SBM Planted Partition — Kesten-Stigum Threshold

Community recovery in the stochastic block model — detectability phase transition and belief propagation
Graph colored by true (outer ring) vs detected (inner) community
Phase diagram: detectability vs (c_in, c_out) — KS threshold boundary

SBM Parameters

Detectability

KS threshold (c_in-c_out)²
k·c̄ (avg degree)
Detectable?
Overlap (detected)
SNR (c_in-c_out)²/kc̄
(c_in − c_out)² > k · c̄
The Kesten-Stigum threshold marks when community structure becomes statistically detectable.

Below: no algorithm can do better than chance. Above: spectral methods and belief propagation recover communities. The transition is sharp for sparse graphs (n→∞).