Is it possible to construct a Lorentz invariant, three rank Levi-Civita tensor in Minkowski Spacetime? If not, why so? I am talking about something like this $\epsilon_{\alpha\beta\gamma}$ or $\epsilon^{\alpha\beta\gamma}$, where each indices run from $0$ to $3$. As in this answer here, which proves the Lorentz co-variance of the Levi-Civita tensor by using the determinant formula, I guess one would run into trouble if we have three rank Levi-Civita tensors. Kindly elaborate on that.

  • $\begingroup$ The levi civita symbol is defined to be this: $\epsilon_{0123}=1$ and every odd permutation of 0123 would have value of $\epsilon$ to be zero. So, how would we define a 3 ranked levi civita symbols in 4 dimensions? $\endgroup$ Commented Jun 6, 2018 at 14:14

2 Answers 2


You can use Young tableaux/diagrams and the permutation group to figure out the symmetries of the general rank-3 tensor. The spaces correspond to the partitions of the rank:


One 20 dimensional total symmetric subspace.


Two 20 dimensional mixed symmetry subspaces.


One 4 dimensional totally antisymmetric subspace:

$$ A_{\alpha\beta\gamma} = \frac 1 6 [T_{\alpha\beta\gamma} + T_{\beta\gamma\alpha} + T_{\gamma\alpha\beta} - T_{\gamma\beta\alpha} - T_{\beta\alpha\gamma} - T_{\alpha\gamma\beta}] $$

That is the only antisymmetric thing you can make according to Schurl-Weyl theory.

To find the dimensions, I used the Hook Length Formula (summing of the boxes $x$ in a diagram $Y(\lambda)$) for the Young diagram corresponding to the integer partition:

$$ {\rm dim}\pi_{\lambda} = \frac {n!}{\prod_{x\in Y}{\rm hook}(x)}$$

If you consider 3 dimensions ($n=3$), you get ${\rm dim} = 1$, that is the standard Levi-Civita symbol $\epsilon_{ijk}$.

If you set $n=4$, the result is ${\rm dim} = 4$.

That means $A_{\alpha\beta\gamma}$ transforms like a 4-vector.

So, the only antisymmetric part of a rank-3 tensor in Minkowski space rotates like a 4-vector, which means it is not invariant and is not a candidate to be Levi-Civita like.

Meanwhile, the dimensions of the 3 other irreducible spaces are all 20--which are certainly not scalars, and thus not candidates to be Levi-Civita like.

Note that if you consider rank-4 tensors, the partitions are as follows:


35 dimensional and symmetric.


Three 45-dimensional mixed symmetry spaces.


Two 20-dimensional mixed symmetry spaces.


Three 15-dimensional mixed symmetry spaces.


One total antisymmetric 1 dimensional space, which is proportional to the Levi-Civita symbol $\epsilon_{\mu\nu\sigma\lambda}$.

In summary, the answer is "No", and the reason why has to do with the representations of the symmetric group on 3-letters. You partition the rank=3, use the Robinson-Schensted correspondence to associate that partition with irreducible representations of the permutation group. (The Young Diagrams make this step a snap). Then, Schur-Weyl duality associates those with irreducible subspaces of and rank-N tensor (signed permutations of indices). Finally, the Hook Length formula tells you the dimensions of those subspaces.

The Levi-Civita symbol needs to be invariant (e.g., dimension 1, like a scalar) and it need to be totally antisymmetric in all indices--and that simply did not exist for rank 3 in 4 dimensions.

  • $\begingroup$ The answer is much more engaged than I thought it would be. It will take me a bit of time to understand the whole thing. Can you please refer something where I can read about Young tableaux, partitions and Hook Length formula etc.? Thank you. $\endgroup$ Commented Jun 5, 2018 at 23:25
  • 1
    $\begingroup$ @fogofmylife I had to cull many references. Most are too mathy for us. Finite group theory is abstract, but a starting point is yufeizhao.com/research/youngtab-hcmr.pdf which covers the permutation group and Young tableaux. That the Young symmetrizer applied to indices yields the irreducible subspaces for all ranks in any dimension over all fields (Schur-Weyl Duality) is not clear and I know of no physics oriented reference. It's best to try it by hand for 0=0 (scalars), 1=1 (vectors), 2=2,1+1 (tensors), 3=3, 2+1, 1+1+1 (the 1st nontrivial one) by hand and see what's happening. Goodluck $\endgroup$
    – JEB
    Commented Jun 6, 2018 at 3:02
  • 1
    $\begingroup$ @fogofmylife: The best physics-oriented reference for the Schur-Weyl duality is probably Hamermesh's Group Theory and Its Application to Physical Problems, but even that one isn't the clearest. $\endgroup$ Commented Jun 6, 2018 at 14:47

Does this satisfy the requirements?

In Minkowski spacetime (signature: -,+,+,+), let $\epsilon_{abcd}$ be the alternating tensor satisfying $\epsilon_{abcd}=\epsilon_{[abcd]}$ and $\epsilon_{abcd}\epsilon^{abcd}=-24$ and let $v^a$ be the 4-velocity of an observer ($v^av_a=-1$).

Define the "spatial alternating tensor seen by the observer $v^a$" $$\epsilon_{abc}=\epsilon_{abcd}v^d,$$ which satisfies $\epsilon_{abc}=\epsilon_{[abc]}$, $\epsilon_{abc}\epsilon^{abc}=6$, and $v^a\epsilon_{abc}=0$.

(This is extracted from Robert Geroch's "General Relativity, 1972 Lecture Notes" [ISBN 978-0987987174] .)

  • $\begingroup$ This wouldn't be Lorentz-invariant, since its definition depends on the four-vector $v^a$ chosen, and this vector will transform non-trivially between reference frames. $\endgroup$ Commented Jun 6, 2018 at 14:45
  • $\begingroup$ @MichaelSeifert , I agree. By construction, this certainly depends on the choice of four-vector. $\endgroup$
    – robphy
    Commented Jun 6, 2018 at 15:03

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.