1: \begin{abstract} 

2: Given an $n$ by $n$ matrix $A$ and an $n$-vector $b$, along with a rational function

3: $R(z) := D(z )^{-1} N(z)$, we show how to find the optimal approximation to $R(A) b$

4: from the Krylov space, $\mbox{span}( b, Ab, \ldots , A^{k-1} b)$, 

5: using the basis vectors

6: produced by the Arnoldi algorithm.  To find this optimal approximation requires

7: running $\max \{ \mbox{deg} (D) , \mbox{deg} (N) \} - 1$ extra Arnoldi steps and

8: solving a $k + \max \{ \mbox{deg} (D) , \mbox{deg} (N) \}$ by $k$ least squares problem.

9: Here {\em optimal} is taken to mean optimal in the 

10: $D(A )^{*} D(A)$-norm.  Similar to the case for linear systems, we show that 

11: eigenvalues alone cannot provide information about the convergence behavior of this 

12: algorithm and we discuss other possible error bounds for highly nonnormal matrices.

13: \end{abstract}

14: