TheoremBase

After recording the norm decomposition and a convergence criterion, completeness follows from that of the square-integrable space together with closure of the weak derivative under limits; mollification supplies the approximating smooth classes, and a countable dense set is obtained by mollifying a countable dense subset of the square-integrable space.

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 x≤yx\le y. We use repeatedly the following consequence of claims 2 and 3 of Elementary Arithmetic in an Ordered Field: if a≤ba\le b and c≤dc\le d then a+c≤b+da+c\le b+d, since 0≤b−a0\le b-a and 0≤d−c0\le d-c give 0≤(b+d)−(a+c)0\le(b+d)-(a+c).

(N0) The absolute value of a nonnegative real. Let xx be a real number with 0≤x0\le x. Then ∣x∣=x|x|=x. Indeed x≤∣x∣x\le|x| by claim 3 of Properties of the Absolute Value in an Ordered Field. Also −x≤−0-x\le -0 by claim 4 of Elementary Order Arithmetic in an Ordered Field, and −0=0-0=0 by claim 4 of Additive Cancellation and Elementary Additive Identities in a Field, so −x≤0-x\le 0 and hence −x≤x-x\le x by transitivity, while x≤xx\le x; hence ∣x∣≤x|x|\le x by claim 6 of Properties of the Absolute Value in an Ordered Field applied with c=xc=x. Antisymmetry gives ∣x∣=x|x|=x.

(N1) Norms, squares and square roots. Let EE be a real inner product space with inner product ⟨ ⋅ ,⋅ ⟩E\langle\,\cdot\,,\cdot\,\rangle_{E}, norm ∥ ⋅ ∥E\lVert\,\cdot\,\rVert_{E} and distance dEd_{E}. By Real Inner Product Spaces (ii), 0≤∥X∥E0\le\lVert X\rVert_{E} and the product ∥X∥E⋅∥X∥E\lVert X\rVert_{E}\cdot\lVert X\rVert_{E} equals ⟨X,X⟩E\langle X,X\rangle_{E}; by Properties of Real Powers (ii) that product is the power ∥X∥E2\lVert X\rVert_{E}^{2} of Real Powers (i), so ∥X∥E2=⟨X,X⟩E\lVert X\rVert_{E}^{2}=\langle X,X\rangle_{E}. By Properties of Real Powers (vi), for nonnegative reals ss and tt we have (t2)1/2=t(t^{2})^{1/2}=t, (t1/2)2=t(t^{1/2})^{2}=t, and s2≤ts^{2}\le t if and only if s≤t1/2s\le t^{1/2}; consequently s2≤t2s^{2}\le t^{2} implies s≤ts\le t. By Properties of Real Powers (i) a power of a nonnegative real is nonnegative, and 01/2=00^{1/2}=0. By Real Inner Product Spaces (iii), dE(X,Y)=∥X−Y∥Ed_{E}(X,Y)=\lVert X-Y\rVert_{E}. We apply all of this to L2(Tn)L^{2}(\mathbb{T}^{n}), which is a real inner product space by L²(μ) Is a Real Hilbert Space (i), and to H1(Tn)H^{1}(\mathbb{T}^{n}), which is one by The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (ii), without further mention.

(N2) The decomposition of the Sobolev norm and distance. Let U,U′∈H1(Tn)U,U'\in H^{1}(\mathbb{T}^{n}). By The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (ii) and (N1),

∥U∥H12=∥U∥L22+∑j=1n∥∂jU∥L22.\lVert U\rVert_{H^{1}}^{2}=\lVert U\rVert_{L^{2}}^{2}+\sum_{j=1}^{n}\lVert\partial_{j}U\rVert_{L^{2}}^{2}.

Moreover U−U′∈H1(Tn)U-U'\in H^{1}(\mathbb{T}^{n}), H1(Tn)H^{1}(\mathbb{T}^{n}) being a linear subspace of L2(Tn)L^{2}(\mathbb{T}^{n}) by The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (i), and ∂j(U−U′)=∂jU−∂jU′\partial_{j}(U-U')=\partial_{j}U-\partial_{j}U' for every j∈[n]j\in[n] by Elementary Properties of the Weak Partial Derivative on the Torus (ii) applied with c=−1c=-1. Applying the display to U−U′U-U' and using (N1) therefore gives

dH1(U,U′)2=dL2(U,U′)2+∑j=1ndL2(∂jU,∂jU′)2.d_{H^{1}}(U,U')^{2}=d_{L^{2}}(U,U')^{2}+\sum_{j=1}^{n}d_{L^{2}}(\partial_{j}U,\partial_{j}U')^{2}.

(N3) Convergence and the sequence of distances. Let (Y,d)(Y,d) be a metric space, let (Xm)m∈N(X_{m})_{m\in\mathbb{N}} be a sequence in YY and let X∈YX\in Y. Then (Xm)(X_{m}) converges to XX in (Y,d)(Y,d) if and only if the sequence of real numbers (d(Xm,X))m∈N\bigl(d(X_{m},X)\bigr)_{m\in\mathbb{N}} converges to 00. Indeed the two conditions are the same assertion about ε\varepsilon and NN, because 0≤d(Xm,X)0\le d(X_{m},X) by Metric Space and hence ∣d(Xm,X)−0∣=d(Xm,X)|d(X_{m},X)-0|=d(X_{m},X) by (N0), the difference d(Xm,X)−0d(X_{m},X)-0 being d(Xm,X)d(X_{m},X) by claim 4 of Additive Cancellation and Elementary Additive Identities in a Field.

(N4) A convergence criterion in the Sobolev space. Let UU and, for every m∈Nm\in\mathbb{N}, UmU_{m} belong to H1(Tn)H^{1}(\mathbb{T}^{n}), and suppose that (Um)(U_{m}) converges to UU in L2(Tn)L^{2}(\mathbb{T}^{n}) and that, for every j∈[n]j\in[n], (∂jUm)(\partial_{j}U_{m}) converges to ∂jU\partial_{j}U in L2(Tn)L^{2}(\mathbb{T}^{n}). Then (Um)(U_{m}) converges to UU in H1(Tn)H^{1}(\mathbb{T}^{n}).

To see this, put bm=dL2(Um,U)b_{m}=d_{L^{2}}(U_{m},U) and bj,m=dL2(∂jUm,∂jU)b_{j,m}=d_{L^{2}}(\partial_{j}U_{m},\partial_{j}U) for j∈[n]j\in[n]. By (N3) the real sequences (bm)m(b_{m})_{m} and (bj,m)m(b_{j,m})_{m} converge to 00, so by claim 2 of Arithmetic of Limits of Real Sequences the sequences (bm2)m(b_{m}^{2})_{m} and (bj,m2)m(b_{j,m}^{2})_{m} converge to 0⋅0=00\cdot 0=0, the squares being products by Properties of Real Powers (ii). By Finite Sums of Real Numbers: Nonnegativity, Domination by the Sum, and Limits (iii), applied with N=nN=n and the constant limits 00, the sequence whose mmth term is ∑j=1nbj,m2\sum_{j=1}^{n}b_{j,m}^{2} converges to ∑j=1n0\sum_{j=1}^{n}0, which is 00 by the second assertion of Finite Sums of Real Numbers: Nonnegativity, Domination by the Sum, and Limits (i). By claim 1 of Arithmetic of Limits of Real Sequences and (N2), the sequence whose mmth term is dH1(Um,U)2d_{H^{1}}(U_{m},U)^{2} converges to 0+0=00+0=0. Applying Properties of Real Powers (vii) with exponent 1/21/2 to this sequence of nonnegative reals, the sequence whose mmth term is (dH1(Um,U)2)1/2\bigl(d_{H^{1}}(U_{m},U)^{2}\bigr)^{1/2} converges to 01/20^{1/2}; by (N1) that mmth term is dH1(Um,U)d_{H^{1}}(U_{m},U) and 01/2=00^{1/2}=0. By (N3) again, (Um)(U_{m}) converges to UU in H1(Tn)H^{1}(\mathbb{T}^{n}).

Proof of claim 1. Let U,U′∈H1(Tn)U,U'\in H^{1}(\mathbb{T}^{n}) and let j∈[n]j\in[n]. For k∈[n]k\in[n] put tk=∥∂kU∥L22t_{k}=\lVert\partial_{k}U\rVert_{L^{2}}^{2}, a nonnegative real by (N1), and put Σ=∑k=1ntk\Sigma=\sum_{k=1}^{n}t_{k}.

By Finite Sums of Real Numbers: Nonnegativity, Domination by the Sum, and Limits (i), 0≤Σ0\le\Sigma. By (N2), ∥U∥H12−∥U∥L22=Σ\lVert U\rVert_{H^{1}}^{2}-\lVert U\rVert_{L^{2}}^{2}=\Sigma, so claim 3 of Elementary Arithmetic in an Ordered Field gives ∥U∥L22≤∥U∥H12\lVert U\rVert_{L^{2}}^{2}\le\lVert U\rVert_{H^{1}}^{2}, and (N1) gives ∥U∥L2≤∥U∥H1\lVert U\rVert_{L^{2}}\le\lVert U\rVert_{H^{1}}.

By Finite Sums of Real Numbers: Nonnegativity, Domination by the Sum, and Limits (ii), applied with N=nN=n, the index nn in place of its jj and the index jj in place of its ii, we get tj≤Σt_{j}\le\Sigma. Also ∥U∥H12−Σ=∥U∥L22\lVert U\rVert_{H^{1}}^{2}-\Sigma=\lVert U\rVert_{L^{2}}^{2} by (N2), which is nonnegative by (N1), so Σ≤∥U∥H12\Sigma\le\lVert U\rVert_{H^{1}}^{2} by claim 3 of Elementary Arithmetic in an Ordered Field. By transitivity ∥∂jU∥L22≤∥U∥H12\lVert\partial_{j}U\rVert_{L^{2}}^{2}\le\lVert U\rVert_{H^{1}}^{2}, and (N1) gives ∥∂jU∥L2≤∥U∥H1\lVert\partial_{j}U\rVert_{L^{2}}\le\lVert U\rVert_{H^{1}}.

For the distance inequalities, apply the two norm inequalities just proved to U−U′U-U', which lies in H1(Tn)H^{1}(\mathbb{T}^{n}) by (N2). Using (N1) and the identity ∂j(U−U′)=∂jU−∂jU′\partial_{j}(U-U')=\partial_{j}U-\partial_{j}U' of (N2),

dL2(U,U′)=∥U−U′∥L2≤∥U−U′∥H1=dH1(U,U′),d_{L^{2}}(U,U')=\lVert U-U'\rVert_{L^{2}}\le\lVert U-U'\rVert_{H^{1}}=d_{H^{1}}(U,U'), dL2(∂jU,∂jU′)=∥∂j(U−U′)∥L2≤∥U−U′∥H1=dH1(U,U′).d_{L^{2}}(\partial_{j}U,\partial_{j}U')=\lVert\partial_{j}(U-U')\rVert_{L^{2}}\le\lVert U-U'\rVert_{H^{1}}=d_{H^{1}}(U,U').

Proof of claim 2. By The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (ii), H1(Tn)H^{1}(\mathbb{T}^{n}) with ⟨ ⋅ ,⋅ ⟩H1\langle\,\cdot\,,\cdot\,\rangle_{H^{1}} is a real inner product space whose distance is dH1d_{H^{1}}, and dH1d_{H^{1}} is a metric on it by The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity (iii). By Real Hilbert Spaces (i) it therefore suffices to prove that the metric space (H1(Tn),dH1)(H^{1}(\mathbb{T}^{n}),d_{H^{1}}) is complete.

Let (Um)m∈N(U_{m})_{m\in\mathbb{N}} be a Cauchy sequence in (H1(Tn),dH1)(H^{1}(\mathbb{T}^{n}),d_{H^{1}}) and let j∈[n]j\in[n] be arbitrary; the argument that follows uses nothing about jj beyond j∈[n]j\in[n], so its conclusion holds for every j∈[n]j\in[n]. Given a real ε>0\varepsilon>0, choose N∈NN\in\mathbb{N} with dH1(Um,Uℓ)<εd_{H^{1}}(U_{m},U_{\ell})<\varepsilon for all m,ℓ∈Nm,\ell\in\mathbb{N} with m≥Nm\ge N and ℓ≥N\ell\ge N. By claim 1, for such mm and ℓ\ell we have dL2(Um,Uℓ)≤dH1(Um,Uℓ)<εd_{L^{2}}(U_{m},U_{\ell})\le d_{H^{1}}(U_{m},U_{\ell})<\varepsilon and dL2(∂jUm,∂jUℓ)≤dH1(Um,Uℓ)<εd_{L^{2}}(\partial_{j}U_{m},\partial_{j}U_{\ell})\le d_{H^{1}}(U_{m},U_{\ell})<\varepsilon. Hence (Um)(U_{m}) and (∂jUm)(\partial_{j}U_{m}) are Cauchy sequences in L2(Tn)L^{2}(\mathbb{T}^{n}) with the distance dL2d_{L^{2}}.

By The Flat Torus: Standing Notation (iv) the space L2(Tn)L^{2}(\mathbb{T}^{n}) is a real Banach space, so by that clause its metric space is complete; by the preamble that metric is dL2d_{L^{2}}. Hence there is U∈L2(Tn)U\in L^{2}(\mathbb{T}^{n}) to which (Um)(U_{m}) converges, and for each j∈[n]j\in[n] there is Gj∈L2(Tn)G_{j}\in L^{2}(\mathbb{T}^{n}) to which (∂jUm)(\partial_{j}U_{m}) converges; the nn choices of GjG_{j} are finite in number and require no appeal to a choice principle beyond selecting one object for each of finitely many indices.

Fix j∈[n]j\in[n]. The classes UmU_{m} lie in L2(Tn)L^{2}(\mathbb{T}^{n}) and ∂jUm\partial_{j}U_{m} is the jj-th weak partial derivative of UmU_{m} for every mm, and (Um)(U_{m}) and (∂jUm)(\partial_{j}U_{m}) converge in L2(Tn)L^{2}(\mathbb{T}^{n}) to UU and to GjG_{j}; so Elementary Properties of the Weak Partial Derivative on the Torus (v) shows that GjG_{j} is the jj-th weak partial derivative of UU. As this holds for every j∈[n]j\in[n], The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (i) gives U∈H1(Tn)U\in H^{1}(\mathbb{T}^{n}), and ∂jU=Gj\partial_{j}U=G_{j} by the uniqueness recorded in The Weak Partial Derivative on the Torus (ii). By (N4), (Um)(U_{m}) converges to UU in H1(Tn)H^{1}(\mathbb{T}^{n}). Every Cauchy sequence therefore converges, as required.

Proof of claim 3. Let kk be a natural number. By Rescaling a Mollifier Kernel, ρ1/k\rho_{1/k} is a mollifier kernel of radius R=(1/k) δR=(1/k)\,\delta on Rn\mathbb{R}^{n}, and 0<R0<R by claim 5 of Elementary Order Arithmetic in an Ordered Field. By Mollifier Kernel of Radius δ\delta on Rn\mathbb{R}^n the map ρ1/k\rho_{1/k} is smooth, satisfies 0≤ρ1/k(y)0\le\rho_{1/k}(y) for every y∈Rny\in\mathbb{R}^{n}, vanishes at every yy with R<∥y∥R<\lVert y\rVert, and has ∫Rnρ1/k dλn=1\int_{\mathbb{R}^{n}}\rho_{1/k}\,d\lambda_{n}=1; being smooth it is continuous by claim 3 of Euclidean Space Is Open in Itself. So ρ1/k\rho_{1/k} meets the standing hypotheses imposed on the kernel in Properties of Periodic Convolution on the Torus and in The Lebesgue Bound for Periodic Convolution and Convergence of Mollifications, with this RR.

We record once that Cper∞⊆Cper1C^{\infty}_{\mathrm{per}}\subseteq C^{1}_{\mathrm{per}}: a smooth map on Rn\mathbb{R}^{n} is of class C1C^{1} by claim 2 of Euclidean Space Is Open in Itself, and periodicity is the same condition for the two classes by Lattice-Periodic Functions and the Periodic Function Classes (iii).

Part (a). Since u∈L2(Tn)u\in\mathcal{L}^{2}(\mathbb{T}^{n}) and ρ1/k\rho_{1/k} is smooth, Properties of Periodic Convolution on the Torus (i) gives ρ1/k⋆u∈Cper∞\rho_{1/k}\star u\in C^{\infty}_{\mathrm{per}}, hence ρ1/k⋆u∈Cper1\rho_{1/k}\star u\in C^{1}_{\mathrm{per}}. By The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (i), Uk∈H1(Tn)U_{k}\in H^{1}(\mathbb{T}^{n}), and by Elementary Properties of the Weak Partial Derivative on the Torus (i) the jj-th weak partial derivative of UkU_{k} is the class of (∂j(ρ1/k⋆u))∣Q\bigl(\partial_{j}(\rho_{1/k}\star u)\bigr)|_{Q} for every j∈[n]j\in[n]. By Elementary Properties of the Weak Partial Derivative on the Torus (iv), applied to UU, to ∂jU\partial_{j}U, to the representatives uu and gjg_{j}, and to the kernel ρ\rho with parameter 1/k1/k, we have ∂j(ρ1/k⋆u)=ρ1/k⋆gj\partial_{j}(\rho_{1/k}\star u)=\rho_{1/k}\star g_{j} as maps on Rn\mathbb{R}^{n}. Hence ∂jUk=[(ρ1/k⋆gj)∣Q]\partial_{j}U_{k}=[(\rho_{1/k}\star g_{j})|_{Q}].

Part (b). Let KK be the integral of ∣ρ1/k∣|\rho_{1/k}| over Rn\mathbb{R}^{n}. Since 0≤ρ1/k(y)0\le\rho_{1/k}(y) for every yy, (N0) gives ∣ρ1/k(y)∣=ρ1/k(y)|\rho_{1/k}(y)|=\rho_{1/k}(y) for every yy, so ∣ρ1/k∣|\rho_{1/k}| and ρ1/k\rho_{1/k} are the same map and K=1K=1. Applying The Lebesgue Bound for Periodic Convolution and Convergence of Mollifications (i) with p=2p=2 to uu and to each gjg_{j}, and using 1⋅t=t1\cdot t=t,

∥(ρ1/k⋆u)∣Q∥2≤∥u∥2,∥(ρ1/k⋆gj)∣Q∥2≤∥gj∥2(j∈[n]).\bigl\lVert(\rho_{1/k}\star u)|_{Q}\bigr\rVert_{2}\le\lVert u\rVert_{2},\qquad\bigl\lVert(\rho_{1/k}\star g_{j})|_{Q}\bigr\rVert_{2}\le\lVert g_{j}\rVert_{2}\quad(j\in[n]).

By The Lebesgue Space Lᵖ(μ) (iii) and L²(μ) Is a Real Hilbert Space (i) these are the inequalities ∥Uk∥L2≤∥U∥L2\lVert U_{k}\rVert_{L^{2}}\le\lVert U\rVert_{L^{2}} and ∥∂jUk∥L2≤∥∂jU∥L2\lVert\partial_{j}U_{k}\rVert_{L^{2}}\le\lVert\partial_{j}U\rVert_{L^{2}}, by part (a). Squaring, which preserves ≤\le between nonnegative reals by Properties of Real Powers (v), and comparing the two sums term by term by claim 1 of Comparison and Absolute Value Bounds for Finite Sums of Real Numbers,

∥Uk∥L22≤∥U∥L22,∑j=1n∥∂jUk∥L22≤∑j=1n∥∂jU∥L22.\lVert U_{k}\rVert_{L^{2}}^{2}\le\lVert U\rVert_{L^{2}}^{2},\qquad\sum_{j=1}^{n}\lVert\partial_{j}U_{k}\rVert_{L^{2}}^{2}\le\sum_{j=1}^{n}\lVert\partial_{j}U\rVert_{L^{2}}^{2}.

Adding these two inequalities and using (N2) gives ∥Uk∥H12≤∥U∥H12\lVert U_{k}\rVert_{H^{1}}^{2}\le\lVert U\rVert_{H^{1}}^{2}, whence ∥Uk∥H1≤∥U∥H1\lVert U_{k}\rVert_{H^{1}}\le\lVert U\rVert_{H^{1}} by (N1).

Convergence. By The Lebesgue Bound for Periodic Convolution and Convergence of Mollifications (ii) with p=2p=2, applied to uu, the sequence whose kkth term is the class of (ρ1/k⋆u)∣Q(\rho_{1/k}\star u)|_{Q}, that is UkU_{k}, converges to [u]=U[u]=U in L2(Tn)L^{2}(\mathbb{T}^{n}); applied to gjg_{j}, the sequence whose kkth term is the class of (ρ1/k⋆gj)∣Q(\rho_{1/k}\star g_{j})|_{Q}, that is ∂jUk\partial_{j}U_{k} by part (a), converges to [gj]=∂jU[g_{j}]=\partial_{j}U. By (N4), (Uk)k∈N(U_{k})_{k\in\mathbb{N}} converges to UU in H1(Tn)H^{1}(\mathbb{T}^{n}).

Proof of claim 4. Write S0={[w∣Q]:w∈Cper∞}\mathcal{S}_{0}=\{[w|_{Q}]:w\in C^{\infty}_{\mathrm{per}}\}. Since Cper∞⊆Cper1C^{\infty}_{\mathrm{per}}\subseteq C^{1}_{\mathrm{per}}, as recorded in the proof of claim 3, The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (i) gives S0⊆H1(Tn)\mathcal{S}_{0}\subseteq H^{1}(\mathbb{T}^{n}).

Let U∈H1(Tn)U\in H^{1}(\mathbb{T}^{n}) and let ε\varepsilon be a real number with 0<ε0<\varepsilon. By Existence of Mollifier Kernels of Every Radius there is a mollifier kernel ρ\rho of radius 11 on Rn\mathbb{R}^{n}. Choose a representative uu of UU and, for each j∈[n]j\in[n], a representative gjg_{j} of ∂jU\partial_{j}U. By claim 3, applied with δ=1\delta=1, the sequence (Uk)k∈N(U_{k})_{k\in\mathbb{N}} converges to UU in H1(Tn)H^{1}(\mathbb{T}^{n}), so by Convergent Sequence in a Metric Space there is k∈Nk\in\mathbb{N} with dH1(Uk,U)<εd_{H^{1}}(U_{k},U)<\varepsilon. By part (a) of claim 3, ρ1/k⋆u∈Cper∞\rho_{1/k}\star u\in C^{\infty}_{\mathrm{per}}, so Uk∈S0U_{k}\in\mathcal{S}_{0}.

Thus for every real ε>0\varepsilon>0 there is an element of S0\mathcal{S}_{0} at dH1d_{H^{1}}-distance less than ε\varepsilon from UU; by condition 3 of Characterization of the Closure in a Metric Space by Open Balls, UU belongs to the closure of S0\mathcal{S}_{0} in (H1(Tn),dH1)(H^{1}(\mathbb{T}^{n}),d_{H^{1}}). As UU was arbitrary, that closure is H1(Tn)H^{1}(\mathbb{T}^{n}), so S0\mathcal{S}_{0} is dense in H1(Tn)H^{1}(\mathbb{T}^{n}).

Proof of claim 5. By Existence of Mollifier Kernels of Every Radius fix a mollifier kernel ρ\rho of radius 11 on Rn\mathbb{R}^{n}, and for k∈Nk\in\mathbb{N} let ρ1/k\rho_{1/k} be its rescaling, a mollifier kernel of radius Rk=1/kR_{k}=1/k on Rn\mathbb{R}^{n} by Rescaling a Mollifier Kernel, with 0<Rk0<R_{k}. Let DD be the set of maps furnished by Separability of the Lebesgue Spaces of the Torus for p=2p=2; by Separability of the Lebesgue Spaces of the Torus (ii) every s∈Ds\in D lies in L2(Tn)\mathcal{L}^{2}(\mathbb{T}^{n}) and the set {[s]:s∈D}\{[s]:s\in D\} is countable. Put

S={ [(ρ1/k⋆s)∣Q] : s∈D, k∈N }.\mathcal{S}=\bigl\{\,[(\rho_{1/k}\star s)|_{Q}]\ :\ s\in D,\ k\in\mathbb{N}\,\bigr\}.

S\mathcal{S} is a subset of H1(Tn)H^{1}(\mathbb{T}^{n}). As in part (a) of the proof of claim 3, ρ1/k⋆s∈Cper∞⊆Cper1\rho_{1/k}\star s\in C^{\infty}_{\mathrm{per}}\subseteq C^{1}_{\mathrm{per}} for every s∈Ds\in D and every kk, so The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (i) applies.

S\mathcal{S} is countable. By Properties of Periodic Convolution on the Torus (ii) the map ρ1/k⋆s\rho_{1/k}\star s depends only on the class [s][s], so there is a well-defined map Φ\Phi from {[s]:s∈D}×N\{[s]:s\in D\}\times\mathbb{N} to S\mathcal{S} with Φ([s],k)=[(ρ1/k⋆s)∣Q]\Phi([s],k)=[(\rho_{1/k}\star s)|_{Q}], and S\mathcal{S} is exactly the set of its values. The set N\mathbb{N} is countable by claim 1 of Basic Properties of Countable Sets, so the Cartesian product {[s]:s∈D}×N\{[s]:s\in D\}\times\mathbb{N} is countable by claim 1 of Products and Powers of Countable Sets, and S\mathcal{S} is countable by claim 4 of Basic Properties of Countable Sets.

S\mathcal{S} is dense. Let U∈H1(Tn)U\in H^{1}(\mathbb{T}^{n}) and let ε\varepsilon be a real number with 0<ε0<\varepsilon. Choose a representative uu of UU and, for each j∈[n]j\in[n], a representative gjg_{j} of ∂jU\partial_{j}U. By claim 3 applied with δ=1\delta=1, the sequence (Uk)k∈N(U_{k})_{k\in\mathbb{N}} converges to UU in H1(Tn)H^{1}(\mathbb{T}^{n}), so there is k∈Nk\in\mathbb{N} with dH1(Uk,U)<ε/2d_{H^{1}}(U_{k},U)<\varepsilon/2, where Uk=[(ρ1/k⋆u)∣Q]U_{k}=[(\rho_{1/k}\star u)|_{Q}]. Fix this kk for the rest of the argument.

For j∈[n]j\in[n], the map ∂jρ1/k\partial_{j}\rho_{1/k} is continuous and vanishes at every yy with Rk<∥y∥R_{k}<\lVert y\rVert, by Convolution of a Locally Integrable Function with a Compactly Supported Kernel (iv) applied to the kernel ρ1/k\rho_{1/k}, which is of class C1C^{1} by claim 2 of Euclidean Space Is Open in Itself. Let KjK_{j} be the integral of ∣∂jρ1/k∣|\partial_{j}\rho_{1/k}| over Rn\mathbb{R}^{n}, a nonnegative real number by that clause, and put

C=(1+∑j=1nKj2)1/2.C=\Bigl(1+\sum_{j=1}^{n}K_{j}^{2}\Bigr)^{1/2}.

The number 1+∑j=1nKj21+\sum_{j=1}^{n}K_{j}^{2} is positive: the sum is nonnegative by Finite Sums of Real Numbers: Nonnegativity, Domination by the Sum, and Limits (i) and 0<10<1 by claim 6 of Elementary Order Arithmetic in an Ordered Field, so claim 3 of Elementary Order Arithmetic in an Ordered Field applies. Hence 0<C0<C by Properties of Real Powers (i), and C2=1+∑j=1nKj2C^{2}=1+\sum_{j=1}^{n}K_{j}^{2} by (N1).

By Separability of the Lebesgue Spaces of the Torus (iii), applied with the positive real ε/(4C)\varepsilon/(4C), there is s∈Ds\in D with ∥[u]−[s]∥L2(Tn)≤ε/(4C)\bigl\lVert[u]-[s]\bigr\rVert_{L^{2}(\mathbb{T}^{n})}\le\varepsilon/(4C). Put w=u−sw=u-s, which lies in L2(Tn)\mathcal{L}^{2}(\mathbb{T}^{n}) by Minkowski's Inequality (ii), and note [w]=[u]−[s][w]=[u]-[s] by The Lebesgue Space Lᵖ(μ) (ii), so that ∥w∥2≤ε/(4C)\lVert w\rVert_{2}\le\varepsilon/(4C) by The Lebesgue Space Lᵖ(μ) (iii).

Put V=[(ρ1/k⋆s)∣Q]V=[(\rho_{1/k}\star s)|_{Q}], an element of S\mathcal{S}, and W=[(ρ1/k⋆w)∣Q]W=[(\rho_{1/k}\star w)|_{Q}]. By Properties of Periodic Convolution on the Torus (iii) applied with c=−1c=-1 we have ρ1/k⋆w=ρ1/k⋆u−ρ1/k⋆s\rho_{1/k}\star w=\rho_{1/k}\star u-\rho_{1/k}\star s as maps on Rn\mathbb{R}^{n}; all three maps lie in Cper∞C^{\infty}_{\mathrm{per}} and hence in CperC_{\mathrm{per}}, so Continuous Periodic Functions are Power-Integrable and Dense on the Torus (iii) gives W=Uk−VW=U_{k}-V in L2(Tn)L^{2}(\mathbb{T}^{n}). As UkU_{k} and VV lie in H1(Tn)H^{1}(\mathbb{T}^{n}), so does WW, and dH1(Uk,V)=∥W∥H1d_{H^{1}}(U_{k},V)=\lVert W\rVert_{H^{1}} by (N1).

We bound ∥W∥H1\lVert W\rVert_{H^{1}}. By Properties of Periodic Convolution on the Torus (i), ∂j(ρ1/k⋆w)=(∂jρ1/k)⋆w\partial_{j}(\rho_{1/k}\star w)=(\partial_{j}\rho_{1/k})\star w for every j∈[n]j\in[n], so by Elementary Properties of the Weak Partial Derivative on the Torus (i) we have ∂jW=[((∂jρ1/k)⋆w)∣Q]\partial_{j}W=[((\partial_{j}\rho_{1/k})\star w)|_{Q}]. Applying The Lebesgue Bound for Periodic Convolution and Convergence of Mollifications (i) with p=2p=2, first with the kernel ρ1/k\rho_{1/k}, whose mass is 11 as computed in part (b) of the proof of claim 3, and then with the kernel ∂jρ1/k\partial_{j}\rho_{1/k}, whose mass is KjK_{j},

∥W∥L2≤∥w∥2,∥∂jW∥L2≤Kj∥w∥2(j∈[n]).\lVert W\rVert_{L^{2}}\le\lVert w\rVert_{2},\qquad\lVert\partial_{j}W\rVert_{L^{2}}\le K_{j}\lVert w\rVert_{2}\quad(j\in[n]).

Squaring by Properties of Real Powers (v), using Properties of Real Powers (iii) to write (Kj∥w∥2)2=Kj2∥w∥22\bigl(K_{j}\lVert w\rVert_{2}\bigr)^{2}=K_{j}^{2}\lVert w\rVert_{2}^{2}, comparing the sums term by term by claim 1 of Comparison and Absolute Value Bounds for Finite Sums of Real Numbers, and adding the resulting inequalities, (N2) gives

∥W∥H12≤∥w∥22+∑j=1nKj2 ∥w∥22.\lVert W\rVert_{H^{1}}^{2}\le\lVert w\rVert_{2}^{2}+\sum_{j=1}^{n}K_{j}^{2}\,\lVert w\rVert_{2}^{2}.

By claims 1 and 4 of Properties of a Sum over a Finite Index Set, applied to the map j↦Kj2j\mapsto K_{j}^{2} on [n][n] with the scalar ∥w∥22\lVert w\rVert_{2}^{2}, the sum on the right equals ∥w∥22∑j=1nKj2\lVert w\rVert_{2}^{2}\sum_{j=1}^{n}K_{j}^{2}, so the right-hand side is (1+∑j=1nKj2)∥w∥22=C2∥w∥22\bigl(1+\sum_{j=1}^{n}K_{j}^{2}\bigr)\lVert w\rVert_{2}^{2}=C^{2}\lVert w\rVert_{2}^{2}, which equals (C∥w∥2)2\bigl(C\lVert w\rVert_{2}\bigr)^{2} by Properties of Real Powers (iii). By (N1), ∥W∥H1≤C∥w∥2\lVert W\rVert_{H^{1}}\le C\lVert w\rVert_{2}. Since 0≤C0\le C and ∥w∥2≤ε/(4C)\lVert w\rVert_{2}\le\varepsilon/(4C), claim 5 of Elementary Arithmetic in an Ordered Field gives C∥w∥2≤C⋅ε/(4C)=ε/4C\lVert w\rVert_{2}\le C\cdot\varepsilon/(4C)=\varepsilon/4.

Finally, dH1d_{H^{1}} is a metric by The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity (iii), so by the triangle inequality of Metric Space,

dH1(U,V)≤dH1(U,Uk)+dH1(Uk,V)<ε/2+ε/4,d_{H^{1}}(U,V)\le d_{H^{1}}(U,U_{k})+d_{H^{1}}(U_{k},V)<\varepsilon/2+\varepsilon/4,

and ε/2+ε/4<ε\varepsilon/2+\varepsilon/4<\varepsilon because ε/4\varepsilon/4 is positive. Hence dH1(U,V)<εd_{H^{1}}(U,V)<\varepsilon with V∈SV\in\mathcal{S}.

Since ε\varepsilon was an arbitrary positive real, condition 3 of Characterization of the Closure in a Metric Space by Open Balls shows that UU lies in the closure of S\mathcal{S}; since UU was arbitrary, S\mathcal{S} is dense in H1(Tn)H^{1}(\mathbb{T}^{n}) by Dense Subset of a Topological Space. Being also countable, it witnesses that (H1(Tn),dH1)(H^{1}(\mathbb{T}^{n}),d_{H^{1}}) is separable.

Proof of claim 6. Let S0={[w∣Q]:w∈Cper∞}\mathcal{S}_{0}=\{[w|_{Q}]:w\in C^{\infty}_{\mathrm{per}}\} as in claim 4. By The Lebesgue Bound for Periodic Convolution and Convergence of Mollifications (iii) with p=2p=2, S0\mathcal{S}_{0} is dense in L2(Tn)L^{2}(\mathbb{T}^{n}), so its closure in L2(Tn)L^{2}(\mathbb{T}^{n}) is L2(Tn)L^{2}(\mathbb{T}^{n}) by Dense Subset of a Topological Space. By claim 4, S0⊆H1(Tn)\mathcal{S}_{0}\subseteq H^{1}(\mathbb{T}^{n}), and H1(Tn)⊆L2(Tn)H^{1}(\mathbb{T}^{n})\subseteq L^{2}(\mathbb{T}^{n}) by The Sobolev Space of Once Weakly Differentiable Square-Integrable Classes on the Torus (i). Hence claim 5 of The Closure is the Smallest Closed Superset, applied in the topological space L2(Tn)L^{2}(\mathbb{T}^{n}), gives

L2(Tn)=cl⁡(S0)⊆cl⁡(H1(Tn))⊆L2(Tn),L^{2}(\mathbb{T}^{n})=\operatorname{cl}(\mathcal{S}_{0})\subseteq\operatorname{cl}\bigl(H^{1}(\mathbb{T}^{n})\bigr)\subseteq L^{2}(\mathbb{T}^{n}),

the closures being taken in L2(Tn)L^{2}(\mathbb{T}^{n}). Therefore the closure of H1(Tn)H^{1}(\mathbb{T}^{n}) in L2(Tn)L^{2}(\mathbb{T}^{n}) is L2(Tn)L^{2}(\mathbb{T}^{n}), that is, H1(Tn)H^{1}(\mathbb{T}^{n}) is dense in L2(Tn)L^{2}(\mathbb{T}^{n}) by Dense Subset of a Topological Space.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…