betaFeedback

WorksMathematics

Extension of surjective isometries between unit spheres of ℓp\ell^p-direct sums

Abstract

We study surjective isometries between the unit spheres of Banach spaces of the form A⊕pBA\oplus_p B, where AA is nonzero and strictly convex, 1<p<∞1<p<\infty, and every unit vector of B≠{0}B\neq\{0\} has at least two distinct unit vectors at distance 22. For two direct sums Ai⊕piBiA_i\oplus_{p_i}B_i, i=1,2i=1,2, satisfying these hypotheses, we prove that every surjective isometry Δ:SA1⊕p1B1→SA2⊕p2B2Δ:S_{A_1\oplus_{p_1}B_1}\to S_{A_2\oplus_{p_2} B_2} between their unit spheres forces equality of the direct-sum exponents pip_i and extends uniquely to a surjective real-linear isometry Δ~:A1⊕pB1→A2⊕pB2\widetildeΔ:A_1\oplus_p B_1\to A_2\oplus_p B_2 preserving the corresponding components, where p:=p1=p2p:=p_1=p_2. The proof identifies the component unit spheres through diametral sets and distance formulas, and then constructs isometries between the component unit balls. As a principal application, we obtain extension and representation results for absolutely continuous functions equipped with norms combining evaluation at a fixed point and the L1L^1-norm of the derivative. We also discuss Lipschitz and differentiable function spaces, distinguishing the consequences of our theorem from those already covered by existing extension results. Endpoint conclusions are derived separately, including the Mazur-Ulam property for the real absolutely continuous space with the sum norm.

Introduction

In 1987, Tingley [14] formulated the following problem: if XX and YY are Banach spaces and

Δ:SX→SY\Delta:S_X\to S_Y

is a surjective isometry between their unit spheres, does Δ\Delta extend to a surjective real-linear isometry from XX onto YY? Here SXS_X denotes the unit sphere of XX. Tingley’s problem has received considerable attention, and positive solutions have been obtained for many classical Banach spaces and spaces arising in operator algebra theory. For example, Wang [15] established an extension theorem for scalar-valued continuous-function spaces, and Tan [13] proved that real L1L^1-spaces have the Mazur–Ulam property. Mori and Ozawa [9] proved the Mazur–Ulam property for unital C*-algebras, meaning that the target space may be arbitrary. Banakh [1] resolved the problem for two-dimensional Banach spaces. Further developments include the Mazur–Ulam property for unital JB∗JB^*-algebras [11]. We refer to [10] for a survey of developments concerning operator algebras.

Despite these advances, Tingley’s problem remains open in general, including for arbitrary finite-dimensional Banach spaces of dimension at least three. The known positive results therefore continue to motivate the search for geometric hypotheses under which the behavior of a surjective isometry on the unit sphere forces linearity. For direct sums, a basic issue is whether the isometry recognizes the component spaces. Once the component unit spheres have been identified, one can ask whether their images determine isometries between the component unit balls and hence linear extensions.

Direct sums of continuous-function spaces provide an important setting for this approach. Wang and Orihara [17] studied surjective isometries between the unit spheres of ℓp\ell^p-sums of C0(Ω,E)C_0(\Omega,E)-type spaces. Wang [16] subsequently established extension and representation theorems for ℓp\ell^p-sums of 2020 Mathematics Subject Classification. Primary 46B04; Secondary 46B20, 46E15, 46E40.

Key words and phrases. Tingley’s problem, direct sum, strictly convex space, absolutely continuous function. C0(Ω,E)C_0(\Omega,E)-type spaces, where 1<p<∞1<p<\infty. In particular, [16 Theorem 1.8] applies to

A⊕pC(K,E)A\oplus_p C(K,E)

where AA is a nonzero strictly convex Banach space, KK is a compact Hausdorff space containing at least two points, and EE is an arbitrary nonzero Banach space. In fact, the first component can be represented as C0({∗},A)C_0(\{\ast\},A). Thus the freedom to choose an arbitrary strictly convex first component is already present in Wang’s theorem.

The present paper considers a geometric condition (D) on the second component that does not require a specified continuous-function representation. Condition (D) is defined in the next section. We state the main theorem as follows.

Theorem 1.1. For i=1,2i=1,2, let AiA_i be a nonzero strictly convex Banach space, let BiB_i be a nonzero Banach space satisfying condition (D), and let 1<pi<∞1<p_i<\infty. Then the following statements are equivalent:

(1) There exists a surjective isometry Δ:SA1⊕p1B1→SA2⊕p2B2\Delta:S_{A_1\oplus_{p_1}B_1}\to S_{A_2\oplus_{p_2}B_2}.

(2) p1=p2=:pp_1=p_2=:p, and there exist surjective real-linear isometries U:A1→A2U:A_1\to A_2 and V:B1→B2V:B_1\to B_2.

Moreover, for each such Δ\Delta, there are unique surjective real-linear isometries UU and VV satisfying

Δ(u,v)=(U(u),V(v)),(u,v)∈SA1⊕pB1.\Delta(u,v)=(U(u),V(v)),\qquad(u,v)\in S_{A_1\oplus_p B_1}.

The operator U⊕VU\oplus V is the unique surjective real-linear isometric extension of $\Delta.

We emphasize absolutely continuous and Lipschitz functions. The main theorem gives real-linear extensions for 1<p<∞1<p<\infty and identifies the outer exponent. Standard representations of the component isometries then yield explicit integral formulas. The endpoint conclusions are obtained separately. Adjoining an atom identifies the real absolutely continuous space at p=1p=1 with a real L1L^1-space, so Tan’s theorem gives the Mazur–Ulam property [13]. For Lipschitz functions, the endpoints follow from continuous-function extension theorems and the ℓ1\ell^1-sum theorem of Wang and Orihara [15, 17]. The preceding endpoint conclusions are consistent with the complex-valued Lipschitz results of Hirota and Miura [5] and the related results in [2]. Atom-adjoining identifications also appear in Lin’s work on positive unit spheres [6, 7]; the conclusions here concern real-linear extensions from the full unit spheres.

For completeness, we also discuss scalar- and vector-valued differentiable functions through the identification

f⟼((f(j)(x0))j=0n−1,f(n)).f\longmapsto\left((f^{(j)}(x_0))_{j=0}^{n-1},f^{(n)}\right).

For the norms specified below, the displayed identification represents the spaces as direct sums involving C(I,E)C(I,E). For each fixed exponent 1<p<∞1<p<\infty, these extension assertions also follow from Wang’s theorem [16 Theorem 1.8]. The same observation applies to the fixed-exponent Lipschitz assertion, since L∞(I,K)L^\infty(I,\mathbb{K}) is isometrically a scalar continuous-function space. The function-space applications illustrate the abstract theorem and its representations.

The paper is organized as follows. Section 2 introduces definitions and notation and establishes preliminary results. Section 3 identifies the components and establishes the main theorem. Section 4 develops applications, including the absolutely continuous AC(I)AC(I), Lipschitz Lip⁡(I)\operatorname{Lip}(I), continuous-functions C(K,E)C(K,E) on compact Hausdorff KK and differentiable functions Cn(I,E)C^n(I,E).

Notation and the diametral hypothesis

Throughout this paper, all Banach spaces are considered as real Banach spaces. When a space is written over C\mathbb{C}, we retain its complex-valued elements and its norm but restrict scalar multiplication to real scalars for this purpose. Unless explicitly qualified as complex-linear or K\mathbb{K}-linear, linearity means real linearity. For a nonzero Banach space EE, write

BE={x∈E:∥x∥≤1},SE={x∈E:∥x∥=1}.B_E=\{x\in E:\|x\|\le1\},\qquad S_E=\{x\in E:\|x\|=1\}.

For x∈SEx \in S_E, define its diametral set by

DE(x)={y∈SE:∥x−y∥=2}.D_E(x)=\{y\in S_E:\|x-y\|=2\}.

Notice that −x∈DE(x)-x\in D_E(x) for every x∈SEx\in S_E. We give the definition of condition (D).

Definition 2.1. A nonzero Banach space BB satisfies condition (D) if DB(b)D_B(b) contains at least two distinct points for every b∈SBb\in S_B.

Proposition 2.2. If E,FE,F are nonzero Banach spaces, both E⊕1FE\oplus_1 F and E⊕∞FE\oplus_\infty F satisfy condition (D).

Proof. Let (e,f)∈SE⊕1F(e,f)\in S_{E\oplus_1F}. If ∥e∥,∥f∥>0\|e\|,\|f\|>0, the distinct unit vectors (−e/∥e∥,0)(-e/\|e\|,0) and (0,−f/∥f∥)(0,-f/\|f\|) are both at distance ∥e∥+∥f∥+1=2\|e\|+\|f\|+1=2 from (e,f)(e,f). If ∥f∥=0\|f\|=0, use (−e,0)(-e,0) and (0,v)(0,v) for any v∈SFv\in S_F. If ∥e∥=0\|e\|=0, use (0,−f)(0,-f) and (u,0)(u,0) for any u∈SEu\in S_E.

For (e,f)∈SE⊕∞F(e,f)\in S_{E\oplus_\infty F}, at least one coordinate has norm one. If ∥e∥=1\|e\|=1, choose distinct v1,v2∈BFv_1,v_2\in B_F. Then (−e,v1)(-e,v_1) and (−e,v2)(-e,v_2) are distinct unit vectors at distance two from (e,f)(e,f). If ∥f∥=1\|f\|=1, argue in the other coordinate. □

Example 2.3. Condition (D) holds for ℓ1(Γ)\ell_1(\Gamma), c0(Γ)c_0(\Gamma), and ℓ∞(Γ)\ell_\infty(\Gamma) when Γ\Gamma contains at least two elements: split off one coordinate and apply Proposition 2.2. Condition (D) also holds for C(K,E)C(K,E) when KK is compact Hausdorff with at least two points and E≠{0}E\ne\{0\}; see § 4.

Notice that a nonzero strictly convex Banach space EE does not satisfy (D), since DE(x)={−x}D_E(x)=\{-x\} for every x∈SEx\in S_E. The direct-sum space A⊕pBA\oplus_p B is endowed with

∥(a,b)∥A⊕pB={(∥a∥Ap+∥b∥Bp)1/p(1≤p<∞),max⁡{∥a∥A,∥b∥B}(p=∞).\|(a,b)\|_{A\oplus_pB}= \begin{cases} \left(\|a\|_A^p+\|b\|_B^p\right)^{1/p} & (1\leq p<\infty),\\ \max\{\|a\|_A,\|b\|_B\} & (p=\infty). \end{cases}

We use the following standard consequence of Mankiewicz’s theorem.

Theorem 2.4 (Mankiewicz [8]). Let E,FE,F be real normed spaces. Every surjective isometry G:BE→BFG:B_E\to B_F with G(0)=0G(0)=0 is the restriction of a unique surjective real-linear isometry G~:E→F\widetilde{G}:E\to F.

In particular, closed unit balls are convex sets with nonempty norm interior. Mankiewicz’s theorem therefore gives an affine isometry of the ambient spaces extending the ball isometry. The range of the affine extension contains the target unit ball; the linear part is consequently onto. Fixing zero makes the affine extension linear. Uniqueness follows because every vector is a positive scalar multiple of a vector in the ball.

Recall that a real normed space AA is strictly convex if ∥(x+y)/2∥<1\|(x+y)/2\|<1 whenever x,y∈SAx,y\in S_A are distinct. Equivalently, for nonzero x,y∈Ax,y\in A, equality ∥x+y∥=∥x∥+∥y∥\|x+y\|=\|x\|+\|y\| implies x/∥x∥=y/∥y∥x/\|x\|=y/\|y\|. We use this characterization below. At the end of this section, we recall a well-known fact.

Lemma 2.5. Let X,YX,Y be nonzero real or complex normed spaces and let Δ:SX→SY\Delta:S_X\to S_Y be any map. There is at most one real-linear operator Δ~:X→Y\widetilde{\Delta}:X\to Y extending Δ\Delta. Any such real-linear extension must satisfy

Δ~(0)=0,Δ~(x)=∥x∥Δ(x/∥x∥)(x≠0).(1)\widetilde{\Delta}(0)=0,\qquad\widetilde{\Delta}(x)=\|x\|\Delta\left(x/\|x\|\right)\quad(x\ne0). \tag*{(1)}

If X,YX,Y are complex and this real-linear extension exists, then Δ~\widetilde{\Delta} is complex-linear if and only if Δ(ix)=iΔ(x)\Delta(ix)=i\Delta(x) for all x∈SXx\in S_X.

The uniqueness in Lemma 2.5 requires no isometry assumption. Also, this lemma does not assert existence. For complex spaces, identity and conjugation give different maps on the full unit sphere, so they do not contradict uniqueness of the real-linear extension of a fixed map.

The main extension theorem

Lemma 3.1. Let AA be a nonzero strictly convex Banach space, let BB satisfy (D), and let X=A⊕pBX=A\oplus_pB, where 1<p<∞1<p<\infty. Then

PX:={x∈SX:DX(x) is a singleton}=SA×{0}.P_X:=\{x\in S_X:D_X(x)\text{ is a singleton}\}=S_A\times\{0\}.

For a∈SAa\in S_A, one has DX((a,0))={(−a,0)}D_X((a,0))=\{(-a,0)\}.

Proof. Suppose that a∈SAa\in S_A and (c,d)∈SX(c,d)\in S_X. Put t=∥c∥At=\lVert c\rVert_A. The triangle inequality gives

∥(a,0)−(c,d)∥Xp=∥a−c∥Ap+∥d∥Bp=∥a−c∥Ap+1−tp≤(∥a∥A+∥c∥A)p+1−tp=(1+t)p+1−tp.\begin{aligned} \lVert(a,0)-(c,d)\rVert_X^p&=\lVert a-c\rVert_A^p+\lVert d\rVert_B^p\\ &=\lVert a-c\rVert_A^p+1-t^p\\ &\leq(\lVert a\rVert_A+\lVert c\rVert_A)^p+1-t^p\\ &=(1+t)^p+1-t^p. \end{aligned}

We observe that, for 1<p<∞1<p<\infty, the function gp(t)=(1+t)p+1−tpg_p(t)=(1+t)^p+1-t^p is strictly increasing on [0,1][0,1], since gp′(t)=p((1+t)p−1−tp−1)>0g'_p(t)=p((1+t)^{p-1}-t^{p-1})>0. Consequently, distance two forces t=1t=1, and hence d=0d=0. It also forces ∥a−c∥A=2\lVert a-c\rVert_A=2 with c∈SAc\in S_A. Strict convexity of AA implies c=−ac=-a. That is, (c,d)=(−a,0)(c,d)=(-a,0), and hence DX((a,0))={(−a,0)}D_X((a,0))=\{(-a,0)\}, proving that SA×{0}⊂PXS_A\times\{0\}\subset P_X.

Conversely, let (a,b)∈SX(a,b)\in S_X with r=∥b∥B≠0r=\lVert b\rVert_B\ne0. By (D), choose distinct v1,v2∈SBv_1,v_2\in S_B with ∥b/r−v1∥B=∥b/r−v2∥B=2\lVert b/r-v_1\rVert_B=\lVert b/r-v_2\rVert_B=2. For k=1,2k=1,2, the two distinct points (−a,rvk)(-a,rv_k) belong to SXS_X, since

∥(−a,rvk)∥Xp=∥a∥Ap+rp∥vk∥Bp=∥a∥Ap+rp=∥a∥Ap+∥b∥Bp=1.\lVert(-a,rv_k)\rVert_X^p=\lVert a\rVert_A^p+r^p\lVert v_k\rVert_B^p=\lVert a\rVert_A^p+r^p=\lVert a\rVert_A^p+\lVert b\rVert_B^p=1.

Observe that

∥(a,b)−(−a,rvk)∥Xp=∥2a∥Ap+∥b−rvk∥Bp=2p∥a∥Ap+rp∥b/r−vk∥Bp=2p∥a∥Ap+2prp=2p(∥a∥Ap+∥b∥Bp)=2p.\begin{aligned} \lVert(a,b)-(-a,rv_k)\rVert_X^p&=\lVert2a\rVert_A^p+\lVert b-rv_k\rVert_B^p\\ &=2^p\lVert a\rVert_A^p+r^p\lVert b/r-v_k\rVert_B^p\\ &=2^p\lVert a\rVert_A^p+2^pr^p\\ &=2^p(\lVert a\rVert_A^p+\lVert b\rVert_B^p)=2^p. \end{aligned}

Thus, DX((a,b))D_X((a,b)) with ∥b∥B≠0\lVert b\rVert_B\ne0 is not a singleton. Equivalently, DX((a,b))D_X((a,b)) being a singleton implies b=0b=0, proving that PX⊂SA×{0}P_X\subset S_A\times\{0\}. Therefore, PX=SA×{0}P_X=S_A\times\{0\}. □

Lemma 3.1 identifies the first component unit sphere using distances on the unit sphere. We next calculate distances to PXP_X to identify the second component unit sphere and the direct-sum exponent.

Lemma 3.2. Under the hypotheses of Lemma 3.1, let (a,b)∈SX(a,b)\in S_X. Then

max⁡c∈SA∥(a,b)−(c,0)∥Xp=(1+∥a∥A)p+1−∥a∥Ap(2)\max_{c\in S_A}\lVert(a,b)-(c,0)\rVert_X^p=(1+\lVert a\rVert_A)^p+1-\lVert a\rVert_A^p \tag*{(2)}
min⁡c∈SA∥(a,b)−(c,0)∥Xp=(1−∥a∥A)p+1−∥a∥Ap.(3)\min_{c\in S_A}\lVert(a,b)-(c,0)\rVert_X^p=(1-\lVert a\rVert_A)^p+1-\lVert a\rVert_A^p. \tag*{(3)}

If ∥a∥A≠0\lVert a\rVert_A\ne0, the unique maximizer in (3) is c=−a/∥a∥Ac=-a/\lVert a\rVert_A.

Proof. For c∈SAc\in S_A, the triangle inequality yields

1−∥a∥A≤∥a−c∥A≤1+∥a∥A,∥b∥Bp=1−∥a∥Ap.1-\lVert a\rVert_A\leq\lVert a-c\rVert_A\leq1+\lVert a\rVert_A,\qquad\lVert b\rVert_B^p=1-\lVert a\rVert_A^p.

Consequently, we have equalities (3) and (4). If ∥a∥A≠0\lVert a\rVert_A\ne0, equality of ∥a−c∥A≤1+∥a∥A\lVert a-c\rVert_A\leq1+\lVert a\rVert_A is obtained at c=−a/∥a∥Ac=-a/\lVert a\rVert_A, and equality of 1−∥a∥A≤∥a−c∥A1-\lVert a\rVert_A\leq\lVert a-c\rVert_A at c=a/∥a∥Ac=a/\lVert a\rVert_A. Strict convexity of AA implies that equality ∥a−c∥A=∥a∥A+∥c∥A\lVert a-c\rVert_A=\lVert a\rVert_A+\lVert c\rVert_A is possible only when aa and −c-c are positively collinear, and thus the maximizer is uniquely c=−a/∥a∥Ac=-a/\lVert a\rVert_A. If ∥a∥A=0\lVert a\rVert_A=0, every distance in question equals 21/p2^{1/p}. □

Lemma 3.3. For i=1,2i=1,2, let AiA_i be nonzero and strictly convex, let BiB_i satisfy (D), and let 1<pi<∞1<p_i<\infty. Set Xi=Ai⊕piBiX_i=A_i\oplus_{p_i}B_i. If Δ:SX1→SX2\Delta:S_{X_1}\to S_{X_2} is a surjective isometry, then p1=p2p_1=p_2. Moreover,

Δ(SA1×{0})=SA2×{0},Δ({0}×SB1)={0}×SB2.\Delta(S_{A_1}\times\{0\})=S_{A_2}\times\{0\},\qquad\Delta(\{0\}\times S_{B_1})=\{0\}\times S_{B_2}.

Proof. Since Δ:SX1→SX2\Delta:S_{X_1}\to S_{X_2} is a surjective isometry, for every x∈SX1x\in S_{X_1} we have Δ(DX1(x))=DX2(Δ(x))\Delta(D_{X_1}(x))=D_{X_2}(\Delta(x)). More precisely, preservation of distances gives one inclusion, and surjectivity gives the reverse inclusion. Thus DX1(x)D_{X_1}(x) is a singleton if and only if DX2(Δ(x))D_{X_2}(\Delta(x)) is a singleton. By Lemma 3.1,

PXi=SAi×{0}(i=1,2),P_{X_i}=S_{A_i}\times\{0\}\qquad(i=1,2),

and consequently Δ(SA1×{0})=SA2×{0}\Delta(S_{A_1}\times\{0\})=S_{A_2}\times\{0\}.

For i=1,2i=1,2, define

QXi={x∈SXi:the function z↦∥x−z∥Xi is constant on PXi}.Q_{X_i}=\{x\in S_{X_i}: \text{the function }z\mapsto\lVert x-z\rVert_{X_i}\text{ is constant on }P_{X_i}\}.

Because Δ\Delta preserves distances and maps PX1P_{X_1} onto PX2P_{X_2}, it also maps QX1Q_{X_1} onto QX2Q_{X_2}. We claim that QXi={0}×SBiQ_{X_i}=\{0\}\times S_{B_i}. To see this, let (a,b)∈SXi(a,b)\in S_{X_i}. By Lemma 3.2,

max⁡c∈SAi∥(a,b)−(c,0)∥Xipi=(1+∥a∥Ai)pi+1−∥a∥Aipi.\max_{c\in S_{A_i}}\lVert(a,b)-(c,0)\rVert_{X_i}^{p_i}=(1+\lVert a\rVert_{A_i})^{p_i}+1-\lVert a\rVert_{A_i}^{p_i}.

whereas

min⁡c∈SAi∥(a,b)−(c,0)∥Xipi=(1−∥a∥Ai)pi+1−∥a∥Aipi.\min_{c\in S_{A_i}}\lVert(a,b)-(c,0)\rVert_{X_i}^{p_i}=(1-\lVert a\rVert_{A_i})^{p_i}+1-\lVert a\rVert_{A_i}^{p_i}.

These maximum and minimum agree if and only if ∥a∥Ai=0\lVert a\rVert_{A_i}=0. Therefore, the distance to points of PXiP_{X_i} is constant exactly when a=0a=0. Since (a,b)∈SXi(a,b)\in S_{X_i}, a=0a=0 implies ∥b∥Bi=1\lVert b\rVert_{B_i}=1, proving the claim. It follows that

Δ({0}×SB1)={0}×SB2.\Delta(\{0\}\times S_{B_1})=\{0\}\times S_{B_2}.

Finally, choose a∈SA1a\in S_{A_1} and b∈SB1b\in S_{B_1}, which is possible because both spaces are nonzero. Their images have the form

Δ(a,0)=(a′,0),Δ(0,b)=(0,b′),\Delta(a,0)=(a',0),\qquad\Delta(0,b)=(0,b'),

with a′∈SA2a'\in S_{A_2} and b′∈SB2b'\in S_{B_2}. Preservation of their distance gives

21/p1=∥(a,0)−(0,b)∥X1=∥(a′,0)−(0,b′)∥X2=21/p2.2^{1/p_1}=\lVert(a,0)-(0,b)\rVert_{X_1}=\lVert(a',0)-(0,b')\rVert_{X_2}=2^{1/p_2}.

Because p↦21/pp\mapsto2^{1/p} is strictly decreasing, we conclude that p1=p2p_1=p_2, completing the proof. □

Proof of Theorem 1.1: (⟹)(\Longrightarrow) Assume that a surjective isometry Δ:SA1⊕p1B1→SA2⊕p2B2\Delta:S_{A_1\oplus_{p_1}B_1}\to S_{A_2\oplus_{p_2}B_2} exists. Lemma 3.3 gives p1=p2=:pp_1=p_2=:p and Δ(SA1×{0})=SA2×{0}\Delta(S_{A_1}\times\{0\})=S_{A_2}\times\{0\}. We write

Δ(a0,0)=(ϕ(a0),0),a0∈SA1,\Delta(a_0,0)=(\phi(a_0),0),\qquad a_0\in S_{A_1},

where ϕ:SA1→SA2\phi:S_{A_1}\to S_{A_2} is a surjective isometry. The last assertion of Lemma 3.1 gives

ϕ(−a0)=−ϕ(a0),a0∈SA1.(4)\phi(-a_0)=-\phi(a_0),\qquad a_0\in S_{A_1}. \tag*{(4)}

Let (a,b)∈SA1⊕p1B1(a,b)\in S_{A_1\oplus_{p_1}B_1}, and write Δ(a,b)=(a′,b′)\Delta(a,b)=(a',b'). By equality (2), the maximum distance to the first summand sphere determines ∥a∥\lVert a\rVert uniquely. It follows that

∥a′∥=∥a∥,∥b′∥=∥b∥.(5)\lVert a'\rVert=\lVert a\rVert,\qquad\lVert b'\rVert=\lVert b\rVert. \tag*{(5)}

If ∥a∥≠0\lVert a\rVert\ne0, by Lemma 3.2, the unique maximizer −a/∥a∥-a/\lVert a\rVert is carried onto the unique maximizer −a′/∥a′∥-a'/\lVert a'\rVert for (a′,b′)(a',b'). Consequently, equalities (5) yield

ϕ(−a/∥a∥)=−a′/∥a′∥.\phi(-a/\lVert a\rVert)=-a'/\lVert a'\rVert.

Together with equality (4), this proves that the first coordinate of Δ(a,b)\Delta(a,b) is U0(a)U_0(a), where

U0(0)=0,U0(a)=∥a∥ϕ(a/∥a∥),0≠a∈BA1.(6)U_0(0)=0,\qquad U_0(a)=\lVert a\rVert\phi(a/\lVert a\rVert),\qquad0\ne a\in B_{A_1}. \tag*{(6)}

For a=0a=0 the same conclusion follows from equalities (5). The definition (6) shows that U0:BA1→BA2U_0:B_{A_1}\to B_{A_2} is a norm-preserving bijection. Its inverse is defined radially from ϕ−1\phi^{-1}, and the first coordinate of Δ−1(c,d)\Delta^{-1}(c,d) is U0−1(c)U_0^{-1}(c).

We now extend U0U_0 to the whole space A1A_1. Take a,c∈BA1a,c\in B_{A_1}. Fix w∈SB1w\in S_{B_1}, and write

Δ(a,(1−∥a∥p)1/pw)=(U0(a),v),Δ(c,(1−∥c∥p)1/pw)=(U0(c),z).\Delta\left(a,\left(1-\lVert a\rVert^p\right)^{1/p}w\right)=(U_0(a),v),\qquad\Delta\left(c,\left(1-\lVert c\rVert^p\right)^{1/p}w\right)=(U_0(c),z).

By equalities (5), ∥v∥=(1−∥a∥p)1/p\lVert v\rVert=\left(1-\lVert a\rVert^p\right)^{1/p} and ∥z∥=(1−∥c∥p)1/p\lVert z\rVert=\left(1-\lVert c\rVert^p\right)^{1/p}. Consequently,

∥U0a−U0c∥p+∣(1−∥a∥p)1/p−(1−∥c∥p)1/p∣p≤∥U0(a)−U0(c)∥p+∥v−z∥p=∥(a,(1−∥a∥p)1/pw)−(c,(1−∥c∥p)1/pw)∥p=∥a−c∥p+∣(1−∥a∥p)1/p−(1−∥c∥p)1/p∣p.\begin{aligned} \lVert U_0a-U_0c\rVert^p+\left|\left(1-\lVert a\rVert^p\right)^{1/p}-\left(1-\lVert c\rVert^p\right)^{1/p}\right|^p\leq\lVert U_0(a)-U_0(c)\rVert^p+\lVert v-z\rVert^p \\ &=\left\lVert\left(a,\left(1-\lVert a\rVert^p\right)^{1/p}w\right)-\left(c,\left(1-\lVert c\rVert^p\right)^{1/p}w\right)\right\rVert^p \\ &=\lVert a-c\rVert^p+\left|\left(1-\lVert a\rVert^p\right)^{1/p}-\left(1-\lVert c\rVert^p\right)^{1/p}\right|^p. \end{aligned}

Thus U0U_0 is nonexpansive, namely, ∥U0(a)−U0(c)∥≤∥a−c∥\lVert U_0(a)-U_0(c)\rVert\leq\lVert a-c\rVert. Applying precisely this argument to Δ−1\Delta^{-1} and a fixed unit vector in B2B_2 shows that U0−1U_0^{-1} is nonexpansive. Hence U0U_0 is a surjective isometry between the closed balls. Theorem 2.4 provides a surjective real-linear isometry U:A1→A2U:A_1\to A_2 extending U0U_0.

We next show that the second coordinate of Δ(a,b)\Delta(a,b) is determined solely by bb. Let (a,b),(c,b)∈SA1⊕pB1(a,b),(c,b)\in S_{A_1\oplus_p B_1} and write their images as (U(a),v)(U(a),v) and (U(c),z)(U(c),z). Since UU is an isometry, we have

∥a−c∥p=∥(a,b)−(c,b)∥p=∥(U(a),v)−(U(c),z)∥p=∥U(a)−U(c)∥p+∥v−z∥p=∥a−c∥p+∥v−z∥p.\begin{aligned} \lVert a-c\rVert^p=\lVert(a,b)-(c,b)\rVert^p \\ &=\lVert(U(a),v)-(U(c),z)\rVert^p \\ &=\lVert U(a)-U(c)\rVert^p+\lVert v-z\rVert^p \\ &=\lVert a-c\rVert^p+\lVert v-z\rVert^p. \end{aligned}

Therefore v=zv=z. For every b∈BB1b\in B_{B_1}, choose a∈A1a\in A_1 of norm (1−∥b∥p)1/p\left(1-\lVert b\rVert^p\right)^{1/p}. We have thus defined a map V0:BB1→BB2V_0:B_{B_1}\to B_{B_2} such that

Δ(a,b)=(U(a),V0(b)),(a,b)∈SA1⊕pB1.(7)\Delta(a,b)=(U(a),V_0(b)),\qquad(a,b)\in S_{A_1\oplus_p B_1}. \tag*{(7)}

Equation (5) implies V0(0)=0V_0(0)=0.

We now extend V0V_0 to the whole space B1B_1, and conclude the required statement. For b,d∈BB1b,d\in B_{B_1} choose a,ca,c so that (a,b),(c,d)∈SA1⊕pB1(a,b),(c,d)\in S_{A_1\oplus_p B_1}. Equation (7) and the isometry of Δ\Delta give

∥a−c∥p+∥b−d∥p=∥U(a)−U(c)∥p+∥V0(b)−V0(d)∥p.\lVert a-c\rVert^p+\lVert b-d\rVert^p=\lVert U(a)-U(c)\rVert^p+\lVert V_0(b)-V_0(d)\rVert^p.

Since UU is an isometry, the preceding equality yields ∥V0(b)−V0(d)∥=∥b−d∥\lVert V_0(b)-V_0(d)\rVert=\lVert b-d\rVert. For z∈BB2z\in B_{B_2} choose c∈A2c\in A_2 such that (c,z)(c,z) lies on the target sphere. A preimage under Δ\Delta shows that zz is in the range of V0V_0. Hence V0V_0 is a surjective ball isometry, and we know that V0(0)=0V_0(0)=0 from the previous paragraph. Theorem 2.4 supplies a surjective real-linear isometry V:B1→B2V:B_1\to B_2 extending V0V_0 such that

Δ(a,b)=(U(a),V(b)),(a,b)∈SA1⊕pB1.\Delta(a,b)=(U(a),V(b)),\qquad(a,b)\in S_{A_1\oplus_p B_1}.

This proves formula (1).

(⟸)(\Longleftarrow) Conversely, if p1=p2=:pp_1=p_2=:p and U,VU,V are surjective real-linear isometries, as stated, then

∥(U⊕V)(a,b)∥A2⊕pB2p=∥U(a)∥A2p+∥V(b)∥B2p=∥a∥A1p+∥b∥B1p=∥(a,b)∥A1⊕pB1p,\lVert(U\oplus V)(a,b)\rVert_{A_2\oplus_p B_2}^p=\lVert U(a)\rVert_{A_2}^p+\lVert V(b)\rVert_{B_2}^p=\lVert a\rVert_{A_1}^p+\lVert b\rVert_{B_1}^p=\lVert(a,b)\rVert_{A_1\oplus_p B_1}^p,

showing that U⊕V:A1⊕pB1→A2⊕pB2U\oplus V:A_1\oplus_p B_1\to A_2\oplus_p B_2 is a surjective real-linear isometry. Immediately, (U⊕V)∣SA1⊕pB1(U\oplus V)|_{S_{A_1\oplus_p B_1}} is a surjective isometry. Finally, a real-linear extension of a sphere map is determined on each x≠0x\ne0 by radial formula (1). Thus, Lemma 2.5 proves all uniqueness claims. □

Remark 3.4. Theorem 1.1 does not assume finite dimensionality, smoothness, or a prior solution of Tingley’s problem for AiA_i or BiB_i. Its conclusion is nevertheless a theorem about pairs of spaces with the specified decompositions; it does not assert the Mazur–Ulam property with an arbitrary target Banach space.

Example 3.5. Let AA be a nonzero strictly convex Banach space and let E,FE,F be arbitrary nonzero Banach spaces. By Proposition 2.2, Theorem 1.1 applies to both

A⊕p(E⊕1F),A⊕p(E⊕∞F),1<p<∞.A \oplus_p (E \oplus_1 F), \qquad A \oplus_p (E \oplus_\infty F), \qquad1 < p < \infty.

Examples include R⊕pℓ13\mathbb{R} \oplus_p \ell_1^3 and H⊕p(ℓ1⊕∞L2[0,1])H \oplus_p (\ell_1 \oplus_\infty L^2[0,1]), where HH is a nonzero Hilbert space. No strict convexity is required of EE or FF.

Applications to function spaces

We first apply Theorem 1.1 to absolutely continuous functions whose norms combine evaluation at a fixed point with the L1L^1-norm of the derivative. We then record shorter applications to Lipschitz and differentiable functions, specifying their relation to existing extension theorems.

Throughout this section, II denotes a nondegenerate compact interval, x0∈Ix_0 \in I, and K∈{R,C}\mathbb{K} \in\{\mathbb{R},\mathbb{C}\}. All extension assertions concern the underlying real Banach spaces. For s,t≥0s,t \ge0, put

Np(s,t)=(sp+tp)1/p(1≤p<∞),N∞(s,t)=max⁡{s,t}.N_p(s,t) = (s^p+t^p)^{1/p} \quad(1 \le p < \infty), \qquad N_\infty(s,t) = \max\{s,t\}.

Integrals with reversed endpoints are understood as oriented integrals. All LqL^q-spaces on intervals use the Lebesgue measure.

Absolutely continuous functions

Let ACp(I,x0;K)\mathrm{AC}^p(I,x_0;\mathbb{K}) denote AC(I,K)\mathrm{AC}(I,\mathbb{K}) equipped with

∥f∥AC,p,x0=Np(∣f(x0)∣,∥f′∥1),1≤p≤∞.(8)\|f\|_{\mathrm{AC},p,x_0}=N_p\bigl(|f(x_0)|,\|f'\|_1\bigr), \qquad1 \le p \le\infty. \tag*{(8)}

Here pp combines the two displayed quantities; the derivative belongs to L1(I,K)L^1(I,\mathbb{K}) for every pp. The fundamental theorem for absolutely continuous functions gives the surjective K\mathbb{K}-linear isometry

JAC:ACp(I,x0;K)→K⊕pL1(I,K),JAC(f)=(f(x0),f′),(9)J_{\mathrm{AC}}:\mathrm{AC}^p(I,x_0;\mathbb{K})\to\mathbb{K}\oplus_p L^1(I,\mathbb{K}), \qquad J_{\mathrm{AC}}(f)=(f(x_0),f'), \tag*{(9)}

whose inverse is

JAC−1(c,u)(x)=c+∫x0xu(t) dt.(10)J_{\mathrm{AC}}^{-1}(c,u)(x)=c+\int_{x_0}^{x}u(t)\,\mathrm{d}t. \tag*{(10)}

In particular, ACp(I,x0;K)\mathrm{AC}^p(I,x_0;\mathbb{K}) is complete for all 1≤p≤∞1\le p\le\infty.

Lemma 4.1. The spaces L1(I,K)L^1(I,\mathbb{K}) and L∞(I,K)L^\infty(I,\mathbb{K}) satisfy condition (D).

Proof. Partition II into measurable sets E,FE,F of positive measure. Restriction gives

L1(I,K)≅L1(E,K)⊕1L1(F,K),L∞(I,K)≅L∞(E,K)⊕∞L∞(F,K).L^1(I,\mathbb{K})\cong L^1(E,\mathbb{K})\oplus_1 L^1(F,\mathbb{K}), \qquad L^\infty(I,\mathbb{K})\cong L^\infty(E,\mathbb{K})\oplus_\infty L^\infty(F,\mathbb{K}).

Every component is nonzero, so Proposition 2.2 applies to both decompositions. □\square

Proposition 4.2. Fix K∈{R,C}\mathbb{K}\in\{\mathbb{R},\mathbb{C}\}. For i=1,2i=1,2, let IiI_i be a nondegenerate compact interval, xi∈Iix_i\in I_i, and 1<pi<∞1<p_i<\infty. Then there exists a surjective isometry Δ:SACp1(I1,x1;K)→SACp2(I2,x2;K)\Delta:S_{\mathrm{AC}^{p_1}}(I_1,x_1;\mathbb{K})\to S_{\mathrm{AC}^{p_2}}(I_2,x_2;\mathbb{K}) if and only if p1=p2p_1=p_2 and there exist surjective real-linear isometries U:K→KU:\mathbb{K}\to\mathbb{K} and V:L1(I1,K)→L1(I2,K)V:L^1(I_1,\mathbb{K})\to L^1(I_2,\mathbb{K}). Every such isometry Δ\Delta extends uniquely to a surjective real-linear isometry Δ~\widetilde{\Delta} of the form

Δ~(f)(x)=U(f(x1))+∫x2xV(f′)(t) dt,x∈I2,(11)\widetilde{\Delta}(f)(x)=U(f(x_1))+\int_{x_2}^{x}V(f')(t)\,\mathrm{d}t,\qquad x\in I_2, \tag*{(11)}

where UU and VV are uniquely determined by Δ\Delta.

Proof. The underlying real space of K\mathbb{K} is strictly convex, and the second components in (9) satisfy (D) by Lemma 4.1. Conjugating Δ\Delta by the restrictions of the maps JACJ_{\mathrm{AC}} to the unit spheres and applying Theorem 1.1 gives p1=p2p_1=p_2 and the unique component isometries U,VU,V. Formula (10) then gives (11). Conversely, since L1(I1,K)L^1(I_1,\mathbb{K}) and L1(I2,K)L^1(I_2,\mathbb{K}) are surjectively real-linearly isometric, the existence assertion follows from Theorem 1.1. □\square

The following representation makes the derivative component in (11) explicit. Put κ0(z)=z\kappa_0(z)=z and, for complex scalars, κ1(z)=z‾\kappa_1(z)=\overline{z}. A measurable bijection modulo null sets means a bimeasurable bijection between full-measure measurable subsets such that both the bijection and its inverse preserve null sets.

Corollary 4.3. Under the hypotheses of Proposition 4.2, every extension has the form

Δ~(f)(x)=ωκε0(f(x1))+∫x2xρ(t)h(t)κε(t)(f′(τ(t))) dt.(12)\widetilde{\Delta}(f)(x)=\omega\kappa_{\varepsilon_0}\bigl(f(x_1)\bigr)+\int_{x_2}^{x}\rho(t)h(t)\kappa_{\varepsilon(t)}\bigl(f'(\tau(t))\bigr)\,\mathrm{d}t. \tag*{(12)}

Here τ:I2→I1\tau:I_2\to I_1 is a measurable bijection modulo null sets, and ρ\rho is measurable, positive and finite almost everywhere, with

∫I2ρ(t)g(τ(t)) dt=∫I1g(s) ds(13)\int_{I_2}\rho(t)g(\tau(t))\,\mathrm{d}t=\int_{I_1}g(s)\,\mathrm{d}s \tag*{(13)}

for every nonnegative measurable function gg on I1I_1. Also, ∣ω∣=1|\omega|=1, ε0∈{0,1}\varepsilon_0\in\{0,1\}, and h:I2→Th:I_2\to\mathbb{T} and ε:I2→{0,1}\varepsilon:I_2\to\{0,1\} are measurable, where T={z∈C:∣z∣=1}\mathbb{T}=\{z\in\mathbb{C}:|z|=1\}. When K=R\mathbb{K}=\mathbb{R}, we require ω∈{−1,1}\omega\in\{-1,1\}, h(t)∈{−1,1}h(t)\in\{-1,1\} almost everywhere, and ε0=ε=0\varepsilon_0=\varepsilon=0.

Proof. By Proposition 4.2, we have

Δ~(f)(x)=U(f(x1))+∫x2xV(f′)(t) dt,\widetilde{\Delta}(f)(x)=U(f(x_1))+\int_{x_2}^{x}V(f')(t)\,\mathrm{d}t,

where U:K→KU:\mathbb{K}\to\mathbb{K} and V:L1(I1,K)→L1(I2,K)V:L^1(I_1,\mathbb{K})\to L^1(I_2,\mathbb{K}) are surjective real-linear isometries. Every surjective real-linear isometry of K\mathbb{K} has the form

U(z)=ωκε0(z),U(z)=\omega\kappa_{\varepsilon_0}(z),

where ∣ω∣=1|\omega|=1, ε0∈{0,1}\varepsilon_0\in\{0,1\}, κ0(z)=z\kappa_0(z)=z, and κ1(z)=z‾\kappa_1(z)=\overline{z}. In the real case, ω∈{−1,1}\omega\in\{-1,1\} and $\varepsilon_0=0.

We next describe VV. Regard K\mathbb{K} as a real Banach space, namely R\mathbb{R} or the Euclidean plane R2\mathbb{R}^2. In either case, K\mathbb{K} is separable and strictly convex, and hence cannot be expressed as an ℓ1\ell^1-direct sum of two nonzero Banach spaces. Therefore Sourour’s representation theorem [12] applies to VV. The two intervals may first be identified by an affine bijection, together with the corresponding normalization of the L1L^1-norm. Sourour’s theorem yields an isomorphism between the underlying measure algebras, a positive measurable weight, and a measurable family of surjective real-linear isometries of K\mathbb{K}.

Since I1I_1 and I2I_2, equipped with the Lebesgue measure, are standard nonatomic finite measure spaces, the measure-algebra isomorphism can be represented by a measurable bijection τ:I2→I1\tau:I_2\to I_1 modulo null sets [4 344C]. Consequently, there exist a measurable function ρ:I2→(0,∞)\rho:I_2\to(0,\infty), finite almost everywhere, and a measurable family of surjective real-linear isometries Tt:K→KT_t:\mathbb{K}\to\mathbb{K} such that

V(u)(t)=ρ(t)Tt(u(τ(t)))V(u)(t)=\rho(t)T_t\bigl(u(\tau(t))\bigr)

for every u∈L1(I1,K)u\in L^1(I_1,\mathbb{K}) and almost every t∈I2t\in I_2.

Each TtT_t is multiplication by a sign when K=R\mathbb{K}=\mathbb{R}. When K=C≃R2\mathbb{K}=\mathbb{C}\simeq\mathbb{R}^2, regarded as a real Banach space, every such isometry is either multiplication by a unimodular scalar or multiplication by a unimodular scalar followed by complex conjugation. Thus there exist measurable functions h:I2→Th:I_2\to\mathbb{T} and ε:I2→{0,1}\varepsilon:I_2\to\{0,1\} such that

Tt(z)=h(t)κε(t)(z)T_t(z)=h(t)\kappa_{\varepsilon(t)}(z)

for almost every t∈I2t\in I_2. In the real case, h(t)∈{−1,1}h(t)\in\{-1,1\} almost everywhere and ε=0\varepsilon=0. Hence

V(u)(t)=ρ(t)h(t)κε(t)(u(τ(t)))a.e.V(u)(t)=\rho(t)h(t)\kappa_{\varepsilon(t)}\bigl(u(\tau(t))\bigr)\quad\text{a.e.}

It remains to verify equality (13). Let gg be a nonnegative integrable function on I1I_1, and choose a fixed e∈Ke\in\mathbb{K} with ∣e∣=1|e|=1. Applying the preceding representation to u(s)=g(s)eu(s)=g(s)e, and using the fact that VV and each TtT_t are isometries, we obtain

∫I1g(s) ds=∥u∥1=∥V(u)∥1=∫I2ρ(t)∣Tt(g(τ(t))e)∣ dt=∫I2ρ(t)g(τ(t)) dt.\int_{I_1}g(s)\,\mathrm{d}s=\lVert u\rVert_1=\lVert V(u)\rVert_1=\int_{I_2}\rho(t)\lvert T_t(g(\tau(t))e)\rvert\,\mathrm{d}t=\int_{I_2}\rho(t)g(\tau(t))\,\mathrm{d}t.

By applying this identity to truncations and using the monotone convergence theorem, the equality extends to every nonnegative measurable function gg on I1I_1. This proves equality (13). Finally, substituting the representations of UU and VV into formula (11) yields (12), completing the proof. □

Remark 4.4. Proposition 4.2 uses condition (D) of the L1L^1-component directly. Wang’s [16 Theorem 1.8] is formulated for specified C0(Ω,E)C_0(\Omega,E)-type components and a common exponent. The applicability of Wang’s theorem to an L1L^1-component does not follow merely from the identity L1(I,K)≅C0({∗},L1(I,K))L^1(I,\mathbb{K}) \cong C_0(\{\ast\},L^1(I,\mathbb{K})): the singleton component in that theorem is required to be strictly convex, whereas L1(I,K)L^1(I,\mathbb{K}) is not strictly convex. The representation in Corollary 4.3 uses the existing classification of linear L1L^1-isometries; the extension assertion is supplied by Theorem 1.1.

The following proposition establishes that AC1(I,x0;R)AC^1(I,x_0;\mathbb{R}) has the Mazur–Ulam property.

Proposition 4.5. The real Banach space AC1(I,x0;R)AC^1(I,x_0;\mathbb{R}) has the Mazur–Ulam property.

Proof. Let I~=I⊔{∗}\widetilde{I}=I\sqcup\{\ast\} and μ=m+δ∗\mu=m+\delta_\ast, where mm is the Lebesgue measure and δ∗\delta_\ast is the Dirac measure at {∗}\{\ast\}. The map

Q(f):=f(x0)χ{∗}+f′χI,f∈AC1(I,x0;R)Q(f):=f(x_0)\chi_{\{\ast\}}+f'\chi_I,\qquad f\in AC^1(I,x_0;\mathbb{R})

is a surjective real-linear isometry onto L1(I~,μ;R)L^1(\widetilde{I},\mu;\mathbb{R}) by formula (10). Since μ\mu is finite, Tan’s theorem [13] applies. Let Δ:SAC1(I,x0;R)→SY\Delta:S_{AC^1(I,x_0;\mathbb{R})}\to S_Y be a surjective isometry, where YY is a real Banach space. Then Δ∘Q−1\Delta\circ Q^{-1} extends to a surjective real-linear isometry

R:L1(I~,μ;R)→Y.R:L^1(\widetilde{I},\mu;\mathbb{R})\to Y.

The operator R∘QR\circ Q extends Δ\Delta. Uniqueness follows from Lemma 2.5. □

Notice that the endpoint conclusion in Proposition 4.5 is a consequence of the existing L1L^1 extension theorem.

Lipschitz functions

For f∈Lip⁡(I,K)f\in\operatorname{Lip}(I,\mathbb{K}), define

L(f):=sup⁡s≠t∣f(s)−f(t)∣∣s−t∣,∥f∥Lip⁡,p,x0:=Np(∣f(x0)∣,L(f)),L(f):=\sup_{s\ne t}\frac{\lvert f(s)-f(t)\rvert}{\lvert s-t\rvert},\qquad\lVert f\rVert_{\operatorname{Lip},p,x_0}:=N_p(\lvert f(x_0)\rvert,L(f)),

and denote the resulting space by Lip⁡p(I,x0;K)\operatorname{Lip}^p(I,x_0;\mathbb{K}). Since II is a compact interval, L(f)=∥f′∥L∞L(f)=\lVert f'\rVert_{L^\infty}. Consequently, the map

JLip⁡(f):=(f(x0),f′):Lip⁡p(I,x0;K)→K⊕pL∞(I,K)(14)J_{\operatorname{Lip}}(f):=(f(x_0),f'):\operatorname{Lip}^p(I,x_0;\mathbb{K})\to\mathbb{K}\oplus_p L^\infty(I,\mathbb{K}) \tag*{(14)}

is a surjective K\mathbb{K}-linear isometry, with inverse (c,u)↦c+∫x0 ⋅u(t) dt(c,u)\mapsto c+\int_{x_0}^{\,\cdot}u(t)\,\mathrm{d}t.

Proposition 4.6. Fix K∈{R,C}\mathbb{K}\in\{\mathbb{R},\mathbb{C}\}. Suppose 1<pi<∞1<p_i<\infty. Then SLip⁡p1(I1,x1;K)S_{\operatorname{Lip}^{p_1}(I_1,x_1;\mathbb{K})} is surjectively isometric to SLip⁡p2(I2,x2;K)S_{\operatorname{Lip}^{p_2}(I_2,x_2;\mathbb{K})} if and only if p1=p2=:pp_1=p_2=:p. For every common exponent 1≤p≤∞1\le p\le\infty, every surjective isometry Δ:SLip⁡p(I1,x1;K)→SLip⁡p(I2,x2;K)\Delta:S_{\operatorname{Lip}^{p}(I_1,x_1;\mathbb{K})}\to S_{\operatorname{Lip}^{p}(I_2,x_2;\mathbb{K})} extends uniquely to a surjective real-linear isometry Δ~\widetilde{\Delta} of the form

Δ~(f)(x)=U(f(x1))+∫x2xV(f′)(t) dt(15)\widetilde{\Delta}(f)(x)=U(f(x_1))+\int_{x_2}^{x}V(f')(t)\,\mathrm{d}t \tag*{(15)}

where U:K→KU:\mathbb{K}\to\mathbb{K} and V:L∞(I1,K)→L∞(I2,K)V:L^\infty(I_1,\mathbb{K})\to L^\infty(I_2,\mathbb{K}) are surjective real-linear isometries uniquely determined by Δ\Delta.

Proof. For 1<pi≤∞1<p_i\le\infty, apply Theorem 1.1 to (14), using Lemma 4.1. Existence for equal exponents follows by taking UU to be the identity and V(u)=u∘τV(u)=u\circ\tau, where τ:I2→I1\tau:I_2\to I_1 is an increasing affine bijection.

For the endpoints, write L∞(Ii,K)≅C(Ki,K)L^\infty(I_i,\mathbb{K})\cong C(K_i,\mathbb{K}) using its scalar continuous-function representation. For p=1p=1, apply [17 Theorem 3.5] to C({∗},G)⊕1C(Ki,G)C(\{\ast\},G)\oplus_1 C(K_i,G), where GG is the underlying real space of K\mathbb{K}. Neither component is isometric to real ℓ∞∞\ell_\infty^\infty: GG is strictly convex, and the other component is infinite-dimensional. The component representation in [17 Theorems 3.3 and 3.4] therefore applies. The finite-dimensional component must be preserved, giving (15).

For p=∞p=\infty, the spaces identify with C(Ki⊔{∗},K)C(K_i\sqcup\{\ast\},\mathbb{K}). The scalar continuous-function extension results [15, 9] give a real-linear extension. The compact space KiK_i has no isolated points, since the Lebesgue measure algebra has no atoms. The real Banach–Stone theorem [3] consequently preserves the unique isolated point ∗\ast, again giving (15).

For a common 1<p<∞1<p<\infty, the extension and component representation above also follow from [16 Theorem 1.8], since L∞(Ii,K)≅C(Ki,K)L^\infty(I_i,\mathbb{K})\cong C(K_i,\mathbb{K}). Theorem 1.1 additionally compares possibly different exponents. The endpoint assertions were obtained from the cited theorems. In particular, the complex p=1p=1 extension problem with norm ∣f(0)∣+∥f′∥L∞|f(0)|+\lVert f'\rVert_{L^\infty} was studied by Hirota and Miura [5]; see also [2].

Remark 4.7 (Explicit Lipschitz extensions). By the real Banach–Stone representation of the L∞L^\infty component and [4 344C], formula (15) can be rewritten as

Δ~(f)(x)=ωκε0(f(x1))+∫x2xh(t)κε(t)(f′(τ(t))) dt.(16)\widetilde{\Delta}(f)(x)=\omega\kappa_{\varepsilon_0}\bigl(f(x_1)\bigr)+\int_{x_2}^{x}h(t)\kappa_{\varepsilon(t)}\bigl(f'(\tau(t))\bigr)\,dt. \tag*{(16)}

The symbols ω,ε0,h,ε,τ\omega,\varepsilon_0,h,\varepsilon,\tau have the meanings specified in Corollary 4.3. Every such choice gives a surjective real-linear isometry. In fact, the null-set preservation in both directions implies ∥u∘τ∥∞=∥u∥∞\lVert u\circ\tau\rVert_\infty=\lVert u\rVert_\infty, and the pointwise scalar isometries and τ\tau are invertible modulo null sets. The representation (16) contains no density factor, since the derivative norm is the essential supremum norm.

Continuous and differentiable functions

We first verify condition (D) for the continuous-function components needed for the application: every C(K,E)C(K,E) with KK compact Hausdorff, ∣K∣≥2|K|\ge2, and E≠{0}E\ne\{0\} satisfies (D). To verify condition (D), let u∈SC(K,E)u\in S_{C(K,E)}, choose t0t_0 with ∥u(t0)∥=1\lVert u(t_0)\rVert=1, a point t1≠t0t_1\ne t_0, and η∈C(K,[0,1])\eta\in C(K,[0,1]) with η(t0)=0\eta(t_0)=0, η(t1)=1\eta(t_1)=1. Writing e=u(t0)e=u(t_0), the distinct functions

v1(t)=−e,v2(t)=−(1−η(t))ev_1(t)=-e,\qquad v_2(t)=-(1-\eta(t))e

belong to SC(K,E)S_{C(K,E)} and satisfy ∥u−v1∥∞=∥u−v2∥∞=2\lVert u-v_1\rVert_\infty=\lVert u-v_2\rVert_\infty=2.

Let EE be a nonzero real or complex Banach space and n≥1n\ge1. Equip Cn(I,E)C^n(I,E) with

∥f∥n,p,x0={(∑j=0n−1∥f(j)(x0)∥Ep+∥f(n)∥∞p)1/p,1≤p<∞,max⁡{max⁡0≤j≤n∥f(j)(x0)∥E,∥f(n)∥∞},p=∞,(17)\lVert f\rVert_{n,p,x_0}= \begin{cases} \left(\sum_{j=0}^{n-1}\lVert f^{(j)}(x_0)\rVert_E^p+\lVert f^{(n)}\rVert_\infty^p\right)^{1/p}, & 1\le p<\infty,\\ \max\left\{\max_{0\le j\le n}\lVert f^{(j)}(x_0)\rVert_E,\lVert f^{(n)}\rVert_\infty\right\}, & p=\infty, \end{cases} \tag*{(17)}

and write Cpn(I,x0;E)C_p^n(I,x_0;E) for the normed space (Cn(I,E),∥⋅∥n,p,x0)(C^n(I,E),\lVert\cdot\rVert_{n,p,x_0}). Derivatives at endpoints are understood as one-sided derivatives.

Proposition 4.8. For 1≤p≤∞1\le p\le\infty, n≥1n\ge1, the map

J(f):=((f(j)(x0))j=0n−1,f(n))J(f):=\left(\left(f^{(j)}(x_0)\right)_{j=0}^{n-1},f^{(n)}\right)

is a surjective real-linear isometry from Cpn(I,x0;E)C_p^n(I,x_0;E) onto ℓpn(E)⊕pC(I,E)\ell_p^n(E)\oplus_p C(I,E). Its inverse is

J−1(a,u)(x)=∑j=0n−1aj(x−x0)jj!+1(n−1)!∫x0x(x−t)n−1u(t) dt.(18)J^{-1}(a,u)(x)=\sum_{j=0}^{n-1}a_j\frac{(x-x_0)^j}{j!}+\frac{1}{(n-1)!}\int_{x_0}^{x}(x-t)^{n-1}u(t)\,dt. \tag*{(18)}

In particular, Cpn(I,x0;E)C_p^n(I,x_0;E) is a Banach space.

Proof. For u∈C(I,E)u \in C(I,E) and m≥1m \ge1, define the Bochner integral

Rmu(x)=1(m−1)!∫x0x(x−t)m−1u(t) dt.R_m u(x)=\frac{1}{(m-1)!}\int_{x_0}^{x}(x-t)^{m-1}u(t)\,\mathrm{d}t.

The fundamental theorem of calculus gives (R1u)′=u(R_1u)'=u. For m≥2m \ge2, differentiation under the integral sign gives (Rmu)′=Rm−1u(R_mu)'=R_{m-1}u; the boundary term vanishes. These identities hold with one-sided derivatives at the endpoints, and Rmu(x0)=0R_mu(x_0)=0 for every m≥1m \ge1. Thus the function FF on the right of (18) belongs to Cn(I,E)C^n(I,E) and satisfies

F(k)(x)=∑j=kn−1aj(x−x0)j−k(j−k)!+Rn−ku(x)(0≤k<n),F(n)=u.F^{(k)}(x)=\sum_{j=k}^{n-1}a_j\frac{(x-x_0)^{j-k}}{(j-k)!}+R_{n-k}u(x)\quad(0\le k<n),\qquad F^{(n)}=u.

Consequently, F(k)(x0)=akF^{(k)}(x_0)=a_k and J(F)=(a,u)J(F)=(a,u). Hence, JJ is surjective. If J(f)=0J(f)=0, successive applications of

f(k)(x)=f(k)(x0)+∫x0xf(k+1)(t) dtf^{(k)}(x)=f^{(k)}(x_0)+\int_{x_0}^{x}f^{(k+1)}(t)\,\mathrm{d}t

for k=n−1,…,0k=n-1,\ldots,0 give f=0f=0, proving injectivity. Linearity is immediate, and the norm (17) gives ∥J(f)∥=∥f∥n,p,x0\|J(f)\|=\|f\|_{n,p,x_0}, including the maximum norm when p=∞p=\infty. Completeness follows from completeness of the two components. □

Corollary 4.9. For i=1,2i=1,2, let EiE_i be nonzero strictly convex Banach spaces, ni≥1n_i\ge1, and 1<pi<∞1<p_i<\infty. Then there exists a surjective isometry Δ:Sp1n1(I1,x1;E1)→Sp2n2(I2,x2;E2)\Delta:S^{n_1}_{p_1}(I_1,x_1;E_1)\to S^{n_2}_{p_2}(I_2,x_2;E_2) if and only if p1=p2=:pp_1=p_2=:p and there exist surjective real-linear isometries U:ℓpn1(E1)→ℓpn2(E2)U:\ell_{p}^{n_1}(E_1)\to\ell_{p}^{n_2}(E_2) and V:C(I1,E1)→C(I2,E2)V:C(I_1,E_1)\to C(I_2,E_2). Moreover, every such surjective isometry Δ\Delta extends uniquely to

Δ~:=J2−1(U⊕V)J1,(19)\widetilde{\Delta}:=J_2^{-1}(U\oplus V)J_1, \tag*{(19)}

where UU and VV are uniquely determined by Δ\Delta. For a fixed scalar field E1=E2=KE_1=E_2=\mathbb{K}, such Δ\Delta exists if and only if (n1,p1)=(n2,p2)(n_1,p_1)=(n_2,p_2).

Proof. For 1<p<∞1<p<\infty, the finite ℓp\ell_p-sum of strictly convex spaces is strictly convex. Apply Theorem 1.1 through Proposition 4.8, using condition (D) of C(Ii,Ei)C(I_i,E_i).

In the scalar case, the real dimensions of the first components give n1=n2n_1=n_2. Conversely, if (n1,p1)=(n2,p2)(n_1,p_1)=(n_2,p_2), take UU to be the identity and define V(u)=u∘τV(u)=u\circ\tau, where τ:I2→I1\tau:I_2\to I_1 is a homeomorphism. Formula (19) gives the required isometry. □

For a common exponent 1<p<∞1<p<\infty, Corollary 4.9 is already covered by [16 Theorem 1.8]. More generally, Wang’s Theorem 1.8 covers A⊕pC(K,E)A\oplus_p C(K,E) when AA is strictly convex and ∣K∣≥2|K|\ge2, without requiring strict convexity of EE. The additional value-space assumptions in Wang’s Theorem 1.10 concern pointwise representations.

Remark 4.10 (Known endpoint consequences). For p=1p=1 and strictly convex EiE_i, the decomposition into nin_i copies of EiE_i and one copy of C(Ii,Ei)C(I_i,E_i) allows the application of [17 Theorems 3.3–3.5]. Each component is of C0(Ω,Ei)C_0(\Omega,E_i) type. None is isometric to real ℓ∞2\ell_\infty^2, and C(Ii,Ei)C(I_i,E_i) is the unique component that is not strictly convex. Consequently, every surjective isometry between the unit spheres extends, n1=n2=nn_1=n_2=n, and the extension is

Δ~(f)(x)=∑j=0n−1Uj(f(π(j))(x1))(x−x2)jj!+1(n−1)!∫x2x(x−t)n−1V(f(n))(t) dt.\widetilde{\Delta}(f)(x)=\sum_{j=0}^{n-1}U_j\left(f^{(\pi(j))}(x_1)\right)\frac{(x-x_2)^j}{j!}+\frac{1}{(n-1)!}\int_{x_2}^{x}(x-t)^{n-1}V\left(f^{(n)}\right)(t)\,\mathrm{d}t.

where π\pi is a permutation and Uj:E1→E2U_j:E_1\to E_2 and V:C(I1,E1)→C(I2,E2)V:C(I_1,E_1)\to C(I_2,E_2) are surjective real-linear isometries. The converse follows from the decomposition.

For scalar-valued functions and p=∞p=\infty,

C∞n(I,x0;K)≅C(I⊔Dn,K),C_\infty^n(I,x_0;\mathbb{K})\cong C(I\sqcup D_n,\mathbb{K}),

where DnD_n is a discrete set of nn points. The scalar continuous-function extension theorems [15, 9], followed by the real Banach–Stone theorem, give extension and equality of the orders, since a homeomorphism preserves the isolated points. For general EE, the analogous identification is with C(I⊔Dn,E)C(I \sqcup D_n,E); an extension conclusion then requires a suitable vector-valued continuous-function theorem.

Remark 4.11 (Scalar representation). For a common n≥1n \ge1 and 1≤p≤∞1 \le p \le\infty, the scalar extensions above can be written as

Δ~(f)(x)=∑j=0n−1(Uaf)j(x−x2)jj!+1(n−1)!∫x2x(x−t)n−1h(t)κε(f(n)(τ(t))) dt,af=(f(j)(x1))j=0n−1.\widetilde{\Delta}(f)(x)=\sum_{j=0}^{n-1}(Ua_f)_j\frac{(x-x_2)^j}{j!}+\frac{1}{(n-1)!}\int_{x_2}^{x}(x-t)^{n-1}h(t)\kappa_{\varepsilon}\left(f^{(n)}(\tau(t))\right)\,\mathrm{d}t,\qquad a_f=\left(f^{(j)}(x_1)\right)_{j=0}^{n-1}.

Here τ:I2→I1\tau:I_2\to I_1 is a homeomorphism, hh is continuous with ∣h∣=1|h|=1, and ε∈{0,1}\varepsilon\in\{0,1\} is constant in the complex case. In the real case, hh is a constant sign and ε=0\varepsilon=0. The real Banach–Stone representation of the continuous-function component is determined by τ\tau, hh and ε\varepsilon [3]. For p≠2p\ne2, the finite-dimensional component has the form

(Ua)j=ωjκεj(aπ(j)),(Ua)_j=\omega_j\kappa_{\varepsilon_j}(a_{\pi(j)}),

with a permutation π\pi, unimodular ωj\omega_j, and independent conjugation choices; over R\mathbb{R} the factors are signs and all εj=0\varepsilon_j=0. The stated representation of UU follows from preservation of disjoint coordinate supports for 1≤p<∞1\le p<\infty, p=2≠∞p=2\ne\infty, and duality for p=∞p=\infty. For p=2p=2, UU is an arbitrary element of O(n)O(n) over R\mathbb{R}, or O(2n)O(2n) over C\mathbb{C} viewed as a real space. Conversely, every choice of UU, τ\tau, hh, and ε\varepsilon satisfying the stated conditions determines a surjective real-linear isometry.

It is worth emphasizing that, in the scalar-valued case, the differentiability order nn is determined by the metric structure of the unit sphere. More precisely, for 1<p1,p2<∞1<p_1,p_2<\infty, Corollary 4.9 shows that the existence of a surjective isometry between the unit spheres of Cp1n1(I1,x1;K)C^{n_1}_{p_1}(I_1,x_1;\mathbb{K}) and Cp2n2(I2,x2;K)C^{n_2}_{p_2}(I_2,x_2;\mathbb{K}) forces (n1,p1)=(n2,p2)(n_1,p_1)=(n_2,p_2), without any linearity assumption on the given isometry. For a fixed exponent 1<p<∞1<p<\infty, the preservation of the differentiability order already follows from Wang’s theorem [16 Theorem 1.8] and the finite-dimensionality of the first component. The additional feature of Corollary 4.9 is that the outer exponents are also allowed to differ initially and are necessarily equal.

Use of AI tools

The author used ChatGPT (OpenAI) to assist with developing and revising mathematical arguments and proof drafts, organizing the manuscript, and improving its exposition. This assistance included the direct-sum extension arguments and the function-space applications in Sections 3 and 4. The author independently checked the mathematical arguments and references, revised the generated material, and takes full responsibility for the final manuscript.

Department of Applied Mathematics, National Sun Yat-sen University, Kaohsiung 80424, Taiwan; and Graduate School of Science and Technology, Niigata University, Niigata 950–2181, Japan

Email address: [email protected]

References

  1. [1]T. Banakh, Every 2-dimensional Banach space has the Mazur–Ulam property, Linear Algebra Appl. 632 (2022), 268–280. doi:10.1016/j.laa.2021.09.020.DOI
  2. [2]D. Cabezas, M. Cueto-Avellaneda, Y. Enami, T. Miura, and A. M. Peralta, Tingley’s problem for complex Banach spaces which do not satisfy the Hausdorff distance condition, Banach J. Math. Anal. 17 (2023), Article 65. doi:10.1007/s43037-023-00291-0. Preprint: arXiv:2210.17131.DOI
  3. [3]R. J. Fleming and J. E. Jamison, Isometries on Banach Spaces: Function Spaces, Chapman & Hall/CRC Monographs and Surveys in Pure and Applied Mathematics, vol. 129, Chapman & Hall/CRC, Boca Raton, FL, 2003.
  4. [4]D. H. Fremlin, Measure Theory, vol. 3: Measure Algebras, Torres Fremlin, Colchester, 2002.
  5. [5]D. Hirota and T. Miura, Tingley’s problem for a Banach space of Lipschitz functions on the closed unit interval, RIMS Kôkyûroku Bessatsu B93 (2023), 157–181.
  6. [6]M.-R. Lin, A variant of Tingley’s problem on ordered Banach spaces of absolutely continuous functions, preprint (2026). arXiv:2607.10685.arxiv.org/abs/2607.10685
  7. [7]M.-R. Lin, _Metric rigidity in anchored Sobolev spaces on intervals_, preprint (2026). arXiv:2607.27646.arxiv.org/abs/2607.27646
  8. [8]P. Mankiewicz, _On extension of isometries in normed linear spaces_, Bull. Acad. Polon. Sci. Sér. Sci. Math. Astronom. Phys. 20 (1972), 367–371.
  9. [9]M. Mori and N. Ozawa, _Mankiewicz’s theorem and the Mazur–Ulam property for C*-algebras_, Studia Math. 250 (2020), no. 3, 265–281. doi:10.4064/sm180727-14-11.DOI
  10. [10]A. M. Peralta, _A survey on Tingley’s problem for operator algebras_, Acta Sci. Math. (Szeged) 84 (2018), nos. 1–2, 81–123. doi:10.14232/actasm-018-255-0.DOI
  11. [11]A. M. Peralta and R. Švarc, _A strengthened Kadison’s transitivity theorem for unital JB*-algebras with applications to the Mazur–Ulam property_, Anal. Math. 15 (2025), Article 91. doi:10.1007/s10473-025-01068-4.arxiv.org/abs/2301.00895
  12. [12]A. R. Sourour, _The isometries of L^p(Ω,X)_, J. Funct. Anal. 30 (1978), no. 2, 276–285. doi:10.1016/0022-1236(78)90075-7.DOI
  13. [13]D.-N. Tan, _On extension of isometries on the unit spheres of L^p-spaces for 0 < p ≤ 1_, Nonlinear Anal. 74 (2011), no. 18, 6981–6987. doi:10.1016/j.na.2011.07.035.
  14. [14]D. Tingley, _Isometries of the unit sphere_, Geom. Dedicata 22 (1987), no. 3, 371–378. doi:10.1007/BF00147942.DOI
  15. [15]R. Wang, _Isometries between the unit spheres of C₀(Ω)-type spaces_, Acta Math. Sci. 14 (1994), no. 1, 82–89.
  16. [16]R. Wang, _Isometries on the ℓ^p-sum of C₀(Ω,E) type spaces_, J. Math. Sci. Univ. Tokyo 3 (1996), no. 2, 471–493.
  17. [17]R. Wang and A. Orihara, _Isometries on the ℓ¹-sum of C₀(Ω,E) type spaces_, J. Math. Sci. Univ. Tokyo 2 (1995), no. 1, 131–154.

Paper details

vlnc:W_ax8kv3by
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 6

downloads · 0

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

  • Mathematics

ChatGPT

View statement
Looking for related papers…
Contents