{"id":"arxiv/2610.15272","url":"https://arxiv.org/abs/2610.15272","tags":["math.ST","math.PR"],"score":0,"title":"Optimal detection of geometry in random graphs and kernel matrices at critical thresholds","domain":"math.ST","source":"arxiv","authors":["Cheng Mao","Timothy L. H. Wee"],"reviews":0,"subject":"math.ST","summary":"Detection of geometry in random graphs and matrices has been extensively studied in recent years. For testing a dense random geometric graph on $n$ vertices with latent points in dimension $d$ against an Erdős--Rényi graph, the detection threshold $d\\asymp n^3$ is well known since Bubeck, Ding, Eldan, and Rácz (2016). We sharpen this result by establishing the optimal asymptotic testing error in the critical regime $d/n^3\\toλ\\in(0,\\infty)$. The log-likelihood ratio is shown to be asymptotically normal and dominated by the signed triangle count. Moreover, we consider testing the rank…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-11","reactions":{}}