关于Moreau-Yosida的一些优良性质

最新推荐文章于 2024-07-15 22:10:26 发布

Full ofstarsl

最新推荐文章于 2024-07-15 22:10:26 发布

阅读量123

点赞数

分类专栏：优化基础知识文章标签：算法

本文链接：https://blog.csdn.net/qq_43884228/article/details/129943888

版权

优化基础知识专栏收录该内容

3 篇文章 0 订阅

订阅专栏

1、Introduction

Moreau -Yosida regularitztion is a preconditioner of a convex function f.

And this preconditioner has second-order properties

共轭： conjugate of a closed convex function
在这里插入图片描述
closed function =lower semi-continuous function

具体表示： We denote by F the Moreau-Yosida regularization of a given closed convex function f associated to the metric de ned by M :
在这里插入图片描述
Denote $p (x)$ is the unique minimizer in $F (x)$

2、Properities of the moreau-yosida regularization

Definition1:
The recession (or asymptotic) function of a closed convex function C is defined by 在这里插入图片描述
(a limit which does not depend on $x\in dom\phi$ ). This function is useful because $\phi$ has a nonempty bounded set of minima if and only if $\phi^{'}_{\infty}(d)>0$ for all $d\not=0$ .
Definition2: Proximal Point: We will extensively use the following system of notation:
$p_{M}(x)=argmin_{y} f(y)+\frac{1}{2}M^{T}(y-x)^{T}(y-x)$
is called the proximal point of x.

Lemma1:
The minimization problem of $F (x)$ has a unique solution, characterized as the unique point $y\in R^{n}$ satisfying
$M(x-y)\in\partial f(y)$ 即 $M(x-p_{M}(x))\in\partial f(p_{M}(x))$
proof is on J.-B. Hiriart-Urruty and C. Lemar echal, Convex Analysis and Minimization Algorithms, Springer-Verlag, Berlin, New York, 1993;XV,Lemma.4.1.1

Theorem1:
The recession functions of $f$ and $F$ are identical.
proof : Ref:Rockafellar,Corollary 9.2.1

Theorem 2.
For a finite-valued convex function $f$ , the following statements are equivalent:
(i) $f$ is strongly convex with modulus $\frac{1}{l}$ ;
(ii) $f$ has a Lipschitzian gradient with Lipschitz constant $l$
(iii) $F$ has a Lipschitzian gradient with Lipschitz constant $L$ ;
(iv) $F$ is strongly convex with modulus $\frac{1}{L}$ .
Furthermore, we have the inequalities $\frac{l-1}{\lambda}<L<\frac{l-1}{\lambda}$ .
Reference

proof:[REF:Lemar C , chal, Sagastiz C A . Practical Aspects of the Moreau–Yosida Regularization[J]. SIAM Journal on Optimization, 1997.]

Theorem3:
The function $F (y)$ is finite everywhere,convex and differentiable ;its gradient is $G(y)=\nabla F(y)=s_{M}(y)=M(y-p(y))$ .
Furthermore,there holds for all $y$ and $y^{'}$ in $R^{n}$ ,
$\langle s_{M}(y)-s_{M}(y^{'}),W(s_{M}(y)-s_{M}(y^{'}))\rangle \leq \langle s_{M}(y)-s_{M}(y^{'}),y-y^{'}\rangle$
and
$\parallel \nabla F(y)-\nabla F(y^{'})\parallel\leq \parallel M\parallel \parallel y-y^{'} \parallel$
proof

Theorem4:
Minimizing $f$ and $F$ are equivalent problems,in the sense
$\inf_{x\in R^{n}} F(x) = \inf_{x\in R^{n}} f(x)$
an equality in $R\cup\lbrace -\infty\rbrace$ ,and the following statements are equvalent.
(i) $x$ minimizes $f$
(ii) $p_{M}(x)=x$
(iii) $s_{M}(x)=0$
(iv) $x$ minimizes $f_{M}(x)$
(v) $f(p_{M}(x))=f(x)$
(vi) $f_{M}(x)=f(x)$
在这里插入图片描述
We can use the corollary 2.1.4 to proov the equivalent of $\inf F(x)$ and $\inf (x)$

proof: On J.-B. Hiriart-Urruty and C. Lemar echal, Convex Analysis and Minimization Algorithms, Springer-Verlag, Berlin, New York, 1993.XV,TH4.1.7

Proposition 1
Assume $f$ is a closed convex function. Then $\nabla F ( )$ has directional derivatives if and only if $p ()$ has directional derivatives. The Hessian $\nabla^{2}F (x)$ exists if and only if the Jacobian $\nabla p(x)$ exists:
$\nabla ^{2}F(x)=M(I-\nabla p(x)),for\quad all \quad x\in R^{n}$

Reference:
[1] Lemar C , chal, Sagastiz C A . Practical Aspects of the Moreau–Yosida Regularization[J]. SIAM Journal on Optimization, 1997.
[2]J.-B. Hiriart-Urruty and C. Lemar echal, Convex Analysis and Minimization Algorithms, Springer-Verlag, Berlin, New York, 1993