Infeasible solution with "fill_tightened_domains_in_response"

[sjindel]

What version of OR-Tools and what language are you using?
Version: main
Language: C++

Which solver are you using (e.g. CP-SAT, Routing Solver, GLOP, BOP, Gurobi)

CP-SAT

What operating system (Linux, Windows, ...) and version?
Linux 6.12.44

What did you do?
Solve the attached model (optimisation)

What did you expect to see
No crash

What did you see instead?
Crash - infeasible solution produced

Model attached. It crashes with or without the 'deterministic' parameter set, provided fill_tightened_domains_in_response=true in the SatParameters. Without that, it solves successfully.

I0000 00:00:1769090174.150522 3968175 cp_model_solver.cc:2875] Dumping infeasible response proto to '/tmp/wrong_response.pb.txt'.
F0000 00:00:1769090174.187860 3968175 cp_model_solver.cc:2879] Infeasible solution! source: 'default_lp' dumped CpSolverResponse to '/tmp/wrong_response.pb.txt'.
*** Check failure stack trace: ***
    @     0x56c195de7cf1  absl::lts_20250814::log_internal::LogMessage::SendToLog()
    @     0x56c1956d72b4  operations_research::sat::SolveCpModel()::{lambda()#1}::operator()()
    @     0x56c195c5fbd9  operations_research::sat::SharedResponseManager::GetResponse()
    @     0x56c1956fcd39  operations_research::sat::SolveCpModel()
    @     0x56c1956bd1fc  main
    @     0x7dfc95027675  (unknown)
    @     0x7dfc95027729  __libc_start_main
    @     0x56c1956bc9e5  _start

model.txt

[lperron]

From the cp_model.proto:

// If the option fill_tightened_domains_in_response is set, then this field
// will be a copy of the CpModelProto.variables where each domain has been
// reduced using the information the solver was able to derive. Note that this
// is only filled with the info derived during a normal search and we do not
// have any dedicated algorithm to improve it.
//
// Warning: if you didn't set keep_all_feasible_solutions_in_presolve, then
// these domains might exclude valid feasible solution. Otherwise for a
// feasibility problem, all feasible solution should be there.
//
// Warning: For an optimization problem, these will correspond to valid bounds
// for the problem of finding an improving solution to the best one found so
// far. It might be better to solve a feasibility version if one just want to
// explore the feasible region.
repeated IntegerVariableProto tightened_variables = 21;

So, the domains are the current ones when the proto is written. If you do not specify anything and have an optimization model, this will be called when the optimality has been proven, meaning the improving solution model is infeasible, thus you have infeasible domains.

Even if you stop after the first solution, as it is optimal, the written domains are infeasible.

The only option that makes sense is use stop_after_presolve:true and report the tightened bound at the beginning of the search.
Everything else is non deterministic, and may fail.

[sjindel]

Understood, thanks for the clarification. Does this imply then that the bounds on the last feasible response prior to the final (status=OPTIMAL) response will contain a feasible solution? (Provided CP-SAT first finds a feasible but non-optimal solution, which is often a reasonable assumption).

Those bounds can be extracted already using the callback interface; however, even so CP-SAT will still crash before giving the final/optimal response.

[lperron]

after a feasible solution is found, the domain is reduced if you have an objective. Laurent Perron | Operations Research | ***@***.*** | (33) 1 42 68 53 00 Le jeu. 22 janv. 2026 à 15:26, Samir Jindel ***@***.***> a écrit :

…

*sjindel* left a comment (google/or-tools#5001) <#5001 (comment)> Understood, thanks for the clarification. Does this imply then that the bounds on the last feasible response *prior to* the final (status=OPTIMAL) response will contain a feasible solution? (E.g., to bootstrap a warm re-start.) Those bounds can be extracted already using the callback interface; however, even so CP-SAT will still crash before giving the final/optimal response. — Reply to this email directly, view it on GitHub <#5001 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACUPL3M6VGUK2VHAUMRLAPL4IDMXRAVCNFSM6AAAAACSRMBUB2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOOBUGY3DOMBRGE> . You are receiving this because you modified the open/close state.Message ID: ***@***.***>