Machine Learning Surrogate Model for Neutron Star Mass and Radius from Piecewise Polytropic Parameters

We have constructed a surrogate model to predict neutron star mass and radius from a three-segment parameterized piecewise polytropic equation of state using an artificial neural network. We have trained the network with the generated data from the fourth-order Runge-Kutta-based solver of the Tolman-Oppenheimer-Volkoff equation. It shows that the neural network predicts the mass-radius with no less than 99% accuracy and significantly reduces the computation time compared to the traditional Runge-Kutta method. However, caution is advised when predicting outside the training data parameter ranges, as the model exhibits poor accuracy in extrapolating data and tends to generate false output values where no stellar solution exists. We have argued that this situation may also occur in other similar neural network-based surrogate models.