BQM QPU solvers do not yield optimal solutions
I'm trying to solve a BQM with 45 variables. The problem in itself isn't too difficult and a simple tabu search easily finds the optimal solution. However, neither the Chimera nor Pegasus QPU seem to be able to find a good solution (using minorminer for both).
I've increased the chain strength just enough to prevent chain breakages. With the Chimera I get many long chains (length >7), but still the Chimera gives me better solutions than Pegasus (still far from optimal though).
I've tried increasing num_reads, playing around with the annealing schedule and the minorminer parameters, but still no good results.
How should one debug such an issue? It seems odd to me that the DWave QPU can't solve such a relatively simple problem with no more than 400 target variables, what am missing here?