Hitting Sets Online and Unique-Max Coloring

We consider the problem of hitting sets online. The hypergraph (i.e., range-space consisting of points and ranges) is known in advance, and the ranges to be stabbed are input one-by-one in an online fashion. The online algorithm must stab each range upon arrival. An online algorithm may add points to the hitting set but may not remove already chosen points. The goal is to use the smallest number of points. The best known competitive ratio for hitting sets online by Alon et al. \cite{alon2009online} is $O(\log n \cdot \log m)$ for general hypergraphs, where $n$ and $m$ denote the number of points and the number of ranges, respectively. We consider hypergraphs in which the union of two intersecting ranges is also a range. Our main result for such hypergraphs is as follows. The competitive ratio of the online hitting set problem is at most the unique-max number and at least this number minus one.

💡 Research Summary

The paper studies the online hitting‑set problem, where a hypergraph (H=(X,\mathcal{R})) is known in advance but the ranges (hyperedges) arrive one by one. An online algorithm must “stab’’ each incoming range by adding points from (X); once a point is chosen it cannot be removed. The performance measure is the competitive ratio, i.e., the worst‑case factor by which the algorithm’s total number of selected points exceeds that of an optimal offline solution that knows the entire sequence in advance.
Alon et al. (2009) gave a general‑purpose algorithm with competitive ratio (O(\log n\cdot\log m)) for arbitrary hypergraphs (where (n=|X|) and (m=|\mathcal{R}|)). This paper shows that for a natural subclass of hypergraphs the competitive ratio can be characterized almost exactly. The subclass consists of hypergraphs that are union‑closed: whenever two ranges intersect, their union is also a range. Many geometric and combinatorial settings satisfy this property, such as intervals on a line, circular arcs, and paths in a tree.

The key technical tool introduced is unique‑max coloring. A coloring assigns an integer label to each point such that, for every range (R\in\mathcal{R}), the maximum label appearing in (R) occurs at exactly one point. The smallest number of colors needed for such a labeling is called the unique‑max number, denoted (\chi_{\text{um}}(H)). This parameter is distinct from classical measures like VC‑dimension or the size of a minimum set cover; it captures a combinatorial “uniqueness’’ property of the hypergraph.

The authors prove two complementary results.

1. Upper bound (algorithm). Using any unique‑max coloring with (\chi_{\text{um}}(H)) colors, they design a deterministic online algorithm. When a new range arrives, the algorithm looks at the point inside the range that carries the highest color. If that point has not been selected yet, the algorithm adds it to the hitting set. Because the hypergraph is union‑closed, previously selected points can never block the existence of a unique‑max point for a new range. Consequently the algorithm never selects more than (\chi_{\text{um}}(H)) times the optimal offline solution, i.e., its competitive ratio is at most (\chi_{\text{um}}(H)).
2. Lower bound (adversary). An adversary can force any online algorithm to incur a cost of at least (\chi_{\text{um}}(H)-1) times optimal. The construction proceeds by repeatedly presenting ranges whose unique‑max point is the highest‑colored point not yet used by the algorithm. Each step forces the algorithm to pick a new point with a previously unseen color, and after (\chi_{\text{um}}(H)-1) steps the adversary stops. This shows that no online algorithm can achieve a competitive ratio better than (\chi_{\text{um}}(H)-1).

Together, these bounds imply that for union‑closed hypergraphs the optimal competitive ratio lies in the interval (