{"id":"arxiv/2602.10566","url":"https://arxiv.org/abs/2602.10566","tags":["math.ST","math.CO","math.RA","math.SP"],"score":0,"title":"Finite-sample certification and operating envelopes for spectral clustering and graph centrality","domain":"math.ST","source":"arxiv","authors":["Chandrasekhar Gokavarapu","Sekhar Babu Gosala","Rajeev Muthu","Vamsi Pasalapudi","Tarakarama Kapakayala"],"reviews":0,"subject":"math.ST","summary":"Spectral clustering and node rankings are commonly reported from one observed network without a finite-sample statement of what the observation supports. We develop a certification protocol that either returns a coverage-guaranteed set or explicitly returns ``no nontrivial certificate.'' For an inhomogeneous Bernoulli graph, a matrix-Bernstein quantile with all numerical constants and its ambient-dimension factor retained is combined with a one-sided spectral-gap certificate. The resulting Grassmann ball is valid at finite \\(n\\), but is reported as informative only when its radius is below…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-02-11","reactions":{}}