Fixed Point and ODE
Discrete Dynamical Systems
Discrete Dynamical Systems Let \(X\subseteq (V,\|\cdot\|), T:X\rightarrow X\) continuous. \((X,T)\) is a discrete dynamical system
Forward orbit \(\forall x\in X\), forward orbit of x is the sequence \(O(x):=\{T^n_x\}_{n\geq 0}\), \(T^n_x=T(T...T(x))\)
Fixed Point \(x^*\) is
attractive fixed point (sink) if \(\exists (a,b)\ni x^*\) s.t. \(\forall x \in (a,b). T^N_x\rightarrow x^*\)
repelling fixed point (source) if \(\forall x\neq x^*\), \(O(x)\) leaves \((a,b)\)
Contraction \(X\subseteq (V,\|\cdot\|),T:X\rightarrow X\) is a contraction on \(X\) if
i.e. \(T\) is \(c\)-Lipschitz \(c<1\)
Banach Contraction Principle
Let \(X\subseteq (V,\|\cdot\|)\) be a closed subset of a complete normed vector space,
IF \(T:X\rightarrow X\) is a contraction on \(X\)
THEN
(i) \(T\) has unique fixed point \(x^*\)
(ii) \(\forall x\in X. x^* = \lim_{n\rightarrow\infty} T^nx\)
(iii) \(\|T^nx-x^*\|\leq c^n\|x-x^*\|\leq \frac{c^n}{1-c}\|x-Tx\|, c\) is the Lipschitz constant
proof. Let \(x_0\in X\), recursively define \(x_{n+1} = Tx_n\), hence form a sequence \(\{x_n\}\).
Then, we can show that \(\{x_n\}\) is Cauchy by observing
Since \(c^n\rightarrow 0\), we can choose \(N\) sufficiently large to have \(\|x_{n+m}-x_n\|\rightarrow 0\)
Existence By completeness of the normed space, Cauchy implies convergent to some \(x^*\in V\), by \(X\) closed, \(x^*\in X\). By Lipshitz, hence continuous of \(T\), \(Tx^* = T(\lim x_n) = \lim Tx_n = x^*\Rightarrow x^*\) is the fixed point.
Uniqueness Suppose \(y\in X\) is also fixed point, \(\|x^*-y\| = \|Tx^* - Ty\| \leq c\|x^*-y\|\Rightarrow \|x^*-y\| = 0\Rightarrow x^*=y\)
Solving ODE
Example given the initial value problem \(f'(x)=1+x-f(x), |x|\leq 1/2\) and \(f(0)=1\)
First convert to the integral problem
Define \(T\) on \(C[-1/2,1/2]\) and \(Tf = f\) and the solution of the integral equation is a fixed point of \(T\) Then, to show \(T\) is a contraction
Thus, by Banach Contraction Principle, choose \(f_0 := 1\in C[-1/2,1/2]\), then
Since \(|x|\leq 1/2\), using M-test, this power series is uniform convergence, and note that \(f_n(x):=e^{-x} + x\)
Thrm. Existence of ODE
If \(\Phi:[a,b]\times \mathbb R^n\rightarrow \mathbb R^n\) is Lipschitz on the second coordinate \(y\in\mathbb R^n\).
Then \(\exists F_*\in C([a,b],\mathbb R^n)\) s.t. \(F_*=TF_*=\Gamma + \int_a^x \Phi(t,F_*(t))dt\)
proof. (Picard iteration)
Recursively define \(F_0(x)=\Gamma, F_{k+1}(x)=TF_k = \Gamma + \int_a^x \Phi(t, F_k(t))dt\)
Consider \(F_{k+1}-F_k\), let \(M = \|\Phi(t,\Gamma)\|_\infty\), by EVT, since \([a,b]\) compact, \(M\) is the maximum.
Then for any \(x\in [a,b]\). \(\|F_1(x) - F_0(x)\| = \int_a^x \Phi(t,\Gamma)dt\leq M(x-a)\)
Then, for \(n\geq N\), for some large \(N\)
Since the infinite sum is finite, the series is Cauchy. so that \(F_n\rightarrow^{u.c.}F_*\in C([a,b], \mathbb R^n)\) by compactness and closeness.
Therefore, \(TF_* = T\lim F_n = \lim TF_n = F_*\) using continuity
Uniqueness Suppose exists \(G_*\), then
Example
\(f'(x)=f(x), f(0)=1, x\in [0,1]\)
proof. Let \(F_0(x)=1, F_1(x) = 1 + \int_0^x F_0(t)dt = 1 + \int_0^x dt = 1 + x\)
Assume \(F_n(x) = \sum_0^k \frac{x^m}{m!}\), then
Therefore by induction, \(F_*(x) = \lim_nF_n(x) = \sum^\infty x^m/m! = e^x\)