4
$\begingroup$

I have a question regarding interchanging the order of three nested summations. My expression looks like

\begin{align} \sum_{n=0}^\infty \sum_{k=0}^n \sum_{\nu=0}^{4n-2k}\frac{C_{nk\nu}}{k!(n-k)!}\binom{4n-2k}{\nu}, \end{align}

where $C_{nk\nu}$ is a term depending on the three indices which, for the purposes of this question is irrelevant. What I want to do (assuming convergence) is to put the $\nu$ sum to the far most left, however I am not sure about the proper summation indices after interchanging the sums. For instance, I know that due to the binomial coefficient $\binom{4n-2k}{\nu}$ and the $(n-k)!$ factor we have the following summation constraints

\begin{align} n-k\geq 0 \hspace{5mm} \Longrightarrow k \leq n, \end{align} \begin{align} 4n-2k-\nu\geq 0 \hspace{5mm} \Longrightarrow k \leq \dfrac{4n-\nu}{2}. \end{align}

From which I can "guess" the new summation indices after interchanging the $k \leftrightarrow \nu$ sums:

\begin{align} \sum_{n=0}^\infty \sum_{k=0}^n \sum_{\nu=0}^{4n-2k}\frac{C_{nk\nu}}{k!(n-k)!}\binom{4n-2k}{\nu}=\sum_{n=0}^\infty \sum_{\nu=0}^\infty \sum_{k=0}^{\max\left(\lfloor \frac{4n-\nu}{2}\rfloor,n\right)}\frac{C_{nk\nu}}{k!(n-k)!}\binom{4n-2k}{\nu}, \end{align}

However this is only true for the cases when $n=0$ and $\forall \ \nu$ but for instance when $n=1$ and $\nu=0$, $\lfloor \dfrac{4(1)-0}{2} \rfloor=\lfloor 2 \rfloor = 2$ so the $k$ sum will run up to $2$ but then the term $(n-k)!=(1-(2))!=(-1)!$ is undefined so the upper limit that I put is correct for some values but incorrect for others, I have been trying to do some modifications to my upper limit but so far I have failed. Do you have any suggestions to properly flip the summations?

Thanks in advance :)

$\endgroup$
1
  • 1
    $\begingroup$ You can attempt to write all 3 sums as infinite series by adding a factor which is zero whenever the indices are not included into the sym and 1 in otherwise, then, assuming your $C$ constants are nonnegative, you should be able to permute the summation signs as you please. In the end you can see how the given constants reduce your other two sums. $\endgroup$
    – Boxonix
    Commented Feb 28, 2023 at 16:46

2 Answers 2

3
$\begingroup$

Since we are interested in a transformation of the index regions only we consider wlog the sum \begin{align*} \sum_{n=0}^{\infty}\sum_{k=0}^n\sum_{\nu=0}^{4n-2k} a_{n,k,\nu} \end{align*}

We obtain \begin{align*} \color{blue}{\sum_{n=0}^{\infty}\sum_{k=0}^{n}\sum_{\nu=0}^{4n-2k}a_{n,k,\nu}} &=\sum_{n=0}^{\infty}\sum_{k=0}^n\sum_{\nu=0}^{2n+2k}a_{n,n-k,\nu}\tag{1}\\ &=\sum_{k=0}^{\infty}\sum_{n=k}^{\infty}\sum_{\nu=0}^{2n+2k}a_{n,n-k,\nu}\tag{2}\\ &=\sum_{k=0}^{\infty}\sum_{n=2k}^{\infty}\sum_{\nu=0}^{2n}a_{n-k,n-2k,\nu}\tag{3}\\ &=\sum_{n=0}^{\infty}\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor}\sum_{\nu=0}^{2n}a_{n-k,n-2k,\nu}\tag{4}\\ &=\sum_{n=0}^{\infty}\sum_{\nu=0}^{2n}\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor}a_{n-k,n-2k,\nu}\tag{5}\\ &\,\,\color{blue}{=\sum_{\nu=0}^{\infty}\sum_{n=\left\lfloor\frac{\nu}{2}\right\rfloor}^{\infty} \sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor}a_{n-k,n-2k,\nu}}\tag{6} \end{align*}

Comment:

  • In (1) we change the order of summation of the middle sum $k\to n-k$.

  • In (2) we exchange left and middle sum noting \begin{align*} \sum_{n=0}^{\infty}\sum_{k=0}^n \Box=\sum_{0\leq k\leq n<\infty}\Box=\sum_{k=0}^{\infty}\sum_{n=k}^{\infty}\Box \end{align*}

  • In (3) we shift the index of the middle sum by $k$ to get rid of $k$ in the upper limit of the right sum.

  • In (4) we exchange left and middle sum noting the equivalence of \begin{align*} &0\leq k<\infty&&0\leq n<\infty\\ &2k\leq n<\infty&&0\leq k\leq \left\lfloor\frac{n}{2}\right\rfloor \end{align*}

  • In (5) we exchange middle and right sum.

  • In (6) we exchange left and middle sum noting the equivalence of \begin{align*} &0\leq n<\infty&&0\leq \nu<\infty\\ &0\leq \nu\leq 2n&&n\leq \left\lfloor\frac{\nu}{2}\right\rfloor<\infty \end{align*}

$\endgroup$
2
  • $\begingroup$ Thanks for your answer. I found very smart the way you got rid of the annoying 1/4 factor in the floor function. I have a question in step (6) though. The boundary for the $n$ sum shouldn't be $n\leq \lfloor \frac{\nu}{2}\rfloor<\infty$ ? $\endgroup$
    – Hey
    Commented Mar 6, 2023 at 15:30
  • $\begingroup$ @Hey: Your're welcome and yes, you're right. :-) Thanks for pointing at the mistake, it is now corrected. $\endgroup$ Commented Mar 6, 2023 at 16:18
1
$\begingroup$

Since by definition $\binom pq=0$ whenever $q>p,$ your sum (assuming absolute convergence) can simply be written $$\sum_{n,k,\nu\ge0}\frac{C_{n,k,\nu}}{n!}\binom nk\binom{4n-2k}\nu$$ or, if you "want to put the $ν$ sum to the far most left": $$\sum_{\nu\ge0}\sum_{n,k\ge0}\frac{C_{n,k,\nu}}{n!}\binom nk\binom{4n-2k}\nu$$ $$=\sum_{\nu\ge0}\sum_{k\ge0}\sum_{n\ge\max(k,(2k+\nu)/4)}\frac{C_{n,k,\nu}}{k!(n-k)!}\binom{4n-2k}\nu$$ $$=\sum_{\nu\ge0}\sum_{n\ge\nu/4}\sum_{0\le k\le\min(n,(4n-\nu)/2)}\frac{C_{n,k,\nu}}{k!(n-k)!}\binom{4n-2k}\nu.$$ Note the $\min$ in this last line, instead of your annoying $\max.$ This is because $k\le a$ and $k\le b$ is not equivalent to $k\le\max(a,b)$ but to $k\le\min(a,b).$

Edit, after seeing epi163sqrt's answer:

The expression can be slightly improved by the change of index $m=2n-k:$ $$\sum_{0\le k\le n,\;0\le\nu\le4n-2k}\frac{C_{n,k,\nu}}{k!(n-k)!}\binom{4n-2k}\nu$$ $$=\sum_{0\le n\le m\le 2n,\;0\le\nu\le2m}\frac{C_{n,2n-m,\nu}}{(2n-m)!(m-n)!}\binom{2m}\nu$$ $$=\sum_{\nu\ge0}\sum_{m\ge\nu/2}\sum_{m/2\le n\le m}\frac{C_{n,2n-m,\nu}}{(2n-m)!(m-n)!}\binom{2m}\nu.$$

$\endgroup$
3
  • $\begingroup$ I see. I wasn't using the min because when $n=0$; $\frac{2n-\nu}{2}$ is negative, but I just realised you "protected" that by starting the n sum at the $\lfloor\nu/4\rfloor$ value. However, for later purposes in my calculation, it would be nice to rewrite the n sum differently, this is because I will need to consider the even and odd cases separately and the 1/4 factor in the floor function is annoying. $\endgroup$
    – Hey
    Commented Mar 1, 2023 at 10:42
  • $\begingroup$ I understand that this answer does not satisfy you, but not what you want precisely. Can you please edit your question (though it is unusual to do so when it has already got answers) to let us know exactly your wish? $\endgroup$ Commented Mar 1, 2023 at 11:27
  • 1
    $\begingroup$ Your answer solved my question. Let's say that now I am asking whether is possible to rewrite the indices such that there isn't a 1/4 factor in the floor functions. This is because later in my calculation I'll require to get rid of the floor function, I do it by separating my $\nu$ sum in odd/even values, however with $\lfloor\nu/4\rfloor$ I can't get rid of the floor function because I get $\lfloor \nu/2\rceil$ for even $\nu$ and $\lfloor (2\nu+1)/2\rfloor$ for odd values. $\endgroup$
    – Hey
    Commented Mar 1, 2023 at 12:36

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .