Multiplying (73) by an arbitrary weight function v(x) and
integrating over the interval [a,b] one obtains
Evidently (73) and (75) are equivalent, because v(x) is an arbitrary
function. Now we seek a numerical solution to (75), (74) in the form
Here , ...,
are functions of x and a1,
..., an are unknown coefficients.
In vector form (76) becomes
where
In (75) we may substitute u by u* to obtain
However, substituting u(x) by its approximation u*(x) in
(73), generally it appears that (73) is not satisfied exactly, e.g.
Here e(x) is a measure for the error.
It follows from (79)-(80) that
Obviously, the residual, e(x), depends on the unknown parameters
given by vector . Therefore the coefficients a1, ..., an
must be determined so, that expression (81) is satisfied.
Generally
where V1, ..., Vn are known functions of x and c1, ..,
cn are certain parameters. In terms of vector notation
(82) reads
where
Evidently
and therefore (see (81))
Relation (86) holds for arbitrary cT- matrices, i.e.
or
Now, we have n equations (88) to determine coefficients
a1,...,an. Inserting (77) in (80) yields
and the condition (86) can be rewritten as
Introducing the matrix K and the vector as
we can write (90) in compact form
Finally, we have n linear equations (92) for determing n
coefficients a1,...,an.