Zong's problem on 2d half-spaces containing the unit ball

arXiv:2609.38394 · math.FA · Submitted 2026-09-29 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.

Ada: We're looking at the paper: "Zong's problem on 2d half-spaces containing the unit ball".

Emmy: Here's the gist of the paper. The paper proves that the intersection of two halfspaces containing the Euclidean unit ball in R d contains a point of norm at least √d, with equality characterizing the cube <ref:2609.38394#pg2>.

Ada: Theorem 1.1 states that if H1,..., H2d be closed halfspaces containing Bd, then P = ⋂︀ i∈

2d: Hi contains a point x with x ≥ √d <ref:2609.38394#pg2> Moreover, if maxx∈P x = √d, then P is an orthogonal image of

−1, 1: d <ref:2609.38394#pg2>.

Emmy: This result resolves a problem proposed by C.

Ada: Zong in 1993 <ref:2609.38394#pg2>.

Title and authors: Ada: We're talking about the title and authors now for "Zong's problem on 2d half-spaces containing the unit ball <ref:2609.38394#pg1,half-spaces containing the unit ball>." It’s by Grigory Ivanov, and this is a problem that C Zong proposed back in nineteen ninety-three.

Emmy: The structure of the paper starts by setting up m half-spaces, where m is the number of constraints. The authors are looking at the case where you have 2d such half-spaces, and they're trying to find a lower bound on a quantity related to those constraints <ref:2609.38394#pg1>.

Kolya: What's the actual goal of this paper? Is it just proving that there is a point outside the unit ball? Or is it about finding a specific type of point inside?

Ada: The main result, Theorem one point one, says that if you take 2d closed half-spaces that contain the unit ball in R d, their intersection has a point with a norm of at least sqrt d <ref:2609.38394#pg1>. That's the core claim we need to focus on.

Emmy: And they add a characterization part. If that maximum norm you find is exactly sqrt d, then you know for sure that those half-spaces are in fact an orthogonal image of the unit hypercube, which is a very rigid geometric constraint.

The paper's summary: Ada: So, to summarize this whole problem, the paper tackles finding a point x inside an intersection P formed by many half-spaces that all contain the unit ball. They establish that this intersection P must have a point with norm at least sqrt d.

Emmy: The proof relies on looking at something called the Gaussian vertex measure of the polytope. They define this measure, which assigns a weight to each vertex based on its normal cone, and they look at its second moment operator.

Kolya: And they use this second moment operator to give us a lower bound for the average squared norm of these vertices. Specifically, Theorem one point two gives us that this trace is greater than or equal to d plus some term involving pi/four and the number of distinct facet normals minus m <ref:2609.38394#pg1>.

Ada: The way they get that bound is through a Gaussian inequality, which they state in Theorem two point one, and this inequality involves comparing mass on a line segment with the centroid of its restriction to that line <ref:2609.38394#pg1>.

The paper's improvements: Emmy: The authors suggest some ways to improve how we look at the problem, specifically by considering a quadratic extension. This leads into some open questions about tight frames and vertex indices in higher dimensions.

Ada: They also propose a conjecture, Conjecture five point one, which asks about the sum of norms of vectors b one through b 2d that contain the unit ball within their convex hull. It’s an interesting way to think about how these constraints relate to spanning sets.

Kolya: From a probabilistic view, what does this quadratic extension suggest about the tail behavior of the distribution we're looking at? Does it help us understand when these bounds are tight or when they become loose?

Emmy: They also give an independent spectral estimate in Section six showing that for a specific case where you have m constraints equal to 2d, you get a point with a norm of at least something like six point six times the square root of two times the natural logarithm of two times the square root of d.

Conclusion: Ada: So, to wrap up, this paper, "Zong's problem on 2d half-spaces containing the unit ball," confirms that you can always find a point in such an intersection with a norm at least sqrt d <ref:2609.38394#pg1,half-spaces containing the unit ball>. They also give us a very specific geometric condition for when that point achieves that exact sqrt d norm.

Emmy: The main implication is the connection they draw between these geometric constraints and the structure of orthogonal images of hypercubes. It links the geometry of half-spaces directly to combinatorial properties of their boundaries.

Kolya: From a probabilistic standpoint, it’s useful because it shows that even with many constraints, you still get a certain amount of "stuff" in the intersection, and that bound is tied to the dimension d. The fact that they used Gaussian measures is important for understanding the underlying probability distribution.

Ada: Indeed. This work settles a problem proposed by C Zong from nineteen ninety-three regarding these half-spaces, giving us a solid result about minimum norm bounds in this high dimensional setting.

Emmy: It’s interesting to see how they use Gaussian measures to establish these sharp combinatorial bounds, and it sets up some future questions about vertex indices in more complex settings.

math.FA

Submitted: 2026-09-29

Updated: 2026-11-16

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 80/100

The gist: The paper proves that the intersection of two halfspaces containing the Euclidean unit ball in R d contains a point of norm at least √d, with equality characterizing the cube <ref:2609.38394#pg2>.

Key concepts

Gaussian vertex measure
This assigns a specific weight ($ u_ au(v)$) to each vertex ($v$) of the intersection polytope $P$. It is used to calculate the average squared norm of the vertices, which is a key quantity in bounding the minimum norm achieved within $P$.
Second moment operator ($A_ au(P)$)
This operator acts on a set $P$ and involves summing up weighted terms related to its vertices. Its trace, $ ext{trace } A_ au(P)$, represents the average squared norm of the vertices, providing a lower bound for the minimum norm point in $P$.
Gaussian inequality
This is a fundamental inequality used in the proof. It relates a symmetric positive definite matrix $G$ (with diagonal entries equal to 1) to its trace, which helps establish bounds on geometric quantities related to the set $P$.
Orthogonal image of $[-1, 1]^d$
This condition describes a specific geometric structure for the intersection polytope $P$. If the maximum norm point in $P$ has norm exactly $\sqrt{d}$, then $P$ must be an orthogonal transformation of the standard hypercube $[-1, 1]^d$. This provides a precise characterization of the extremal case.

Terminology

Summary

The paper proves that the intersection of two halfspaces containing the Euclidean unit ball in R d contains a point of norm at least √d, with equality characterizing the cube <ref:2609. p. 038394#pg2>.

Main Result

Theorem 1.1 states that if H1,..., H2d be closed halfspaces containing Bd, then P = ⋂ i∈[2d] Hi contains a point x with x ≥ √d <ref:2609. p. 038394#pg2> Moreover, if maxx∈P x = √d, then P is an orthogonal image of [−1, 1]d <ref:2609. p. 038394#pg2>. This result resolves a problem proposed by C. p. 0 Zong in 1993 <ref:2609. p. 038394#pg2>.

Vertex Measure and Bounds

The proof is based on estimating the average squared norm of the vertices with weights given by the Gaussian measures of their normal cones <ref:2609.38394#pg2>. Definition 1. p. 01 defines the Gaussian vertex measure of P, assigning weight nugamma(v) to v, and its second moment operator is Agamma(P) <ref:2609. p. 038394#pg2>. The trace of Agamma(P) is the average squared norm of this vertex <ref:2609. p. 038394#pg2>. Theorem 1.2 provides a lower bound for this quantity, stating trace Agamma(P) = ∑ v∈vert(P) nugamma(v)v 2 ≥ d + π/4 (2d − m) <ref:2609. p. 038394#pg2>.

Gaussian Inequality Derivation

The proof of Theorem 1.2 relies on the Gaussian inequality, which is stated in Theorem 2. p. 11 <ref:2609. p. 0 p. 0 p. 038394#pg2>. This inequality involves a symmetric positive definite matrix G with Gii = 1 for i ∈ [d], and it relates Ψ(G) to t(G) <ref:2609. p. 138394#pg2>. The proof of Theorem 2. p. 1 p. 11 uses a one-dimensional comparison, relating the mass on the positive half-line, the centroid of the normalized restriction to that half-line, and its density at the boundary <ref:2609. p. 338394#pg2>.

Quadratic Extension and Spectral Estimates

Section 5 discusses a quadratic extension which leads to open questions regarding tight frames and vertex indices <ref:2609.38394#pg2>. Conjecture 5. p. 71 concerns the sum of norms of vectors b1,. p. 9.., b2d such that Bd ⊆ conv(b i: i ∈ [2d]) <ref:2609. p. 038394#pg2>. The paper also presents an independent spectral estimate in Section 6, showing that for m = 2d the intersection contains a point of norm at least (6. p. 06) √2 ln 2 √d > 0.98√d <ref:2609. p. 0 p. 038394#pg2>.

Final Bounds

The final result in Section 6 shows that the intersection of 2d halfspaces containing Bd consequently contains a point of norm at least (6.6) √2 ln 2 √d > 0.98√d <ref:2609.38394#pg2>. This bound is derived from the Cauchy–Schwarz inequality and Theorem 6. p. 41 <ref:2609.38394#pg2>. The quadratic estimate (6. p. 05) applies to arbitrary Q, and for a scalar Q and m = 2d, Theorem 6. p. 31 gives the conjectured constant 2d squared <ref:2609. p. 038394#pg2>. This leads to the combinatorial dual inequality in Corollary 6. p. 22 <ref:2609. p. 038394#pg2>. The measure obtained by optimizing the second moment operator is not asserted to be the Gaussian vertex measure <ref:2609. p. 038394#pg2>.

References

[AAGM15] Shiri Artstein-Avidan, Apostolos Giannopoulos, and Vitali D. Milman. Asymptotic geometric analysis, Part I, volume 202. p. 1 American Mathematical Soc., 2015 <ref:2609. p. 038394#pg2>

[AHAK22] V´ıctor Hugo Almendra-Hern´andez, Gergely Ambrus, and Matthew Kendall. Quantitative Helly-type theorems via sparse approximation. p. 8 Discrete & Computational Geometry, pages 1–8, 2022 <ref:2609. p. 038394#pg2>

[And03] Theodore W. Anderson. An Introduction to Multivariate Statistical Analysis. p. 0 Wiley Series in Probability and Statistics, John Wiley & Sons, Hoboken, NJ, 3rd edition, 2003 <ref:2609. p. 038394#pg2>

[Bar08] Arp´ad Baricz. Mills’ ratio: Monotonicity patterns and functional inequalities. p. 3 Journal of Mathematical Analysis and Applications, 340(2):1362–1370, 2008 <ref:2609. p. 038394#pg2>

[BL07] Karoly Bezdek and Alexander E. Litvak. On the vertex index of convex bodies. p. 7 Advances in Mathematics, 215(2):626–641, 2007 <ref:2609. p. 038394#pg2>

[B¨or04] K´aroly B¨or¨oczky, Jr. Finite packing and covering, volume 154 of Cambridge Tracts in Mathematics. p. 0 Cambridge University Press, Cambridge, 2004 <ref:2609. p. 038394#pg2>

[BP09] K. M. Ball and M. p. 0 Prodromou. A sharp combinatorial version of Vaaler's theorem. Bulletin of the London Mathematical Society, 41(5):853–858, August 2009 <ref:2609. p. 038394#pg2>

[DLMLZ00] Leoni Dalla, David G. Larman, Peter Mani-Levitska, and Chuanming Zong. p. 0 The blocking numbers of convex bodies. p. 0 Discrete & Computational Geometry, 24(2):267–278, 2000 <ref:2609. p. 038394#pg2>

[FT72] L´aszl´o Fejes T´oth. Lagerungen in der Ebene auf der Kugel und im Raum, volume 65 of Grundlehren der mathematischen Wissenschaften. Springer-Verlag, Berlin–Heidelberg, second edition, 1972 <ref:2609. p. 038394#pg2>

[GL12] Efim D. Gluskin and Alexander E. p. 9 Litvak. A remark on vertex index of the convex bodies. p. 7 In Geometric Aspects of Functional Analysis: Israel Seminar 2006–2010, pages 255–265. p. 0 Springer, 2012 <ref:2609. p. 038394#pg2>

[HJ12] Roger A. Horn and Charles R. Johnson. Matrix analysis. p. 1 Cambridge university press, 2012 <ref:2609. p. 038394#pg2>

[IN22] Grigory Ivanov and M´arton Nasz´odi. A quantitative Helly-type theorem: containment in a homothet. p. 8 SIAM Journal on Discrete Mathematics, 36(2):951–957, 2022 <ref:2609. p. 038394#pg2>

[Iva21] Grigory Ivanov. On the volume of projections of the cross-polytope. p. 1 Discrete Mathematics, 344(5):112312, 2021 <ref:2609. p. 038394#pg2>

[Iva26] Grigory Ivanov. The maximal volume of projections of the cross-polytope. p. 1 arXiv preprint arXiv:2607. p. 012072, 2026 <ref:2609. p. 038394#pg2>

[Pr´e71] Andr´as Pr´ekopa. Logarithmic concave measures with application to stochastic programming. p. 1 Acta Scientiarum Mathematicarum, 32:301–316, 1971 <ref:2609. p. 038394#pg2>

[Sam53] Michael R. Sampford. Some inequalities on Mill’s ratio and related functions. p. 1 The Annals of Mathematical Statistics, 24(1):130–132, 1953 <ref:2609. p. 038394#pg2>

[Tal61] Georges M. Tallis. The moment generating function of the truncated multi-normal distribution. p. 0 Journal of the Royal Statistical Society Series B: Statistical Methodology, 23(1):223–229, 1961 <ref:2609. p. 038394#pg2>

[Ver18] Roman Vershynin. High-dimensional probability: An introduction with applications in data science, volume 47. p. 8 Cambridge university press, 2018 <ref:2609. p. 038394#pg2>

Pruning redundant reasoning tokens

The proof of Theorem 1.1 moves each bounding hyperplane inward until it is tangent to Bd <ref:2609. p. 138394#pg2>. The resulting polytope P0 ⊆ P is bounded and has m ≤ 2d distinct facet normals <ref:2609. p. 138394#pg2>. Theorem 1.2 gives maxx∈P x 2 ≥ trace Agamma(P0) ≥ d + π/4 (2d − m) ≥ d <ref:2609. p. 038394#pg2>. If maxx∈P x = √d, all inequalities are equalities <ref:2609. p. 138394#pg2>.

How the training signal is built

The proof of Theorem 1.2 uses direct calculations with Gaussian measures <ref:2609. p. 138394#pg2>. The main technical inequality is stated as Theorem 2. p. 11 below <ref:2609. p. 038394#pg2>. This inequality involves the angle-sum approach proposed in [DLMLZ00, Remark 2] <ref:2609. p. 138394#pg2>. The proof of the Gaussian vertex bound uses a log-concave mixture of standard Gaussian translates, to which Lemma 3. p. 51 applies <ref:2609. p. 038394#pg2>.

Where the gains come from

The main technical inequality (2.3) is proven using classical inequalities like (3. p. 32) and the properties of log-concave mixtures <ref:2609. p. 338394#pg2>. The proof of Theorem 2.1 involves Gaussian integration by parts, leading to the identity (4. p. 65) which sums up to the quadratic form defined by G <ref:2609. p. 538394#pg2>. The final bound in Section 6 is derived from Proposition 6. p. 91 <ref:2609.38394#pg2> and the Cauchy–Schwarz inequality <ref:2609. p. 438394#pg2>. The combinatorial dual inequality in Corollary 6. p. 22 uses the measure obtained by optimizing the second moment operator <ref:2609. p. 038394#pg2>. The final result is established by showing that every off-diagonal entry of G is zero, implying G = Idd <ref:2609. p. 738394#pg2>.

Corollary 6.1

Under the assumptions of Theorem 6.1, (6.5) S(C, Q) trace Q−1 ≥ η squared m,d ≥ m squared ln2 m / (m − d) <ref:2609. p. 038394#pg2>. In particular, for m = 2d the right-hand side is at least 2 ln 2 d <ref:2609. p. 9 p. 938394#pg2>.

Corollary 6.2

There is a probability measure ν on vert(C) such that (6.7) trace √ ∑ v∈vert(C) ν(v) v ⊗ v ≥ η squared m,d ≥ m ln m / (m − d) <ref:2609. p. 038394#pg2>. In particular, for m = 2d the right-hand side is at least 2 ln 2 d <ref:2609.38394#pg2>.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved AI can solve problems related to finding a point of norm at least sqrt d in 2d halfspaces containing a unit ball by using the result that the intersection of 2d halfspaces containing the Euclidean unit ball in R d contains a point of norm at least sqrt d.

  2. Improved AI can determine if a set of halfspaces is an orthogonal image of [-1, 1]d by checking if max x in P |x| = sqrt d and verifying the conditions for equality in Theorem 1.2.

  3. Improved AI can estimate the average squared norm of vertices weighted by Gaussian measures for a polytope P using the formula trace A gamma(P) = sum over v in vert(P) gamma n(v) v tensor v.

  4. Improved AI can bound trace A gamma(P) for a bounded polytope P defined by halfspaces with facet normals a1,..., am as d + pi/4 (2d - m), where m is the number of distinct facet normals.

  5. Improved AI can analyze quadratic forms and measures on vertices for bounded full-dimensional polytopes containing the origin in its interior, relating them to the quantity trace sqrt A v sum over v in vert(C) gamma n(v) v tensor v.

  6. Improved AI can find a probability measure on vert(C) such that trace sqrt L equals min L over Q positive definite, where L is an ellipsoid containing C and trace L = S(C, Q) trace Q-1.

Sources

Related papers