TGSF-DA: DATA ANALYSIS OF THE TURBULENCE AND GRAVITY ROLES IN STELLAR FORMATION

Partnering Team 1 (Project Lead) in TGSF-DA: 
Rubén Cabezón PI (University of Basel) 
Florina Ciorba co-PI (University of Basel)
Osman Seckin Simsek (University of Basel)

Partnering Team 2:
Lucio Mayer (University of Zurich)

Partnering Team 3:
Sebastian Keller (ETH Zurich/CSCS)
Jean M. Favre (ETH Zurich/CSCS)

Partnering Team 4:
Ralf Klessen (University of Heidelberg)
Oliver Avril (University of Heidelberg)

SDSC Collaborators:
Mathieu Sazmann (SDSC)
Georgios Kissas (SDSC)
Quentin Duchemin (SDSC)
Geoffrey Chinot (SDSC)

Funding agency: The Swiss Data Science Center (https://datascience.ch/)

Duration: 23.06.2025-22.06.2027

Software: The SPH-EXA simulation framework is publicly available here.

Project Summary

How do turbulence and gravity influence the formation of stars? The TGSF-DA project addresses this fundamental question by analysing large-scale astrophysical simulations generated with the SPH-EXA framework, which models interstellar turbulence and self-gravity at unprecedented scale and resolution. The project combines astrophysics, data science, and high-performance computing to extract new scientific insights from the vast amounts of data produced by these simulations.

A major focus of the project is understanding how pre-stellar cores form and evolve, and how their mass distribution relates to the properties of the surrounding turbulent flow. By analysing core formation events and mass accretion histories, TGSF-DA aims to characterize the stellar initial mass function and investigate how it depends on the physical conditions of the interstellar environment. The project also studies turbulent mixing and the distribution of chemical elements, seeking to understand how turbulence contributes to the apparent chemical homogeneity observed in stellar clusters.

The simulations provide a unique opportunity to investigate the chaotic nature of turbulence. Taking advantage of the Lagrangian nature of SPH, TGSF-DA will develop methods to track particle trajectories and measure Lyapunov exponents, providing quantitative information about how initially close particles diverge within turbulent flows. At the same time, the project analyses the computational performance of these extreme-scale simulations, investigating load imbalance, computational bottlenecks, and energy consumption to understand how the simulations behave on modern supercomputers and how their efficiency can be improved.

The TGSF-DA project forms part of the broader Swiss contribution to the Square Kilometre Array (SKA) through SKACH which support the development of extreme-scale simulation capabilities for astrophysics and cosmology. Within SKACH, the SPH-EXA framework is being extended toward cosmological applications and trillion-particle simulations on next-generation Tier-0 supercomputers. The simulations of turbulence, gravity, and stellar formation provide an important test case for these capabilities, pushing SPH-EXA to unprecedented scales while addressing fundamental questions about how stars form. TGSF-DA extends this effort by bringing in the data-science expertise of the Swiss Data Science Center (SDSC) to analyse the enormous datasets generated by these simulations, using machine learning, scalable HPC workflows, and scientific visualization. In this way, the two efforts are closely connected: SKACH provides the scientific and computational framework for producing extreme-scale astrophysical simulations, while TGSF-DA develops the data-analysis approaches needed to extract scientific and computational insights from them.

The scale of the data presents a challenge in itself. The TGSF simulations generate physics datasets ranging from terabytes to tens of terabytes per checkpoint, together with performance data describing the execution of the simulations. TGSF-DA therefore develops scalable approaches combining statistical analysis, machine learning, HPC methods, and scientific visualization to efficiently explore and interpret these datasets. Visualization and in-situ analysis will help identify and communicate important physical and computational phenomena without requiring the entire dataset to be processed conventionally.

The project brings together researchers from astrophysics, data science, high-performance computing, software engineering, and visualization at the University of Basel, University of Zürich, CSCS, and University of Heidelberg, with support from the Swiss Data Science Center. Together, the collaboration aims to advance our understanding of stellar formation and turbulence while developing data-analysis and computational techniques that can benefit other areas of astrophysics, cosmology, computational fluid dynamics, and large-scale scientific computing.

TGSF-DA builds on simulations performed on Europe’s leading supercomputing infrastructures, including LUMI and Alps, and embraces open science and reproducibility. The resulting tools, analyses, and insights are intended to be shared with the scientific community, creating new opportunities for research and enabling future generations of extreme-scale simulations to be analysed more effectively.

Publications

Coming soon