Toward a Multimodal Foundation Model for Plasma State Reconstruction
Abstract
Plasma experiments observe coupled physical states through sparse, heterogeneous, and incomplete diagnostics. We present a proof of concept toward a multimodal foundation model for plasma state reconstruction during magnetic reconnection, motivated by the Facility for Laboratory Reconnection Experiments (FLARE). Using 250 synthetic VPIC simulations at fixed reference electron beta βe0=0.2, with 225 training runs and 25 disjoint validation runs, we train a 5.5M-parameter residual 3D U-Net to reconstruct three magnetic-field components and electron density from incomplete observations. The model combines spatial-only pooling, which preserves all 24 temporal frames, with global spatiotemporal self-attention and rotary positional encoding. Independently masking magnetic and density modalities across random, grid, block, and temporal patterns enables sparse sensing, cross-modal completion, density super-resolution, and temporal completion with a single set of weights and no task-specific retraining. With density fully hidden, just 0.1% magnetic spatial visibility reduces median density NRMSE from approximately 0.9 to 0.1. Conversely, observed density improves current-density reconstruction when magnetic data are absent or extremely sparse. With no magnetic observations, roughly 1% density visibility approaches the density reconstruction accuracy achieved with fully observed magnetic fields, although current structure remains more dependent on direct magnetic measurements. Full magnetic conditioning also stabilizes completion of missing future density frames. These results demonstrate bidirectional cross-modal predictive coupling within this simulation regime. Generalization across plasma regimes or to experimental measurements remains untested.
Related
- A generative foundation model for 2D magnetized plasma turbulence
- Experimental validation of a fast control-oriented, physics-informed surrogate model for plasma equilibrium reconstruction in the TCV tokamak
- Kibble-Zurek Dynamics in Two-dimensional Frustrated Systems with a Neural Foundation-state Subspace Method
- Reconstruction of Multiscale Plasma Dynamics Across Operating Regimes
- Comment on "Long-range magnetic fields in the ground state of the Standard Model plasma"
BibTeX
@misc{xia2026toward,
title = {{Toward a Multimodal Foundation Model for Plasma State Reconstruction}},
author = {Bin Xia and Chuanfei Dong and Boting Li and Yilan Qin and Zikang Xie and Sung Hyun Son and Adam J. Stanier and Jongsoo Yoo and Hantao Ji and Ahmed Diallo and Han Ding and John Wise},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.15076},
eprint = {2610.15076},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.15076},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Toward a Multimodal Foundation Model for Plasma State Reconstruction
AU - Bin Xia
AU - Chuanfei Dong
AU - Boting Li
AU - Yilan Qin
AU - Zikang Xie
AU - Sung Hyun Son
AU - Adam J. Stanier
AU - Jongsoo Yoo
AU - Hantao Ji
AU - Ahmed Diallo
AU - Han Ding
AU - John Wise
PY - 2026
DA - 2026/10/10
PB - arXiv
DO - 10.48550/arxiv.2610.15076
UR - https://arxiv.org/abs/2610.15076
AB - Plasma experiments observe coupled physical states through sparse, heterogeneous, and incomplete diagnostics. We present a proof of concept toward a multimodal foundation model for plasma state reconstruction during magnetic reconnection, motivated by the Facility for Laboratory Reconnection Experiments (FLARE). Using 250 synthetic VPIC simulations at fixed reference electron beta β_e0=0.2, with 225 training runs and 25 disjoint validation runs, we train a 5.5M-parameter residual 3D U-Net to reconstruct three magnetic-field components and electron density from incomplete observations. The model combines spatial-only pooling, which preserves all 24 temporal frames, with global spatiotemporal self-attention and rotary positional encoding. Independently masking magnetic and density modalities across random, grid, block, and temporal patterns enables sparse sensing, cross-modal completion, density super-resolution, and temporal completion with a single set of weights and no task-specific retraining. With density fully hidden, just 0.1% magnetic spatial visibility reduces median density NRMSE from approximately 0.9 to 0.1. Conversely, observed density improves current-density reconstruction when magnetic data are absent or extremely sparse. With no magnetic observations, roughly 1% density visibility approaches the density reconstruction accuracy achieved with fully observed magnetic fields, although current structure remains more dependent on direct magnetic measurements. Full magnetic conditioning also stabilizes completion of missing future density frames. These results demonstrate bidirectional cross-modal predictive coupling within this simulation regime. Generalization across plasma regimes or to experimental measurements remains untested.
ER -