SAT solver gives different result with the same input

[brucefeng10]

What version of OR-Tools and what language are you using?
Version: v8.2
Language: Java

Which solver are you using
CP-SAT

What operating system (Linux, Windows, ...) and version?
macOS 13.5.1

What did you do?
I'm solving a complex pure Integer Problem, and i use CP-SAT solver, which performs much much better than SCIP. Since it's hard to solve the model to optimal, I set the solving time to 3 minutes. Part of my codes:

Loader.loadNativeLibraries();
MPSolver solver = MPSolver.createSolver("CP-SAT");
.......
MPSolver.ResultStatus status = solver.solve();
solver.setTimeLimit(3 * 60 * 1000);

What did you see instead?
However, everytime i run the problem with the same input, the result is different. I mean the objective value is different, sometimes it varies much.

Anything else we should know about your project / environment
I also tried to change the solver to "SCIP", the 3min result is always the same, though the result is much worse than "CP-SAT".
The model file is attached.
file.txt

[lperron]

1. with time limit, you can never be truly be deterministic.
2. why not use 9,9 ? It is much better
3. There are parameters to make a run deterministic, but it is much slower.

So I agree the solver is non deterministic.

[brucefeng10]

It's non deterministic, i've just tried v9.9.

1. "with time limit, you can never be truly be deterministic", do you mean the model has not converged to a stabe status? If i set the time limit longer, the nondeterminacy should relieve?
2. Are there any running termination parameters other than time limit?

[lperron]

by default, in parallel, cp-sat is non deterministic. If you use runtime, the amount of work you can do depends on the load of the machine. no other termination parameters in parallel. Laurent Perron | Operations Research | ***@***.*** | (33) 1 42 68 53 00 Le mar. 23 avr. 2024 à 09:25, Bruce Feng ***@***.***> a écrit :

…

It's non deterministic, i've just tried v9.9. 1. "with time limit, you can never be truly be deterministic", do you mean the model has not converged to a stabe status? If i set the time limit longer, the nondeterminacy should relieve? 2. Are there any running termination parameters other than time limit? — Reply to this email directly, view it on GitHub <#4196 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUPL3K5VFSSHPQKO3RGAITY6YEGFAVCNFSM6AAAAABGUI2CJCVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TCOJXGQZTO> . You are receiving this because you commented.Message ID: ***@***.***>

[brucefeng10]

I got it, thanks for your answer.