Purifying Electron Spectra from Noisy Pulses with Machine Learning Using Synthetic Hamilton Matrices

Area of Science:

• Quantum optics
• Attosecond science
• Computational physics

Background:

• Intense laser pulses from free-electron lasers (FELs) produce noisy photoelectron spectra.
• Shot-to-shot fluctuations in FEL pulses complicate spectral analysis.
• Accurate spectral data is crucial for understanding atomic and molecular dynamics.

Purpose of the Study:

• To develop a computational method for purifying noisy photoelectron spectra.
• To extract the intrinsic Fourier-limited pulse spectrum from FEL data.
• To demonstrate the general applicability of the purification method to various ionization processes.

Main Methods:

• Utilized a deep neural network trained on simulated photoelectron spectra.
• Employed efficient propagation of the Schrödinger equation with synthetic Hamilton matrices.
• Generated numerous random realizations of fluctuating FEL pulses for training.

Main Results:

• The deep neural network successfully purified noisy spectra from FELs.
• The method accurately recovered the Fourier-limited pulse spectrum.
• The trained network generalized to purify spectra from resonant two- and three-photon ionization processes without specific training.

Conclusions:

• Deep neural networks offer a powerful tool for processing noisy FEL data.
• The developed method enhances the reliability of photoelectron spectroscopy.
• This approach facilitates the study of nonlinear atomic and molecular processes sensitive to pulse characteristics.