Machine learning optimizes high-power laser experiments

Commercial fusion energy plants and advanced compact radiation sources may rely on high-intensity high-repetition rate lasers, capable of firing multiple times per second, but humans could be a limiting factor in reacting to changes at these shot rates. 

Applying advanced computing to this problem, a team of international scientists from Lawrence Livermore National Laboratory (LLNL), Fraunhofer Institute for Laser Technology (ILT) and the Extreme Light Infrastructure (ELI ERIC) collaborated on an experiment to optimize a high-intensity, high-repetition-rate laser using machine learning.   

"Our goal was to demonstrate robust diagnosis of laser-accelerated ions and electrons from solid targets at a high intensity and repetition rate," said LLNL’s Matthew Hill, the lead researcher. "Supported by rapid feedback from a machine-learning optimization algorithm to the laser front end, it was possible to maximize the total ion yield of the system." 

The researchers trained a closed-loop machine learning code developed by LLNL’s Cognitive Simulation team on laser-target interaction data to optimize the laser pulse shape, allowing it to make adjustments as the experiment ran. Data generated during the experiment was fed back into the machine learning-based optimizer, allowing it to tweak the pulse shape on the fly.  

The laser fired every 5 seconds, consistently exceeding laser intensities of 3x10²¹ W/cm² at its focus, stopping after 120 shots when the copper target foil had to be replaced. During this time, the researchers also inspected the diagnostics for damage and assessed debris accumulation from vaporized targets. The team conducted experiments at ELI for three weeks, with experimental runs lasting approximately 12 hours per day, during which the laser would fire up to 500 shots.   

The experiment took place at the ELI Beamlines Facility in the Czech Republic, where the researchers utilized the state-of-the-art High-Repetition-Rate Advanced Petawatt Laser System (L3-HAPLS) to generate protons in the ELIMAIA laser-plasma ion accelerator. Focusing on the goal of applying machine learning to a high-rate-laser experiment, the team simplified aspects of the experiment as much as possible, like using a robust, simple copper foil target. 

"By harnessing the HAPLS and pioneering machine learning techniques, we embarked on a remarkable endeavor to further comprehend the intricate physics of laser-plasma interactions,” said Constantin Haefner, managing director of Fraunhofer ILT. 

More than 4,000 shots were fired during the campaign, allowing statistical analysis to be performed on the results and demonstrating optimization of ion yield above the already-impressive nominal baseline performance. 

Using machine learning was a new experience for the experimental physicists. “It becomes a spectator sport,” Hill said. “We watched the data coming in and tried to guess what the optimizer would do. It’s very different than an experiment with manual intervention.” 

LLNL becomes a L3-HAPLS user  

The L3-HAPLS laser has excellent laser performance repeatability, displaying exceptionally stable alignment, focal spot quality and the ability to generate intense laser pulses at a high repetition rate to drive the generation of secondary sources such as electrons, ions and x-rays.  

"The successful execution of such a complex experiment showcases the cutting-edge quality and reliability of the L3-HAPLS laser system,” said Bedrich Rus, chief laser scientist at ELI Beamlines.  

LLNL developed the HAPLS laser as part of a bilateral agreement with ELI Beamlines, with first light from the system after delivery and installation in the Czech Republic in 2017. This was only the second user experiment at the facility, having been awarded time through a competitive worldwide call for proposals, now issued twice annually and attracting hundreds of applications. 

Lengthy preparation pays off 

In addition to Hill, the LLNL team of Elizabeth Grace, Franziska Treffert, James McLoughlin, Isabella Pagano, Abhik Sarkar, Raspberry Simpson, Blagoje Djordjevic, Matthew Selwood, Derek Mariscal, Jackson Williams and Tammy Ma spent about a year preparing for the experiment with the Fraunhofer ILT and ELI Beamlines teams. In addition to local facility diagnostics, the Livermore team fielded several instruments developed under the Laboratory Directed Research and Development Program, including the REPPS magnetic spectrometer, PROBIES ion beam imaging spectrometer, a rep-rated scintillator imaging system and rep-rated X-ray spectrometer. 

To follow export control regulations, the machine learning code had to be run at LLNL to segregate it from computer systems at ELI Beamlines. This created a challenge of rapidly transferring data between the Czech Republic to LLNL to keep pace with the laser.  

Sarkar helped engineer the trans-Atlantic coupling between computers at LLNL and at ELI for real-time closed-loop operation. “Thanks to hard work on both sides, we came up with a solution,” he said. “On the computing side, this opens up new possibilities for remote operation.”  

The experiment was very successful, generating robust data from the first day. In future experiments, the team hopes to allow machine learning to control additional parameters like the focal spot and to use the approach at different experimental facilities.

More broadly, the success of coupling state-of-the-art laser technology with machine learning techniques could open new avenues for advancements in various fields such as fusion energy, medical therapy, materials science and non-destructive analysis in the field of cultural heritage and archaeology.