参考文献:first-order methods in optimization: Amir Beck
1实值函数、domain、epi、下半连续、水平集 几个等价性质 函数的闭性 保闭的运算 (下半连续和闭几乎等价)
2函数的闭性和连续性 函数在紧集上连续那么在紧集上可取到minimum 、紧集
3函数的凸性(2个凸函数定义的等价性) 保凸的运算
4支撑函数 支撑集
5非线性优化 里前面的内容可能多了锥的详细介绍 本书多了函数连续、闭、凸之间的一些关系
2.1 Extended Real-Valued Functions and Closedness
- real-valued function: can take any real value without null value and
- extended real-valued function:can take any real value, as well as
- domain:
ex2.1 .
then
- epigraph of f:
[f is closed if epi(f) is closed]
- lower semicontinuity:
- level set:
Note :f is lower semicontinuous<=>f is closed<=>![{\color{Red} for \,any\,\alpha \in \mathbb{R},\, Lev(f,\alpha )=\left \{ x\in\mathbb{E}:f(x)\leq \alpha \right \}\,is \,closed.}](https://i-blog.csdnimg.cn/blog_migrate/756ff82bedf6549de45260e0d69683e6.png)
2.2 Closedness versus Continuity
T2.8 如果f在其domain连续且dom(f)是闭的,那么f是闭的。(利用f闭等价于epi(f)是闭证明)
但是,反过来,f闭不能推出f在其domain上连续
0 范数 不是范数(不满足绝对值齐次性)
A set is called compact if it is closed and bounded.
2.3 Convex Function
Def 2.15 (convex functions)An extended real-valued function f : E →[−∞, ∞] is called convex if epi(f) is a convex set.
f(λx + (1 − λ)y) ≤ λf(x) + (1 − λ)f(y) for all x, y ∈ E, λ ∈ [0, 1],
介绍了几种保凸运算:
Infimal Convolution(卷积):
是凸的。
凸函数在其domain内部(int(dom f))是连续的,
凸函数的局部Lipschitz连续性:
凸函数在边界点不一定连续。即使函数是闭凸的,也不能保证。
单变量的闭凸函数f在dom(f)上是连续的【包括边界】
2.4 Support Functions
C非空,则C的支撑集是闭凸的。
严格分离定理 (strict separation theorem)
注:泛函里面提到 此处C是包含内点(零元)的