Suppose we would like to anti-symmetrize a tensor $$T^{\mu_1, \mu_2,\ldots, \mu_n} = G^{[\mu_1, \mu_2,\ldots, \mu_r]} H^{[\mu_{r+1},\ldots, \mu_n]},$$ where $G$ and $H$ are anti-symmetric. One could do this iteratively by applying the anti-symmetrizer $$\Lambda_{1,2,\cdots, n} = \frac{1}{n}\left(1-\sum_{i=1}^{n-1} P_{i,n}\right)\Lambda_{1,2,\cdots, n-1}, \quad \Lambda_{1,2} = \frac{1 - P_{1,2}}{2}$$ where $P_{i,n}$ swaps the indices $i$ and $n$, for instance $$P_{1,2} T^{\mu_{1} \mu_2} = T^{\mu_2 \mu_1}$$ however this method makes no use of the fact that $G$ and $H$ are already anti-symmetrized. My question is whether there exists a simplified anti-symmtrizer which anti-symmetrizes $T$ with a minimal amount of operations.

One immediate simplification is that we know that $\Lambda_{1,\ldots,r}$ will have no effect on the tensor $G$ and thus no effect on $T$ either. And something similar for $H$.

One idea is that this problem is somehow related to finding all partitions like $\{1,\ldots, r\}$ and $\{r+1,\ldots, n\}$ via swapping elements between them, but I cannot figure out what the correct algorithm should be. I do however know that there are a total of $\binom{n}{r}$ such partitions meaning that we should have a sum of that many products of operators in the end. In the end, this problem is related to a few problems in both relativity and fermionic statistics, but I cannot seem to find a solution anywhere.

  • $\begingroup$ To antisymmetrize $T$ you can use the generalized Kronecker Delta en.wikipedia.org/wiki/… $\endgroup$ Nov 22, 2020 at 16:59
  • $\begingroup$ Your suggestion amount to using the anti-symmetrizer, which is already mentioned in the question. The question is how to make use of the parts which are already anti-symmetrized to reduce the expression in the end. $\endgroup$ Nov 22, 2020 at 18:15

1 Answer 1


One can think of this problem as all different swaps with the components of the two tensors. With this in mind, we start by defining the index ordered tensors $$G^{\{\mu_1,\ldots, \mu_r\}} = G^{[\mu_{j_1},\ldots, \mu_{j_r}]},$$ where $(j_1,\ldots, j_r)$ is a permutation of $(1,\ldots,r)$ such that $$\mu_{j_1}< \mu_{j_2} <\ldots < \mu_{j_r}.$$ This definition is important since we then not longer have to worry about the ordering in a particular anti-symmetrized tensor, for instance $$G^{\{\mu_1, \mu_2, \ldots\}} = G^{\{\mu_2, \mu_1, \ldots\}}.$$ Then the anti-symmetrizing process is done with $$T^{[\mu_1\ldots \mu_n]} = \sum_{i_r = r}^{n} \sum_{i_{r-1}}^{i_r-1} \cdots \sum_{i_1 = 1}^{i_2 - 1} \text{sgn}\binom{1\cdots n}{i_1 \cdots i_n} P_{r,i_r} P_{r-1,i_{r-1}}\cdots P_{2,i_2} P_{1,i_1} G^{\{\mu_1,\ldots\mu_r\}} T^{\{\mu_{r+1},\ldots\mu_n\}}$$ where $i_{r+1},\ldots,i_n$ is the ordered list of all indices occurring in $T$. SO for instance if we had the case $$G^{\{\mu_3,\mu_4, \mu_6, \mu_7\}} T^{\{\mu_2,\mu_5,\mu_1\}},$$ then $(i_1,i_2,i_3,i_4,i_5,i_6, i_7) = (3,4,6,7, 1,2,5)$, since we first order the sublists. It turns out that under this swapping process the first $(i_1,\ldots, i_r)$ will in fact remain ordered. Ideally I would still like to simplify this somewhat into a recursive form such as that of the anti-symmetrizer.


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.