This preview shows pages 1–4. Sign up to view the full content.
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 Document
Unformatted text preview: 10725 Optimization, Spring 2008: Homework 5 Solutions Due: Wednesday, April 23, beginning of the class (Acknowledgement: The solutions for Q1 and Q2 are adapted from Prasanna Veglagapudi) 1 Conjugate functions [Han, 20 points] Derive the conjugates of the following functions (4 points each): (a) Max function. f ( x ) = max i =1 ,...,n x i on R n . F SOLUTION: f * ( y ) = , i y i = 1 , y i i , otherwise (b) Sum of largest elements. f ( x ) = r i =1 x [ i ] on R n . (where x [ i ] denotes the ith largest element in x ). F SOLUTION: f * ( y ) = , i y i = r, y i i , otherwise (c) Power function. f ( x ) = x p on R ++ , where p > 1. F SOLUTION: f * ( y ) = y y p 1 / ( p 1) y p p/ 1 p , y > , y (d) Geometric mean. f ( x ) = ( Q n i =1 x i ) 1 /n on R n ++ . F SOLUTION: f * ( y ) = , Q i ( y i ) (1 /n ) n y i < i , otherwise (e) Negative generalized logarithm for secondorder cone. f ( x, t ) = log( t 2 x T x ) on { ( x, t ) R n R k x k 2 < t } . 1 F SOLUTION: f * ( y, s ) = 2 + log 4 s 2 y T y , s < s 2 > y T y s R , y R n , otherwise 2 Dual of SOCP [Han, 20 points] Show that the dual of the SOCP minimize f T x s . t . k A i x + b i k 2 c T i x + d i , i, = 1 , . . . , m with variables x R n can be expressed as maximize m X i =1 ( b T i u i + d i v i ) s . t . m X i =1 ( A T i u i + c i v i ) + f = 0 k u i k 2  v i , i = 1 , . . . , m, with variables u i R n , v i R , i = 1 , . . . , m . The problem data are f R n , A i R n i n , b i R n i , c i R n and d i R , i = 1 , . . . , m . Derive the dual in the following two ways. (a) [10 points] Introduce new variables y i R n i and t i R and equalities y i = A i x + b i , t i = c T i x + d i , and derive the Lagrange dual. F SOLUTION: First, restate the problem with new variables y i and t i : min x,y,t f T x s.t. y i = A i x + b i t i = c T i x + d i  y i  2 t i t i Next, solve for the Lagrangian: L ( x, y, t, u, v, , ) = f T x + m X i =1 u T i ( A i x + b i y i ) + m X i =1 v i ( c T i x + d i t i ) + m X i =1 i (  y i  2 t i ) + m X i =1 i t i L ( x, y, t, u, v, , ) = f T x + m X i =1 u T i ( A i x + b i y i ) + m X i =1 v i ( c T i x + d i t i ) + m X i =1 i q y T i y i t i + m X i =1 i t i i i i i Use the derivatives of the Lagrangian to get dual constraints: 2 L x = f + m X i =1 ( A T i u i + c i v i ) = 0 m X i =1 ( A T i u i + c i v i ) + f = 0 L y i = u i + i y i p y T i y i = 0 u i = i y i p y T i y i  u i  2 = q u T i u i = v u u t i y i p y T i y i ! T i y i p y T i y i !  u i  2 = i v u u t y T i y i p y T i y i 2 !...
View Full
Document
 Spring '07
 CarlosGustin

Click to edit the document details