Mathematical analysis demonstrates sharp convergence rates for Gauss–Hermite Chernoff approximations of the heat semigroup, revealing computational trade-offs governed by input smoothness.
For the heat semigroup T(t) = exp(t d²/dx²) on R the q-point Gauss–Hermite Chernoff functions C_q(t)f(x) = Σ_j p_j f(x + 2√t z_j) form a hierarchy of positive contractions with the sharp rate ||C_q(t/n)^n f − T(t)f||_∞ ≤ t^q n−(q−1) ||f^(2q)||_∞ / (2q−1)!!, whose 3-point member coincides with the fast Chernoff function S of Vedenin, Voevodkin, Galkin, Karatetskaya and Remizov. The price of order is the dimension r = floor(q/2) of the shift lattice: after n steps the iterate is an exact atomic measure whose evaluation costs N_supp = Θ(n^r) readings of the datum and whose heaviest atom obeys the proved local-limit law max_k W_k ~ c_q n−r/2 with explicit constants (c_5 = 5/π), verified to 0.5% at n = 1024. A cost-fair benchmark yields a hybrid verdict with an explicit decision rule: C_5 wins on spectrally narrow data at sup-norm targets ≤ 1e−5 (up to 6.7× fewer datum evaluations), C_3 = S wins on broadband and kinked data, and C_4 is never the cost-fair choice in the tested sup-norm benchmarks; in weak (mollified) metrics the full order n−(q−1) returns for arbitrarily rough L¹ data by duality, so the nonsmooth ceiling is a sup-norm phenomenon. Ancillary studies are collected in a supplement; every number cited in either text is checked against the archived data by an automated verifier (152 claims).
No takes yet. Share an insight, caveat, or question.
Sergey V. Shpital (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: