2

The binomial transform is the shift operator for the Bell numbers. That is, $$ \sum_{j=0}^k {k\choose j} B_j =B_{k+1} $$ where the $B_n$ are the Bell numbers.

Is there a somewhat similar expressions involving incomplete Bell polynomials: $$ \sum_{j=1}^k(-1)^{j-1}\binom{k}{j} {\hat B}_{m,j}(x_1,x_2,...,x_{m-j+1})=\;? $$

Maybe the definition of incomplete Bell polynomials helps to find an answer: $$ {\hat B}_{m,j}(x_1,x_2,...,x_{m-j+1})=\sum_{{k_0+k_1+\cdots+k_N=j}\atop{k_1+2k_2+\cdots+Nk_N=m}}\binom{j}{k_0,k_1,\ldots,k_N} \prod_{i=1}^Nx_i^{k_i} $$

RobPratt
  • 45,619
draks ...
  • 18,449
  • 1
    The incomplete Bell polynomials $B_{n,k}(x_1,x_2,\ldots,x_{n-k+1})$ become Stirling numbers of the second kind $S(n,k)$ if we put all $x_i=1$. Therefore the Bell numbers come from the complete Bell polynomials by putting all $x_i=1$. The shift formula for Bell numbers is thus more directly related to the formula $$\sum_{i=0}^{n}\binom{n}{i}B_{n-i}(x_1,x_2,\ldots,x_{n-i})x_{i+1}=B_{n+1}(x_1,x_2,\ldots,x_{n+1})$$ for complete Bell polynomials, see the wiki link for all the details. – N. Shales Apr 24 '17 at 13:29

2 Answers2

2

Hint: We could look at the situation with the help of exponential generating functions.

The following holds: If $$A(z)=\sum_{k=0}^\infty a_k \frac{z^k}{k!}$$ is an exponential generating function of the sequence $(a_k)$, then \begin{align*} A(z)e^z=\sum_{k=0}^\infty \left(\sum_{j=0}^k\binom{k}{j} a_j \right)\frac{z^k}{k!} \end{align*} and applying the differential operator $D_z=\frac{d}{dz}$ we obtain \begin{align*} D_z A(z)=\sum_{k=0}^\infty a_{k+1}\frac{z^k}{k!} \end{align*}

Since the exponential generating function of the Bell numbers is \begin{align*} B(z)=\exp\left({e^z-1}\right)=\sum_{n=0}^\infty B_n\frac{z^n}{n!} \end{align*} the relation \begin{align*} \ \qquad\qquad\sum_{j=0}^k\binom{k}{j}B_j=B_{k+1}\qquad\qquad k\geq 0 \end{align*} translates into \begin{align*} B(z)e^z&=D_z B(z)\\ \end{align*}

We could now try to find an exponential generating function $P(z)$ for the partial ordinary Bell polynomials. The expression \begin{align*} \sum_{j=0}^k(-1)^{j-1}\binom{k}{j}\hat{B}_{m,j}(x_1,x_2,\ldots,m-j+1) \end{align*} looks like the coefficient of \begin{align*} (-1)^{k-1}P(z)e^{-z} \end{align*} We could now check if this series could also be represented using operators like $D_z, e^z$, etc. Here we use the notation $\hat{B}_{m,j}$ in accordance with the Wiki page.

[Add-on 2017-04-30] Some information regarding ordinary generating functions of partial ordinary Bell polynomials. We use the coefficient of operator $[t^n]$ to denote the coefficient of $t^n$ of a series $A(t)$.

According to the referred Wiki page are the partial ordinary Bell polynomials \begin{align*} \hat{B}_{m,j}(x_1,x_2,\ldots,x_{m-j+1}) \end{align*}

the coefficients of a bivariate generating function $\hat{\Phi}(t,u)$ with

\begin{align*} \hat{\Phi}(t,u)&=\exp\left(u\sum_{l=1}^\infty x_lt^l\right)\\ &=\sum_{k=0}^\infty\left(\sum_{l=1}^\infty x_lt^l\right)^k\frac{u^k}{k!}\\ &=\sum_{k=0}^\infty\left(\sum_{n=k}^\infty\hat{B}_{n,k}\left(x_1,x_2,\ldots,x_{n-k+1}\right)t^n\right)\frac{u^k}{k!} \end{align*}

We can write \begin{align*} \hat{B}_{m,j}(x_1,x_2,\ldots,x_{m-j+1})&=[t^mu^j]\hat{\Phi}(t,u)\\ &=[t^m]\underbrace{\left(\sum_{l=1}^\infty x_lt^l\right)^j}_{:= \hat{B}_j(t)}\tag{1} \end{align*}

We recall the Euler transformation formula of a series \begin{align*} A(t)=\sum_{n= 0}^\infty a_nt^n\qquad\qquad \frac{1}{1-t}A\left(\frac{-t}{1-t}\right)=\sum_{n= 0}^\infty \left(\sum_{j=0}^m\binom{m}{j}(-1)^ja_j\right)t^n\tag{2} \end{align*} This transformation formula together with a proof can be found e.g. in Harmonic Number Identities Via Euler's transform by K.N. Boyadzhiev. See remark 2 with $\lambda=1, \mu=-1$.

We conclude according to (1) and (2) \begin{align*} \sum_{j=0}^m(-1)^{j}\binom{m}{j}\hat{B}_{m,j}(x_1,x_2,\ldots,x_{m-j+1}) &=[t^mu^j]\frac{1}{1-t}\hat{\Phi}\left(\frac{-t}{1-t},u\right)\\ &=[t^m]\frac{1}{1-t}\hat{B}_j\left(\frac{-t}{1-t}\right)\tag{3} \end{align*}

Note: Regrettably there seems to be no simple representation of the coefficients in (3) based on the ordinary generating function. An exponential generating function \begin{align*} \sum_{n=k}^\infty\hat{B}_{n,k}\left(x_1,x_2,\ldots,x_{n-k+1}\right)\frac{t^n}{n!} \end{align*} of the partial ordinary Bell polynomials is not stated at the Wiki-page about Bell polynomials and useful represenations are not known to me.

Markus Scheuer
  • 108,315
  • ok, I think I start to see what you mean. I also checked the wiki page again and found $\Phi (t,u)=\exp \left(u\sum {r=1}^{\infty }x{r}{\frac {t^{r}}{r!}}\right)=\sum {n,k\geq 0}B{n,k}(x_{1},\ldots ,x_{n-k+1}){\frac {t^{n}}{n!}}u^{k}$. Putting your suggestions this becomes $\Phi (t,u)e^{-t}=\sum {n,k\geq 0}\left(\sum{j=0}^k(-1)^{j-1}\binom{k}{j}B_{n,j}(x_1,x_2,\ldots)\right)u^{k}{\frac {t^{n}}{n!}}$, where I'm not sure if I've chosen the correct variable $t$ or $u$... – draks ... Apr 26 '17 at 13:46
  • You mean using this: $\frac{1}{1-z}A\left(\frac{z}{1-z}\right)=\sum_{n= 0}^\infty \left(\sum_{j=0}^n\binom{n}{j}a_j\right)z^n$? But where do the $x_k$'s come from then? And $m$ looks like a free parameter so it should be $A_m(z)$ then right? – draks ... Apr 27 '17 at 10:28
  • Thanks a lot so far. Cheers, – draks ... Apr 27 '17 at 14:35
  • My comment on the summation was crap. I deleted it. You also deleted one. May I ask why? – draks ... Apr 29 '17 at 18:59
  • @draks...: I sometimes remove comments which are outdated and do not have useful information any longer. Did you make some progress? – Markus Scheuer Apr 29 '17 at 19:45
  • I have to compile the hints, I found on here and there. It doesn't look like a result right away, but I'm optimistic... – draks ... Apr 30 '17 at 07:20
  • @draks...: I've added some info to clarify aspects about ordinary and exponential generating functions, regrettably without substantial information regarding a nice formula. – Markus Scheuer Apr 30 '17 at 21:23
  • Be careful to note the difference between the Euler transformation of an o.g.f. and an e.g.f. – Tom Copeland May 01 '17 at 17:20
  • @TomCopeland: I typically apply Euler's transformation formula to an ogf as indicated in (2). Anything wrong with my answer? – Markus Scheuer May 01 '17 at 17:51
  • 1
    @MarkusScheuer, nope. Just made the comment so others would not be confused and become aware of a difference. Both forms are rather important. – Tom Copeland May 01 '17 at 19:48
  • @TomCopeland: I agree, both forms were more than once pretty useful in my calculations. Many thanks for checking my answer. – Markus Scheuer May 01 '17 at 19:56
  • Hi Markus, you assume all $x_{m-j+2}=0$ in your general definition of ${\hat B}j(t)$, right? Further $j=0$ doesn't really make sense; does it follow that ${\hat B}{m,0}=0$? – draks ... May 02 '17 at 15:13
  • Hi again, your ${\hat B}j(t)$ reminded me on power symmetric functions, which are mentioned at Wiki:Bell polynomial. This is insofar remarkable because, the appearing Bell polynomials are written as $B{n,k}(e_{1},2!e_{2},3!e_{3},\ldots ,(n-k+1)!e_{n-k+1})$. Here $e_k$ are elementary symmetric functions, which are the scaled coefficients of the polynomial where we once started here... – draks ... May 03 '17 at 06:05
  • @draks...: Hi! With respect to the comment before: Yes, $\hat{B}_m,0=0$ whenever the index $m$ is not in the appropriate range. With respect to the activities up to now: Despite the connection with the elementary symmetric functions, I don't expect any nice formula similarly to the identity with Bell numbers. Regards, – Markus Scheuer May 03 '17 at 13:11
  • Hi, Tom Copeland just remarked that the expression the derivative of the log is related to Faber polynomials (see here), so $\frac{d^m}{dx^m}\log(1+\sum_{k=1}^N n_kx^k)\Biggr|_{x=0}=F_m(n_1,n_2,...n_k)$. Further I found $p_k(x_1,\ldots,x_n) = F_k\bigl(e_1(x_1,\ldots,x_n),\ldots,e_k(x_1,\ldots,x_n)\bigr)$ here. So if $x_k$ are the roots of the polynomial inside the log (i.e. eigenvalues of the adjacency matrix), we end the power symmetric functions of them. Wouldn't this be a wonderful result? – draks ... May 17 '17 at 06:11
  • @draks...: Very nice! Thanks for the info. This is an interesting connection which I wasn't aware of and is worth investigating. – Markus Scheuer May 17 '17 at 09:54
  • @MarkScheuer could you please carify, does $[t^m]f(t)$ mean "coefficient of $t^m$ in $f(t)$", e.g. c.f. equation (1) ? Thanks. – pshmath0 Jun 21 '20 at 09:42
  • @Pixel: Yes, $[t^n]$ denotes the coefficient of operator. I've added a reference which might be interesting. – Markus Scheuer Jun 21 '20 at 14:40
1

Let \begin{equation*}%\label{Fall-Factorial-Dfn-Eq} \langle\alpha\rangle_n= \prod_{k=0}^{n-1}(\alpha-k)= \begin{cases} \alpha(\alpha-1)\dotsm(\alpha-n+1), & n\ge1\\ 1,& n=0 \end{cases} \end{equation*} and \begin{equation*} (\alpha)_n=\prod_{\ell=0}^{n-1}(\alpha+\ell) = \begin{cases} \alpha(\alpha+1)\dotsm(\alpha+n-1), & n\ge1\\ 1, & n=0 \end{cases} \end{equation*} denote the falling and rising factorials of $\alpha\in\mathbb{C}$ respectively.

In Theorem 2.1 of the paper [1] below, the formula \begin{equation}\label{Bell-fall-Eq-inv}\tag{3} \sum_{\ell=0}^{k}\frac{B_{n,\ell}(\langle\alpha\rangle_1, \langle\alpha\rangle_2, \dotsc,\langle\alpha\rangle_{n-\ell+1})}{(k-\ell)!} =\frac{\langle\alpha k\rangle_n}{k!} \end{equation} for $n\ge k\ge0$ and $\alpha\in\mathbb{R}$ was established.

As a consequence of \eqref{Bell-fall-Eq-inv}, the formula \begin{equation}\label{Bell-rise-Eq-inv}\tag{5} \sum_{\ell=0}^{k}\frac{B_{n,\ell}((\alpha)_1, (\alpha)_2, \dotsc, (\alpha)_{n-\ell+1})}{(k-\ell)!} =\frac{(\alpha k)_n}{k!} \end{equation} for $\alpha\in\mathbb{C}$ and $n\ge k\ge0$ was derived in Corollary 2.1 of the paper [1] below.

References

  1. Feng Qi, Da-Wei Niu, Dongkyu Lim, and Bai-Ni Guo, Closed formulas and identities for the Bell polynomials and falling factorials, Contributions to Discrete Mathematics 15 (2020), no. 1, 163--174; available online at https://doi.org/10.11575/cdm.v15i1.68111.
  2. https://mathoverflow.net/a/405003/147732
  3. https://math.stackexchange.com/a/4262657/945479
  4. https://math.stackexchange.com/a/4261764/945479
qifeng618
  • 1,691