Pseudo-random with 8 clusters of 50 vertices, intra-cluster density 1.0
(i.e. the clusters are complete subgraphs), expected inter-cluster density 0.2.

For this graph, OptimizerModularity finds fewer than 8 clusters.
While the Modularity is indeed optimal for the clustering with 8 clusters, 
the algorithm fails to find the optimum.  Of course, suggestions 
for improvements of the algorithm are welcome.