Rapid Magnetic Field Inference for Ocean Characterization at Europa via Transformers

Jupiter’s moon Europa is a key target in the search for extraterrestrial life, but assessing its habitability via magnetic induction requires a well-grounded understanding of the plasma environment. At Europa, this task relies on magnetohydrodynamic (MHD) models that have been developed over many years. While powerful, these models are computationally expensive, with some codes requiring > 12 hours on a 2,000-core machine. Fitting spacecraft observations to MHD models demands many runs, resulting in a potentially days- or weeks-long process. This presents a major bottleneck for missions such as Europa Clipper and JUICE, potentially limiting their scientific return.
In this talk, I introduce a transformer-based surrogate for a state-of-the-art MHD model used to help characterize Europa’s subsurface ocean. The surrogate evaluates in milliseconds on a laptop rather than hours on a supercomputer, achieving a speed-up of approximately 40,000x while delivering high-fidelity, uncertainty-aware magnetic field predictions. This acceleration enables three new scientific pathways that are feasible with MHD alone: large-scale parameter surveys, simulation-based inference, and feature-importance analysis. These analyses are important as the environment is poorly constrained observationally. Overall, this approach represents a paradigm shift in the investigation of space plasmas and opens the door to a host of novel science investigations.