betaFeedback

WorksMathematics

Convex-body volume counterexamples: Convex-body volume counterexamples to hyperplane equipartition

Abstract

For every d≥4d\ge4 and every ellipsoid in Rd\R^d, we construct real-analytic convex bodies of positive curvature, converging smoothly to that ellipsoid and having the same volume, whose volume admits no equipartition by dd affine hyperplanes. For d≥6d\ge6, the bodies can be chosen centrally symmetric. The four-dimensional construction uses Soberón's cubic–quartic tensor nonvanishing theorem. We prove a boundary-perturbation criterion that realizes this obstruction for the uniform volume of a convex body; higher-dimensional examples follow from a finite-dimensional transversality argument. We also construct counterexamples to orthogonal equipartition near every ellipsoid with distinct semiaxes in dimensions d≥3d\ge3, with central symmetry when d≥4d\ge4. The counterexample property is open in the Hausdorff topology, giving polytopal examples as well.

Introduction

An equipartition of a convex body K⊂RdK \subset\mathbb{R}^d by dd affine hyperplanes requires each of the 2d2^d sign regions to have Lebesgue volume 2−dvol⁡d(K)2^{-d}\operatorname{vol}_d(K). The hyperplanes have arbitrary offsets and directions. A convex body is a compact convex set with nonempty interior.

Grünbaum’s question originates in [3 Section 4(v)], where it is phrased for continuous mass distributions. Its convex-body volume formulation is stated explicitly in [2 Section 1]. The general mass problem has a positive answer in dimensions at most three by the elementary planar argument and Hadwiger’s theorem [4]. Avis [1] constructed high-dimensional point-set counterexamples, and Soberón [8] recently constructed a smooth positive density counterexample in dimension four. The surveys [2, 6] discuss the mass-partition framework; the measurable-set volume formulation in dimension four is recorded in [9 Problem 21.2.6].

Here the measure is required to be the uniform volume of one convex body. A negative result for unrestricted densities does not by itself supply such a body. We realize the four-dimensional tensor obstruction of [8] by perturbing the boundary of a ball, while retaining convexity and positive curvature.

If c∈int⁡Kc \in\operatorname{int} K, write

ρK,c(v)=max⁡{r≥0:c+rv∈K},v∈Sd−1.\rho_{K,c}(v) = \max\{r \ge0 : c + rv \in K\}, \qquad v \in\mathbb{S}^{d-1}.

Smooth convergence below means convergence of these radial functions in every fixed Ck(Sd−1)C^k(\mathbb{S}^{d-1}) norm.

Theorem 1.1. Let d≥4d \ge4, and let E⊂RdE \subset\mathbb{R}^d be an ellipsoid with center cc. There are ε0>0\varepsilon_0 > 0 and convex bodies KεK_\varepsilon, ∣ε∣<ε0|\varepsilon| < \varepsilon_0, such that K0=EK_0 = E and:

(1) ∂Kε\partial K_\varepsilon is real analytic and all its principal curvatures are positive;

(2) vol⁡d(Kε)=vol⁡d(E)\operatorname{vol}_d(K_\varepsilon) = \operatorname{vol}_d(E), and ρKε,c→ρE,c\rho_{K_\varepsilon,c} \to\rho_{E,c} in C∞(Sd−1)C^\infty(\mathbb{S}^{d-1});

(3) for every 0<∣ε∣<ε00 < |\varepsilon| < \varepsilon_0, KεK_\varepsilon has no equipartition by dd affine hyperplanes.

If d≥6d \ge6, the bodies may be chosen centrally symmetric about cc.

Together with the low-dimensional positive results, this gives the dimension threshold for uniform convex-body volume.

Corollary 1.2. Every convex body in Rd\mathbb{R}^d admits an equipartition by dd affine hyperplanes if and only if d≤3d \le3.

We also consider the restriction that the hyperplane normals be mutually orthogonal. Maldonado and Roldán-Pensado [5] constructed measure counterexamples in dimensions at least three. The following result realizes uniform-volume counterexamples near a prescribed nonspherical reference shape.

Theorem 1.3. Let d≥3d \ge3, and let E⊂RdE \subset\mathbb{R}^d be an ellipsoid with pairwise distinct semiaxis lengths and center cc. There are real-analytic, positively curved convex bodies KεK_\varepsilon, smoothly converging to EE, of the same volume as EE, which admit no equipartition by dd mutually perpendicular affine hyperplanes whenever 0<∣ε∣0 < |\varepsilon| is sufficiently small. For d≥4d \ge4, they may be chosen centrally symmetric about cc.

Corollary 1.4. In every dimension d≥4d \ge4 there are convex polytopes whose volume has no equipartition by dd affine hyperplanes. They can be chosen centrally symmetric for d≥6d \ge6. For orthogonal equipartitions, polytopal counterexamples exist for d≥3d \ge3, with central symmetry for d≥4d \ge4. In each assertion the polytopes can be chosen arbitrarily close in Hausdorff distance to the reference ellipsoid in the corresponding theorem, and with its volume.

Contribution and proof outline

Contribution and proof outline. The use of sign correlations, reduction to frames near O(4)O(4), and the cubic–quartic obstruction in dimension four come from Soberón’s work. We use his tensor theorem as an explicit external input, restated in Theorem 4.1. The contribution here is its boundary realization for uniform convex-body volume, together with the proximity, symmetry, higher-dimensional and orthogonal statements above.

The main technical point is Proposition 3.4. Any sequence of equipartitions of small radial perturbations of a ball must converge to an orthogonal frame through the origin. For sign correlations involving at least three hyperplanes, changes of angles and offsets contribute only o(∣ε∣)o(|\varepsilon|). The remaining first-order term is the spherical boundary perturbation. A nonvanishing boundary obstruction therefore rules out every affine equipartition, including configurations whose hyperplanes move with ε\varepsilon.

§ 2 establishes the geometry and configuration estimates. § 3 proves the transfer criterion with the offset and boundary limits written separately. § 4 and § 5 prove Theorem 1.1. The orthogonal and polytopal assertions are proved in § 6 and § 7.

Boundary perturbations and equipartition coefficients

Throughout, BdB^d is the unit ball, κd=vol⁡d(Bd)\kappa_d = \operatorname{vol}_d(B^d), and dσd\sigma is unnormalized surface measure on Sd−1S^{d-1}. A frame q=(q1,…,qd)∈O(d)q = (q_1,\ldots,q_d) \in O(d) is identified with the orthogonal matrix having these columns. Constants may depend on dd and on a fixed perturbation function, but not on its parameter.

A volume-exact radial perturbation

Let ϕ∈C(Sd−1)\phi\in C(S^{d-1}). For sufficiently small real ε\varepsilon, define the radial body

ρε(v)=(1+dεϕ(v))1/d,Kε={rv:v∈Sd−1, 0≤r≤ρε(v)}.(1)\rho_\varepsilon(v) = \left(1+d\varepsilon\phi(v)\right)^{1/d}, \qquad K_\varepsilon= \{rv : v \in S^{d-1},\ 0 \le r \le\rho_\varepsilon(v)\}. \tag*{(1)}

We take ∣dε∣∥ϕ∥∞<1/2|d\varepsilon|\|\phi\|_\infty< 1/2. Polar integration gives the exact identities

vol⁡d(Kε)=κd+ε∫Sd−1ϕ dσ.(2)\operatorname{vol}_d(K_\varepsilon) = \kappa_d + \varepsilon\int_{S^{d-1}} \phi\,d\sigma. \tag*{(2)}
vol⁡d(Kε△Bd)=∣ε∣∫Sd−1∣ϕ∣ dσ.(3)\operatorname{vol}_d(K_\varepsilon\mathbin{\triangle} B^d) = |\varepsilon| \int_{S^{d-1}} |\phi|\,d\sigma. \tag*{(3)}

Lemma 2.1. If ϕ\phi is C∞C^\infty, then for sufficiently small ∣ϵ∣|\epsilon|, KϵK_\epsilon is a convex body with positive principal curvatures and ρϵ→1\rho_\epsilon\to1 in C∞C^\infty. If ϕ\phi is real analytic, its boundary is real analytic. If ϕ\phi is even, Kϵ=−KϵK_\epsilon=-K_\epsilon. If ∫ϕ dσ=0\int\phi\,d\sigma=0, its volume is exactly κd\kappa_d.

Proof. Convergence in each fixed smooth norm follows from (1). To justify convexity directly, define the squared gauge

gϵ(x)=∣x∣2ρϵ(x/∣x∣)2(x≠0),gϵ(0)=0.g_\epsilon(x)=\frac{|x|^2}{\rho_\epsilon(x/|x|)^2}\quad(x\ne0),\qquad g_\epsilon(0)=0.

Its Hessian on the unit sphere converges uniformly to 2Id2I_d. Positive homogeneity makes the Hessian uniformly positive definite on Rd∖{0}\mathbb{R}^d\setminus\{0\}. Also gϵg_\epsilon is C1C^1 at zero, with gradient zero. Its restriction to any line is therefore convex, including a line through zero, so its sublevel set Kϵ={gϵ≤1}K_\epsilon=\{g_\epsilon\le1\} is convex. On gϵ=1g_\epsilon=1, Euler’s identity gives x⋅∇gϵ=2x\cdot\nabla g_\epsilon=2, and the second fundamental form is the positive Hessian restricted to the tangent space, divided by ∣∇gϵ∣|\nabla g_\epsilon|. This proves positive curvature. Analyticity follows from the analytic radial parametrization; symmetry follows from evenness. The volume assertion is (2).

Sign correlations and all affine configurations

For unit normals u=(u1,…,ud)u=(u_1,\ldots,u_d), offsets t=(t1,…,td)t=(t_1,\ldots,t_d), and I⊂[d]={1,…,d}I\subset[d]=\{1,\ldots,d\}, put

WI(K;u,t)=∫K∏i∈Isgn⁡(ui⋅x−ti) dx.(4)W_I(K;u,t)=\int_K\prod_{i\in I}\operatorname{sgn}(u_i\cdot x-t_i)\,dx. \tag*{(4)}

The empty product is one. The value of sgn⁡(0)\operatorname{sgn}(0) is irrelevant.

Lemma 2.2. The hyperplanes ui⋅x=tiu_i\cdot x=t_i equipartition KK if and only if WI(K;u,t)=0W_I(K;u,t)=0 for every nonempty I⊂[d]I\subset[d].

Proof. For s=(s1,…,sd)∈{−1,1}ds=(s_1,\ldots,s_d)\in\{-1,1\}^d, the indicator of its sign region equals, outside the hyperplanes,

2−d∏i=1d(1+sisgn⁡(ui⋅x−ti)).2^{-d}\prod_{i=1}^{d}(1+s_i\operatorname{sgn}(u_i\cdot x-t_i)).

Its integral is 2−d∑I⊂[d](∏i∈Isi)WI(K;u,t)2^{-d}\sum_{I\subset[d]}(\prod_{i\in I}s_i)W_I(K;u,t). The characters ∏i∈Isi\prod_{i\in I}s_i form a basis on the sign cube; their orthogonality proves both implications.

Lemma 2.3. Suppose ϵm→0\epsilon_m\to0 through nonzero values and (um,tm)(u^m,t^m) equipartitions the radial bodies (1). Then

max⁡i∣tim∣=O(∣ϵm∣),max⁡i≠j∣uim⋅ujm∣=O(∣ϵm∣).(5)\max_i|t_i^m|=O(|\epsilon_m|),\qquad\max_{i\ne j}|u_i^m\cdot u_j^m|=O(|\epsilon_m|). \tag*{(5)}

After passage to a subsequence, uim→qiu_i^m\to q_i, where q∈O(d)q\in O(d).

Proof. Write D=BdD=B^d. Since every hyperplane bisects KϵmK_{\epsilon_m}, (3) and Lemma 2.2 imply

∣W{i}(D;um,tm)∣≤vol⁡d(Kϵm△D)≤C∣ϵm∣.|W_{\{i\}}(D;u^m,t^m)|\le\operatorname{vol}_d(K_{\epsilon_m}\mathbin{\triangle}D)\le C|\epsilon_m|.

For a unit normal and ∣s∣<1|s|<1,

W{i}(D;u,s)=−2κd−1∫0s(1−r2)(d−1)/2 dr.W_{\{i\}}(D;u,s)=-2\kappa_{d-1}\int_0^s(1-r^2)^{(d-1)/2}\,dr.

This is strictly decreasing, vanishes only at zero, and has derivative bounded away from zero near zero. Consequently each tim=O(∣ϵm∣)t_i^m=O(|\epsilon_m|). The same conclusion follows first by forcing ∣tim∣<1/2|t_i^m|<1/2 and then using the displayed derivative bound, so the estimate is uniform in the normal. Changing one offset to zero changes a sign product only in the corresponding slab. A slab of width ∣s∣|s| in DD has volume at most κd−1∣s∣\kappa_{d-1}|s|. Hence

∣W{i,j}(D;um,tm)−W{i,j}(D;um,0)∣≤2κd−1(∣tim∣+∣tjm∣).|W_{\{i,j\}}(D;u^m,t^m)-W_{\{i,j\}}(D;u^m,0)|\le2\kappa_{d-1}(|t_i^m|+|t_j^m|).

::: The two-sign correlation in a rotationally invariant ball is

κd−1W{i,j}(D;u,0)=1−2πarccos⁡(ui⋅uj)=2πarcsin⁡(ui⋅uj).(6)\kappa_d^{-1}W_{\{i,j\}}(D;u,0)=1-\frac{2}{\pi}\arccos(u_i\cdot u_j)=\frac{2}{\pi}\arcsin(u_i\cdot u_j). \tag*{(6)}

Indeed, in the plane spanned by the two normals, the angular proportion on which the signs disagree is arccos⁡(ui⋅uj)/π\arccos(u_i\cdot u_j)/\pi; rotational invariance gives the same proportion of ball volume. The coefficient over KεmK_{\varepsilon_m} is zero, and its difference from the coefficient over DD is bounded by (3). Since ∣arcsin⁡r∣≥∣r∣|\arcsin r|\ge|r| for ∣r∣≤1|r|\le1, this proves the second estimate in (5). Compactness of (Sd−1)d(S^{d-1})^d gives a convergent subsequence, whose limiting normals are pairwise orthogonal. □

The first-order boundary obstruction

For ∣I∣≥3|I|\ge3, define

TI(ϕ,q)=∫Sd−1ϕ(v)∏i∈Isgn⁡(qi⋅v) dσ(v),q∈O(d).(7)T_I(\phi,q)=\int_{S^{d-1}}\phi(v)\prod_{i\in I}\operatorname{sgn}(q_i\cdot v)\,d\sigma(v),\qquad q\in O(d). \tag*{(7)}

The map q↦TI(ϕ,q)q\mapsto T_I(\phi,q) is continuous by dominated convergence: the exceptional great spheres have surface measure zero.

Centered correlations have no linear angular term

Lemma 3.1. Let u1,…,udu_1,\ldots,u_d be unit vectors whose Gram matrix is G=Id+EG=\mathrm{Id}+E, with ∥E∥\|E\| sufficiently small. For ∣I∣≥3|I|\ge3,

∣WI(Bd;u,0)∣≤Cd∥E∥2.|W_I(B^d;u,0)|\le C_d\|E\|^2.

Proof. A uniformly distributed point of the ball and a standard Gaussian have the same uniformly distributed direction. Signs of homogeneous linear forms depend only on this direction. Thus, if YGY_G is a centered Gaussian vector with covariance GG,

κd−1WI(Bd;u,0)=E∏i∈Isgn⁡((YG)i).\kappa_d^{-1}W_I(B^d;u,0)=\mathbb{E}\prod_{i\in I}\operatorname{sgn}((Y_G)_i).

The covariance is positive definite near Id\mathrm{Id}, and its density is

pG(y)=(2π)−d/2(det⁡G)−1/2exp⁡(−12yTG−1y).p_G(y)=(2\pi)^{-d/2}(\det G)^{-1/2}\exp\left(-\frac{1}{2}y^TG^{-1}y\right).

At G=IdG=\mathrm{Id}, its directional derivative in a symmetric direction EE is

DpId[E](y)=12(yTEy−tr⁡E)pId(y).(8)Dp_{\mathrm{Id}}[E](y)=\frac{1}{2}(y^TEy-\operatorname{tr}E)p_{\mathrm{Id}}(y). \tag*{(8)}

The integral of the sign product against pIdp_{\mathrm{Id}} is zero. Its integral against every term in (8) is also zero: a monomial of degree at most two leaves at least one of the three or more specified coordinates available for an odd reflection.

For completeness, on a fixed neighborhood ∥G−Id∥<1/2\|G-\mathrm{Id}\|<1/2, second derivatives of pGp_G in the matrix entries are bounded in absolute value by Cd(1+∣y∣4)e−cd∣y∣2C_d(1+|y|^4)e^{-c_d|y|^2}. This is integrable uniformly in GG. Taylor’s formula in L1L^1 therefore bounds the remaining integral by Cd∥E∥2C_d\|E\|^2. □

Small offsets do not restore the higher correlations

Lemma 3.2. Under the assumptions of Lemma 2.3, for every ∣I∣≥3|I|\ge3,

WI(Bmd;um,tm)=o(∣εm∣).(9)W_I(B^d_m;u^m,t^m)=o(|\varepsilon_m|). \tag*{(9)}

Proof. By Lemmas 2.3 and 3.1,

WI(Bd;um,0)=O(∣εm∣2).W_I(B^d;u^m,0)=O(|\varepsilon_m|^2).

Change the offsets with indices in II, one at a time. At the step changing index ii, the change of the coefficient is exactly

−2∫0timFi,m(s) ds,Fi,m(s)=∫Bd∩{uim⋅x=s}∏j∈I∖{i}sgn⁡(ujm⋅x−t^jm) dHd−1(x).(10)-2\int_{0}^{t_i^m} F_{i,m}(s)\,ds,\qquad F_{i,m}(s)=\int_{B^d\cap\{u_i^m\cdot x=s\}}\prod_{j\in I\setminus\{i\}}\operatorname{sgn}(u_j^m\cdot x-\widehat{t}_j^m)\,d\mathcal{H}^{d-1}(x). \tag*{(10)}

Here each t^jm\widehat{t}_j^m is either zero or tjmt_j^m. The signed integral also covers tim<0t_i^m<0.

We claim that Fi,m(s)=o(1)F_{i,m}(s)=o(1) uniformly for ss between zero and timt_i^m. To see this, choose any sequence of such s=sms=s_m. Choose orthogonal maps Rm→Id⁡R_m\to\operatorname{Id} with Rmqi=uimR_mq_i=u_i^m. The parametrization

x=smuim+1−sm2Rmy,y∈Bd∩qi⊥x=s_mu_i^m+\sqrt{1-s_m^2}R_my,\qquad y\in B^d\cap q_i^\perp

has surface Jacobian (1−sm2)(d−1)/2(1-s_m^2)^{(d-1)/2}. For almost every yy, its sign product converges to ∏j∈I∖{i}sgn⁡(qj⋅y)\prod_{j\in I\setminus\{i\}}\operatorname{sgn}(q_j\cdot y). The excluded sets qj⋅y=0q_j\cdot y=0 have zero (d−1)(d-1)-dimensional measure in this disk. The limiting integral is zero: reflection in any one of the directions qjq_j, j∈I∖{i}j\in I\setminus\{i\}, reverses the product and preserves the disk. Dominated convergence proves convergence to zero for every choice of sms_m, which also proves the claimed uniformity. Since tim=O(∣εm∣)t_i^m=O(|\varepsilon_m|), (10) is o(∣εm∣)o(|\varepsilon_m|). There are only finitely many changes of offsets, proving (9). □

The shell term and the transfer criterion

Lemma 3.3. Suppose um→q∈O(d)u^m\to q\in O(d), tm→0t^m\to0, and εm→0\varepsilon_m\to0 through nonzero values. For every I⊂[d]I\subset[d],

WI(Kεm;um,tm)−WI(Bd;um,tm)εm⟶TI(ϕ,q),\frac{W_I(K_{\varepsilon_m};u^m,t^m)-W_I(B^d;u^m,t^m)}{\varepsilon_m}\longrightarrow T_I(\phi,q),

where (7) is used also for smaller index sets.

Proof. The difference is the signed shell integral

∫Sd−1∫1ρεm(v)rd−1∏i∈Isgn⁡(r uim⋅v−tim) dr dσ(v).\int_{\mathbb{S}^{d-1}}\int_{1}^{\rho_{\varepsilon_m}(v)}r^{d-1}\prod_{i\in I}\operatorname{sgn}(r\,u_i^m\cdot v-t_i^m)\,dr\,d\sigma(v).

For vv outside the finitely many great spheres qi⋅v=0q_i\cdot v=0, all signs on the shrinking radial interval eventually agree with the limiting signs. Its signed radial volume is exactly

∫1ρεm(v)rd−1 dr=εmϕ(v).\int_{1}^{\rho_{\varepsilon_m}(v)}r^{d-1}\,dr=\varepsilon_m\phi(v).

The inner integral divided by εm\varepsilon_m is bounded in absolute value by ∣ϕ(v)∣|\phi(v)|, including negative parameters and inward perturbations. Dominated convergence gives the result. □

Proposition 3.4 (Boundary transfer). Let ϕ∈C(Sd−1)\phi\in C(\mathbb{S}^{d-1}). If a sequence εm→0\varepsilon_m\to0, εm≠0\varepsilon_m\ne0, admits affine equipartitions of KεmK_{\varepsilon_m}, then some q∈O(d)q\in O(d) satisfies

TI(ϕ,q)=0for every ∣I∣≥3.(11)T_I(\phi,q)=0\qquad\text{for every }|I|\ge3. \tag*{(11)}

Consequently, if the functions in (11) have no common zero on O(d)O(d), then KεK_\varepsilon has no affine equipartition whenever 0<∣ε∣0<|\varepsilon| is sufficiently small.

Proof. Take the limiting orthogonal frame supplied by Lemma 2.3. The coefficient over KεmK_{\varepsilon_m} is zero. Divide it by εm\varepsilon_m, use Lemma 3.2 for the ball contribution and Lemma 3.3 for the difference. This proves (11). If the last conclusion failed, one could choose such a sequence with εm→0\varepsilon_m\to0, contradicting the absence of a common zero. □

Remark 3.5. The argument excludes moving affine configurations, not merely centered orthogonal ones. Lemma 3.1 controls the change of directions and Lemma 3.2 controls the change of offsets. These estimates are what permit the spherical obstruction to be used for the full affine problem.

The four-dimensional construction

The sole nonvanishing input needed in dimension four is the following theorem. We state its attribution separately from the boundary-transfer argument.

Theorem 4.1. (Soberón [8 Theorem 2]). There exist symmetric multilinear forms A:(R4)3→RA : (\mathbb{R}^{4})^{3} \to\mathbb{R} and B:(R4)4→RB : (\mathbb{R}^{4})^{4} \to\mathbb{R} such that, for every q=(q1,q2,q3,q4)∈O(4)q = (q_{1},q_{2},q_{3},q_{4}) \in O(4), the five numbers

A(q1,q2,q3),A(q1,q2,q4),A(q1,q3,q4),A(q2,q3,q4),B(q1,q2,q3,q4)A(q_{1},q_{2},q_{3}), \quad A(q_{1},q_{2},q_{4}), \quad A(q_{1},q_{3},q_{4}), \quad A(q_{2},q_{3},q_{4}), \quad B(q_{1},q_{2},q_{3},q_{4})

do not all vanish.

The cited version is v2, whose Section 3 gives a non-computational proof of this theorem. We do not reprove that external theorem here.

Remark 4.2. (An explicit choice of the perturbation.) The tensors in [8 Theorem 3] make the construction concrete. In coordinates x=(x1,x2,x3,x4)x = (x_{1},x_{2},x_{3},x_{4}), set

M=(0111110010201004),N=M+e1e1T,M = \begin{pmatrix} 0 & 1 & 1 & 1 \\ 1 & 1 & 0 & 0 \\ 1 & 0 & 2 & 0 \\ 1 & 0 & 0 & 4 \end{pmatrix}, \qquad N = M + e_{1}e_{1}^{T},

and define the symmetric tensors by their diagonal polynomials

Aτ(x,x,x)=x1xTMx+6τx2x3x4,B(x,x,x,x)=(xTNx)2.A_{\tau}(x,x,x) = x_{1}x^{T}Mx + 6\tau x_{2}x_{3}x_{4}, \qquad B(x,x,x,x) = (x^{T}Nx)^{2}.

Soberón’s theorem states that Aτ,BA_{\tau}, B have the required nonvanishing property for every sufficiently small fixed τ>0\tau> 0. For such a fixed τ\tau, one may therefore use

ϕ(v)=v1vTMv+6τv2v3v4+(vTNv)2−5,v∈S3.(12)\phi(v) = v_{1}v^{T}Mv + 6\tau v_{2}v_{3}v_{4} + (v^{T}Nv)^{2} - 5, \qquad v \in S^{3}. \tag*{(12)}

The cubic terms have zero mean. The standard fourth spherical moments give

1σ(S3)∫S3(vTNv)2 dσ(v)=(tr⁡N)2+2tr⁡(N2)4⋅6=64+5624=5.\frac{1}{\sigma(S^{3})}\int_{S^{3}} (v^{T}Nv)^{2}\,d\sigma(v) = \frac{(\operatorname{tr}N)^{2} + 2\operatorname{tr}(N^{2})}{4 \cdot6} = \frac{64 + 56}{24} = 5.

Thus (12) has zero mean, and ρε(v)=(1+4εϕ(v))1/4\rho_{\varepsilon}(v) = (1 + 4\varepsilon\phi(v))^{1/4} is an explicit defining formula for the family in dimension four. The parameter τ\tau is fixed first; ε\varepsilon is then chosen sufficiently small. Both smallness assertions are existence statements, with no numerical lower bound on their allowed ranges.

Proposition 4.3. There is a real-analytic family of positively curved convex bodies in R4\mathbb{R}^{4}, of volume κ4\kappa_{4}, converging smoothly to B4B^{4}, with no equipartition by four affine hyperplanes for all sufficiently small nonzero parameters.

Proof. Take A,BA, B from Theorem 4.1, and set

ϕ0(v)=A(v,v,v)+B(v,v,v,v),ϕ(v)=ϕ0(v)−1σ(S3)∫S3ϕ0 dσ.\phi_{0}(v) = A(v,v,v) + B(v,v,v,v), \qquad\phi(v) = \phi_{0}(v) - \frac{1}{\sigma(S^{3})}\int_{S^{3}} \phi_{0}\,d\sigma.

The subtracted constant has zero correlation with every nonempty orthogonal sign product, so TI(ϕ,q)=TI(ϕ0,q)T_{I}(\phi,q) = T_{I}(\phi_{0},q).

To compute these correlations, write v=∑i=14ziqiv = \sum_{i=1}^{4} z_{i}q_{i}. Multiplication by a triple sign product leaves, in a cubic polynomial, only the monomial with exponent one in the three specified coordinates. All other monomials integrate to zero by coordinate reflection. The quartic contribution to a triple correlation is zero by the antipodal map. Thus, for distinct i,j,ki,j,k,

T{i,j,k}(ϕ,q)=c3A(qi,qj,qk),c3=6∫S3∣z1z2z3∣ dσ(z)>0.(13)T_{\{i,j,k\}}(\phi,q) = c_{3}A(q_{i},q_{j},q_{k}), \qquad c_{3} = 6\int_{S^{3}} |z_{1}z_{2}z_{3}|\,d\sigma(z) > 0. \tag*{(13)}

For the fourfold correlation, only the quartic monomial with all four exponents one survives:

T[4](ϕ,q)=c4B(q1,q2,q3,q4),c4=24∫S3∣z1z2z3z4∣ dσ(z)>0.(14)T_{[4]}(\phi,q)=c_4B(q_1,q_2,q_3,q_4),\qquad c_4=24\int_{\mathbb{S}^3}|z_1z_2z_3z_4|\,\mathrm{d}\sigma(z)>0. \tag*{(14)}

The constants 6 and 24 are the numbers of permutations of the corresponding distinct arguments in the symmetric multilinear expansions.

Equations (13)–(14) and Theorem 4.1 show that the higher correlations have no common zero. Proposition 3.4 rules out all affine equipartitions of (1) for small nonzero parameters. Lemma 2.1 gives analyticity, positive curvature, smooth convergence and the exact volume. □

Higher dimensions and prescribed ellipsoids

A finite-dimensional genericity argument

Let Pd0\mathcal{P}_d^0 be the finite-dimensional space of restrictions to Sd−1\mathbb{S}^{d-1} of real polynomials of degree at most dd, with zero spherical mean. No choice of polynomial representation is needed; this is a space of functions on the sphere.

Proposition 5.1. If d≥5d\ge5, there is an open subset of full Lebesgue measure in Pd0\mathcal{P}_d^0 whose elements have no common zero in (11). If d≥6d\ge6, the same assertion holds in the even subspace of Pd0\mathcal{P}_d^0, using only even index sets ∣I∣≥4|I|\ge4. Lebesgue measure here means Lebesgue measure after any linear identification of the relevant function space with a Euclidean space.

Proof. Put I={I⊂[d]:∣I∣≥3}\mathcal{I}=\{I\subset[d]:|I|\ge3\} and consider

T:Pd0×O(d)⟶RI,T(ϕ,q)=(TI(ϕ,q))I∈I.T:\mathcal{P}_d^0\times O(d)\longrightarrow\mathbb{R}^{\mathcal{I}},\qquad T(\phi,q)=(T_I(\phi,q))_{I\in\mathcal{I}}.

Changing variables v=qzv=qz gives

TI(ϕ,q)=∫Sd−1ϕ(qz)∏i∈Isgn⁡zi dσ(z).T_I(\phi,q)=\int_{\mathbb{S}^{d-1}}\phi(qz)\prod_{i\in I}\operatorname{sgn}z_i\,\mathrm{d}\sigma(z).

Since ϕ\phi is a polynomial restriction, this is smooth jointly in its coefficients and in qq, despite the sign functions.

For a fixed qq and J∈IJ\in\mathcal{I}, use the test polynomial

pJ,q(v)=∏j∈J(qj⋅v).p_{J,q}(v)=\prod_{j\in J}(q_j\cdot v).

Its mean is zero and its degree is at most dd. Coordinate reflection gives

TI(pJ,q,q)={0,I≠J,∫Sd−1∏j∈J∣zj∣ dσ(z)>0,I=J.(15)T_I(p_{J,q},q)= \begin{cases} 0, & I\ne J,\\ \int_{\mathbb{S}^{d-1}}\prod_{j\in J}|z_j|\,\mathrm{d}\sigma(z)>0, & I=J. \end{cases} \tag*{(15)}

Therefore the derivative of TT in the polynomial variable is surjective at every (ϕ,q)(\phi,q). Its zero set ZZ is a smooth submanifold of codimension

r=∣I∣=2d−1−d−(d2).r=|\mathcal{I}|=2^d-1-d-\binom{d}{2}.

For d≥5d\ge5,

r>dim⁡O(d)=(d2),equivalently2d>d2+1.(16)r>\dim O(d)=\binom{d}{2},\qquad\text{equivalently}\qquad2^d>d^2+1. \tag*{(16)}

The inequality holds at d=5d=5, and its strictness persists on increasing dd. The projection Z→Pd0Z\to\mathcal{P}_d^0 has lower-dimensional domain than target. Its image has Lebesgue measure zero by Sard’s theorem, applied on a countable atlas. That image is also closed: from a convergent sequence of polynomials in the image, extract a convergent sequence of frames using compactness of O(d)O(d). Its complement is therefore open and has full measure.

For even polynomials, let Iev={I⊂[d]:∣I∣≥4, ∣I∣ even}I_{\mathrm{ev}}=\{I\subset[d]: |I|\ge4,\ |I|\ \text{even}\}. The polynomials pJ,Qp_{J,Q} are now even, and the same diagonal test proves surjectivity. The target dimension is

rev=2d−1−1−(d2)>(d2)(d≥6).r_{\mathrm{ev}}=2^{d-1}-1-\binom{d}{2}>\binom{d}{2}\qquad(d\ge6).

At d=6d=6 this is 16>1516>15; the equivalent inequality 2d−1>d(d−1)+12^{d-1}>d(d-1)+1 persists for larger dd. The identical projection argument proves the assertion.

Affine transfer and proof of the main theorem

Lemma 5.2. An invertible affine map preserves the existence of an equipartition by dd affine hyperplanes. It preserves real-analytic boundaries and positive principal curvatures of convex bodies. For a fixed such map, it also preserves smooth convergence of radial functions about the corresponding interior points.

Proof. Let x=c+Ayx=c+Ay. The inverse image of a hyperplane is a hyperplane, and all sign-region volumes are multiplied by the same number ∣det⁡A∣|\det A|. For a smooth boundary parametrization, the second fundamental form after the change of variables is the original positive form divided by a positive normalizing factor. This proves preservation of positive curvature; analyticity follows from the analytic change of variables. If KK is radial about zero, then

ρc+AK,c(v)=ρK,0(A−1v/∣A−1v∣)∣A−1v∣.\rho_{c+AK,c}(v)=\frac{\rho_{K,0}(A^{-1}v/|A^{-1}v|)}{|A^{-1}v|}.

The sphere map and its positive denominator are fixed and smooth, so the formula proves the convergence assertion.

Proof of Theorem 1.1. For d=4d=4, Proposition 4.3 provides the required mean-zero polynomial perturbation of the unit ball. For d≥5d\ge5, choose a polynomial from Proposition 5.1. If d≥6d\ge6, choose it in the even subspace. Proposition 3.4 excludes equipartitions for all sufficiently small nonzero parameters, and Lemma 2.1 supplies the remaining properties.

Write the prescribed ellipsoid as E=c+ABdE=c+AB^d, with AA invertible, and apply x=c+Ayx=c+Ay to the constructed family. Lemma 5.2 gives the assertions about equipartitions, curvature and smooth convergence. Equation (2) and the zero mean give the exact volume of EE. An even perturbation gives symmetry about cc.

Proof of Corollary 1.2. The one-dimensional assertion follows by bisecting an interval. In the plane, bisect the body by one line and apply the ham-sandwich theorem to its two parts to find a second line bisecting both. Hadwiger’s theorem [4] gives the assertion in dimension three. Theorem 1.1 supplies counterexamples in each dimension d≥4d\ge4.

Orthogonal equipartitions near ellipsoids

An affine map need not preserve orthogonality. The distinct axes of the reference ellipsoid allow us to control precisely which orthogonal configurations can survive in the limit.

Proof of Theorem 1.3. Translations and orthogonal transformations preserve the problem. We may therefore assume that the center is zero and

E=ABd,A=diag⁡(a1,…,ad),ai>0,ai≠aj (i≠j).E=AB^d,\qquad A=\operatorname{diag}(a_1,\ldots,a_d),\qquad a_i>0,\qquad a_i\ne a_j\ (i\ne j).

For the first assertion take ϕ(v)=v1v2v3\phi(v)=v_1v_2v_3. This polynomial has zero mean. Construct KεK_\varepsilon by (1), and consider AKεAK_\varepsilon. Suppose that along a sequence of nonzero parameters tending to zero, AKεmAK_{\varepsilon_m} admits an orthogonal equipartition. Its unit normals form an orthogonal matrix Um=(u1m,…,udm)U^m=(u_1^m,\ldots,u_d^m). Pulling back its hyperplanes under AA gives an equipartition of KεmK_{\varepsilon_m}, with unit normals

vim=Auim∣Auim∣.v_i^m=\frac{Au_i^m}{\lvert Au_i^m\rvert}.

The offsets are divided by the same positive normalization factors. Take a subsequence Um→U∈O(d)U^m\to U\in O(d). Proposition 3.4 says that the limiting viv_i are orthonormal and that their higher sign correlations with ϕ\phi all vanish.

For i≠ji\ne j, their orthogonality implies

(ui)TA2uj=0.(u_i)^T A^2u_j=0.

Thus UTA2UU^T A^2U is diagonal. Since the eigenvalues ai2a_i^2 are distinct, the columns of UU are a signed permutation of the coordinate vectors. Positive diagonal normalization leaves these vectors unchanged, so vv is that same signed coordinate frame.

Choose the three indices II whose limiting normals correspond to the first three coordinate axes. Then

TI(ϕ,v)=±∫Sd−1∣z1z2z3∣ dσ(z)≠0,T_I(\phi,v)=\pm\int_S^{d-1}\lvert z_1z_2z_3\rvert\,d\sigma(z)\ne0,

contradicting Proposition 3.4. This excludes all orthogonal equipartitions for sufficiently small nonzero parameters.

For d≥4d\ge4, take instead ϕ(v)=v1v2v3v4\phi(v)=v_1v_2v_3v_4. It is even and has zero mean. The same argument uses the four indices corresponding to those axes and the nonzero integral of ∣z1z2z3z4∣\lvert z_1z_2z_3z_4\rvert. The resulting body is centrally symmetric. Lemma 2.1, scaling by AA, and the radial formula in Lemma 5.2 prove all geometric and volume assertions. Finally undo the initial translation and rotation. □

Hausdorff stability and polytopal examples

Lemma 7.1. Within the space of convex bodies with the Hausdorff metric, the bodies admitting an equipartition by dd affine hyperplanes form a closed set. The same is true when the hyperplanes are required to be mutually perpendicular.

Proof. Suppose Km→KK_m\to K in Hausdorff distance and KK has nonempty interior. All bodies lie in one ball of radius RR. For an equipartition (um,tm)(u^m,t^m), each hyperplane bisects a body of positive volume and therefore intersects that ball. Consequently ∣tim∣≤R\lvert t_i^m\rvert\le R. The normals and offsets have a convergent subsequence, say (um,tm)→(u,t)(u^m,t^m)\to(u,t).

Indicators of Hausdorff-convergent convex bodies with an interior limit converge almost everywhere; their boundaries have Lebesgue measure zero. Outside the finitely many limiting hyperplanes, the sign products converge as well. Dominated convergence in a common bounding ball gives

WI(Km;um,tm)⟶WI(K;u,t)(I⊂[d]).W_I(K_m;u^m,t^m)\longrightarrow W_I(K;u,t)\qquad(I\subset[d]).

Every nonempty coefficient on the left is zero. The limiting configuration therefore equipartitions KK, by Lemma 2.2. If the normals were orthonormal, their limits remain orthonormal. No independence assumption on a general limiting configuration is needed. □

Proof of Corollary 1.4. By Lemma 7.1, the absence of an equipartition is open. Approximate a counterexample in Theorem 1.1 or Theorem 1.3 by convex polytopes in Hausdorff distance. Such approximations are standard for convex bodies [7]; in the centrally symmetric case, take convex hulls of finite antipodally paired boundary nets. All sufficiently close approximants retain the absence of the corresponding equipartition.

Their volumes converge to the desired volume. Dilating each approximant about the reference center by a factor tending to one makes the volumes exactly equal. Positive dilations preserve both ordinary and orthogonal equipartitionability, and preserve central symmetry. First choosing the smooth counterexample close to the reference ellipsoid and then choosing the polytopal approximation proves the stated proximity. □

Remark 7.2. The smooth approximations in the main theorems are in every fixed CkC^k norm. The polytopal approximation is in Hausdorff distance. The central-symmetry assertion for unrestricted affine hyperplanes is proved here only for d≥6d \ge6; the even-polynomial dimension count used above does not establish it in dimensions four or five.

Use of artificial intelligence

ChatGPT assisted with mathematical exploration, checking arguments, organizing the exposition, and preparing the LaTeX source. The mathematical claims, the attribution of prior work, and the final manuscript are the responsibility of the author. The four-dimensional proof invokes Soberón’s cited tensor nonvanishing theorem; the present article introduces no numerical certificate or computer-assisted existence step.

Chongqing Bashu Science City Secondary School, Chongqing, China Email address: [email protected]

References

  1. [1]D. Avis, Non-partitionable point sets, Inform. Process. Lett. 19 (1984), no. 3, 125–129. https://doi.org/10.1016/0020-0190(84)90090-5.
  2. [2]P. V. M. Blagojević, F. Frick, A. Haase and G. M. Ziegler, Topology of the Grünbaum–Hadwiger–Ramos hyperplane mass partition problem, Trans. Amer. Math. Soc. 370 (2018), no. 10, 6795–6824. https://arxiv.org/abs/1502.02975.
  3. [3]B. Grünbaum, Partitions of mass-distributions and of convex bodies by hyperplanes, Pacific J. Math. 10 (1960), no. 4, 1257–1261. https://doi.org/10.2140/pjm.1960.10.1257.
  4. [4]H. Hadwiger, Simultane Vierteilung zweier Körper, Arch. Math. (Basel) 17 (1966), 274–278.DOI
  5. [5]G. L. Maldonado and E. Roldán-Pensado, On the orthogonal Grünbaum partition problem in dimension three, Comput. Geom. 126 (2025), Paper No. 102149. https://doi.org/10.1016/j.comgeo.2024.102149.
  6. [6]E. Roldán-Pensado and P. Soberón, A survey of mass partitions, Bull. Amer. Math. Soc. (N.S.) 59 (2022), no. 2, 227–267. https://doi.org/10.1090/bull/1725.
  7. [7]R. Schneider, Convex Bodies: The Brunn–Minkowski Theory, second expanded edition, Encyclopedia of Mathematics and its Applications 151, Cambridge University Press, Cambridge, 2014.
  8. [8]P. Soberón, Four hyperplanes do not always equipartition a mass in ℝ⁴, arXiv:2608.23312v2, 2026. https://arxiv.org/abs/2608.23312v2.
  9. [9]R. T. Živaljević, Topological methods in discrete geometry, in Handbook of Discrete and Computational Geometry, third edition, J. E. Goodman, J. O’Rourke and C. D. Tóth (eds.), CRC Press, Boca Raton, FL, 2017, 551–580. https://www.csun.edu/~ctoth/Handbook/chap21.pdf.

Paper details

vlnc:W_w9z99pzh
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

OpenAI ChatGPT GPT 6.1 Sol

View statement
Looking for related papers…
Contents