ISE 426
Optimization models and applications
instructor:
phone:
office hrs:
email:
web page:
Katya Scheinberg
Mohler Lab #479
(610) 758 4039
after 5pm on Tuesdays (if I am available)
and by appointment
katyas at lehigh dot edu
http://coral.ie.lehigh.edu/ katyas
Introduction
What do I expect from this course?
Homeworks, TAs (Sahar Tahernejad and Secil Sozuer) and
lab time.
Use of cell phones and laptops.
How to take lecture notes.
Introductions.
A few practical issues.
Evaluation
Homework:
Quiz #1:
Quiz #2:
Case study:
Final exam:
In-class participation:
25%
10%
10%
20%
25%
10%
(one every two weeks?)
(first part of October)
(early November)
(assigned in November)
Case studies
study an Optimization problem,
propose a model, and
solve it using a tool1 of their choice.
An optimization tool. . .
Lecture plan
Convexity, relaxations, lower/upper bounds
Linear Programming (+ duality)
Integer Programming
Nonlinear Programming
Relaxation and Decomposition
Stochastic Programming
Robust Optimization
Multicriteria Optimization
Material
Introduction to Mathematical Programming:
Applications and Algorithms, Volume 1, by W.L. Winston
and M. Venkataramanan;
Select chapters of Introduction to Operations Research
by F.S. Hillier and G.J. Lieberman, McGraw-Hill: New
York, NY, 1990;
Select chapters of Operations Research: Applications and
Algorithms by Wayne L. Winston, PWS-Kent Pub. Co.,
1991;
modeling language: AMPL: A Modeling Language for
Mathematical Programming by Robert Fourer, David M.
Gay, and Brian W. Kernighan.
Lecture slides.
Modeling languages
They are similar, and each has its own pros/cons. All have
limited version available to students.
Mosel: very nice Graphical User Interface (GUI)
AMPL: preferred. No GUI, but I and TAs know it better (can help)
also helps the formulation process.
GAMS: Has version with even nicer GUI (Aimms)
What is optimization?
When to use it?
What to use it for?
How to use it?
Optimization models. . .
Optimization aims at finding the best configuration of
processes, systems, products, etc.
It relies on a theory developed mostly in the past 50 years
Applying Optimization in an industrial, financial, logistic
context yields a better use of budget/resources ($$$) or a
higher revenue ($$$)
. . . and applications
Source: http://www.informs.com
(see also http://www.ScienceOfBetter.org)
yr
86
90
92
95
96
99
00
company
Eletrobras (hydroelectric energy)
Taco Bell (human resources)
Harris semicond. prod. planning
GM Car Rental
HP printers re-designed prod.
IBM supply chain
Syngenta corn production
result
43M$ saved
7.6M$ saved
50% 95% orders on time
+50M$
2x production
750M$ saved
5M$ saved
A few examples from IBM Research
Circuit design
Sorting facility optimization
Limousine driver/fleet optimization
An example
You work at a company that sells food in glass containers
R
. Its
only. Today, your boss has a bright idea! The tin can
a cylinder made of tin.
The can must contain V = 20 cu.in. (11 fl.oz., 33 cl)
Cut and solder tin foil to produce cans
Tin (foil) is expensive, use as little as possible
Boss: What is the ideal can? Tall and thin or short and fat?
You: A cylinder with volume V using as little tin as possible.
Example
r
If we knew radius r and height h,
h
the volume would be r2 h
qty of tin would be 2r2 + 2rh
r2 h must be V = 20in3 h =
Rewrite the quantity of tin as Q(r) = 2r2 + 2r rV2 , or
Q(r) = 2r2 +
Find the minimum of Q(r)!
2V
r
V
r2
Minimize the quantity of tin
Q(r)[in2 ]
250
200
150
100
50
minimum
0
r = 1.471 in
h=
V
(1.471)2
= 2.942 in
r [in]
Aims of this course
model Optimization problems
so that they can be solved
learn a modeling language
apply modeling languages to real-world problems
Your first Optimization model
Variables
Objective
Constraints
r: radius of the cans base
h: height of the can
2rh + 2r2 (minimize)
r2 h = V
Your first Optimization model
Variables
Objective
Constraints
r: radius of the cans base
h: height of the can
2rh + 2r2 (minimize)
r2 h = V
h>0
r>0
Optimization Models, in general, have:
Variables: Height and radius, number of trucks, . . . The
unknown (and desired) part of the problem (one thing your
boss cares about).
Constraints: Physical, explicit (V = 20in3 ), imposed by law,
budget limits. . . They define all and only values of the
variables that give possible solutions.
Objective function: what the boss really cares about.
Quantity of tin, total cost of trucks, total estimated
revenue, . . . a function of the variables
The general optimization problem
Consider a vector x Rn of variables.
An optimization problem can be expressed as:
P:
minimize f0 (x)
such that f1 (x) b1
f2 (x) b2
..
.
fm (x) bm
Convexity
Convex sets
Def.: A set S Rn is convex if any two points x and x of S are
joined by a segment entirely contained in S:
Convex sets
Def.: A set S Rn is convex if any two points x and x of S are
joined by a segment entirely contained in S:
x , x S, [0, 1],
x + (1 )x S
Convex sets
Def.: A set S Rn is convex if any two points x and x of S are
joined by a segment entirely contained in S:
x , x S, [0, 1],
x + (1 )x S
The intersection of two convex sets is convex.
Examples: Convex sets
Convex
Examples: Convex sets
Convex
R, R2 , R3 , etc. are convex
[a, b] is convex
{4} is convex
Examples: Nonconvex sets
Examples: Nonconvex sets
x
x
{0, 1} is nonconvex
{x R : x 2 x 3} is nonconvex
Z is nonconvex
Examples: Nonconvex sets
x
x
x
x
{0, 1} is nonconvex
{x R : x 2 x 3} is nonconvex
Z is nonconvex
Examples: Nonconvex sets
x
x
x
x
x
x
{0, 1} is nonconvex
{x R : x 2 x 3} is nonconvex
Z is nonconvex
Convex functions
Def.: A function f : Rn R is convex if, for any two points x
and x Rn and for any [0, 1]
f (x + (1 )x )
f (x ) + (1 )f (x )
Convex functions
Def.: A function f : Rn R is convex if, for any two points x
and x Rn and for any [0, 1]
f (x + (1 )x )
f (x ) + (1 )f (x )
The sum of convex functions is a convex function
Convex functions
Def.: A function f : Rn R is convex if, for any two points x
and x Rn and for any [0, 1]
f (x + (1 )x )
f (x ) + (1 )f (x )
The sum of convex functions is a convex function
Multiplying a convex function by a positive scalar gives a
convex function
Convex functions
Def.: A function f : Rn R is convex if, for any two points x
and x Rn and for any [0, 1]
f (x + (1 )x )
f (x ) + (1 )f (x )
The sum of convex functions is a convex function
Multiplying a convex function by a positive scalar gives a
convex function
P
linear functions ki=1 ai xi are convex, irrespective of the
sign of ai s.
