I would like to prove that for any positive semidefinite matrix $\Sigma_{n\times n}$ and any vector $\mu_{n\times1}$, are there always data points (no matter how many, as long as finite) whose empirical covariance matrix is $\Sigma_{n\times n}$ and the empirical vector is $\mu_{n\times1}$. What is the way to prove it rigorously?
Asked
Active
Viewed 27 times
0
-
Feels like it could be shown by giving a way to construct such a sample - easy along one dimension, perhaps not much more difficult with more dimensions? – Björn Jun 16 '16 at 05:37
-
@josliber and Björn thanks for your comments! I figured out that by Cholesky decomposition, one can easily show $\Simga$ and $\mu$ can be reconstructed by $n+1$ data points. And I will mark this question as duplicated. – user112758 Jun 16 '16 at 05:46