• Accepted Paper

Predicting the characteristic polynomials of rogue wave patterns from initial-boundary data via deep learning

Zhe Lin and Yong Chen

Phys. Rev. E - Accepted 16 November, 2026

DOI: https://doi.org/10.1103/qdlp-2gzb

Abstract

Rogue waves in integrable systems organize into structured patterns whose geometry is governed by the root distributions of characteristic polynomial hierarchies, such as the Yablonskii–Vorob’ev polynomials of the nonlinear Schrödinger equation. These polynomial structures are conventionally obtained through asymptotic analysis of exact solutions defined over the complete spatiotemporal field. In realistic settings, however, only initial and boundary data are physically accessible, and inferring the underlying pattern structure from such limited observations remains an open problem. Here we show that the characteristic polynomial representation of rogue wave patterns can be recovered directly from initial-boundary data using a deep learning framework. A unified classification–regression architecture simultaneously identifies the pattern type and estimates the associated polynomial parameters. Trained on high-fidelity data generated via the Darboux transformation and large-parameter asymptotic analysis, the framework achieves high classification accuracy and precise parameter estimation, with the predicted polynomial root distributions exhibiting errors consistent with theoretical asymptotic expectations in the large-parameter regime.

Export citation

Export citation

Choose format for download:

Download Citation

If the author has provided any supplemental materials with this article they will be available upon publication of the version of record.

Sign In to Your Journals Account

Filter

Filter

Article Lookup

Enter a citation