Adopt Differential Calculus and Convexity on Euclidean Open Sets: Standing Notation , Euclidean Space and Lebesgue Measure: Standing Notation and Measure Spaces and the Lebesgue Integral: Standing Notation , and fix a dimension q q q . Points of R q \mathbb{R}^{q} R q are read as q q q -tuples by Euclidean Points as Tuples of Real Numbers , the i i i th component of z z z being z i z_{i} z i . Borel is as fixed in the preamble of Pairs of Euclidean Points: Coordinate Projections, Pairings, the Product Measure on a Euclidean Space, Borel Norm Functions and Finite Sets , and for A ∈ B ( R q ) A\in\mathcal{B}(\mathbb{R}^{q}) A ∈ B ( R q ) the indicator 1 A \mathbf{1}_{A} 1 A is the Borel function of claim 1 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions . Let λ q \lambda_{q} λ q be Lebesgue measure on B ( R q ) \mathcal{B}(\mathbb{R}^{q}) B ( R q ) ; integrals with respect to it are those of Measure Spaces and the Lebesgue Integral: Standing Notation §integral , a nonnegative Borel function being integrated as a [ 0 , ∞ ] [0,\infty] [ 0 , ∞ ] -valued map, and ∫ R q f d λ q \int_{\mathbb{R}^{q}}f\,d\lambda_{q} ∫ R q f d λ q is also written ∫ R q f ( z ) d z \int_{\mathbb{R}^{q}}f(z)\,dz ∫ R q f ( z ) d z ; integrable means integrable with respect to λ q \lambda_{q} λ q . Smoothness of a function R q → R \mathbb{R}^{q}\to\mathbb{R} R q → R is that of Smooth Map on a Euclidean Open Set on R q \mathbb{R}^{q} R q , which is open by claim 1 of Euclidean Space is Open in Itself, and C k C^k C k Maps are Continuous ; δ i j \delta_{ij} δ ij is the entry in row i i i and column j j j of the identity matrix I q I_{q} I q ; t \sqrt{t} t is the nonnegative square root of a nonnegative real number t t t ; for a positive real number s s s and a natural number m m m , s − m / 2 s^{-m/2} s − m /2 denotes the multiplicative inverse of the m m m th natural power ( s ) m (\sqrt{s})^{m} ( s ) m , and s − 1 / 2 z s^{-1/2}z s − 1/2 z the scalar multiple of z ∈ R q z\in\mathbb{R}^{q} z ∈ R q by s − 1 / 2 s^{-1/2} s − 1/2 ; and the natural number q q q is read in R \mathbb{R} R as in The Real Numbers: Standing Notation and Background §numbers where a real number is required.
For every real number s s s with 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 let g s : R q → R g_{s}:\mathbb{R}^{q}\to\mathbb{R} g s : R q → R be the Gaussian smoothing weight φ η \varphi_{\eta} φ η of claim 1 of The Gaussian Smoothing Weight: Normalization, Derivatives, Exponential Tilting, Moments, and First-Order Remainder , read in dimension m = q m=q m = q with η = s \eta=s η = s ; that is,
g s ( z ) = c s exp ( − ∥ z ∥ 2 2 s ) , g_{s}(z)=c_{s}\exp\Bigl(-\frac{\lVert z\rVert^{2}}{2s}\Bigr), g s ( z ) = c s exp ( − 2 s ∥ z ∥ 2 ) ,
where c s c_{s} c s is the unique positive real number with ∫ R q g s d λ q = 1 \int_{\mathbb{R}^{q}}g_{s}\,d\lambda_{q}=1 ∫ R q g s d λ q = 1 . Then the following hold.
1. (Scaling) ¶ For every s s s with 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 and every z ∈ R q z\in\mathbb{R}^{q} z ∈ R q ,
c s = c 1 s − q / 2 and g s ( z ) = s − q / 2 g 1 ( s − 1 / 2 z ) . c_{s}=c_{1}\,s^{-q/2}\qquad\text{and}\qquad g_{s}(z)=s^{-q/2}\,g_{1}\bigl(s^{-1/2}z\bigr). c s = c 1 s − q /2 and g s ( z ) = s − q /2 g 1 ( s − 1/2 z ) .
2. (Derivatives) ¶ For every s s s with 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 the function g s g_{s} g s is smooth on R q \mathbb{R}^{q} R q and even, g s ( − z ) = g s ( z ) g_{s}(-z)=g_{s}(z) g s ( − z ) = g s ( z ) , and for all z ∈ R q z\in\mathbb{R}^{q} z ∈ R q and i , j , k ∈ [ q ] i,j,k\in[q] i , j , k ∈ [ q ] ,
∂ i g s ( z ) = − z i s g s ( z ) , ∂ j ∂ i g s ( z ) = ( z i z j s 2 − δ i j s ) g s ( z ) , \partial_{i}g_{s}(z)=-\frac{z_{i}}{s}\,g_{s}(z),\qquad
\partial_{j}\partial_{i}g_{s}(z)=\Bigl(\frac{z_{i}z_{j}}{s^{2}}-\frac{\delta_{ij}}{s}\Bigr)g_{s}(z), ∂ i g s ( z ) = − s z i g s ( z ) , ∂ j ∂ i g s ( z ) = ( s 2 z i z j − s δ ij ) g s ( z ) ,
∂ k ∂ j ∂ i g s ( z ) = ( − z i z j z k s 3 + δ i j z k + δ i k z j + δ j k z i s 2 ) g s ( z ) . \partial_{k}\partial_{j}\partial_{i}g_{s}(z)=\Bigl(-\frac{z_{i}z_{j}z_{k}}{s^{3}}+\frac{\delta_{ij}z_{k}+\delta_{ik}z_{j}+\delta_{jk}z_{i}}{s^{2}}\Bigr)g_{s}(z). ∂ k ∂ j ∂ i g s ( z ) = ( − s 3 z i z j z k + s 2 δ ij z k + δ ik z j + δ jk z i ) g s ( z ) .
3. (Uniform and square-integral bounds) ¶ There are nonnegative real numbers A 0 , A 1 , A 2 , A 3 A_{0},A_{1},A_{2},A_{3} A 0 , A 1 , A 2 , A 3 and B 0 , B 1 , B 2 , B 3 B_{0},B_{1},B_{2},B_{3} B 0 , B 1 , B 2 , B 3 , depending only on q q q , such that for every s s s with 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 , every z ∈ R q z\in\mathbb{R}^{q} z ∈ R q and all i , j , k ∈ [ q ] i,j,k\in[q] i , j , k ∈ [ q ] ,
g s ( z ) ≤ A 0 s − q / 2 , ∣ ∂ i g s ( z ) ∣ ≤ A 1 s − ( q + 1 ) / 2 , ∣ ∂ j ∂ i g s ( z ) ∣ ≤ A 2 s − ( q + 2 ) / 2 , ∣ ∂ k ∂ j ∂ i g s ( z ) ∣ ≤ A 3 s − ( q + 3 ) / 2 , g_{s}(z)\le A_{0}\,s^{-q/2},\qquad |\partial_{i}g_{s}(z)|\le A_{1}\,s^{-(q+1)/2},\qquad |\partial_{j}\partial_{i}g_{s}(z)|\le A_{2}\,s^{-(q+2)/2},\qquad |\partial_{k}\partial_{j}\partial_{i}g_{s}(z)|\le A_{3}\,s^{-(q+3)/2}, g s ( z ) ≤ A 0 s − q /2 , ∣ ∂ i g s ( z ) ∣ ≤ A 1 s − ( q + 1 ) /2 , ∣ ∂ j ∂ i g s ( z ) ∣ ≤ A 2 s − ( q + 2 ) /2 , ∣ ∂ k ∂ j ∂ i g s ( z ) ∣ ≤ A 3 s − ( q + 3 ) /2 ,
and the squares of g s g_{s} g s , ∂ i g s \partial_{i}g_{s} ∂ i g s , ∂ j ∂ i g s \partial_{j}\partial_{i}g_{s} ∂ j ∂ i g s and ∂ k ∂ j ∂ i g s \partial_{k}\partial_{j}\partial_{i}g_{s} ∂ k ∂ j ∂ i g s are integrable, with
∫ R q g s 2 d λ q ≤ B 0 s − q / 2 , ∫ R q ( ∂ i g s ) 2 d λ q ≤ B 1 s − ( q + 2 ) / 2 , \int_{\mathbb{R}^{q}}g_{s}^{2}\,d\lambda_{q}\le B_{0}\,s^{-q/2},\qquad
\int_{\mathbb{R}^{q}}(\partial_{i}g_{s})^{2}\,d\lambda_{q}\le B_{1}\,s^{-(q+2)/2}, ∫ R q g s 2 d λ q ≤ B 0 s − q /2 , ∫ R q ( ∂ i g s ) 2 d λ q ≤ B 1 s − ( q + 2 ) /2 ,
∫ R q ( ∂ j ∂ i g s ) 2 d λ q ≤ B 2 s − ( q + 4 ) / 2 , ∫ R q ( ∂ k ∂ j ∂ i g s ) 2 d λ q ≤ B 3 s − ( q + 6 ) / 2 . \int_{\mathbb{R}^{q}}(\partial_{j}\partial_{i}g_{s})^{2}\,d\lambda_{q}\le B_{2}\,s^{-(q+4)/2},\qquad
\int_{\mathbb{R}^{q}}(\partial_{k}\partial_{j}\partial_{i}g_{s})^{2}\,d\lambda_{q}\le B_{3}\,s^{-(q+6)/2}. ∫ R q ( ∂ j ∂ i g s ) 2 d λ q ≤ B 2 s − ( q + 4 ) /2 , ∫ R q ( ∂ k ∂ j ∂ i g s ) 2 d λ q ≤ B 3 s − ( q + 6 ) /2 .
4. (Convolution identity) ¶ For every s s s with 0 < s ≤ 1 2 0<s\le\tfrac12 0 < s ≤ 2 1 and all x , x ′ ∈ R q x,x'\in\mathbb{R}^{q} x , x ′ ∈ R q , the function y ↦ g s ( y − x ) g s ( y − x ′ ) y\mapsto g_{s}(y-x)\,g_{s}(y-x') y ↦ g s ( y − x ) g s ( y − x ′ ) on R q \mathbb{R}^{q} R q is integrable and
∫ R q g s ( y − x ) g s ( y − x ′ ) d y = g 2 s ( x − x ′ ) . \int_{\mathbb{R}^{q}}g_{s}(y-x)\,g_{s}(y-x')\,dy=g_{2s}(x-x'). ∫ R q g s ( y − x ) g s ( y − x ′ ) d y = g 2 s ( x − x ′ ) .
5. (Moments and tails) ¶ For every s s s with 0 < s ≤ 1 0<s\le1 0 < s ≤ 1 the functions z ↦ ∥ z ∥ 2 g s ( z ) z\mapsto\lVert z\rVert^{2}g_{s}(z) z ↦ ∥ z ∥ 2 g s ( z ) and z ↦ ∥ z ∥ g s ( z ) z\mapsto\lVert z\rVert\,g_{s}(z) z ↦ ∥ z ∥ g s ( z ) are integrable,
∫ R q ∥ z ∥ 2 g s ( z ) d z = q s , ∫ R q ∥ z ∥ g s ( z ) d z ≤ q s , \int_{\mathbb{R}^{q}}\lVert z\rVert^{2}g_{s}(z)\,dz=q\,s,\qquad\int_{\mathbb{R}^{q}}\lVert z\rVert\,g_{s}(z)\,dz\le\sqrt{q\,s}, ∫ R q ∥ z ∥ 2 g s ( z ) d z = q s , ∫ R q ∥ z ∥ g s ( z ) d z ≤ q s ,
and for every positive real number R R R , the set A R = { z ∈ R q : R < ∥ z ∥ } A_{R}=\{z\in\mathbb{R}^{q}:R<\lVert z\rVert\} A R = { z ∈ R q : R < ∥ z ∥} being Borel as the preimage of the Borel set ( R , ∞ ) (R,\infty) ( R , ∞ ) under the Borel map z ↦ ∥ z ∥ z\mapsto\lVert z\rVert z ↦ ∥ z ∥ of Pairs of Euclidean Points: Coordinate Projections, Pairings, the Product Measure on a Euclidean Space, Borel Norm Functions and Finite Sets §functions ,
∫ R q 1 A R g s d λ q ≤ q s R 2 . \int_{\mathbb{R}^{q}}\mathbf{1}_{A_{R}}\,g_{s}\,d\lambda_{q}\le\frac{q\,s}{R^{2}} . ∫ R q 1 A R g s d λ q ≤ R 2 q s .