HITS Lab Reports

Making synergies work for astronomy – “Emulation in Simulation” in the HITS Lab

Kiril Maltsev, HITS Alumnus

The first projects started in 2020, with the aim to foster collaboration across groups and disciplines and to serve as incubator for joint research within HITS and beyond. Now, five years on, we zoom in on a series of outcomes of one of the first HITS Lab projects “Emulation in Simulation”.

For the research works below, we talked to Kiril Maltsev, former PHD student in the project and HITS alumnus of the Physics of Stellar Objects group (PSO). He and his colleagues combined the institute’s expertise in computational statistics and numerical simulation to develop a new generation of surrogate modeling tools.

Emulation, or surrogate modeling, is a technique to approximate the behavior of a complex computer code by a computationally less expensive model that is typically constructed using supervised machine learning. The three participating groups in  “Emulation in Simulation” at HITS were “Physics of Stellar Objects” (PSO), “Molecular Biomechanics” (MBM) and “Computational Statistics” (CST).           


Sub-Project 1:
Catch a failing star – an optimistic model for neutrino-driven supernovae


Massive stars undergoing iron core-collapse at the end of their evolution terminate their lives either in a successful or a failed supernova (SN) explosion. But whereas explosions of Type Ia SN or nuclear bombs are always thermonuclear, these core-collapse explosions are neutrino-driven. If they fail, they result in black hole formation, in case they are successful, they leave behind either a neutron star or a fallback black hole.

A recent study now introduces a new statistical model for predicting the outcomes of core-collapse supernovae (CCSNe). The model is in good agreement with detailed 3D simulations of CCSNe, though it is more optimistic about successful explosions than previous statistical models and seems to be in better agreement with SN observations. At the same time it partially addresses the missing red supergiant problem by black hole formation.        

“Our SN model is the first to distinguish between envelope-retaining and envelope-stripped stars that can evolve into very different pre-SN core structures, and connects to a recently established theory explaining properties of pre-SN structure by the core evolution through the late burning phases,” says Kiril Maltsev, first author of the paper. “When used in concert with population synthesis codes, the model allows us to re-estimate the rates of binary black hole mergers observable by gravitational wave astronomy as well as the rates of Type II, Type Ib and Type Ic SNe in our galaxy and beyond.”

To set up the model, the researchers started by formulating explodability criteria that allow them to predict the final fate of massive stars based on stellar structure parameters at the onset of iron-core infall and thus found a way to relate these to the carbon-oxygen core mass and metallicity of the SN progenitors.

Sub-Project 2:
Not only the weather – efficient stellar evolution forecasting        


Another research outcome of “Emulation in Simulation” is a surrogate model of stellar evolution, its unique aspect being the width of the stellar parameter space over which it casts predictions. A feedforward neural network traces the evolution of stellar observables up to the end of core-helium burning, covering more than 99% of stellar lifetimes. Moreover, it predicts these over a zero-age-main-sequence (ZAMS) birth mass range from the cool and faint sub-solar mass red dwarves up to the heaviest 300 solar mass stars that later become the hot, bright and violent Wolf-Rayet stars.

The model allows for the estimation of a star’s age and its ZAMS mass (given the observations of a star’s color, brightness and surface gravity). This is an optimization problem that typically requires hundreds or thousands of evaluations in the stellar parameter space for convergence to the best-fit parameters. Similarly, the surrogate model can be used to estimate the age of a stellar population, such as an open cluster.          



Sub-Project 3:
Invisible Ripples in space – gravitational waves from hypermassive neutron stars


The third outcome involving emulation techniques is a model for gravitational waves from binary neutron star merger remnants. As two neutron stars (NSs) merge, they emit gravitational waves (GWs). In future, interferometry data recorded by the LIGO collaboration for real-time GW signal searches on Earth can be scanned with the help of an efficient, accurate and noise-robust template model. “In order to detect GWs on Earth, one needs to know what to search for, i.e. how the GW signals look like as a function of astrophysical source parameters, in order to dig them out of a noise background,” says Theodoros Soultanis, former PhD student at HITS and first-author of the paper in which the model was first presented.

The merger of two NSs results either in a prompt collapse to a black hole or in a quasi-stable hypermassive neutron star (HMNS). The HMNS then either stabilizes or undergoes a delayed collapse to a black hole. The interest in detecting GW signals from a HMNS is that these do not only depend on the binary NS progenitor masses but also on the high-density Equation of State (EoS) governing the behavior of matter in the ultra-dense core of the HMNS. Therefore, detection of such a signal has the potential to constrain theoretical models of the high-density EoS and thus inform fundamental physics. GW signals from the binary NS post-merger phase have a high oscillation frequency (hundreds of kilohertz) and the necessary sensitivity to detect these by interferometry will be reached prospectively in a few years’ time.

“The emulation technique has proven to be highly versatile, as the applications to stellar evolution, core-collapse supernova and gravitational wave modeling in context of the HITS Lab project show. These efforts have already inspired a range of follow-up studies, and that momentum is expected to continue well into the future,” concludes Kiril Maltsev.

Switch to the German homepage or stay on this page