Speaker
Description
Transport processes are virtually ubiquitous in engineering fluid and plasma problems but their properties are not always well-determined, particularly when complex microphysics is at play. One outstanding example is heat flux, which according to both laser plasma experiments performed at NIF and more recently measurements of astrophysical plasmas becomes strongly suppressed with respect to predictions from Spitzer-Härm when the electron mean-free-path approaches the temperature gradient scale-length. While such information is contained in the results of microscopic-scale numerical simulations close to first principles or experiments it remains in a form that is not suitable for macroscopic modelling. Here we leverage machine learning to produce micro-physics informed transport flux representations applicable to a macroscopic/fluid model description. We address convergence issues arising from noisiness of deep neural networks representations in numerical schemes. Our results apply most specifically to astrophysical plasmas where severe flux reduction may occur while arguably locality of the heat flux function is maintained and represent a promising initial step towards filling the gap between micro and macro description in this important area of modelling.
