Online Matching with Size-Based and Convex Delays

Published 1 Jul 2026 in cs.DS | (2607.00536v1)

Abstract: We study the online min-cost perfect matching with delay (MPMD) problem where $m$ requests arrive in a metric space of $n$ points. In MPMD, an algorithm can choose to match a request or to delay, and the objective is to minimise the sum of connection and delay costs. The connection cost of a match is the distance between the locations of two matched requests in the metric, and the increase of the delay cost is a function of the set of unmatched requests at every moment. In this paper, we study two different types of delay functions, size-based (MPMD-Size) and convex delays (MPMD-Convex). The study of MPMD-Size was initiated by Deryckere and Umboh (APPROX/RANDOM 2023) where the instantaneous delay increment is a non-negative monotone function of the number of unmatched requests. Our bounds are in terms of $n$ , as opposed to Deryckere and Umboh's bounds that depend on $m$ . Our results settle the deterministic competitive ratio (up to constants). At the heart of these results is a succinct encoding scheme of MPMD-Size on a given $n$ -point metric as a metrical task system problem on a $2<sup>{n-1}$ -point metric. We also consider MPMD-Convex proposed by Liu et al. (ISAAC 2018) where the delay cost incurred by each request is a uniform convex delay function of the time difference between its arrival time and the moment that it is matched by the algorithm. They focused on delay functions $f$ that are unbounded, non-decreasing, continuous, and satisfy $f(0)=f'(0)=0$, and showed that the deterministic competitive ratio is $Ω(n)$ for $n$ -point uniform metrics. We show that, surprisingly, when $f$ is a non-negative, monotone polynomial with $f'(0)>0$, there is an $O(1)$ -competitive deterministic algorithm for uniform metrics. Our result completes our understanding of MPMD-Convex on uniform metrics for a broad class of functions.

Abstract PDF Upgrade to Chat

Authors (3)

Summary

The paper presents a novel reduction of the size-based delay problem to Metrical Task Systems using a compact Boolean parity encoding, improving competitive bounds.
It establishes tight deterministic and randomized competitive ratios, showing exponential cost dependency on n and near-linear bounds in randomized settings.
For convex delay models, the introduction of T-impatient algorithms achieves constant competitive ratios when an infinitesimal linear waiting cost is added, revealing key practical trade-offs.

Online Matching with Size-Based and Convex Delays: An Authoritative Analysis

Introduction and Problem Landscape

The paper presents significant theoretical advances in the online Min-Cost Perfect Matching with Delay (MPMD) paradigm. The MPMD model, operating over an $n$ -point metric space with $m$ online requests, requires the algorithm to strike a tradeoff between prompt pairings (thus minimizing connection distance) and delayed matches that accrue penalizing delay costs. The cost model is modular, consisting of the connection metric and a delay functional, with recent attention shifting from classic linear delay models to more general size-based and convex delay regimes. The work addresses two challenging settings:

MPMD-Size: Delay increments are arbitrary monotone functions of the number of pending requests (“size-based”).
MPMD-Convex: Delay for each request is governed by a convex, monotone, uniform function of waiting time.

The authors contribute tight bounds for MPMD-Size in terms of $n$ , as opposed to prior results reliant on $m$ , and settle the competitive ratio landscape for MPMD-Convex on uniform metrics for a wide family of convex polynomials $f$ with $f'(0)>0$ . This essay provides an in-depth analysis of the technical formulations, reductions, and implications of these results.

MPMD-Size: Encoding and Competitive Ratios

Succinct Encoding via Metrical Task Systems

A central technical achievement is a succinct reduction of MPMD-Size to Metrical Task Systems (MTS) on a $2^{n-1}$ -point metric, as opposed to an $O(2^m)$ -point instance. The reduction exploits the observation that, under “smart” policies (never leaving more than one unmatched request per location), the pending state can be encoded as an $n$ -dimensional parity vector. Each bit tracks whether an odd or even number of arrivals at a location have yet to be matched off, so the entire matching history is tracked compactly as a Boolean parity vector.

Through this tight encoding, MPMD-Size is mapped to an MTS where the distance between two states corresponds to the minimal cost of flipping pending parities (i.e., performing matches), and the instantaneous cost vector at each timestep exactly mirrors the delay incurred by the corresponding pending pattern.

Deterministic and Randomized Bounds

By leveraging the classical MTS lower bound of $\Theta(N)$ for deterministic algorithms and $m$ 0 (upper) and $m$ 1 (lower) randomized competitive ratios (recent separation from the randomized $m$ 2-server conjecture (Xi et al., 2023)), the paper tightens MPMD-Size bounds in $m$ 3:

Deterministic: Tight $m$ 4 competitive ratio, closing the gap from $m$ 5 in prior work (Abellán et al., 2023) and showing that no deterministic algorithm can avoid exponential-in- $m$ 6 cost, even when $m$ 7 is polynomially large in $m$ 8 (“long” sequences).
Randomized: $m$ 9 upper bound and $n$ 0 lower bound. The lower bound is universal (applying to all metrics, not only uniform metrics as in previous results), but there remains a gap compared to the tight metric-dependent lower bounds in general MTS.

Crucially, the reduction is bidirectional: MTS-Single also reduces directly to MPMD-Size, which not only implies strong lower bounds for matching but also establishes the inherent online combinatorial complexity.

MPMD-Convex: Resolution for Monotone Polynomials

Convex Delay with $n$ 1: Threshold Phenomena

Prior work demonstrated that with convex $n$ 2, specifically those satisfying $n$ 3, connection costs grow $n$ 4 in the uniform metric, inducing heavy penalties for delay and making batching approaches (leveraged by linear and concave cases) non-competitive (Li et al., 2018). The central new result is that when $n$ 5, e.g., for shifts of monomials with a linear component, it is possible to construct $n$ 6-competitive deterministic algorithms (for fixed $n$ 7 and fixed metric diameter):

Theorem: For any monotone convex polynomial $n$ 8 with $n$ 9, an $m$ 0-competitive online algorithm exists for MPMD-Convex on $m$ 1-point uniform metrics.

This exposes a threshold in algorithmic complexity: infinitesimal linear contribution at $m$ 2 (i.e., $m$ 3) fundamentally alters the cost structure, permitting reductions to the linear delay regime.

Construction of T-Impatient Algorithms

The algorithmic design hinges on “ $m$ 4-impatient” algorithms: policies which ensure that any request pending more than $m$ 5 time steps (where $m$ 6 is controlled as a function of the degree and coefficients of $m$ 7) will be matched instantly upon the next availability. The analysis shows that the portion of time when long-delayed requests could accumulate excessive cost is amortized against the offline solution’s cost, via properties of convexity and the fact that the delay function “blows up” fast enough.

A novel upper bound is shown for $m$ 8 of the form $m$ 9: for such $f$ 0 and appropriate $f$ 1, the competitive ratio (relative to OPT) is $f$ 2, where constants depend on $f$ 3 and the uniform metric’s diameter.

Notably, algorithms tailored for linear delay (e.g., (Xiang et al., 2017)) do not possess the required impatience property and are not transferable as black-boxes, necessitating new algorithmic instrumentation focused on impatient counter systems.

Algorithmic and Structural Implications

Structural Complexity and Metric Space Reductions

The tight reductions between MPMD-Size and MTS reveal deep connections between online matching, task systems, and Boolean function representations over parity spaces. The cost function on the lifted metric space $f$ 4 is a non-trivial extension of earthmover/Wasserstein distances, and its relationship to classical metric embeddings suggests lines for further metric-space structural exploration.

Delay Function Sensitivity

The strong separation between the $f$ 5 and $f$ 6 regimes provides evidence that fine distinctions in the analytic properties of the delay functional (even its derivative at $f$ 7) can have first-order effects on online competitiveness. This is an important theoretical warning for the design and analysis of delay-sensitive online algorithms across application domains—both the polynomial degree and any “base” linear term map to regime shifts in computational complexity.

Practical Relevance

The refined dependence on $f$ 8 and metric structure improves the worst-case behavioral guarantees of online matching mechanisms, relevant in real systems with large numbers of locations (e.g., cloud, logistics, online gaming). Algorithms no longer suffer arbitrarily badly under high-throughput “long” input streams (large $f$ 9), provided the network structure ( $f'(0)>0$ 0) is moderate.

The convex delay results suggest that system designers can tune delay penality functions: introducing an infinitesimal linear waiting cost is sufficient to avoid the blowup in latency penalties encountered under “purely” superlinear delay cost regimes.

Future Directions

Open problems highlighted in this work include closing the remaining gap for randomized algorithms (possibly metric-dependent), characterizing the structure of the matching metric $f'(0)>0$ 1 for general spaces, investigating competitive ratios for bipartite matching under size-based and convex delays, and extending impatient algorithms to broader classes of convex (or even general) delays and to non-uniform metrics or aspect ratios. The techniques here additionally invite further connections to combinatorial and algebraic structure in online optimization settings.

Conclusion

This paper settles the deterministic competitive ratio for online matching with size-based delays (MPMD-Size), making substantial technical improvements via an exponentially more efficient encoding and universal lower bounds across all metrics. It rigorously maps the landscape for convex delay cost (MPMD-Convex), identifying critical analytic thresholds ( $f'(0)>0$ 2) and providing construction of constant-competitive algorithms via newly formalized impatient policies. The results sharpen the theoretical understanding of delay-sensitive online matching and metrical task systems, and open new avenues for metric-based algorithmic research in online combinatorial optimization.