1: \begin{abstract}

2:         We consider deflation and augmentation techniques for accelerating the

3:         convergence of Krylov subspace methods for the solution of nonsingular

4:         linear algebraic systems.  Despite some formal similarity, the two

5:         techniques are conceptually different from preconditioning.  Deflation

6:         (in the sense the term is used here) ``removes'' certain parts from the

7:         operator making it singular, while augmentation adds a subspace to the

8:         Krylov subspace (often the one that is generated by the singular

9:         operator); in contrast, preconditioning changes the spectrum of the

10:         operator without making it singular.  Deflation and augmentation have

11:         been used in a variety of methods and settings.  Typically, deflation is

12:         combined with augmentation to compensate for the singularity of the

13:         operator, but both techniques can be applied separately.

14: 

15:         We introduce a framework of Krylov subspace methods that satisfy a

16:         Galerkin condition. It includes the families of orthogonal residual (OR)

17:         and minimal residual (MR) methods.  We show that in this framework

18:         augmentation can be achieved either explicitly or, equivalently,

19:         implicitly by projecting the residuals appropriately and correcting the

20:         approximate solutions in a final step.

21:         We study conditions for a breakdown of the deflated methods, and we show

22:         several possibilities to avoid such breakdowns for the

23:         deflated \MINRES{} method. Numerical experiments illustrate properties

24:         of different variants of deflated \MINRES{} analyzed in this paper.

25: 	\end{abstract}

26: