Clique in 3-track interval graphs is APX-hard

Butman, Hermelin, Lewenstein, and Rawitz proved that Clique in t-interval graphs is NP-hard for t >= 3. We strengthen this result to show that Clique in 3-track interval graphs is APX-hard.

💡 Research Summary

The paper addresses the approximation complexity of the Clique problem in a very specific class of intersection graphs: 3‑track interval graphs. A t‑interval graph is defined by assigning each vertex a set of t intervals on the real line; two vertices are adjacent if at least one interval from each set intersect. When the intervals are placed on t parallel “tracks” (horizontal lines) and each vertex occupies exactly one interval per track, the resulting graph is a t‑track interval graph. Prior work by Butman, Hermelin, Lewenstein, and Rawitz established that for any t ≥ 3 the Clique problem is NP‑hard, but they left open the question of whether the problem admits a constant‑factor approximation algorithm.

The authors close this gap by proving that Clique in 3‑track interval graphs is APX‑hard, i.e., there exists a constant ε > 0 such that no polynomial‑time algorithm can achieve a (1 + ε)‑approximation unless P = NP. Their reduction proceeds in two stages and satisfies the stringent requirements of an L‑reduction, preserving approximation ratios up to constant factors.

Stage 1 – Source APX‑hard problem.
The reduction starts from the Maximum Independent Set (MIS) problem on bounded‑degree graphs, specifically 3‑regular graphs. MIS on such graphs is known to be APX‑hard (Feige, 1998). An independent set in the original graph corresponds to a set of vertices with no edges between them.

Stage 2 – Mapping to a 3‑track interval representation.
For each vertex v of the source graph, the construction creates three intervals v¹, v², v³, one on each of the three tracks. The intervals are placed so that:

1. If two vertices u and v are adjacent in the source graph, then at least one pair of their intervals (e.g., u¹ and v¹) overlap, guaranteeing an edge in the 3‑track interval graph.
2. If u and v are non‑adjacent, the intervals are arranged so that none of the three track‑pairs intersect, ensuring no edge.

The precise coordinates are chosen to keep the intervals short and to avoid unintended intersections across tracks. This “track‑allocation” technique is a refinement of the classic interval‑graph reduction used for NP‑hardness, now calibrated to control the size of the optimal solution.

L‑reduction parameters.
The authors prove two constants α and β such that:

• α‑bound: The size of a maximum clique in the constructed 3‑track graph is at most α times the size of a maximum independent set in the original graph (α = 2). Thus optimal solutions are linearly related.
• β‑bound: Any feasible clique of size C in the 3‑track graph can be transformed back into an independent set of size at least C/β in the source graph (β = 3). Consequently, an (1 − δ)‑approximation for Clique yields a (1 − βδ)‑approximation for MIS.

Because MIS on 3‑regular graphs does not admit a PTAS unless P = NP, the existence of such α and β implies that Clique on 3‑track interval graphs also lacks a PTAS and is APX‑hard.

Implications and broader context.
The result shows that even when the number of tracks is fixed to the smallest value that still permits NP‑hardness (t = 3), the problem remains resistant to arbitrarily good approximation. This contrasts with the classic interval graph case (t = 1), where Clique is trivial (the graph is chordal and the maximum clique can be found in linear time). The paper therefore delineates a sharp boundary in the tractability landscape of interval‑based graph classes.

From an application standpoint, many scheduling, bioinformatics, and wireless‑network problems can be modeled as t‑track interval graphs, where each track represents a resource or a time line. The APX‑hardness result warns practitioners that, for three or more parallel resources, one should not expect algorithms that guarantee solutions arbitrarily close to optimal; instead, research should focus on constant‑factor heuristics, parameterized algorithms, or special‑case restrictions (e.g., bounded interval length or limited overlap patterns).

Future directions.
The authors suggest that their reduction can be extended to any t ≥ 3, implying APX‑hardness for all higher‑track interval graphs. Moreover, the technique invites investigation into whether tighter approximation lower bounds (e.g., specific inapproximability ratios) can be derived, possibly by reducing from problems with known hardness of approximation factors. Another promising line is to identify subclasses of 3‑track interval graphs (such as those with monotone interval ordering or limited nesting depth) that might admit better approximations or even exact polynomial‑time algorithms.

In summary, the paper delivers a rigorous proof that Clique in 3‑track interval graphs is APX‑hard, filling a notable gap in the literature and establishing a clear approximation barrier for a natural and widely applicable graph class.