Evaluating Emulation Strategies in Heavy-Ion Collisions: Gaussian Processes vs. Taylor Expansions

The application of model emulators is making it possible for Bayesian analyses to generate rigorous quantitative conclusions from model-data comparisons in nuclear physics. Emulator strategies allow one to search high-dimension parameter space with a minimal number of full-model runs. Given the numerical cost of running at a single point in model-parameter space, more efficient emulator forms or more optimized choices of design points can save significant CPU time, storage and cost. I will compare two different schemes for emulation, Gaussian processes and Taylor expansions, and discuss under which circumstances once scheme might be more appropriate. As an example, both approaches will be applied to simulations of jet quenching for heavy-ion data from RHIC and the LHC.