factorizations are widely applied in Data Science for dimensionality and noise reduction as well as for feature extraction. Often, additional constraints are imposed on the factors in order to improve uniqueness and interpretability of the results. We consider several specific factorization formats with smooth and nonsmooth constraints that can be computed using techniques from Riemannian optimization. For this, existing methods need to be adapted according to the problem at hand. Furthermore, we aim to apply our methods also for Data Fusion, where several data sets are factorized simultaneously.
Time/place: Thursday, January 18, 2024, 14:15-15:00, Simastue 656, SBII, Campus Gløshaugen
Speaker: Joel Dahne (Uppsala Univ., Sweden)
Title: Highest Cusped Waves for the Burgers-Hilbert and Fractional KdV equations
Abstract: In this talk I'll present a proof of the existence of a periodic highest, cusped, travelling wave solution for the Burgers-Hilbert equation and a family of fractional KdV equations. The proof is to a large extent computer-assisted and the focus will be on how the required inequalities can be proved with the help of the computer.

Previous semesters:

Time/place: Monday, December 4, 2023, 14:15-15:00, room F3, Gamle fysikk, Campus Gløshaugen
Speaker: Boris Buffoni (EPFL, Switzerland)
Title: On minimizers of a functional describing convection patterns
Abstract: The system \[\epsilon A''''+(A^2+ B^2-1) A+(g-1) A B^2=0,\] \[B''+(A^2+ B^2-1) B+(g-1) A^2 B=0,\] where \(\epsilon>0\) is a small parameter, describes orthogonal convective structures in the B\'enard-Rayleigh problem just above the onset of instability. Associated to it is the functional \[\int_{\mathbb R}\Big( \frac \epsilon 2 |A''|^2+\frac 1 2 |B'|^2+\frac 1 4 (A^2+B^2-1)^2+\frac 1 2 (g-1)A^2B^2\Big)dx\] for which minimizers are known to exist. Their behaviour is described as \(\epsilon\rightarrow 0^+\), confirming an inner-outer analysis by P. Manneville and Y. Pomeau and their limiting equation \[A''''+ \left(\sqrt {2/g} (g-1) x + A^2\right) A=0\] for the \(A\)-component (after rescaling).
Time/place: Monday, October 30, 2023, 14:15-15:00, room F3, Gamle fysikk, Campus Gløshaugen
Speaker: Rainey Lyons (Karlstads Univ., Sweden)
Title: Analysis and simulation of morphology formation in ternary mixtures
Abstract: We study the ability of a coupled nonlocal system of two quasilinear parabolic partial differential equations to produce morphologies. This system is derived in the literature as the rigorous hydrodynamic limit of a suitably scaled interacting particle system of Blume-Capel-type interactions driven by Kawasaki dynamics. Such a model has the potential to describe the evolution of two interacting species of polymers contained in a sea of solvent. In this talk, we will, via simulation, discuss the potential of the model to produce morphologies and the effect of the solvent ratio on their coarsening rates. Analytically, we will explore the well-posedness of the model, regularity of solutions, and the preservation of a physically meaningful inequality. Finally, we will explore potential adjustments to the model so that the evaporation of the solvent can be considered, a desired aspect in many applications. This research was funded in part by the Carl Tryggers Stiftelse grant no. CTS: 21:1656.
Time/place: Wednesday, October 4, 2023, 11:15-12:00, zoom
Speaker: Marius Zeinhofer (SIMULA, Oslo)
Title: Natural Gradient Methods for Physics Informed Machine Learning
Abstract: We discuss natural gradient methods as a promising choice for the training of physics-informed neural networks (PINN) and the deep Ritz method. As a main motivation, we show that the update direction in function space resulting from the energy natural gradient corresponds to the Newton direction modulo an orthogonal projection onto the model’s tangent space. Empirical results demonstrate that natural gradient optimization is able to produce highly accurate solutions in the PINN approach with errors several orders of magnitude smaller than what is obtained when training PINNs with standard optimizers like gradient descent, Adam or BFGS, even when those are allowed significantly more computation time.
Time/place: Monday, October 2, 2023, 14:15-15:00, room F3, Gamle fysikk, Campus Gløshaugen
Speaker: Long Pei (Sun Yat-sen Univ., China)
Title: The symmetry of periodic travelling waves for weakly dispersive equations
Abstract: We will present results on the a priori symmetry of travelling wave solutions of a general class of weakly dispersive equations. In particular, a reflection criterion, which does not presuppose a monotone structure on the wave profile, will be introduced. The result applies not only to smooth travelling wave solutions but also to those with singular crests around which some cancellation structure appears, including in particular waves with peaks or cusps. In addition, we talk about the general relation between symmetric structure and the steady structure of solutions to general evolution equations.
Time/place: Wednesday, September 6, 2023, 13:15-14:00, Simastuen 656, Campus Gløshaugen
Speaker: Sami Al-Izzi (UiO)
Title: A twist on active membranes
Abstract: Living systems are chiral on multiple scales, from constituent biopolymers to large scale morphology, and their active mechanics is both driven by chiral components and serve to generate chiral morphologies. I will describe the mechanics of active fluid membranes in coordinate-free form, with a focus on chiral contributions to the stress. These generate geometric 'odd elastic' forces in response to mean curvature gradients but are directed perpendicularly. As a result, they induce tangential membrane flows that circulate around maxima and minima of membrane curvature. When the normal viscous force amplifies perturbations the membrane shape can become linearly unstable giving rise to shape instabilities controlled by an active Scriven-Love number. I will describe examples for spheroids, membrane tubes and helicoids, discussing the relevance and predictions such examples make for a variety of biological systems from the sub-cellular to tissue level. Time permitting, I will also discuss the relation between these active isotropic membranes and active nematics liquid crystals.
Time/place: Friday, August 25, 2023, 11:15-12:00, Kjelhuset: KJL23, Campus Gløshaugen
Speaker: Jifeng Chu (Shanghai Normal Univ., China)
Title: Sharp bounds for eigenvalues of Camassa-Holm equations
Abstract: We present some recent results on sharp bounds for periodic eigenvalues, Dirichlet eigenvalues of a spectral problem derived from the famous Camassa-Holm equations, given the \(L^1\) norm of the potentials.
Time/place: Friday, June 23, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Bastian Hilder (Lund University)
Title: Moving modulating pulse and front solutions in a FPU-type model
Abstract: Consider a nonlinear chain of coupled oscillators with nearest and next-to-nearest neighbor interaction. This is a direct generalization of the classical Fermi-Pasta-Ulam (FPU) lattice. In this talk, I will discuss the existence of moving modulating front and pulse solutions. Specifically, these solutions are traveling waves and have small converging tails at infinity. Additionally, their amplitude can be approximated by solitary wave solutions of the Nonlinear Schrödinger (NLS) equation. I will outline the idea of the proof, which is based on spatial dynamics and center manifold theory, and point out connections to related results for the water wave problem. The talk is based on joint work with Björn de Rijk and Guido Schneider.
Time/place: Friday, June 2, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Erik Jansson (Chalmers Univ.)
Title: A Lie–Poisson preserving spatial discretization of the spherical EPDiff equation
Abstract: The EPDiff equation is well-known in the field of shape analysis. It arises as the geodesic equation of the \(H^k\)-metric on the group of diffeomorphisms. In this presentation, we propose a discretization scheme for the EPDiff equation that preserves its Lie–Poisson structure. This method is based on Berezin–Toeplitz quantization and builds on foundational work by Hoppe and Yau. The scheme was used by Vladimir Zeitlin to discretize the incompressible Euler equations, and is known in the hydrodynamics community as Zeitlin's model. We describe the underlying geometry of the equations and the quantization scheme. We provide some numerical examples and provide an outlook on how this discretization may be used in fields such as hydrodynamics, optimal transport, and shape analysis.
Time/place: Friday, May 5, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Douglas Svensson Seth (NTNU)
Title: Internal doubly periodic gravity-capillary waves with vorticity
Abstract: In this talk we consider a multi-fluid system in three dimensions with an arbitrary number of free interfaces. We prove existence of steady gravity-capillary waves where the fluids can be prescribed non-zero vorticity. This result is an multi-parameter bifurcation theorem for small amplitude waves given in two variants: a first guaranteeing existence under some general parameter assumptions; and a second specific, but less exhaustive theorem, for which the assumptions can easily be verified. The latter also yields existence of both in-phase and off-phase waves in the different Interfaces.
Time/place: Friday, April 28, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Adrien Laurent (UiB)
Title: The aromatic bicomplex for the exact numerical preservation of invariant measures
Abstract: For the approximation of ODEs and ergodic SDEs, one can create schemes that preserve the volume or the invariant measure up to any high-order. One then wonders if there exists a scheme that preserves the volume/the invariant measure exactly. While B-series (resp. exotic B-series) are used to represent the Taylor expansion of the solution of ODEs (resp. SDEs), aromatic B-series (resp. exotic aromatic B-series) appear in the creation of integrators that preserve the invariant measure of ODEs (resp. ergodic SDEs). In this talk, we define aromatic forms and the aromatic bicomplex, in the spirit of the variational bicomplex in differential geometry. We prove the exactness of this bicomplex and use it to give an explicit description of volume-preserving methods. We use this description to show that no aromatic modification of Runge-Kutta methods preserves volume in general and to discuss the possible ansatz for creating methods that preserve the invariant measure exactly.
Time/place: Friday, April 14, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Miles Wheeler (Univ. of Bath, UK)
Title: Desingularization and global continuation for hollow vortices
Abstract: Hollow vortices are regions of constant pressure with finite circulation embedded into an otherwise irrotational flow. We prove that non-degenerate configurations of point vortices which steadily translate or rotate can be desingularized into analogous configurations of hollow vortices. The resulting local curves of solutions can then be extended using global bifurcation theory. As examples, we give what appear to be the first rigorous existence results for rotating hollow vortex pairs and for stationary hollow vortex tripoles.
This is joint work with Ming Chen and Samuel Walsh.
Time/place: Friday, March 24, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Thomas Christiansen (NTNU)
Title: A numerical algorithm for \(\alpha\)-dissipative solutions of the Hunter–Saxton equation
Abstract: We present a convergent numerical method for \(\alpha\)-dissipative solution of the Cauchy problem for the Hunter–Saxton equation \[u_t(t, x) + uu_x(t, x) = \frac{1}{4}\left(\int_{-\infty}^xu_x^2(t, z)dz - \int_{x}^{\infty}u_x^2(t, z)dz \right), \quad u|_{t=0} = u_0.\] The method is based on applying a tailor-made projection operator to the initial data, and then solving exactly using the generalized method of characteristics. Thus, the projection step is the only step that introduces any approximation error.
We elaborate on the construction of this projection operator, and indicate how to prove convergence towards \(\alpha\)-dissipative solutions. Finally, we illustrate how the method performs on a few numerical examples.
This presentation is based on joint work with Katrin Grunert (NTNU), Anders S. Nordli (UIT) and Susanne Solem (NMBU). For further details, see our recent arXiv preprint: "A numerical algorithm for $\alpha$-dissipative solutions of the Hunter–Saxton equation". arXiv: 2303.08763
Time/place: Friday, March 3, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Ganesh Kiran Vaidya (NTNU)
Title: On the accuracy of the finite volume approximations to nonlocal conservation laws
Abstract: This talk focuses on the error analysis for a certain class of monotone finite volume schemes approximating nonlocal scalar conservation laws, modelling traffic flow/crowd dynamics. We first discuss a Kuznetov type lemma for such PDEs and thereby prove that the finite volume approximations converge to the entropy solution at an optimal rate of \(\sqrt{\Delta t}\) in \(L^1(\mathbb{R})\).
This is a joint work with Aekta Aggarwal (IIM-Indore) and Helge Holden (NTNU).
Time/place: Friday, February 10, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Frédéric Valet (UiB)
Title: Collision of two solitary waves for the Zakharov-Kuznetsov equation
Abstract: The Zakharov-Kuznetsov (ZK) equation in dimension 2 is a generalization in plasma physics of the one- dimensional Korteweg de Vries equation (KdV). Both equations admit solitary waves, that are solutions moving in one direction at a constant velocity, vanishing at infinity in space. When two solitary waves collide, two phe- nomena can occur: either the structure of two solitary waves is conserved without any loss of energy and change of sizes (elastic collision), or the structure is lost or modified (inelastic collision). As a completely integrable equation, KdV only admits elastic collisions. The goal of this talk is to explain the collision phenomenon for two solitary waves having almost the same size for ZK, and to describe the inelasticity of the collision. The talk is based on current works with Didier Pilod.
Time/place: Friday, January 27, 2023, 11:15-12:00, Room 656, SBII, Campus Gløshaugen
Speaker: Peter Lindqvist (NTNU)
Title: The Eigenvalue Problem for the Infinity - Laplace Operator
Abstract: The eigenvalue problem associated with the \(\infty\) - Laplace Operator has fascinating properties. The corresponding equation \[\max\left\{\Lambda - \frac{|\nabla u|}{u},\, \sum_{i,j=1}^{n}\frac{\partial u}{\partial x_{i}} \, \frac{\partial u}{\partial x_{j}}\,\frac{\partial ^{2} u}{\partial x_{i} \partial x_{j}}\right\}\,=\, 0\] has a positive viscosity solution \(u \in W^{1,\infty}_{0}(\Omega)\) if and only if \[\Lambda =\Lambda_{\infty} \equiv \frac{1}{\underset{x \in \Omega}{\max}\,\rm{dist}(x,\partial \Omega)}.\] In convex domains in the plane, the streamlines are known to exhibit an arcane pattern. \(\Lambda_{\infty}\) is the first eigenvalue. The higher eigenvalues are, as it were, related to sphere packing.
Time/place: Wednesday, December 14, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Jun Xue (NTNU)
Title: Periodic Travelling water waves with point vortices
Abstract: In this talk, we prove the existence of two-dimensional small-amplitude periodic travelling gravity-capillary water waves on a finite depth, with a finite number of point vortices along vertical lines. In particular, we consider the periodic condition to establish the stream function in specific form. The main result is showed by using a local bifurcation argument for the periodic point vortex problem.
Time/place: Wednesday, December 7, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Johanna Marstrander (NTNU)
Title: Existence of solitary waves in equations with nonlocal nonlinearities via constrained minimization
Abstract: We establish the existence of solitary waves in a class of nonlinear, dispersive evolution equations of the form \(\frac{d}{dt} u + \frac{d}{dx} (Lu + B(u,u)) = 0\). Here \(L\) is a linear Fourier multiplier of typically small fractional positive order, while \(B\) is a bilinear Fourier multiplier, also called pseudo-product. These are fully nonlocal equations, where even product interaction takes place in frequency space. The problem is related to the mathematical theory of water waves, where such operators arise. In this talk, I provide background on these types of nonlocal operators and present an existence proof for smooth solitary wave solutions to the given equation. Methods include constrained minimization, bilinear Fourier estimates, and concentration-compactness.
Time/place: Wednesday, November 30, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Annika Lang (Chalmers University of Technology, Sweden)
Title: Simulation of random fields on Riemannian manifolds.
Abstract: Random fields are important building blocks in spatial models disturbed by randomness such as solutions to stochastic partial differential equations. The fast simulation of random fields is therefore crucial for efficient algorithms in uncertainty quantification. In this talk I present numerical methods for Gaussian random fields on Riemannian manifolds and discuss their convergence. Simulations illustrate the theoretical findings.
This talk is based on joint work with Erik Jansson, Mihály Kovács, and Mike Pereira.
Time/place: Friday, November 25, 2022, 14:15-15:00, Room 656, SBII, Campus Gløshaugen
Speaker: Oliver Sutton (King's College London, UK)
Title: Learning from few examples with nonlinear feature maps
Abstract: Machine Learning and Artificial Intelligence have seemingly become synonymous with Big Data - we imagine neural networks and other tools being trained on vast datasets to learn to perform tasks. But is it possible for these systems to learn from small datasets consisting of just a few (e.g. 5 or 10) examples of a new class of objects?
In this talk we investigate just this question. We build on existing work showing that implicit properties of high dimensional spaces can make learning from few examples possible in certain circumstances using very simple linear classifiers (a so-called blessing of dimensionality). We are particularly interested in whether nonlinear feature maps, i.e. nonlinear mappings of the dataset into much higher dimensional spaces, can accelerate the onset of this blessing of dimensionality, thereby enabling algorithms to learn from few examples. We provide specific relationships between the distributions of the data classes we are learning, the geometry of the feature mappings, and the probability of successfully learning from just a few examples – without catastrophically forgetting any previous learning.
For further details, see our recent arXiv preprint: O. J. Sutton, A. N. Gorban and I. Y. Tyukin (2022). Towards a mathematical understanding of learning from few examples with nonlinear feature maps. https://arxiv.org/abs/2211.03607
Time/place: Wednesday, November 16, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Martin Ludvigsen (NTNU)
Title: Adverserial Non-negative Matrix Factorization for Single Channel Source Separation
Abstract: This talk concerns data-driven methods for solving the Single Channel Source Separation (SCSS) problem, which is an inverse problem that concerns recovering individual source signals from a measured mixed signal. A traditional data-driven approach to solving this problem is to use Non-negative Matrix Factorization (NMF) to learn convex cones that best approximate the distribution of data of the individual sources, and then project mixed signals onto these cones during testing. This approach is particularily useful in audio applications, as well as some imaging applications. Recent advancements in machine learning and learnt regularization suggests using the Wasserstein distance between distributions as a metric for learning structures between distributions of signals. I use this as a framework for deriving a novel NMF method that I call Adverserial Non-Negative Matrix Factorization (ANMF). In contrast to normal NMF, ANMF not only attempts to fit certain signals, but also discriminates against other signals. I show the application of this method for the SCSS problem and provide numerical results. I also discuss how the ideas in this talk generalize to non-linear and non-convex methods, as well as different data settings.
Time/place: Monday, November 7, 2022, 13:15-14:00, Room 656, SBII, Campus Gløshaugen
Speaker: Liviu I. Ignat (Institute of Mathematics Simion Stoilow, Bucharest, Romania)
Title: Asymptotic behavior of solutions for some diffusion problems on metric graphs
Abstract: In this talk we present some recent result about the long time behavior of the solutions for some diffusion processes on a metric graph. We study evolution problems on a metric connected finite graph in which some of the edges have infinity length. We show that the asymptotic behaviour of the solutions of the heat equation (or even some nonlocal diffusion problems) is given by the solution of the heat equation, but on a star shaped graph in which there is only one node and as many infinite edges as in the original graph. In this way we obtain that the compact component that consists in all the vertices and all the edges of finite length can be reduced to a single point when looking at the asymptotic behaviour of the solutions. We prove that when time is large the solution behaves like a gaussian profile on the infinite edges. When the nonlinear convective part is present we obtain similar results but only on a star shaped tree.
This is a joint work with Cristian Cazacu (University of Bucharest), Ademir Pazoto (Federal University of Rio de Janeiro), Julio D. Rossi (University of Buenos Aires) and Angel San Antolin (University of Alicante).
Time/place: Tuesday, October 25, 2022, 14:15-15:00, Room KJL22, Kjelhuset, Campus Gløshaugen
Speaker: Oliver Sander (Tech. Univ. Dresden, Germany)
Title: Geometric finite elements for rod and shell models
Abstract: Thin sheets show typical wrinkling patterns when loaded in shear. The simulation of such patterns is challenging, because they are highly nonlinear phenomena. One contender for a mathematical model for such wrinkling are geometrically nonlinear shell models of Reissner-Mindlin or Cosserat type. They describe configurations of the sheet as the deformation of a surface together with a field of directions or microrotations. These additional degrees of freedom allow to model transverse shear.
The numerical treatment of such models is difficult, because the set of admissible configurations of Sobolev smoothness does not form a vector space. In the talk, we present generalizations of standard finite element methods to function spaces with a nonlinear codomain. These generalized FE spaces are \(H^1\) conforming, and we show optimal a priori discretization error bounds. When used to simulate wrinkling patterns of a Cosserat sheet model, we obtain good quantitative agreement with experimental results from the literature.
Time/place: Wednesday, October 19, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Oana Silvia Serea (Western Norway Univ. of Applied Sciences)
Title: Reflected Dynamics: Viscosity Analysis for \(\mathbb{L}^{\infty}\) Cost, Relaxation and Abstract Dynamic Programming.
Abstract: We study an optimal control problem consisting in minimizing the \(\mathbb{L}^{\infty}\) norm of a Borel measurable cost function, in finite time, and over all trajectories associated with a controlled dynamics which is reflected in a compact prox-regular set. The first part of the presentation provides the viscosity characterization of the value function for uniformly continuous costs. The second part is concerned with linear programming formulations of the problem and the ensued by-products as e.g. dynamic programming principle for merely measurable costs.
Time/place: Wednesday, October 12, 2022, 14:15-15:00, zoom
Speaker: Alessandro Scagliotti (TU Munich, Germany)
Title: Ensemble optimal control: ResNets, diffeomorphisms approximation and Normalizing Flows
Abstract: In the last years it was observed that Residual Neural Networks (ResNets) can be interpreted as discretizations of control systems, bridging ResNets (and, more generally, Deep Learning) with Control Theory. In the first part of this seminar we formulate the task of a data-driven reconstruction of a diffeomorphism as an ensemble optimal control problem. In the second part we adapt this machinery to address the problem of Normalizing Flows: after observing some samplings of an unknown probability measure, we want to (approximately) construct a transport map that brings a “simple” distribution (e.g., a Gaussian) onto the unknown target distribution. In both problems we use tools from \(\Gamma\)-convergence to study the limiting case when the size of the data-set tends to infinity.
Time/place: Wednesday, October 5, 2022, 14:15-15:00, zoom
Speaker: Nicola Sansonetto (Univ. of Verona, Italy)
Title: Ball on a rotating 'cup'
Abstract: In this seminar, I will introduce and study the class of nonholonomic mechanical systems formed by a heavy symmetric ball that rolls without sliding on a surface of revolution, which is either at rest or rotates about its (vertical) figure axis with uniform angular velocity \(\Omega\). I will show that the reduced system is Hamiltonizable even if \(\Omega\) is not vanishing and, exploiting the recently introduced ‘moving energy’, we give sufficient conditions on the profile of the surface that ensure the periodicity of the reduced dynamics and hence the quasi-periodicity of the unreduced dynamics on tori of dimension up to three. Furthermore, we determine all the equilibria of the reduced system, which are classified in three distinct families, and determine their stability properties.
The seminar is mainly based on the collaboration with F. Fasso and M. Dalla Via "On the dynamics of a heavy symmetric ball that rolls without sliding on a uniformly rotating surface of revolution.” J Nonlinear Sci 32, 84 (2022). https://doi.org/10.1007/s00332-022-09842-5
Time/place: Monday, October 3, 2022, 14:15-15:00, zoom
Speaker: Jaume Alonso (TU Berlin, Germany)
Title: A 3D generalisation of QRT and the construction of integrable birational systems
Abstract: When completely integrable Hamiltonian systems are discretised, the resulting discrete-time systems are often no longer integrable themselves. This is the so-called problem of integrable discretisation. Two known exceptions to this situation in 3D are the so-called Kahan discretisations of the Euler top and some particular cases of the Zhukovski-Volterra gyrostat, both birational maps of degree 3. The integrals of these systems define pencils of quadrics. By analysing the geometry of these pencils, we develop a framework that generalises QRT maps and QRT roots to 3D, which allows us to create new integrable maps as a composition of two involutions. We show that under certain geometric conditions, the new maps become of degree 3. We use these results to create new families of discrete integrable maps and we solve the problem of integrability of the Zhukovski- Volterra gyrostat with two \(\beta\)'s.
This is a joint work with Yuri Suris and Kangning Wei.
Time/place: Monday, September 19, 2022, 14:15-15:00, zoom
Speaker: Yuri Suris (TU Berlin, Germany)
Title: How geometry helps to repair non-integrable Kahan discretizations
Abstract: Kahan discretization method is applicable to any system of ordinary differential equations on \(R^n\) with a quadratic vector field. When applied to integrable systems, Kahan discretization preserves integrability much more frequently than one would expect a priori, however not always. In some cases where the original recipe fails to preserve integrability, one can nevertheless adjust the Kahan discretization to ensure its integrability. Such adjustments are highly non-trivial and it is practically impossible to find them via guesswork. I will explain how to find these adjustments in a systematic way, via a thorough investigation of the underlying geometry.
Time/place: Wednesday, September 14, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Eugen Bronasco (Univ. of Geneva, Switzerland)
Title: Exotic B-series and S-series: algebraic structures and order conditions for invariant measure sampling
Abstract: B-series and generalizations are a powerful tool for the analysis of numerical integrators. An extension named exotic aromatic B-series was introduced to study the order conditions for sampling the invariant measure of ergodic SDEs. In this talk, we analyze the algebraic structures related to exotic B-series and S-series. Applying this algebraic framework to the derivation of order conditions for a class of stochastic Runge-Kutta methods, we present a multiplicative property that ensures some order conditions to be satisfied automatically.
Time/place: Wednesday, September 7, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Katerina Nik (Univ. of Vienna, Austria)
Title: On a three-dimensional model for MEMS with a hinged upper plate
Abstract: An idealized electrostatic microelectromechanical system (MEMS) consists of a rigid ground plate above which a thin elastic plate is suspended. The elastic plate is assumed to be hinged on its boundary. Applying a voltage difference between the two plates induces a Coulomb force that deforms the elastic plate. The corresponding mathematical model couples a fourth-order parabolic equation for the vertical deformation of the elastic plate to the harmonic electrostatic potential in the free domain between the two plates.
In this talk, I will present some recent results on local and global well-posedness of the model as well as on existence and non-existence of stationary solutions.
Time/place: Wednesday, August 31, 2022, 14:15-15:00, Room F2, Gamle Fysikk, Campus Gløshaugen
Speaker: Louis Emerald (Univ. of Rennes 1, France)
Title: On the rigorous justification of the Whitam equations
Abstract: The Whitham equations was introduced by G. Whitham in 1967 to study the Stokes waves of maximal amplitude and the wavebreaking phenomenon. It is a modification of the Korteweg-de Vries equations which have the same dispersion relation as the general water waves model. In this talk we will see a method to derive the Whitham equations directly from the water waves equations using a generalisation of Birkhoff's normal form algorithm. It will allow us to rigorously justify the Whitham equations as an asymptotic shallow water model for water waves at a nontrivial order of precision.
Time/place: Tuesday, August 16, 2022, 14:15-15:00, Room 734, SBII, Campus Gløshaugen
Speaker: Robert Klöfkorn (Lund Univ., Sweden)
Title: Solving PDEs with DUNE-FEM
Abstract: For about two decades the Distributed and Unified Numerics Environment (DUNE:https://dune-project.org) has been an active part in the scientific development of computational software and technology and it's C++ routines are the basis for several other well established open source projects, for example, DuMux (https://dumux.org) or the Open Porous Media Initiative (https://opm-project.org).
Although the C++ interfaces of DUNE are highly flexible and customizable, a solid knowledge of C++ is necessary to make use of this powerful tool.
In this talk we give an overview on recent development towards a Python interface for DUNE and in particular DUNE-FEM, a module which provides highly efficient implementations of hp-adaptive Discontinuous Galerkin (DG) methods for solving a wide range of non linear partial differential equations. Providing easier user interfaces based on Python and the Unified Form Language (UFL) opens DUNE-FEM to a broader audience, for example, Bachelor and Master students. This talk will also briefly discuss how Python and DUNE are embedded in teaching of Numerical Analysis courses at Lund University.
Time/place: Monday, June 27, 2022, 14:15-15:00, Zoom
Speaker: Salvador Ortiz-Latorre (UiO)
Title: High order discretizations to the nonlinear filtering problem
Abstract: The solution of the continuous time stochastic filtering problem can be represented as a ratio of two expectations of certain functionals of the signal process that are parametrized by the observation path. In this talk I will introduce a class of discretization schemes of these functionals of arbitrary order. For a given time interval partition, we construct discretization schemes with convergence rates that are proportional with the mth power of the mesh of the partition for arbitrary natural number m. The result generalizes the classical work of Picard, who introduced first order discretizations to the filtering functionals. Moreover, the result paves the way for constructing high order numerical approximation for the solution of the filtering problem.
Time/place: Monday, June 20, 2022, 14:15-15:00, Zoom
Speaker: Davide Murari (NTNU)
Title: Neural networks modelled by dynamical systems
Abstract: Neural networks have gained much interest because of their effectiveness in many applications. However, they usually lack an imposed a priori structure, and their properties are not generally well understood. When one knows that the target function to approximate or the data being processed has some properties, it might be desirable to reproduce them into the design of the neural network. This talk presents a framework that makes ODEs and numerical methods work together to model neural networks having prescribed properties. The ODE models we work with are non-autonomous dynamical systems with piecewise (in time) autonomous dynamics. First, we report a result about the approximation capabilities of a class of networks generated in this way. After some examples of network preserving networks, we focus on those with controllable and prescribable Lipschitz constant. We conclude showing how these last architectures allow to get mathematical guarantees of adversarial robustness.
Time/place: Monday, June 13, 2022, 14:15-15:00, Zoom
Speaker: Adrian Celestino (NTNU)
Title: Magnus expansion and series of rooted trees
Abstract: The Magnus expansion was introduced in the 1950s with the aim of expressing the solution of non-autonomous linear initial-value problem \(Y' = A(t)Y\), \(Y(0)=I\), where \(A(t)\) and \(Y=Y(t)\) are \(n \times n\) matrix-valued functions, as a matrix-exponential of the function \(\Omega[A](t)\). The latter is characterized as the solution of a highly non-linear differential equation with initial value \(\Omega[A](0)=0\), and can be expressed as a series \(\Omega[A](t) = \sum_{k\geq1} \Omega_k[A](t)\) satisfying nice geometric properties. On the other hand, rooted trees have appeared in the works of Iserles and Nørsett as a means of expressing the terms \(\Omega_k[A](t)\) of the Magnus expansion, and also in the works of Butcher on writing the solution of non-linear differential equations as series indexed by rooted trees. In this talk, we will study the pre-Lie counterpart of the Magnus expansion, first proposed by Agrachev and Gamkrelidze in control theory, together with an approach to computing it using rooted trees. Based on joint work with F. Patras (arXiv:2203.11968).
Time/place: Monday, May 30, 2022, 14:15-15:00, Zoom & Room 656, SBII, Campus Gløshaugen
Speaker: Shipra Singh (NTNU)
Title: A Multidimensional Approach to Non-Cooperative Strategic Games
Abstract: The talk intends to give a concise view of my following ongoing research work; this research aims to study (1) the formulation of a multidimensional generalized Nash equilibrium problem, (2) its equivalent relation with a multidimensional quasi-variational inequality problem, (3) existence of the formulated equilibria, (4) the reformulation of a spot electricity market problem in the terms of the proposed multidimensional generalized Nash equilibrium problem, (5) and numerical experiments by using the projected dynamical system theory. Some dynamic generalizations of the generalized Nash equilibrium problem have already been investigated in the literature, for instance, a time-dependent generalized Nash equilibrium problem is studied for a dynamic spot electricity market problem involving a scalar time parameter. However, even with the knowledge of every constraint and argument with respect to the time parameter in an economic problem, it may happen that other economic parameters may also affect the values of constraints and arguments pertaining to this economic problem. For example, it is quite natural, generally speaking, that the production of electricity depends on several economic parameters, not just on the time parameter related to the given time period of the production of electricity. For example, parameters related to the capacity of a generator to produce electricity, parameters related to manpower, and other economic parameters may also affect the production of electricity and the earned profit of each producer participating in the electricity market. To fully understand the performance of the multidimensional parameter of evolution (multi-parameters of evolution) in non-cooperative strategic game problems, this research intends to study a multidimensional generalized Nash equilibrium problem with an application in electricity markets.
Time/place: Monday, May 23, 2022, 14:15-15:00, Zoom & Room 656, SBII, Campus Gløshaugen
Speaker: Jörg Weber (Lund Univ., Sweden)
Title: Large-amplitude steady gravity water waves with general vorticity and critical layers
Abstract: While the research on water waves modelled by Euler's equations has a long history, mainly in the last two decades travelling periodic rotational waves have been constructed with mathematical rigour by means of bifurcation theorems. In this talk, I will present a new reformulation of this travelling periodic water wave problem in two dimensions. Using conformal mappings and a new Babenko-type reformulation of Bernoulli's equation, the problem is equivalently cast into the form “identity plus compact”, which is amenable for Rabinowitz' global bifurcation theorem. The main advantages (and the novelty) of this new reformulation are that no simplifying restrictions on the geometry of the surface profile and no simplifying assumptions on the vorticity distribution (and thus no assumptions regarding the absence of stagnation points or critical layers) have to be made. Within the scope of this new formulation, local and global families of solutions, bifurcating from laminar flows with a flat surface, are constructed. Moreover, I will further discuss the condition for local bifurcation and the possible alternatives for the global set of solutions, as well as their nodal properties. This is joint work with Erik Wahlén.
Time/place: Monday, May 2, 2022, 14:15-15:00, Zoom & Room 656, SBII, Campus Gløshaugen
Speaker: Markus Köbis (NTNU)
Title: Time-Optimal Adaptation in Metabolic Network Models
Abstract: We introduce time-optimal adaptation (TOA), a constraint-based modeling approach for the simulation of microbial life where the objective lies in reaching a pre-defined goal state in as short time as possible. Mathematically, TOA is a time-optimal control problem for differential-algebraic equations. After a short introduction into the problem class, we will propose a unifying mathematical framework for many existing constraint-based methods within systems biology and exemplify how it captures established resource allocation-type methods like resource balance analysis and dynamic enzyme-cost flux-balance analysis. We will then shortly discuss some numerical details, before TOA will be illustrated using a coarse-grained self-replicator model. Here, we show that TOA can explain phenomena such as storage accumulation in bacteria without taking competition and/or time-varying environments into account. The talk is based on a joint work with Alexander Bockmayr (FU Berlin) and Ralf Steuer (HU Berlin).
Time/place: Monday, April 25, 2022, 14:15-15:00, Zoom
Speaker: Adeline Fermanian (Mines ParisTech, France)
Title: Framing RNN as a kernel method: A neural ODE approach
Abstract: Building on the interpretation of a recurrent neural network (RNN) as a continuous- time neural differential equation, we show, under appropriate conditions, that the solution of a RNN can be viewed as a linear function of a specific feature set of the input sequence, known as the signature. This connection allows us to frame a RNN as a kernel method in a suitable reproducing kernel Hilbert space. As a consequence, we obtain theoretical guarantees on generalization and stability for a large class of recurrent networks. Our results are illustrated on simulated datasets.
Time/place: Monday, April 4, 2022, 14:15-15:00, Zoom
Speaker: Ronny Bergmann (NTNU)
Title: Splitting Methods for Non-smooth Optimization on Manifolds
Abstract: Many modern image or data acquisitions capture data whose values lie on a Riemannian manifold. For example, phase-valued data appears in interferometric synthetic aperture radar (InSAR), diffusion tensors represented as 3-by-3 symmetric positive definite matrices appear in magnetic resonance imaging (DT-MRI), and in electron backscattered diffraction (EBSD), each measurement point is an orientation of a crystal lattice.
This leads to data given on a nonlinear space, that has some geometric structure, i.e. at each pixel we have a data point from some Riemannian manifold. Variational methods are a popular tool in image processing, where the most prominent variational model for denoising is total variation regularization. In general, we obtain a high-dimensional, non-smooth optimization problem on a (structured or power) manifold. In classical image processing for color images, splitting methods are a widely used technique to derive fast algorithms.
This talk gives an overview of (Euclidean) splitting methods and how they can be transferred to nonsmooth optimization tasks on a Riemannian manifold.
Time/place: Monday, March 28, 2022, 14:15-15:00, Zoom
Speaker: Fredrik Hildrum (NTNU)
Title: Periodic Hölder waves in nonsmooth fractional KdV equations
Abstract: Joint with Jun Xue (NTNU)
We prove the existence of highest, cusped, periodic travelling-wave solutions with exact and optimal \({ \alpha }\)-Hölder continuity in a family of negative-order fractional Korteweg–de Vries equations of the form \[ u_t + (|\mathrm{D} |^{- \alpha} u + n(u) )_x = 0 \] for every \({ \alpha \in (0, 1) }\) with homogeneous Fourier multiplier \({| \mathrm{D} |^{ - \alpha} }\). We tackle nonlinearities \({ n(u) }\) of the type \({ | u |^p }\) or \({ u | u |^{p - 1} }\) for all real \({ p > 1 }\), and show that when \({ n }\) is odd, the waves also feature antisymmetry and thus contain inverted cusps. The analysis for nonsmooth \({ n }\) is applicable to other negative-order nonlocal equations. Both the construction of highest antisymmetric waves and the regularisation of nonsmooth terms to an analytic bifurcation setting, we believe are new in this context.
Time/place: Monday, March 21, 2022, 14:15-15:00, Zoom & Room 656, SBII, Campus Gløshaugen
Speaker: Joscha Diehl (Univ. of Greifswald, Germany)
Title: Multiparameter iterated sums
Abstract: Iterated sums (and integrals) have proven very beneficial in time series analysis. I show how ideas from this one-parameter setting can be used to study multiparameter data, for example, images. I will also sketch the Hopf algebraic background. This is a work in progress with Leonard Schmitz (University of Greifswald).
Time/place: Monday, March 14, 2022, Zoom & Room 656, SBII, Campus Gløshaugen
13:15-14:00
Speaker: Dietmar Hömberg (WIAS, Berlin, Germany)
Title: A phasefield approach to two-scale topology optimization
Abstract: Subject of my presentation is a novel approach for optimizing both the macroscopic shape and the porous mesoscopic structure of components. In the first part of my presentation I will introduce the concept of phasefield based topology optimization.
The second part of my presentation is devoted to two-scale topology optimization. The key feature here is the introduction of an additional local volume control (LVC), which allows to adjust the desired spatial scales. The main novelty is that the radius of the LVC may depend both on space and a local stress measure. This allows for creating optimal topologies with heterogeneous mesostructures enforcing any desired spatial grading and accommodating stress concentrations by stress dependent pore size. I will present some analytical results for the resulting optimal control problem and conclude with numerical results showing the versatility of our approach for creating optimal macroscopic designs with tailored mesostructures.
14:15-15:00
Speaker: Ralf Zimmermann (University of Southern Denmark)
Title: Efficient algorithms for computing the Riemannian normal coordinates on the Stiefel manifold
Abstract: We address the problem of computing Riemannian normal coordinates on the real, compact Stiefel manifold of orthonormal frames. The Riemannian normal coordinates are based on the so-called Riemannian exponential and the associated Riemannian logarithm map and enable to transfer almost any computational procedure to the realm of the Stiefel manifold. To compute the Riemannian logarithm is to solve the (local) geodesic endpoint problem. Instead of restricting the consideration to geodesics with respect to a single selected metric, we consider a family of Riemannian metrics introduced by Hüper, Markina and Silva-Leite that includes the Euclidean and the canonical metric as prominent examples.
As main contributions, we provide (1) a unified, structured, reduced formula for the Stiefel geodesics. The formula is unified in the sense that it works for the full family of metrics under consideration. It is structured in the sense that it relies on matrix exponentials of skew-symmetric matrices exclusively. It is reduced in relation to the dimension of the matrices of which matrix exponentials have to be calculated. We provide (2) a unified method to tackle the geodesic endpoint problem numerically and (3) we improve the existing Riemannian log algorithm under the canonical metric in terms of the computational efficiency. The findings are illustrated by means of numerical examples.
This is joint work with Knut Hüper from Julius-Maximilians-Universität, Würzburg.
Time/place: Monday, March 7, 2022, 14:15-15:00, Zoom
Speaker: Christa Cuchiero (Univ. of Vienna, Austria)
Title: Randomized signature for approximation of dynamic processes
Abstract: We consider the question whether the time evolution of controlled (or stochastic) differential equations on general state spaces can be arbitrarily well approximated by regressions on certain randomly chosen dynamical systems of moderately high dimension. On the one hand this is motivated by paradigms of reservoir computing, on the other hand by signature methods from rough paths theory. Appropriately interpreted this yields provable approximation results for generic dynamic processes by regressions on states of random and thus untrained dynamical systems. The results have important implications for efficiency of training.
The talk is based on joint work with Lukas Gonon, Lyudmila Grigoryeva, Juan-Pablo Ortega and Josef Teichmann.
Time/place: Monday, February 21, 2022, 14:15-15:00, Zoom
Speaker: David Cohen (Chalmers Univ., Sweden)
Title: Efficient discretisations of stochastic Hamiltonian and Poisson systems
Abstract: We start by recalling classical results on time discretisations of (deterministic) Hamiltonian and Poisson systems. We will then randomly perturb such systems, and present and analyse various time integrators for an efficient simulation of stochastic Hamiltonian and Poisson systems.
The presentation is based on joint works with C-E. Bréhier, C. Chen, R. D'Ambrosio, K. Debrabant, T. Jahnke, A. Lang, A. Rößler and G. Vilmart.
Time/place: Monday, February 14, 2022, 14:15-15:00, Zoom & Room 656, SBII, Campus Gløshaugen
Speaker: Charles Curry (NTNU)
Title: Stochastic quantization: theory and practice
Abstract: We present an introduction to stochastic quantization, demonstrating how SDEs and SPDEs are useful tools in understanding phenomena ranging from analysis of Quantum Field Theories to solid state physics. In doing so we discuss aspects of renormalization and conclude with an outlook on numerical computations and challenges arising in generalizations of present techniques, particularly to gauge theories such as QED and QCD.
Time/place: Monday, February 7, 2022, 14:15-15:00, Zoom
Speaker: Avi Mayorcas (Univ. of Cambridge, UK)
Title: Blow-up criteria for SPDE models of chemotaxis
Abstract: Chemotaxis and related phenomena have been an active area of mathematical research since statistical and PDE models were first proposed by C. Patlak (’53) and E. Keller & L. Segel (’71). A common feature of these models is possibility of finite time blow-up of solutions under given model parameters. From a biological perspective this is often interpreted as an aggregation effect. From a mathematical point of view the blow-up phenomena are complex enough to be non-trivial but simple enough to be relatively well understood; providing useful insight into less well understood blow-up phenomena. Stochastic PDE (SPDE) models provide a natural extension to these models and in this talk I will discuss new results (obtained with M. Tomašević) regarding blow-up for a class of conservative, SPDE models of chemotaxis.
Time/place: Monday, January 24, 2022, 14:15-15:00, Zoom
Speaker: Milo Viviani (Scuola Normale Superiore, Pisa, Italy)
Title: Canonical Scale Separation in 2D Incompressible Hydrodynamics
Abstract: The fundamental rules governing a two-dimensional inviscid incompressible fluid are simple. Yet, to characterize the long-time behavior is a knotty problem. The fluid's motion is described by Euler's equations: a non-linear Hamiltonian system with infinitely many conservation laws. In both experiments and numerical simulations, coherent vortex structures, or blobs, emerge after an initial stage. These formations dominate the large-scale dynamics, but small scales also persist.
In his classical work, Kraichnan qualitatively describes a forward cascade of enstrophy into smaller scales and a backward cascade of energy into larger scales. Previous attempts to model Kraichnan's double cascade use filtering techniques that enforce separation from the outset. Here we show that Euler's equations possess an intrinsic, canonical splitting of the vorticity function. The splitting is remarkable in four ways:
(i) it is defined solely via the Poisson bracket and the Hamiltonian,
(ii) it characterizes steady flows,
(iii) without imposition it yields a separation of scales, enabling the dynamics behind Kraichnan's qualitative description,
(iv) it accounts for the "broken line” in the power law for the energy spectrum (observed in both experiments and numerical simulations).
The splitting originates from Zeitlin's truncated model of Euler's equations in combination with a standard quantum-tool: the spectral decomposition of Hermitian matrices. In addition to theoretical insight, the scale separation dynamics could be used for stochastic model reduction, where small scales are modelled by multiplicative noise. Preprint available at: https://arxiv.org/abs/2102.01451
Time/place: Monday, December 13, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Karl K. Brustad (NTNU)
Title: One-dimensional Dirichlet problem for nonlocal equations
Abstract: Explicit solutions to the Dirichlet problem for a class of mean value equations on the real line are derived. This is a special case of a problem considered in joint work with Peter Lindqvist and Juan Manfredi, where existence, uniqueness, and uniform convergence of the nonlocal solutions towards the solution of the corresponding PDE was established.
Time/place: Monday, December 6, 2021, 14:15-15:00, Zoom
Speaker: Subbarao Venkatesh Guggilam (Old Dominion Univ., Norfolk, USA)
Title: Additive Static Feedback Connection: A Chen–Fliess Series Viewpoint
Abstract: Michel Fliess, in 1981, proved that an analytic dynamical system with input entering linearly has an input-output description written purely in terms of sum of weighted iterated integrals, called Chen-Fliess series. The expressions of iterated integrals can be symbolized by words formed out of noncommutative letters or indeterminates. Hence, the input-output description of such analytic vector fields can be described by a Chen-Fliess series whose underlying descriptor is a formal power series. The problem considered in the seminar is when does a Chen-Fliess series in an additive static feedback connection with a formal static map (definition of the map does not inherently depend on time) yield a closed-loop system with a Chen-Fliess series expansion? One can prove that such a closed-loop system always has a Chen-Fliess series representation. Furthermore, an algorithm based on the Hopf algebras for the shuffle group and the dynamic output feedback group is designed to compute the generating series of the closed-loop system. It is proved that the additive static feedback connection preserves local convergence and relative degree, but a counterexample shows that the additive static feedback does not preserve global convergence in general. The results presented are a part of the Ph.D. dissertation of the speaker. The talk will begin with a slight introduction to Chen-Fliess series.
Time/place: Thursday, December 2, 2021, Zoom & room 656 (SIMA-stuen)
14:15 - 15:00
Speaker: Olivier Ley (Rennes)
Title: Infinite and finite horizon stochastic Mean Field Games on networks
Abstract: The purpose of the talk is to describe several results obtained in collaboration with Yves Achdou (Paris), Manh-Khang Dao (Rouen) and Nicoletta Tchou (Rennes) about Mean Field Games on general networks. We study the associated coupled system of a Hamilton-Jacobi-Bellman PDE and a Fokker-Planck PDE. The system is complemented with Kirchhoff conditions for the HJB equations and dual transmission conditions for the FP equation at the vertices of the network. When dealing with general general Kirchhoff conditions, the value function associated with the stochastic control problem become discontinuous at the vertices, which is one the main difficulty. We prove the well-posedness of of the system both in the stationary and parabolic case.
15:15-16:00
Speaker: Annalisa Cesaroni (Padova)
Title: Fractional mean curvature flow of entire Lipschitz graphs.
Abstract: I will discuss some results on the fractional mean curvature flow of entire Lipschitz graphs, in particular regularity results, and long time asymptotics. By the level set method, it is possible to describe this evolution by looking at viscosity solutions of a fractional quasilinear parabolic problem. As in the classical case, in a suitable rescaled framework, if the initial graph is a sublinear perturbation of a cone, the evolution asymptotically approaches an expanding self-similar solution. We also discuss some results in the unrescaled case, such as stability of hyperplanes and stability of mean convex cones.
Time/place: Monday, November 29, 2021, 14:15-15:00, Zoom
Speaker: Markus Tempelmayr (MPI Leipzig)
Title: A multi-index based regularity structure for quasi-linear SPDEs
Abstract: We give an overview of the solution theory for singular SPDEs in case of a quasi-linear equation, following the recent approach of Otto, Sauer, Smith, Weber. The basic idea is to parametrize the model, which captures the local solution behavior, by partial derivatives w.r.t. the non-linearity. This allows for an efficient bookkeeping and an inductive construction of the model. Then we construct the structure group, which is needed to ``re-center'' the model, based on a Lie algebra consisting of infinitesimal generators of actions in the space of non-linearities. Although the approach is tree-free, we show morphism properties w.r.t. well-known tree-based structures in branched rough paths and regularity structures. Based on joint work with Pablo Linares and Felix Otto.
Time/place: Monday, November 22, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Douglas Svensson Seth (NTNU)
Title: The three dimensional water wave problem with vorticity
Abstract: In 1981 Reeder and Shinbrot published the first rigorous existence result for the three dimensional water wave problem. Since then the theory has evolved and we will begin with a brief overview to highlight some of the differences between the two and the three dimensional problem. The rest of the talk will be dedicated to two more recent existence results for the three dimensional problem where the vorticity is nonzero. The first is based on the assumption that the velocity field of the water is a Beltrami field. In the other, the vorticity is given by an assumption that stems from magnetohydrodynamics. This is joint work with Erik Wahlén, Evgeniy Lokharu and Kristoffer Varholm.
Time/place: Monday, November 15, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Susanne Solem (NMBU)
Title: A system of PDEs modelling noisy grid cells
Abstract: Grid cells are neurons which play an important role in the internal navigational system of mammals. Even though grid cell networks, and the patterns they create, have been extensively studied within several disciplines in the last years, understanding the effects of noise on the network remains a challenge. In this talk, I will discuss an upscaled noisy grid cell model in the form of a system of partial differential equations. Current results on the robustness of network activity patterns with respect to noise, in terms of bifurcations from unstable homogeneous states, will be presented. Based on joint work with José A. Carrillo (Oxford) and Helge Holden (NTNU).
Time/place: Monday, November 8, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: James Jackaman (NTNU)
Title: Inexact geometric linear solvers
Abstract: In this talk we will investigate the conservation properties of non-exact linear solvers before convergence. We shall study the famous GMRES (generalized residual method) and propose a modified algorithm which may preserve arbitrarily many conserved quantities. We will investigate this method for a variety of linear finite element discretisations and discuss the benefits and costs of this methodology. In addition, we discuss a user-friendly python implementation of this methodology.
Time/place: Monday, November 1, 2021, 14:15-15:00, Zoom
Speaker: Tien Truong (Lund Univ., Sweden)
Title: Solitary waves in a Whitham equation with small surface tension
Abstract: The talk attempts to connect a weakly dispersive toy model with the original two-dimensional gravity-capillary water wave problem; the proposal of Whitham in 1970 with the discovery of Kirchgässner in 1988. It is interesting because these two problems are structurally different, and the toolbox for nonlocal equations has recently seen a new development inspired by spatial dynamics. We show that the gravity-capillary Whitham equation features generalized solitary waves and modulated solitary waves. This agrees with the findings for the water wave problem. Finally, we discuss a missing ingredient for finding multipulse solitary waves.
Time/place: Monday, October 25, 2021, 14:15-15:00, Zoom
Speaker: William Salkeld (Université Cote d'Azur, France)
Title: Lions calculus and coupled Hopf algebras with applications to probabilistic rough paths
Abstract: In this talk, I will explain some of the foundation results for a new regularity structure developed to study interactive systems of equations and their mean-field limits. At the heart of this solution theory is a Taylor expansion using the so-called Lions measure derivative. This quantifies infinitesimal perturbations of probability measures induced by infinitesimal variations in a linear space of random variables. I will explore how basic properties of Lions derivatives evolve into the structures of a coupled Hopf algebra and describe the implications for Runge-Kutta methods and probabilistic rough paths. These lead to a new solution theory and simulation techniques for rough mean-field equations. This talk is based on ArXiv:2106.09801 and ongoing work with my supervisor Francois Delarue at Universite Cote d'Azur.
Time/place: Wednesday, October 20, 2021, 13:15-14:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Mario Maurelli
Title: Stochastic Euler equations: a geometric approach
Abstract: In their celebrated work [Ann. Math. 1970], Ebin and Marsden have shown local well-posedness of the incompressible Euler equations in any dimension by solving a smooth ODE on the infinite-dimensional space of volume-preserving Sobolev diffeomorphisms. In this talk, we will develop this approach for the incompressible Euler equations driven by an additive, stochastic force term: we will solve a stochastic ODE with smooth coefficients on the space of volume-preserving Sobolev diffeomorphisms and get in turn local well-posedness of the stochastic Euler equations. This approach is quite flexible and we believe it can be used for other stochastic PDEs. No particular prerequisite on geometry or stochastic analysis is needed, we will try to give the intuition behind the arguments. Based on the joint work with Klas Modin and Alexander Schmeding arXiv:1909.09982 .
PS: Mario will give short introduction to Regularisation by noise in ODEs and PDEs on Friday October 22, 11:15-12:15 and 13:15-14:00 in room 822 SB2.
Physical course only. Interested? Send an email to Espen.
Time/place: Monday, October 18, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Luca Galimberti (NTNU)
Title: Neural Networks in Fréchet spaces
Abstract: In this talk we present some novel results obtained by Fred Espen Benth (UiO), Nils Detering (University of California Santa Barbara) and myself on abstract neural networks and deep learning. More precisely, we derive an approximation result for continuous functions from a Fréchet space \(X\) into its field \(\mathbb{F}, (\mathbb{F}\in\{\mathbb{R},\mathbb{C} \})\). The approximation is similar to the well-known universal approximation theorems for continuous functions from \(\mathbb{R}^n\) to \(\mathbb{R}\) with (multilayer) neural networks [Cyb,Hor,Fun,Les]. Similar to classical neural networks, the approximating function is easy to implement and allows for fast computation and fitting. Few applications geared toward derivative pricing and numerical solutions of parabolic partial differential equations will be outlined.
[Cyb] G. Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of Control, Signals and Systems, 2(4):303-314, 1989.
[Hor] K. Hornik, M. Stinchcombe, and H. White. Multilayer feedforward networks are universal approximators. Neural Networks, 2(5):359-366, 1989.
[Fun] K.-I. Funahashi. On the approximate realization of continuous mappings by neural networks. NeuralNetworks, 2(3):183-192, 1989.
[Les] M. Leshno, V. Y. Lin, A. Pinkus, and S. Schocken. Multilayer feedforward networks with a nonpolynomial activation function can approximate any function. Neural Networks, 6(6):861-867, 1993.
Time/place: Monday, October 11, 2021, 14:15-15:00, Zoom
Speaker: Ilya Chevyrev (Univ. of Edinburgh UK)
Title: Feature Engineering with Regularity Structures
Abstract: In this talk, I present a recent work in which we investigate models from the theory of regularity structures as features in machine learning tasks. A model is a polynomial function of a space-time signal designed to well-approximate solutions to partial differential equations (PDEs). Models can be seen as multi-dimensional generalisations of signatures of paths; this work therefore aims to extend the use of signatures in data science beyond the context of time-ordered data. I will introduce a flexible definition of a model feature vector and two algorithms which combine these features with supervised linear regression. I will also present several numerical experiments in which we use these algorithms to predict solutions to parabolic and hyperbolic PDEs with a given forcing and boundary conditions. Interestingly, in the hyperbolic case, the prediction power relies heavily on whether the boundary conditions are appropriately included in the model. Based on joint work with Andris Gerasimovics and Hendrik Weber.
Time/place: Monday, October 4, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Kristoffer Varholm (NTNU)
Title: On the precise local behavior of extreme solutions of nonlocal dispersive equations
Abstract: We prove exact asymptotic behaviour at the origin for nontrivial solutions of a family of nonlocal equations. This family of equations includes those satisfied by the cusped highest steady waves for both the uni- and bidirectional Whitham equations. In particular, our results partially settle conjectures for such extreme waves posed in [1,2,3].
[1] M. Ehrnström and E. Wahlén, On Whitham’s conjecture of a highest cusped wave for a nonlocal dispersive equation, Ann. Inst. H. Poincaré Anal. Non Linéaire 36 (2019), no. 6, 1603–1637.
[2] M. Ehrnström, M. A. Johnson, and K. M. Claassen, Existence of a highest wave in a fully dispersive two-way shallow water model, Arch. Ration. Mech. Anal. 231 (2019), no. 3, 1635–1673.
[3] T. Truong, E. Wahlén, and M. H. Wheeler, Global bifurcation of solitary waves for the Whitham equation, arXiv:2009.04713 (2020).
Time/place: Monday, September 20, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Adrien Laurent (Bergen)
Title: Exotic aromatic B-series for sampling the invariant measure of ergodic stochastic differential equations in \(R^d\) and on manifolds
Abstract: For sampling the invariant measure of ergodic systems in large dimensions, it is known that the order conditions are different from those for the standard weak order conditions over short time, and they can be intricate to compute. We propose a new methodology for creating high order integrators for sampling the invariant measure of ergodic SDEs in \(R^d\) and on manifolds. In the particular case of overdamped Langevin dynamics, we obtain the order conditions for a class of Runge-Kutta methods, thanks to a new extension of the Butcher-series, called exotic aromatic B-series. To illustrate the methodology, an integrator of order two is introduced, and numerical experiments on the sphere, the torus and the special linear group confirm the theoretical findings.
Time/place: Monday, September 13, 2021, 14:15-15:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Artur Rutkowski (NTNU)
Title: Dirichlet problem, extension and trace, and Douglas identities for nonlocal operators
Abstract: We give the existence and uniqueness of the variational solutions for the Dirichlet problem governed by general nonlocal operators built on symmetric Lévy measures. We also discuss the extension and trace problem for the related Sobolev-type spaces, motivated by the problem of determining minimal possible assumptions for the exterior condition. In the process, we obtain a nonlocal Douglas identity which connects the energy of the harmonic function with a certain energy of its exterior condition.
Time/place: Wednesday, September 1, 2021, 13:15-14:00, Zoom & room 734, SBII, Campus Gløshaugen
Speaker: Helge Glockner (Paderborn Univ., Germany)
Title: Flows, parameter dependence, and diffeomorphism groups
Abstract: Flows of complete time-dependent vector fields give rise to curves in diffeomorphism groups and thus furnish a link between solutions to ordinary differential equations and infinite-dimensional Lie groups. This sheds a new light on the parameter-dependence of solutions. Notably, it enables approximations to be studied which arise when general vector fields are replaced by vector fields which are piecewise constant in time. The general context of the studies are evolution equations on infinite-dimensional Lie groups (regularity theory). I shall give an introduction to the topic and examples.
Time/place: Friday, June 25, 2021, 14:00–14:45, Zoom
Speaker: Matthew Tandy (NTNU)
Title: Lipschitz Stability for the Hunter-Saxton Equation
Abstract: We study the Lipschitz stability of solutions to the Hunter-Saxton equation, \begin{equation} u_t(x,t) + u(x,t)u_x(x,t) = \frac{1}{4}\left(\int_{-\infty}^{x} u_x^2(y, t)\ dy - \int_x^{+\infty} u_x^2(y,t)\ dy\right). \end{equation} This equation was introduced by Hunter and Saxton in 1991 as a model for the nonlinear instability in the director field of a nematic liquid crystal. Solutions of this equation are of interest because they may develop singularities, called wave breaking, in finite time. That is, the derivative \(u_x \to -\infty\) spatially pointwise, while \(u\) remains bounded and continuous. Different solutions then exist depending on how much of the energy is lost at these singularities, with the loss of energy defined by some given \(\alpha\). In this work we take \(\alpha\) to be a constant between \(0\) and \(1\). To construct solutions, we will transform from the Eulerian formulation to a Lagrangian formulation using a generalised method of characteristics. We will then define a metric in the Lagrangian formulation, which is Lipschitz continuous in time. Using this, we define a metric in Eulerian coordinates that is also Lipschitz continuous in time. This is a joint work with Katrin Grunert.
Time/place: Friday, June 18, 2021, 14:00–14:45, Zoom
Speaker: Ola Isaac Høgåsen Mæhlen (NTNU)
Title: One-sided Hölder conditions for solutions of weakly dispersive equations
Abstract: The majority of dispersive equations in one space-dimension can be realised as dispersive perturbations of the Burgers equation \begin{equation}u_t + u u_x = L u_x,\end{equation} where \(L\) is a local or nonlocal symmetric operator. For sufficiently weak dispersion, the Burgers’ nonlinearity dominates and classical solutions break down due to shock-formation/wave-breaking. Using hyperbolic techniques we establish global existence and uniqueness of entropy solutions for weakly dispersive equations, but our main focus will be on a new generalization of the classical Oleinik estimate for Burgers’ equation. We obtain one-sided Hölder conditions for the solutions, which in turn controls their height and provides a novel bound of the lifespan of classical solutions based on their initial skewness.
This is joint work with Jun Xue (NTNU).
Time/place: Friday, June 11, 2021, 14:00–14:45, Zoom
Speaker: Davide Murari (NTNU)
Title: Lie group integrator’s approach to the N-fold pendulum
Abstract: Lie group integrators have been proven to be effective in many applications, where the preservation of geometrical properties is of interest. In the talk, after a brief introduction to these integrators, this approach to the mechanical system of the N-fold 3D pendulum is proposed. We introduce the main points in the intrinsic derivation of the system on a manifold and present a transitive group action which allows us to reframe it into the Lie group integrators setting. This problem can be considered a toy model to understand how to workout more intricate multi-body systems.
Time/place: Friday, June 4, 2021, 14:00–14:45, Zoom
Speaker: Olav Ersland (NTNU)
Title: On numerical approximations of fractional and nonlocal Mean Field Games
Abstract: In this talk we present a recent work, where we construct numerical approximations for Mean Field Games with fractional or nonlocal diffusions. The schemes are based on semi-Lagrangian approximations of the underlying control problems/games along with dual approximations of the distributions of agents. The methods are monotone, stable, and consistent, and we prove several convergence results: Convergence along subsequences to viscosity-very weak solutions for (i) degenerate equations in dimension d=1 and (ii) nondegenerate equations in arbitrary dimensions. We also give results on full convergence and convergence to classical solutions. Numerical tests are implemented for a range of different nonlocal diffusions and support our analytical findings.
This is a joint work with Indranil Chowdhury and Espen R. Jakobsen.
Time/place: Friday, May 28, 2021, 14:00–14:45, Zoom
Speaker: Esten Nicolai Wøien (NTNU)
Title: A PDE-based Method for Shape Registration
Abstract: In the square root velocity framework, the computation of shape space distances and the registration of curves requires solution of a non-convex variational problem. In the talk, we present a new PDE-based method for solving this problem numerically. The method is constructed from numerical approximation of the Hamilton-Jacobi-Bellman equation for the variational problem, and has quadratic complexity and global convergence for the distance estimate. In conjunction, we propose a backtracking scheme for approximating solutions of the registration problem, which additionally can be used to compute shape space geodesics. The methods have linear numerical convergence, and improved efficiency compared previous global solvers.
Time/place: Friday, May 21, 2021, 14:00–14:45, Zoom
Speaker: Evelyn Buckwar (Johannes Kepler Univ. Linz, Austria)
Title: A couple of ideas on splitting methods for SDEs
Abstract: We discuss developing splitting methods for stochastic differential equations. Splitting methods are a well-known type of numerical methods in the context of Geometric Numerical Integration of ordinary differential equations, in particular, they are known to be structure preserving schemes in various situations. Extensions of these methods to the case of stochastic differential equations exist for a considerable time already and they currently appear to become quite popular. In this talk I will present examples illustrating some benefits of splitting methods for SDEs. Illustrative examples include SDEs employed in neuroscience and computation-based inference.
Time/place: Friday, May 14, 2021, 14:00–14:45, Zoom
Speaker: Karolina Kropielnicka (Univ. of Gdańsk, Poland)
Title: Solving the linear semiclassical Schrödinger equation on the real line
Abstract: The numerical solution of a linear Schrödinger equation in the semiclassical regime is very well understood in a torus \(T^d\). A raft of modern computational methods are precise and affordable while conserving energy and resolving high oscillations very well. This, however, is far from the case with regard to its solution in \(\mathbb{R}^d\), a setting more suitable for many applications.
In this talk we will be concerned with the extension of the theory of splitting methods to this end. We start our journey with numerical analysis on the real line. The main idea is to derive the solution using a spectral method from a combination of solutions of the free Schrödinger equation and of linear scalar ordinary differential equations, in a symmetric Zassenhaus splitting method. This necessitates a detailed analysis of certain orthonormal spectral bases on the real line and their evolution under the free Schrödinger operator.
This talk is based on results obtained with Arieh Iserles, Marcus Webb and Katharina Schratz.
Time/place: Friday, May 7, 2021, 14:00–14:45, Zoom
Speaker: Samuel Walsh (Univ. of Missouri, USA)
Title: Orbital stability of internal waves
Abstract: In this talk, I will discuss new results on the nonlinear stability of capillary-gravity waves propagating along the interface dividing two immiscible fluid layers of finite depth. The motion in both regions is governed by the incompressible and irrotational Euler equations, with the density of each fluid being constant but distinct. We prove that for supercritical surface tension, all known small-amplitude localized waves are (conditionally) orbitally stable in the natural energy space. Moreover, the trivial solution is shown to be conditionally stable when the Bond and Froude numbers lie in a certain unbounded parameter region. For the near critical surface tension regime, we show that one can infer conditional orbital stability or orbital instability of small-amplitude traveling waves solutions to the full Euler system from considerations of a dispersive PDE similar to the steady Kawahara equation.
These results are obtained by reformulating the problem as an infinite-dimensional Hamiltonian system, then applying a version of the Grillakis–Shatah–Strauss method recently developed with K. Varholm and E. Wahlén. A key part of the analysis consists of computing the spectrum of the linearized augmented Hamiltonian at a shear flow or small-amplitude wave. For this, we generalize an idea used by Mielke to treat capillary-gravity water waves beneath vacuum. This is joint work with R. M. Chen.
Time/place: Friday, April 30, 2021, 14:00–14:45, Zoom
Speaker: Matthias Ehrhardt (Univ. of Bath, UK)
Title: Bilevel Learning for Inverse Problems
Abstract: Variational regularization techniques are dominant in the field of inverse problems. A drawback of these techniques is that they are dependent on a number of parameters which have to be set by the user. This issue can be approached by machine learning where we estimate these parameters from data. This is known as "Bilevel Learning" and has been successfully applied to many tasks, some as small-dimensional as learning a regularization parameter, others as high-dimensional as learning a sampling pattern in MRI. While mathematically appealing this strategy leads to a nested optimization problem which is computationally difficult to handle. In this talk, we discuss a number of applications of bilevel learning for imaging as well as new computational approaches. There are a number of open problems in this relatively recent field of study, some of which I will highlight along the way.
Time/place: Friday, April 23, 2021, 14:00–14:45, Zoom
Speaker: Catherine Higham (Univ. of Glasgow, UK)
Title: Applications of Deep Learning for Quantum Technologies
Abstract: Deep learning is being applied and obtaining impressive results in many novel emerging Quantum Technologies. In this talk, I will introduce some machine learning approaches, in a range of quantum physics photonics applications (single-pixel camera/video/LiDAR), where neural networks including convolutional autoencoders and generative adversarial networks are being used to solve experimental optimization, inverse and classification/regression problems.
Time/place: Friday, April 16, 2021, 14:00–14:45, Zoom
Speaker: Desmond Higham (Univ. of Edinburgh, UK)
Title: A Brief Introduction to Deep Learning for Applied Mathematicians
Abstract: Multilayered artificial neural networks are becoming a pervasive tool in a host of application fields. At the heart of this deep learning revolution are familiar concepts from calculus, approximation theory, optimization and linear algebra. I will provide a very brief introduction to the basic ideas that underlie deep learning from the perspective of a numerical analyst. I will also highlight areas where applied and computational mathematicians are well-placed to make research contributions.
Time/place: Friday, April 9, 2021, 14:00–14:45, Zoom
Speaker: Lars Ruthotto (Emory Univ., USA)
Title: Numerical Analysis Perspectives on Deep Neural Networks
Abstract: The resurging interest of deep learning are commonly attributed to advances in hardware and growing data sizes and less so to new algorithmic improvements. However, cutting edge numerical methods are needed to tackle ever larger and more complex learning problems. In this talk, I will illustrate the use of numerical analysis tools for improving the effectiveness of deep learning algorithms. With a focus on deep neural networks that can be modeled as differential equations, I will highlight the importance of choosing an adequate time integrator. I will also compare, using a numerical example, the difference of the first-discretize-then-optimize and the first-optimize-then-discretize paradigms for training residual neural networks. Finally, I show that exploiting the separable structure of most learning problems can increase the efficiency and the accuracy of training.
Time/place: Friday, March 26, 2021, 14:00–14:45, Zoom
Speaker: Nils Berglund (Univ. d'Orléans, France)
Title: Metastable dynamics of stochastic Allen-Cahn PDEs on the torus
Abstract: Allen-Cahn equations are parabolic PDEs modelling phase separation. While their solution theory is the same as for the \(\Phi^4\) model of Quantum Field Theory, their dynamics when forced by weak space-time white noise is very different, because of the phenomenon of metastability: solutions tend to spend exponentially long (in the noise intensity) times near local minima of the energy function. The exponent of the mean transition time can be deduced from a large-deviation principle, but computing the subexponential prefactor requires more work.
Using a potential-theoretic approach, we show that in one spatial dimension, the prefactor can be expressed in terms of a Fredholm determinant. In two spatial dimensions, a formal computation of the prefactor by the same approach fails to converge. However, taking into account results on renormalisation of singular PDEs, we show that the prefactor involves a renormalised Carleman-Fredholm determinant.
Based on joint works with Barbara Gentz, and with Giacomo Di Gesù and Hendrik Weber.
Time/place: Friday, March 19, 2021, 14:00–14:45, Zoom
Speaker: Antoine Lejay (INRIA, Univ. de Lorraine, France)
Title: General rough differential equations through flow approximations: a constructive tool
Abstract: The theory of rough paths allows one to consider differential equations driven by rough signals. Many alternative constructions now exist, all relying on the various forms of so-called sewing lemma. Among them, the non-linear sewing lemma transforms a family of approximations which is "close to" be flow, defined in term of composition, to a real flow related to the rough differential equation. This approach also quantifies the quality of numerical schemes.
In this talk, we explore some approximations based on regular flows parametrized by objects living in a suitable algebraic structure that determines the nature of the driving path. To apply the non-linear sewing lemma, we control their compositions by repeated use of the Newton formula. To work with minimal assumptions on the regularity of the flow, a key tool is to establish a multivariate Taylor formula with an integral remainder. This latter mixes both analytic and algebraic considerations.
We then give several examples and applications to rough differential equations driven by geometric rough paths, branched rough paths and also by paths living in possibly other structures.
Time/place: Friday, March 12, 2021, 14:00–14:45, Zoom
Speaker: Nikolas Tapia (WIAS and TU Berlin)
Title: Numerical schemes for Rough Partial Differential Equations
Abstract: In the first part of the talk, I will introduce the basic tools from rough analysis including Davie expansions. Then, I will show how a previous result, joint with C. Bellingeri, A. Djurdjevac and P. K. Friz (arXiv:2002.10432), can be applied to obtain a numerical scheme for the transport equation driven by irregular multiplicative noise. Finally, I will concentrate on ongoing work with A. Djurdjevac and C. Bayer on a Finite Element Method for linear parabolic PDEs with multiplicative noise.
Time/place: Friday, March 5, 2021, 14:00–14:45, Zoom
Speaker: Elisabeth Köbis (NTNU)
Title: Introduction to Set Optimization
Abstract: Set optimization is concerned with finding minimal solutions of set-valued mappings, where the outcome sets are compared by binary relations among sets, so-called set relations. In this talk, we give a characterization of set relations by means of nonlinear scalarization functionals. We will in addition formulate two notions of approximate solutions and give conditions for approximate minimal solutions by means of single inequalities. Furthermore, we propose a generalized Jahn-Graef-Younes method to compute the set of (approximate) minimal elements. Our methods do not rely on any convexity assumptions on the considered sets. Moreover, an application to uncertain programming, in particular, robustness, is presented.
Time/place: Friday, Feb. 19, 2021, 14:00–14:45, Zoom
Speaker: Carina Geldhauser (Lund Univ., Sweden)
Title: Space-discretizations of reaction-diffusion SPDEs
Abstract: In this talk we will discuss two different viewpoints on a space-discrete reaction-diffusion equation with noise: First, as an interacting particle system in a bistable potential, and second, as a lattice differential equation. Each viewpoint sheds light on a different phenomenon, which will be highlighted in the talk. Based on joint works with A. Bovier (Bonn) and Ch. Kuehn (TU Munich).
Time/place: Friday, Feb. 12, 2021, 14:00–14:45, Zoom
Speaker: James Jackaman (NTNU)
Title: Geometric space-time finite element discretizations for multi-symplectic PDEs
Abstract: In this talk we shall focus on the motivation, design and properties of geometric space-time finite element methods, utilising multisymplectic partial differential equations as prototypical example.
A multisymplectic PDE in one time and one space dimension is a PDE which can be written as a first order system in the form \begin{equation} \nonumber K\,\mathbf{z}_t+L\,\mathbf{z}_x=\nabla S(\mathbf{z}) \end{equation} \(K\) and \(L\) are skew-symmetric matrices, \(\mathbf{z}\in \mathbb{R}^d\), \(S(\mathbf{z})\) smooth. Historically multisymplectic methods are space-time approximations of such PDEs, often based on finite differences. These methods are designed to preserve a discrete variant of the conservation law of multisymplecticity. Unfortunately, discretizations of multisymplectic PDEs in strong form (e.g. multisymplectic discretizations) can sometimes be not well defined locally, and/or globally, or not have solutions/unique solutions.
We shall investigate the preservation of discrete local momentum and energy conservation laws, deriving space-time finite element integrators with a focus on preserving these invariants. We conclude with the presentation of numerical experiments and an outlook on how the proposed finite element method may be utilised for geometric space-time adaptive algorithms.
Time/place: Friday, Feb. 5, 2021, 14:00–14:45, Zoom
Speaker: Fabian Harang (UiO)
Title: Pathwise regularization by noise for SDEs and SPDEs with multiplicative noise
Abstract: In this talk we will discuss two results relating to pathwise regularization by noise for stochastic differential equations (SDEs) and stochastic partial differential equations (SPDEs) with multiplicative noise. We will begin with the wellposedness of an SDE controlled by a fractional Brownian motion with H>1/2. Similarly, as for the case of ODEs, it is well known that existence and uniqueness for such equations are granted under the assumption of Lipschitz diffusion coefficients. There have been few results that extend this assumption to a wider class of coefficients. Based on the recent progress in pathwise regularization by noise, we investigate the effect that perturbations of the SDE by a continuous path may have on the well-posedness of the equation. We find that that for a large class of continuous processes perturbing the equation, the SDE is well-posed even when the diffusion coefficient is only a distribution. This extends the pathwise regularization techniques developed by Catellier and Gubinelli (2016), to the case of stochastic differential equations with multiplicative noise. This result is based on joint work with Lucio Galeatii (university of Bonn). We will continue with a discussion of a recent result which is similar in spirit. We investigate the nonlinear stochastic heat equation with multiplicative noise. The multiplicative noise considered here is only spatially dependent, and we investigate the effects of a perturbation of the equation by a measurable path. We will see that also in this case, with a suitable choice of perturbation, the equation is well-posed, even when the nonlinear coefficient is only a distribution, and the multiplicative noise is considered to be a spatial white noise. This work is based on a recent preprint together with Rémi Catellier (University of Nice Sophia Antipolis).
Time/place: Friday, Jan. 29, 2021, 14:00–14:45, Zoom
Speaker: Olivier Verdier (Huawei)
Title: Automatic Differentiation on Manifolds
Abstract: In order to solve an optimisation problem, one has to compute gradients. A very popular and convenient way to do this is using automatic differentiation.
I will explain what automatic differentiation is in general, discuss some of its implementation details, and how it can be adapted to work for optimization problems on manifolds. I will focus a little on problems involving deformations (image matching, template reconstruction in tomography), where optimization is carried out on the group of diffeomorphisms.
Time/place: Wednesday, Dec. 16, 2020, 10:15–11:00, Zoom
Speaker: Brynjulf Owren (NTNU)
Title: Deep neural networks as structure preserving optimal control problems
Abstract: A deep neural network model consists of a large number of layers, each with a number of parameters or controls associated to them. In supervised learning, these parameters are optimised to match training data in the best possible way. The data are propagated through the layers by nonlinear transformations, and in an important subclass of models (ResNet) the transformation can be seen as the numerical flow of some continuous vector field. Ruthotto and Haber (2017), as well as Cheng et al., have experimented in using a different type of vector fields to improve the deep learning model. In particular, it is of interest that the trained model has good long time behaviour and is stable in the deep limit when the number of layers tends to infinity. The models presented in the literature have certain built-in structural properties, they can for instance be gradient flows or Hamiltonian vector fields. A difficulty is however that the models are not autonomous and therefore it is less clear what their flows actually preserve. Starting from such ResNet vector fields, we shall discuss their properties and derive some new nonlinear stability bounds. The long time behaviour of these neural ODE flows is important in the generalisation mode, i.e. after the model has been trained. But also in the training algorithm itself, structure preserving numerical schemes are important. In deep learning models, the use of gradient flows for optimisation is prevalent, and there exists a number of different algorithms that can be used, some of them can be interpreted as approximations of the flow of certain vector fields with dissipations, such as conformal Hamiltonian systems. If time permits, we will briefly discuss also these algorithms and in particular the need for and efficiency of regularisation.
Joint work with: Martin Benning, Elena Celledoni, Matthias Ehrhardt, Christian Etmann, Robert McLachlan, Carola-Bibiane Schönlieb and Ferdia Sherry.
Time/place: Wednesday, Dec. 9, 2020, 10:15–11:00, Zoom
Speaker: Gilles Vilmart (Univ. Genève)
Title: Second kind explicit stabilized integrators for ergodic and stiff stochastic (partial) differential equations
Abstract: For large dimensional and stiff diffusion problems, explicit stabilized integrators are an efficient alternative to implicit or semi-implicit methods to avoid the severe timestep restriction faced by standard explicit time integrators.
We present a family of explicit stabilized integrators for ergodic and stiff problems, based on second kind Chebyshev polynomials, that yield an optimal size of extended mean-square stability domain that grows at the same quadratic rate as the optimal family for deterministic problems. We also show that the new explicit stabilized schemes converge in the strong sense when applied to stochastic semilinear diffusion partial differential equations.
Based on joint works with A. Abdulle (Lausanne), I. Almuslimani (Rennes), and C.-E. Bréhier (Lyon).
Time/place: Wednesday, Dec. 2, 2020, 10:15–11:00, Zoom
Speaker: Fabian Harang (Univ. Oslo)
Title: Infinitely regularizing paths, and regularization by noise
Abstract: In this talk we will discuss regularization by noise from a pathwise perspective using non-linear Young integration, and discuss the relations with occupation measures and local times. This methodology of pathwise regularization by noise was originally proposed by Gubinelli and Catellier (2016), who use the concept of averaging operators and non-linear Young integration to give meaning to certain ill-posed SDEs.
In a recent work together with Prof. Nicolas Perkowski at Freie University, Berlin, we show that there exists a class of paths with exceptional regularizing effects on ODEs, using the framework of Gubinelli and Catellier. In particular, we prove existence and uniqueness of ODEs perturbed by such a path, even when the drift is given as a Scwartz distribution. Moreover, the flow associated to such ODEs are proven to be infinitely differentiable. Our analysis can be seen as purely pathwise, and is only depending on the existence of a sufficiently regular occupation measure associated to the path added to the ODE.
As an example, we show that a certain type of Gaussian process has infinitely differentiable local times, whose paths then can be used to obtain the infinitely regularizing effect on ODEs. This gives insight into the powerful effect that noise may have on certain equations. If time permits, we will also discuss an ongoing extension of these results towards regularization of certain PDE/SPDEs by noise.​
Time/place: Wednesday, Nov. 25, 2020, 10:15–11:00, Zoom
Speaker: Pranav Singh (Univ. Bath)
Title: Convergence of Magnus based methods for Schrödinger equations
Abstract: Magnus expansion based methods are an efficient class of integrators for solving Schrödinger equations that feature time dependent potentials such as lasers. These methods have been found to be highly effective in computational quantum chemistry since the pioneering work of Tal Ezer, Kosloff and Cerjan in the early 90s.
The convergence of the Magnus expansion, however, is understood only for ODEs and traditional analysis suggests a much poorer performance of these methods than observed experimentally. It was not till the work of Hochbruck and Lubich in 2003 that a rigorous analysis justifying the application to PDEs with unbounded operators, such as the Schrödinger equation, was presented.
In this talk I will present the extension of this analysis to the semiclassical regime, where the highly oscillatory solution conventionally suggests large errors and a requirement for very small time steps.
Time/place: Wednesday, Nov. 11, 2020, 10:15–11:00, Zoom
Speaker: Adrian M. Ruf (ETH Zürich)
Title: Convergence rates of numerical methods for conservation laws with discontinuous flux
Abstract: The subject of conservation laws with discontinuous flux has been an active re-search area during the last several decades. Many different selection criteria to single out a unique weak solution have been proposed and several numerical schemes have been designed and analyzed in the literature. Surprisingly, the preexisting literature on convergence rates for such schemes is practically nonexistent. In this talk, focusing on so-called adapted entropy solutions, I will present recent developments in this direction and prove convergence rates for finite volume and front tracking methods. As an application, I will briefly describe how these results can be used for uncertainty quantification in two-phase reservoir simulations for reservoirs with varying geological properties.
Time/place: Thursday, Nov. 5, 2020, 13:15–14:00, Zoom
Speaker: Massimiliano Gubinelli (Univ. Bonn)
Title: Grassmann stochastic analysis and stochastic quantisation of Euclidean Fermions
Abstract: This talk is about the extension of some probabilistic construction to the case of Grassmann valued random variables, i.e. random variables which anticommute. This requires setting up the problem in the context of non-commutative probability. Moreover we study some simple stochastic differential equations for Grassmann variables and derive information on their invariant states.
Joint work with S. Albeverio, L. Borasi and F. de Vecchi. Based on the paper: Albeverio, Sergio, Luigi Borasi, Francesco C. De Vecchi, and Massimiliano Gubinelli. ‘Grassmannian Stochastic Analysis and the Stochastic Quantization of Euclidean Fermions’. arXiv:2004.09637
Time/place: Wednesday, Oct. 28, 2020, 10:15–11:00, Zoom
Speaker: Yuya Suzuki (NTNU)
Title: A quasi-Monte Carlo method combined with operator splitting for time-dependent Schrödinger equations
Abstract: In the first half of this talk I will give a brief introduction of quasi-Monte Carlo (QMC) methods which are originally used for numerical integration over a multidimensional unit cube. I will show some known results including the classical Koksma–Hlawka inequality and also modern results in reproducing kernel Hilbert space settings.
In the second half of the talk, I will apply one important branch of QMC methods, namely rank-1 lattice rules, for numerically solving time-dependent Schrödinger equations in multidimensional settings. We combine operator splitting (time discretization) and a pseudospectral method on rank-1 lattice (space discretization). We show the theoretical error bound of the scheme and also numerical results which confirm the theory. This second part is based on joint work with Dirk Nuyens and Gowri Suryanarayana.
Time/place: Wednesday, Oct. 21, 2020, 10:15–11:00, Zoom
Speaker: Ben Tapley (NTNU)
Title: Simulating slender particles in viscous flow
Abstract: A slender particle is a particle that is long but with a small cross sectional radius. Being able to accurately and cheaply simulate slender particles in viscous flow is essential for our understanding of micro swimming organisms, for example. In this presentation, we present a new model for calculating the forces and torques on slender particles. The model gives rise to a Fredholm integral equation whose result needs to be integrated to find the forces and torques on the particle. We then propose and analyse a fast numerical method for solving these integral equations. Finally, numerical experiments are given to validate the model and method.
Time/place: Thursday*, Oct. 15, 2020, 13:15–14:00, Zoom
Speaker: Lorenzo Zambotti (Sorbonne Univ. Paris, France)
Title: Hairer's Reconstruction Theorem without Regularity Structures
Abstract: This talk, based on joint work with Francesco Caravenna 2005.09287, is devoted to Martin Hairer's Reconstruction Theorem, which is one of the cornerstones of his theory of Regularity Structures [Hairer 2014]. Our aim is to give a new self-contained and elementary proof of this Theorem and of some applications. We present it as a general result in the theory of distributions that can be understood without any knowledge of Regularity Structures themselves, which we do not even need to define.
*Note: the talk was moved -exceptionally- to Thursday due to the unavailability of the speaker on Wednesdays.
Time/place: Wednesday, Oct. 7, 2020, 10:15–11:00, Zoom
Speaker: Weinan E (Princeton Univ., USA)
Title: Machine Learning and Numerical Analysis
Abstract: The heart of machine learning is the approximation of functions using finite pieces of data. This is one of the main pillars of numerical analysis. Thus it is not surprising that the success of machine learning in dealing with functions in very high dimensions has opened up some brand new territories in computational mathematics, with potentially unprecedented impact for years to come.
In the first part of this talk, I will review some of the most exciting advances of using machine learning to address problems in scientific computing and computational science.
In the second part of this talk, I will discuss how machine learning can be formulated as a problem in numerical analysis and how ideas from numerical analysis can be used to understand machine learning as well as construct new machine learning models and algorithms.
Time/place: Wednesday, Sept. 30, 2020, 10:15–11:00, Zoom
Speaker: Tale Bakken Ulfsby (NTNU)
Title: Stabilized cut discontinuous Galerkin methods for advection-diffusion-reaction problems on surfaces
Abstract: Advection-diffusion-reaction (ADR) problems on surfaces with complicated or moving geometries appear in many important problems in science and engineering, including the modeling of cell motility in cell biology and the study of transport phenomena in fractured porous media. Standard numerical approaches such as the finite element method demand the construction of meshes fitted to the geometry. If the domain geometry is very complicated or undergoes large deformations, the required generation of fitted meshes can be very time consuming and computationally expensive, and account for a large percentage of the overall simulation time.
As a potential remedy, new generations of unfitted or cut finite element methods have been developed in the last decade. For surface problems, these methods use a fixed background mesh while the embedded surface is described implicitly. In this talk, we present a new Cut Discontinuous Galerkin (CutDG) method for the discretization of advection-dominant problems on surfaces. The discrete function space is defined as the restriction of the basis functions associated with the underlying background mesh to the surface. Geometrically robustness problems usually caused by troublesome cut configurations are avoided by adding carefully designed CutFEM stabilizations. As a result, we can prove geometrically robust stability and optimal a priori error estimates. Finally, numerical results are presented to illustrate our theoretical findings.
Time/place: Wednesday, Sept. 23, 2020, 10:15–11:00, Zoom
Speaker: Torstein Nilssen (UiA)
Title: Rough path variational principles for fluid equations
Abstract: In this presentation we will first recall the insight of Arnold that Euler's equation can be understood as a geodesic equation on an infinite dimensional manifold. Using Lagrange multipliers, this yields a natural framework for understanding the structure of random/irregular perturbations of fluid equations. In this presentation, we will consider rough path perturbations and the corresponding variational principles.
Time/place: Wednesday, Sept. 16, 2020, 10:15–11:00, Zoom
Speaker: Sondre Tesdal Galtung (NTNU)
Title: A semi-discrete Camassa–Holm system based on variational principles
Abstract: In this talk, based on joint works with Katrin Grunert and Xavier Raynaud (2003.03114, 2006.15562). I will present a discretization of a Camassa-Holm system (2CH). This system is an extension of the well-known Camassa–Holm equation and serves as a model for shallow water waves. The discretization is based on a derivation of 2CH from variational principles in Lagrangian coordinates: Discretizing the energy functionals, an analogous procedure yields a semi-discrete, Lagrangian 2CH system with its own conserved quantities. Some of the main challenges encountered in the study of the discrete system will be highlighted, and I will indicate how we prove convergence to solutions of 2CH. Finally, some numerical examples illustrate the performance of the discretization when paired with explicit time integrators.
Time/place: Wednesday, June 24, 2020, 14:15–15:00, Zoom
Speaker: Sebastian Riedel (TU Berlin)
Title: Runge-Kutta methods for rough differential equations
Abstract: In the 60s, J.C. Butcher used a power series expansion over rooted trees for ordinary differential equations to study numerical methods of a high order. Nowadays, such series are called B-series. We derive a B-series representation for rough differential equations and calculate the local order of the truncated expansion. En passant, we introduce Gubinelli's concept of a branched rough path which is closely related. As a result, we obtain a class of numerical methods that can be used to solve random rough differential equations numerically. They apply, for instance, to stochastic differential equations driven by a fractional Brownian motion. Joint work with Martin Redmann (Halle).
Time/place: Wednesday, June 17, 2020, 14:15–15:00, Zoom
Speaker: Miłosz Krupski (NTNU)
Title: Control of the jump process in non-local mean field games PART 2
Abstract: We study a version of non-local "mean field game" (MFG) system of equations, derived from a stochastic game model, in which players control the underlying Lévy process.
The MFG system consists of two equations: in type of Hamilton-Jacobi-Bellman and Fokker-Planck-Kolmogorov. We prove well-posedness of the two equations individually, and then study solutions of the MFG system itself.
Part 1 (presented by Indranil Chowdhury) will introduce the MFG model of our interest, starting from a stochastic game, and then focus on the results on the Fokker-Planck equation.
Part 2 (presented by Miłosz Krupski) will describe the stochastic model in more detail and develop the results in regard of the MFG system as a whole.
Time/place: Tuesday, June 16, 2020, 14:15–15:00, Zoom
Speaker: Indranil Chowdhury (NTNU)
Title: Control of the jump process in non-local mean field games PART 1
Abstract: We study a version of non-local "mean field game" (MFG) system of equations, derived from a stochastic game model, in which players control the underlying Lévy process.
The MFG system consists of two equations: in type of Hamilton-Jacobi-Bellman and Fokker-Planck-Kolmogorov. We prove well-posedness of the two equations individually, and then study solutions of the MFG system itself.
Part 1 (presented by Indranil Chowdhury) will introduce the MFG model of our interest, starting from a stochastic game, and then focus on the results on the Fokker-Planck equation.
Part 2 (presented by Miłosz Krupski) will describe the stochastic model in more detail and develop the results in regard of the MFG system as a whole.
Time/place: Wednesday, May 27, 2020, 14:15–15:00, Zoom
Speaker: Yvain Bruned (Univ. of Edinburgh, UK)
Title: Resonance based schemes for dispersive equations via decorated trees PART II
Abstract: We introduce a numerical framework for dispersive equations embedding their underlying resonance structure into the discretisation. This will allow us to resolve the nonlinear oscillations of the PDE and to approximate with high order accuracy a large class of equations under lower regularity assumptions than classical techniques require. The key idea to control the nonlinear frequency interactions in the system up to arbitrary high order thereby lies in a tailored decorated tree formalism. Our algebraic structures are close to the ones developed for singular SPDEs with Regularity Structures. We adapt them to the context of dispersive PDEs by using a novel class of decorations (which encode the dominant frequencies). The structure proposed in this work is new and gives a variant of the Butcher-Connes-Kreimer Hopf algebra on decorated trees.
Part I: We will introduce the general concept of the numerical method and present some examples.
Part II: We will develop the algebraic structures and present the decorated trees needed for writing the general numerical scheme.
Time/place: Tuesday, May 26, 2020, 14:15–15:00, Zoom
Speaker: Katharina Schratz (Heriot-Watt Univ., Edinburgh, UK, & Sorbonne Univ. Paris, France)
Title: Resonance based schemes for dispersive equations via decorated trees PART I
Abstract: We introduce a numerical framework for dispersive equations embedding their underlying resonance structure into the discretisation. This will allow us to resolve the nonlinear oscillations of the PDE and to approximate with high order accuracy a large class of equations under lower regularity assumptions than classical techniques require. The key idea to control the nonlinear frequency interactions in the system up to arbitrary high order thereby lies in a tailored decorated tree formalism. Our algebraic structures are close to the ones developed for singular SPDEs with Regularity Structures. We adapt them to the context of dispersive PDEs by using a novel class of decorations (which encode the dominant frequencies). The structure proposed in this work is new and gives a variant of the Butcher-Connes-Kreimer Hopf algebra on decorated trees.
Part I: We will introduce the general concept of the numerical method and present some examples.
Part II: We will develop the algebraic structures and present the decorated trees needed for writing the general numerical scheme.
Time/place: Wednesday, April 22, 2020, 14:15–15:00, Zoom
Speaker: Nikolas Tapia (TU Berlin and WIAS)
Title: Transport and continuity equations with (very) rough noise
Abstract: We show existence and uniqueness of strong solutions (in the rough paths sense) to the stochastic transport equation with general Hölder-continuous noise. We also study existence and uniqueness of the associated continuity equation. (Based on joint work with C. Bellingeri, A. Djurdjevac, P. K. Friz.)
Time/place: Wednesday, April 15, 2020, 14:15–15:00, Zoom
Speaker: Peter Pang (NTNU)
Title: The Hunter-Saxton equation with noise
Abstract: This talk will be on some results derived for a stochastic Hunter-Saxton equation. The overarching theme will be an existence theory. Rather than dwelling heavily on technicalities, the intention is to give a relatively self-contained talk centred around phenomena exhibited by solutions. In particular, we shall look at stochastic blow-up/wave-breaking, the stopping times for which explicit laws can be derived. We shall also discuss continuation beyond blow-up. The primary tool is a stochastic method-of-characteristics. Results are based on joint work with Helge Holden and Kenneth H. Karlsen (https://arxiv.org/abs/2003.13984).
Time/place: Tuesday, March 10, 2020, 14:15–15:00, room 656
Speaker: Benjamin Fahs (Imperial College, London)
Title: Sine-kernel determinant on two large intervals
Abstract: The sine-kernel determinant on a set \(A\) represents the probability that there are no eigenvalues in the set \(A\) in the bulk scaling limit of the Gaussian Unitary Ensemble.
The asymptotic study of the determinant when \(A\) is a single large interval was initiated by Dyson in 1962 and was solved 50 years later through the work of numerous authors.
We consider the case where \(A\) is composed of 2 large intervals. In this case, a study of the asymptotics was started by Widom. The most detailed formula up to now is due to Deift, Its, Zhou in 1997, who obtained a formula up to an unknown multiplicative constant. We present a formula for the multiplicative constant, and additionally the uniform transition between 1 and 2 intervals.
Time/place: Wednesday, February 19, 2020, 14:15–15:00, room F2
Speaker: Jørgen Endal (NTNU)
Title: The one-phase fractional Stefan problem
Abstract: We study the existence, properties of solutions, and free boundaries of the one-phase Stefan problem with fractional diffusion posed in \(\mathbb{R}^N\). The equation for the enthalpy \(h\) reads \(\partial_t h+ (-\Delta)^{s}\Phi(h) =0\) where the temperature \(u:=\Phi(h):=\max\{h-L,0\}\) is defined for some constant \(L>0\) called the latent heat, and \((-\Delta)^{s}\) is the fractional Laplacian with exponent \(s\in(0,1)\). We prove the existence of a continuous and bounded selfsimilar solution of the form \(h(x,t)=H(x\,t^{-1/(2s)})\) which exhibits a free boundary at the change-of-phase level \(h(x,t)=L\) located at \(x(t)=\xi_0 t^{1/(2s)}\) for some \(\xi_0>0\). This special solution will be an important tool to obtain that the temperature has finite speed of propagation while the enthalpy has infinite speed, and that the support of the temperature never recedes. Other interesting properties like e.g. \(L\to0^+\) and \(L\to\infty\) will also be discussed, and the theory itself is illustrated by convergent finite-difference schemes.
This is a joint work with Félix del Teso and Juan Luis Vázquez.
Time/place: Tuesday, February 18, 2020, 14:15–15:00, room 734
Speaker: Emmanuel Chasseigne (Tours)
Title: Ergodic problems for viscous Hamilton-Jacobi equations with inward drift
Abstract: I will first give a general overview of ergodic-type problems for viscous Hamilton-Jacobi equations in the whole space. We will also study brifely the associated stochastic control optimization problem as well as the links between the elliptic and parabolic cases.

Then I will focus on the existence of critical (additive) eigenvalues and their qualitative properties depending on the potential/forcing term appearing in the equation.

Finally, I will discuss some more recent advances in presence of an additional inward drift.
Time/place: Thursday, February 13, 2020, 14:15–15:00, room 734
Speaker: Jacek Jendrej (Paris)
Title: Strongly interacting kink-antikink pairs for scalar fields on a line
Abstract: I will present a recent joint work with Michał Kowalczyk and Andrew Lawrie. A nonlinear wave equation with a double-well potential in 1+1 dimension admits stationary solutions called kinks and antikinks, which are minimal energy solutions connecting the two minima of the potential. We study solutions whose energy is equal to twice the energy of a kink, which is the threshold energy for a formation of a kink-antikink pair. We prove that, up to translations in space and time, there is exactly one kink-antikink pair having this threshold energy. I will explain the main ingredients of the proof.
Mini-workshop on numerical analysis
Time/place: Thursday, February 06, 2020, 11:15–12:00, room F4
Speaker: Geir Bogfjellmo (NTNU)
Title: Interpolation on curved spaces
Abstract: Cubic spline interpolation on Euclidean space is a standard topic in numerical analysis, with countless applications in science and technology. The generalization of cubic splines to manifolds is not self-evident, with several distinct approaches. One possibility is to mimic De Casteljau's algorithm, which leads to generalized Bézier curves. To construct \(C^2\)-splines from such curves is a complicated non-linear problem. We derived an iterative algorithm for \(C^2\)-splines on Riemannian symmetric spaces, and proven convergence. In terms of numerical tractability and computational efficiency, the new method surpasses methods based on Riemannian cubics. The algorithm is demonstrated for three geometries of interest: the \(n\)-sphere, complex projective space, and real Grassmannians.
Time/place: Thursday, February 06, 2020, 14:15–15:00, room R10
Speaker: Nick Trefethen (Oxford)
Title: From the Faraday cage to lightning Laplace and Helmholtz solvers
Abstract: We begin with the story of the Faraday cage used for shielding electrostatic fields and electromagnetic waves. Feynman in his Lectures claims the shielding is exponential with respect to the gap between wires and that it works with wires of infinitesimal radius. In fact, the shielding is much weaker than this and requires wires of finite radius (which is why it's hard to see into your microwave oven). How can we compute the field inside a 2D cage? This brings us to the numerical part of the talk. When the boundaries are smooth, series expansions (going back to Runge in 1885) converge exponentially. When there are corners and associated singularities, the new technique of lightning Laplace and Helmholtz solvers, depending on rational or Hankel functions with poles exponentially clustered near the corners, converges root-exponentially. The name "lightning" comes from the fact that this method exploits the same mathematics that makes lightning strike at sharp points. Lightning solvers and the related AAA approximation algorithm are bringing in a new era of application of rational functions and their relatives to PDEs, conformal mapping, and other numerical problems.
Time/place: Thursday, February 06, 2020, 15:15–16:00, room R10
Speaker: André Massing (NTNU)
Title: Unfitted finite element methods: discretizing geometry and partial differential equations
Abstract: Many advanced engineering problems require the numerical solution of multidomain, multidimension, multiphysics and multimaterial problems with interfaces. When the interface geometry is highly complex or evolving in time, the generation of conforming meshes may become prohibitively expensive, thereby severely limiting the scope of conventional discretization methods.
In this talk, we focus on recent unfitted finite element technologies as one possible remedy. The main idea is to design discretization methods which allow for flexible representations of complex or rapidly changing geometries by decomposing the computational domain into several, possibly overlapping domains. Alternatively, complex geometries only described by some surface representation can easily be embedded into a structured background mesh. In the first part of this talk, we briefly review how finite element schemes on cut and composite meshes can be designed by either using a Nitsche-type imposition of interface and boundary conditions or, alternatively, a partition of unity approach. Some theoretical and implementational challenges and their rectifications are highlighted. In the second part we demonstrate how unfitted finite element techniques can be employed to address various challenges from mesh generation to fluid-structure interaction problems, solving PDE systems on embedded manifolds of arbitrary co-dimension and PDE systems posed on and coupled through domains of different topological dimensionality.
Time/place: Wednesday, February 05, 2020, 14:15–15:00, room F2
Speaker: Cristopher Salvi (Oxford)
Title: Capturing similarities between streams: the warping \(p\)-variation distance
Abstract: We introduce a new metric on the space of unparameterized paths which originates from fundamental results in rough paths theory. We describe why natural signals can be described as rough paths and how the new distance is well-designed to capture (dis)similarities of streams. We then turn to the challenge of designing an efficient algorithm to compute the inroduced metric, which consists in the solution of a min-max problem. As a solution to this probem, we propose a branch-and-bound algorithm that drastically reduces the number of nodes to explore and therefore increases computational performance. We compare our metric to other distances classically used in data science, such as dynamic time warping, for two simple classification and a clustering tasks, and show experimentally that the new metric achieve better performance.
Time/place: Wednesday, January 22, 2020, 14:15–15:00, room F2
Speaker: Rinaldo M. Colombo (Brescia)
Title: Conservation Laws: Analysis & Modeling
Abstract: Conservation Laws are first order partial differential equations whose analytic developments has always been interwined with the questions posed by their applications. Originally, fluid dynamics and more recently vehicular traffic and crowd dynamics have provided an incredible amount of problems. The present talk will describe a few recent developments: a "purely analytic" one and others inspired by applications. The latter also pose new optimization and game theoretic questions.
Time/place: Tuesday, January 14, 2020, 15:15–16:00, room 734
Speaker: Luca Galimberti (Oslo)
Title: Stochastic continuity equations on Riemannian manifolds: renormalization and uniqueness
Abstract: We are given a \(d\)-dimensional (\(d \ge 1\)) smooth closed manifold \(M\), endowed with a smooth Riemannian metric \(h\). We study the Cauchy problem for the following stochastic continuity equation \[ (1) \qquad d\rho + \operatorname{div}_h (\rho u)dt + \sum_{i=1}^N \operatorname{div}_h (\rho a_i) \circ dW^i (t) = 0 \quad\text{ on }\ [0, T ] \times M, \] and \(\rho(0) = \rho_0\) on \(M\). Typically, \(u \colon [0, T ] \times M \to T M\) is a time-dependent irregular vector field on \(M\) which is interpreted as a velocity field, while \(\rho = \rho(\omega, t, x)\) is a density for a mass distribution. \(a_1,\ldots,a_N\) are arbitrary smooth vector fields on \(M\), \(W^1,\cdots,W^N\) are independent real Brownian motions, and the symbol \(\circ\) means that the equation is understood in the Stratonovich sense.
After introducing a proper concept of weak solution, we prove a very delicate renormalization theorem, in the spirit of DiPerna and Lions (1989), which at once provides us with some a priori estimates for the solution \(\rho\) as well as with an easy uniqueness result, under the mild assumption \[\operatorname{div}_h u \in L^1(0,T;L^\infty(M)).\] Crucial ingredients in this renormalization argument are the introduction of an ad hoc regularization procedure for mixed tensor fields of arbitrary order on the manifold \(M\), as well as a “second order” commutator argument, which is necessary because of the appearance of a second order differential operator in the equivalent Itô formulation of (1).
Time/place: Tuesday 03rd of December, 2019, 14:15–15:00, room 734
Speaker: Hung Le (NTNU)
Title: On the existence and instability of solitary waves with a finite dipole
Abstract: In this talk, we consider the existence and stability properties of two-dimensional solitary waves traversing an infinitely deep body of water. We assume that above the water is air and that the waves are acted upon by gravity with surface tension effects on the air-water interface. In particular, we study the case where there is a finite dipole in the bulk of the fluid, that is, the vorticity is a sum of two weighted \(\delta\)-functions. Using an implicit function theorem argument, we construct a family of solitary waves solutions for this system that is exhaustive in a neighborhood of 0. Our main result is that this family is conditionally orbitally unstable. This is proved using a modification of the Grillakis—Shatah–Strauss method recently introduced by Varholm, Wahlén, and Walsh.
Time/place: Thursday 28th of November, 2019, 10:15–11:00, room 656
Speaker: Balázs Kovács (Tübingen)
Title: A convergent algorithm for mean curvature flow with and without forcing
Abstract: We will sketch a proof of convergence for semi- and full discretizations of mean curvature flow of closed two-dimensional surfaces. The proposed and studied numerical method combines evolving surface finite elements, whose nodes determine the discrete surface like in Dziuk's algorithm proposed in 1990, and linearly implicit backward difference formulae for time integration. The proposed method differs from Dziuk's approach in that it discretizes Huisken's evolution equations (from [Huisken (1984)]) for the normal vector and mean curvature and uses these evolving geometric quantities in the velocity law projected to the finite element space. This numerical method admits a convergence analysis, which combines stability estimates and consistency estimates to yield optimal-order \(H^1\)-norm error bounds for the computed surface position, velocity, normal vector and mean curvature. The stability analysis is based on the matrix–vector formulation of the finite element method and does not use geometric arguments. The geometry enters only into the consistency estimates. We will also present various numerical experiments to illustrate and complement the theoretical results. Furthermore, we will give an outlook towards forced mean curvature flow, that is for problems coupling mean curvature flow with a surface PDE.
The talk is based on joint work with B. Li (Hong Kong) and Ch. Lubich (Tübingen).
Time/place: Tuesday 8th of October, 2019, 14:15–15:00, room 734
Speaker: Gunnar Taraldsen (NTNU)
Title: The spectrum of a random operator is a random set
Abstract: The theory of random sets is demonstrated to prove useful for the theory of random operators. A random operator is here defined by requiring the graph to be a random set. It is proved that the spectrum and the set of eigenvalues of random operators are random sets. These results seem to be a novelty even in the case of random bounded operators. The main technical tools are given by the measurable selection theorem, the measurable projection theorem, and a characterisation of the spectrum by approximate eigenvalues of the operator and the adjoint operator. A discussion of some of the existing definitions of the concept of a random operator is included at the end of the paper.
Time/place: Tuesday 1st of October, 2019, 14:15–15:00, room 734
Speaker: Mark Groves (Saarland University)
Title: Solitary-wave solutions to the full dispersion Kadomtsev-Petviashvili equation
Abstract: The KP-I equation \[u_t + m(D) u_x - 2uu_x = 0,\] where \(m(D)\) is the Fourier multiplier operator with multiplier \[m(k)=1+\frac{k_2^2}{2k_1^2} + \frac{1}{2}(\beta-{\textstyle\frac{1}{3}})k_1^2,\] arises as a weakly nonlinear model equation for gravity-capillary waves with strong surface tension (Bond number \(\beta>\frac{1}{3}\)). This equation admits – as an explicit solution – a "fully localised" or "lump" solitary wave which decays to zero in all spatial directions.
Recently there has been interest in the full dispersion KP-I equation \[u_t + \tilde{m}(D) u_x - 2uu_x = 0\] obtained by retaining the exact dispersion relation from the water-wave problem, that is, replacing \(m\) by \[\tilde{m}(k)=\left((1+\beta|k|^2)\frac{\tanh |k|}{|k|}\right)^{1/2}\left(1+\frac{k_2^2}{k_1^2}\right).\] In this talk I show that the full dispersion KP-I equation also has a fully localised solitary-wave solution. The existence theory is variational and perturbative in nature.
This project is joint work with Mats Ehrnström (NTNU, Norway).
Time/place: Thursday 26th of September, 2019, 14:15–15:00, room F3 (gamle fysikk)
Speaker: David Ambrose (Drexel University, Philadelphia)
Title: Existence Theory for a Mean Field Games Model of Household Wealth
Abstract: We study a nonlinear system of partial differential equations arising in macroeconomics which utilizes a mean field approximation. This equation together with the corresponding data, subject to two moment constraints, is a model for debt and wealth across a large number of similar households, and was introduced in a recent paper of Achdou, Burea, Lasry, Lions, and Moll. We introduce a relaxation of their problem, generalizing one of the moment constraints; any solution of the original model is a solution of this relaxed problem. We prove existence and uniqueness of strong solutions to the relaxed problem, under the assumption that the time horizon is small. Since these solutions are unique and since solutions of the original problem are also solutions of the relaxed problem, we conclude that if the original problem does have solutions, then such solutions must be the solutions we prove to exist. Furthermore, for some initial data and for sufficiently small time horizons, we are able to show that solutions of the relaxed problem are not solutions of the original problem. In this way we demonstrate nonexistence of solutions for the original problem in certain cases.
Time/place: Thursday 26th of September, 2019, 15:15–16:00, room F3 (gamle fysikk)
Speaker: Miles Wheeler (University of Bath)
Title: New exact solutions to the steady 2D Euler equations
Abstract: We present a large class of explicit "hybrid" equilibria for the 2D Euler equations, consisting of point vortices embedded in a smooth sea of "Stuart-type" vorticity. Mathematically, these are singular solutions of the elliptic Liouville equation satisfying some additional constraints at each singularity.
This is joint work with Vikas Krishnamurthy, Darren Crowdy, and Adrian Constantin.
2024-04-24, Kurusch Ebrahimi-Fard