The Similarity Transform Song

(to the tune of "If You're Happy and You Know It")

Note: This song describes what to do for a similarity transformation to find the eigenvalues of a matrix A. A similarity transform changes the matrix but does not alter its eigenvalues. Ideally, if you could find a similar matrix that was diagonal, you could pick the eigenvalues right off the diagonal. A similarity transform is of this form: B = T^-1 A T, where T is the similarity transform matrix, and B is the matrix that is similar to A.

If your matrix has distinct eigenvalues,
It is very much apparent what to use:
Choose nonsingular for T,
and get a diagonal B,
If your matrix has distinct eigenvalues.

If you have a real symmetric matrix A,
Choose orthogonal for T: it's the best way!
Because, after all
B is real diagonal
If you have a real symmetric matrix A.

If it's complex and Hermitian, never fear!
The solution to your problem is quite clear:
Use a unitary T,
to get a real diagonal B
If it's complex and Hermitian, never fear!

If your matrix A is normal as can be,
Take the following advice from me:
If your T is unitary
B's diagonal (how scary!)
If your matrix A is normal as can be.

If A is chosen arbitrarily,
and you find a T that's unitary,
Then B has the form of Schur
(that's upper triangular)
If A is chosen arbitrarily.

If A is once again arbitrary,
Then you choose "nonsingular" for T,
B is almost diagonal,
(that's Jordan form, y'all!)
If A is once again arbitrary.

"The Similarity Transform Song" Copyright (c) 2000-2007 Rebecca Hartman-Baker.

Last updated January 22, 2007

hartmanbakrj@ornl.gov