Active learning-assisted neutron spectroscopy with log-Gaussian processes.

Area of Science:

• Condensed Matter Physics
• Materials Science
• Neutron Scattering

Background:

• Three-axes spectrometers (TAS) are crucial for studying magnetic and lattice excitations in materials.
• High demand and limited beam time for TAS experiments necessitate improved efficiency.
• Manual searching for weak signals in uninformative regions is time-consuming and inefficient.

Purpose of the Study:

• To develop a more efficient method for conducting neutron scattering experiments.
• To reduce the time and resources required for TAS experiments.
• To autonomously identify optimal measurement locations for signal detection.

Main Methods:

• A probabilistic active learning approach was employed.
• Log-Gaussian processes were utilized to guide measurements.
• The method operates autonomously without human intervention.

Main Results:

• The active learning approach successfully identified informative measurement locations.
• Demonstrated benefits on a real three-axes spectrometer experiment.
• Validated performance on a benchmark dataset with diverse excitations.

Conclusions:

• The proposed method significantly enhances the efficiency of neutron scattering experiments.
• Autonomous data acquisition using active learning is feasible and effective.
• This approach optimizes the use of limited beam time for materials property investigations.