Past seminars 2024
4/12/2024 Philipp Birken, https://www.maths.lu.se/staff/philipp-birken Links to an external site.
Title: Waveform methods for partitioned coupling
Abstract: We consider coupled time dependent partial differential equations on separate domains, This can be used to model conjugate heat transfer, but also the transfer of wind stress between ocean and atmosphere or flutter of airplanes. Our interest is in so called waveform relaxation. There, the subproblems are solved iteratively on a time window, given an approximation of the solution to the other problem. These methods open up variable time steps and time adaptivity while using separate codes for the subproblem. They can be combined with a relaxation step or Quasi-Newton acceleration.
We first present a novel time adaptive Quasi-Newton waveform method. We then discuss convergence theory of this method. This is quite limited for the general case. For the simpler case of linear coupled heat equation with discontinuous coefficients, much more can be said in dependence on the material parameters, the mesh width and the time steps. Over the years, this has been approached on different levels, meaning fully discrete, semidiscrete, or fully continuous. Thereby, one has obtained both superlinear convergence results, and norm estimates. Currently, these results give an incomplete and seemingly conflicting picture. We review these results, and then fill some of the gaps.
This is joint work with Niklas Kotarsky (Lund) and Martin Gander (Geneva).
27/11/2024 Erik Sjöqvist, https://www.uu.se/kontakt-och-organisation/personal?query=XX4071 Links to an external site.
Title: Nonlinear quantum spin dynamics
Abstract: The nonlinear Landau-Lifshitz-Gilbert (LLG) equation [1] plays an essential role when describing the dynamics of magnetization in solids. While this equation treats the magnetisation as a classical vector, the underlying degrees of freedom are quantum spins (described by operators and tensor products), which justifies the search for a quantum-mechanical description of magnetisation dynamics. Here, we discuss our recently proposed [2] quantum analog of the LLG equation that mimics the expected dynamics of a single classical spin for a certain class of states, but may differ significantly in the multi-spin case. We illustrate this latter point by examining the quantum LLG dynamics in the simplest nontrivial case consisting of two interacting spin-1/2 particles. Our analysis reveals that, in the case of antiferromagnetic coupling, the evolution of initially uncorrelated spins leads to pronounced deviations from classical behavior, such as the unique dynamics of becoming a spinless state in the asymptotic time limit. We further examine the classical large-spin limit of the quantum LLG dynamics of the dimer. We discuss the extension to several spins, which is a computationally challenging problem due to the exponential growth of coupled nonlinear equations in the quantum case.
[1] M. Lakshmanan, The fascinating world of the Landau-Lifshitz-Gilbert equation, Phil. Trans. R. Soc. A 369, 1280 (2011).
[2] Y. Liu et al., Quantum analog of the Landau-Lifshitz-Gilbert dynamics, arXiv:2403.09255
20/11/2024 Markus Kowalewski, https://www.su.se/english/profiles/mako5582-1.379187 Links to an external site.
Title: The Theory of Light-Matter Interactions - From Ultrafast X-Ray Spectroscopy to Photochemistry in Nano Resonators
Abstract: Light and matter interact in countless ways, from the biological processes of vision and photosynthesis to the engineered systems of solar cells and lasers. My group explores these interactions through computational simulations, aiming to understand and predict ultra-fast phenomena and control photochemical reactions. The quantum nature of these interactions poses significant computational challenges, which makes the computations inherently expensive.
In the last few minutes will give a brief overview over our approach to maintain a reproducible software environment to perform all our calculations. We leverage the Nix package manager to maintain a consistent software environment across different platforms and projects. This allows us to accurately reproduce our results and share our computational workflows with the scientific community.
13/11/2024 Anna-Karin Tornberg, https://www.kth.se/profile/akto Links to an external site.
Title: Layer potentials - quadrature error estimates and approximation with error control
Abstract: When numerically solving PDEs reformulated as integral equations, so called layer potentials must be evaluated. The quadrature error associated with a regular quadrature rule for evaluation of such integrals increases rapidly when the evaluation point approaches the surface and the integrand becomes sharply peaked. Error estimates are needed to determine when the accuracy becomes insufficient, and then, a sufficiently accruate special quadrature method needs to be employed.
In this talk, we discuss how to estimate quadrature errors, building up from simple integrals in one dimension to layer potentials over smooth surfaces in three dimensions. We also discuss a new special quadrature technique for axisymmetric surfcaces with error control. The underlying technique is so-called interpolatory semi-analytical quadrature in conjunction with a singularity swap technique. Here, adaptive discretizations and parameters are set automatically given an error tolerance, utilizing further quadrature and interpolation error estimates derived for this purpose.
6/11/2024 Murtazo Nazarov, https://murtazo.github.io/ Links to an external site.
Title: A Nodal-Based High-Order Nonlinear Stabilization for Finite Element
Approximation of Conservation Laws
Abstract: We present a novel high-order nodal nonlinear artificial viscosity
method for solving conservation laws numerically. The traditional
artificial viscosity method usually requires several ad hoc parameters
and mesh sizes for each element. However, the precise choice of these
parameters and mesh-size functions is unclear for fully unstructured
meshes. Our new approach eliminates the need for ad hoc parameters. The
viscosity is mesh-dependent, yet an explicit definition of the mesh-size
is unnecessary. Our method employs a multimesh strategy: the viscosity
coefficient is constructed from a linear polynomial space defined on the
fine mesh, corresponding to the nodal values of the finite element
approximation space. In this way, the method adds a minimal amount of
viscosity for high-order polynomial spaces. We use the residual of the
underlying PDE to capture sharp discontinuities and shocks. This
approach is designed to precisely capture and resolve shocks. Then,
high-order Runge-Kutta methods are employed to discretize the temporal
domain. Through a comprehensive set of challenging test problems, we
validate the robustness and high-order accuracy of our proposed approach
for solving several nonlinear systems, including Burger's,
shallow-water, Euler, and magnetohydrodynamic equations.
10/10/2024 (TDB "1/4"-seminar)
Usama Zafar https://www.uu.se/kontakt-och-organisation/personal?query=N22-2641 Links to an external site.
Title: Secure Federated Machine Learning
Abstract: Artificial intelligence (AI) is central to modern-day applications and relies on three key components: data access, machine learning models, and training processes. Over time, the focus in AI development has shifted from mathematical modeling to improving training methods, and more recently, to addressing security, privacy, and trust concerns. While significant efforts have been made to address privacy and security challenges in Federated Machine Learning (FedML), achieving scalable, secure, and privacy-preserving federated environments remains an ongoing challenge.
In this seminar, I will briefly introduce the field of Federated Machine Learning, discuss the associated challenges, and present one of our recent works focused on enhancing security by addressing poisoning and adversarial attacks in FedML. A key challenge lies in detecting and mitigating model poisoning attacks, where compromised models from local agents can degrade the overall system. Our work explores methods for identifying and removing poisoned models during aggregation, closely related to anomaly detection techniques.
Junjie Wen https://www.uu.se/kontakt-och-organisation/personal?query=N22-2736 Links to an external site.
Title: Stabilized Finite Element Approximations for Vlasov-Maxwell Equations
Abstract: The Vlasov-Maxwell equation is a PDE that describes the distribution of plasma under self-consistent electromagnetic fields. This equation is important in plasma physics, and it is crucial to investigate the numerical solutions of it. In this talk, I plan to give a presentation about my research, and the main objective is to develop a structure-preserving, low-rank tensor discretization for high-dimensional Vlasov-Maxwell equations. I will introduce the numerical approximations for Vlasov-Maxwell equations using continuous finite element (FE) methods and the artificial viscosity scheme for stabilizing the FE approximations. Another focus of my talk is on conserving the invariant properties of the Vlasov-Maxwell equation, such as the divergence-free nature of the magnetic field and the Gauss’s law. Additionally, I will also talk about the future plan of my research, which involves positivity perserving and low-rank compression.
9/10/2024 Fredrik Liljeros, https://www.su.se/profiles/liljeros-1.184158 Links to an external site.
Title: On the Importance of Socially Structured Diffusion in the Modeling of Completely New Pandemics
Abstract: Traditional differential equation models have historically proven effective for modeling highly contagious diseases such as measles with good precision. However, they have been less successful in modeling sexually transmitted infections, tuberculosis, and MRSA. To model this type of disease, it is necessary to take into account the structural characteristics of the network of contacts through which the disease spreads.
In the seminar, I will, from a network epidemiological perspective, discuss factors that were missed in most attempts to model Covid-19 and highlight the systematic biases this led to. I will also discuss the epistemological challenges that the study of entirely new pandemics entails in the calibration of different types of pandemic models.
The talk will be based on the review article below as well as more recent publications.
Liljeros, F. (2009) Human sexual networks. In Encyclopedia Complexity and System Science: Springer Science and Business Media Link:
https://stockholmuniversity.box.com/s/mz248woil3xuxnhwbw4mydwd9p8mdqg6
Links to an external site.
2/10/2024 Fredrik Fryklund, https://www.uu.se/kontakt-och-organisation/personal?query=N24-777 Links to an external site.
Title: A Fast Algorithm for the Evaluation of Layer and Volume Potentials
Abstract: Over the last two decades, several fast, robust, and high-order accurate methods have been developed for solving the Poisson equation in complicated geometry using potential theory. In this approach, rather than discretizing the partial differential equation itself, one first evaluates a volume integral to account for the source distribution within the domain, followed by solving a boundary integral equation to impose the specified boundary conditions.
Here, I present a new fast algorithm which is easy to implement and compatible with virtually any discretization technique, including unstructured domain triangulations, such as those used in standard finite element or finite volume methods. The approach combines earlier work on potential theory for the heat equation, asymptotic analysis, and the nonuniform fast Fourier transform (NUFFT). It is insensitive to flaws in the triangulation, permitting not just nonconforming elements, but arbitrary aspect ratio triangles, gaps and various other degeneracies.
26/9/2024 (extra "TDB++" seminar) Xin Huang, https://www.kth.se/profile/xinhuang Links to an external site.
Title: Mean-Field Molecular Dynamics Approximation of Quantum Statistical Observables with Machine Learning
Abstract: This talk discusses the numerical approximation of quantum statistical observables from the perspective of mean-field molecular dynamics and machine learning modeling. We first introduce the approximation of canonical quantum correlation observables using classical molecular dynamics. Specifically by introducing a mean-field molecular dynamics approach where the nuclei's Hamiltonian is derived from the partial trace over electron degrees of freedom, we show that this method approximates quantum correlation observables with an accuracy of O(M^{-1}+ ε^2 t), with the nucleus-electron mass ratio parameter M, correlation time t, and parameter ε2 related to the variance of the mean-field approximation. We then extend the discussion to machine learning-based approaches, specifically using random Fourier feature neural networks to approximate the mean-field potential in Hamiltonian systems. Employing K nodes in the network and J data points in the training set, these neural networks yield molecular dynamics approximation of correlation observables with an expected error of O((K^{-1}+ J^{-1/2})^{1/2}). Numerical experiments validate the accuracy of the proposed approximation approach and demonstrate consistency with analytical error estimates.
18/9/2024 Annica Black-Schaffer, https://www.uu.se/en/contact-and-organisation/staff?query=N10-585 Links to an external site.
Title: Topological superconductivity
Abstract: Topological states of matter have emerged in the last decade as likely the most vibrant area of condensed matter physics. These range from topological insulators and Weyl semimetals to an abundance of different topological states in superconductors. The defining property for all topological phases is a global non-trivial topology of their electronic structure. This is fundamentally different from the traditional Landau paradigm traditionally used to classify matter, where local order parameters appearing due to spontaneous symmetry breaking play the key role. Topological superconductors are particularly interesting as they automatically join these two fundamentally different views of matter having both a global topological order and a local superconducting order parameter. Combining this with the distinctive particle-hole mixing in superconductors gives rise to emergent Majorana fermion states, that can be viewed as half electron quasiparticles, which offers unique possibilities to realize robust quantum computation.
In this talk I will give an introduction to topology in matter leading to topological superconductivity and the formation of Majorana fermions. I will also discuss some computational challenges appearing when modeling topological superconductors and how we can work to address these in order to be able to access properties of realistic systems.
11/9/2024 Eva Darulova, https://www.uu.se/kontakt-och-organisation/personal?query=N21-880 Links to an external site.
Title: Recent Advances in Floating-point Static Analyses
Abstract: Reasoning about floating-point rounding errors is essentially impossible to do manually for anyone but experts. Static analyzers for floating-point code compute a guaranteed bound on the worst-case rounding errors automatically at compile time, and can thus help (expert or non-expert) developers to ensure that their numerical computations are accurate enough. Doing this efficiently, while computing tight error bounds is still an open challenge for many programs. In this talk, I aim to give an intuition for the state-of-the-art static rounding error analysis techniques as well as an idea for what they can and (yet) cannot do.
5/9/2024 TDB Workshop on Computational Fusion & Plasma
Speakers: Remi Abgrall UZH, Mats Holmström IRF, Louis Richard UU/IRF, Murtazo Nazarov UU
Workshop program: download as pdf
Download download as pdf
4/9/2024 Rémi Abgrall, https://www.math.uzh.ch/people?key1=8882 Links to an external site.
Title: Approximation of non linear hyperbolic problems and the property of conservation
Abstract: download as pdf Download download as pdf
12/6/2024 Daniel Appelö, https://math.vt.edu/people/faculty/appelo-daniel.html Links to an external site.
Title: Quantum Digital Twins - a numerical methodist’s adventure in the land of quantum computers
Abstract: In this talk I will introducing the most basic concepts in quantum computing and describe one type of quantum computing hardware (a transmon) and how it is modeled. We will then outline the computationally challenging tasks that are needed for making a quantum computer run and introduce numerical methods tailored especially for these tasks. Time permitting I will take you on a comprehensive journey through a real-world example involving characterization, control, and experimental validation, showcasing our experiences with a qutrit device within the Lawrence Livermore QUDIT testbed.
11/6/2024 (extra "TDB++" seminar) Lukas Einkemmer, http://www.einkemmer.net/ Links to an external site.
Title: Higher order and implicit dynamical low-rank algorithms
Abstract: Dynamical low-rank algorithms have developed into an efficient way to solve high-dimensional problems ranging from plasma physics to quantum mechanics. Those methods are attractive because they reduce a high-dimensional problem into a set of lower-dimensional equations and thus can overcome the curse of dimensionality. This, for example, enables 6D Vlasov simulation on a desktop computer that otherwise would require large supercomputers.
Mathematically, dynamical low-rank methods project the true dynamics onto a low-rank manifold (the approximation space). To obtain a robust (i.e. well behaved) time integrator care has to be taken. Until very recently the literature has mostly focused on first-order and explicit methods. In this talk, we consider our recent work that allows us to construct higher-order and implicit robust dynamical low-rank schemes. Since each 'stage' of these methods solves a lower dimensional PDE it can be readily combined with standard iterative methods. We also provide some background on the state of the art of dynamical low-rank methods and show numerical examples from the field of plasma physics and radiative transfer.
5/6/2024 Adrian Muntean, https://www.kau.se/forskare/adrian-muntean Links to an external site.
Title: Phase separation and morphology formation in interacting ternary mixtures under evaporation: Well-posedness and numerical simulation of a non-local evolution system
Abstract: We study a nonlinear coupled parabolic system with non-local drift terms modeling at the continuum level the inter-species interaction within a ternary mixture that allows the evaporation of one of the species. In the absence of evaporation, the proposed system coincides with the hydrodynamic limit of a stochastic interacting particle system of Blume--Capel--type driven by the Kawasaki dynamics. Similar governing dynamics are found in models used to study morphology formation in the design of organic solar cells, thin adhesive bands, and other applications.
We investigate the well-posedness of the target system and present preliminary numerical simulations which incorporate `from the top' evaporation into the model. We employ a finite volumes scheme to construct approximations of the weak solution and illustrate how the evaporation process can affect the shape and connectivity of the evolving-in-time morphologies.
This is a report on recent joint work with Rainey Lyons (Karlstad), Andrea Muntean (Karlstad), and Emilio N.M. Cirillo (La Sapienza University, Rome).
29/5/2024 Erik Lindström, https://www.maths.lu.se/english/research/staff/erik-lindstroem Links to an external site.
Title: Feature Selection in Jump Models
Abstract: Hidden Markov models are often used for inferring hidden states as these can be interpreted as regimes. However, that interpretation implicitly assumes that the underlying state process has a certain level of persistence, which is not always the case, especially for high dimensional models. We propose a novel estimation approach based on temporal clustering of features by penalizing jumps between clusters. The advantages of the proposed jump estimator include that it learns the hidden state sequence and model parameters simultaneously while providing control over the transition rate, it is less sensitive to initialization, it performs better when the number of states increases, and is robust to misspecified conditional distributions.
Feature selection becomes necessary in high-dimensional settings where the number of features is large compared to the number of observations and/or when the underlying states differ only with respect to a subset of the features. We consequently develop a closed form coordinate descent algorithm for the extended feature selection problem that scales well to large data sets with large numbers of (noisy) features. The usefulness of the proposed framework is demonstrated by comparing it with several other methods, indicating that sparse jump model outperforms all other methods considered and is remarkably robust to noise.
Finally, the framework is applied to five leading cryptocurrencies, where more than 400 features derived from financial markets, sentiment variables and cryptomarkets are considered. The resulting model uses fewer and more persistent states than comparable hidden Markov Models. Furthermore, each state provides a coherent interpretation as a corresponding market regime, indicating the usefulness for applications.
28/5/2024 (extra "TDB++" seminar)
Anders M. N. Niklasson, Theoretical Division, Los Alamos National Laboratory, New Mexico, USA
Title: Machine learned shadow molecular dynamics with flexible charge models
Abstract: Atomistic simulation methods are currently undergoing a revolution thanks to new data-driven machine-learning techniques that can provide highly accurate short-range interatomic potentials. However, to truly unlock the full potential of these methods, we need to incorporate long-range electrostatic interactions and the associated charge relaxations. Including flexible atomic charges and long-range charge relaxations is crucial for many real-world problems, where we need to account for polarization, chemical reactions with charge transfer, and ionic fragmentation. Unfortunately, doing so dramatically increases the computational cost over traditional charge-independent force fields, requiring expensive iterative solvers in each integration time step, which may lead to non-conservative forces and instabilities. In this presentation, I will describe how a new form of approximate shadow molecular dynamics can circumvent these shortcomings with a significant reduction in computational cost. I will demonstrate examples, including neural network predictions of electronegativities and chemical hardness parameters, as well as calculations of the infrared spectrum from dipole auto-correlation functions. These advancements pave the way for efficient and accurate simulations of complex chemical systems, unlocking new frontiers in computational chemistry and materials science.
23/5/2024 (extra "TDB++" seminar)
Vladimir Druskin, https://www.wpi.edu/people/staff/vdruskin
Links to an external site.
Title: Matrix S-fractions and enhanced block-quadrature computation of MIMI transfer functions
Abstract: In this talk, we explore the quadratures B^T φ(A)B where A is a symmetric nonnegative-definite real NxN matrix, B is a tall, real N×p matrix, and φ(·) represents a matrix function, that is regular enough in the neighborhood of A’s spectrum, e.g., a Stieltjes or exponential function. These formulations, for example, commonly arise in the computation of multiple-input multiple-output (MIMO) transfer functions for diffusion PDEs.
We propose an approximation scheme for B^T φ(A)B leveraging the block Lanczos algorithm and its equivalent representation through Stieltjes matrix continued fractions. We extend the notion of Gauss-Radau quadrature to the block case, facilitating the derivation of easily computable, tight error bounds.
Our observations reveal that these error bounds are approximately twice the true error for problems stemming from the discretization of diffusion Maxwell’s equations in infinite domains. This insight enables us to develop cost-free extrapolation algorithms, which significantly enhance the accuracy of the block Lanczos approximation. We provide qualitative reasoning for such extrapolation, grounded in potential theory for Padé approximations.
22/5/2024 Johan Hoffman, https://www.kth.se/profile/jhoffman Links to an external site.
Title: The structure and evolution of turbulence as three fundamental flow structures and their stability
Abstract: We propose a model of turbulence in the form of three fundamental flow structures, each with its own distinct stability property: (i) stable rigid body rotational flow, (ii) exponentially unstable straining flow, and (iii) linearly unstable shear flow. According to this model, these are the only possible flow structures, and the evolution of any turbulent flow is determined by the stability and interaction of such flow structures. We describe how this triple decomposition of any velocity field can be computed in a unique and efficient way, and we present a refined energy stability analysis based on the triple decomposition. The model and methods are illustrated in computational examples.
References:
https://pubs.aip.org/aip/pof/article/35/3/031703/2881587
Links to an external site.
https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2022.806534/full
Links to an external site.
https://pubs.aip.org/aip/pof/article/33/8/081707/1056688/Energy-stability-analysis-of-turbulent
Links to an external site.
21/5/2024 (extra "TDB++" seminar)
Maeddeh Pourbagher, https://www.researchgate.net/profile/Maeddeh-Pourbagher
Links to an external site.
Title: On the iterative solution of complex symmetric systems of equations
Abstract: Complex systems of linear equations arise in important applications, for instance, in certain time-harmonic forms of some partial differential equations such as the Maxwell’s equation and the Helmholtz equation. In this talk, we will present a two-parameter iteration method to solve this type of system, specifically when the coefficient matrix of the corresponding system is indefinite. We will also show some estimates for the involved method parameters.
Further, we will present three efficient preconditioners for solving the Helmholtz equation and show their robustness with respect to discretization, problem and method parameters.
If time permits, we will discuss the iterative methods in the context of the system of linear equations arising from a distributed optimal control problem with a time-periodic equation as control.
15/5/2024 Ylva Ljungberg Rydin, https://www.foi.se/en/foi/research/underwater-technology Links to an external site.
Title: Numerical methods in underwater acoustics at the Swedish Defence Research Agency
Abstract: In air, the use of radio, radar, and optics is heavily relied on for communication, surveillance, and detection. Due to water's strong attenuation of electromagnetic waves, none of the above are available in the underwater domain. Instead, especially at longer ranges, sound waves are used in similar ways. Thus, accurate modeling and prediction of the propagation of sound waves is of great importance when operating underwater.
In this talk, we will give an overview of FOI's interests in the area of underwater sound propagation. We will describe some characteristics of the modeling problems and how they are approached using numerical methods.
8/5/2024 Martin Buhmann, https://www.uni-giessen.de/de/fbz/fb07/fachgebiete/mathematik/mathematik/ags/numerik/personen/prof/buhmann Links to an external site.
Title: New multiquadric-type interpolation and on a regular domain
Abstract: We study multivariable approximation in large dimensions by so-called radial basis functions ("kernel functions"). Their purpose is to approximate by multivariate interpolation mostly.
The applications of these methods are manifold, and they include big data approximation, image processing, quadrature, medical applications and optimisation.
Our main themes in this talk are two-fold: first we study a classical selection of kernels which are most useful in approximations -- which we then generalise into new classes of a so-far unknown fashion.
Secondly, we study these radial basis function approximations on special structures which take care of particular geometries of the underlying spaces. This can greatly improve the efficiency of the method, because it reduces the dimensions we work with when processing data in several dimensions.
17/4/2024 Niklas Wahlström, https://www.katalog.uu.se/profile/?id=N16-250 Links to an external site.
Title: Physics-informed neural networks with unknown measurement noise
Abstract: Physics-informed neural networks (PINNs) constitute a flexible approach to both finding solutions and identifying parameters of partial differential equations (PDEs). The main advantage of PINNs over traditional solvers lies in their flexibility: as neural networks, they have the capacity for universal function approximation, they are mesh-free, and they can directly be applied to very different kinds of PDEs, without the need to use a custom solver.
In this talk I will give an introduction to PINNs and present a recent work by ours, where we consider data contaminated by unknown measurement noise. Joint work with Philipp Pilar.
11/4/2024 (extra "TDB++" seminar)
Philip Gerlee, https://www.gu.se/om-universitetet/hitta-person/philipgerlee
Links to an external site.
Title: From single cells to tumours: bridging scales with individual-based models
Abstract: Glioblastoma is the most common type of primary brain tumour, and despite intense research efforts in the last 30 years, little improvement has occurred in the treatment of this tumour. One reason for this failure is the genetic and morphological diversity of tumours from different patients. While genetic information is readily available we still lack an understanding of how it translates to cellular properties and further into morphological characteristics of the tumour. In this talk I will give an overview of our recent attempts at understanding the link between single cell properties and the behaviour of populations of cancer cells. To this end we have created mathematical models of specific experiments, which ,given data in terms of microscopy images, provide us with estimates of cellular parameters such as the migration rate and rate of division. I will also discuss how parameter estimates from different assays can be compared and how they connect to the human disease.
10/4/2024 Peter Broquist, https://www.katalog.uu.se/profile?id=N11-1403
Links to an external site.
Title: Multi-scale Modelling of Battery Chemistries
Abstract: Multi-scale modelling has emerged as a powerful tool in advancing our understanding of battery chemistries. By integrating information from electronic and atomistic scales to mesoscopic and macroscopic levels, multi-scale models provide insights into complex phenomena such as charge transport, electrode-electrolyte interactions, and degradation mechanisms. However, many challenges concerning the coupling and linking of different entities and models used in the multi-scale approach need to be overcome. The future direction here is to empower the multi-scale models using AI and machine learning approaches, which hold immense promise for advancing battery technologies.
In this presentation, I will present how we use the multi-scale approach to i) understand intercalation processes in common electrode materials, and ii) how to hinder the degradation of electrolytes using radical scavengers. Practical considerations concerning the development of hybrid-multiscale models, combining physics-based models with machine learning, will be discussed.
9/4/2024 (extra "TDB++" seminar)
Bruno Blais, https://www.polymtl.ca/expertises/en/blais-bruno
Links to an external site.
Title: High-performance and open-source Euler-Lagrange modeling of dense particle-laden flows
Abstract: More than 50% of the products sold worldwide are in granular form or involve material in a granular state during their production process. Modelling granular materials remains highly challenging due to their complex rheology. Furthermore, in several unit operations (dryers, fluidized bed reactors, etc.), the granular material interacts with a fluid (gas or liquid). The dynamics of these particle-laden flows are difficult to predict, as the particle-fluid interaction at the particle scale (e.g. drag, lift) greatly affects the flow at the unit operation scale (macroscopic scale).
In this talk, we present a parallel open-source software (Lethe) that is based on the deal.II library. This software couples Computational Fluid Dynamics (CFD) and the Discrete Element Method (DEM) to solve dense particle-laden flow using high-order methods. First, we discuss the DEM model implemented within Lethe. We present the strategies used to ensure efficient data structure, parallelization and load balancing strategies within a Lagrangian framework. We then extend the model to particle-laden flow. We introduce the coupling strategy between the fluid and the particles as well as the modeling challenges that arise from the solution of the Volume-Averaged Navier-Stokes (VANS) equations.
Through the investigation of both simple and complex cases, we demonstrate that high-order CFD-DEM has significant potential as an accurate and robust strategy to simulate dense particle-laden flow. We conclude by providing a high-level perspective of the direction in which we are heading, the challenges that we are currently facing and the key lessons that have been learned through this endeavor to develop an open-source CFD/DEM/CFD-DEM software that scales to very large systems.
27/3/2024 Benny Avelin, https://www.katalog.uu.se/profile/?id=N11-2503 Links to an external site.
Title: Understanding the geometry of data sets with the Graph Laplacian
Abstract: High dimensional data sets are today ubiquitous in many areas of research, and understanding them is of central importance. Such a data set is commonly assumed to lie on some unknown manifold, and in this talk I will describe and introduce methods one can use to study the geometry of these manifolds, based mainly on the Graph Laplacian. In particular I will focus on how one can detect singularities, such as self intersections, and also methods to estimate the dimension of these data sets.
This is based on joint work with Martin Andersson.
20/3/2024 Pritpal 'Pip' Matharu https://people.kth.se/~pritpal/
Links to an external site.
Title: Adjoint-Based Enforcement of State Constraints in PDE Optimization Problems
Abstract: Adjoint-based methods have become a workhorse in the solution of unconstrained PDE optimization problems. They make it possible to conveniently determine the gradient (sensitivity) of the objective functional with respect to a control variable, which can then be used in various gradient descent algorithms. Unlike most constraints imposed on the control variable, constraints on the state variables are generally harder to satisfy since they define, via solutions of the governing system, complicated manifolds in the space of control variables. In this talk, we will demonstrate how this traditional adjoint-based framework can be extended to handle general constraints on the state variables. This is accomplished by constructing a projection of the gradient of the objective functional onto a subspace tangent to the manifold defined by the constraint. We focus on the “optimize-then-discretize” paradigm in the infinite-dimensional setting where the required regularity of both the gradient and of the projection is ensured. This proposed approach will be illustrated first with a simple test problem describing optimization of heat transfer in one direction and then a more involved problem where an optimal closure is found for a turbulent flow described by the Navier-Stokes system in two dimensions.
Joint work with Bartosz Protas.
13/3/2024 Tobias Wrigstad, http://wrigstad.com/ Links to an external site.
Title: The Nature of Programming
Abstract: In this talk, I will ask and possibly try to answer the question "what is programming". This question should have bearing on our teaching, but it may be that this question is answered differently depending on, say, whether you are in scientific computing or in computing science. And on that note: What are the foundational principles in programming that all our students should learn, how do we ensure that two students form the same mental model of the semantics of a programming language, and is it possible to get a 60k/month salary without knowing whether your programming language uses call by value or call by reference?
In the end, given time, I will also talk about some recent work on Verona, a programming language somewhere between C++ and Java, that we are developing together with Microsoft Research.
6/3/2024 Joakim Lindblad, https://www.katalog.uu.se/profile/?id=N5-1054 Links to an external site.
Title: Automated alignment of image data – not as easy as it seems
Abstract: Image registration is the process of finding the spatial transformation which aligns images of the same or similar scenes, mapping them into the same coordinate system. The data to align may be acquired from different viewpoints, times, under different conditions, or with different types of sensors. Image registration is necessary in order to compare or integrate the information from these different sources and is a common operation in computer vision and medical imaging. A well defined problem which should be fairly easy to solve – or so it seems.
In this talk, I will present a few state-of-the-art methods for mono- and multi-modal image registration, demonstrate a way to overcome problems with deep learning-based image-to-image translation, show a clever way to compute mutual information in the frequency domain, and invite to further discussion on how to best address the still open problem of efficient and reliable image registration.
14/2/2024 Anna Rising, https://ki.se/en/bionut/spider-silk-biology-for-biomedical-applications-anna-rising, Links to an external site.https://www.slu.se/cv/anna-rising Links to an external site.
Title: How to make artificial spider silk
Abstract: Spiders have over the last 400 million years developed the ability to spin the toughest fibers known to man in a process that is completely water based and void of high temperatures. Thus, spider silk represents an attractive material for many different applications and a viable alternative to many environmentally harmful synthetic fibers. However, production of the spider silk proteins (spidroins) is problematic due to their repetitiveness and propensity to aggregate [1].
We have developed an E.coli based production method that generates unprecedented amounts of correctly folded and soluble spidroins [3]. A biomimetic spinning method combined with a protein engineering strategy, result in artificial spider silk fibers that match the toughness of native spider silk [3--5]. The fibers have succesfully been used to guide the extension of neurites in cell culture assays [6] and their usefulness in textile applications is being explored.
1. Bourzac, K. Spiders: Web of intrigue. Nature 519, S4-6 (2015). https://doi.org:10.1038/519S4a
Links to an external site.
2. Rising A and Harrington MJ. Biological Materials Processing: Time-tested tricks for sustainable fiber fabrication. Chem Rev. 123, 5, 2155-2199 (2023) https://doi.org/10.1021/acs.chemrev.2c00465
Links to an external site.
3. Schmuck, B. et al. High-yield production of a super-soluble miniature spidroin for biomimetic high-performance materials. Mater Today 50, 16-23 (2021). https://doi.org:10.1016/j.mattod.2021.07.020
Links to an external site.
4. Arndt, T. et al. Engineered Spider Silk Proteins for Biomimetic Spinning of Fibers with Toughness Equal to Dragline Silks. Adv Funct Mater 32 (2022). https://doi.org/10.1002/adfm.202200986
Links to an external site.
5. Andersson, M. et al. Biomimetic spinning of artificial spider silk from a chimeric minispidroin. Nat Chem Biol 13, 262-264 (2017). https://doi.org:10.1038/nchembio.2269
Links to an external site.
6. Hansson, M. L. et al. Artificial spider silk supports and guides neurite extension in vitro. FASEB J 35, e21896 (2021). https://doi.org:10.1096/fj.202100916R
Links to an external site.
7/2/2024 Ricardo Vinuesa, https://www.vinuesalab.com/ Links to an external site.
Title: Explaining and controlling turbulent flows through deep learning
Abstract: In this contribution we first use a framework for deep-learning explainability to identify the most important Q events in two wall-bounded turbulent flows: a numerical channel at Retau=125 and an experimental boundary layer at Retau=1,000. This objective way to assess importance reveals that the most important Q events are not necessarily the ones with the highest Reynolds shear stress. Interestingly, this framework is used to identify completely new coherent structures: we find that the most important coherent regions in the flow only have an overlap of 70% with the classical Q events.
In the second part of the presentation we use deep reinforcement learning (DRL) to discover completely new strategies of active flow control. We show that DRL applied to a blowing-and-suction scheme significantly outperforms the classical opposition control in a turbulent channel at Retau=180: while the former yields 30% drag reduction, the latter only 20%. We conclude that DRL has tremendous potential for drag reduction in a wide range of complex turbulent-flow configurations.
31/1/2024 Jonas Lindemann, https://portal.research.lu.se/sv/persons/jonas-lindemann Links to an external site.
Title: Modernizing a Desktop C++ Application: A Refactoring Journey
Abstract: This talk delves into the challenges and strategies of refactoring and modernizing ObjectiveFrame, a large-scale C++ desktop application initially developed in the early 2000s. ObjectiveFrame has undergone many updates over the years as an interactive finite element tool to explore new concepts in real-time computation and user interaction. However, the aging infrastructure of the application will require significant upgrades to be able to add more features.
Key focus areas included the user interface library, which lacked contemporary features and experienced compatibility issues with OpenGL. The computational engine, reliant on a single-threaded matrix library, also required modernization. The domain object model also needed to incorporate advanced C++ features like smart pointers.
This talk describes the journey from the application's early versions to its latest iteration in 2023, highlighting the transformative process and the major upgrades undertaken. It also offers insights into adapting a legacy application to meet current technological standards and user expectations.
24/1/2024 Gabriel Skantze, https://www.kth.se/profile/skantze Links to an external site.
Title: Predictive Modelling of Spoken Interaction for Conversational AI
Abstract: Being able to communicate with machines through spoken conversation, in the same way we naturally communicate with each other, has been a long-standing vision in both science fiction and research labs, and it has been considered a hallmark of human intelligence. In recent years, so-called Conversational AI has started to become a reality, in the form of smart speakers, voice assistants, chatbots and social robots. However, the experience of talking to these systems is still very far from what it is like talking to another person. Part of the reason for this is that current systems do not understand the intricate facial and vocal coordination signals humans use in conversation. In this talk, I will present our work on predictive modelling of spoken interaction. Through self-supervised learning, we train a model to continuously predict the next few seconds of a spoken interaction, and in that process the model learns to pick up communicative signals necessary for coordinating fluid spoken interaction.
17/1/2024 Sven Nelander, https://www.igp.uu.se/forskning/neuroonkologi-neurodegeneration/sven-nelander Links to an external site.
Title: Uncovering the genetic circuitry behind brain tumor invasion from interventional single-cell data