Assertion failed at the end of a solve

[lperron]

SCIP 8.0.4 Soplex 6.0.4 No papilo, no gmp

Here is the output after the solve:

SCIP Status        : problem is solved [optimal solution found]
Solving Time (sec) : 339.68
Solving Nodes      : 617205
Primal Bound       : -2.95872115235364e+05 (54 solutions)
Dual Bound         : -2.95872115235364e+05
Gap                : 0.00 %
WARNING: Original variable <> not released when freeing SCIP problem <>.
WARNING: Original variable <> not released when freeing SCIP problem <>.
Assertion failed: (! unreleasedvar), function SCIPprobFree, file prob.c, line 506.

Here is the CIP file:

STATISTICS
  Problem name     : 
  Variables        : 13 (0 binary, 4 integer, 0 implicit integer, 9 continuous)
  Constraints      : 0 initial, 7 maximal
OBJECTIVE
  Sense            : minimize
VARIABLES
  [integer] <x0>: obj=0, original bounds=[-inf,+inf]
  [integer] <x1>: obj=0, original bounds=[-inf,+inf]
  [integer] <x2>: obj=0, original bounds=[-inf,+inf]
  [integer] <x3>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x4>: obj=2366976.4609375, original bounds=[-inf,+inf]
  [continuous] <x5>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x6>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x7>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x8>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x9>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x10>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x11>: obj=0, original bounds=[-inf,+inf]
  [continuous] <x12>: obj=1, original bounds=[-inf,+inf]
CONSTRAINTS
  [disjunction] <c0>: disjunction(  [linear] <>: <x4>[C] -<x4>[C] == 0);
  [linear] <c1>: <x4>[C] -<x4>[C] >= 0;
  [disjunction] <c2>: disjunction(  [linear] <>: <x4>[C] -<x4>[C] == 0);
  [linear] <c3>: <x4>[C] -<x4>[C] >= 0;
  [disjunction] <c4>: disjunction(  [linear] <>: <x9>[C] -<x4>[C] == 0);
  [linear] <c5>: <x9>[C] -<x4>[C] >= 0;
  [nonlinear] <c6>: 2.36698e+06*(<x4>)^2+2.36698e+06*<x9>*<x4>-<x12> == 0;
END

[matbesancon]

with presolving on, I get LP errors both with SCIP and HiGHS on this problem, the nonlinear constraint with that scaling seems tricky, although it could just be rescaled no?

[matbesancon]

without presolving on master, I get no error

[svigerske]

I cannot reproduce with the cip file and

SCIP version 8.0.4 [precision: 8 byte] [memory: block] [mode: debug] [LP solver: Soplex 6.0.4] [GitHash: a8e51afd1e]
Copyright (c) 2002-2023 Zuse Institute Berlin (ZIB)

External libraries: 
  Readline 8.2         GNU library for command line editing (gnu.org/s/readline)
  Soplex 6.0.4         Linear Programming Solver developed at Zuse Institute Berlin (soplex.zib.de) [GitHash: 950b1658]
  CppAD 20180000.0     Algorithmic Differentiation of C++ algorithms developed by B. Bell (github.com/coin-or/CppAD)
  ZLIB 1.3             General purpose compression library by J. Gailly and M. Adler (zlib.net)
  AMPL/MP 4e2d45c4     AMPL .nl file reader library (github.com/ampl/mp)

either. It eventually stopped with

SCIP Status        : problem is solved [optimal solution found]
Solving Time (sec) : 984.53
Solving Nodes      : 620438
Primal Bound       : -2.95871615236004e+05 (33 solutions)
Dual Bound         : -2.95871615236004e+05
Gap                : 0.00 %

and no warning.

Since it says "original problem", it could make a big difference if SCIP was not actually run on the .cip file but the problem was input differently.

I guess that the variable name in the warning is empty because of the empty names of constraints given to the disjunction constraint, e.g.,

disjunction(  [linear] <>: <x4>[C] -<x4>[C] == 0);

If you can get some unique names into there and the warning, you could try enabling the code to debug variable capture/release in var.c (https://github.com/scipopt/scip/blob/v804/src/scip/var.c#L82-L92). This would enable printing when a variable of a specific name is captured or released, which may help to see at which place it was forgotten to release a variable, in particular if you have addr2line installed.

You may want to try whether setting, e.g., a nodelimit of 1, can speed things up.

[svigerske]

Not enough information to reproduce or fix.