Scribd
Upload
0 views

Optimization_Problems

The document discusses optimization problems, specifically focusing on a piecewise function f(z) and its graphical representation. It outlines the transformation of a minimization problem into a linear programming format by introducing auxiliary variables and constraints. Additionally, it details the derivation of the dual problem using Lagrangian formulation, including the necessary constraints and the dual objective function.

Uploaded by

daokhanhlinh0523
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0 views

Optimization_Problems

The document discusses optimization problems, specifically focusing on a piecewise function f(z) and its graphical representation. It outlines the transformation of a minimization problem into a linear programming format by introducing auxiliary variables and constraints. Additionally, it details the derivation of the dual problem using Lagrangian formulation, including the necessary constraints and the dual objective function.

Uploaded by

daokhanhlinh0523
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Optimization Problems

from: Lecturer: Vuong Van yen


2023

(a) Graph and formula for f (z) The function f (z) is defined piecewise:
(
0, if |z| ≤ 1,
f (z) =
|z| − 1, if |z| > 1.
To graph f (z):
• For |z| ≤ 1, the function is constant, equal to 0.
• For |z| > 1, the function grows linearly as |z| − 1.
The given hint suggests expressing f (z) as f (z) = max{0, z − 1, −z − 1}, which simplifies to:
f (z) = max{(z − 1), 0, (−z − 1)}.
Graphing f (z) involves plotting:
• A flat line at 0 between −1 ≤ z ≤ 1,
• A line of slope 1 starting from z = 1 for z > 1,
• A line of slope -1 starting from z = −1 for z < −1.
(b) Express the problem as linear programming The original minimization problem is:
m
X
min f (aTi x − bi ),
x
i=1

where f (z) is defined above. To express this as a linear programming problem, we need to replace the
absolute value and the max with linear inequalities.
We introduce auxiliary variables ti ≥ 0 to handle the linear nature of f (z). The constraints to ensure
ti captures the behavior of f (aTi x − bi ) are:
ti ≥ aTi x − bi − 1,
ti ≥ −(aTi x − bi ) − 1,
ti ≥ 0.
Thus, the linear programming problem becomes:
m
X
min ti ,
x
i=1

subject to the constraints mentioned above.

(c) Deriving the Dual Problem


We begin by expressing the primal linear programming problem as:
m
X
min ti
x
i=1
subject to:
ti ≥ aTi x − bi − 1,
ti ≥ −(aTi x − bi ) − 1,
ti ≥ 0.

1
Lagrangian Formulation
We introduce Lagrange multipliers λi ≥ 0 and µi ≥ 0 for the two inequalities, and νi ≥ 0 for the
non-negativity condition ti ≥ 0. The Lagrangian is:

m
X m
X m
X m
X
L(x, t, λ, µ, ν) = ti + λi (aTi x − bi − 1 − ti ) + µi (−(aTi x − bi ) − 1 − ti ) + νi (−ti ).
i=1 i=1 i=1 i=1

Simplifying the Lagrangian:


m
X
ti (1 − λi − µi − νi ) + λi (aTi x − bi − 1) + µi (−(aTi x − bi ) − 1) .
 
L(x, t, λ, µ, ν) =
i=1

Dual Problem
To derive the dual problem, we minimize the Lagrangian with respect to the primal variables ti and x.

1. Minimization with respect to ti :


The coefficient of ti in the Lagrangian is 1 − λi − µi − νi . For the Lagrangian to be minimized, this
coefficient must be zero:

1 − λi − µi − νi = 0.
Thus, we obtain the constraint:

λi + µi + νi = 1.

2. Minimization with respect to x:


Focusing on the terms involving x, we have:
m
X
λi aTi x − µi aTi x.
i=1

To maximize the dual, the coefficients of x must sum to zero, leading to:
m
X
(λi − µi )aTi = 0.
i=1

This gives the dual constraint:


m
X
(λi − µi )ai = 0.
i=1

3. Dual Objective Function:


The remaining terms in the Lagrangian provide the dual objective function:
m
X
max (−λi (bi + 1) − µi (bi + 1)) ,
λ,µ
i=1

subject to the constraints:


m
X
λi , µi ≥ 0, λi + µi ≤ 1, (λi − µi )ai = 0.
i=1

Thus, the dual problem is:


m
X
max (−(λi + µi )(bi + 1)) ,
λ,µ
i=1

2
subject to the constraints:
m
X
λi , µi ≥ 0, λi + µi ≤ 1, (λi − µi )ai = 0.
i=1

You might also like