Sarah Jaffa
Simulations are a powerful tool to test our theoretical understanding of astrophysical processes. Star formation involves the interplay of many complex processes such including gravity, turbulence, chemistry, magnetic fields and stellar feedback, and the relative importance of each of these in the star formation process is currently unclear. Simulations are often pushed to their computational limit to include as many as possible, leading to very long run times. Consequently, many published works are based on only a single simulation or a handful of repeated setups with varied parameters.
This leads us to a crisis in theoretical star formation, as these chaotic systems are highly dependent on the exact details of initial conditions. The turbulent field used to set the velocity of the gas is randomly generated, and while many works have examined the effects of varying the nature of the turbulent field (strength, power spectrum, solenoidal or compressive, etc.) there is not yet a comprehensive study of how the randomly generated velocities affect the outcome of the simulations just for different random states with the same parameters.
We aim to fill this gap by performing a suite of cloud simulations with identical setups except for the random seed used to generate the turbulent field. We examine how this one change affects commonly used metrics of star formation, such as the SFE, SFR and IMF. This will provide a measure of the uncertainty in any of these quantities when calculated from only a single simulation and allow us to understand whether the measured variation in larger simulations is due to the parameters under investigation or just the random variation of initial conditions.
A poster by Sarah Jaffa and Jim Dale of the University of Hertfordshire, UK.
Star formation involves several non-linear processes acting over a large range of size, density and energy scales. This makes it formally chaotic - a small change in initial conditions can have a large effect on the outcome.
Simulators often try to include as much physics as possible and maximise resolution, leading to long computational time. Most simulation studies perform only one or a handful of realisations, particularly above the scale of individual star-forming cores.
A graph of the size against mass of several studies from the literature shows the smaller scales have been reasonably well explored with some authors doing 10 or 20 repeat realisations, but systems more massive than 100 solar masses have not been repeated by many authors. This work sits at a medium scale and has more repeats than otehrs of this size.
Methods: Turbulent velocities are randomly generated
Most simulations include at least self-gravity and turbulence. The turbulent velocity field is randomly generated, and its exact morphology can affect where dense gas collects and therefore how and where star formation proceeds.
I have performed 30 SPH simulations of one physical setup, changing only the random seed used to generate the turbulent velocity field. This changes the detail of the turbulent velocity field but the power spectrum, solenoidal or compressive fraction, Mach number and everything else is kept the same.
Images of 4 of my simulations at 5 megayears show a variety of structures, with different amounts of star formation. Some have one central cluster of more than 100 sinks, while others have more diffuse structures with half as many sinks.
Preliminary results: IMF and SFE show significant variation
Bertelli Motta et al. 2016 (BM16) performed 5 simulations, changing the Mach number of the turbulence. Their Initial Mass Function shifted to higher masses, interpreted as the Mach number causing larger stars to form.
My simulations show a variation of the same order without changing the Mach number, just due to the change in random number seed.
My simulations have much more massive stars than BM16, but it is well known that the density, resolution and choice of physics can shift the IMF. I plan to repeat this experiment with different setups to investigate the parameter space.
The star formation efficiency and number of sinks produced also varied greatly in my simulations.
Geen et al. (2018) performed 13 simulations using an AMR code, and found a larger variation in the time of the onset of star formation and smaller variation in the star formation efficiency compared to my work. I plan to test different numberical methods to provide a benchmark.
Conclusions: Randomly generated turbulence affects simulation in unexplored ways
The long runtime for state-of-the-art simulations means repeated experiments are rare. Therefore, we have not thoroughly investigated how randomly generated turbulent velocity fields affect the results. My preliminary work suggests the variation in the mass function and star formation efficiency can be of the same order as results attributed to changing physical setups.
Future plans:
1. Investigate different metrics used in literature such as the Mach number, virial ratio, morphology of gas and clusters, dense gas fraction, etc.
2. Change initial conditions such as the mass, density profile, resolution and sink density.
3. Simulate additional physics such as radiative and mechanical feedback or stellar evolution.
4. Create identical setups using different codes including SPH and AMR.
If you think this work is interesting, important or wrong, please contact me to collaborate at s.jaffa@herts.ac.uk, or hire me in August 2021.