凸優化基本概念-仿射集,凸集,凸錐

1)凸集,凸函數,凸優化
仿射集
例1:任何一個線性方程的解集一定是一個仿射集
c={xAX=b},ARm×n,bRm,xRnc=\{x|AX = b\},A \in R^{m\times n},b \in R^m,x \in R^n

證明如下:
X1,X2c\forall X_1,X_2 \in c,AX1=b,AX2=bAX_1 = b,AX_2 = b
θR\theta \in R,θX1+(1θ)X2c\theta X_1 + (1-\theta)X2 \in c
A(θX1+(1θ)X2)=bA(\theta X_1 + (1-\theta)X2) = b
=θAX1+(1θ)AX2=\theta AX_1 +(1-\theta)AX_2
=b=b

例2:
v={XX0Xc},X0cv = \{ X-X_0|X \in c\},\forall X_0 \in c
={XX0AX=b},AX0=b = \{X-X_0|AX = b\},AX_0 = b
{XX0A(XX0)=0}\{X-X_0|A(X-X_0) = 0\}
XX0=yX-X_0 = y,則上式可以寫爲:
{yAy=0}\{ y|Ay = 0\}
yyAA的畫零空間

二 給定任意集合cc,構造儘可能小的仿射集

仿射包:
aff c={θX1+θX2+....θRXRxiR,θ1+.....θR=1}c = \{\theta X_1 + \theta X_2 + ....\theta_RX_R|\forall x_i \in R,\forall \theta_1+.....\theta_R = 1\}

凸集(convex set):
一個集合cc是凸集,當任意兩點之間的線段仍然在cc內,
cc爲凸集x1,x2c,θ,θ0,1,θx1+(1θ)x2c\Leftrightarrow \forall x_1,x_2 \in c,\forall \theta,\theta \in \lceil 0,1\rceil, \theta x_1 + (1-\theta)x_2 \in c

x1,x2,.........xnx_1,x_2,.........x_n的凸組合,θ1,θ2........θRR,\theta_1,\theta_2........\theta_R \in R,,
θ1+θ2+θ3....+θn=1\theta_1+\theta2+\theta_3....+\theta_n = 1
θ1......θR0,1\theta_1......\theta_R \in \lceil 0,1 \rceil
θ1x1+θ2x2+.......θkxk\theta_1x_1+\theta_2x_2+.......\theta_kx_k稱爲凸組合

cc爲凸集 \Leftrightarrowcc中任意元素的凸組合一定在cc內。
凸包:conv C={θ1x1+θ2x2.......+θRxRx1,x2,....xrc,θ1,θ2,......θr0,1,θ1+.......θR=1}C = \{ \theta_1x_1+\theta_2x_2.......+\theta_Rx_R| \forall x_1,x_2,....x_r \in c,\forall \theta_1,\theta_2,......\theta_r \in \lceil0,1\rceil,\theta_1+.......\theta_R =1\}

錐Cone ,凸錐convex Cone
C 是錐xc,θ0,θxC\Leftrightarrow \forall x \in c ,\theta \geqslant 0,\theta x \in C
C是凸錐x1,x2c,θ1,θ20,x1θ1+x2θ2c\Leftrightarrow \forall x_1,x_2 \in c ,\theta_1,\theta_2 \geqslant 0,x_1\theta_1 + x_2\theta_2 \in c

凸錐組合
θ1x1+θ2x2+.....+θkxk,θ1,θ2,θ3.....θk0\theta_1x_1+\theta_2x_2+.....+\theta_kx_k,\theta_1,\theta_2,\theta_3.....\theta_k \geqslant 0
凸錐包
x1,x2,......xkc,{θ1x1+θ2x2+....θkxkx1,x2....xkc,θ1,θ2,.......θk0x_1,x_2,......x_k \in c,\{\theta_1x_1+\theta2 x_2+....\theta_kx_k|x_1,x_2....x_k \in c,\theta_1,\theta_2,.......\theta_k \geqslant 0

總結:
仿射組合
θ1,........θk,θ1+....+θk=1\forall \theta_1,........\theta_k,\theta_1+....+\theta_k = 1
凸組合
θ1.......θk,θ1+......+θk=1,θ1.....θk0,1\forall \theta_1.......\theta_k,\theta_1+......+\theta_k = 1,\theta_1.....\theta_k \in \lceil 0,1\rceil

凸錐組合
θ1,........θk,θ1.....θk0\forall \theta_1,........\theta_k,\theta_1.....\theta_k \geqslant 0

課程鏈接

發表評論
所有評論
還沒有人評論,想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章