Evaluation of matrix element of a linear differential matrix equation by constrained least square fitting

Hi All,

I was trying to evaluate the matrix coefficient by the method of constrained least square fitting. The linear matrix differential equation looks like X'=AX, where X' and X are two vectors and A is matrix. Moreover, there is a constraint on diagonal matrix element and which is a(i,i)=-Sum(i/=j)a(i,j). The elements of X' and X vectors are known. I was trying to use MATLAB to do that. But I didn't figure out which figure out which MATLAB function is suitable for that. Can you guide me in this regard.

Thanks in advance Sudipta

Answer 1

[Richard Brown]

Assuming you have multiple X and X' vectors that you're trying to fit your matrix to, your problem can be posed as a constrained linear least squares problem in the entries of A, which can be solved using lsqlin

[Richard Brown]

Please reply to answers with comments, not further answers. I won't write you a script to do it, but it's pretty straightforward - I suggest you do it with pen and paper first.

Treat the entries of A as a vector comprised of its columns stacked together. Each row of each X' = A*x equation is linear in the entries of A. It just takes a bit of thought and messing around to construct the right matrices.

The diagonal sum constraint is an equality constraint.

[Sudipta Sinha]

Thanks for your continuing help. I am facing a confusion. The function lsqlin function evaluates the values of X. But I need the matrix elements. The arguments of the lsqlin function are A, X' and the constraint equation and it evalutes the X vector. In my case the matrix is unknown and need to evaluate that. So what will be the format of matlab lsqlin function. Another thing if Sum a(i,j)=0 then the constraint equation look like a11+a12+....a1n=0 and so on for other rows. But what will be the matlab lsqlin function format of this constraint equation.

[Richard Brown]

The vector that you get out from lsqlin will be the columns of A, stacked into a vector. The matrix Ahat that you provide to lsqlin will be big (it's not A). For example, assuming your vectors have m entries and A is square, the first m rows of the matrix Ahat you provide to lsqlin will be

tmp = repmat({x1},m,1);
Ahat(1:m,:) = blkdiag(tmp{:})

where x1 is your first X vector. Likewise the first m entries of the RHS will be xp1 (your first X' vector)

Answer 2

[Sudipta Sinha]

Thanks for your reply. Yes, I have multiple x(1).....x(n) and x'(1).....x'(n) vectors. Can you provide a small script for that? That will be great for me.