MAX-CUT is a classic combinatorial optimization problem, but its binary formulation is notoriously difficult to solve directly. According to the arXiv paper, the landscape contains exponentially many local optima, which makes the outcome strongly dependent on the chosen starting point.
The authors study Particle Polynomial Relaxation (PPR), a moment-based nonconvex continuous reformulation of MAX-CUT. The abstract does not provide details of the method itself, but frames it as a way to approach the problem without relying on binary variables.
The paper's title adds that graph coarsening enables escaping local optima in compute-constrained scenarios. Because the provided abstract is brief, the exact relationship between coarsening and PPR is not specified in the excerpt, but the overall claim is that this combination helps mitigate the initialization problem under limited computational resources.