Trying to understand a proof relating Stirling number of the second kind to its exponential generating function

165 Views Asked by At

In Cameron's Combinatorics, he set out to proof the below identity:

$$\sum_{n\ge0}\frac{S(n,k)t^n}{n!}=\frac{(exp(t)-1)^k}{k!}$$

He proceeds as follows:

$$\begin{align*}\sum_{n\ge0}\frac{S(n,k)t^n}{n!}&=\sum_{n\ge0}\frac{1}{k!}\sum_{j=1}^k{k\choose j}(-1)^{k-j}j^n\frac{t^n}{n!}\\ &=\frac{1}{k!}\sum_{j=1}^k{k\choose j}(-1)^{k-j}\sum_{n\ge0}\frac{{(jt)}^n}{n!}\\ &=\frac{1}{k!}\sum_{j=1}^k{k\choose j}(-1)^{k-j}exp(jt)\\ &=\frac{(exp(t)-1)^k}{k!}\end{align*}$$

I had no trouble following author's proof until the second-to-last line. I believed the author used the binomial theorem to prove the equivalence of that expression to the last one. However, the starting index of the summation in the binomial theorem is $0$ instead of $1$: $$(1+t)^n=\sum_{k=0}^n{n \choose k}t^k$$

That is, if I tried to convert the numerator of the aforementioned expression using binomial theorem: $$\begin{align*}(exp(t)-1)^k&=(-1)^k(1-exp(t))^k\\ &=(-1)^k(1+(-exp(t)))^k\\ &=(-1)^k\sum_{j=0}^k{{n\choose j}(-1)^{-j}(exp(t))^j}\\ &=\sum_{j=0}^k{k\choose j}(-1)^{k-j}exp(jt) \end{align*}$$

I think I made a mistake somewhere but I don't know where. I am completed stuck here in trying to understand the proof. Any help is greatly appreciated. Thank you in advance.

2

There are 2 best solutions below

0
On BEST ANSWER

You did not make any mistakes, but the author did. The author's mistake was when they wrote $$ \sum_{n\ge0}\frac{S(n,k)t^n}{n!} \stackrel{?}=\sum_{n\ge0}\frac{1}{k!}\sum_{\color{red}{j=1}}^k{k\choose j}(-1)^{k-j}j^n\frac{t^n}{n!}\tag1 $$ They are, presumably, basing this off of the formula $$ S(n,k)\stackrel?=\frac1{k!}\sum_{\color{red}{j=1}}^k \binom kj(-1)^{k-j}j^n,\tag2 $$ which is incorrect. The lower index for the summation here is supposed to be $j=0$. That is, the correct formula is $$ S(n,k)=\frac1{k!}\sum_{\color{green}{j=0}}^k \binom kj(-1)^{k-j}j^n\tag3 $$ When you make that replacement, everything works out just fine. Note that $(2)$ and $(3)$ are almost exactly equal to each other; most of the time, $j^n=0$ when $j=0$, so it is fine to omit the $j=0$ term. However, since $0^0=1$ (in the context of combinatorics), you cannot omit this term when $n=0$. This makes a difference in $(1)$, because we are summing over all $n\ge 0$.

0
On

Use the identity $\sum_{j=1}^k C(k,j) e^{jt} =(1+ e^{t})^k -1$.

The last term $-1$ accounts for the missing index value $j=0$.