Advanced PySPAM: An Infrastructure to Constrain Underlying Interacting Galaxy Parameters Synthetic Results

<p>This database contains the results for the Chapter 3 of DOR's thesis. For a full description of these results and the way they were built, please see&nbsp;<em>Link to be added on publication</em>.</p> <p>The aim of this Chapter was to use MCMC methods with a fast, efficient simulation algorithm (APySPAM) to constrain the underyling parameters of observed interacting galaxy systems. This algorithm used a Chi-Squared distance minimisation between morphology distributions of observed and simulated images to constrain 13 underlying parameters of galaxy interaction. We applied our algorithm to to 50 of the 62 systems described in <a href="https://ui.adsabs.harvard.edu/abs/2016MNRAS.459..720H/abstract">Holincheck et al. (2016).</a></p> <p>We opted to use the Holincheck et al. sample as the underlying parameters of these systems had already been constrained using a Citizen Science project named <a href="https://mergers.galaxyzoo.org/">Galaxy Zoo: Mergers</a>. This gave us a ground truth to which compare our constraints to. We created synthetic observations of each image, and then ran our MCMC over them, achieving constraint across the sample and parameter space. However, when applied to observational data (we opted to use SDSS images of these systems) we are unable to constrain the full parameter space. This is particularily true of the orientations of the interacting system and their relative sizes.</p> <p>Exploring using velocity information in our constraints find that we improve almost all our constrains considerably. Therefore, adding in spectroscopic information to this method could drastically improve it. The main limitation of this approach, however, is computation time with each system taking approximately 20 hours on a well parallelised HPC to converge. Alternatives to improve performance lie in simulation based inference (SBI, an introduction can be found <a href="https://arxiv.org/pdf/2009.08459">here</a>) or including the use of GPUs (such as done by NVIDEA in fluid dynamics <a href="https://developer.nvidia.com/blog/ai-powered-simulation-tools-for-surrogate-modeling-engineering-workflows-with-siml-ai-and-nvidia-modulus/">here</a>)</p> <p>The results are portrayed as corner plots, with contour plots showing the distribution of likelihoods found in each MCMC run and the histograms on the side showing the marginalised posterior distributions.</p>