I am trying to understand why $$\det(F^{\mu\nu})=(\vec{E}\cdot\vec{B})^2\tag{1}.$$ Of course one can just calculate the determinant of $F^{\mu\nu}$ expressed as a matrix with components given in terms of $E_x, E_y, ...$, etc., but I am looking for something a bit more insightful. In particular, I understand that $\vec{E}\cdot\vec{B}$ can be written as


This seems promising because then


which looks awfully like the expression for the determinant of $F$ in terms of Levi-Civita tensors,


But I can't figure out how to connect these two expressions. Since $F$ is antisymmetric, one can flip the two indices of any single $F$ at the cost of a minus sign, but I need a way to permute indices between different $F$'s. I'm also not sure how a factor of 3 could possibly enter in to change the 1/64 into a 1/24.

  • $\begingroup$ You might be able to get away with using symmetrization/antisymmettization, i.e. that you can try and use $$F^{\mu \nu}F^{\alpha \beta} = 2! (F^{\mu [\nu}F^{\alpha] \beta} + F^{\mu \alpha}F^{\nu \beta})$$ and perhaps using the epsilons to get rid of certain parts. I haven't totally looked this over but the idea popped up in my head $\endgroup$ – Triatticus Sep 30 '18 at 21:17
  • $\begingroup$ By factor of $3$ you mean factor of $8/3$ $\endgroup$ – J.G. Sep 30 '18 at 21:28
  • 1
    $\begingroup$ @J.G. true, I just meant I can imagine how factors of 2 might pop up, but not factors of 3. $\endgroup$ – WillG Sep 30 '18 at 21:36
  • $\begingroup$ Related: Fundamental invariants of the electromagnetic field. $\endgroup$ – Emilio Pisanty Oct 9 '18 at 11:35

The two expressions are only superficially similar; they have to be by dimensional analysis. The index structure is completely different, and I don't think you can convert one to the other without essentially undoing everything you did to get to the first result, then inserting the standard proof of the second result. Such a proof would be unenlightening since it'd just be unnecessarily complicated.

So let me just give a simple derivation of the result you want, directly. The determinant is $$\det F^{\mu\nu} = \epsilon_{\mu\nu\rho\sigma} F^{\mu 0} F^{\nu 1} F^{\rho 2} F^{\sigma 3}$$ There are naively $4!$ terms, but the only nonzero terms are those where $\mu \neq 0$, $\nu \neq 1$, $\rho \neq 2$, $\sigma \neq 3$. That is, we need to count the number of derangements of a 4-element set. It is easy to show by casework there are $9$.

Each of these $9$ terms is quadratic in $\mathbf{E}$, since there are two indices equal to zero, and hence quadratic in $\mathbf{B}$. Moreover, the sum of all $9$ terms is a tensorial invariant. There are only two independent invariants: $E^2 - B^2$ and $\mathbf{E} \cdot \mathbf{B}$. We can't use the first one, because otherwise our terms wouldn't all have the same degree in $E$ and $B$. Then the answer must be proportional to $(\mathbf{E} \cdot \mathbf{B})^2$, but since $(\mathbf{E} \cdot \mathbf{B})^2$ has $9$ terms, they must simply be equal.

You might complain I used components, but I had to because your expression is not tensorial. The fields $\mathbf{E}$ and $\mathbf{B}$ are not Lorentz tensors, but rather a way of writing components of $F_{\mu\nu}$. You can't expect to prove a statement about components without expanding in components.

  • $\begingroup$ ok thanks, I'm fine to accept the component-based argument. However, could you expand on / justify the fact that $E^2-B^2$ and $\vec{E}\cdot\vec{B}$ are the only two possible invariants? This fact is not obvious to me. $\endgroup$ – WillG Sep 30 '18 at 22:29
  • $\begingroup$ @WillG It's well-known these are the only linearly independent scalars obtainable from $F_{\mu\nu}$, and are respectively proportional to $F_{\mu\nu}F^{\mu\nu}$ and $\epsilon_{\mu\nu\rho\sigma}F^{\mu\nu}F^{\rho\sigma}$. $\endgroup$ – J.G. Oct 1 '18 at 5:34

OP's eq. (2) is (minus) the Pfaffian ${\rm Pf}(F)$, and OP's eq. (4) is the determinant ${\rm Det}(F)$. The sought-for relation (1) follows because the square of the Pfaffian is the determinant:

$$ {\rm Pf}(F)^2~=~{\rm Det}(F). \tag{A}$$

A proof of eq. (A) is given in my Math.SE answer here.


The determinant on the left is invariant under (proper) Lorentz transformations, so you can easily calculate it in a frame of reference where $E^2=E^3=0$, it equals $(E^1 H^1)^2=(\vec{E}\vec{H})^2 $. As $\vec{E}\vec{H} $ is an invariant under proper Lorentz transformations, this proves your formula in a general case.


Quoting myself from an old PhysicsForums post

If I am not mistaken, (someone check my math) you can calculate the principal invariants of a tensor: $$F^a{}_a,\quad F^a{}_{[a} F^b{}_{b]},\quad F^a{}_{[a} F^b{}_{b}F^c{}_{c]},\quad F^a{}_{[a} F^b{}_{b} F^c{}_{c}F^d{}_{d]}$$ in 4-dimensions. You'd get the trace (sum of the eigenvalues), sum of products-of-pairs of eigenvalues, sum of products-of-triples, and finally, the product-of-the-4-eigenvalues (the determinant).

For a real antisymmetric matrix, the eigenvalues are imaginary. So, only those with sums of products of even-numbers-of-eigenvalues will be nonzero.

(Higher combinations like $F^a{}_{[a} F^b{}_{b} F^c{}_{c}F^d{}_{d}F^e{}_{e}F^f{}_{f]}$ vanish since it would have more than 4 factors being antisymmetrized). So, all but two are identically zero.

At some point in your calculation, you would probably have to make use of the epsilon-delta identities [sometimes written as determinants with delta-entries] (like this contracted combination $\epsilon_{abmn}\epsilon^{cdmn}=-4\delta_a{}^{[c}\delta_b{}^{d]}$ (up to sign conventions, based on signature and dimensionality)). Together with the antisymmetry of $F_{ab}=F_{[ab]}$, I think the index-gynmastics will work out.


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.