This is a question on the nitty-gritty bits of general relativity.

Would anybody mind teaching me how to work these indices?


Throughout the following, repeated indices are to be summed over.

Hodge dual of a p-form $X$: $$(*X)_{a_1...a_{n-p}}\equiv \frac{1}{p!}\epsilon_{a_1...a_{n-p}b_1...b_p}X^{b_1...b_p}$$ Exterior derivative of p-form $X$: $$(dX)_{a_1...a_{p+1}}\equiv (p+1) \nabla_{[a_1}X_{a_2...a_{p+1}]}$$

Given the relation $$\epsilon^{a_1...a_p c_{p+1}...c_n}\epsilon_{b_1...b_pc_{p+1}...c_n}\equiv p!(n-p)! \delta^{a_1}_{[b_1}...\delta^{a_p}_{b_p]}\,\,\,\,\,\,\,\,\,(\dagger)$$ where $\epsilon_{a_1...a_n}$ is an orientation of the manifold.

Why then is $$(*d*X)_{a_1...a_{p-1}}=(-1)^{p(n-p)}\nabla^b X_{a_1...a_{p-1}b}$$?

How far I've got myself:

Firstly, I believe $(*d*X)$ means $*(d(*X))$? $$(d*X)_{c_1...c_{n-p+1}}=\frac{n-p+1}{p!}\nabla_{[c_1}\epsilon_{c_2...c_{n-p+1}]b_1...b_p}X^{b_1...b_p}$$ Then $$*(d*X)_{d_1...d_{p-1}}=\frac{n-p+1}{(n-p+1)!p!}\epsilon_{d_1...d_{p-1}c_1...c_{n-p+1}}\nabla^{[c_1}\epsilon^{c_2...c_{n-p+1}]b_1...b_p}X_{b_1...b_p}$$ $$=\frac{1}{(n-p)!p!}\epsilon_{d_1...d_{p-1}c_1...c_{n-p+1}}\nabla^{[c_1}\epsilon^{c_2...c_{n-p+1}]b_1...b_p}X_{b_1...b_p}$$

Now I know that I should apply $(\dagger)$ but I don't know how to given the antisymmetrisation brackets. Would someone mind explaining it to me please? Thank you!


2 Answers 2


I don't want to enter in the detail of the calculus, but the error is in the order of $\epsilon$ terms, it is not:

$$\epsilon^{\large A_pC_{n-p}} \epsilon_{\large B_pC_{n-p}}$$

but :

$$\epsilon^{\large A_pC_{n-p}} \epsilon_{\large C_{n-p}B_p}$$

Where indices like $A_p$ means $a_1...a_p$

So, before applying general formulae concerning Levi-Civita tensor, you have to invert, for instance, $B_p$ and $C_{n-p}$. If you move a term belonging to $B_p$ tp the left, you got a term $(-1)^{n-p}$. Because you have $p$ terms in $B_p$, you get a total term $(-1)^{p(n-p)}$

You have the same type of problem in looking at the difference between $*(*X)$ and $X$, so, if you want, begin by this calculus.

Secondly, you have to take in account the metrics (here I consider constant metrics), so your first formula for the Hodge dual is not true, you have to use $\sqrt{|g|} \epsilon_{AB}$ instead of $\epsilon_{AB}$, and $ - sgn(g) \frac{1}{\sqrt{|g|}} \epsilon^{AB}$, instead of $\epsilon^{AB}$. So, there is a difference between an euclidean metrics and a Minkowski metrics (You will get a minus sign with the Minkowski metrics). Do not forget that your raise or lower the indices with the metrics.

Finally, You mix $2$ things, ordinary exterior derivative $d$, and covariant exterior derivative $D$. I think, that, if you use the notation $d$, it concerns only standard derivatives.


Note also, that all your factorial terms disappear if you don't take an totally antisymmetric expression, but just an ordered expression, for instance :

$\frac{1}{2!} \epsilon^{ij} X_{ij} = X_{12}$ ($X$ is supposed anti-symmetric)


Yes, (∗d∗X) means ∗(d(∗X))


I will only give a solution in a constant euclidean metrics.

Multi-Indices $A_p$ means $a_1...a_p$, I used a mono-indice $C$ which means $c$.

So, we have :

$$(*X)_{\large B_{n-p}} = \frac{1}{p!} \epsilon^{\quad \quad\large A_{p}}_{\large B_{n-p}} \quad (X)_{\large A_{p}} \quad \quad \quad \quad(1)$$

$$(d*X)_{\large CB_{n-p}} = (n-p+1) \quad \partial_{\large [C}(*X)_{\large B_{n-p}]} \quad \quad \quad \quad(2)$$

$$*(d*X)_{\large D_{p-1}} = \frac{1}{(n-p+1)!} \epsilon^{\quad \quad\large CB_{n-p}}_{\large D_{p-1}} \quad (d*X)_{\large CB_{n-p}} \quad \quad \quad \quad(3)$$

So, using $(1), (2), and (3)$, you have :

$$*(d*X)_{\large D_{p-1}} = \frac {n-p+1}{(n-p+1)! p!} \epsilon^{\quad \quad\large CB_{n-p}}_{\large D_{p-1}} ~~\epsilon^{\quad \quad\large A_{p}}_{\large B_{n-p}} \quad \partial_{\large [C}(X)_{\large A_{p}]} \quad \quad \quad \quad(4)$$

That is :

$$*(d*X)_{\large D_{p-1}} = \frac {1}{(n-p)! p!} \epsilon_{\large D_{p-1}C'B_{n-p}} ~~\epsilon^{\large \large B_{n-p}A_{p}} g^{CC'} \quad \partial_{\large [C}(X)_{\large A_{p}]} \quad(5)$$

Putting $\large B_{n-p}$ at left, we get:

$$*(d*X)_{\large D_{p-1}} = (-1)^{p(n-p)} \frac {1}{(n-p)! p!} \epsilon_{\large B_{n-p}D_{p-1}C'} ~~\epsilon^{\large \large B_{n-p}A_{p}} g^{CC'} \quad \partial_{\large [C}(X)_{\large A_{p}]} \quad(6)$$

Using the contraction formulae of the Levi-Civita symbol, we get :

$$*(d*X)_{\large D_{p-1}} = (-1)^{p(n-p)} \delta_{d_1}^{[a_1} \delta_{d_2}^{a_2} ...\delta_{d_{p-1}}^{a_{p-1}} \delta^{a_p]}_{c'} g^{CC'} \quad \partial_{\large [C}(X)_{\large A_{p}]} \quad \quad(7)$$

Because, both terms $\delta_{\large D_{p-1}C'} ^{\large [A_{p}]}$ (that is $\delta_{d_1}^{[a_1} \delta_{d_2}^{a_2} ...\delta_{d_{p-1}}^{a_{p-1}}$) and $\partial_{\large [C}(X)_{\large A_{p}]}$ are antisymmetric in $\large A_{p}$, we could write :

$$*(d*X)_{\large D_{p-1}} = (-1)^{p(n-p)} \delta_{d_1}^{a_1} \delta_{d_2}^{a_2} ...\delta_{d_{p-1}}^{a_{p-1}} \delta^{a_p}_{c'} g^{CC'} \quad \partial_{\large [C}(X)_{\large A_{p}]} \quad \quad(8)$$

So, finally,

$$*(d*X)_{\large D_{p-1}} = (-1)^{p(n-p)} g^{CC'} \quad \partial_{\large [C}(X)_{\large D_{p-1}C']} \quad \quad(9)$$

With a (constant) Lorentzian metrics, with determinant $-1$, you will have a $sgn(g) = -1$ supplementary term, because, in this case, you have $\epsilon^{\large F_n} = sgn(g) \epsilon_{\large F_n} = - \epsilon_{\large F_n}$, and because contractions formulae for Levi-Civita symbol work only for euclidean metrics.


You got to $$*(d*X)_{d_1...d_{p-1}}=\frac{n-p+1}{(n-p+1)!p!}\epsilon_{d_1...d_{p-1}c_1...c_{n-p+1}}\nabla^{[c_1}\epsilon^{c_2...c_{n-p+1}]b_1...b_p}X_{b_1...b_p}$$ $$=\frac{1}{(n-p)!p!}\epsilon_{d_1...d_{p-1}c_1...c_{n-p+1}}\nabla^{[c_1}\epsilon^{c_2...c_{n-p+1}]b_1...b_p}X_{b_1...b_p}$$ Lets denote the index $c_{n-p+1}$ by $e$ instead (or rename the indices otherwise...). Then \begin{align} \epsilon_{d_1...d_{p-1}c_1...c_{n-p+1}}\nabla^{[c_1}\epsilon^{c_2...c_{n-p+1}]b_1...b_p} &= \epsilon_{d_1...d_{p-1}ec_1...c_{n-p}}\nabla^{[e}\epsilon^{c_1...c_{n-p}]b_1...b_p} \\ &= (-1)^{p(n-p)}\epsilon_{d_1...d_{p-1}ec_1...c_{n-p}}\epsilon^{b_1...b_p[c_1...c_{n-p}}\nabla^{e]}\\ &= (-1)^{p(n-p)}p!(n-p)!\delta^{b_1}_{[d_1}\dots \delta^{b_{p-1}}_{d_{p-1}}\delta^{b_p}_{e]}\nabla^{e} \end{align} (modulo some details, which you'll still need to fill in I guess...) It follows that \begin{align} \ast (d\ast X)_{d_1 \dots d_{p-1}} &= (-1)^{p(n-p)}\delta^{b_1}_{[d_1}\dots \delta^{b_{p-1}}_{d_{p-1}}\delta^{b_p}_{e]}\nabla^{e}X_{b_1 \dots b_p} \\ &= (-1)^{p(n-p)} \nabla^e X_{d_1\dots d_{p-1}e}. \end{align}


Your Answer

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

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