2022 Abstract from Simone

Mass Observable Relationship for Machine Learning

Student: Simone Shevchuk

Scientist Mentor: Yuanyuan Zhang AstroPhysics/Galaxy Clusters

Dark matter and energy constitute 95% of our universe, yet little is known about it. Fortunately, galaxy cluster or halo simulations can constrain cosmological parameters and cosmology, allowing us to gain a more accurate understanding of the properties of dark matter and energy. Masses from the Latin Hypercube Quijote simulation are used to calculate a richness value and the number of clusters within various richness ranges—“bins”—are used to train an ML algorithm. However, the LH Quijote simulation does not have halos below 10¹³ solar masses, and this leaves a gap in the ML. I began by investigating the mass observable relationship of halos in the Quijote simulation to simplify data interpretation and develop my python toolset. Unexpectedly, there was no correlation between mass and velocity. After performing sanity checks by plotting the Number of CDM particles vs. mass and mean velocities, a possible bug in the simulation was found that warrants additional investigation. To add masses below 10¹³, I first used the NumPy random number generator but it failed to give the cosmologically accurate random distribution. Therefore, I used Hanzhi Tan’s code to generate the correct distribution of samples but was still inputting a sample number that was disproportional to the Quijote data. I found a ratio to calculate the correct sample number and input it to generate masses, resulting in a difference of 0%-15% in richness counts across all bins. It is not immediately clear whether this affects the ML as the uncertainty range in the richness accounts requires consideration, but it is possible.