Terminate with first feasible point

[ChristophNeumann]

I hope that this is the right place to post my question/issue. If not, I would be thankful if you pointed me to another place.

I wish to obtain runtimes that Bonmin algorithms (b-hyb, b-oa) need to find a feasible point at least as good as a certain objective value val. I use the pyomo framework and do this by setting the options

• solution_limit = 1
• cutoff = val

I would expect the algorithm to terminate with the first feasible point and the query (in the pyomo framework) solver_message.time to give me the runtime of solver (maybe + some model building times?).

Yet, if I set these options e.g. for b-hyb and the MINLPLib instance cvxnonsep_normcon30 with val = -33.04175, I get the output

OA0003I New best feasible of -33.095184 found after 1504.2358 sec and
OA0003I New best feasible of -33.159661 found after 1592.0921 sec and

so the method seems to terminate after the second feasible point, and when I query solver_message.solver.time, I obtain 1656.4675970077515 sec.

Same with OA:

OA0012I After 100.14456.1f seconds, 1003 iterations upper bound -33.0417990g, lower bound -34.4287510g
OA0012I After 200.30932.1f seconds, 1415 iterations upper bound -33.0417990g, lower bound -34.4287510g
OA0012I After 300.42422.1f seconds, 1731 iterations upper bound -33.0417990g, lower bound -34.4287510g
OA0012I After 400.98337.1f seconds, 1917 iterations upper bound -33.0417990g, lower bound -34.4287510g
OA0012I After 501.29921.1f seconds, 2075 iterations upper bound -33.0417990g, lower bound -34.4287510

The options seem to be correctly set, given the output:

Bonmin 1.8.7 using Cbc 2.10.0 and Ipopt 3.12.12
bonmin: bonmin.cutoff=-33.04175
bonmin.solution_limit=1
bonmin.algorithm=b-oa

So, which one is the correct runtime for finding a feasible point of a given objective value? The first that shows up in the log file, or the time when the algorithm terminates?

Cheers,
Christoph