arXiv subjects

Matthias Kirchhart

Publications and source records attributed to Matthias Kirchhart.

10 recordsLinked to original sources

arXiv2026

Extending the Numerical Flow Iteration to the multi-species Vlasov-Maxwell system through Hamiltonian Splitting

The Numerical Flow Iteration (NuFI) method has recently been proposed as a memory-slim while accurate in phase-space method for the electro-static Vlasov--Poisson system. It stores the temporal evolution of the electric field, instead of the distribution functions, and reconstructs the solution in each time step by following the characteristics backwards in time and reconstructing the solution from the initial distribution. NuFI has been shown to be more accurate than other state-of-the-art electro-static Vlasov solvers given the same amount of degrees of freedom. In this paper, we build on the Hamiltonian structure of the full Vlasov--Maxwell system to extend NuFI to handle electro-magnetic kinetic plasma dynamics. We show that the structure-preserving properties of the NuFI time-stepping are preserved when extending to the electro-magnetic case. Furthermore we discuss how NuFI can be incorporated into existing Semi-Lagrangian codes as an efficient while accurate subcycling technique.

physics.plasm-ph↗

arXiv2024

Fully-fused Multi-Layer Perceptrons on Intel Data Center GPUs

This paper presents a SYCL implementation of Multi-Layer Perceptrons (MLPs), which targets and is optimized for the Intel Data Center GPU Max 1550. To increase the performance, our implementation minimizes the slow global memory accesses by maximizing the data reuse within the general register file and the shared local memory by fusing the operations in each layer of the MLP. We show with a simple roofline model that this results in a significant increase in the arithmetic intensity, leading to improved performance, especially for inference. We compare our approach to a similar CUDA implementation for MLPs and show that our implementation on the Intel Data Center GPU outperforms the CUDA implementation on Nvidia's H100 GPU by a factor up to 2.84 in inference and 1.75 in training. The paper also showcases the efficiency of our SYCL implementation in three significant areas: Image Compression, Neural Radiance Fields, and Physics-Informed Machine Learning. In all cases, our implementation outperforms the off-the-shelf Intel Extension for PyTorch (IPEX) implementation on the same Intel GPU by up to a factor of 30 and the CUDA PyTorch version on Nvidia's H100 GPU by up to a factor 19. The code can be found at https://github.com/intel/tiny-dpcpp-nn.

cs.AI↗

arXiv2023

The Numerical Flow Iteration for the Vlasov-Poisson equation

We present the numerical flow iteration (NuFI) for solving the Vlasov--Poisson equation. In a certain sense specified later herein, NuFI provides infinite resolution of the distribution function. NuFI exactly preserves positivity, all $L^p$-norms, charge, and entropy. Numerical experiments show no energy drift. NuFI is fast, requires several orders of magnitude less memory than conventional approaches, and can very efficiently be parallelised on GPU clusters. Low fidelity simulations provide good qualitative results for extended periods of time and can be computed on low-cost workstations.

math.NA↗

arXiv2022

An interpolating particle method for the Vlasov-Poisson equation

In this paper we present a novel particle method for the Vlasov--Poisson equation. Unlike in conventional particle methods, the particles are not interpreted as point charges, but as point values of the distribution function. In between the particles, the distribution function is reconstructed using mesh-free interpolation. Our numerical experiments confirm that this approach results in significantly increased accuracy and noise reduction. At the same time, many benefits of the conventional schemes are preserved.

math.NA↗

arXiv2022

Solidfmm: A highly optimised library of operations on the solid harmonics for use in fast multipole methods

We present solidfmm, a highly optimised C++ library for the solid harmonics as they are needed in fast multipole methods. The library provides efficient, vectorised implementations of the translation operations M2M, M2L, and L2L, and is available as free software. While asymptotically of complexity $O(P^3)$, for all practically relevant expansion orders, the translation operators display an empirical complexity of $O(P^2)$, outperforming the naïve implementation by orders of magnitude.

cs.MS↗

arXiv2021

Div-Curl Problems and $\mathbf{H}^1$-regular Stream Functions in 3D Lipschitz Domains

We consider the problem of recovering the divergence-free velocity field ${\mathbf U}\in\mathbf{L}^2(Ω)$ of a given vorticity ${\mathbf F}=\mathrm{curl}\,{\mathbf U}$ on a bounded Lipschitz domain $Ω\subset\mathbb{R}^3$. To that end, we solve the "div-curl problem" for a given ${\mathbf F}\in{\mathbf H}^{-1}(Ω)$. The solution is expressed in terms of a vector potential (or stream function) ${\mathbf A}\in{\mathbf H}^1(Ω)$ such that ${\mathbf U}=\mathrm{curl}\,{\mathbf A}$. After discussing existence and uniqueness of solutions and associated vector potentials, we propose a well-posed construction for the stream function. A numerical method based on this construction is presented, and experiments confirm that the resulting approximations display higher regularity than those of another common approach.

math.AP↗

arXiv2020

Analytic Integration of the Newton Potential over Cuboids and an Application to Fast Multipole Methods

We present simplified formulae for the analytic integration of the Newton potential of polynomials over boxes in two- and three-dimensional space. These are implemented in an easy-to-use C++ library that allows computations in arbitrary precision arithmetic which is also documented here. We describe how these results can be combined with fast multipole methods for general, non-polynomial data.

math.NA↗

arXiv2019

On Particles and Splines in Bounded Domains

We propose numerical schemes that enable the application of particle methods for advection problems in general bounded domains. These schemes combine particle fields with Cartesian tensor product splines and a fictitious domain approach. Their implementation only requires a fitted mesh of the domain's boundary, and not the domain itself, where an unfitted Cartesian grid is used. We establish the stability and consistency of these schemes in $W^{s,p}$-norms, $s\in\mathbb{R}$, $1<p\leq\infty$.

math.NA↗

arXiv2017

A Smooth Partition of Unity Finite Element Method for Vortex Particle Regularization

We present a new class of $C^\infty$-smooth finite element spaces on Cartesian grids, based on a partition of unity approach. We use these spaces to construct smooth approximations of particle fields, i.e., finite sums of weighted Dirac deltas. In order to use the spaces on general domains, we propose a fictitious domain formulation, together with a new high-order accurate stabilization. Stability, convergence, and conservation properties of the scheme are established. Numerical experiments confirm the analysis and show that the Cartesian grid-size $σ$ should be taken proportional to the square-root of the particle spacing $h$, resulting in significant speed-ups in vortex methods.

math.NA↗

arXiv2016

A Splitting-free Vorticity Redistribution Method

We present a splitting-free variant of the vorticity redistribution method. Spatial consistency and stability when combined with a time-stepping scheme are proven. We propose a new strategy preventing excessive growth in the number of particles while retaining the order of consistency. The novel concept of small neighbourhoods significantly reduces the method's computational cost. In numerical experiments the method showed second order convergence, one order higher than predicted by the analysis. Compared to the fast multipole code used in the velocity computation, the method is about three times faster.

math.NA↗