Re: Matrix derivative
- From: "seppl" <basti2@xxxxxxxxxxxxxx>
- Date: 23 Dec 2006 02:45:34 -0800
Thanks to all of you.
I forget to mention that A is symmetric and positive semidefinite (if
thats important here).
Thanks for the literature-hints but until now I didn't get the final
idea to solve this.
By the way, the whole formula is
tr( A * d/dA logA)
and I guess this has to evaluate to the identity.
Herman Rubin schrieb:
In article <1166789958.652578.169980@xxxxxxxxxxxxxxxxxxxxxxxxxxx>,
seppl <basti2@xxxxxxxxxxxxxx> wrote:
Hi all,
I'd like to know the derivative of
log A w.r.t A ,where log is the matrix logarithm and how to compute it.
Generally, one uses differentials. But log(A+dA) is not
an easy expression otherwise.
To see this, consider log (I - X - dX), where X is small
(all characteristic roots less than 1 in absolute value.
This is -\sum (X+dX)^n/n. Now if we expand this, and
just keep the first order terms in dX, we find the
negative of the differential of the logarithm is
dX + (X*dX + dX*X)/2 + (X^2*dX + X*dX*X + dX*X^2)/3 + ...
This does not simplify unless X and dX commute.
Furthermore I would like to know if there is a good online-available
reference for matrix derivatives.
Thanks in advance.
Seppl
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@xxxxxxxxxxxxxxx Phone: (765)494-6054 FAX: (765)494-0558
.
- References:
- Matrix derivative
- From: seppl
- Re: Matrix derivative
- From: Herman Rubin
- Matrix derivative
- Prev by Date: Re: How can I order a aggregation of items from multiple lists
- Next by Date: Re: Comparing fractions (or proportions)
- Previous by thread: Re: Matrix derivative
- Next by thread: Simple Linear Regression and Area
- Index(es):
Relevant Pages
|