(42)

I discussed shifting eigenvalues in Chapter 5. If the eigenvalues of Sx are ?1, ···, ?n. then the eigenvalues of (Sx+a I) are ?1+ a , …, ?n+a . So the inverse given by

equation (45) has been achieved by shifting the eigenvalues of S by an amount sufficient to make R well behaved.