Local Max-Cut on Sparse Graphs

Gregory Schwartzman
2024-04-22
Abstract:We bound the smoothed running time of the FLIP algorithm for local Max-Cut as a function of $\alpha$, the arboricity of the input graph. We show that, with high probability and in expectation, the following holds (where $n$ is the number of nodes and $\phi$ is the smoothing parameter): 1) When $\alpha = O(\log^{1-\delta} n)$ FLIP terminates in $\phi poly(n)$ iterations, where $\delta \in (0,1]$ is an arbitrarily small constant. Previous to our results the only graph families for which FLIP was known to achieve a smoothed polynomial running time were complete graphs and graphs with logarithmic maximum degree. 2) For arbitrary values of $\alpha$ we get a running time of $\phi n^{O(\frac{\alpha}{\log n} + \log \alpha)}$. This improves over the best known running time for general graphs of $\phi n^{O(\sqrt{ \log n })}$ for $\alpha = o(\log^{1.5} n)$. Specifically, when $\alpha = O(\log n)$ we get a significantly faster running time of $\phi n^{O(\log \log n)}$.
Data Structures and Algorithms
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the smoothed complexity analysis of the Local Max - Cut problem on sparse graphs. Specifically, the author studies the smoothed running time of the FLIP algorithm on graphs with different arboricity. ### Background and Problem Definition - **Local Max - Cut Problem**: Given a weighted graph \(G(V, E)\), where \(|V| = n\) and \(|E| = m\). The Local Max - Cut problem requires finding a locally maximal cut \(C\subset V\), that is, the weight of the cut cannot be increased by moving any vertex to the other side. - **FLIP Algorithm**: Starting from any initial cut, select a vertex and move it to the other side as long as this can increase the weight of the cut. The FLIP algorithm will eventually terminate at a locally maximal cut. - **Smoothed Analysis**: Smoothed analysis is an analysis method between the worst - case and the average - case. It evaluates the performance of an algorithm by adding random noise to the input data. ### Main Contributions of the Paper 1. **For graphs with arboricity \(\alpha = O(\log^{1-\delta}n)\)**: - When \(\alpha = O(\log^{1-\delta}n)\), the expected and high - probability running time of the FLIP algorithm under the smoothing parameter \(\phi\) is \(\phi\cdot n^{O(\log n)}\). - This result improves the previously known results only for complete graphs and logarithmic - degree graphs. 2. **For graphs with any arboricity \(\alpha\)**: - For any \(\alpha\), the running time of the FLIP algorithm is \(\phi\cdot n^{O(\alpha\log n+\log\alpha)}\). - When \(\alpha = O(\log n)\), the running time is significantly faster, which is \(\phi\cdot n^{O(\log\log n)}\). ### Technical Details - **Vertex Partition**: Construct a hierarchical partition of vertices by the arboricity \(\alpha\) of the graph, such that the number of neighbors of each vertex \(v\in V_i\) in \(V_i\) and subsequent levels does not exceed \(2\beta\alpha\). - **Good Movement Sequences**: Using the above hierarchical partition, it is proved that certain specific movement sequences (for example, moving the same vertex twice in a row without any other relevant vertex movement in between) can significantly increase the weight of the cut. - **Probability Analysis**: Through probability analysis and the union - bound technique, it is ensured that these good movement sequences will inevitably appear in a long enough movement sequence, thereby deriving the running time of the FLIP algorithm. ### Conclusion This paper provides a new framework to analyze the smoothed complexity of the FLIP algorithm on sparse graphs by introducing the concept of graph arboricity. This result not only extends the known graph classes with smoothed polynomial running time but also provides theoretical support for understanding the efficiency of local search algorithms in practical applications.