TheoremBase

Proof of The One-Dimensional Slice Bound for a Continuously Differentiable Periodic Function

lemmalem:one-dimensional-sup-bound-torus-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 10,247 chars · 20 deps · depth 25 Reason: First publication: proof of the one-dimensional slice bound, by the fundamental theorem of calculus on a slice, an increment bound, and averaging over one period.

The slice map is continuous and differentiable with the partial derivative as its derivative, so the fundamental theorem of calculus bounds the increment of the slice by the integral of the derivative; averaging over one period in that coordinate gives the bound, and periodicity reduces a general point to the period.

Proof

Each result cited is universally quantified over the data in its own statement, and is applied here to the data named below. We use silently that the order of R\mathbb{R} is reflexive, transitive and antisymmetric, that equal real numbers satisfy \le in both directions, and that x<yx<y entails xyx\le y.

Let xRnx\in\mathbb{R}^{n} be fixed until the end of the proof and abbreviate x[s]=x[i:s]x[s]=x[i{:}s] for sRs\in\mathbb{R}. Write

θx(s)=Θi(x[s])(sR),\theta_{x}(s)=\Theta_{i}(x[s])\qquad(s\in\mathbb{R}),

let θ~x=(Θi)~i,x\tilde{\theta}_{x}=\widetilde{(\Theta_{i})}_{i,x} be the zero extension of θx\theta_{x} taking the value θx(s)\theta_{x}(s) at s[0,1]s\in[0,1] and the value 00 at every sRs\in\mathbb{R} with s[0,1]s\notin[0,1], and put

F=(PiΘi)(x)=Rθ~xdλ.F=(P_{i}\Theta_{i})(x)=\int_{\mathbb{R}}\tilde{\theta}_{x}\,d\lambda .

By The Slice Average of a Continuous Periodic Function §defined, applied to ΘiCper\Theta_{i}\in C_{\mathrm{per}}, the map θ~x\tilde{\theta}_{x} is measurable with respect to B(R)\mathcal{B}(\mathbb{R}) and integrable with respect to λ\lambda, and 0F0\le F, the map Θi\Theta_{i} taking nonnegative values. The assertion to be proved is u(x)F|u(x)|\le F.

Since uCper1u\in C^{1}_{\mathrm{per}}, the map uu is of class C1C^{1} on Rn\mathbb{R}^{n} and Zn\mathbb{Z}^{n}-periodic, and uCperu\in C_{\mathrm{per}} and iuCper\partial_{i}u\in C_{\mathrm{per}}: a map of class C1C^{1} is continuous by claim 3 of Euclidean Space is Open in Itself, and CkC^k Maps are Continuous, periodicity is the same condition for the classes concerned by Lattice-Periodic Functions and the Periodic Function Classes §classes, and iuCper\partial_{i}u\in C_{\mathrm{per}} by Elementary Properties of Lattice-Periodic Functions §derivative. By The Slice Average of a Continuous Periodic Function §closure the maps u|u| and iu|\partial_{i}u| also lie in CperC_{\mathrm{per}}. Let p,q:RRp,q:\mathbb{R}\to\mathbb{R} denote the maps

p(s)=u(x[s]),q(s)=iu(x[s])(sR),p(s)=u(x[s]),\qquad q(s)=\partial_{i}u(x[s])\qquad(s\in\mathbb{R}),

so that θx(s)=p(s)+q(s)\theta_{x}(s)=|p(s)|+|q(s)|. By The Slice Average of a Continuous Periodic Function §defined, applied in turn to the members uu, iu\partial_{i}u, u|u| and iu|\partial_{i}u| of CperC_{\mathrm{per}}, the maps pp, qq, p|p| and q|q| are continuous from (R,dR)(\mathbb{R},d_{\mathbb{R}}) to (R,dR)(\mathbb{R},d_{\mathbb{R}}), and the zero extension off [0,1][0,1] of each of them is measurable with respect to B(R)\mathcal{B}(\mathbb{R}) and integrable with respect to λ\lambda; the zero extensions of p|p| and q|q| take nonnegative values, so their integrals are nonnegative by the monotonicity in claim 2 of Linearity and Monotonicity of the Lebesgue Integral, the map with constant value 00 being integrable with integral 00 by that same claim applied with both coefficients 00.

(P1) The slice is differentiable with derivative qq. Let a,bRa,b\in\mathbb{R} with a<ba<b and let tRt\in\mathbb{R} with a<t<ba<t<b. Then the restriction of pp to [a,b][a,b] is differentiable at tt with derivative q(t)q(t).

Indeed, apply Slice Function and the Partial Derivative with the open set Rn\mathbb{R}^{n}, which is open in itself by claim 1 of Euclidean Space is Open in Itself, and CkC^k Maps are Continuous, the map uu, the point a=x[t]a^{\ast}=x[t] whose iith coordinate is tt, and the index ii. Claim 1 of that lemma provides a real ρ>0\rho>0 with a[s]Rna^{\ast}[s]\in\mathbb{R}^{n} for st<ρ|s-t|<\rho, and claim 2 provides the interval I={sR:tρ<s and s<t+ρ}I=\{s\in\mathbb{R}:t-\rho<s\text{ and }s<t+\rho\}, with tt an interior point of II, and the slice function g:IRg:I\to\mathbb{R}, g(s)=u(a[s])g(s)=u(a^{\ast}[s]). Since a=x[t]a^{\ast}=x[t] has the same kkth coordinate as xx for every kik\ne i, we have a[s]=x[s]a^{\ast}[s]=x[s] and hence g(s)=p(s)g(s)=p(s) for every sIs\in I. The partial derivative of uu with respect to the iith variable exists at aa^{\ast}, since uu is of class C1C^{1} on Rn\mathbb{R}^{n} and that definition requires exactly the existence of the first-order partial derivatives at every point, so claim 2 of that lemma gives that gg is differentiable at tt with g(t)=iu(a)=q(t)g'(t)=\partial_{i}u(a^{\ast})=q(t).

It remains to transfer this to the restriction PP of pp to [a,b][a,b], for which tt is an interior point since a<t<ba<t<b. Let ε\varepsilon be a real number with 0<ε0<\varepsilon. By Derivative at an Interior Point applied to gg there is a real δ>0\delta>0 such that every hRh\in\mathbb{R} with 0<h<δ0<|h|<\delta and t+hIt+h\in I satisfies

g(t+h)g(t)hq(t)<ε.\Bigl|\frac{g(t+h)-g(t)}{h}-q(t)\Bigr|<\varepsilon .

Put δ=min{δ,ρ}\delta'=\min\{\delta,\rho\}, a positive real by claim 9 of Elementary Order Arithmetic in an Ordered Field. Let hRh\in\mathbb{R} with 0<h<δ0<|h|<\delta' and t+h[a,b]t+h\in[a,b]. Then h<ρ|h|<\rho, so t+hIt+h\in I, and h<δ|h|<\delta; moreover P(t+h)=p(t+h)=g(t+h)P(t+h)=p(t+h)=g(t+h) and P(t)=g(t)P(t)=g(t). Hence the displayed inequality holds with PP in place of gg. As ε\varepsilon was arbitrary, PP is differentiable at tt with derivative q(t)q(t), by Derivative at an Interior Point.

(P2) The increment bound. Put

M=Rκ~dλ,M=\int_{\mathbb{R}}\tilde{\kappa}\,d\lambda ,

where κ~\tilde{\kappa} is the map taking the value q(s)|q(s)| at s[0,1]s\in[0,1] and 00 elsewhere; it is the zero extension off [0,1][0,1] of q|q|, so it is measurable and integrable, and 0M0\le M, by the opening paragraphs of this proof. Then

p(t)p(s)Mfor all s,t[0,1].|p(t)-p(s)|\le M\qquad\text{for all }s,t\in[0,1].

To see this, let s,t[0,1]s,t\in[0,1]. If s=ts=t the left-hand side is 00 and 0M0\le M, as just noted. Otherwise let aa be the smaller and bb the larger of ss and tt, so a<ba<b and [a,b][0,1][a,b]\subseteq[0,1]. The restriction of qq to [a,b][a,b] is continuous, since for a point of [a,b][a,b] and a real ε>0\varepsilon>0 any δ\delta witnessing the continuity of qq at that point also witnesses that of the restriction, the requirement being imposed at fewer points; hence Riemann integrable on [a,b][a,b] by Continuous Functions on Compact Intervals are Riemann Integrable. The restriction PP of pp to [a,b][a,b] is continuous by the same argument, and by (P1) it is differentiable at every tt' with a<t<ba<t'<b, with derivative q(t)q(t'). So Fundamental Theorem of Calculus, Part II, on a Closed Real Interval applies and gives

abq(r)dr=P(b)P(a)=p(b)p(a).\int_{a}^{b}q(r)\,dr=P(b)-P(a)=p(b)-p(a).

By claim 3 of Agreement of the Riemann and Lebesgue Integrals for Continuous Functions on a Closed Interval, applied to the restriction of qq to [a,b][a,b], the Riemann integral on the left equals Rq~[a,b]dλ\int_{\mathbb{R}}\tilde{q}_{[a,b]}\,d\lambda, where q~[a,b]\tilde{q}_{[a,b]} takes the value q(r)q(r) on [a,b][a,b] and 00 elsewhere. Hence, using claim 2 of Linearity and Monotonicity of the Lebesgue Integral,

p(b)p(a)=Rq~[a,b]dλRq~[a,b]dλ.|p(b)-p(a)|=\Bigl|\int_{\mathbb{R}}\tilde{q}_{[a,b]}\,d\lambda\Bigr|\le\int_{\mathbb{R}}\bigl|\tilde{q}_{[a,b]}\bigr|\,d\lambda .

Now q~[a,b](r)κ~(r)\bigl|\tilde{q}_{[a,b]}(r)\bigr|\le\tilde{\kappa}(r) for every rRr\in\mathbb{R}: for r[a,b]r\in[a,b] both sides equal q(r)|q(r)|, since [a,b][0,1][a,b]\subseteq[0,1]; for r[a,b]r\notin[a,b] the left-hand side is 0=0|0|=0 and the right-hand side is nonnegative. The map q~[a,b]\bigl|\tilde{q}_{[a,b]}\bigr| is measurable by claim 4 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions and integrable, being dominated by the integrable κ~\tilde{\kappa}: this follows from claim 1 of Linearity and Monotonicity of the Lebesgue Integral together with the criterion in Integrable Function and the Lebesgue Integral. So the monotonicity in claim 2 of Linearity and Monotonicity of the Lebesgue Integral gives Rq~[a,b]dλM\int_{\mathbb{R}}\bigl|\tilde{q}_{[a,b]}\bigr|\,d\lambda\le M. Since p(t)p(s)=p(b)p(a)|p(t)-p(s)|=|p(b)-p(a)| by claim 2 of Properties of the Absolute Value in an Ordered Field, the bound follows.

(P3) Averaging over one period. Let t[0,1]t\in[0,1]. Then p(t)F|p(t)|\le F.

Let s[0,1]s\in[0,1]. By claim 5 of Properties of the Absolute Value in an Ordered Field and (P2),

p(t)=p(s)+(p(t)p(s))p(s)+p(t)p(s)p(s)+M.|p(t)|=|p(s)+(p(t)-p(s))|\le|p(s)|+|p(t)-p(s)|\le|p(s)|+M .

Let π~\tilde{\pi} be the map taking the value p(s)|p(s)| at s[0,1]s\in[0,1] and 00 elsewhere, measurable and integrable as above, and let cc denote the real number p(t)|p(t)|. Let 1[0,1]\mathbf{1}_{[0,1]} be the indicator of [0,1][0,1], which lies in B(R)\mathcal{B}(\mathbb{R}) and satisfies λ([0,1])=1\lambda([0,1])=1 by claim 4 of Existence of Lebesgue Measure on the Real Line. It is measurable by claim 1 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions, and R1[0,1]dλ=λ([0,1])=1\int_{\mathbb{R}}\mathbf{1}_{[0,1]}\,d\lambda=\lambda([0,1])=1 by The Integral of an Indicator Function is the Measure of the Set. Since 1[0,1]\mathbf{1}_{[0,1]} takes only the values 00 and 11, which are nonnegative, 1[0,1]=1[0,1]|\mathbf{1}_{[0,1]}|=\mathbf{1}_{[0,1]} by claim 1 of Properties of the Absolute Value in an Ordered Field together with the sign of those values, so R1[0,1]dλ=1\int_{\mathbb{R}}|\mathbf{1}_{[0,1]}|\,d\lambda=1 is finite and 1[0,1]\mathbf{1}_{[0,1]} is integrable by the criterion recorded in Integrable Function and the Lebesgue Integral.

The displayed inequality says that c1[0,1](s)π~(s)+M1[0,1](s)c\,\mathbf{1}_{[0,1]}(s)\le\tilde{\pi}(s)+M\,\mathbf{1}_{[0,1]}(s) for every sRs\in\mathbb{R}: for s[0,1]s\in[0,1] this is the inequality just proved, and for s[0,1]s\notin[0,1] both sides are 00, the right-hand side because π~(s)=0\tilde\pi(s)=0. Both sides are integrable by claim 2 of Linearity and Monotonicity of the Lebesgue Integral, so the monotonicity and linearity in that claim give

c=c1=Rc1[0,1]dλRπ~dλ+MR1[0,1]dλ=Rπ~dλ+M.c=c\cdot 1=\int_{\mathbb{R}}c\,\mathbf{1}_{[0,1]}\,d\lambda\le\int_{\mathbb{R}}\tilde{\pi}\,d\lambda+M\int_{\mathbb{R}}\mathbf{1}_{[0,1]}\,d\lambda=\int_{\mathbb{R}}\tilde{\pi}\,d\lambda+M .

Finally π~(s)+κ~(s)=θ~x(s)\tilde{\pi}(s)+\tilde{\kappa}(s)=\tilde{\theta}_{x}(s) for every sRs\in\mathbb{R}, both sides being p(s)+q(s)|p(s)|+|q(s)| on [0,1][0,1] and 00 elsewhere, so claim 2 of Linearity and Monotonicity of the Lebesgue Integral gives Rπ~dλ+M=F\int_{\mathbb{R}}\tilde{\pi}\,d\lambda+M=F. Hence p(t)F|p(t)|\le F.

Proof of claim 1. By Existence and Uniqueness of the Integer Part of a Real Number there is an integer mm with mxim\le x_{i} and xi<m+1x_{i}<m+1; put t=ximt=x_{i}-m, so that 0t0\le t and t<1t<1, whence t[0,1]t\in[0,1]. Let ee be the point of Rn\mathbb{R}^{n} whose iith coordinate is mm and whose other coordinates are 00; then eZne\in\mathbb{Z}^{n}, the integer lattice, and x[t]+e=x[t+m]=x[xi]=xx[t]+e=x[t+m]=x[x_{i}]=x. Since uu is Zn\mathbb{Z}^{n}-periodic,

u(x)=u(x[t]+e)=u(x[t])=p(t).u(x)=u(x[t]+e)=u(x[t])=p(t).

By (P3), u(x)=p(t)F|u(x)|=|p(t)|\le F. Since xRnx\in\mathbb{R}^{n} was arbitrary, claim 1 is proved.

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Prerequisites

Loading...

Comments

Loading…