![]() |
We consider a test point
having the following coordinates:
![]() |
We assume we know only
and
and want to compute the unknown
and
.
Equations (6) give us the tangent of the half slope:
Applying equation (7), we get the signed distance
between
and the intersection of the first line and the ellipse:
>From this value of
, we deduce the first coefficients
,
,
and
of the quartic (we don't compute
). We reduce it to a cubic using the already known root
:
![]() |
The middle point between
and
is the
best approximation we have at the end of this first iteration:
We know an upper bound for the error of this approximation is the
half-width of the interval, which is
(in fact the
error is about 30 times smaller). As we consider this is too much, we
compute the starting values for a second iteration:
The steps of the second iteration are similar to the steps of the first one. They lead to the following result:
So at the end of the second iteration, we have an estimate
and an upper bound of its error of
(the real error being almost 900 times
smaller).
Luc Maisonobe 2006-02-04