Sorting as Gradient Flow on the Permutohedron

TL;DR

This paper reframes sorting as a gradient-flow process on the permutohedron, offering a new continuous-time view that explains how structure contracts the exponential permutation space. The main result shows that the gradient flow toward the sorted vertex induces exponential decay of disorder, yielding an independent proof of the $Ω(n \log n)$ lower bound when discretized. By weaving algebraic, geometric, and continuous perspectives, the authors connect decision-tree arguments, polytope constraints, and energy minimization to illuminate why comparison-based sorting achieves optimal polynomial-time performance. The work suggests a unifying principle: exploiting intrinsic structure converts inherently exponential search spaces into efficiently navigable landscapes, with potential implications for a broad class of combinatorial problems.

Abstract

We investigate how sorting algorithms efficiently overcome the exponential size of the permutation space. Our main contribution is a new continuous-time formulation of sorting as a gradient flow on the permutohedron, yielding an independent proof of the classical $Ω(n \log n)$ lower bound for comparison-based sorting. This formulation reveals how exponential contraction of disorder occurs under simple geometric dynamics. In support of this analysis, we present algebraic, combinatorial, and geometric perspectives, including decision-tree arguments and linear constraints on the permutohedron. The idea that efficient sorting arises from structure-guided logarithmic reduction offers a unifying lens for how comparisons tame exponential spaces. These observations connect to broader questions in theoretical computer science, such as whether the existence of structure can explain why certain computational problems permit efficient solutions.

Figures (3)

• Figure 1: Binary decision tree representing all comparison-based paths for sorting three elements. Each internal node encodes a binary comparison, and each leaf corresponds to one of the six permutations in $S_3$. The highlighted path traces the sorted permutation $[1, 2, 3]$, requiring only three comparisons— the information-theoretic minimum $\lceil \log_2(3!) \rceil = 3$. This structure exemplifies recursive logarithmic reduction, where each comparison halves the remaining search space and reduces factorial complexity to the optimal $O(n log n)$ regime.
• Figure 2: Geometric interpretation of sorting as structured contraction of the permutohedron $\mathcal{P}_3$, the convex hull of all permutations of $(1,2,3)$. Each vertex corresponds to a permutation, and edges connect permutations that differ by adjacent swaps. The panels illustrate how successive linear constraints—$x_1 < x_2$ and $x_2 < x_3$—shrink the feasible region of $\mathcal{P}_3$, ultimately isolating the sorted permutation. This sequence illustrates recursive logarithmic reduction, with comparisons acting as hyperplane cuts that halve the solution space, reducing factorial complexity to the optimal $\mathcal{O}(n \log n)$.
• Figure 3: Gradient flow on the permutohedron illustrating exponential contraction toward the sorted permutation. Vertices correspond to unique permutations, with intermediate lines depicting trajectories of gradient descent driven by the disorder potential $V(x) = \frac{1}{2}\|x - v_s\|^2$. The highlighted paths demonstrate rapid convergence from initial unsorted states via exponential decay of disorder, linking continuous geometric dynamics directly to discrete sorting steps. For instance, starting at vertex $(3, 2, 1)$, the gradient flow moves continuously along adjacent edges toward the sorted vertex $(1, 2, 3)$, resolving inversions step-by-step. This visualization highlights the geometric foundations underlying the classical $\Omega(n \log n)$ sorting complexity bound established in Theorem \ref{['thm:contraction']}.

Theorems & Definitions (11)

• Lemma 3.1: Decision-Tree Lower Bound
• proof
• Definition 4.1: Permutohedron
• Proposition 4.2: Dimensionality and Affine Containment
• proof
• Theorem 4.3: Geometric Contraction via Linear Constraints
• proof
• Theorem 5.1
• proof
• Lemma 5.2