MTL103: Optimization Methods and Applications Spring 2023

Lecture Number 1 — January 3, 2023

Lecturer: Prof. Minati De

Scribe:
Arushi Goyal - 2020CS50418
Vidushi Agarwal - 2020CS10410
Supreeti Kumari - 2020CS10396
Goonjan Saha - 2020CS10494
Tushar Gujral - 2020CS10400
Ishaan Govil - 2020CS10497

1 Overview

This is the first lecture of the course MTL103: Optimization Methods and Applications.
In this lecture, we took a look at the course logistics, learned about the linear and convex program-
ming problems, and took a look at few real life examples and applications of problems which can
be re-formulated as linear programming problems.

2 Course Logistics

Textbooks

1. Introduction to Linear Optimization by Dimitris Bertsimas and John Tsitsiklis

2. Nonlinear Programming Theory and Algorithms by Mokhtar S. Bazaraa, Hanif D. Sherali

and C. M. Shetty

Other References

1. Convex Optimization by Stephen Boyd and Lieven Vandenberghe

2. Linear Programming and Network Flows by Mokhtar S. Bazaraa, John J. Jarvis and Hanif
D. Sherali

3. Combinatorial Optimization: Algorithms and Complexity by CH Papadimitriou and K. Stei-

glitz

1
Evaluation

Component Weightage Remarks

Lecture Scribing 5% In teams of 5-6
Assignment 5% In teams of 2
Minor 1 25% -
Minor 2 25% -
Major 40% -

3 Introduction

3.1 Definition

In this course we will learn about methods and algorithms to solve optimisation problems. An
optimisation problem deals with finding values to variables that minimises/maximises a given ex-
pression.
Any optimisation problem can be posed as the following:

Minimize f0 (x)
Subject to fi (x) ≤ bi

Here,

1. x = (x1 , . . . , xn ) is the vector of decision variables or optimisation variables. These are the
variables whose value is to be found such that it minimises the objective function, defined
next.

2. f0 : Rn → R is the objective function, the expression that needs to be minimised. Note that,
any problem of maximising an expression can also posed as a problem for minimising the
negative of that expression.

3. fi : Rn → R are the constraint functions. A solution has to be found to the problem obeying
these constraints.

4. b1 , . . . , bm are limits/bounds of the constraints defined above.

The goal is to find a solution to such an optimisation problem, known as an optimal solution. An
optimal solution can be defined as a vector x∗ such that it satisfies the following properties:

1. It satisfies all the constraints of the optimisation problem.

2. Its objective value is smaller than that of any other vector satisfying the constraints.

Mathematically, for any vector z in Rn such that z ̸= x∗ and fi (z) ≤ bi ∀ i = 1, . . . , m, we must

have f0 (z) ≥ f0 (x∗ ).

2
3.2 Linear & Convex Programming

Till now we discussed about what optimization problems are, intuitively and mathematically. Now
let’s discuss about the classes of optimization problems:

3.2.1 Linear Program

Linear Programming is a class of optimization problems in which the objective and constraint
functions are linear functions. Mathematically, for the optimization problem:

Minimize f0 (x)
Subject to fi (x) ≤ bi

f0 , f1 , ..., fn are linear functions, i.e. they satisfy the following property:

fi (αx + βy) = αfi (x) + βfi (y)

where i ϵ {0, 1, .., n}
∀ α, β ϵ R, ∀ x, y ϵ Rn

Note If any of the objective or constraint functions is non linear then the optimization problem
comes under non linear programming.

3.2.2 Convex Program

Convex Programming is a class of optimization problems in which the objective and constraint
functions are convex functions. Mathematically, for the optimization problem:

Minimize f0 (x), Subject to fi (x) ≤ bi

f0 , f1 , ..., fn are convex functions, i.e. they satisfy the following property:

fi (αx + βy) ≤ αfi (x) + βfi (y)

where i ϵ {0, 1, .., n}, ∀ x, y ϵ Rn
∀ α, β ϵ R with α + β = 1, α ≥ 0, β ≥ 0

Note Convex Optimization is a generalization of Linear Programming since in Linear Programming

the constraint and objective functions are constrained to be linear functions and we know that linear
functions are subset of convex functions as used in convex optimization. Hence the constraints under
which linear programs are optimized fits well into the constraints of convex optimization also.

4 Some simple examples of Optimization In Real Life

4.1 Range Assignment In Wireless Networks

We have been given certain set of base stations as the input. Each base-station is assigned with
a range, and it covers a circular area with the center at that base-station with a radius equal to

3
its assigned range. The objective is to maximize the area covered by these base-stations without
any interference between any 2 base stations. In other words, the area covered by two different
base-stations should not overlap.

Solution:
Now our objective is to formulate the above problem as an optimization problem. Let the number
of base stations be n.
Let P denotes the set of all base stations.
Let us say that ri denotes the radius of the base station i.
Let us say that ρi denotes the position of the base station i and the 2d plane.
Let us say dist(ρi , ρjP
) denotes the distance between base station i and base station j.
Hence we maximize ni=1 (ri )2
Subject to ri + rj ≤ dist(ρi , ρj ), ∀ρi , ρj ∈ P , i ̸= j, and ri ≥ 0 ∀i = 1, 2, ..n.

4.2 Scheduling Problem in Hospital

A hospital wants to make a weekly night shift schedule for its nurses. For night shift on day j the
demand is dj which meand we need to have minimum dj number of nurses and day j, for j=1,2..7.
Every nurse works 5 days in a row on the night shift. Our objective is to find the minimum number
of nurses which the hospital needs to hire to satisfy its demands.

Solution:
Our objective is to hire the minimum possible number of nurses.
Let xj be P
the number of nurses starting their week on day j.
Minimize 7i=1 xi
Subject to:

x1 + x4 + x5 + x6 + x7 ≥ d1
x1 + x2 + x5 + x6 + x7 ≥ d2
x1 + x2 + x3 + x6 + x7 ≥ d3
x1 + x2 + x3 + x4 + x7 ≥ d4
x1 + x2 + x3 + x4 + x5 ≥ d5
x2 + x3 + x4 + x5 + x6 ≥ d6
x3 + x4 + x5 + x6 + x7 ≥ d7
xj ≥ 0, xj is an integer

4.3 Portfolio Optimization

We seek the best way to invest some capital in a set of n assets. We will invest xi amount in ith
asset so the vector x ∈ Rn describes the overall portfolio allocation across assets.

Objective: Our objective is to choose a portfolio allocation that minimizes the risk among all
possible allocations that meet the firm requirement. The constraints represent a limit on the
budget. An alternate objective could be to minimize the variance of portfolio return.

4
4.4 Device Sizing in Electronic Design

We need to choose the width and length of each device in an electronic circuit. Here the variables
are the widths and lengths of devices.

Objective: Our objective is to find the device sizes to obtain maximum power efficiency. The
devices need to satisfy design requirements such as limits on device sizes imposed by manufactur-
ing processes, timing requirements that ensure that the circuit can operate reliably at a specified
speed, and a limit on the total area of the circuit.

4.5 Rocket Control

Consider a rocket that travels along a straight path. Let xt , vt , at be its position, velocity, and
acceleration, respectively, at time t. The rocket needs to start at t = 0 with x0 = 0, v0 = 0 and
land smoothly to the destination at time t = T such that xT = 1, vT = 0.
We have that,

xt+1 = xt + vt
vt+1 = vt + at

The magnitude |at | of acceleration is proportional to the fuel consumption at time t.

Objective: Our objective is to minimize total fuel consumption or minimize maximum thrust
required.

4.6 Choosing Paths in a Communication Network

Consider a communication network consisting of n nodes.

Let A = {(i, j) | There is a link allowing one-way transmission from node i to node j}.
Capacity of link (i, j) is uij bits per second.
Transmission cost along the link (i, j) is cij per bit (cij > 0).
bkl bits per second are generated at node k that need to be transmitted to node l. (direct or indirect
transmission)

Objective: Our objective is to minimize the total cost to transmit all data to their destinations.

5 A non-trivial example

5.1 Multiperiod planning of electric power capacity

For the following T years, a state intends to plan its electrical capacity. The state has a projection
of the demand for power for the years t = 1, ..., T of dt megawatts, which is assumed to be accurate.
The capacity that is currently in use, located in oil-fired facilities, and that will be accessible in
year t is et . Nuclear or coal-fired power stations are the two options for increasing electric capacity.

5
Each megawatt of coal-fired capacity that starts operating at the start of year t has a capital cost
of ct . Nuclear power facilities’ associated capital expense is nt . It has been decided that no more
than 20% of the total capacity should ever be nuclear for a variety of political and safety concerns.
Coal plants last for 20 years, while nuclear plants last for 15 years. A least cost capacity expansion
plan is desired.

Solution:

–> We will follow the following steps inorder to formulate this problem as a linear programming
problem:

1. Define the decision variables.

2. Let xt and yt be the amount of coal and nuclear capacity (respectively) brought on line at the
beginning of year t.
3. Let wt and zt be the total coal and nuclear capacity (respectively) available in year t.
4. The cost of a capacity expansion plan is therefore,

T
X
(ct xt + nt yt )
t=1

This is our objective function and our aim is to minimise it subject to some constraints.
5. We will now work out the constraints based on the problem statement.

1. Since coal-fired plants last for 20 years, we have the following capacity available for coal in
year t:
Pt
wt = s=max{1,t−19} xs , t = 1, ...., T

2. Similarly, since nuclear power plants last for 15 years, we have

Pt
zt = s=max{1,t−14} ys , t = 1, ...., T

3. Since the capacity available must satisfy the anticipated demand, we require

wt + zt + et ≥ dt , t = 1, ...., T

Here, dt is the demand projection for the year t and et is the capacity that is currently in use,
located in oil-fired facilities, which will be accessible in the same year.

4. Finally, since no more than 20% of the total capacity should ever be nuclear, we have
zt
≤ 0.2,
wt + z t + e t

which can be simplified to

0.8zt − 0.2wt ≤ 0.2et

6
Summarizing, the capacity expansion problem is as follows:
T
X
minimize (ct xt + nt yt )
t=1
t
X
given wt − xs = 0,
s=max{1,t−19}
t
X
zt − ys = 0,
s=max{1,t−14}

wt + zt ≥ dt − et ,
0.8zt − 0.2wt ≤ 0.2et ,
xt , yt , wt , zt ≥ 0
where t = 1, ...., T

References

[1] Osama Yaseen M. Al-Rawi (Gulf University), Taniya Mukherjee (Gulf University)
Application of Linear Programming in Optimizing Labour Scheduling, Journal of Mathematical
Finance 09(03):272-285.

[2] Introduction to Linear Optimization by Dimitris Bertsimas and John Tsitsiklis

[3] Lecture slides for the course, MTL103: Optimization Methods and Applications by Prof. Minati
De