Speaker
Description
Deep learning is one of the machine learning technologies developed in computer science. Recently, it has been implemented to various research areas in physics, including search of gravitational lens [1], identifying and classifying the phases of Ising model [2], the search of Higgs and exotic particles [3], classification jet structure [4], etc. In this talk, we will implement deep learning to relativistic hydrodynamics, which is a useful tool to simulate the evolution of relativistic systems in high energy nuclear physics and astrophysics [5].
Using 10000 initial and final energy density and flow velocity profiles generated from 2+1-d hydrodynamics with MC-Glauber initial conditions, we train the network and use it to predict the final profiles associated with various initial conditions, including MC-Glauber, MC-KLN and AMPT and TRENTo. A comparison with the hydrodynamic calculations shows that the network predictions can nicely capture the magnitude and inhomogeneous structures of these final profiles, as well as the related eccentricity distributions $P(\varepsilon_n)$ (n=2, 3, 4). These results indicate that deep learning can capture the main feature of the non-linear evolution of hydrodynamics, which shows the potential of largely accelerate the simulations of relativistic hydrodynamics.
Reference
[1] Y. D. Hezaveh, et.al., Nature 548, 555 (2017).
[2] J. Carrasquilla and G. R. Melko, Nature Phys. 13, 431 (2017).
[3] P. Baldi, et.al, Nature Commun. 5, 4308 (2014).
[4] P. T. Komiske, et. al., JHEP 1701, 110 (2017).
[5] H. Huang, B. Xiao, H. Xiong, Z. Wu, Y. Mu and H. Song, in preparation.
| Content type | Theory |
|---|---|
| Centralised submission by Collaboration | Presenter name already specified |
