A hybrid genetic-Levenberg Marquardt algorithm for automated spectrometer design optimization

Area of Science:

• Physics
• Engineering
• Computational Science

Background:

• Computational tools are increasingly used for automated design and optimization of electron-optical systems.
• Gradient-based methods have been explored, but modern alternatives like genetic algorithms are underutilized.

Purpose of the Study:

• To propose a novel, fully-automated hybrid optimization method for electron-optical systems.
• To combine the efficiency of gradient-based methods with the robustness of genetic algorithms.

Main Methods:

• Developed a hybrid optimization algorithm with an adaptive switching scheme between Levenberg-Marquardt and genetic sub-algorithms.
• Tested the algorithm on two electron-optical systems: parallel cylindrical mirror analyzer and parallel magnetic sector analyzer.

Main Results:

• The hybrid algorithm outperformed individual Levenberg-Marquardt and genetic algorithms in optimizing the test systems.
• Demonstrated simultaneous exploitation of computational efficiency and robustness.

Conclusions:

• The proposed hybrid method is a versatile tool for designing parallel energy spectrometers.
• This approach can significantly aid the development of complex parallel energy analyzers for the semiconductor industry.