Explain why the community-structure eigenvalue is real-dominant with near-zero imaginary part

Characterize and prove the theoretical reasons under which, for directed graphs, the eigenvalue associated with the eigenvector encoding cluster structure for the complex non-backtracking matrix B_alpha has the largest real part and an almost zero imaginary part.

Background

Empirically, the authors observe that the eigenvalue tied to community-structure eigenvectors tends to have the largest real part and nearly zero imaginary part, which underlies the effectiveness of their spectral method.

A theoretical explanation would clarify the spectral signatures of community structure in CNBT, potentially yielding diagnostic metrics for cluster strength and stability.

References

Furthermore, while it is empirically known that the eigenvalue corresponding to the eigenvector with cluster structure has the largest real part and an almost zero imaginary part, the underlying connection remains unclear.

Complex non-backtracking matrix for directed graphs (2507.12503 - Sando et al., 16 Jul 2025) in Section 5 (Conclusion)