Assume $M$ is an invertible positive matrix of rank $N$. The Von Neumann entropy $H$ of a matrix $M$ with eigenvalues $\{ \lambda_n \}$ is

$H[M] = -\sum_{n=1}^N \lambda_n \ln \lambda_n$.

In principle, the eigenvalues are encoded in the characteristic polynomial

$\phi_t (M) = \mathrm{det}(tI-M) = \prod_n(t-\lambda_n) = t^{N} + a_{N-1} t^{N-1} \cdots + a_1 t +a_0 $.

The trace $\mathrm{Tr}\, M$ is given by the coefficient $a_{N-1}$ in the characteristic polynomial:

$\lim\limits_{t \to \infty} \dfrac{\phi_t (M)-t^N}{t^{N-1}} = a_{N-1} = (-1)^{N-1} \, \mathrm{Tr}\, M = (-1)^{N-1} \sum_n \lambda_n$.

Is there a similar relationship between the entropy and the characteristic polynomial?

  • 1
    To define $H$ you also need that the eigenvalues are positive, right? – Qiaochu Yuan Jul 18 '12 at 18:10
  • Sorry, yes, I will fix that. Thanks. – Jess Riedel Jul 18 '12 at 18:28

I'm not sure if this qualifies as simple (or if this is helpful at all), but we have $$ \frac{\phi'_M(t)}{\phi_M(t)}=\sum_n\frac{1}{t-\lambda_n} $$

Using the residue theorem, we can write $$ H[M]=\frac{-1}{2\pi i}\oint\frac{\phi'_M(z)}{\phi_M(z)}z\log(z)\,dz $$ where the integral is taken over a closed contour containing all of the eigenvalues of $M$ (I guess we're either working on the Riemann surface of $\log(z)$, or we chose a branch of $\log(z)$).

Yes. The entropy is a power series in the coefficients of the characteristic polynomial. (If the eigenvalues all lie on the same side of $0$.)

Expand $\lambda_n \ln \lambda_n$ as a power series in $\lambda_n$. This allows you to write $H[M]$ as an infinite sum of symmetric polynomials in the $\lambda_n$. A symmetric polynomial in the $\lambda_n$ can be written as a polynomial of the elementary symmetric polynomials, which are $a_0,...,a_{N-1}$. So you can write $H[M]$ as an infinite sum of polynomials in the $\lambda_n$.

Is that a simple enough relation?

  • $f(x) = x \ln x$ is not differentiable at $x=0$, so I don't think I can do the power series, right? – Jess Riedel Jul 18 '12 at 18:35
  • You do a Taylor series around some nonzero number, depending on your eigenvalue spread. If, as I understand, they're all positive real, you take it around some number that is at at least half the largest eigenvalue. – Will Sawin Jul 18 '12 at 19:27

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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