Visualization of Solar Cell Library Space by Dimensionality Reduction Methods.

Area of Science:

• Materials Informatics
• Computational Materials Science
• Renewable Energy Research

Background:

• Dimensionality reduction is widely used in drug design but underexplored in materials informatics.
• Visualizing high-dimensional materials data is crucial for understanding trends and discovering new materials.
• Metal-oxide solar cells are promising for clean energy but their complex materials space remains challenging to navigate.

Purpose of the Study:

• To present the first application of four dimensionality reduction techniques (PCA, kernel PCA, Isomap, diffusion map) to materials informatics.
• To visualize and analyze the materials space of metal-oxide-based solar cells.
• To compare the performance of these techniques in maintaining data structure and identifying material trends.

Main Methods:

• Integrated five metal-oxide solar cell libraries into a unified database.
• Applied four dimensionality reduction techniques: Principal Component Analysis (PCA), Kernel PCA, Isomap, and Diffusion Map.
• Evaluated methods based on preserving local sample environments, clustering structure, and outlier identification.

Main Results:

• Principal Component Analysis (PCA) demonstrated the best performance in maintaining the local environment of samples.
• Isomap proved most effective for classifying materials based on nearest neighbor identities, acting as a superior classifier.
• Identified and rationalized numerous outliers across the tested dimensionality reduction methods.

Conclusions:

• Dimensionality reduction techniques offer valuable tools for analyzing and navigating materials space, particularly for solar cells.
• PCA and Isomap show distinct strengths for different analytical tasks in materials informatics.
• The presented methodology can be extended to other solar cell types and broader materials science applications.