Machine learning for laser-induced electron diffraction imaging of molecular structures.

Area of Science:

• Physical Chemistry
• Molecular Imaging
• Computational Chemistry

Background:

• Ultrafast diffraction imaging offers picometre spatial and attosecond temporal resolution for gas-phase molecular structures.
• Determining structures of complex molecules is challenging due to multi-dimensional solution spaces and computational intractability.
• Existing ultrafast electron diffraction methods struggle with large and complex molecular systems.

Purpose of the Study:

• To develop a machine learning approach for accurate structural retrieval of complex gas-phase molecules.
• To overcome the limitations of traditional methods in handling large molecular configurations and orientations.
• To demonstrate the feasibility of using machine learning with ultrafast diffraction imaging for complex molecular structures.

Main Methods:

• Implementation of a machine learning algorithm utilizing a convolutional neural network.
• Training the neural network with a limited dataset of molecular configurations.
• Application of the trained model to laser-induced electron diffraction (LIED) data for structural analysis.

Main Results:

• Successful structural retrieval of the complex molecule Fenchone (C10H16O) from LIED data.
• Demonstrated capability without relying on traditional fitting algorithms or ab initio calculations.
• Achieved structural determination for a molecule size not previously possible with other LIED variants.

Conclusions:

• Combining machine learning with ultrafast electron diffraction provides a powerful new avenue for molecular imaging.
• This approach significantly enhances the ability to study complex and large molecules in static and time-resolved experiments.
• Opens new possibilities for investigating intricate molecular dynamics and structures previously inaccessible.