An Experimental Evaluation of Semidefinite Programming and Spectral Algorithms for Max Cut

We experimentally evaluate the performance of several Max Cut approximation algorithms. In particular, we compare results Goemans and Williamson algorithm using semidefinite programming with Trevisan’s spectral partitioning. The former has a known.878 guarantee whereas latter a.614 guarantee. investigate whether this gap in guarantees is evident practice or performs as well SDP. also performances to standard greedy which a.5 guarantee, two additional algorithms, heuristic from Burer, Monteiro, Zhang. algorithms are tested on Erdős-Renyi random graphs, complete graphs TSPLIB, real-world Network Repository. find, unsurprisingly, that provide significant speed advantage over our experiments, BMZ return cuts values competitive those