From Simulations to Reality: Dark Energy Reconstruction with Simulated SNIa data from the Vera C. Rubin Observatory

In this paper, we present an Artificial Neural Network (ANN) based reconstruction analysis of the Supernova Ia (SNIa) distance moduli (μ(z)), and hence dark energy, using LSST simulated three-year SNIa data. Our ANN reconstruction architecture can model both the distance moduli and their corresponding error estimates. For this we employ astroANN and incorporate Monte Carlo dropout techniques to quantify uncertainties in our predictions. We tune our hyperparameters through advanced genetic algorithms, including elitism, utilizing the DEAP library. We compared the performance of the ANN based reconstruction with two theoretical descriptions of dark energy models, ΛCDM and Chevallier-Linder-Polarski (CPL). We perform a Bayesian analysis for these two theoretical models using the LSST simulations and also compare with observations from Pantheon and Pantheon+ SNIa real data. We show that our model-independent reconstruction using ANN is consistent with both of them. We assessed the performance using mean squared error (MSE) and showed that the ANN can produce distance estimates in better agreement with the LSST dataset than either ΛCDM or CPL, albeit very small. We included an additional residual analysis and a null test with F-scores to show that the reconstructed distances from the ANN model, are in excellent agreement with the ΛCDM or CPL model.