1
$\begingroup$

Here are two n-dimensional vectors: $V_1$ and $V_2$

$V_1 (v_1,v_2, \dots ,v_n)$

$V_2 (v_1,v_2, \dots ,v_n)$

$V_1 \cos(\theta) + V_2 \sin(\theta)$ is an ellipse in the $n$-D space. (Its center is the origin.)

I want to find the directions and magnitudes of the major and minor axes.

I have a $2$-D example here: Ellipse with non-orthogonal minor and major axes?

People used SVD to find the minor and major axes of that $2$-D example, but I'm not familiar with SVD. So I'm having problem extending it to $n$-dimension or arbitrary $V_1$ and $V_2$.

Could anybody give a formula for it? Thank you very much in advance!

$\endgroup$
1
  • 1
    $\begingroup$ It's still possible to compute the singular value decomposition of the matrix whose columns are your two vectors; the singular values will then correspond to the two axes. $\endgroup$ Aug 23 '12 at 17:52
2
$\begingroup$

For example, let's take the 3-dimensional vectors $$ V_1 = \pmatrix{1\cr 2\cr 1\cr},\ V_2 = \pmatrix{3\cr -3\cr 2\cr} $$ With these as columns we form the matrix $$A = \pmatrix{1 & 3\cr 2 & -3\cr 1 & 2\cr}$$ Now $$ A A^T = \pmatrix{1 & 3\cr 2 & -3\cr 1 & 2\cr} \pmatrix{1 & 2 & 1\cr 3 & -3 & 2\cr} = \pmatrix{10 & -7 & 9\cr -7 & 13 & 0\cr 9 & 0 & 13\cr}$$ which has eigenvalues $23, 13, 0$ corresponding to normalized eigenvectors $$ W_1 = \frac{1}{\sqrt{230}} \pmatrix{10\cr -7 \cr 9\cr},\ W_2 = \frac{1}{\sqrt{130}} \pmatrix{0 \cr 9\cr 7\cr},\ W_3 = \frac{1}{\sqrt{299}} \pmatrix{13\cr 7\cr -9\cr} $$ Note that these are orthogonal since $A A^T$ is symmetric. The major and minor axes have lengths $\sqrt{23}$ and $\sqrt{13}$ (the singular values, which are the square roots of the nonzero eigenvalues of $A A^T$) and directions $W_1$ and $W_2$ respectively.

$\endgroup$
3
  • $\begingroup$ Thank you so much. So for n-dimensional space (where n>3), Is it possible that we get more eigenvalues? Do we always pick the first two eigenvalues and corresponding eigenvectors as major and minor axes? $\endgroup$
    – Hi271
    Aug 24 '12 at 1:55
  • $\begingroup$ If we're talking about an ellipse, $A$ has rank $2$, so $A A^T$ has rank $2$, and there are only $2$ nonzero eigenvalues. The largest eigenvalue gives you the major axis and the second gives you the minor axis. $\endgroup$ Aug 24 '12 at 5:36
  • $\begingroup$ Thank you very much! I get it. $\endgroup$
    – Hi271
    Aug 24 '12 at 14:57

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.