SIGMA must be a symmetric positive semi-definite matrix error in matlab

HI all, I have been trying to use the mvnrnd function to generate samples of alpha using the truncated gaussian distribution.mvnrnd function needs sigma which must be positive semi-definite and symmetric.My matrix is 1.0e-006 *

    0.3822         0         0    0.1831         0         0         0
         0    0.2081    0.0417         0    0.0759         0    0.0426
         0    0.0417    0.3505         0         0    0.1096         0
    0.1831         0         0    0.2050         0    0.0492         0
         0    0.0759         0         0    0.0721         0    0.0220
         0         0    0.1096    0.0492         0    0.1952         0
         0    0.0426         0         0    0.0220         0    0.0903
the matrix is as per requirements bt still i m still getting the error.Any suggestions will be of great help.
Thanks 
wid regards,
Ekta

Answer 1

[Walter Roberson]

Your matrix is probably not exactly right due to round-off. For example have a look at

YourMatrix - YourMatrix.'

and you will likely see some non-zero values.

The usual trick is to use

YourMatrix = (YourMatrix + YourMatrix.') / 2;

[emre parlak]

i am sorry, i know this is an old comment but i couldnt find the difference on the internet.

what is the difference between YourMatrix' and YourMatrix.' ?

[Bruno Luong]

xxx' Conjugate transpose

xxx.' transpose

The correct formula is Conjugate transpose is your matrix is complex (and Hermitian).

YourMatrix = (YourMatrix + YourMatrix') / 2;

The correction using simple transpose is wrong.

It doesn't matter if matrix is real, hence OK for random gaussian covariance matrix.

[emre parlak]

thank you. it didnt solve my problem though

[Bruno Luong]

More complete discussion, see here

[emre parlak]

thank you very much

[Abinaya Sankarakumar]

I use mvnrnd to generate a normal distribution . But when I find its mean I get different value from the one I used . Why is it so ?