Researchers propose GyroFlow, a latent generative model that directly estimates steady-state statistics of gyrokinetic turbulence in 5D phase space without resolving the transient phase. The approach bypasses explicit time evolution by modeling the distribution of saturated states under an ergodicity assumption.
- GyroFlow generates saturated snapshots from noise conditioned on dimensionless operating parameters.
- It outperforms autoregressive, reduced-order, and other generative approaches while providing substantial speedup.
- A new distributional metric called FGyD is introduced to evaluate generation quality in the latent space of a pretrained gyrokinetic model.
- FGyD correlates with downstream flux accuracy and solver convergence.
- GyroFlow can be used to warm-start the numerical code used to produce the data.
This method addresses the high computational cost of direct numerical simulations by avoiding the resolution of full transient dynamics, offering a faster alternative for systems where effective closures are not available.