LINEAR, INTEGER, NONLINEAR PROGRAMMING - PowerPoint PPT Presentation

1 / 34
About This Presentation
Title:

LINEAR, INTEGER, NONLINEAR PROGRAMMING

Description:

Create two new problems: x6 15, x6 16. Solve the new problems ... Gap between UB and LB in IP (Start high, then reduce) A few tips/reminders: ... – PowerPoint PPT presentation

Number of Views:106
Avg rating:3.0/5.0
Slides: 35
Provided by: erhan1
Category:

less

Transcript and Presenter's Notes

Title: LINEAR, INTEGER, NONLINEAR PROGRAMMING


1
LINEAR, INTEGER, NONLINEAR PROGRAMMING
  • A practitioners perspective

2
OPTIMIZATION
  • min f(x)s.t. G(x) b.

3
CLASSIFICATION
  • Decision Variables
  • Functions Fractional Integer
  • Linear LP ILP
  • Nonlinear NLP INLP

4
LP
  • SIMPLEX method (linear algebra)
  • Corner point optimality
  • Move from corner-to-corner, improve f(x)
  • Very efficient
  • Can solve problems with thousands of variables
    and constraints

5
(No Transcript)
6
ILP
  • BranchBound (divide-and-conquer)
  • Solve the LP, ignoring the integrality
  • Select a fractional variable, x6 15.7
  • Create two new problems x6 lt 15, x6 gt 16
  • Solve the new problems
  • Continue until all branches exhausted
  • No of branches is exponential in no. of var.

7
(No Transcript)
8
NLP
  • Gradient method (uses derivatives)
  • Repeat until convergence
  • Find an improving direction
  • Move in the improving direction
  • Can converge to a local optimum
  • Multiple starts recommended

9
(No Transcript)
10
INLP
  • Ignore integrality, solve the NLP
  • Use Branch Bound
  • Solve a series of NLPs
  • Computationally demanding
  • No guarantee of optimality
  • YUCK!

11
(No Transcript)
12
  • Now that we have reduced half of an MSc in
    OR program to 10 slides, we can move on to our
    regularly scheduled program.

13
Optimization on Spreadsheets
  • Erhan Erkut
  • Faculty of Business, University of Alberta
  • February 2, 1999
  • Session 4, HEC-Montreal

14
Spreadsheet solver
  • Modeling language
  • ... with graphical user interface

15
First solver
  • Whats Best? (Lotus 1-2-3 add on)
  • Birth date 1985
  • Father Sam Savage

16
Excel solver
  • Birth date 1991 (Excel 3.0)
  • Producer Frontline Systems
  • Current Population 35M
  • Optimizers
  • Linear simplex w. bounds
  • Nonlinear GRG
  • Integer BB

17
How large a problem?
  • Standard 200 variables
  • Premium 800
  • Platform 16,000

18
Speed?
  • Excel 3.0
  • 5x
  • Excel 5.0
  • 20x
  • Excel 97
  • 25x
  • Premium

19
Bugs?
  • Bugs in earlier versions
  • Frontline offers 500 for a new bug discovery
  • Two of our students received the prize...

20
No competition for CPLEX,...
  • but real-world problems are within reach.

21
Frontline says
  • 7M/year savings via a 200-variable application
    (Fortune 50 company)
  • No. of applications with lt 200 variables
    (5-10) x (No. of larger applications)
  • OR/MS Pros. are involved in lt 1 of all solver
    applications in use

22
Excel 97 advertising
  • Auto scaling applies to LP
  • Can set convergence tolerance for IP
  • Improved linearity tests
  • Direct input of binary variables (bin)
  • Much faster

23
Premium advertising
  • Altius, fortius, celsius
  • Linearity and feasibility reports
  • Standard IP breadth-first search
  • Premium IP
  • depth-first followed by breadth-first
  • sophisticated selection rules
  • dual simplex

24
The solver is..
  • not designed for the OR Pros,
  • but created with the spreadsheet user in mind.
  • wider use
  • - wider abuse

25
Some of the abuses
  • Using IF instead of 0/1 variables
  • Not reading the dialogue boxes
  • Not scaling the problem
  • Using a high tolerance for IP
  • Not keeping everything on one sheet
  • Finding local optimum for NLP
  • f(x)/g(x) lt c
  • Assume Linear Model (Y/N)
  • 2 tons

26
A few tips/reminders
  • 1) Use names, color, layout to deal with
    documentability problem

27
A few tips/reminders
  • 2) Use auto scaling when necessary
  • 15000A 20000B lt 160000 (1)
  • 3A 4B lt 8 (2)
  • Scaling Divide (1) by 10,000

28
A few tips/reminders
  • 3) Precision
  • a b means...
  • b ? gt a gt b ?,
  • where ? precision

29
A few tips/reminders
  • 4) Tolerance
  • Gap between UB and LB in IP
  • (Start high, then reduce)

30
A few tips/reminders
  • 5) NLP multiple runs with different starting
    points (if OF not convex)

31
A few tips/reminders
  • 6) Dual variables are formatted the same way as
    variable and constraint values.

32
A few curiosity items
  • Set target cell equal to ??
  • Leftover from Goal Seek

33
A few curiosity items
  • min (2SUMPRODUCT)
  • min (SUMPRODUCT2)
  • 2(min SUMPRODUCT)
  • Whats the difference?

34
A few curiosity items
  • Limits report
  • (What is the max/min each variable can take?)
  • Why have one?
  • cause Lotus 1-2-3 had one...
Write a Comment
User Comments (0)
About PowerShow.com