2
$\begingroup$

I'm trying to construct an error ellipsoid from a covariance matrix (which exists for a 3D point) and then sample consistent xyz points in this region. (This question succeeds this one.)

What I'm currently doing is:

1) Calculate the Cholesky decomposition of the covariance matrix.
2) Sample each initial vertex point as a Gaussian with width 1 to generate (x', y', z')
3) Multiply (x',y',z') by the Cholesky decomposition matrix for the newly generated point.
4) Add this result to a matrix of the mean values.

To add some more concrete details, the covariance matrix of the initial point is:

$ \begin{pmatrix} 10.0115 & -10.6835 & 5.18024 \\ -10.6835 & 11.4009 & -5.52798 \\ 5.18024 & -5.52798 & 2.77646 \\ \end{pmatrix} $

I then calculate the Cholesky decomposition, which is:

$ \begin{pmatrix} 3.164095& 0& 0\\ -3.376478& 0.017131& 0\\ 1.637195& -0.001619& 0.309921\\ \end{pmatrix} $

And the initial xyz point is at (35.5361, -37.2661, 22.521).

Generating three random numbers from a Gaussian of mean 0 and width 1 yielded: -0.377495, -0.933623, 0.241011, and at this point I'm not completely clear on the correct procedure.

My assumption is that it is correct to multiply the Cholesky decomposition matrix by a matrix containing the randomly generated Gaussian points, and then add this to a matrix containing the initial values. This would look like this:

$ \begin{pmatrix} 35.5361& 0& 0\\ 0& -37.2661& 0\\ 0& 0& 22.521\\ \end{pmatrix}+ \begin{pmatrix} 3.164095& 0& 0\\ -3.376478& 0.017131& 0\\ 1.637195& -0.001619& 0.309921\\ \end{pmatrix} \begin{pmatrix} -0.377495& 0& 0\\ 0 & -0.933623 & 0 \\ 0 & 0 & 0.241011\\ \end{pmatrix} $

For which the result is:

$ \begin{pmatrix} 34.3417�& 0& 0\\ 0 & -37.281 & 0 \\ 0 & 0 & 22.5957\\ \end{pmatrix} $

And I could definitely believe that a new xyz point at (34.3417,-37.281,22.5957) could be consistent with what I want to generate, but I'm still not completely confident I have followed a valid procedure to generate this.

Any comments would be much appreciated!

Improve this question
$\endgroup$
  • 1
    $\begingroup$ I can't understand how what you are doing crosses with your question title. Multiplying orthonormal data by Cholesky root of a covariance matrix makes the data covariate right according that cov matrix. And if the data is a random sample from orthonormal population, then the same trick yields data which is a random sample from population with the cov matrix. But sorry, I can't trace, how it connects to drawing an ellips. $\endgroup$ – ttnphns Jun 12 '14 at 7:09
  • $\begingroup$ I was trying to follow something similar to the procedure described here, where you use the Cholesky decomposition to calculate points on an ellipse corresponding to a covariance matrix. Have I got this horribly wrong? $\endgroup$ – anthr Jun 12 '14 at 15:22
  • $\begingroup$ Maybe it is useful to add that my concern is the off-diagonals of the Cholesky matrix don't have an effect on the final result. I can't see whether or not this is important. $\endgroup$ – anthr Jun 12 '14 at 17:06
2
$\begingroup$

The initial point and the sample should be vectors, not matrices. So you should get: $$ \begin{bmatrix} 35.5361\\ -37.2661\\ 22.521\\ \end{bmatrix} + \begin{bmatrix} 3.164095& 0& 0\\ -3.376478& 0.017131& 0\\ 1.637195& -0.001619& 0.309921\\ \end{bmatrix} \begin{bmatrix} -0.377495\\ -0.933623 \\ 0.241011\\ \end{bmatrix} = \begin{bmatrix} 34.3417\\ -37.281 \\ 22.5957\\ \end{bmatrix} $$ But other than that, yes, your procedure is correct.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.