Convex sets -- Convex optimization

Hyperplanes and halfspaces

超平面和半空间的几何理解, 取任意在超平面上的点x0x_0, 那么存在点xx, 然后向量x−x0x-x_0与超平面的法向量aa做对比,乘积为钝角或者锐角即可划分半平面空间了。

{x∣a⊺x=b}\{x | a^\intercal x = b\}

{x∣a⊺(x−x0)=0},a⊺x0=b\{x | a^\intercal(x -x_0) = 0\}, a^\intercal x_0=b


Hyperplane in R2\mathcal{R}^2, with normal vector a and a point x0x_0 in the hyperplane. For any point x in the hyperplane, x−x0x − x_0 (shown as the darker arrow) is orthogonal to aa.

A hyperplane defined by a⊺x=ba^\intercal x = b in R2\mathcal{R}^2 determines two halfspaces. The halfspace determined by
a⊺x≥ba^\intercal x \geq b (not shaded) is the halfspace extending in the direction aa. The halfspace determined by a⊺x≤ba^\intercal x \leq b (which is shown shaded) extends in the direction −a-a. The vector a is the outward normal of this halfspace.

The shaded set is the halfspace determined by a⊺(x−x0)≤0a^\intercal (x − x_0) \leq 0. The vector x1−x0x_1 −x_0 makes an acute angle with a, so x1x_1 is not in the halfspace. The vector x2−x0x_2 − x_0 makes an obtuse angle with a, and so is in the halfspace.

Proper cones and generalized inequalities

  • KK is pointed, which means that it contains no line (or equivalently, x∈K,−x∈K⇏x=0x \in K, −x \in K \nRightarrow x=0). 几何意思是一个cone里面不存在直线,使得穿过0点后,既在cone中,又在cone外。

minimum and minimal

极小元,最小元
minimum和minimal的区别: 广义的x+Kx + K及x−Kx - K是否与S只是相交于一点。KK is a proper cone. 在广义的不等式中,被当做符号来使用⪯K\preceq_K.
以下是两个重要概念

  • 最小元的定义只考察在KK中的所有广义大于xx的点,如下图左边子图所示广义大于x1x_1的点组成了浅灰色的区域

S⊆x+KS \subseteq x + K

where x+Kx+K denotes all the points that are comparable to xx and greater than or equal to xx (according to ⪯K\preceq K ).

  • 极小元的定义只考察在KK中的所有广义小于xx的点,如下图右边子图所示广义小于x1x_1的点组成了浅灰色的区域

(x−K)∩S={x}(x -K) \cap S = \{x\}

where x−Kx − K denotes all the points that are comparable to xx and less than or equal to xx (according to ⪯K\preceq K).

A∈S++nA \in \mathcal{S}^n_{++}表示一个圆心在圆点的椭圆

εA={x∣x⊺A−1x≤1}\varepsilon_A = \{x | x^\intercal A^{-1}x \leq 1\}

且定义A⪯BA \preceq B等价于εA⊆εB\varepsilon_A \subseteq \varepsilon_B
定义对称矩阵的结合为SS:

S={P∈S++n∣vi⊺P−1vi,i=1,...,k}S = \{P \in \mathcal{S}^n_{++} | v_i^\intercal P^{-1} v_i, i = 1,...,k\}

下图表示:
对比椭圆1和3都经过了图中的两个点,等价于了A⪯BA \preceq B, 但是没有一个椭圆可以被椭圆2包含,且经过图中的那个点(两个椭圆圆心相同)。说明了椭圆2表示的对称矩阵PP是SS的最小元, 当k=1k=1。