Generative diffusion surrogates with analytical variance schedule.

Prescribing the variance derivative as a transport clock enables generative diffusion models to accurately emulate turbulent plasma transport without schedule tuning.

• Why it matters: Accurate modeling of stochastic transport systems is crucial for understanding physical phenomena where distributional structures are complex and non-Gaussian, yet current models lack physical calibration.
• What they did: The authors developed a generative diffusion approach that uses the known variance or mean-square displacement to define the forward noising rate, enforcing the variance path and learning the non-Gaussian structure without intermediate data.
• The result: The surrogate reproduces test-particle distributions, matches laboratory-measured variance scales, and tracks kurtosis evolution in turbulent plasmas, facilitating calibrated emulation and likelihood inference.