广义二项系数 / 指数函数
一、定义
定义广义二项系数(Generalized binomial series)为:
\[\mathcal{B}_t(z) = \sum\limits_{n \geq 0}\dbinom{tn+1}{n}\dfrac{z^n}{tn+1} \]或等价写为:
\[\mathcal{B}_t(z) = \sum\limits_{n \geq 0}(tn)^{\underline{n - 1}}\dfrac{z^n}{n!} \]结论 1:
\[\mathcal B_t(z) = z\mathcal B_t(z)^t+1 \]结论 2:
\[\mathcal B_t(z)^r = \sum\limits_{n \geq 0}\dbinom{tn+r}{n}\dfrac{r}{tn+r}z^n \]结论 3:
\[\dfrac{\mathcal B_t(z)^r}{1 - t+t\mathcal B_t(z)^{-1} } = \sum\limits_{n \geq 0}\dbinom{tn+r}nz^n \]代数证明
利用拉格朗日反演逆证明,由于 \(\mathcal B_t(z)\) 有常数项,设 \(F(z) = \mathcal B_t(z) - 1\),则
\[F(z) = z(1+F(z))^t \]设 \(G(z) = \dfrac{z}{(1+z)^t}\),套用普通拉格朗日反演:
\[[z^n]F(z) = [x^{n - 1}]\dfrac 1 n(1+x)^{nt} \\ =\dbinom{nt}{n - 1} \times \dfrac 1n = \dbinom{nt+1}{n}\dfrac{1}{nt+1} \]显然 \([x^0]\mathcal B_t(z)\) 也满足条件。这就证明了结论 1。
对于结论 2,设 \(H(z) = (1+z)^r\),根据扩展拉格朗日反演 \([x^n]H(F(x)) = [x^{n - 1}]\dfrac 1 n H'(\dfrac x{G(x)})^n\) 得:
\[[z^n]\mathcal B_t(z)^r = \dfrac r n(1+z)^{r - 1}(1+z)^{nt} \\ =\dfrac r n \dbinom{nt+r - 1}{n - 1} = \dfrac{r}{nt+r}\dbinom{nt+r}{n} \]显然 \([x^0]\mathcal B_t(z)^r\) 满足条件。这就证明了结论 2。
对于结论 3,设 \(H(x) = \dfrac{(1+z)^r}{1 - t+t(1+z)^{-1}}\)。
根据扩展另类拉格朗日反演:
\[[z^n]\dfrac{\mathcal B_t(z)^r}{1 - t+t\mathcal B_t(z)^{-1}} = [z^n]G'(\dfrac{z}{G(z)})^{n+1}H(z)=[z^n]\dfrac{(1+z)^t - t(1+z)^{t - 1}z}{(1+z)^{2t}}\dfrac{(1+z)^r}{1 - t+t(1+z)^{-1}}(1+z)^{t(n+1)} \]稍作化简,得到:
\[[z^n](1+z)^{tn+r} \]若使用普通扩展拉格朗日反演(长篇警告):
\[[z^n]\dfrac{\mathcal B_t(z)^r}{1 - t+t\mathcal B_t(z)^{-1}} = [z^{n - 1}] \dfrac 1 n \dfrac{r(1+z)^{r - 1}(1-t+t(1+z)^{-1}) +(1+z)^rt(1+z)^{-2}}{(1 - t+t(1+z)^{-1})^2}(1+z)^{nt} \\ =[z^{n - 1}](1+z)^{nt+r}\dfrac 1 n \dfrac{(1+z)^{-1}r(1-t+t(1+z)^{-1})+(1+z)^{-2}t}{(1-t+t(1+z)^{-1})^2} \\ =[z^{n - 1}](1+z)^{nt+r}\dfrac 1 n(\dfrac{r}{1-(t -1)z}+\dfrac{t}{(1 - (t - 1)z)^2}) \\ =\dfrac 1 n\sum\limits_{i = 0}^{n - 1}\dbinom{nt+r}i(r(t - 1)^{n - 1 - i}t(n - i)(t - 1)^{n - 1- i}) \\ =\dfrac 1 n ((nt+r)\sum\limits_{i = 0}^{n - 1}\dbinom {nt+r}i(t - 1)^{n - 1 - i} - t\sum\limits_{i = 0}^{n - 1}\dbinom{nt+r}i i(t - 1)^{n - 1- i}) \\ =\dfrac 1 n ((nt+r)\sum\limits_{i = 0}^{n - 1}\dbinom{nt+r}{i}(t - 1)^{n - 1 - i} -t(nt+r)\sum\limits_{i = 0}^{n - 1}\dbinom{nt+r - 1}{i- 1}(t - 1)^{n -1 - i} ) \]考虑 \(c_k = \sum\limits_{i = 0}^{k}\dbinom{nt+r}i(t - 1)^{k - i} - t\sum\limits_{i = 0}^k\dbinom {nt+r - 1}{i - 1}(t - 1)^{k - i}\),有:
\[c_k = \sum\limits_{i = 0}^{k}\dbinom{nt+r} i(t - 1)^{k - i}- t\sum\limits_{i = 0}^{k}\dbinom{nt+r - 1}{i - 1}(t - 1)^{k - i - 1} \\ =\dbinom{nt+r}k+\sum\limits_{i = 0}^{k - 1}(t - 1)^{k - i - 1}((t - 1)\dbinom{nt+r}i - t\dbinom{nt+r - 1}{i - 1}) \\ =\dbinom{nt+r}k+\sum\limits_{i = 0}^{k - 1}(t - 1)^{k - i - 1}(t\dbinom{nt+r - 1}{i - 1} - \dbinom{nt+r}i) \\ \dbinom{nt+r}k-(\sum\limits_{i = 0}^{k - 1}\dbinom {nt+r}i(t - 1)^{k - i - 1} -t\sum\limits_{i = 0}^{k - 1}\dbinom{nt+r - 1}{i - 1}) = \dbinom{nt+r}k - c_{k - 1} \]因此,\(c_k = \sum\limits_{i = 0}^{k}\dbinom{nt+r}{i}(-1)^{k - i}\),根据翻转求和得到 \(c_k = \dbinom{nt+r - 1}{k}\),因此 \(c_{n - 1} = \dbinom{nt+r - 1}{n - 1}\)。
回带,$ = \dfrac{1}n(nt+r)\dbinom{nt+r - 1}{n - 1} = \dbinom{nt+r}{n}$
这就证明了结论 3。