TheoremBase

Proof of Transfer of a Partial Derivative in Periodic Convolution

lemmalem:periodic-convolution-derivative-torus-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 6,035 chars · 10 deps · depth 27 Reason: First publication: integration by parts on the torus against the periodised kernel, combined with the derivative of the periodised kernel, and uniform continuity of the function and of its first derivatives.

The periodised kernel turns each convolution into an integral over the cell; integration by parts on the torus moves the derivative onto the kernel, where differentiating the periodised kernel returns the periodised kernel of the derivative. The uniform statement then follows from uniform continuity of the function and of its first derivatives.

Proof

Each result cited below is universally quantified over the data appearing in its own statement, and is applied here to the data named in the statement of this lemma. Integrals over Tn\mathbb{T}^{n} are written with the integration variable named yy, and for wCperw\in C_{\mathrm{per}} we use the abbreviation Tnwdy=TnwQdy\int_{\mathbb{T}^{n}}w\,dy=\int_{\mathbb{T}^{n}}w|_{Q}\,dy of Vanishing Mean of a Derivative and Integration by Parts on the Torus.

Step 1. Proof of claim 1.

Since ψ\psi is smooth it is of class C1C^{1} on Rn\mathbb{R}^{n} by claim 2 of Euclidean Space is Open in Itself, and CkC^k Maps are Continuous, so iψ\partial_{i}\psi is continuous and vanishes at every yy with R<yR<\lVert y\rVert, as recorded in claim 1; thus iψ\partial_{i}\psi satisfies the hypotheses placed on a kernel in The Periodised Kernel of a Periodic Convolution, as does ψ\psi. Let Ψ\Psi be the periodised kernel of ψ\psi and let Ψ(i)\Psi^{(i)} be that of iψ\partial_{i}\psi; for xRnx\in\mathbb{R}^{n} let Ψx\Psi_{x} and Ψx(i)\Psi^{(i)}_{x} denote the restrictions to QQ of Ψ(x,)\Psi(x,\cdot) and Ψ(i)(x,)\Psi^{(i)}(x,\cdot).

Fix xRnx\in\mathbb{R}^{n}. By The Periodised Kernel of a Periodic Convolution §regularity the map Ψ(x,)\Psi(x,\cdot) lies in CperC^{\infty}_{\mathrm{per}}, hence in Cper1C^{1}_{\mathrm{per}}, a smooth map on Rn\mathbb{R}^{n} being of class C1C^{1} by claim 2 of Euclidean Space is Open in Itself, and CkC^k Maps are Continuous and periodicity being the same condition for the two classes by Lattice-Periodic Functions and the Periodic Function Classes §classes. By The Periodised Kernel of a Periodic Convolution §representation, applied to the kernel ψ\psi and the function (iφ)Q(\partial_{i}\varphi)|_{Q},

(ψ((iφ)Q))(x)=TnΨx(iφ)Qdy=TnΨ(x,)(iφ)dy,\bigl(\psi\star((\partial_{i}\varphi)|_{Q})\bigr)(x)=\int_{\mathbb{T}^{n}}\Psi_{x}\,(\partial_{i}\varphi)|_{Q}\,dy=\int_{\mathbb{T}^{n}}\Psi(x,\cdot)\,(\partial_{i}\varphi)\,dy ,

the second equality because the restriction of a pointwise product is the product of the restrictions. Both Ψ(x,)\Psi(x,\cdot) and φ\varphi lie in Cper1C^{1}_{\mathrm{per}}, so Vanishing Mean of a Derivative and Integration by Parts on the Torus §parts, applied with g=Ψ(x,)g=\Psi(x,\cdot) and h=φh=\varphi, gives

Tn(iΨ(x,))φdy=TnΨ(x,)(iφ)dy.\int_{\mathbb{T}^{n}}\bigl(\partial_{i}\Psi(x,\cdot)\bigr)\,\varphi\,dy=-\int_{\mathbb{T}^{n}}\Psi(x,\cdot)\,(\partial_{i}\varphi)\,dy .

By The Periodised Kernel of a Periodic Convolution §derivative, iΨ(x,)=Ψ(i)(x,)\partial_{i}\Psi(x,\cdot)=-\Psi^{(i)}(x,\cdot) as maps on Rn\mathbb{R}^{n}, so the left-hand side equals TnΨ(i)(x,)φdy-\int_{\mathbb{T}^{n}}\Psi^{(i)}(x,\cdot)\,\varphi\,dy by claim 2 of Linearity and Monotonicity of the Lebesgue Integral, used with the scalar 1-1. Combining the two displays and cancelling the common sign,

(ψ((iφ)Q))(x)=TnΨx(i)φQdy=((iψ)(φQ))(x),\bigl(\psi\star((\partial_{i}\varphi)|_{Q})\bigr)(x)=\int_{\mathbb{T}^{n}}\Psi^{(i)}_{x}\,\varphi|_{Q}\,dy=\bigl((\partial_{i}\psi)\star(\varphi|_{Q})\bigr)(x),

the second equality by The Periodised Kernel of a Periodic Convolution §representation applied to the kernel iψ\partial_{i}\psi and the function φQ\varphi|_{Q}. Finally, ψ\psi is of class CkC^{k} on Rn\mathbb{R}^{n} for every natural number kk by Smooth Map on a Euclidean Open Set, so Properties of Periodic Convolution on the Torus §smooth, applied with the kernel ψ\psi and the function φQ\varphi|_{Q}, gives i(ψ(φQ))=(iψ)(φQ)\partial_{i}(\psi\star(\varphi|_{Q}))=(\partial_{i}\psi)\star(\varphi|_{Q}). As xx was arbitrary, claim 1 follows.

Step 2. Proof of claim 2.

Let η\eta be a real number with 0<η0<\eta. The maps φ\varphi and 1φ,,nφ\partial_{1}\varphi,\dots,\partial_{n}\varphi all lie in CperC_{\mathrm{per}}, so A Continuous Lattice-Periodic Function is Uniformly Continuous §uniform, applied to each of them with the tolerance η\eta, provides positive real numbers r0,r1,,rnr_{0},r_{1},\dots,r_{n} such that φ(x)φ(z)η|\varphi(x)-\varphi(z)|\le\eta for all x,zx,z with xzr0\lVert x-z\rVert\le r_{0}, and iφ(x)iφ(z)η|\partial_{i}\varphi(x)-\partial_{i}\varphi(z)|\le\eta for all x,zx,z with xzri\lVert x-z\rVert\le r_{i}, for each i[n]i\in[n]. Let rr be the least of r0,r1,,rnr_{0},r_{1},\dots,r_{n}; it is a positive real number not exceeding any of them, by claim 9 of Elementary Order Arithmetic in an Ordered Field for two numbers together with an induction over the finitely many indices. Put ε0=rδ1\varepsilon_{0}=r\,\delta^{-1}, a positive real number by claims 7 and 5 of Elementary Order Arithmetic in an Ordered Field.

Let ε\varepsilon be a real number with 0<ε<ε00<\varepsilon<\varepsilon_{0}. Multiplying by the positive factor δ\delta, claim 10 of Elementary Order Arithmetic in an Ordered Field gives εδ<ε0δ=r\varepsilon\delta<\varepsilon_{0}\delta=r. Consequently any pair x,zx,z with xzεδ\lVert x-z\rVert\le\varepsilon\delta also satisfies xzr\lVert x-z\rVert\le r, hence xzr0\lVert x-z\rVert\le r_{0} and xzri\lVert x-z\rVert\le r_{i} for every ii, so the estimates of the previous paragraph hold at the scale εδ\varepsilon\delta.

The rescaled kernel ρε\rho_{\varepsilon} is a mollifier kernel of radius εδ\varepsilon\delta, hence smooth and vanishing at every yy with εδ<y\varepsilon\delta<\lVert y\rVert by Mollifier Kernel of Radius δ\delta on Rn\mathbb{R}^n; so the hypotheses placed on ψ\psi in this lemma hold for ρε\rho_{\varepsilon} with the radius εδ\varepsilon\delta in place of RR, and claim 1 applies to it. By Properties of Periodic Convolution on the Torus §uniform, applied with the mollifier kernel ρε\rho_{\varepsilon} of radius εδ\varepsilon\delta, the function w=φw=\varphi and the bound η\eta,

(ρε(φQ))(x)φ(x)ηfor every xRn,\bigl|\bigl(\rho_{\varepsilon}\star(\varphi|_{Q})\bigr)(x)-\varphi(x)\bigr|\le\eta\qquad\text{for every }x\in\mathbb{R}^{n},

which is the first estimate of claim 2. Applying the same claim with w=iφw=\partial_{i}\varphi gives

(ρε((iφ)Q))(x)iφ(x)ηfor every xRn,\bigl|\bigl(\rho_{\varepsilon}\star((\partial_{i}\varphi)|_{Q})\bigr)(x)-\partial_{i}\varphi(x)\bigr|\le\eta\qquad\text{for every }x\in\mathbb{R}^{n},

and claim 1, applied with the kernel ρε\rho_{\varepsilon}, identifies ρε((iφ)Q)\rho_{\varepsilon}\star((\partial_{i}\varphi)|_{Q}) with i(ρε(φQ))\partial_{i}(\rho_{\varepsilon}\star(\varphi|_{Q})). This is the second estimate of claim 2, and ε0\varepsilon_{0} was chosen independently of xx and ii. \blacksquare

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…