[Paper Review] Improved Bi-point Rounding Algorithms and a Golden Barrier for $k$-Median

Kishen N. Gowda, Thomas Pensyl+2 more|arXiv (Cornell University)|Oct 24, 2022

Advanced Optimization Algorithms Research4 citations

TL;DR

This paper improves bi-point rounding algorithms for the k-median problem by introducing a hierarchical framework of partitioning schemes and corresponding non-linear programs, achieving a 2.613-approximation—improving upon the prior best of 2.675. It also establishes a theoretical barrier: no bi-point rounding approach can achieve better than 2.544-approximation due to an integrality gap of √ϕ, limiting future progress under current methods.

ABSTRACT

The current best approximation algorithms for $k$-median rely on first obtaining a structured fractional solution known as a bi-point solution, and then rounding it to an integer solution. We improve this second step by unifying and refining previous approaches. We describe a hierarchy of increasingly-complex partitioning schemes for the facilities, along with corresponding sets of algorithms and factor-revealing non-linear programs. We prove that the third layer of this hierarchy is a $2.613$-approximation, improving upon the current best ratio of $2.675$, while no layer can be proved better than $2.588$ under the proposed analysis. On the negative side, we give a family of bi-point solutions which cannot be approximated better than the square root of the golden ratio, even if allowed to open $k+o(k)$ facilities. This gives a barrier to current approaches for obtaining an approximation better than $2 \sqrtϕ \approx 2.544$. Altogether we reduce the approximation gap of bi-point solutions by two thirds.

Motivation & Objective

• To improve the bi-point rounding factor in k-median approximation algorithms beyond the prior best of 2.675.
• To develop a systematic hierarchy of partitioning schemes and algorithm families for increasingly tighter bi-point rounding guarantees.
• To establish a fundamental lower bound on the performance of bi-point rounding techniques, showing that no such method can achieve better than 2.544-approximation.
• To analyze the resilience of this barrier under relaxed facility opening constraints (k + o(k) facilities), confirming it remains tight.

Proposed method

• Proposes a three-layer hierarchy of partitioning schemes for facilities, each with tailored algorithms and factor-revealing non-linear programs (NLPs).
• Uses a novel NLP formulation to analyze the best-performing algorithm within each layer, optimizing over facility partitioning and rounding probabilities.
• Introduces a dual-based analysis that interprets the solution as a probability distribution over algorithms, enabling a randomized interpretation of the hierarchical approach.
• Employs a constructive instance with irrational parameters approximated rationally to demonstrate the integrality gap of √ϕ, leveraging the golden ratio φ.
• Analyzes the impact of rational approximations and relaxed facility counts (k + o(k)) on the integrality gap, showing it remains within ε of √ϕ for large k.
• Draws connections to star-rounding and pseudo-tree structures, generalizing Li and Svensson’s techniques to improve cost bounds in high-correlation cases.

Experimental results

Research questions

• RQ1 Can the bi-point rounding factor be improved beyond 2.675 using a structured hierarchy of partitioning schemes and algorithm families?
• RQ2 What is the tightest possible approximation ratio achievable by any bi-point rounding algorithm under current analysis frameworks?
• RQ3 Does the integrality gap of √ϕ represent a fundamental limit for bi-point rounding, even when allowing k + o(k) facility openings?
• RQ4 Can the hierarchical algorithm framework be interpreted as a randomized algorithm that matches the theoretical limit as the hierarchy depth increases?
• RQ5 Is there a structural reason why current bi-point rounding techniques cannot surpass the 2.544-approximation barrier?

Key findings

• The proposed three-layer hierarchy achieves a 2.613-approximation for the k-median problem, improving upon the prior best of 2.675.
• The analysis proves that no bi-point rounding approach can achieve a factor better than 2.588 under the current framework, establishing a new upper bound on achievable performance.
• A new integrality gap instance demonstrates that the bi-point rounding factor cannot be better than √ϕ ≈ 1.272, implying a fundamental 2.544-approximation barrier.
• The integrality gap of √ϕ persists even when allowing k + o(k) facilities, showing that this barrier is robust to relaxed facility counts.
• Theoretical analysis confirms that the gap is tight for large k, as rational approximations and o(k) facility relaxations only affect the bound by o(1).
• The results suggest that further improvements beyond 2.544 require either better bi-point generation or a holistic analysis combining generation and rounding, as in prior works [1, 4].