Let T be as in the preceeding section, and let be a solution to
We consider the inverse problem of recovering the diagonal matrix Q in from
Again we put with L from Theorem 2.1. We obtain
With the recursion can be written as
We claim that for and
where
We prove (5.3) by induction with respect to k. The case k = 0 is obvious. Assume (5.3) to be correct up to some k with . From (5.2) we get
Since S is tridiagonal, only components 1 through n-k of enter for i < n - k. Hence, by the induction hypothesis,
This is (5.3) with k replaced by k+1. Thus (5.3) is established.
We use (5.3) for i = 1 only, yielding for
Introducing the row vectors
we have
and, since S, commute,
Thus is the analogue to with T replaced by :
This shows that the matrix is upper triangular with diagonal elements
It follows that the systems
where are uniquely solvable for provided that , .
Given that , we can determine the matrix from our data. The matrix is upper triangular with diagonal elements
Thus computing L, U amounts to doing an LU-Decomposition on the matrix Z,
with the diagonal of U being known. Once L, U are known, Q is computed very much in the same fashion as in the preceding section.