Speaker
Description
Type I X-ray bursts (XRBs) are explosive astrophysical phenomena powered by hundreds of thermonuclear reactions in the rapid proton capture process ($rp$-process). Sensitivity studies with XRB simulation codes have been used to identify nuclear reactions that have the most impact on observables and should be prioritized for future studies. Due to the high computational cost and time-consuming nature of hydrodynamic simulations, previous sensitivity studies only considered the impact of variations of one reaction rate at a time. Consequently, the impacts of reaction correlations by simultaneous variation of multiple rates have not been well investigated.
We propose a novel deep learning approach to emulate XRB simulations and significantly accelerate predictions of XRB observables. By training a deep neural network on datasets of XRB properties generated with the multi-zone hydrodynamic code MESA, we can explore the impact of simultaneous variations of multiple reaction rates. This enables us to identify unexplored combinations of reactions that have substantial influence on XRB properties. Details of the method and preliminary results will be presented.