Learning surrogate models for time‑dependent partial differential equations usually requires rebuilding a simulation corpus whenever the governing operator changes. We introduce RD‑JEPA, a joint‑embedding predictive architecture for self‑supervised pretraining on reaction‑diffusion trajectories. A single model is pretrained on five parametrized systems and then adapted to three held‑out systems whose reaction operators and trajectories are excluded from pretraining.\
During adaptation, using only one, five, or ten complete trajectories from a held‑out system, RD‑JEPA achieves lower mean relative discrete $\ell^2$ field error and lower mean absolute spatial first‑difference error than five supervised surrogate baselines, an independently trained control that removes the trajectory‑dependent predictive latent pathway, and an architecture‑matched model trained from scratch.\
Across the evaluated equations, output resolutions, forecast horizons, and choices of adaptation trajectories, the results indicate that predicting future‑state representations can support data‑efficient adaptation across related reaction‑diffusion systems.\
Review