MATH 6026 - UPRMacademic.uprm.edu/~pvasquez/mate6026/clases1516II/2.1.pdf · MATH 6026 What is a...
Transcript of MATH 6026 - UPRMacademic.uprm.edu/~pvasquez/mate6026/clases1516II/2.1.pdf · MATH 6026 What is a...
MATH 6026
Dr. Pedro Vásquez
UPRM
P. Vásquez (UPRM) Conferencia 1 / 17
MATH 6026
Fundamentals of unconstrained optimization
The mathematical formulation of an unconstrained optimization problemis:
minxf (x)
where x 2 Rn is a real vector with n ≥ 1 components and f : Rn ! R isa smoothed function.
Usually, there is no enough information related to f . All we know are thevalues of f and maybe some of its derivatives at a set of pointsx0, x1, x2, · · · .
P. Vásquez (UPRM) Conferencia 2 / 17
MATH 6026
Example
P. Vásquez (UPRM) Conferencia 3 / 17
MATH 6026
What is a solution?
Ideally we would like to obtain a global minimizer of f , a point where thefunction obtains its least value.
Definitiona point x∗ is a global minimizer if f (x∗) ≤ f (x) for all x 2 Rn
or over the domain of interest.
Definitiona point x∗ is a local minimizer if there is a neighborhood @ ofx∗ such that f (x∗) ≤ f (x) for all x 2 @.
Definitiona point x∗ is a strict local minimizer (also called strong localminimizer) if there is a neighborhood @ of x∗ such that f (x∗) < f (x) forall x 2 @ with x 6= x∗.
Definitiona point x∗ is an isolated local minimizer if there is aneighborhood @ of x∗ such that x∗ is the only minimizer in @.
P. Vásquez (UPRM) Conferencia 4 / 17
MATH 6026
Example
P. Vásquez (UPRM) Conferencia 5 / 17
MATH 6026
Reconizing a local minimumOne way to find out if a point x∗ is a local minimum is to examine all thepoints in its inmediate vicinity, to make sure that none of them has asmaller function value. If the function f is smoothed, in particular if itstwice continuously di§erentiable, we will be able to tell that x∗ is a localmimizer by examining the gradient rf (x∗) and the Hessian r2f (x∗) .
Theorem
(Taylor 0s Theorem) Suppose that f : Rn ! R is continuouslydi§erentiable and that p 2 Rn. Then we have that:f (x + p) = f (x) +rf (x + tp)T pfor some t 2 (0, 1) . Moreover, if f is twice continuously di§erentiable, wehave that:rf (x + p) = rf (x) +
R 10 r
2f (x + tp) pdtand thatf (x + p) = f (x) +rf (x)T p + 1
2pTr2 (x + tp) p
for some t 2 (0, 1) .
P. Vásquez (UPRM) Conferencia 6 / 17
MATH 6026
Theorem
(First order necessary conditions)If x∗ is a local minimizer and f is continuously di§erentiable in an openneighborhood of x∗, then rf (x∗) = 0.
Proof.
P. Vásquez (UPRM) Conferencia 7 / 17
MATH 6026
Theorem
(Second order necessary conditions)If x∗ is a local minimizer of f and r2f exists and is continuouslydi§erentiable in an open neighborhood of x∗, then rf (x∗) = 0 andr2f (x∗) is positive semidefinite.
Proof.
P. Vásquez (UPRM) Conferencia 8 / 17
MATH 6026
Theorem
(Second order su¢cient conditions)Suppose that r2f is continuously di§erentiable in an open neighborhoodof x∗ and that rf (x∗) = 0 and r2f (x∗) is positive definite. Then x∗ isa local minimizer of f .
Proof.
P. Vásquez (UPRM) Conferencia 9 / 17
MATH 6026
Theorem
When f is convex, any local minimizer x∗ is a global minimizer of f . If inaddition f is di§erentiable, then any stationary point x∗ is a globalminimizer of f .
Proof.
P. Vásquez (UPRM) Conferencia 10 / 17
MATH 6026
Example
P. Vásquez (UPRM) Conferencia 11 / 17
MATH 6026
P. Vásquez (UPRM) Conferencia 12 / 17
MATH 6026
Example
P. Vásquez (UPRM) Conferencia 13 / 17
MATH 6026
P. Vásquez (UPRM) Conferencia 14 / 17
MATH 6026
P. Vásquez (UPRM) Conferencia 15 / 17
MATH 6026
P. Vásquez (UPRM) Conferencia 16 / 17
MATH 6026
P. Vásquez (UPRM) Conferencia 17 / 17