I can turn-the-crank and show that $\frac{1}{2}\otimes \frac{1}{2} = 1\oplus 0$ etc, but what would be a strategy to proving the general statement for spin representations that $j\otimes s =\bigoplus_{l=|s-j|}^{|s+j|} l$.


It's easy to prove the formula if you just look at the individual basis vectors of the tensor product. Let's use $(2j+1)$ and $(2s+1)$ eigenvectors of $j^2, j_z$ and $s^2, s_z$ called $|j,j_z\rangle$ and so on.

Now let's ask about the multiplicity of basis vectors of the tensor product with a given eigenvalue of $J_z = j_z+s_z$. The maximum eigenvalue of $J_z$ in the tensor product is $j+s$: it can be obtained if we choose $$ |J,J_z=j+s\rangle = |j,j\rangle \otimes |s,s\rangle $$ There are no higher eigenvalues of $J_z$; this proves that no representation with $J>j+s$ is included in the tensor product. However, the $J=j+s$ representation must be included exactly once to obtain one basis vector with $J_z=j+s$; representations with lower values of $J<j+s$ wouldn't contribute any vectors with $J_z=j+s$.

So the tensor product $$ Rep(j) \otimes Rep(s) = Rep(j+s)\oplus Rep(rest) $$ I have used the fact that reducible representations of simple compact groups may be written as direct sums. Now, what about the remaining representation(s) $Rep(rest)$? It is the linear envelope of a set of basis vectors in which we have already removed all the $(2J+1)$ basis vectors with $J=j+s$.

Well, in the rest, the maximum allowed $J$ is $j+s-1$. From the original bases, we see that the original space was 2-dimensional: the old basis included $$|j,j-1\rangle \otimes |s,s\rangle, \qquad |j,j\rangle \otimes |s,s-1\rangle $$ But we have already included one combination to $Rep(j+s)$; so the $Rep(rest)$ representation only contains the other one. By the same argument as above, we may see that the multiplicity of $Rep(j+s-1)$ in the tensor product is also one.

By induction, this algorithm may continue: at the beginning, the number of eigenvectors with a given $J_z$ is increasing by one every time we decrease $J_z$ by one. However, this behavior stops once we get to too low values of $J_z$ that would require too negative values of $j_z$, either $j_z<-j$ or $s_z<-s$. When that happens, the number of basis vectors no longer jumps by one; it stays constant. It happens when $$J_z^{max} \equiv J = |j-s|$$ so $J=|j-s$ is the lowest-$J$ representation included in the decomposition of the tensor product. Another way to see that at this moment, we have already written down all components, is either to notice that $J$ can't be smaller than $|j-s|$ because the minimum is obtained by adding "oppositely directed vectors" and can't be further shortened; or, alternatively, we may check that the dimensions of your formula work: $$ (2j+1)(2s+1) = \sum_{J=|j-s|}^{(j+s)} (2J+1) $$


I think the best way is not the textbook way, but the one described in the warm-up problem in this answer: Mathematically, what is color charge?

I will repeat the main point: the irreducible represenations of SU(2) are given by all complex completely symmetric tensors with all indices down, where each index takes two values 0,1. This is because you have invariant $\epsilon_{ij}$ tensor, which you can use as a metric to raise and lower indices, and you can remove the antisymmetric parts using the $\epsilon$ tensor. The fully symmetric k-index tensor is the spin k/2 representation.

When you tensor two of these together of size k and m, you just put the two tensors end to end, which gives a reducible k+m tensor. You need to remove the antisymmetric parts using the $\epsilon$ tensor, and this steps down by 2 each time, producing exactly one representation of every size between k+m (where you start) and k-m (when you run out of indices to contract).

  • 1
    $\begingroup$ Right, this is the most elegant answer for this particular group and set of reps. $\endgroup$ – Luboš Motl Oct 24 '11 at 8:41

You can solve this by examining characters. Any finite dimensional representation of $SU(2)$ breaks up into a sum of 1-dimensional representations of $U(1)$, which are classified by an integer called weight (you may divide by 2 to get a half-integer if you need to conform to physics conventions). You can then write the decomposition of a representation as a generating function, by adding the monomial $q^n$ for each representation of $U(1)$ of weight $n$ that you see. For example, the representation $\frac12$ corresponds to $q + q^{-1}$, and the representation $1$ corresponds to $q^2 + 1 + q^{-2}$.

The two facts you need are then:

  1. For any non-negative integer $k$, the irreducible representation $k/2$ has character of the form $$\frac{q^{k+1} - q^{-k-1}}{q - q^{-1}} = q^{k} + q^{k-2} + \cdots + q^{-k}$$

  2. The character of a tensor product is the product of characters.

In other words, the tensor product decomposition can be reconstructed by forgetting the $SU(2)$ action, taking the tensor product of the underlying $U(1)$ representations, then remembering that the characters of irreducible $SU(2)$ representations have a special form.

In your example, squaring $q + q^{-1}$ yields $q^2 + 2 + q^{-2}$. You then subtract $q^2 + 1 + q^{-2}$ to get $1$. For your more general question, you want to show that: $$\frac{q^{j+1} - q^{-j-1}}{q - q^{-1}}\frac{q^{s+1} - q^{-s-1}}{q - q^{-1}} = \sum_{\ell = |s-j|}^{s+j} \frac{q^{l+1} - q^{-l-1}}{q - q^{-1}}$$ You can prove this by induction on $s$: Your base cases are $s=0$ and $s=1$, which are relatively easy to check. For larger $s$, you can do a reduction by splitting the sum $\frac{q^{s+1} - q^{-s-1}}{q - q^{-1}}$ as $(q^s + q^{-s}) + \frac{q^{s-1} - q^{-s+1}}{q - q^{-1}}$. Note that $(q^s + q^{-s})(q^j + q^{j-2} + \cdots + q^{-j})$ is expanded as $$(q^{s+j} + q^{s+j-2} + \cdots + q^{-s-j}) + (q^{|s-j|} + q^{|s-j|-2} + \cdots + q^{-|s-j|}).$$ These are the extreme summands in $\sum_{\ell = |s-j|}^{s+j} \frac{q^{l+1} - q^{-l-1}}{q - q^{-1}}$. The remaining summands are what you get by replacing $s$ by $s-2$.

There is also a combinatorial method (which would be easier to communicate if I could draw): view the Laurent polynomial $q^j + \cdots + q^{-j}$ as a set of evenly spaced dots on the number line, and view each of the monomials in $q^{s} + \cdots + q^{-s}$ as a shifting operator. You get a bunch of shifted sets of dots, and you can reorganize them into a sort of symmetric pile. Each layer of the pile is one of the new irreducible representations.


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.