[Paper Review] On Continuity Properties for Infinite Rectangle Packing

Zhiheng Liu|arXiv (Cornell University)|May 6, 2017

Optimization and Packing Problems5 references3 citations

TL;DR

This paper establishes continuity properties for infinite rectangle packing by proving that the optimal packing efficiency $\eta_0(A)$ varies continuously with rectangle dimensions, enabling finite-step verification of negative results for perfect packing. It introduces retraction and extension operations to bound changes in bounding area, showing that packing efficiency cannot drop below a positive lower bound when rectangles expand, thus providing a framework to rule out perfect packings algorithmically.

ABSTRACT

By rectangle packing we mean putting a set of rectangles into an enclosing rectangle, without any overlapping. We begin with perfect rectangle packing problems, then prove two continuity properties for parallel rectangle packing problems, and discuss how they might be used to obtain negative results for perfect rectangle packing problems.

Motivation & Objective

• To establish continuity properties of optimal packing efficiency $\eta_0(A)$ in infinite rectangle packing problems.
• To develop tools—retraction and extension operations—that control changes in bounding area $T_0(A)$ under dimension perturbations.
• To provide a method for determining whether perfect packing is impossible by estimating $\eta_0(A)$ with error bounds.
• To show that packing efficiency remains bounded away from zero under rectangle expansion, offering a structural barrier to perfect packing.

Proposed method

• Uses retraction and extension operations on rectangle positionings to analyze how changes in rectangle widths affect the bounding area $T(A_M)$.
• Defines $T_0(A) = \inf T(A_M)$ as the minimal bounding area over all valid positionings, equivalent to $S(A)/\eta_0(A)$.
• Applies Meir and Moser's theorem [2] to embed remaining rectangles into a box, bounding the error in $T_0(A^n)$ as $n \to \infty$.
• Derives bounds on $T_0(A^n)$ using $R_n = \sum_{k=n+1}^\infty l_k^2 \to 0$, ensuring convergence of the infinite series packing.
• Uses the inequality $\eta_0(A') \geq \ell_1 / q(A_0)$ for small positive width perturbations $\Delta x > 0$, establishing a universal lower bound.
• Leverages scaling invariance of $\eta_0$ to extend local efficiency estimates to nearby configurations in the parameter space.

Experimental results

Research questions

• RQ1 Can the optimal packing efficiency $\eta_0(A)$ be continuously extended to infinite rectangle sets with convergent total area?
• RQ2 What is the effect of small perturbations in rectangle dimensions on the minimal bounding area $T_0(A)$?
• RQ3 Can the continuity of $\eta_0(A)$ be used to algorithmically rule out perfect packings for certain rectangle families?
• RQ4 Is there a positive lower bound on $\eta_0(A)$ when rectangle dimensions are slightly increased, regardless of initial efficiency?
• RQ5 How can error bounds on $\eta_0(A)$ be constructed from finite subpackings $A^n$ to infer properties of the full infinite system?

Key findings

• The optimal packing efficiency $\eta_0(A)$ is continuous with respect to rectangle dimensions: $T_0(A^n) \to T_0(A)$ as $n \to \infty$ under $\ell^2$-summability of lengths.
• For any rectangle family $A$, the packing efficiency satisfies $\eta_0(A) \geq \ell_1 / q(A_0)$, providing a universal lower bound under width expansion.
• When a rectangle's width increases by $\Delta x > 0$, the new efficiency satisfies $\eta_0(A + \Delta A) \geq \ell_1 / q(A_0)$, even if efficiency decreases.
• The difference $\eta_0(A) - \eta_0(A + \Delta A)$ is bounded by $\eta_0(A) - \ell_1 / q(A_0)$, ensuring efficiency does not vanish under perturbation.
• The convergence of $T_0(A^n)$ to $T_0(A)$ is established via embedding tail rectangles into a box of size $b \times a$, with $a = 4\sqrt{R_n}$, yielding error $O(\sqrt{R_n})$.
• The method allows finite-step verification of non-perfect packing: if $\eta_0(A^n)$ is bounded away from 1 with controlled error, perfect packing is ruled out.