Employing deep-learning techniques for the conservative-to-primitive recovery in binary neutron star simulations

Area of Science:

• Nuclear Astrophysics
• Computational Physics
• Gravitational Wave Astronomy

Background:

• Binary neutron star mergers are crucial for understanding nucleosynthesis and the equation of state of dense matter.
• Numerical-relativity simulations are essential for studying binary neutron star mergers but are computationally expensive.
• The conversion of conservative to primitive variables is a key step in these simulations.

Purpose of the Study:

• To investigate the use of neural networks for converting conservative to primitive hydrodynamical variables in binary neutron star merger simulations.
• To assess the stability, accuracy, and computational cost of this novel deep learning approach.

Main Methods:

• Implementation of neural network techniques for variable conversion in numerical-relativity simulations of binary neutron star mergers.
• Comparison of simulation results with traditional methods in terms of stability, accuracy, and computational resource utilization.

Main Results:

• The study presents the first binary neutron star merger simulations utilizing neural networks for variable conversion.
• Simulations demonstrated stability and achieved accuracy comparable to traditional methods.
• The computational cost was found to be comparable to existing techniques.

Conclusions:

• Neural networks show promise for use in numerical-relativity simulations, offering a potential pathway for future advancements.
• These deep learning techniques can be employed for stable and accurate variable conversion in binary neutron star merger simulations.
• Further research and optimization are needed to achieve a computational advantage over traditional methods.