betaFeedback

WorksMathematics

A simple proof of the characterization of ℓ∞n\ell_\infty^n by norm-one projections and a stability estimate

Abstract

We give a short and elementary proof of a classical fact that an nn-dimensional real or complex normed space with the absolute projection constant equal to one is linearly isometric to ℓ∞n\ell_\infty^n. The proof uses only finite-dimensional convexity and linear algebra, with no infinite-dimensional ingredients. The same approach yields a quantitative stability estimate of order ε13\varepsilon^{\frac{1}{3}} when the absolute projection constant is at most 1+ε1+\varepsilon. This improves Zippin's estimate of order ε18\varepsilon^{\frac{1}{8}} in each fixed dimension as ε→0\varepsilon \to 0. The real and complex cases are treated by the same arguments.

Introduction

All normed spaces considered in this paper are nonzero, and nn always denotes a positive integer. For a normed space XX over K\mathbb{K}, where K=R\mathbb{K} = \mathbb{R} or K=C\mathbb{K} = \mathbb{C}, and a nonzero linear subspace Y⊆XY \subseteq X, the relative projection constant is defined by

λ(Y,X)=inf⁡{∥P∥:P:X→Y is a bounded linear projection onto Y}.\lambda(Y,X)=\inf\{\lVert P\rVert: P:X\to Y\text{ is a bounded linear projection onto }Y\}.

with the convention that the infimum is ∞\infty if no such projection exists. If YY is finite-dimensional, this infimum is always attained by a certain projection. The absolute projection constant λ(X)\lambda(X) is the supremum of λ(T(X),Z)\lambda(T(X),Z) over all normed spaces ZZ and all linear isometric embeddings T:X→ZT:X\to Z. A normed space XX is called 1-injective if every bounded linear operator from a subspace of a normed space into XX extends to the whole space without increasing its norm. This is known to be equivalent to the requirement that every isometric copy T(X)⊆ZT(X)\subseteq Z of XX is the range of a norm-one projection, or, in other words, that λ(T(X),Z)=1\lambda(T(X),Z)=1 and the infimum is attained. Consequently, for finite-dimensional spaces XX, 1-injectivity of XX is equivalent to the equality λ(X)=1\lambda(X)=1.

The characterization of 1-injective spaces is a classical result in Banach space theory. By definition, these spaces provide an operator-valued analogue of the Hahn–Banach extension theorem for linear functionals. Whereas norm-preserving extensions need not exist for general operators, they always exist when the range space is 1-injective. The classical characterization states that a normed space XX over K\mathbb{K} is 1-injective if and only if it is linearly isometric to the space of continuous K\mathbb{K}-valued functions C(K)C(K), equipped with the supremum norm, for some extremally disconnected compact Hausdorff space KK. In the real case, Goodner [4] and Nachbin [8] independently proved the characterization under the additional assumption that the unit ball has an extreme point. Kelley [7] subsequently removed this assumption, and Hasumi [5] extended Kelley’s argument to the complex case.

In the finite-dimensional case we shall consider here, the characterization implies that every nn-dimensional normed space XX satisfying λ(X)=1\lambda(X)=1 is linearly isometric to ℓ∞n\ell_\infty^n, that is, Kn\mathbb{K}^n equipped with the maximum norm. The converse follows immediately from the Hahn–Banach theorem. Indeed, suppose that ℓ∞n\ell_\infty^n is an isometric subspace of a normed space YY. For each j=1,…,nj=1,\ldots,n, the coordinate functional fj:ℓ∞n→Kf_j:\ell_\infty^n\to\mathbb{K} (defined by f~j(x)=xj\tilde f_j(x)=x_j) has norm one and admits an extension f~j:Y→K\widetilde f_j:Y\to\mathbb{K} of norm one. The required projection P:Y→ℓ∞nP:Y\to\ell_\infty^n of norm 1 is then given by P(y)=(f~1(y),…,f~n(y))∈ℓ∞nP(y)=(\widetilde f_1(y),\ldots,\widetilde f_n(y))\in\ell_\infty^n.

As we shall see, it suffices to assume that λ(X,Y)=1\lambda(X,Y)=1 for every (n+1)(n+1)-dimensional normed superspace YY of XX to conclude that XX is linearly isometric to ℓ∞n\ell_\infty^n. Thus the characterization can be formulated entirely in finite-dimensional terms, making it natural to seek a direct finite-dimensional proof. The original proofs establish substantially more general results, but their restriction to finite dimensions still involves additional structural machinery. Nachbin [8] first derives the binary intersection property for closed balls and then uses an extreme point of the unit ball to construct a vector-lattice order. The finite-dimensional characterization follows from the classification of finite-dimensional vector lattices. Although this route does not require a representation as a space of continuous functions, it involves several intermediate results; in particular, the passage from the projection property to the binary intersection property uses a minimal-function construction based on Zorn’s lemma. Goodner [4] also constructs one-dimensional superspaces to obtain geometric restrictions on the unit ball, but subsequently introduces an order structure and applies Clarkson’s characterization of real C(K)C(K)-spaces. Kelley’s proof [7], which removes the extreme-point assumption from the general real characterization, uses the weak-star topology on the dual, the Krein–Milman theorem, and embeddings into spaces of continuous functions. These arguments concern real spaces only. Hasumi [5] established the complex analogue by adapting Kelley’s method and introducing a continuous-selection theorem.

A different approach is provided by Figiel and Johnson in Section 4 of [2], who obtain a stronger isometric characterization in terms of absolutely summing operators. Their proof uses operator-theoretic machinery such as the Pietsch factorization theorem and trace duality between absolutely summing and integral operator norms. Zippin in Lemma 1 of [12] gives a particularly short convexity proof that the range of a norm-one projection on real ℓ1n\ell_1^n is isometric to ℓ1n\ell_1^n, where nn is the dimension of the range. By duality, this yields an elementary proof of the characterization for real polyhedral spaces, which admit isometric embeddings into finite-dimensional ℓ∞\ell_\infty-spaces. This argument does not directly cover arbitrary finite-dimensional spaces, since such embeddings need not exist. Zippin also established a substantially stronger quantitative result: a finite-dimensional real space with absolute projection constant sufficiently close to one is close to ℓ∞n\ell_\infty^n in the Banach–Mazur distance, with a bound independent of the dimension [12, 13] (see also the subsequent erratum [14]). This stability theorem yields the exact characterization by applying it with arbitrarily small positive ε\varepsilon to a space with projection constant one. Its proof, however, requires a substantial quantitative analysis of projections on finite-dimensional ℓ1\ell_1-spaces, developed across the two papers.

The question of finding a simple direct proof of the finite-dimensional characterization was raised by M. Argerami on MathOverflow in September 2024 [15]. In his answer, W. B. Johnson observed that there were “no really simple proofs” and pointed to several alternative approaches.

Our first goal is to present such a proof. We believe that the argument merits being called simple, in the sense that it is short, direct, and entirely elementary, using only basic finite-dimensional convexity and linear algebra, including the hyperplane separation theorem. Moreover, the same argument covers the real and complex cases simultaneously. As already mentioned, it suffices to consider (n+1)(n+1)-dimensional superspaces of XX only. In the real case, this criterion is already implicit in the arguments of Goodner [4] and Nachbin [8]. We are not aware of an explicit proof of the corresponding criterion in the complex case.

Theorem 1.1. Let XX be an nn-dimensional normed space over K\mathbb{K}. If the equality λ(X,Y)=1\lambda(X,Y)=1 holds for every (n+1)(n+1)-dimensional normed superspace X⊆YX\subseteq Y over K\mathbb{K}, then XX is linearly isometric to ℓ∞n\ell_\infty^n.

The proof is based on a general construction of isometric embeddings into (n+1)(n+1)-dimensional superspaces, given in Lemma 2.1. It can be viewed as a finite-dimensional adaptation of the canonical embedding into C(BX∗)C(B_{X^*}): instead of using the whole function space, we retain only the evaluation functions associated with vectors of XX and one additional continuous equivariant function. The argument also draws on some other ideas from earlier proofs, including the use of suitably normed one-dimensional extensions and the geometry of the dual unit ball. In the finite-dimensional setting, the relevant description of the dual unit ball is obtained by an elementary convex hull computation. In this sense, it is a synthesis of existing ideas tailored to the finite-dimensional setting. Its main new ingredient is the construction of an embedding in Lemma 2.1, which is used to prescribe scalar values at selected extreme functionals. Extremality then forces a norm-one projection to respect these prescribed values, which are chosen to be incompatible with linearity. The construction thus retains some of the advantages of the function-space embedding without leaving the finite-dimensional setting. For this reason, we believe that the underlying mechanism may have further applications.

The proof of Theorem 1.1 is presented in § 2. Our approach can be adapted in a relatively straightforward way to obtain a quantitative stability estimate. In § 3, we therefore revisit the stability of ℓ∞n\ell_\infty^n as the unique minimizer (up to linear isometry) of the absolute projection constant among nn-dimensional normed spaces. We recall that the Banach–Mazur distance between two nn-dimensional normed spaces XX and YY over K\mathbb{K} is defined by

dBM(X,Y)=inf⁡∥T∥⋅∥T−1∥,d_{\mathrm{BM}}(X,Y)=\inf\lVert T\rVert\cdot\lVert T^{-1}\rVert,

where the infimum is taken over all invertible linear operators T:X→YT:X\to Y. Since we consider only finite-dimensional spaces, this infimum is always attained. Zippin [12, 13, 14] proved that a finite-dimensional real normed space whose absolute projection constant is sufficiently close to one is close to ℓ∞n\ell_\infty^n, with a bound independent of the dimension. An explicit form of the estimate, attributed to Bernau [1], is given in [3] (see the proof of Fact 3.3) and [10] (Theorem 33.3): if XX is an nn-dimensional real normed space with λ(X)≤1+ε\lambda(X)\leq1+\varepsilon, where 0≤ε<17−80\leq\varepsilon<17^{-8}, then

dBM(X,ℓ∞n)≤1+6ε18(1−6ε18)(1−17ε18).d_{\mathrm{BM}}(X,\ell_\infty^n)\leq\frac{1+6\varepsilon^{\frac{1}{8}}}{(1-6\varepsilon^{\frac{1}{8}})(1-17\varepsilon^{\frac{1}{8}})}.

In particular, dBM(X,ℓ∞n)≤1+64ε18d_{\mathrm{BM}}(X,\ell_\infty^n)\leq1+64\varepsilon^{\frac{1}{8}} for ε≤34−8\varepsilon\leq34^{-8}. Both the admissible range of ε\varepsilon and the bound are thus independent of nn. Our result goes as follows.

Theorem 1.2. Let XX be an nn-dimensional normed space over K\mathbb{K} and let ε∈(0,(2560n)−1]\varepsilon\in(0,(2560n)^{-1}]. If λ(X,Y)≤1+ε\lambda(X,Y)\leq1+\varepsilon for every (n+1)(n+1)-dimensional normed superspace X⊆YX\subseteq Y over K\mathbb{K}, then

dBM(X,ℓ∞n)<1+14(nε)13.d_{\mathrm{BM}}(X,\ell_\infty^n)<1+14(n\varepsilon)^{\frac{1}{3}}.

Thus, in each fixed dimension, our estimate improves the order ε18\varepsilon^{\frac{1}{8}} to ε13\varepsilon^{\frac{1}{3}} as ε→0\varepsilon\to0. However, both the admissible range of ε\varepsilon and the coefficient in the bound do depend on nn. Moreover, our hypothesis concerns only (n+1)(n+1)-dimensional superspaces of XX, and the same argument covers real and complex scalars. The proof again relies on the embedding construction in Lemma 2.1. The key additional ingredient is a quantitative exposed-point lemma (Lemma 3.1), related to the classical theorem of Straszewicz that exposed points are dense among extreme points of a compact convex set in Kn\mathbb{K}^n. Our lemma constructs an exposed functional with a small norming slice that is also almost norming at a prescribed unit vector. To establish the lemma, we choose an auxiliary Euclidean structure using the classical bound dBM(X,ℓ2n)≤nd_{\mathrm{BM}}(X,\ell_2^n)\leq\sqrt{n}, where ℓ2n\ell_2^n denotes Kn\mathbb{K}^n equipped with the Euclidean norm. Apart from this standard normalization, the proof uses only elementary finite-dimensional arguments.

Throughout the paper, nn-dimensional normed spaces are always identified with Kn\mathbb{K}^{n} equipped with a norm. The norm of a space XX is denoted by ∥⋅∥X\lVert\cdot\rVert_{X}. The usual operator norm is denoted simply by ∥⋅∥\lVert\cdot\rVert when the context is clear. In particular, we write ∥f∥\lVert f\rVert for the dual norm of a functional f∈X∗f\in X^{*}. We identify X∗X^{*} with Kn\mathbb{K}^{n} using the dual coordinates, i.e., the evaluation is given by f(x)=∑j=1nfjxjf(x)=\sum_{j=1}^{n}f_{j}x_{j} for x=(x1,…,xn)∈Xx=(x_{1},\ldots,x_{n})\in X and f=(f1,…,fn)∈X∗f=(f_{1},\ldots,f_{n})\in X^{*}. The closed unit ball and the unit sphere of XX are denoted by BXB_{X} and SXS_{X}, respectively. For a nonempty set C⊆KnC\subseteq\mathbb{K}^{n}, its convex hull and absolutely convex hull are defined by

conv⁡C={∑j=1Nαjxj:N≥1, xj∈C, αj∈[0,∞), ∑j=1Nαj=1}\operatorname{conv} C=\left\{\sum_{j=1}^{N}\alpha_{j}x_{j}:N\geq1,\ x_{j}\in C,\ \alpha_{j}\in[0,\infty),\ \sum_{j=1}^{N}\alpha_{j}=1\right\}

and

absconv⁡C={∑j=1Nαjxj:N≥1, xj∈C, αj∈K, ∑j=1N∣αj∣≤1},\operatorname{absconv} C=\left\{\sum_{j=1}^{N}\alpha_{j}x_{j}:N\geq1,\ x_{j}\in C,\ \alpha_{j}\in\mathbb{K},\ \sum_{j=1}^{N}|\alpha_{j}|\leq1\right\},

respectively. If CC is a convex subset of Kn\mathbb{K}^{n}, then a point x∈Cx\in C is called an extreme point of CC if the equality x=ty+(1−t)zx=ty+(1-t)z for y,z∈Cy,z\in C and t∈(0,1)t\in(0,1) implies x=y=zx=y=z. The set of extreme points of CC is denoted by ext⁡C\operatorname{ext} C. By T⊆K\mathbb{T}\subseteq\mathbb{K} we denote the set of all scalars of modulus 11, i.e., the set {−1,1}\{-1,1\} for K=R\mathbb{K}=\mathbb{R} and the complex unit circle for K=C\mathbb{K}=\mathbb{C}.

Proof of Theorem 1.1

As mentioned in the introduction, the main idea of the proof of Theorem Theorem 1.1 is to construct suitably normed one-dimensional extensions of XX. Clearly, for every x∈Xx\in X we have ∥x∥X=max⁡g∈BX∗∣g(x)∣\lVert x\rVert_{X}=\max_{g\in B_{X^{*}}}|g(x)|. Writing functionals in the standard coordinates of Kn\mathbb{K}^{n}, we add to each g∈BX∗g\in B_{X^{*}} an additional coordinate ψ(g)\psi(g) that depends continuously on gg. Taking the maximum of the moduli of the resulting functionals (x,t)↦g(x)+tψ(g)(x,t)\mapsto g(x)+t\psi(g) defines a norm on X×KX\times\mathbb{K} under rather mild assumptions on ψ\psi. This norm preserves the original norm on X×{0}X\times\{0\}. Crucially, ψ\psi need not be linear, leaving considerable freedom in its choice. In the proof a very specific choice of ψ\psi is used. Geometrically, the construction lifts the dual unit ball into one additional coordinate and takes the convex hull of the resulting graph. Under similar assumptions, the same construction yields (n+m)(n+m)-dimensional superspaces by taking ψ\psi with values in Km\mathbb{K}^{m}. A map ψ:BX∗→K\psi:B_{X^{*}}\to\mathbb{K} is called equivariant if ψ(αg)=αψ(g)\psi(\alpha g)=\alpha\psi(g) holds for every α∈T\alpha\in\mathbb{T} and g∈BX∗g\in B_{X^{*}}.

Lemma 2.1. Let XX be an nn-dimensional normed space over K\mathbb{K}, and let ψ:BX∗→K\psi:B_{X^{*}}\to\mathbb{K} be a continuous equivariant map that is not identically zero. Suppose that there exists r∈(0,1)r\in(0,1) such that ψ(g)=0\psi(g)=0 whenever ∥g∥<r\lVert g\rVert<r. Then

∥(x,t)∥Y=max⁡g∈BX∗∣g(x)+tψ(g)∣\lVert(x,t)\rVert_{Y}=\max_{g\in B_{X^{*}}}|g(x)+t\psi(g)|

defines a norm on Y=X×KY=X\times\mathbb{K}. Moreover, the embedding X∋x↦(x,0)∈YX\ni x\mapsto(x,0)\in Y is isometric, and, under the identification of Y∗Y^{*} with X∗×KX^{*}\times\mathbb{K} given by (g,s)(x,t)=g(x)+st(g,s)(x,t)=g(x)+st, we have

BY∗=conv⁡{(g,ψ(g)):g∈BX∗}.B_{Y^{*}}=\operatorname{conv}\{(g,\psi(g)):g\in B_{X^{*}}\}.

Proof. We first justify that ∥⋅∥Y\lVert\cdot\rVert_{Y} is indeed a norm on YY. It is well defined, since the set BX∗B_{X^{*}} is compact and ψ\psi is continuous, so the maximum exists. It is homogeneous and the triangle inequality holds, so it is a seminorm on YY. To prove that it is positive for nonzero vectors, suppose first that x∈Xx \in X is nonzero. We can choose g∈BX∗g \in B_{X^*} such that g(x)≠0g(x) \ne0 and ∥g∥<r\|g\| < r, so that

∥(x,t)∥Y≥∣g(x)+tψ(g)∣=∣g(x)∣>0.\|(x,t)\|_Y \ge|g(x)+t\psi(g)|=|g(x)|>0.

On the other hand, if x=0x=0, since ψ\psi is not identically zero, we can choose g∈BX∗g \in B_{X^*} such that ψ(g)≠0\psi(g) \ne0. Then ∥(0,t)∥Y≥∣tψ(g)∣>0\|(0,t)\|_Y \ge|t\psi(g)|>0 for t≠0t \ne0, and hence ∥⋅∥Y\|\cdot\|_Y is a norm. Furthermore, it is clear that ∥(x,0)∥Y=∥x∥X\|(x,0)\|_Y=\|x\|_X holds for any x∈Xx \in X, and hence, XX embeds isometrically in YY.

It remains to prove the representation for BY∗B_{Y^*}. Let C:={(g,ψ(g)):g∈BX∗}C:=\{(g,\psi(g)):g \in B_{X^*}\}. Clearly, conv⁡C⊆BY∗\operatorname{conv} C \subseteq B_{Y^*} and it is enough to establish the reverse inclusion. Indeed, suppose, towards a contradiction, that φ∈BY∗∖conv⁡C\varphi\in B_{Y^*}\setminus\operatorname{conv} C. Since ψ\psi is continuous, CC is compact and by the hyperplane separation theorem, there exists y∈Yy \in Y such that

Re⁡(φ(y))>max⁡γ∈CRe⁡(γ(y))=max⁡γ∈C∣γ(y)∣=∥y∥Y≥∣φ(y)∣.\operatorname{Re}(\varphi(y))>\max_{\gamma\in C}\operatorname{Re}(\gamma(y))=\max_{\gamma\in C}|\gamma(y)|=\|y\|_Y\ge|\varphi(y)|.

where the first equality follows from the fact that αC=C\alpha C=C for any α∈T\alpha\in\mathbb{T} by the equivariance of ψ\psi. This contradiction shows that conv⁡C=BY∗\operatorname{conv} C=B_{Y^*}.

Proof of Theorem 1.1. Since the dual unit ball BX∗⊆KnB_{X^*}\subseteq\mathbb{K}^n has a nonempty interior and BX∗=conv⁡ext⁡BX∗B_{X^*}=\operatorname{conv}\operatorname{ext}B_{X^*} (by the finite-dimensional Krein–Milman theorem), the set ext⁡BX∗\operatorname{ext}B_{X^*} spans Kn\mathbb{K}^n. In particular, we can choose nn linearly independent functionals f1,…,fn∈ext⁡BX∗f_1,\ldots,f_n\in\operatorname{ext}B_{X^*}. If the equality absconv⁡{f1,…,fn}=BX∗\operatorname{absconv}\{f_1,\ldots,f_n\}=B_{X^*} holds, then

∥x∥X=max⁡{∣fj(x)∣:j=1,…,n},\|x\|_X=\max\{|f_j(x)|:j=1,\ldots,n\},

and consequently, the space XX is isometric to ℓ∞n\ell_\infty^n with the isometry T:X→KnT:X\to\mathbb{K}^n given by T(x)=(f1(x),…,fn(x))T(x)=(f_1(x),\ldots,f_n(x)). Otherwise, since BX∗=conv⁡ext⁡BX∗B_{X^*}=\operatorname{conv}\operatorname{ext}B_{X^*}, there exists fn+1∈ext⁡BX∗f_{n+1}\in\operatorname{ext}B_{X^*} such that fn+1∉absconv⁡{f1,…,fn}f_{n+1}\notin\operatorname{absconv}\{f_1,\ldots,f_n\}. We shall derive a contradiction with this assumption. In particular, for all j=1,…,nj=1,\ldots,n we have Tfj∩Tfn+1=∅\mathbb{T}f_j\cap\mathbb{T}f_{n+1}=\varnothing. Since the sets Tfj\mathbb{T}f_j are compact, we can pick 0<d<120<d<\frac{1}{2} such that dist⁡(fj,Tfn+1)>d\operatorname{dist}(f_j,\mathbb{T}f_{n+1})>d for all j=1,…,nj=1,\ldots,n (with the distance being measured using the dual norm of XX).

For s≥0s\ge0 let u(s)=max⁡{0,1−sd}u(s)=\max\left\{0,1-\frac{s}{d}\right\}. Thus, uu is a continuous function satisfying u(0)=1u(0)=1 and u(s)=0u(s)=0 for s≥ds\ge d. Since ∥fn+1∥=1\|f_{n+1}\|=1, we can choose xn+1∈SXx_{n+1}\in S_X satisfying fn+1(xn+1)=1f_{n+1}(x_{n+1})=1. We define the mapping ψ:BX∗→K\psi:B_{X^*}\to\mathbb{K} by

ψ(g)=g(xn+1)u(dist⁡(g,Tfn+1)).\psi(g)=g(x_{n+1})u\left(\operatorname{dist}(g,\mathbb{T}f_{n+1})\right).

The mapping ψ\psi has the following properties:

  1. ψ\psi is continuous.

  1. ψ\psi is equivariant, as dist⁡(αg,Tfn+1)=dist⁡(g,Tfn+1)\operatorname{dist}(\alpha g,\mathbb{T}f_{n+1})=\operatorname{dist}(g,\mathbb{T}f_{n+1}) holds for every α∈T\alpha\in\mathbb{T}.

  1. ψ(fj)=0\psi(f_j)=0 for every j=1,…,nj=1,\ldots,n, since dist⁡(fj,Tfn+1)>d\operatorname{dist}(f_j,\mathbb{T}f_{n+1})>d. Moreover, ψ(fn+1)=fn+1(xn+1)u(0)=1\psi(f_{n+1})=f_{n+1}(x_{n+1})u(0)=1.

  1. ψ(g)=0\psi(g)=0 for all g∈BX∗g\in B_{X^*} with ∥g∥<12\|g\|<\frac{1}{2}. Indeed, for such gg we have dist⁡(g,Tfn+1)≥∥fn+1∥−∥g∥>12>d\operatorname{dist}(g,\mathbb{T}f_{n+1})\ge\|f_{n+1}\|-\|g\|>\frac{1}{2}>d.

Let us now consider the space Y=X×KY=X\times\mathbb{K} with the norm ∥⋅∥Y\|\cdot\|_Y defined in Lemma 2.1. Since X×{0}X\times\{0\} is a subspace of YY isometric to XX, by the assumption there exists a linear map P:Y→XP:Y\to X such that P(x,0)=xP(x,0)=x for all x∈Xx\in X and ∥P∥=1\|P\|=1. Fix j∈{1,…,n+1}j\in\{1,\ldots,n+1\} and let φj=fj∘P∈Y∗\varphi_j=f_j\circ P\in Y^*. Then ∥φj∥≤∥fj∥⋅∥P∥=1\|\varphi_j\|\le\|f_j\|\cdot\|P\|=1, so that φj∈BY∗\varphi_j\in B_{Y^*}. Therefore, by Lemma 2.1, there exist functionals g1,…,gN∈BX∗g_1,\ldots,g_N\in B_{X^*} and positive real numbers λ1,…,λN\lambda_1,\ldots,\lambda_N with sum 11 such that

φj=∑k=1Nλk(gk,ψ(gk)).\varphi_j=\sum_{k=1}^{N}\lambda_k(g_k,\psi(g_k)).

Evaluating both sides of this equality on X×{0}X \times\{0\} yields fj=∑k=1Nλkgkf_j = \sum_{k=1}^{N} \lambda_k g_k. Since fjf_j is an extreme point of BX∗B_{X^*}, this gives gk=fjg_k = f_j for all k=1,…,Nk = 1, \ldots, N. Thus, from the fact that ψ(fj)=0\psi(f_j) = 0 for j=1,…,nj = 1, \ldots, n, it now follows that

φj=(fj,ψ(fj))=(fj,0),\varphi_j = (f_j, \psi(f_j)) = (f_j, 0),

for such jj. Similarly, φn+1=(fn+1,1)\varphi_{n+1} = (f_{n+1}, 1). Let us now take y=P(0,1)∈Xy = P(0, 1) \in X. For j=1,…,nj = 1, \ldots, n, we have

fj(y)=φj(0,1)=0.f_j(y) = \varphi_j(0, 1) = 0.

Since f1,…,fnf_1, \ldots, f_n were chosen to form a basis of X∗X^*, this yields y=0y = 0. On the other hand,

fn+1(y)=φn+1(0,1)=1,f_{n+1}(y) = \varphi_{n+1}(0, 1) = 1,

a contradiction. This completes the proof.

Stability

In this section we prove Theorem 1.2 concerning the stability of ℓ∞\ell_\infty. In the proof of Theorem 1.1, we used extremality of functionals to force a norm-one projection to respect prescribed scalar values at arbitrarily selected extreme functionals. Turning this argument into a quantitative estimate requires a more precise choice of functionals. Specifically, we seek a sufficiently rich family of functionals admitting small norming slices, from which we can select a well-separated subfamily. We obtain such functionals by a quantitative adaptation of the classical proof of Straszewicz’s theorem (see Theorem 18.6 in [9]), which states that exposed points are dense among the extreme points of a compact convex set in a finite-dimensional space. The main idea is to choose a functional in the dual unit ball whose Euclidean distance from an appropriately chosen point is maximal. This requires an auxiliary Euclidean structure well-comparable to the given norm. For this purpose, we use the classical consequence of the John Ellipsoid Theorem [6], which gives dBM(X,ℓ2n)≤nd_{\mathrm{BM}}(X, \ell_2^n) \le\sqrt{n} for every nn-dimensional normed space XX over K\mathbb{K}. Although the John Ellipsoid Theorem and the resulting upper bound on the Banach–Mazur distance are best known in the real setting, they also hold over the complex field (see Theorem 15.3 in [10]).

Lemma 3.1. Let XX be an nn-dimensional normed space over K\mathbb{K}. Let x∈Xx \in X be a unit vector and let L>0L > 0 be a real number. Then there exist a functional f∈SX∗f \in S_{X^*} and a vector v∈SXv \in S_X such that f(v)=1f(v) = 1 and the following conditions are satisfied:

(i)f(x)∈R and f(x)≥1−n−12L.(ii)If g∈BX∗ and Re⁡g(v)≥1−δ for some δ>0, then ∥f−g∥≤2δ(L+n).\begin{aligned} \text{(i)}\quad&f(x) \in\mathbb{R}\ \text{and}\ f(x) \ge1 - \frac{n-1}{2L}.\\ \text{(ii)}\quad&\text{If }g \in B_{X^*}\text{ and }\operatorname{Re} g(v) \ge1-\delta\text{ for some }\delta> 0,\text{ then }\|f-g\| \le\sqrt{2\delta(L+n)}. \end{aligned}

Proof. We recall that we identify both XX and X∗X^* with Kn\mathbb{K}^n. For u∈Xu \in X, we denote its norm by ∥u∥X\|u\|_X, while for g∈X∗g \in X^* we denote its dual norm simply by ∥g∥\|g\|. It is a classical consequence of the John Ellipsoid Theorem that E⊆BX∗⊆nE\mathcal{E} \subseteq B_{X^*} \subseteq\sqrt{n}\mathcal{E}, where E\mathcal{E} is the ellipsoid of maximal volume contained in BX∗B_{X^*}. We choose coordinates on X∗X^*, and the corresponding dual coordinates on XX, such that E=Bn\mathcal{E} = B^n is the standard Euclidean unit ball. Then ∥g∥≤∥g∥2≤n∥g∥\|g\| \le\|g\|_2 \le\sqrt{n}\|g\| for g∈X∗g \in X^* and the standard Euclidean norm ∥⋅∥2\|\cdot\|_2 on Kn\mathbb{K}^n. Let also [⋅,⋅][\cdot,\cdot] be the standard inner product on Kn\mathbb{K}^n, linear in the first variable. Thus [g,h]=∑j=1ngjhj‾[g,h] = \sum_{j=1}^{n} g_j\overline{h_j} and [g,g]=∥g∥22[g,g] = \|g\|_2^2. With these coordinate conventions, for every g∈X∗g \in X^* and u∈Xu \in X we have g(u)=[g,u‾]g(u) = [g,\overline{u}], where u‾=(u1‾,…,un‾)\overline{u} = (\overline{u_1},\ldots,\overline{u_n}) is the coordinatewise conjugate of uu.

Let f∈BX∗f \in B_{X^*} be a functional with maximal Euclidean distance from the vector −Lx‾∈Kn-L\overline{x} \in\mathbb{K}^n, that is

∥f+Lx‾∥2=max⁡g∈BX∗∥g+Lx‾∥2.\|f+L\overline{x}\|_2 = \max_{g\in B_{X^*}} \|g+L\overline{x}\|_2.

Thus, for every g∈BX∗g \in B_{X^*} we have ∥f+Lx‾∥22≥∥g+Lx‾∥22\|f+L\overline{x}\|_2^2 \ge\|g+L\overline{x}\|_2^2. This inequality can be rewritten as

∥f∥22+2LRe⁡f(x)≥∥g∥22+2LRe⁡g(x).\|f\|_2^2+2L\operatorname{Re} f(x) \ge\|g\|_2^2+2L\operatorname{Re} g(x).

In particular, for g=αfg=\alpha f, where α∈T\alpha\in\mathbb{T} is such that αf(x)=∣f(x)∣\alpha f(x)=|f(x)| we obtain f(x)∈Rf(x)\in\mathbb{R}. Moreover, for g∈SX∗g\in S_{X^*} satisfying g(x)=1g(x)=1 we get

n+2Lf(x)≥∥f∥22+2Lf(x)≥1+2L,n+2Lf(x)\ge\|f\|_2^2+2Lf(x)\ge1+2L,

which yields (i).

For (ii), let us note that for any g∈BX∗g\in B_{X^*}

∥f+Lx‾∥22≥∥g+Lx‾∥22=∥(g−f)+f+Lx‾∥22=∥g−f∥22+2Re⁡[g−f,f+Lx‾]+∥f+Lx‾∥22,\|f+L\overline{x}\|_2^2\ge\|g+L\overline{x}\|_2^2=\|(g-f)+f+L\overline{x}\|_2^2=\|g-f\|_2^2+2\operatorname{Re}[g-f,f+L\overline{x}]+\|f+L\overline{x}\|_2^2,

so that

Re⁡[f−g,f+Lx‾]≥12∥f−g∥22.\operatorname{Re}[f-g,f+L\overline{x}]\ge\frac{1}{2}\|f-g\|_2^2.

If w:=f‾+Lx∈Knw:=\overline{f}+Lx\in K^n, then the above inequality becomes

Re⁡f(w)≥Re⁡g(w)+12∥f−g∥22,\operatorname{Re}f(w)\ge\operatorname{Re}g(w)+\frac{1}{2}\|f-g\|_2^2,

for any g∈BX∗g\in B_{X^*}. In particular, it is clear that w≠0w\ne0, since w=0w=0 in the preceding inequality would imply f=gf=g for every g∈BX∗g\in B_{X^*}, which is impossible. Therefore, for g0∈SX∗g_0\in S_{X^*} such that g0(w)=∥w∥Xg_0(w)=\|w\|_X we obtain

∥w∥X≥∣f(w)∣≥Re⁡(f(w))≥g0(w)=∥w∥X,\|w\|_X\ge|f(w)|\ge\operatorname{Re}(f(w))\ge g_0(w)=\|w\|_X,

and hence f(w)=∥w∥Xf(w)=\|w\|_X. Thus, we have ∥f∥=1\|f\|=1. Let v:=w∥w∥X∈Xv:=\frac{w}{\|w\|_X}\in X. Then f(v)=1f(v)=1 and therefore

1≥Re⁡g(v)+12∥w∥X∥f−g∥22,1\ge\operatorname{Re}g(v)+\frac{1}{2\|w\|_X}\|f-g\|_2^2,

for any g∈BX∗g\in B_{X^*}. In particular, if Re⁡g(v)≥1−δ\operatorname{Re}g(v)\ge1-\delta, then

∥f−g∥22≤2δ∥w∥X.\|f-g\|_2^2\le2\delta\|w\|_X.

Since ∥f−g∥≤∥f−g∥2\|f-g\|\le\|f-g\|_2, to conclude the proof it remains to bound the norm ∥w∥X\|w\|_X from above. We note that

∥w∥X=f(w)=[f,f+Lx‾]=∥f∥2+Lf(x)≤n∥f∥2+L∥f∥=n+L.\|w\|_X=f(w)=[f,f+L\overline{x}]=\|f\|^2+Lf(x)\le n\|f\|^2+L\|f\|=n+L.

This completes the proof.

We are ready to prove Theorem 1.2.

Proof of Theorem 1.2. We set

ρ=(40nε)13,δ=ρ36n,L=n2ρ.\rho=(40n\varepsilon)^{\frac{1}{3}},\qquad\delta=\frac{\rho^3}{6n},\qquad L=\frac{n}{2\rho}.

From the assumed range of ε\varepsilon we have

0<ρ≤14,0<δ≤1384,L>n2.0<\rho\le\frac{1}{4},\qquad0<\delta\le\frac{1}{384},\qquad L>\frac{n}{2}.

For each x∈SXx\in S_X, we choose a functional fx∈SX∗f_x\in S_{X^*} and a corresponding point vx∈SXv_x\in S_X as in Lemma 3.1. In particular, fx(vx)=1f_x(v_x)=1 and

fx(x)=Re⁡fx(x)≥1−n−12L≥1−ρ.f_x(x)=\operatorname{Re}f_x(x)\ge1-\frac{n-1}{2L}\ge1-\rho.

Moreover,

2δ(L+n)=ρ2(1+2ρ)6≤ρ24,2\delta(L+n)=\frac{\rho^2(1+2\rho)}{6}\le\frac{\rho^2}{4},

so the same lemma gives

Re⁡g(vx)≥1−δ⇒∥fx−g∥≤ρ2,(1)\operatorname{Re} g(v_x) \ge1-\delta\Rightarrow\|f_x-g\| \le\frac{\rho}{2}, \tag*{(1)}

for all x∈SXx \in S_X and g∈BX∗g \in B_{X^*}. Let F={fx:x∈SX}⊆SX∗F=\{f_x:x\in S_X\}\subseteq S_{X^*}. Let us choose a subset {f1,…,fm}⊆F\{f_1,\ldots,f_m\}\subseteq F of the largest possible cardinality that satisfies ∥fi−αfj∥>ρ\|f_i-\alpha f_j\|>\rho for all 1≤i<j≤m1\le i<j\le m and α∈T\alpha\in\mathbb{T}. For each ii, let xi∈SXx_i\in S_X be such that fi=fxif_i=f_{x_i}, and set vi=vxiv_i=v_{x_i}.

We observe that for a given g∈BX∗g\in B_{X^*}, there exists at most one index i∈{1,…,m}i\in\{1,\ldots,m\} for which ∣g(vi)∣≥1−δ|g(v_i)|\ge1-\delta. Indeed, let us assume that there are two such indices i<ji<j and choose αi,αj∈T\alpha_i,\alpha_j\in\mathbb{T} such that Re⁡(αig(vi))=∣g(vi)∣\operatorname{Re}(\alpha_i g(v_i))=|g(v_i)| and Re⁡(αjg(vj))=∣g(vj)∣\operatorname{Re}(\alpha_j g(v_j))=|g(v_j)|. Then by (1) we get ∥fi−αig∥≤ρ2\|f_i-\alpha_i g\|\le\frac{\rho}{2} and ∥fj−αjg∥≤ρ2\|f_j-\alpha_jg\|\le\frac{\rho}{2}, so that

∥fi−αiαj−1fj∥=∥fi−αig+αi(g−αj−1fj)∥≤∥fi−αig∥+∥g−αj−1fj∥≤ρ,\|f_i-\alpha_i\alpha_j^{-1}f_j\|=\|f_i-\alpha_i g+\alpha_i(g-\alpha_j^{-1}f_j)\|\le\|f_i-\alpha_i g\|+\|g-\alpha_j^{-1}f_j\|\le\rho,

which contradicts the assumed strict separation by ρ\rho. Because {f1,…,fm}\{f_1,\ldots,f_m\} is a maximal ρ\rho-separated subset, for every g∈Fg\in F there exists j∈{1,…,m}j\in\{1,\ldots,m\} and α∈T\alpha\in\mathbb{T} such that ∥g−αfj∥≤ρ\|g-\alpha f_j\|\le\rho. For a given vector x∈SXx\in S_X, applying this for g:=fxg:=f_x yields

∣fj(x)∣≥Re⁡(αfj(x))≥Re⁡(fx(x))−∥fx−αfj∥≥1−ρ−ρ=1−2ρ.|f_j(x)|\ge\operatorname{Re}(\alpha f_j(x))\ge\operatorname{Re}(f_x(x))-\|f_x-\alpha f_j\|\ge1-\rho-\rho=1-2\rho.

It follows that

(1−2ρ)∥x∥X≤max⁡j=1,…,m∣fj(x)∣≤∥x∥X,(1-2\rho)\|x\|_X\le\max_{j=1,\ldots,m}|f_j(x)|\le\|x\|_X,

for every x∈Xx\in X. Hence, the linear map T:X→ℓ∞mT:X\to\ell_\infty^m given by

T(x)=(f1(x),…,fm(x))∈ℓ∞mT(x)=(f_1(x),\ldots,f_m(x))\in\ell_\infty^m

is injective, and, consequently, m≥nm\ge n. If m=nm=n, then TT is an isomorphism onto ℓ∞n\ell_\infty^n yielding the estimate

dBM(X,ℓ∞n)≤∥T∥⋅∥T−1∥≤(1−2ρ)−1≤1+4ρ=1+4(40nε)13<1+14(nε)13.d_{\mathrm{BM}}(X,\ell_\infty^n)\le\|T\|\cdot\|T^{-1}\|\le(1-2\rho)^{-1}\le1+4\rho=1+4(40n\varepsilon)^{\frac{1}{3}}<1+14(n\varepsilon)^{\frac{1}{3}}.

Therefore, the result will follow once we establish that m≤nm\le n. Suppose, towards a contradiction, that m≥n+1m\ge n+1. Since dim⁡X∗=n\dim X^*=n, the functionals f1,…,fn+1f_1,\ldots,f_{n+1} are linearly dependent. Let c1,…,cn+1∈Kc_1,\ldots,c_{n+1}\in\mathbb{K}, not all zero, be such that ∑j=1n+1cjfj=0\sum_{j=1}^{n+1}c_jf_j=0. For each j=1,…,n+1j=1,\ldots,n+1 choose γj∈T\gamma_j\in\mathbb{T} such that γjcj=∣cj∣\gamma_jc_j=|c_j| (if cj=0c_j=0, we choose γj\gamma_j arbitrarily). We define a function σ:K→K\sigma:\mathbb{K}\to\mathbb{K} by σ(0)=0\sigma(0)=0 and

σ(s)=s∣s∣⋅min⁡{1,max⁡{0,2∣s∣+2δ−2δ}}for s≠0.\sigma(s)=\frac{s}{|s|}\cdot\min\left\{1,\max\left\{0,\frac{2|s|+2\delta-2}{\delta}\right\}\right\}\qquad\text{for }s\ne0.

It is straightforward to verify that σ\sigma is a continuous function satisfying σ(s)=0\sigma(s)=0 for ∣s∣≤1−δ|s|\le1-\delta and σ(s)=s∣s∣\sigma(s)=\frac{s}{|s|} for ∣s∣≥1−δ2|s|\ge1-\frac{\delta}{2}. Moreover, ∣σ(s)∣≤1|\sigma(s)|\le1 and σ(αs)=ασ(s)\sigma(\alpha s)=\alpha\sigma(s) for all s∈Ks\in\mathbb{K}, α∈T\alpha\in\mathbb{T}. Next, we define a mapping ψ:BX∗→K\psi:B_{X^*}\to\mathbb{K} by

ψ(g)=∑j=1n+1γjσ(g(vj)).\psi(g)=\sum_{j=1}^{n+1}\gamma_j\sigma(g(v_j)).

The mapping ψ\psi is continuous, equivariant, and satisfies ψ(g)=0\psi(g)=0 for ∥g∥≤1−δ\|g\|\le1-\delta (as then ∣g(vj)∣≤∥g∥⋅∥vj∥X≤1−δ|g(v_j)|\le\|g\|\cdot\|v_j\|_X\le1-\delta for all jj). Furthermore, as observed in the previous step, for every g∈BX∗g\in B_{X^*}, at most one summand in the definition of ψ(g)\psi(g) is nonzero. Since ∣γj∣=1|\gamma_j|=1 and ∣σ(s)∣≤1|\sigma(s)|\le1, it follows that

∣ψ(g)∣≤1.(2)|\psi(g)|\le1. \tag*{(2)}

In particular, since fj(vj)=1f_j(v_j)=1 we have ψ(fj)=γj\psi(f_j)=\gamma_j for every j=1,…,n+1j=1,\ldots,n+1. This shows that ψ\psi is not identically zero and by Lemma 2.1 the formula

∥(x,t)∥Y=max⁡g∈BX∗∣g(x)+tψ(g)∣\|(x,t)\|_Y=\max_{g\in B_{X^*}}|g(x)+t\psi(g)|

defines a norm on Y=X×KY=X\times\mathbb{K}. Moreover, the embedding X∋x↦(x,0)∈YX\ni x\mapsto(x,0)\in Y is isometric, and

BY∗=conv⁡{(g,ψ(g)):g∈BX∗}.B_{Y^*}=\operatorname{conv}\{(g,\psi(g)):g\in B_{X^*}\}.

By the assumption, there exists a linear map P:Y→XP:Y\to X such that P(x,0)=xP(x,0)=x for every x∈Xx\in X and ∥P∥≤1+ε\lVert P\rVert\leq1+\varepsilon. Set y=P(0,1)∈Xy=P(0,1)\in X. Under the identification (g,s)(x,t)=g(x)+st(g,s)(x,t)=g(x)+st of Y∗Y^* with X∗×KX^*\times\mathbb{K}, we have f∘P=(f,f(y))f\circ P=(f,f(y)) for f∈X∗f\in X^*. Let us fix j∈{1,…,n+1}j\in\{1,\ldots,n+1\}. Since ∥fj∘P∥≤1+ε\lVert f_j\circ P\rVert\leq1+\varepsilon, there exist functionals g1,…,gN∈BX∗g_1,\ldots,g_N\in B_{X^*} and positive real numbers λ1,…,λN\lambda_1,\ldots,\lambda_N with sum one such that

(fj,fj(y))=(1+ε)∑k=1Nλk(gk,ψ(gk)).(f_j,f_j(y))=(1+\varepsilon)\sum_{k=1}^{N}\lambda_k(g_k,\psi(g_k)).

Thus

fj=(1+ε)∑k=1Nλkgkandfj(y)=(1+ε)∑k=1Nλkψ(gk).(3)f_j=(1+\varepsilon)\sum_{k=1}^{N}\lambda_k g_k \qquad\text{and}\qquad f_j(y)=(1+\varepsilon)\sum_{k=1}^{N}\lambda_k\psi(g_k). \tag*{(3)}

Set A={k:Re⁡gk(vj)≥1−δ2}⊆{1,…,N}A=\{k:\operatorname{Re}g_k(v_j)\geq1-\frac{\delta}{2}\}\subseteq\{1,\ldots,N\} and β=∑k∉Aλk\beta=\sum_{k\notin A}\lambda_k. Since Re⁡(gk(vj))≤∣gk(vj)∣≤∥gk∥⋅∥vj∥X≤1\operatorname{Re}(g_k(v_j))\leq\lvert g_k(v_j)\rvert\leq\lVert g_k\rVert\cdot\lVert v_j\rVert_X\leq1 for all k=1,…,Nk=1,\ldots,N, (3) yields

1=Re⁡(fj(vj))=(1+ε)∑k=1NλkRe⁡(gk(vj))=(1+ε)(∑k∈AλkRe⁡(gk(vj))+∑k∉AλkRe⁡(gk(vj)))≤(1+ε)(1−β+(1−δ2)β)=(1+ε)(1−δβ2),\begin{aligned} 1=\operatorname{Re}(f_j(v_j)) &=(1+\varepsilon)\sum_{k=1}^{N}\lambda_k\operatorname{Re}(g_k(v_j))\\ &=(1+\varepsilon)\left(\sum_{k\in A}\lambda_k\operatorname{Re}(g_k(v_j))+\sum_{k\notin A}\lambda_k\operatorname{Re}(g_k(v_j))\right)\\ &\leq(1+\varepsilon)\left(1-\beta+\left(1-\frac{\delta}{2}\right)\beta\right) =(1+\varepsilon)\left(1-\frac{\delta\beta}{2}\right), \end{aligned}

and hence

(1+ε)β≤2εδ.(4)(1+\varepsilon)\beta\leq\frac{2\varepsilon}{\delta}. \tag*{(4)}

For k∈Ak\in A we have ∣gk(vj)∣≥1−δ2>1−δ\lvert g_k(v_j)\rvert\geq1-\frac{\delta}{2}>1-\delta. Thus the only nonzero summand in ψ(gk)\psi(g_k) is the jj-th summand, and consequently, ψ(gk)=γjgk(vj)∣gk(vj)∣\psi(g_k)=\gamma_j\frac{g_k(v_j)}{\lvert g_k(v_j)\rvert}. Therefore

Re⁡(γj‾ψ(gk))=Re⁡(gk(vj))∣gk(vj)∣≥1−δ2.\operatorname{Re}(\overline{\gamma_j}\psi(g_k)) =\frac{\operatorname{Re}(g_k(v_j))}{\lvert g_k(v_j)\rvert} \geq1-\frac{\delta}{2}.

For k∉Ak\notin A we can simply estimate Re⁡(γj‾ψ(gk))≥−∣ψ(gk)∣≥−1\operatorname{Re}(\overline{\gamma_j}\psi(g_k))\geq-\lvert\psi(g_k)\rvert\geq-1 by (2). Using (3) and (4), we obtain

Re⁡(γj‾fj(y))≥(1+ε)((1−β)(1−δ2)−β)=(1+ε)(1−δ2)−(1+ε)β(2−δ2)≥(1+ε)(1−δ2)−(2−δ2)2εδ≥(1−δ2)−4εδ≥1−ρ312n−24nερ3≥1−1768−35>13.\begin{aligned} \operatorname{Re}(\overline{\gamma_j}f_j(y)) &\geq(1+\varepsilon)\left((1-\beta)\left(1-\frac{\delta}{2}\right)-\beta\right)\\ &=(1+\varepsilon)\left(1-\frac{\delta}{2}\right)-(1+\varepsilon)\beta\left(2-\frac{\delta}{2}\right)\\ &\geq(1+\varepsilon)\left(1-\frac{\delta}{2}\right)-\left(2-\frac{\delta}{2}\right)\frac{2\varepsilon}{\delta}\\ &\geq\left(1-\frac{\delta}{2}\right)-\frac{4\varepsilon}{\delta}\\ &\geq1-\frac{\rho^3}{12n}-\frac{24n\varepsilon}{\rho^3} \geq1-\frac{1}{768}-\frac{3}{5}>\frac{1}{3}. \end{aligned}

This holds for every j=1,…,n+1j=1,\ldots,n+1. Since cj=∣cj∣γj‾c_j=\lvert c_j\rvert\overline{\gamma_j}, the assumed linear dependence of the functionals fjf_j yields

0=Re⁡(∑j=1n+1cjfj(y))=∑j=1n+1∣cj∣Re⁡(γj‾fj(y))>13∑j=1n+1∣cj∣>0.0=\operatorname{Re}\left(\sum_{j=1}^{n+1}c_jf_j(y)\right) =\sum_{j=1}^{n+1}\lvert c_j\rvert\operatorname{Re}(\overline{\gamma_j}f_j(y)) >\frac{1}{3}\sum_{j=1}^{n+1}\lvert c_j\rvert>0.

which is the desired contradiction. This completes the proof. □

We do not know the optimal order in this stability problem. Since

λ(X)λ(Y)≤dBM(X,Y)\frac{\lambda(X)}{\lambda(Y)} \le d_{\mathrm{BM}}(X,Y)

holds for all nn-dimensional normed spaces XX and YY over K\mathbb{K} (see Corollary 6 in Section III.B of [11]), for every nn-dimensional XX we have dBM(X,ℓ∞n)≥λ(X)d_{\mathrm{BM}}(X,\ell_\infty^n) \ge\lambda(X). Thus, if λ(X)=1+ε\lambda(X)=1+\varepsilon, then dBM(X,ℓ∞n)≥1+εd_{\mathrm{BM}}(X,\ell_\infty^n) \ge1+\varepsilon, so linear order is the best one could hope for. Moreover, even if a stability estimate of this optimal order could be obtained, it is not clear whether the relevant constants could be chosen independently of nn, as in Zippin’s estimate.

AI Declaration

The author acknowledges substantial assistance from AI tools in developing the arguments and preparing this paper. The author posed the problem of finding a simple proof of the characterization of ℓ∞\ell_\infty and a quantitative stability estimate to Claude Opus 5.5 (Anthropic). The mathematical arguments presented in this paper were developed through interaction between the author and Claude Opus 5.5. The author subsequently checked and revised the arguments and rewrote the whole paper in his own words. Microsoft Copilot was used to check and refine the revised arguments, explore further simplifications and assist with language editing. The author independently verified the mathematical arguments and references and takes full responsibility for the content of the publication.

Faculty of Mathematics and Computer Science, Jagiellonian University, Łojasiewicza 6, 30-348 Kraków, Poland

Email address: [email protected]

References

  1. [1]S. J. Bernau, Small Projections on ℓ₁(n), Longhorn Notes, The University of Texas, 1983/84.
  2. [2]T. Figiel, W. B. Johnson, Large subspaces of ℓ∞ and estimates of the Gordon–Lewis constant, Israel J. Math. 37 (1980), 92–112.DOI
  3. [3]G. Godefroy, N. J. Kalton, and G. Lancien, Subspaces of c₀(ℕ) and Lipschitz isomorphisms, Geom. Funct. Anal. 10 (2000), 798–820.
  4. [4]D. B. Goodner, Projections in normed linear spaces, Trans. Amer. Math. Soc. 69 (1950), 89–108.DOI
  5. [5]M. Hasumi, The extension property of complex Banach spaces, Tôhoku Math. J. 10 (1958), 135–142.DOI
  6. [6]F. John, Extremum problems with inequalities as subsidiary conditions, in: Studies and Essays Presented to R. Courant on his 60th Birthday, Interscience, New York, 1948, 187–204.
  7. [7]J. L. Kelley, Banach spaces with the extension property, Trans. Amer. Math. Soc. 72 (1952), 323–326.DOI
  8. [8]L. Nachbin, A theorem of the Hahn–Banach type for linear transformations, Trans. Amer. Math. Soc. 68 (1950), 28–46.
  9. [9]R. T. Rockafellar, Convex Analysis, Princeton Mathematical Series, vol. 28, Princeton University Press, Princeton, NJ, 1970.
  10. [10]N. Tomczak-Jaegermann, Banach–Mazur Distances and Finite-Dimensional Operator Ideals, Pitman Monographs and Surveys in Pure and Applied Mathematics, vol. 38, Longman Scientific & Technical, Harlow, 1989.
  11. [11]P. Wojtaszczyk, Banach Spaces for Analysts, Cambridge University Press, Cambridge, 1991.DOI
  12. [12]M. Zippin, The range of a projection of small norm in ℓ¹₁, Israel J. Math. 39 (1981), 349–358.
  13. [13]M. Zippin, The finite-dimensional Pλ spaces with small λ, Israel J. Math. 39 (1981), 359–364.
  14. [14]M. Zippin, Correction to “The finite-dimensional Pλ spaces with small λ”, Israel J. Math. 48 (1984), 255–256.
  15. [15]Direct characterization of finite-dimensional 1-injective Banach spaces, MathOverflow, question 478664, September 11, 2024, https://mathoverflow.net/questions/478664/direct-characterization-of-finite-dimensional-1-injective-banach-spaces.

Paper details

vlnc:W_797txnch
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

Not recorded for this paper

View statement
Looking for related papers…
Contents