I would like to compute a partial inverse of a symmetric semi-definite matrix.

I read about computing the pseudoinverse of a rectangular matrix by using SVD, however with a symmetric matrix I could apply a similar technique using instead the eigenvalue decomposition, i.e. compute the eigenvalues, discard the smallest and invert the remaining.

Does this make sense? If so can you point me to a reference that explains how to achieve this in detail?

Any help appreciated.

  • 1
    $\begingroup$ if it is a positive definite singular matrix it is called positive semi-definite matrix. $\endgroup$ – user13838 Nov 28 '11 at 23:14
  • $\begingroup$ thanks, edited the question. $\endgroup$ – Andrea Zonca Nov 28 '11 at 23:17

The idea is that the singular value decomposition,

$$\mathbf A=\mathbf U\mathbf \Sigma\mathbf V^\top$$

and the eigendecomposition

$$\mathbf A=\mathbf Q\mathbf D\mathbf Q^\top$$

of a symmetric matrix are one and the same.

Thus, if one wants the Moore-Penrose pseudoinverse of $\mathbf A$, either decomposition could be used. (However, an SVD routine generally wouldn't exploit the nice structure of a symmetric matrix, so a bit more computational effort than what is actually needed will be used in that case; thus, use the eigendecomposition.)

The idea is that, letting $\mathbf A^\dagger$ be the Moore-Penrose pseudoinverse, we have the property

$$\mathbf A^\dagger=\mathbf Q\mathbf D^\dagger\mathbf Q^\top$$

where $\mathbf D^\dagger$ is (usually) computed via the following procedure: take $d_1$ to be the largest eigenvalue, and let $\varepsilon$ be machine epsilon. Reciprocate any entry of $\mathbf D$ that is greater than $\varepsilon\cdot d_1$, and set all other entries to zero.

  • $\begingroup$ thanks! do you know of any library that has this already implemented? $\endgroup$ – Andrea Zonca Nov 28 '11 at 23:27
  • $\begingroup$ LAPACK is the standard library for these. $\endgroup$ – J. M. is a poor mathematician Nov 28 '11 at 23:29
  • $\begingroup$ in LAPACK I found only the eigenvalue computation, I wondered if any library had all the pseudoinverse process implemented, it would think this is a quite common situation. $\endgroup$ – Andrea Zonca Nov 28 '11 at 23:31
  • $\begingroup$ Sure, but the problem is that the canned routines for pseudoinverses would not exploit the fact that your matrix is symmetric. You're fine with wasted effort? (How big are your matrices, anyway?) $\endgroup$ – J. M. is a poor mathematician Nov 28 '11 at 23:38
  • $\begingroup$ matrices are tiny, 3x3, but I have few thousands of them. $\endgroup$ – Andrea Zonca Nov 28 '11 at 23:41

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.