RadiiPolynomial

1.4 Radii Polynomials in Finite Dimensions

This section develops the radii polynomial method for proving existence of zeros in finite dimensions. The method provides explicit domains of existence and uniqueness.

Theorem 1.4.1 Fixed point theorem with radii polynomial (Theorem 2.4.1)

Consider a map \(T \in C^1(\mathbb {R}^n, \mathbb {R}^n)\) and let \(\bar{x} \in \mathbb {R}^n\). Let \(Y_0 \geq 0\) and \(Z: (0,\infty ) \rightarrow [0,\infty )\) be a non-negative function satisfying

\begin{equation*} \| T(\bar{x}) - \bar{x}\| \leq Y_0 \end{equation*}
\begin{equation*} \| \text{DT}(c)\| \leq Z(r), \quad \text{for all } c \in \overline{B}_r(\bar{x}) \text{ and all } r {\gt} 0. \end{equation*}

Define

\begin{equation*} p(r) := (Z(r) - 1)r + Y_0. \end{equation*}

If there exists \(r_0 {\gt} 0\) such that \(p(r_0) {\lt} 0\), then there exists a unique \(\tilde{x} \in \overline{B}_{r_0}(\bar{x})\) such that \(T(\tilde{x}) = \tilde{x}\).

Proof

The assumption \(p(r_0) {\lt} 0\) implies \(Z(r_0)r_0 + Y_0 {\lt} r_0\) and hence \(Z(r_0) {\lt} 1\).

Step 1: \(T\) maps \(\overline{B}_{r_0}(\bar{x})\) into itself. Let \(x \in \overline{B}_{r_0}(\bar{x})\). By the Mean Value Inequality:

\begin{equation*} \| T(x) - \bar{x}\| \leq \| T(x) - T(\bar{x})\| + \| T(\bar{x}) - \bar{x}\| \leq Z(r_0)\| x - \bar{x}\| + Y_0 \leq Z(r_0)r_0 + Y_0 {\lt} r_0. \end{equation*}

Step 2: \(T\) is a contraction on \(\overline{B}_{r_0}(\bar{x})\). For \(a, b \in \overline{B}_{r_0}(\bar{x})\), by the Mean Value Inequality: \(\| T(a) - T(b)\| \leq Z(r_0)\| a - b\| \). Since \(Z(r_0) {\lt} 1\), \(T\) is a contraction.

Step 3: Apply Contraction Mapping Theorem. By Theorem 1.1.3, there exists a unique \(\tilde{x} \in \overline{B}_{r_0}(\bar{x})\) with \(T(\tilde{x}) = \tilde{x}\).

Definition 1.4.2 Radii polynomial
#

Given constants \(Y_0, Z_0 \geq 0\) and a function \(Z_2: (0,\infty ) \rightarrow [0,\infty )\), the radii polynomial is defined by

\begin{equation*} p(r) := Z_2(r)r^2 - (1-Z_0)r + Y_0. \end{equation*}
Definition 1.4.3 Combined bound
#

The combined bound is defined as

\begin{equation*} Z(r) := Z_0 + Z_2(r) \cdot r. \end{equation*}
Lemma 1.4.4 Alternative form of radii polynomial

The radii polynomial can be rewritten as

\begin{equation*} p(r) = (Z(r) - 1)r + Y_0. \end{equation*}
Proof

Direct calculation: \((Z_0 + Z_2(r)r - 1)r + Y_0 = Z_2(r)r^2 + Z_0 r - r + Y_0 = Z_2(r)r^2 - (1-Z_0)r + Y_0\).

Lemma 1.4.5 Polynomial negativity implies contraction

If \(Y_0 \geq 0\), \(r_0 {\gt} 0\), and \(p(r_0) {\lt} 0\), then \(Z(r_0) {\lt} 1\).

Proof

By Lemma 1.4.4, \(p(r_0) = (Z(r_0) - 1)r_0 + Y_0 {\lt} 0\). Since \(Y_0 \geq 0\), we have \((Z(r_0) - 1)r_0 {\lt} 0\). Since \(r_0 {\gt} 0\), dividing gives \(Z(r_0) - 1 {\lt} 0\), i.e., \(Z(r_0) {\lt} 1\).

The main radii polynomial theorem (Theorem 2.4.2) is stated and proved in full generality for Banach spaces in Section 2.2 as Theorem 2.2.23. Here we apply it to the finite-dimensional case \(E = \mathbb {R}^n\).

1.4.1 Example 2.4.5: Finding \(\sqrt{2}\)

We demonstrate the radii polynomial method on the simplest nonlinear function \(f(x) = x^2 - 2\), verifying the existence of a unique zero near \(\bar{x} = 1.3\).

Theorem 1.4.6 Square root verification (Example 2.4.5)
#

Consider \(f(x) = x^2 - 2\). Choose initial guess \(\bar{x} = \frac{13}{10} = 1.3\), approximate inverse \(A = \frac{19}{50} = 0.38 \approx (f'(\bar{x}))^{-1} = (2\bar{x})^{-1}\), bounds \(Y_0 = \frac{3}{25} = 0.12\), \(Z_0 = \frac{3}{250} = 0.012\), \(Z_2 = \frac{19}{25} = 0.76\), and radius \(r_0 = \frac{3}{20} = 0.15\).

The radii polynomial \(p(r) = 0.76r^2 - 0.988r + 0.12\) satisfies \(p(0.15) {\lt} 0\).

Therefore, there exists a unique \(\tilde{x} \in \overline{B}_{0.15}(1.3) = [1.15, 1.45]\) with \(\tilde{x}^2 = 2\) and \(f'(\tilde{x}) = 2\tilde{x}\) invertible.

Proof

Step 1: Compute bounds. For the \(Y_0\) bound: \(|Af(\bar{x})| = |0.38(1.3^2 - 2)| = |0.38 \cdot (-0.31)| = 0.1178 \leq 0.12 = Y_0\).

For the \(Z_0\) bound: \(|1 - A \cdot 2\bar{x}| = |1 - 0.38 \cdot 2.6| = |1 - 0.988| = 0.012 = Z_0\).

For the \(Z_2\) bound: If \(c \in \overline{B}_r(\bar{x})\), then \(|A[f'(c) - f'(\bar{x})]| = |A(2c - 2\bar{x})| = 2|A||c - \bar{x}| \leq 2 \cdot 0.38 \cdot r = 0.76r = Z_2 r\).

Step 2: Verify polynomial negativity. \(p(0.15) = 0.76 \cdot 0.15^2 - 0.988 \cdot 0.15 + 0.12 = 0.0171 - 0.1482 + 0.12 = -0.0111 {\lt} 0\).

Step 3: Apply Theorem 2.2.23. The theorem guarantees a unique zero \(\tilde{x} \in [1.15, 1.45]\) with invertible derivative. Since \(\sqrt{2} \approx 1.414 \in [1.15, 1.45]\), this zero is \(\sqrt{2}\).

Corollary 1.4.7
#

There exists a unique \(\tilde{x} \in \overline{B}_{3/20}(13/10)\) with \(\tilde{x}^2 = 2\).

Remark 1.4.8

Optimal choice: If \(\bar{x} = \sqrt{2}\) (exact) and \(A = (2\bar{x})^{-1}\) (exact inverse), the radii polynomial becomes \(p(r) = \frac{\sqrt{2}}{2}r^2 - r\), giving \(\text{EI}(p) = (0, \sqrt{2})\).

Approximate choice: With \(\bar{x} = 1.3\) and \(A = 0.38\), the existence interval is approximately \(\text{EI}(p) \approx (0.136, 1.164)\), still sufficient to verify the zero.