Is it possible to "invert" an exponential generating function?

201 Views Asked by At

Consider the exponential generating function for a sequence $\mathbf{a}$, given by:

$$\text{EG}_\mathbf{a}(x) = \sum_{n=0}^\infty a_n \frac{x^n}{n!}.$$

In order to find the sequence $\mathbf{a}$ that corresponds with certain specified forms for the generating function, it would be useful to be able to "invert" this expression to a form like this:

$$\frac{x^n}{n!} = \sum_{k \in \mathbb{Z}} b_{n,k} e^{kx} \quad \quad \quad \text{for all }n=0,1,2,....$$

In this latter form we have an infinite matrix $\mathbf{b} = [b_{n,k} ]$ with indices $n =0,1,2,...$ and $k \in \mathbb{Z}$. In this expression, a weighted sum of exponentials gives a single term in the exponential generating function. Taking $c_k \equiv \sum_{n=0}^\infty a_n b_{n,k}$ we can write the generating function $\text{EG}_\mathbf{a}(x) = \sum_{k \in \mathbb{Z}} c_k e^{kx}$, so this form allows us to obtain an alternative expression for the generating function using a weighted sum of exponentials at regular intervals.


My question: Is there any $\mathbf{b}$ that solves with set of equations? If so, how do we find it? (If there is a solution in only some cases, what cases are those?)


My working so far: Suppose we can find a matrix $\mathbf{b}$ which satisfies:

$$\sum_{k \in \mathbb{Z}}^\infty b_{n,k} k^r = \mathbb{I}(r=n) \quad \quad \quad \text{for all } r,n =0,1,2,....$$

Then we have:

$$\begin{equation} \begin{aligned} \sum_{k \in \mathbb{Z}}^\infty b_{n,k} e^{kx} &= \sum_{k \in \mathbb{Z}}^\infty b_{n,k} \sum_{r=0}^\infty \frac{(kx)^r}{r!} \\[6pt] &= \sum_{r=0}^\infty \frac{x^r}{r!} \sum_{k \in \mathbb{Z}}^\infty b_{n,k} k^r \\[6pt] &= \sum_{r=0}^\infty \frac{x^r}{r!} \cdot \mathbb{I}(r=n) \\[6pt] &= \frac{x^n}{n!}, \\[6pt] \end{aligned} \end{equation}$$

which gives the desired form. This means that the desired form can be obtained by solving a series of equations involving sums of powers. I'm not sure if this constitutes "progress" since this set of equations seems just as difficult to solve as the initial problem.

1

There are 1 best solutions below

0
On

This is not an answer.

Suppose that we have a limited summation $$y=\sum_{i=0}^n a_i \frac {x^i}{i!}$$ that we could consider as a truncated Taylor series $$y=\sum_{i=0}^n a_i \frac {x^i}{i!}+O(x^{n+1})$$ Using series reversion, we could write $$x=\sum_{i=1}^n b_i t^i+O(t^{n+1})\qquad \text{where}\qquad t=\frac{y-a_0}{a_1}$$ and then use the binomial expansion to $$\frac{x^p}{p!}=\frac{1}{p!}\left(\sum_{i=1}^n b_i t^i+O(t^{n+1}) \right)^p$$ in which, later, we should to re-expand the powers of $t$ in terms of $y$ to finally obtain $$\frac{x^p}{p!}=\sum_{i=0}^n c_i y^i+O(y^{n+1})$$

Trying to do it with $y=e^x$ and $n=6$, we should obtain $$\frac {x^1}{1!}=-\frac{49}{20}+6 y-\frac{15 y^2}{2}+\frac{20 y^3}{3}-\frac{15 y^4}{4}+\frac{6 y^5}{5}-\frac{y^6}{6}+O\left(y^7\right)$$ $$\frac {x^2}{2!}=\frac{469}{180}-\frac{223 y}{20}+\frac{879 y^2}{40}-\frac{949 y^3}{36}+\frac{41 y^4}{2}-\frac{201 y^5}{20}+\frac{1019 y^6}{360}+O\left(y^7\right)$$ $$\frac {x^3}{3!}=-\frac{267}{160}+\frac{349 y}{36}-\frac{18353 y^2}{720}+\frac{797 y^3}{20}-\frac{1457 y^4}{36}+\frac{4891 y^5}{180}-\frac{187 y^6}{16}+O\left(y^7\right)$$ $$\frac {x^4}{4!}=\frac{95}{128}-\frac{7667 y}{1440}+\frac{24901 y^2}{1440}-\frac{4013 y^3}{120}+\frac{122249 y^4}{2880}-\frac{5273 y^5}{144}+\frac{10279 y^6}{480}+O\left(y^7\right)$$ $$\frac {x^5}{5!}=-\frac{8591}{34560}+\frac{6041 y}{2880}-\frac{92771 y^2}{11520}+\frac{13349 y^3}{720}-\frac{163313 y^4}{5760}+\frac{43319 y^5}{1440}-\frac{129067 y^6}{5760}+O\left(y^7\right)$$

May be, this could give you some ideas (I hope and wish).