It is often desirable to work with transformations
that are as nearly unitary as possible,
i.e., their transpose is their pseudoinverse.
These transformations are pseudounitary.
Let us make NMO with *linear* interpolation
into a pseudounitary transformation.
We need to factor the tridiagonal
matrix into bidiagonal parts,
.One such factorization is the well-known **Cholesky decomposition**;
which is like spectral factorization.
(We never really need to look at square roots of matrices).
1

10/21/1998