betaFeedback

WorksMathematics

Isometric Classification of Sobolev Spaces

Abstract

We show that the norm obtained by summing the L1L^1 norms of all weak derivatives through order kk, including the function itself, determines both nn and kk on the real Sobolev space Wk,1(Rn)W^{k,1}(\R^n). Every surjective linear isometry for this norm is given by a translation, a signed permutation of the coordinates and multiplication by a fixed sign c∈{−1,1}c\in\{-1,1\}.

Introduction

Throughout, functions are real-valued and operators are real-linear. For integers n,k≥1n,k \ge1, we equip Wk,1(Rn)W^{k,1}(\mathbb{R}^n) with the norm

Nn,k(f)=∑α∈An,k∥Dαf∥1,An,k={α∈N0n:∣α∣≤k}.(1)N_{n,k}(f)=\sum_{\alpha\in\mathcal{A}_{n,k}}\|D^\alpha f\|_1,\qquad\mathcal{A}_{n,k}=\{\alpha\in\mathbb{N}_0^n:|\alpha|\le k\}. \tag*{(1)}

Here ∣α∣=α1+⋯+αn|\alpha|=\alpha_1+\cdots+\alpha_n, and D0f=fD^0f=f. In particular, for k=1k=1 the derivative part is ∑j∥∂jf∥1\sum_j\|\partial_jf\|_1. A signed permutation of the axes permutes these terms. We shall see that the geometry of the norm recovers the axes even when neither the dimension nor the order of the target space is fixed in advance.

Theorem 1.1. Suppose that n,m,k,sn,m,k,s are positive integers and

T:(Wk,1(Rn),Nn,k)⟶(Ws,1(Rm),Nm,s)T:(W^{k,1}(\mathbb{R}^n),N_{n,k})\longrightarrow(W^{s,1}(\mathbb{R}^m),N_{m,s})

is a surjective linear isometry. Then m=nm=n and s=ks=k. There exist a signed permutation matrix PP, a vector b∈Rnb\in\mathbb{R}^n and a sign c∈{−1,1}c\in\{-1,1\} such that, for every f∈Wk,1(Rn)f\in W^{k,1}(\mathbb{R}^n),

(Tf)(y)=cf(Py+b)for almost every y∈Rn.(2)(Tf)(y)=cf(Py+b)\qquad\text{for almost every }y\in\mathbb{R}^n. \tag*{(2)}

A related question is whether, for n,m≥2n,m\ge2 and k,s≥1k,s\ge1, a Banach-space isomorphism

Wk,1(Rn)≅Ws,1(Rm)W^{k,1}(\mathbb{R}^n)\cong W^{s,1}(\mathbb{R}^m)

necessarily implies n=mn=m and k=sk=s. For background on the isomorphic classification of Sobolev spaces, see Pełczyński and Wojciechowski [1]. Diestel [4 Theorem 1] used an LpL^p extension theorem to classify surjective linear isometries of Wk,p(U)W^{k,p}(U) on bounded connected Lipschitz domains U⊂RnU\subset\mathbb{R}^n in the sense of (2). The result applies for 1≤p<∞1\le p<\infty with p≠2p\ne2, provided that the isometry maps Wk,p(U)∩C(U‾)W^{k,p}(U)\cap C(\overline{U}) onto itself. When k>n/pk>n/p, every function in Wk,p(U)W^{k,p}(U) has a continuous representative on U‾\overline{U} by Sobolev embedding, so this condition is automatic [4 Corollary 1]. M. Biegert and R. Nittka [3] proved that, for 1≤p<∞1\le p<\infty with p≠2p\ne2, every linear isometry T:W1,p(Ω1)→W1,p(Ω2)T:W^{1,p}(\Omega_1)\to W^{1,p}(\Omega_2) preserves disjointness. Here Ω1,Ω2⊂Rn\Omega_1,\Omega_2\subset\mathbb{R}^n are open and the Sobolev norms use the Euclidean length of the gradient. If both sets are bounded and TT is order bounded as a map from W1,p(Ω1)W^{1,p}(\Omega_1) into Lp(Ω2)L^p(\Omega_2), they also proved that there exist a measurable map ξ:Ω2→Ω1\xi:\Omega_2\to\Omega_1 and a real-valued weight gg on Ω2\Omega_2 such that, for each u∈W1,p(Ω1)u\in W^{1,p}(\Omega_1),

(Tu)(y)=g(y)u(ξ(y))for almost every y∈Ω2.(Tu)(y)=g(y)u(\xi(y))\qquad\text{for almost every }y\in\Omega_2.

Taking u=1u=1 gives g=T1g=T1 almost everywhere.

Printed: 2020 Mathematics Subject Classification. Primary 46B04, 46E35.

Preliminaries

We use the Lebesgue measure on Euclidean spaces and identify functions modulo null sets. The support of a measurable function will always mean the class of its nonzero set. Two functions are disjoint if their product vanishes almost everywhere.

Fact 2.1. [11 Theorem 6.19] If u∈Wloc1,1(Rn;R)u \in W^{1,1}_{\mathrm{loc}}(\mathbb{R}^{n};\mathbb{R}), then ∇u=0\nabla u = 0 almost everywhere on {u=0}\{u = 0\}. Consequently, for f∈Wk,1(Rn)f \in W^{k,1}(\mathbb{R}^{n}) every DαfD^{\alpha}f with ∣α∣≤k|\alpha| \le k vanishes almost everywhere on {f=0}\{f = 0\}.

Lemma 2.2. For f,g∈Wk,1(Rn)f,g \in W^{k,1}(\mathbb{R}^{n}) one has

fg=0 a.e.⟺Nn,k(f+g)+Nn,k(f−g)=2Nn,k(f)+2Nn,k(g).(3)fg = 0 \ \mathrm{a.e.} \quad\Longleftrightarrow\quad N_{n,k}(f+g) + N_{n,k}(f-g) = 2N_{n,k}(f) + 2N_{n,k}(g). \tag*{(3)}

In particular TT and T−1T^{-1} preserve disjointness.

To see this, suppose that fg=0fg = 0 almost everywhere. By Fact 2.1, DαfD^{\alpha}f vanishes almost everywhere on {f=0}\{f = 0\}, and DαgD^{\alpha}g vanishes almost everywhere on {g=0}\{g = 0\}. Since these two sets cover Rn\mathbb{R}^{n} up to a null set, DαfDαg=0D^{\alpha}fD^{\alpha}g = 0 almost everywhere for every ∣α∣≤k|\alpha| \le k. Lamperti’s equality criterion [10 Corollary 2.1] therefore gives

∥Dαf+Dαg∥1+∥Dαf−Dαg∥1=2∥Dαf∥1+2∥Dαg∥1.\lVert D^{\alpha}f + D^{\alpha}g\rVert_{1} + \lVert D^{\alpha}f - D^{\alpha}g\rVert_{1} = 2\lVert D^{\alpha}f\rVert_{1} + 2\lVert D^{\alpha}g\rVert_{1}.

Summing over α\alpha gives the norm identity in (3). Conversely, the triangle inequality bounds each term on the left by the corresponding term on the right. Equality in the sum thus implies equality for α=0\alpha= 0. Since D0f=fD^{0}f = f and D0g=gD^{0}g = g, Lamperti’s criterion gives fg=0fg = 0 almost everywhere. Both TT and T−1T^{-1} preserve the norm identity, so both preserve disjointness.

The Sobolev jet

We use the product of Lebesgue and counting measure on Rn×An,k\mathbb{R}^{n} \times A_{n,k}. A sheet is a set Rn×{α}\mathbb{R}^{n} \times\{\alpha\} with its index fixed. We use γ≤α\gamma\le\alpha for coordinatewise comparison, α!=∏iαi!\alpha! = \prod_i \alpha_i! and (αγ)=∏i(αiγi)\binom{\alpha}{\gamma} = \prod_i \binom{\alpha_i}{\gamma_i}. Our approach of embedding Sobolev spaces into scalar LpL^p spaces, extending the induced isometry, and applying Lamperti’s theorem follows the strategy of Diestel [4 Sections 2–3].

Lemma 3.1. Give An,kA_{n,k} counting measure. The map

Jn,kf(x,α)=Dαf(x),Jn,k:Wk,1(Rn)⟶L1(Rn×An,k),J_{n,k}f(x,\alpha) = D^{\alpha}f(x), \qquad J_{n,k}: W^{k,1}(\mathbb{R}^{n}) \longrightarrow L^{1}(\mathbb{R}^{n} \times A_{n,k}),

is an isometry onto a closed subspace. For g(x)=e−∣x∣2g(x) = e^{-|x|^{2}}, the jet Jn,kgJ_{n,k}g is nonzero almost everywhere. The ratios Jn,kf/Jn,kgJ_{n,k}f/J_{n,k}g, as ff ranges over Wk,1(Rn)W^{k,1}(\mathbb{R}^{n}), generate the full product σ\sigma-algebra of Rn×An,k\mathbb{R}^{n} \times A_{n,k} modulo null sets.

Proof. Integrating ∣Jn,kf∣\lvert J_{n,k}f\rvert over the sheet Rn×{α}\mathbb{R}^{n} \times\{\alpha\} gives ∥Dαf∥1\lVert D^{\alpha}f\rVert_{1}. Summing over the sheets gives precisely Nn,k(f)N_{n,k}(f), i.e., Jn,kJ_{n,k} is an isometry. Closedness now follows from completeness. We must find one jet which is nonzero almost everywhere on every sheet, and show that ratios by this jet distinguish the measurable sets of the whole product.

Let g(x)=e−∣x∣2g(x) = e^{-|x|^{2}}. We show by induction on ∣α∣|\alpha| that

Dαg(x)=Hα(x)g(x),D^{\alpha}g(x) = H_{\alpha}(x)g(x),

where HαH_{\alpha} is a polynomial whose unique term of highest total degree is (−2)∣α∣xα(-2)^{|\alpha|}x^{\alpha}. For α=0\alpha= 0 this holds with H0=1H_{0} = 1, since D0g=gD^{0}g = g.

Suppose the assertion holds for α\alpha and write d=∣α∣d = |\alpha|. Since ∂jg=−2xjg\partial_jg = -2x_jg, the product rule gives

Dα+ejg=∂j(Hαg)=(∂jHα)g+Hα(−2xjg)=(∂jHα−2xjHα)g.D^{\alpha+e_j}g = \partial_j(H_{\alpha}g) = (\partial_jH_{\alpha})g + H_{\alpha}(-2x_jg) = (\partial_jH_{\alpha} - 2x_jH_{\alpha})g.

Thus Hα+ej=∂jHα−2xjHαH_{\alpha+e_j} = \partial_jH_{\alpha} - 2x_jH_{\alpha} is again a polynomial. To identify its highest-degree term, note that ∂jHα\partial_jH_{\alpha} contains no term of degree d+1d+1. In −2xjHα-2x_jH_{\alpha}, the only term of degree d+1d+1 comes from multiplying the highest-degree term of HαH_{\alpha} by −2xj-2x_j. This gives

(−2xj)(−2)dxα=(−2)d+1xα+ej.(4)(-2x_j)(-2)^d x^{\alpha} = (-2)^{d+1}x^{\alpha+e_j}. \tag*{(4)}

All other terms have degree at most dd, so this term cannot cancel. This proves the induction step. In particular, HαH_\alpha is a nonzero polynomial of degree ∣α∣|\alpha| for every α\alpha. Remove its zero set for each of the finitely many α\alpha, and let GG be the common remainder. Since each HαH_\alpha is a nonzero polynomial, GG has full Lebesgue measure. On GG the jet of gg supplies a nonzero denominator on every sheet.

For a∈Qna \in\mathbb{Q}^n put ga(x)=ea⋅xg(x)g_a(x)=e^{a\cdot x}g(x). It is easy to see that ga∈Wk,1g_a \in W^{k,1}. On GG the product rule gives

Dαga(x)Dαg(x)=ea⋅xPx,α(a),Px,α(a)=∑γ≤α(αγ)aα−γDγg(x)Dαg(x).(5)\frac{D^\alpha g_a(x)}{D^\alpha g(x)}=e^{a\cdot x}P_{x,\alpha}(a), \qquad P_{x,\alpha}(a)=\sum_{\gamma\leq\alpha}\binom{\alpha}{\gamma}a^{\alpha-\gamma}\frac{D^\gamma g(x)}{D^\alpha g(x)}. \tag*{(5)}

Suppose two pairs (x,α)(x,\alpha) and (x′,α′)(x',\alpha') give the same quotients for all a∈Qna \in\mathbb{Q}^n. The expressions are continuous in aa, so they agree for all a∈Rna \in\mathbb{R}^n. If x≠x′x\ne x', set v=x−x′v=x-x' and substitute a=tva=tv in the equality of the two quotients. By (3.1), this gives

etv⋅xPx,α(tv)=etv⋅x′Px′,α′(tv)(t∈R).e^{tv\cdot x}P_{x,\alpha}(tv)=e^{tv\cdot x'}P_{x',\alpha'}(tv)\qquad(t\in\mathbb{R}).

Dividing both sides by etv⋅x′e^{tv\cdot x'} and using v⋅x−v⋅x′=v⋅(x−x′)=v⋅v=∣v∣2v\cdot x-v\cdot x'=v\cdot(x-x')=v\cdot v=|v|^2, we obtain

et∣v∣2Px,α(tv)=Px′,α′(tv)(t∈R).e^{t|v|^2}P_{x,\alpha}(tv)=P_{x',\alpha'}(tv)\qquad(t\in\mathbb{R}).

Both sides have a polynomial factor taking value 11 at t=0t=0. As t→+∞t\to+\infty, the left side grows in absolute value faster than any polynomial, a contradiction. Thus the quotients determine xx. With x=x′x=x' the polynomial factors themselves agree. The unique term of highest total degree in Px,α(a)P_{x,\alpha}(a) is aαg(x)/Dαg(x)a^\alpha g(x)/D^\alpha g(x), with nonzero coefficient. Comparing its multiindex with that on the other sheet yields α=α′\alpha=\alpha'.

The quotients in (3.1) therefore define an injective Borel map

Φ:X:=G×An,k⟶RQn,Φ(x,α)=(Dαga(x)Dαg(x))a∈Qn.\Phi:X:=G\times A_{n,k}\longrightarrow\mathbb{R}^{\mathbb{Q}^n},\qquad\Phi(x,\alpha)=\left(\frac{D^\alpha g_a(x)}{D^\alpha g(x)}\right)_{a\in\mathbb{Q}^n}.

By the Lusin–Souslin theorem in its standard Borel form [9 Corollary 15.2], the image Φ(X)\Phi(X) is Borel in RQn\mathbb{R}^{\mathbb{Q}^n}, and Φ\Phi is a Borel isomorphism from XX onto Φ(X)\Phi(X).

Let

S:=σ((x,α)⟼Dαga(x)Dαg(x):a∈Qn).\mathcal{S}:=\sigma\left((x,\alpha)\longmapsto\frac{D^\alpha g_a(x)}{D^\alpha g(x)}:a\in\mathbb{Q}^n\right).

We claim that S=B(X)\mathcal{S}=\mathcal{B}(X), where B(X)\mathcal{B}(X) denotes the Borel σ\sigma-algebra of XX. Since each quotient is Borel measurable, we immediately have S⊆B(X)\mathcal{S}\subseteq\mathcal{B}(X).

For the reverse inclusion, let E∈B(X)E\in\mathcal{B}(X). The Borel isomorphism property, together with the fact that Φ(X)\Phi(X) is Borel, implies that Φ(E)\Phi(E) is Borel in RQn\mathbb{R}^{\mathbb{Q}^n}. Moreover, injectivity gives

E=Φ−1(Φ(E)).E=\Phi^{-1}(\Phi(E)).

Since Qn\mathbb{Q}^n is countable, the Borel σ\sigma-algebra of RQn\mathbb{R}^{\mathbb{Q}^n} is generated by the coordinate projections πa\pi_a, a∈Qna\in\mathbb{Q}^n. For each such projection,

(πa∘Φ)(x,α)=Dαga(x)Dαg(x).(\pi_a\circ\Phi)(x,\alpha)=\frac{D^\alpha g_a(x)}{D^\alpha g(x)}.

Consequently, the inverse image under Φ\Phi of every Borel subset of RQn\mathbb{R}^{\mathbb{Q}^n} belongs to S\mathcal{S}. In particular, E=Φ−1(Φ(E))∈SE=\Phi^{-1}(\Phi(E))\in\mathcal{S}. Thus B(X)⊆S\mathcal{B}(X)\subseteq\mathcal{S}, proving that the coordinate quotients generate the Borel σ\sigma-algebra on G×An,kG\times A_{n,k}. The omitted part of the product is null. This proves the claimed generation modulo null sets. □

By Lemma 3.1 and Hardin’s extension theorem [7 Theorem 4.2], the isometry Jm,sTJn,k−1J_{m,s}T J_{n,k}^{-1} between the jet ranges extends to a surjective L1L^1 isometry UU of the ambient spaces satisfying

UJn,k=Jm,sT.(6)U J_{n,k}=J_{m,s}T. \tag*{(6)}

Put X=Rn×An,kX=\mathbb{R}^n\times A_{n,k} and Y=Rm×Am,sY=\mathbb{R}^m\times A_{m,s}, and denote by μ\mu and ν\nu their respective products of Lebesgue measure and counting measure. Write MX\mathcal{M}_X for the algebra of measurable subsets of $X modulo μ\mu-null sets. Thus [E]=[F][E]=[F] in MX\mathcal{M}_X precisely when μ(E△F)=0\mu(E\mathbin{\triangle}F)=0. Define MY\mathcal{M}_Y in the same way using ν\nu.

By Lamperti’s representation theorem [10 Theorem 3.1], the surjective linear isometry U:L1(X,μ)→L1(Y,ν)U:L^1(X,\mu)\to L^1(Y,\nu) admits a representation

Uu=h τ^uUu=h\,\widehat{\tau}u

where

τ:MX⟶MY\tau:\mathcal{M}_X\longrightarrow\mathcal{M}_Y

is a Boolean σ\sigma-isomorphism and τ^\widehat{\tau} is the induced transformation on measurable functions, characterized by

τ^1E=1Fwhenever [F]=τ([E]).\widehat{\tau}1_E=1_F\qquad\text{whenever }[F]=\tau([E]).

Surjectivity implies that h≠0h\ne0 ν\nu-almost everywhere. Indeed, every function in the range of UU vanishes almost everywhere on Z:={h=0}Z:=\{h=0\}. If ν(Z)>0\nu(Z)>0, σ\sigma-finiteness provides a measurable set B⊆ZB\subseteq Z with 0<ν(B)<∞0<\nu(B)<\infty. Then 1B∈L1(Y,ν)1_B\in L^1(Y,\nu) cannot belong to the range of UU, contradicting surjectivity.

By construction, the induced transformation transports measurable supports according to τ\tau:

ess supp⁡(τ^u)=τ(ess supp⁡u),ess supp⁡u:=[{u≠0}].\operatorname{ess\,supp}(\widehat{\tau}u)=\tau(\operatorname{ess\,supp}u),\qquad\operatorname{ess\,supp}u:=[\{u\ne0\}].

Since multiplication by hh does not change the measurable support, we conclude that

ess supp⁡(Uu)=τ(ess supp⁡u)(u∈L1(X,μ)).\operatorname{ess\,supp}(Uu)=\tau(\operatorname{ess\,supp}u)\qquad(u\in L^1(X,\mu)).

Recovering the spatial algebra and the finite fibers

Let (Ei)i∈I(E_i)_{i\in I} be a family of measurable subsets of (X,μ)(X,\mu). Its essential union is the least measurable upper bound modulo null sets: it is represented by a measurable set EE such that

μ(Ei∖E)=0(i∈I),\mu(E_i\setminus E)=0\qquad(i\in I),

and whenever a measurable set FF satisfies μ(Ei∖F)=0\mu(E_i\setminus F)=0 for every i∈Ii\in I, one has μ(E∖F)=0\mu(E\setminus F)=0. Such an EE is unique modulo null sets. On a σ\sigma-finite space, the essential union exists and is represented by the union of a countable subfamily [5 Proposition 215B(iv)].

Lemma 4.1. Let F⊂Ws,1(Rm)F\subset W^{s,1}(\mathbb{R}^m) be stable under multiplication by Cc∞(Rm)C^\infty_c(\mathbb{R}^m) functions. If EE is the essential union of the supports of the functions in FF, then E×Am,sE\times\mathcal{A}_{m,s} is the essential union of the supports of their jets.

Proof. For f∈Ff\in F, Fact 2.1 gives Dβf=0D^\beta f=0 almost everywhere off EE for every β\beta. This proves containment in E×Am,sE\times\mathcal{A}_{m,s}. For the reverse containment, the difficulty is that f(y)≠0f(y)\ne0 need not imply Dβf(y)≠0D^\beta f(y)\ne0. We may, however, replace ff by an appropriate smooth multiple without leaving FF.

Choose fℓ∈Ff_\ell\in F whose supports realize EE, the case of empty EE being immediate. For r∈Nr\in\mathbb{N} take χr∈Cc∞(Rm)\chi_r\in C^\infty_c(\mathbb{R}^m) equal to one on B(0,r)B(0,r). For every t∈Qmt\in\mathbb{Q}^m the multiplier y↦et⋅yχr(y)y\mapsto e^{t\cdot y}\chi_r(y) is smooth with compact support. Fix r,ℓ∈Nr,\ell\in\mathbb{N}, t∈Qmt\in\mathbb{Q}^m, and β∈Am,s\beta\in\mathcal{A}_{m,s}. Since χr=1\chi_r=1 on B(0,r)B(0,r), the weak product rule gives

Dβ(et⋅(⋅)χrfℓ)(y)=et⋅y∑γ≤β(βγ)tβ−γDγfℓ(y).D^\beta\left(e^{t\cdot(\cdot)}\chi_r f_\ell\right)(y) =e^{t\cdot y}\sum_{\gamma\le\beta}\binom{\beta}{\gamma}t^{\beta-\gamma}D^\gamma f_\ell(y).

for almost every y∈B(0,r)y\in B(0,r). The exceptional set may depend on r,ℓ,t,βr,\ell,t,\beta. Since there are only countably many such choices, the union of all exceptional sets is a null set N⊂RmN\subset\mathbb{R}^m. Thus, for each y∉Ny\notin N, the identity holds for every ℓ∈N\ell\in\mathbb{N}, t∈Qmt\in\mathbb{Q}^m, and β∈Am,s\beta\in\mathcal{A}_{m,s} whenever r>∣y∣r>|y|. If fℓ(y)≠0f_\ell(y)\ne0, the coefficient of tβt^\beta in the sum is fℓ(y)f_\ell(y), so the polynomial in tt is nonzero for some rational tt. Choosing r>∣y∣r>|y|, we have found a member of FF whose jet is nonzero at (y,β)(y,\beta). Since r∈Nr\in\mathbb{N} was arbitrary and almost every y∈E∖Ny\in E\setminus N satisfies fp(y)≠0f_p(y)\ne0 for some p∈Np\in\mathbb{N}, we obtain the conclusion.

For a family F⊂Wk,1(Rn)F \subset W^{k,1}(\mathbb{R}^{n}), define its disjoint complement in Wk,1(Rn)W^{k,1}(\mathbb{R}^{n}) by

Fd={u∈Wk,1(Rn):uf=0 a.e. for every f∈F}.F^{d}=\{u\in W^{k,1}(\mathbb{R}^{n}):uf=0\ \text{a.e. for every }f\in F\}.

For families in Ws,1(Rm)W^{s,1}(\mathbb{R}^{m}), the disjoint complement is defined analogously within Ws,1(Rm)W^{s,1}(\mathbb{R}^{m}). For open O⊂RnO\subset\mathbb{R}^{n} let XOX_O consist of the Sobolev functions vanishing almost everywhere outside OO. For an open ball BB one has

XB=(XRn∖B‾)d.(7)X_B=\left(X_{\mathbb{R}^{n}\setminus\overline{B}}\right)^{d}. \tag*{(7)}

Indeed, membership in XBX_B implies disjointness from the exterior family. Conversely, choose smooth bumps supported in the exterior whose positive sets cover it, using a countable family of balls with closure in the exterior. A function disjoint from these bumps vanishes almost everywhere there. The only remaining set outside BB is ∂B\partial B, which is null.

By surjectivity and Lemma 2.2, T(Fd)=(TF)dT(F^{d})=(TF)^{d}. In particular TXBTX_B is a disjoint complement by (7). Such a complement is stable under multiplication by smooth compactly supported functions, since multiplying uu by a smooth factor preserves every relation uf=0uf=0. On the source side, XBX_B already has this stability and its essential support union is BB, as is seen using bumps inside BB. Write EBE_B for the essential support union of TXBTX_B. Fix f∈XBf\in X_B. Applying Lamperti’s support formula τ(ess supp⁡u)=ess supp⁡(Uu)\tau(\operatorname{ess\,supp}u)=\operatorname{ess\,supp}(Uu) to the function u=Jn,kfu=J_{n,k}f gives

τ(ess supp⁡(Jn,kf))=ess supp⁡(U(Jn,kf)).\tau\left(\operatorname{ess\,supp}(J_{n,k}f)\right)=\operatorname{ess\,supp}\left(U(J_{n,k}f)\right).

By (3.2), U(Jn,kf)U(J_{n,k}f) and Jm,s(Tf)J_{m,s}(Tf) agree almost everywhere on Rm×Am,s\mathbb{R}^{m}\times A_{m,s}. Thus, for every f∈XBf\in X_B,

τ(ess supp⁡(Jn,kf))=ess supp⁡(Jm,s(Tf)).\tau\left(\operatorname{ess\,supp}(J_{n,k}f)\right)=\operatorname{ess\,supp}\left(J_{m,s}(Tf)\right).

We now take the essential union over f∈XBf\in X_B. Since τ\tau is an order isomorphism of the measure algebras, it carries the least upper bound of the supports on the left to the least upper bound of their images. By Lemma 4.1, the essential union of the supports of Jn,kfJ_{n,k}f for f∈XBf\in X_B is B×An,kB\times A_{n,k}. As ff ranges over XBX_B, TfTf ranges over TXBTX_B, so the same lemma on the target gives EB×Am,sE_B\times A_{m,s} as the essential union of the supports of Jm,s(Tf)J_{m,s}(Tf). Consequently,

τ(B×An,k)=EB×Am,smodulo null sets.\tau(B\times A_{n,k})=E_B\times A_{m,s}\quad\text{modulo null sets}.

We claim that the same formula holds for every Lebesgue measurable set C⊂RnC\subset\mathbb{R}^{n}. Indeed, the sets CC for which there exists a measurable D⊂RmD\subset\mathbb{R}^{m} satisfying

τ(C×An,k)=D×Am,smodulo null sets\tau(C\times A_{n,k})=D\times A_{m,s}\quad\text{modulo null sets}

form a σ\sigma-algebra which contains the balls with centers in Qn\mathbb{Q}^{n} and positive rational radii, so it contains every Borel subset of Rn\mathbb{R}^{n}. Every Lebesgue measurable set agrees with a Borel set modulo a null set, and taking the product with the finite set An,kA_{n,k} preserves this agreement. This proves the claim.

Conversely, let D⊂RmD\subset\mathbb{R}^{m} be Lebesgue measurable. Applying the same argument to T−1T^{-1} and its extension on jets U−1U^{-1} gives a measurable set C⊂RnC\subset\mathbb{R}^{n} such that

τ−1(D×Am,s)=C×An,kmodulo null sets.\tau^{-1}(D\times A_{m,s})=C\times A_{n,k}\quad\text{modulo null sets}.

Thus every set of the form D×Am,sD\times A_{m,s} is the image under τ\tau of a set of the form C×An,kC\times A_{n,k}, modulo null sets.

Lemma 4.2 (The permutation of the derivative indices). The isomorphism τ\tau on the jet spaces acts by inverse images under a map of the form

(y,β)⟼(φ(y),πy(β)).(y,\beta)\longmapsto(\varphi(y),\pi_y(\beta)).

Here φ:Rm→Rn\varphi:\mathbb{R}^{m}\to\mathbb{R}^{n} is a nonsingular isomorphism modulo null sets and πy:Am,s→An,k\pi_y:A_{m,s}\to A_{n,k} is a bijection for almost every yy.

Proof. The preceding argument gives an isomorphism between the Lebesgue measure algebras of Rn\mathbb{R}^{n} and Rm\mathbb{R}^{m}, determined by

τ(C×An,k)=D×Am,s.\tau(C \times\mathcal{A}_{n,k}) = D \times\mathcal{A}_{m,s}.

By [6 Theorem 343B(v) and Proposition 343F], the isomorphism between the Lebesgue measure algebras of Rn\mathbb{R}^{n} and Rm\mathbb{R}^{m} induced by τ\tau is represented by mutually inverse nonsingular measurable maps

φ:Rm→Rn,ψ:Rn→Rm\varphi: \mathbb{R}^{m} \to\mathbb{R}^{n}, \qquad\psi: \mathbb{R}^{n} \to\mathbb{R}^{m}

modulo null sets. In particular, for all Lebesgue measurable sets C⊂RnC \subset\mathbb{R}^{n} and D⊂RmD \subset\mathbb{R}^{m},

τ(C×An,k)=φ−1(C)×Am,s,\tau(C \times\mathcal{A}_{n,k}) = \varphi^{-1}(C) \times\mathcal{A}_{m,s},
τ−1(D×Am,s)=ψ−1(D)×An,k,\tau^{-1}(D \times\mathcal{A}_{m,s}) = \psi^{-1}(D) \times\mathcal{A}_{n,k},

modulo null sets, and

ψ(φ(y))=yfor almost every y∈Rm,φ(ψ(x))=xfor almost every x∈Rn.\psi(\varphi(y)) = y \quad\text{for almost every } y \in\mathbb{R}^{m}, \qquad\varphi(\psi(x)) = x \quad\text{for almost every } x \in\mathbb{R}^{n}.

For each α∈An,k\alpha\in\mathcal{A}_{n,k}, choose a measurable representative EαE_{\alpha} of τ(Rn×{α})\tau(\mathbb{R}^{n} \times\{\alpha\}). Since τ\tau preserves unions and intersections, the sets EαE_{\alpha} are pairwise disjoint modulo null sets and cover Rm×Am,s\mathbb{R}^{m} \times\mathcal{A}_{m,s} modulo a null set. There are only finitely many α\alpha, so we may modify these sets on a single null set to obtain an actual partition. Define πy:Am,s→An,k\pi_y : \mathcal{A}_{m,s} \to\mathcal{A}_{n,k} by

πy(β)=α⟺(y,β)∈Eα.\pi_y(\beta) = \alpha\quad\Longleftrightarrow\quad(y,\beta) \in E_{\alpha}.

For every measurable C⊂RnC \subset\mathbb{R}^{n}, we have

C×{α}=(C×An,k)∩(Rn×{α}),C \times\{\alpha\} = (C \times\mathcal{A}_{n,k}) \cap(\mathbb{R}^{n} \times\{\alpha\}),

and hence

τ(C×{α})=(φ−1(C)×Am,s)∩Eαmodulo null sets.\tau(C \times\{\alpha\}) = (\varphi^{-1}(C) \times\mathcal{A}_{m,s}) \cap E_{\alpha} \quad\text{modulo null sets}.

The expression on the right is precisely the inverse image of C×{α}C \times\{\alpha\} under the measurable map

F(y,β)=(φ(y),πy(β)).F(y,\beta) = (\varphi(y),\pi_y(\beta)).

Every measurable subset of Rn×An,k\mathbb{R}^{n} \times\mathcal{A}_{n,k} agrees modulo a null set with a finite union of sets Cα×{α}C_{\alpha} \times\{\alpha\}. Thus FF represents τ\tau by inverse images.

Applying the same construction to τ−1\tau^{-1} gives a measurable map

G(x,α)=(ψ(x),ρx(α)),ρx:An,k→Am,s,G(x,\alpha) = (\psi(x),\rho_x(\alpha)), \qquad\rho_x : \mathcal{A}_{n,k} \to\mathcal{A}_{m,s},

which represents τ−1\tau^{-1} by inverse images. Consequently, G∘FG \circ F and F∘GF \circ G induce the identity on the respective measure algebras. Each composition therefore equals the identity almost everywhere. Indeed, since G∘FG \circ F induces the identity on the measure algebra of Rm×Am,s\mathbb{R}^{m} \times\mathcal{A}_{m,s}, for every measurable set D⊂Rm×Am,sD \subset\mathbb{R}^{m} \times\mathcal{A}_{m,s} the symmetric difference

ND=(G∘F)−1(D)△DN_D = (G \circ F)^{-1}(D) \mathbin{\triangle} D

is a null set with respect to Lebesgue measure times counting measure. Thus, for every z∉NDz \notin N_D,

z∈D⟺G(F(z))∈D.z \in D \quad\Longleftrightarrow\quad G(F(z)) \in D.

The exceptional set NDN_D may depend on DD. To obtain this equivalence simultaneously for a family of sets that distinguishes every pair of points, consider

D={B(q,r)×{β}:q∈Qm, r∈Q, r>0, β∈Am,s}.\mathcal{D} = \{B(q,r) \times\{\beta\} : q \in\mathbb{Q}^{m},\ r \in\mathbb{Q},\ r > 0,\ \beta\in\mathcal{A}_{m,s}\}.

Here B(q,r)B(q,r) denotes the open ball with center qq and radius rr. This family is countable, since the rational centers and radii form countable sets and Am,s\mathcal{A}_{m,s} is finite. It also separates points. The countable union

N=⋃D∈DNDN = \bigcup_{D \in\mathcal{D}} N_D

is a null set. Consequently, for every z∉Nz \notin N and every D∈DD \in\mathcal{D},

z∈D⟺G(F(z))∈D.z \in D \quad\Longleftrightarrow\quad G(F(z)) \in D.

If G(F(z))≠zG(F(z)) \ne z, the separating property of D\mathcal{D} gives a set D∈DD \in\mathcal{D} containing exactly one of these two points, contradicting this equivalence. Hence G(F(z))=zG(F(z)) = z for every z∉Nz \notin N, which proves

G∘F=id⁡Rm×Am,sG \circ F = \operatorname{id}_{\mathbb{R}^{m} \times\mathcal{A}_{m,s}}

almost everywhere.

Applying the same argument to F∘GF \circ G, using rational balls in Rn\mathbb{R}^{n} and the finite index set An,k\mathcal{A}_{n,k}, gives

F∘G=id⁡Rn×An,kF \circ G = \operatorname{id}_{\mathbb{R}^{n} \times\mathcal{A}_{n,k}}

almost everywhere.

Since Am,s\mathcal{A}_{m,s} is finite, for almost every y∈Rmy \in\mathbb{R}^{m} we have

G(F(y,β))=(y,β)G(F(y,\beta)) = (y,\beta)

for every β∈Am,s\beta\in\mathcal{A}_{m,s}.

Comparing the second coordinates gives

ρφ(y)∘πy=id⁡Am,s.\rho_{\varphi(y)} \circ\pi_y = \operatorname{id}_{\mathcal{A}_{m,s}}.

Similarly, for almost every x∈Rnx \in\mathbb{R}^{n},

F(G(x,α))=(x,α)F(G(x,\alpha)) = (x,\alpha)

for every α∈An,k\alpha\in\mathcal{A}_{n,k}, so

πψ(x)∘ρx=id⁡An,k.\pi_{\psi(x)} \circ\rho_x = \operatorname{id}_{\mathcal{A}_{n,k}}.

By nonsingularity of φ\varphi, the latter identity remains valid for almost every yy after substituting x=φ(y)x = \varphi(y). Using ψ(φ(y))=y\psi(\varphi(y)) = y almost everywhere, we obtain

πy∘ρφ(y)=id⁡An,k.\pi_y \circ\rho_{\varphi(y)} = \operatorname{id}_{\mathcal{A}_{n,k}}.

Thus πy\pi_y and ρφ(y)\rho_{\varphi(y)} are mutually inverse for almost every yy, which proves the asserted bijectivity. □\square

The weighted-composition representation of the ambient isometry can now be written, for u∈L1(Rn×An,k)u \in L^1(\mathbb{R}^{n} \times\mathcal{A}_{n,k}), as

(Uu)(y,β)=hβ(y)u(φ(y),πy(β)).(Uu)(y,\beta) = h_\beta(y)u(\varphi(y),\pi_y(\beta)).

For a=Jn,kga = J_{n,k}g from Lemma 3.1, the weight on sheet β\beta is

hβ(y)=(Ua)(y,β)a(φ(y),πy(β)).h_\beta(y) = \frac{(Ua)(y,\beta)}{a(\varphi(y),\pi_y(\beta))}.

Both aa and UaUa have full support, so every hβh_\beta is finite and nonzero almost everywhere. Taking u=Jn,kfu = J_{n,k}f and using (3.2) gives for almost all y∈Rmy \in\mathbb{R}^{m}:

Dβ(Tf)(y)=hβ(y)Dπy(β)f(φ(y)),β∈Am,s.(8)D^\beta(Tf)(y) = h_\beta(y)D^{\pi_y(\beta)}f(\varphi(y)), \qquad\beta\in\mathcal{A}_{m,s}. \tag*{(8)}

Also ∣An,k∣=∣Am,s∣|\mathcal{A}_{n,k}| = |\mathcal{A}_{m,s}|. Appending the coordinate k−∣α∣k - |\alpha| to α∈An,k\alpha\in\mathcal{A}_{n,k} gives a bijection with the (n+1)(n+1)-tuples of nonnegative integers of sum kk. The fiber count is therefore

(n+kk)=(m+ss).(9)\binom{n+k}{k} = \binom{m+s}{s}. \tag*{(9)}

In (4.2) the exceptional null set may depend on ff. We shall compare this identity only for countable families of tests.

The zeroth and first derivative levels

So far the bijection πy\pi_y may interchange derivatives of different orders. Even the cardinality in (4.3) does not determine nn and kk separately. We first identify the index 00, and then identify the indices of length one.

Lemma 5.1. One has πy(0)=0\pi_y(0) = 0 almost everywhere. In particular, if h=h0h = h_0, then for every f∈Wk,1(Rn)f \in W^{k,1}(\mathbb{R}^{n}),

(Tf)(y)=h(y)f(φ(y))(10)(Tf)(y) = h(y)f(\varphi(y)) \tag*{(10)}

for almost every yy.

Proof. Take a ball BB with rational center and radius. Choose a cutoff θ∈Cc∞(Rn)\theta\in C^\infty_c(\mathbb{R}^n) equal to one on BB. Consider

E=φ−1(B)∩{y:πy(0)≠0}.E=\varphi^{-1}(B)\cap\{y:\pi_y(0)\ne0\}.

On EE the target value TθT\theta reads a positive-order derivative of θ\theta by (8). Hence Tθ=0T\theta=0 almost everywhere on EE. By locality all its derivatives vanish there as well.

On the other hand, partition EE into the finitely many sets Eβ=E∩{y:πy(β)=0}E_\beta=E\cap\{y:\pi_y(\beta)=0\}. They cover EE up to a null set because πy\pi_y is onto. On EβE_\beta, the same jet formula gives DβTθ=hβθ∘φ=hβD^\beta T\theta=h_\beta\theta\circ\varphi=h_\beta. Since hβh_\beta is nonzero almost everywhere, EβE_\beta is null. We conclude that EE is null. Choosing a cover of Rn\mathbb{R}^n by balls with rational centers and positive rational radii gives the conclusion. □

Lemma 5.2. For 1≤j≤m1\le j\le m, one has ∣πy(ej)∣=1|\pi_y(e_j)|=1 almost everywhere. Consequently m=nm=n and s=ks=k.

Proof. Fix jj, a rational ball B⊂RnB\subset\mathbb{R}^n, and a cutoff θ\theta as in the preceding proof. Applying (10) to n+1n+1 functions gives

v0=Tθ=h,vi=T(xiθ)=hφi(1≤i≤n)a.e. on φ−1(B).v_0=T\theta=h,\qquad v_i=T(x_i\theta)=h\varphi_i\quad(1\le i\le n)\quad\text{a.e. on }\varphi^{-1}(B).

All the viv_i belong to W1,1W^{1,1}. These identities allow us to compute ∂jTf\partial_jTf locally without assuming weak differentiability of hh or of φ\varphi.

Define the vector-valued function v=(v1,…,vn)v=(v_1,\ldots,v_n) and fix ε>0\varepsilon>0. We work on Eε=φ−1(B)∩{∣h∣≥ε}E_\varepsilon=\varphi^{-1}(B)\cap\{|h|\ge\varepsilon\}. Fix f∈C∞(Rn)f\in C^\infty(\mathbb{R}^n). Choose a smooth cutoff ηε\eta_\varepsilon on [0,∞)[0,\infty) equal to zero on [0,ε/2][0,\varepsilon/2] and to one on [ε,∞)[\varepsilon,\infty) with 0≤ηε≤10\le\eta_\varepsilon\le1. For (a,b)∈R×Rn(a,b)\in\mathbb{R}\times\mathbb{R}^n put

Fε(a,b)={ηε(∣a∣) af(b/a),a≠0,0,a=0.F_\varepsilon(a,b)= \begin{cases} \eta_\varepsilon(|a|)\,a f(b/a), & a\ne0,\\ 0, & a=0. \end{cases}

To apply the Sobolev chain rule [2 Theorem 4.2], we show that FεF_\varepsilon is globally Lipschitz by establishing the estimate

∣DFε(a,b)[u,w]∣≤Cε(∣u∣+∣w∣)(11)|DF_\varepsilon(a,b)[u,w]|\le C_\varepsilon(|u|+|w|) \tag*{(11)}

for all (a,b)∈R×Rn(a,b)\in\mathbb{R}\times\mathbb{R}^n and all directions (u,w)∈R×Rn(u,w)\in\mathbb{R}\times\mathbb{R}^n. The constant may depend on ff, ε\varepsilon and the chosen cutoff, but is independent of a,b,u,wa,b,u,w. First fix a≠0a\ne0. To compute the derivative in the direction (u,w)(u,w), consider the curve

t⟼(a+tu,b+tw)t\longmapsto(a+tu,b+tw)

and set

z(t)=b+twa+tu,z=z(0)=ba.z(t)=\frac{b+tw}{a+tu},\qquad z=z(0)=\frac{b}{a}.

The denominator is nonzero for all sufficiently small tt. Applying the quotient rule to each coordinate gives

z′(0)=aw−uba2=w−uza.z'(0)=\frac{aw-ub}{a^2}=\frac{w-uz}{a}.

The product rule and the chain rule therefore yield

ddt∣t=0[(a+tu)f(b+twa+tu)]=uf(z)+a∇f(z)⋅z′(0)=uf(z)+a∇f(z)⋅w−uza=uf(z)+∇f(z)⋅w−uz⋅∇f(z)=(f(z)−z⋅∇f(z))u+∇f(z)⋅w.\begin{aligned} \left.\frac{d}{dt}\right|_{t=0}\left[(a+tu)f\left(\frac{b+tw}{a+tu}\right)\right] =uf(z)+a\nabla f(z)\cdot z'(0) \\ &=uf(z)+a\nabla f(z)\cdot\frac{w-uz}{a} \\ &=uf(z)+\nabla f(z)\cdot w-uz\cdot\nabla f(z) \\ &=(f(z)-z\cdot\nabla f(z))u+\nabla f(z)\cdot w. \end{aligned}

We also differentiate the factor ηε(∣a∣)\eta_\varepsilon(|a|). Since a≠0a\ne0,

ddt∣t=0∣a+tu∣=sgn⁡(a)u.\left.\frac{d}{dt}\right|_{t=0}|a+tu|=\operatorname{sgn}(a)u.

Another application of the product rule gives

DFε(a,b)[u,w]=ηε(∣a∣)[(f(z)−z⋅∇f(z))u+∇f(z)⋅w]+ηε′(∣a∣)sgn⁡(a) u af(z).DF_{\varepsilon}(a,b)[u,w]=\eta_{\varepsilon}(|a|)\left[(f(z)-z\cdot\nabla f(z))u+\nabla f(z)\cdot w\right] +\eta_{\varepsilon}'(|a|)\operatorname{sgn}(a)\,u\,a f(z).

Using asgn⁡(a)=∣a∣a\operatorname{sgn}(a)=|a|, the last summand becomes

ηε′(∣a∣)∣a∣f(z) u.\eta_{\varepsilon}'(|a|)|a|f(z)\,u.

To bound these expressions, choose R>0R>0 such that supp⁡f⊂B(0,R)\operatorname{supp}f\subset B(0,R) and put

M0=sup⁡z∈Rn∣f(z)∣,M1=sup⁡z∈Rn∣∇f(z)∣.M_0=\sup_{z\in\mathbb{R}^n}|f(z)|,\qquad M_1=\sup_{z\in\mathbb{R}^n}|\nabla f(z)|.

Both constants are finite. Moreover, ∇f(z)=0\nabla f(z)=0 outside B(0,R)B(0,R), so for every z∈Rnz\in\mathbb{R}^n,

∣z⋅∇f(z)∣≤RM1,∣f(z)−z⋅∇f(z)∣≤M0+RM1.|z\cdot\nabla f(z)|\le RM_1,\qquad|f(z)-z\cdot\nabla f(z)|\le M_0+RM_1.

The derivative ηε′(∣a∣)\eta_{\varepsilon}'(|a|) can be nonzero only when ε/2≤∣a∣≤ε\varepsilon/2\le|a|\le\varepsilon. Consequently,

∣ηε′(∣a∣)∣ ∣a∣ ∣f(z)∣ ∣u∣≤ε∥ηε′∥∞M0∣u∣.|\eta_{\varepsilon}'(|a|)|\,|a|\,|f(z)|\,|u|\le\varepsilon\|\eta_{\varepsilon}'\|_{\infty}M_0|u|.

Combining these estimates and using 0≤ηε≤10\le\eta_{\varepsilon}\le1, we obtain

∣DFε(a,b)[u,w]∣≤(M0+RM1+ε∥ηε′∥∞M0)∣u∣+M1∣w∣.|DF_{\varepsilon}(a,b)[u,w]|\le(M_0+RM_1+\varepsilon\|\eta_{\varepsilon}'\|_{\infty}M_0)|u|+M_1|w|.

This proves the desired estimate (∗)(*) for a≠0a\ne0. For ∣a∣<ε/2|a|<\varepsilon/2, the function Fε(a,b)F_{\varepsilon}(a,b) vanishes for every b∈Rnb\in\mathbb{R}^n. Hence FεF_{\varepsilon} is also smooth in a neighborhood of every point (0,b)(0,b), with derivative zero there. Thus (∗)(*) holds on all of Rn+1\mathbb{R}^{n+1}.

Integrating the bounded derivative along line segments shows that FεF_{\varepsilon} is globally Lipschitz. Since Fε(0,0)=0F_{\varepsilon}(0,0)=0, the Sobolev chain rule [2 Theorem 4.2] gives

Fε∘(v0,v)∈W1,1(Rm).F_{\varepsilon}\circ(v_0,v)\in W^{1,1}(\mathbb{R}^m).

For almost every y∈Eεy\in E_{\varepsilon}, we have

(Fε∘(v0,v))(y)=ηε(∣v0(y)∣)v0(y)f(v(y)v0(y))=ηε(∣h(y)∣)h(y)f(φ(y))=h(y)f(φ(y))=(Tf)(y).(F_{\varepsilon}\circ(v_0,v))(y)=\eta_{\varepsilon}(|v_0(y)|)v_0(y)f\left(\frac{v(y)}{v_0(y)}\right) =\eta_{\varepsilon}(|h(y)|)h(y)f(\varphi(y)) =h(y)f(\varphi(y)) =(Tf)(y).

Here we used v0=hv_0=h, v/v0=φv/v_0=\varphi, and ηε(∣h∣)=1\eta_{\varepsilon}(|h|)=1 on EεE_{\varepsilon}, since ∣h∣≥ε|h|\ge\varepsilon there. Both TfTf and Fε∘(v0,v)F_{\varepsilon}\circ(v_0,v) belong to W1,1(Rm)W^{1,1}(\mathbb{R}^m), and their difference vanishes almost everywhere on EεE_{\varepsilon}. By locality, the first weak derivatives of this difference also vanish almost everywhere on EεE_{\varepsilon}. Hence

∂j(Tf)(y)=∂j(Fε∘(v0,v))(y)\partial_j(Tf)(y)=\partial_j(F_{\varepsilon}\circ(v_0,v))(y)

for almost every y∈Eεy\in E_{\varepsilon}.

Combining this identity with the Sobolev chain rule gives

∂j(Tf)(y)=DFε(v0(y),v(y))[∂jv0(y),∂jv(y)]\partial_j(Tf)(y)=DF_{\varepsilon}(v_0(y),v(y))[\partial_jv_0(y),\partial_jv(y)]

for almost every y∈Eεy\in E_{\varepsilon}.

On EεE_{\varepsilon}, we have

ηε(∣v0(y)∣)=1,ηε′(∣v0(y)∣)=0.\eta_{\varepsilon}(|v_0(y)|)=1,\qquad\eta_{\varepsilon}'(|v_0(y)|)=0.

Consequently, the formula for DFε(a,b)[u,w]DF_{\varepsilon}(a,b)[u,w] delivers

∂j(Tf)(y)=(f(φ(y))−φ(y)⋅∇f(φ(y)))∂jv0(y)+∇f(φ(y))⋅∂jv(y).\partial_j(Tf)(y)=(f(\varphi(y))-\varphi(y)\cdot\nabla f(\varphi(y)))\partial_jv_0(y) +\nabla f(\varphi(y))\cdot\partial_jv(y).

Expanding the two scalar products yields

∂j(Tf)(y)=f(φ(y))∂jv0(y)−∑i=1nφi(y)(∂if)(φ(y))∂jv0(y)+∑i=1n(∂if)(φ(y))∂jvi(y).\partial_j(Tf)(y) = f(\varphi(y))\partial_jv_0(y) - \sum_{i=1}^{n}\varphi_i(y)(\partial_i f)(\varphi(y))\partial_jv_0(y) + \sum_{i=1}^{n}(\partial_i f)(\varphi(y))\partial_jv_i(y).

Collecting the terms multiplying each (∂if)(φ(y))(\partial_i f)(\varphi(y)), we obtain

∂j(Tf)(y)=∂jv0(y)f(φ(y))+∑i=1n(∂jvi(y)−φi(y)∂jv0(y))(∂if)(φ(y)).\partial_j(Tf)(y) = \partial_jv_0(y)f(\varphi(y)) + \sum_{i=1}^{n}\left(\partial_jv_i(y)-\varphi_i(y)\partial_jv_0(y)\right)(\partial_i f)(\varphi(y)).

Define the coefficient functions on EεE_\varepsilon by

a0(y)=∂jv0(y),ai(y)=∂jvi(y)−φi(y)∂jv0(y),1≤i≤n.a_0(y) = \partial_jv_0(y), \qquad a_i(y) = \partial_jv_i(y)-\varphi_i(y)\partial_jv_0(y), \qquad1 \leq i \leq n.

Then

∂j(Tf)(y)=a0(y)f(φ(y))+∑i=1nai(y)(∂if)(φ(y))a.e. on Eε.(12)\partial_j(Tf)(y) = a_0(y)f(\varphi(y))+\sum_{i=1}^{n}a_i(y)(\partial_i f)(\varphi(y)) \qquad\text{a.e. on } E_\varepsilon. \tag*{(12)}

For the fixed ball BB, cutoff θ\theta and index jj, the functions v0=Tθv_0=T\theta and vi=T(xiθ)v_i=T(x_i\theta) are fixed independently of ff. The formulas defining a0,…,ana_0,\ldots,a_n therefore give the same coefficient functions for every f∈Cc∞(Rn)f\in C^\infty_c(\mathbb{R}^n). For each such ff, identity (12) holds almost everywhere on EεE_\varepsilon.

Let α∈An,k\alpha\in\mathcal{A}_{n,k} have length at least two and consider E=Eε∩{y:πy(ej)=α}E=E_\varepsilon\cap\{y:\pi_y(e_j)=\alpha\}. On this set (8) identifies the left side of (12) with hej(Dαf)∘φh_{e_j}(D^\alpha f)\circ\varphi. Take successively the following test functions.

(i) For f=θf=\theta, DαfD^\alpha f and all first derivatives vanish on BB, while f=1f=1 there. Thus a0=0a_0=0 on EE almost everywhere.

(ii) For f=xiθf=x_i\theta, the derivative DαfD^\alpha f still vanishes on BB, whereas ∂ℓf=δiℓ\partial_\ell f=\delta_{i\ell} there. With a0=0a_0=0, the identity gives ai=0a_i=0 for every ii.

(iii) For f=xαθf=x^\alpha\theta, the right side of (12) is now zero on EE, while Dαf=α!D^\alpha f=\alpha! on BB. Hence hejα!=0h_{e_j}\alpha!=0 almost everywhere on EE.

Since hejh_{e_j} is nonzero almost everywhere, EE has measure zero. Only finitely many tests were needed for each fixed BB, jj, α\alpha and ε\varepsilon. Take ε=1/l\varepsilon=1/l for l∈Nl\in\mathbb{N} and let BB range over the rational balls. The sets EεE_\varepsilon obtained in this way cover the target up to a null set because h≠0h\neq0 almost everywhere. This excludes every α\alpha of length at least two. The value 00 is already used by πy(0)\pi_y(0), so injectivity gives ∣πy(ej)∣=1|\pi_y(e_j)|=1.

The mm distinct first-order target indices consequently have distinct images among the nn first-order source indices. Hence m≤nm\leq n. Applying the same reasoning to T−1T^{-1} gives n≤mn\leq m. For fixed n≥1n\geq1 the number ∣An,k∣|\mathcal{A}_{n,k}| is strictly increasing in kk, since each higher order adds at least one multiindex. Thus (9) also gives s=ks=k. □

Henceforth m=nm=n and s=ks=k. There is a measurable permutation σy\sigma_y of {1,…,n}\{1,\ldots,n\} at almost every yy such that

πy(ej)=eσy(j),∂j(Tf)(y)=hj(y)(∂σy(j)f)(φ(y)),hj=hej.(13)\pi_y(e_j)=e_{\sigma_y(j)}, \qquad\partial_j(Tf)(y)=h_j(y)(\partial_{\sigma_y(j)}f)(\varphi(y)), \qquad h_j=h_{e_j}. \tag*{(13)}

Equal weights, truncation, and rigidity

Lemma 6.1. For every β∈An,k\beta\in\mathcal{A}_{n,k}, one has ∣hβ∣=∣h∣|h_\beta|=|h| almost everywhere. For every nonnegative measurable function vv,

∫Rn∣h(y)∣v(φ(y)) dy=∫Rnv(x) dx.(14)\int_{\mathbb{R}^n}|h(y)|v(\varphi(y))\,dy=\int_{\mathbb{R}^n}v(x)\,dx. \tag*{(14)}

Proof. Fix α∈An,k\alpha\in\mathcal{A}_{n,k} and a nonnegative function v∈L1(Rn)v \in L^1(\mathbb{R}^n). Define u∈L1(Rn×An,k)u \in L^1(\mathbb{R}^n \times\mathcal{A}_{n,k}) by

u(x,γ)={v(x),γ=α,0,γ≠α.u(x,\gamma)= \begin{cases} v(x), & \gamma=\alpha,\\ 0, & \gamma\ne\alpha. \end{cases}

Then

∥u∥L1(Rn×An,k)=∑γ∈An,k∫Rn∣u(x,γ)∣ dx=∫Rnv(x) dx.\lVert u\rVert_{L^1(\mathbb{R}^n\times\mathcal{A}_{n,k})} =\sum_{\gamma\in\mathcal{A}_{n,k}}\int_{\mathbb{R}^n}|u(x,\gamma)|\,dx =\int_{\mathbb{R}^n}v(x)\,dx.

For almost every y∈Rny\in\mathbb{R}^n, the map πy:An,k→An,k\pi_y:\mathcal{A}_{n,k}\to\mathcal{A}_{n,k} is bijective. There is therefore a unique index βα(y)∈An,k\beta_\alpha(y)\in\mathcal{A}_{n,k} such that

πy(βα(y))=α.\pi_y(\beta_\alpha(y))=\alpha.

The representation of UU consequently gives, for almost every y∈Rmy\in\mathbb{R}^m and every β∈An,k\beta\in\mathcal{A}_{n,k},

(Uu)(y,β)=hβ(y)u(φ(y),πy(β))={hβ(y)v(φ(y)),β=βα(y),0,β≠βα(y).(Uu)(y,\beta)=h_\beta(y)u(\varphi(y),\pi_y(\beta)) = \begin{cases} h_\beta(y)v(\varphi(y)), & \beta=\beta_\alpha(y),\\ 0, & \beta\ne\beta_\alpha(y). \end{cases}

The isometry of UU gives

∫Rn∣hβα(y)(y)∣v(φ(y)) dy=∫Rnv(x) dx.\int_{\mathbb{R}^n}|h_{\beta_\alpha(y)}(y)|v(\varphi(y))\,dy =\int_{\mathbb{R}^n}v(x)\,dx.

For α=0\alpha=0 this is (6.1), since β0(y)=0\beta_0(y)=0. Monotone convergence extends the formula from integrable vv to all nonnegative vv.

For any R>0R>0, use as vv the indicator of a measurable subset of B(0,R)B(0,R). Comparing the formulas for α\alpha and 00 shows that the densities ∣hβα(y)(y)∣|h_{\beta_\alpha(y)}(y)| and ∣h(y)∣|h(y)| have equal integrals on each such preimage. Since φ\varphi is invertible modulo null sets, these preimages include, up to null sets, every measurable subset of φ−1(B(0,R))\varphi^{-1}(B(0,R)). Both densities have finite integral on this set, so they agree almost everywhere there. Letting RR grow yields the equality for a fixed α\alpha, i.e., for each α∈An,k\alpha\in\mathcal{A}_{n,k}, we have a null set Nα⊂RmN_\alpha\subset\mathbb{R}^m such that

∣hβα(y)(y)∣=∣h(y)∣(y∉Nα).|h_{\beta_\alpha(y)}(y)|=|h(y)|\qquad(y\notin N_\alpha).

Since An,k\mathcal{A}_{n,k} is finite, the union

N:=⋃α∈An,kNαN:=\bigcup_{\alpha\in\mathcal{A}_{n,k}}N_\alpha

is again a null set. Enlarge NN by the null set on which πy\pi_y fails to be a bijection. Thus, for every y∉Ny\notin N, the preceding equality holds simultaneously for all α∈An,k\alpha\in\mathcal{A}_{n,k}.

Now fix β∈Am,s\beta\in\mathcal{A}_{m,s} and y∉Ny\notin N, and set αy:=πy(β)\alpha_y:=\pi_y(\beta). By the definition of the inverse indices,

βαy(y)=πy−1(πy(β))=β.\beta_{\alpha_y}(y)=\pi_y^{-1}(\pi_y(\beta))=\beta.

Since the equality is valid for every source index at this point yy, we may apply it with α=αy\alpha=\alpha_y to obtain

∣hβ(y)∣=∣hβαy(y)(y)∣=∣h(y)∣.|h_\beta(y)|=|h_{\beta_{\alpha_y}(y)}(y)|=|h(y)|.

Consequently, ∣hβ∣=∣h∣|h_\beta|=|h| almost everywhere for every β∈Am,s\beta\in\mathcal{A}_{m,s}, with a common exceptional null set.

Lemma 6.2. The isometry TT extends to a surjective linear isometry of (W1,1(Rn),Nn,1)(W^{1,1}(\mathbb{R}^n),N_{n,1}). The extension satisfies (5.1) and (5.3) with the same base map and weights.

Proof. For f∈Wk,1f\in W^{k,1}, (6.1) gives ∥Tf∥1=∥f∥1\lVert Tf\rVert_1=\lVert f\rVert_1. Moreover, summing (5.3) over jj and using ∣hj∣=∣h∣|h_j|=|h| and the fact that σy\sigma_y is a permutation for almost all yy, it follows that

∑j=1n∥∂jTf∥1=∫Rn∣h(y)∣∑i=1n∣(∂if)(φ(y))∣ dy=∑i=1n∥∂if∥1,\sum_{j=1}^{n}\lVert\partial_jTf\rVert_1 =\int_{\mathbb{R}^n}|h(y)|\sum_{i=1}^{n}|(\partial_i f)(\varphi(y))|\,dy =\sum_{i=1}^{n}\lVert\partial_i f\rVert_1,

where the last equality follows from (14). Thus TT also preserves Nn,1N_{n,1} on Wk,1W^{k,1}. This subspace is dense in W1,1W^{1,1} because it contains Cc∞C^\infty_c. Extending TT and T−1T^{-1} by continuity for the first-order norm gives inverse isometries, as the compositions agree with the identity on Wk,1W^{k,1}. Write T~\widetilde{T} for the continuous extension constructed above. Fix f∈W1,1(Rn)f \in W^{1,1}(\mathbb{R}^n) and choose fℓ∈Cc∞(Rn)f_\ell\in C^\infty_c(\mathbb{R}^n) such that fℓ→ff_\ell\to f in W1,1(Rn)W^{1,1}(\mathbb{R}^n).

By (14), every q∈L1(Rn)q \in L^1(\mathbb{R}^n) satisfies

∥h(q∘φ)∥1=∥q∥1.\lVert h(q \circ\varphi)\rVert_1 = \lVert q\rVert_1.

Thus q↦h(q∘φ)q \mapsto h(q \circ\varphi) is a continuous linear map on L1(Rn)L^1(\mathbb{R}^n). Moreover, for any tuple (qi)i=1n(q_i)_{i=1}^n with qi∈L1(Rn)q_i \in L^1(\mathbb{R}^n), the identities ∣hj∣=∣h∣\lvert h_j\rvert= \lvert h\rvert and the fact that σy\sigma_y is a permutation for almost every yy give

∑j=1n∫Rn∣hj(y)∣∣qσy(j)(φ(y))∣ dy=∫Rn∣h(y)∣∑i=1n∣qi(φ(y))∣ dy=∑i=1n∥qi∥1,\begin{aligned} \sum_{j=1}^{n}\int_{\mathbb{R}^n} \lvert h_j(y)\rvert\lvert q_{\sigma_y(j)}(\varphi(y))\rvert\,dy = \int_{\mathbb{R}^n} \lvert h(y)\rvert\sum_{i=1}^{n}\lvert q_i(\varphi(y))\rvert\,dy \\ &= \sum_{i=1}^{n}\lVert q_i\rVert_1, \end{aligned}

where the last equality follows by applying (14) to each ∣qi∣\lvert q_i\rvert.

Define

g(y)=h(y)f(φ(y)),Gj(y)=hj(y)(∂σy(j)f)(φ(y)),1≤j≤n.g(y)=h(y)f(\varphi(y)),\qquad G_j(y)=h_j(y)(\partial_{\sigma_y(j)}f)(\varphi(y)),\qquad1\leq j\leq n.

The preceding identities show that g,G1,…,Gng,G_1,\ldots,G_n belong to L1(Rn)L^1(\mathbb{R}^n). The value and derivative formulas already established for TT apply to every fℓf_\ell, since fℓ∈Wk,1(Rn)f_\ell\in W^{k,1}(\mathbb{R}^n). Applying the first norm identity to q=fℓ−fq=f_\ell-f, we therefore obtain

∥Tfℓ−g∥1=∥fℓ−f∥1⟶0.\lVert Tf_\ell-g\rVert_1=\lVert f_\ell-f\rVert_1\longrightarrow0.

Applying the identity for tuples to qi=∂ifℓ−∂ifq_i=\partial_i f_\ell-\partial_i f gives

∑j=1n∥∂j(Tfℓ)−Gj∥1=∑i=1n∥∂ifℓ−∂if∥1⟶0.\sum_{j=1}^{n}\lVert\partial_j(Tf_\ell)-G_j\rVert_1 = \sum_{i=1}^{n}\lVert\partial_i f_\ell-\partial_i f\rVert_1 \longrightarrow0.

In particular, ∂j(Tfℓ)→Gj\partial_j(Tf_\ell)\to G_j in L1(Rn)L^1(\mathbb{R}^n) for each jj.

On the other hand, the continuity of T~\widetilde{T} on W1,1(Rn)W^{1,1}(\mathbb{R}^n) and its agreement with TT on Wk,1(Rn)W^{k,1}(\mathbb{R}^n) imply

Tfℓ=T~fℓ⟶T~fin W1,1(Rn).Tf_\ell=\widetilde{T}f_\ell\longrightarrow\widetilde{T}f \qquad\text{in }W^{1,1}(\mathbb{R}^n).

Consequently,

Tfℓ⟶T~f,∂j(Tfℓ)⟶∂j(T~f)in L1(Rn).Tf_\ell\longrightarrow\widetilde{T}f,\qquad \partial_j(Tf_\ell)\longrightarrow\partial_j(\widetilde{T}f) \qquad\text{in }L^1(\mathbb{R}^n).

Uniqueness of the L1L^1 limit now gives T~f=g\widetilde{T}f=g and ∂j(T~f)=Gj\partial_j(\widetilde{T}f)=G_j. Hence

(T~f)(y)=h(y)f(φ(y))(\widetilde{T}f)(y)=h(y)f(\varphi(y))

and

∂j(T~f)(y)=hj(y)(∂σy(j)f)(φ(y)),1≤j≤n,\partial_j(\widetilde{T}f)(y)=h_j(y)(\partial_{\sigma_y(j)}f)(\varphi(y)), \qquad1\leq j\leq n,

for almost every y∈Rny\in\mathbb{R}^n. Thus both formulas remain valid for the extension, with the same map φ\varphi, the same permutations σy\sigma_y, and the same weights.

We now use the first-order extension to truncate a preimage under TT. A Lipschitz truncation preserves W1,1W^{1,1}, whereas for k>1k>1 there is no such general rule on Wk,1W^{k,1}.

Lemma 6.3 (The weight is constant). There is a constant c∈R∖{0}c\in\mathbb{R}\setminus\{0\} such that h=ch=c almost everywhere.

Proof. We work with the first-order extension. Fix a rational ball K⊂RnK \subset\mathbb{R}^{n} and a positive integer aa. Choose u∈Cc∞(Rn)u \in C^{\infty}_{c}(\mathbb{R}^{n}) equal to aa on KK. By surjectivity there exists f∈W1,1f \in W^{1,1} with Tf=uTf=u. The value formula gives f(φ(y))=a/h(y)f(\varphi(y))=a/h(y) on KK. Also ∂ju=0\partial_{j}u=0 there for each jj. Since all hjh_{j} are nonzero and σy\sigma_{y} is a permutation, (13) yields (∂if)(φ(y))=0(\partial_{i}f)(\varphi(y))=0 for all ii at almost every y∈Ky \in K.

We now truncate ∣f∣|f|. The function ψ(t)=min⁡(∣t∣,1)\psi(t)=\min(|t|,1) is 11-Lipschitz on R\mathbb{R} and ψ(0)=0\psi(0)=0. The Sobolev composition estimate [2 Theorem 4.2] gives ψ(f)∈W1,1\psi(f)\in W^{1,1} and ∣∂i(ψ(f))∣≤∣∂if∣|\partial_{i}(\psi(f))|\leq|\partial_{i}f| almost everywhere. The preceding inequalities hold for all i=1,…,ni=1,\ldots,n outside a common Lebesgue null set N⊂RnN\subset\mathbb{R}^{n}. Since φ\varphi is nonsingular, the inverse image φ−1(N)\varphi^{-1}(N) is also a Lebesgue null set. Thus, for almost every y∈Rny\in\mathbb{R}^{n}, we have φ(y)∉N\varphi(y)\notin N, and consequently

∣(∂i(ψ(f)))(φ(y))∣≤∣(∂if)(φ(y))∣,1≤i≤n.\left|(\partial_{i}(\psi(f)))(\varphi(y))\right|\leq\left|(\partial_{i}f)(\varphi(y))\right|,\qquad1\leq i\leq n.

Applying (13) to ψ(f)\psi(f) therefore gives ∇T(ψ(f))=0\nabla T(\psi(f))=0 almost everywhere on KK, since (∂if)(φ(y))=0(\partial_{i}f)(\varphi(y))=0 for all ii at almost every y∈Ky\in K. Its value on KK is, by (10),

T(ψ(f))(y)=h(y)min⁡(a∣h(y)∣,1)=:ha(y).T(\psi(f))(y)=h(y)\min\left(\frac{a}{|h(y)|},1\right)=:h_{a}(y).

The expression hah_{a} is independent of the choice of uu and of KK. Since ∇T(ψ(f))=0\nabla T(\psi(f))=0 almost everywhere on KK, we have that T(ψ(f))T(\psi(f)) is almost everywhere constant there. For a fixed aa, apply this conclusion on the countable nested balls B(0,R)B(0,R), R∈NR\in\mathbb{N}. Their constants agree on overlaps, so ha=cah_{a}=c_{a} almost everywhere on Rn\mathbb{R}^{n} for one scalar cac_{a}.

Finally take a common full-measure set for all integers a≥1a\geq1 on which 0<∣h∣<∞0<|h|<\infty. At every point of this set ha(y)=h(y)h_{a}(y)=h(y) once a≥∣h(y)∣a\geq|h(y)|. Thus the same scalar sequence (ca)(c_{a}) converges to h(y)h(y) at every such yy. Its limit gives the required nonzero cc. □

Lemma 6.4. The base map has a representative φ(y)=Py+b\varphi(y)=Py+b, where PP is a signed permutation matrix. The constant in Lemma 6.3 belongs to {−1,1}\{-1,1\}.

Proof. Choose θ∈Cc∞(Rn)\theta\in C^{\infty}_{c}(\mathbb{R}^{n}) equal to one on B(0,1)B(0,1). For integers R≥1R\geq1 and 1≤i≤n1\leq i\leq n, put fR,i(x)=xiθ(x/R)f_{R,i}(x)=x_{i}\theta(x/R). Their derivative is

∂ℓfR,i(x)=δiℓθ(x/R)+(xi/R)(∂ℓθ)(x/R).\partial_{\ell}f_{R,i}(x)=\delta_{i\ell}\theta(x/R)+(x_{i}/R)(\partial_{\ell}\theta)(x/R).

With z=x/Rz=x/R, the second term is zi(∂ℓθ)(z)z_{i}(\partial_{\ell}\theta)(z), which has a bound independent of RR. Put FR,i=c−1TfR,i=fR,i∘φF_{R,i}=c^{-1}Tf_{R,i}=f_{R,i}\circ\varphi. Each FR,iF_{R,i} is bounded and its weak first derivatives are bounded uniformly in RR by (13), since ∣hj/c∣=1|h_{j}/c|=1. The W1,∞W^{1,\infty} representative theorem [8 Theorem 4.1] gives globally Lipschitz representatives with a common Lipschitz bound.

Take a full-measure set on which all these representatives agree with fR,i∘φf_{R,i}\circ\varphi. For any yy in this set, FR,i(y)=φi(y)F_{R,i}(y)=\varphi_{i}(y) whenever R>∣φ(y)∣R>|\varphi(y)|. Passing to the limit R→∞R\to\infty in the Lipschitz inequality at two points of this set gives the Lipschitz bound for φi\varphi_{i}. Extension from this dense set gives a globally Lipschitz representative of each coordinate of φ\varphi. On {y:∣φ(y)∣<R}\{y:|\varphi(y)|<R\} the functions FR,iF_{R,i} and φi\varphi_{i} agree. By locality their derivatives agree there almost everywhere, and ∂ℓFR,i=δiℓ\partial_{\ell}F_{R,i}=\delta_{i\ell} on B(0,R)B(0,R). Using (13) and exhausting in RR, we obtain

∂jφi(y)=hj(y)c1{i=σy(j)}almost everywhere.(15)\partial_{j}\varphi_{i}(y)=\frac{h_{j}(y)}{c}\mathbf{1}_{\{i=\sigma_{y}(j)\}}\qquad\text{almost everywhere}. \tag*{(15)}

Each column of DφD\varphi has one nonzero entry, and these entries occupy distinct rows, since σy\sigma_{y} is a permutation for almost every yy. Each nonzero entry is a ratio hj(y)/ch_{j}(y)/c of real numbers with ∣hj(y)∣=∣c∣|h_{j}(y)|=|c|, so it is either 11 or −1-1. Thus Dφ(y)D\varphi(y) is a signed permutation matrix almost everywhere. Its Euclidean operator norm is one and therefore ∣φ(y)−φ(z)∣≤∣y−z∣|\varphi(y)-\varphi(z)|\leq|y-z|.

Apply the same argument to T−1T^{-1}, using U−1U^{-1} as its ambient extension. This gives a globally 11-Lipschitz representative ψ\psi of the inverse base map. Since both original maps are nonsingular, replacing them by their continuous representatives preserves both inverse identities almost everywhere. By continuity the identities hold everywhere. Applying the bound for ψ\psi at φ(y)\varphi(y) and φ(z)\varphi(z) gives the opposite inequality, so that

∣y−z∣=∣ψ(φ(y))−ψ(φ(z))∣≤∣φ(y)−φ(z)∣≤∣y−z∣.|y-z|=|\psi(\varphi(y))-\psi(\varphi(z))|\leq|\varphi(y)-\varphi(z)|\leq|y-z|.

Thus φ\varphi is a surjective Euclidean isometry. By the Mazur–Ulam theorem [12], φ(y)=Py+b\varphi(y) = P y + b. Its linear part PP preserves the Euclidean norm and is therefore orthogonal. Its derivative is the constant matrix PP, and (15) therefore makes PP a signed permutation matrix. Since h=ch = c and φ(y)=Py+b\varphi(y) = P y + b, identity (14) reads

∣c∣∫Rnv(Py+b) dy=∫Rnv(x) dx.|c| \int_{\mathbb{R}^n} v(P y + b)\,dy = \int_{\mathbb{R}^n} v(x)\,dx.

Let B=B(0,1)B = B(0,1) and take v=1Bv = \mathbf{1}_B. Then

∣c∣∫Rn1B(Py+b) dy=∣B∣.|c| \int_{\mathbb{R}^n} \mathbf{1}_B(P y + b)\,dy = |B|.

The change of variables x=Py+bx = P y + b gives dy=∣det⁡P∣−1 dxdy = |\det P|^{-1}\,dx, and hence

∣c∣∫Rn1B(Py+b) dy=∣c∣ ∣det⁡P∣−1∫Rn1B(x) dx=∣c∣ ∣det⁡P∣−1∣B∣.\begin{aligned} |c| \int_{\mathbb{R}^n} \mathbf{1}_B(P y + b)\,dy &= |c|\,|\det P|^{-1} \int_{\mathbb{R}^n} \mathbf{1}_B(x)\,dx \\ &= |c|\,|\det P|^{-1}|B|. \end{aligned}

Consequently,

∣c∣ ∣det⁡P∣−1∣B∣=∣B∣.|c|\,|\det P|^{-1}|B| = |B|.

Since 0<∣B∣<∞0 < |B| < \infty, division by ∣B∣|B| yields

∣c∣ ∣det⁡P∣−1=1.|c|\,|\det P|^{-1} = 1.

The matrix PP is orthogonal, so ∣det⁡P∣=1|\det P| = 1. Thus ∣c∣=1|c| = 1, and since cc is real, we conclude that c∈{−1,1}c \in\{-1,1\}. □\square

Formula (10) now gives (2), completing the proof of Theorem 1.1.

Acknowledgments

The author used OpenAI’s GPT-6 Astra to assist with mathematical exploration, the development and checking of proofs, and the presentation of arguments. The author takes full responsibility for the mathematical content and any remaining errors.

References

References

  1. [1]A. Pełczyński and M. Wojciechowski, Sobolev spaces, in Handbook of the Geometry of Banach Spaces, Vol. 2, W. B. Johnson and J. Lindenstrauss (eds.), North-Holland, Amsterdam, 2003, pp. 1361–1423.
  2. [2]W. Arendt and M. Kreuter, Mapping theorems for Sobolev spaces of vector-valued functions, Studia Math. 240 (2018), 275–299.DOI
  3. [3]M. Biegert and R. Nittka, Isometries between Sobolev spaces, Math. Nachr. 285 (2012), 2059–2077. Preprint version available as arXiv:0807.0360v4.
  4. [4]G. Diestel, Sobolev spaces with only trivial isometries, II, Positivity 13 (2009), 621–630.DOI
  5. [5]D. H. Fremlin, Measure Theory, vol. 2, Broad Foundations, Torres Fremlin, Colchester, 2001.
  6. [6]D. H. Fremlin, Measure Theory, vol. 3, Measure Algebras, Chapter 34, author’s online edition.
  7. [7]C. D. Hardin, Jr., Isometries on subspaces of L^p, Indiana Univ. Math. J. 30 (1981), 449–465.
  8. [8]J. Heinonen, Lectures on Lipschitz Analysis, Report 100, Department of Mathematics and Statistics, University of Jyväskylä, 2005.
  9. [9]A. S. Kechris, Classical Descriptive Set Theory, Graduate Texts in Mathematics, vol. 156, Springer, New York, 1995.DOI
  10. [10]J. Lamperti, On the isometries of certain function-spaces, Pacific J. Math. 8 (1958), 459–466.DOI
  11. [11]E. H. Lieb and M. Loss, Analysis, 2nd ed., Graduate Studies in Mathematics, vol. 14, American Mathematical Society, Providence, RI, 2001.
  12. [12]S. Mazur and S. Ulam, Sur les transformations isométriques d’espaces vectoriels normés, C. R. Acad. Sci. Paris 194 (1932), 946–948.

Paper details

vlnc:W_hrtd7kkz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

OpenAI’s GPT-6 Astra

View statement
Looking for related papers…
Contents