Every square matrix can be expressed as a linear combination of
commuting projection matrices plus a nilpotent matrix.  Specifically,
given a square matrix M, there are scalars z1, z2, ... zn and matrices 
P1, P2, ... , Pn, D, such that

(1) M = z1 P1 + z2 P2 + ... + zn Pn + D,
(2) spectrum(M) = {z1,z2, ..., zn},
(3) Pi Pj = Pj Pi = kron_delta(i,j) Pi, for 1 <= i,j <= n,
(4) D is nilpotent,
(5) D Pi = Pi D, for 1 <= i <= n,
(6) P1 + P2 + ... + Pn = I.

We call (1) the spectral representation of M.  Recall that a matrix
is D is nilpotent means that there is a positive integer k such that
D^k = 0.

The spectral representation of a matrix is not a continuous function of
the matrix arguments. Here is an example

(C1) spectral_rep(matrix([1,0],[0,x]));

Proviso: assuming x-1 # 0
			     [ 0  0 ]  [ 1  0 ]	  [ 0  0 ]
(D1) 		   [[x, 1], [[ 	    ], [      ]], [ 	 ]]
			     [ 0  1 ]  [ 0  0 ]	  [ 0  0 ]

Maxima does give a warning that it has assumed that x does not equal 1.
Unfortunately, the assumption x-1 # 0 isn't available to the user. When  
x = 1, the spectral representation is 

(C2) spectral_rep(matrix([1,0],[0,1]));

				 [ 1  0 ]   [ 0	 0 ]
(D2) 			  [[1], [[      ]], [ 	   ]]
				 [ 0  1 ]   [ 0	 0 ]
(C3) 

Functions:

spectral_rep(mat)

When mat is a square matrix, return a list of the form 

   [[z1,z2, ... zn], [P1, P2, ... ,Pn], D],

where z1 P1 + z2 P2 + ... + zn Pn + D is the spectral representation of M.

matrixexp(mat, t)

When mat is a square matrix, return exp(mat * t).  The second argument
is optional and it defaults to 1.

matrixfun(lambda-expr, mat)

When mat is a square matrix and lambda-expr is a Maxima lambda expression,
extend the function lambda-expr to matrices and return lambda-expr(mat). 
Thus matrixfun(lambda([x],x^2), mat) == mat . mat and  
matrixfun(lambda([x], 1/x), mat) == mat^^-1.  

If the lambda-expr involves an unknown function, you may need to load 
pdiff; otherwise, you'll get an error. Here is an example

(C1) load("matexp.lisp")$
(C2) m : matrix([1,2],[0,1])$
(C3) matrixfun(lambda([x], f(x)),m);

Attempt to differentiate with respect to a number:
1 -- an error.  Quitting.  To debug this try debugmode(true);
(C4) load("pdiff")$
(C5)  matrixfun(lambda([x], f(x)),m);

			      [ f(1)  2 f   (1) ]
(D5) 			      [ 	 (1)    ]
			      [		        ]
			      [	 0      f(1)    ]
(C6) 


Barton Willis wrote matexp. This code is covered by the GNU public
license. Please send bug reports to the Maxima mailing list.