MathGroup Archive 2003

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Matrix functions

  • To: mathgroup at smc.vnet.net
  • Subject: [mg42422] Re: Matrix functions
  • From: "Richard I. Pelletier" <bitbucket at attbi.com>
  • Date: Sun, 6 Jul 2003 06:57:04 -0400 (EDT)
  • References: <be5ua2$jhu$1@smc.vnet.net>
  • Sender: owner-wri-mathgroup at wolfram.com

In article <be5ua2$jhu$1 at smc.vnet.net>, "Dr. Wolfgang Hintze"
<weh at snafu.de> wrote:

> We can define functions of an nxn matrix M as (infinite) series and 
> study their convergence using a suitable matrix norm and/or numerically.
> ....
> Question: Are there standard functions (within some package) of this 
> kind available? If not, what would be the best way to define e.g. the 
> exponential matrix function (2), i.e. how to terminate the series 
> appropriately?

there are two std functions available: MatrixExp will compute the
exponential of a matrix; MatrixPower will compute a power (e.g. sqrt)
of a matrix.

for other matrix functions, you have at least three choices other than
summing the infinite series. 

   (1) the spectral decomposition theorem, which says that a normal
transformation can be wriiten as a linear combination of orthogonal
projections, with its eigenvalues as coefficients, and that functions
(i.e. power series) of it can be computed by applying those functions
just to the eigenvalues.

   if A = lambda1 E1 + lambda2 E2
then f(A) = f(lambda1) E1 + f(lambda2) E2

(2) since a matrix A satisfies its own characteristic equation, any
infinite series in A collapses to a polynomial. that is, if the minimal
polynomial of A is of degree 2, then

   sqrt(A) = aI + bA

for some a and b. (solving for a and b is easily done since the
eigenvalues of A must also satisfy an eq of that form. ok, if there are
repeated eigenvalues, we have to match derivatives too.)

(3) easiest of all is that if the matrix can be diagonalized (D =
Pinv.A.P) then f(A) = P f(D) pinv.

i hope this outline helps you find more detail.

rip

-- 
NOTE: email address is rip1 AT comcast DOT net


  • Prev by Date: Why is it so???
  • Next by Date: Re: Speed improvements in Mathematica 5 ??
  • Previous by thread: Matrix functions
  • Next by thread: Re: 3D plot of error function of neural network