Inversion of simulated positron annihilation lifetime spectrum using a neural network

Area of Science:

• Physics
• Computational Chemistry
• Spectroscopy

Background:

• Positron annihilation lifetime spectroscopy (PALS) is a sensitive technique for probing free volume in materials.
• Inverting PALS data to obtain material density functions can be challenging due to inherent data complexities.
• Developing robust computational methods is crucial for accurate PALS data analysis.

Purpose of the Study:

• To present a novel method for inverting positron annihilation lifetime spectroscopy data using a neural network Hopfield model.
• To evaluate the precision and accuracy of the neural network model in reconstructing density functions.
• To investigate the influence of network parameters, such as the number of neurons and learning time, on inversion accuracy.

Main Methods:

• A simulated PALS spectrum was generated from a known lysozyme density function, free from experimental noise and resolution effects.
• A neural network Hopfield model was trained using this simulated spectrum as the exact target data.
• The performance of the trained neural network was assessed by comparing its inverted density function with the original exact function.

Main Results:

• The neural network Hopfield model demonstrated a fair agreement with the exact density function.
• The precision of the inverted density function was found to be dependent on the number of neurons and learning time.
• Specifically, achieving a percentual relative error of 0.4% for the maximum of the density function was possible with 64 neurons.

Conclusions:

• The developed neural network Hopfield model offers a promising approach for the inversion of positron annihilation lifetime spectroscopy data.
• The model's precision is influenced by its architecture and training parameters, highlighting the importance of optimization.
• This method provides a viable pathway for accurate characterization of material properties through PALS data analysis.