A Shortcut to Statistically Steady-State Turbulence with Flow Matching

Many nonlinear physical systems, like those in fluid dynamics, require extensive computational resources to simulate their initial transient phases before reaching a steady state. Current methods often rely on autoregressive models that accumulate errors over time, leading to inefficiencies. This paper introduces GyroFlow, a generative model that directly estimates the steady-state behavior of gyrokinetic turbulence, avoiding the costly transient phase. Builders in computational fluid dynamics might find this approach beneficial as it provides faster simulations without sacrificing accuracy.

Introduces a novel generative model for estimating steady-state statistics without resolving transients.

Demonstrates performance against existing methods with a new evaluation metric.