Physics-informed deep learning for three dimensional black holes

TL;DR

This paper introduces a neural network approach inspired by AdS/DL correspondence to model black hole metrics, demonstrating how quantum scalar fields influence the accuracy of holographic metric predictions.

Contribution

It develops a novel neural network architecture that incorporates holography principles and quantum fluctuations to accurately produce black hole metrics from data.

Findings

Loss function convergence depends on epochs and learning rate.

Regularization term critically affects metric matching.

Optimal learning parameters improve model accuracy.

Abstract

According to AdS/DL (Anti de Sitter/ Deep Learning) correspondence given by \cite{Has}, in this paper with a data-driven approach and leveraging holography principle we have designed an artificial neural network architecture to produce metric field of planar BTZ and quintessence black holes. Data has been collected by choosing minimally coupled massive scalar field with quantum fluctuations and try to process two emergent and ground-truth metrics versus the holographic parameter which plays role of depth of the neural network. Loss or error function which shows rate of deviation of these two metrics in presence of penalty regularization term reaches to its minimum value when values of the learning rate approach to the observed steepest gradient point. Values of the regularization or penalty term of the quantum scalar field has critical role to matching this two mentioned metric. Also we design an algorithm which helps us to find optimum value for learning parameter and at last we understand that loss function convergence heavily depends on the number of epochs and learning rate.