betaFeedback

Reading v1View latest version

WorksMathematics

The Zaporozhets-Tarasov Conjecture on Mean Distances

Abstract

For a convex body K⊂RdK \subset \mathbb R^d let Δ(K)Δ(K) be the expected distance between two independent uniform points of KK, and let θ(K)θ(K) be the corresponding expectation for normalized surface measure on ∂K\partial K. The Zaporozhets-Tarasov conjecture asserts Δ(K)≤θ(K)Δ(K) \le θ(K). We prove this conjecture in case d=2d=2. In addition, we give a six-vertex convex polytope in R3\mathbb R^3 for which the reverse strict inequality holds, and obtain counterexamples in every dimension d≥3d \ge 3 by taking products with segments. Finally, we show that θ(K)>per⁡K6θ(K)> \frac{\operatorname{per}K}{6} for every planar convex body.

Introduction

Let K⊂RdK \subset\mathbb{R}^{d} be a compact convex body. The conjecture of Zaporozhets and Tarasov asks whether

Δ(K):=E∣X1−X2∣≤E∣B1−B2∣=:θ(K)\Delta(K) := \mathbb{E}|X_{1}-X_{2}| \le\mathbb{E}|B_{1}-B_{2}| =: \theta(K)

for every convex body K⊂RdK \subset\mathbb{R}^{d}, where X1,X2X_{1},X_{2} are uniform in KK with respect to the normalized Lebesgue measure and B1,B2B_{1},B_{2} are uniform on ∂K\partial K with respect to the normalized arclength measure.

Lotnikov proved the planar centrally symmetric case and an analogue for sufficiently high moments in arbitrary dimension [2]. Tokmachev proved the conjecture, in fact a convex-order statement, for planar bodies whose area and boundary centroids coincide and for a class of circumscribed polytopes [4]. Related extremal questions for Δ\Delta and θ\theta were studied in [1, 3]. In particular, Bonnet, Gusakova, Thäle and Zaporozhets proved the sharp inequalities

760<Δ(K)per⁡K<16\frac{7}{60} < \frac{\Delta(K)}{\operatorname{per} K} < \frac{1}{6}

for planar convex bodies [1], where per⁡K\operatorname{per} K stands for perimeter of KK. It is also shown that the upper bound is approached by thin rectangles converging to a doubled segment. According to Tokmachev [4 Eq. (5)], Gusakova and Zaporozhets suggested an analogous pair of inequalities for θ\theta

16<θ(K)per⁡K≤2π2.\frac{1}{6} < \frac{\theta(K)}{\operatorname{per} K} \le\frac{2}{\pi^{2}}.

Email address: [email protected] (Eric Shen) The upper bound, with equality only for the disk, was proved by Tokmachev [3]; we prove the lower bound, see Theorem 5.1.

Prior to that, we prove the Zaporozhets-Tarasov conjecture in the case d=2d=2, see Corollary 3.2. The result is mostly self-contained, in particular, it does not use the aforementioned inequalities on Δ(K)\Delta(K) from [1] and is obtained as a corollary of a more general statement, see Theorem 2.1. We also prove an analogous inequality for the second moment, see Corollary 3.3. In addition to that, we show that the conjecture is false in every dimension d≥3d \ge3, see Example 4.2, Figure 1 and Proposition 4.5.

The polytope $K_{L,\varepsilon}$ in the rescaled coordinates $(t,x,y/\varepsilon)$

Figure 1. The polytope KL,εK_{L,\varepsilon} in the rescaled coordinates (t,x,y/ε)(t,x,y/\varepsilon). One transverse width grows with tt, while the other decreases.

Definition 1.1. For a probability measure ρ\rho on R\mathbb{R}, write

D(ρ)=∬R2∣x−y∣ dρ(x)dρ(y).\mathcal{D}(\rho) = \iint_{\mathbb{R}^{2}} |x-y| \, \mathrm{d}\rho(x)\mathrm{d}\rho(y).

This is the Gini mean difference of ρ\rho.

We slightly abuse this notation throughout the text, writing D(X)\mathcal{D}(X) with XX being a random variable or a probability density. In either case, D(X)\mathcal{D}(X) is understood as D(ρX)\mathcal{D}(\rho_X), where ρX\rho_X denotes the corresponding probability measure. For basic properties of the Gini mean difference used throughout the paper we refer the reader to [6]. The main result immediately follows from a stronger directional inequality.

Theorem 1.2. Let K⊂R2K \subset\mathbb{R}^{2} be a convex body. Fix a unit vector uu. Let pup_u be the pushforward of normalized area measure on KK under x↦⟨x,u⟩x \mapsto\langle x,u\rangle, and let quq_u be the corresponding pushforward of normalized arclength measure on ∂K\partial K. Then

D(qu)≥D(pu).(1)\mathcal{D}(q_u) \ge\mathcal{D}(p_u). \tag*{(1)}

We first prove the theorem for smooth strictly convex bodies and the general case follows by approximation.

Plan. After rotating and rescaling, the projection of KK is [0,1][0,1] and

K={(x,y):0≤x≤1, g(x)≤y≤f(x)},K = \{(x,y): 0 \le x \le1,\ g(x) \le y \le f(x)\},

where ff is concave and gg is convex. Put

W=f−g,a=f′,b=g′.W = f-g,\qquad a=f',\qquad b=g'.

The projected area density is proportional to WW, while the projected boundary density is proportional to

1+a2+1+b2.\sqrt{1+a^{2}}+\sqrt{1+b^{2}}.

The proof has two steps. First, we find two densities φ\varphi and ψ\psi, nonincreasing and nondecreasing, respectively, such that the projected arclength density qq satisfies q=φ+ψ2q = \frac{\varphi+\psi}{2} while the projected area density pp is proportional to the difference of the CDFs of φ\varphi and ψ\psi. Then we apply a one-dimensional Theorem saying that an endpoint mixture is more dispersed than the normalized occupation measure between the endpoints. The only substantial step is the second one, to which we now turn.

The monotone-endpoint theorem

Theorem 2.1. Let φ\varphi be a nonincreasing probability density and ψ\psi a nondecreasing probability density on [0,1][0,1]. Let

Φ(x)=∫0xφ(s) ds,Ψ(x)=∫0xψ(s) ds.\Phi(x)=\int_0^x \varphi(s)\,\mathrm{d}s,\qquad\Psi(x)=\int_0^x \psi(s)\,\mathrm{d}s.

Put

H=Φ−Ψ,A=∫01H(x) dx.H=\Phi-\Psi,\qquad A=\int_0^1 H(x)\,\mathrm{d}x.

and assume A>0A>0. Define

p(x)=H(x)A,q(x)=φ(x)+ψ(x)2.p(x)=\frac{H(x)}{A},\qquad q(x)=\frac{\varphi(x)+\psi(x)}{2}.

Then

D(q)≥D(p).\mathcal{D}(q)\ge\mathcal{D}(p).

Since Φ\Phi is concave and Ψ\Psi is convex, Φ(x)≥x≥Ψ(x)\Phi(x)\ge x\ge\Psi(x). Thus H≥0H\ge0, and pp is indeed a probability density.

Remark 2.2. There is also a geometric reformulation. Let x(u)=Φ−1(u)x(u)=\Phi^{-1}(u) and y(u)=Ψ−1(u)y(u)=\Psi^{-1}(u). If U∼Unif⁡[0,1]U\sim\operatorname{Unif}[0,1], then x(U)∼φx(U)\sim\varphi, y(U)∼ψy(U)\sim\psi. Consider

L={(u,z):0≤u≤1, x(u)≤z≤y(u)}.L=\{(u,z):0\le u\le1,\ x(u)\le z\le y(u)\}.

Since xx is convex and yy is concave, LL is convex. Choosing one of the two endpoints of the (uniformly distributed) vertical slice [x(U),y(U)][x(U),y(U)] with equal probability gives the law qq. Schematically,

q(z) dz=12∫01(δx(u)+δy(u))(dz) du.q(z)\,\mathrm{d}z=\frac{1}{2}\int_0^1\left(\delta_{x(u)}+\delta_{y(u)}\right)(\mathrm{d}z)\,\mathrm{d}u.

On the other hand, the length of the horizontal section of LL at height zz is

∣{u:x(u)≤z≤y(u)}∣=Φ(z)−Ψ(z)=H(z).\left|\{u:x(u)\le z\le y(u)\}\right|=\Phi(z)-\Psi(z)=H(z).

Consequently, pp is the second-coordinate distribution of a uniform point of LL, that is, choose a slice with a probability proportional to its length and then a random point in it. Explicitly,

p(z) dz=∫011{x(u)≤z≤y(u)} du∫01(y(u)−x(u)) du dz.p(z)\,\mathrm{d}z=\frac{\int_0^1 \mathbf{1}_{\{x(u)\le z\le y(u)\}}\,\mathrm{d}u}{\int_0^1(y(u)-x(u))\,\mathrm{d}u}\,\mathrm{d}z.

Every decreasing probability density on [0,1][0,1] is a mixture of uniform densities on intervals [0,t][0,t], as shown by the following Proposition. This is a one-sided form of Khintchine’s representation theorem for unimodal distributions, see [5].

Proposition 2.3. Let φ\varphi be a probability density on [0,1][0,1] admitting a nonincreasing representative, which we also denote by φ\varphi. There is a random variable T∈[0,1]T \in[0,1] such that, for U∼Unif⁡[0,1]U \sim\operatorname{Unif}[0,1] independent of TT, the variable

X=TUX = TU

has density φ\varphi. Likewise, for every nondecreasing density ψ\psi, there is S∈[0,1]S \in[0,1] such that

Y=1−SV,V∼Unif⁡[0,1],Y = 1 - SV, \qquad V \sim\operatorname{Unif}[0,1],

has density ψ\psi.

Proof. As the law of TT one may take the Stieltjes mixing measure

ρ(dt)=−t dφ(t)+φ(1−)δ1(dt),\rho(dt) = -t\,d\varphi(t) + \varphi(1-)\delta_1(dt),

for which

φ(x)=∫[x,1]1t ρ(dt)for a.e. x∈(0,1).\varphi(x) = \int_{[x,1]} \frac{1}{t}\,\rho(dt) \quad\text{for a.e. } x \in(0,1).

As φ\varphi is nonincreasing, ρ\rho is nonnegative and the Stieltjes integration by parts gives

ρ([0,1])=∫01φ=1.\rho([0,1]) = \int_0^1 \varphi= 1.

Next lemma gives a lower bound on the Gini mean difference of qq in terms of its mean and auxiliary variables.

Lemma 2.4. In the notation of Theorem 2.1 and Proposition 2.3, denote

A=∫01(Φ−Ψ)=EY−EX=1−ET+ES2.A = \int_0^1 (\Phi-\Psi) = \mathbb{E}Y-\mathbb{E}X = 1-\frac{\mathbb{E}T+\mathbb{E}S}{2}.

Then D(q)≥13B:=13(1−A+32A2+Var⁡(T)+Var⁡(S))\mathcal{D}(q) \ge\frac{1}{3}B := \frac{1}{3}\left(1-A+\frac{3}{2}A^2+\operatorname{Var}(T)+\operatorname{Var}(S)\right).

Proof. Take U,U′∼Unif⁡[0,1]U,U' \sim\operatorname{Unif}[0,1] independent. Then for s,t∈[0,1]s,t \in[0,1], direct integration gives

E∣tU−sU′∣=t+s6+(t−s)23max⁡(t,s),\mathbb{E}|tU-sU'|=\frac{t+s}{6}+\frac{(t-s)^2}{3\max(t,s)},

with the second term read as zero when s=t=0s=t=0. Since max⁡(s,t)≤1\max(s,t) \le1,

E∣tU−sU′∣≥t+s6+(t−s)23.\mathbb{E}|tU-sU'|\ge\frac{t+s}{6}+\frac{(t-s)^2}{3}.

Now invoke Proposition 2.3:

D(X)=E∣TU−T′U′∣=ET,T′E(∣TU−T′U′∣∣T,T′)≥E(T+T′6+(T−T′)23).\mathcal{D}(X)=\mathbb{E}|TU-T'U'|=\mathbb{E}_{T,T'}\mathbb{E}(|TU-T'U'|\mid T,T')\ge\mathbb{E}\left(\frac{T+T'}{6}+\frac{(T-T')^2}{3}\right).

Using that TT and T′T' are independent and applying the same argument to YY one obtains

D(X)≥ET3+23Var⁡T,D(Y)≥ES3+23Var⁡S.(2)\mathcal{D}(X)\ge\frac{\mathbb{E}T}{3}+\frac{2}{3}\operatorname{Var}T,\qquad\mathcal{D}(Y)\ge\frac{\mathbb{E}S}{3}+\frac{2}{3}\operatorname{Var}S. \tag*{(2)}

For independent laws on the line with finite first moments with distribution functions Φ\Phi, Ψ\Psi,

2E∣X−Y∣−D(X)−D(Y)=2∫R(Φ−Ψ)2.2\mathbb{E}|X-Y|-\mathcal{D}(X)-\mathcal{D}(Y)=2\int_{\mathbb{R}}(\Phi-\Psi)^2.

This follows immediately from the standard CDF formulas for the three terms. Since qq is an equal-probability mixture of XX and YY, we have

D(q)=D(X)+D(Y)2+12∫01H(x)2 dx.(3)\mathcal{D}(q)=\frac{\mathcal{D}(X)+\mathcal{D}(Y)}{2}+\frac{1}{2}\int_0^1 H(x)^2\,\mathrm{d}x. \tag*{(3)}

By Cauchy–Schwarz, ∫H2≥A2\int H^2\ge A^2. Finally, using (2), and (3), we obtain

D(q)≥16(ET+ES+2Var⁡T+2Var⁡S)+12A2=13B.(4)\mathcal{D}(q)\ge\frac{1}{6}\bigl(\mathbb{E}T+\mathbb{E}S+2\operatorname{Var}T+2\operatorname{Var}S\bigr)+\frac{1}{2}A^2=\frac{1}{3}B. \tag*{(4)}

Lemma 2.5. Let pp be as in Theorem 2.1, let ZZ have density pp, and assume BB as in Lemma 2.4. Then

12Var⁡Z≤B2.12\operatorname{Var}Z\le B^2.

If PP denotes the CDF of ZZ, then by the Gini covariance identity

D(p)=4Cov⁡(Z,P(Z)).\mathcal{D}(p)=4\operatorname{Cov}(Z,P(Z)).

Since pp is absolutely continuous, P(Z)P(Z) is uniform on [0,1][0,1]. Consequently, using Cauchy–Schwarz again,

D(p)≤4Var⁡(Z)Var⁡(P(Z))=4Var⁡(Z)112=23Var⁡Z.\mathcal{D}(p)\le4\sqrt{\operatorname{Var}(Z)\operatorname{Var}(P(Z))}=4\sqrt{\operatorname{Var}(Z)\frac{1}{12}}=\frac{2}{\sqrt{3}}\sqrt{\operatorname{Var}Z}.

Hence with (4), Lemma 2.5 finishes the proof of Theorem 2.1.

Proof of Lemma 2.5. This part is rather technical, though quite elementary. For every integer m≥0m\ge0, Fubini’s theorem gives

EZm=EYm+1−EXm+1(m+1)A.\mathbb{E}Z^m=\frac{\mathbb{E}Y^{m+1}-\mathbb{E}X^{m+1}}{(m+1)A}.

Indeed,

∫01xmΦ(x) dx=∫01xmE1{X≤x} dx=E∫X1xm dx=1−EXm+1m+1,\int_0^1x^m\Phi(x)\,\mathrm{d}x=\int_0^1x^m\mathbb{E}\mathbf{1}_{\{X\le x\}}\,\mathrm{d}x=\mathbb{E}\int_X^1x^m\,\mathrm{d}x=\frac{1-\mathbb{E}X^{m+1}}{m+1},

and since p=(Φ−Ψ)/Ap=(\Phi-\Psi)/A, the analogous identity for Ψ\Psi gives the claim after subtraction. What follows is mostly algebraic identities involving TT and SS.

Recall that X=TUX=TU, and Y=1−SVY=1-SV, and introduce

z=ET−ES2,e=E[T(1−T)]+E[S(1−S)],k=E[T(1−T)]−E[S(1−S)].z=\frac{\mathbb{E}T-\mathbb{E}S}{2},\qquad e=\mathbb{E}[T(1-T)]+\mathbb{E}[S(1-S)],\qquad k=\mathbb{E}[T(1-T)]-\mathbb{E}[S(1-S)].

Using the general moment formula with m=1m=1, and using

EX2=E[T2]3,EY2=1−ES+E[S2]3,\mathbb{E}X^2=\frac{\mathbb{E}[T^2]}{3},\qquad\mathbb{E}Y^2=1-\mathbb{E}S+\frac{\mathbb{E}[S^2]}{3},

we obtain

EZ=1−ES+13(E[S2]−E[T2])2A=12+z+k6A.\mathbb{E}Z=\frac{1-\mathbb{E}S+\frac{1}{3}\left(\mathbb{E}[S^2]-\mathbb{E}[T^2]\right)}{2A}=\frac{1}{2}+\frac{z+k}{6A}.

Similarly, using the moment formula with m=2m=2,

EX3=E[T3]4,\mathbb{E}X^3=\frac{\mathbb{E}[T^3]}{4},

while

EY3=E(1−SV)3=1−32ES+E[S2]−14E[S3].\mathbb{E}Y^3=\mathbb{E}(1-SV)^3=1-\frac{3}{2}\mathbb{E}S+\mathbb{E}[S^2]-\frac{1}{4}\mathbb{E}[S^3].

Hence

EZ2=1−32ES+E[S2]−14(E[S3]+E[T3])3A.\mathbb{E}Z^2=\frac{1-\frac{3}{2}\mathbb{E}S+\mathbb{E}[S^2]-\frac{1}{4}\left(\mathbb{E}[S^3]+\mathbb{E}[T^3]\right)}{3A}.

The numerator can be regrouped as

1−32ES+E[S2]−14(E[S3]+E[T3])=A+z+k2−14(E[T(1−T)2]+E[S(1−S)2]).1-\frac{3}{2}\mathbb{E}S+\mathbb{E}[S^2]-\frac{1}{4}\left(\mathbb{E}[S^3]+\mathbb{E}[T^3]\right)=A+\frac{z+k}{2}-\frac{1}{4}\left(\mathbb{E}[T(1-T)^2]+\mathbb{E}[S(1-S)^2]\right).

Consequently,

EZ2=13+z+k6A−E[T(1−T)2]+E[S(1−S)2]12A.\mathbb{E}Z^2=\frac{1}{3}+\frac{z+k}{6A}-\frac{\mathbb{E}[T(1-T)^2]+\mathbb{E}[S(1-S)^2]}{12A}.

Subtracting (EZ)2(\mathbb{E}Z)^2 gives

12Var⁡Z=12EZ2−12(EZ)2=1−E[T(1−T)2]+E[S(1−S)2]A−(z+k)23A2.12\operatorname{Var} Z=12\mathbb{E}Z^2-12(\mathbb{E}Z)^2=1-\frac{\mathbb{E}[T(1-T)^2]+\mathbb{E}[S(1-S)^2]}{A}-\frac{(z+k)^2}{3A^2}.

We now estimate the two expectations which occur with a minus sign. By Cauchy–Schwarz,

(E[T(1−T)])2≤ETE[T(1−T)2].\left(\mathbb{E}[T(1-T)]\right)^2\leq\mathbb{E}T\mathbb{E}[T(1-T)^2].

Since

E[T(1−T)]=e+k2,ET=1−A+z,\mathbb{E}[T(1-T)]=\frac{e+k}{2},\qquad\mathbb{E}T=1-A+z,

this gives

E[T(1−T)2]≥(e+k)24(1−A+z).\mathbb{E}[T(1-T)^2]\geq\frac{(e+k)^2}{4(1-A+z)}.

Likewise,

E[S(1−S)2]≥(e−k)24(1−A−z).\mathbb{E}[S(1-S)^2]\geq\frac{(e-k)^2}{4(1-A-z)}.

Substitution into the preceding formula yields

12Var⁡Z≤1−Λ,12\operatorname{Var} Z\leq1-\Lambda,

where

Λ=(e+k)24A(1−A+z)+(e−k)24A(1−A−z)+(z+k)23A2.\Lambda=\frac{(e+k)^2}{4A(1-A+z)}+\frac{(e-k)^2}{4A(1-A-z)}+\frac{(z+k)^2}{3A^2}.

We claim that

Λ≥2(1−B),\Lambda\geq2(1-B),

where B=1−A+32A2+Var⁡(T)+Var⁡(S)=1+A−12A2−2z2−eB = 1 - A + \frac{3}{2}A^2 + \operatorname{Var}(T) + \operatorname{Var}(S) = 1 + A - \frac{1}{2}A^2 - 2z^2 - e.

Once this is proved, the result follows immediately. Indeed, B≥0B \ge0 from its original definition, and therefore

12Var⁡Z≤1−Λ≤2B−1≤B2.12\operatorname{Var} Z \le1 - \Lambda\le2B - 1 \le B^2.

It remains to prove the claim. Since ET=1−A+z\mathbb{E}T = 1 - A + z and ES=1−A−z\mathbb{E}S = 1 - A - z lie in [0,1][0,1],

0≤z2≤min⁡{A2,(1−A)2}.0 \le z^2 \le\min\{A^2,(1-A)^2\}.

Using the latter formula for BB, we obtain

Λ−2(1−B)=(e+k)24A(1−A+z)+(e−k)24A(1−A−z)+(z+k)23A2+2A−A2−4z2−2e.\Lambda- 2(1-B) = \frac{(e+k)^2}{4A(1-A+z)} + \frac{(e-k)^2}{4A(1-A-z)} + \frac{(z+k)^2}{3A^2} + 2A - A^2 - 4z^2 - 2e.

This is a convex quadratic polynomial in kk and ee, which we estimate from below in a standard way. Recall 1−A+z=ET1-A+z = \mathbb{E}T, 1−A−z=ES1-A-z = \mathbb{E}S and, to simplify the analysis, introduce new variables e=x+ye=x+y, k=x−yk=x-y. Then

Λ−2(1−B)=F(x,y)=x2AET+y2AES+(z+x−y)23A2+2A−A2−4z2−2x−2y.\Lambda- 2(1-B) = F(x,y) = \frac{x^2}{A\mathbb{E}T} + \frac{y^2}{A\mathbb{E}S} + \frac{(z+x-y)^2}{3A^2} + 2A - A^2 - 4z^2 - 2x - 2y.

Minimizing that in x,yx,y gives

F(x,y)≥A3(A+2)−(4A−1)z2A(A+2).F(x,y) \ge\frac{A^3(A+2) - (4A-1)z^2}{A(A+2)}.

Thus it only remains to verify that the numerator is nonnegative. If 0<A≤140 < A \le\frac{1}{4}, then 4A−1≤04A-1 \le0, so

A3(A+2)−(4A−1)z2≥A3(A+2)≥0.A^3(A+2) - (4A-1)z^2 \ge A^3(A+2) \ge0.

If 14≤A≤12\frac{1}{4} \le A \le\frac{1}{2}, then z2≤A2z^2 \le A^2 and

A3(A+2)−(4A−1)z2≥A3(A+2)−(4A−1)A2=A2(1−A)2≥0.A^3(A+2) - (4A-1)z^2 \ge A^3(A+2) - (4A-1)A^2 = A^2(1-A)^2 \ge0.

Finally, if 12≤A<1\frac{1}{2} \le A < 1, then z2≤(1−A)2z^2 \le(1-A)^2 and

A3(A+2)−(4A−1)z2≥A3(A+2)−(4A−1)(1−A)2=(A2−A)2+(4A−1)(2A−1)≥0.\begin{aligned} A^3(A+2) - (4A-1)z^2 &\ge A^3(A+2) - (4A-1)(1-A)^2 \\ &= (A^2-A)^2 + (4A-1)(2A-1) \ge0. \end{aligned}

Therefore Λ−2(1−B)≥0\Lambda- 2(1-B) \ge0, and hence 12Var⁡Z≤B212\operatorname{Var} Z \le B^2.

The analogous result for the second moment is immediate.

Corollary 2.6. In the setting of Theorem 2.1, let QQ be a random variable with density qq, and let ZZ, as before, have density pp. Then Var⁡(Q)≥Var⁡(Z)\operatorname{Var}(Q) \ge\operatorname{Var}(Z).

Proof. Indeed, we have 13B≤D(q)≤23Var⁡(Q)\frac{1}{3}B \le D(q) \le\frac{2}{\sqrt{3}}\sqrt{\operatorname{Var}(Q)}, therefore 12Var⁡(Q)≥B212\operatorname{Var}(Q) \ge B^2. At the same time 12Var⁡(Z)≤B212\operatorname{Var}(Z) \le B^2, by Lemma 2.5.

Proof of Theorem 1.2

Assume for now that KK is smooth and strictly convex. Fix a direction, normalize its projection to [0,1][0,1], and use the notation from the plan:

W=f−g,a=f′,b=g′.W=f-g,\qquad a=f',\qquad b=g'.

The functions aa and bb are respectively nonincreasing and nondecreasing, while

W′=a−b,W(0)=W(1)=0.W'=a-b,\qquad W(0)=W(1)=0.

The normalized projected area density is

p(x)=W(x)A,A=∫01W(x) dx.(5)p(x)=\frac{W(x)}{\mathcal{A}},\qquad\mathcal{A}=\int_{0}^{1}W(x)\,\mathrm{d}x. \tag*{(5)}

The unnormalized projected arclength boundary density is

r(x)=1+a(x)2+1+b(x)2.r(x)=\sqrt{1+a(x)^2}+\sqrt{1+b(x)^2}.

Let

R=∫01r(x) dx.R=\int_{0}^{1}r(x)\,\mathrm{d}x.

so that the normalized projected boundary density is q(x)=r(x)Rq(x)=\frac{r(x)}{R}. Define

φ(x)=1+a(x)2+a(x)+1+b(x)2−b(x)R\varphi(x)=\frac{\sqrt{1+a(x)^2}+a(x)+\sqrt{1+b(x)^2}-b(x)}{R}

and

ψ(x)=1+a(x)2−a(x)+1+b(x)2+b(x)R.\psi(x)=\frac{\sqrt{1+a(x)^2}-a(x)+\sqrt{1+b(x)^2}+b(x)}{R}.

Clearly,

q=φ+ψ2.q=\frac{\varphi+\psi}{2}.

Since aa is nonincreasing and bb is nondecreasing, it follows that φ\varphi is nonincreasing and ψ\psi is nondecreasing. Both are nonnegative. Moreover,

∫01(a−b) dx=W(1)−W(0)=0,\int_{0}^{1}(a-b)\,\mathrm{d}x=W(1)-W(0)=0,

and hence

∫01φ(x) dx=1R∫01(r(x)+a(x)−b(x)) dx=1,\int_{0}^{1}\varphi(x)\,\mathrm{d}x=\frac{1}{R}\int_{0}^{1}\bigl(r(x)+a(x)-b(x)\bigr)\,\mathrm{d}x=1,

and similarly

∫01ψ(x) dx=1.\int_{0}^{1}\psi(x)\,\mathrm{d}x=1.

Thus φ\varphi and ψ\psi are probability densities of the type appearing in Theorem 2.1. Let Φ\Phi and Ψ\Psi be their distribution functions. Then

Φ(x)−Ψ(x)=2R∫0x(a(s)−b(s)) ds=2W(x)R.\Phi(x)-\Psi(x)=\frac{2}{R}\int_{0}^{x}\bigl(a(s)-b(s)\bigr)\,\mathrm{d}s=\frac{2W(x)}{R}.

Consequently,

∫01(Φ(x)−Ψ(x)) dx=2AR,\int_{0}^{1} \left(\Phi(x)-\Psi(x)\right)\,dx=\frac{2\mathcal{A}}{R},

and therefore

Φ(x)−Ψ(x)∫01(Φ−Ψ)=W(x)A=p(x).\frac{\Phi(x)-\Psi(x)}{\int_{0}^{1}(\Phi-\Psi)}=\frac{W(x)}{\mathcal{A}}=p(x).

Since q=(φ+ψ)/2q=(\varphi+\psi)/2, Theorem 2.1 gives

D(q)≥D(p).\mathcal{D}(q)\geq\mathcal{D}(p).

This proves the directional inequality (1.1) for every smooth strictly convex planar body.

Let KnK_n be smooth strictly convex bodies converging to KK in the Hausdorff metric. Then the normalized area measures on KnK_n converge weakly to normalized area measure on KK, see e.g. [8]. Likewise, the normalized arclength measures on ∂Kn\partial K_n converge weakly to normalized arclength measure on ∂K\partial K; this follows from the weak continuity of curvature measures under Hausdorff convergence, see e.g. [7, 8]. Consequently, their pushforwards under x↦⟨x,u⟩x\mapsto\langle x,u\rangle converge weakly as well. Since all these measures are supported in a common compact interval, the Gini functionals converge, and passing to the limit proves the result for KK.

Remark 3.1. One can avoid the approximation by keeping the vertical boundary faces as atoms at the endpoints of the projection interval. The limiting argument is used here only to keep the main exposition lighter.

Corollary 3.2. For every planar convex body KK,

E∣X1−X2∣≤E∣B1−B2∣,\mathbb{E}|X_1-X_2|\leq\mathbb{E}|B_1-B_2|,

where X1,X2X_1,X_2 are uniform in KK and B1,B2B_1,B_2 are uniform on ∂K\partial K.

Proof. Indeed, for normalized angular measure σˉ\bar{\sigma} on S1S^1,

∣z∣=π2∫S1∣⟨z,u⟩∣ dσˉ(u).(6)|z|=\frac{\pi}{2}\int_{S^1}|\langle z,u\rangle|\,d\bar{\sigma}(u). \tag*{(6)}

See e.g. [8]. Integrating (3.2) against the two-point area and boundary distributions, and then using (1.1), gives the result.

Corollary 3.3. For every planar convex body KK,

E∣X1−X2∣2≤E∣B1−B2∣2,\mathbb{E}|X_1-X_2|^2\leq\mathbb{E}|B_1-B_2|^2,

where X1,X2X_1,X_2 are uniform in KK and B1,B2B_1,B_2 are uniform on ∂K\partial K.

Proof. For a smooth strictly convex body, the proof of Theorem 1.2 shows that, for every u∈S1u\in S^1, the projected area and boundary densities are obtained from the same pair of decreasing and increasing densities as in Theorem 2.1. Corollary 2.6 gives

Var⁡⟨X1,u⟩≤Var⁡⟨B1,u⟩.\operatorname{Var}\langle X_1,u\rangle\leq\operatorname{Var}\langle B_1,u\rangle.

Observe that

E⟨X1−X2,u⟩2=2Var⁡⟨X1,u⟩,\mathbb{E}\langle X_1-X_2,u\rangle^2=2\operatorname{Var}\langle X_1,u\rangle,

and similarly

E⟨B1−B2,u⟩2=2Var⁡⟨B1,u⟩.\mathbb{E}\langle B_1-B_2,u\rangle^2=2\operatorname{Var}\langle B_1,u\rangle.

Therefore

E⟨X1−X2,u⟩2≤E⟨B1−B2,u⟩2\mathbb{E}\langle X_1-X_2,u\rangle^2\leq\mathbb{E}\langle B_1-B_2,u\rangle^2

for every u∈S1u\in S^1. For normalized angular measure σˉ\bar{\sigma} on S1S^1,

∣z∣2=2∫S1⟨z,u⟩2 dσˉ(u).|z|^2=2\int_{S^1}\langle z,u\rangle^2\,d\bar{\sigma}(u).

Integrating the preceding directional inequality over uu therefore gives

E∣X1−X2∣2≤E∣B1−B2∣2.\mathbb{E}|X_1-X_2|^2\leq\mathbb{E}|B_1-B_2|^2.

The general case follows by the same smooth approximation argument as in the proof of Theorem 1.2, since variance is continuous under weak convergence for measures supported in a common compact interval.

A long-body observation

In this section, we construct a three-dimensional counterexample to the Zaporozhets-Tarasov conjecture, see Example 4.2. This example is then used to construct counterexamples in higher dimensions, see Proposition 4.5. The main observation, discussed in Remark 4.4, is that in dimension three the area and perimeter profiles of sections orthogonal to a fixed direction can be controlled almost independently. We exploit this freedom to make the projected Gini mean difference strictly larger for the volume measure than for the surface measure. A sufficiently large stretch in the projection direction then yields the required counterexample.

Lemma 4.1. Suppose a random point has the form Z=(LT,U)∈R×RmZ=(LT,U)\in\mathbb{R}\times\mathbb{R}^m, where T∈[0,1]T\in[0,1] and the support of UU has diameter at most DD. If Z1,Z2Z_1,Z_2 are independent copies, then

LE∣T1−T2∣≤E∣Z1−Z2∣≤LE∣T1−T2∣+D.L\mathbb{E}|T_1-T_2|\leq\mathbb{E}|Z_1-Z_2|\leq L\mathbb{E}|T_1-T_2|+D.

Proof. The lower bound follows by projection onto the first coordinate. For the upper bound, use

∣(L(T1−T2),U1−U2)∣≤L∣T1−T2∣+∣U1−U2∣.|(L(T_1-T_2),U_1-U_2)|\leq L|T_1-T_2|+|U_1-U_2|.

Thus, after division by LL, mean distances in a long body converge to mean absolute differences of the corresponding axial coordinates.

Example 4.2. For L>0L>0 and ε>0\varepsilon>0, define

KL,ε:={(Lt,x,y):0≤t≤1, ∣x∣≤t, ∣y∣≤ε(1−2t3)}.K_{L,\varepsilon}:=\left\{(Lt,x,y):0\leq t\leq1,\ |x|\leq t,\ |y|\leq\varepsilon\left(1-\frac{2t}{3}\right)\right\}.

This is the convex hull of (0,0,±ε)(0,0,\pm\varepsilon) and the four points (L,±1,±ε/3)(L,\pm1,\pm\varepsilon/3).

Theorem 4.3. For L=1000L=1000 and ε=1/100\varepsilon=1/100, one has

Δ(KL,ε)>θ(KL,ε).\Delta(K_{L,\varepsilon})>\theta(K_{L,\varepsilon}).

Proof. The area of the section at level tt equals 4εt(1−2t3)4\varepsilon t\left(1-\frac{2t}{3}\right). Hence the axial coordinate of a uniform interior point has density

pI(t)=185t(1−2t3)p_I(t)=\frac{18}{5}t\left(1-\frac{2t}{3}\right)

and distribution function

FI(t)=9t2−4t35.F_I(t)=\frac{9t^2-4t^3}{5}.

Using D(p)=2∫01F(t)(1−F(t)) dt\mathcal{D}(p)=2\int_0^1 F(t)(1-F(t))\,\mathrm{d}t, we obtain

D(pI)=246875=0.281142857….\mathcal{D}(p_I)=\frac{246}{875}=0.281142857\ldots.

Lemma 4.1 therefore gives

Δ(KL,ε)L≥246875.\frac{\Delta(K_{L,\varepsilon})}{L}\ge\frac{246}{875}.

We next compute the axial distribution of normalized surface measure. The two faces x=±tx=\pm t have combined area density

4ε(1−2t3)L2+1,4\varepsilon\left(1-\frac{2t}{3}\right)\sqrt{L^2+1},

while the two faces y=±ε(1−2t/3)y=\pm\varepsilon(1-2t/3) have combined area density

4tL2+4ε29.4t\sqrt{L^2+\frac{4\varepsilon^2}{9}}.

The end face t=1t=1 has area 4ε/34\varepsilon/3. Put

a=εL2+1,b=L2+4ε29−2ε3L2+1,a=\varepsilon\sqrt{L^2+1},\qquad b=\sqrt{L^2+\frac{4\varepsilon^2}{9}}-\frac{2\varepsilon}{3}\sqrt{L^2+1},
D=a+b2+ε3,α=aD,β=b2D.D=a+\frac{b}{2}+\frac{\varepsilon}{3},\qquad\alpha=\frac{a}{D},\qquad\beta=\frac{b}{2D}.

For 0≤t<10\leq t<1, the boundary axial distribution function is

F∂(t)=αt+βt2;F_{\partial}(t)=\alpha t+\beta t^2;

the remaining mass is an atom at t=1t=1. Consequently its Gini mean difference is exactly

gL,ε=α+2β3−2α23−αβ−2β25.g_{L,\varepsilon}=\alpha+\frac{2\beta}{3}-\frac{2\alpha^2}{3}-\alpha\beta-\frac{2\beta^2}{5}.

The transverse diameter of KL,εK_{L,\varepsilon} is at most 21+ε22\sqrt{1+\varepsilon^2}, so Lemma 4.1 gives

θ(KL,ε)L≤gL,ε+21+ε2L.\frac{\theta(K_{L,\varepsilon})}{L}\leq g_{L,\varepsilon}+\frac{2\sqrt{1+\varepsilon^2}}{L}.

For the values suggested in the statement, g1000,1/100=0.2692731635…g_{1000,1/100}=0.2692731635\ldots and hence

θ(K1000,1/100)1000≤0.2712732636…<0.2811428571…≤Δ(K1000,1/100)1000.\frac{\theta(K_{1000,1/100})}{1000}\leq0.2712732636\ldots<0.2811428571\ldots\leq\frac{\Delta(K_{1000,1/100})}{1000}.

The family also has a clean asymptotic criterion. As L→∞L\to\infty, the boundary axial density tends to the normalization of

t+ε(1−2t3),t+\varepsilon\left(1-\frac{2t}{3}\right),

and its Gini mean difference is

g∂(ε)=4(19ε2+33ε+9)15(4ε+3)2.g_{\partial}(\varepsilon)=\frac{4(19\varepsilon^2+33\varepsilon+9)}{15(4\varepsilon+3)^2}.

Moreover,

g∂(ε)−246875=2(746ε2+2694ε−171)2625(4ε+3)2.g_{\partial}(\varepsilon)-\frac{246}{875}=\frac{2(746\varepsilon^2+2694\varepsilon-171)}{2625(4\varepsilon+3)^2}.

Thus every

0<ε<ε∗:=−1347+45959746=0.062396288…0<\varepsilon<\varepsilon_*:=\frac{-1347+45\sqrt{959}}{746}=0.062396288\ldots

produces counterexamples for all sufficiently large LL.

Remark 4.4. Consider a long body with rectangular transverse sections

Qt=∏i=1d−1[−ai(t),ai(t)].(7)Q_t=\prod_{i=1}^{d-1}[-a_i(t),a_i(t)]. \tag*{(7)}

Up to constant factors, the axial density of volume measure is

∏i=1d−1ai(t),\prod_{i=1}^{d-1} a_i(t),

whereas the lateral surface density is

∑i=1d−1∏j≠iaj(t).\sum_{i=1}^{d-1}\prod_{j\ne i} a_j(t).

In dimension three this is the difference between a product a1a2a_1a_2 and a sum a1+a2a_1+a_2. Choosing one width increasing and one decreasing can make the normalized surface density more concentrated along the long axis than the normalized volume density. In dimension two there is only one transverse width: the second expression is the empty product.

Proposition 4.5. There are counterexamples in every dimension d≥3d\ge3.

Proof. The case d=3d=3 was established above. Let d=3+md=3+m, where m≥1m\ge1, and put

KL:=KL,1/100,CM:=[−M,M]m,K~L,M:=KL×CM.K_L:=K_{L,1/100},\qquad C_M:=[-M,M]^m,\qquad\widetilde{K}_{L,M}:=K_L\times C_M.

Since 1/100<ε∗1/100<\varepsilon_*, the preceding asymptotic computation shows that there exists c>0c>0 such that, for all sufficiently large LL,

Δ(KL)−θ(KL)≥cL.\Delta(K_L)-\theta(K_L)\ge cL.

The boundary of K~L,M\widetilde{K}_{L,M}, up to a set of surface measure zero, is the disjoint union

(∂KL×CM)∪(KL×∂CM).(\partial K_L\times C_M)\cup(K_L\times\partial C_M).

The first part has surface measure

H2(∂KL)(2M)m,\mathcal{H}^2(\partial K_L)(2M)^m,

whereas the second has surface measure

∣KL∣Hm−1(∂CM)=2m∣KL∣(2M)m−1.|K_L|\mathcal{H}^{m-1}(\partial C_M)=2m|K_L|(2M)^{m-1}.

Since both ∣KL∣|K_L| and H2(∂KL)\mathcal{H}^2(\partial K_L) are of order LL, the relative surface mass of KL×∂CMK_L\times\partial C_M is

ηL,M=O(M−1),\eta_{L,M}=O(M^{-1}),

uniformly for sufficiently large LL.

Let μL,M\mu_{L,M} be normalized surface measure on ∂KL×CM\partial K_L\times C_M. It is the product of normalized surface measure on ∂KL\partial K_L and normalized volume measure on CMC_M. If Z1,Z2Z_1,Z_2 are independent with law μL,M\mu_{L,M}, then

E∣Z1−Z2∣≤θ(KL)+2Mm.(8)\mathbb{E}|Z_1-Z_2|\le\theta(K_L)+2M\sqrt{m}. \tag*{(8)}

The diameter of K~L,M\widetilde{K}_{L,M} is O(L+M)O(L+M). Replacing normalized surface measure on ∂K~L,M\partial\widetilde{K}_{L,M} by μL,M\mu_{L,M} changes the corresponding mean distance by at most

O(ηL,M(L+M))=O(LM+1).O(\eta_{L,M}(L+M))=O\left(\frac{L}{M}+1\right).

Consequently,

θ(K~L,M)≤θ(KL)+O(M)+O(LM+1).\theta(\widetilde{K}_{L,M})\leq\theta(K_L)+O(M)+O\left(\frac{L}{M}+1\right).

Choose M=M(L)M=M(L) so that

M⟶∞,ML⟶0(L→∞),M\longrightarrow\infty,\qquad\frac{M}{L}\longrightarrow0\qquad(L\to\infty),

then the error terms above are o(L)o(L). On the other hand, projection onto the first three coordinates does not increase distances, so

Δ(K~L,M)≥Δ(KL).\Delta(\widetilde{K}_{L,M})\geq\Delta(K_L).

Combining all inequalities above, we obtain

Δ(K~L,M)−θ(K~L,M)≥cL−o(L)>0.\Delta(\widetilde{K}_{L,M})-\theta(\widetilde{K}_{L,M})\geq cL-o(L)>0.

for all sufficiently large LL. Thus K~L,M\widetilde{K}_{L,M} is a counterexample in dimension dd.

The lower bound for boundary mean distance

We now prove the lower estimate stated in the Introduction. The argument is again a comparison of one-dimensional distributions. Here the relevant laws are obtained by choosing either one random tangent direction or the median of three random tangent directions on a moving half-boundary.

Theorem 5.1. For every planar convex body KK,

θ(K)>per⁡K6.\theta(K)>\frac{\operatorname{per} K}{6}.

The constant 1/61/6 is sharp.

Assume first that ∂K\partial K is C2C^2, strictly convex, has positive curvature, and oriented counter-clockwise. Parameterize it by the tangent angle α∈R/2πZ\alpha\in\mathbb{R}/2\pi\mathbb{Z}:

γ′(α)=τ(α)(cos⁡α,sin⁡α),ds=τ(α) dα.(9)\gamma'(\alpha)=\tau(\alpha)(\cos\alpha,\sin\alpha),\qquad ds=\tau(\alpha)\,d\alpha. \tag*{(9)}

where τ>0\tau>0 is the radius of curvature. Put

P=per⁡K=∫02πτ(α) dα,S(t)=∫tt+πτ(α) dα.P=\operatorname{per} K=\int_{0}^{2\pi}\tau(\alpha)\,d\alpha,\qquad S(t)=\int_{t}^{t+\pi}\tau(\alpha)\,d\alpha.

For fixed tt, choose a point uniformly with respect to arclength on the half-boundary whose tangent angles lie in [t,t+π][t,t+\pi]. If its tangent angle is α\alpha, put It=α−t∈[0,π]I_t=\alpha-t\in[0,\pi]. Then ItI_t has density

ft(r)=τ(t+r)S(t).(10)f_t(r)=\frac{\tau(t+r)}{S(t)}. \tag*{(10)}

Let MtM_t be the median of three independent copies of ItI_t.

Proposition 5.2. With the preceding notation,

θ(K)≥16P∫02πS(t)2Esin⁡Mt dt.(11)\theta(K) \ge\frac{1}{6P}\int_{0}^{2\pi} S(t)^2 \mathbb{E}\sin M_t\,dt. \tag*{(11)}

Proof. Fix tt and project the body onto

ut=(−sin⁡t,cos⁡t).u_t=(-\sin t,\cos t).

The boundary ∂K\partial K splits into two arcs joining the two support points with tangent vectors orthogonal to utu_t. Let Γt\Gamma_t and Γt+π\Gamma_{t+\pi} be the two boundary arcs whose tangent angles range over [t,t+π][t,t+\pi] and [t+π,t+2π][t+\pi,t+2\pi]. Their arclengths are S(t)S(t) and S(t+π)=P−S(t)S(t+\pi)=P-S(t). Let μt,μt+π\mu_t,\mu_{t+\pi} be the pushforwards of normalized arclength measures on these arcs under projection onto utu_t. Finally, let νt\nu_t be the pushforward of normalized arclength measure on the whole boundary ∂K\partial K under projection onto utu_t. Since

∂K=Γt∪Γt+π,S(t)+S(t+π)=P,\partial K=\Gamma_t\cup\Gamma_{t+\pi},\qquad S(t)+S(t+\pi)=P,

we have

νt=S(t)Pμt+S(t+π)Pμt+π.\nu_t=\frac{S(t)}{P}\mu_t+\frac{S(t+\pi)}{P}\mu_{t+\pi}.

As the Gini functional is concave under mixtures,

D(νt)≥S(t)PD(μt)+S(t+π)PD(μt+π).(12)\mathcal{D}(\nu_t)\ge\frac{S(t)}{P}\mathcal{D}(\mu_t)+\frac{S(t+\pi)}{P}\mathcal{D}(\mu_{t+\pi}). \tag*{(12)}

Now

∣z∣=12∫0π∣⟨z,ut⟩∣ dt,|z|=\frac{1}{2}\int_{0}^{\pi}|\langle z,u_t\rangle|\,dt,

and hence

θ(K)=12∫0πD(νt) dt.\theta(K)=\frac{1}{2}\int_{0}^{\pi}\mathcal{D}(\nu_t)\,dt.

Using 5.4,

θ(K)≥12P∫0π[S(t)D(μt)+S(t+π)D(μt+π)] dt.\theta(K)\ge\frac{1}{2P}\int_{0}^{\pi}\left[S(t)\mathcal{D}(\mu_t)+S(t+\pi)\mathcal{D}(\mu_{t+\pi})\right]\,dt.

Projection of Γt+π\Gamma_{t+\pi} onto ut+π=−utu_{t+\pi}=-u_t reflects the law obtained by projection onto utu_t. Since the Gini mean difference is invariant under reflection, the change of variable r=t+πr=t+\pi gives

θ(K)≥12P[∫0πS(t)D(μt) dt+∫π2πS(t)D(μt) dt]=12P∫02πS(t)D(μt) dt.(13)\theta(K)\ge\frac{1}{2P}\left[\int_{0}^{\pi}S(t)\mathcal{D}(\mu_t)\,dt+\int_{\pi}^{2\pi}S(t)\mathcal{D}(\mu_t)\,dt\right]=\frac{1}{2P}\int_{0}^{2\pi}S(t)\mathcal{D}(\mu_t)\,dt. \tag*{(13)}

Parametrize Γt\Gamma_t by arclength as

rt:[0,S(t)]⟶Γt,r_t:[0,S(t)]\longrightarrow\Gamma_t,

starting at the point with tangent angle tt, and put

xt(s)=⟨rt(s),ut⟩.x_t(s)=\langle r_t(s),u_t\rangle.

If αt(s)\alpha_t(s) denotes the tangent angle at rt(s)r_t(s), then

rt′(s)=(cos⁡αt(s),sin⁡αt(s))r'_t(s)=(\cos\alpha_t(s),\sin\alpha_t(s))

and therefore

xt′(s)=⟨rt′(s),ut⟩=sin⁡(αt(s)−t)≥0.x'_t(s)=\langle r'_t(s),u_t\rangle=\sin(\alpha_t(s)-t)\ge0.

Thus xtx_t is nondecreasing on [0,S(t)][0,S(t)]. Using that xtx_t is nondecreasing and applying Fubini’s theorem we get

S(t)2D(μt)=∫0S(t)∫0S(t)∣xt(s)−xt(r)∣ dr ds=2∬0≤r≤s≤S(t)xt(s)−xt(r) dr ds=2∬0≤r≤s≤S(t)∫rsxt′(z) dz dr ds=2∫0S(t)z(S(t)−z)xt′(z) dz.(14)\begin{aligned} S(t)^2\mathcal{D}(\mu_t)&=\int_0^{S(t)}\int_0^{S(t)}\lvert x_t(s)-x_t(r)\rvert\,dr\,ds=2\iint_{0\le r\le s\le S(t)}x_t(s)-x_t(r)\,dr\,ds\\ &=2\iint_{0\le r\le s\le S(t)}\int_r^s x'_t(z)\,dz\,dr\,ds=2\int_0^{S(t)}z(S(t)-z)x'_t(z)\,dz. \tag*{(14)} \end{aligned}

Observe that the median Z(2)Z_{(2)} of three independent uniform points Z1,Z2,Z3Z_1,Z_2,Z_3 on [0,S(t)][0,S(t)] has density

6z(S(t)−z)S(t)3 dz,\frac{6z(S(t)-z)}{S(t)^3}\,dz,

so

S(t)2D(μt)=S(t)33Ext′(Z(2)).S(t)^2\mathcal{D}(\mu_t)=\frac{S(t)^3}{3}\mathbb{E}x'_t(Z_{(2)}).

Since s↦αt(s)s\mapsto\alpha_t(s) is increasing on the half-boundary, taking the median commutes with change of variable, i.e.

αt(Z(2))−t=Mt\alpha_t(Z_{(2)})-t=M_t

and, by (5.1), xt′(s)=sin⁡(αt(s)−t)x'_t(s)=\sin(\alpha_t(s)-t). Hence

S(t)2D(μt)=S(t)33Esin⁡Mt.S(t)^2\mathcal{D}(\mu_t)=\frac{S(t)^3}{3}\mathbb{E}\sin M_t.

Finally, plugging this into 5.5 gives

θ(K)≥16P∫02πS(t)2Esin⁡Mt dt.(15)\theta(K)\ge\frac{1}{6P}\int_0^{2\pi}S(t)^2\mathbb{E}\sin M_t\,dt. \tag*{(15)}

Let us replace MtM_t in the integral above with ItI_t. Then

∫02πS(t)2Esin⁡It dt=∫02πS(t)∫0πτ(t+ω)sin⁡ω dω dt.\int_0^{2\pi}S(t)^2\mathbb{E}\sin I_t\,dt=\int_0^{2\pi}S(t)\int_0^\pi\tau(t+\omega)\sin\omega\,d\omega\,dt.

The inner integral is the width w(t)w(t) of KK in the direction utu_t. Hence

∫02πS(t)2Esin⁡It dt=∫02πS(t)w(t) dt=∫0π(S(t)+S(t+π))w(t) dt=P∫0πw(t) dt=P2.(16)\int_0^{2\pi}S(t)^2\mathbb{E}\sin I_t\,dt=\int_0^{2\pi}S(t)w(t)\,dt=\int_0^\pi(S(t)+S(t+\pi))w(t)\,dt=P\int_0^\pi w(t)\,dt=P^2. \tag*{(16)}

where in the last step we used Cauchy’s perimeter formula. Therefore, the following Proposition together with Proposition 5.2 would conclude the proof of Theorem 5.1 with non-strict inequality.

Proposition 5.3. For every 0≤a≤π/20 \le a \le\pi/2,

∫02πS(t)2P{Mt∈[a,π−a]} dt≥∫02πS(t)2P{It∈[a,π−a]} dt.(17)\int_{0}^{2\pi} S(t)^2 \mathbb{P}\{M_t \in[a,\pi-a]\}\,\mathrm{d}t \ge\int_{0}^{2\pi} S(t)^2 \mathbb{P}\{I_t \in[a,\pi-a]\}\,\mathrm{d}t. \tag*{(17)}

Consequently,

∫02πS(t)2Esin⁡Mt dt≥∫02πS(t)2Esin⁡It dt.(18)\int_{0}^{2\pi} S(t)^2 \mathbb{E}\sin M_t\,\mathrm{d}t \ge\int_{0}^{2\pi} S(t)^2 \mathbb{E}\sin I_t\,\mathrm{d}t. \tag*{(18)}

For a randomly chosen half-boundary, the median of three random points tends to move away from its ends. If both angular end-pieces [t,t+a][t,t+a] and [t+π−a,t+π][t+\pi-a,t+\pi] contain at most half of the arclength, this already gives the statement by pointwise comparison in tt. The difficulty is that tangent angle and arclength need not be distributed uniformly: a short angular end-piece may carry more than half of the arclength. Proof of this Proposition relies on the following Lemma.

Lemma 5.4. Let μ\mu be a finite nonnegative Borel measure on the circle of circumference 2h2h. Suppose 0<a≤b0<a\le b and a+b=ha+b=h. For x<y≤x+2hx<y\le x+2h, write μ[x,y)\mu[x,y) for the μ\mu-mass of the positively oriented half-open arc from xx to yy. Put

Φ(x,y)=xy(x−y)x+y,with Φ(0,0)=0,\Phi(x,y)=\frac{xy(x-y)}{x+y},\quad\text{with }\Phi(0,0)=0,

and define

Fa,b(μ)=∫02hΦ(μ[t,t+a),μ[t+a,t+a+b)) dt.(19)\mathcal{F}_{a,b}(\mu)=\int_{0}^{2h}\Phi\bigl(\mu[t,t+a),\mu[t+a,t+a+b)\bigr)\,\mathrm{d}t. \tag*{(19)}

Then

Fa,b(μ)≤Fb,a(μ).(20)\mathcal{F}_{a,b}(\mu)\le\mathcal{F}_{b,a}(\mu). \tag*{(20)}

Proof. For 0<r<h0<r<h, put

Lr(t)=μ[t,t+r),Rr(t)=μ[t+r,t+h),H(t)=μ[t,t+h).L_r(t)=\mu[t,t+r),\qquad R_r(t)=\mu[t+r,t+h),\qquad H(t)=\mu[t,t+h).

Thus

H(t)=Lr(t)+Rr(t),H(t)=L_r(t)+R_r(t),

and

Φ(Lr(t),Rr(t))=Lr(t)Rr(t)(1−2Rr(t)H(t)).\Phi(L_r(t),R_r(t))=L_r(t)R_r(t)\left(1-2\frac{R_r(t)}{H(t)}\right).

Whenever H(t)=0H(t)=0, we interpret Rr(t)/H(t)R_r(t)/H(t) as 00; this is harmless since then Lr(t)Rr(t)=0L_r(t)R_r(t)=0.

Since

Lr(t)Rr(t)=∫[t,t+r)∫[t+r,t+h)dμ(y) dμ(x),L_r(t)R_r(t)=\int_{[t,t+r)}\int_{[t+r,t+h)}\mathrm{d}\mu(y)\,\mathrm{d}\mu(x),

Fubini’s theorem gives

Fr,h−r(μ)=∬[∫Tr(x,y)(1−2Rr(t)H(t))dt]dμ(x) dμ(y).(21)\mathcal{F}_{r,h-r}(\mu)=\iint\left[\int_{\mathcal{T}_r(x,y)}\left(1-2\frac{R_r(t)}{H(t)}\right)\mathrm{d}t\right]\mathrm{d}\mu(x)\,\mathrm{d}\mu(y). \tag*{(21)}

where

Tr(x,y)={t:x∈[t,t+r), y∈[t+r,t+h)}.\mathcal{T}_r(x,y)=\{t:x\in[t,t+r),\ y\in[t+r,t+h)\}.

We shall call the decomposition

[t,t+h)=[t,t+r)∪[t+r,t+h)[t,t+h)=[t,t+r)\cup[t+r,t+h)

the rr-splitting of the half-arc. Thus Fa,bF_{a,b} corresponds to the aa-splitting and Fb,aF_{b,a} to the bb-splitting. Fix an ordered pair (x,y)(x,y) and let dd be the positively oriented distance from xx to yy. Choose the angular origin at xx, and represent yy by the distance d∈[0,2h)d\in[0,2h) from xx. Thus x=0x=0 and y=dy=d on the periodic lift to R\mathbb{R}.

The admissible values of tt, namely those for which 0∈[t,t+r)0\in[t,t+r) and d∈[t+r,t+h)d\in[t+r,t+h), are therefore

Tr(d)=(−r,0]∩(d−h,d−r].T_r(d)=(-r,0]\cap(d-h,d-r].

Since a+b=ha+b=h and a≤ba\leq b, one has

0≤d≤a:Ta(d)=(−a,d−a],Tb(d)=(−b,d−b],a≤d≤b:Ta(d)=(−a,0],Tb(d)=(d−h,d−b],b≤d≤h:Ta(d)=(d−h,0],Tb(d)=(d−h,0].\begin{aligned} 0\leq d\leq a &: \quad T_a(d)=(-a,d-a], \qquad T_b(d)=(-b,d-b],\\ a\leq d\leq b &: \quad T_a(d)=(-a,0], \qquad T_b(d)=(d-h,d-b],\\ b\leq d\leq h &: \quad T_a(d)=(d-h,0], \qquad T_b(d)=(d-h,0]. \end{aligned}

Hence

Tb(d)=Ta(d)−δ(d),(22)T_b(d)=T_a(d)-\delta(d), \tag*{(22)}

where

δ(d)={b−a,0≤d≤a,b−d,a≤d≤b,0,b≤d≤h.\delta(d)= \begin{cases} b-a, & 0\leq d\leq a,\\ b-d, & a\leq d\leq b,\\ 0, & b\leq d\leq h. \end{cases}

If h<d<2hh<d<2h, then Tr(d)=∅T_r(d)=\varnothing for every 0<r<h0<r<h, so there is nothing to prove. We may therefore assume 0≤d≤h0\leq d\leq h. Thus the admissible starting positions for the aa- and bb-splittings are paired by

t⟼t′=t−δ(d).t\longmapsto t'=t-\delta(d).

In other words, for a fixed t∈Ta(d)t\in T_a(d) the cut point of aa-splitting is q=t+aq=t+a, whereas for the corresponding bb-splitting it is q′=t′+b=t+b−δ(d)q'=t'+b=t+b-\delta(d). As xx and yy are fixed we write δ=δ(d)\delta=\delta(d). Note that q≤q′q\leq q', since δ≤b−a\delta\leq b-a. Set

λ=μ[t−δ,t),C=μ[t,q′),V=μ[q′,t+h−δ),η=μ[t+h−δ,t+h).\lambda=\mu[t-\delta,t),\qquad C=\mu[t,q'),\qquad V=\mu[q',t+h-\delta),\qquad\eta=\mu[t+h-\delta,t+h).

Then

Ra(t)H(t)=μ[q,t+h)μ[t,t+h)≥μ[q′,t+h)μ[t,t+h)=V+ηC+V+η≥Vλ+C+V=μ[q′,t+h−δ)μ[t−δ,t+h−δ)=Rb(t′)H(t′).\begin{aligned} \frac{R_a(t)}{H(t)} =\frac{\mu[q,t+h)}{\mu[t,t+h)} \geq\frac{\mu[q',t+h)}{\mu[t,t+h)} =\frac{V+\eta}{C+V+\eta} \\ &\geq\frac{V}{\lambda+C+V} =\frac{\mu[q',t+h-\delta)}{\mu[t-\delta,t+h-\delta)} =\frac{R_b(t')}{H(t')}. \end{aligned}

The second inequality follows by cross-multiplication:

(V+η)(λ+C+V)−V(C+V+η)=λ(V+η)+Cη≥0.(V+\eta)(\lambda+C+V)-V(C+V+\eta)=\lambda(V+\eta)+C\eta\geq0.

Consequently,

1−2Ra(t)H(t)≤1−2Rb(t′)H(t′).1-2\frac{R_a(t)}{H(t)}\leq1-2\frac{R_b(t')}{H(t')}.

We defined TrT_r so that Tr(x,y)=x+Tr(d)T_r(x,y)=x+T_r(d). From (5.14) one gets

Tb(x,y)=Ta(x,y)−δ(d).T_b(x,y)=T_a(x,y)-\delta(d).

Thus, for the fixed ordered pair (x,y)(x,y), the substitution s=t−δ(d)s=t-\delta(d) gives

∫Tb(x,y)(1−2Rb(s)H(s)) ds=∫Ta(x,y)(1−2Rb(t−δ(d))H(t−δ(d))) dt≥∫Ta(x,y)(1−2Ra(t)H(t)) dt.\int_{\mathcal{T}_{b}(x,y)}\left(1-2\frac{R_b(s)}{H(s)}\right)\,ds=\int_{\mathcal{T}_{a}(x,y)}\left(1-2\frac{R_b(t-\delta(d))}{H(t-\delta(d))}\right)\,dt\ge\int_{\mathcal{T}_{a}(x,y)}\left(1-2\frac{R_a(t)}{H(t)}\right)\,dt.

Integrating this inequality over (x,y)(x,y) in (21) gives

Fa,b(μ)≤Fb,a(μ).\mathcal{F}_{a,b}(\mu)\le\mathcal{F}_{b,a}(\mu).

Now we are ready to prove Proposition 5.3.

Proof of Proposition 5.3. Denote tangent-angle measure of the body KK by μ\mu. That is, for a Borel set AA of angles, μ(A)=H1{x∈∂K:α(x)∈A}\mu(A)=\mathcal{H}^{1}\{x\in\partial K:\alpha(x)\in A\}, where α(x)\alpha(x) is the tangent angle at xx, as in 5.1, and H1\mathcal{H}^{1} is the arclength measure. For fixed tt, let

Ft(w)=P(It<Sw)=1S(t)μ[t,t+w),S(t)=μ[t,t+π).F_t(w)=\mathbb{P}(I_t<Sw)=\frac{1}{S(t)}\mu[t,t+w),\qquad S(t)=\mu[t,t+\pi).

The CDF of MtM_t is

Gt(w)=3Ft(w)2−2Ft(w)3,G_t(w)=3F_t(w)^2-2F_t(w)^3,

so

Gt(w)−Ft(w)=Ft(w)(1−Ft(w))(2Ft(w)−1).(23)G_t(w)-F_t(w)=F_t(w)(1-F_t(w))(2F_t(w)-1). \tag*{(23)}

Put

At(w)=S(t)Ft(w)=μ[t,t+w),Bt(w)=S(t+w)Ft+w(π−w)=μ[t+w,t+π),A_t(w)=S(t)F_t(w)=\mu[t,t+w),\qquad B_t(w)=S(t+w)F_{t+w}(\pi-w)=\mu[t+w,t+\pi),

and then (23) says

S(t)2(Gt(w)−Ft(w))=Φ(At(w),Bt(w)).(24)S(t)^2(G_t(w)-F_t(w))=\Phi(A_t(w),B_t(w)). \tag*{(24)}

After evaluating (24) at w=aw=a and w=π−aw=\pi-a, the left-hand side of (17) minus the right-hand side equals

Fπ−a,a(μ)−Fa,π−a(μ),\mathcal{F}_{\pi-a,a}(\mu)-\mathcal{F}_{a,\pi-a}(\mu),

which is nonnegative by Lemma 5.4. Finally,

sin⁡w=∫0π/21[a,π−a](w)cos⁡a da,0≤w≤π.(25)\sin w=\int_{0}^{\pi/2}\mathbf{1}_{[a,\pi-a]}(w)\cos a\,da,\qquad0\le w\le\pi. \tag*{(25)}

Integrating (17) against cos⁡a da\cos a\,da proves (18).

Proof of Theorem 5.1. As we outlined before Proposition 5.3,

∫02πS(t)2Esin⁡It dt=∫02πS(t)(∫tt+πτ(α)sin⁡(α−t) dα) dt=∫02πS(t)w(t) dt,(26)\begin{aligned} \int_{0}^{2\pi}S(t)^2\mathbb{E}\sin I_t\,dt =\int_{0}^{2\pi}S(t)\left(\int_t^{t+\pi}\tau(\alpha)\sin(\alpha-t)\,d\alpha\right)\,dt \\ &=\int_{0}^{2\pi}S(t)w(t)\,dt, \tag*{(26)} \end{aligned}

where w(t)w(t) is the width of KK in direction utu_t. Indeed,

w(t)=⟨γ(t+π)−γ(t),ut⟩=∫tt+π⟨γ′(α),ut⟩ dα=∫tt+πτ(α)sin⁡(α−t) dα.w(t)=\langle\gamma(t+\pi)-\gamma(t),u_t\rangle =\int_t^{t+\pi}\langle\gamma'(\alpha),u_t\rangle\,d\alpha =\int_t^{t+\pi}\tau(\alpha)\sin(\alpha-t)\,d\alpha.

Since

S(t)+S(t+π)=P,w(t+π)=w(t),S(t) + S(t+\pi) = P,\qquad w(t+\pi) = w(t),

Cauchy’s perimeter formula gives

∫02πS(t)w(t) dt=P∫0πw(t) dt=P2.(27)\int_{0}^{2\pi} S(t)w(t)\,\mathrm{d}t = P\int_{0}^{\pi} w(t)\,\mathrm{d}t = P^2. \tag*{(27)}

Combining Propositions 5.2 and 5.3 with (5.19), we obtain

θ(K)≥P6.\theta(K) \ge\frac{P}{6}.

By smooth approximation, the same chain of inequalities holds for every planar convex body, with τ(α) dα\tau(\alpha)\,\mathrm{d}\alpha replaced by its tangent-angle measure. Thus θ(K)≥P/6\theta(K) \ge P/6 in general. It remains to exclude equality for a nondegenerate convex body. Suppose therefore that

θ(K)=P6.\theta(K) = \frac{P}{6}.

Then both inequalities supplied by Propositions 5.2 and 5.3 must be equalities.

Let μ\mu be the tangent-angle measure of ∂K\partial K. We first observe that equality in Proposition 5.3 implies that no open semicircle contains three points of supp⁡μ\operatorname{supp}\mu. Indeed, suppose that, after lifting the angles to R\mathbb{R},

0=x<z<y=d<π,x,z,y∈supp⁡μ.0=x<z<y=d<\pi,\qquad x,z,y\in\operatorname{supp}\mu.

Choose a>0a>0 so small that

a<z,a<π−d,a<z,\qquad a<\pi-d,

and put b=π−ab=\pi-a. Then Ta(d)=(−a,0]T_a(d)=(-a,0], δ=b−d\delta=b-d. The cut points of aa- and bb-splittings are q=t+aq=t+a and q′=t′+b=t+dq'=t'+b=t+d, respectively. Now, we want zz to lie strictly between them. For that purpose any tt in the non-empty interval (max⁡{−a,z−d},0)(\max\{-a,z-d\},0) works. For every such tt the inequality

μ[q,t+h)μ[t,t+h)≥μ[q′,t+h)μ[t,t+h)\frac{\mu[q,t+h)}{\mu[t,t+h)} \ge\frac{\mu[q',t+h)}{\mu[t,t+h)}

is in fact strict, since z∈[q,q′)z\in[q,q'). The same remains true for xx and yy in sufficiently small neighborhoods of the chosen points. Consequently,

Fa,b(μ)<Fb,a(μ).F_{a,b}(\mu)<F_{b,a}(\mu).

Moreover, this remains strict for all aa in a sufficiently small open interval. That makes inequality 5.10 strict. Now, we have shown that

∣supp⁡μ∣≤4.|\operatorname{supp}\mu|\le4.

Indeed, if five points occurred in cyclic order and g1,…,g5g_1,\ldots,g_5 were the intervening angular gaps, then the absence of three points in an open semicircle would imply

gi+gi+1≥π(imod5).g_i+g_{i+1}\ge\pi\qquad(i\mathrel{\mathrm{mod}}5).

Summing gives 4π≥5π4\pi\ge5\pi, since ∑igi=2π\sum_i g_i=2\pi, which is impossible.

Hence KK is a polygon with at most four sides. We now show that Proposition 5.2 is strict for every nondegenerate polygon. Choose an open interval JJ of angles tt so that for every t∈Jt \in J

ut=(−sin⁡t,cos⁡t)u_t = (-\sin t,\cos t)

is not normal to any edge of KK. For t∈Jt \in J the linear functional x↦⟨x,ut⟩x \mapsto\langle x,u_t\rangle attains its minimum and maximum at unique vertices vv and ww, respectively. Those vertices split ∂K\partial K in two complementary polygonal chains. Denote their lengths by S1,S2S_1,S_2, and let ν1,t,ν2,t\nu_{1,t},\nu_{2,t} be their normalized arclength measures projected onto utu_t.

The equality in the concavity step of Proposition 5.2 is equivalent to

ν1,t=ν2,t.\nu_{1,t} = \nu_{2,t}.

Let e1,e2e_1,e_2 be the unit vectors along the edges of the two chains incident to vv and directed away from vv. For t∈Jt \in J,

⟨ei,ut⟩>0.\langle e_i,u_t\rangle> 0.

Near the projection of vv, the densities of ν1,t\nu_{1,t} and ν2,t\nu_{2,t} are respectively

1S1⟨e1,ut⟩,1S2⟨e2,ut⟩.\frac{1}{S_1\langle e_1,u_t\rangle},\qquad\frac{1}{S_2\langle e_2,u_t\rangle}.

Therefore equality of the two projected laws would imply

⟨S1e1−S2e2,ut⟩=0.\langle S_1e_1-S_2e_2,u_t\rangle= 0.

As S1e1−S2e2≠0S_1e_1-S_2e_2 \ne0, this can hold only for isolated values of tt, whereas JJ is an interval. Hence the Gini concavity inequality is strict on a set of positive measure, and so Proposition 5.2 is strict for every nondegenerate polygon. Therefore,

θ(K)>per⁡K6.\theta(K) > \frac{\operatorname{per} K}{6}.

Sharpness follows from the rectangles Kε=[0,1]×[0,ε]K_\varepsilon=[0,1]\times[0,\varepsilon]. As ε↓0\varepsilon\downarrow0, normalized boundary arclength concentrates on the two horizontal sides, and hence θ(Kε)→E∣U−V∣=1/3\theta(K_\varepsilon)\to\mathbb{E}|U-V|=1/3, where U,V∼Unif⁡[0,1]U,V\sim\operatorname{Unif}[0,1], while per⁡Kε→2\operatorname{per} K_\varepsilon\to2.

Note added

After the first version of our work was posted on arXiv, two related independent works became publicly available. Kukushkin [9] gave a proof of Theorem 2.1 and the planar Zaporozhets–Tarasov conjecture using a quite different approach. Lotnikov [10], whose work is based on his bachelor’s thesis defended in June 2026, constructed centrally symmetric counterexamples in every dimension d≥3d \ge3, for every moment of the distance. These results overlap, respectively, with our Theorem 1.2 and the higher-dimensional counterexamples.

Acknowledgements

The author thanks A. Lotnikov and D. Zaporozhets for useful discussions.

Declaration of AI use

The author used ChatGPT with the GPT-5.6 Sol model during the preparation of this work. The first proof of the planar case of Zaporozhets-Tarasov conjecture, namely Theorem 1.2, was obtained by the author without the use of AI tools; it was based on KKT conditions and an exhaustive case analysis and was substantially more technical. The argument was subsequently simplified through discussions with ChatGPT. The author also used ChatGPT to search for the explicit construction in Example 4.2 from the author’s underlying idea described in Remark 4.4 (that includes generation of relevant code). The author reviewed and verified all AI-assisted output and takes full responsibility for the content of the article.

References

  1. [1]G. Bonnet, A. Gusakova, C. Thäle, and D. Zaporozhets, Sharp inequalities for the mean distance of random points in convex bodies, Adv. Math. 386 (2021), 107813.DOI
  2. [2]A. S. Lotnikov, Mean distance between random points in centrally symmetric convex bodies, Zap. Nauchn. Sem. POMI 544 (2025), 211–235.
  3. [3]A. S. Tokmachev, The Mean Distance Between Random Points on the Boundary of a Convex Figure, J. Math. Sci. 286 (2024), 798–806.DOI
  4. [4]A. S. Tokmachev, Inequalities for convex functions of random points inside and on the boundary of convex bodies, arXiv:2607.08869, 2026.arxiv.org/abs/2607.08869
  5. [5]A. Ya. Khintchine, On unimodal distributions, Izv. Nauchno-Issled. Inst. Mat. Mekh. Tomsk. Gos. Univ. 2 (1938), 1–7. (In Russian.)
  6. [6]S. Yitzhaki, Gini’s mean difference: a superior measure of variability for non-normal distributions, Metron 61 (2003), no. 2, 285–316.
  7. [7]D. Hug and R. Schneider, Hölder continuity for support measures of convex bodies, Arch. Math. (Basel) 104 (2015), 83–92.DOI
  8. [8]R. Schneider, Convex Bodies: The Brunn–Minkowski Theory, 2nd expanded ed., Cambridge University Press, Cambridge, 2014.
  9. [9]M. Kukushkin, The Zaporozhets–Tarasov inequality for an arbitrary planar convex body, arXiv:2608.14833, 2026.arxiv.org/abs/2608.14833
  10. [10]A. S. Lotnikov, Mean distance between points inside and on the boundary of a convex body, arXiv:2608.14848, 2026.arxiv.org/abs/2608.14848

Paper details

vlnc:W_rmamea92
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

GPT-5.6 Sol (bare)

View statement
Looking for related papers…
Contents