Beruflich Dokumente
Kultur Dokumente
Francesco Nava
September 2013
Nava (LSE)
EC400B Lecture 2
September 2013
1 / 25
Taylor Expansion
The rst order Taylor Expansion of a function f : R1 R1 is f (a + h) f (a) + f (a)h Dene the residual from the approximation as: R (h|a) = f (a + h) f (a) f (a)h Note that the approximation holds since by the denition of the derivative f (a), we have that: R (h|a) =0 h0 h lim
Nava (LSE)
EC400B Lecture 2
September 2013
2 / 25
Taylor Expansion
The rst order Taylor Expansion of a function f : R1 R1 is f (a + h) f (a) + f (a)h Dene the residual from the approximation as: R (h|a) = f (a + h) f (a) f (a)h Note that the approximation holds since by the denition of the derivative f (a), we have that: R (h|a) =0 h0 h lim
Nava (LSE)
EC400B Lecture 2
September 2013
2 / 25
Taylor Expansion
Geometrically: this is the formalization of the approximation of the graph of f (x ) by its tangent line at (a, f (a)).
Nava (LSE)
EC400B Lecture 2
September 2013
3 / 25
the corresponding k th order residual is dened by: Rk (h|a) = f (a + h) Pk (a|h) and satises (by Taylors Theorem): Rk (h|a) =0 h0 hk lim
Nava (LSE)
EC400B Lecture 2
September 2013
4 / 25
A Simple Example
Consider the rst and second order Taylor polynomial of the exponential function f (x ) = e x at x = 0. All the derivatives of f (x ) at x = 0 equal 1. Therefore: P1 (0|h) = 1 + h P2 (0|h) = 1 + h + For h = 0.2, then: P1 (0|0.2) = 1.2, P2 (0|0.2) = 1.22, f (0.2) = 1.22140275816017 h2 2
Nava (LSE)
EC400B Lecture 2
September 2013
5 / 25
A Simple Example
Consider the rst and second order Taylor polynomial of the exponential function f (x ) = e x at x = 0. All the derivatives of f (x ) at x = 0 equal 1. Therefore: P1 (0|h) = 1 + h P2 (0|h) = 1 + h + For h = 0.2, then: P1 (0|0.2) = 1.2, P2 (0|0.2) = 1.22, f (0.2) = 1.22140275816017 h2 2
Nava (LSE)
EC400B Lecture 2
September 2013
5 / 25
Alternatively, let R1 (h|a) denote the residual, to get: F (a + h) = F (a) + dF (a) h + R1 (h|a) where dF (a) = ( F / x1 , . . . , F / xn ) is the Jacobian of F at a. As before R1 (h|a)/||h|| 0 as h 0.
Nava (LSE)
EC400B Lecture 2
September 2013
6 / 25
Alternatively, let R1 (h|a) denote the residual, to get: F (a + h) = F (a) + dF (a) h + R1 (h|a) where dF (a) = ( F / x1 , . . . , F / xn ) is the Jacobian of F at a. As before R1 (h|a)/||h|| 0 as h 0.
Nava (LSE)
EC400B Lecture 2
September 2013
6 / 25
...
xn xn (a) 2F
Nava (LSE)
EC400B Lecture 2
September 2013
7 / 25
...
xn xn (a) 2F
Nava (LSE)
EC400B Lecture 2
September 2013
7 / 25
Denition
Let f (x) be a real valued function dened on a subset C of Rn . A point x in C is: a global maximizer for f (x) on C if f (x ) f (x) for all x C a strict global maximizer for f (x) on C if f (x ) > f (x) for all x C such that x = x
Nava (LSE) EC400B Lecture 2 September 2013 8 / 25
Denition
Let f (x) be a real valued function dened on a subset C of Rn . A point x in C is: a global maximizer for f (x) on C if f (x ) f (x) for all x C a strict global maximizer for f (x) on C if f (x ) > f (x) for all x C such that x = x
Nava (LSE) EC400B Lecture 2 September 2013 8 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
10 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
11 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
12 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
13 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
14 / 25
How can we determine whether a critical point is a local maximum or a local minimum in this more general setup?
To this end we have to consider the Hessian of the map f (ie the matrix of the second order partial derivatives).
Note that the Hessian is always a symmetric matrix, since cross-partial derivatives are equal (Clairaut-Schwarz Theorem).
Nava (LSE)
EC400B Lecture 2
September 2013
15 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
16 / 25
The rst order leading principle minor is 6x and the second order leading principal minor is 36xy 81. At (0, 0) the two minors are 0 and 81. Hence the Hessian is indenite and (0, 0) is not an extreme point, it is a saddle point. At (3, 3) these two minors are positive. Hence, (3, 3) is a strict local minimum of F . However, (3, 3) is not a global minimum of F (to see why this is the case consider (x , y ) (0, inf)).
Nava (LSE) EC400B Lecture 2 September 2013 18 / 25
The rst order leading principle minor is 6x and the second order leading principal minor is 36xy 81. At (0, 0) the two minors are 0 and 81. Hence the Hessian is indenite and (0, 0) is not an extreme point, it is a saddle point. At (3, 3) these two minors are positive. Hence, (3, 3) is a strict local minimum of F . However, (3, 3) is not a global minimum of F (to see why this is the case consider (x , y ) (0, inf)).
Nava (LSE) EC400B Lecture 2 September 2013 18 / 25
The rst order leading principle minor is 6x and the second order leading principal minor is 36xy 81. At (0, 0) the two minors are 0 and 81. Hence the Hessian is indenite and (0, 0) is not an extreme point, it is a saddle point. At (3, 3) these two minors are positive. Hence, (3, 3) is a strict local minimum of F . However, (3, 3) is not a global minimum of F (to see why this is the case consider (x , y ) (0, inf)).
Nava (LSE) EC400B Lecture 2 September 2013 18 / 25
If d 2 F (x ) is negative denite, then for all small enough h = 0, the right hand side is negative. This in turn implies that for small enough h: F (x + h) < F (x ) Hence, x is a strict local maximizer of F .
Nava (LSE) EC400B Lecture 2 September 2013 19 / 25
Denition
A real valued function f dened on a convex subset U of Rn is concave, if for all x, y in U and for all t [0, 1] : f (t x + (1 t )y) t f (x) + (1 t ) f (y)
Denition
A real valued function g dened on a convex subset U of Rn is convex, if for all x, y in U and for all t [0, 1] : g (t x + (1 t )y) t g (x) + (1 t ) g (y)
Nava (LSE) EC400B Lecture 2 September 2013 20 / 25
Denition
A real valued function f dened on a convex subset U of Rn is concave, if for all x, y in U and for all t [0, 1] : f (t x + (1 t )y) t f (x) + (1 t ) f (y)
Denition
A real valued function g dened on a convex subset U of Rn is convex, if for all x, y in U and for all t [0, 1] : g (t x + (1 t )y) t g (x) + (1 t ) g (y)
Nava (LSE) EC400B Lecture 2 September 2013 20 / 25
Denition
A real valued function f dened on a convex subset U of Rn is concave, if for all x, y in U and for all t [0, 1] : f (t x + (1 t )y) t f (x) + (1 t ) f (y)
Denition
A real valued function g dened on a convex subset U of Rn is convex, if for all x, y in U and for all t [0, 1] : g (t x + (1 t )y) t g (x) + (1 t ) g (y)
Nava (LSE) EC400B Lecture 2 September 2013 20 / 25
Theorem
Let f be a continuous dierentiable function on a convex subset U of Rn . If so, f is concave on U if and only if for all x, y in U: f (y) f (x) df (x) (y x) = f (x) f (x) = (y1 x1 ) + ... + (yn xn ) x1 xn
Nava (LSE) EC400B Lecture 2 September 2013 21 / 25
Theorem
Let f be a continuous dierentiable function on a convex subset U of Rn . If so, f is concave on U if and only if for all x, y in U: f (y) f (x) df (x) (y x) = f (x) f (x) = (y1 x1 ) + ... + (yn xn ) x1 xn
Nava (LSE) EC400B Lecture 2 September 2013 21 / 25
Proof for R1
Since f is concave, then t f (y ) + (1 t ) f (x ) f (ty + (1 t )x ) f (x ) + t (f (y ) f (x )) f (x + t (y x )) f (x + t (y x )) f (x ) f (y ) f (x ) t f (x + h) f (x ) f (y ) f (x ) (y x ) h where the latter holds for h = t (y x ). The limit the last expression when h 0 gives: f (y ) f (x ) f (x )(y x )
Nava (LSE)
EC400B Lecture 2
September 2013
22 / 25
Theorem
If f is a continuous, dierentiable, concave function on a convex set U and if x0 U, then: df (x0 )(y x0 ) 0 f (y) f (x0 ) and if this holds for all y U, then x0 is a global maximizer of f .
Theorem
Let f be a twice continuously dierentiable function whose domain is a convex open subset U of Rn . If f is a concave function on U and df (x0 ) = 0 for some x0 , then x0 is a global maximum of f on U.
Theorem
A twice continuously dierentiable function f on an open convex subset U of Rn is concave on U if and only if the Hessian d 2 f (x) is negative semidenite for all x in U. The function f is a convex function if and only if d 2 f (x) is positive semidenite for all x in U.
Nava (LSE) EC400B Lecture 2 September 2013 23 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
24 / 25
Nava (LSE)
EC400B Lecture 2
September 2013
25 / 25