I want to create a normal distribution with a MU = 0 and SIGMA = cov(A); A is a 4000*5460 matrix. I get this error: SIGMA must be a square, symmetric, positive definite matrix.
Since the error is related to the Cholesky transformation and negative eignvalues, I used SVD and retained first positive 745 eigenvalues which describe 95% of variance. Using this new SIGMA, I get the same error.