Category: Network Sci.

Summary: Testing whether targeting antagonistic cross-community edges to intermediate degree shells preserves signed-community recovery better than core-heavy or leaf-heavy placement.

Signed stochastic block models add both positive and negative edges to community structure, making detection sensitive not only to signal strength but also to where antagonistic links are placed. This experiment asks whether concentrating those negative cross-community edges on intermediate degree shells creates a better tradeoff than targeting either the network core or its weakly connected periphery.

The simulation uses batched GPU spectral sweeps near the signed detectability boundary and compares localization of the informative vector with label-recovery quality. The central idea is that core-heavy targeting may over-localize the signal, while leaf-heavy targeting may disperse it into nodes that carry little useful structure.

That turns degree-shell placement into a concrete design variable in signed community detection. The result tests for a genuinely nonmonotone optimum rather than assuming that more concentration on the most important nodes is always better.

Method: Batched GPU spectral analysis of signed stochastic block models across degree-shell targeting schemes for antagonistic cross-community edges.

What is measured: Community recovery, informative-eigenvector localization, signed detectability-boundary behavior, and shell-targeting crossover effects.