I've read the definition of the electromagnetic field tensor to be \begin{equation}F^{\mu\nu}\equiv\begin{pmatrix}0&E_x&E_y&E_z\\-E_x&0&B_z&-B_y\\-E_y&-B_z&0&B_x\\-E_z&B_y&-B_x&0\end{pmatrix}\tag{*}\end{equation} in Introduction to Electrodynamics by David Griffiths, or as $$F_{\mu\nu}\equiv\begin{pmatrix}0&-E_x&-E_y&-E_z\\E_x&0&B_z&-B_y\\E_y&-B_z&0&B_x\\E_z&B_y&-B_x&0\end{pmatrix}$$ on the Lecture Notes on GR by Sean Carroll, which I know to be consistent via ${F_{\mu\nu}=\eta_{\mu\alpha}F^{\alpha\beta}\eta_{\beta\nu}}$ where the metric $\eta_{\rho\sigma}$ has a $(-+++)$ signature.

However on Wikipedia and other sources (sorry I can't remember) they use a $(+---)$ signature and they define the EM tensor to be the negative of ${(*)}$.

These are my thoughts about it: The antisymmetry $F^{\mu\nu}=-F^{\nu\mu}$ may point out that it's just an unfortunate mix of index letters and that for the sources notation to be consistent, either the first two or Wikipedia should change $\mu\nu$ to $\nu\mu$. If not the case, the properties seem to be the same; at first I thought the inner product would pop out a minus sign of difference, but it of course didn't happen, and as for other entities I've worked with, e. g. the 4-velocity, though the metric signature can change, the contravariant vector is the same in either case. However again, I've read the stress-energy tensor does change sign depending on the signature.

So is the signature of the metric involved in the definition of ${F^{\mu\nu}}$ or any tensor whatsoever? If so, how can I know what signature is involved? or if not, what's the matter with the minus sign difference on the definitions?


3 Answers 3


Let $$ \eta_{\mu\nu}={\rm diag}(+1,-1,-1,-1) \qquad \bar\eta_{\mu\nu}={\rm diag}(-1,+1,+1,+1) $$ with corresponding Lorentz force laws (in units where mass equals charge) $$ \ddot x^\mu=\eta_{\nu\lambda}F^{\mu\nu}\dot x^\lambda \qquad \ddot{\bar x}^\mu=\bar\eta_{\nu\lambda}\bar F^{\mu\nu}\dot{\bar x}^\lambda $$

As the trajectories $x^\mu, \bar x^\mu$ should agree (and so will all its derivatives) for all initial conditions, we can equate the terms $$ \tag{1} \eta_{\nu\lambda}F^{\mu\nu} = \bar\eta_{\nu\lambda}\bar F^{\mu\nu} $$ Contracting with the inverse $\eta^{\lambda\sigma}$ of $\eta_{\nu\lambda}$ finally yields $$ F^{\mu\sigma} = -\bar F^{\mu\sigma} $$ as $$ \bar\eta_{\nu\lambda}\eta^{\lambda\sigma} = -\delta_\nu^{\sigma} $$ This means the signs of the components of the electromagnetic tensor $F^{\mu\nu}$ do indeed depend on the metric convention. This also applies to $F_{\mu\nu}$, whereas the tensor of mixed rank $F^\mu{}_\nu$ is independant of this choice (which is just (1)).

  • $\begingroup$ Great! Just one thing, you used ${u^\lambda=\bar u^\lambda}$ which is known by definition of the 4-position, but you also used ${\dot{u}^\mu=\dot{\bar{u}}^\mu}$, (as one supposedly doesn't know beforehand that ${F^{\mu\sigma}=-\bar F^{\mu\sigma}}$) right? So how is this justified? Can one always define contravariant vectors to be the same whatever the metric signature is? $\endgroup$
    – user24999
    Commented Nov 20, 2013 at 17:48
  • 1
    $\begingroup$ @PedroFigueroa: same as velocities, accelerations (as well as any higher derivatives of position) agree - we're dealing with the same trajectory; I'll clarify $\endgroup$
    – Christoph
    Commented Nov 20, 2013 at 19:28

We will work in unit with $c=1$. In both sign conventions for the metric $\eta_{\mu\nu}$ we define the field strength as

$$\tag{1} A^{\mu}~=~(\Phi,{\bf A}). $$

$$\tag{2} F_{\mu\nu}~:=~ \partial_{\mu} A_{\nu} -\partial_{\nu} A_{\mu}, \qquad \mu,\nu~\in~\{0,1,2,3\}. $$

$$\tag{3} E_i~:=~- \partial_i\Phi -\partial_0 A^i, \qquad i~\in~\{1,2,3\}. $$

[The relation (3) can be partially remembered by the fact that in electrostatics, one demands that ${\bf E}~=~-{\bf \nabla}\Phi$. It turns out that the rest of eq. (3) is then fixed by consistency.] Tensors are raised and lowered with the metric tensor $\eta_{\mu\nu}$.

It is then straightforward to check that this implies that in signature

$$\tag{4} (+,-,-,-)\qquad \text{resp.} \qquad(-,+,+,+), $$

the $4$-potential $A_{\mu}$ with lower index is

$$\tag{5} A_{\mu}~=~(\Phi,-{\bf A}) \qquad \text{resp.} \qquad A_{\mu}~=~(-\Phi,{\bf A}),$$

and the electric field ${\bf E}$ is

$$\tag{6} E_i~=~F_{0i} \qquad \text{resp.} \qquad E_i~=~F_{i0}. $$

See also this related Phys.SE post.

  • $\begingroup$ So you postulate a vector field $A$ on $M$ and then you define $F:=\mathrm d(A^\flat)$? $\endgroup$
    – Filippo
    Commented Oct 19, 2022 at 8:51
  • 1
    $\begingroup$ Hi Filippo. Thanks for your feedback. The priority is admittedly not elegance, but agreement with clunky historic sign conventions in the E&M literature. $\endgroup$
    – Qmechanic
    Commented Oct 19, 2022 at 8:59

Though this question was asked over 10 years ago, I believe the answer can be more clear and explicit.

The most common notation is the West Coast notation based on thee metric signature (+---). I however prefer the East Coast signature (-+++). For this choice the signs of the electric field components of the field tensor remain the same, while the magnetic field components change sign as given by the OP in (*).

This can be verified by comparing the covariant Maxwell equation $\partial_\mu F^{\mu\nu}=-j^\mu/\varepsilon_0$ with the Coulomb equation and the standard Maxwell equation containing the current (Ampère's circuital law).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.