{n+1m+1}=∑k(nk){km}(6.15)\begin{Bmatrix}n+1 \\m+1\end{Bmatrix}=\sum_k \binom{n}{k}\begin{Bmatrix}k \\m\end{Bmatrix} \tag{6.15}{n+1m+1}=k∑(kn){km}(6.15)
We enumerate the size of the subset which n+1n+1n+1 belongs to, which equals n−k+1n-k+1n−k+1. We first choose n−kn-kn−k elements that is in the same subset with n+1n+1n+1, which is (nk)\dbinom{n}{k}(kn), then allocate the rest kkk elements into mmm non-empty subsets.
[n+1m+1]=∑k[nk](km)(6.16)\begin{bmatrix}n+1 \\m+1\end{bmatrix}=\sum_k \begin{bmatrix}n \\k\end{bmatrix}\binom{k}{m} \tag{6.16}[n+1m+1]=k∑[nk](mk)(6.16)
This looks almost the same as the previous one, but it’s much harder to consider the combinatorial explanation.
What the formula do is obvious. First divide the first nnn elements into kkk cycles, then choose mmm cycles from it, and combine the rest k−mk-mk−m cycles with n+1n+1n+1 to form a new cycle. But the question is how to proof this approach is not heavy and not leaky.
Consider that we first divide the elements into m+1m+1m+1 subsets A1,A2,⋯ ,Am+1A_1,A_2,\cdots,A_{m+1}A1,A2,⋯,Am+1. We assume that n+1∈A1n+1\in A_1n+1∈A1. Say C(Ai)C(A_i)C(Ai) is a cycle formed with elements in AiA_iAi. Take any specific C(A2),C(A3),⋯ ,C(Am+1)C(A_2),C(A_3),\cdots,C(A_{m+1})C(A2),C(A3),⋯,C(Am+1),there are (∣A1∣−1)!(|A_1|-1)!(∣A1∣−1)! ways to choose C(A1)C(A_1)C(A1). In our formula, with C(A2),C(A3),⋯ ,C(Am+1)C(A_2),C(A_3),\cdots,C(A_{m+1})C(A2),C(A3),⋯,C(Am+1) fixed(after we times kkk choose mmm), the subset A1A_1A1 is constitute of some of the cycles of its elements, which sum up to exactly ∑k[∣A1∣−1k]\sum_k\begin{bmatrix}|A_1|-1 \\k\end{bmatrix}∑k[∣A1∣−1k] times, which equals (∣A1∣−1)!(|A_1|-1)!(∣A1∣−1)!.