{"id":"arxiv/2606.07931","url":"https://arxiv.org/abs/2606.07931","tags":["math.PR","cond-mat.stat-mech","cs.IT","cs.LG","math.ST"],"score":0,"title":"Pointwise Complexity for Gaussian Fields: Upper Envelopes, Algorithmic Lower Bounds, and Separation","domain":"math.PR","source":"arxiv","authors":["Yunbei Xu"],"reviews":0,"subject":"math.PR","summary":"We prove a variance-aware pointwise majorizing-measure theorem for centered Gaussian processes. Classical generic chaining characterizes the scalar quantity $\\mathbb E\\sup_{x\\in T}X_x$; the theorem here gives a simultaneous high-probability envelope for the entire field. For an ambient prior $μ$, the envelope at $x$ is governed by a pointwise Fernique-Talagrand functional \\[Φ_μ(x):=\\int_0^{4σ(x)}\\sqrt{\\log\\frac{1}{μ(B_d(x,\\varepsilon))}}\\,d\\varepsilon,\\] together with the corresponding Gaussian tail term. The theorem provides a reusable field-level refinement of classical generic chaining…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-06","reactions":{}}