Boussinesq Vorticity Predictor — multi-ε

A Fourier Neural Operator (FNO) trained on 2D Boussinesq convection. It maps initial vorticity $\omega_0$ and temperature $\theta_0$ to the vorticity field at $T = 100$ in milliseconds.

New in this version: the input amplitude $\varepsilon$ is a user parameter. Initial data are normalized so that $|u_0|{H^2} = |\theta_0|{H^2} = \varepsilon / 2$ (with $u_0$ from $\omega_0$ by Biot–Savart). Trained range: $\varepsilon \in [0.021,, 0.476]$.

Conventions. Array axis 0 is $x_1$ (horizontal), axis 1 is $x_2$ (vertical). Grid $128 \times 128$ on $[0, 2\pi]^2$, $\nu = \eta = 1$, final time $T = 100$. Weights: moust1am/boussinesq-fno-weights.