In Sean Carroll's GR book, pg 84, the exterior derivative $d$ is defined as $$(dA)_{\mu_1\mu_2...\mu_{p+1}} = (p+1) \partial_{[\mu_1}A_{\mu_2...\mu_{p+1}]}\tag{2.76}$$ where $A$ is a $p$-form and the RHS is the appropiately normalized and antisymmetrized partial derivative.

But on pg. 454, equation E.5, he replaced the partial derivative with a covariant derivative:

$(dA)_{\mu_1\mu_2...\mu_{p+1}} = (p+1) \nabla_{[\mu_1}A_{\mu_2...\mu_{p+1}]}\tag{2.76}$

Are these two definitions the same?


Provided that the connection is torsion-free, $\Gamma^{\sigma}_{[\mu\nu]}=0$, there's a neat cancellation that happens. I'll work out a few examples to get a feel for this: $$ \require{cancel} \newcommand\ccancel[2][black]{\color{#1}{\cancel{\color{black}{#2}}}} (dA)_{\mu_1\mu_2\dots\mu_{p+1}}\sim\nabla_{[\mu_1}A_{\mu_2\dots\mu_{p+1}]} $$

$$ \\(dA)_{\mu\nu}\sim\nabla_{[\mu}A_{\nu]} \tag{p = 1} \\\sim\nabla_\mu A_\nu-\nabla_\nu A_\mu \\=\partial_\mu A_\nu-\ccancel[red]{\Gamma^{\sigma}{}_{\mu\nu}A_\sigma} \\-\partial_\nu A_\mu+\ccancel[red]{\Gamma^{\sigma}{}_{\nu\mu}A_\sigma} \\=\partial_{[\mu}A_{\nu]} $$

$$ (dA)_{\mu\nu\sigma}\sim\nabla_{[\mu}A_{\nu\sigma]} \tag{p = 2} \\\sim\nabla_\mu A_{\nu\sigma}-\nabla_\nu A_{\sigma\mu}+\nabla_\sigma A_{\mu\nu} \\=\partial_\mu A_{\nu\sigma}-\ccancel[red]{\Gamma^\rho{}_{\mu\nu}A_{\rho\sigma}}-\ccancel[blue]{\Gamma^\rho{}_{\mu\sigma}A_{\rho\nu}} \\-\partial_\nu A_{\sigma\mu}+\ccancel[green]{\Gamma^\rho{}_{\nu\sigma}A_{\rho\mu}}+\ccancel[red]{\Gamma^\rho{}_{\nu\mu}A_{\rho\sigma}} \\+\partial_\sigma A_{\mu\nu}-\ccancel[blue]{\Gamma^\rho{}_{\sigma\mu}A_{\rho\nu}}-\ccancel[green]{\Gamma^\rho{}_{\sigma\nu}A_{\rho\mu}} \\=\partial_{[\mu}A_{\nu\sigma]} $$

since the Christoffel symbols cancel out diagonally due to the absence of torsion. It is not the greatest leap of faith to assume that this result generalises for all $p$-forms. However, we have not yet made use of the crucial point - the antisymmetry of the form! (although we didn't actually need to for the $p=1$ and $p=2$ cases)

There are prima facie three antisymmetric "factors" here: the torsion-free Christoffel symbols, the antisymmetrisation over all the indices, and the antisymmetric nature of the forms.

Let $(a,b,c,...n)$ be a permutation of $(1,2,...p+1)$. For the exterior derivative of a $p$-form, $(dA)_{\mu_1\mu_2\dots\mu_{p+1}}$, compare the terms $\Gamma^\rho{}_{\mu_a\mu_b}A_{\rho\mu_c...\mu_n}$ and $\Gamma^\rho{}_{\mu_b\mu_a}A_{\rho\,\sigma(\mu_c...\mu_n)}$ where $\sigma$ is some permutation (remember, $a$ and $b$ need not be adjacent numbers). Both terms are unique, and are specified up to a sign, which we would like to be opposite in order to get them to cancel out. There are two cases:

  • $\sigma$ is even: the Christoffel symbols give a minus sign, the antisymmetry of the form does nothing and the overall antisymmetrisation does nothing - in total, a relative minus sign

  • $\sigma$ is odd: the Christoffel symbols give us a minus sign, the antisymmetry of the form gives us a minus sign and the overall antisymmetrisation also yields a minus sign - again in total, a relative minus sign

We conclude that $$\Gamma^\rho{}_{\mu_b\mu_a}A_{\rho\,\sigma(\mu_c...\mu_n)}=-\Gamma^\rho{}_{\mu_a\mu_b}A_{\rho\mu_c...\mu_n},$$

for the specific permutation $\sigma$ of the indices that appears in the exterior derivative. So all the Christoffel symbol terms vanish. Thus, depending on the normalisation conventions of the author, $(dA)_{\mu_1\mu_2\dots\mu_{p+1}}\sim\nabla_{[\mu_1}A_{\mu_2\dots\mu_{p+1}]}$ up to a prefactor.


Nihar Karve provided an explanation why the two formulas which are denoted by $(\text{2.76})$ in the original post are identical in a curved spacetime with no torsion, i.e. with a symmetric connection. Here is the missing part to his post (which is useful in GR extensions with torsion):

Let $\nabla$ be an operator which generalizes the $d$ from curved spacetime with no torsion. This acts on a p-form on a (cotangent bundle of a) manifold with torsion $T^{\mu}_{~~\nu\rho} :=\Gamma^{\mu}_{~~\nu\rho} - \Gamma^{\mu}_{~~\rho\nu}$.

Then: $$ \displaystyle{\left(\nabla\omega\right)_{\sigma\mu_1 \mu_2...\mu_p} :=\left(d\omega\right)_{\sigma\mu_1 \mu_2...\mu_p} + (-)^p \frac{p(p+1)}{2}T^{\nu}_{~~[\sigma\mu_1} \omega_{\mu_2...\mu_p]\nu}} $$

The first term is given by $(\text{2.76})$ in the original post with $\nabla_s$ a symmetric connection. Of course, since $d^2 =0$, we have that $\nabla\nabla\omega \neq 0$.


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.