Newsroom

World’s largest simulation of magnetized turbulence reveals major deviations from canonical models.

Groundbreaking simulations led by an international team of researchers, including STRUC­TURES member Ralf Klessen (Centre for Astronomy Hei­del­berg, ZAH, and Interdisciplinary Center for Scientific Computing, IWR) have unveiled previously unknown details about turbulence in the magnetized environment of the interstellar medium (ISM). This complex, turbulent plasma, composed of hot, electrically charged gas and dust between the stars in our galaxy, profoundly influences fundamental cosmic processes such as star formation and cosmic-ray transport. Unlike everyday turbulence, the interstellar medium's plasma is magnetized, making it particularly difficult to observe and model.

The new study, published in Nature Astronomy, utilized simulations with unprecedented grid resolutions to model highly compressible, magnetized turbulence akin to that occurring in the ISM.  “This is the first time we can study these phenomena at this level of precision and at these different scales,” says James R. Beattie, the first author of the study. A key finding is the discovery of two distinct, coexisting kinetic energy cascades within the turbulence – a process where kinetic energy flows from large to small scales within the turbulent ISM. The interstellar medium can be envisioned as a fluid with swirling motions of various sizes, powered by energy often injected on large scales (e.g., from exploding stars). According to the new simulations, this energy transfer happens in two separate ways depending on the scale of these motions. On the largest scales, where gas is moving faster than sound (supersonic) and magnetic fields are relatively weak, the energy transfer is more direct and can “jump” across different sizes of swirls. However, on smaller scales, where gas moves slower than sound (subsonic) and is strongly influenced by magnetic fields woven in the ISM, energy flows more gradually down through successive scales. This two-level process of energy dissipation is a significant departure from simpler, older models of turbulence. The team also discovered that energy stored in the magnetic energy itself has its own unique way of cascading through these turbulent regions, following a pattern that current theories do not fully explain.

These findings challenge fundamental theories of magnetohydrodynamic (MHD) turbulence and provide crucial insights into the properties of the ISM. In particular, they help to explain how turbulence affects the initial conditions for star formation. Whereas the coupling of turbulence and magnetic fields enhances filamentary structures along which stars preferentially form, the strong magnetic influence on smaller scales acts like a form of pressure that can help support gas clouds against gravitational collapse – suppressing the rate of star formation at the same time. The simulations used over 80 million CPU hours on nearly 140,000 cores of the SuperMUC-NG supercomputer at the Leibniz Supercomputing Centre.

Further information:

• Original Publication: Beattie, J.R., Federrath, C., Klessen, R.S. et al. The spectrum of magnetized turbulence in the interstellar medium. Nat Astron, 2025 (Link to Nature Astronomy)
• Re­search homepage of Prof. Ralf Klessen

We are delighted to announce the next event in our Ma­chine Learning Galore! series, focusing on Scientific Ma­chine Learning, which will take place on Thursday, April 24, from 4:30 to 6:00 pm at INF 205 Mathematikon (5th floor). The event will feature lab presentations by principal investigators, followed by brief presentations from junior scientists showcasing their latest work. Extended discussions will offer ample opportunity for in-depth exchanges.

In order to participate, please register for free at the ML-AI portal by April 22.

Event Details:

• Lab presentations:
  • Stephanie Hansmann-Menzemer
  • Jan Stühmer
  • Frank Zöllner
• Science Talks:
  • Christoph Langenbruch (Hansmann-Menzemer lab): Flavourful Ma­chine Learning at LHCb
  • Leif Seute (Stühmer lab) Generative Ma­chine Learning for the Design of Dynamical Proteins
  • Anika Strittmatter (Zöllner lab) Multimodal Medical Image Registration Using Deep Learning 

About Scientific Ma­chine Learning
Scientific Ma­chine Learning is a collaborative initiative by the Interdisciplinary Center for Scientific Computing (IWR) and the STRUC­TURES Cluster of Excellence. Its mission is to foster interaction and exchange within the local ma­chine learning community, and to support its development by consolidating activities and resources that might otherwise remain scattered across individual institutions or disciplines. The initiative aligns closely with the objectives of STRUCTURES, which aims to advance fundamental research, and with IWR’s focus on applying ma­chine learning to address long-standing challenges in the natural and life sciences, engineering, and the humanities.

Further information:

• Scientific Ma­chine Learning Hei­del­berg
• Ma­chine Learning Talks on Campus – Information service and mailing list
• Interdisciplinary Center for Scientific Computing (IWR)

We are pleased to announce the workshop Geometry, Topology and Ma­chine Learning (GTML), jointly organized by the Max Planck Institute for Mathematics in the Sciences and the STRUC­TURES Cluster of Excellence. The event, taking place from November 10 to 14, 2025, will bring together two rapidly evolving fields central to modern ma­chine learning. Geometry and topology offer fundamental methods for understanding data structure and provide powerful frameworks for analyzing, unifying, and generalizing ma­chine learning techniques across diverse applications.

The workshop will feature 10 keynote talks and 20 presentations by leading experts. Topics range from mathematical foundations of ma­chine learning to geometric & topological deep learning, as well as multidisciplinary applications of geometry & topology. By merging the previous Workshop on Geometry in Ma­chine Learning (GaML) and the Workshop on Topological Methods in Data Analysis (TMDA), GTML creates a platform to foster collaboration and to explore the synergistic relationship between geometry, topology, and ma­chine learning.

For further details, including an overview of talks and speakers, and to register, please visit:
https://www.mis.mpg.de/events/series/workshop-on-geometry-topology-and-machine-learning-gtml-2025

Applications for lightning talks can be submitted via the registration form by May 31, 2025.

Further information:

• Conference homepage with more details & registration
• Max Planck Institute for Mathematics in the Sciences, Leipzig

Hei­del­berg Uni­ver­si­ty is set to pioneer a transformative approach to mathematics education with the launch of its new master’s programme, “Mathematics of Ma­chine Learning and Data Science.” Starting in the winter term 2025/2026, the new research-oriented course aims to lay the mathematical and methodological groundwork that will enable future generations of mathematicians to advance the frontiers of ma­chine learning and scientific data analysis. The programme, which is taught in English, is based in the Faculty of Mathematics and Computer Science and coordinated by the Institute for Mathematics in cooperation with the Interdisciplinary Center for Scientific Computing. Prospective students are encouraged to apply by the deadline of 15 May 2025.

“Ma­chine learning and data science are currently revolutionizing the sciences. This equally concerns basic re­search to better understand established methods, such as the learning-based analysis of data, as well as projects to apply these methods in by now almost all branches of science,” underlines Prof. Dr Christoph Schnörr from the Institute for Mathematics of Hei­del­berg Uni­ver­si­ty. This is where the new four-semester master's programme starts: it imparts an advanced understanding of how pure and applied mathematics intersect to innovatively expand on the methodology of ma­chine learning and scientific data analysis. By integrating core areas such as topology, differential geometry, dynamic systems, statistics, optimization, numerics, and functional analysis, the curriculum creates a robust framework for developing novel methodologies. “Compared to the regular mathematics master’s, this programme is extremely interdisciplinary in approach. We see this, for example, in the lecture series for the first semester, which gives a broad overview of core fields of mathematics,” says Prof. Schnörr.

Students will gain practical experience by working on the implementation and application of theo­re­ti­cal models in a data science lab, and acquire core competences analytical & structural thinking, scientific problem-solving and interdisciplinary collaboration. The option to spend a semester abroad offers valuable opportunities to engage in international teamwork and collaborate with leading researchers worldwide. Additionally, specialization modules extending over the two years will prepare the students for their master’s thesis and equip the graduates for doctoral positions in Germany and abroad, as well as for research-oriented work in industry.

Further information:

• Hei­del­berg Uni­ver­si­ty Degree Programme: Mathematics of Ma­chine Learning and Data Science
• Faculty of Mathematics and Computer Science
• Interdisciplinary Center for Scientific Computing

Most important re­search advancement prize honors the experimental work on integrated photonics by Prof. Wolfram Pernice and his team.

We are proud to announce that our member Wolfram Pernice has been awarded the prestigious Gottfried Wilhelm Leibniz Prize of the German Re­search Foundation (DFG). The award honours his groundbreaking pioneering work on neuromorphic photonic computing, a transformative field at the intersection of physics, computer science, and engineering.

Prof Wolfram Pernice heads the re­search group Neuromorphic Quan­tum Photonics at Kirchhoff Institute for Physics and is part of STRUCTURES' Comprehensive Project CP 5: Quan­tum Systems and Neural Networks: Computation in Physical Structures. The goal of his re­search in the field of integrated photonics is to develop new methods for information processing and rapid computation using light. By developing nanoscale chip systems, his re­search has far-reaching implications for artificial intelligence and quan­tum technologies. The DFG underlines that his interdisciplinary re­search crosses traditional boundaries; it impacts on various disciplines – from natural sciences to computer science to engineering sciences. “His re­search results point the way to innovative, sustainable methods for reducing energy consumption of AI computer hardware and still enabling rapid calculations. Furthermore, he is known worldwide as a pioneer in the field of integrated quan­tum photonics,” the German Re­search Foundation adds.

The Gottfried Wilhelm Leibniz Prize – the most important re­search award in Germany – has been awarded annually by the German Re­search Foundation since 1986. Up to ten prizes can be awarded each year with prize money of 2.5 million euros each. The awards for 2025 go to four female and six male researchers, including Wolfram Pernice. An award also goes to mathematician Prof. Angkana Rüland, a former member of STRUC­TURES who did re­search on applied mathematics at Hei­del­berg Uni­ver­si­ty from 2020 to 2023. The purpose of the Leibniz Programme, established in 1985, is to honor outstanding scientists, to expand their re­search opportunities and facilitate employment of particularly qualified early-career researchers. The award ceremony takes place on 19 March 2025 in Berlin. 

Further information:

• Wolfram Pernice's re­search group
• Press release by the German Re­search Foundation

For his re­search project “Cosmological Geometry and Gravity with Non-linear Physics (GeoGrav),” the German Re­search Foundation (DFG) has granted Prof. Luca Amendola, professor of physics at the Institute for Theo­re­ti­cal Physics (ITP) in Hei­del­berg, 340,000 euros for three years. GeoGrav will investigate fundamental questions about the universe’s geometry and gravity.

Cosmology has seen a tremendous development in recent years, driven by high-quality data from the cosmic microwave background, large-scale structure, and distance indicators. Nevertheless, several fundamental questions about our Universe are still open: What are the properties of dark energy? Is the Universe spatially flat? Is gravity Einsteinian at all scales and epochs? Often these questions are addressed within restricted classes of models, e.g. simple extensions of ΛCDM with inflationary initial conditions. In this case, the results will unavoidably depend on the model assumption. In a recent series of papers, Luca Amendola and colleagues have shown how to reach accurate and precise cosmological conclusions regardless of assumptions about the evolution of the background, the initial conditions, the linear perturbations, and the bias functions. To achieve this, they combined non-linear correlators with distance indicators (supernovae Ia or standard sirens) as additional tracers of large-scale structure. Their new project aims at the next logical step: applying this methodology to the real data that will soon be provided by Euclid and other surveys. The ultimate goal of GeoGrav is to measure geometry and gravity at cosmological scales in a way that is both precise (high statistical significance) and accurate (weakly dependent on cosmological assumptions).

Luca Amendola is professor of physics at the Institute for Theo­re­ti­cal Physics in Hei­del­berg, Germany. His area of re­search is Cosmology and Astrophysics, with a particular focus on topics related to Dark Energy, Large Scale Structure, Cosmic Microwave Background and Statistics. He is a member of the Euclid collaboration and joined the STRUC­TURES Cluster of Excellence in 2024.

Further information:

• Luca Amendola's Webpage
• Institute for Theo­re­ti­cal Physics (ITP)