2
$\begingroup$

EDIT: I thought on rephrasing the question in another way:

I have been working recently with a tensor that satisfies

$A_{ijkl}=A_{i+b,j+b,k+b,l+b}$ $\forall$ $i,j,k,l$ $\in$ Z

$$dist(i,j,k,l)\leq M$$

where all indices are meant to be integers (also b with $b\geq 0$), and dis(i,j,k,l) is the distance between all pair of indices, so $|i-j|\leq M$, $|i-k|\leq M$ etc... with 6 total distances. Because of this symmetry, it is said that one can just fix one of the values of the indices, say $i=0,1,..,b-1$ and generate the other elements from the symmetry relation above.

I am struggling to see this, as for example, considering the case of a matrix $C_{i+b,j+b}=C_{i,j}$ it is clear to me how can one do this: one just calculates values for a single row, say $i=0$ and since $C_{ij}=0\forall |i-j|>M$, one is left with $2M+1$ independent terms. Then, the rest of matrix elements can be derived by using the symmetry relation:

$$C_{i+b,j+b}=C_{i,j}$$

However, for the $A$ tensor above I am having difficulties to see this and how one could in principle recover all the missing elements of the set, if I set $i=0$ and calculate for the other indices. How can one get, for example $A_{1,1,2,3}$ if we only now those terms for $i=0$ ( that is, we know $A_{0,jkl}$ only ) for the case $b=1$?

Thanks !!

$\endgroup$
4
  • $\begingroup$ It is not true; you are only allowing simultaneous addition to all indices at once, not each index separately. Think again about matrices (which, in your setting, are I suppose of infinite dimension). Draw matrix entry $C_{ij}$ at position $(i,j)$. Your symmetry allows you to equate entries as you move by a vector $(r,r)$ in the plane, not by $(0,r)$ or $(r,0)$. $\endgroup$
    – Ben McKay
    May 18, 2020 at 12:48
  • $\begingroup$ I forgot to add that the tensor must satisfy $dist(i,j,k,l)\leq N$. This is also confusing me, as what is meant here by dist(i,j,k,l) is meant by the pair combinations of indices, so $|i-j|\leq N$ , $|i-k|\leq N$. $|i-l|\leq N$ ... and so on. Also, the matrix is required to satisfy the symmetry above with $|i-j| \leq N$. In that case, you can fix one of the matrix indices $i$ and compute a single row; the rest follows by symmery but this is what I don't see in the tensor case $\endgroup$ May 18, 2020 at 13:00
  • $\begingroup$ So, you want to count the set of integer $4$-tuples $(i,j,k,l)$ with $|i|,|j|,|k|,|l|,|i-j|,\dotsc,|k-l|\leq N$, modulo the equivalence relation $(i,j,k,l)\sim(i+b,j+b,k+b,l+b)$? $\endgroup$ May 18, 2020 at 15:50
  • $\begingroup$ Yes, but what is confusing me is the notation $dist(i,j,k,l)$, is this supposed to represent pair-wise distances between the integers? For the case of a matrix, this would mean that elements of distance $|i-j|>N$ away from the diagonal are zero. $\endgroup$ May 18, 2020 at 16:02

1 Answer 1

0
$\begingroup$

Because of this symmetry, it is said that one can just fix one of the values of the indices, say $i=0,1,..,b-1$ and generate the other elements from the symmetry relation above.

The issue is very simple. For each integer $i’$ there exists $i $ defined above such that $i’=i+tb$ for some integer $t$. Then $$A_{i’j’k’l’}=A_{i+tb,(j’-tb)+tb, (k’-tb)+tb,(l’-tb)+tb}= A_{i,j’-tb, k’-tb,l’-tb}.$$

Remark that the addition of $tb$ to each index keeps the distances between the indices.

for example $A_{1,1,2,3}$ if we only now those terms for $i=0$ ( that is, we know $A_{0,jkl}$ only ) for the case $b=1$?

$A_{1,1,2,3}=A_{0+b,0+b,1+b,2+b}= A_{0,0,1,2}$.

$\endgroup$
1
  • 1
    $\begingroup$ Yes indeed, many thanks for the reply! $\endgroup$ May 22, 2020 at 11:09

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.