I have a Mirror-Circe, ser-=21 problem which Popeye v4.63, 32bit, solved in 23 minutes. Also Popeye v4.65, 64bit, found the solution, but Popeye v4.65, 32bit found no solution in four hours.
What could be the reason?
Is the problem published? If not: can you post a shortened version that demonstrates the behavior without affecting the publication status of the full-length problem?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
Anonymous
-
2014-04-15
Sorry, I didn't see that you had open the ticket. I send you the problem by mail (14.3.14 and 2.4.14 with the subject Popeye 4.65), because it's an original.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The solving time is dominated by attempts which move the white bishop.
Replace it by the obvious 2 white pawns, and solving will take half the time with 4.65 than with 4.63. This means that you are punished for being clever :-)
One contributing factor seems to be that 4.65 generates diagonal king moves before orthogonal ones while 4.63 did the opposite. This means that for your problem, 4.65 has to do much more check tests than 4.63 once the bishop has been moved.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Impossible to tell without knowing the problem.
Is the problem published? If not: can you post a shortened version that demonstrates the behavior without affecting the publication status of the full-length problem?
Sorry, I didn't see that you had open the ticket. I send you the problem by mail (14.3.14 and 2.4.14 with the subject Popeye 4.65), because it's an original.
I can reproduce your results, both with the 32 and 64 bit Windows executables.
Strangely enough, the results are exactly reversed when using the Linux executable!
The solving time is dominated by attempts which move the white bishop.
Replace it by the obvious 2 white pawns, and solving will take half the time with 4.65 than with 4.63. This means that you are punished for being clever :-)
One contributing factor seems to be that 4.65 generates diagonal king moves before orthogonal ones while 4.63 did the opposite. This means that for your problem, 4.65 has to do much more check tests than 4.63 once the bishop has been moved.