4

I worked on the summation for natural $a$ $$ \Omega_a(n)=\sum_{k=0}^n (-1)^{n-k} \binom{n}{k}k^{n+a}$$ and since $$ k^{n+a}=\lim_{x\to 0} \frac{d^{n+a}}{dx^{n+a}}e^{k x}$$ So I got $$ \Omega_a(n)=\lim_{x\to 0} \frac{d^{n+a}}{dx^{n+a}} \left(e^x-1\right)^n$$ and to find nth derivative I get firstly $$ \Omega_a(n)=\lim_{x\to 0} \frac{d^{n+a-1}}{dx^{n+a-1}} ne^x \left(e^x-1\right)^{n-1}$$ then by Leibniz rule $$ \Omega_a(n)=n\sum_{k=0}^{n+a-1} \binom{n+a-1}{k} \lim_{x\to 0} \frac{d^k}{dx^k}\left(e^x-1\right)^{n-1}$$ and for natural $n,k$ we have $$\lim_{x\to 0} \frac{d^k}{dx^k}\left(e^x-1\right)^{n-1}=0 , k<n-1$$ then $$ \Omega_a(n)=n\sum_{k=n-1}^{n+a-1} \binom{n+a-1}{k} \lim_{x\to 0} \frac{d^k}{dx^k}\left(e^x-1\right)^{n-1}$$ Or $$ \Omega_a(n)=n\sum_{k=0}^{a} \binom{n+a-1}{k+n-1} \lim_{x\to 0} \frac{d^{k+n-1}}{dx^{k+n-1}}\left(e^x-1\right)^{n-1}$$ So I found a Recurrence Relation $$ \Omega_a(n)=n\sum_{k=0}^{a} \binom{n+a-1}{k+n-1}\Omega_k(n-1) , \Omega_n(0)=0$$ from this relation I got $$ \Omega_0(n)=n! \\ \Omega_1(n)=\frac{n}{2} (n+1)! \\ \Omega_2(n)=\frac{n(3n+1)}{24} (n+2)! \\ \Omega_3(n)=\frac{n^2(n+1)}{48} (n+3)! \\ \Omega_4(n)=\frac{n(15n^3+30n^2+5n-2)}{5760} (n+4)! \\ \Omega_5(n)=\frac{n^2(3n^3+10n^2+5n-2)}{11520} (n+5)!$$ So we see that $$\Omega_a(n)=P_{a}(n) (n+a)!$$ where $P_{a}(n)$ it's polynomial with $a$ degree and its coefficients are relate to A053657 and A100655

which makes sure that it's relate to Bernoulli polynomials.

So my Question is how to find the polynomial $P_{a}(n)$ ?

EDITION

from the answers I got that

$$ \Omega_a(n)=n! S_{n+a}^{(n)}$$

but I still stuck on HOW to find $P_{a}(n)$

also here there are a similar polynomial but different signs .

Faoler
  • 2,568
  • 6
  • 18

2 Answers2

4

You can use a “Function from exponent” from here:

$$\frac{d^nf(e^z)}{dz^n}=\sum_{k=0}^ne^{kz}s_n^{(k)}f^{(k)}(e^z)$$

which is also a pattern that can be found by making a table for $n$ and noticing the Stirling numbers of the second kind $s_n^{(k)}$. This gives:

$$\Omega_a(n)=\left.\frac{d^{n+a}}{dx^{n+a}} \left(e^x-1\right)^n\right|_0=\sum_{k=0}^{n+a}s_{n+a}^{(k)}\left.\frac{d^k(x-1)^n}{dx^k}\right|_1$$

but only the $k=n$ term survives since the other derivatives at $x=1$ give $0$: $$\sum_{k=0}^{n+a}s_{n+a}^{(k)}\left.\frac{d^k(x-1)^n}{dx^k}\right|_1=s_{n+a}^{(n)}\frac{d^n(x-1)^n}{dx^n}$$

Therefore:

$$\boxed{\Omega_a(n)= \sum_{k=0}^n (-1)^{n-k} \binom{n}{k}k^{n+a} =n!s_{n+a}^{(n)}}$$

For example, the $a=4$ case is shown here. The formula can also be found here.

Тyma Gaidash
  • 13,576
  • You can also see this, purely combinatorially, by realizing the left side counts, by inclusion-exclusion, the number of onto functions from a set of $n+a$ elements to a set of $n$ elements. Such functions anre equivalent to partitioning the $n+a$-set and sending each part to a distinct element of the $n$-set, so you get $n!s_{n+a}^{(a)}.$ – Thomas Andrews Aug 20 '24 at 15:49
  • Sorry, should have been "partitioning the $n+a$-set into $n$ parts, and sending..." – Thomas Andrews Aug 20 '24 at 16:36
  • nice! thank you – Faoler Aug 20 '24 at 17:39
  • its nice idea but it need more work to find the polynomial I showed – Faoler Aug 20 '24 at 17:55
3

We are interested in a closed form of

$$\Omega_a(n) = \sum_{k=0}^n (-1)^{n-k} {n\choose k} k^{n+a}$$

for natural $a.$ More precisely we seek $P_a(n)$ where $$P_a(n) = \frac{\Omega_a(n)}{(n+a)!},$$ which is a polynomial. Keeping it simple we find

$$(n+a)! [z^{n+a}] \sum_{k=0}^n (-1)^{n-k} {n\choose k} \exp(kz) \\ = (n+a)! [z^{n+a}] (\exp(z)-1)^n = n! {n+a\brace n}.$$

With $a$ our constant we have combinatorially that we are counting the number of partitions of $[n+a]$ into an ordered sequence of $n$ sets. We can think of this as first choosing $n$ values to go into our $n$ slots for a factor of $n! {n+a\choose n}.$ Then we partition the $a$ remaining values into a sequence of $q$ sets where $1\le q\le a$ and attach them at ${n\choose q}$ points to the bottom layer of singletons. However we cannot use $\exp(z)-1$ to select the sets that form the partition of $a$ because we overcount by a factor of $m$ on $z^m$. This is because if we select a set containing $m$ elements the attachment point has $m$ possibilities. Hence we must divide by $m.$ Integrate to get $\exp(z)-1-z$ and divide by $z$ for an EGF of $(\exp(z)-1)/z-1.$ Putting it all together,

$$n! {n+a\choose n} \sum_{q=1}^a {n\choose q} a! [z^a] ((\exp(z)-1)/z-1)^q \\ = (n+a)! \sum_{q=1}^a {n\choose q} [z^a] \frac{1}{z^q} (\exp(z)-1-z)^q.$$

We have our factorial in front which is canceled. Here we already see that we have a polynomial in $n$ because the factors on the binomial coefficients do not depend on $n$ and the binomial coefficients yield polynomials in $n$ upon expansion. Continuing,

$$\sum_{q=1}^a {n\choose q} [z^{a+q}] \sum_{p=0}^{q} {q\choose p} (-1)^{q-p} z^{q-p} (\exp(z)-1)^p \\ = \sum_{q=1}^a {n\choose q} \sum_{p=0}^{q} {q\choose p} p! (-1)^{q-p} [z^{a+p}] \frac{(\exp(z)-1)^p}{p!} \\ = \sum_{q=1}^a {n\choose q} \sum_{p=0}^{q} {q\choose p} p! (-1)^{q-p} \frac{1}{(a+p)!} {a+p\brace p}.$$

The conclusion is that

$$\bbox[5px,border:2px solid #00A000]{ P_a(n) = \frac{1}{a!} \sum_{q=1}^a {n\choose q} \sum_{p=0}^{q} {q\choose p} (-1)^{q-p} {a+p\choose p}^{-1} {a+p\brace p}.}$$

We can compute an explicit form of the coefficients of $P_a(n)$ by using

$$[n^r] {n\choose q} = (-1)^{q+r} \frac{1}{q!} {q\brack r}$$

which then yields

$$\bbox[5px,border:2px solid #00A000]{ P_a(n) = \frac{1}{a!} \sum_{r=1}^a n^r (-1)^r \sum_{q=1}^a \frac{1}{q!} {q\brack r} \sum_{p=0}^{q} {q\choose p} (-1)^p {a+p\choose p}^{-1} {a+p\brace p}.}$$

The first sum in $r$ iterates over the powers of $n$ up to $n^a$, forming a polynomial in $n$.

Remark. Once we have seen the closed form from the combinatorial argument it becomes readily apparent that we can also get it by simple algebraic manipulation, as suggested in the comment, which goes as follows. We have

$$(n+a)! [z^{n+a}] (\exp(z)-1)^n = (n+a)! [z^{n+a}] \sum_{q=0}^n {n\choose q} (\exp(z)-z-1)^q z^{n-q} \\ = (n+a)! \sum_{q=0}^n {n\choose q} [z^{a+q}] (\exp(z)-z-1)^q.$$

Note however that we may set the upper range to $a.$ This is because when $a\gt n$ the binomial coefficient is zero in the added range and we are not adding anything and may raise to $a$. On the other hand the sum term starts at $z^{2q}$ so we get zero when $2q\gt q+a$ or $q\gt a.$ Hence if $a\lt n$ we may lower to $a$ because we are not losing anything. We find

$$(n+a)! \sum_{q=1}^a {n\choose q} [z^{a+q}] (\exp(z)-z-1)^q$$

and may then continue with the computation as before. (The term for $q=0$ does not go past the coefficient extractor.)

Marko Riedel
  • 64,728