Deep neural network for modeling soliton dynamics in the mode-locked laser

Area of Science:

• Nonlinear optics
• Ultrafast photonics
• Machine learning applications in physics

Background:

• Predicting soliton dynamics in ultrafast fiber lasers is crucial for controlling laser output.
• Traditional modeling methods can be computationally intensive and may not capture complex nonlinear behaviors effectively.

Purpose of the Study:

• To develop a novel neural network model for predicting soliton dynamics in ultrafast fiber lasers.
• To accurately forecast the transition from detuning steady state to stable mode-locking.

Main Methods:

• Utilizing a Bidirectional Long Short-Term Memory (Bi_LSTM) recurrent neural network (RNN) with an attention mechanism.
• Training and testing the model on conventional soliton and soliton molecule evolution under varying parameters (saturation energy, group velocity dispersion).

Main Results:

• The Bi_LSTM RNN achieved a root mean square error (RMSE) below 15% for 80% of test samples.
• Predicted conventional soliton pulse width and soliton molecule pulse interval showed strong agreement with experimental results.

Conclusions:

• The proposed neural network model offers an effective approach for nonlinear dynamics modeling of ultrafast fiber lasers.
• This work provides a new perspective on understanding and predicting complex soliton behaviors in optical systems.