Hi,
I tried to run a VPC with a covariance term between Volume and clearance but got an error message. It ran ok without the covariance so is this a known problem or specific to this project?
Thanks
Teresa
Posted 05 March 2015 - 03:19 PM
Dear Teresa
Your variance covariance matrix is not positive definite.
A correlation between V and Cl must be between -1 and 1
We have
correlation=covariance/(product of standard deviations)
Here standard deviation=square root of 0.1=0.3162
Therefore product of stdev=0.1
OK now correlation=0.3/0.1=3 >1
Therefore the variance covariance is not positive definite.
If you use let say covariance of 0.03, it should work.
in summary, there is nothing in the program. The problem is that mathematically you cannot put any covariance value in a variance covariance matrix and expecting to simulate random value form the corresponding normal distribution.
Only variance covariance matrices that are positive definite can be used for simulations.
Best Regards
Serge
0 members, 0 guests, 0 anonymous users