[Paper Review] Sphere Packings in 3 Dimensions

Thomas Hales|ArXiv.org|May 19, 2002

Computational Geometry and Mesh Generation4 references3 citations

TL;DR

This paper presents a second-generation proof of the Kepler conjecture, proposing a simplified approach using a function $ f_\lambda $ that combines Voronoi cell volumes with correction terms to bound packing density. By formulating a conjecture that $ f_\lambda \geq \nu_0 $—where $ \nu_0 $ is the volume of the rhombic dodecahedron—the method reduces the problem to verifying inequalities via multi-simplex optimization, interval arithmetic, and generative programming, aiming for a more automated and generalizable proof than the original 1998 proof.

ABSTRACT

This short note describes the tentative form of a finite-dimensional optimization problem that may be of use in a second-generation proof of the Kepler conjecture. In the original 1998 proof of the Kepler conjecture, the form of the optimization problem was constrained by limits to computer power and by the speed of the algorithms that were available in 1994 to prove inequalities by computer. The computational resources have changed considerably since then, and much has been learned by an analysis of the 1998 proof. This analysis has lead to the proposed new form for an optimization problem.

Motivation & Objective

• To develop a simplified, automated, and generalizable proof of the Kepler conjecture in three dimensions.
• To replace the 270-page 1998 proof with a more structured and algorithmic approach based on inequality verification.
• To enable broader application to related problems in discrete geometry, such as the kissing number problem or Kelvin’s problem.
• To leverage generative programming and multi-simplex optimization to automate repetitive proof components.
• To establish a finite, decidable framework for verifying the conjecture using algebraic and numerical methods.

Proposed method

• Define a function $ f_\lambda $ at each ball center $ \lambda $, combining the Voronoi cell volume with correction terms $ \epsilon_\lambda $.
• The correction term $ \epsilon_\lambda $ includes sums over triangles $ T $ with edge-length constraints $ \leq r \in [2, \sqrt{8}] $, using a quadratic polynomial $ L $ to adjust edge-length discrepancies.
• Introduce a second correction term involving triangles $ S $ with a distinguished edge of length $ w(S) $, weighted by coefficients $ \mu_{S,\lambda} $ and scaled by a constant $ M $.
• Use multi-simplex optimization to automate inequality proving via linear programming and interval arithmetic, reducing reliance on manual case analysis.
• Apply generative programming techniques to produce reusable code for inequality verification, improving scalability and maintainability.
• Leverage Tarski’s decidability result to ensure the conjecture is finitely verifiable, even with high-dimensional parameter spaces.

Experimental results

Research questions

• RQ1 Can the Kepler conjecture be re-proven using a function-based approach that bounds Voronoi cell volumes with correction terms?
• RQ2 How can interval arithmetic and linear programming be systematically combined to automate inequality verification in dense sphere packing?
• RQ3 To what extent can generative programming techniques simplify and modularize the proof process in discrete geometry?
• RQ4 Can the conjecture that $ f_\lambda \geq \nu_0 $ be verified efficiently using modern computational methods?
• RQ5 What is the minimal set of geometric and algebraic conditions needed to ensure the average Voronoi cell size matches that of the face-centered cubic packing?

Key findings

• Conjecture 1—that $ f_\lambda \geq \nu_0 $ for a suitable choice of $ L $, $ M $, and $ r $—implies the Kepler conjecture due to cancellation of correction terms across the packing.
• The correction terms $ \epsilon_\lambda $ are designed so that their average over all centers vanishes, preserving the total volume bound.
• The method reduces the 270-page 1998 proof to a finite, decidable problem using Tarski’s theorem, though practical verification requires advanced algorithms.
• Multi-simplex optimization enables automated verification of inequalities across complex geometric configurations, replacing manual case analysis.
• The use of generative programming allows for code generation tailored to inequality proving, increasing reproducibility and reducing proof complexity.
• The approach provides a framework that could be extended to other problems in discrete geometry, such as the kissing number in 4D or Kelvin’s problem.