betaFeedback

WorksMathematics

A solution to Petty's polar projection problem

Abstract

In 1971, Petty asked whether a convex body in Rn\mathbb{R}^n, n⩾3n\geqslant 3, must be an ellipsoid if its projection body is a dilate of its polar. In this paper, we give an affirmative answer to this question.

2020 Mathematics Subject Classification: 52A20, 52A21

Introduction

The setting of this article is the nn-dimensional Euclidean space Rn\mathbb{R}^n, n≥2n \ge2. A convex body in Rn\mathbb{R}^n is a compact convex set with nonempty interior. Denote by Kn\mathcal{K}^n and Kon\mathcal{K}^n_o the set of convex bodies and the set of convex bodies with the origin oo in their interiors in Rn\mathbb{R}^n, respectively.

Projection bodies, introduced by Minkowski at the turn of the twentieth century, have since become a central notion in convex geometry [8, 15, 18]. For K∈KnK \in\mathcal{K}^n, its projection body ΠK\Pi K is the origin-symmetric convex body defined by

hΠK(u)=∣Pu⊥K∣,u∈Sn−1.h_{\Pi K}(u)=|P_{u^\perp}K|,\qquad u\in S^{n-1}.

Here hL(x)=max⁡{x⋅y:y∈L}h_L(x)=\max\{x\cdot y:y\in L\}, x∈Rnx\in\mathbb{R}^n, is the support function of L∈KnL\in\mathcal{K}^n, and ∣Pu⊥K∣|P_{u^\perp}K| is the (n−1)(n-1)-dimensional volume of the orthogonal projection of KK onto the hyperplane orthogonal to uu. For background and further results on projection bodies, see, e.g., [1, 6, 9, 11, 13, 15, 16, 17, 19, 20].

For K∈KonK\in\mathcal{K}^n_o, let K∗={x∈Rn:x⋅y≤1 for all y∈K}K^*=\{x\in\mathbb{R}^n:x\cdot y\le1\text{ for all }y\in K\} denote its polar body. It is known that every origin-centered ellipsoid KK satisfies

ΠK=λK∗(1)\Pi K=\lambda K^* \tag*{(1)}

for some λ>0\lambda>0. In 1971, Petty [16] asked whether the converse holds for n≥3n\ge3 (see also [3 p. 182] and [10 p. 170]).

Martini and Mustafaev [12 Theorem 3] proved that (1) fails when K∗K^* is a cylinder. Mustafaev [14 Theorem 4.1] answered Petty’s question affirmatively under an additional condition on projection volumes and central chord lengths.

E-mail address: [email protected].

G. Horváth and Lángi [4 Conjecture 2] conjectured that an origin-symmetric convex body KK in Rn\mathbb{R}^{n}, n≥3n \ge3, is an ellipsoid if the volume of the convex hull of KK and K+xK+x is independent of xx whenever the two bodies intersect but have disjoint interiors. They [5 Theorem 3] proved that this conjecture is equivalent to Petty’s problem.

In this article, we completely solve Petty’s problem.

Theorem 1.1. Let n≥3n \ge3 and K∈KonK \in\mathcal{K}_{o}^{n}. If ΠK=λK∗\Pi K = \lambda K^{*} for some λ>0\lambda> 0, then KK is an origin-centered ellipsoid.

This conclusion fails for n=2n=2, as shown by a regular hexagon. See Remark 3.1.

The key geometric observation in the proof of Theorem 1.1 is that if ΠK=λK∗\Pi K = \lambda K^{*}, then for every x∈∂Kx \in\partial K, there exists an outer unit normal uu at xx such that the projection of KK onto u⊥u^{\perp} along Rx\mathbb{R}x is exactly K∩u⊥K \cap u^{\perp}. Once this is proved, Theorem 1.1 follows from James’s characterization of inner product spaces in Lemma 2.1.

AI usage statement. GPT-6 was used to assist with proof development, language editing, and the preparation of Figure 1. The author has checked the arguments and rewritten the text, and takes full responsibility for the content of this paper.

Figure 1. Projection of KK onto u⊥u^\perp along Rx\mathbb{R}x

Preliminaries

For quick reference, we collect some basic facts on convex bodies. As usual, ∣x∣|x| denotes the standard Euclidean norm of xx in Rn\mathbb{R}^{n}; x⋅yx \cdot y denotes the standard inner product of xx and yy. Write Sn−1={x∈Rn:∣x∣=1}S^{n-1}=\{x \in\mathbb{R}^{n}: |x|=1\} for the unit sphere in Rn\mathbb{R}^{n}. For x∈Rn∖{o}x \in\mathbb{R}^{n}\setminus\{o\}, write Rx={tx:t∈R}\mathbb{R}x=\{tx:t \in\mathbb{R}\}. For a convex set KK, write ∣K∣|K| for its volume in its affine hull.

A linear map P:Rn→RnP:\mathbb{R}^{n}\to\mathbb{R}^{n} is called a projection if P2=PP^{2}=P. For a linear subspace ξ\xi of Rn\mathbb{R}^{n}, write Pξ:Rn→ξP_{\xi}:\mathbb{R}^{n}\to\xi for the orthogonal projection onto ξ\xi and ξ⊥\xi^{\perp} for its orthogonal complement in Rn\mathbb{R}^{n}.

For K∈KonK \in\mathcal{K}_{o}^{n}, its Minkowski functional ∥⋅∥K\|\cdot\|_{K} is defined by

∥x∥K=inf⁡{λ>0:x∈λK},x∈Rn.\|x\|_{K}=\inf\{\lambda>0:x \in\lambda K\}, \qquad x \in\mathbb{R}^{n}.

Note that ∥x∥K=hK∗(x)\|x\|_{K}=h_{K^{*}}(x) for x∈Rnx \in\mathbb{R}^{n}. In particular, hK∗(x)=1h_{K^{*}}(x)=1 for x∈∂Kx \in\partial K. If KK is origin-symmetric, ∥⋅∥K\|\cdot\|_{K} is a norm with unit ball KK.

The surface area measure SKS_{K} of K∈KnK \in\mathcal{K}^{n} is a finite Borel measure on Sn−1S^{n-1}, defined for Borel sets ω⊆Sn−1\omega\subseteq S^{n-1} by SK(ω)=Hn−1(νK−1(ω))S_{K}(\omega)=\mathcal{H}^{n-1}(\nu_{K}^{-1}(\omega)), where Hn−1\mathcal{H}^{n-1} is the (n−1)(n-1)-dimensional Hausdorff measure, and νK:∂′K→Sn−1\nu_{K}:\partial'K\to S^{n-1} is the Gauss map of KK, defined on ∂′K\partial'K, the set of points of ∂K\partial K that have a unique outer unit normal. Recall that Hn−1(∂K∖∂′K)=0\mathcal{H}^{n-1}(\partial K\setminus\partial'K)=0; see, e.g., [18 p. 84].

We need the following result of James [7 Theorem 5].

Lemma 2.1. Let n≥3n \ge3 and ∥⋅∥\|\cdot\| be a norm on Rn\mathbb{R}^{n}. If for every x∈Rn∖{o}x \in\mathbb{R}^{n} \setminus\{o\} there exists an (n−1)(n-1)-dimensional linear subspace ξ\xi of Rn\mathbb{R}^{n} such that ∥y+tx∥≥∥y∥\|y+tx\| \ge\|y\| for y∈ξy \in\xi and t∈Rt \in\mathbb{R}, then ∥⋅∥\|\cdot\| is induced by an inner product.

Proof of the main result

Proof of Theorem 1.1. We divide the proof into three steps.

Step 1. Prove that

∫Sn−1∣K∩v⊥∣ dSK(v)=λ∫Sn−11hK(v) dSK(v).(2)\int_{S^{n-1}} |K \cap v^\perp|\,dS_K(v) = \lambda\int_{S^{n-1}} \frac{1}{h_K(v)}\,dS_K(v). \tag*{(2)}

Let v∈Sn−1v \in S^{n-1}. Applying the divergence theorem [2 Theorem 4.6(ii)] to the constant vector field vv on K∩{x:x⋅v≥0}K \cap\{x:x \cdot v \ge0\}, we obtain

0=∫∂K∩{x:x⋅v>0}v⋅νK(x) dHn−1(x)+∫K∩v⊥v⋅(−v) dHn−1(x)=∫∂K∩{x:x⋅v>0}v⋅νK(x) dHn−1(x)−∣K∩v⊥∣.\begin{aligned} 0 = \int_{\partial K \cap\{x:x\cdot v>0\}} v \cdot\nu_K(x)\,d\mathcal{H}^{n-1}(x) + \int_{K\cap v^\perp} v\cdot(-v)\,d\mathcal{H}^{n-1}(x) \\ &= \int_{\partial K \cap\{x:x\cdot v>0\}} v \cdot\nu_K(x)\,d\mathcal{H}^{n-1}(x) - |K\cap v^\perp|. \end{aligned}

Similarly, applying it to K∩{x:x⋅v≤0}K \cap\{x:x \cdot v \le0\} gives

0=∫∂K∩{x:x⋅v<0}v⋅νK(x) dHn−1(x)+∣K∩v⊥∣.0 = \int_{\partial K \cap\{x:x\cdot v<0\}} v \cdot\nu_K(x)\,d\mathcal{H}^{n-1}(x) + |K\cap v^\perp|.

Thus,

∣K∩v⊥∣=12∫∂Ksign⁡(x⋅v) v⋅νK(x) dHn−1(x).(3)|K\cap v^\perp| = \frac{1}{2}\int_{\partial K}\operatorname{sign}(x\cdot v)\,v\cdot\nu_K(x)\,d\mathcal{H}^{n-1}(x). \tag*{(3)}

To integrate (3) against SKS_K, define

w(x)=12∫Sn−1sign⁡(x⋅v)v dSK(v),x∈Rn.(4)w(x)=\frac{1}{2}\int_{S^{n-1}}\operatorname{sign}(x\cdot v)v\,dS_K(v), \qquad x\in\mathbb{R}^{n}. \tag*{(4)}

For x∈∂Kx\in\partial K and y∈Ky\in K, the Cauchy projection formula and ΠK=λK∗\Pi K=\lambda K^* yield

x⋅w(x)=12∫Sn−1∣x⋅v∣ dSK(v)=hΠK(x)=λhK∗(x)=λ,x\cdot w(x)=\frac{1}{2}\int_{S^{n-1}}|x\cdot v|\,dS_K(v)=h_{\Pi K}(x)=\lambda h_{K^*}(x)=\lambda,
y⋅w(x)≤12∫Sn−1∣y⋅v∣ dSK(v)=hΠK(y)=λhK∗(y)≤λ.y\cdot w(x)\le\frac{1}{2}\int_{S^{n-1}}|y\cdot v|\,dS_K(v)=h_{\Pi K}(y)=\lambda h_{K^*}(y)\le\lambda.

Thus w(x)w(x) is an outer normal to KK at xx. For x∈∂′Kx\in\partial'K, it is a positive multiple of νK(x)\nu_K(x). This, together with x⋅w(x)=λx\cdot w(x)=\lambda and x⋅νK(x)=hK(νK(x))x\cdot\nu_K(x)=h_K(\nu_K(x)), yields w(x)=λνK(x)hK(νK(x))w(x)=\frac{\lambda\nu_K(x)}{h_K(\nu_K(x))}. By (3), Fubini’s theorem, (4), this formula and the definition of SKS_K, it follows that

∫Sn−1∣K∩v⊥∣ dSK(v)=12∫Sn−1∫∂Ksign⁡(x⋅v) v⋅νK(x) dHn−1(x) dSK(v)=∫∂Kw(x)⋅νK(x) dHn−1(x)=λ∫∂K1hK(νK(x)) dHn−1(x)=λ∫Sn−11hK(v) dSK(v).\begin{aligned} \int_{\mathbb{S}^{n-1}} |K \cap v^\perp|\,dS_K(v) &= \frac{1}{2}\int_{\mathbb{S}^{n-1}}\int_{\partial K} \operatorname{sign}(x\cdot v)\,v\cdot\nu_K(x)\,d\mathcal{H}^{n-1}(x)\,dS_K(v) \\ &= \int_{\partial K} w(x)\cdot\nu_K(x)\,d\mathcal{H}^{n-1}(x) \\ &= \lambda\int_{\partial K}\frac{1}{h_K(\nu_K(x))}\,d\mathcal{H}^{n-1}(x) \\ &= \lambda\int_{\mathbb{S}^{n-1}}\frac{1}{h_K(v)}\,dS_K(v). \end{aligned}

Step 2. Prove that for every x∈∂Kx \in\partial K there exists an outer unit normal uu to KK at xx such that Px,uK=K∩u⊥P_{x,u}K=K\cap u^\perp. Here, Px,uP_{x,u} is the projection onto u⊥u^\perp along Rx\mathbb{R}x defined by

Px,uy=y−y⋅ux⋅ux,y∈Rn.(5)P_{x,u}y=y-\frac{y\cdot u}{x\cdot u}x,\qquad y\in\mathbb{R}^n. \tag*{(5)}

We first consider x∈∂′Kx\in\partial'K. Write u=νK(x)u=\nu_K(x). Then x⋅u=hK(u)>0x\cdot u=h_K(u)>0. The projection Px,uP_{x,u} is the identity on u⊥u^\perp, so

K∩u⊥=Px,u(K∩u⊥)⊆Px,uK.(6)K\cap u^\perp=P_{x,u}(K\cap u^\perp)\subseteq P_{x,u}K. \tag*{(6)}

Since x⋅u>0x\cdot u>0, the restriction Px⊥∣u⊥:u⊥→x⊥P_{x^\perp}|_{u^\perp}:u^\perp\to x^\perp has kernel Rx∩u⊥={o}\mathbb{R}x\cap u^\perp=\{o\} and is therefore an isomorphism. Its Jacobian is the cosine of the angle between u⊥u^\perp and x⊥x^\perp, that is, x∣x∣⋅u=hK(u)∣x∣\frac{x}{|x|}\cdot u=\frac{h_K(u)}{|x|}. We also have Px⊥Px,u=Px⊥P_{x^\perp}P_{x,u}=P_{x^\perp} by (5). The definition of ΠK\Pi K, together with these two facts, yields

hΠK(x)∣x∣=∣Px⊥K∣=∣Px⊥Px,uK∣=hK(u)∣x∣∣Px,uK∣.\frac{h_{\Pi K}(x)}{|x|}=|P_{x^\perp}K|=|P_{x^\perp}P_{x,u}K|=\frac{h_K(u)}{|x|}|P_{x,u}K|.

By (6), this equality, ΠK=λK∗\Pi K=\lambda K^* and hK∗(x)=1h_{K^*}(x)=1, it follows that

∣K∩u⊥∣≤∣Px,uK∣=hΠK(x)hK(u)=λhK∗(x)hK(u)=λhK(u).|K\cap u^\perp|\leq|P_{x,u}K|=\frac{h_{\Pi K}(x)}{h_K(u)}=\frac{\lambda h_{K^*}(x)}{h_K(u)}=\frac{\lambda}{h_K(u)}.
endsmid=1{ends_mid=1}

This, together with (2) and the definition of SKS_K, implies that ∣K∩u⊥∣=∣Px,uK∣|K \cap u^\perp| = |P_{x,u}K| for Hn−1\mathcal{H}^{n-1}-a.e. x∈∂′Kx \in\partial'K. Combining this with (6), we obtain

Px,uK=K∩u⊥.(7)P_{x,u}K = K \cap u^\perp. \tag*{(7)}

We now extend the assertion to every boundary point. Let E⊆∂′KE \subseteq\partial'K be the set of points xx for which (7) holds. Since Hn−1(∂K∖E)=0\mathcal{H}^{n-1}(\partial K \setminus E) = 0, EE is dense in ∂K\partial K. Fix x∈∂Kx \in\partial K, choose xj∈Ex_j \in E with xj→xx_j \to x, and write uj=νK(xj)u_j = \nu_K(x_j). By compactness of Sn−1\mathbb{S}^{n-1}, we may pass to a subsequence such that uj→u∈Sn−1u_j \to u \in\mathbb{S}^{n-1}. Taking limits in xj⋅uj=hK(uj)x_j \cdot u_j = h_K(u_j) yields x⋅u=hK(u)>0x \cdot u = h_K(u) > 0. Hence uu is an outer unit normal to KK at xx. Since xj⋅uj→x⋅u>0x_j \cdot u_j \to x \cdot u > 0, it follows from (5) that

Pxj,ujy=y−y⋅ujxj⋅ujxj⟶y−y⋅ux⋅ux=Px,uy,y∈K.P_{x_j,u_j}y = y - \frac{y \cdot u_j}{x_j \cdot u_j}x_j \longrightarrow y - \frac{y \cdot u}{x \cdot u}x = P_{x,u}y,\qquad y \in K.

As KK is closed and Pxj,ujK⊆KP_{x_j,u_j}K \subseteq K by (7), we have Px,uK⊆KP_{x,u}K \subseteq K. Since Px,uP_{x,u} is the identity on u⊥u^\perp and Px,uK⊆u⊥P_{x,u}K \subseteq u^\perp, it follows that

K∩u⊥=Px,u(K∩u⊥)⊆Px,uK⊆K∩u⊥.K \cap u^\perp= P_{x,u}(K \cap u^\perp) \subseteq P_{x,u}K \subseteq K \cap u^\perp.

Thus Px,uK=K∩u⊥P_{x,u}K = K \cap u^\perp.

Step 3. Prove that KK is an origin-centered ellipsoid.

Since ΠK\Pi K is origin-symmetric and ΠK=λK∗\Pi K = \lambda K^*, it follows that K∗K^* is origin-symmetric, and so is KK. Thus ∥⋅∥K\|\cdot\|_K is a norm on Rn\mathbb{R}^n with unit ball KK.

By Step 2, for every x∈∂Kx \in\partial K, there exists u∈Sn−1u \in\mathbb{S}^{n-1} such that Px,uK⊆KP_{x,u}K \subseteq K. Hence ∥Px,uz∥K≤∥z∥K\|P_{x,u}z\|_K \leq\|z\|_K for z∈Rnz \in\mathbb{R}^n. This, together with (5), yields that

∥y+tx∥K≥∥Px,u(y+tx)∥K=∥y∥K,y∈u⊥, t∈R.\|y + tx\|_K \geq\|P_{x,u}(y + tx)\|_K = \|y\|_K,\qquad y \in u^\perp,\ t \in\mathbb{R}.

Since each nonzero vector is a positive multiple of a point of ∂K\partial K, it follows from Lemma 2.1 that ∥⋅∥K\|\cdot\|_K is induced by an inner product. That is, KK is an origin-centered ellipsoid. □\square

Theorem 1.1 also yields the following translated version. For n≥3n \geq3, suppose that K∈KnK \in\mathcal{K}_n satisfies ΠK=λ(K−z)∗\Pi K = \lambda(K-z)^* for some λ>0\lambda> 0 and z∈int⁡Kz \in\operatorname{int} K. Let L=K−zL = K-z. Then L∈KnoL \in\mathcal{K}_n^o, and ΠL=ΠK=λL∗\Pi L = \Pi K = \lambda L^*. Theorem 1.1 implies that LL is an origin-centered ellipsoid, and hence KK is an ellipsoid centered at zz.

Remark 3.1. In the proof of Theorem 1.1, the assumption n≥3n \geq3 is used only in applying Lemma 2.1. This assumption cannot be removed. Let H⊆R2H \subseteq\mathbb{R}^2 be a regular hexagon of circumradius one centered at the origin and RR be a rotation through π2\frac{\pi}{2}. Then for u∈S1u \in\mathbb{S}^1,

hΠH(u)=∣Pu⊥H∣=2hH(R−1u)=2hRH(u).h_{\Pi H}(u) = |P_{u^\perp}H| = 2h_H(R^{-1}u) = 2h_{RH}(u).

The inradius of HH is 32\frac{\sqrt{3}}{2}, so H∗H^{*} is obtained by rotating HH through π6\frac{\pi}{6} and dilating it by 23\frac{2}{\sqrt{3}}. Since HH is invariant under rotation through π3\frac{\pi}{3}, we have H∗=23RHH^{*}=\frac{2}{\sqrt{3}}RH, and hence ΠH=2RH=3H∗\Pi H=2RH=\sqrt{3}H^{*}. Thus HH satisfies (1) with λ=3\lambda=\sqrt{3}, although it is not an ellipse.

References

  1. [1]A. Aleksandrov, Zur Theorie der gemischten Volumina von konvexen Körpern. II. Neue Ungleichungen zwischen den gemischten Volumina und ihre Anwendungen, Mat. Sb. N.S. 2(44) (1937) 1205–1238. 1
  2. [2]L. Evans, R. Gariepy, Measure theory and fine properties of functions, revised ed., CRC Press, Boca Raton, 2015. 3
  3. [3]R. Gardner, Geometric tomography, 2nd ed., Cambridge University Press, Cambridge, 2006. 1DOI
  4. [4]Á. G.Horváth, Z. Lángi, On the volume of the convex hull of two convex bodies, Monatsh. Math. 174 (2014) 219–229. 2
  5. [5]Á. G.Horváth, Z. Lángi, On the convex hull and homothetic convex hull functions of a convex body, Geom. Dedicata 216 (2022), Paper No. 10, 12 pp. 2arxiv.org/abs/2012.08955
  6. [6]C. Haberl, F. Schuster, Affine vs. Euclidean isoperimetric inequalities, Adv. Math. 356 (2019), Paper No. 106811, 26 pp. 1arxiv.org/abs/1804.11165
  7. [7]R. James, Inner products in normed linear spaces, Bull. Amer. Math. Soc. 53 (1947) 559–566. 3DOI
  8. [8]M. Ludwig, Projection bodies and valuations, Adv. Math. 172 (2002) 158–168. 1DOI
  9. [9]E. Lutwak, Mixed projection inequalities, Trans. Amer. Math. Soc. 287 (1985) 91–106. 1DOI
  10. [10]E. Lutwak, Selected affine isoperimetric inequalities, in: P. Gruber, J. Wills (Eds.), Handbook of convex geometry, Vol. A, North-Holland, Amsterdam, 1993, pp. 151–176. 1DOI
  11. [11]E. Lutwak, D. Yang, G. Zhang, L_p affine isoperimetric inequalities, J. Differential Geom. 56 (2000) 111–132. 1
  12. [12]H. Martini, Z. Mustafaaev, On unit balls and isoperimetrics in normed spaces, Colloq. Math. 127 (2012) 133–142. 1DOI
  13. [13]E. Milman, A. Yehudayoff, Sharp isoperimetric inequalities for affine quermassintegrals, J. Amer. Math. Soc. 36 (2023) 1061–1101. 1
  14. [14]Z. Mustafaaev, On minimax and maxmin inequalities for centered convex bodies, Math. Inequal. Appl. 25 (2022) 903–911. 1
  15. [15]C. Petty, Projection bodies, in: Proceedings of the Colloquium on Convexity (Copenhagen, 1965), København Univ. Mat. Inst., Copenhagen, 1967, pp. 234–241. 1
  16. [16]C. Petty, Isoperimetric problems, in: Proceedings of the Conference on Convexity and Combinatorial Geometry (Norman, 1971), University of Oklahoma, Norman, 1972, pp. 26–41. 1
  17. [17]R. Schneider, Zu einem Problem von Shephard über die Projektionen konvexer Körper, Math. Z. 101 (1967) 71–82. 1DOI
  18. [18]R. Schneider, Convex bodies: the Brunn–Minkowski theory, 2nd expanded ed., Cambridge University Press, Cambridge, 2014. 1, 2
  19. [19]G. Zhang, Restricted chord projection and affine inequalities, Geom. Dedicata 39 (1991) 213–222. 1DOI
  20. [20]D. Zou, G. Xiong, The Orlicz Brunn–Minkowski inequality for the projection body, J. Geom. Anal. 30 (2020) 2253–2272. 1DOI

Paper details

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

views · 0

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents