Asymptotic Factor of Iterative Rounding on d-Dimensional Gasoline Instances

Computational study of the conjectured limiting ratio 2d for the iterative rounding algorithm on hard instances discovered via Co-FunSearch. Based on Nikoleit et al., arXiv:2601.16849 (Jan 2026).

Key Results

Conjecture
Ratio → 2d
d=1 limit
2
d=2 limit
4
d=3 limit
6
d=4 limit
8
Status
Open Problem
Conjecture: limk→∞ APX(k,d) / OPT(k,d) = 2d, where each coordinate contributes factor ≈ 2.

1D Scaling: APX vs Instance Size

APX Growth Across Dimensions

LP Tracking: Value at Each Rounding Step

Per-Coordinate Decomposition

1D Scaling Data

knAPXOPT (LP)APX/n
266.0-2.01.00
31425.0-4.01.79
430113.0-8.03.77
562481.0-16.07.76

Multi-Dimensional Results

dknAPXOPT (LP)Predicted 2d
1266.0-2.02
131425.0-4.02
1430113.0-8.02
1562481.0-16.02
2268.0-4.04
231429.0-6.04
321219.0-4.06

Brute-Force Verification (Small Instances)

dknAPXOPT (exact)OPT (LP)Ratio (exact)
1266.04.0-2.01.5
131425.0-4.0-4.0---
2268.08.0-4.01.0

Proposed Proof Strategies

Direction 1: Per-Coordinate Potential Decomposition

Decompose the objective into d independent coordinate contributions. Each coordinate contributes ratio approaching 2: coordinate 1 via Lorieau's 1D analysis, auxiliary coordinates via the 4 vs 2 coefficient asymmetry. Total: 2 + (d-1) x 2 = 2d. Challenge: proving LP relaxation does not couple coordinates to improve fractional solution.

Direction 2: LP Dual Certificate Tracking

Track dual feasible solutions (shadow prices) across rounding steps to lower-bound OPT at every step. The geometric progression u_i = 2^k(1 - 2^{-i}) creates self-similar dual structure. Most rigorous but technically demanding due to O(nd) dual constraints.

Direction 3: Self-Similar Recurrence

Exploit the recursive structure: instance at parameter k embeds instance at k-1. Derive recurrences APX(k,d) = f(APX(k-1,d), k, d) and OPT(k,d) = g(OPT(k-1,d), k, d). Most constructive approach but fragile under tie-breaking rule changes.

Reference

Nikoleit, Allouah, Kleer, Marchetti-Bowick, Talgam-Cohen. The Art of Being Difficult: Combining Human and AI Strengths to Find Adversarial Instances for Heuristics. arXiv:2601.16849, January 2026. [arXiv]