Gaussian Process Regression Applied to Atom Probe Tomography Data Reconstruction.

Area of Science:

• Materials Science
• Data Science
• Computational Physics

Background:

• Atom probe tomography (APT) is a powerful technique for 3D material characterization at the atomic scale.
• Traditional reconstruction algorithms in APT can be limited by geometric assumptions and may not fully capture complex material behaviors.
• Enhancing APT data reconstruction is crucial for accurate material analysis and discovery.

Purpose of the Study:

• To introduce Gaussian process regression (GPR) as a novel Bayesian inference approach for improving APT data reconstruction.
• To evaluate the performance of GPR with different kernels (radial basis function and Matérn) against standard reconstruction methods.
• To demonstrate the potential of GPR for reconstructing simulated APT data, including surface diffusion effects.

Main Methods:

• Simulated evaporation of single-crystal and polycrystalline specimens was performed.
• Radial basis function and Matérn kernels within a GPR framework were employed for data reconstruction.
• Reconstructed data was compared with results from the standard wide field-of-view reconstruction algorithm.

Main Results:

• Both radial basis function and Matérn kernels in GPR outperformed the traditional geometry-based reconstruction algorithm.
• The Matérn kernel demonstrated particular effectiveness in reconstructing long-range periodic datasets.
• GPR successfully incorporated surface diffusion effects into the reconstruction process.

Conclusions:

• Gaussian process regression offers a flexible and powerful Bayesian approach to enhance atom probe tomography data reconstruction.
• The findings highlight the potential of GPR for more accurate and detailed material analysis using APT.
• This study provides a foundation for applying Bayesian inference methods to experimental APT data.