next up previous
Next: Accuracy of HRFA Up: Accuracy Analysis of Hybrid Previous: Hybrid Rational Function Approximation

Approximate-GCD by Hribernig and Stetter

The approximate-GCD, which is called near-GCD in [8], of two polynomials $f_1, f_2 \in C[x]$, where C[x] is a complex number coefficient polynomial, is stated as the following concept.

Definition 3.1 (Hribernig and Stetter)   At the accuracy level $\alpha$, two polynomials $\tilde{f}_i\in C[x]$, i=1,2, possess a near-GCD $\tilde{g}$ if there exist polynomials $f^{\ast}_i\in C[x]$, i=1,2, which satisfy

 \begin{displaymath}GCD(f^{\ast}_1,f^{\ast}_2)=\tilde{g},\ and\ \Vert\tilde{f}_i-f^{\ast}_i\Vert\leq\alpha, i=1,2,
\end{displaymath} (5)

and deg $(\tilde{g})$ is the highest possible degree for (5). Equivalently, a near-GCD $\tilde{g}$ of $\tilde{f}_1$ and $\tilde{f}_2$satisfies

\begin{displaymath}\tilde{f}_i=\tilde{g}\cdot\tilde{q}_i+\tilde{r}_i\ with\ \Vert\tilde{r}_i\Vert\leq\alpha, i=1,2.
\end{displaymath} (6)

A near-GCD at accuracy level $\alpha$ will be denoted by $\alpha\mbox{-}GCD(\tilde{f}_1,\tilde{f}_2)$.

Here, the norm of the polynomial means L1 norm. A near-GCD $\tilde{g}$ is computed via Euclidean Algorithm and a relaxed termination criterion which should be imposed at a given accuracy level $\alpha$. The algorithm to compute $\alpha$-GCD of f1 and f2 is summarized as follows:

Algorithm 3.1 (tex2html_wrap_inline$$-GCD)  
Input: $f_1, f_2, \alpha$
Output: $\alpha$-GCD of f1 and f2
Method:
1.
Compute

\begin{displaymath}f_{j-1}=f_j\cdot q_j+f_{j+1}\end{displaymath}

and

\begin{displaymath}s_j^{(i)}=q_js_{j-1}^{(i)}+s_{j-2}^{(i)},\ j>i\end{displaymath}

for $j=2,3,\cdots$ and i=1,2 with

si-1(i)=0, si(i)=1.

2.
if $\Vert s_{j-1}^{(i)}f_{j+1}\Vert\leq\alpha$, then

\begin{displaymath}f_j=\alpha\mbox{-}GCD(f_1,f_2).\end{displaymath}

 

The polynomials fj generated by the algorithm, f1 and f2 are represented as

 \begin{displaymath}f_i=s^{(i)}_jf_j+s^{(i)}_{j-1}f_{j+1},\ j>i\geq 1.
\end{displaymath} (7)

In our case, f1 and f2 should be read as P(x) and Q(x), respectively. Also s(1)j,s(2)j, and fj should be read as p(x), q(x) and g(x), respectively. Therefore (7) is expressed as

 \begin{displaymath}\left\{\begin{array}{ccc}
P(x) &=& p(x)g(x)+\delta P(x),\\
Q(x) &=& q(x)g(x)+\delta Q(x),
\end{array}\right.
\end{displaymath} (8)

where $\delta P(x)=s^{(1)}_{j-1}f_{j+1}$ and $\delta Q(x)=s^{(2)}_{j-1}f_{j+1}$.


next up previous
Next: Accuracy of HRFA Up: Accuracy Analysis of Hybrid Previous: Hybrid Rational Function Approximation
IMACS ACA'98 Electronic Proceedings