非线性优化 基础概念（方述诚1

nonlinear programming 讨论两种问题：unconstrained problem、constrained optimization

n个变数， 二阶导连续

 

 

 

从简单问题入手，可以有意识的发掘问题所在，motivation、hypothesis、valid、推广

Unconstrained vs. Constrained Problems

In a sense, a “ constrained optimization " problem can be reformulated as an “ unconstrained

optimization " problem.

Theory and algorithms of unconstrained NLP leads to that of constrained NLP

Example

 

当x落在S之内（feasible），没有penalty、

掉在外面处罚太高，最后会留在S里，    用罚函数 限制x不要跑出去，

c足够大，不敢落出去，然后再把c逐渐降下来，

constraint问题   化为unconstraint问题

Nonlinear vs. Linear Problems   

   In a sense, “nonlinearity” is locally close to “linearity,”  but it much better reflects reality.

泰勒展开，永远可以做一阶的逼近，linear approximation，一个非线性函数，不求精确解时，可以用线性函数来approximate

Example

 问：会在哪里相遇？

 当guard比我速度快一点点  v+ε，非线性    二次的

放松”equal speed“ ”equal strength“ ”equal intelligence“...等条件..

Discrete vs. Continuous Problems  

   In a sense, “discreteness” is not far from

“continuity,” but much more abundant of   local information.

discrete     可以变成 双整数优化问题

离散时，函数断掉了，性质没法从