I'm trying to learn about relativistic electrodynamics on my own, and I am struggling with derivatives of the 4-potential and index (Einstein) notation.

I think I understand expressions such as $\partial_\mu A^\mu$. The index is repeated and is once up and once down, so I would expand the sum as: $\partial_0 A^0 + \partial_1 A^1 + \partial_2 A^2 + \partial_3 A^3$, which gives me a scalar.

  1. How am I to interpret something like this: $(\partial_\mu A_\nu)(\partial^\mu A^\nu)$ ? We are summing over the two indices this time, which is fine. What confuses me is that we are taking a covariant derivative of a covariant vector. Does one need to "convert" $A_\nu$ in the first term to contravariant, like so: $(\partial_\mu A^\rho \eta_{\nu\rho})(\partial^\mu A_\sigma\eta^{\nu\sigma})$?

    I guess my doubts arise from the fact that I see a covariant vector as being an entirely different object from a contravariant one. The covariant derivative $\partial_\mu = \frac{\partial}{\partial x^\mu}$ differentiates with respect to the components of the contravariant vector $x$. So I don't understand how such an operation can be applied to a vector that isn't also contravariant.

  2. How should I interpret terms such as $(\partial_\mu A^\mu)^2$ ? Is it just $\left(\partial_0 A^0 + \partial_1 A^1 + \partial_2 A^2 + \partial_3 A^3\right)^2$ or is there something else going on?
  3. According to some textbook, $(\partial_\mu \phi)^2 = \eta^{\mu\nu}\partial_\mu \phi\partial_\nu\phi$, but I don't understand why. For me $\partial_\mu \phi$ is just the derivative of a scalar $\phi$ with respect to some (unspecified) component $\mu$ of a contravariant 4-vector $x$. Instead, judging from the right-hand side, it is to be interpreted as a vector $(\frac{\partial}{\partial x^0},\boldsymbol{\nabla})\phi$ which is then squared. Is it just sloppy notation, or am I being stupid?



  1. Are the following then true? $$\frac{\partial}{\partial(\partial_\mu A_\nu)} \left(\partial_\mu A_\nu\right) = 1$$ $$\frac{\partial}{\partial(\partial_\mu A_\nu)} \left(\partial^\mu A^\nu\right) = 0$$
  2. Can I also raise and lower the indices of a partial derivative?
  • $\begingroup$ yes, you can definitely raise and lower the indices of a partial derivative, but $\partial_{\mu}$ is the one that you know and love. $\endgroup$ Oct 12, 2012 at 22:08

1 Answer 1


1) You have the right idea about $\partial_{\mu}A^{\mu}$ and 2) $(\partial_{\mu}A^{\mu})^2$.

For the rest of it, you're goign to have to be careful about covariant and contravariant indices of $\partial_\mu$ and $A^{\mu}$. Since you are already going with $A$ having vector indices, I think you should stick to that as your "base" version of $A$ for now. Now, you correctly seem to note that lowering and raising is done with $\eta_{\mu\nu}$ and its inverse $\eta^{\mu\nu}$ (which are the same matrix for Mikowskian coordinates). Thus, you will find that an expression like $A_{\mu}A^{\mu} = \eta_{\mu\nu}A^{\mu}A^{\nu} = -(A^{0})^{2} + (A^{1})^{2} + (A^{2})^{2} + (A^{3})^{2}$, and you can extend this to most of the other examples you cite.

3) applying $\partial_\mu$ to a scalar $\phi$ gives you a result $\partial_{\mu}\phi$ that is best interpreted as a one form, not as a scalar--that's because you have the free index $\mu$ floating around, and if you change coordinates, you will have to apply the chain rule to get the right answer for your new $\partial_{\mu}\phi$. this is why you have to apply the inverse metric to "square" $\partial_{\mu}\phi$.

4) If you are taking variations of vector valued quantities, you should leave a dummy index for your variational term:

$\begin{equation} \frac{\delta}{\delta A^{\mu}}A^{\nu} = \delta^{\nu}{}_{\mu} \end{equation}$

This will give you terms like

$\begin{equation} \frac{\delta}{\delta A^{\alpha}} A^{\mu}A_{\mu} = \frac{\delta}{\delta A^{\alpha}}(\eta_{\mu\nu}A^\mu A^{\nu}) = \eta_{\mu\nu}\delta^{\mu}_{\alpha}A^{\nu} +\eta_{\mu\nu}A^{\mu}\delta^{\nu}_{\alpha} = 2 A_{\alpha} \end{equation}$

which should seem pretty reasonable from your calculus I based intuition. Similarly, you should probably think of $\frac{\delta}{\delta \partial_{\alpha}A_{\beta}}\partial_{\mu}A_{\nu} = \delta^{\alpha}_{\mu}\delta^{\beta}_{\nu}$. If you're taking a variation of the up version of a quantity with respect to the down version of the quantity, just use the metric to raise/lower the target before taking the variation, and fix everything after the fact using raising and lowering conventions and the Kroneker deltas.

  • $\begingroup$ Thank you so much, that was most helpful and I am finally starting to makes sense of it. I have added a #4 to my questions, I hope I am not abusing your kindness. $\endgroup$
    – kdfc
    Oct 12, 2012 at 21:48
  • $\begingroup$ @kdfc: there's a lazy attempt at an answer. $\endgroup$ Oct 12, 2012 at 22:08

Your Answer

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

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