10
$\begingroup$

I understand the proof of the identity in the title for $A$ Hermitian. One uses that any Hermitian matrix can be diagonalized as $A = X \Lambda X^{-1}$, such that $$ \det{A} = \prod_i \lambda_i, $$ and we have $$ \exp(Tr(\log(A)) = \exp(Tr(X\log\Lambda X^ {-1}) = \exp(\sum_i\log(\lambda_i)) = \prod_i \lambda_i. $$

However, is it possible to show the identity for $A$ not Hermitian? My motivation for this question is that in physics the identity is often used without it being clear that $A$ is Hermitian.

$\endgroup$
3
  • $\begingroup$ You do need to assume $A$ is nonsingular, else $\ln(A)$ doesn't make sense. $\endgroup$ Oct 19, 2015 at 17:24
  • $\begingroup$ this is true for any non-singular A as per doi.org/10.1080/0020739X.2010.500700 $\endgroup$
    – Car Loz
    May 27, 2021 at 17:36
  • $\begingroup$ I think you should remove $X$ and $X^{-1}$ and write $\exp(Tr(\log(A)) = \exp(Tr(\log\Lambda)) = \exp(\sum_i\log(\lambda_i)) = \prod_i \lambda_i$ instead, since $\log(A) = \log\Lambda$, since $\log(A) = \log(X\Lambda X^ {-1}) = \log X + \log\Lambda + \log X^ {-1} = \log X + \log\Lambda - \log X = \log\Lambda$. $\endgroup$ Sep 28, 2021 at 14:12

2 Answers 2

18
$\begingroup$

If $\ln(A) = B$, the identity says

$$ \det(\exp(B)) = \exp(\text{Tr}(B)) $$

which is more usual form for this identity, true for all $n \times n$ matrices $B$ over $\mathbb C$ (avoiding questions about whether $\ln(A)$ is defined, and which of the possible logarithms to use).

One way to do this is to show it first for diagonalizable matrices $B$, then use the fact that diagonalizable matrices are dense and both sides of the equation are continuous functions of $B$.

A second way is to use Jordan canonical form.

A third way is to note that both $\det(\exp(tB))$ and $\exp(t \text{Tr}(B))$ satisfy the differential equation $y' = \text{Tr}(B) y$ with initial value $y(0) = 1$.

$\endgroup$
2
  • 1
    $\begingroup$ Wikipedia's page on Jacobi's identity treats this identity as a corollary of Jacobi's identity using the third proof method, with the Jacobi identity being used to obtain $\frac{d}{dt}\det(\exp(t B))$. $\endgroup$ Sep 2, 2019 at 16:48
  • $\begingroup$ You are using that $\exp(\ln A)=A$, which is clear if $A$ is a number (since then $\ln$ is usually defined to be the inverse $\exp$), but in this case it is not clear. $\endgroup$
    – Filippo
    Feb 1 at 21:01
1
$\begingroup$

Here is an elegant proof with minimal assumptions (only non-singular and square):

\begin{align} \rm{Det}(e^{B}) &= \displaystyle \lim_{N \to \infty} \rm{Det}\Big(e^{B/N}\Big)^{N} \\ &= \displaystyle \lim_{N \to \infty} \rm{Det} \Big(1 + \frac{B}{N}\Big)^{N} \\ &= \displaystyle \lim_{N \to \infty} \Big(1 + \frac{\rm{Tr}B}{N}\Big)^{N} \\ &= e^{\rm{Tr}B} \end{align}

Now take log on both sides and put $ e^{B} = A$ to get the result.

$\endgroup$
2
  • $\begingroup$ How do you get the third equality ? $\endgroup$
    – Bibek_G
    Jun 9, 2020 at 9:35
  • 1
    $\begingroup$ Using a variation of Jacobi formula i.e. det(A + hB) for small h is equal to det (A) + h Tr(adj(A)B) + O(h^2) terms $\endgroup$
    – R.G.J
    Jun 9, 2020 at 18:47

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .