HiGHS 1.13 Terminates MIP Prematurely Similar to Issue #1865 Using Same Problem

[jeffreydeankelly2]

Dear HiGHS;

Below is the log output for same issue reported in #1865.

The correct objective function is 937000.0.

In addition, 1.13 takes 319-seconds to solve prematurely and 1.12 solves to 937000.0 in 34-seconds.

All the best - Jeff

Running HiGHS 1.13.0 (git hash: 1bce6d5): Copyright (c) 2026 under MIT licence terms
Cols: 2 lower bounds less than or equal to -1e+20 are treated as -Infinity
Cols: 2 upper bounds greater than or equal to 1e+20 are treated as +Infinity
Rows: 25170 lower bounds less than or equal to -1e+20 are treated as -Infinity
SOLVE
MIP has 33392 rows; 15608 cols; 318754 nonzeros; 5538 integer variables (5538 binary)
Coefficient ranges:
Matrix [1e-04, 2e+04]
Cost [1e+00, 1e+00]
Bound [5e-01, 1e+04]
RHS [1e+00, 2e+04]
Presolving model
27851 rows, 10067 cols, 306660 nonzeros 0s
19594 rows, 9535 cols, 288379 nonzeros 0s
15495 rows, 7822 cols, 279873 nonzeros 0s
8821 rows, 5153 cols, 265826 nonzeros 0s
4916 rows, 3037 cols, 64501 nonzeros 75s
4483 rows, 2366 cols, 61441 nonzeros 106s
Presolve reductions: rows 4483(-28909); columns 2366(-13242); nonzeros 61441(-257313)

Src: B => Branching; C => Central rounding; F => Feasibility pump; H => Heuristic;
I => Shifting; J => Feasibility jump; L => Sub-MIP; P => Empty MIP; R => Randomized rounding;
S => Solve LP; T => Evaluate node; U => Unbounded; X => User solution; Y => HiGHS solution;
Z => ZI Round; l => Trivial lower; p => Trivial point; u => Trivial upper; z => Trivial zero
0 0 0 0.00% 1573466.666667 -inf inf 0 0 0 0 106.7s
0 0 0 0.00% 1083591.573449 -inf inf 0 0 4 2244 107.0s
0 0 0 0.00% 1082738.9208 -inf inf 4208 306 204 2885 113.0s
0 0 0 0.00% 1081803.469916 -inf inf 5819 438 340 15497 121.3s

0.3% inactive integer columns, restarting
Model after restart has 3491 rows, 2024 cols (537 bin., 0 int., 790 impl., 697 cont., 0 dom.fix.), and 48014 nonzeros
0 0 0 0.00% 1081803.469916 -inf inf 214 0 0 18062 153.9s
0 0 0 0.00% 1076639.783768 -inf inf 214 87 4 19208 154.3s
0 0 0 0.00% 1076009.630827 -inf inf 2696 181 533 23564 210.6s
2 0 1 50.00% 1076009.630827 -inf inf 2837 181 13484 102095 261.3s
4 0 2 75.00% 1076009.630827 -inf inf 2973 181 19506 126743 293.0s
8 0 4 93.75% 1076009.630827 -inf inf 3037 181 21353 140098 301.1s
impl> Logistics-Feasible Solution # 1 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT1
T 8 0 4 93.75% 1076009.630827 791854.268293 35.88% 3052 181 21468 140098 301.8s
impl> Logistics-Feasible Solution # 2 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT2
T 19 0 5 94.14% 1076009.630827 792470.609756 35.78% 3055 181 20944 142365 301.9s
impl> Logistics-Feasible Solution # 3 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT3
T 29 0 7 94.92% 1076009.630827 801845.609756 34.19% 3064 181 8276 143923 303.1s
impl> Logistics-Feasible Solution # 4 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT4
T 38 0 8 95.12% 1076009.630827 810970.609756 32.68% 3066 181 8285 143995 303.2s
impl> Logistics-Feasible Solution # 5 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT5
T 45 0 9 95.31% 1076009.630827 811231.829268 32.64% 3068 181 8388 144072 303.5s
impl> Logistics-Feasible Solution # 6 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT6
T 55 0 10 95.51% 1076009.630827 820106.829268 31.20% 3069 181 8391 144243 303.6s
impl> Logistics-Feasible Solution # 7 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT7
T 62 0 11 95.70% 1076009.630827 820339.02439 31.17% 3070 181 8410 144274 303.8s
impl> Logistics-Feasible Solution # 8 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT8
T 69 0 12 96.09% 1076009.630827 821500 30.98% 3071 181 8422 144325 303.9s
84 0 15 98.54% 1076009.630827 821500 30.98% 3258 181 12294 177161 317.8s
impl> Logistics-Feasible Solution # 9 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT9
T 84 0 15 98.54% 1076009.630827 828964.02439 29.80% 3259 181 12296 177161 318.2s
impl> Logistics-Feasible Solution # 10 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT10
T 91 0 16 99.22% 1076009.630827 830125 29.62% 3262 181 12205 177192 318.6s
impl> Logistics-Feasible Solution # 11 with Objective Function = 0.0000000000E+000 ( 0.917028000000000E+006)
SOLUTIONSPOT11
T 105 0 17 99.61% 931850 931850 0.00% 3506 186 10492 177655 319.2s
110 0 18 100.00% 931850 931850 0.00% 3507 186 10492 178147 319.2s

Solving report
Status Optimal
Primal bound 931850
Dual bound 931850
Gap 0% (tolerance: 0.01%)
P-D integral 5.36753678461
Solution status feasible
Timing 319.17
Max sub-MIP depth 2
Nodes 110
Repair LPs 0
LP iterations 178147
87425 (strong br.)
2054 (separation)
18134 (heuristics)

[jajhall]

Thanks, we'll look into this and be sure to add the model to our test set.

[Opt-Mucca]

I can look into this tomorrow for a proper diagnosis. (@fwesselm purely from the logs this looks like it comes from a presolve change)

[Opt-Mucca]

@jeffreydeankelly2 How did you build HiGHS (I can't recreate your solution path from the .lp or .mps file)? I'm also not seeing the infinity warnings at the top of my log, so I assume this is being solved in some form of pipeline? I can't recreate getting an incorrect objective locally.

As for the increased solving time: This is coming from our new enumerateSolutions presolve (the different MIP paths between releases on my machine after presolve seem to be just be noise and not major time differences). We need to more finely tune the stopping criteria for this algorithm apparently. I will look into this more tomorrow.

[jeffreydeankelly2]

This problem is generated from our modeling system where the *.LP / *.MPS files are generated directly from the HIGHS call to write the *.LP / *.MPS files.

These files were used to resolve the issue with 1.7.2 and from 1.8.0 to 1.12.0 the issue was fixed but now it has re-appeared.

[fwesselm]

@jeffreydeankelly2 How did you build HiGHS (I can't recreate your solution path from the .lp or .mps file)? I'm also not seeing the infinity warnings at the top of my log, so I assume this is being solved in some form of pipeline? I can't recreate getting an incorrect objective locally.

As for the increased solving time: This is coming from our new enumerateSolutions presolve (the different MIP paths between releases on my machine after presolve seem to be just be noise and not major time differences). We need to more finely tune the stopping criteria for this algorithm apparently. I will look into this more tomorrow.

@Opt-Mucca, I can also have a look, if you like. Let me know.

[Opt-Mucca]

@jeffreydeankelly2 you're not re-reading in the instance directly from the created .mps or .lp file though? (The warnings make me concerned that there's going to be some minor difference between what you've modelled and what HiGHS is writing to file)
@fwesselm If you have the time then please! From what I found yesterday it looks like we'd need to add some second stopping criteria to enumerateSolutions, e.g., if enough consecutive constraints have been probed and found to be useless or something like the dynamic budget in standard probing.

[jeffreydeankelly2]

Yes, the LP and MPS files are those generated after my modeling system has transferred the matrix, etc. to HiGHS. Then, I call the HiGHS routine to generate the LP and MPS files.

When you run the LP or MPS files, do you get the objective function value of 937000.0? If you do not, then this is enough reproducibility on your side I believe.

[Opt-Mucca]

I am getting 937000 (the correct objective function) when I run both the .mps and .lp file. The huge amount of time spent in enumerateSolutions is only the .lp file though.