BestBd not updating in a MIQCP problem
AnsweredHi! I am trying to solve a MIQCP problem with 10 continuous variables and 1000 binary variables. I have one quadratic constraint that restricts the continuous variables on a unit ball.
I find that the number in the Obj column has been increasing but the number in the BestBd column does not change at all. I was wondering if this makes sense...
Thank you!
Here is the output that
Gurobi Optimizer version 10.0.2 build v10.0.2rc0 (mac64[rosetta2])
CPU model: Apple M1
Thread count: 8 physical cores, 8 logical processors, using up to 8 threads
Optimize a model with 12001 rows, 1010 columns and 127972 nonzeros
Model fingerprint: 0xc29fa255
Model has 1 quadratic constraint
Variable types: 10 continuous, 1000 integer (1000 binary)
Coefficient statistics:
Matrix range [2e-05, 1e+01]
QMatrix range [1e+00, 1e+00]
Objective range [1e+00, 1e+00]
Bounds range [1e+00, 4e+00]
RHS range [1e+00, 5e+01]
QRHS range [1e+00, 1e+00]
Presolve time: 0.07s
Presolved: 12002 rows, 1019 columns, 127981 nonzeros
Presolved model has 9 quadratic constraint(s)
Variable types: 19 continuous, 1000 integer (1000 binary)
Starting NoRel heuristic
Found phase-1 solution: relaxation 4.76804
Found phase-1 solution: relaxation 1.76804
Found heuristic solution: objective 3.0000000
Transition to phase 2
Found heuristic solution: objective 2.0000000
Elapsed time for NoRel heuristic: 5s (best bound -1.5)
Elapsed time for NoRel heuristic: 11s (best bound -1.5)
Elapsed time for NoRel heuristic: 19s (best bound -1.5)
Elapsed time for NoRel heuristic: 26s (best bound -1.5)
Elapsed time for NoRel heuristic: 33s (best bound -1.5)
Elapsed time for NoRel heuristic: 39s (best bound -1.5)
Elapsed time for NoRel heuristic: 45s (best bound -1.5)
Elapsed time for NoRel heuristic: 53s (best bound -1.5)
Elapsed time for NoRel heuristic: 59s (best bound -1.5)
Elapsed time for NoRel heuristic: 66s (best bound -1.5)
Elapsed time for NoRel heuristic: 72s (best bound -1.5)
Elapsed time for NoRel heuristic: 83s (best bound -1.5)
Elapsed time for NoRel heuristic: 89s (best bound -1.5)
Elapsed time for NoRel heuristic: 95s (best bound -1.5)
Found heuristic solution: objective 1.5085565
Elapsed time for NoRel heuristic: 108s (best bound -1.5)
Found heuristic solution: objective 1.4859919
Elapsed time for NoRel heuristic: 114s (best bound -1.5)
Found heuristic solution: objective 1.1999996
Elapsed time for NoRel heuristic: 121s (best bound -1.5)
Root relaxation presolved: 11957 rows, 1054 columns, 127557 nonzeros
Root simplex log...
Iteration Objective Primal Inf. Dual Inf. Time
0 -9.8036828e+01 2.450518e+04 0.000000e+00 121s
353 -1.5000000e+00 0.000000e+00 0.000000e+00 121s
353 -1.5000000e+00 0.000000e+00 0.000000e+00 121s
Root relaxation: objective -1.500000e+00, 353 iterations, 0.16 seconds (0.35 work units)
Nodes | Current Node | Objective Bounds | Work
Expl Unexpl | Obj Depth IntInf | Incumbent BestBd Gap | It/Node Time
0 0 -1.50000 0 94 1.20000 -1.50000 225% - 121s
0 0 -1.50000 0 93 1.20000 -1.50000 225% - 122s
0 0 -1.50000 0 93 1.20000 -1.50000 225% - 122s
0 0 -1.50000 0 86 1.20000 -1.50000 225% - 124s
0 0 -1.50000 0 87 1.20000 -1.50000 225% - 125s
0 0 -1.50000 0 86 1.20000 -1.50000 225% - 125s
0 0 -1.50000 0 86 1.20000 -1.50000 225% - 126s
0 0 -1.50000 0 86 1.20000 -1.50000 225% - 126s
0 0 -1.50000 0 86 1.20000 -1.50000 225% - 127s
H 0 0 1.1506758 -1.50000 230% - 127s
0 0 -1.50000 0 86 1.15068 -1.50000 230% - 127s
0 0 -1.50000 0 88 1.15068 -1.50000 230% - 128s
0 0 -1.50000 0 87 1.15068 -1.50000 230% - 129s
0 0 -1.50000 0 87 1.15068 -1.50000 230% - 130s
0 0 -1.50000 0 87 1.15068 -1.50000 230% - 130s
0 0 -1.50000 0 87 1.15068 -1.50000 230% - 131s
0 0 -1.50000 0 87 1.15068 -1.50000 230% - 131s
0 0 -1.50000 0 87 1.15068 -1.50000 230% - 131s
H 0 0 1.1506746 -1.50000 230% - 154s
0 2 -1.50000 0 87 1.15067 -1.50000 230% - 155s
H 31 33 1.1506741 -1.50000 230% 381 157s
H 33 33 1.1341423 -1.50000 232% 371 157s
H 67 67 1.0599894 -1.50000 242% 210 181s
936 883 0.83765 173 - 1.05999 -1.50000 242% 24.4 189s
971 892 -1.02710 37 200 1.05999 -1.50000 242% 25.4 229s
H 976 892 1.0599883 -1.50000 242% 25.4 233s
979 926 -1.50000 37 68 1.05999 -1.50000 242% 25.6 237s
1264 1177 -1.50000 41 73 1.05999 -1.50000 242% 22.0 255s
2063 1822 -1.50000 73 72 1.05999 -1.50000 242% 16.3 266s
2382 2084 0.11644 224 87 1.05999 -1.50000 242% 15.0 523s
2385 2086 0.23868 49 89 1.05999 -1.50000 242% 15.0 526s
2386 2087 0.03451 128 87 1.05999 -1.50000 242% 15.0 534s
2388 2088 0.75025 242 87 1.05999 -1.50000 242% 15.0 535s
2398 2095 0.07609 27 88 1.05999 -1.50000 242% 14.9 540s
2406 2100 0.76476 223 87 1.05999 -1.50000 242% 14.9 545s
2408 2101 0.36287 100 87 1.05999 -1.50000 242% 14.9 555s
2439 2123 -0.37483 19 92 1.05999 -1.50000 242% 25.3 567s
2472 2137 0.92600 23 1 1.05999 -1.50000 242% 26.2 571s
2949 2432 -0.09963 29 72 1.05999 -1.50000 242% 28.5 586s
3014 2418 -1.50000 30 78 1.05999 -1.50000 242% 28.5 596s
3109 2481 -1.50000 31 78 1.05999 -1.50000 242% 28.2 610s
5333 3593 -1.50000 42 80 1.05999 -1.50000 242% 22.6 615s
11334 7971 -1.50000 55 76 1.05999 -1.50000 242% 16.6 620s
14262 10227 -0.37567 61 67 1.05999 -1.50000 242% 15.2 643s
14570 10919 -1.50000 63 77 1.05999 -1.50000 242% 15.2 645s
19699 15376 -1.50000 72 77 1.05999 -1.50000 242% 12.7 667s
22595 20539 -1.50000 82 77 1.05999 -1.50000 242% 12.5 670s
35736 31687 -1.50000 92 79 1.05999 -1.50000 242% 9.3 675s
44542 38883 1.05253 592 1 1.05999 -1.50000 242% 8.5 681s
49957 43975 -1.50000 115 82 1.05999 -1.50000 242% 8.3 685s
H53127 47018 1.0438069 -1.50000 244% 8.7 689s
54518 48097 -1.50000 133 83 1.04381 -1.50000 244% 8.8 690s
H54954 47178 1.0034430 -1.50000 249% 8.8 714s
H54958 47178 1.0034430 -1.50000 249% 8.8 714s
H54974 46764 0.9587022 -1.50000 256% 8.8 719s
H54975 46743 0.9506084 -1.50000 258% 8.8 722s
56496 47885 -1.50000 139 83 0.95061 -1.50000 258% 8.8 745s
56498 47885 -0.03215 657 63 0.95061 -1.50000 258% 8.8 755s
H56501 47885 0.9506084 -1.50000 258% 8.8 761s
H56502 47885 0.9506084 -1.50000 258% 8.8 762s
56505 47885 0.67621 405 43 0.95061 -1.50000 258% 8.8 767s
57963 49646 -1.50000 144 84 0.95061 -1.50000 258% 8.8 778s
61328 52080 -0.06079 150 77 0.95061 -1.50000 258% 8.8 780s
63605 52740 -1.50000 173 82 0.95061 -1.50000 258% 10.0 785s
66440 55026 -0.05786 180 66 0.95061 -1.50000 258% 10.2 808s
H66446 55026 0.9506073 -1.50000 258% 10.2 811s
66874 55546 -1.50000 183 84 0.95061 -1.50000 258% 10.2 816s
67567 55547 -1.50000 186 85 0.95061 -1.50000 258% 10.2 845s
67582 55572 -0.04898 186 72 0.95061 -1.50000 258% 10.3 854s
67633 55631 -0.03314 188 87 0.95061 -1.50000 258% 10.3 855s
73768 60793 -1.50000 209 90 0.95061 -1.50000 258% 10.5 866s
80804 66275 -1.50000 222 81 0.95061 -1.50000 258% 10.2 870s
86051 70213 -1.50000 231 83 0.95061 -1.50000 258% 10.1 880s
92723 75260 -1.50000 246 81 0.95061 -1.50000 258% 9.8 885s
97870 78776 -1.50000 267 81 0.95061 -1.50000 258% 9.8 890s
98347 78830 -1.50000 270 81 0.95061 -1.50000 258% 9.9 897s
100816 80393 -1.50000 282 80 0.95061 -1.50000 258% 10.0 900s
101278 80448 -1.50000 287 81 0.95061 -1.50000 258% 10.1 907s
101992 80630 -0.04179 297 84 0.95061 -1.50000 258% 10.3 910s
105466 83567 -1.50000 307 80 0.95061 -1.50000 258% 10.3 915s
107827 85415 -1.50000 321 79 0.95061 -1.50000 258% 10.5 923s
109129 86386 -1.50000 327 80 0.95061 -1.50000 258% 10.6 932s
110270 87010 -1.50000 335 79 0.95061 -1.50000 258% 10.6 935s
112989 89503 -0.02829 355 71 0.95061 -1.50000 258% 10.8 948s
114463 90314 -1.50000 363 80 0.95061 -1.50000 258% 10.8 950s
116065 91776 0.90710 578 1 0.95061 -1.50000 258% 10.9 955s
118505 94119 -1.50000 375 81 0.95061 -1.50000 258% 10.9 965s
121165 94975 -0.02855 390 67 0.95061 -1.50000 258% 11.1 971s
121850 95163 -1.50000 403 80 0.95061 -1.50000 258% 11.4 980s
122347 95604 -1.50000 428 82 0.95061 -1.50000 258% 11.9 985s
122442 95667 -1.50000 433 80 0.95061 -1.50000 258% 12.0 991s
122679 95848 0.67621 686 41 0.95061 -1.50000 258% 12.2 996s
123005 96132 -1.50000 451 80 0.95061 -1.50000 258% 12.5 1004s
123098 96359 -1.50000 453 82 0.95061 -1.50000 258% 12.6 1006s
123385 96739 -0.01571 457 81 0.95061 -1.50000 258% 12.7 1012s
124479 97266 -0.07440 470 63 0.95061 -1.50000 258% 12.8 1045s
124488 97870 -1.50000 471 82 0.95061 -1.50000 258% 12.8 1051s
125147 98216 -0.05160 476 76 0.95061 -1.50000 258% 12.9 1058s
125690 98432 -1.50000 489 81 0.95061 -1.50000 258% 13.1 1068s
125813 98596 -1.50000 492 80 0.95061 -1.50000 258% 13.2 1070s
126022 99617 0.00000 496 71 0.95061 -1.50000 258% 13.3 1078s
127398 100719 -1.50000 499 81 0.95061 -1.50000 258% 13.3 1080s
128706 102133 -1.50000 505 81 0.95061 -1.50000 258% 13.3 1098s
130442 103552 -1.50000 508 81 0.95061 -1.50000 258% 13.2 1100s
133928 105479 -1.50000 517 81 0.95061 -1.50000 258% 13.0 1108s
134406 105479 0.94858 524 - 0.95061 -1.50000 258% 13.0 1111s
136033 107843 -1.50000 523 81 0.95061 -1.50000 258% 13.0 1115s
141214 111960 -1.50000 535 81 0.95061 -1.50000 258% 12.8 1120s
145784 114096 -1.50000 539 81 0.95061 -1.50000 258% 12.6 1125s
149282 115804 -1.50000 550 81 0.95061 -1.50000 258% 12.6 1130s
149658 116516 -1.50000 553 86 0.95061 -1.50000 258% 12.6 1139s
150522 116775 -0.02673 553 78 0.95061 -1.50000 258% 12.5 1140s
152920 118837 0.94041 318 - 0.95061 -1.50000 258% 12.5 1145s
155163 119929 -1.50000 578 82 0.95061 -1.50000 258% 12.7 1153s
155356 120406 -1.50000 584 82 0.95061 -1.50000 258% 12.7 1155s
157081 121521 -1.50000 594 82 0.95061 -1.50000 258% 12.8 1168s
157502 122058 -1.50000 599 83 0.95061 -1.50000 258% 12.8 1170s
158862 123116 -1.50000 620 85 0.95061 -1.50000 258% 13.1 1175s
160978 124630 -1.50000 633 83 0.95061 -1.50000 258% 13.1 1193s
161379 125091 -1.50000 641 84 0.95061 -1.50000 258% 13.2 1195s
163078 127280 -1.50000 658 84 0.95061 -1.50000 258% 13.3 1200s
165363 128394 -1.50000 675 83 0.95061 -1.50000 258% 13.5 1206s
166011 128498 -1.50000 680 83 0.95061 -1.50000 258% 13.5 1215s
167371 129797 -1.50000 690 84 0.95061 -1.50000 258% 13.6 1220s
168332 130008 -1.50000 708 82 0.95061 -1.50000 258% 13.8 1236s
169232 130655 -0.02625 723 79 0.95061 -1.50000 258% 14.0 1240s
170131 130829 -0.01912 740 78 0.95061 -1.50000 258% 14.2 1271s
170354 131097 -1.50000 754 82 0.95061 -1.50000 258% 14.4 1280s
170594 131283 -1.50000 768 82 0.95061 -1.50000 258% 14.6 1290s
170850 131640 -1.50000 781 83 0.95061 -1.50000 258% 14.8 1297s
171371 132530 -1.50000 798 82 0.95061 -1.50000 258% 15.0 1301s
173265 133829 -1.50000 817 82 0.95061 -1.50000 258% 15.1 1310s
174112 134115 -0.22648 826 76 0.95061 -1.50000 258% 15.2 1315s
(some outputs omitted)
393327 289951 0.67621 741 41 0.95061 -0.44566 147% 16.3 1650s
396032 291924 0.67621 814 42 0.95061 -0.44566 147% 16.4 1655s
401612 295609 0.67020 726 53 0.95061 -0.44532 147% 16.4 1660s
405102 297616 0.94630 583 38 0.95061 -0.44532 147% 16.4 1666s
406626 298396 0.67621 589 59 0.95061 -0.44532 147% 16.5 1670s
407063 298784 0.67621 634 57 0.95061 -0.44532 147% 16.7 1675s
407692 299857 0.67621 664 56 0.95061 -0.44532 147% 16.8 1680s
411486 302915 0.67621 703 61 0.95061 -0.44532 147% 16.8 1685s
412450 304404 0.67621 746 58 0.95061 -0.44532 147% 16.9 1691s
414950 305107 0.67621 777 62 0.95061 -0.44532 147% 17.0 1695s
415475 305872 0.67621 880 60 0.95061 -0.44532 147% 17.1 1701s
417410 308203 0.67621 919 62 0.95061 -0.44532 147% 17.2 1705s
420465 309857 infeasible 811 0.95061 -0.44532 147% 17.2 1710s
424224 312630 cutoff 594 0.95061 -0.44532 147% 17.2 1715s
427590 314970 cutoff 625 0.95061 -0.44532 147% 17.2 1720s
429110 315458 infeasible 651 0.95061 -0.44532 147% 17.3 1725s
429909 316534 0.82955 674 43 0.95061 -0.44532 147% 17.3 1730s
433892 320071 cutoff 703 0.95061 -0.44532 147% 17.3 1736s
437042 322472 infeasible 726 0.95061 -0.44532 147% 17.3 1740s
440585 324858 infeasible 753 0.95061 -0.44532 147% 17.3 1745s
445095 327789 0.67621 781 53 0.95061 -0.44532 147% 17.3 1751s
447444 329485 0.67621 799 52 0.95061 -0.44532 147% 17.3 1755s
449167 330020 infeasible 824 0.95061 -0.44532 147% 17.4 1760s
450680 330387 0.67621 860 51 0.95061 -0.44532 147% 17.5 1765s
451421 331466 cutoff 911 0.95061 -0.44532 147% 17.7 1771s
454540 334164 0.66787 537 47 0.95061 -0.43885 146% 17.7 1775s
458555 338941 0.67621 634 49 0.95061 -0.43885 146% 17.6 1780s
466376 345746 0.67621 680 45 0.95061 -0.43885 146% 17.5 1785s
472708 349662 0.67621 725 48 0.95061 -0.43885 146% 17.3 1790s
476946 353203 0.67621 754 47 0.95061 -0.43885 146% 17.3 1795s
478382 353984 0.67621 798 48 0.95061 -0.43885 146% 17.4 1800s
480203 356491 0.67621 846 47 0.95061 -0.43885 146% 17.4 1806s
485464 359849 0.67621 891 46 0.95061 -0.43885 146% 17.4 1810s
488949 361879 0.94978 683 1 0.95061 -0.43885 146% 17.3 1816s
493025 366248 0.94812 246 15 0.95061 -0.39584 142% 17.2 1820s
496289 367622 cutoff 565 0.95061 -0.39584 142% 17.3 1825s
502034 371290 0.67371 834 41 0.95061 -0.39584 142% 17.2 1830s
506138 373726 0.67621 712 50 0.95061 -0.38527 141% 17.3 1835s
509753 376338 0.94095 629 - 0.95061 -0.38527 141% 17.3 1843s
511502 376568 0.67621 753 45 0.95061 -0.38527 141% 17.3 1845s
514779 377775 0.67621 777 46 0.95061 -0.38527 141% 17.3 1850s
517118 379273 0.67621 818 45 0.95061 -0.38527 141% 17.3 1855s
520117 381059 0.67621 858 42 0.95061 -0.38527 141% 17.4 1860s
520860 381588 0.67621 906 45 0.95061 -0.38527 141% 17.5 1865s
521404 382903 0.80473 135 30 0.95061 -0.36791 139% 17.6 1870s
521888 382903 0.95045 576 - 0.95061 -0.36791 139% 17.6 1876s
524787 384193 0.67621 676 59 0.95061 -0.36791 139% 17.6 1880s
525672 384837 0.67621 716 56 0.95061 -0.36791 139% 17.7 1885s
526190 385399 0.67621 756 57 0.95061 -0.36791 139% 17.8 1890s
527980 386535 0.67621 805 55 0.95061 -0.36791 139% 17.9 1895s
529036 386939 0.67621 710 57 0.95061 -0.36791 139% 18.0 1900s
529393 387325 0.67621 751 56 0.95061 -0.36791 139% 18.1 1905s
531012 388371 0.67621 798 55 0.95061 -0.36791 139% 18.2 1911s
531813 389307 0.67621 817 57 0.95061 -0.36791 139% 18.3 1915s
534616 390678 0.67621 867 57 0.95061 -0.36791 139% 18.3 1921s
535426 391354 0.67621 900 55 0.95061 -0.36791 139% 18.4 1925s
536931 392077 0.67621 763 58 0.95061 -0.36791 139% 18.5 1931s
538351 392928 0.67621 806 54 0.95061 -0.36791 139% 18.6 1937s
539280 393723 0.67621 845 54 0.95061 -0.36791 139% 18.7 1940s
540654 394622 0.67621 901 52 0.95061 -0.36791 139% 18.8 1947s
541854 395929 0.67621 867 64 0.95061 -0.36791 139% 18.9 1952s
543316 396149 0.67621 902 63 0.95061 -0.36791 139% 19.0 1957s
543482 397065 0.67621 919 61 0.95061 -0.36791 139% 19.0 1960s
546293 398201 0.67621 894 62 0.95061 -0.36791 139% 19.1 1966s
549598 401412 0.67621 928 60 0.95061 -0.36791 139% 19.1 1971s
553437 404072 0.66837 912 55 0.95061 -0.36791 139% 19.1 1975s
562973 409989 0.70922 939 59 0.95061 -0.36791 139% 18.9 1980s
564539 410755 0.94576 269 4 0.95061 -0.36312 138% 18.9 1986s
568570 413530 0.30379 667 62 0.95061 -0.34097 136% 18.8 1990s
(omitted some outputs)
640102 466701 0.67621 414 33 0.95061 -0.31018 133% 17.7 2065s
643970 468851 0.67621 446 32 0.95061 -0.31018 133% 17.7 2071s
647524 470995 0.67466 618 31 0.95061 -0.31018 133% 17.7 2075s
650860 473266 cutoff 379 0.95061 -0.31018 133% 17.6 2080s
651427 473791 0.94901 594 1 0.95061 -0.31018 133% 17.7 2085s
655718 477449 0.67621 115 51 0.95061 -0.31018 133% 17.6 2091s
657847 478521 0.67621 166 39 0.95061 -0.31018 133% 17.7 2096s
661605 481173 0.67621 223 41 0.95061 -0.31018 133% 17.6 2100s
665404 483570 0.67621 254 36 0.95061 -0.31018 133% 17.6 2105s
666743 484422 0.67621 297 36 0.95061 -0.31018 133% 17.6 2110s
669031 486411 0.67621 341 36 0.95061 -0.31018 133% 17.6 2115s
670921 488016 0.67621 409 37 0.95061 -0.31018 133% 17.7 2121s
672740 488188 0.67621 448 37 0.95061 -0.31018 133% 17.7 2125s
675613 490514 0.67621 493 36 0.95061 -0.31018 133% 17.7 2130s
678214 491779 0.67621 531 36 0.95061 -0.31018 133% 17.7 2135s
683605 495470 0.67621 568 36 0.95061 -0.31018 133% 17.6 2140s
686678 496136 0.67621 604 36 0.95061 -0.31018 133% 17.6 2145s
688221 497943 0.67621 619 36 0.95061 -0.31018 133% 17.6 2151s
690150 498841 0.67621 672 36 0.95061 -0.31018 133% 17.7 2155s
691703 499777 0.67621 734 38 0.95061 -0.31018 133% 17.7 2160s
9 410755 0.94576 269 4 0.95061 -0.36312 138% 18.9 1986s
568570 413530 0.30379 667 62 0.95061 -0.34097 136% 18.8 1990s
576947 420502 0.67621 787 56 0.95061 -0.33162 135% 18.7 1995s
578320 421075 0.67621 796 53 0.95061 -0.33162 135% 18.6 2000s
578628 421262 0.94989 624 - 0.95061 -0.33162 135% 18.7 2005s
582445 423811 0.67621 833 51 0.95061 -0.33162 135% 18.6 2010s
584802 425089 0.67621 861 50 0.95061 -0.33162 135% 18.6 2015s
(omitted some outputs)
615729 445408 0.95047 691 1 0.95061 -0.31018 133% 18.1 2045s
624466 453034 0.94784 515 25 0.95061 -0.31018 133% 17.9 2050s
627303 455855 0.67621 327 46 0.95061 -0.31018 133% 17.9 2055s
633859 460941 0.67621 362 34 0.95061 -0.31018 133% 17.8 2060s
640102 466701 0.67621 414 33 0.95061 -0.31018 133% 17.7 2065s
643970 468851 0.67621 446 32 0.95061 -0.31018 133% 17.7 2071s
647524 470995 0.67466 618 31 0.95061 -0.31018 133% 17.7 2075s
650860 473266 cutoff 379 0.95061 -0.31018 133% 17.6 2080s
651427 473791 0.94901 594 1 0.95061 -0.31018 133% 17.7 2085s
655718 477449 0.67621 115 51 0.95061 -0.31018 133% 17.6 2091s
657847 478521 0.67621 166 39 0.95061 -0.31018 133% 17.7 2096s
661605 481173 0.67621 223 41 0.95061 -0.31018 133% 17.6 2100s
665404 483570 0.67621 254 36 0.95061 -0.31018 133% 17.6 2105s
-
Hi,
The behavior you see makes sense. The reason is the constraint
I have one quadratic constraint that restricts the continuous variables on a unit ball.
which most likely introduces a lot of symmetry into the model. The best way to deal with the symmetry is to introduce an ordering of variables, e.g., \(x_1 \leq x_2 \leq x_3 \leq ...\).
Out of curiosity, could you please share the model? Maybe there is something we could improve on. Note that uploading files in the Community Forum is not possible but we discuss an alternative in Posting to the Community Forum.
Best regards,
Jaromił0 -
Hi Jaromił,
Thank you for your prompt response!
1. I want to check how the BestBound is generated. Since my problem is a minimization problem, the Bestbound is the minimum (not the maximum) of all relaxations at the leaf nodes, right? Does it mean that, if the Bestbound is not improving, there are lots of leaf nodes that need to be explored?
2. Regarding the ordering constraints, there is no guarantee that a particular ordering is optimal for my problem. Trying out all (#continuous variable)! seems impossible.
3. Here is the link to the replication files
https://drive.google.com/drive/folders/1mt-2IkEsLdh5vk4hafYNmu5Y1hj_PduV?usp=drive_link
'Trial_0919.m' will replicate the results I posted above.
The file 'CR_draw.m', 'D_cr.m', and 'gen_diag_matrix_cr.m' are auxiliary files that generate inputs to the main file.
'Trial_0911.m' solves the same problem as the one in 'Trial_0919.m', but with fewer constraints. Unfortunately, Trial_0911.m works better than Trial_0919.m based on my experiments. Trial_0911.m converges in one hour on a computer node with 48 cores. I have not tried Trial_0919.m, but on my computer Trial_0919.m performs worse.
Finally, let me briefly mention the setup of my problem. This is a Value-at-Risk problem, where I want to minimize the quantile of a linear combination of random variables with weights that square-sum to 1.
Decision Variable:
1. w (continuous): a n-dimensional vector restricted on a unit ball;
2. D_i, i=1,...,n_s: binary variable
3. t (continuous): a scalar: We have a priori knowledge that t is between -1.5 to 3.
Parameters:
1. \{Z_i \}_{i=1}^{ns} a set of n-dimensional vector, cardinality=ns
2. qmax: a scalar: it would be on the order of sqrt( |#continuous variable|)
4. qmin: a scalar: it would be on the order of -sqrt( |#continuous variable|)
5. q_alpha: a scalar: it is set to -3
6. ns: number of binary variables
7. alpha(ns): a small positive number: for example, 0.05 or 0.025
The problem Trial_0911.m solves
The first set of constraints requires D_i=0 if w'Z_i-t > q_{max}. The second constraint restricts the number of D_i.
The file Trial_0919.m solves
where cij = ||Z_i-Z_j||, the Euclidean distance between Z_i and Z_j. The set of constraint (3) requires D_i and D_j to obey the order of w'Z_i and w'Z_j. There are ns * (ns-1) such constraints. I know the optimal solution will satisfy. There are too many of them so I subsampled some in my implementation.
The additional set of constraints (2) adds on (1), which requires that the condition D_i=0 and w'Z_i-t > q_{max} mutually imply each other (up to the equality sign), as oppose to a one-way direction if we only impose (1).
Thank you!
Haoge
0 -
Hi Haoge,
1. I want to check how the BestBound is generated. Since my problem is a minimization problem, the Bestbound is the minimum (not the maximum) of all relaxations at the leaf nodes, right?
Correct.
Does it mean that, if the Bestbound is not improving, there are lots of leaf nodes that need to be explored?
Yes, additionally this means that there may be symmetry involved and the solver has to explore all symmetrical areas before the lower bound improves. Usually, adding more user knowledge as constraint helps to overcome such cases. Of course this is not always possible.
2. Regarding the ordering constraints, there is no guarantee that a particular ordering is optimal for my problem. Trying out all (#continuous variable)! seems impossible.
There may be no guarantee that a particular ordering is optimal, but is there a guarantee that if an ordering is optimal, that you can swap some variable values while remaining the same optimal objective value? If so, then you should try adding this knowledge into the model.
3. Here is the link to the replication files
Could you please generate an MPS or LP file via the gurobi_write() function and share the generated model file? This would avoid user error on my side.
Best regards,
Jaromił0 -
Hi Jaromił,
Thank you very much for the clarification and explanation!
I uploaded the mps file on to the
https://drive.google.com/drive/folders/1mt-2IkEsLdh5vk4hafYNmu5Y1hj_PduV?usp=drive_link
They are under the name trial_0919.mps and trial_0911.mps.
Looking forward to hearing from you!
Thank you!
Haoge
0 -
Hi Haoge,
I guess that sticking to the trial_911 formulation looks more promising. I cannot explain why the trial_919 formulation performs so much worse.
I think that the best way to improve performance now is to provide more model specific knowledge that you have. For example, add bounds \([-1.5, 3]\) for variable \(t\). In best case, you might be able to deduce a better lower bound for variable \(t\), maybe it is always \(t\geq \delta > 0\) for some \(\delta > 0\)?
Do you think that one could say something about the values of the \(Z_i\) variables? For example, could you say that the values in an optimal solution for some of those variables are rather close to \(1\) and for others close to \(0\) or should all always be close to \(\approx 0.5\) (or any other value)? It might make sense to introduce finite bounds, e.g., \(-0.75 \leq Z_i \leq 0.75\), if any are available. It would be even better if you somehow have the information which of the variables should be positive and which negative.Playing with the MIPFocus might also bring some performance.
Best regards,
Jaromił0 -
Hi Jaromił,
Thank you very much for the suggestions! Would you mind if I ask you three additional questions?
1. I saw in the documentation of the MIPFocus parameter that it only affects MIP problems. Will it also affect the MIQCP problem that I am solving?
2. Would you mind explaining what the Symmetry parameter does? I wonder if it may be an additional parameter that I can play around, since you mentioned symmetry in your comment.
3. If I have a guess (that is, a feasible solution that, I think, is close to the optimal value), is there a way that I can incorporate it into my problem? Do you think it would increase the performance, in particular, in terms of closing the distance with the BestBound?
Thank you!
Haoge
Haoge
0 -
Hi Haoge,
I saw in the documentation of the MIPFocus parameter that it only affects MIP problems. Will it also affect the MIQCP problem that I am solving?
Yes, it will always affect quadratic models since these are handled in the same way as MIPs.
Would you mind explaining what the Symmetry parameter does? I wonder if it may be an additional parameter that I can play around, since you mentioned symmetry in your comment.
The Symmetry parameter controls Gurobi's symmetry detection level. A higher setting means that more time is spent to find symmetries. I don't think that this parameter will help your model, because the symmetry you construct through the "border of an n-dimensional ball" constraint is not something that can be exploited in general. It is just often the case that with additional model knowledge, in such a case, it is actually correct to introduce symmetry breaking constraints of the form \(x_1 \leq x_2 \leq \dots\). However, this has to be decided by the modeler.
If I have a guess (that is, a feasible solution that, I think, is close to the optimal value), is there a way that I can incorporate it into my problem? Do you think it would increase the performance, in particular, in terms of closing the distance with the BestBound?
You can provide your guess as described in the Knowledge Base article How do I use MIP starts? Often a good initial feasible point improves performance. However, there is no guarantee.
I think that shrinking the domains of the variables with the knowledge you have together with a good initial point might be the way to go here.
Maybe you could also do some manual presolving and check whether there are some constraints which can be dropped from the beginning due to some properties that you know about.
Best regards,
Jaromił0
Please sign in to leave a comment.
Comments
7 comments