Next: Bibliography
Up: Exact Real Computation in
Previous: Vectors, matrices
Let
be a nonsingular matrix with a known
approximation
. We
determine first the forward error:
|
(1) |
We have to find an upper bound for
, so by using
inequality 1 we get
yielding
whenever
. Because is nonsingular,
converges to
as goes
to zero, thus the previous condition can be fulfilled within finitely
many iterative refinements of the approximation of .
Assuming that
is an approximation with
, for the forward error we get
In the backward error computation, given
, we determine
such that
implies
. We apply inequality 1 for and
:
and because
,
using the upper bound for
, we write:
If we choose
such that the above product is
less than
, we can compute an
-approximate
for from
. For instance, we may set
Next: Bibliography
Up: Exact Real Computation in
Previous: Vectors, matrices