An improved lower bound for one-dimensional online unit clustering

The online unit clustering problem was proposed by Chan and Zarrabi-Zadeh (WAOA2007 and Theory of Computing Systems 45(3), 2009), which is defined as follows: “Points” are given online in the $d$-dimensional Euclidean space one by one. An algorithm creates a “cluster,” which is a $d$-dimensional rectangle. The initial length of each edge of a cluster is 0. An algorithm can extend an edge until it reaches unit length independently of other dimensions. The task of an algorithm is to cover a new given point either by creating a new cluster and assigning it to the point, or by extending edges of an existing cluster created in past times. The goal is to minimize the total number of created clusters. Chan and Zarrabi-Zadeh proposed some method to obtain a competitive algorithm for the $d$-dimensional case using an algorithm for the one-dimensional case, and thus the one-dimensional case has been extensively studied including some variants of the unit clustering problem. In this paper, we show a lower bound of $13/8 = 1.625$ on the competitive ratio of any deterministic online algorithm for the one-dimensional unit clustering, improving the previous lower bound $8/5 (=1.6)$ presented by Epstein and van Stee (WAOA2007 and ACM Transactions on Algorithms 7(1), 2010). Note that Ehmsen and Larsen (SWAT2010 and Theoretical Computer Science, 500, 2013) showed the current best upper bound of $5/3$, and conjectured that the exact competitive ratio in the one-dimensional case may be $13/8$.

💡 Research Summary

The paper addresses the one‑dimensional online unit‑clustering problem, a fundamental variant of online clustering where points arrive one by one on the real line and must be covered by intervals (clusters) of unit length. Each cluster starts with length zero and can be extended independently along the line up to length one; once placed, a cluster cannot be moved or removed. The objective is to minimize the total number of clusters created. Performance is measured by competitive analysis: an online algorithm is c‑competitive if, for every input sequence σ, the number of clusters it creates, C_ON(σ), is at most c times the number created by an optimal offline algorithm, C_OPT(σ).

Previous work established a deterministic lower bound of 8/5 (≈1.6) by Epstein and van Stee (2010) and an upper bound of 5/3 (≈1.667) by Ehmsen and Larsen (2013). The latter authors conjectured that the exact deterministic competitive ratio might be 13/8 (1.625). This paper confirms that conjecture by proving a new deterministic lower bound of 13/8, thereby narrowing the gap to just 0.042.

The core contribution is a carefully crafted adversarial input sequence (presented in Table 2) that forces any deterministic online algorithm to incur at least a 13/8 cost ratio. The sequence consists of 15 points placed at specific real‑valued positions. At each arrival, the online algorithm faces a binary decision: either create a new cluster or assign the point to an existing cluster. The authors analyze every possible decision path. For each choice, they compute the minimal number of clusters an optimal offline algorithm would need to cover the points that follow. By arranging the points so that the “worst‑case” decision at each step leads to the highest possible cumulative ratio, they guarantee that the final ratio cannot be smaller than 13/8.

Key steps of the construction include:

1. Early points (3, 4) force the algorithm to create a new cluster D, yielding a local ratio of 2:1.
2. Points 5 and 6 introduce a fork: creating a new cluster F or reusing an existing cluster E. Both branches are designed so that later points will penalize the algorithm if it chooses the apparently cheaper option.
3. Points 2 and 2.5 force the creation of cluster B, establishing a local ratio of 5/3.
4. Subsequent points (7, 8, 8.5) manipulate the algorithm into using clusters G and H in a way that later points (9.5, 10, 11, 11.5, 12, 13, 14) force the algorithm to open many more clusters than the offline optimum.
5. In the final critical stage, the online algorithm must have created 13 clusters while the optimal offline solution uses only 6, giving an exact ratio of 13/8.

The proof relies on exhaustive case analysis. For each decision node, the authors argue that any deviation from the prescribed worst‑case path would either increase the algorithm’s immediate cost or allow the offline optimum to catch up, never improving the overall ratio beyond 13/8. They also discuss the combinatorial structure of possible ratios x/y with 13/8 < x/y < 5/3, noting that (13, 8), (18, 11), (21, 13), etc., are the only integer pairs in this interval. Their construction covers the smallest pair (13, 8); extending the technique to larger pairs would require substantially more intricate analysis.

In the discussion, the authors point out that achieving a competitive ratio of c essentially requires showing that for every x clusters created by the online algorithm, the optimal offline algorithm must create at least x/c clusters. This principle underlies earlier analyses for ratios 7/4 and 5/3. They argue that to push the lower bound down to 13/8, the adversarial instance must force the online algorithm to behave in a very specific way, as captured in Table 2.

The paper concludes that the deterministic competitive ratio for one‑dimensional online unit clustering is at least 13/8, improving the previous bound by 0.025 and supporting the conjecture that 13/8 is the exact value. This result also has implications for higher dimensions, because any improvement in the one‑dimensional case directly translates into better algorithms for d‑dimensional clustering via the reduction shown by Chan and Zarrabi‑Zadeh. Future work may explore whether the lower bound can be generalized to larger integer pairs or extended to randomized algorithms.