1: \begin{abstract}

2: We consider three mathematically equivalent variants of the conjugate gradient

3: (CG) algorithm and how they perform in finite precision arithmetic.  It was

4: shown in [{\em Behavior of slightly perturbed Lanczos and conjugate-gradient

5: recurrences}, Lin.~Alg.~Appl., 113 (1989), pp.~7-63] that under certain

6: coditions, that {\em may} be satisfied by a finite precision CG computation,

7: the convergence of that computation is like that

8: of exact CG for a matrix with many eigenvalues distributed throughout tiny

9: intervals about the eigenvalues of the given matrix.  We determine to what

10: extent each of these variants satisfies the desired conditions, using a set

11: of test problems, and show that there is significant correlation

12: between how well these conditions are satisfied and how well the finite

13: precision computation converges before reaching its ultimately attainable

14: accuracy.  We show that for problems where the interval width makes a

15: significant difference in the behavior of exact CG, the different CG variants

16: behave differently in finite precision arithmetic.

17: For problems where the interval width makes little difference or where the

18: convergence of exact CG is essentially governed by the upper bound based

19: on the square root of the condition number of the matrix, the different CG

20: variants converge similarly in finite precision arithmetic until the ultimate

21: level of accuracy is achieved.

22: \end{abstract}

23: