EAGER: An Inference Methodology to Illuminate Nonlinear Neutrino Flavor Transformation for Nuclear Astrophysics

Probing the physics underlying cosmic explosions is vital for understanding the makeup of the observable Universe. Specifically, the explosions of massive stars are candidate sites for the nucleosynthesis of some heavy elements – the building blocks of structures including life on Earth. Meanwhile, important aspects of the physics of these explosions are difficult to access via traditional approaches in nuclear astrophysics.

This is due both to a lack of adaptability of existing codes to the required mathematical framework and to computational complexity. Moreover, important features of these explosions remain artificially hidden from the tools built to describe them. Meanwhile, inference is an alternative methodology, related to machine learning techniques.

In the geosciences and neurobiology, inference has demonstrated success in illuminating problems akin to those noted to hinder progress within nuclear astrophysics. The potential for inference to illuminate these problems is high, and thus this project will explore inference to bear upon them. Innovations cultivated within one scientific arena can be transformative when applied to disjoint fields.

Integral to the research is the training of undergraduates, many with socio-economic backgrounds under-represented in science. Students also engage in comedic science public outreach.

The physics noted as “artificially hidden” from traditional techniques is direction-changing backscattering in the neutrino flavor field in these high-density environments. Neutrinos are elementary particles whose “flavor” dictates the manner in which they interact with other particles. Flavor in large part sets the neutron-to-proton ratio as well as energy and entropy deposition, thereby in-part dictating the mechanism of explosion and nucleosynthesis.

Direction-changing backscattering can occur in the flavor field and significantly shape the explosion, but it presents a two-point boundary-value problem: a framework that traditional numerical integration is ill-equipped to handle. This project investigates the ability of statistical data assimilation (SDA) to illuminate this problem. SDA is a Bayesian inference methodology, invented for numerical weather prediction, to predict sparsely-sampled nonlinear systems.

In principle, SDA is well-suited for solving boundary-value problems, and it is expected to outperform integration in computational efficiency. This project seeks to establish whether SDA can outperform integration in terms of 1) solving the direction-changing backscattering problem, and 2) efficiency so as to avoid sacrificing physical detail. In preliminary results, SDA efficiently recovers solutions obtained by integration in the non-backscattering regime.

These findings call for a deep examination of SDA’s ability to handle the back-scattering problem.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.