Optimizing over the Split Closure Anureet Saxena ACO PhD Student, Tepper School of Business, Carnegie Mellon University. (Joint Work with Egon Balas) Talk Outiline • Cutting Planes Commercial • Split Closure Separation Problem • PMILP & Deparametrization • Computational Results • Support Size & Sparsity • Support Coefficients • Cuts Statistics • arki001 solved Anureet Saxena, TSoB 1 MIP Model min cx Ax ¸ b xj 2 Z 8 j2N1 Contains xj ¸ 0 j2N xj · uj j2N1 N1: set of integer variables Incumbent Fractional Solution Anureet Saxena, TSoB 2 Taxonomy of Cutting Planes Fractional Basic MIR Fractional Gomory Split Cuts Fractional Basic Feas Chvatal MIG Mixed Integer Basic Feas Mixed Integer Basic Intersection Basic + Strengthening Intersection Basic L&P L&P + Strengthening Intersection Basic Feas + Strengthening Intersection Basic Feas Anureet Saxena, TSoB Simple Disjunctive 3 Taxonomy of Cutting Planes Fractional Basic MIR Fractional Gomory Split Cuts Fractional Basic Feas Chvatal MIG Mixed Integer Basic Feas Mixed Integer Basic Intersection Basic + Strengthening Intersection Basic L&P L&P + Strengthening Intersection Basic Feas + Strengthening Intersection Basic Feas Anureet Saxena, TSoB Simple Disjunctive 4 Taxonomy of Cutting Planes Fractional Basic MIR Fractional Gomory Split Cuts Fractional Basic Feas Chvatal MIG Mixed Integer Basic Feas Mixed Integer Basic Intersection Basic + Strengthening Intersection Basic L&P L&P + Strengthening Intersection Basic Feas + Strengthening Intersection Basic Feas Anureet Saxena, TSoB Simple Disjunctive 5 Taxonomy of Cutting Planes Fractional Basic MIR Fractional Gomory Split Cuts Fractional Basic Feas Chvatal MIG Mixed Integer Basic Feas Mixed Integer Basic Intersection Basic + Strengthening Intersection Basic L&P L&P + Strengthening Intersection Basic Feas + Strengthening Intersection Basic Feas Anureet Saxena, TSoB Simple Disjunctive 6 Taxonomy of Cutting Planes Fractional Basic MIR Fractional Gomory Split Cuts Elementary Chvatal Closure Fractional Basic Feas MIG Elementary closure of P w.r.t a family of cutting planes is defined by Mixed Integer Basic Feas intersecting P with all rank-1 cuts in Intersection . Mixed Integer Basic Eg: CG Intersection Basic Feas + Strengthening Intersection Basic Feas Basic + Strengthening L&P + Strengthening Closure, Split Closure Intersection Basic Anureet Saxena, TSoB L&P Simple Disjunctive 7 Elementary Closures Fractional Basic MIR Fractional Gomory Split Cuts Fractional Basic Feas CG Closure Chvatal Split Closure MIG Mixed Integer Basic Feas Mixed Integer Basic Intersection Basic Feas + Strengthening Intersection Basic Feas Intersection Basic + Strengthening L&P + Strengthening L&P Closure Intersection Basic Anureet Saxena, TSoB L&P Simple Disjunctive 8 Elementary Closures Operations Research Inference Dual max v x2PI )P cx¸v P2 Elementary Closures Rank-1 cuts have short polynomial length proofs Proof Family Constraint Programming Complexity Theory Anureet Saxena, TSoB 9 Elementary Closures How much duality gap can be closed by optimizing over elementary closures? L&P Closure CG Closure Split Closure Bonami and Minoux Fischetti and Lodi ? Anureet Saxena, TSoB 10 Elementary Closures How much duality gap can be closed by optimizing over elementary closures? L&P Closure CG Closure Split Closure Bonami and Minoux Fischetti and Lodi Balas and Saxena Anureet Saxena, TSoB 11 Split Disjunctions • • • 2 ZN, 0 2 Z j = 0, j2 N2 0 < < 0 + 1 x · 0 x ¸ 0 + 1 Split Disjunction Anureet Saxena, TSoB 12 Split Cuts u u0 Ax ¸ b x · 0 Ax ¸ b x ¸ 0+1 L x ¸ L R x ¸ R x¸ Anureet Saxena, TSoB v v0 Split Cut 13 Split Closure Elementary Split Closure of P = { x | Ax ¸ b } is the polyhedral set defined by intersecting P with the valid rank-1 split cuts. C = { x2 P | x ¸ 8 rank-1 split cuts x¸ } Without Recursion Anureet Saxena, TSoB 14 Algorithmic Framework Add Cuts min cx Ax ¸ b t x¸ t t2 Solve Master LP Integral Sol? Unbounded? Infeasible? Yes MIP Solved No Split Cuts Generated Rank-1 Split Cut Separation No Split Cuts Generated Anureet Saxena, TSoB Optimum over Split Closure attained 15 Algorithmic Framework Add Cuts min cx Ax ¸ b t x¸ t t2 Solve Master LP Integral Sol? Unbounded? Infeasible? Yes MIP Solved No Split Cuts Generated Rank-1 Split Cut Separation No Split Cuts Generated Anureet Saxena, TSoB Optimum over Split Closure attained 16 Split Closure Separation Problem Theorem: lies in the split closure of P if and only if the optimal value of the following program is non-negative. Cut Violation • • • • • =1 Disjunctive Cut u.e + v.e + u0 + v0 = 1 u0 + v0 = 1 y=1 ||2=1 Split Disjunction Normalization Set Anureet Saxena, TSoB 17 Split Closure Separation Problem Theorem: lies in the split closure of P if and only if the optimal value of the following program is non-negative. Mixed Integer Non-Convex Quadratic Program u0 + v0 = 1 Anureet Saxena, TSoB 18 SC Separation Theorem Theorem: lies in the split closure of P if and only if the optimal value of the following parametric mixed integer linear program is non-negative. Parameter Parametric Mixed Integer Linear Program Anureet Saxena, TSoB 19 Deparametrization Parameteric Mixed Integer Linear Program Anureet Saxena, TSoB 20 Deparametrization Parameteric Mixed Integer Linear Program If is fixed, then PMILP reduces to a MILP Anureet Saxena, TSoB 21 Deparametrization MILP( ) Deparametrized Mixed Integer Linear Program Maintain a dynamically updated grid of parameters Anureet Saxena, TSoB 22 Separation Algorithm Initialize Parameter Grid ( ) For 2 , Diversification •Solve MILP() using CPLEX 9.0 • Enumerate branch and bound nodes • Store all the separating split disjunctions which are discovered Strengthening At least one Grid Enrichment no split disjunction yes STOP discovered? Bifurcation Anureet Saxena, TSoB 23 Implementation Details Processor Details • Pentium IV • 2Ghz, 2GB RAM COIN-OR CPLEX 9.0 Core Implementation • Solving Master LP • Setting up MILP • Disjunctions/Cuts Management • L&P cut generation+strengthening Anureet Saxena, TSoB Solving MILP( ) 24 Computational Results • MIPLIB 3.0 instances • OR-Lib (Beasley) Capacitated Warehouse Location Problems Anureet Saxena, TSoB 25 MIPLIB 3.0 MIP Instances Instance 10teams dcmulti egout flugpl gen gesa2_o khb05250 misc06 qnet1 qnet1_o rgn vpm1 fixnet6 gesa2 14 Instances # Int Var 1800 75 55 11 150 720 24 112 1417 1417 100 168 378 408 # Cont Var 225 473 86 7 720 504 1326 1696 124 124 80 210 500 816 % Gap Closed 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 99.76% 98.66% 98-100% Gap Closed Anureet Saxena, TSoB 26 MIPLIB 3.0 MIP Instances Instance 10teams dcmulti egout flugpl gen gesa2_o khb05250 misc06 qnet1 qnet1_o rgn vpm1 fixnet6 gesa2 14 Instances # Int Var 1800 75 55 11 150 720 24 112 1417 1417 100 168 378 408 # Cont Var 225 473 86 7 720 504 1326 1696 124 124 80 210 500 816 % Gap Closed 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 99.76% 98.66% 98-100% Gap Closed Anureet Saxena, TSoB 27 MIPLIB 3.0 MIP Instances Instance pp08aCUTS modglob pp08a gesa3 gesa3_o set1ch bell5 arki001 vpm2 mod011 qiu 11 Instances # Int Var 64 98 64 384 672 240 58 538 168 96 48 Unsolved MIP Instance In MIPLIB 3.0 # Cont Var 176 324 176 768 480 472 46 850 210 10862 792 % Gap Closed 97.01% 96.48% 95.81% 95.78% 95.31% 89.41% 87.44% 83.05% 81.22% 80.72% 77.51% 75-98% Gap Closed Anureet Saxena, TSoB 28 MIPLIB 3.0 MIP Instances Instance rout bell3a blend2 3 Instances # Int Var 315 71 264 # Cont Var 241 62 89 % Gap Closed 70.73% 55.19% 46.77% 25-75% Gap Closed Anureet Saxena, TSoB 29 MIPLIB 3.0 MIP Instances Instance danoint dano3mip pk1 3 Instances # Int Var 56 552 55 # Cont Var 465 13321 31 % Gap Closed 7.44% 0.12% 0.00% 0-25% Gap Closed Anureet Saxena, TSoB 30 MIPLIB 3.0 MIP Instances Summary of MIP Instances (MIPLIB 3.0) Total Number of Instances: 34 Number of Instances included: 33 No duality gap: noswot, dsbmip Instance not included: rentacar Results 98-100% Gap closed in 14 instances 75-98% Gap closed in 11 instances 25-75% Gap closed in 3 instances 0-25% Gap closed in 3 instances Average Gap Closed: 82.53% Anureet Saxena, TSoB 31 MIPLIB 3.0 Pure IP Instances Instance air03 gt2 mitre mod008 mod010 nw04 p0548 p0282 fiber 9 Instances # Variables 10757 188 10724 319 2655 87482 548 282 1254 %Gap Closed 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 100.00% 99.90% 98.50% 98-100% Gap Closed Anureet Saxena, TSoB 32 MIPLIB 3.0 Pure IP Instances Instance lseu p2756 l152lav p0033 4 Instances # Variables 89 2756 1989 33 %Gap Closed 93.75% 92.32% 87.56% 87.42% 75-98% Gap Closed Anureet Saxena, TSoB 33 MIPLIB 3.0 Pure IP Instances Instance p0201 air04 air05 seymour misc03 cap6000 6 Instances Ceria, Pataki et al closed around 50% of the gap using 10 rounds of L&P cuts # Variables 201 8904 7195 1372 159 6000 %Gap Closed 74.93% 62.42% 62.05% 61.94% 51.47% 37.63% 25-75% Gap Closed Anureet Saxena, TSoB 34 MIPLIB 3.0 Pure IP Instances Instance misc07 fast0507 stein27 stein45 4 Instances # Variables 259 63009 27 45 %Gap Closed 19.48% 18.08% 0.00% 0.00% 0-25% Gap Closed Anureet Saxena, TSoB 35 MIPLIB 3.0 Pure IP Instances Summary of Pure IP Instances (MIPLIB 3.0) Total Number of Instances: 25 Number of Instances included: 24 No duality gap: enigma Instance not included: harp2 Results 98-100% Gap closed in 9 instances 75-98% Gap closed in 4 instances 25-75% Gap closed in 6 instances 0-25% Gap closed in 4 instances Average Gap Closed: 71.63% Anureet Saxena, TSoB 36 MIPLIB 3.0 Pure IP Instances Instance air03 air04 air05 cap6000 enigma fast0507 fiber gt2 l152lav lseu misc03 misc07 mitre mod008 mod010 nw04 p0033 p0201 p0282 p0548 p2756 seymour stein27 stein45 24 Instances # Variables 10757 8904 7195 6000 100 63009 1254 188 1989 89 159 259 10724 319 2655 87482 33 201 282 548 2756 1372 27 45 %Gap Closed (SC) 100.00% 62.42% 62.05% 37.63% 18.08% 98.50% 100.00% 87.56% 93.75% 51.47% 19.48% 100.00% 100.00% 100.00% 100.00% 87.42% 74.93% 99.90% 100.00% 92.32% 61.94% 0.00% 0.00% 71.63% %Gap Closed (CG) 100.00% 27.60% 15.50% 26.90% 4.70% 98.50% 100.00% 69.20% 91.30% 51.20% 16.10% 100.00% 100.00% 100.00% 100.00% 85.40% 60.50% 99.90% 100.00% 69.20% 23.50% 0.00% 0.00% 62.59% % Gap Closed by First Chvatal Closure (Fischetti-Lodi Bound) Anureet Saxena, TSoB 37 MIPLIB 3.0 Pure IP Instances Instance air03 air04 air05 cap6000 enigma fast0507 fiber gt2 l152lav lseu misc03 misc07 mitre mod008 mod010 nw04 p0033 p0201 p0282 p0548 p2756 seymour stein27 stein45 24 Instances # Variables 10757 8904 7195 6000 100 63009 1254 188 1989 89 159 259 10724 319 2655 87482 33 201 282 548 2756 1372 27 45 %Gap Closed (SC) 100.00% 62.42% 62.05% 37.63% 18.08% 98.50% 100.00% 87.56% 93.75% 51.47% 19.48% 100.00% 100.00% 100.00% 100.00% 87.42% 74.93% 99.90% 100.00% 92.32% 61.94% 0.00% 0.00% 71.63% Anureet Saxena, TSoB %Gap Closed (CG) 100.00% 27.60% 15.50% 26.90% 4.70% 98.50% 100.00% 69.20% 91.30% 51.20% 16.10% 100.00% 100.00% 100.00% 100.00% 85.40% 60.50% 99.90% 100.00% 69.20% 23.50% 0.00% 0.00% 62.59% 38 MIPLIB 3.0 Pure IP Instances Instance air03 air04 air05 cap6000 enigma fast0507 fiber gt2 l152lav lseu misc03 misc07 mitre mod008 mod010 nw04 p0033 p0201 p0282 p0548 p2756 seymour stein27 stein45 24 Instances # Variables 10757 8904 7195 6000 100 63009 1254 188 1989 89 159 259 10724 319 2655 87482 33 201 282 548 2756 1372 27 45 %Gap Closed (SC) 100.00% 62.42% 62.05% 37.63% 18.08% 98.50% 100.00% 87.56% 93.75% 51.47% 19.48% 100.00% 100.00% 100.00% 100.00% 87.42% 74.93% 99.90% 100.00% 92.32% 61.94% 0.00% 0.00% 71.63% Anureet Saxena, TSoB %Gap Closed (CG) 100.00% 27.60% 15.50% 26.90% 4.70% 98.50% 100.00% 69.20% 91.30% 51.20% 16.10% 100.00% 100.00% 100.00% 100.00% 85.40% 60.50% 99.90% 100.00% 69.20% 23.50% 0.00% 0.00% 62.59% 39 MIPLIB 3.0 Pure IP Instances Instance air04 air05 cap6000 fast0507 l152lav lseu misc03 misc07 p0033 p0201 p2756 seymour 12 Instances # Variables 8904 7195 6000 63009 1989 89 159 259 33 201 2756 1372 %Gap Closed (SC) 62.42% 62.05% 37.63% 18.08% 87.56% 93.75% 51.47% 19.48% 87.42% 74.93% 92.32% 61.94% 62.42% %Gap Closed (CG) 27.60% 15.50% 26.90% 4.70% 69.20% 91.30% 51.20% 16.10% 85.40% 60.50% 69.20% 23.50% 45.09% Comparison of Split Closure vs CG Closure Total Number of Instances: 24 CG closure closes >98% Gap: 9 Ratio 2.261 4.003 1.347 3.847 1.265 1.027 1.005 1.210 1.024 1.239 1.334 2.636 1.850 Results (Remaining 15 Instances) Split Closure closes significantly more gap in 9 instances Both Closures close almost same gap in 6 instances Anureet Saxena, TSoB 40 OrLib CWLP • Set 1 – 37 Real-World Instances – 50 Customers, 16-25-50 Warehouses • Set 2 – 12 Real-World Instances – 1000 Customers, 100 Warehouses Anureet Saxena, TSoB 41 OrLib CWLP Set 1 Summary CWLP Instances (Set 1) Instance of OrLib % Gap Closed Instance % Gap Closed 100.000% 100.000% cap41 cap93 Number of Instances: 37 100.000% 100.000% cap42 cap94 100.000% 100.000% cap43 cap101 Number of Instances included: 37 100.000% 100.000% cap44 cap102 cap51 cap61 cap62 cap63 cap64 cap71 cap72 cap73 cap74 cap81 cap82 cap83 cap84 cap91 cap92 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% cap103 cap104 cap111 cap112 cap113 cap114 cap121 cap122 cap123 cap124 cap131 cap132 cap133 cap134 Results 100% Gap closed in 37 instances Anureet Saxena, TSoB 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 100.000% 42 OrLib CWLP Set 2 Instance % Gap Closed Summary of capa_8000 OrLib CWFL Instances (Set 2) 87.00% 87.57% Number of Instances:capa_10000 12 91.53% Number of Instancescapa_12000 included: 12 capa_14000 capb_5000 Results capb_6000 capb_7000 >90% Gap closed in 10 instances 85-90% Gap closed incapb_8000 2 instances capc_5000 Average Gap Closed:capc_5750 92.82% capc_6500 capc_7250 97.53% 94.18% 92.80% 92.42% 93.74% 93.42% 93.90% 94.69% 95.06% Anureet Saxena, TSoB 43 Algorithmic Framework Add Cuts min cx Ax ¸ b t x¸ t t2 Solve Master LP Integral Sol? Unbounded? Infeasible? Yes MIP Solved No Split Cuts Generated Rank-1 Split Cut Separation No Split Cuts Generated Anureet Saxena, TSoB Optimum over Split Closure attained 44 Algorithmic Framework Add Cuts min cx Ax ¸ b t x¸ t t2 Solve Master LP What are the characteristics of the cuts which are binding at the final optimal solution? Integral Sol? Yes MIP Solved Unbounded? Infeasible? What can one say about the split disjunctions which were used to generate cuts? No Split Cuts Generated Rank-1 Split Cut Separation No Split Cuts Generated Anureet Saxena, TSoB Optimum over Split Closure attained 45 Support Size & Sparsity The support of a split disjunction D(, 0) is the set of non-zero components of x · 0 x ¸ 0 + 1 (2x1 + 3x3 – x5 · 1) Ç (2x1 + 3x3 – x5 ¸ 2) Support Size = 3 Anureet Saxena, TSoB 46 Support Size & Sparsity The support of a split disjunction D(, 0) is the set of non-zero components of • Computationally Faster • Avoid fill-in Sparse Split Disjunctions Disjunctive argument Non-negative row combinations Sparse Split Cuts Anureet Saxena, TSoB Basis Factorization Sparse Matrix Op 47 Support Size & Sparsity air05 [ MIPLIB 3.0 Pure IP instance, 7195 Int Var, 62.05% Gap closed by Split Closure ] Mean Support Size Standard Deviation 30 25 15 10 5 157 153 149 145 141 137 133 129 125 121 117 113 109 105 101 97 93 89 85 81 77 73 69 65 61 57 53 49 45 41 37 33 29 25 21 17 9 13 5 0 1 Support Size 20 Iteration Anureet Saxena, TSoB 48 Support Size & Sparsity arki001** [ MIPLIB 3.0 Mixed IP instance, 538 Int Var, 83.35% Gap closed by Split Closure ] Mean Support Size Standard Deviation 10 9 8 6 5 4 3 2 1 91 88 85 82 79 76 73 70 67 64 61 58 55 52 49 46 43 40 37 34 31 28 25 22 19 16 13 10 7 4 0 1 Support Size 7 Iteration Anureet Saxena, TSoB 49 Support Size & Sparsity Pure IP Instance nw04 air05 seymour misc03 p0033 # Int Variables 87482 7195 1372 159 33 Mean Support Size 2.084 8.210 5.263 3.771 4.847 Empirical Observation Substantial Duality gap can be closed Mixed IP Instance # Int Variables Mean Support Size by using split cuts generated from 6.690 qnet1_o 1417 gesa2_o 720 4.937 sparse split disjunctions arki001 538 3.146 vpm1 168 4.503 pp08aCUTS 64 3.850 Anureet Saxena, TSoB 50 Support Coefficients Practice Theory • Determinants of sub-matrices • Andersen, Cornuejols & Li (’05) • Cook, Kannan & Scrhijver (’90) • Elementary 0/1 disjunctions • Mixed Integer Gomory Cuts • Lift-and-project cuts Huge Gap det (B) 1 Anureet Saxena, TSoB 51 Support Coefficients air05 [ MIPLIB 3.0 Pure IP instance, 7195 Int Var, 62.05% Gap closed by Split Closure ] Mean Coef Size Standard Deviation 8 7 5 4 3 2 1 157 153 149 145 141 137 133 129 125 121 117 113 109 105 101 97 93 89 85 81 77 73 69 65 61 57 53 49 45 41 37 33 29 25 21 17 9 13 5 0 1 Coefficient Size 6 Iteration Anureet Saxena, TSoB 52 Support Coefficients arki001** [ MIPLIB 3.0 Mixed IP instance, 538 Int Var, 83.35% Gap closed by Split Closure ] Mean Coef Size Standard Deviation 40 35 Coefficient Size 30 25 20 15 10 5 0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 Iteration Anureet Saxena, TSoB 53 Support Coefficients Pure IP Instance nw04 air05 seymour misc03 p0033 #Int Variables 87482 7195 1372 159 33 Mean Coef Size 1.228 1.156 1.099 1.227 2.099 Empirical Observation Substantial Duality gap can be closed Mixed by IP Instance Variables Mean Coef Size using split#Int cuts generated from qnet1_o 1417 2.381 split disjunctions containing gesa2_o 720 1.767 small support coefficients. arki001 538 3.044 vpm1 pp08aCUTS 168 64 Anureet Saxena, TSoB 1.833 1.418 54 Cuts Statistics Instance Name 10teams air03 air04 air05 arki001 bell3a bell5 blend2 cap6000 dano3mip dsbmip egout enigma fiber fixnet6 flugpl gen gesa2 gesa2_o gesa3 khb05250 l152lav lseu misc03 n 2025 10757 7564 7195 959 100 104 334 5911 13873 2320 141 100 1298 878 35 870 1224 1224 1152 1350 1989 89 160 m 230 124 614 426 782 86 91 184 2095 3202 1182 98 21 363 478 24 780 1392 1248 1368 101 97 28 96 ncuts 37 3 185 192 227 19 26 3 22 123 32 57 4 113 321 6 32 116 140 129 72 80 19 71 Avg Cut Density 37.75% 26.74% 24.81% 31.48% 15.96% 5.79% 10.80% 26.35% 99.90% 50.92% 4.16% 2.75% 55.50% 6.92% 2.75% 20.95% 1.09% 1.10% 1.73% 1.62% 5.98% 67.62% 31.87% 31.63% Instance Name misc06 mitre mod008 mod010 mod011 modglob noswot nw04 p0033 p0201 p0282 p2756 pk1 pp08a pp08aCUTS qiu qnet1 qnet1_o rgn rout set1ch seymour stein27 stein45 vpm1 vpm2 Anureet Saxena, TSoB n 1808 10724 319 2655 6872 384 128 87482 33 201 282 2756 86 240 240 840 1541 1541 180 556 712 1372 27 45 378 378 m 820 2054 6 146 1535 286 182 36 16 133 241 755 45 136 246 1192 503 456 24 291 492 4944 118 331 234 234 ncuts 31 523 16 9 674 105 33 136 16 36 69 298 20 127 113 220 113 111 39 124 223 413 20 34 47 111 Avg Cut Density 1.53% 0.50% 76.25% 35.76% 5.29% 4.93% 15.63% 63.46% 26.52% 30.21% 5.15% 1.72% 61.05% 4.21% 4.64% 13.00% 5.23% 2.41% 20.11% 19.22% 1.06% 1.88% 27.59% 43.14% 2.02% 2.16% 55 isc u Anureet Saxena, TSoB vp m 1 rg n se t1 ch st ei n2 7 qn et 1 pk pp 1 08 aC UT S m 06 od 00 8 m od 01 1 no sw ot p0 03 3 p0 28 2 m lse ge n ge sa 2_ o kh b0 52 50 ncuts 600 be ll5 ca p6 00 0 ds bm ip en ig m a f ix ne t6 4 ar ki0 01 ai r0 10 te am s Number of Cuts #Cuts 800 700 Average: 113.80 500 400 300 200 100 0 instances 56 #Cuts/m vs log(n) #Cuts/m 400.00% 350.00% 300.00% ncuts/m 250.00% Average: 45.76% 200.00% 150.00% 100.00% 50.00% 0.00% 0.000 1.000 2.000 3.000 4.000 5.000 6.000 log(n) Anureet Saxena, TSoB 57 Average Cut Density vs log(n) Avg Cut Density 120.00% Avg Cut density 100.00% 80.00% Average: 20.82% 60.00% 40.00% 20.00% 0.00% 0.000 1.000 2.000 3.000 4.000 5.000 6.000 log(n) Anureet Saxena, TSoB 58 Cuts Statistics Internet Checkable Proofs Strengthened formulations for MIPLIB 3.0 instances available at www.andrew.cmu.edu/user/anureets/osc/osc.htm Google Query: anureet saxena (I’m feeling lucky) Anureet Saxena, TSoB 59 arki001 • MIPLIB 3.0 & 2003 instance • Metallurgical Industry Problem Stats • Unsolved for the past 10 years [1996-2000-2005] 1048 Rows 1388 Columns 123 Gen Integer Vars 415 Binary Vars 850 Continuous Vars Anureet Saxena, TSoB 60 Solution Strategy Original Problem CPLEX 9.0 Presolver Preprocessed Problem CPLEX 9.0 Emphasis on optimality Strong Branching Rank-1 Split Cut Generation Anureet Saxena, TSoB Strengthened Formulation 61 Strengthening + CPLEX 9.0 Solved to optimality Crossover Point (227 rank-1 cuts) Anureet Saxena, TSoB 62 Strengthening + CPLEX 9.0 Solved to optimality arki001 Solution Statistics % Gap closed by rank-1 split cuts: 83.05% Time spent in generating rank-1 split cuts: 53.76 hrs Time taken by CPLEX 9.0 after strengthening: 10.94 hrs No. of branch-and-bound nodes enumerated by CPLEX: 643425 Total time taken to solve the instance to optimality: 64.70 hrs Anureet Saxena, TSoB 63 CPLEX 9.0 After 100 hours: 43 million B&B nodes 22 million active nodes 12GB B&B Tree Anureet Saxena, TSoB 64 Comparison Crossover Point Anureet Saxena, TSoB 65 Thank You Anureet Saxena, TSoB 66