Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Correcting the Foundational Analysis of Karp--Vazirani--Vazirani (STOC 1990): A Rigorous Revision of the $1-1/e$ Upper Bound

Pan Xu

TL;DR

The paper revisits the online bipartite matching problem and the KV(V) $1 - 1/e$ upper bound, identifying technical gaps in the original analysis. It introduces a rigorous reformulation of the evolution of available neighbors as a discrete-time death process and couples it with a factor-revealing linear program to bound the expected online performance. The main result is a precise bound $\mathbb{E}[\sigma(\mathsf{RGA}, M^*)] \le \left\lceil (1 - 1/e)\, n + 2 - 1/e \right\rceil$ for all $n$, achieved by solving an auxiliary sequence with a closed-form involving harmonic numbers. This work provides a transparent, mathematically sound correction to the classical KV(V) analysis while preserving its combinatorial framework, offering a simpler yet rigorous tool for analyzing online matching problems and related competitive analyses.

Abstract

We revisit the classical analysis of Karp, Vazirani, and Vazirani (KVV, STOC~1990), which established the well-known upper bound of $1 - 1/e$ as the limiting proportion of vertices that can be matched by any online procedure in a canonical bipartite structure. Although foundational, the original analysis contains several inaccuracies, including a fundamental technical gap in the treatment of the underlying discrete process. We give a transparent and fully rigorous reconstruction of the KVV argument by reformulating the evolution of available neighbors as a discrete-time death process and deriving a sharp upper bound via a simple factor-revealing linear program that captures the correct recurrence structure. This yields a precise bound $\lceil n(1 - 1/e) + 2 - 1/e \rceil$ on the expected number of matched vertices, refining the classical claim $n(1 - 1/e) + o(n)$. Our goal is not to optimize this upper bound, but to provide a mathematically sound and conceptually clean correction of the classical KVV analysis, while remaining faithful to its original combinatorial framework.

Correcting the Foundational Analysis of Karp--Vazirani--Vazirani (STOC 1990): A Rigorous Revision of the $1-1/e$ Upper Bound

TL;DR

The paper revisits the online bipartite matching problem and the KV(V) upper bound, identifying technical gaps in the original analysis. It introduces a rigorous reformulation of the evolution of available neighbors as a discrete-time death process and couples it with a factor-revealing linear program to bound the expected online performance. The main result is a precise bound for all , achieved by solving an auxiliary sequence with a closed-form involving harmonic numbers. This work provides a transparent, mathematically sound correction to the classical KV(V) analysis while preserving its combinatorial framework, offering a simpler yet rigorous tool for analyzing online matching problems and related competitive analyses.

Abstract

We revisit the classical analysis of Karp, Vazirani, and Vazirani (KVV, STOC~1990), which established the well-known upper bound of as the limiting proportion of vertices that can be matched by any online procedure in a canonical bipartite structure. Although foundational, the original analysis contains several inaccuracies, including a fundamental technical gap in the treatment of the underlying discrete process. We give a transparent and fully rigorous reconstruction of the KVV argument by reformulating the evolution of available neighbors as a discrete-time death process and deriving a sharp upper bound via a simple factor-revealing linear program that captures the correct recurrence structure. This yields a precise bound on the expected number of matched vertices, refining the classical claim . Our goal is not to optimize this upper bound, but to provide a mathematically sound and conceptually clean correction of the classical KVV analysis, while remaining faithful to its original combinatorial framework.

Paper Structure

This paper contains 10 sections, 6 theorems, 28 equations, 1 figure, 1 algorithm.

Table of Contents

  1. Introduction
  2. Review of the Approach Establishing the $1 - 1/e$ Upper Bound in $\mathsf{KVV}$
  3. Review of the Approach of Computing the Value $\mathop{\mathrm{\mathsf{E}}}\limits[\sigma(\mathsf{RGA}\xspace, M^*)]$ in $\mathsf{KVV}$
  4. Fundamental Errors in the Approach of Computing $\mathop{\mathrm{\mathsf{E}}}\limits[\sigma(\mathsf{RGA}\xspace, M^*)]$ in $\mathsf{KVV}$
  5. Main Contributions
  6. Formal Approach to Upper Bounding the Value $\mathop{\mathrm{\mathsf{E}}}\limits[\sigma(\mathsf{RGA}\xspace, M^*)]$
  7. Upper Bounding the Value $\mathop{\mathrm{\mathsf{E}}}\limits[T]$ with $T$ Defined in \ref{['def:T']}
  8. Proof of the Main Theorem \ref{['thm:main']}
  9. Conclusion
  10. Solving the Sequence $(\delta^*_t)_{t \in [n]}$ as Defined in \ref{['seq:26-a']}

Key Result

Lemma 1

For any deterministic algorithm $\mathsf{D\hbox{-}ALG}\xspace$, its expected performance (the expected size of the matching obtained) on $\mathcal{U}(\mathcal{M})$, denoted $\mathop{\mathrm{\mathsf{E}}}\limits_{M \sim \mathcal{U}(\mathcal{M})}[\sigma(\mathsf{D\hbox{-}ALG}\xspace,M)]$, satisfies where $\mathsf{RGA}\xspace$ is formally defined in alg:rad as the Randomized Greedy Algorithm that unif

Figures (1)

  • Figure 1: A basic module instance proposed by $\mathsf{KVV}$ to prove the upper bound of $1 - 1/e$. The instance (Left) can be represented using a unique $n \times n$ binary matrix $M^*$ (Right), where $M^*(i,j) = 1$ if and only if the online agent $j$ is connected to the offline agent $i$, i.e.,$(i,j) \in E$.

Theorems & Definitions (12)

  • Lemma 1: kvv
  • Theorem 2
  • Definition 1
  • Lemma 2
  • Proof 1
  • Lemma 3
  • Proof 2
  • Lemma 4
  • Proof 3
  • Proof 4: Proof of Theorem \ref{['thm:main']}
  • ...and 2 more