Next: Error Bounds for Computed
Up: Stability and Accuracy Assessments
Previous: Residual Vector.
  Contents
  Index
Transfer Residual Error to Backward Error.
There are Hermitian matrices
such that
and
are an exact eigenvalue and
its corresponding eigenvector of
, i.e.,
One such
is
![\begin{displaymath}
E = -r\wtd x^*-\wtd x r^* +\left(\wtd x^*A\wtd x -\wtd\lambda\right)
\wtd x\wtd x^*.
\end{displaymath}](img1451.png) |
(61) |
We are interested in such matrices
with smallest possible norms.
It turns out the best possible
for the spectral norm
and the best possible
for Frobenius norm
satisfy
![\begin{displaymath}
\Vert E_2\Vert _2=\Vert r\Vert _2 \quad \mbox{and} \quad
\Ve...
...t r\Vert^2_2 - \left(\wtd x^* A\wtd x
-\wtd\lambda\right)^2};
\end{displaymath}](img1456.png) |
(62) |
see, e.g., [256,431].
In fact,
is given explicitly by (4.52).
So if
is small, the computed
and
are an exact eigenpair of nearby matrices. Error analysis of
this kind is called backward error analysis, and
matrices
are backward errors.
We say an algorithm that delivers an approximate
eigenpair
is
-backward stable for the pair with respect to the norm
if it is an exact eigenpair for
with
.
With this definition in mind,
statements can be made about the numerical stability of the algorithm which
computes the eigenpair
.
By convention, an algorithm is called backward stable
if
, where
is the machine
precision.
Next: Error Bounds for Computed
Up: Stability and Accuracy Assessments
Previous: Residual Vector.
  Contents
  Index
Susan Blackford
2000-11-20