When traditional numerical methods face challenges in meeting the demand for fast, accurate analysis of optical devices, artificial intelligence (AI) offers a helpful alternative to the existing approaches—this time, for Kerr resonators, which play an important role in technologies like optical frequency combs and ultrafast pulses generation.
Kerr resonators are tiny but powerful: these devices generate temporal cavity solitons (CSs)—stable, pulse-like light structures—and optical frequency combs, which are critical for technologies like high-precision sensing, optical data storage, and ultrafast communications. For decades, scientists have relied on the Lugiato-Lefever Equation (LLE) solved via the split-step Fourier method (SSFM) to study the resonators’ nonlinear dynamics. But there’s a catch: SSFM is slow, often taking minutes to run simulations, creating a bottleneck for researchers and engineers trying to test new device designs or optimize parameters.

“Traditional simulations hold back innovation—when you need to tweak pump power, pulse width, or detuning, waiting minutes for each result slows down the entire process,” said Tianye Huang, lead researcher at the China University of Geosciences (Wuhan) and the study’s corresponding author. “We wanted to build an AI model that keeps the accuracy of LLE but cuts the time down to seconds.” The team’s solution focuses on RNNs, a type of AI built to handle time-series data—perfect for tracking how light evolves inside Kerr resonators over hundreds or thousands of “round trips” (when light circulates through the resonator). After testing different RNN architectures, they found that GRUs outperformed others, balancing accuracy and speed. This method is nearly 20 times faster than traditional numerical methods. The work entitled “Rapid prediction of complex nonlinear dynamics in Kerr resonators using the recurrent neural network” was published in Frontiers of Optoelectronics (published on Sept. 25, 2025).
DOI：10.1007/s12200-025-00164-4