Lessons from MIP Search. John Hooker Carnegie Mellon University November 2009
|
|
- Harriet Scott
- 5 years ago
- Views:
Transcription
1 Lessons from MIP Search John Hooker Carnegie Mellon University November 2009
2 Outline MIP search The main ideas Duality and nogoods From MIP to AI (and back) Binary decision diagrams From MIP to constraint programming Banff workshop. Slide 2
3 Background reading MIP search General Survey J. T. Linderoth and M. W. P. Savelsbergh, A computational study of search strategies for mied integer programming, INFORMS Journal on Computing 11 (1999) Survey of Cutting Planes H. Marchand, A. Martin, R. Weismantel, and L. Wolsey. Cutting planes in integer and mied integer programming. Discrete Applied Mathematics, 123 (2002) Some Recent Developments M. Fischetti, F. Glover, A. Lodi, The feasibility pump, Mathematical Programming 104 (2005) M. Fischetti and A. Lodi, Repairing MIP infeasibility through local branching, Computers and Operations Research 35 (2008) , T. Sandholm, R. Shields, Nogood learning for mied integer programming, CMU, November 2006 Banff workshop. Slide (#)
4 Background reading BDDs Basic Ideas H. R. Andersen, T. Hadzic, J. N. Hooker, and P. Tiedemann, A constraint store based on multivalued decision diagrams, in C. Bessiere, ed., CP 2007, Refinements Tarik Hadzic and J. N. Hooker, Cost-bounded binary decision diagrams for 0-1 programming, in E. Loute and L. Wolsey, eds., CPAIOR 2007, T. Hadzic, J. N. Hooker, and P. Tiedemann, Approimate compilation of constraints into multivalued decision diagrams, in P. J. Stuckey, ed., CP 2007, Banff workshop. Slide (#)
5 MIP search MIP and the LP relaation Branch and bound Variable selection Node selection Feasibility heuristics Banff workshop. Slide 5
6 Mied integer programming We focus on mied integer/linear programming (MILP) min c A b, 0 Z, j J j integers Banff workshop. Slide 6
7 LP relaation An advantage of MIP: ready-made global relaation Linear programming (LP) problem min A j c b, 0 R, all j Drop integrality condition Banff workshop. Slide 7
8 LP relaation An advantage of MIP: ready-made global relaation Linear programming (LP) problem min c A b, 0 j R, all j Cutting planes can strengthen the relaation Drop integrality condition Banff workshop. Slide 8
9 LP relaation An advantage of MIP: ready-made global relaation Linear programming (LP) problem min A j c b, 0 R, all j Feasible set of LP relaation Banff workshop. Slide 9
10 LP relaation An advantage of MIP: ready-made global relaation Linear programming (LP) problem min A j c b, 0 R, all j Feasible set of MIP Banff workshop. Slide 10
11 LP relaation An advantage of MIP: ready-made global relaation Linear programming (LP) problem min c A b, 0 j R, all j Simple method finds a verte solution of the LP relaation Banff workshop. Slide 11
12 Branch and bound Used for 50 years Land and Doig Creates search tree by branching on variables Solves continuous relaation (LP) at each node Basic search framework for all general-purpose MIP codes Branch and cut adds cutting planes at some nodes. Banff workshop. Slide 12
13 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Initially, there is one open node (the original problem)
14 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Initially, there is one open node (the original problem) Optimal value of LP relaation is lower bound on optimal value of original problem
15 Branch and bound - illustration Because solution is fractional, branch on a fractional variable Root node = (2, 3.5, 4.6) LB = Remember: we are minimizing
16 Branch and bound - illustration Because solution is fractional, branch on a fractional variable Root node = (2, 3.5, 4.6) LB = Close after branching Two new open nodes
17 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Integer solution = (2, 3, 4) UB = 170 Select open node and solve LP
18 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Integer solution = (2, 3, 4) UB = 170 Select open node and solve LP Close node, because solution is integer
19 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Integer solution = (2, 3, 4) UB = 170 Select open node Close node, and solve LP because solution is integer Objective function value is upper bound on optimal value
20 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Integer solution = (2, 3, 4) UB = 170 = (2.3, 4, 4.7) LB = 165 Incumbent solution Select open node and solve LP
21 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Incumbent solution Integer solution = (2, 3, 4) UB = 170 = (2.3, 4, 4.7) LB = 165 Must branch again
22 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Incumbent solution Integer solution = (2, 3, 4) UB = 170 = (2.3, 4, 4.7) LB = Select open node = (2, 4.1, 4.7) LB = 171
23 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Incumbent solution Integer solution = (2, 3, 4) UB = 170 = (2.3, 4, 4.7) LB = 165 Close node, because relaation value is no better than incumbent = (2, 4.1, 4.7) LB = 171
24 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Incumbent solution Integer solution = (2, 3, 4) UB = 170 = (2.3, 4, 4.7) LB = 165 Select open node = (2, 4.1, 4.7) LB = 171 Integer solution = (3, 4, 5) LB = UB = 169
25 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Incumbent solution Integer solution = (2, 3, 4) UB = 170 = (2.3, 4, 4.7) LB = Integer solution is better than incumbent, becomes new incumbent = (2, 4.1, 4.7) LB = 171 Integer solution = (3, 4, 5) LB = UB = 169
26 Branch and bound - illustration Root node = (2, 3.5, 4.6) LB = 160 Remember: we are minimizing Integer solution = (2, 3, 4) UB = 170 = (2.3, 4, 4.7) LB = 165 No remaining open nodes. Incumbent is optimal = (2, 4.1, 4.7) LB = 171 Integer solution = (3, 4, 5) LB = UB = 169
27 Branch and bound procedure Let original problem be an open node Let UB = While open nodes remain Select an open node N and solve its continuous relaation R Let LB = value of R and close N If solution of R is integer then If LB < UB then let UB = LB Else if LB < UB then Branch on fractional variable and create 2 new open nodes Banff workshop. Slide 27
28 Why it works Luck Most variables tend to be integer in solution of relaation. Banff workshop. Slide 28
29 Why it works Luck Most variables tend to be integer in solution of relaation. Global continuous relaation Cutting planes, good formulations Banff workshop. Slide 29
30 Why it works Luck Most variables tend to be integer in solution of relaation. Global continuous relaation Cutting planes, good formulations Presolve Bag of tricks to simplify problem Banff workshop. Slide 30
31 Why it works Luck Most variables tend to be integer in solution of relaation. Global continuous relaation Cutting planes, good formulations Presolve Bag of tricks to simplify problem Feasibility heuristics Find feasible solutions before branching Banff workshop. Slide 31
32 Why it works Luck Most variables tend to be integer in solution of relaation. Global continuous relaation Cutting planes, good formulations Presolve Bag of tricks to simplify problem Feasibility heuristics Find feasible solutions before branching Good engineering 50 years of eperience Banff workshop. Slide 32
33 Basic issues Variable selection Branch on which variable at current node? Node selection Process which node net? Feasibility heuristics How to find integer solutions early? to get good upper bounds Banff workshop. Slide 33
34 Variable selection Goal is to tighten lower bound after branching We don t know how to increase probability of finding integer solutions. Pseudocosts Estimate change in objective function after branching Strong branching Put bound on change in objective function after branching Banff workshop. Slide 34
35 Pseudocosts Pseudocost = rate of increase of LP lower bound when fractional variable j is rounded up or down Current LP bound j = LB j f LB j + LB + j j 1 f j = LB Fractional part of j Banff workshop. Slide 35
36 Pseudocosts Pseudocost = rate of increase of LP lower bound when fractional variable j is rounded up or down Bound after branching down Current LP bound Bound after branching up j = LB j f LB j + LB + j j 1 f j = LB Fractional part of j Banff workshop. Slide 36
37 Pseudocosts Using pseudocosts One method: branch on variable j with largest + j f j + j f j (1 ) j = LB j f LB j + LB + j j 1 f j = LB Banff workshop. Slide 37
38 Pseudocosts Computing pseudocosts at root node + One method: Eplicitly compute LB for fractional j, LBj variables by solving 2 LPs. Perhaps with limited number of simple iterations. j = LB j f LB j + + j j 1 f j = LB LB Banff workshop. Slide 38
39 Pseudocosts Computing pseudocosts lower in the tree For variables not yet branched on or evaluated: + Eplicitly compute LBj, LBj For variables previously branched on:, + Use average of values obtained after previous j j branches. j = LB j f LB j + LB + j j 1 f j = LB Banff workshop. Slide 39
40 Strong branching We need the concept of LP duality Will eplain origin of dual later ma ua u i ub c, u 0 R, all i min c A j b, 0 R, all j Dual problem Primal problem Banff workshop. Slide 40
41 LP dual ma ua u i Dual ub c, u 0 R, all i min c A j Primal b, 0 R, all j Objective function value Dual feasible solutions Primal feasible solutions Weak duality Banff workshop. Slide 41
42 LP dual ma ua u i Dual ub c, u 0 R, all i min c A j Primal b, 0 R, all j Objective function value When primal and dual feasible solutions have the same value, both are optimal Banff workshop. Slide 42 Strong duality
43 LP dual ma ua u i Dual ub c, u 0 R, all i min c A j Primal b, 0 R, all j Objective function value Simple method maintains primal feasibility and strives for dual feasibility Banff workshop. Slide 43
44 LP dual ma ua u i Dual ub c, u 0 R, all i min c A j Primal b, 0 R, all j Objective function value Dual simple method maintains dual feasibility and strives for primal feasibility Simple method maintains primal feasibility and strives for dual feasibility Banff workshop. Slide 44
45 Dual simple Dual simple is used to reoptimize after adding a constraint such as new bound on variable after branching. Solution remains dual feasible after adding constraint. Because this adds a variable in the dual. Dual simple restores primal feasibility. Premature termination of dual simple provides lower bound on optimal value. Due to weak duality. Banff workshop. Slide 45
46 Strong branching For fractional variables j : Compute lower bounds j j on increase in LP value after branching down or up. Branch on j with largest L, L + j j Obtain by running a few iterations of dual simple. Perhaps only one. L, L + L + L + j j To save time, evaluate only a few fractional variables. Banff workshop. Slide 46
47 Node selection Select which of the open nodes to process. Goal is to find good feasible solution, or prove that there is no solution better than incumbent. In either case, it is reasonable to look at quality of LP bounds. Two common methods: Depth first Best bound Banff workshop. Slide 47
48 Depth first Advantages: Easy to implement. Fast reoptimization of LP. Minimal memory requirements. Disadvantages: Blind to what is going on. Must search entire subtree before moving to a more promising subtree. Tends to eplode in hard problems. Banff workshop. Slide 48
49 Best bound Select open node with best LP value. For a given branching rule, this minimizes number of nodes processed. Banff workshop. Slide 49
50 Best bound LB = 10 Complete search tree for a given variable selection strategy LB = 13 LB = 11 LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Banff workshop. Slide 50
51 Best bound LB = 10 Complete search tree for a given variable selection strategy LB = 13 LB = 11 LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Banff workshop. Slide 51 Leaf nodes represent integer solutions
52 Best bound Black = open node LB = 10 Select open node with best bound. LB = 13 LB = 11 LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Banff workshop. Slide 52 Leaf nodes represent integer solutions
53 Best bound LB = 10 Select node with best bound. We are in the right subtree. LB = 13 LB = 11 LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Optimal solution Banff workshop. Slide 53
54 Best bound LB = 10 Select node with best bound. We are still in the right subtree. LB = 13 LB = 11 LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Optimal solution Banff workshop. Slide 54
55 Best bound LB = 10 Wrong subtree, but we would have to select this node in any method LB = 13 LB = 11 LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Optimal solution Banff workshop. Slide 55
56 Best bound LB = 10 Net choice is optimal solution. LB = 13 LB = 11 LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Optimal solution Banff workshop. Slide 56
57 Best bound LB = 10 Net choice is optimal solution. LB = 13 LB = 11 No need to consider other open nodes. LB = 24 LB = 15 LB = 22 LB = 12 LB = 20 LB = 16 LB = 18 LB = 14 Optimal solution Banff workshop. Slide 57
58 Feasibility heuristics We wish find good integer solutions as early as possible. To get useful upper bounds. Two approaches (among others): Local branching. Feasibility pump. Banff workshop. Slide 58
59 Local branching Must begin with known integer solution Two level branching procedure Strategic branches define neighborhoods of known integer solutions. Low-level branching searches each neighborhood using MIP Banff workshop. Slide 59
60 Local branching Assume integer variables are binary. + (1 ) k j j: v = 0 j: v = 1 j j j Current node + (1 ) > k j j: v = 0 j: v = 1 j j j Solutions within k-neighborhood of known integer solution v Strategic branch Other solutions Banff workshop. Slide 60
61 Local branching Assume integer variables are binary. + (1 ) k j j: v = 0 j: v = 1 j j j Current node + (1 ) > k j j: v = 0 j: v = 1 j j j Solutions within k-neighborhood of known integer solution v Strategic branch Other solutions Solve this MIP to optimality. Perhaps obtain better integer solutions. Banff workshop. Slide 61
62 Local branching Assume integer variables are binary. + (1 ) k j j: v = 0 j: v = 1 j j j Current node + (1 ) > k j j: v = 0 j: v = 1 j j j Solutions within k-neighborhood of known integer solution v Strategic branch Other solutions Solve this MIP to optimality. Perhaps obtain better integer solutions. Repeat process with another integer solution v Banff workshop. Slide 62
63 Feasibility pump Can find initial integer solution Alternate between integrality and linear feasibility Solve LP Round to integer Find closest LP solution Banff workshop. Slide 63
64 Feasibility pump Solve LP relaation Banff workshop. Slide 64
65 Feasibility pump Round to nearest integer Banff workshop. Slide 65
66 Feasibility pump Round to nearest integer Banff workshop. Slide 66
67 Feasibility pump Find closest LP-feasible point using L 1 norm Banff workshop. Slide 67
68 Feasibility pump Find closest LP-feasible point using L 1 norm Banff workshop. Slide 68 This problem is itself an LP
69 Feasibility pump Round to nearest integer Banff workshop. Slide 69
70 Feasibility pump Round to nearest integer Procedure terminates Banff workshop. Slide 70
71 Feasibility pump Round to nearest integer Use random perturbations to break out of loops Procedure terminates Banff workshop. Slide 71
72 Duality and Nogoods Inference duality Nogood-based search Eample: SAT Eample: MIP Banff workshop. Slide 72
73 Inference duality Duality in optimization is closely related to nogood-based search in AI. Nogood = set of solutions to avoid in the rest of the search. All optimization duals are inference duals LP, Lagrangean, surrogate, subadditive, etc. Solution of inference dual is proof of optimality Dual solution provides nogoods Including Benders cuts, conflict clauses for SAT Nogood = conditions under which proof is still valid Slide 73
74 Inference duality Primal min f ( ) S Feasible set ma v P S v f ( ) P Dual P Family of proofs (inference method) Follows using proof P Dual is defined relative to an inference method. Strong duality applies if inference method is complete. Slide 74
75 Eample: LP dual Primal min c A b 0 ma Dual v A b P c v 0 P P = ma λb λa c λ 0 Nonnegative linear combination + domination 0 λa λb dominates c v A b c v iff for some λ 0 λa c and λb v Slide 75 Inference method is complete (assuming feasibility) due to Farkas Lemma. So we have strong duality (assuming feasibility).
76 Duals for MIP Lagrangean dual Inference = same as LP Inference method incomplete duality gap Surrogate dual Inference = same as LP, ecept domination is weaker Therefore smaller duality gap than Lagrangean dual Subadditive dual Inference = subadditive, homogeneous function + domination Inference method complete (cf. Gomory s method) no gap Slide 76
77 Nogood-based search All search is nogood-based search Each solution eamined generates a nogood. Net solution must satisfy current nogood set. Nogoods are derived by solving inference dual of the subproblem. Subproblem is normally defined by fiing variables to current values in the search. Slide 77
78 Eample: Logic-based Benders Partition variables,y and search over values of Subproblem results from fiing min f (, y ) min f (, y ) (, y ) S (, y ) S Let proof P be solution of subproblem dual for = Let B(P,) be lower bound obtained by P for given. Add Benders cut v B(P,) to master problem: Solve master problem for net min v Benders cuts Slide 78
79 Eample: Logic-based Benders Master and subproblem may be solved by different techniques MIP for master problem CP for subproblem Benders cuts are problem-specific Based on type of inference in dual Some applications Planning & scheduling (e.g., assembly lines) Computer processor allocation and scheduling Chemical batch sizing and scheduling Sports scheduling Location/allocation Slide 79
80 Eample: Classical Benders Partition variables,y and search over values of min f ( ) + cy g( ) + Ay b D, 0 y Subproblem results from fiing k min f ( ) + cy k Ay b g( ) y 0 ( λ) Let proof λ be solution of subproblem dual for Let B(λ,) = f() + λ(b g()) be bound obtained by λ for given. Add Benders cut v B(λ,) to master problem: Solve master problem for net = min v Benders cuts Slide 80
81 Eample: SAT Solve SAT by chronological backtracking + unit clause rule = DPL. Chronological = fied branching order. To get nogood, solve subproblem at current node. Solve with unit clause rule Nogood identifies branches that create infeasibility. Simplest scheme: nogood rules out path to current leaf node. Process nogood set with parallel resolution Nogood set is a relaation of the problem. Solve relaation without branching Select solution with preference for 0 Slide 81
82 DPL with chronological backtracking 1 = 0 2 = 0 3 = 0 5 = 0 4 = 0 Branch to here. Solve subproblem with unit clause, which proves infeasibility. ( 1, 5 ) = (0,, 0) creates the infeasibility. Banff workshop. Slide 82
83 DPL with chronological backtracking 1 = 0 2 = 0 3 = 0 5 = 0 4 = Branch to here. Solve subproblem with unit clause, which proves infeasibility. ( 1, 5 ) = (0,, 0) creates the infeasibility. Generate nogood. Banff workshop. Slide 83
84 DPL with chronological backtracking 1 = 0 Consists of processed nogoods 3 = 0 2 = 0 k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) k k = 0 5 = 0 Conflict clause appears as nogood induced by solution of R k Banff workshop. Slide 84
85 DPL with chronological backtracking 1 = 0 Consists of processed nogoods 3 = 0 2 = 0 k Relaation R Solution of R Nogoods k 0 (0,0,0,0,0, ) (0,0,0,0,1, ) k 5 = 0 4 = 0 5 = 1 Go to solution that solves relaation, with priority to 0 Banff workshop. Slide 85
86 DPL with chronological backtracking 1 = 0 Consists of processed nogoods 3 = 0 2 = 0 k Relaation R Solution of R Nogoods k 0 (0,0,0,0,0, ) (0,0,0,0,1, ) k 5 = 0 4 = 0 5 = Process nogood set with parallel resolution parallel-absorbs Banff workshop. Slide 86
87 DPL with chronological backtracking 1 = 0 Consists of processed nogoods 5 = 0 4 = 0 3 = 0 2 = 0 5 = 1 k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (0,0,0,0,1, ) k k Process nogood set with parallel resolution parallel-absorbs Banff workshop. Slide 87
88 DPL with chronological backtracking 1 = 0 Consists of processed nogoods 5 = 0 4 = 0 3 = 0 2 = 0 k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (0,0,0,0,1, ) 2 (0,0,0,1,0, ) k Solve relaation again, continue. So backtracking is nogood-based search with parallel resolution k Banff workshop. Slide 88
89 Eample: SAT and conflict clauses Nogoods = conflict clauses. Nogoods rule out only branches that play a role in unit clause refutation. Slide 89
90 DPL with conflict clauses 1 = 0 2 = 0 3 = 0 4 = 0 = 0 5 Branch to here. Unit clause rule proves infeasibility. ( 1, 5 ) = (0,0) is only premise of unit clause proof. Banff workshop. Slide 90
91 DPL with conflict clauses 1 = 0 3 = 0 2 = 0 k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) k k = 0 5 = 0 Conflict clause appears as nogood induced by solution of R k. 1 5 Banff workshop. Slide 91
92 DPL with conflict clauses 1 = 0 Consists of processed nogoods 3 = 0 2 = 0 k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) k 1 (0,0,0,0,1, ) k 4 = 0 5 = 0 5 = Banff workshop. Slide 92
93 DPL with conflict clauses 1 = 0 Consists of processed nogoods 4 = 0 3 = 0 2 = 0 5 = 0 5 = k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (0,0,0,0,1, ) k parallel-resolve to yield 1 2 parallel-absorbs k Banff workshop. Slide 93
94 DPL with conflict clauses 1 = 0 2 = 0 2 = 1 4 = 0 3 = 0 5 = 0 = Banff workshop. Slide k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (0,0,0,0,1, ) 2 (0,1,,,, ) k parallel-resolve to yield k
95 DPL with conflict clauses 1 = 0 1 = 1 4 = 0 3 = 0 2 = 0 5 = 0 5 = Banff workshop. Slide k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (0,0,0,0,1, ) 2 (0,1,,,, ) 3 (1,,,,, ) 4 k Search terminates k
96 Eample: SAT and partial-order dynamic backtracking Solve relaation by selecting a solution that conforms to nogoods. Conform = takes opposite sign than in nogoods. More freedom than in branching. Slide 96
97 Partial Order Dynamic Backtracking k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) k k 5 1 Arbitrarily choose one variable to be last Banff workshop. Slide 97
98 Partial Order Dynamic Backtracking k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) k k 5 1 Other variables are penultimate Arbitrarily choose one variable to be last Banff workshop. Slide 98
99 Partial Order Dynamic Backtracking k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (1,,,,0, ) k k 5 1 Since 5 is penultimate in at least one nogood, it must conform to nogoods. It must take value opposite its sign in the nogoods. 5 will have the same sign in all nogoods where it is penultimate. Banff workshop. Slide 99 This allows more freedom than chronological backtracking.
100 Partial Order Dynamic Backtracking k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (1,,,,0, ) 2 k k Choice of last variable is arbitrary but must be consistent with partial order implied by previous choices. Since 5 is penultimate in at least one nogood, it must conform to nogoods. It must take value opposite its sign in the nogoods. 5 will have the same sign in all nogoods where it is penultimate. Banff workshop. Slide 100 This allows more freedom than chronological backtracking.
101 Partial Order Dynamic Backtracking k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (1,,,,0, ) k k Choice of last variable is arbitrary but must be consistent with partial order implied by previous choices Parallel-resolve to yield 5 Banff workshop. Slide 101 Since 5 is penultimate in at least one nogood, it must conform to nogoods. It must take value opposite its sign in the nogoods. 5 will have the same sign in all nogoods where it is penultimate. This allows more freedom than chronological backtracking.
102 Partial Order Dynamic Backtracking k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (1,,,,0, ) 2 (,0,,,1, ) k k 5 does not parallel-resolve with because 5 is not last in both clauses 5 2 Banff workshop. Slide 102
103 Partial Order Dynamic Backtracking k Relaation R Solution of R Nogoods 0 (0,0,0,0,0, ) 1 (1,,,,0, ) 2 (,0,,,1, ) (,1,,,1, ) 4 k k 2 Search terminates Must conform Banff workshop. Slide 103
104 Eample: MIP Nogoods now being used in MIP For eample, in SIMPL and SCIP. So we come full circle MIP AI MIP. Slide 104
105 Binary Decision Diagrams Motivating idea Among constraint BDDs Propagation through a BDD Computational results Banff workshop. Slide 105
106 Motivating idea Domain store is key element of constraint programming (CP) Consists of current variable domains. Domain store propagates results of filtering domains. Constraint-specific filtering algorithms remove infeasible values from domains. Reduced domains are handed over to net constraint. Slide 106
107 Motivating idea Problem: Domain store is a weak relaation It encodes limited information. Not worth investing time to process nodes of search tree. LP is a strong relaation for MIP It pays to invest time at nodes (solve LP, add cutting planes) Needed: A strong, discrete relaation for CP Adaptable to any constraint Domain store should be a degenerate case Slide 107
108 Among constraint We illustrate the idea using the among constraint. Important in sequencing and scheduling. ( X Sl u) among,,, Set of variables Nonnegative integers Set of values Slide 108
109 Among constraint We illustrate the idea using the among constraint. Important in sequencing and scheduling. ( X Sl u) among,,, Set of variables Nonnegative integers Set of values The constraint requires that at least l and at most u variables in X take a value in S Slide 109
110 Binary decision diagrams A BDD is a graphical representation of a boolean function It can also represent constraint on binary variables. The boolean function is 1 when the constraint is satisfied. A (reduced) BDD is basically a compact representation of a branching tree. Superimpose isomorphic subtrees. Remove unnecessary nodes. There is a unique reduced BDD for a given branching order. Generalizes to multivalued decision diagrams (MDDs) for arbitrary discrete variables. Slide 110
111 Search tree for ( ) among {,, },{1},2, Infeasible Feasible
112 ( ) among {,, },{1},2, Banff workshop. Slide 112 Remove infeasible paths
113 ( ) among {,, },{1},2, Banff workshop. Slide 113 Remove infeasible paths
114 ( ) among {,, },{1},2, Banff workshop. Slide 114 Merge isomorphic subtrees
115 ( ) among {,, },{1},2, {0,1} {0,1} Banff workshop. Slide 115 Merge isomorphic subtrees
116 ( ) among {,, },{1},2, {0,1} {0,1} Banff workshop. Slide 116 Can also remove redundant nodes
117 ( ) among {,, },{1},2, Banff workshop. Slide 117 Can also remove redundant nodes
118 ( ) among {,, },{1},2, {0,1} {0,1} Banff workshop. Slide 118 But this is unnecessary.
119 ( ) among {,, },{1},2, {0} {1} 2 {0,1} {0,1} 3 {0} {1} {0} 4 {1} 1 {0} Each path corresponds to a cartesian product of solutions. {0} {0,1} {0} {1} Banff workshop. Slide 119
120 New constraint: ( ) among {,,, },{1},2, {0} u 1 {1} 2 {0} u 2 u 3 {1} {0} {1} 3 u 4 u 5 u 6 4 {1} {0} {1} {0} u 7 u 8 {1} {0} 1 Banff workshop. Slide 120 Eact BDD representation: 6 solutions
121 ( ) among {,,, },{1},2, {0,1} u 1 2 u 2 {0} {1} 3 u 3 u 4 {0,1} {0} {1} 4 u 5 u 6 {0,1} {0} 1 Banff workshop. Slide 121 Relaed BDD of width 2: 14 solutions
122 ( ) among {,,, },{1},2, {0,1} u 1 2 {0,1} u 2 3 {0,1} u 3 4 {0,1} u 4 Relaed BDD of width 1 is just the domain store 1
123 Propagation through a BDD At each node of the search tree: Apply each constraint to current BDD. Refine BDD to incorporate relaation of the constraint. Refine by splitting nodes, subject to ma width. Width limit keeps BDD size within bounds. Greater width yields tighter relaation. Slide 123
124 Propagation of among through a BDD 1 {0,1} u 1 2 {0,1} u 2 3 {0,1} u 3 4 {0,1} u 4 Arbitrarily start with BDD of width 1 and refine it 1
125 ( ) among {,,, },{1},2, u 1 {0,1} 2 {0,1} u 2 Two incoming edges are not equivalent. So split u 2 3 {0,1} u 3 4 {0,1} u 4 1 Banff workshop. Slide 125 Use a maimum width of 3
126 ( ) among {,,, },{1},2, u 1 {0} {1} 2 u 2 {0,1} u 2 Two incoming paths are not equivalent. So split u 2 3 {0,1} u 3 4 {0,1} u 4 1 Banff workshop. Slide 126 Use a maimum width of 3
127 ( ) among {,,, },{1},2, u 1 {0} {1} 2 u 2 {0,1} u 2 {0,1} Duplicate outgoing edges 3 u 3 {0,1} 4 u 4 {0,1} 1 Banff workshop. Slide 127 Use a maimum width of 3
128 ( ) among {,,, },{1},2, {0} u 1 {1} 2 u 2 u 2 3 {0,1} {0,1} u 3 {0,1} 3 incoming paths are nonequivalent 4 {0,1} u 4 1 Banff workshop. Slide 128 Use a maimum width of 3
129 ( ) among {,,, },{1},2, u 1 {0} {1} 2 {0} u 2 {1} {0} u 2 {1} 4 3 u 3 u 3 u 3 {0,1} {0,1} {0,1} {0,1} u 4 Split into 3 Banff workshop. Slide 129 Use a maimum width of 3 1
130 ( ) among {,,, },{1},2, u 1 {0} {1} 2 {0} u 2 {1} {0} u 2 {1} 4 3 u 3 u 3 u 3 {0,1} {0,1} {0,1} {0,1} u 4 Filter edge domains Banff workshop. Slide 130 Use a maimum width of 3 1
131 ( ) among {,,, },{1},2, u 1 {0} {1} 2 {0} u 2 {1} {0} u 2 {1} 4 3 u 3 u 3 u 3 {0,1} {1} {0} {0,1} u 4 Filter edge domains Banff workshop. Slide 131 Use a maimum width of 3 1
132 ( ) among {,,, },{1},2, u 1 {0} {1} 2 {0} u 2 u 3 {1} {0} {1} 3 u 4 u 5 u 6 4 {1} {0} {0} {1} u 7 u 8 {1} {0} Another split generates eact BDD, which has width 3 1 Banff workshop. Slide 132 Use a maimum width of 3
133 Computational results Multiple all-different constraints Multivalued variables encoded as binary variables Using domain propagation only: Some instances has search tree of > 1 million nodes Using BDD relaation: All instances solved at the root node. Slide 133
134 Number of branches + number of splits Domain store BDD-based Banff workshop. Slide 134 Ma width of BDD
135 Computation time (milliseconds) Domain store MDD-based Banff workshop. Slide 135 Ma width of MDD
136 Computational results Multiple among constraints Nurse scheduling problems When checking for equivalence of incoming paths Paths are viewed as nonequivalent if distance measure eceeds a threshold Results shown for finding first feasible solution Results are very similar for finding all feasible solutions Slide 136
137 Domain store only Search tree nodes 1.E+05 Class 2 (n=80) Search Tree Nodes 1.E+04 1.E+03 1.E+02 Threshold 1 Threshold 2 Threshold 3 Threshold 4 Threshold 5 1.E Banff workshop. Slide 137 Maimum Width
138 Domain store only Computation time Class 2 (n=80) Time (sec) Threshold 1 Threshold 2 Threshold 3 Threshold 4 Threshold Banff workshop. Slide 138 Maimum Width
Integrating Solution Methods. through Duality. US-Mexico Workshop on Optimization. Oaxaca, January John Hooker
Integrating Solution Methods through Duality John Hooker US-Meico Workshop on Optimization Oaaca, January 2011 Zapotec Duality mask Late classical period Represents life/death, day/night, heat/cold See
More informationA Scheme for Integrated Optimization
A Scheme for Integrated Optimization John Hooker ZIB November 2009 Slide 1 Outline Overview of integrated methods A taste of the underlying theory Eamples, with results from SIMPL Proposal for SCIP/SIMPL
More informationConsistency as Projection
Consistency as Projection John Hooker Carnegie Mellon University INFORMS 2015, Philadelphia USA Consistency as Projection Reconceive consistency in constraint programming as a form of projection. For eample,
More informationProjection in Logic, CP, and Optimization
Projection in Logic, CP, and Optimization John Hooker Carnegie Mellon University Workshop on Logic and Search Melbourne, 2017 Projection as a Unifying Concept Projection is a fundamental concept in logic,
More informationThe Separation Problem for Binary Decision Diagrams
The Separation Problem for Binary Decision Diagrams J. N. Hooker Joint work with André Ciré Carnegie Mellon University ISAIM 2014 Separation Problem in Optimization Given a relaxation of an optimization
More informationA Framework for Integrating Optimization and Constraint Programming
A Framework for Integrating Optimization and Constraint Programming John Hooker Carnegie Mellon University SARA 2007 SARA 07 Slide Underlying theme Model vs. solution method. How can we match the model
More informationProjection, Consistency, and George Boole
Projection, Consistency, and George Boole John Hooker Carnegie Mellon University CP 2015, Cork, Ireland Projection as a Unifying Concept Projection underlies both optimization and logical inference. Optimization
More informationProjection, Inference, and Consistency
Projection, Inference, and Consistency John Hooker Carnegie Mellon University IJCAI 2016, New York City A high-level presentation. Don t worry about the details. 2 Projection as a Unifying Concept Projection
More informationA Framework for Integrating Exact and Heuristic Optimization Methods
A Framework for Integrating Eact and Heuristic Optimization Methods John Hooker Carnegie Mellon University Matheuristics 2012 Eact and Heuristic Methods Generally regarded as very different. Eact methods
More informationDecision Diagrams: Tutorial
Decision Diagrams: Tutorial John Hooker Carnegie Mellon University CP Summer School Cork, Ireland, June 2016 Decision Diagrams Used in computer science and AI for decades Logic circuit design Product configuration
More informationLogic, Optimization and Data Analytics
Logic, Optimization and Data Analytics John Hooker Carnegie Mellon University United Technologies Research Center, Cork, Ireland August 2015 Thesis Logic and optimization have an underlying unity. Ideas
More informationDecision Diagrams for Discrete Optimization
Decision Diagrams for Discrete Optimization Willem Jan van Hoeve Tepper School of Business Carnegie Mellon University www.andrew.cmu.edu/user/vanhoeve/mdd/ Acknowledgments: David Bergman, Andre Cire, Samid
More informationOptimization Bounds from Binary Decision Diagrams
Optimization Bounds from Binary Decision Diagrams J. N. Hooker Joint work with David Bergman, André Ciré, Willem van Hoeve Carnegie Mellon University ICS 203 Binary Decision Diagrams BDDs historically
More informationHow to Relax. CP 2008 Slide 1. John Hooker Carnegie Mellon University September 2008
How to Relax Slide 1 John Hooker Carnegie Mellon University September 2008 Two ways to relax Relax your mind and body. Relax your problem formulations. Slide 2 Relaxing a problem Feasible set of original
More informationAn Integrated Approach to Truss Structure Design
Slide 1 An Integrated Approach to Truss Structure Design J. N. Hooker Tallys Yunes CPAIOR Workshop on Hybrid Methods for Nonlinear Combinatorial Problems Bologna, June 2010 How to Solve Nonlinear Combinatorial
More informationMultivalued Decision Diagrams. Postoptimality Analysis Using. J. N. Hooker. Tarik Hadzic. Cork Constraint Computation Centre
Postoptimality Analysis Using Multivalued Decision Diagrams Tarik Hadzic Cork Constraint Computation Centre J. N. Hooker Carnegie Mellon University London School of Economics June 2008 Postoptimality Analysis
More informationAlternative Methods for Obtaining. Optimization Bounds. AFOSR Program Review, April Carnegie Mellon University. Grant FA
Alternative Methods for Obtaining Optimization Bounds J. N. Hooker Carnegie Mellon University AFOSR Program Review, April 2012 Grant FA9550-11-1-0180 Integrating OR and CP/AI Early support by AFOSR First
More informationDuality in Optimization and Constraint Satisfaction
Duality in Optimization and Constraint Satisfaction J. N. Hooker Carnegie Mellon University, Pittsburgh, USA john@hooker.tepper.cmu.edu Abstract. We show that various duals that occur in optimization and
More informationScheduling Home Hospice Care with Logic-Based Benders Decomposition
Scheduling Home Hospice Care with Logic-Based Benders Decomposition John Hooker Carnegie Mellon University Joint work with Aliza Heching Ryo Kimura Compassionate Care Hospice CMU Lehigh University October
More informationOptimization in Process Systems Engineering
Optimization in Process Systems Engineering M.Sc. Jan Kronqvist Process Design & Systems Engineering Laboratory Faculty of Science and Engineering Åbo Akademi University Most optimization problems in production
More informationOperations Research Methods in Constraint Programming
Operations Research Methods in Constraint Programming John Hooker Carnegie Mellon University Prepared for Lloret de Mar, Spain June 2007 2007 Slide 1 CP and OR Have Complementary Strengths CP: Inference
More informationSingle-Facility Scheduling by Logic-Based Benders Decomposition
Single-Facility Scheduling by Logic-Based Benders Decomposition Elvin Coban J. N. Hooker Tepper School of Business Carnegie Mellon University ecoban@andrew.cmu.edu john@hooker.tepper.cmu.edu September
More informationDecision Diagram Relaxations for Integer Programming
Decision Diagram Relaxations for Integer Programming Christian Tjandraatmadja April, 2018 Tepper School of Business Carnegie Mellon University Submitted to the Tepper School of Business in Partial Fulfillment
More informationA Search-Infer-and-Relax Framework for Integrating Solution Methods
Carnegie Mellon University Research Showcase @ CMU Tepper School of Business 2005 A Search-Infer-and-Relax Framework for Integrating Solution Methods John N. Hooker Carnegie Mellon University, john@hooker.tepper.cmu.edu
More informationJob Sequencing Bounds from Decision Diagrams
Job Sequencing Bounds from Decision Diagrams J. N. Hooker Carnegie Mellon University June 2017 Abstract. In recent research, decision diagrams have proved useful for the solution of discrete optimization
More informationStochastic Decision Diagrams
Stochastic Decision Diagrams John Hooker CORS/INFORMS Montréal June 2015 Objective Relaxed decision diagrams provide an generalpurpose method for discrete optimization. When the problem has a dynamic programming
More informationInteger Programming and Branch and Bound
Courtesy of Sommer Gentry. Used with permission. Integer Programming and Branch and Bound Sommer Gentry November 4 th, 003 Adapted from slides by Eric Feron and Brian Williams, 6.40, 00. Integer Programming
More informationLogic-Based Benders Decomposition
Logic-Based Benders Decomposition J. N. Hooker Graduate School of Industrial Administration Carnegie Mellon University, Pittsburgh, PA 15213 USA G. Ottosson Department of Information Technology Computing
More informationThe CPLEX Library: Mixed Integer Programming
The CPLEX Library: Mixed Programming Ed Rothberg, ILOG, Inc. 1 The Diet Problem Revisited Nutritional values Bob considered the following foods: Food Serving Size Energy (kcal) Protein (g) Calcium (mg)
More informationwith Binary Decision Diagrams Integer Programming J. N. Hooker Tarik Hadzic IT University of Copenhagen Carnegie Mellon University ICS 2007, January
Integer Programming with Binary Decision Diagrams Tarik Hadzic IT University of Copenhagen J. N. Hooker Carnegie Mellon University ICS 2007, January Integer Programming with BDDs Goal: Use binary decision
More informationOptimization Methods in Logic
Optimization Methods in Logic John Hooker Carnegie Mellon University February 2003, Revised December 2008 1 Numerical Semantics for Logic Optimization can make at least two contributions to boolean logic.
More informationDecision Diagrams for Sequencing and Scheduling
Decision Diagrams for Sequencing and Scheduling Willem-Jan van Hoeve Tepper School of Business Carnegie Mellon University www.andrew.cmu.edu/user/vanhoeve/mdd/ Plan What can MDDs do for Combinatorial Optimization?
More informationBenders decomposition [7, 17] uses a problem-solving strategy that can be generalized to a larger context. It assigns some of the variables trial valu
Logic-Based Benders Decomposition Λ J. N. Hooker Graduate School of Industrial Administration Carnegie Mellon University, Pittsburgh, PA 15213 USA G. Ottosson Department of Information Technology Computing
More informationmaxz = 3x 1 +4x 2 2x 1 +x 2 6 2x 1 +3x 2 9 x 1,x 2
ex-5.-5. Foundations of Operations Research Prof. E. Amaldi 5. Branch-and-Bound Given the integer linear program maxz = x +x x +x 6 x +x 9 x,x integer solve it via the Branch-and-Bound method (solving
More informationarxiv: v1 [cs.cc] 5 Dec 2018
Consistency for 0 1 Programming Danial Davarnia 1 and J. N. Hooker 2 1 Iowa state University davarnia@iastate.edu 2 Carnegie Mellon University jh38@andrew.cmu.edu arxiv:1812.02215v1 [cs.cc] 5 Dec 2018
More informationAdvances in Bayesian Network Learning using Integer Programming
Advances in Bayesian Network Learning using Integer Programming Mark Bartlett and James Cussens UAI-13, 2013-07-12 Supported by the UK Medical Research Council (Project Grant G1002312) Mark Bartlett and
More informationmin3x 1 + 4x 2 + 5x 3 2x 1 + 2x 2 + x 3 6 x 1 + 2x 2 + 3x 3 5 x 1, x 2, x 3 0.
ex-.-. Foundations of Operations Research Prof. E. Amaldi. Dual simplex algorithm Given the linear program minx + x + x x + x + x 6 x + x + x x, x, x. solve it via the dual simplex algorithm. Describe
More informationMDD-based Postoptimality Analysis for Mixed-integer Programs
MDD-based Postoptimality Analysis for Mixed-integer Programs John Hooker, Ryo Kimura Carnegie Mellon University Thiago Serra Mitsubishi Electric Research Laboratories Symposium on Decision Diagrams for
More informationIntegrating CP and Mathematical Programming
Integrating CP and Mathematical Programming John Hooker Carnegie Mellon University June 2011 June 2011 Slide 1 Why Integrate CP and MP? Complementary strengths Computational advantages Outline of the Tutorial
More informationSolving Mixed-Integer Nonlinear Programs
Solving Mixed-Integer Nonlinear Programs (with SCIP) Ambros M. Gleixner Zuse Institute Berlin MATHEON Berlin Mathematical School 5th Porto Meeting on Mathematics for Industry, April 10 11, 2014, Porto
More informationMixed Integer Programming Solvers: from Where to Where. Andrea Lodi University of Bologna, Italy
Mixed Integer Programming Solvers: from Where to Where Andrea Lodi University of Bologna, Italy andrea.lodi@unibo.it November 30, 2011 @ Explanatory Workshop on Locational Analysis, Sevilla A. Lodi, MIP
More informationLift-and-Project cuts: an efficient solution method for mixed-integer programs
Lift-and-Project cuts: an efficient solution method for mied-integer programs Sebastian Ceria Graduate School of Business and Computational Optimization Research Center http://www.columbia.edu/~sc44 Columbia
More informationSection Notes 8. Integer Programming II. Applied Math 121. Week of April 5, expand your knowledge of big M s and logical constraints.
Section Notes 8 Integer Programming II Applied Math 121 Week of April 5, 2010 Goals for the week understand IP relaxations be able to determine the relative strength of formulations understand the branch
More informationDisconnecting Networks via Node Deletions
1 / 27 Disconnecting Networks via Node Deletions Exact Interdiction Models and Algorithms Siqian Shen 1 J. Cole Smith 2 R. Goli 2 1 IOE, University of Michigan 2 ISE, University of Florida 2012 INFORMS
More informationwhere X is the feasible region, i.e., the set of the feasible solutions.
3.5 Branch and Bound Consider a generic Discrete Optimization problem (P) z = max{c(x) : x X }, where X is the feasible region, i.e., the set of the feasible solutions. Branch and Bound is a general semi-enumerative
More informationCombining Optimization and Constraint Programming
Combining Optimization and Constraint Programming John Hooker Carnegie Mellon University GE Research Center 7 September 2007 GE 7 Sep 07 Slide 1 Optimization and Constraint Programming Optimization: focus
More informationSystem Planning Lecture 7, F7: Optimization
System Planning 04 Lecture 7, F7: Optimization System Planning 04 Lecture 7, F7: Optimization Course goals Appendi A Content: Generally about optimization Formulate optimization problems Linear Programming
More informationRepresentations of All Solutions of Boolean Programming Problems
Representations of All Solutions of Boolean Programming Problems Utz-Uwe Haus and Carla Michini Institute for Operations Research Department of Mathematics ETH Zurich Rämistr. 101, 8092 Zürich, Switzerland
More informationInteger Programming for Bayesian Network Structure Learning
Integer Programming for Bayesian Network Structure Learning James Cussens Prague, 2013-09-02 Supported by the UK Medical Research Council (Project Grant G1002312) James Cussens IP for BNs Prague, 2013-09-02
More informationLecture 23 Branch-and-Bound Algorithm. November 3, 2009
Branch-and-Bound Algorithm November 3, 2009 Outline Lecture 23 Modeling aspect: Either-Or requirement Special ILPs: Totally unimodular matrices Branch-and-Bound Algorithm Underlying idea Terminology Formal
More informationIP Duality. Menal Guzelsoy. Seminar Series, /21-07/28-08/04-08/11. Department of Industrial and Systems Engineering Lehigh University
IP Duality Department of Industrial and Systems Engineering Lehigh University COR@L Seminar Series, 2005 07/21-07/28-08/04-08/11 Outline Duality Theorem 1 Duality Theorem Introduction Optimality Conditions
More informationEXACT ALGORITHMS FOR THE ATSP
EXACT ALGORITHMS FOR THE ATSP Branch-and-Bound Algorithms: Little-Murty-Sweeney-Karel (Operations Research, ); Bellmore-Malone (Operations Research, ); Garfinkel (Operations Research, ); Smith-Srinivasan-Thompson
More informationSection Notes 9. IP: Cutting Planes. Applied Math 121. Week of April 12, 2010
Section Notes 9 IP: Cutting Planes Applied Math 121 Week of April 12, 2010 Goals for the week understand what a strong formulations is. be familiar with the cutting planes algorithm and the types of cuts
More informationInformation-Based Branching Schemes for Binary Linear Mixed Integer Problems
Information-Based Branching Schemes for Binary Linear Mixed Integer Problems Fatma Kılınç Karzan, George L. Nemhauser, Martin W.P. Savelsbergh H. Milton Stewart Industrial and Systems Engineering, Georgia
More informationIntroduction to Mathematical Programming IE406. Lecture 21. Dr. Ted Ralphs
Introduction to Mathematical Programming IE406 Lecture 21 Dr. Ted Ralphs IE406 Lecture 21 1 Reading for This Lecture Bertsimas Sections 10.2, 10.3, 11.1, 11.2 IE406 Lecture 21 2 Branch and Bound Branch
More informationFeasibility Pump Heuristics for Column Generation Approaches
1 / 29 Feasibility Pump Heuristics for Column Generation Approaches Ruslan Sadykov 2 Pierre Pesneau 1,2 Francois Vanderbeck 1,2 1 University Bordeaux I 2 INRIA Bordeaux Sud-Ouest SEA 2012 Bordeaux, France,
More informationFrom structures to heuristics to global solvers
From structures to heuristics to global solvers Timo Berthold Zuse Institute Berlin DFG Research Center MATHEON Mathematics for key technologies OR2013, 04/Sep/13, Rotterdam Outline From structures to
More informationIE418 Integer Programming
IE418: Integer Programming Department of Industrial and Systems Engineering Lehigh University 2nd February 2005 Boring Stuff Extra Linux Class: 8AM 11AM, Wednesday February 9. Room??? Accounts and Passwords
More informationPolyhedral Approach to Integer Linear Programming. Tepper School of Business Carnegie Mellon University, Pittsburgh
Polyhedral Approach to Integer Linear Programming Gérard Cornuéjols Tepper School of Business Carnegie Mellon University, Pittsburgh 1 / 30 Brief history First Algorithms Polynomial Algorithms Solving
More informationMath Models of OR: Branch-and-Bound
Math Models of OR: Branch-and-Bound John E. Mitchell Department of Mathematical Sciences RPI, Troy, NY 12180 USA November 2018 Mitchell Branch-and-Bound 1 / 15 Branch-and-Bound Outline 1 Branch-and-Bound
More informationDecision Procedures An Algorithmic Point of View
An Algorithmic Point of View ILP References: Integer Programming / Laurence Wolsey Deciding ILPs with Branch & Bound Intro. To mathematical programming / Hillier, Lieberman Daniel Kroening and Ofer Strichman
More informationInterior-Point versus Simplex methods for Integer Programming Branch-and-Bound
Interior-Point versus Simplex methods for Integer Programming Branch-and-Bound Samir Elhedhli elhedhli@uwaterloo.ca Department of Management Sciences, University of Waterloo, Canada Page of 4 McMaster
More informationSolvers for the Problem of Boolean Satisfiability (SAT) Will Klieber Aug 31, 2011
Solvers for the Problem of Boolean Satisfiability (SAT) Will Klieber 15-414 Aug 31, 2011 Why study SAT solvers? Many problems reduce to SAT. Formal verification CAD, VLSI Optimization AI, planning, automated
More informationComputational Logic. Davide Martinenghi. Spring Free University of Bozen-Bolzano. Computational Logic Davide Martinenghi (1/30)
Computational Logic Davide Martinenghi Free University of Bozen-Bolzano Spring 2010 Computational Logic Davide Martinenghi (1/30) Propositional Logic - sequent calculus To overcome the problems of natural
More informationParallel PIPS-SBB Multi-level parallelism for 2-stage SMIPS. Lluís-Miquel Munguia, Geoffrey M. Oxberry, Deepak Rajan, Yuji Shinano
Parallel PIPS-SBB Multi-level parallelism for 2-stage SMIPS Lluís-Miquel Munguia, Geoffrey M. Oxberry, Deepak Rajan, Yuji Shinano ... Our contribution PIPS-PSBB*: Multi-level parallelism for Stochastic
More information1 Solution of a Large-Scale Traveling-Salesman Problem... 7 George B. Dantzig, Delbert R. Fulkerson, and Selmer M. Johnson
Part I The Early Years 1 Solution of a Large-Scale Traveling-Salesman Problem............ 7 George B. Dantzig, Delbert R. Fulkerson, and Selmer M. Johnson 2 The Hungarian Method for the Assignment Problem..............
More informationInteger Programming. Wolfram Wiesemann. December 6, 2007
Integer Programming Wolfram Wiesemann December 6, 2007 Contents of this Lecture Revision: Mixed Integer Programming Problems Branch & Bound Algorithms: The Big Picture Solving MIP s: Complete Enumeration
More informationInteger Programming ISE 418. Lecture 8. Dr. Ted Ralphs
Integer Programming ISE 418 Lecture 8 Dr. Ted Ralphs ISE 418 Lecture 8 1 Reading for This Lecture Wolsey Chapter 2 Nemhauser and Wolsey Sections II.3.1, II.3.6, II.4.1, II.4.2, II.5.4 Duality for Mixed-Integer
More informationUsing Sparsity to Design Primal Heuristics for MILPs: Two Stories
for MILPs: Two Stories Santanu S. Dey Joint work with: Andres Iroume, Marco Molinaro, Domenico Salvagnin, Qianyi Wang MIP Workshop, 2017 Sparsity in real" Integer Programs (IPs) Real" IPs are sparse: The
More informationA Lifted Linear Programming Branch-and-Bound Algorithm for Mixed Integer Conic Quadratic Programs
A Lifted Linear Programming Branch-and-Bound Algorithm for Mied Integer Conic Quadratic Programs Juan Pablo Vielma Shabbir Ahmed George L. Nemhauser H. Milton Stewart School of Industrial and Systems Engineering
More informationInteger Programming. The focus of this chapter is on solution techniques for integer programming models.
Integer Programming Introduction The general linear programming model depends on the assumption of divisibility. In other words, the decision variables are allowed to take non-negative integer as well
More informationAnalyzing the computational impact of individual MINLP solver components
Analyzing the computational impact of individual MINLP solver components Ambros M. Gleixner joint work with Stefan Vigerske Zuse Institute Berlin MATHEON Berlin Mathematical School MINLP 2014, June 4,
More information18 hours nodes, first feasible 3.7% gap Time: 92 days!! LP relaxation at root node: Branch and bound
The MIP Landscape 1 Example 1: LP still can be HARD SGM: Schedule Generation Model Example 157323 1: LP rows, still can 182812 be HARD columns, 6348437 nzs LP relaxation at root node: 18 hours Branch and
More informationCutting Planes in SCIP
Cutting Planes in SCIP Kati Wolter Zuse-Institute Berlin Department Optimization Berlin, 6th June 2007 Outline 1 Cutting Planes in SCIP 2 Cutting Planes for the 0-1 Knapsack Problem 2.1 Cover Cuts 2.2
More informationAn Adaptive Partition-based Approach for Solving Two-stage Stochastic Programs with Fixed Recourse
An Adaptive Partition-based Approach for Solving Two-stage Stochastic Programs with Fixed Recourse Yongjia Song, James Luedtke Virginia Commonwealth University, Richmond, VA, ysong3@vcu.edu University
More informationIntroduction to Integer Linear Programming
Lecture 7/12/2006 p. 1/30 Introduction to Integer Linear Programming Leo Liberti, Ruslan Sadykov LIX, École Polytechnique liberti@lix.polytechnique.fr sadykov@lix.polytechnique.fr Lecture 7/12/2006 p.
More informationInexact Solution of NLP Subproblems in MINLP
Ineact Solution of NLP Subproblems in MINLP M. Li L. N. Vicente April 4, 2011 Abstract In the contet of conve mied-integer nonlinear programming (MINLP, we investigate how the outer approimation method
More informationSome Recent Advances in Mixed-Integer Nonlinear Programming
Some Recent Advances in Mixed-Integer Nonlinear Programming Andreas Wächter IBM T.J. Watson Research Center Yorktown Heights, New York andreasw@us.ibm.com SIAM Conference on Optimization 2008 Boston, MA
More informationis called an integer programming (IP) problem. model is called a mixed integer programming (MIP)
INTEGER PROGRAMMING Integer Programming g In many problems the decision variables must have integer values. Example: assign people, machines, and vehicles to activities in integer quantities. If this is
More informationOrbital Shrinking: A New Tool for Hybrid MIP/CP Methods
Orbital Shrinking: A New Tool for Hybrid MIP/CP Methods Domenico Salvagnin DEI, University of Padova salvagni@dei.unipd.it Abstract. Orbital shrinking is a newly developed technique in the MIP community
More informationIndicator Constraints in Mixed-Integer Programming
Indicator Constraints in Mixed-Integer Programming Andrea Lodi University of Bologna, Italy - andrea.lodi@unibo.it Amaya Nogales-Gómez, Universidad de Sevilla, Spain Pietro Belotti, FICO, UK Matteo Fischetti,
More informationInteger Programming, Constraint Programming, and their Combination
Integer Programming, Constraint Programming, and their Combination Alexander Bockmayr Freie Universität Berlin & DFG Research Center Matheon Eindhoven, 27 January 2006 Discrete Optimization General framework
More informationLinear & Integer programming
ELL 894 Performance Evaluation on Communication Networks Standard form I Lecture 5 Linear & Integer programming subject to where b is a vector of length m c T A = b (decision variables) and c are vectors
More informationImproving the Randomization Step in Feasibility Pump using WalkSAT
Improving the Randomization Step in Feasibility Pump using Santanu S. Dey Joint work with: Andres Iroume, Marco Molinaro, Domenico Salvagnin Discrepancy & IP workshop, 2018 Sparsity in real" Integer Programs
More informationComputational Mixed-Integer Programming
Computational Mixed-Integer Programming Ambros Gleixner and the SCIP team Zuse Institute Berlin gleixner@zib.de SCIP Optimization Suite http://scip.zib.de Theory and Practice of Satisfiability Solving
More informationSection Notes 9. Midterm 2 Review. Applied Math / Engineering Sciences 121. Week of December 3, 2018
Section Notes 9 Midterm 2 Review Applied Math / Engineering Sciences 121 Week of December 3, 2018 The following list of topics is an overview of the material that was covered in the lectures and sections
More informationInteger Linear Programming (ILP)
Integer Linear Programming (ILP) Zdeněk Hanzálek, Přemysl Šůcha hanzalek@fel.cvut.cz CTU in Prague March 8, 2017 Z. Hanzálek (CTU) Integer Linear Programming (ILP) March 8, 2017 1 / 43 Table of contents
More informationIBM Research Report. A Feasibility Pump for Mixed Integer Nonlinear Programs
RC386 (W060-09) February, 006 Mathematics IBM Research Report A Feasibility Pump for Mied Integer Nonlinear Programs Pierre Bonami, Gérard Cornuéjols, Andrea Lodi*, François Margot Tepper School of Business
More informationTutorial: Operations Research in Constraint Programming
Tutorial: Operations Research in Constraint Programming John Hooker Carnegie Mellon University May 2009 Revised June 2009 May 2009 Slide 1 Why Integrate OR and CP? Complementary strengths Computational
More informationAn Integrated Method for Planning and Scheduling to Minimize Tardiness
An Integrated Method for Planning and Scheduling to Minimize Tardiness J N Hooker Carnegie Mellon University john@hookerteppercmuedu Abstract We combine mixed integer linear programming (MILP) and constraint
More informationLogic-based Benders Decomposition
Logic-based Benders Decomposition A short Introduction Martin Riedler AC Retreat Contents 1 Introduction 2 Motivation 3 Further Notes MR Logic-based Benders Decomposition June 29 July 1 2 / 15 Basic idea
More informationOperations Research Lecture 6: Integer Programming
Operations Research Lecture 6: Integer Programming Notes taken by Kaiquan Xu@Business School, Nanjing University May 12th 2016 1 Integer programming (IP) formulations The integer programming (IP) is the
More informationDuality, Warm Starting, and Sensitivity Analysis for MILP
Duality, Warm Starting, and Sensitivity Analysis for MILP Ted Ralphs and Menal Guzelsoy Industrial and Systems Engineering Lehigh University SAS Institute, Cary, NC, Friday, August 19, 2005 SAS Institute
More informationHighly-scalable branch and bound for maximum monomial agreement
Highly-scalable branch and bound for maximum monomial agreement Jonathan Eckstein (Rutgers) William Hart Cynthia A. Phillips Sandia National Laboratories Sandia National Laboratories is a multi-program
More informationCan Li a, Ignacio E. Grossmann a,
A generalized Benders decomposition-based branch and cut algorithm for two-stage stochastic programs with nonconvex constraints and mixed-binary first and second stage variables Can Li a, Ignacio E. Grossmann
More informationGeneration and Representation of Piecewise Polyhedral Value Functions
Generation and Representation of Piecewise Polyhedral Value Functions Ted Ralphs 1 Joint work with Menal Güzelsoy 2 and Anahita Hassanzadeh 1 1 COR@L Lab, Department of Industrial and Systems Engineering,
More informationLecture 2. Split Inequalities and Gomory Mixed Integer Cuts. Tepper School of Business Carnegie Mellon University, Pittsburgh
Lecture 2 Split Inequalities and Gomory Mixed Integer Cuts Gérard Cornuéjols Tepper School of Business Carnegie Mellon University, Pittsburgh Mixed Integer Cuts Gomory 1963 Consider a single constraint
More information23. Cutting planes and branch & bound
CS/ECE/ISyE 524 Introduction to Optimization Spring 207 8 23. Cutting planes and branch & bound ˆ Algorithms for solving MIPs ˆ Cutting plane methods ˆ Branch and bound methods Laurent Lessard (www.laurentlessard.com)
More informationSAT Modulo ODE: A Direct SAT Approach to Hybrid Systems
Welcome ATVA 2008 SAT Modulo ODE: A Direct SAT Approach to Hbrid Sstems Andreas Eggers, Martin Fränzle, and Christian Herde DFG Transregional Collaborative Research Center AVACS Outline Motivation Bounded
More informationCan Li a, Ignacio E. Grossmann a,
A generalized Benders decomposition-based branch and cut algorithm for two-stage stochastic programs with nonconvex constraints and mixed-binary first and second stage variables Can Li a, Ignacio E. Grossmann
More information