1: \begin{abstract} 

2: The speed of convergence of the $\mathbb{R}$-linear GMRES 

3: method is bounded in terms of a

4: polynomial approximation problem on a finite subset of the spectrum. 

5: This result resembles the classical GMRES convergence estimate

6: except that the matrix involved is assumed to be condiagonalizable.

7: The bounds obtained are applicable to the CSYM method, in which case they

8: are sharp. Then

9: a three term recurrence for generating a family of 

10: orthogonal polynomials is shown to exist, yielding a natural link with complex

11: symmetric Jacobi matrices. 

12: This shows that a mathematical framework analogous to the one

13: appearing with the Hermitian Lanczos method

14: exists in the complex symmetric case.

15: The probability of being  condiagonalizable is 

16: estimated with random matrices.

17: \end{abstract}

18: