TheoremBase

Powers of the function are fed into the Gagliardo-Nirenberg inequality, with the resulting integrals estimated by Hoelder's inequality, and the common factor is cancelled; the constant five works in each of the three dimensions. The bound passes to the Sobolev space by mollification, an almost-everywhere convergent subsequence and Fatou's lemma.

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, and that equal real numbers satisfy ≤\le in both directions. Powers of nonnegative real numbers and their properties are those of Measurable Functions and the Lebesgue Integral: Standing Notation (vii); for a measurable ff on (Q,BQ,λQ)(Q,\mathcal{B}_{Q},\lambda_{Q}) and a real number rr with 0<r0<r, ∣f∣r|f|^{r} denotes the map y↦(∣f(y)∣)ry\mapsto(|f(y)|)^{r} of Power-Integrable Functions and the p-Seminorm (i).

Proof of claim 1. Let u∈Cper1u\in C^{1}_{\mathrm{per}}. By Elementary Properties of the Weak Partial Derivative on the Torus (i), u∈Cperu\in C_{\mathrm{per}} and ∂ju∈Cper\partial_{j}u\in C_{\mathrm{per}} for every j∈[n]j\in[n], the restrictions u∣Qu|_{Q} and (∂ju)∣Q(\partial_{j}u)|_{Q} lie in Lt(Tn)\mathcal{L}^{t}(\mathbb{T}^{n}) for every real tt with 1≤t1\le t, and [(∂ju)∣Q][(\partial_{j}u)|_{Q}] is the jj-th weak partial derivative of [u∣Q][u|_{Q}]; so [u∣Q]∈H1(Tn)[u|_{Q}]\in H^{1}(\mathbb{T}^{n}) by The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (i). Write

N=∥[u∣Q]∥H1,S=∥u∣Q∥6,N=\bigl\lVert[u|_{Q}]\bigr\rVert_{H^{1}},\qquad S=\lVert u|_{Q}\rVert_{6},

both nonnegative real numbers, the first by The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (ii) and the second by Power-Integrable Functions and the p-Seminorm (iii).

(R1) The two elementary bounds. By The Sobolev Space on the Torus is a Separable Hilbert Space Densely Embedded in the Square-Integrable Space (i), applied to U=[u∣Q]U=[u|_{Q}], and by The Lebesgue Space Lᵖ(μ) (iii), which identifies the norm of a class with the seminorm of a representative,

∥u∣Q∥2≤N,∥(∂ju)∣Q∥2≤N(j∈[n]),\lVert u|_{Q}\rVert_{2}\le N,\qquad\bigl\lVert(\partial_{j}u)|_{Q}\bigr\rVert_{2}\le N\qquad(j\in[n]),

the norm ∥ ⋅ ∥L2\lVert\,\cdot\,\rVert_{L^{2}} of that theorem being the norm of L2(Tn)L^{2}(\mathbb{T}^{n}) fixed in The Flat Torus: Standing Notation (iv) and the jj-th weak partial derivative of [u∣Q][u|_{Q}] being [(∂ju)∣Q][(\partial_{j}u)|_{Q}].

(R2) Powers of uu. For k∈{2,3,4}k\in\{2,3,4\} let uku^{k} denote the pointwise product of kk copies of uu, formed as u2=uuu^{2}=uu, u3=u2uu^{3}=u^{2}u and u4=u3uu^{4}=u^{3}u. By claim 3 of Constants, Affine Functions, Sums, Products and Quotients of Cᵏ Functions and Elementary Properties of Lattice-Periodic Functions (i), each uku^{k} lies in Cper1C^{1}_{\mathrm{per}}, and by the pointwise product rule of claim 1 of Constants, Affine Functions, Sums, Products and Quotients of Cᵏ Functions,

∂j(u2)=2u ∂ju,∂j(u3)=3u2 ∂ju,∂j(u4)=4u3 ∂ju(j∈[n]),\partial_{j}(u^{2})=2u\,\partial_{j}u,\qquad\partial_{j}(u^{3})=3u^{2}\,\partial_{j}u,\qquad\partial_{j}(u^{4})=4u^{3}\,\partial_{j}u\qquad(j\in[n]),

each obtained from the previous one, since for instance ∂j(u3u)=(3u2∂ju)u+u3∂ju=4u3∂ju\partial_{j}(u^{3}u)=\bigl(3u^{2}\partial_{j}u\bigr)u+u^{3}\partial_{j}u=4u^{3}\partial_{j}u. Moreover ∣uk(y)∣=(∣u(y)∣)k|u^{k}(y)|=(|u(y)|)^{k} for every yy, by claim 4 of Properties of the Absolute Value in an Ordered Field together with Properties of Real Powers (iv) and Properties of Real Powers (ii), which give (∣u(y)∣)k(|u(y)|)^{k} as the product of kk copies of ∣u(y)∣|u(y)|.

Consequently, for k∈{2,3,4}k\in\{2,3,4\} and any real tt with 1≤t1\le t, the restriction (uk)∣Q(u^{k})|_{Q} and the map ∣u∣Q∣k|u|_{Q}|^{k} have the same absolute value at every point of QQ, so the seminorms ∥(uk)∣Q∥t\lVert(u^{k})|_{Q}\rVert_{t} and ∥∣u∣Q∣k∥t\bigl\lVert|u|_{Q}|^{k}\bigr\rVert_{t} are equal, both being (∫Tn(∣u(y)∣)kt dx)1/t\bigl(\int_{\mathbb{T}^{n}}(|u(y)|)^{kt}\,dx\bigr)^{1/t} by Power-Integrable Functions and the p-Seminorm (iii) and Properties of Real Powers (iv). By Elementary Properties of the p-Seminorm (vi), applied to h=u∣Qh=u|_{Q} with the exponents r=kr=k and s=ts=t,

∥∣u∣Q∣k∥t=(∥u∣Q∥kt)k.\bigl\lVert|u|_{Q}|^{k}\bigr\rVert_{t}=\bigl(\lVert u|_{Q}\rVert_{kt}\bigr)^{k}.

In particular ∥(u3)∣Q∥2=S3\lVert(u^{3})|_{Q}\rVert_{2}=S^{3}, ∥(u4)∣Q∥3/2=S4\lVert(u^{4})|_{Q}\rVert_{3/2}=S^{4}, ∥∣u∣Q∣3∥2=S3\bigl\lVert|u|_{Q}|^{3}\bigr\rVert_{2}=S^{3} and ∥∣u∣Q∣2∥2=(∥u∣Q∥4)2\bigl\lVert|u|_{Q}|^{2}\bigr\rVert_{2}=\bigl(\lVert u|_{Q}\rVert_{4}\bigr)^{2}.

(R3) Four integral estimates. The numbers 22 and 22 are conjugate exponents, since 1<21<2 and 12+12=1\tfrac12+\tfrac12=1. Hence Hölder's Inequality (i), applied to the measure space (Q,BQ,λQ)(Q,\mathcal{B}_{Q},\lambda_{Q}) with p=q=2p=q=2, gives ∫Tn∣fg∣ dx≤∥f∥2∥g∥2\int_{\mathbb{T}^{n}}|fg|\,dx\le\lVert f\rVert_{2}\lVert g\rVert_{2} for all f,g∈L2(Tn)f,g\in\mathcal{L}^{2}(\mathbb{T}^{n}). The maps ∣u∣Q∣2|u|_{Q}|^{2} and ∣u∣Q∣3|u|_{Q}|^{3} belong to L2(Tn)\mathcal{L}^{2}(\mathbb{T}^{n}): by Elementary Properties of the p-Seminorm (vi), applied to h=u∣Qh=u|_{Q} with s=2s=2 and with r=2r=2 and r=3r=3 in turn, the map ∣h∣r|h|^{r} lies in L2(Tn)\mathcal{L}^{2}(\mathbb{T}^{n}) if and only if hh lies in L2r(Tn)\mathcal{L}^{2r}(\mathbb{T}^{n}), which it does, u∣Qu|_{Q} belonging to Lt(Tn)\mathcal{L}^{t}(\mathbb{T}^{n}) for every real tt with 1≤t1\le t. Taking in turn f=∣u∣Q∣3f=|u|_{Q}|^{3} with g=(∂ju)∣Qg=(\partial_{j}u)|_{Q}, f=∣u∣Q∣3f=|u|_{Q}|^{3} with g=u∣Qg=u|_{Q}, f=∣u∣Q∣2f=|u|_{Q}|^{2} with g=(∂ju)∣Qg=(\partial_{j}u)|_{Q} and f=∣u∣Q∣2f=|u|_{Q}|^{2} with g=u∣Qg=u|_{Q}, and using (R1), (R2) and the fact that the values of ∣u∣Q∣k|u|_{Q}|^{k} are nonnegative by Properties of Real Powers (i), we obtain for every j∈[n]j\in[n]

∫Tn∣u∣Q∣3 ∣(∂ju)∣Q∣ dx≤S3N,∫Tn∣u∣Q∣4 dx≤S3N,\int_{\mathbb{T}^{n}}|u|_{Q}|^{3}\,\bigl|(\partial_{j}u)|_{Q}\bigr|\,dx\le S^{3}N,\qquad\int_{\mathbb{T}^{n}}|u|_{Q}|^{4}\,dx\le S^{3}N, ∫Tn∣u∣Q∣2 ∣(∂ju)∣Q∣ dx≤(∥u∣Q∥4)2N,∫Tn∣u∣Q∣3 dx≤(∥u∣Q∥4)2N,\int_{\mathbb{T}^{n}}|u|_{Q}|^{2}\,\bigl|(\partial_{j}u)|_{Q}\bigr|\,dx\le\bigl(\lVert u|_{Q}\rVert_{4}\bigr)^{2}N,\qquad\int_{\mathbb{T}^{n}}|u|_{Q}|^{3}\,dx\le\bigl(\lVert u|_{Q}\rVert_{4}\bigr)^{2}N ,

the second and fourth because the pointwise products ∣u∣Q∣3 u∣Q|u|_{Q}|^{3}\,u|_{Q} and ∣u∣Q∣2 u∣Q|u|_{Q}|^{2}\,u|_{Q} have absolute values ∣u∣Q∣4|u|_{Q}|^{4} and ∣u∣Q∣3|u|_{Q}|^{3}, by claim 4 of Properties of the Absolute Value in an Ordered Field and Properties of Real Powers (iv). By Comparison of the Lebesgue Seminorms on the Torus (i), applied with r=4r=4 and s=6s=6, we have ∥u∣Q∥4≤S\lVert u|_{Q}\rVert_{4}\le S, so Properties of Real Powers (v) turns the last two bounds into S2NS^{2}N.

(R4) Cancelling a power. Let σ\sigma and τ\tau be nonnegative real numbers and let aa and bb be positive real numbers with σa+b≤τa σb\sigma^{a+b}\le\tau^{a}\,\sigma^{b}. Then σ≤τ\sigma\le\tau.

Indeed, if σ=0\sigma=0 then σ≤τ\sigma\le\tau since τ\tau is nonnegative. Otherwise 0<σ0<\sigma, so 0<σb0<\sigma^{b} by Properties of Real Powers (i), and the multiplicative inverse cc of σb\sigma^{b} is positive by claim 7 of Elementary Order Arithmetic in an Ordered Field. Since σa+b=σaσb\sigma^{a+b}=\sigma^{a}\sigma^{b} by Properties of Real Powers (iv), multiplying the hypothesis by cc, which is legitimate by claim 5 of Elementary Arithmetic in an Ordered Field, gives σa≤τa\sigma^{a}\le\tau^{a}. Now Properties of Real Powers (vi), in the form sa≤t  ⟺  s≤t1/as^{a}\le t\iff s\le t^{1/a} with t=τat=\tau^{a}, gives σ≤(τa)1/a=τ\sigma\le(\tau^{a})^{1/a}=\tau, the last equality by that same clause.

The case n=1n=1. By The Gagliardo-Nirenberg Inequality on the Torus in Dimensions at Most Three (i), ∣u(x)∣≤A1|u(x)|\le A_{1} for every x∈Rnx\in\mathbb{R}^{n}, where A1=∥u∣Q∥1+∥(∂1u)∣Q∥1A_{1}=\lVert u|_{Q}\rVert_{1}+\lVert(\partial_{1}u)|_{Q}\rVert_{1}. By Comparison of the Lebesgue Seminorms on the Torus (i), applied with r=1r=1 and s=2s=2, and by (R1), each of the two seminorms is at most NN, so A1≤N+N=2NA_{1}\le N+N=2N. Hence ∣u(x)∣≤2N|u(x)|\le2N for every xx, and Continuous Periodic Functions are Power-Integrable and Dense on the Torus (i), applied with p=6p=6 and M=2NM=2N, gives ∥u∣Q∥6≤2N\lVert u|_{Q}\rVert_{6}\le2N. Finally 2N≤5N2N\le5N by claim 5 of Elementary Arithmetic in an Ordered Field, since 2≤52\le5 and 0≤N0\le N.

The case n=2n=2. Put v=u3v=u^{3}. By Elementary Properties of the p-Seminorm (i), applied with p=1p=1, and Properties of Real Powers (ii),

∥v∣Q∥1=∫Tn∣u∣Q∣3 dx≤S2N,∥(∂jv)∣Q∥1=3∫Tn∣u∣Q∣2 ∣(∂ju)∣Q∣ dx≤3S2N,\lVert v|_{Q}\rVert_{1}=\int_{\mathbb{T}^{n}}|u|_{Q}|^{3}\,dx\le S^{2}N,\qquad\bigl\lVert(\partial_{j}v)|_{Q}\bigr\rVert_{1}=3\int_{\mathbb{T}^{n}}|u|_{Q}|^{2}\,\bigl|(\partial_{j}u)|_{Q}\bigr|\,dx\le3S^{2}N,

by (R2), (R3) and claim 4 of Properties of the Absolute Value in an Ordered Field, which gives ∣3u2∂ju∣=3∣u∣2∣∂ju∣|3u^{2}\partial_{j}u|=3|u|^{2}|\partial_{j}u|, together with the linearity in claim 2 of Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion. So the numbers Aj=∥v∣Q∥1+∥(∂jv)∣Q∥1A_{j}=\lVert v|_{Q}\rVert_{1}+\lVert(\partial_{j}v)|_{Q}\rVert_{1} of The Gagliardo-Nirenberg Inequality on the Torus in Dimensions at Most Three satisfy Aj≤4S2NA_{j}\le4S^{2}N for j∈[2]j\in[2]. By The Gagliardo-Nirenberg Inequality on the Torus in Dimensions at Most Three (ii), applied to v∈Cper1v\in C^{1}_{\mathrm{per}}, and by (R2),

S6=(S3)2=(∥v∣Q∥2)2≤A1A2≤(4S2N)2=16 S4N2,S^{6}=\bigl(S^{3}\bigr)^{2}=\bigl(\lVert v|_{Q}\rVert_{2}\bigr)^{2}\le A_{1}A_{2}\le\bigl(4S^{2}N\bigr)^{2}=16\,S^{4}N^{2},

using Properties of Real Powers (iv) for (S3)2=S6(S^{3})^{2}=S^{6}, claim 5 of Elementary Arithmetic in an Ordered Field twice for the product of the two bounds, and Properties of Real Powers (iii) for the last equality. Since 16N2=(4N)216N^{2}=(4N)^{2} by Properties of Real Powers (iii), (R4) applied with σ=S\sigma=S, τ=4N\tau=4N, a=2a=2 and b=4b=4 gives S≤4NS\le4N, and 4N≤5N4N\le5N by claim 5 of Elementary Arithmetic in an Ordered Field.

The case n=3n=3. Put v=u4v=u^{4}. Exactly as in the previous case, with the first and second estimates of (R3) in place of the third and fourth,

∥v∣Q∥1=∫Tn∣u∣Q∣4 dx≤S3N,∥(∂jv)∣Q∥1=4∫Tn∣u∣Q∣3 ∣(∂ju)∣Q∣ dx≤4S3N,\lVert v|_{Q}\rVert_{1}=\int_{\mathbb{T}^{n}}|u|_{Q}|^{4}\,dx\le S^{3}N,\qquad\bigl\lVert(\partial_{j}v)|_{Q}\bigr\rVert_{1}=4\int_{\mathbb{T}^{n}}|u|_{Q}|^{3}\,\bigl|(\partial_{j}u)|_{Q}\bigr|\,dx\le4S^{3}N,

so Aj≤5S3NA_{j}\le5S^{3}N for j∈[3]j\in[3]. By The Gagliardo-Nirenberg Inequality on the Torus in Dimensions at Most Three (iii), applied to vv, together with (R2), Properties of Real Powers (v), Properties of Real Powers (iv) and Properties of Real Powers (iii),

S6=(S4)3/2=(∥v∣Q∥3/2)3/2≤(A1A2A3)1/2≤((5S3N)3)1/2=(5N)3/2 S9/2,S^{6}=\bigl(S^{4}\bigr)^{3/2}=\bigl(\lVert v|_{Q}\rVert_{3/2}\bigr)^{3/2}\le\bigl(A_{1}A_{2}A_{3}\bigr)^{1/2}\le\Bigl(\bigl(5S^{3}N\bigr)^{3}\Bigr)^{1/2}=\bigl(5N\bigr)^{3/2}\,S^{9/2},

the second inequality because A1A2A3≤(5S3N)3A_{1}A_{2}A_{3}\le(5S^{3}N)^{3} by claim 5 of Elementary Arithmetic in an Ordered Field applied twice, all factors being nonnegative. Now (R4), applied with σ=S\sigma=S, τ=5N\tau=5N, a=3/2a=3/2 and b=9/2b=9/2, gives S≤5NS\le5N.

In each of the three cases ∥u∣Q∥6=S≤5N\lVert u|_{Q}\rVert_{6}=S\le5N, which is claim 1.

Proof of claim 2. Let U∈H1(Tn)U\in H^{1}(\mathbb{T}^{n}), let vv be a representative of UU and put N=∥U∥H1N=\lVert U\rVert_{H^{1}}, a nonnegative real number.

Step 1: mollification. Let ρ\rho be a mollifier kernel of radius 11 on Rn\mathbb{R}^{n}, which exists by Existence of Mollifier Kernels of Every Radius, and for a natural number kk let ρ1/k\rho_{1/k} be its rescaling with parameter 1/k1/k. By The Sobolev Space on the Torus is a Separable Hilbert Space Densely Embedded in the Square-Integrable Space (iii), applied with δ=1\delta=1, with UU and with the representative vv, the maps uk=ρ1/k⋆vu_{k}=\rho_{1/k}\star v lie in Cper∞C^{\infty}_{\mathrm{per}} and the classes Uk=[uk∣Q]U_{k}=[u_{k}|_{Q}] lie in H1(Tn)H^{1}(\mathbb{T}^{n}); moreover ∥Uk∥H1≤N\lVert U_{k}\rVert_{H^{1}}\le N by The Sobolev Space on the Torus is a Separable Hilbert Space Densely Embedded in the Square-Integrable Space (v), and the sequence (Uk)k∈N(U_{k})_{k\in\mathbb{N}} converges to UU in H1(Tn)H^{1}(\mathbb{T}^{n}) by The Sobolev Space on the Torus is a Separable Hilbert Space Densely Embedded in the Square-Integrable Space (vi).

A smooth map on Rn\mathbb{R}^{n} is of class C1C^{1} by Smooth Functions on an Open Subset of ℝⁿ, so each uku_{k}, being Zn\mathbb{Z}^{n}-periodic and smooth, lies in Cper1C^{1}_{\mathrm{per}} by Lattice-Periodic Functions and the Periodic Function Classes (iii). Claim 1, applied to uku_{k}, therefore gives

∥uk∣Q∥6≤5∥Uk∥H1≤5N,\lVert u_{k}|_{Q}\rVert_{6}\le5\lVert U_{k}\rVert_{H^{1}}\le5N ,

the second inequality by claim 5 of Elementary Arithmetic in an Ordered Field.

Step 2: an almost everywhere convergent subsequence. By The Sobolev Space on the Torus is a Separable Hilbert Space Densely Embedded in the Square-Integrable Space (i) the distance in L2(Tn)L^{2}(\mathbb{T}^{n}) between two members of H1(Tn)H^{1}(\mathbb{T}^{n}) is at most their distance in H1(Tn)H^{1}(\mathbb{T}^{n}), so any positive real witnessing the convergence of (Uk)(U_{k}) to UU in H1(Tn)H^{1}(\mathbb{T}^{n}) witnesses it in L2(Tn)L^{2}(\mathbb{T}^{n}) as well; thus (Uk)(U_{k}) converges to UU in L2(Tn)L^{2}(\mathbb{T}^{n}) by Convergent Sequence in a Metric Space. By The Riesz–Fischer Theorem (iii), applied with p=2p=2 to this sequence, to the representatives uk∣Qu_{k}|_{Q} of UkU_{k} and to the representative vv of UU, there are natural numbers m1<m2<…m_{1}<m_{2}<\dots and a null set E0E_{0} such that for every y∈Q∖E0y\in Q\setminus E_{0} the sequence (umj(y))j∈N(u_{m_{j}}(y))_{j\in\mathbb{N}} converges to v(y)v(y). That clause also provides a dominating function, which is not needed here. A null set is not required to lie in BQ\mathcal{B}_{Q}, so we enlarge it: by Null Set of a Measure there is E∈BQE\in\mathcal{B}_{Q} with E0⊆EE_{0}\subseteq E and λQ(E)=0\lambda_{Q}(E)=0, and this EE is itself a null set. Since Q∖E⊆Q∖E0Q\setminus E\subseteq Q\setminus E_{0}, the sequence (umj(y))j∈N(u_{m_{j}}(y))_{j\in\mathbb{N}} converges to v(y)v(y) for every y∈Q∖Ey\in Q\setminus E.

Step 3: two maps into the extended half-line. For j∈Nj\in\mathbb{N} let gj:Q→[0,∞]g_{j}:Q\to[0,\infty] be the map taking the value (∣umj(y)∣)6(|u_{m_{j}}(y)|)^{6} at y∈Q∖Ey\in Q\setminus E and the value 00 at y∈Ey\in E, and let g:Q→[0,∞]g:Q\to[0,\infty] take the value (∣v(y)∣)6(|v(y)|)^{6} at y∈Q∖Ey\in Q\setminus E and the value 00 at y∈Ey\in E. The maps ∣umj∣Q∣6|u_{m_{j}}|_{Q}|^{6} and ∣v∣6|v|^{6} are measurable with respect to BQ\mathcal{B}_{Q} by Power-Integrable Functions and the p-Seminorm (i), the set Q∖EQ\setminus E lies in BQ\mathcal{B}_{Q}, that being a σ\sigma-algebra on QQ of which EE is a member, and its indicator is measurable by claim 1 of Arithmetic of Measurable Real- and [0,∞]-Valued Maps, and gjg_{j} and gg are the pointwise products of these, hence measurable by claim 3 of that lemma; their values are nonnegative by Properties of Real Powers (i), so they are measurable as maps into [0,∞][0,\infty] by Measurable Functions and the Lebesgue Integral: Standing Notation (iii).

We claim that g(y)≤lim inf⁡jgj(y)g(y)\le\liminf_{j}g_{j}(y) for every y∈Qy\in Q, the limit inferior being the one defined in Fatou's Lemma. For y∈Ey\in E this holds because g(y)=0g(y)=0 and every value gj(y)g_{j}(y) is nonnegative, so every greatest lower bound occurring in the definition is nonnegative and hence so is their least upper bound. Let then y∈Q∖Ey\in Q\setminus E and write αj=gj(y)\alpha_{j}=g_{j}(y) and α=g(y)\alpha=g(y). By claim 7 of Properties of the Absolute Value in an Ordered Field the sequence (∣umj(y)∣)j(|u_{m_{j}}(y)|)_{j} converges to ∣v(y)∣|v(y)|, and then Properties of Real Powers (vii), applied with the exponent 66, shows that (αj)j(\alpha_{j})_{j} converges to α\alpha. Let ε\varepsilon be a real number with 0<ε0<\varepsilon. There is k∈Nk\in\mathbb{N} with ∣αj−α∣<ε|\alpha_{j}-\alpha|<\varepsilon for every j∈Nj\in\mathbb{N} with k≤jk\le j, so α−ε≤αj\alpha-\varepsilon\le\alpha_{j} for all such jj by claim 3 of Properties of the Absolute Value in an Ordered Field; hence α−ε\alpha-\varepsilon is a lower bound of the values αj\alpha_{j} with k≤jk\le j, so α−ε≤inf⁡j≥kαj≤lim inf⁡jαj\alpha-\varepsilon\le\inf_{j\ge k}\alpha_{j}\le\liminf_{j}\alpha_{j}, the last step because that greatest lower bound is one of the numbers whose least upper bound is the limit inferior. Write L=lim inf⁡jαjL=\liminf_{j}\alpha_{j}. If L=∞L=\infty then α≤L\alpha\le L; otherwise LL is a nonnegative real number, and if L<αL<\alpha we may take ε=(α−L)/2\varepsilon=(\alpha-L)/2, which is positive by claim 3 of Elementary Arithmetic in an Ordered Field and claim 8 of Elementary Order Arithmetic in an Ordered Field, and then α−ε≤L\alpha-\varepsilon\le L together with ε<α−L\varepsilon<\alpha-L, again by claim 8, gives L<α−ε≤LL<\alpha-\varepsilon\le L, a contradiction. So α≤L\alpha\le L in every case.

Step 4: Fatou's lemma. For every jj and every y∈Qy\in Q we have gj(y)≤∣umj∣Q∣6(y)g_{j}(y)\le|u_{m_{j}}|_{Q}|^{6}(y), the two being equal off EE and the right-hand side being nonnegative on EE. So The Lebesgue Integral and Null Sets (iii), Elementary Properties of the p-Seminorm (i) applied with p=6p=6, Step 1 and Properties of Real Powers (v) give

∫Tngj dx≤∫Tn∣umj∣Q∣6 dx=(∥umj∣Q∥6)6≤(5N)6(j∈N).\int_{\mathbb{T}^{n}}g_{j}\,dx\le\int_{\mathbb{T}^{n}}|u_{m_{j}}|_{Q}|^{6}\,dx=\bigl(\lVert u_{m_{j}}|_{Q}\rVert_{6}\bigr)^{6}\le(5N)^{6}\qquad(j\in\mathbb{N}).

Consequently (5N)6(5N)^{6} is an upper bound of every greatest lower bound occurring in the definition of lim inf⁡j∫Tngj dx\liminf_{j}\int_{\mathbb{T}^{n}}g_{j}\,dx, so that limit inferior is at most (5N)6(5N)^{6}. By Fatou's Lemma, applied to the measure space (Q,BQ,λQ)(Q,\mathcal{B}_{Q},\lambda_{Q}) and the sequence (gj)(g_{j}), and by Step 3 with The Lebesgue Integral and Null Sets (iii),

∫Tng dx≤∫Tn(lim inf⁡jgj)dx≤lim inf⁡j∫Tngj dx≤(5N)6.\int_{\mathbb{T}^{n}}g\,dx\le\int_{\mathbb{T}^{n}}\Bigl(\liminf_{j}g_{j}\Bigr)dx\le\liminf_{j}\int_{\mathbb{T}^{n}}g_{j}\,dx\le(5N)^{6}.

Step 5: conclusion. The maps gg and ∣v∣6|v|^{6} agree at every point of Q∖EQ\setminus E, and EE is null, so they agree almost everywhere and The Lebesgue Integral and Null Sets (iii) gives

∫Tn∣v∣6 dx=∫Tng dx≤(5N)6,\int_{\mathbb{T}^{n}}|v|^{6}\,dx=\int_{\mathbb{T}^{n}}g\,dx\le(5N)^{6},

which is finite. The map vv is measurable with respect to BQ\mathcal{B}_{Q}, being a member of L2(Tn)\mathcal{L}^{2}(\mathbb{T}^{n}), so v∈L6(Tn)v\in\mathcal{L}^{6}(\mathbb{T}^{n}) by Power-Integrable Functions and the p-Seminorm (ii). Finally, by Power-Integrable Functions and the p-Seminorm (iii), Properties of Real Powers (v) and Properties of Real Powers (vi),

∥v∥6=(∫Tn∣v∣6 dx)1/6≤((5N)6)1/6=5N,\lVert v\rVert_{6}=\Bigl(\int_{\mathbb{T}^{n}}|v|^{6}\,dx\Bigr)^{1/6}\le\bigl((5N)^{6}\bigr)^{1/6}=5N ,

which is claim 2.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…