Accelerating Reionization Constraints: An ANN-Emulator Framework for the SCRIPT Semi-numerical Model

TL;DR

Constraining the Epoch of Reionization with high-fidelity simulations is computationally expensive, limiting exploration of multi-parameter models. The paper presents an ANN-based emulator framework that combines a coarse-resolution MCMC to locate the high-likelihood region with adaptive, targeted sampling to train an emulator for the SCRIPT semi-numerical reionization model. The resulting emulators achieve $R^2$ around $0.97$–$0.99$ using roughly $10^3$ high-resolution simulations and reproduce full high-resolution posteriors within an MCMC while reducing expensive simulations by ~100× and CPU cost by up to ~70×. This approach enables efficient inference in higher-dimensional EoR models and provides a general strategy for next-generation reionization constraints, including JWST and upcoming 21 cm data.

Abstract

Constraining the Epoch of Reionization (EoR) with physically motivated simulations is hampered by the high cost of conventional parameter inference. We present an efficient emulator-based framework that dramatically reduces this bottleneck for the photon-conserving semi-numerical code SCRIPT. Our approach combines (i) a reliable coarse-resolution MCMC to locate the high-likelihood region (exploiting the large-scale convergence of SCRIPT) with (ii) an adaptive, targeted sampling strategy to build a compact high-resolution training set for an artificial neural network based emulator of the model likelihood. With only $\approx 10^3$ high-resolution simulations, the trained emulators achieve excellent predictive accuracy ($R^2 \approx 0.97$--$0.99$) and, when embedded within an MCMC framework, reproduce posterior distributions from full high-resolution runs. Compared to conventional MCMC, our pipeline reduces the number of expensive simulations by a factor of $\sim 100$ and lowers total CPU cost by up to a factor of $\sim 70$, while retaining statistical fidelity. This computational speedup makes inference in much higher-dimensional models tractable (e.g., those needed to incorporate JWST and upcoming 21 cm datasets) and provides a general strategy for building efficient emulators for next generation of EoR constraints.

Figures (5)

• Figure 1: Parameter constraints from the full high-resolution ($N_\text{grid}=32$) MCMC run (red) and the ANN-based MCMC runs with the emulator trained on $10^4$ parameter vectors sampled using LH sampling from the full prior, and with the priors reduced by factors of 2 and 4. The dashed lines mark the prior boundaries.
• Figure 2: Training diagnostic plots for the best-performing $N_\text{grid}=32$ emulator. Left: Evolution of the training and model-validation MSE as a function of training epochs. The dashed line marks the epoch at which the model-validation MSE reached its minimum, at which point the trained network was saved. Right: True $\chi^2$ values versus the $\chi^2$ values predicted by the emulator for the test set.
• Figure 3: Training diagnostic plots for the best-performing $N_\text{grid}=64$ emulator. Formatting is the same as in figure \ref{['fig:model_perf_32']}.
• Figure 4: Parameter constraints from the full high-resolution ($N_\text{grid}=32$) MCMC run (red) and the ANN-based MCMC run (blue). The diagonal panels show 1D posterior probability distributions while the off-diagonal panels show joint 2D posteriors. The contours represent 68 percent and 95 percent confidence intervals. The dashed lines denote the best-fitting parameter values. The quoted values on each parameter show the mean along with the 1$\sigma$ uncertainties.
• Figure 5: Parameter constraints from the full high-resolution ($N_\text{grid}=64$) MCMC run (red) and the ANN-based MCMC run (blue). Formatting is the same as in figure \ref{['fig:constraints_32']}.