4
$\begingroup$

0. Background. This question is linked to a previous one: https://math.stackexchange.com/questions/3950321/computing-sums-of-exponential-partial-bell-polynomials. Based on the computation of the exponential partial Bell polynomial $B_{n,k}(2!,\ldots,(n-k+2)!)$ there (that I hope is correct), I managed to rewrite the problem and this led me to ask this new question.

1. The question. Can we explicitly compute $$S(\alpha',\beta')=\sum_{\gamma=0}^{\min\{\beta',\,\alpha'-\beta'\}}\frac{2^{\beta'-\gamma}(-1)^\gamma}{\gamma!(\beta'-\gamma)!}\frac{(2\beta')^{\overline{\alpha'-\beta'-\gamma}}}{(\alpha'-\beta'-\gamma)!}$$ where $\beta'\geq1$ and $2\beta'\geq\alpha'\geq0$? Here $x^{\overline{k}}:=x(x+1)\ldots(x+k-1)$ denotes the rising factorial. If no closed form can be found, an estimate will be enough. Below are displayed two attempts to deal with the problem ; any advice to go further would be very appreciated.

2.1. First attempt. Introduce the signed Lah number (see https://en.wikipedia.org/wiki/Lah_number): $$L(n,k):=(-1)^n\frac{n!(n-1)!}{k!(k-1)!(n-k)!}.$$ Then, putting $\gamma':=\alpha'-\beta'-\gamma$, we can write: \begin{align*} S(\alpha',\beta')&=\sum_{\gamma'=\min\{\alpha'-2\beta',\,0\}}^{\alpha'-\beta'}\frac{2^{2\beta'-\alpha'+\gamma'}(-1)^{\alpha'-\beta'-\gamma'}}{(\alpha'-\beta'-\gamma')!(2\beta'-\alpha'+\gamma')!}\frac{(2\beta')^{\overline{\gamma'}}}{\gamma'!}\\ % &=2^{2\beta'-\alpha'}\sum_{\gamma'=\min\{\alpha'-2\beta',\,0\}}^{\alpha'-\beta'}\frac{2^{\gamma'}(-1)^{\gamma'}}{(\alpha'-\beta'-\gamma')!(2\beta'-\alpha'+\gamma')!}\frac{(2\beta')^{\overline{\gamma'}}}{\gamma'!}\\ % &=\frac{2^{2\beta'-\alpha'}(-1)^{\alpha'-\beta'}}{(\alpha'-\beta')!(\alpha'-\beta'-1)!}\\&\quad\times\sum_{\gamma'=\min\{\alpha'-2\beta',\,0\}}^{\alpha'-\beta'}\frac{2^{\gamma'}(\gamma'-1)!}{(2\beta'-\alpha'+\gamma')!}(-1)^{\gamma'}\underbrace{(-1)^{\alpha'-\beta'}\frac{(\alpha'-\beta')!(\alpha'-\beta'-1)!}{\gamma'!(\gamma'-1)!(\alpha'-\beta'-\gamma')!}}_{=L(\alpha'-\beta',\gamma')}(2\beta')^{\overline{\gamma'}}\\ % &=\frac{2^{2\beta'-\alpha'}(-1)^{\alpha'-\beta'}}{(\alpha'-\beta')!(\alpha'-\beta'-1)!}\sum_{\gamma'=\min\{\alpha'-2\beta',\,0\}}^{\alpha'-\beta'}\frac{2^{\gamma'}(\gamma'-1)!}{(2\beta'-\alpha'+\gamma')!}(-1)^{\gamma'}L(\alpha'-\beta',\gamma')(2\beta')^{\overline{\gamma'}} \end{align*} The term $\frac{2^{\gamma'}(\gamma'-1)!}{(2\beta'-\alpha'+\gamma')!}$ and the lower bound $\min\{\alpha'-2\beta',\,0\}$ for the sum are annoying ; here we have something pretty close to $$\sum_{\gamma'=0}^{\alpha'-\beta'}(-1)^{\gamma'}L(\alpha'-\beta',\gamma')(2\beta')^{\overline{\gamma'}}=(2\beta')^{\underline{\alpha'-\beta'}}.$$ Here $x^{\underline{k}}:=x(x-1)\ldots(x-k+1)$ denotes the falling factorial.

2.2. Second attempt. We could as well write: \begin{align*} S(\alpha',\beta')&=\sum_{\gamma=0}^{\min\{\beta',\,\alpha'-\beta'\}}\frac{2^{\beta'-\gamma}(-1)^\gamma}{\gamma!(\beta'-\gamma)!}\frac{2\beta'(2\beta'+1)\ldots(\alpha'+\beta'-\gamma)}{(\alpha'-\beta'-\gamma)!}\\ % &=\frac{2}{(\beta'-1)!}\sum_{\gamma=0}^{\min\{\beta',\,\alpha'-\beta'\}}2^{\beta'-\gamma}(-1)^\gamma\binom{\beta'}{\gamma}\binom{\alpha'+\beta'-\gamma}{\alpha'-\beta'-\gamma}. \end{align*} Now this looks like a Vandermonde's identity: https://en.wikipedia.org/wiki/Vandermonde%27s_identity. This time the annoying term is $2^{\beta'-\gamma}(-1)^\gamma$.

EDIT. After having read some comments/answers, I think I can handle my problem IF I can get a bound such as $B_{n,k}(2!,3!,\ldots,(n-k+2)!)\leq n!$ or $-$ even better $-$ $B_{n,k}(2!,3!,\ldots,(n-k+2)!)\leq(n-k+1)!$, where I computed:


\begin{align*} B_{n,k}(2!,\ldots,(n-k+2)!)&=\sum_{j=0}^{\min\{k,\,n-k\}}\binom{k}{j}2^{k-j}(-1)^j\frac{(2k)^{(n-k-j)}n!}{(n-k-j)!k!}. \end{align*}
$\endgroup$
9
  • $\begingroup$ One thing I can assure is that: there is no closed form for this sum. $\endgroup$ Commented Dec 21, 2020 at 20:58
  • $\begingroup$ @T.Amdeberhan Because of $\min$? $\endgroup$ Commented Dec 21, 2020 at 21:01
  • $\begingroup$ I think there's no closed form too. But who knows... anyway, an estimate would be enough for my purposes. In particular, something like $(2\beta')!$ would be good. $\endgroup$
    – Nicolas
    Commented Dec 21, 2020 at 21:04
  • 1
    $\begingroup$ @მამუკაჯიბლაძე: the sum vanishes at values $\gamma>\min$ if all parameters are integers. So, $\min$ does not hurt. $\endgroup$ Commented Dec 21, 2020 at 22:23
  • 1
    $\begingroup$ Perhaps the weighted version of Lah numbers are useful, or at least simplify the expression. See here doi.org/10.1016/j.jcta.2020.105365 $\endgroup$ Commented Dec 21, 2020 at 23:43

2 Answers 2

4
+50
$\begingroup$

Probably this is not very helpful, but it is an explicit expression after all.

I get $$ S(a,b)=\frac{(-1)^{a+1}}{b!}{}_2F_1(-a+1,b+1;2b-a+1;2)\binom b{2b-a}2^{2b-a}. $$ This follows from $$ S(a,b)=\sum_{\ell=2b-a}^b(-1)^\ell\binom{a+\ell-1}{2b-1}\binom b\ell2^\ell, $$ which in turn I derived from the generating function $$ \sum_{a,b}S(a,b)t^au^b=e^{\frac{2-t}{(1-t)^2}tu}. $$

Note also that $S(a,b)$ is defined for other values of $a$ and $b$. In particular, for $a\geqslant2b$ one obtains $$ S(a,b)=\frac{(-1)^b}{b!}{}_2F_1(a,-b;a-2b+1;2)\binom{a-1}{2b-1} $$

In fact, from that generating function, $b!S(a,b)$ is the coefficient of $\left(2t+3t^2+4t^3+5t^4+...\right)^b$ at $t^a$. Here is the table of few of the $b!S(a,b)$: $$ \begin{array}{ccccccccccc} 1 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 2 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 3 & 4 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 4 & 12 & 8 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 5 & 25 & 36 & 16 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 6 & 44 & 102 & 96 & 32 & 0 & 0 & 0 & 0 & 0 \\ 0 & 7 & 70 & 231 & 344 & 240 & 64 & 0 & 0 & 0 & 0 \\ 0 & 8 & 104 & 456 & 952 & 1040 & 576 & 128 & 0 & 0 & 0 \\ 0 & 9 & 147 & 819 & 2241 & 3400 & 2928 & 1344 & 256 & 0 & 0 \\ 0 & 10 & 200 & 1372 & 4712 & 9290 & 11040 & 7840 & 3072 & 512 & 0 \\ 0 & 11 & 264 & 2178 & 9108 & 22363 & 34332 & 33488 & 20224 & 6912 & 1024 \\ \end{array} $$

Here is, in fact, a version that works for any $a$ and $b$, and does not contain any powers of $-1$ or $2$: represent $$ \frac{2-t}{(1-t)^2}=(1-t)^{-2}+(1-t)^{-1}; $$ then from the above we obtain that $b!S(a,b)$ is the coefficient at $t^a$ of the series $t^b((1-t)^{-2}+(1-t)^{-1})^b$, i. e. of $$ t^b\sum_{j=0}^b\binom bj(1-t)^{-2j}(1-t)^{-(b-j)}=t^b\sum_{j=0}^b\binom bj(1-t)^{-(b+j)} $$ It then follows easily that $$ S(a,b)=\frac1{b!}\sum_{j=0}^b\binom bj\binom{a+j-1}{a-b}=\frac1{b!}{}_2\!\!\ F_1(a,-b;b;-1)\binom{a-1}{a-b}. $$ Another observation that might be useful: $$ \cosh^{a-2}(t)\sinh((a+1)t)=\sum_{b=1}^ab!S(a,b)\sinh^{2b-1}(t) $$

$\endgroup$
8
  • $\begingroup$ Thanks for your efforts! Could elaborate a bit more please (combinatorics is not my field)? In particular, how did you get the formula for $S(a,b)$ after 'This follows from'? I would naively say that you set $\beta'-\gamma=:\ell$ but it does not fit, and I do not see where the $\min$ has gone (or negative values are taken to be 0?). Is it a well-known fact that the hyper-geometric function you've written is linked to the GF $e^{\frac{2-t}{(1-t)^2}tu}$? $\endgroup$
    – Nicolas
    Commented Dec 23, 2020 at 19:17
  • $\begingroup$ @Nicolas The key observation is a version of your second attempt:$$\frac{(2\beta')^{\overline{\alpha'-\beta'-\gamma}}}{(\alpha'-\beta'-\gamma)!}=\binom{\alpha'+\beta'-\gamma-1}{2\beta'-1}$$ $\endgroup$ Commented Dec 23, 2020 at 19:52
  • $\begingroup$ As for $\min$, note that $2\beta'\geqslant\alpha'$ if and only if $\alpha'-\beta'\leqslant\beta'$, so that your sum is actually to $\alpha'-\beta'$. $\endgroup$ Commented Dec 23, 2020 at 20:18
  • $\begingroup$ Many thanks for your developed answer! It helps to see a bot clearer. In particular, the sum is small in comparison to $\beta'!$. $\endgroup$
    – Nicolas
    Commented Dec 24, 2020 at 9:17
  • $\begingroup$ I have accepted your answer. I doubt one can say something more precise than that. Thank you for your efforts! $\endgroup$
    – Nicolas
    Commented Jan 3, 2021 at 10:03
1
$\begingroup$

For $n\ge k\ge0$, the Bell polynomials of the second kind $B_{n,k}$ satisfy \begin{equation}\label{Bell-2!-3!}\tag{+} B_{n,k}(2!,3!,\dotsc,(n-k+2)!) =\frac{n!}{k!}\sum_{\ell=0}^k(-1)^{k-\ell} \binom{k}{\ell}\binom{n+2\ell-1}{n}, \end{equation} where $\binom{q}{0}=1$ for all $q\in\mathbb{C}$.

The formula \eqref{Bell-2!-3!} was cited in Lemma 3.4 of the paper [1], reviewed in Section 1.8 (pp. 8--9) of the paper [2], and proved in Lemma 6 of the paper [3] below.

References

  1. F. Qi and B.-N. Guo, Explicit and recursive formulas, integral representations, and properties of the large Schroder numbers, Kragujevac J. Math. 41 (2017), no. 1, 121--141; available online at https://doi.org/10.5937/KgJMath1701121F.
  2. Feng Qi, Da-Wei Niu, Dongkyu Lim, and Yong-Hong Yao, Special values of the Bell polynomials of the second kind for some sequences and functions, Journal of Mathematical Analysis and Applications 491 (2020), no. 2, Paper No. 124382, 31 pages; available online at https://doi.org/10.1016/j.jmaa.2020.124382.
  3. F. Qi, X.-T. Shi, and B.-N. Guo, Two explicit formulas of the Schroder numbers, Integers 16 (2016), Paper No. A23, 15 pages.
$\endgroup$
1

You must log in to answer this question.

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