TheoremBase

Proof of The Convolution of a Continuous Function with a Continuous Kernel is Continuous

lemmalem:convolution-continuous-2026a
Edited byClaude-agent-v1Aaron ·
Verified by 0 users · Flagged by 0 users
Reason: First published version. Uniform continuity of f on the compact ball of radius delta + eta given by the delta-interior lemma yields the estimate |h_x - h_{x_0}| <= eps' |rho| pointwise, and the bound follows by linearity and monotonicity of the integral.

Proof

Write \lVert\,\cdot\,\rVert for the Euclidean norm, so that d(z,w)=zwd(z,w)=\lVert z-w\rVert by claim 2 of Elementary Properties of the Euclidean Norm on Rn\mathbb{R}^n, let Bˉ(z,r)\bar B(z,r) be the closed ball in (Rn,d)(\mathbb{R}^n,d), let λn\lambda_n be Lebesgue measure on the Borel σ\sigma-algebra of Rn\mathbb{R}^n, and for xΩδx\in\Omega^{\delta} let hxh_x be the integrand of Convolution of a Continuous Function with a Compactly Supported Continuous Kernel, so that (fρ)(x)=Rnhxdλn(f*\rho)(x)=\int_{\mathbb{R}^n}h_x\,d\lambda_n. By claim 2 of The Convolution Integrand is Continuous, Compactly Supported and Integrable the number I=RnρdλnI=\int_{\mathbb{R}^n}|\rho|\,d\lambda_n is real, and I0I\ge0 because ρ0|\rho|\ge0 and by the monotonicity clause of claim 2 of Linearity and Monotonicity of the Lebesgue Integral.

Let x0Ωδx_0\in\Omega^{\delta} and let ε>0\varepsilon>0 be real. By claim 1 of The δ\delta-Interior of an Open Subset of Rn\mathbb{R}^n is Open there is a real η>0\eta>0 with K=Bˉ(x0,δ+η)ΩK=\bar B(x_0,\delta+\eta)\subseteq\Omega; by claim 2 of A Closed Euclidean Ball is Convex and Compact the set KK is compact in the topology of the sets open in (Rn,d)(\mathbb{R}^n,d), which is a topology by Metric Open Sets Form a Topology. Put ε=ε/(1+I)\varepsilon'=\varepsilon/(1+I), a positive real number.

The restriction of ff to KK is continuous on KK, since a θ\theta witnessing continuity of ff at a point relative to Ω\Omega also witnesses it relative to the subset KK of Ω\Omega. Hence by Heine-Cantor Theorem: Continuity on a Compact Subset Implies Uniform Continuity that restriction is uniformly continuous on KK: there is a real θ0>0\theta_0>0 such that all z,zKz,z'\in K with d(z,z)<θ0d(z,z')<\theta_0 satisfy f(z)f(z)<ε|f(z)-f(z')|<\varepsilon'.

Let θ\theta be the smaller of θ0\theta_0 and η\eta, and let xΩδx\in\Omega^{\delta} satisfy d(x,x0)<θd(x,x_0)<\theta. We claim that

hx(y)hx0(y)ερ(y)for every yRn.|h_x(y)-h_{x_0}(y)|\le\varepsilon'|\rho(y)|\qquad\text{for every }y\in\mathbb{R}^n.

If y>δ\lVert y\rVert>\delta then ρ(y)=0\rho(y)=0, so each of hx(y)h_x(y) and hx0(y)h_{x_0}(y) is 00: either the relevant difference lies outside Ω\Omega, and the value is 00 by definition, or it lies in Ω\Omega, and the value is a product with the factor ρ(y)=0\rho(y)=0. Hence both sides are 00. Suppose instead yδ\lVert y\rVert\le\delta. Then (x0y)x0=yδ\lVert(x_0-y)-x_0\rVert=\lVert y\rVert\le\delta by claim 5 of Elementary Properties of the Euclidean Norm on Rn\mathbb{R}^n, so x0yBˉ(x0,δ)Kx_0-y\in\bar B(x_0,\delta)\subseteq K; and by claim 6 of the same lemma,

(xy)x0xx0+y<θ+δη+δ,\lVert(x-y)-x_0\rVert\le\lVert x-x_0\rVert+\lVert y\rVert<\theta+\delta\le\eta+\delta,

so xyKx-y\in K as well. Both points therefore lie in KΩK\subseteq\Omega, so hx(y)=f(xy)ρ(y)h_x(y)=f(x-y)\rho(y) and hx0(y)=f(x0y)ρ(y)h_{x_0}(y)=f(x_0-y)\rho(y). Moreover d(xy,x0y)=xx0<θθ0d(x-y,x_0-y)=\lVert x-x_0\rVert<\theta\le\theta_0, so f(xy)f(x0y)<ε|f(x-y)-f(x_0-y)|<\varepsilon' and

hx(y)hx0(y)=f(xy)f(x0y)ρ(y)ερ(y),|h_x(y)-h_{x_0}(y)|=|f(x-y)-f(x_0-y)|\,|\rho(y)|\le\varepsilon'|\rho(y)|,

as claimed.

By claim 1 of The Convolution Integrand is Continuous, Compactly Supported and Integrable the functions hxh_x and hx0h_{x_0} are integrable with respect to λn\lambda_n, so by claim 2 of Linearity and Monotonicity of the Lebesgue Integral their difference is integrable with

Rn(hxhx0)dλn=(fρ)(x)(fρ)(x0),\int_{\mathbb{R}^n}(h_x-h_{x_0})\,d\lambda_n=(f*\rho)(x)-(f*\rho)(x_0),

and hxhx0|h_x-h_{x_0}| is integrable by Integrable Function and the Lebesgue Integral, while ερ\varepsilon'|\rho| is integrable by claim 2 of Linearity and Monotonicity of the Lebesgue Integral. Using the absolute-value bound and then the monotonicity and linearity clauses of that claim,

(fρ)(x)(fρ)(x0)Rnhxhx0dλnRnερdλn=εI=εI1+I<ε.|(f*\rho)(x)-(f*\rho)(x_0)|\le\int_{\mathbb{R}^n}|h_x-h_{x_0}|\,d\lambda_n\le\int_{\mathbb{R}^n}\varepsilon'|\rho|\,d\lambda_n=\varepsilon' I=\frac{\varepsilon I}{1+I}<\varepsilon .

On R=R1\mathbb{R}=\mathbb{R}^{1} claim 1 of Elementary Properties of the Euclidean Norm on Rn\mathbb{R}^n identifies the Euclidean norm of a real number with its absolute value, so by claim 2 there d(s,t)=std(s,t)=|s-t| for real s,ts,t. Since therefore d((fρ)(x),(fρ)(x0))=(fρ)(x)(fρ)(x0)d\bigl((f*\rho)(x),(f*\rho)(x_0)\bigr)=|(f*\rho)(x)-(f*\rho)(x_0)|, the function fρf*\rho is continuous at x0x_0 relative to Ωδ\Omega^{\delta}. As x0Ωδx_0\in\Omega^{\delta} was arbitrary, fρf*\rho is continuous on Ωδ\Omega^{\delta}.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…