How to get rid of Nan in Pseudo-inverse matrix

[Mahmoud Zeydabadinezhad]

Hi all,
To solve a system of linear equations I have to compute the matrix inverse. I did use both the inv(A'*A)*A' and pinv(A) but in all cases I got some NaNs.
I read some related posts here but finally couldn't come up with a consistent answer to my questions?
Does anybody know another method for solving a system of linear equations?
Is there anyway to avoid this NaNs in matrix inverse computing ?

Thank you,

[Bruno Luong]

"Mahmoud Zeydabadinezhad" <zeyd...@yahoo.com> wrote in message <ha5i8b$frr$1...@fred.mathworks.com>...

Does your matrix contain NaN?

Check with the command
find(isnan(A(:)))

Bruno

[Mahmoud Zeydabadinezhad]

> Does your matrix contain NaN?
>
> Check with the command
> find(isnan(A(:)))
>
> Bruno

No, The original matrix A, doesn't have any NaN value. the result of pinv(A) or pseudo inverse matrix of A have some NaN values.

[Bruno Luong]

"Mahmoud Zeydabadinezhad" <zeyd...@yahoo.com> wrote in message <ha5mqj$gou$1...@fred.mathworks.com>...

> No, The original matrix A, doesn't have any NaN value. the result of pinv(A) or pseudo inverse matrix of A have some NaN values.

Strange. I think it is a bug.

Try then my PSEUDOINVERSE of Fex if your Matlab has oriented object feature:

http://www.mathworks.com/matlabcentral/fileexchange/25453

Bruno

[Bruno Luong]

What this command gives?

norm(A,'fro')

Bruno

[Tim Davis]

"Mahmoud Zeydabadinezhad" <zeyd...@yahoo.com> wrote in message <ha5i8b$frr$1...@fred.mathworks.com>...

Why compute the inverse to solve a linear system? inv(A'*A)*A' is a really bad way to solve the normal equations. Just use x=A\b.