arXiv subjects

Matthias Bolten

Publications and source records attributed to Matthias Bolten.

At least 19 recordsLinked to original sources

arXiv2026

Spectral deferred corrections parallelized across the method for differential-algebraic equations

In this work, the performance of different spectral deferred corrections (SDC) methods applied to initial value problems for differential-algebraic equations (DAEs) of index one is analyzed. The SDC method solves a series of correction equations, and after each iteration, the numerical solution is corrected by adding the obtained approximation of the error. The formulation of the SDC method allows for a parallelization across the method to obtain small-scale parallelism, for which a number of processes equal to the number of collocation nodes can be used. Recently, an analytical approach to finding optimal diagonal coefficients for parallel SDC was proposed. So far, no analysis with the new coefficients for parallel SDC applied to DAEs was done. We demonstrate that parallel SDC methods solve DAE problems with high accuracy faster than the associated sequential SDC schemes, obtaining speedup in a small-scale parallel regime.

math.NA↗

arXiv2026

On the analysis of spectral deferred corrections for differential-algebraic equations of index one

In this paper, we present a new spectral deferred corrections (SDC) method to solve semi-explicit differential-algebraic equations (DAEs) with the ability to be parallelized. The new scheme restricts numerical integration to differential equations. In Y. Xia et al. (2007), it was shown that each correction elevates the order of the solution by one. We show that this carries over to the new SDC scheme. The derivation of the method combines the approach of SDC and the idea to enforce the algebraic constraints without numerical integration as shown in the $\varepsilon$-embedding method by E. Hairer and G. Wanner (1996). Keeping the algebraic equations as an implicit condition of the system allows an efficient solve of semi-explicit DAEs with high-accuracy. The proposed scheme is compared with other DAE methods. We demonstrate that the proposed SDC scheme is competitive with Runge-Kutta methods for DAEs in terms of accuracy and its parallelized versions are very efficient compared to their associated sequential SDC variants.

math.NA↗

arXiv2025

Event-driven eligibility propagation in large sparse networks: efficiency shaped by biological realism

Despite remarkable technological advances, AI systems may still benefit from biological principles, such as recurrent connectivity and energy-efficient mechanisms. Drawing inspiration from the brain, we present a biologically plausible extension of the eligibility propagation (e-prop) learning rule for recurrent spiking networks. By translating the time-driven update scheme into an event-driven one, we integrate the learning rule into a simulation platform for large-scale spiking neural networks and demonstrate its applicability to tasks such as neuromorphic MNIST. We extend the model with prominent biological features such as continuous dynamics and weight updates, strict locality, and sparse connectivity. Our results show that biologically grounded constraints can inform the design of computationally efficient AI algorithms, offering scalability to millions of neurons without compromising learning performance. This work bridges machine learning and computational neuroscience, paving the way for sustainable, biologically inspired AI systems while advancing our understanding of brain-like learning.

cs.NE↗

arXiv2024

Analysis on aggregation and block smoothers in multigrid methods for block Toeplitz linear systems

We present novel improvements in the context of symbol-based multigrid procedures for solving large block structured linear systems. We study the application of an aggregation-based grid transfer operator that transforms the symbol of a block Toeplitz matrix from matrix-valued to scalar-valued at the coarser level. Our convergence analysis of the Two-Grid Method (TGM) reveals the connection between the features of the scalar-valued symbol at the coarser level and the properties of the original matrix-valued one. This allows us to prove the convergence of a V-cycle multigrid with standard grid transfer operators for scalar Toeplitz systems at the coarser levels. Consequently, we extend the class of suitable smoothers for block Toeplitz matrices, focusing on the efficiency of block strategies, particularly the relaxed block Jacobi method. General conditions on smoothing parameters are derived, with emphasis on practical applications where these parameters can be calculated with negligible computational cost. We test the proposed strategies on linear systems stemming from the discretization of differential problems with $\mathbb{Q}_{d} $ Lagrangian FEM or B-spline with non-maximal regularity. The numerical results show in both cases computational advantages compared to existing methods for block structured linear systems.

math.NA↗

arXiv2023

Symbol based Convergence Analysis in Block Multigrid Methods with Applications for Stokes Problems

The main focus of this paper is the study of efficient multigrid methods for large linear systems with a particular saddle-point structure. Indeed, when the system matrix is symmetric, but indefinite, the variational convergence theory that is usually used to prove multigrid convergence cannot be directly applied. However, different algebraic approaches analyze properly preconditioned saddle-point problems, proving convergence of the Two-Grid method. In particular, this is efficient when the blocks of the coefficient matrix possess a Toeplitz or circulant structure. Indeed, it is possible to derive sufficient conditions for convergence and provide optimal parameters for the preconditioning of the saddle-point problem in terms of the associated generating symbols. In this paper, we propose a symbol-based convergence analysis for problems that have a hidden block Toeplitz structure. Then, they can be investigated focusing on the properties of the associated generating function f, which consequently is a matrix-valued function with dimension depending on the block size of the problem. As numerical tests we focus on the matrix sequence stemming from the finite element approximation of the Stokes problem. We show the efficiency of the methods studying the hidden 9-by-9 block multilevel structure of the obtained matrix sequence. Moreover, we propose an efficient algebraic multigrid method with convergence rate independent of the matrix size. Finally, we present several numerical tests comparing the results with state-of-the-art strategies.

math.NA↗

arXiv2023

Multigrid preconditioning for regularized least-squares problems

In this paper, we are concerned with efficiently solving the sequences of regularized linear least squares problems associated with employing Tikhonov-type regularization with regularization operators designed to enforce edge recovery. An optimal regularization parameter, which balances the fidelity to the data with the edge-enforcing constraint term, is typically not known a priori. This adds to the total number of regularized linear least squares problems that must be solved before the final image can be recovered. Therefore, in this paper, we determine effective multigrid preconditioners for these sequences of systems. We focus our approach on the sequences that arise as a result of the edge-preserving method introduced in [6], where we can exploit an interpretation of the regularization term as a diffusion operator; however, our methods are also applicable in other edge-preserving settings, such as iteratively reweighted least squares problems. Particular attention is paid to the selection of components of the multigrid preconditioner in order to achieve robustness for different ranges of the regularization parameter value. In addition, we present a parameter culling approach that, when used with the L-curve heuristic, reduces the total number of solves required. We demonstrate our preconditioning and parameter culling routines on examples in computed tomography and image deblurring.

math.NA↗

arXiv2023

Non-convex shape optimization by dissipative Hamiltonian flows

Shape optimization with constraints given by partial differential equations (PDE) is a highly developed field of optimization theory. The elegant adjoint formalism allows to compute shape gradients at the computational cost of a further PDE solve. Thus, gradient descent methods can be applied to shape optimization problems. However, gradient descent methods that can be understood as approximation to gradient flows get stuck in local minima, if the optimization problem is non-convex. In machine learning, the optimization in high dimensional non-convex energy landscapes has been successfully tackled by momentum methods, which can be understood as passing from gradient flow to dissipative Hamiltonian flows. In this paper, we adopt this strategy for non-convex shape optimization. In particular, we provide a mechanical shape optimization problem that is motivated by optimal reliability considering also material cost and the necessity to avoid certain obstructions in installation space. We then show how this problem can be solved effectively by port Hamiltonian shape flows.

math.OC↗

arXiv2022

Symbol based convergence analysis in multigrid methods for saddle point problems

Saddle point problems arise in a variety of applications, e.g., when solving the Stokes equations. They can be formulated such that the system matrix is symmetric, but indefinite, so the variational convergence theory that is usually used to prove multigrid convergence cannot be applied. In a 2016 paper in Numerische Mathematik Notay has presented a different algebraic approach that analyzes properly preconditioned saddle point problems, proving convergence of the Two-Grid method. In the present paper we analyze saddle point problems where the blocks are circulant within this framework. We are able to derive sufficient conditions for convergence and provide optimal parameters for the preconditioning of the saddle point problem and for the point smoother that is used. The analysis is based on the generating symbols of the circulant blocks. Further, we show that the structure can be kept on the coarse level, allowing for a recursive application of the approach in a W- or V-cycle and proving the "level independency" property. Numerical results demonstrate the efficiency of the proposed method in the circulant and the Toeplitz case.

math.NA↗

arXiv2022

A note on the spectral analysis of matrix sequences via GLT momentary symbols: from all-at-once solution of parabolic problems to distributed fractional order matrices

The first focus of this paper is the characterization of the spectrum and the singular values of the coefficient matrix stemming from the discretization with space-time grid for a parabolic diffusion problem and from the approximation of distributed order fractional equations. For this purpose we will use the classical GLT theory and the new concept of GLT momentary symbols. The first permits to describe the singular value or eigenvalue asymptotic distribution of the sequence of the coefficient matrices, the latter permits to derive a function, which describes the singular value or eigenvalue distribution of the matrix of the sequence, even for small matrix-sizes but under given assumptions. The note is concluded with a list of open problems, including the use of our machinery in the study of iteration matrices, especially those concerning multigrid-type techniques.

math.NA↗

arXiv2022

Toeplitz Momentary Symbols: definition, results, and limitations in the spectral analysis of Structured Matrices

A powerful tool for analyzing and approximating the singular values and eigenvalues of structured matrices is the theory of GLT sequences. By the GLT theory one can derive a function, which describes the singular value or the eigenvalue distribution of the sequence, the latter under precise assumptions. However, for small values of the matrix size of the considered sequence, the approximations may not be as good as it is desirable, since in the construction of the GLT symbol one disregards small norm and low-rank perturbations. On the other hand, LFA can be used to construct polynomial symbols in a similar manner for discretizations, where the geometric information is present, but the small norm perturbations are retained. The main focus of this paper is the introduction of the concept of sequence of "Toeplitz momentary symbols", associated with a given sequence of truncated Toeplitz-like matrices. We construct the symbol in the same way as in the GLT theory, but we keep the information of the small norm contributions. The low-rank contributions are still disregarded, and we give an idea on the reason why this is negligible in certain cases and why it is not in other cases, being aware that in presence of high nonnormality the same low-rank perturbation can produce a dramatic change in the eigenvalue distribution. Moreover, a difference with respect to the LFA symbols is that GLT symbols and Toeplitz momentary symbols are more general and are applicable to a larger class of matrices. We show the applicability of the approach which leads to higher accuracy in some cases when compared with the GLT symbol. Finally, since for many applications and their analysis it is often necessary to consider non-square Toeplitz matrices, we formalize and provide some useful definitions, applicable for non-square Toeplitz momentary symbols.

math.NA↗

arXiv2021

A symbol based analysis for multigrid methods for Block-Circulant and Block-Toeplitz Systems

In the literature, there exist several studies on symbol-based multigrid methods for the solution of linear systems having structured coefficient matrices. In particular, the convergence analysis for such methods has been obtained in an elegant form in the case of Toeplitz matrices generated by a scalar-valued function. In the block-Toeplitz setting, that is, in the case where the matrix entries are small generic matrices instead of scalars, some algorithms have already been proposed regarding specific applications and a first rigorous convergence analysis has been performed in [7]. However, with the existent symbol-based theoretical tools, it is still not possible to prove the convergence of many multigrid methods known in the literature. This paper aims to generalize the previous results giving more general sufficient conditions on the symbol of the grid transfer operators.In particular, we treat matrix-valued trigonometric polynomials which can be non-diagonalizable and singular at all points and we express the new conditions in terms of the eigenvectors associated with the ill-conditioned subspace. Moreover, we extend the analysis to the V-cycle method proving a linear convergence rate under stronger conditions, which resemble those given in the scalar case. In order to validate our theoretical findings, we present a classical block structured problem stemming from a FEM approximation of a second order differential problem. We focus on two multigrid strategies that use the geometric and the standard bisection grid transfer operators and we prove that both fall into the category of projectors satisfying the proposed conditions. In addition, using a tensor product argument, we provide a strategy to construct efficient V-cycle procedures in the block multilevel setting.

math.NA↗

arXiv2021

Using ODE waveform-relaxation methods to efficiently include gap junctions in distributed neural network simulations

Waveform-relaxation methods divide systems of differential equations into subsystems and therefore allow for parallelization across the system. Here we present an application for ODE waveform-relaxation methods in the context of spiking neural network simulators. Parallel spiking neural network simulators make use of the fact that the dynamics of neurons with chemical synapses is decoupled for the duration of the minimal network delay and thus can be solved independently for this duration. The inclusion of electrical synapses, so-called gap junctions, requires continuous interaction between neurons and therefore constitutes a conceptional problem for those simulators. We present a suitable waveform-relaxation method for an efficient integration of gap junctions and demonstrate that the use of the waveform-relaxation method improves both, accuracy and performance, compared to a non-iterative solution of the problem. We investigate the employed method in a reference implementation in the parallel spiking neural network simulator NEST.

math.NA↗

arXiv2021

Event-based update of synapses in voltage-based learning rules

Due to the point-like nature of neuronal spiking, efficient neural network simulators often employ event-based simulation schemes for synapses. Yet many types of synaptic plasticity rely on the membrane potential of the postsynaptic cell as a third factor in addition to pre- and postsynaptic spike times. Synapses therefore require continuous information to update their strength which a priori necessitates a continuous update in a time-driven manner. The latter hinders scaling of simulations to realistic cortical network sizes and relevant time scales for learning. Here, we derive two efficient algorithms for archiving postsynaptic membrane potentials, both compatible with modern simulation engines based on event-based synapse updates. We theoretically contrast the two algorithms with a time-driven synapse update scheme to analyze advantages in terms of memory and computations. We further present a reference implementation in the spiking neural network simulator NEST for two prototypical voltage-based plasticity rules: the Clopath rule and the Urbanczik-Senn rule. For both rules, the two event-based algorithms significantly outperform the time-driven scheme. Depending on the amount of data to be stored for plasticity, which heavily differs between the rules, a strong performance increase can be achieved by compressing or sampling of information on membrane potentials. Our results on computational efficiency related to archiving of information provide guidelines for the design of learning rules in order to make them practically usable in large-scale networks.

q-bio.NC↗

arXiv2020

Towards Real-Time Magnetic Dosimetry Simulations for Inductive Charging Systems

The exposure of a human by magneto-quasistatic fields from wireless charging systems is to be determined from magnetic field measurements in near real-time. This requires a fast linear equations solver for the discrete Poisson system of the Co-Simulation Scalar Potential Finite Difference (Co-Sim. SPFD) scheme. Here, the use of the AmgX library on NVIDIA GPUs is presented for this task. It enables solving the equation system resulting from an ICNIRP recommended human voxel model resolution of 2 mm in less than 0.5 seconds on a single NVIDIA Tesla V100 GPU.

cs.CE↗

arXiv2020

Krylov Subspace Recycling for Evolving Structures

Krylov subspace recycling is a powerful tool for solving long series of large, sparse linear systems that change slowly. In PDE constrained shape optimization, these appear naturally, as hundreds or more optimization steps are needed with only small changes in the geometry. In this setting, however, applying Krylov subspace recycling can be difficult. As the geometry evolves, so does the finite element mesh, especially if re-meshing is needed. As a result, the number of algebraic degrees of freedom in the system may change from one optimization step to the next, and with it the size of the finite element system matrix. Changes in the mesh also lead to structural changes in the matrices. In the case of remeshing, even if the geometry changes only a little, the corresponding mesh might differ substantially from the previous one. This prevents any straightforward mapping of the approximate invariant subspace of the linear system matrix (the focus of recycling in this paper) from one step to the next; similar problems arise for other selected subspaces. We present an algorithm for general meshes to map an approximate invariant subspace of the system matrix for the previous optimization step to an approximate invariant subspace of the system matrix for the current optimization step. We exploit the map from coefficient vectors to finite element functions on the mesh combined with function approximation on the finite element mesh. In addition, we develop a straightforward warm-start adaptation of the Krylov-Schur algorithm [G.W. Stewart, SIAM J. Matrix Anal. Appl. 23, 2001] to improve the approximate invariant subspace at the start of a new optimization step if needed. We demonstrate the effectiveness of our approach numerically with several proof of concept studies for a specific meshing technique.

math.NA↗

arXiv2020

Non-overlapping block smoothers for the Stokes equations

Overlapping block smoothers efficiently damp the error contributions from highly oscillatory components within multigrid methods for the Stokes equations but they are computationally expensive. This paper is concentrated on the development and analysis of new block smoothers for the Stokes equations that are discretized on staggered grids. These smoothers are non-overlapping and therefore desirable due to reduced computational costs. Traditional geometric multigrid methods are based on simple pointwise smoothers. However, the efficiency of multigrid methods for solving more difficult problems such as the Stokes equations lead to computationally more expensive smoothers, e.g., overlapping block smoothers. Non-overlapping smoothers are less expensive, but have been considered less efficient in the literature. In this paper, we develop new non-overlapping smoothers, the so-called triad-wise smoothers, and show their efficiency within multigrid methods to solve the Stokes equations. In addition, we compare overlapping and non-overlapping smoothers by measuring their computational costs and analyzing their behavior by the use of local Fourier analysis.

math.NA↗

arXiv2020

PyMGRIT: A Python Package for the parallel-in-time method MGRIT

In this paper, we introduce the Python framework PyMGRIT, which implements the multigrid-reduction-in-time (MGRIT) algorithm for solving the (non-)linear systems arising from the discretization of time-dependent problems. The MGRIT algorithm is a reduction-based iterative method that allows parallel-in-time simulations, i. e., calculating multiple time steps simultaneously in a simulation, by using a time-grid hierarchy. The PyMGRIT framework features many different variants of the MGRIT algorithm, ranging from different multigrid cycle types and relaxation schemes, as well as various coarsening strategies, including time-only and space-time coarsening, to using different time integrators on different levels in the multigrid hierachy. PyMGRIT allows serial runs for prototyping and testing of new approaches, as well as parallel runs using the Message Passing Interface (MPI). Here, we describe the implementation of the MGRIT algorithm in PyMGRIT and present the usage from both user and developer point of views. Three examples illustrate different aspects of the package, including pure time parallelism as well as space-time parallelism by coupling PyMGRIT with PETSc or Firedrake, which enable spatial parallelism through MPI.

cs.MS↗

arXiv2020

Parallel-in-Time Simulation of an Electrical Machine using MGRIT

We apply the multigrid-reduction-in-time (MGRIT) algorithm to an eddy current simulation of a two-dimensional induction machine supplied by a pulse-width-modulation signal. To resolve the fast-switching excitations, small time steps are needed, such that parallelization in time becomes highly relevant for reducing the simulation time. The MGRIT algorithm is well suited for introducing time parallelism in the simulation of electrical machines using existing application codes, as MGRIT is a non-intrusive approach that essentially uses the same time integrator as a traditional time-stepping algorithm. We investigate effects of spatial coarsening on MGRIT convergence when applied to two numerical models of an induction machine, one with linear material laws and a full nonlinear model. Parallel results demonstrate significant speedup in the simulation time compared to sequential time stepping, even for moderate numbers of processors.

math.NA↗