We use norms for error analysis. The natural question arises: which norm should we use?
Example: Given in meters and
in meters. Then
is a good approximation to
because the rel. error
And
is considered a bad approximation as
But suppose the first component of each vector is measured in kms. In this case,
. The relative error measured
as:
If we use a different norm, say the 1-norm, the results would be different.
Definiton: Let B be real linear space,
:
is an inner product if
Example: Over ,
Lemma: Cauchy-Schwartz: .
Lemma: is a norm.
Definition: are orthogonal.
Lemma: Let and
be two norms on
, then
constants
such that
.
For some particular norms, the following results are useful: