15.05.2015 Views

Introduction to Unconstrained Optimization - Scilab

Introduction to Unconstrained Optimization - Scilab

Introduction to Unconstrained Optimization - Scilab

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

f(x)<br />

y 1<br />

x y 2<br />

Figure 17: Proof of the first order condition on convex functions - part 2<br />

By the convexity hypothesis of f, we deduce:<br />

f(θy + (1 − θ)x) ≤ θf(y) + (1 − θ)f(x). (32)<br />

The left hand side can be written as f(y + θ(x − y)) while the right hand side of the<br />

previous inequality can be written in the form f(x) + θ(f(y) − f(x)). This leads <strong>to</strong><br />

the following inequality:<br />

f(x + θ(y − x)) ≤ f(x) + θ(f(y) − f(x)). (33)<br />

We move the term f(x) <strong>to</strong> the left hand side, divide by θ > 0 and get the inequality:<br />

f(x + θ(y − x)) − f(x)<br />

θ<br />

≤ f(y) − f(x). (34)<br />

We can take the limit as θ → 0 since, by hypothesis, the function f is continuously<br />

differentiable. Therefore,<br />

g(x) T (y − x) ≤ f(y) − f(x), (35)<br />

which concludes the first part of the proof.<br />

In the second part of the proof, let us assume that the function f satisfies the<br />

inequality 31. Let us prove that f is convex.<br />

Let y 1 , y 2 be two points in C and let θ be a scalar such that 0 ≤ θ ≤ 1. The<br />

idea is <strong>to</strong> use the inequality 31, with the carefully chosen point x = θy 1 + (1 − θ)y 2 .<br />

This idea is presented in figure 17.<br />

The first order condition 31 at the two points y 1 and y 2 gives:<br />

f(y 1 ) ≥ f(x) + g(x) T (y 1 − x) (36)<br />

f(y 2 ) ≥ f(x) + g(x) T (y 2 − x) (37)<br />

We can form a convex combination of the two inequalities and get:<br />

θf(y 1 ) + (1 − θ)f(y 2 ) ≥ θf(x) + (1 − θ)f(x) (38)<br />

The previous inequality can be simplified in<strong>to</strong>:<br />

+g(x) T (θ(y 1 − x) + (1 − θ)(y 2 − x)) . (39)<br />

θf(y 1 ) + (1 − θ)f(y 2 ) ≥ f(x) + g(x) T (θy 1 + (1 − θ)y 2 − x) . (40)<br />

22

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!