凸优化—凸松弛(Convex Relaxation)


目标(Objective)

Our objective is to transform non-convex functions to a convex functions, i.e.,
min ⁡   f 0 ( x ) → C o n v e x   f u n c t i o n s . t .    f i ( x ) ≤ 0 , i = 1 , . . . , m → C o n v e x   f u n c t i o n s s . t .    a i T x = b i , ( h i ( x ) = 0 )   i = 1 , . . . , p → A f f i n e   f u n c t i o n s \begin{array}{lll} & \min ~ f_0(x) \rightarrow \mathrm{Convex ~ function} \\ & s.t. ~~ f_i(x) \le 0, i=1,...,m \rightarrow \mathrm{Convex ~ functions}\\ & s.t. ~~ a_i^\mathrm{T} x = b_i, ( h_i(x) = 0) ~ i=1,...,p \rightarrow \mathrm{Affine ~ functions} \end{array} min f0(x)Convex functions.t.  fi(x)0,i=1,...,mConvex functionss.t.  aiTx=bi,(hi(x)=0) i=1,...,pAffine functions


等价变换(Equivalent Transformation for Conditions)

min ⁡   f 0 ( x ) = x 1 2 + x 2 2 s . t .     f 1 ( x ) = x 1 1 + x 2 2 ( n o n − c o n v e x ) h i ( x ) = x 1 2 + x 2 2 = 0 ( n o n − a f f i n e ) \begin{array}{lll} & \min ~ f_0(x) =x_1^2 +x_2^2 \\ & s.t. ~~~ f_1(x) = \frac{x_1}{1+x_2^2} (\mathrm{non-convex}) \\ & \qquad h_i(x) = x_1^2 + x_2^2 = 0 (\mathrm{non-affine}) \end{array} min f0(x)=x12+x22s.t.   f1(x)=1+x22x1(nonconvex)hi(x)=x12+x22=0(nonaffine) The above equation can be rewrriten as
min ⁡   f 0 ( x ) = x 1 2 + x 2 2 s . t .     f 1 ( x ) = x 1 ≤ 0 h i ( x ) = x 1 + x 2 = 0 \begin{array}{lll} & \min ~ f_0(x) =x_1^2 +x_2^2 \\ & s.t. ~~~ f_1(x) = x_1 \le 0 \\ & \qquad h_i(x) = x_1 + x_2 = 0 \end{array} min f0(x)=x12+x22s.t.   f1(x)=x10hi(x)=x1+x2=0


降维(Dimension Reduction)

Generally, for reducing the complication of the problem, we need to reduce the dimensionality of functions. But, sometime, dimensionality reduction will lead the probloem to be difficult.
For example, a convex problem can be expressed as
P 0 : min ⁡   f 0 ( x ) s . t .    f i ( x ) ≤ 0 , i = 1 , . . . , m s . t .    a i T x = b i ,   i = 1 , . . . , p \begin{array}{lll} P0: & \min ~ f_0(x) \\ & s.t. ~~ f_i(x) \le 0, i=1,...,m\\ & s.t. ~~ a_i^\mathrm{T} x = b_i, ~ i=1,...,p \end{array} P0:min f0(x)s.t.  fi(x)0,i=1,...,ms.t.  aiTx=bi, i=1,...,p Letting F z = x 0 Fz =x_0 Fz=x0, the above equation can be rewritten as
min ⁡   f 0 ( F z = x 0 ) s . t .    f i ( F z = x 0 ) ≤ 0 , i = 1 , . . . , m \begin{array}{lll} & \min ~ f_0(Fz =x_0)\\ & s.t. ~~ f_i(Fz =x_0) \le 0, i=1,...,m \end{array} min f0(Fz=x0)s.t.  fi(Fz=x0)0,i=1,...,m


升维(Dimension Raising)

Here, we introduce slack variable s i s_i si to raise the dimensionality of the problem.
Problem P 0 P0 P0 can be rewritten as
min ⁡   f 0 ( x ) s . t .   s i ≤ 0 , i = 1 , . . . , m    f i ( x ) − s i = 0 , i = 1 , . . . , m    a i T x = b i ,   i = 1 , . . . , p \begin{array}{lll} & \min ~ f_0(x) \\ & s.t. ~ s_i \le 0, i=1,...,m\\ & \quad ~~f_i(x) -s_i = 0, i=1,...,m\\ & \quad ~~a_i^\mathrm{T} x = b_i, ~ i=1,...,p \end{array} min f0(x)s.t. si0,i=1,...,m  fi(x)si=0,i=1,...,m  aiTx=bi, i=1,...,p


评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值