Physics Informed Deep Learning Model for Dimensionality Reduction and Temporal 
Evolution of Collisional-Radiative Model

As the world’s largest tokamak fusion reactor, the international thermonuclear experimental
reactor (ITER) will be the first fusion device to produce net energy. Currently, the biggest threat to
ITER’s safe operation is plasma disruptions. A disruption is a sudden termination of the plasma and
can lead to extreme heat deposition and relativistic electron beam impact on reactor walls. Accurate
modeling of fusion plasmas with added impurities is a necessary component to understanding the
physics and mitigation plan of tokamak disruptions.

A collisional-radiative (CR) model is a high-dimensional nonlinear ODE that describes transitions
into or out of ionic excited state populations. The solution of this system can be computationally
challenging when ion populations are large and the rate matrix is near singular. Furthermore,
solving the CR model at each time-step and iteration in a plasma transport code leads to high
and undesirable computational costs particularly in higher dimensions. We seek an alternative
approach to allow rapid evaluation of necessary quantities. 

At Los Alamos National Laboratory, as an research intern, my role in this project is to develop a 
physics-informed deep learning model for dimensionality reduction and temporal evolution of the 
CR model.