Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I don't understand how tensors work, can someone please explain?

In particular, in the context of Electromagnetism, the dual of the field tensor $F$ is $$(*F)^{\mu\nu}:={1\over 2}\epsilon^{\mu\nu\rho\sigma}F_{\rho \sigma}$$

I am wondering what happens if we take the dual twice. It is given that $$**F=-F$$ But I don't see it trivially.

What I think (probably wrong): I suppose $$(**F)^{ab}={1\over 4} \epsilon^{ab\mu\nu}\epsilon_{\mu\nu\rho\sigma}F^{\rho\sigma}$$

Actually I am very confused as to whether to put the indices upstairs or downstairs.For example, I don't know whether to write the above or $$(**F)^{ab}={1\over 4} \epsilon^{ab\mu\nu}\epsilon_{\mu\nu}.^{\rho\sigma}F_{\rho\sigma}$$ Or in fact other variants, providing that indices to be summed over are such that one's up and one's down.

Help would be appreciated. Thank you.

share|cite|improve this question
Isn't the dual simply defined as $(\ast F)^{ab}=\eta^{ac}\eta^{bd}F_{cd}$, where $\eta$ is the Minkowski metric? – wisefool Jan 5 '13 at 22:21
@wisefool : I think that is just $F^{ab}$. It is like changing covariant and contravariant components... no? – user55309 Jan 5 '13 at 22:38
up vote 4 down vote accepted

$\def\a{\alpha} \def\b{\beta} \def\e{\epsilon} \def\d{\delta} \def\m{\mu} \def\n{\nu} \def\r{\rho} \def\s{\sigma}$We have $(*F)^{\m\n} = \frac{1}{2}\e^{\m\n\r\s} F_{\r\s}$, so $(*F)_{\m\n} = \frac{1}{2}\e_{\m\n\r\s} F^{\r\s}.$ Indices are raised and lowered with the Minkowski metric $\eta_{\m\n}$ using the fact that $$\eta^{\m\n}\eta_{\n\r} = \d^\m_\r.$$ Thus, $$\begin{eqnarray*} (**F)^{\m\n} &=& \frac{1}{4} \e^{\m\n\r\s}\e_{\r\s\a\b} F^{\a\b} \\ &=& \frac{1}{4} \e^{\r\s\m\n}\e_{\r\s\a\b} F^{\a\b} \end{eqnarray*}$$ as you've written. To evaluate this we need a result about the Levi-Civita symbol, $$\e^{a_1\cdots a_k a_{k+1}\cdots a_d}\e_{a_1\cdots a_k b_{k+1}\cdots b_d} = (-1)^s k!(d-k)!\ \d^{[a_{k+1}}_{b_{k+1}}\cdots \d^{a_d]}_{b_d},$$ where $(-1)^s = \textrm{sign}\,\det\eta$ and where indices in square brackets are antisymmetrized. In particular, for $d=4$, $k=2$, and $(-1)^s = -1$ we have $$\e^{\r\s\m\n}\e_{\r\s\a\b} = -4 \d_\a^{[\m} \d_\b^{\n]}.$$ Thus, $$\begin{eqnarray*} (**F)^{\m\n} &=& -\d_\a^{[\m} \d_\b^{\n]} F^{\a\b} \\ &=& -\frac{1}{2}(\d_\a^\m \d_\b^\n - \d_\a^\n \d_\b^\m) F^{\a\b} \\ &=& - \frac{1}{2}(F^{\m\n} - F^{\n\m}) \\ &=& -F^{\m\n}, \end{eqnarray*}$$ where we have used the fact that $F$ is antisymmetric.

It is possible to write all this in the language of differential forms in which case we are verifying that the inverse of the Hodge star operation $*$ on 2-forms is $-*$.

share|cite|improve this answer
Thank you very much, oen. This is a very comprehensive explanation. – user55309 Jan 6 '13 at 0:46
@user55309: Glad to help. Good luck in your studies. – user26872 Jan 6 '13 at 0:57

There are two concepts here. $F^{ab}$ is dual to $F_{ab}$ in the sense of being contravariant components vs covariant. This isn't denoted with the Hodge star, however. The Hodge star constructs a field that is orthogonal to the original.

The various expressions with different indices upstairs vs downstairs are equivalent.

You will need to evaluate $\epsilon_{abcd} \epsilon^{abef}$ to see -1 come out. In geometric algebra, the -1 is obvious--duality is enforced by contraction with the unit pseudoscalar. This object squares to -1 in 3+1 spacetime, so no more work is needed.

Edit: if you're familiar with clifford algebra, in the guise of Dirac matrices of quantum mechanics, this should be somewhat intelligible. We can interpret the gamma matrices $\gamma_a$ as (contravariant) Cartesian basis vectors for Minkowski space. The algebra they follow by multiplication is the clifford algebra for 3+1 spacetime. Let $\gamma_0 \gamma_0 = -1$ and $\gamma_i \gamma_i = +1$ for $i=1\ldots 3$.

Now then, let's look at the duality operation. Let $G^{cd} = \epsilon^{abcd} F_{ab}$ so that $G = \star F$. We can build up $F$ using the Clifford algebra as

$$F = \frac{1}{2} F_{ab} \gamma^a \gamma^b \implies F_{ab} = F \cdot (\gamma_b \wedge \gamma_a)$$

and $\epsilon$ as

$$\epsilon \equiv \frac{1}{4!} \epsilon^{abcd} \gamma_a \gamma_b \gamma_c \gamma_d = \gamma_0 \gamma_1 \gamma_2 \gamma_3 \implies \epsilon^{abcd} = \epsilon \cdot (\gamma^d \wedge \gamma^c \wedge \gamma^b \wedge \gamma^a)$$

The object $\epsilon$ is called the pseudoscalar (and also sometimes denoted $\gamma_5$), and it represents a unit 4-volume.

Using this notions, we can attack index notation expressions more or less "directly". Let's look at our duality operation:

$$\begin{align*}G^{cd} = G \cdot (\gamma^d \gamma^c) &= \frac{1}{2} F \cdot (\gamma_b \wedge \gamma_a) \epsilon \cdot (\gamma^d \wedge \gamma^c \wedge \gamma^b \wedge \gamma^a) \\ &= \frac{1}{2} [F \cdot (\gamma_b \wedge \gamma_a) \gamma^b \wedge \gamma^a] \epsilon \cdot (\gamma^d \wedge \gamma^c) \\ &= (F \epsilon) \cdot (\gamma^d \wedge \gamma^c)\end{align*}$$

So $G = F \epsilon$, and so $\star G = G \epsilon = F \epsilon \epsilon = -F$. All this follows from the algebraic properties of the $\gamma_a$.

Similarly, proving the proper identity for the Levi-Civita tensors is easy to do with Clifford algebra.

$$\begin{align*} \epsilon_{abcd} \epsilon^{abef} &= \epsilon (\gamma^d \wedge \gamma^c \wedge\gamma^b \wedge \gamma^a) \epsilon (\gamma_f \wedge \gamma_e \wedge \gamma_b \wedge \gamma_a) \\ &= (-2)(-1) (\gamma^d \gamma^c) \cdot ( \gamma_f \gamma_e) \\ &= 2(\eta^c_f \eta^d_e - \eta^c_e \eta^d_f ) \\ &= 4 \eta^{[c}_f \eta^{d]}_e \end{align*} $$

share|cite|improve this answer
Thank you, Muphrid. How do I evaluate $\epsilon_{abcd} \epsilon^{abef}$? – user55309 Jan 5 '13 at 22:51
@user55309 I've added a section on how to do this, in supplement to oen's explanation. – Muphrid Jan 6 '13 at 5:12
Thank you, Muphrid. This is very helpful! :) – user55309 Jan 7 '13 at 20:06

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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