covariance matrix data sampling

[Pasquale Tricarico <tricaric at pd dot infn dot it>]

Hi All,

 The task is very simple: if you have a set of experimental data, you 
can compute mean values and standard deviation for each derived physical 
quantity. The covariance matrix can also be computed without any 
problem. If the number of physical quantities is i.e. 6, we will have 6 
mean values with the corresponding standard deviations, and the 
covariance matrix will be a 6x6 real symmetric matrix. (I have in mind 
an orbit, with 6 free parameters, and a set of sky observations).
 Now suppose that you want to generate, using a program based on the 
GSL, one million of data points using the covariance matrix. Those 
points must 'agree' with the experimental data in the sense that mean 
values, standard deviation and covariance matrix computed using only the 
generated data points must be as close as possible to the original 
experimental one. (In the orbit analogy, the generated data points 
represent a plausible orbit given the observations.) This is a common 
problem in many simulation programs, and at the moment I use a 'not very 
reliable' set of source code found somewhere in the net to achieve this, 
but I'm not satisfied with this solution.
 I don't know of any GSL function call to achieve this. Any solution?

 Thanks.

  --Pasquale
    http://orsa.sf.net