Count degrees of freedom symmetric tensor

807 Views Asked by At

How do I count the numer of degrees of freedom in a totally symmetric tensor with $m$ lower (or upper) indices taking values in $\{ 1,2,3 \}$?

I know that the result is $$dim(m,0)=\frac{(m+2)(m+1)}{2}$$

For a symmetric tensor that takes values $\{ 1,2 \}$ I get the counting: indices $1$ can be $m,m-1,...,1,0$ and therefore $dim=m+1$.

For three possible values my assignment says we can split m objects into three sets (first splitting m objects into two sets, and then splitting the second part in two again). Can someone, in more detail, explain the counting or splitting in sets?

1

There are 1 best solutions below

0
On BEST ANSWER

Values in the tensor are indexed by $m$-tuples $(a_1,a_2...a_m)$, where $a_p \in \{1..k\}$ (in your case k = 3). The symmetry imposes an equivalence relation on these tuples, where two tuples are equivalent if they contain the same values, regardless or order. For example, in a totally symmetric $4$-tensor, the index $(5,1,2,2)$ gives the same value as the index $(1,2,5,2)$. Now from each such equivalence class of tuples we select a "canonical" member which will represent the class.

This will be the $m$-tuple for which all the values are ordered. For our example this gives $(1,2,2,5)$. We need to count the $m$-tuples $(a_1,a_2...a_m)$ such that $(j<s)\rightarrow (a_j \leq a_s)$ . For $a_p \in \{1..k\}$ this gives us exactly $C^{k+m-1}_{m}$:

We make a bijection taking $(a_1,a_2,a_3,...a_m)$ to $(a_1,a_2+1,a_3+2,...a_m+(m-1))$. If $(a_1,a_2,a_3,...a_m)$ is (non-strictly) ordered then $(b_1,b_2,b_3,...b_m) =(a_1,a_2+1,a_3+2,...a_m+(m-1))$ is strictly ordered an each of its terms is distinct.

As we have $a_p \in \{1..k\}$, we have $b_p \in \{1,..k+m-1\}$. The counting of $m$ distinct ordered terms with values from $1$ to $k+m-1$ is done by combinations as $C^{k+m-1}_{m}$. By the above bijection, this is equivalent to counting $m$ (not-necessarily distinct) ordered terms with values form $1$ to $k$, which index the degrees of freedom of a totally symmetric tensor with $m$ indices in dimension $k$.

Setting $k=3$ we get $C^{3+m-1}_{m} = C^{2+m}_{m} = \frac{(m+2)!}{m!2!} = \frac{(m+1)(m+2)}{2}$.