Complete metric space and Banach fixed point theorem(Complete space와 바나흐 고정점 정리)

DongYoung Kim·2022년 7월 26일
0

Nonlinear Dynamics

목록 보기
1/6
post-thumbnail

This article referenced the lecture notes below.
https://nptel.ac.in/courses/108106024

  In this article, we are going to talk about the definition of complete metric space and Banach fixed point theorem, also prove the Banach fixed point theorem. So first, we need to talk about the definition of Cauchy sequence and complete metric space.

Definition of Cauchy sequence) Let X\mathcal{X} be a metric space and {pn}\{p_n\} be a sequence in X\mathcal{X}; Then {pn}\{p_n\} is a Cauchy sequence if and only if \forall ϵ>0\epsilon>0, \exist NNN \in \mathbb{N} such that

m,nNimpliesd(pm,pn)<ϵ;m, n \ge N \, \, implies \, \, d(p_m, p_n)<\epsilon;

Definition of complete metric space) A metric space X\mathcal{X} is complete if and only if every Cauchy sequence converges to a point of X\mathcal{X};

  And we can show that closed subset of a complete metric space is also complete. So, we are going to prove Banach fixed point theorem. Before proving the theorem, we need to define contractive mapping.

Definition of contractive mapping) A map P:XXP:\mathbb{X} \rightarrow \mathbb{X} is said to be contractive if and only if there exists a 0ρ<10\le\rho<1 such that

dX(P(x1),P(x2))ρdX(x1,x2)forx1,x2X;d_{\mathbb{X}}(P(x_1 ), P(x_2)) \le \rho \cdot d_{\mathbb{X}}(x_1, x_2) \, \, for \, \, \forall x_1, x_2 \in \mathbb{X};

  Thus we say that the sequence from x2x_2 to x1x_1 gets contracted under the action of P. Now we are ready to prove Banach fixed point theorem.

Theorem) Let X\mathbb{X} be a complete metric space and let SS be a closed subset of X\mathbb{X}; Suppose T:SST:S \rightarrow S is contractive. Then :

1. There exists a unique fixed point xx^* (i.e., T(x)=xT(x^*)=x^*) in SS.

2. xx^* can be found by successively operating TT on any x0Sx_0 \in S.

Proof of the theorem)

  Because SS is a closed subset of the complete metric space X\mathbb{X}, we can say that SS is also complete. For a sequence xn{x_n} in SS, let xn+1=T(xn)x_{n+1} = T(x_{n}) for any n0n\ge0. We first need to show that {xn}\{x_n\} is a Cauchy hence xx^* exists. See that

dS(xn+1,xn)=dS(T(xn),T(xn1))ρdS(xn,xn1)...ρndS(x1,x0)=ρndS(T(x0),x0);d_{\mathcal{S}}(x_{n+1}, x_n) = d_{\mathcal{S}}(T(x_n), T(x_{n-1})) \le \rho \cdot d_{\mathcal{S}}(x_{n}, x_{n-1}) \\ \le ... \le \rho^n \cdot d_{\mathcal{S}}(x_1, x_0) = \rho^n \cdot d_{\mathcal{S}}(T(x_0), x_0);

  Then, for any given rNr \in \mathbb{N}, let m=n+r;m=n+r; by triangular inequality:

dS(xm,xn)=dS(xn+r,xn)dS(xn+r,xn+r1)+dS(xn+r1,xn)dS(xn+r,xn+r1)+dS(xn+r1,xn+r2)+dS(xn+r2,xn)dS(xn+r,xn+r1)+...+dS(xn+1,xn)ρn+r1dS(T(x0),x0)+ρn+r2dS(T(x0),x0)+...+ρndS(T(x0),x0)=i=0r1ρn+idS(T(x0),x0)i=0ρn+idS(T(x0),x0)=ρn1ρdS(T(x0),x0);d_{\mathcal{S}}(x_m, x_n) = d_{\mathcal{S}}(x_{n+r}, x_n) \le d_{\mathcal{S}}(x_{n+r}, x_{n+r-1}) + d_{\mathcal{S}}(x_{n+r-1}, x_n) \\ \le d_{\mathcal{S}}(x_{n+r}, x_{n+r-1}) + d_{\mathcal{S}}(x_{n+r-1}, x_{n+r-2}) + d_{\mathcal{S}}(x_{n+r-2}, x_n) \\ \le d_{\mathcal{S}}(x_{n+r}, x_{n+r-1}) + ... + d_{\mathcal{S}}(x_{n+1}, x_n) \\ \le \rho^{n+r-1} \cdot d_{\mathcal{S}}(T(x_0), x_0) + \rho^{n+r-2} \cdot d_{\mathcal{S}}(T(x_0), x_0) + ... + \rho^{n} \cdot d_{\mathcal{S}}(T(x_0), x_0) \\ = \sum_{i=0}^{r-1} \rho^{n+i} \cdot d_{\mathcal{S}}(T(x_0), x_0) \le \sum_{i=0}^{\infty} \rho^{n+i} \cdot d_{\mathcal{S}}(T(x_0), x_0)= \frac{\rho^n}{1-\rho} \cdot d_{\mathcal{S}}(T(x_0), x_0);

Thus

dS(xm,xn)ρn1ρdS(T(x0),x0);d_{\mathcal{S}}(x_m, x_n) \le \frac{\rho^n}{1-\rho} \cdot d_{\mathcal{S}}(T(x_0), x_0);

  Note that as nn \rightarrow \infty then dS(xm,xn)0d_{\mathcal{S}}(x_m, x_n) \rightarrow 0 for arbitrary nn and mm. Thus we show that {xn}\{x_n\} is a Cauchy. Because SS is complete, {xn}\{x_n\} converges to a limit xSx^*\in S.

  Now we need to show that xx^* is a fixed point :T(x)=x: T(x^*) = x^*. See that

x=limnxn=limnxn+1=limnT(xn)=T(x);x^* = \lim_{n \rightarrow \infty}x_{n} = \lim_{n \rightarrow \infty}x_{n+1} = \lim_{n \rightarrow \infty}T(x_{n}) = T(x^*) ;

hence we can say that xx^* is fixed.

  And we show that xx^* is the unique fixed point. Let xˉ\bar{x} be another fixed point:

dS(x,xˉ)=dS(T(x),T(xˉ))ρdS(x,xˉ)=ρdS(T(x),T(xˉ))ρ2dS(x,xˉ)...ρndS(x,xˉ)d_{\mathcal{S}}(x^*, \bar{x}) = d_{\mathcal{S}}(T(x^*), T(\bar{x})) \le \rho \cdot d_{\mathcal{S}}(x^*, \bar{x}) = \rho \cdot d_{\mathcal{S}}(T(x^*), T(\bar{x})) \\ \le \rho^2 \cdot d_{\mathcal{S}}(x^*, \bar{x}) \le ... \le \rho^n \cdot d_{\mathcal{S}}(x^*, \bar{x})

  So we can conclude that x=xˉx^* = \bar{x}.

  Hence we have proved that if there is a contraction mapping on SS, then there exists a unique fixed point xx^* where we can find it by successively operating TT on any x0Sx_0 \in S; \blacksquare

  Finally we can define Banach space.

Definintion of Banach space) A complete normed vector space is called Banach space;

profile
Bayesian, System engineer, Evangelist

0개의 댓글