Network analysis using Krylov subspace trajectories

Area of Science:

• Network analysis
• Graph theory
• Computational mathematics

Background:

• Power iteration is commonly used for eigenvector centrality, but typically employs random initial vectors and only utilizes the final converged result.
• Intermediate results from power iteration with random vectors lack clear interpretation and have seen limited use in network analysis.
• Existing methods using intermediate power iteration results often focus on single pre-convergence solutions or node similarity.

Purpose of the Study:

• To introduce and explore novel network analysis methods based on Krylov subspace trajectories.
• To leverage intermediate results from power iteration using non-random initial vectors for enhanced network understanding.
• To demonstrate the utility of these trajectories in characterizing network structure, node importance, and response to perturbations.

Main Methods:

• Computation of the Krylov subspace matrix from a network adjacency matrix via power iteration.
• Application of a non-random initial vector in the power iteration process.
• Generation of node-specific Krylov subspace trajectories from the rows of the computed matrix.

Main Results:

• Krylov subspace trajectories derived from non-random initial vectors provide rich information about network properties.
• These trajectories offer insights into network structure, node importance, and system dynamics under perturbation.
• The proposed methods extend the utility of power iteration beyond traditional eigenvector centrality calculations.

Conclusions:

• Krylov subspace trajectories generated with non-random initial vectors offer a powerful new approach to network analysis.
• These methods enhance the understanding of complex networks by utilizing previously overlooked intermediate computational data.
• The framework presented has implications for various fields relying on network science, including systems biology and social network analysis.