POSTER P501: Simulating stellar color-magnitude diagrams on the GPU

ADASS posters are displayed all week

When

10:01 p.m., Nov. 8, 2023

Theme: GPU implementations for core astronomical libraries

pretalxeposter

Simulating stellar color-magnitude diagrams on the GPU

L. Dobos, C. Filion, T. Budavári, R. Wyse, A.S. Szalay
Dept. of Physics & Astronomy, The Johns Hopkins University


Generating synthetic photometric catalogs of fields containing a large number of stars belonging to multiple stellar populations, such as the composite stellar populations of nearby satellite galaxies with resolved stars and the various sub-populations of the Galactic foreground, is computationally intensive. Stellar populations are usually characterized by their metallicity, age and distance distributions, which are often correlated, whereas the initial mass of each star is thought to come from a universal Initial Mass Function (IMF). Precomputed isochrone grids for a large number of magnitude systems are available to convert these fundamental stellar parameters to magnitudes. Interpolation between isochrones, however, must be done in the parameter called the Equivalent Evolutionary Phase (EEP), instead of initial mass, since stars with the same initial mass but slightly different age (or metallicity) can be in entirely different evolutionary phases. Hence, isochrone grids are tabulated for metallicity, age and EEP and interpolating the magnitudes from metallicity, age and initial mass requires an implicit interpolation scheme where one has to find the EEP as well, as part of the solution. On the other hand, synthetic catalogs are often generated with given magnitude and color cuts which must be observed during random sampling. Due to the steep IMF, the large number of low-mass, faint stars are likely to cause high random sampling rejection rates. While intricate heuristics can help mitigate the latter problem, our GPU-based solution to synthetic catalog generation is fast enough to generate hundreds of thousands of stars on the time scale of tens of seconds with broadly available hardware. Our software library, built with eager mode TensorFlow, solves the implicit isochrone interpolation problem, as well as provides a relatively simple, yet very flexible way of describing mixtures composite stellar populations for detailed modelling.

Contacts

Laszlo Dobos, Johns Hopkins University