Why the covariance matrix of outputs in PC-Kriging may not symmetric or positive definite

UQ with UQLab Community Q&A and How To
,
1

I build a PC-Kriging and calculate the covariance matrix of outputs of Monte-Carlo samples using uq_evalModel. But sometimes, the obtained covariance matrix may be not symmetric or positive definite.
image

Why? How can I solve this problem？