Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 0 views

    Branch and Bound Algorithm Implications and Adv Pb

    Uploaded by

    Virat Mundra

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Branch and Bound Algorithm Implications and Adv Pb

    Uploaded by

    Virat Mundra
    0 views19 pages

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    0 views19 pages

    Branch and Bound Algorithm Implications and Adv Pb

    Uploaded by

    Virat Mundra

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 19

    Solving Integer Programming (IP) Problems

    Max 3x1 + 4x2


    s.t.
    5x1 + 3x2  15
    3x1 + 5x2  15
    x1, x2 Integers
    X2

    5 Max 3X1 + 4X2

    s.t.
    4
    5 X1 + 3 X2 <= 15

    3 X1, X2 >=0
    I
    2

    X1
    1 2 3 4 5
    X2

    5 Max 3X1 + 4X2

    s.t.
    4
    5 X1+ 3X2 <= 15
    3X1 + 5X2 <= 15
    3 X1, X2 >=0
    I
    2

    X1
    1 2 3 4 5
    X2

    5 5 X1+ 3X2 <= 15


    3X1 + 5X2 <= 15
    X1, X2 >=0
    4
    X1, X2 Integers

    3
    I
    2

    X1
    1 2 3 4 5
    X2

    5 5 X1+ 3X2 <= 15


    3X1 + 5X2 <= 15
    X1, X2 >=0
    4
    X1, X2 Integers

    3
    I
    2

    X1
    1 2 3 4 5
    X2

    5 5 X1+ 3X2 <= 15


    3X1 + 5X2 <= 15
    X1, X2 >=0
    4
    X1, X2 Integers

    3
    I
    2

    X1
    1 2 3 4 5
    X2

    5 5 X1+ 3X2 <= 15


    3X1 + 5X2 <= 15
    X1, X2 >=0
    4
    X1, X2 Integers

    3
    I
    2

    X1
    1 2 3 4 5
    X2

    5 LP1:
    5 X1+ 3X2 <= 15
    3X1 + 5X2 <= 15
    4
    X1 <= 1

    3 LP2:
    I
    5 X1+ 3X2 <= 15
    2 3X1 + 5X2 <= 15
    X1 >= 2
    1

    X1
    1 2 3 4 5
    Branch-and Bound Algorithm
    Ex: Max problem with general integer variables
    1. Solve the LP-Relaxation (LP0)
    - LP0 Optimal OF value is an upper bound (UB) for the
    Optimal IP OF value.
    - Current lower bound (LB) = -infinity/0

    2. If all variables are integer valued, STOP


    Optimal Solution found ☺

    3. Else branch on one of the non-integer variables

    4. Obtain LP solutions at new nodes generated


    Branch-and-Bound Algorithm Contd.

    5. If no feasible solution found at a Node, Fathom the Node


    6. If integer solution found at a Node, Fathom the Node
    - If OF value > current lower bound (LB), update the lower
    bound and update the candidate for optimality (Incumbent)
    - Fathom all pending nodes with OF Value <= current LB

    6. Else branch on one of the non-integer variables and go to


    Step 4

    7. Continue till all nodes fathomed


    - Optimal Solution: Current Incumbent
    - Optimal OF Value: Current LB
    Criteria for Fathoming a Node

    1. If integer solution found at a Node

    2. If any pending Node has OF value <= current lower bound

    3. If any Node is infeasible

    ➢ STOP when all Nodes are fathomed


    B-a-B Algorithm Considerations

    ➢ Breadth first vs. Depth first

    ➢ Which variable to branch on?

    ➢ Which Node to solve first?

    ➢ Ordering of variables can affect the tree


    B-a-B Algorithm In Practice

    ➢ Stop early. Pick a Tolerance value. If (ub-lb)/ub < Tol, STOP


    ➢ Stop once k solutions found
    ➢ Stop once specified time has elapsed
    ➢ Specify a lower bound if a feasible solution is available
    ➢ Note: Tight formulations are desirable
    - Choose constraints that will result in a smaller (tighter)
    feasible region
    - Add relevant constraints from observation/experience
    Branch-and Bound Algorithm – Minimization Problem

    What if the problem is a Minimization one? Modifications:


    ➢ Solution to LP-Relaxation (Node 0) is a lower bound
    ➢ Integer solution found at any Node is an upper bound
    ➢ If OF value < current upper bound(UB), update the candidate
    for optimality (incumbent) and update the current UB
    ➢ If any pending node has OF value >= upper bound, Fathom
    Formulating Integer Programming Problems
    1. Add constraints that reduce feasible options

    Example: Advertisement Problem

    Min 200C1 + 180C2 + 230C3 + 15000M1 + 20000M2 + 25000M3


    s.t.
    300C2 +250C2 + 500C3 >= 25000 (Min Reach)
    C1 <= 30*M1 (Max Length)
    C2 <= 30*M2 (Max Length)
    C3 <= 30*M3 (Max Length)
    C1, C2, C3 >= 0 and M1, M2, M3 binary
    Formulating Integer Programming Problems

    2. Add ‘tight’ constraints that result in a smaller feasible region for


    the LP Relaxation of the IP

    Example: Can take Project 1 only if both Projects 3 and 4 are taken
    Solving Integer Programming Problems - Preprocessing
    - Very useful to identify variables whose values can be fixed (at
    one of their possible values).
    - Such variables can then be eliminated, reducing the complexity
    of the problem to be solved.
    - Which variable(s) value can be fixed and eliminated in the
    formulation below?

    Example: Consider the following binary integer programming


    (BIP) problem:

    Max 5x1 + 3x2 + 4x3 + 2x4


    s.t. 4x1 + x2 + x3 + 2x4 <= 3 (1)
    x1 + 3x2 + x3 + 3x4 <= 5 (2)
    2x1 + x2 - 2x3 + 4x4 >= 4 (3)
    x1, x2, x3, x4 binary
    Solving Integer Programming (Mixed BIP) Problems -
    Advertisement Decisions

    Min 200C1 + 180C2 + 230C3 + 15000M1 + 20000M2 + 25000M3


    s.t.
    300C2 +250C2 + 500C3 >= 25000 (Min Reach)
    C1 <= 30*M1 (Max Length)
    C2 <= 30*M2 (Max Length)
    C3 <= 30*M3 (Max Length)
    C1, C2, C3 >= 0 and M1, M2, M3 binary
    Solving Integer Programming (Mixed BIP)
    Problems (Advertisement Decisions)
    43566.67
    LP (0.556, 0 , 1)
    0 (16.667, 0 , 30)

    M1 = 0 M1 = 1

    48833.33 LP 44393.33
    LP
    (0, 0.667, 1) 1 2 (1, 0, 0.733)
    (1, 20, 30) (30, 0, 22)
    M2 = 0 M2 = 1
    M3 = M3 = 1
    *Node 6 Fathomed: 0
    OF value > OF value of incumbent
    (integer Soln) at Node 4
    50233.33
    LP 51983.3 LP LP LP (1,0,1)
    5 (0,1,0.833) 6 3 4 (16.67,0,30)
    (0,30,25)
    Infeasible- Fatho Infeasible-
    fathom m* fathom Integer Soln-
    Fathom

    You might also like