
Scientists at Stanford University, the University of California, Los Angeles and SLAC National Accelerator Laboratory have developed an artificial intelligence model that accelerates simulations of light behaviour in laser systems, according to a press release from SPIE, the International Society for Optics and Photonics. The work is described in the peer-reviewed journal Advanced Photonics.
The model targets second-order nonlinear optics, or χ(2) processes, in which light waves exchange energy while travelling through specially designed crystals, allowing the frequency of light and the shape of laser pulses to be controlled. At SLAC's upgraded Linac Coherent Light Source (LCLS-II), infrared light is converted into green light and then into ultraviolet light; the ultraviolet pulse strikes a cathode, releasing electrons that are accelerated to generate X-rays. The timing and shape of the ultraviolet pulse help determine the characteristics of the electron bunch and hence the quality of the X-rays.
Conventionally, these interactions are simulated by solving the nonlinear Schrödinger equation with the split-step Fourier method, which repeatedly switches between time and frequency representations of the pulse. In the complete laser simulation examined by the researchers, this calculation alone accounted for approximately 95 percent of total processing time, the press release states.
The team's model is a recurrent neural network of the long short-term memory (LSTM) type, adapted from earlier work predicting pulse propagation in optical fibres. It works entirely with a compressed representation of the light's frequency characteristics instead of converting repeatedly between time and frequency. The model was tested on noncollinear sum frequency generation, in which three interconnected light fields evolve together inside a crystal, across a wide variety of pulse shapes, including pulses with pronounced gaps in their frequency spectra and strong variations in optical phase.
Processed together on graphics processing units, the average calculation time per simulation fell to a few milliseconds, which the paper describes as over 250× speedup while maintaining high fidelity. When the model predicted the main output of the sum frequency generation process accurately, its predictions for the other interacting fields were generally accurate as well, the researchers report.
The paper states that the work enables future real-time optimisation and lays the foundation for data-integrated modelling frameworks and digital twins of laser systems. The press release describes these as envisioned applications, including monitoring experiments and adjusting laser settings in response to changing conditions; no operating-system demonstration is reported.
Key facts
- Over 250× speedup of χ(2) nonlinear optics simulations, per the peer-reviewed paper
- Average calculation time of a few milliseconds per simulation when processed together on GPUs
- Split-step Fourier method accounted for approximately 95 percent of total processing time in the complete laser simulation examined
- Model tested on noncollinear sum frequency generation with three interconnected light fields in a crystal
Sources
- Deep learning-assisted modeling for χ(2) nonlinear optics — Advanced Photonics — Jack Hirschman, Erfan Abedi, Minyang Wang et al.
- AI supercharges laser physics with a stunning 250x speed boost — ScienceDaily (Engineering)
Discussion
Technical questions and corrections are welcome. Comments are signed and checked before they appear.
No comments yet.
Sign in to add a typed review comment. Reviews are signed with your name.



