The torus heat kernel is twice continuously differentiable, even, positive and bounded with bounded derivatives. Convolving it with a measure gives the smooth positive density of the heat-smoothed measure, whose score is the logarithmic gradient of that density, and this density depends Lipschitz-continuously on the measure in the torus Wasserstein distance.
1. (Regularity and bounds)Θs∈Cper2, Θs(−v)=Θs(v) for every v∈Rd, and ∫QΘsdλd=1. There are real numbers cs and Cs with 0<cs≤Cs such that cs≤Θs(v)≤Cs, ∣∂iΘs(v)∣≤Cs and ∣∂j∂iΘs(v)∣≤Cs for all v∈Rd and i,j∈[d].
2. (Density of a heat-smoothed measure) Let μ∈P(Td). For every y∈Rd the function x↦Θs(y−x) is continuous and bounded by clause 1, so pμ(y)=∫Θs(y−x)μ(dx) is a real number. The function pμ belongs to Cper2, with ∂ipμ(y)=∫∂iΘs(y−x)μ(dx) and cs≤pμ(y)≤Cs for all y∈Rd and i∈[d]; the function 1Qpμ is a density of Ssμ with respect to λd; and Ssμ∈PI(Td), with ξSsμ the class of the map y↦pμ(y)−1∇pμ(y).
3. (Lipschitz dependence on the measure) There is a real number Ls≥0 such that, with pμ and pν as in clause 2,
Curated associations between results. These are editable and subjective — they do not replace the dependency graph, which is derived from the references in the text.