Search papers, labs, and topics across Lattice.
The paper introduces APEX, a framework for predicting high-frequency wave fields when target data is scarce, leveraging the asymmetry in cross-frequency transfer where amplitude structure is more stable than phase. APEX uses a lower-frequency neural operator to predict a coarse field, extracts the amplitude as an anchor, and then employs a conditional flow-matching enhancer guided by a Green's-function-inspired phase prior to reconstruct the high-frequency field. Experiments on wave benchmarks demonstrate that APEX outperforms direct extrapolation and other baselines under limited high-frequency supervision.
Don't try to directly extrapolate wave fields to higher frequencies when data is scarce; instead, anchor on the transferable amplitude structure and separately recover the oscillatory details.
Learning-based surrogates have become increasingly effective for wave-field prediction, and neural operators in particular have shown strong performance within observed frequency regimes. However, higher-frequency prediction under scarce target supervision remains comparatively underexplored, especially in wave problems where higher-frequency data are substantially more expensive to simulate or measure than lower-frequency data. A central difficulty is that cross-frequency transfer is inherently asymmetric: coarse amplitude structure remains relatively stable across frequencies, whereas phase-sensitive oscillatory structure deteriorates much more rapidly as frequency increases. Motivated by this asymmetry, we propose APEX, Amplitude-anchored and Phase-prior-guided Enhancement from eXtrapolated coarse predictions, a framework for target-scarce higher-frequency wave-field prediction. A lower-frequency neural operator first provides a coarse prediction in the target-frequency regime, from which we retain only the amplitude as a transferable structural anchor. A conditional flow-matching enhancer then reconstructs the target higher-frequency field under the guidance of a Green's-function-inspired phase prior. Experiments on SimpleWave, Helmholtz, and Maxwell benchmarks show that APEX consistently outperforms direct lower-to-higher extrapolation, target-adapted operator, and joint generative baselines under limited target-frequency supervision. Our results suggest that reliable higher-frequency prediction of oscillatory wave fields should not rely on direct end-to-end transfer of the full complex field, but instead on explicitly reusing transferable coarse structure while separately recovering the missing oscillatory detail.
