SVR Algorithm as a Tool for More Optimal Intergalactic Medium Simulation in the Epoch of Reionization

Abstract

All kinds of simulations of the intergalactic medium, such as hydrodynamic simulation, N-body simulation, numerical and semi-numerical simulation, etc., have been used to realize the history of this medium. In addition, emulators also have become widely used in recent years. Especially because they are very fast, they are used in many works instead of actual complete simulations. Emulators are trained with the help of simulator results and can work with very high accuracy similar to real simulators. Usually, reliable machine learning algorithms, such as neural networks, are used to program these emulators. In this work, we show that one of the best machine learning algorithms that can be used in general and have extremely high accuracy is the Support Vector Regression algorithm. This algorithm works better than the Multilayer Perceptron algorithm not only from the practical point of view but also from the theoretical point of view. To prove this issue, we have considered the 21SSD hydrodynamical simulation in addition to the theoretical arguments, and we have used both algorithms to be trained by this simulation for comparison. Finally, we have shown that the Support Vector Regression algorithm can be a much better and more accurate algorithm than neural network algorithms in future emulations, and by mastering it, one can neglect other algorithms with full confidence. Although this work is focused only on an intergalactic medium simulation in the Epoch of Reionization, it can be used for all other simulations regardless of their type. In other words, it can be employed for all hydrodynamic simulations, N-body simulations, etc. in all cosmic epochs and rely on its results.