This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: MATHEMATICS 116, FALL 2007 CONVEXITY AND OPTIMIZATION WITH APPLICATIONS Outline #7 (Calculus of Variations) Last modified: December 6, 2007 Reading. Luenberger, Chapter 7, sections 7.17.5 and 7.7. 1 Lecture topics. 1. Review of the twovariable case Let f ( u,v ) = u 2 v u 2 + v 2 with f (0 , 0) = 0 First we use the Gateaux approach, without introducing a norm into R 2 Calculate ∂f ∂u and ∂f ∂v Calculate the directional derivative at (0,0) along the vector h = (1 ,m ): δf ((0 , 0);(1 ,m )) = lim α → 1 α f ( α (1 ,m )) f (0 , 0)) True or false?: the directional derivative in this case is a linear of its second argument (the increment) What is the only linear function of h , δf ( θ ; h ) that could possibly work in this case, given the values of the partial derivatives? Introduce the Euclidean norm, so r =  x  = √ u 2 + v 2 . Show that lim  x → 1  x  ( f ( x ) f (0 , 0)) 6 = 0 Example 1 on page 171, specialized to this case, says that if X = E 2 and f ( u,v ) is a functional with continuous partial derivatives, then δf ( x ; h ) = ∂f ∂u h 1 + ∂f ∂v h 2 . Is there any contradiction? 2 2. Gateaux and Frechet differentials Consider a transformation T : X → Y . The space Y has to be normed, because we need to take a limit to define the differential. Gateaux generalizes the concept of directional derivative from multivariable calculus. It defines a functional of the increment vector h at any point x ∈ X , provided the limit exists. δT ( x ; h ) = lim α → 1 α ( T ( x + αh ) T ( x )) Assuming that the limit exists for all h , is the Gateaux differential guaran teed to be linear in h ? Frechet generalizes the concept of derivative from multivariable calculus. It requires X to be normed. It defines a continuous linear functional δT ( x ; h ) of the increment vector h at any point x ∈ X , provided the error in the linear approximation goes to zero faster than  h  . lim  h → 1  h   T ( x + h ) T ( x ) δT ( x ; h ))  = 0 In the example on the preceding page, does the Frechet differential of f at (0,0) exist? 3 3. A few familiar thoerems The following results are proved in Luenberger. What are the corresponding theorems from multivariable calculus? • If T has a Frechet differential, it is unique. • If T has a Frechet differential at x , it has a Gateaux differential at x , and the two are equal. • If T has a Frechet differential at x , it is continuous at x . Prove this one. • If function f on E n has continuous partial derivatives, then it has the Frechet differential δf ( x ; h ) = n X i =1 ∂f ∂x i h i . Just beacuse f has a Gateaux differential at x , it is not necessarily even con tinuous at x . For a counterexample, define f ( u,v ) = u 2 v u 4 + v 2 with f (0 , 0) = 0 and consider the sequence x i = { 1 i , 1 i 2 } 4 4. Differentials in infinitedimensional spaces Consider the nonlinear functional f ( x ) = Z a x ( t ) 2 dt ....
View
Full Document
 Fall '09
 Calculus, Derivative, dt, continuous partial derivatives, Luenberger

Click to edit the document details