5

I want to know that if the equation in the title always holds? I have generated a random Hermitian matrix $A$, and then compute $\log(\det A)$ and $ tr(\log(A))$ in matlab, it is not equal. So I'm really puzzled.

Can you give me the proof of this equation if it is true. Thanks in advance

I just show an example:

$$A=\begin{bmatrix}30.9186 + 0.0000i & -0.5120 - 0.0197i \\-0.5120 + 0.0197i & 10.3822 + 0.0000i \end{bmatrix}$$

$K>> log2(det(A))$

$ans = 8.3253 + 0.0000i$

$K>> trace(log2(A))$

$ans =8.3264 + 0.0000i$

what's wrong about it?

it is essentially different from these problem such as

How to prove $\det(e^A) = e^{\operatorname{tr}(A)}$?

because I think this equation is totally wrong, so I don't know why it can be proved.

LinTIna
  • 295
  • 1
    Do you know a formula for the determinant or trace, maybe in terms of eigenvalues? – Stan Tendijck Feb 28 '19 at 12:13
  • yes, but I don't know how to proof this equation – LinTIna Feb 28 '19 at 12:16
  • 2
    First you have to ensure that the expression $\log(A)$ makes sense (this is not the case for all hermitian matrices). – Alpha001 Feb 28 '19 at 12:17
  • Continuing the @Alpha001 comment: To answer this, you need a definition of $\log(A)$. So begin there. – GEdgar Feb 28 '19 at 12:25
  • I just add an example, do you know what's wrong with it – LinTIna Feb 28 '19 at 12:25
  • I'm voting to close this question as off-topic because the OP does not read answers and comments. –  Mar 01 '19 at 01:11
  • @user91684 OP has read and even answered to comments, and even if OP wouldn't read comments, that still wouldn't make the question off-topic, so your comment doesn't make much sense to me. – HelloGoodbye Sep 28 '21 at 12:37

1 Answers1

1

If $A$ is hermitian $>0$, then your formula is valid is we choose the principal logarithm $(\log(re^{i\theta})=\log(r)+i\theta$ where $\theta\in (-\pi,\pi)$).

If $A$ is invertible hermitian, your formula is -in general- not valid; example

$A=-I_2$. The required equality is (*) $\log(1)=2\log(-1)$.

If we choose the following $\log$: $\log(re^{i\theta})=\log(r)+i\theta$ where $\theta\in (-\pi+1/10,\pi+1/10)$, then $\log(1)=0,\log(-1)=i\pi$ and (*) is not satisfied.

EDIT. About your edit, you are not serious.

If $spectrum(A)=(\lambda_i)$, then $spectrum(e^A)=(e^{\lambda_i})$. Therefore, $\det(e^A)=\Pi_i e^{\lambda_i}$ and $e^{tr(A)}=e^{\Sigma_i \lambda_i}$. I hope that you are convinced...