- The paper introduces a strong configuration LP relaxation with local consistency constraints that reduces the integrality gap for WTAP to below 1.5.
- It employs a two-phase randomized rounding algorithm with an innovative clean-up step to efficiently control overlapping coverage costs.
- The approach advances network design approximations by translating LP-hierarchy insights into practical algorithms for complex tree augmentation challenges.
A Strong LP Relaxation and (1.49)-Approximation for the Weighted Tree Augmentation Problem
Introduction and Background
The Weighted Tree Augmentation Problem (WTAP) asks for the minimum-cost set of non-tree edges (links) that must be added to a given weighted tree to achieve 2-edge-connectivity. WTAP is a canonical problem in network design, central to both theoretical and algorithmic combinatorial optimization. Despite its simple statement, WTAP exhibits a persistent barrier: classical LP-based techniques (specifically the "Cut LP") and combinatorial algorithms have not broken the 2-approximation factor for decades. Notably, for the standard Cut LP, the integrality gap is known to be between 1.5 and 2, with the best established upper bound remaining at 2.
Progress for the unweighted case (TAP) has recently yielded ratios below 1.5 using sophisticated structural decompositions and semidefinite programming [Cecchetto-Traub-Zenklusen 2021, Fiorini et al. 2018, Cheriyan & Gao 2018]. However, the weighted case has proven substantially more challenging, with the first (2−ϵ)-approximation only achieved in the past few years via intricate local search and greedy approaches [Traub & Zenklusen 2022]. The absence of strong LP upper bounds for WTAP, as well as the lack of a generic LP-based algorithm beating factor-2, represent fundamental obstacles in the literature.
Main Results
This work introduces a randomized (1.49)-approximation algorithm for WTAP, obtaining the first approximation guarantee strictly below $1.5$ for the weighted case and demonstrating the existence of an LP relaxation with integrality gap at most $1.49$ for WTAP. The approach is grounded in the design and rounding of a novel, hierarchically strengthened LP relaxation inspired by the Sherali-Adams hierarchy but specially tailored to the combinatorial structure of tree augmentation.
Specifically, the authors provide:
- A strengthened configuration-style LP relaxation ("Strong LP") that enforces local consistency across overlapping tree substructures, thereby "lifting" beyond the Odd-Cut LP by encoding higher-order correlations analogous to early levels of Sherali-Adams.
- A reduction showing any fractional solution to the Strong LP can be efficiently (and nearly losslessly) transformed into a structured fractional solution (termed a "Structured Fractional Solution"), capturing key local dependencies while enabling tractable rounding.
- A two-phase randomized rounding algorithm leveraging this structure: one component builds on the integrality of the Odd Cut LP for up-link/cross-link only instances for some parts of the solution, while a second phase applies sophisticated local correlation and clean-up arguments to minimize redundancy among cross-links and in-links.
- A provably correct clean-up procedure that exploits correlations in substructure coverage to ensure that uncorrelated links are not counted in the solution more than necessary, producing strict savings over earlier approaches that paid a full cost for "double" coverage.
- A tight technical analysis showing that, via a careful blend of the above algorithms and a subtle cost allocation, the expected cost achieves at most $1.4887$ times the cost of the Strong LP solution.
Technical Approach
1. Hierarchical LP Strengthening
The authors extend the Odd Cut LP by introducing configuration variables indexed by events associated with small subtrees (the "stars") and link sets. The crucial structural innovation is the enforcement of consistency constraints: for any two overlapping stars, the marginals over their shared edges/links must agree in the fractional solution. This globalizes dependency information, allowing supermodular correlations between link choices to be reflected in the fractional relaxation.
They define two main LPs:
- Strong LP: A configuration LP with variables for covering arrangements on subtrees with bounded numbers of leaves, enforcing marginal and consistency constraints via an explicit extension mechanism. This LP is efficiently solvable for any fixed parameters.
- Structured LP: A simplified version focusing only on constant-size stars, which is used as the interface for rounding algorithms after preprocessing and reduction.
Their main technical reduction shows that every feasible solution to Strong LP can, via local transformations (sparsification, link shadow completion, and shadow splitting), be transformed into a corresponding solution to the Structured LP with only a negligible cost increase.
2. Rounding and Approximation Guarantee
2.1 Odd-Cut Rounding and Decomposition
For a specific partitioning of the instance into correlated and uncorrelated edges, they leverage the result of [Fiorini et al. 2018]—which asserts integrality of the Odd Cut LP for specific instances—to perform "cheap" rounding on up-links and cross-links, while paying an extra factor on in-links.
2.2 Structured Rounding with Clean-Up
They introduce a novel clean-up phase that, after an initial structured (possibly over-counting) rounding, identifies cross-links or in-links whose coverage is "redundant" (i.e., more than one link covers the same cut in expectation) and replaces them with a minimum up-link cover when advantageous. This reduces the expected double-counting cost for uncorrelated in-links to a quantity strictly less than two, with precise control via a parameter γ.
2.3 Randomized Algorithm and Cost Analysis
By running two complementary algorithms (one based on the Odd Cut integrality, one on the clean-up-augmented structured rounding) with carefully optimized probabilities, the upper bound is shown to be ≈1.4887 times the LP value, for all sufficiently small ϵ>0.
Numerical Guarantee and Contradictory Claims
- The paper proves for the first time that the integrality gap of an LP relaxation for WTAP can be made strictly smaller than $1.5$.
- The (1.49)-approximation ratio holds for general instances, strictly improving upon the prior (1.49)0 guarantees achieved via local search or relative greedy methods [Traub & Zenklusen 2022], and surpassing all previous LP-based bounds.
- The algorithm works in polynomial time for any fixed accuracy parameter, and the structure of the relaxation is amenable to explicit rounding (rather than only existential statements as in SDP lift-and-project approaches).
Implications and Theoretical Significance
The conceptual contribution is the construction of the first LP relaxation for WTAP that breaches the "barrier of 2" and, more remarkably, the "1.5 threshold," which was formerly only overcome in the unweighted case or in the metric/SDP context. This demonstrates that lift-and-project inspired configuration LPs, augmented with local consistency constraints and structural preprocessing, suffice for precise cost control in complex augmentation problems.
The analytic techniques developed here (conditional sampling, consistent shadow removal, explicit control of double-counting via subtree partitioning, event-based configurations) are robust and have immediate applicability to a broader class of network augmentation and covering problems, particularly for problems with significant symmetry or where local-global dependencies inhibit naive LP rounding.
This work also narrows the gap between what is achievable with LP/LP-based rounding and the best combinatorial or SDP-based results for closely related problems (e.g., 2-Edge-Connected Spanning Subgraph, Steiner Network Design), potentially signaling new lines for LP-strengthening for those settings.
Practical and Future Directions
While the current relaxation is polynomial for fixed parameters, a significant practical advance would be the development of more efficient separation routines or further LP compressions for larger instances, perhaps by blending these configuration ideas with cutting-plane methods or by exploiting sparsity in the event structure.
The approach also suggests the possibility of further progress: The analytic framework for clean-up via positive correlation is not fully exhausted, and improved dependencies or alternative clean-up/rewiring schemes could conceivably bring the gap even closer to the conjectured integrality lower bound of (1.49)1. Furthermore, similar configuration LPs could be constructed for directed variants and for generalized connectivity augmentation settings, where combinatorial complexity and current integrality gaps are even larger.
Conclusion
This paper establishes a sub-1.5 LP-based approximation algorithm for WTAP, via a new class of strong configuration LPs equipped with local consistency constraints reminiscent of hierarchy-based methods. The results demonstrate both a conceptual shift in how structural dependencies are encoded in relaxations and a concrete improvement in the achievable worst-case ratio for a central network design problem. The techniques are broadly applicable and clarify the preeminence of LP-based approaches for edge-connectivity augmentation in the presence of complex topological and cost-dependent interactions among links (2603.29582).