A Comprehensive Survey of Data Reduction Rules for the Maximum Weighted Independent Set Problem

The Maximum Weight Independent Set (MWIS) problem, as well as its related problems such as Minimum Weight Vertex Cover, are fundamental NP-hard problems with numerous practical applications. Due to their computational complexity, a variety of data reduction rules have been proposed in recent years to simplify instances of these problems, enabling exact solvers and heuristics to handle them more effectively. Data reduction rules are polynomial time procedures that can reduce an instance while ensuring that an optimal solution on the reduced instance can be easily extended to an optimal solution for the original instance. Data reduction rules have proven to be especially useful in branch-and-reduce methods, where successful reductions often lead to problem instances that can be solved exactly. This survey provides a comprehensive overview of data reduction rules for the MWIS problem. We also provide a reference implementation for these reductions. This survey will be updated as new reduction techniques are developed, serving as a centralized resource for researchers and practitioners.

💡 Research Summary

The paper presents a comprehensive survey of exact data‑reduction (kernelization) rules for the Maximum‑Weight Independent Set (MWIS) problem and its closely related variants, namely Minimum‑Weight Vertex Cover (MWVC) and Maximum‑Weight Clique (MWC). It begins by motivating the importance of MWIS in a variety of real‑world applications such as long‑haul vehicle routing, winner determination in combinatorial auctions, and protein‑structure prediction, emphasizing that the problem’s NP‑hard nature makes preprocessing indispensable for both exact and heuristic solvers.

After establishing basic graph terminology (neighbourhood, degree, simplicial vertices, etc.) and formal definitions of independent sets, vertex covers, and cliques, the authors review the historical development of reduction techniques. They distinguish two broad categories: (1) theoretically‑grounded kernelization rules that guarantee a kernel size bounded by a function of a chosen parameter (typically the solution size k), and (2) practical, often heuristic‑driven rules that achieve high reduction ratios on real‑world instances without formal size guarantees.

The theoretical rules include classic degree‑1 and degree‑2 reductions, twin‑vertex removal, domination, and simplicial‑vertex elimination. Each of these can be applied in polynomial time and, together, they can shrink many sparse graphs to kernels of size O(k) or O(k²). The practical side introduces meta‑reductions such as neighborhood removal and neighborhood folding, which combine several structural observations and weight information to collapse larger substructures in a single step. The “struction” transformation, originally proposed for unweighted graphs, does not necessarily reduce vertex count but rewires the graph to expose further reductions.

The survey then systematically catalogs exact solvers that rely on these reductions. Early branch‑and‑bound algorithms (e.g., C_ℓ, MWCLQ, TSM‑MWC) used only bounding techniques. Starting with K‑MIS, the community shifted toward branch‑and‑reduce methods that interleave reductions with branching. K‑MIS introduced the two meta‑reductions mentioned above and combined them with weighted clique‑cover upper bounds and an adapted ARW local search for lower bounds. Subsequent solvers (S‑BMS, MWCRedu, S‑BMS, C_ℓ‑W) extended K‑MIS by adding twin, domination, simplicial, and struction rules, often achieving dramatic reductions (up to 90 % of vertices/edges removed) on benchmark graphs with millions of vertices. A recent theoretical contribution by Xiao et al. provides a measure‑and‑conquer analysis showing an O*(1.1443^{(0.624x‑0.872)n}) running time for sparse graphs (average degree x ≤ 3), illustrating that reductions can also improve worst‑case bounds.

On the heuristic side, the authors discuss that virtually all high‑performance heuristics for MWIS/MWVC/MWC are based on local search. For unweighted instances, the ARW algorithm’s (1,2)‑swap mechanism set the standard. Weighted variants incorporate tabu search, adaptive perturbation, configuration checking, smart restarts, and push operators to escape local optima. The “reduce‑and‑peel” paradigm, used in recent MWCP‑PP, repeatedly applies exact reductions until no further progress is possible, then resorts to a greedy tie‑breaking rule to continue. In the MWVC domain, ant‑colony optimization (ACO, ACO+SEE), multi‑start iterated tabu search (MS‑ITS), and population‑based iterated greedy (PBIG) have been enhanced with dynamic edge‑weight schemes (originating from DLSWCC) and with preprocessing reductions, leading to rapid convergence to high‑quality solutions.

A key contribution of the paper is a taxonomy that classifies all known reduction rules into four groups: (1) degree‑based, (2) structural (twin, domination, simplicial, struction), (3) meta‑reductions (neighbourhood removal/folding), and (4) weight‑based upper/lower bound rules. For each rule the authors provide the precise applicability condition, computational complexity, and expected impact on graph size. Implementation considerations such as adjacency‑list representation, priority queues for fast degree updates, bit‑set operations for dense subgraphs, and opportunities for parallel execution are discussed in detail.

The authors also release a reference implementation (https://github.com/KarlsruheMIS/DataReductions) written in C++. The library modularizes each reduction rule, offers a unified interface for integration with existing solvers, and includes an automated test suite for correctness and performance benchmarking. This open‑source resource is intended to become a central hub for future research, allowing new reduction ideas to be evaluated against a common baseline.

In the concluding section, the paper highlights several open research directions: (i) scaling reductions to ultra‑large graphs (hundreds of millions of vertices) while respecting memory constraints, (ii) developing dynamic reductions for streaming or online graph updates, and (iii) leveraging machine‑learning techniques to predict which reductions will be most effective on a given instance and to tune their parameters automatically. The authors argue that continued advances in data‑reduction will remain pivotal for pushing the frontier of exact and heuristic solutions to MWIS and its related problems.