Path: news.mathworks.com!not-for-mail
From: "leo nidas" <bleonidas25@yahoo.gr>
Newsgroups: comp.soft-sys.matlab
Subject: matrix derivative
Date: Fri, 19 Mar 2010 10:04:06 +0000 (UTC)
Organization: The MathWorks, Inc.
Lines: 44
Message-ID: <hnvi6m$2jn$1@fred.mathworks.com>
Reply-To: "leo nidas" <bleonidas25@yahoo.gr>
NNTP-Posting-Host: webapp-03-blr.mathworks.com
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
X-Trace: fred.mathworks.com 1268993046 2679 172.30.248.38 (19 Mar 2010 10:04:06 GMT)
X-Complaints-To: news@mathworks.com
NNTP-Posting-Date: Fri, 19 Mar 2010 10:04:06 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 1194208
Xref: news.mathworks.com comp.soft-sys.matlab:618188


I am confused with something rather simple which is embarrassing..

Let G=A(Y-&#956;). 

G is p by 1 matrix depending on &#946;=[&#946;0 &#946;1 &#946;(p-1)]'. (&#946; is p by 1)
&#913; is p by n matrix depending on &#946;
&#956; is n by 1 matrix depending on &#946;
&#933; is n by 1 matrix independent of &#946;

I thought the derivative of G with respect to &#946; (let dG) would be 

dG=(dA)*(Y-&#956;)-&#913;*d&#956;.

which is obviously wrong because the dG should be a p by p matrix.

&#913;*d&#956; is p by p (which is encouraging!)
but (Y-&#956;) is n by 1 and the dimensions would not agree.

I checked the following:

http://en.wikipedia.org/wiki/Matrix_calculus

but did't get it..

I checked my calculus book of Ross and Finley and doesn't have anything.
I checked my algebra book of Gilbert Strang and doesn't have anything.
I searched the net with no luck
and I decided to make an example in matlab, so just see how can I get the derivative of a matrix with respect to a vector I typed

>> syms b0 b1 real
>> b=[b0 b1]';
>> A=[b0*b1 b0*b1];
>> diff(A,'b')
 
ans =
 
[ 0, 0]

which is not what I want. diff just fount=d the difference of the elements right? I would like the 2 by 2 derivative matrix of A with respect to  b.

My final goal is to get right the derivative of G=A(Y-&#956;) right. 

Thanx in advance for any answers.