I am currently working on my bachelor's thesis on the anapole / toroidal moment and it seems that I am stuck with a tensor decomposition problem.

I have actually never had a course about tensors, so I am a complete newbie.

I need to expand a localized current density, which is done by expressing the current via delta distribution and expanding the latter:

$$\vec{j}(\vec{r},t) = \int\vec{j}(\vec{\xi},t) \delta(\vec{\xi}-\vec{r}) d^3\xi$$ $$\delta(\vec{\xi}-\vec{r}) = \sum_{l=0}^{\infty} \frac{(-1)^l}{l!} \xi_i ...\xi_k \nabla_i ... \nabla_k \delta(\vec{r}) $$

So I get some result containing the following tensor:

$$B_{ij...k}^{(l)}(t) := \frac{(-1)^{l-1}}{(l-1)!} \int j_i \xi_j ... \xi_k d^3\xi$$

So far, I have understood the math. But now comes the tricky part. In the paper, it says that "we can decompose the tensors $B_{ij...k}^{(l)}$ into irreducible tensors, separating the various multipole moments and radii." and further "...of third rank, $B_{ijk}^{(3)}$ can obviously reduced according to the scheme $1 \times (2+0) = (3+1)+2+1$. It can be seen that the representation of weight $l=1$ is extracted twice from $B_{ijk}^{(3)}$." And then follows what seems like the decomposition and I am hopelessly lost.

$$j_i\xi_j\xi_k = \frac{1}{3} \left[ j_i\xi_j\xi_k + j_k\xi_i\xi_j + j_j\xi_k\xi_i - \frac{1}{5} \left( \delta_{ij}\theta_k + \delta_{ik}\theta_j + \delta_{jk}\theta_i \right) \right] - \frac{1}{3} \left( \epsilon_{ijl} \mu_{kl} + \epsilon_{ikl}\mu_{jl}\right)$$ $$+ \frac{1}{6} \left( \delta_{ij}\lambda_k + \delta_{ik}\lambda_j - 2 \delta_{jk}\lambda_i \right) + \frac{1}{5} \left( \delta_{ij}\theta_k + \delta_{ik}\theta_j + \delta_{jk}\theta_i \right)$$


$$\mu_{ik} = \mu_i\xi_k + \mu_k\xi_i \ , \ \mu_i=\frac{1}{2} \epsilon_{ijk}\xi_j j_k$$ $$\theta_i=2\xi_i \vec{\xi}\cdot \vec{j} + \xi^2 j_i$$ $$\lambda_i=\xi_i\vec{\xi}\cdot \vec{j} - \xi^2 j_i$$

This decomposition obviously contains many quantities that later on appear also in the multipole expansion, e.g. the magnetic quadrupole moment $\mu_{ik}$. So on the physics side of things, this makes sense to me.

But not on the mathematical side. On this board I found some questions regarding tensor decomposition and in the answers I learned something about symmetric and antisymmetric tensors and that every tensor can be decomposed in several irreducible ones, which better represent physical properties of the system and symmetries.

But I still, some questions are still there... 1.) What do the numbers $\frac{1}{3}$, $\frac{1}{5}$, etc. mean? Is this some kind of normalization? 2.) How exactly does one decompose the tensor? How can I reconstruct what exactly has been done, which steps one has to follow to decompose it like this?

  • $\begingroup$ $\uparrow$ Which paper? $\endgroup$ – Qmechanic May 12 '14 at 23:20
  • $\begingroup$ "Toroid Moments in Electrodynamics and Solid-State Physics" by V.M. Dubovik and V.V. Tugushev. Should I add this every time? $\endgroup$ – user46208 May 13 '14 at 11:01

This appears to be related to the decomposition of a totally symmetric tensor into traceless parts, which is a fairly involved process. The general equation is $$\mathcal{C} Q_{a_1 a_2\cdots a_s} = \sum_{k=0}^{[\frac{s}{2}]} (-1)^s \frac{\binom{s}{k} \binom{s}{2k}}{ \binom{2s}{2k}} \delta_{(a_1 a_2} \cdots \delta_{a_{2k-1} a_{2k}} Q_{a_{2k+1}\cdots a_s)}{}^{c_1} {}_{c_1} {}^{c_2}{}_{c_2} {}^{\cdots c_k}{}_{\cdots c_k},$$ where $[\cdot]$ denotes the integer part, Einstein summation is implied and $Q_{(a_1 a_2 \cdots a_s)} \equiv \frac{1}{s!} \sum_{\sigma\in S_s} Q_{a_{\sigma(1)} a_{\sigma(2)} a_{\sigma(3)} \cdots a_{\sigma(s)}}$. For the quadrupole moment it is $\mathcal{C}Q_{ab} = Q_{ab} - \frac{1}{3} Q^c{}_c \delta_{ab}$, for the octupole $\mathcal{C}Q_{abc} = Q_{abc} - \frac{1}{5} (Q^d{}_{dc}\delta_{ab} + Q^d{}_{da} \delta_{bc}+ Q^d{}_{db}\delta_{ac})$; these yield the factors in your question.

An indication (perhaps proof, although I'm not certain about this at the moment) that the traceless part of a totally symmetric tensor is an irreducible representation is easy to see if one uses the hook formula in dimension 3. A totally symmetric tensor of rank $s$ has $\frac{1}{2}(s+1)(s+2)$ degrees of freedom and the traceless one has the latter minus the number of ways to obtain the traces, $\binom{s}{2}$, which yields $2s+1$. This is the dimension of the irreducible representation of the algebra of SO(3) with spin $s$. A full proof of this statement is in Maggiore Gravitational waves - theory and experiments.

Reference: F.A.E. Pirani Lectures on General Relativity 1965.

| cite | improve this answer | |
  • $\begingroup$ To add to this answer: see Manheim's appendix in "Brane Localized Gravity" for a full derivation of various decompositions of tensors. $\endgroup$ – JamalS May 10 '14 at 21:16
  • $\begingroup$ Thanks for your answer! This has definitely helped, I have taken a look in the decomposition part of Maggiore's lecture. A general question I still have, though, is how to calculate the trace itself, e.g. in $$Q_{\alpha \beta \gamma} = I_{\alpha\beta\gamma} - a (\delta_{(\alpha\beta}I_{\gamma)})_{\mu\mu}$$ Why does the author then calculate $$Q_{\beta\beta\gamma} = I_{\beta\beta\gamma} - \frac{1}{3} a (\delta_{\beta\beta} I_{\gamma\mu\mu} + \delta_{\beta\gamma} I_{\beta\mu\mu} + \delta_{\gamma\beta} I_{\beta\mu\mu}) = I_{\beta\beta\gamma} - \frac{5}{3} a I_{\beta\beta\gamma}$$ ? $\endgroup$ – user46208 May 16 '14 at 13:35
  • $\begingroup$ To specify my question more, things that are unclear to me are: + what does the notation $(\alpha \beta \quad \gamma)\mu\mu$ mean? + why is the trace calculated via $Q_{\beta\beta\gamma}$ and why is the "weight" $\frac{1}{3}$? $\endgroup$ – user46208 May 16 '14 at 13:53
  • $\begingroup$ @user46208 I didn't answer your question directly, why did you accept my answer? I'll have a look at the article over the weekend. $\endgroup$ – auxsvr May 16 '14 at 15:21
  • $\begingroup$ So sorry, I am absolutely new to this platform! I promise to think twice next time! $\endgroup$ – user46208 May 16 '14 at 20:47

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.