Optimizing over the Split Closure Anureet Saxena ACO PhD Student, Tepper School of Business, Carnegie Mellon University. (Joint Work with Egon Balas)

Slides:



Advertisements
Similar presentations
IEOR 4004 Final Review part II.
Advertisements

1 LP Duality Lecture 13: Feb Min-Max Theorems In bipartite graph, Maximum matching = Minimum Vertex Cover In every graph, Maximum Flow = Minimum.
Gomory’s cutting plane algorithm for integer programming Prepared by Shin-ichi Tanigawa.
EE 553 Integer Programming
Konrad-Zuse-Zentrum für Informationstechnik Berlin (ZIB) Tobias Achterberg Conflict Analysis SCIP Workshop at ZIB October 2007.
Progress in Linear Programming Based Branch-and-Bound Algorithms
EMIS 8373: Integer Programming Valid Inequalities updated 4April 2011.
Dash Optimization IMA 2005 “Constraint Branching and Disjunctive Cuts for Mixed Integer Programs” 1 Constraint Branching and Disjunctive Cuts for Mixed.
1 One Size Fits All? : Computational Tradeoffs in Mixed Integer Programming Software Bob Bixby, Mary Fenelon, Zonghao Gu, Ed Rothberg, and Roland Wunderling.
Looking inside Gomory Aussois, January Looking inside Gomory Matteo Fischetti, DEI University of Padova (joint work with Egon Balas and Arrigo.
1 Logic-Based Methods for Global Optimization J. N. Hooker Carnegie Mellon University, USA November 2003.
Solving Integer Programs. Natural solution ideas that don’t work well Solution idea #1: Explicit enumeration: Try all possible solutions and pick the.
Duality in Optimization and Constraint Satisfaction J. N. Hooker Carnegie Mellon Univ. Pittsburgh, USA September 2006.
Aussois, 4-8/1/20101 = ? Matteo Fischetti and Domenico Salvagnin University of Padova +
Computational Methods for Management and Economics Carla Gomes
Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy under contract.
1 Contents college 3 en 4 Book: Appendix A.1, A.3, A.4, §3.4, §3.5, §4.1, §4.2, §4.4, §4.6 (not: §3.6 - §3.8, §4.2 - §4.3) Extra literature on resource.
1 Robustness by cutting planes and the Uncertain Set Covering Problem AIRO 2008, Ischia, 10 September 2008 (work supported my MiUR and EU) Matteo Fischetti.
1 Combinatorial Problems in Cooperative Control: Complexity and Scalability Carla Gomes and Bart Selman Cornell University Muri Meeting March 2002.
Lift-and-Project cuts: an efficient solution method for mixed-integer programs Sebastian Ceria Graduate School of Business and Computational Optimization.
LP formulation of Economic Dispatch
The Problem with Integer Programming H.P.Williams London School of Economics.
Daniel Kroening and Ofer Strichman Decision Procedures An Algorithmic Point of View Deciding ILPs with Branch & Bound ILP References: ‘Integer Programming’
How to Play Sudoku & Win Integer Programming Formulation of a Popular Game Sven LeyfferSven Leyffer, Argonne, Feb. 15, 2005 (windoze powerpoint sumi painting.
Decision Procedures An Algorithmic Point of View
A Decomposition Heuristic for Stochastic Programming Natashia Boland +, Matteo Fischetti*, Michele Monaci*, Martin Savelsbergh + + Georgia Institute of.
1 EL736 Communications Networks II: Design and Algorithms Class5: Optimization Methods Yong Liu 10/10/2007.
The Chvátal Dual of a Pure Integer Programme H.P.Williams London School of Economics.
Strong Bounds for Linear Programs with Cardinality Limited Violation (CLV) Constraint Systems Ronald L. Rardin University of Arkansas
C&O 355 Mathematical Programming Fall 2010 Lecture 18 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A.
MILP algorithms: branch-and-bound and branch-and-cut
Tobias Achterberg Konrad-Zuse-Zentrum für Informationstechnik Berlin Branching SCIP Workshop at ZIB October 2007.
1 Chapter 4: Integer and Mixed-Integer Linear Programming Problems 4.1 Introduction to Integer and Mixed-Integer Linear Programming 4.2 Solving Integer.
Integer programming, MA Operational Research1 Integer Programming Operational Research -Level 4 Prepared by T.M.J.A.Cooray Department of Mathematics.
15.053Tuesday, April 9 Branch and Bound Handouts: Lecture Notes.
Gomory Cuts Updated 25 March Example ILP Example taken from “Operations Research: An Introduction” by Hamdy A. Taha (8 th Edition)“Operations Research:
Branch-and-Cut Valid inequality: an inequality satisfied by all feasible solutions Cut: a valid inequality that is not part of the current formulation.
1 A polynomial relaxation-type algorithm for linear programming Sergei Chubanov University of Siegen, Germany
FATCOP: A Mixed Integer Program Solver Michael FerrisQun Chen Department of Computer Sciences University of Wisconsin-Madison Jeff Linderoth, Argonne.
Embedding Formulations, Complexity and Representability for Unions of Convex Sets Juan Pablo Vielma Massachusetts Institute of Technology CMO-BIRS Workshop:
Implicit Hitting Set Problems Richard M. Karp Erick Moreno Centeno DIMACS 20 th Anniversary.
CPSC 536N Sparse Approximations Winter 2013 Lecture 1 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AAAAAAAAAA.
OR Chapter 7. The Revised Simplex Method  Recall Theorem 3.1, same basis  same dictionary Entire dictionary can be constructed as long as we.
FATCOP: A Mixed Integer Program Solver Michael FerrisQun Chen University of Wisconsin-Madison Jeffrey Linderoth Argonne National Laboratories.
Integer Programming, Branch & Bound Method
Intersection Cuts for Bilevel Optimization
Linear Programming Piyush Kumar Welcome to CIS5930.
Konrad-Zuse-Zentrum für Informationstechnik Berlin (ZIB) Tobias Achterberg Conflict Analysis in Mixed Integer Programming.
Discrete Optimization MA2827 Fondements de l’optimisation discrète Material from P. Van Hentenryck’s course.
Sebastian Ceria Graduate School of Business and
Integer Programming An integer linear program (ILP) is defined exactly as a linear program except that values of variables in a feasible solution have.
From Mixed-Integer Linear to Mixed-Integer Bilevel Linear Programming
Exact Algorithms for Mixed-Integer Bilevel Linear Programming
The CPLEX Library: Mixed Integer Programming
EMGT 6412/MATH 6665 Mathematical Programming Spring 2016
Matteo Fischetti, University of Padova
MILP algorithms: branch-and-bound and branch-and-cut
Gomory Cuts Updated 25 March 2009.
Chap 9. General LP problems: Duality and Infeasibility
Chapter 6. Large Scale Optimization
Chapter 5. Sensitivity Analysis
MIP Tools Branch and Cut with Callbacks Lazy Constraint Callback
Chapter 6. Large Scale Optimization
2. Generating All Valid Inequalities
Matteo Fischetti, University of Padova
Flow Feasibility Problems
Part II General Integer Programming
Intersection Cuts from Bilinear Disjunctions
Chapter 6. Large Scale Optimization
Intersection Cuts for Quadratic Mixed-Integer Optimization
Presentation transcript:

Optimizing over the Split Closure Anureet Saxena ACO PhD Student, Tepper School of Business, Carnegie Mellon University. (Joint Work with Egon Balas)

Anureet Saxena, TSoB1 Talk Outiline Cutting Planes Commercial Split Closure Separation Problem PMILP & Deparametrization Computational Results Support Size & Sparsity Support Coefficients Cuts Statistics arki001 solved

Anureet Saxena, TSoB2 MIP Model min cx Ax ¸ b x j 2 Z 8 j2N 1 N 1 : set of integer variables Contains x j ¸ 0 j2N x j · u j j2N 1 Incumbent Fractional Solution

Anureet Saxena, TSoB3 Taxonomy of Cutting Planes Fractional Basic Feas Fractional Basic Mixed Integer Basic Feas Intersection Basic Feas Intersection Basic L&P Simple Disjunctive Chvatal Fractional Gomory Mixed Integer Basic Intersection Basic + Strengthening L&P + Strengthening MIG Split Cuts Intersection Basic Feas + Strengthening MIR

Anureet Saxena, TSoB4 Taxonomy of Cutting Planes Fractional Basic Feas Fractional Basic Mixed Integer Basic Feas Intersection Basic Feas Intersection Basic L&P Simple Disjunctive Chvatal Fractional Gomory Mixed Integer Basic Intersection Basic + Strengthening L&P + Strengthening MIG Split Cuts Intersection Basic Feas + Strengthening MIR

Anureet Saxena, TSoB5 Taxonomy of Cutting Planes Fractional Basic Feas Fractional Basic Mixed Integer Basic Feas Intersection Basic Feas Intersection Basic L&P Simple Disjunctive Chvatal Fractional Gomory Mixed Integer Basic Intersection Basic + Strengthening L&P + Strengthening MIG Split Cuts Intersection Basic Feas + Strengthening MIR

Anureet Saxena, TSoB6 Taxonomy of Cutting Planes Fractional Basic Feas Fractional Basic Mixed Integer Basic Feas Intersection Basic Feas Intersection Basic L&P Simple Disjunctive Chvatal Fractional Gomory Mixed Integer Basic Intersection Basic + Strengthening L&P + Strengthening MIG Split Cuts Intersection Basic Feas + Strengthening MIR

Anureet Saxena, TSoB7 Taxonomy of Cutting Planes Elementary Closure Elementary closure of P w.r.t a family  of cutting planes is defined by intersecting P with all rank-1 cuts in  Eg: CG Closure, Split Closure

Anureet Saxena, TSoB8 Elementary Closures Intersection Basic L&P Simple Disjunctive Chvatal Fractional Gomory MIG Split Cuts MIR Split Closure CG Closure L&P Closure

Anureet Saxena, TSoB9 Elementary Closures Operations Research Constraint Programming Complexity Theory max v x2P I ) P cx¸v P2  Inference Dual Proof Family Rank-1 cuts have short polynomial length proofs

Anureet Saxena, TSoB10 Elementary Closures How much duality gap can be closed by optimizing over elementary closures? L&P Closure Bonami and Minoux CG Closure Fischetti and Lodi Split Closure ?

Anureet Saxena, TSoB11 Elementary Closures How much duality gap can be closed by optimizing over elementary closures? L&P Closure Bonami and Minoux CG Closure Fischetti and Lodi Split Closure Balas and Saxena

Anureet Saxena, TSoB12 Split Disjunctions  2 Z N,  0 2 Z  j = 0, j2 N 2  0 <  <  Split Disjunction  x ·  0  x ¸  0 + 1

Anureet Saxena, TSoB13 Split Cuts Ax ¸ b  x ·  0 Ax ¸ b  x ¸  0 +1 u u0u0 v0v0 v  L x ¸  L  R x ¸  R  x ¸  Split Cut

Anureet Saxena, TSoB14 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, TSoB15 Algorithmic Framework Solve Master LP Integral Sol? Unbounded? Infeasible? Rank-1 Split Cut Separation MIP Solved Optimum over Split Closure attained Split Cuts Generated No Split Cuts Generated min cx Ax ¸ b  t x¸  t t2  Yes No Add Cuts

Anureet Saxena, TSoB16 Algorithmic Framework Solve Master LP Integral Sol? Unbounded? Infeasible? Rank-1 Split Cut Separation Rank-1 Split Cut Separation MIP Solved Optimum over Split Closure attained Split Cuts Generated No Split Cuts Generated min cx Ax ¸ b  t x¸  t t2  Yes No Add Cuts

Anureet Saxena, TSoB17 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. Disjunctive Cut Cut Violation Split Disjunction Normalization Set  = 1 u.e + v.e + u 0 + v 0 = 1 u 0 + v 0 = 1  y = 1 |  | 2 =1

Anureet Saxena, TSoB18 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 u 0 + v 0 = 1

Anureet Saxena, TSoB19 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, TSoB20 Deparametrization Parameteric Mixed Integer Linear Program

Anureet Saxena, TSoB21 Deparametrization Parameteric Mixed Integer Linear Program If  is fixed, then PMILP reduces to a MILP

Anureet Saxena, TSoB22 Deparametrization MILP( ) Deparametrized Mixed Integer Linear Program Maintain a dynamically updated grid of parameters

Anureet Saxena, TSoB23 Separation Algorithm Initialize Parameter Grid (  ) For  2 , Solve MILP(  ) using CPLEX 9.0 Enumerate  branch and bound nodes Store all the separating split disjunctions which are discovered At least one split disjunction discovered? Grid Enrichment Diversification Strengthening STOP Bifurcation yes no

Anureet Saxena, TSoB24 Implementation Details Processor Details Pentium IV 2Ghz, 2GB RAM COIN-ORCPLEX 9.0 Core Implementation Solving Master LP Setting up MILP Disjunctions/Cuts Management L&P cut generation+strengthening Solving MILP(  )

Anureet Saxena, TSoB25 Computational Results MIPLIB 3.0 instances OR-Lib (Beasley) Capacitated Warehouse Location Problems

Anureet Saxena, TSoB26 MIPLIB 3.0 MIP Instances % Gap Closed

Anureet Saxena, TSoB27 MIPLIB 3.0 MIP Instances % Gap Closed

Anureet Saxena, TSoB28 MIPLIB 3.0 MIP Instances 75-98% Gap Closed Unsolved MIP Instance In MIPLIB 3.0

Anureet Saxena, TSoB29 MIPLIB 3.0 MIP Instances 25-75% Gap Closed

Anureet Saxena, TSoB30 MIPLIB 3.0 MIP Instances 0-25% Gap Closed

Anureet Saxena, TSoB31 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 % 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, TSoB32 MIPLIB 3.0 Pure IP Instances % Gap Closed

Anureet Saxena, TSoB33 MIPLIB 3.0 Pure IP Instances 75-98% Gap Closed

Anureet Saxena, TSoB34 MIPLIB 3.0 Pure IP Instances 25-75% Gap Closed Ceria, Pataki et al closed around 50% of the gap using 10 rounds of L&P cuts

Anureet Saxena, TSoB35 MIPLIB 3.0 Pure IP Instances 0-25% Gap Closed

Anureet Saxena, TSoB36 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 % 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, TSoB37 MIPLIB 3.0 Pure IP Instances % Gap Closed by First Chvatal Closure (Fischetti-Lodi Bound)

Anureet Saxena, TSoB38 MIPLIB 3.0 Pure IP Instances

Anureet Saxena, TSoB39 MIPLIB 3.0 Pure IP Instances

Anureet Saxena, TSoB40 MIPLIB 3.0 Pure IP Instances Comparison of Split Closure vs CG Closure Total Number of Instances: 24 CG closure closes >98% Gap: 9 Results (Remaining 15 Instances) Split Closure closes significantly more gap in 9 instances Both Closures close almost same gap in 6 instances

Anureet Saxena, TSoB41 OrLib CWLP Set 1 –37 Real-World Instances –50 Customers, Warehouses Set 2 –12 Real-World Instances –1000 Customers, 100 Warehouses

Anureet Saxena, TSoB42 OrLib CWLP Set 1 Summary of OrLib CWLP Instances (Set 1) Number of Instances: 37 Number of Instances included: 37 Results 100% Gap closed in 37 instances

Anureet Saxena, TSoB43 OrLib CWLP Set 2 Summary of OrLib CWFL Instances (Set 2) Number of Instances: 12 Number of Instances included: 12 Results >90% Gap closed in 10 instances 85-90% Gap closed in 2 instances Average Gap Closed: 92.82%

Anureet Saxena, TSoB44 Algorithmic Framework Solve Master LP Integral Sol? Unbounded? Infeasible? Rank-1 Split Cut Separation MIP Solved Optimum over Split Closure attained Split Cuts Generated No Split Cuts Generated min cx Ax ¸ b  t x¸  t t2  Yes No Add Cuts

Anureet Saxena, TSoB45 Algorithmic Framework What can one say about the split disjunctions which were used to generate cuts? What are the characteristics of the cuts which are binding at the final optimal solution?

Anureet Saxena, TSoB46 Support Size & Sparsity The support of a split disjunction D( ,  0 ) is the set of non-zero components of   x ·  0  x ¸  (2x 1 + 3x 3 – x 5 · 1) Ç (2x 1 + 3x 3 – x 5 ¸ 2) Support Size = 3

Anureet Saxena, TSoB47 Support Size & Sparsity The support of a split disjunction D( ,  0 ) is the set of non-zero components of  Sparse Split Disjunctions Sparse Split Cuts Computationally Faster Avoid fill-in Disjunctive argument Non-negative row combinations Basis Factorization Sparse Matrix Op

Anureet Saxena, TSoB48 Support Size & Sparsity

Anureet Saxena, TSoB49 Support Size & Sparsity

Anureet Saxena, TSoB50 Support Size & Sparsity Empirical Observation Substantial Duality gap can be closed by using split cuts generated from sparse split disjunctions

Anureet Saxena, TSoB51 Support Coefficients Practice Elementary 0/1 disjunctions Mixed Integer Gomory Cuts Lift-and-project cuts Theory Determinants of sub-matrices Andersen, Cornuejols & Li (’05) Cook, Kannan & Scrhijver (’90) 1 det (B) Huge Gap

Anureet Saxena, TSoB52 Support Coefficients

Anureet Saxena, TSoB53 Support Coefficients

Anureet Saxena, TSoB54 Support Coefficients Empirical Observation Substantial Duality gap can be closed by using split cuts generated from split disjunctions containing small support coefficients.

Anureet Saxena, TSoB55 Cuts Statistics

Anureet Saxena, TSoB56 Number of Cuts Average:

Anureet Saxena, TSoB57 #Cuts/m vs log(n) Average: 45.76%

Anureet Saxena, TSoB58 Average Cut Density vs log(n) Average: 20.82%

Anureet Saxena, TSoB59 Cuts Statistics Internet Checkable Proofs Strengthened formulations for MIPLIB 3.0 instances available at Google Query: anureet saxena (I’m feeling lucky)

Anureet Saxena, TSoB60 arki001 MIPLIB 3.0 & 2003 instance Metallurgical Industry Unsolved for the past 10 years [ ] Problem Stats 1048 Rows 1388 Columns 123 Gen Integer Vars 415 Binary Vars 850 Continuous Vars

Anureet Saxena, TSoB61 Solution Strategy Original Problem Strengthened Formulation Preprocessed Problem CPLEX 9.0 Presolver Rank-1 Split Cut Generation Emphasis on optimality Strong Branching

Anureet Saxena, TSoB62 Strengthening + CPLEX 9.0 Crossover Point (227 rank-1 cuts) Solved to optimality

Anureet Saxena, TSoB63 Strengthening + CPLEX 9.0 arki001 Solution Statistics % Gap closed by rank-1 split cuts: 83.05% Time spent in generating rank-1 split cuts: hrs Time taken by CPLEX 9.0 after strengthening: hrs No. of branch-and-bound nodes enumerated by CPLEX: Total time taken to solve the instance to optimality: hrs

Anureet Saxena, TSoB64 CPLEX 9.0 After 100 hours: 43 million B&B nodes 22 million active nodes 12GB B&B Tree

Anureet Saxena, TSoB65 Comparison Crossover Point

Anureet Saxena, TSoB66 Thank You