1: \begin{abstract}

2: An algorithm  for numerically computing  the exponential of a  matrix

3: is presented. We  have derived a  polynomial expansion of $e^x$  by

4: computing it  as an  initial value  problem using  a symbolic

5: programming language. This algorithm  is shown  to be  comparable in

6: operation  count and convergence with the  state--of--the--art method

7: which is  based on a Pade approximation of  the exponential matrix

8: function. The  present polynomial form,  however, is more  reliable

9: because the  evaluation  requires only  linear combinations  of the

10: input matrix.  We also show that  the technique  used to  solve  the

11: differential  equation, when implemented symbolically, leads to a

12: rational as well as a  polynomial form of the solution function. The

13: rational form is the well-known diagonal Pade approximation  of

14: $e^x$. The polynomial form, after some rearranging to minimize

15: operation count, will be used to evaluate the exponential of a matrix

16: so as to illustrate its advantages as compared with the Pade form. 

17: \end{abstract}

18: