betaFeedback

WorksMathematics

The symmetric Mahler conjecture and its equality cases

Abstract

We resolve the symmetric Mahler conjecture positively, including its equality classification. Every origin-symmetric convex body in Rn\mathbb R ^n has volume product at least 4n/n!4^n/n!, with equality exactly for invertible linear images of Hanner polytopes.

Introduction

A convex body is a compact convex subset of Rn\mathbb{R}^n with nonempty interior. If K=−KK=-K, its center 00 is an interior point, and its polar body is

K∘={y∈Rn:⟨x,y⟩≤1 for every x∈K}.K^\circ= \left\{y \in\mathbb{R}^n : \langle x,y\rangle\le1 \text{ for every } x \in K\right\}.

Write ∣K∣|K| for nn-dimensional Lebesgue volume and P(K)=∣K∣∣K∘∣P(K)=|K||K^{\circ}| for the volume product. The two determinant factors cancel under an invertible linear change of coordinates, so P(TK)=P(K)P(TK)=P(K). Mahler’s symmetric volume-product problem asks how small this linear invariant can be. The conjectured value is 4n/n!4^n/n!, the product of a cube and its polar cross-polytope. We prove this bound in every dimension and classify the equality cases.

The extremal bodies are defined recursively. A centered nondegenerate interval is a one-dimensional Hanner polytope. Given Hanner polytopes H1⊂RkH_1\subset\mathbb{R}^k and H2⊂RlH_2\subset\mathbb{R}^l, their Cartesian product and the convex hull

H1⊕∞H2=H1×H2,H_1\oplus_{\infty}H_2=H_1\times H_2,
H1⊕1H2=conv⁡((H1×{0})∪({0}×H2))H_1\oplus_1 H_2=\operatorname{conv}\bigl((H_1\times\{0\})\cup(\{0\}\times H_2)\bigr)

are Hanner polytopes in Rk+l\mathbb{R}^{k+l}. These are the ℓ∞\ell_{\infty} and ℓ1\ell_1 sums, respectively. This class contains cubes, cross-polytopes, and mixed recursive sums. We call an invertible linear image of a Hanner polytope a linear Hanner body.

Theorem 1.1. For every integer n≥1n\ge1 and every origin-symmetric convex body K⊂RnK\subset\mathbb{R}^n,

∣K∣∣K∘∣≥4nn!(1)|K||K^{\circ}|\ge\frac{4^n}{n!} \tag*{(1)}

Equality holds if and only if KK is a linear Hanner body.

The linear equivalence in the statement respects the fixed origin used for polarity.

History and context

Mahler introduced the volume-product problem in his work on convex bodies and transference in the geometry of numbers [22]; the planar inequality also goes back to him [21]. Reisner later characterized its equality cases as parallelograms, as a consequence of his zonoid theorem [28], p. 340. Iriyeh and Shibata proved the complete three-dimensional symmetric theorem, including equality [14], Theorem 1.1. Fradelizi, Hubard, Meyer, Roldán-Pensado, and Zvavitch gave a shorter proof based on equipartitions, together with another equality argument and stability [9]. Chen, Li, Xi, and Xu recently gave a shadow-flow proof of that theorem in a preprint that also treats the nonsymmetric three-dimensional problem [4], Theorem 7.1.

Hanner’s recursively constructed spaces [12] already play a central role in the known higher-dimensional cases. Saint-Raymond proved the sharp inequality for unconditional bodies (those invariant under coordinate sign changes) [30]; Meyer and Reisner identified the Hanner equality cases in that class [23, 29]. Reisner also proved the sharp inequality for zonoids, which are Hausdorff limits of Minkowski sums of centered segments, and showed that equality within that class means a parallelotope [27, 28]; Gordon, Meyer, and Reisner later gave a short geometric proof [10]. For local questions in Banach–Mazur distance, Nazarov, Petrov, Ryabogin, and Zvavitch established strict local minimality, up to linear equivalence, of the cube [25]. Kim extended this to all Hanner polytopes [16]. Kim and Zvavitch obtained stability in the unconditional class and the inequality in a neighborhood of that class [17].

A separate line of work seeks dimension-independent exponential bounds without the sharp constant. Bourgain and Milman established such a reverse Blaschke–Santaló inequality [3]; Kuperberg obtained an explicit stronger bound by a geometric argument using Gauss linking integrals [18]. Complex-analytic approaches provide another setting for these questions. Nazarov used Hörmander’s ∂ˉ\bar{\partial} theorem and Bergman kernels for the reverse Blaschke–Santaló inequality [24]. Berndtsson recast Kuperberg’s proof using complex integrals [1]. In a symplectic approach, Karasev proved the sharp inequality for hyperplane sections and projections of ℓp\ell_p balls (1≤p≤∞1 \le p \le\infty) and Hanner polytopes [15], Theorems 4.2 and 4.4. These works give analytic and geometric precedents for volume-product estimates; their estimates are not hypotheses of our proof.

The analytic ingredient is a conformal map of the disk for which uniform angular measure on the boundary maps to a uniform imaginary coordinate. It is a rotation of Gross’s example for the uniform distribution in his conformal Skorokhod embedding [11], Section 3.2. A holomorphic mass estimate turns that boundary law into a lower bound for an integral of feasible-simplex volumes over the primal body. We prove the mass estimate directly by Stokes’ theorem, with its precise normalization; its general background is the theory of generalized Lelong numbers [6], Sections 3 and 5. The symmetric-convex extremal-function framework of Lundin and Baran was a further motivation for the analytic search; see [20] and the formula recorded in [2], Proposition 1.2, pp. 246–247. The proof here instead uses the direct isolated-zero mass calculation, not those extremal-function formulas as premises. The equality argument leads to a classical property of normed spaces: any three pairwise-intersecting closed balls have a common point. Hanner studied the associated recursively constructed convex bodies [12]. Lima’s norm-additive decomposition criterion [19], Theorem 1.2, also recorded in Hansen–Lima [13], Theorem 3.6(4), is exactly the metric-median condition used below. Reisner’s unconditional equality theorem already connects minimal volume product to this ball intersection structure [29], Theorem 1. Our endpoint argument obtains the structure without assuming unconditionality; Hansen–Lima’s finite-dimensional classification is the final input [13], Corollary 7.4.

For a body without a prescribed center, the general Mahler problem instead minimizes ∣K∣∣(K−z)∘∣|K| |(K-z)^{\circ}| over z∈int⁡Kz \in\operatorname{int} K. Its conjectured constant (n+1)n+1/(n!)2(n+1)^{n+1}/(n!)^2 is smaller than the symmetric constant for n≥2n \ge2; a result for that general problem alone therefore does not establish Theorem 1.1. The argument here uses symmetry throughout and is independent of the general theorem proved in the companion paper [26], Theorem 1.1.

Proof overview

Section 2 records polarity and the two norm-sum formulas. They show that every Hanner body has the stated volume product. The remaining work is the lower bound and its converse equality statement.

First let

A={X∈Rd:∣bi⋅X∣≤1, 1≤i≤m},A∘=conv⁡{±b1,…,±bm}.A = \{X \in\mathbb{R}^d : |b_i \cdot X| \le1,\ 1 \le i \le m\}, \qquad A^\circ= \operatorname{conv}\{\pm b_1,\ldots,\pm b_m\}.

where the nonzero, pairwise nonproportional rows bib_i span Rd\mathbb{R}^d. Section 3 constructs a planar lens with horizontal half-width λ(t)\lambda(t) at height tt. For X∈int⁡AX \in\operatorname{int} A, use this width to form

LX={Y∈Rd:∣bi⋅Y∣≤λ(bi⋅X), 1≤i≤m}.L_X = \{Y \in\mathbb{R}^d : |b_i \cdot Y| \le\lambda(b_i \cdot X),\ 1 \le i \le m\}.

Each independent dd-tuple of signed constraints that are tight at one point of LXL_X determines a simplex: take the convex hull of 00 and those signed rows. Let ΣX\Sigma_X be the union of these simplices. It lies in A∘A^\circ.

Write FF for the conformal map from the unit disk to the lens. For each set II of dd independent rows, sample independent uniform angles θi\theta_i and solve biZ=F(eiθi)b_i Z = F(e^{i\theta_i}), i∈Ii \in I, for the complex vector ZZ. Let PIP_I be the probability that all remaining row coordinates also lie in the closed lens. Section 5 applies the mass estimate of Section 4 to high powers of the inverse map: after a change of radial variables, the Jacobian integrals concentrate on these boundary samples and force S:=∑IPI≥1S := \sum_I P_I \ge1. The uniform imaginary coordinate then identifies this probability sum with the simplex integral in Section 6:

1≤S=d!4d∫A∣ΣX∣ dX≤d!4d∣A∣ ∣A∘∣.1 \le S = \frac{d!}{4^d}\int_A |\Sigma_X|\,dX \le\frac{d!}{4^d}|A|\,|A^\circ|.

The first inequality is analytic; the last is the inclusion just described. The exact identity behind this bound also controls the missing volume ∣A∘∣−∣ΣX∣|A^\circ|-|\Sigma_X|. Approximation extends the inequality to all origin-symmetric convex bodies.

For equality, set Q=K∘Q=K^\circ and pass to the body

B=K⊕1[−1,1],B∘=Q×[−1,1]B=K\oplus_1[-1,1],\qquad B^\circ=Q\times[-1,1]

in dimension d=n+1d=n+1. The norm-sum formula preserves equality at the new dimension. Inner polytope approximations to QQ therefore have integrated missing volume tending to zero. Section 7 uses product neighborhoods to obtain, for each interior point of BB and each nonzero direction, a representation whose rows attain their lens bounds at one vector satisfying all the constraints. This includes degenerate limiting representations. Finally, Section 8 examines points approaching an apex of BB. The lens asymptotic gives a support-function inequality, and convex separation gives a metric median for every triple in the norm with unit ball QQ: a point whose distances to each pair sum to that pair’s distance. The Hansen–Lima theorem then identifies QQ, and hence KK, as a linear Hanner body [13], Corollary 7.4.

We next derive functional and entropy–transport consequences of the all-dimensional theorem. The geometric proof begins in Section 2 and ends in Section 8.

Functional Mahler inequality

The volume-product problem has a functional form in which polarity is replaced by convex conjugation. For a proper lower-semicontinuous convex function φ:Rn→R∪{+∞}\varphi:\mathbb{R}^n\to\mathbb{R}\cup\{+\infty\}, its Fenchel–Legendre transform is

φ∗(y)=sup⁡x∈Rn{⟨x,y⟩−φ(x)},y∈Rn.\varphi^*(y)=\sup_{x\in\mathbb{R}^n}\{\langle x,y\rangle-\varphi(x)\},\qquad y\in\mathbb{R}^n.

We use the convention e−∞=0e^{-\infty}=0.

Corollary 1.2 (Even functional Mahler inequality). For every integer n≥1n\ge1 and every even proper lower-semicontinuous convex function φ:Rn→R∪{+∞}\varphi:\mathbb{R}^n\to\mathbb{R}\cup\{+\infty\} such that 0<∫Rne−φ(x) dx<∞0<\int_{\mathbb{R}^n}e^{-\varphi(x)}\,dx<\infty,

(∫Rne−φ(x) dx)(∫Rne−φ∗(y) dy)≥4n.\left(\int_{\mathbb{R}^n}e^{-\varphi(x)}\,dx\right)\left(\int_{\mathbb{R}^n}e^{-\varphi^*(y)}\,dy\right)\ge4^n.

The second integral is allowed to be infinite, in which case the inequality is immediate.

Proof. Set f=e−φf=e^{-\varphi}, a nonzero integrable log-concave function. For z∈Rnz\in\mathbb{R}^n, Fradelizi and Meyer use the polar function

fz(y)=inf⁡{x:f(x)>0}e−⟨x−z,y−z⟩f(x)f^z(y)=\inf_{\{x:f(x)>0\}}\frac{e^{-\langle x-z,y-z\rangle}}{f(x)}

and the translation-minimized functional volume product

P(f)=(∫Rnf(x) dx)inf⁡z∈Rn∫Rnfz(y) dy.\mathcal{P}(f)=\left(\int_{\mathbb{R}^n}f(x)\,dx\right)\inf_{z\in\mathbb{R}^n}\int_{\mathbb{R}^n}f^z(y)\,dy.

Their Proposition 1(a) states that the symmetric geometric Mahler inequality in every dimension implies P(f)≥4n\mathcal{P}(f)\ge4^n for every even log-concave function on Rn\mathbb{R}^n with 0<∫Rnf<∞0<\int_{\mathbb{R}^n}f<\infty [7]. Theorem 1.1 supplies this all-dimensional hypothesis. Since f0=e−φ∗f^0=e^{-\varphi^*}, the product in Corollary 1.2 is at least P(f)\mathcal{P}(f).

The cited implication uses auxiliary convex bodies in dimensions n+mn+m and lets mm tend to infinity; it is not an application only of the geometric theorem in dimension nn. The constant is sharp: for φ(x)=∑i=1n∣xi∣\varphi(x)=\sum_{i=1}^{n}|x_i|, the transform φ∗\varphi^{*} is zero on [−1,1]n[-1,1]^n and +∞+\infty outside, and both integrals in the corollary equal 2n2^n. The equality classification in Theorem 1.1 concerns convex bodies. The limiting implication does not by itself classify the even potentials for which equality holds [7], p. 1437.

For potentials without evenness, the companion paper on general convex bodies gives the sharp lower bound ene^n [26], Corollary 1.2.

The functional inequality also has an entropy–transport formulation. For a log-concave probability measure η\eta with a density, let H(η)=∫log⁡(dη/dx) dηH(\eta)=\int\log(\mathrm{d}\eta/\mathrm{d}x)\,\mathrm{d}\eta denote its entropy relative to Lebesgue measure, the negative of differential Shannon entropy. For probability measures μ1,μ2\mu_1,\mu_2 with finite first moments, set

T(μ1,μ2)=inf⁡f∈F{∫f dμ1+∫f∗ dμ2}.\mathcal{T}(\mu_1,\mu_2)=\inf_{f\in\mathcal{F}}\left\{\int f\,\mathrm{d}\mu_1+\int f^{*}\,\mathrm{d}\mu_2\right\}.

where F\mathcal{F} is the class of proper lower-semicontinuous convex functions Rn→R∪{+∞}\mathbb{R}^{n}\to\mathbb{R}\cup\{+\infty\} and f∗f^{*} is the Fenchel–Legendre transform. The cost T\mathcal{T} is allowed to be +∞+\infty.

Corollary 1.3 (Symmetric entropy–transport inequality). For every integer n≥1n\ge1, let ηi(dx)=e−Vi(x) dx\eta_i(\mathrm{d}x)=e^{-V_i(x)}\,\mathrm{d}x, i=1,2i=1,2, be full-dimensional log-concave probability measures, both symmetric about the origin, with proper lower-semicontinuous convex potentials ViV_i. Suppose their densities are essentially continuous, meaning that e−Vi=0e^{-V_i}=0 at Hn−1\mathcal{H}^{n-1}-almost every point of ∂supp⁡ηi\partial\operatorname{supp}\eta_i. Their moment measures νi=(∇Vi)#ηi\nu_i=(\nabla V_i)_{\#}\eta_i have finite first moments, H(ηi)=−∫Vi dηiH(\eta_i)=-\int V_i\,\mathrm{d}\eta_i is finite, and

H(η1)+H(η2)≤−nlog⁡(4e2)+T(ν1,ν2).H(\eta_1)+H(\eta_2)\le-n\log(4e^2)+\mathcal{T}(\nu_1,\nu_2).

Proof. For ρi=e−Vi\rho_i=e^{-V_i}, the identity ∣∇ρi∣=∣∇Vi∣e−Vi|\nabla\rho_i|=|\nabla V_i|e^{-V_i} holds almost everywhere on the interior of the support. Lemma 4 and the proof of Proposition 7 of [5] give ∫∣∇ρi∣<∞\int|\nabla\rho_i|<\infty and Vi∈L1(ηi)V_i\in L^1(\eta_i), respectively. The pushforward definition of νi\nu_i therefore gives its finite first moment, and H(ηi)=−∫Vi dηiH(\eta_i)=-\int V_i\,\mathrm{d}\eta_i is finite. Gozlan’s equivalence between the symmetric functional inverse Santaló and entropy–transport inequalities, as stated in [8], Theorem 1.3, applies to Corollary 1.2 with c=4c=4. □

Polarity, norm sums, and Hanner polytopes

All bodies below are symmetric about the origin. In addition to the polar and volume product defined in Section 1, we use the gauge and support function

pA(x)=inf⁡{t>0:x∈tA},hA(y)=max⁡x∈A⟨x,y⟩.p_A(x)=\inf\{t>0:x\in tA\},\qquad h_A(y)=\max_{x\in A}\langle x,y\rangle.

Dual spaces are identified with Euclidean coordinate spaces, and ∣A∣|A| always denotes volume in the dimension of AA. The notation hAh_A also applies to any nonempty compact convex set. The gauge is a norm with closed unit ball AA. Convex separation gives the bipolar identity A∘∘=AA^{\circ\circ}=A, and hence

pA=hA∘,hA=pA∘.(2)p_A=h_{A^\circ},\qquad h_A=p_{A^\circ}. \tag*{(2)}

Indeed x∈tAx\in tA is equivalent to ⟨x,y⟩≤t\langle x,y\rangle\le t for every y∈A∘y\in A^\circ. For an invertible linear map TT,

(TA)∘=T−TA∘,P(TA)=P(A),(3)(TA)^\circ=T^{-T}A^\circ,\qquad P(TA)=P(A), \tag*{(3)}

since the two volumes acquire reciprocal determinant factors.

For bodies A⊂RkA\subset\mathbb{R}^k and A′⊂RlA'\subset\mathbb{R}^l, define

A⊕∞A′=A×A′,A⊕1A′=conv⁡((A×{0})∪({0}×A′)).A\oplus_\infty A'=A\times A',\qquad A\oplus_1 A'=\operatorname{conv}((A\times\{0\})\cup(\{0\}\times A')).

These operations are taken in the product space. More generally, complementary subspaces are identified with product coordinates by an invertible linear map. Their gauges are

pA⊕∞A′(x,x′)=max⁡{pA(x),pA′(x′)},pA⊕1A′(x,x′)=pA(x)+pA′(x′).(4)\begin{aligned} p_{A\oplus_{\infty} A'}(x,x') &= \max\{p_A(x),p_{A'}(x')\},\\ p_{A\oplus_1 A'}(x,x') &= p_A(x)+p_{A'}(x'). \tag*{(4)} \end{aligned}

For the second identity, a convex combination from the two coordinate bodies has gauge sum at most one. Conversely, when the sum is at most one, use these two gauges as the coefficients of the normalized coordinate vectors and place any remaining weight at 0. A zero coordinate contributes no term.

Lemma 2.1 (Polars and volumes of the two sums). For origin-symmetric convex bodies of dimensions k,l≥1k,l \ge1,

(A⊕1A′)∘=A∘⊕∞(A′)∘,(A⊕∞A′)∘=A∘⊕1(A′)∘.(5)(A\oplus_1 A')^\circ=A^\circ\oplus_\infty(A')^\circ,\qquad(A\oplus_\infty A')^\circ=A^\circ\oplus_1 (A')^\circ. \tag*{(5)}

Moreover,

∣A⊕∞A′∣=∣A∣∣A′∣,∣A⊕1A′∣=k!l!(k+l)!∣A∣∣A′∣.(6)|A\oplus_\infty A'|=|A||A'|,\qquad|A\oplus_1 A'|=\frac{k!l!}{(k+l)!}|A||A'|. \tag*{(6)}

Consequently, for s∈{1,∞}s\in\{1,\infty\},

P(A⊕sA′)=P(A)P(A′)(k+lk).(7)P(A\oplus_s A')=\frac{P(A)P(A')}{\binom{k+l}{k}}. \tag*{(7)}

Proof. Testing a functional on the two coordinate bodies gives the first polar identity; taking polars again gives the second. The product volume follows from Fubini. By (4), the fiber of A⊕1A′A\oplus_1 A' over x∈Ax\in A is (1−pA(x))A′(1-p_A(x))A'. Since ∣{x:pA(x)≤t}∣=tk∣A∣|\{x:p_A(x)\le t\}|=t^k|A| for 0≤t≤10\le t\le1, integration gives

∣A⊕1A′∣=k∣A∣∣A′∣∫01tk−1(1−t)l dt=k!l!(k+l)!∣A∣∣A′∣.|A\oplus_1 A'|=k|A||A'|\int_0^1 t^{k-1}(1-t)^l\,dt=\frac{k!l!}{(k+l)!}|A||A'|.

The last integral is evaluated by repeated integration by parts. Combining the polar and volume identities proves (7).

The Hanner recursion from Section 1 uses precisely these two operations, following Hanner [12]. Their volume and polarity formulas give the attaining family immediately.

Lemma 2.2 (The Hanner volume product). Every dd-dimensional linear Hanner body HH satisfies

P(H)=4dd!.P(H)=\frac{4^d}{d!}.

The polar of a linear Hanner body is again a linear Hanner body.

Proof. A centered interval [−a,a][-a,a], a>0a>0, has polar [−1/a,1/a][-1/a,1/a] and volume product 4. Equation (7) propagates 4k/k!4^k/k! and 4l/l!4^l/l! to 4k+l/(k+l)!4^{k+l}/(k+l)! under either operation. Induction and linear invariance prove the first assertion. The polar identities exchange the two operations and preserve the interval base case. Equation (3) then proves closure under polarity also for linear images.

In dimension one every origin-symmetric convex body is a centered nondegenerate interval. Thus both the sharp inequality and the full equality classification in that dimension follow directly from the base case above.

A conformal lens with a uniform boundary coordinate

We construct a conformal image of the disk whose boundary has a uniformly distributed imaginary coordinate. This law will turn complex feasibility probabilities into real volumes. The horizontal half-width has further roles: strict curvature will make additional boundary equalities have probability zero, while concavity will extend feasibility from polar vertices to their convex hull. Its endpoint scaling will enter the equality argument near the apex of the lifted body. The map is a rotation of the uniform-distribution example in Gross’s conformal Skorokhod embedding [11], Section 3.2; we prove all its required properties directly.

Write D={z∈C:∣z∣<1}\mathbb{D} = \{z \in\mathbb{C} : |z| < 1\} and define

F(z)=8π2∑j=0∞(−1)jz2j+1(2j+1)2.(8)F(z) = \frac{8}{\pi^2} \sum_{j=0}^{\infty} \frac{(-1)^j z^{2j+1}}{(2j+1)^2}. \tag*{(8)}

The series converges uniformly on D‾\overline{\mathbb{D}}.

Lemma 3.1 (The planar lens). The function FF is a biholomorphism from D\mathbb{D} onto a bounded convex domain DD, with inverse gg, and F′(0)=8/π2F'(0) = 8/\pi^2. There is a continuous even function λ:[−1,1]→[0,∞)\lambda: [-1,1] \to[0,\infty), positive on (−1,1)(-1,1) and zero at ±1\pm1, such that

D‾={v+it:∣t∣≤1, ∣v∣≤λ(t)}.(9)\overline{D} = \{v + it : |t| \le1,\ |v| \le\lambda(t)\}. \tag*{(9)}

The function λ\lambda is smooth on (−1,1)(-1,1) and satisfies

λ′(t)=−2πarctanh⁡(sin⁡πt2),λ′′(t)=−sec⁡πt2<0.(10)\lambda'(t) = -\frac{2}{\pi}\operatorname{arctanh}\left(\sin\frac{\pi t}{2}\right), \qquad\lambda''(t) = -\sec\frac{\pi t}{2} < 0. \tag*{(10)}

If Θ\Theta is uniform on [0,2π)[0,2\pi), then

F(eiΘ)=lawελ(T)+iT,(11)F(e^{i\Theta}) \mathrel{\stackrel{\mathrm{law}}{=}} \varepsilon\lambda(T) + iT, \tag*{(11)}

where TT is uniform on [−1,1][-1,1] and ε\varepsilon is an independent uniform sign.

Proof. The map is odd and commutes with conjugation. Put

w=arctan⁡z=12ilog⁡1+iz1−iz,z∈D.w = \arctan z = \frac{1}{2i}\log\frac{1+iz}{1-iz}, \qquad z \in\mathbb{D}.

The fraction inside the logarithm lies in the right half-plane, where we use the principal logarithm. Thus ∣ℜw∣<π/4|\Re w| < \pi/4, tan⁡w=z\tan w = z, and w=0w = 0 only when z=0z = 0. Differentiating the series gives

F′(z)=8π2wz,1+zF′′(z)F′(z)=sin⁡(2w)2w,(12)F'(z) = \frac{8}{\pi^2}\frac{w}{z}, \qquad1 + \frac{zF''(z)}{F'(z)} = \frac{\sin(2w)}{2w}, \tag*{(12)}

with their limiting values at zero. In particular F′F' has no zeros. For w=a+ib≠0w = a + ib \ne0, the real part of the last expression is

asin⁡(2a)cosh⁡(2b)+bcos⁡(2a)sinh⁡(2b)2(a2+b2)>0,\frac{a\sin(2a)\cosh(2b) + b\cos(2a)\sinh(2b)}{2(a^2+b^2)} > 0,

since ∣a∣<π/4|a| < \pi/4; its value at zero is 1. Fix 0<r<10 < r < 1 and put γr(θ)=F(reiθ)\gamma_r(\theta) = F(re^{i\theta}). This curve is regular, and its tangent angle β\beta satisfies

β′(θ)=ℜ(1+zF′′(z)F′(z))>0,β(θ+2π)=β(θ)+2π.\beta'(\theta) = \Re\left(1 + \frac{zF''(z)}{F'(z)}\right) > 0, \qquad\beta(\theta+ 2\pi) = \beta(\theta) + 2\pi.

The second identity uses the zero-free derivative in the disk. At a fixed parameter θ0\theta_0, project the curve onto its right unit normal there. The derivative of this projection has the sign of −sin⁡(β(θ)−β(θ0))-\sin(\beta(\theta)-\beta(\theta_0)). It is negative until the tangent has turned through π\pi, and positive thereafter, until the curve returns to θ0\theta_0. Hence the projection has its unique maximum at θ0\theta_0. This proves simplicity and shows that every point of the curve is the unique supporting point of its convex hull in the corresponding normal direction. All normal directions occur, so the curve is the boundary of that convex hull and is positively oriented. The harmonic maximum principle for these supporting projections places F(rD)F(r\mathbb{D}) strictly inside the curve. The argument principle then gives exactly one preimage for every point inside and none outside. Thus FF maps rDr\mathbb{D} biholomorphically onto its bounded convex interior DrD_r.

As rr increases, these domains exhaust D=F(D)D=F(\mathbb{D}). Consequently FF is univalent on D\mathbb{D}, DD is convex, and its inverse gg is holomorphic. For each 0<r<10<r<1, continuity also gives Dr‾=F(rD‾)⊂D\overline{D_r}=F(r\overline{\mathbb{D}})\subset D. Continuity on the closed disk gives boundedness and D‾=F(D‾)\overline{D}=F(\overline{\mathbb{D}}). No boundary value F(ζ)F(\zeta), ∣ζ∣=1|\zeta|=1, lies in DD: otherwise continuity of gg there and g(F(rζ))=rζg(F(r\zeta))=r\zeta would give g(F(ζ))=ζ∉Dg(F(\zeta))=\zeta\notin\mathbb{D}. It follows that ∂D=F(∂D)\partial D=F(\partial\mathbb{D}).

We now compute this boundary. On the right semicircle, z=eisz=e^{is} with −π/2<s<π/2-\pi/2<s<\pi/2, one has

1+iz1−iz=icos⁡s1+sin⁡s,ℜw=π4,ℑw=12arctanh⁡(sin⁡s).\frac{1+iz}{1-iz}=\frac{i\cos s}{1+\sin s},\qquad\Re w=\frac{\pi}{4},\qquad\Im w=\frac{1}{2}\operatorname{arctanh}(\sin s).

The formulas for ww and F′F' extend analytically across each compact subarc, so we may differentiate the boundary values. Writing F(eis)=v(s)+it(s)F(e^{is})=v(s)+it(s) gives

t′(s)=2π,v′(s)=−4π2arctanh⁡(sin⁡s).t'(s)=\frac{2}{\pi},\qquad v'(s)=-\frac{4}{\pi^2}\operatorname{arctanh}(\sin s).

Since t(0)=0t(0)=0, we have t(s)=2s/πt(s)=2s/\pi. Define λ(t)=v(πt/2)\lambda(t)=v(\pi t/2) and extend it to [−1,1][-1,1] by continuity. Conjugation makes λ\lambda even. The values F(±i)F(\pm i) are purely imaginary by the series, so λ(±1)=0\lambda(\pm1)=0. Differentiation gives (10); strict concavity and the endpoint values then give positivity in the interior. Oddness and conjugation show that the other semicircle traces the graph v=−λ(t)v=-\lambda(t). These two graphs give (9). On either semicircle the imaginary coordinate traverses [−1,1][-1,1] at constant absolute speed 2/π2/\pi. Each semicircle has probability 1/21/2, which proves (11).

Figure 1 illustrates the coordinate law.

Figure 1. The conformal lens and its boundary coordinate. Equally spaced angles on either semicircle give equally spaced imaginary coordinates on the corresponding boundary branch. On the right semicircle, F(eis)=λ(2s/π)+2is/πF(e^{is})=\lambda(2s/\pi)+2is/\pi; reflection gives the left branch. Thus a uniform angle gives a uniform coordinate in [−1,1][-1,1] and an independent fair branch sign. The horizontal arrow marks the half-width λ(t0)\lambda(t_0), with s0=π/6s_0=\pi/6 and t0=1/3t_0=1/3.

The two boundary graphs are smooth away from ±i\pm i and have strictly turning tangent directions by (10). In particular, a fixed tangent line direction occurs at only finitely many parameters away from these two endpoints. The boundary has planar Lebesgue measure zero, as is also immediate from its graph description.

Lemma 3.2 (Endpoint scaling). As η↓0\eta\downarrow0,

λ(1−η)=2πηlog⁡1η+O(η).(13)\lambda(1-\eta)=\frac{2}{\pi}\eta\log\frac{1}{\eta}+O(\eta). \tag*{(13)}

For every 0<a≤b<∞0<a\le b<\infty,

sup⁡a≤c≤b∣λ(1−δc)λ(1−δ)−c∣⟶0(δ↓0).(14)\sup_{a\le c\le b}\left|\frac{\lambda(1-\delta c)}{\lambda(1-\delta)}-c\right|\longrightarrow0\qquad(\delta\downarrow0). \tag*{(14)}

Proof. Equation (10) and arctanh⁡(cos⁡x)=log⁡cot⁡(x/2)\operatorname{arctanh}(\cos x)=\log\cot(x/2) give

λ(1−η)=2π∫0ηlog⁡cot⁡πs4 ds.\lambda(1-\eta)=\frac{2}{\pi}\int_0^\eta\log\cot\frac{\pi s}{4}\,ds.

Here log⁡cot⁡(πs/4)=log⁡(1/s)+O(1)\log\cot(\pi s/4)=\log(1/s)+O(1) uniformly as s↓0s\downarrow0. Integration proves (13). Writing L=log⁡(1/δ)L=\log(1/\delta), the same uniform remainder gives

λ(1−δc)λ(1−δ)=cL−log⁡c+O(1)L+O(1)=c+Oa,b(L−1)\frac{\lambda(1-\delta c)}{\lambda(1-\delta)}=c\frac{L-\log c+O(1)}{L+O(1)}=c+O_{a,b}(L^{-1})

uniformly for a≤c≤ba\le c\le b, which proves (14).

Mass at an isolated holomorphic zero

The next estimate turns the order of a holomorphic zero into an integral bound for its Jacobian minors. Its application will use high powers of the inverse planar map. In complex dimension dd, the mass bound grows as the ddth power of their exponent, matching the radial rescaling that will produce boundary feasibility probabilities in Section 5. The estimate is a special case of the sublevel-mass and comparison formulas for generalized Lelong numbers; see Demailly [6], Sections 3 and 5. We give a direct Stokes proof in the normalization used here.

On Cd\mathbb{C}^d, write zj=xj+iyjz_j=x_j+iy_j and use the complex orientation

dV=dx1∧dy1∧⋯∧dxd∧dyd.dV=dx_1\wedge dy_1\wedge\cdots\wedge dx_d\wedge dy_d.

For a smooth real function vv, our convention is

dc=i4(∂ˉ−∂),ddc=i2∂∂ˉ.(15)d^c=\frac{i}{4}(\bar\partial-\partial),\qquad dd^c=\frac{i}{2}\partial\bar\partial. \tag*{(15)}

Thus dcv(ξ)=−14dv(iξ)d^c v(\xi)=-\frac{1}{4}dv(i\xi) for a real tangent vector ξ\xi, and

ddc∣z∣2=∑j=1ddxj∧dyj,(ddcv)d=d!det⁡(vzˉizj)dV.(16)dd^c|z|^2=\sum_{j=1}^{d}dx_j\wedge dy_j,\qquad(dd^c v)^d=d!\det(v_{\bar z_i z_j})dV. \tag*{(16)}

The second identity follows by expanding the exterior product, or by a unitary diagonalization of the Hermitian Hessian.

Lemma 4.1 (Holomorphic mass). Let d≥1d\ge1, N≥dN\ge d, and let U⊂Cd\mathcal{U}\subset\mathbb{C}^d be open and contain 00. Suppose that f=(f1,…,fN):U→CNf=(f_1,\ldots,f_N):\mathcal{U}\to\mathbb{C}^N is holomorphic and has the following properties:

  1. f−1(0)={0}f^{-1}(0)=\{0\};

  1. for an integer k≥1k\ge1,

f(z)=H(z)+O(∣z∣k+1)near 0,f(z)=H(z)+O(|z|^{k+1})\quad\text{near }0,

where each component of HH is a homogeneous polynomial of degree kk and H(z)≠0H(z)\ne0 whenever z≠0z\ne0;

  1. for every 0<s<10<s<1, the set {z∈U:∣f(z)∣2≤s}\{z\in\mathcal{U}:|f(z)|^2\le s\} is compact in U\mathcal{U}.

Then, with τ=∣f∣2\tau=|f|^2,

∫{τ<1}∑I⊂{1,…,N}∣I∣=d∣det⁡(∂fi∂zj)i∈I,1≤j≤d∣2dV≥(πk)dd!.(17)\int_{\{\tau<1\}}\sum_{\substack{I\subset\{1,\ldots,N\}\\ |I|=d}}\left|\det\left(\frac{\partial f_i}{\partial z_j}\right)_{\substack{i\in I,\,1\le j\le d}}\right|^2dV\ge\frac{(\pi k)^d}{d!}. \tag*{(17)}

All norms are Euclidean. Rows in each determinant are in increasing index order.

Proof. We first bound the integral of (ddcτ)d(dd^c\tau)^d. The conversion to Jacobian minors will then follow from Cauchy–Binet. Both τ\tau and log⁡τ\log\tau away from 00 have positive semidefinite complex Hessians. Indeed, if ZZ is the complex derivative matrix of ff, the Hessian of τ\tau is Z∗ZZ^*Z. For a complex direction ξ\xi, the Hessian of log⁡τ\log\tau evaluates to

∣f∣2∣df(ξ)∣2−∣⟨df(ξ),f⟩∣2∣f∣4≥0\frac{|f|^2|df(\xi)|^2-|\langle df(\xi),f\rangle|^2}{|f|^4}\ge0

by Cauchy–Schwarz. Their top exterior powers are therefore nonnegative.

Comparison with a small sphere. Fix a regular value s∈(0,1)s\in(0,1) of τ\tau and set Es={z∈U:τ(z)<s}E_s=\{z\in\mathcal{U}:\tau(z)<s\}. By compactness of the sublevel and the regular-value theorem, Es‾\overline{E_s} is a compact smooth manifold with boundary {τ=s}\{\tau=s\}. Stokes’ theorem gives

∫Es(ddcτ)d=∫∂Esdcτ∧(ddcτ)d−1.\int_{E_s}(dd^c\tau)^d=\int_{\partial E_s}d^c\tau\wedge(dd^c\tau)^{d-1}.

Put γ=dclog⁡τ\gamma=d^c\log\tau on U∖{0}\mathcal{U}\setminus\{0\}. The identities

dclog⁡τ=τ−1dcτ,ddclog⁡τ=τ−1ddcτ−τ−2dτ∧dcτd^c\log\tau=\tau^{-1}d^c\tau,\qquad dd^c\log\tau=\tau^{-1}dd^c\tau-\tau^{-2}d\tau\wedge d^c\tau

and the vanishing of the pullback of dτd\tau to ∂Es\partial E_s imply

s−d∫Es(ddcτ)d=∫∂Esγ∧(dγ)d−1.(18)s^{-d}\int_{E_s}(dd^c\tau)^d=\int_{\partial E_s}\gamma\wedge(d\gamma)^{d-1}. \tag*{(18)}

For sufficiently small ε>0\varepsilon> 0, the closed Euclidean ball B‾ε\overline{B}_{\varepsilon} is contained in EsE_s. Orient Sε=∂BεS_{\varepsilon} = \partial B_{\varepsilon} outward from this ball. Applying Stokes on Es∖B‾εE_s \setminus\overline{B}_{\varepsilon} gives

s−d∫Es(ddcτ)d−∫Sεγ∧(dγ)d−1=∫Es∖B‾ε(ddclog⁡τ)d≥0.(19)s^{-d}\int_{E_s}(dd^c\tau)^d-\int_{S_{\varepsilon}}\gamma\wedge(d\gamma)^{d-1}=\int_{E_s\setminus\overline{B}_{\varepsilon}}(dd^c\log\tau)^d\geq0. \tag*{(19)}

Thus it remains to determine the flux through a shrinking sphere.

The homogeneous flux. Let T(z)=∣H(z)∣2T(z)=|H(z)|^2 and S={z:∣z∣=1}S=\{z:|z|=1\}. The holomorphic Taylor expansion implies

ε−2kτ(εz)⟶T(z)in C2\varepsilon^{-2k}\tau(\varepsilon z)\longrightarrow T(z)\quad\text{in }C^2

on a fixed closed annulus containing SS.

For example, termwise differentiation of the convergent Taylor series gives this convergence for the rescaled map and then for its squared norm. Since TT is strictly positive on that annulus, the logarithms converge in C2C^2 as well. Pulling back to SS by dilation, the additive constant 2klog⁡ε2k\log\varepsilon disappears under dcd^c. Consequently

lim⁡ε↓0∫Sεγ∧(dγ)d−1=∫Sα∧(dα)d−1,α=dclog⁡T∣S.(20)\lim_{\varepsilon\downarrow0}\int_{S_{\varepsilon}}\gamma\wedge(d\gamma)^{d-1}=\int_S\alpha\wedge(d\alpha)^{d-1},\qquad\alpha=d^c\log T\big|_S. \tag*{(20)}

We compare this flux with that of the radial function ∣z∣2k|z|^{2k}. Homogeneity gives

T(rz)=r2kT(z)(r>0),T(eiθz)=T(z).T(rz)=r^{2k}T(z)\quad(r>0),\qquad T(e^{i\theta}z)=T(z).

Thus log⁡T−klog⁡∣z∣2\log T-k\log|z|^2 is unchanged by dilation and by common phase rotation. These invariances will make its contribution to the flux vanish. On SS, put

α0=dclog⁡∣z∣2∣S,β=dc(log⁡T−klog⁡∣z∣2)∣S,αt=kα0+tβ(0≤t≤1).\alpha_0=d^c\log|z|^2\big|_S,\qquad\beta=d^c(\log T-k\log|z|^2)\big|_S,\qquad\alpha_t=k\alpha_0+t\beta\quad(0\leq t\leq1).

Thus α1=α\alpha_1=\alpha. For the nowhere-zero tangent vector field V(z)=izV(z)=iz on SS, the dilation identity and (15) give β(V)=0\beta(V)=0 and α0(V)=1/2\alpha_0(V)=1/2. The phase-rotation identity makes every αt\alpha_t invariant under the flow of VV. Cartan’s identity therefore gives

ιVdαt=LVαt−d(αt(V))=0,\iota_Vd\alpha_t=\mathcal{L}_V\alpha_t-d(\alpha_t(V))=0,

where ιV\iota_V denotes contraction and LV\mathcal{L}_V the Lie derivative. It follows that the top-degree form β∧(dαt)d−1\beta\wedge(d\alpha_t)^{d-1} on the (2d−1)(2d-1)-dimensional sphere vanishes: its contraction with the nonzero vector VV is zero. Differentiation and integration by parts on the closed sphere now yield

ddt∫Sαt∧(dαt)d−1=d∫Sβ∧(dαt)d−1=0.\frac{d}{dt}\int_S\alpha_t\wedge(d\alpha_t)^{d-1}=d\int_S\beta\wedge(d\alpha_t)^{d-1}=0.

For d=1d=1 this is simply the derivative of ∫Sαt\int_S\alpha_t; the same argument applies.

On the unit sphere, α0\alpha_0 and dα0d\alpha_0 are the restrictions of dc∣z∣2d^c|z|^2 and ddc∣z∣2dd^c|z|^2, respectively. Hence a final use of Stokes and (16) gives

∫Sα∧(dα)d−1=kd∫Sα0∧(dα0)d−1=kd∫{∣z∣<1}(ddc∣z∣2)d=kdd! vol⁡2d({∣z∣<1})=(πk)d.\begin{aligned} \int_S\alpha\wedge(d\alpha)^{d-1} &=k^d\int_S\alpha_0\wedge(d\alpha_0)^{d-1}\\ &=k^d\int_{\{|z|<1\}}(dd^c|z|^2)^d\\ &=k^d d!\,\operatorname{vol}_{2d}(\{|z|<1\})=(\pi k)^d. \end{aligned}

The minor bound. Equations (19) and (20) imply

∫Es(ddcτ)d≥sd(πk)d\int_{E_s}(dd^c\tau)^d \ge s^d(\pi k)^d

for every regular s∈(0,1)s \in(0,1). Sard’s theorem supplies an increasing sequence of such values tending to 11. Since the integrand is nonnegative, monotone convergence gives

∫{τ<1}(ddcτ)d≥(πk)d.\int_{\{\tau<1\}}(dd^c\tau)^d \ge(\pi k)^d.

Finally, the complex Hessian Z∗ZZ^*Z and Cauchy–Binet give the identity

(ddcτ)d=d!∑∣I∣=d∣det⁡ZI∣2 dV.(dd^c\tau)^d = d! \sum_{|I|=d} \lvert\det Z_I\rvert^2\,dV.

Division by d!d! proves (17).

Feasible bases

We now apply the holomorphic mass estimate to finitely many real strip constraints. It gives a lower bound for the sum of their boundary feasibility probabilities. For the lens of Lemma 3.1, these probabilities will become volumes of real simplices.

Fix d≥1d \ge1 and nonzero real row vectors b1,…,bmb_1,\ldots,b_m spanning (Rd)∗(\mathbb{R}^d)^*. Write

A={X∈Rd:∣biX∣≤1 for every i},C=A∘=conv⁡{±b1,…,±bm}.(21)A=\{X\in\mathbb{R}^d:\lvert b_iX\rvert\le1\text{ for every }i\},\qquad C=A^\circ=\operatorname{conv}\{\pm b_1,\ldots,\pm b_m\}. \tag*{(21)}

We identify rows with vectors when taking a convex hull, and extend their action complex-linearly to Cd\mathbb{C}^d. The spanning assumption makes AA bounded, and 0∈int⁡A0\in\operatorname{int}A. The polar identity follows from the bipolar theorem: the polar of the displayed convex hull is exactly AA. Let B\mathcal{B} be the collection of dd-element sets of independent rows. For I∈BI\in\mathcal{B}, put BI=(bi)i∈IB_I=(b_i)_{i\in I} in increasing index order and DI=∣det⁡BI∣D_I=\lvert\det B_I\rvert. Dependent row sets are omitted throughout.

For the next lemma, FF may be any biholomorphism of the unit disk onto a bounded planar domain DD, with F(0)=0F(0)=0 and a continuous extension to the closed disk. For independent uniform angles θi∈[0,2π)\theta_i\in[0,2\pi), i∈Ii\in I, define

ZI(θ)=BI−1(F(eiθi))i∈I,PI=P{bjZI(θ)∈D‾ for all j},S=∑I∈BPI.(22)Z_I(\theta)=B_I^{-1}\bigl(F(e^{i\theta_i})\bigr)_{i\in I},\qquad P_I=\mathbb{P}\{b_jZ_I(\theta)\in\overline{D}\text{ for all }j\},\qquad S=\sum_{I\in\mathcal{B}}P_I. \tag*{(22)}

The active coordinates i∈Ii\in I already belong to D‾\overline{D}; thus the probability tests only the remaining rows.

Lemma 5.1 (Boundary feasibility). Let d≥1d\ge1 and let b1,…,bmb_1,\ldots,b_m be nonzero real rows spanning (Rd)∗(\mathbb{R}^d)^*. Let F:D→DF:\mathbb{D}\to D be a biholomorphism onto a bounded planar domain, continuous on D‾\overline{\mathbb{D}}, with F(0)=0F(0)=0. Then the sum SS in (22) satisfies S≥1S\ge1. No condition on extra-row boundary equalities or on dependent row subsets is required.

Proof. Let g=F−1g=F^{-1} and set

Ω={z∈Cd:bjz∈D for every j},hj(z)=g(bjz).\Omega=\{z\in\mathbb{C}^d:b_jz\in D\text{ for every }j\},\qquad h_j(z)=g(b_jz).

The set Ω\Omega is open, bounded, and contains 0. For an integer k≥1k \ge1, apply Lemma 4.1 to the holomorphic map

fk=(h1k,…,hmk),τk=∑j=1m∣hj∣2k.f_k = (h_1^k,\ldots,h_m^k), \qquad\tau_k = \sum_{j=1}^{m} \lvert h_j\rvert^{2k}.

We shall let kk tend to infinity. The mass lower bound has scale kdk^d. The radial substitution below removes the same factor from the Jacobian integrals and leaves a fixed domain on which the inferred points tend to boundary samples. The unique zero of fkf_k is 0: gg vanishes only at 0, and the rows span. Writing c=g′(0)≠0c = g'(0) \ne0, its degree-kk leading term is ((cb1z)k,…,(cbmz)k)((cb_1z)^k,\ldots,(cb_mz)^k), also nonzero whenever z≠0z \ne0. For 0<s<10 < s < 1, the set {τk≤s}\{\tau_k \le s\} is compact in Ω\Omega. Indeed each of its row coordinates belongs to the fixed compact set F({∣w∣≤s1/(2k)})⊂DF(\{\lvert w\rvert\le s^{1/(2k)}\}) \subset D. A basis of rows bounds zz; limits retain all these inclusions and the inequality τk≤s\tau_k \le s. All hypotheses of the mass lemma therefore hold. The derivative row of hjkh_j^k is a scalar multiple of bjb_j, so dependent row subsets have zero minors. Consequently

∑I∈BMI,k≥kdπdd!,MI,k:=∫{τk<1}∣det⁡∂(hik)i∈I∂(z1,…,zd)∣2dV2d(z).(23)\sum_{I\in\mathcal{B}} M_{I,k} \ge k^d\frac{\pi^d}{d!}, \qquad M_{I,k} := \int_{\{\tau_k<1\}} \left\lvert\det\frac{\partial(h_i^k)_{i\in I}}{\partial(z_1,\ldots,z_d)} \right\rvert^2 dV_{2d}(z). \tag*{(23)}

Fix II. The change of variables ui=hi(z)u_i = h_i(z), i∈Ii \in I, is one-to-one onto its image, with inverse z=BI−1(F(ui))i∈Iz = B_I^{-1}(F(u_i))_{i\in I}. Its squared complex Jacobian is its real volume Jacobian. The squared determinant in MI,kM_{I,k} is this Jacobian factor times k2d∏i∣ui∣2k−2k^{2d}\prod_i\lvert u_i\rvert^{2k-2}. The transformed integration domain is contained in

{u∈Dd:∑i∈I∣ui∣2k<1,bjBI−1(F(ui))i∈I∈D for all j}.\left\{u \in \mathbb{D}^d : \sum_{i\in I}\lvert u_i\rvert^{2k} < 1,\quad b_jB_I^{-1}(F(u_i))_{i\in I} \in D\ \text{for all }j\right\}.

We use this larger domain for an upper bound, and then put ui=ρi1/keiθiu_i = \rho_i^{1/k}e^{i\theta_i}. The coordinate hyperplanes, on which polar coordinates are undefined, have Lebesgue measure zero. Since

k2∣ui∣2k−2dV2(ui)=kρi dρi dθi,k^2\lvert u_i\rvert^{2k-2}dV_2(u_i) = k\rho_i\,d\rho_i\,d\theta_i,

we obtain

MI,kkd≤∫ρi>0, ∑iρi2<1θ∈[0,2π)d1{bjBI−1(F(ρi1/keiθi))i∈I∈D for every j}∏i∈Iρi dρi dθi.(24)\frac{M_{I,k}}{k^d} \le\int_{\substack{\rho_i>0,\ \sum_i\rho_i^2<1\\ \theta\in[0,2\pi)^d}} \mathbf{1}\left\{b_jB_I^{-1}\left(F\left(\rho_i^{1/k}e^{i\theta_i}\right)\right)_{i\in I}\in D\ \text{for every }j\right\}\prod_{i\in I}\rho_i\,d\rho_i\,d\theta_i. \tag*{(24)}

The measure in this integral is finite: its total mass is the volume πd/d!\pi^d/d! of the unit ball in Cd\mathbb{C}^d. For every positive ρi\rho_i, continuity of FF on the closed disk makes the inferred points converge to ZI(θ)Z_I(\theta). If a limiting row coordinate lies outside D‾\overline{D}, that constraint fails for all sufficiently large kk. Thus the upper limit of the indicator is bounded by the indicator of the event defining PIP_I, including all possible boundary equalities. The upper-bound form of Fatou’s lemma, applied to functions bounded by 1 on this finite measure space, gives

lim sup⁡k→∞MI,kkd≤πdd!PI.\limsup_{k\to\infty}\frac{M_{I,k}}{k^d} \le\frac{\pi^d}{d!}P_I.

There are only finitely many bases, so (23) and these upper bounds give

πdd!≤lim sup⁡k→∞∑I∈BMI,kkd≤∑I∈Blim sup⁡k→∞MI,kkd≤πdd!∑I∈BPI.\frac{\pi^d}{d!} \le\limsup_{k\to\infty} \sum_{I\in\mathcal{B}} \frac{M_{I,k}}{k^d} \le\sum_{I\in\mathcal{B}} \limsup_{k\to\infty} \frac{M_{I,k}}{k^d} \le\frac{\pi^d}{d!}\sum_{I\in\mathcal{B}} P_I.

Thus S≥1S \ge1.

Boundary equalities for the lens

The lower bound for SS allows additional boundary equalities. The real volume identity in Section 6 will need the stronger fact that these equalities have probability zero. From now on, FF, DD, and λ\lambda are those of Lemma 3.1. We may also require that no two rows are proportional. For any finite strip presentation this is achieved without changing AA by keeping, on each row line, a row of largest magnitude; its strip implies all the discarded strips on that line. Zero rows, if present initially, impose no constraint and are discarded. In dimension one the reduced list has a single row, so there are no extra-row boundary equalities to consider.

Lemma 5.2 (Additional boundary equalities have probability zero). Let F:D→DF : \mathbb{D} \to D be the lens map of Lemma 3.1. Assume that the nonzero spanning rows b1,…,bmb_1,\ldots,b_m are pairwise nonproportional. For every I∈BI \in\mathcal{B} and j∉Ij \notin I,

P{bjZI(θ)∈∂D}=0.\mathbb{P}\{b_j Z_I(\theta) \in\partial D\}=0.

Proof. Write bjBI−1=(ci)i∈Ib_j B_I^{-1}=(c_i)_{i\in I}. At least two coefficients, say ca,cbc_a,c_b, are nonzero; otherwise bjb_j would be proportional to a basis row. Let γ(θ)=F(eiθ)\gamma(\theta)=F(e^{i\theta}). Apart from its two endpoints, the lens boundary consists of the smooth regular branches (±λ(t),t)(\pm\lambda(t),t), −1<t<1-1<t<1. Their tangent directions have at most one occurrence on each branch for any fixed unoriented line direction, because λ′\lambda' is strictly monotone. The boundary parameter has constant nonzero speed in tt on each branch, as established in the proof of Lemma 3.1.

Hold all phases other than θa,θb\theta_a,\theta_b fixed. The map

(θa,θb)⟼caγ(θa)+cbγ(θb)+∑i≠a,bciγ(θi)(\theta_a,\theta_b)\longmapsto c_a\gamma(\theta_a)+c_b\gamma(\theta_b)+\sum_{i\ne a,b}c_i\gamma(\theta_i)

has real Jacobian determinant det⁡(caγ′(θa),cbγ′(θb))\det(c_a\gamma'(\theta_a),c_b\gamma'(\theta_b)). For every nonendpoint θa\theta_a, it vanishes at only finitely many nonendpoint values of θb\theta_b. Its zero set, including the excluded endpoint lines, therefore has planar measure zero by Fubini’s theorem. Off that set the inverse function theorem gives local diffeomorphism charts. A countable subcover of such charts suffices, and in each chart the inverse is locally Lipschitz, so it sends a planar null set to a null set. The set ∂D\partial D is planar-null, being a countable union of compact smooth subarcs and two points. Its inverse image consequently has measure zero in the two remaining phases. A final application of Fubini over the fixed phases proves the assertion.

The real simplex identity

The lens boundary law now turns the feasibility sum into an exact integral over the strip body. The simplices in this integral lie in the polar body; their missing volume will also be the quantity used in the equality argument.

Keep the nonzero spanning, pairwise nonproportional rows of (21). For X∈int⁡AX \in\operatorname{int} A, define

LX={Y∈Rd:∣bjY∣≤λ(bjX) for every j}.(25)L_X=\{Y\in\mathbb{R}^d: |b_jY|\leq\lambda(b_jX)\text{ for every }j\}. \tag*{(25)}

All widths are strictly positive, so LXL_X contains a neighborhood of 00; a basis of rows shows that it is bounded. It is therefore a convex polytope with nonempty interior. For a basis II and signs ϵ=(ϵi)i∈I∈{−1,1}I\epsilon=(\epsilon_i)_{i\in I}\in\{-1,1\}^I, let

YI,ϵ(X)=BI−1(ϵiλ(biX))i∈I.Y_{I,\epsilon}(X)=B_I^{-1}\bigl(\epsilon_i\lambda(b_iX)\bigr)_{i\in I}.

We call (I,ϵ)(I,\epsilon) feasible at XX when this vector belongs to LXL_X. This definition specifies a unique vector for every choice, whether or not it is feasible. Set

TI,ϵ=conv⁡(0,{ϵibi:i∈I}),ΣX=⋃(I,ϵ) feasible at XTI,ϵ.(26)T_{I,\epsilon}=\operatorname{conv}\bigl(0,\{\epsilon_i b_i:i\in I\}\bigr),\qquad \Sigma_X=\bigcup_{(I,\epsilon)\text{ feasible at }X}T_{I,\epsilon}. \tag*{(26)}

Thus ΣX\Sigma_X is a finite union of closed dd-simplices contained in C=A∘C=A^\circ. This definition applies at every interior XX, including points with additional tight constraints. For integration only, declare every pair infeasible and set ΣX=∅\Sigma_X=\varnothing when X∈∂AX\in\partial A.

A feasible pair specifies a vertex YI,ϵ(X)Y_{I,\epsilon}(X) of LXL_X. A strictly positive combination of its signed tight rows defines a linear functional maximized uniquely at that vertex. This is why different feasible pairs give simplices with disjoint interiors whenever their vertices are distinct. The proof below removes, outside a null set of XX, the boundary ties that could make two pairs specify the same vertex.

Proposition 6.1 (Feasible-simplex identity). Let b1,…,bmb_1,\ldots,b_m be nonzero, pairwise nonproportional real rows spanning (Rd)∗(\mathbb{R}^d)^*, and let F,D,λF,D,\lambda be as in Lemma 3.1. With AA, SS, and ΣX\Sigma_X defined in (21), (22), and (26), respectively,

1≤S=d!4d∫A∣ΣX∣ dX≤d!4d∣A∣ ∣A∘∣.(27)1\leq S=\frac{d!}{4^d}\int_A|\Sigma_X|\,dX\leq\frac{d!}{4^d}|A|\,|A^\circ|. \tag*{(27)}

In particular, ∣A∣ ∣A∘∣≥4d/d!|A|\,|A^\circ|\geq4^d/d!, and

0≤∫A(∣A∘∣−∣ΣX∣) dX≤∣A∣ ∣A∘∣−4dd!.(28)0\leq\int_A\bigl(|A^\circ|-|\Sigma_X|\bigr)\,dX\leq|A|\,|A^\circ|-\frac{4^d}{d!}. \tag*{(28)}

Proof. We first convert each probability into a real integral. By (11), each active boundary coordinate has the law ϵiλ(ti)+iti\epsilon_i\lambda(t_i)+it_i, with tit_i uniform on [−1,1][-1,1] and ϵi\epsilon_i an independent fair sign. Consequently each fixed sign choice has joint measure 4−d dtI4^{-d}\,dt_I on [−1,1]d[-1,1]^d. Writing the inferred complex vector as Y+iXY+iX gives

tI=BIX,Y=YI,ϵ(X).t_I=B_IX,\qquad Y=Y_{I,\epsilon}(X).

The constraints that all row coordinates lie in D‾\overline{D} are exactly X∈AX\in A and Y∈LXY\in L_X, apart from the immaterial boundary ∂A\partial A. This boundary has measure zero, since it is contained in the finitely many hyperplanes bjX=±1b_jX=\pm1. The change of variables tI=BIXt_I=B_IX therefore gives

PI=DI4d∫A∑ϵ∈{−1,1}I1{(I,ϵ) is feasible at X} dX.(29)P_I=\frac{D_I}{4^d}\int_A\sum_{\epsilon\in\{-1,1\}^I}\mathbf{1}\{(I,\epsilon)\text{ is feasible at }X\}\,dX. \tag*{(29)}

For almost every X∈int⁡AX\in\operatorname{int} A, every feasible pair (I,ϵ)(I,\epsilon) has no tight constraint outside II. Indeed, such an additional equality means that an extra row of Y+iXY+iX lies on ∂D\partial D. Lemma 5.2 makes this event null in the basis phases; the boundary law and the invertible change tI=BIXt_I=B_IX make its set of XX null for each fixed sign choice. There are finitely many bases, sign choices, and extra rows, so a single null set removes all these equalities.

Fix XX outside this null set. Distinct feasible pairs have distinct vectors YI,ε(X)Y_{I,\varepsilon}(X). If their bases differ, a shared vector would have a tight row outside one of the bases. If their bases agree but some signs differ, the shared vector would satisfy opposite equations with the same strictly positive width, which is impossible.

The interiors of their simplices are disjoint. To see this, let e∈int⁡TI,εe \in\operatorname{int} T_{I,\varepsilon}. It has a representation e=∑i∈Iαiεibie = \sum_{i \in I} \alpha_i \varepsilon_i b_i with all αi>0\alpha_i > 0 and ∑iαi<1\sum_i \alpha_i < 1. For every Y∈LXY \in L_X,

eY≤∑i∈Iαiλ(biX),eY \le\sum_{i \in I} \alpha_i \lambda(b_iX),

and equality holds at YI,ε(X)Y_{I,\varepsilon}(X). Equality forces every independent equation εibiY=λ(biX)\varepsilon_i b_iY = \lambda(b_iX), so this is the unique maximizer of the linear functional ee on LXL_X. If ee belonged to the interiors of two feasible simplices, their distinct vectors would both be its unique maximizer, a contradiction. Each simplex has volume DI/d!D_I/d!, and their boundaries have measure zero. It follows that, for almost every XX,

d!∣ΣX∣=∑I∈B∑ε∈{−1,1}IDI1{(I,ε) is feasible at X}.(30)d! \lvert\Sigma_X\rvert= \sum_{I \in\mathcal{B}} \sum_{\varepsilon\in\{-1,1\}^{I}} D_I 1\{(I,\varepsilon)\text{ is feasible at }X\}. \tag*{(30)}

All integrals here are measurable. For each pair, feasibility is a finite collection of closed inequalities between continuous functions of X∈int⁡AX \in\operatorname{int} A. In particular the set

{(X,e):X∈int⁡A, e∈ΣX}\{(X,e): X \in\operatorname{int} A,\ e \in\Sigma_X\}

is a finite union of products of Borel feasibility sets and fixed closed simplices. Fubini’s theorem makes X↦∣ΣX∣X \mapsto\lvert\Sigma_X\rvert measurable, and it is bounded by ∣C∣\lvert C\rvert. Sum (29), use (30), and apply Lemma 5.1. This proves the first equality and lower bound in (27). The containment ΣX⊂C\Sigma_X \subset C proves the upper bound. In particular, the missing volume has the exact expression

∫A(∣C∣−∣ΣX∣) dX=∣A∣∣C∣−4dd!S.\int_A (\lvert C\rvert-\lvert\Sigma_X\rvert)\,dX = \lvert A\rvert\lvert C\rvert-\frac{4^d}{d!}S.

It is nonnegative by containment, and S≥1S \ge1 bounds it above by ∣A∣∣C∣−4d/d!\lvert A\rvert\lvert C\rvert-4^d/d!. This proves (28) and identifies the quantity that will tend to zero in the equality argument. □

For a concrete union with positive missing volume, take b1=(1,0)b_1=(1,0), b2=(1/2,3/2)b_2=(1/2,\sqrt{3}/2), b3=b2−b1b_3=b_2-b_1, and X=0.9(1,1/3)X=0.9(1,1/\sqrt{3}), as in Figure 2. Then b1X=b2X=0.9b_1X=b_2X=0.9 and b3X=0b_3X=0. Write a=λ(0.9)a=\lambda(0.9). The endpoint integral for λ\lambda and cot⁡x<1/x\cot x<1/x for 0<x<π/20<x<\pi/2 give

Figure 2. A feasible-simplex union can leave part of the polar uncovered. Here b1=(1,0)b_1=(1,0), b2=(1/2,3/2)b_2=(1/2,\sqrt{3}/2), b3=(−1/2,3/2)b_3=(-1/2,\sqrt{3}/2), C=conv⁡{±b1,±b2,±b3}C=\operatorname{conv}\{\pm b_1,\pm b_2,\pm b_3\}, and X=0.9(1,1/3)X=0.9(1,1/\sqrt{3}). The third strip is redundant in LXL_X, as verified above. The four witnesses Yϵ1ϵ2Y_{\epsilon_1\epsilon_2}, defined by biY=ϵiab_iY=\epsilon_i a for i=1,2i=1,2, correspond to the four triangles conv⁡(0,ϵ1b1,ϵ2b2)\operatorname{conv}(0,\epsilon_1b_1,\epsilon_2b_2) with matching sign labels. Their union is ΣX=conv⁡{±b1,±b2}\Sigma_X=\operatorname{conv}\{\pm b_1,\pm b_2\}. The two hatched caps have total area 3/2\sqrt{3}/2, one third of ∣C∣|C|.

a<log⁡(40/π)+15π<415,λ(0)>8π2(1−19)>3245.a < \frac{\log(40/\pi)+1}{5\pi}<\frac{4}{15}, \qquad\lambda(0)>\frac{8}{\pi^2}\left(1-\frac{1}{9}\right)>\frac{32}{45}.

The second bound uses the alternating series for F(1)F(1). In particular, 2a<λ(0)2a<\lambda(0), so the third strip in the example is strictly redundant.

Approximation of an arbitrary convex body

Corollary 6.2 (The symmetric volume-product inequality). For every origin-symmetric convex body K⊂RdK \subset\mathbb{R}^d, d≥1d \ge1,

∣K∣∣K∘∣≥4dd!.\lvert K\rvert\lvert K^\circ\rvert\ge\frac{4^d}{d!}.

Proof. Put Q=K∘Q=K^\circ and choose a>0a>0 with aB2d⊂QaB_2^d\subset Q, where B2dB_2^d is the Euclidean unit ball. For εj↓0\varepsilon_j\downarrow0, take a finite εj\varepsilon_j-net in QQ, adjoin its negatives, and let QjQ_j be its convex hull. Thus Qj⊂QQ_j\subset Q, and their support functions satisfy, for every Euclidean unit vector uu,

hQ(u)−εj≤hQj(u)≤hQ(u),hQ(u)≥a.h_Q(u)-\varepsilon_j\le h_{Q_j}(u)\le h_Q(u),\qquad h_Q(u)\ge a.

For δj=εj/a<1\delta_j=\varepsilon_j/a<1, this implies

(1−δj)Q⊂Qj⊂Q,K⊂Aj:=Qj∘⊂(1−δj)−1K.(31)(1-\delta_j)Q\subset Q_j\subset Q,\qquad K\subset A_j:=Q_j^\circ\subset(1-\delta_j)^{-1}K. \tag*{(31)}

The first inclusions follow from the support-function inequalities and convex separation; the second follow by polarity. In particular QjQ_j and AjA_j are full-dimensional symmetric polytopes for large jj.

Choose one vertex from each antipodal vertex pair of QjQ_j as a row list. These rows are nonzero and span. They are pairwise nonproportional: two distinct boundary vertices on the same ray from an interior origin cannot both be extreme, and the antipodal duplication has already been removed. Their strip body is AjA_j. Proposition 6.1 therefore gives ∣Aj∣ ∣Qj∣≥4d/d!|A_j|\,|Q_j|\ge4^d/d!. The homothetic bounds (31) show both ∣Qj∣→∣Q∣|Q_j|\to|Q| and ∣Aj∣→∣K∣|A_j|\to|K|, because dilation by t>0t>0 multiplies volume by tdt^d. Passing to the limit proves the assertion. No smoothness, simplicity, or general-position assumption is imposed on KK. In dimension one the assertion can also be read directly: K=[−a,a]K=[-a,a] has polar [−1/a,1/a][-1/a,1/a] and product 4. □

Equality and a common feasibility witness

The integral in (27) measures how much of the polar body is covered by feasible simplices. For a body attaining the sharp volume product, we shall make the uncovered volume tend to zero after adjoining one segment. Compactness then gives a feasible witness at every interior point and in every direction. The resulting representation will be the input to the equality classification.

Throughout this section K=−K⊂RnK=-K\subset\mathbb{R}^{n} is a convex body, n≥1n\geq1, satisfying

∣K∣∣K∘∣=4nn!.|K||K^{\circ}|=\frac{4^{n}}{n!}.

Set Q=K∘Q=K^{\circ}, d=n+1d=n+1, and

B=K⊕1[−1,1]={(x,s):pK(x)+∣s∣≤1},C=B∘=Q×[−1,1].(32)B=K\oplus_{1}[-1,1]=\{(x,s):p_{K}(x)+|s|\leq1\},\qquad C=B^{\circ}=Q\times[-1,1]. \tag*{(32)}

The polar CC has the two horizontal faces Q×{1}Q\times\{1\} and Q×{−1}Q\times\{-1\}. The representations constructed below will use vectors from these faces. In this lifted space, XX and YY belong to Rd\mathbb{R}^{d}, while elements of CC act on them by the Euclidean pairing. Recall that the lens width λ:[−1,1]→[0,∞)\lambda:[-1,1]\to[0,\infty) is continuous, even and concave, with λ(0)>0\lambda(0)>0 and λ(t)≤λ(0)\lambda(t)\leq\lambda(0).

Proposition 7.1 (A representation minimizing the lens cost). Let K⊂RnK\subset\mathbb{R}^{n}, n≥1n\geq1, be an origin-symmetric convex body with P(K)=4n/n!P(K)=4^{n}/n!. Put Q=K∘Q=K^{\circ}, d=n+1d=n+1, and define the lifted bodies B,CB,C by (32). For every X∈int⁡BX\in\operatorname{int}B and every nonzero ξ=(q,a)∈Rn×R\xi=(q,a)\in\mathbb{R}^{n}\times\mathbb{R}, put

T=pC(ξ)=max⁡{pQ(q),∣a∣}.T=p_{C}(\xi)=\max\{p_{Q}(q),|a|\}.

There are 1≤r≤d1\leq r\leq d vectors ci=(bi,σi)c_i=(b_i,\sigma_i) with bi∈Qb_i\in Q and σi∈{−1,1}\sigma_i\in\{-1,1\}, nonnegative numbers αi\alpha_i, and a vector Y∈RdY\in\mathbb{R}^{d}, such that

ξ=∑i=1rαici,∑i=1rαi=T,(33)\xi=\sum_{i=1}^{r}\alpha_i c_i,\qquad\sum_{i=1}^{r}\alpha_i=T, \tag*{(33)}

and

∣h⋅Y∣≤λ(h⋅X)(h∈C),ci⋅Y=λ(ci⋅X)(1≤i≤r).(34)|h\cdot Y|\leq\lambda(h\cdot X)\quad(h\in C),\qquad c_i\cdot Y=\lambda(c_i\cdot X)\quad(1\leq i\leq r). \tag*{(34)}

In particular, for every finite representation ξ=∑j=1sβjhj\xi=\sum_{j=1}^{s}\beta_jh_j with βj≥0\beta_j\geq0 and hj∈Ch_j\in C,

∑i=1rαiλ(ci⋅X)≤∑j=1sβjλ(hj⋅X).(35)\sum_{i=1}^{r}\alpha_i\lambda(c_i\cdot X)\leq\sum_{j=1}^{s}\beta_j\lambda(h_j\cdot X). \tag*{(35)}

The choices may depend on XX and ξ\xi; the same choice satisfies (35) for all the competing representations.

The last coordinates σi\sigma_i split the chosen rows into two groups. In Section 8, we compare representation costs as XX approaches the apex (0,1)(0,1) of BB; the endpoint behavior of λ\lambda makes the two signs contribute differently. We now prove Proposition 7.1 in three steps: approximate CC by polytopes, locate feasible simplices near each prescribed point and direction, and pass to a common witness to compare costs.

Inner polytopes and vanishing missing volume

The horizontal section of BB at height s∈[−1,1]s\in[-1,1] is (1−∣s∣)K(1-|s|)K. Hence

∣B∣=∣K∣∫−11(1−∣s∣)n ds=2∣K∣n+1,∣C∣=2∣Q∣,|B|=|K|\int_{-1}^{1}(1-|s|)^{n}\,\mathrm{d}s=\frac{2|K|}{n+1},\qquad|C|=2|Q|,

and therefore

∣B∣∣C∣=4n+1∣K∣∣Q∣=4dd!.(36)|B||C|=\frac{4}{n+1}|K||Q|=\frac{4^{d}}{d!}. \tag*{(36)}

Choose full-dimensional symmetric polytopes Qj⊂QQ_j \subset Q converging to QQ in Hausdorff distance. For example, take the convex hulls of increasingly fine finite symmetric nets in QQ, including a fixed spanning symmetric set. There are numbers tj∈(0,1)t_j \in(0,1) with tj→1t_j \to1 such that

tjQ⊆Qj⊆Q.t_jQ \subseteq Q_j \subseteq Q.

Indeed, if QQ contains a Euclidean ball of radius r0>0r_0 > 0 and the Hausdorff error is εj\varepsilon_j, then on Euclidean unit directions

hQj≥hQ−εj≥(1−εj/r0)hQ.h_{Q_j} \ge h_Q-\varepsilon_j \ge(1-\varepsilon_j/r_0)h_Q.

The support-function criterion for containment gives the claim, after discarding finitely many indices and choosing tjt_j slightly smaller if necessary. Put Cj=Qj×[−1,1]C_j=Q_j\times[-1,1] and Bj=Cj∘B_j=C_j^\circ. The inclusions above give

tjC⊆Cj⊆C,B⊆Bj⊆tj−1B.(37)t_jC \subseteq C_j \subseteq C,\qquad B\subseteq B_j\subseteq t_j^{-1}B. \tag*{(37)}

In particular, ∣Bj∣∣Cj∣→∣B∣∣C∣|B_j||C_j|\to|B||C|. Use one row from each antipodal vertex pair of CjC_j to describe BjB_j as an intersection of strips. These rows span Rd\mathbb{R}^d, and no two are proportional: a ray from the interior point 00 meets the boundary of a convex body in only one point. Every signed row is a vertex of CjC_j, and thus belongs to Qj×{−1,1}Q_j\times\{-1,1\}.

Let Σj,X\Sigma_{j,X} be the union of the feasible signed basis simplices for this strip description, as in (27). In particular, for every X∈int⁡BjX\in\operatorname{int}B_j it is a finite union of closed simplices contained in CjC_j. Its definition is retained also at the exceptional interior points where several constraints are simultaneously tight. The integral estimate and (36) imply

0≤Dj:=∫Bj(∣Cj∣−∣Σj,X∣) dX≤∣Bj∣∣Cj∣−4dd!⟶0.(38)\begin{aligned} 0\le D_j&:=\int_{B_j}(|C_j|-|\Sigma_{j,X}|)\,dX\\ &\le|B_j||C_j|-\frac{4^d}{d!}\longrightarrow0. \tag*{(38)} \end{aligned}

From an integral deficit to each point and direction

Fix X∈int⁡BX\in\operatorname{int}B and e∈∂Ce\in\partial C. We first construct indices jk→∞j_k\to\infty and points

Xk∈int⁡Bjk,ek∈Σjk,Xk,(Xk,ek)⟶(X,e).(39)X_k\in\operatorname{int}B_{j_k},\qquad e_k\in\Sigma_{j_k,X_k},\qquad(X_k,e_k)\longrightarrow(X,e). \tag*{(39)}

This is a consequence of (38) on product neighborhoods; it does not require a pointwise limit of the functions X↦∣Σj,X∣X\mapsto|\Sigma_{j,X}|.

For each kk, choose a Euclidean ball UkU_k of positive volume contained in int⁡B∩B(X,1/k)\operatorname{int}B\cap B(X,1/k). There is also a ball VkV_k of positive volume with closure contained in int⁡C∩B(e,1/k)\operatorname{int}C\cap B(e,1/k). To see this at the boundary point ee, first move a small distance from ee toward 00; the resulting point is interior because 0∈int⁡C0\in\operatorname{int}C. Take a sufficiently small ball about that point. By (37), Uk⊂int⁡BjU_k\subset\operatorname{int}B_j for every jj, and Vk⊂CjV_k\subset C_j for all sufficiently large jj. The second assertion follows also from max⁡h∈Vk‾pC(h)<1\max_{h\in\overline{V_k}}p_C(h)<1 and tj→1t_j\to1.

Choose jk>jk−1j_k>j_{k-1} so large that these inclusions hold and Djk<∣Uk∣∣Vk∣D_{j_k}<|U_k||V_k|. If Σjk,Z\Sigma_{j_k,Z} missed VkV_k for every Z∈UkZ\in U_k, its missing volume would be at least ∣Vk∣|V_k| throughout UkU_k, contradicting this inequality. Thus choose Xk∈UkX_k\in U_k and ek∈Σjk,Xk∩Vke_k\in\Sigma_{j_k,X_k}\cap V_k. This proves (39).

Choose one feasible simplex containing eke_k. Its dd signed basis rows, written c1k,…,cdk∈Qjk×{−1,1}c_{1k},\ldots,c_{dk}\in Q_{j_k}\times\{-1,1\}, give weights θik≥0\theta_{ik}\ge0 satisfying

ek=∑i=1dθikcik,∑i=1dθik≤1.(40)e_k=\sum_{i=1}^{d}\theta_{ik}c_{ik},\qquad\sum_{i=1}^{d}\theta_{ik}\le1. \tag*{(40)}

The coefficient left over is the weight of the simplex vertex 0. Feasibility supplies one vector YkY_k such that

cik⋅Yk=λ(cik⋅Xk)(1≤i≤d),∣v⋅Yk∣≤λ(v⋅Xk)for every vertex v of Cjk.c_{ik}\cdot Y_k=\lambda(c_{ik}\cdot X_k)\quad(1\le i\le d),\qquad|v\cdot Y_k|\le\lambda(v\cdot X_k)\quad\text{for every vertex }v\text{ of }C_{j_k}.

Here a sign on a basis row has been absorbed using the evenness of λ\lambda.

The last inequality extends to all h∈Cjkh\in C_{j_k}. Indeed, write h=∑ℓsℓvℓh=\sum_\ell s_\ell v_\ell as a convex combination of vertices. The triangle inequality followed by concavity gives

∣h⋅Yk∣≤∑ℓsℓ∣vℓ⋅Yk∣≤∑ℓsℓλ(vℓ⋅Xk)≤λ(h⋅Xk).|h\cdot Y_k|\le\sum_\ell s_\ell|v_\ell\cdot Y_k|\le\sum_\ell s_\ell\lambda(v_\ell\cdot X_k)\le\lambda(h\cdot X_k).

Consequently Yk∈λ(0)BjkY_k\in\lambda(0)B_{j_k}. The polar sandwich (37) makes these witnesses uniformly bounded, even if the selected bases approach singularity.

Pass to a subsequence on which Yk→YY_k\to Y, every cik→cic_{ik}\to c_i, and every θik→θi\theta_{ik}\to\theta_i. This uses a fixed finite product of compact sets: all rows belong to Q×{−1,1}Q\times\{-1,1\} and all weights belong to [0,1][0,1]. No independence of the limiting rows is needed. We have

ci∈Q×{−1,1},e=∑i=1dθici,θi≥0,∑iθi≤1.c_i\in Q\times\{-1,1\},\qquad e=\sum_{i=1}^{d}\theta_i c_i,\qquad\theta_i\ge0,\qquad\sum_i\theta_i\le1.

and continuity of λ\lambda gives ci⋅Y=λ(ci⋅X)c_i\cdot Y=\lambda(c_i\cdot X) for all ii. For arbitrary h∈Ch\in C, use hk=tjkh∈Cjkh_k=t_{j_k}h\in C_{j_k} in the feasibility inequality and pass to the limit. It follows that

∣h⋅Y∣≤λ(h⋅X)(h∈C).(41)|h\cdot Y|\le\lambda(h\cdot X)\qquad(h\in C). \tag*{(41)}

The subsequence has already been fixed; this argument proves (41) for all hh with the same YY.

Since e∈∂Ce\in\partial C, its gauge is one. Therefore

1=pC(e)≤∑i=1dθipC(ci)≤∑i=1dθi≤1.1=p_C(e)\le\sum_{i=1}^{d}\theta_i p_C(c_i)\le\sum_{i=1}^{d}\theta_i\le1.

Thus ∑iθi=1\sum_i\theta_i=1: none of the limiting mass remains at the simplex vertex 0.

Comparison with every positive representation

To prove Proposition 7.1, fix its XX and nonzero ξ\xi, and apply the preceding construction to e=ξ/T∈∂Ce=\xi/T\in\partial C. Set αi=Tθi\alpha_i=T\theta_i and discard any zero weights if desired. This gives at most dd rows and proves (33) and (34). For any finite competing representation ξ=∑jβjhj\xi=\sum_j\beta_jh_j, βj≥0\beta_j\ge0, hj∈Ch_j\in C, the common witness gives

∑iαiλ(ci⋅X)=ξ⋅Y=∑jβjhj⋅Y≤∑jβjλ(hj⋅X).\sum_i\alpha_i\lambda(c_i\cdot X)=\xi\cdot Y=\sum_j\beta_jh_j\cdot Y\le\sum_j\beta_j\lambda(h_j\cdot X).

This proves (35) and completes the proof of Proposition 7.1. In particular, the conclusion holds for each prescribed XX and ξ\xi, including directions for which every limiting representation is degenerate.

From endpoint costs to metric medians

We now convert the optimal representations into a geometric property of the polar body. A metric median of three points in a normed space is a point that lies between every pair: if the points are x0,x1,x2x_0,x_1,x_2, a median mm satisfies

∥xi−xj∥=∥xi−m∥+∥m−xj∥(0≤i<j≤2).\lVert x_i-x_j\rVert=\lVert x_i-m\rVert+\lVert m-x_j\rVert\qquad(0\le i<j\le2).

The median need not be unique. Its existence will imply the ball intersection property that characterizes linear Hanner bodies.

Proposition 8.1 (Medians of an equality body). Let K⊂RnK\subset\mathbb{R}^n, n≥1n\ge1, be an origin-symmetric convex body with P(K)=4n/n!P(K)=4^n/n!, and put Q=K∘Q=K^\circ. Every triple of points in (Rn,pQ)(\mathbb{R}^n,p_Q) has a metric median.

Proof. Write ∥⋅∥=pQ\lVert\cdot\rVert=p_Q. Translation reduces the triple to 0,x,y0,x,y. If two points coincide, that repeated point is a median, so assume that 0,x,y0,x,y are distinct. Set

p1=∥x∥,p2=∥y∥,p=∥x−y∥,s=p1+p2,a=p1−p2.p_1=\lVert x\rVert,\qquad p_2=\lVert y\rVert,\qquad p=\lVert x-y\rVert,\qquad s=p_1+p_2,\qquad a=p_1-p_2.

All three distances are positive. The triangle and reverse triangle inequalities give s≥p≥∣a∣s\ge p\ge\lvert a\rvert. Define

A+=p+a2,A−=p−a2,c=s−p2.(42)A_+=\frac{p+a}{2},\qquad A_-=\frac{p-a}{2},\qquad c=\frac{s-p}{2}. \tag*{(42)}

These numbers are nonnegative, with A++A−=pA_++A_-=p, c+A+=p1c+A_+=p_1, and c+A−=p2c+A_-=p_2.

The compact set of aggregate directions. Let

D0={r+−r−:r++r−=x−y,∥r+∥≤A+,∥r−∥≤A−}.(43)D_0=\{r_+-r_-:r_++r_-=x-y,\quad\lVert r_+\rVert\le A_+,\quad\lVert r_-\rVert\le A_-\}. \tag*{(43)}

Our goal is to prove x+y∈D0+(s−p)Qx+y\in D_0+(s-p)Q. Such membership supplies r+,r−r_+,r_- for which m=x−r+=y+r−m=x-r_+=y+r_- has distances to 0,x,y0,x,y bounded by c,A+,A−c,A_+,A_-, respectively; the triangle inequalities then force the three median equalities. We shall prove this membership by estimating support functions in each direction.

The set D0D_0 is compact and convex: it is the linear image of the intersection of A+Q×A−QA_+Q\times A_-Q with the affine subspace r++r−=x−yr_++r_-=x-y. It is nonempty, since r±=(A±/p)(x−y)r_\pm=(A_\pm/p)(x-y) satisfy the constraints. This also covers A+=0A_+=0 or A−=0A_-=0.

Endpoint comparison. Fix u∈int⁡Ku\in\operatorname{int}K and let Xδ=(δu,1−δ)X_\delta=(\delta u,1-\delta), 0<δ<10<\delta<1. The lifted body B=K⊕1[−1,1]B=K\oplus_1[-1,1] contains XδX_\delta in its interior because

pB(Xδ)=δpK(u)+1−δ<1.p_B(X_\delta)=\delta p_K(u)+1-\delta<1.

Apply Proposition 7.1 with this XδX_\delta and (q,a)=(x−y,p1−p2)(q,a)=(x-y,p_1-p_2). Its total weight is T=max⁡{∥q∥,∣a∣}=pT=\max\{\lVert q\rVert,\lvert a\rvert\}=p. We obtain rows (bi,σi)∈Q×{−1,1}(b_i,\sigma_i)\in Q\times\{-1,1\} and weights αi≥0\alpha_i\ge0 satisfying (33) and (35). Their aggregates

r±=∑σi=±1αibir_\pm=\sum_{\sigma_i=\pm1}\alpha_i b_i

have r++r−=x−yr_++r_-=x-y. Since ∑σi=±1αi=(p±a)/2=A±\sum_{\sigma_i=\pm1}\alpha_i=(p\pm a)/2=A_\pm, they satisfy ∥r±∥≤A±\lVert r_\pm\rVert\le A_\pm and hence r+−r−∈D0r_+-r_-\in D_0. No choice of these representations across different uu or δ\delta is required.

For b∈Qb \in Q and σ=±1\sigma= \pm1, evenness of λ\lambda gives

λ(⟨(b,σ),Xδ⟩)=λ(1−δ(1−σ⟨b,u⟩)).\lambda(\langle(b,\sigma),X_\delta\rangle)=\lambda(1-\delta(1-\sigma\langle b,u\rangle)).

The factors 1−σ⟨b,u⟩1-\sigma\langle b,u\rangle belong to [1−pK(u),1+pK(u)][1-p_K(u),1+p_K(u)], a compact subinterval of (0,∞)(0,\infty). Lemma 3.2 therefore gives an error εu(δ)→0\varepsilon_u(\delta)\to0 such that

∣λ(⟨(b,σ),Xδ⟩)λ(1−δ)−1+σ⟨b,u⟩∣≤εu(δ)(b∈Q, σ=±1).(44)\left|\frac{\lambda(\langle(b,\sigma),X_\delta\rangle)}{\lambda(1-\delta)}-1+\sigma\langle b,u\rangle\right|\leq\varepsilon_u(\delta)\quad(b\in Q,\ \sigma=\pm1). \tag*{(44)}

In particular, the normalized cost of the optimal representation is at least

p−⟨r+−r−,u⟩−pεu(δ)≥p−hD0(u)−pεu(δ).p-\langle r_+-r_-,u\rangle-p\varepsilon_u(\delta)\geq p-h_{D_0}(u)-p\varepsilon_u(\delta).

Compare this cost, using (35), with the two-term representation

(x−y,a)=p1(x/p1,1)+p2(−y/p2,−1).(x-y,a)=p_1(x/p_1,1)+p_2(-y/p_2,-1).

Both rows belong to Q×{−1,1}Q\times\{-1,1\}. By (44), their normalized cost tends to s−⟨x+y,u⟩s-\langle x+y,u\rangle. Letting δ↓0\delta\downarrow0 yields

⟨x+y,u⟩≤hD0(u)+s−p(u∈int⁡K).(45)\langle x+y,u\rangle\leq h_{D_0}(u)+s-p\quad(u\in\operatorname{int}K). \tag*{(45)}

The error was uniform only over b,σb,\sigma for the fixed uu; that is all this limit requires.

Separation and the median. Continuity extends (45) to u∈Ku\in K. For any nonzero v∈Rnv\in\mathbb{R}^n, the point u=v/hQ(v)u=v/h_Q(v) lies on ∂K\partial K, by (2). Homogeneity of support functions gives

⟨x+y,v⟩≤hD0(v)+(s−p)hQ(v).\langle x+y,v\rangle\leq h_{D_0}(v)+(s-p)h_Q(v).

The same inequality holds at v=0v=0. The right side is the support function of the compact convex set D0+(s−p)QD_0+(s-p)Q, with the convention 0Q={0}0Q=\{0\}. Convex separation therefore implies

x+y∈D0+(s−p)Q.(46)x+y\in D_0+(s-p)Q. \tag*{(46)}

Choose the corresponding pair r+,r−r_+,r_- in (43), and put

m=x−r+=y+r−.m=x-r_+=y+r_-.

Then 2m=x+y−(r+−r−)2m=x+y-(r_+-r_-), so (46) gives

∥m∥≤c,∥x−m∥≤A+,∥y−m∥≤A−.\lVert m\rVert\leq c,\qquad\lVert x-m\rVert\leq A_+,\qquad\lVert y-m\rVert\leq A_-.

The triangle inequality forces all three bounds to be equalities:

p1≤∥m∥+∥x−m∥≤c+A+=p1,p_1\leq\lVert m\rVert+\lVert x-m\rVert\leq c+A_+=p_1,
p2≤∥m∥+∥y−m∥≤c+A−=p2.p_2\leq\lVert m\rVert+\lVert y-m\rVert\leq c+A_-=p_2.

Thus mm lies between 0,x0,x and between 0,y0,y, and

∥x−m∥+∥m−y∥=A++A−=p=∥x−y∥.\lVert x-m\rVert+\lVert m-y\rVert=A_++A_-=p=\lVert x-y\rVert.

This proves the median property. The argument permits c=0c=0 or either A±=0A_\pm=0, so saturated triangle inequalities require no limiting argument.

Three balls and the equality classification

A normed space has the three-ball intersection property, also called the 3.23.2 intersection property, if any three closed balls that intersect pairwise have a common point. Their radii may differ. The equivalence between this intersection property and the metric-median condition is Lima’s classical norm-additive decomposition criterion [19], Theorem 1.2; see also Hansen–Lima [13], Theorem 3.6(4). We include the short direction needed here, with arbitrary radii.

Lemma 8.2 (From medians to three balls). Suppose every triple of points in a real normed space has a metric median. Then the space has the three-ball intersection property, including balls of radius zero.

Proof. Consider pairwise-intersecting balls with centers x1x_1, x2x_2, x3x_3 and radii r1r_1, r2r_2, r3≥0r_3 \ge0. Let mm be a metric median of their centers, and set di=∥m−xi∥d_i = \lVert m - x_i\rVert. Pairwise intersection and the median property imply

di+dj=∥xi−xj∥≤ri+rj(i≠j).d_i + d_j = \lVert x_i - x_j\rVert\le r_i + r_j \qquad(i \ne j).

If every di≤rid_i \le r_i, the median is a common point. Otherwise, after relabeling, t=d1−r1>0t = d_1 - r_1 > 0. For j=2,3j = 2, 3 the displayed inequalities give dj+t≤rjd_j + t \le r_j, so no other ball excludes mm. Since 0<t≤d10 < t \le d_1, the point

z=m+td1(x1−m)z = m + \frac{t}{d_1}(x_1 - m)

is well defined and lies on the segment from mm to x1x_1. It satisfies

∥z−x1∥=d1−t=r1,\lVert z - x_1\rVert= d_1 - t = r_1,
∥z−xj∥≤∥z−m∥+dj=t+dj≤rj(j=2,3).\lVert z - x_j\rVert\le\lVert z - m\rVert+ d_j = t + d_j \le r_j \qquad(j = 2,3).

Thus zz lies in all three balls. This includes r1=0r_1 = 0, when z=x1z = x_1.

The required structure theorem is due to Hansen and Lima. In their terminology the balls are closed and have arbitrary radii [13], Section 1.

Theorem 8.3 (Hansen–Lima). A nonzero finite-dimensional real Banach space with the 3.23.2 intersection property is linearly isometric to a space obtained from the real line by finitely many ℓ1\ell_1 and ℓ∞\ell_\infty direct sums.

This is [13], Corollary 7.4. The two direct-sum norms are exactly those in (4), so their unit balls are Hanner polytopes.

Proposition 8.4 (Classification of equality). If an origin-symmetric convex body K⊂RnK \subset\mathbb{R}^n, n≥1n \ge1, satisfies P(K)=4n/n!P(K) = 4^n/n!, then it is a linear Hanner body.

Proof. Put Q=K∘Q = K^\circ. Proposition 8.1 and Lemma 8.2 give the 3.23.2 intersection property for (Rn,pQ)(\mathbb{R}^n,p_Q). This is a real Banach space, since every finite-dimensional normed space is complete. Theorem 8.3 gives an invertible linear map TT and a Hanner polytope HH with Q=THQ = TH. By (3) and Lemma 2.2,

K=Q∘=T−TH∘K = Q^\circ= T^{-T}H^\circ

is a linear Hanner body. The converse was proved in Lemma 2.2; together with Corollary 6.2, this completes Theorem 1.1.

References

  1. [1]Bo Berndtsson. Complex integrals and Kuperberg’s proof of the Bourgain–Milman theorem. Advances in Mathematics, 388:107927, 2021. doi: 10.1016/j.aim.2021.107927. URL https://arxiv.org/abs/2008.00838.
  2. [2]Len Bos, Jean-Paul Calvi, and Norman Levenberg. On the Siciak extremal function for real compact convex sets. Arkiv för Matematik, 39:245–262, 2001. URL https://archive.ymsc.tsinghua.edu.cn/pacm_download/116/7771-11512_2006_Article_BF02384556.pdf.DOI
  3. [3]Jean Bourgain and Vitali D. Milman. New volume ratio properties for convex symmetric bodies in ℝⁿ. Inventiones Mathematicae, 88(2):319–340, 1987. doi: 10.1007/BF01388911. URL https://link.springer.com/article/10.1007/BF01388911.DOI
  4. [4]Shibing Chen, Yuanyuan Li, Dongmeng Xi, and Zhe-Feng Xu. The Mahler conjecture in three dimensions, 2026. URL https://arxiv.org/abs/2605.09334v3. Version 3, 15 June 2026.
  5. [5]Dario Cordero-Erausquin and Bo’az Klartag. Moment measures. Journal of Functional Analysis, 268(12):3834–3866, 2015. doi: 10.1016/j.jfa.2015.04.001. URL https://doi.org/10.1016/j.jfa.2015.04.001.
  6. [6]Jean-Pierre Demailly. Monge–Ampère operators, Lelong numbers and intersection theory. In Vincenzo Ancona and Alessandro Silva, editors, Complex Analysis and Geometry, University Series in Mathematics, pages 115–193. Plenum Press, New York, 1993. doi: 10.1007/978-1-4757-9771-8_4. URL https://link.springer.com/chapter/10.1007/978-1-4757-9771-8_4.DOI
  7. [7]Matthieu Fradelizi and Mathieu Meyer. Some functional inverse Santaló inequalities. Advances in Mathematics, 218(5):1430–1452, 2008. doi: 10.1016/j.aim.2008.03.013. URL https://doi.org/10.1016/j.aim.2008.03.013.
  8. [8]Matthieu Fradelizi, Nathael Gozlan, and Simon Zugmeyer. Transport proofs of some functional inverse Santaló inequalities, 2021. arXiv:2109.00871v3. Manuscript PDF.arxiv.org/abs/2109.00871
  9. [9]Matthieu Fradelizi, Alfredo Hubard, Mathieu Meyer, Edgardo Roldán-Pensado, and Artem Zvavitch. Equipartitions and Mahler volumes of symmetric convex bodies. American Journal of Mathematics, 144(5):1201–1219, 2022. doi: 10.1353/ajm.2022.0027. URL https://arxiv.org/abs/1904.10765.
  10. [10]Yehoram Gordon, Mathieu Meyer, and Shlomo Reisner. Zonoids with minimal volume-product—a new proof. Proceedings of the American Mathematical Society, 104(1):273–276, 1988. doi: 10.1090/S0002-9939-1988-0958082-9.DOI
  11. [11]Renan Gross. A conformal Skorokhod embedding. Electronic Communications in Probability, 24, 2019. doi: 10.1214/19-ECP272. URL https://arxiv.org/abs/1905.00852. Article no. 68, 11 pp.
  12. [12]Olof Hanner. Intersections of translates of convex bodies. Mathematica Scandinavica, 4:65–87, 1956. doi: 10.7146/math.scand.a-10456. URL https://journals.msp.org/mscand/article/view/2863.DOI
  13. [13]Allan B. Hansen and Åsvald Lima. The structure of finite dimensional Banach spaces with the 3.2. intersection property. Acta Mathematica, 146:1–23, 1981. doi: 10.1007/BF02392457. URL https://doi.org/10.1007/BF02392457.
  14. [14]Hiroshi Iriyeh and Masataka Shibata. Symmetric Mahler’s conjecture for the volume product in the 3-dimensional case. Duke Mathematical Journal, 169(6):1077–1134, 2020. doi: 10.1215/00127094-2019-0072. URL https://arxiv.org/abs/1706.01749v3.
  15. [15]Roman Karasev. Mahler’s conjecture for some hyperplane sections. Israel Journal of Mathematics, 241:795–815, 2021. doi: 10.1007/s11856-021-2114-4. URL https://arxiv.org/abs/1902.08971.
  16. [16]Jaegil Kim. Minimal volume product near Hanner polytopes. Journal of Functional Analysis, 266(4):2360–2402, 2014. doi: 10.1016/j.jfa.2013.08.008. URL https://arxiv.org/abs/1212.2544v1.
  17. [17]Jaegil Kim and Artem Zvavitch. Stability of the reverse Blaschke–Santaló inequality for unconditional convex bodies. Proceedings of the American Mathematical Society, 143(4):1705–1717, 2015. doi: 10.1090/S0002-9939-2014-12334-3. URL https://arxiv.org/abs/1302.5719v1.
  18. [18]Greg Kuperberg. From the Mahler conjecture to Gauss linking integrals. Geometric and Functional Analysis, 18(3):870–892, 2008. doi: 10.1007/s00039-008-0669-4. URL https://arxiv.org/abs/math/0610904.
  19. [19]Åsvald Lima. Intersection properties of balls in spaces of compact operators. Annales de l’Institut Fourier, 28(3):35–65, 1978. doi: 10.5802/aif.700. URL https://aif.centre-mersenne.org/articles/10.5802/aif.700/.DOI
  20. [20]Magnus Lundin. The extremal PSH for the complement of convex, symmetric subsets of ℝᴺ. Michigan Mathematical Journal, 32(2):197–201, 1985. doi: 10.1307/mmj/1029003186.DOI
  21. [21]Kurt Mahler. Ein Minimalproblem für konvexe Polygone. Mathematica (Zutphen), B7:118–127, 1938.
  22. [22]Kurt Mahler. Ein Übertragungsprinzip für konvexe Körper. Časopis pro pěstování matematiky a fysiky, 68(3–4):93–102, 1939. doi: 10.21136/CPMF.1939.109441. URL https://dml.cz/handle/10338.dmlcz/109441.
  23. [23]Mathieu Meyer. Une caractérisation volumique de certains espaces normés de dimension finie. Israel Journal of Mathematics, 55(3):317–326, 1986. doi: 10.1007/BF02765029. URL https://link.springer.com/article/10.1007/BF02765029.DOI
  24. [24]Fedor Nazarov. The Hörmander proof of the Bourgain–Milman theorem. In Bo’az Klartag, Shahar Mendelson, and Vitali D. Milman, editors, Geometric Aspects of Functional Analysis, volume 2050 of Lecture Notes in Mathematics, pages 335–343. Springer, Berlin, Heidelberg, 2012. doi: 10.1007/978-3-642-29849-3_20. URL https://doi.org/10.1007/978-3-642-29849-3_20.
  25. [25]Fedor Nazarov, Fedor Petrov, Dmitry Ryabogin, and Artem Zvavitch. A remark on the Mahler conjecture: local minimality of the unit cube. Duke Mathematical Journal, 154(3):419–430, 2010. doi: 10.1215/00127094-2010-042. URL https://arxiv.org/abs/0905.0867.
  26. [26]OpenAI. The Mahler Conjecture for General Convex Bodies. OpenAI Math Release preprint OAI:The-Mahler-Conjecture-for-General-Convex-Bodies-September-22-2026, 2026.
  27. [27]Shlomo Reisner. Random polytopes and the volume-product of symmetric convex bodies. Mathematica Scandinavica, 57:386–392, 1985. doi: 10.7146/math.scand.a-12124. URL https://journals.msp.org/mscand/article/view/1544.DOI
  28. [28]Shlomo Reisner. Zonoids with minimal volume-product. Mathematische Zeitschrift, 192(3): 339–346, 1986. doi: 10.1007/BF01164009.DOI
  29. [29]Shlomo Reisner. Minimal volume-product in Banach spaces with a 1-unconditional basis. Journal of the London Mathematical Society, 36(1):126–136, 1987. doi: 10.1112/jlms/s2-36.1.126. URL https://doi.org/10.1112/jlms/s2-36.1.126.
  30. [30]Jean Saint-Raymond. Sur le volume des corps convexes symétriques. In Séminaire d’initiation à l’analyse, 20e année, 1980–1981. Université Paris VI, Paris, 1981. Exposé no. 11.

Paper details

vlnc:W_s23t58km
Version
v1 · latest
Published
License
LicenseRef-OpenAI-public
Usage

views · 42

downloads · 2

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

  • Mathematics

unreleased internal OpenAI model

View statement
Looking for related papers…
Contents