请教一个似然函数求导的问题

snowblink

如图，这个组合数里含参数的求导不太会处理，想请教一下各位
设 $X_1, \cdots, X_n$ 为来自负二项总体的简单随机样本，即
\[
P\left(X_i=k\right)=\binom{\alpha+k-1}{k}\left(\frac{\beta}{1+\beta}\right)^a\left(\frac{1}{1+\beta}\right)^k, i=1, \cdots, n ; k=0,1,2, \cdots
\]
其中 $\alpha>0, \beta>0$ 为分布参数．记 $n_k$ 为样本中取值为 $k$ 的个数，$\sum_{k=0}^{\infty} n_k=n \cdot \bar{X}=\frac{1}{n} \sum_{i=1}^n X_i$．证明 $\alpha$ 的极大似然估计满足 $\sum_{k=1}^{\infty} n_k \sum_{j=1}^k \frac{1}{\alpha+j-1}=n \ln \left(1+\frac{\bar{X}}{\alpha}\right)$．

战巡

\[P(X=k)=C_{\alpha+k-1}^k\left(\frac{\beta}{1+\beta}\right)^\alpha\left(\frac{1}{1+\beta}\right)^k\]
\[L(\boldsymbol{X}|\alpha,\beta)=\prod_{i=1}^nC_{\alpha+X_i-1}^{X_i}\left(\frac{\beta}{1+\beta}\right)^\alpha\left(\frac{1}{1+\beta}\right)^{X_i}=\left(\frac{\beta}{1+\beta}\right)^{n\alpha}\left(\frac{1}{1+\beta}\right)^{n\bar{X}}\prod_{i=1}^nC_{\alpha+X_i-1}^{X_i}\]
\[=\left(\frac{\beta}{1+\beta}\right)^{n\alpha}\left(\frac{1}{1+\beta}\right)^{n\bar{X}}\prod_{k=0}^{\infty}(C_{\alpha+k-1}^{k})^{n_k}\]

\[l(\boldsymbol{X}|\alpha,\beta)=\ln(L(\boldsymbol{X}|\alpha,\beta))=n\alpha\ln(\frac{\beta}{1+\beta})+n\bar{X}\ln(\frac{1}{1+\beta})+\sum_{k=0}^\infty n_k\ln(C_{\alpha+k-1}^k)\]


这里面
\[\ln(C_{\alpha+k-1}^k)=\ln(\frac{(\alpha+k-1)(\alpha+k-2)...(\alpha+1)\alpha}{k!})=\sum_{j=1}^k\ln(\alpha+j-1)-\ln(k!)\]
\[\frac{d}{d\alpha}\ln(C_{\alpha+k-1}^k)=\frac{d}{d\alpha}[\sum_{j=1}^k\ln(\alpha+j-1)-\ln(k!)]=\sum_{j=1}^k\frac{1}{\alpha+j-1}\]

故此
\[
\begin{cases}\frac{d}{d\alpha}l(\boldsymbol{X}|\alpha,\beta)=n\ln(\frac{\beta}{1+\beta})+\sum_{k=0}^\infty n_k\sum_{j=1}^k\frac{1}{\alpha+j-1}=0\\\frac{d}{d\beta}l(\boldsymbol{X}|\alpha,\beta)=(\alpha-\beta\bar{X})\cdot\frac{n}{\beta(1+\beta)}=0\end{cases}\]

然后
\[\hat{\beta}=\frac{\hat{\alpha}}{\bar{X}}\]
\[n\ln(\frac{\hat{\alpha}}{\hat{\alpha}+\bar{X}})+\sum_{k=0}^\infty n_k\sum_{j=1}^k\frac{1}{\hat{\alpha}+j-1}=0\]
下略