Integer and Combinatorial Optimization: Introduction

Similar documents
Math Models of OR: Branch-and-Bound

Lecture 23 Branch-and-Bound Algorithm. November 3, 2009

Math Models of OR: Sensitivity Analysis

Linear integer programming and its application

The Ellipsoid Algorithm

Math Models of OR: Extreme Points and Basic Feasible Solutions

Structured Problems and Algorithms

LP Relaxations of Mixed Integer Programs

Travelling Salesman Problem

Network Flows. 6. Lagrangian Relaxation. Programming. Fall 2010 Instructor: Dr. Masoud Yaghini

Part 4. Decomposition Algorithms

Chapter 6 Constraint Satisfaction Problems

Integer Programming. Wolfram Wiesemann. December 6, 2007

Section Notes 8. Integer Programming II. Applied Math 121. Week of April 5, expand your knowledge of big M s and logical constraints.

Learning to Branch. Ellen Vitercik. Joint work with Nina Balcan, Travis Dick, and Tuomas Sandholm. Published in ICML 2018

AM 121: Intro to Optimization! Models and Methods! Fall 2018!

Geometric Steiner Trees

Math Models of OR: Traveling Salesman Problem

Math Models of OR: Some Definitions

Introduction to integer programming II

23. Cutting planes and branch & bound

Math Models of OR: Handling Upper Bounds in Simplex

LMI Methods in Optimal and Robust Control

1 The linear algebra of linear programs (March 15 and 22, 2015)

Module 04 Optimization Problems KKT Conditions & Solvers

Semidefinite Programming Basics and Applications

Section #2: Linear and Integer Programming

Column Generation. i = 1,, 255;

Integer Programming Part II

Week Cuts, Branch & Bound, and Lagrangean Relaxation

Basic notions of Mixed Integer Non-Linear Programming

to work with) can be solved by solving their LP relaxations with the Simplex method I Cutting plane algorithms, e.g., Gomory s fractional cutting

1 Strict local optimality in unconstrained optimization

Appendix A Taylor Approximations and Definite Matrices

The final is cumulative, but with more emphasis on chapters 3 and 4. There will be two parts.

3.3 Easy ILP problems and totally unimodular matrices

Dual bounds: can t get any better than...

4y Springer NONLINEAR INTEGER PROGRAMMING

APPLIED MECHANISM DESIGN FOR SOCIAL GOOD

MIT Algebraic techniques and semidefinite optimization February 14, Lecture 3

Decision Procedures An Algorithmic Point of View

16.1 L.P. Duality Applied to the Minimax Theorem

Linear Programming: Simplex

Conic optimization under combinatorial sparsity constraints

On the Power of Robust Solutions in Two-Stage Stochastic and Adaptive Optimization Problems

Computational and Statistical Tradeoffs via Convex Relaxation

Separation Techniques for Constrained Nonlinear 0 1 Programming

where X is the feasible region, i.e., the set of the feasible solutions.

Bounds on the Traveling Salesman Problem

Nonconvex Quadratic Programming: Return of the Boolean Quadric Polytope

Introduction to Mathematical Programming IE406. Lecture 21. Dr. Ted Ralphs

Acyclic Semidefinite Approximations of Quadratically Constrained Quadratic Programs

Chapter 3 Deterministic planning

Intractable Problems Part Two

Discrete (and Continuous) Optimization WI4 131

CS675: Convex and Combinatorial Optimization Fall 2016 Combinatorial Problems as Linear and Convex Programs. Instructor: Shaddin Dughmi

Operations Research Lecture 6: Integer Programming

Approximation Algorithms

IP Cut Homework from J and B Chapter 9: 14, 15, 16, 23, 24, You wish to solve the IP below with a cutting plane technique.

16.410/413 Principles of Autonomy and Decision Making

Data Mining and Analysis: Fundamental Concepts and Algorithms

POLYNOMIAL OPTIMIZATION WITH SUMS-OF-SQUARES INTERPOLANTS

15.083J/6.859J Integer Optimization. Lecture 2: Efficient Algorithms and Computational Complexity

Relations between Semidefinite, Copositive, Semi-infinite and Integer Programming

CS675: Convex and Combinatorial Optimization Fall 2014 Combinatorial Problems as Linear Programs. Instructor: Shaddin Dughmi

Section Notes 9. IP: Cutting Planes. Applied Math 121. Week of April 12, 2010

From structures to heuristics to global solvers

Math Models of OR: The Revised Simplex Method

Multi-objective branch-and-cut algorithm and multi-modal traveling salesman problem

Constraint Satisfaction Problems

1 Column Generation and the Cutting Stock Problem

NP-Completeness I. Lecture Overview Introduction: Reduction and Expressiveness

Bilevel Integer Optimization: Theory and Algorithms

Principal Components Analysis (PCA)

Limitations of Algorithm Power

Multivalued Decision Diagrams. Postoptimality Analysis Using. J. N. Hooker. Tarik Hadzic. Cork Constraint Computation Centre

On the Computational Hardness of Graph Coloring

Branch-and-Bound. Leo Liberti. LIX, École Polytechnique, France. INF , Lecture p. 1

Complexity and Simplicity of Optimization Problems

The Class NP. NP is the problems that can be solved in polynomial time by a nondeterministic machine.

NP-problems continued

Integer Quadratic Programming is in NP

PART 4 INTEGER PROGRAMMING

Reconnect 04 Introduction to Integer Programming

4-1 The Algebraic-Geometric Dictionary P. Parrilo and S. Lall, ECC

CHAPTER 3: INTEGER PROGRAMMING

Copositive Programming and Combinatorial Optimization

Online generation via offline selection - Low dimensional linear cuts from QP SDP relaxation -

MDD-based Postoptimality Analysis for Mixed-integer Programs

Stochastic Integer Programming

Lecture 6: Conic Optimization September 8

A new family of facet defining inequalities for the maximum edge-weighted clique problem

Integer Programming and Branch and Bound

ELE539A: Optimization of Communication Systems Lecture 16: Pareto Optimization and Nonconvex Optimization

Lecture 2. MATH3220 Operations Research and Logistics Jan. 8, Pan Li The Chinese University of Hong Kong. Integer Programming Formulations

MATHEMATICS. Units Topics Marks I Relations and Functions 10

Integer Linear Programs

A A x i x j i j (i, j) (j, i) Let. Compute the value of for and

4/12/2011. Chapter 8. NP and Computational Intractability. Directed Hamiltonian Cycle. Traveling Salesman Problem. Directed Hamiltonian Cycle

Development of an algorithm for solving mixed integer and nonconvex problems arising in electrical supply networks

Transcription:

Integer and Combinatorial Optimization: Introduction John E. Mitchell Department of Mathematical Sciences RPI, Troy, NY 12180 USA November 2018 Mitchell Introduction 1 / 18

Integer and Combinatorial Optimization Outline 1 Integer and Combinatorial Optimization Complexity Solution techniques Quadratic constraints and objective functions 2 Cutting planes 3 Branch-and-bound: An example Mitchell Introduction 2 / 18

Integer and Combinatorial Optimization Combinatorial optimization An optimization problem is a problem of the form min x subject to f (x) x 2 S where f (x) is the objective function and S is the feasible region. A combinatorial optimization problem is one where there is only a finite number of points in S. For example: colorings of the countries of a map, so that no two adjacent coutries have the same color. Mitchell Introduction 3 / 18

Integer and Combinatorial Optimization Integer optimization In an integer optimization problem, the variables are constrained to take integer values. In a mixed integer optimization problem, some of the variables are required to take integer values, and the remainder can take fractional values. Mitchell Introduction 4 / 18

Integer and Combinatorial Optimization Integer optimization For example, a mixed integer linear optimization problem has the form min x2r n,y2r p ct x + d T y subject to Ax + By apple g x, y 0, y integer Here, A and B are given matrices, A 2 R m n, B 2 R m p. The remaining parameters c, d, g are vectors, c 2 R n, d 2 R p, g 2 R m. All vectors are understood to be column vectors in this course. Transpose is denoted by the superscript T. The dot product between two vectors c and x can be denoted c T x. In a mixed integer nonlinear optimization problem, the objective function and/or the constraints may be nonlinear functions. Mitchell Introduction 5 / 18

Integer and Combinatorial Optimization Complexity Outline 1 Integer and Combinatorial Optimization Complexity Solution techniques Quadratic constraints and objective functions 2 Cutting planes 3 Branch-and-bound: An example Mitchell Introduction 6 / 18

Integer and Combinatorial Optimization Complexity Computational complexity Linear optimization problems can be solved in polynomial time. This implies the required runtime doesn t grow too quickly as the problem size grows. By contrast, integer optimization problems are often NP-hard (to be defined later), which means that the runtime can grow very quickly in the problem size in the worst case. So mixed integer linear optimization problems are a lot harder to solve to global optimality than linear optimization problems. Mitchell Introduction 7 / 18

Integer and Combinatorial Optimization Solution techniques Outline 1 Integer and Combinatorial Optimization Complexity Solution techniques Quadratic constraints and objective functions 2 Cutting planes 3 Branch-and-bound: An example Mitchell Introduction 8 / 18

Integer and Combinatorial Optimization Solution techniques Solution techniques Our emphasis will be on methods for finding global optimal solutions to integer optimization problems. If we have a global minimizer x then there is no other feasible point x 2 S with f (x) < f (x ). In order to prove optimality, we will look at relaxations. The simplest relaxation of a mixed integer linear optimization problem is to ignore integrality, giving a linear optimization problem. The principal solution techniques we then examine are cutting planes: try to improve the relaxation branching: subdivide the feasible region. Mitchell Introduction 9 / 18

Integer and Combinatorial Optimization Solution techniques Solution techniques Our emphasis will be on methods for finding global optimal solutions to integer optimization problems. If we have a global minimizer x then there is no other feasible point x 2 S with f (x) < f (x ). In order to prove optimality, we will look at relaxations. The simplest relaxation of a mixed integer linear optimization problem is to ignore integrality, giving a linear optimization problem. The principal solution techniques we then examine are cutting planes: try to improve the relaxation branching: subdivide the feasible region. Mitchell Introduction 9 / 18

Integer and Combinatorial Optimization Quadratic constraints and objective functions Outline 1 Integer and Combinatorial Optimization Complexity Solution techniques Quadratic constraints and objective functions 2 Cutting planes 3 Branch-and-bound: An example Mitchell Introduction 10 / 18

Integer and Combinatorial Optimization Quadratic constraints and objective functions Quadratic constraints and objective functions f (x) = I x t I x t Qx Some combinatorial optimization problems can be formulated with quadratic constraints and/or a quadratic objective function. For other problems, it can be helpful to look at a quadratic relaxation. These problems can sometimes be attacked as semidefinite optimization problems, which have a matrix of variables that must be symmetric and positive semidefinite. A useful observation: we can replace the requirement that x is a binary variable by the requirement that x = x 2. :::::::::: -1 Mitchell Introduction 11 / 18

Cutting planes Outline 1 Integer and Combinatorial Optimization Complexity Solution techniques Quadratic constraints and objective functions 2 Cutting planes 3 Branch-and-bound: An example Mitchell Introduction 12 / 18

Cutting planes A cutting plane example Initial relaxation: Solve linear optimization relaxation min{c T x : Ax apple b, x 0} - c Solution to linear optimization relaxation Mitchell Introduction 13 / 18

Cutting planes A cutting plane example Initial relaxation: Solve linear optimization relaxation min{c T x : Ax apple b, x 0} Cutting plane a T 1 x = b 1 Solution to linear optimization relaxation Mitchell Introduction 13 / 18

Cutting planes Add a cutting plane and reoptimize Solve linear optimization relaxation min{c T x : Ax apple b, x 0, a T 1 x apple b 1} Cutting plane a T 1 x = b 1 Solution to linear optimization relaxation Mitchell Introduction 14 / 18

Cutting planes Add a cutting plane and reoptimize Solve linear optimization relaxation min{c T x : Ax apple b, x 0, a T 1 x apple b 1} Cutting plane a T 1 x = b 1 Solution to linear optimization relaxation Cutting plane a T 2 x = b 2 Mitchell Introduction 14 / 18

Cutting planes Add a cutting plane and reoptimize Solve linear optimization relaxation min{c T x : Ax apple b, x 0, a T 1 x apple b 1, a T 2 x apple b 2} Cutting plane a T 1 x = b 1 Solution to linear optimization relaxation Cutting plane a T 2 x = b 2 Solution to relaxation is integral, so it solves the integer optimization problem. Mitchell Introduction 15 / 18

Branch-and-bound: An example Outline 1 Integer and Combinatorial Optimization Complexity Solution techniques Quadratic constraints and objective functions 2 Cutting planes 3 Branch-and-bound: An example Mitchell Introduction 16 / 18

An example Branch-and-bound: An example We look at the integer optimization problem min x2r 3 5x 1 + 5x 2 13x 3 subject to x 1 + x 2 + x 3 apple 6 10x 1 8x 2 apple 15 (IOP) 6x 1 x 2 + 9x 3 apple 9 x i 0, x i integer, i = 1, 2, 3 It s easy to find a feasible solution for this problem: x =(0, 0, 0). Hence we can initialize with z u = 0. We let F = {x 2 R 3 : x 1 + x 2 + x 3 apple 6, 10x 1 8x 2 apple 15, 6x 1 x 2 + 9x 3 apple 9, x 1, x 2, x 3 0}. Mitchell Introduction 17 / 18

Branch-and-bound: An example The tree for the example Root node: min{c T x : x 2 F } x 0 =(1.5, 0, 2), z 0 = 18.5 Mitchell Introduction 18 / 18

Branch-and-bound: An example The tree for the example Root node: min{c T x : x 2 F } x 0 =(1.5, 0, 2), z 0 = 18.5 x 1 apple 1 x 1 2 x 2 =(2, 5 8, 2 29 72 ), z 2 = 18 1 9 Mitchell Introduction 18 / 18

Branch-and-bound: An example The tree for the example x 1 =(1, 0, 1 2 3 ), z 1 = 16 2 3 Root node: min{c T x : x 2 F } x 0 =(1.5, 0, 2), z 0 = 18.5 x 1 apple 1 x 1 2 x 2 =(2, 5 8, 2 29 72 ), z 2 = 18 1 9 Mitchell Introduction 18 / 18

Branch-and-bound: An example The tree for the example x 1 =(1, 0, 1 2 3 ), z 1 = 16 2 3 Root node: min{c T x : x 2 F} x 0 =(1.5, 0, 2), z 0 = 18.5 x 1 apple 1 x 1 2 x 2 =(2, 5 8, 2 29 72 ), z 2 = 18 1 9 x 3 3 infeasible, z 6 =+1 fathom by infeasibility Mitchell Introduction 18 / 18

Branch-and-bound: An example The tree for the example x 1 =(1, 0, 1 2 3 ), z 1 = 16 2 3 Root node: min{c T x : x 2 F} x 0 =(1.5, 0, 2), z 0 = 18.5 x 1 apple 1 x 1 2 x 5 =(2, 5 8, 2), z 5 = 12 7 8 x 2 =(2, 5 8, 2 29 72 ), z 2 = 18 1 9 x 3 apple 2 x 3 3 infeasible, z 6 =+1 fathom by infeasibility Mitchell Introduction 18 / 18

Branch-and-bound: An example The tree for the example x 1 =(1, 0, 1 2 3 ), z 1 = 16 2 3 Root node: min{c T x : x 2 F } x 0 =(1.5, 0, 2), z 0 = 18.5 x 1 apple 1 *. x 1 2 x 2 =(2, 5 8, 2 29 72 ), z 2 = 18 1 9 x 3 apple 1 x 3 apple 2 x 3 3 x 3 =(0, 0, 1), z 3 = 13 fathom by integrality x 5 =(2, 5 8, 2), z 5 = 12 7 8 fathom by bounds infeasible, z 6 =+1 fathom by infeasibility Mitchell Introduction 18 / 18

Branch-and-bound: An example The tree for the example Root node: min{c T x : x 2 F } x 1 =(1, 0, 1 2 3 ), z 1 = 16 2 3 x 0 =(1.5, 0, 2), z 0 = 18.5 x 1 apple 1 x 1 2 x 2 =(2, 5 8, 2 29 72 ), z 2 = 18 1 9 x 3 apple 1 x 3 2 x 3 apple 2 x 3 3 x 3 =(0, 0, 1), z 3 = 13 x 4 =(1, 3, 2), z 4 = 6 x 5 =(2, 5 8, 2), z 5 = 12 7 8 infeasible, z 6 =+1 fathom by integrality fathom by bounds fathom by bounds fathom by infeasibility Mitchell Introduction 18 / 18

Branch-and-bound: An example The tree for the example Root node: min{c T x : x 2 F } x 1 =(1, 0, 1 2 3 ), z 1 = 16 2 3 x 0 =(1.5, 0, 2), z 0 = 18.5 x 1 apple 1 x 1 2 x 2 =(2, 5 8, 2 29 72 ), z 2 = 18 1 9 x 3 apple 1 x 3 2 x 3 apple 2 x 3 3 x 3 =(0, 0, 1), z 3 = 13 x 4 =(1, 3, 2), z 4 = 6 x 5 =(2, 5 8, 2), z 5 = 12 7 8 infeasible, z 6 =+1 fathom by integrality fathom by bounds fathom by bounds fathom by infeasibility The optimal solution to (IOP) is x =(0, 0, 1), with value z = 13. Mitchell Introduction 18 / 18