Abstract:Bayesian parameter estimation provides a systematic approach to compare heavy ion collision models with measurements, leading to constraints on the properties of nuclear matter with proper accounting of experimental and theoretical uncertainties. Aside from statistical and systematic model uncertainties, interpolation uncertainties can also play a role in Bayesian inference, if the model's predictions can only be calculated at a limited set of model parameters. This uncertainty originates from using an emulator to interpolate the model's prediction across a continuous space of parameters. In this work, we study the trade-offs between the emulator (interpolation) and statistical uncertainties. We perform the analysis using spatial eccentricities from the T$_\mathrm{R}$ENTo model of initial conditions for nuclear collisions. Given a fixed computational budget, we study the optimal compromise between the number of parameter samples and the number of collisions simulated per parameter sample. For the observables and parameters used in the present study, we find that the best constraints are achieved when the number of parameter samples is slightly smaller than the number of collisions simulated per parameter sample.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to optimize the computational budget allocation when performing Bayesian parameter estimation in heavy - ion collisions. Specifically, the authors studied how to find the best balance between the number of parameter samples and the number of collision simulations for each parameter sample under a given computational budget, in order to minimize the effects of interpolation (emuator) uncertainty and statistical uncertainty. By using the initial - condition spatial eccentricity generated by the T RENTo model as an observable, the authors explored the trade - off between the number of different design points (i.e., parameter samples) and the number of simulated events per design point to obtain the best parameter constraints. ### Main problems 1. **Trade - off between interpolation uncertainty and statistical uncertainty**: When model predictions can only be calculated on a limited set of parameters, interpolation uncertainty will appear in Bayesian inference. This uncertainty stems from using an interpolator (such as a Gaussian process) to interpolate model predictions across a continuous parameter space. The authors studied how to find the optimal balance between interpolation uncertainty and statistical uncertainty under a fixed computational budget. 2. **Optimal number of parameter samples**: The authors evaluated the impact of different numbers of parameter samples and the number of simulated events per parameter sample on the accuracy of parameter estimation through closure tests. A closure test is a self - consistency check method that performs Bayesian parameter estimation by replacing experimental data with model calculation results to verify the accuracy of the model. ### Research methods - **T RENTo model**: Used to generate the initial conditions of heavy - ion collisions, especially spatial eccentricities. - **Gaussian process interpolator**: Used to interpolate the output of the T RENTo model and provide interpolation uncertainty estimates. - **Bayesian parameter estimation**: By defining the likelihood function and prior distribution, calculate the posterior distribution, thereby estimating the model parameters. - **Closure test**: By using the model calculation results as "experimental data" for Bayesian parameter estimation, evaluate the accuracy and precision of parameter estimation. ### Main findings - **Optimal number of design points**: Under a fixed computational budget, when the number of parameter samples is slightly less than the number of simulated events per parameter sample, the parameter estimation has the best constraint effect. Specifically, the optimal number of parameter samples is close to the square root of the total number of events, that is, $ \frac{N_d}{M_{\text{ev}}} \approx 0.1 - 1 $. - **Robustness**: Even if the interpolation uncertainty and statistical uncertainty are large, Bayesian parameter estimation can still provide constraints consistent with the true parameter values. This indicates that the Gaussian process interpolator is robust in Bayesian inference in heavy - ion collisions. ### Conclusion This study provides a strategy for optimizing computational budget allocation in Bayesian parameter estimation in heavy - ion collisions. By rationally choosing the number of parameter samples and the number of simulated events per parameter sample, the accuracy of parameter estimation can be significantly improved at the same computational cost. This finding is of great significance for improving the efficiency and accuracy of heavy - ion collision physics research.

Computational budget optimization for Bayesian parameter estimation in heavy ion collisions

Parameter optimisation using Bayesian inference for spallation models

A Bayesian approach for parameter estimation and prediction using a computationally intensive model

Enhancing Bayesian parameter estimation by adapting to multiple energy scales in RHIC and LHC heavy-ion collisions

Bayesian parameter estimation for relativistic heavy-ion collisions

Bayesian parameter estimation with a new three-dimensional initial-conditions model for ultrarelativistic heavy-ion collisions

Bayesian analysis of (3+1)D relativistic nuclear dynamics with the RHIC beam energy scan data

Examination of nucleon distribution with Bayesian imaging for isobar collisions

Applications of emulation and Bayesian methods in heavy-ion physics

Toward Accelerated Nuclear-physics Parameter Estimation from Binary Neutron Star Mergers: Emulators for the Tolman–Oppenheimer–Volkoff Equations

Exploring parameter spaces with artificial intelligence and machine learning black-box optimization algorithms

Towards accelerated nuclear-physics parameter estimation from binary neutron star mergers: Emulators for the Tolman-Oppenheimer-Volkoff equations

On model emulation and closure tests for 3+1D relativistic heavy-ion collisions

Bayesian quantification of strongly-interacting matter with color glass condensate initial conditions

Bayesian quantification of strongly interacting matter with color glass condensate initial conditions

Bayesian reconstruction of impact parameter distributions from two observables for intermediate energy heavy ion collisions

3D Multi-system Bayesian Calibration with Energy Conservation to Study Rapidity-dependent Dynamics of Nuclear Collisions

Bayesian optimisation approach to quantify the effect of input parameter uncertainty on predictions of numerical physics simulations

Interpolating between small- and large-$g$ expansions using Bayesian Model Mixing

Bayesian approach to determine the initial condition for the Balitsky-Kovchegov equation