Credit: NASA’s Conceptual Image Laboratory
Current Research
Unnormalized Density Estimation of the Solar Wind
I am developing a normalizing-flow–based model to estimate the unnormalized velocity distribution function, \( f(v) \), represented by the flux with units \( \# / (\mathrm{m}^3 \, (\mathrm{km/s})^3) \), for the solar wind using Parker Solar Probe (SPAN-I) particle flux data.
This research focuses on reconstructing velocity-space particle distributions from low-count data, where classical Gaussian assumptions fail due to the anisotropic, weakly collisional nature of solar-wind plasma. Traditional analytical models are fast but often inaccurate. My approach uses normalizing flows to reconstruct the full velocity distribution while accounting for instrumental geometry and uncertainty.
Interactive 3D Velocity Distribution Function
Below is an interactive visualization of the solar-wind velocity distribution function (VDF) on 16-Dec-2021 13:14. You can rotate, zoom, and explore the structure of \( f(v) \), represented by color intensity estimated using a normalizing-flow model. The axes represent particle velocities relative to the magnetic field direction (e.g., \(v_\parallel\) is the parallel component). The “fan-shaped” velocity grid reflects the SPAN-I instrument’s limited field-of-view on Parker Solar Probe, a challenge this model addresses by reconstructing missing regions in velocity space.
Interactive 3-D visualization of the modeled VDF, generated with Plotly in Python.
A Peek into the Solar Wind and Parker Solar Probe
Below are two NASA visualizations showcasing the solar wind flowing from the Sun and the Parker Solar Probe. For more information, visit NASA’s overview of the solar wind and the official Parker Solar Probe mission page .
Research Goals
- Estimate the true, continuous un-normalized velocity distribution function with flux units \( \# / (\mathrm{m}^3 \, (\mathrm{km/s})^3) \)
- Quantify aleatoric uncertainty from Poisson counting statistics
- Compare performance of flow-based estimators against KDE, parametric, and non-parametric models
Acknowledgements
This research is supervised by
Dr. Jay Newby
and
Dr. Abigail Azari.
Conducted at the University of Alberta as part of ongoing research into
unnormalized density estimation for solar-wind data.