arXiv subjects

Simon Selg

Publications and source records attributed to Simon Selg.

3 recordsLinked to original sources

arXiv2024

Magnetic field and kinetic helicity evolution in simulations of interacting disk galaxies

We carried out a parameter study of interacting disk galaxies with impact parameters ranging from central collisions to weakly interacting scenarios. The orientations of the disks were also varied. In particular, we investigated how magnetic field amplification depends on these parameters. We used magnetohydrodynamics for gas disks in combination with live dark-matter halos in adaptive mesh refinement simulations. The disks were initialized using a setup for isolated disks in hydrostatic equilibrium. Small-scale filtering of the velocity and magnetic field allowed us to estimate the turbulent electromotive force (EMF) and kinetic helicity. Time series of the average magnetic field in central and outer disk regions show pronounced peaks during close encounters and mergers. This agrees with observed magnetic fields at different interaction stages. The central field strength exceeds 10 microgauss (corresponding to an amplification factor of 2 to 3) for small impact parameters. As the disks are increasingly disrupted and turbulence is produced by tidal forces, the small-scale EMF reaches a significant fraction of the total EMF. The small-scale kinetic helicity is initially antisymmetric across the disk plane. Though its evolution is sensitive to both the impact parameter and inclinations of the rotation axes with respect to the relative motion of the disks, antisymmetry is generally broken through interactions and the merger remnant loses most of the initial helicity. The EMF and the magnetic field also decay rapidly after coalescence. The strong amplification during close encounters of the interacting galaxies is mostly driven by helical flows and a mean-field dynamo. The small-scale dynamo contributes significantly in post-interaction phases. However, the amplification of the magnetic field cannot be sustained.

astro-ph.GA↗

arXiv2022

Proceedings of the International Astronomical Union: Studying Magnetic Field Amplification in Interacting Galaxies Using Numerical Simulations

There are indications that the magnetic field evolution in galaxies might be massively shaped by tidal interactions and mergers between galaxies. The details of the connection between the evolution of magnetic fields and that of their host galaxies is still a field of research. We use a combined approach of magnetohydrodynamics for the baryons and an N-body scheme for the dark matter to investigate magnetic field amplification and evolution in interacting galaxies. We find that, for two colliding equal-mass galaxies and for varying initial relative spatial orientations, magnetic fields are amplified during interactions, yet cannot be sustained. Furthermore, we find clues for an active mean-field dynamo.

astro-ph.GA↗

arXiv2020

Hyperparameter Transfer Across Developer Adjustments

After developer adjustments to a machine learning (ML) algorithm, how can the results of an old hyperparameter optimization (HPO) automatically be used to speedup a new HPO? This question poses a challenging problem, as developer adjustments can change which hyperparameter settings perform well, or even the hyperparameter search space itself. While many approaches exist that leverage knowledge obtained on previous tasks, so far, knowledge from previous development steps remains entirely untapped. In this work, we remedy this situation and propose a new research framework: hyperparameter transfer across adjustments (HT-AA). To lay a solid foundation for this research framework, we provide four simple HT-AA baseline algorithms and eight benchmarks changing various aspects of ML algorithms, their hyperparameter search spaces, and the neural architectures used. The best baseline, on average and depending on the budgets for the old and new HPO, reaches a given performance 1.2--2.6x faster than a prominent HPO algorithm without transfer. As HPO is a crucial step in ML development but requires extensive computational resources, this speedup would lead to faster development cycles, lower costs, and reduced environmental impacts. To make these benefits available to ML developers off-the-shelf and to facilitate future research on HT-AA, we provide python packages for our baselines and benchmarks.

cs.LG↗