主定理证明

证明主方法（主定理）

注意：本文部分内容参考了知乎上数学答主的解答

主方法内容：

$$\begin{gathered} 设a\ge 1和b>1为常数，f\left(n\right)是一个函数，T(n)由下面的递推式定义 \\ T(n)=aT(\frac{n}{b})+f(n) \\ 其中，\frac{n}{b}指\lfloor \frac{n}{b}\rfloor 或\lceil \frac{n}{b}\rceil ，则T(n)有如下的渐进界： \\ 若对某常数ϵ>0，有f(n)=O(n^{{\log }_{b}a-ϵ})，则T(n)=\Theta (n^{{\log }_{b}a}) \\ 若f(n)=\Theta (n^{{\log }_{b}a})，T(n)=\Theta (n^{{\log }_{b}a}\log n) \\ 若对某常数ϵ＞0，有f(n)=\Omega (n^{{\log }_{b}a+\epsilon })，且对某个常数c<1和所有足够大的n， \\ 有af(\frac{n}{b})\le cf(n)，则T(n)=\Theta (f(n)) \end{gathered}$$

证明：由题有
$$T(n)=aT(\frac{n}{b})+f(n)$$

首先考虑当n为b的k次幂的情况

​ 原递归式改写为：
$$T(n)=\{\begin{array}{l}\Theta (1),n=1\\ \\ aT(\frac{n}{b})+f(n),n=b^i,i=1,2,\dots k\end{array}$$
​ 根据迭代法：
$$\begin{array}{rl}T(n)& =a^{2}T(\frac{n}{b})+f(n)+af(\frac{n}{b})\\ & =\dots \end{array}$$
​ 令 $\frac{n}{b^i}=1$(临界条件)，则有：
$$\begin{array}{rl}T(n)& =a^{i}T(\frac{n}{b^i})+\sum _{i=0}^{{\log }_{b}n-1}{a}_{i}f(\frac{n}{b^i})\\ & =n^{{\log }_{b}a}T(1)+\sum _{i=0}^{{\log }_{b}n-1}{a}_{i}f(\frac{n}{b^i})\\ & =n^{{\log }_{b}a}\Theta (1)+\sum _{i=0}^{{\log }_{b}n-1}{a}_{i}f(\frac{n}{b^i})\\ & =\Theta (n^{{\log }_{b}a})+\sum _{i=0}^{{\log }_{b}n-1}{a}_{i}f(\frac{n}{b^i})\end{array}$$
​ 故只需要考虑$\sum _{i=0}^{{\log }_{b}n-1}{a}_{i}f(\frac{n}{b^i})$的渐进表达式即可

下面从三种情况展开证明：

​ (1)当$f(n)=O(n^{{\log }_{b}a-ϵ})$时
$$\begin{gathered} f(\frac{n}{b^i})=O((\frac{n}{b^i}{)}^{{\log }_{b}a-ϵ}) \\ \begin{array}{rl}\therefore g(n)& =\sum _{i=0}^{{\log }_{b}n-1}{a}^{i}O\left({\left(\frac{n}{b^i}\right)}^{{\log }_{b}a-ϵ}\right)\\ & =O\left(\sum _{i=0}^{{\log }_{b}n-1}{a}^i{\left(\frac{n}{b^i}\right)}^{{\log }_{b}a-ϵ}\right)\\ & =O\left(n^{{\log }_{b}a-ϵ}\cdot \sum _{i=0}^{{\log }_{b}n-1}{\left(\frac{a}{b^{{\log }_{b}a-ϵ}}\right)}^i\right)\\ & =O\left(n^{{\log }_{b}a-ϵ}\cdot \sum _{i=0}^{{\log }_{b}n-1}{b}^{iϵ}\right)\\ & =O\left(n^{{\log }_{b}a-ϵ}\cdot \frac{n^{ϵ}-1}{b^{ϵ}-1}\right)\\ & =O\left(\frac{n^{{\log }_{b}a}-n^{{\log }_{b}a-ϵ}}{b^{ϵ}-1}\right)\\ & =O\left(n^{{\log }_{b}a}\right)\end{array} \end{gathered}$$
​ (2)当$f(n)=\Theta (n^{{\log }_{b}a}{\lg }^{k}n)$时
$$\begin{gathered} f(\frac{n}{b^i})=\Theta ((\frac{n}{b^i}{)}^{{\log }_{b}a}{\lg }^k(\frac{n}{b^i})) \\ \begin{array}{rl}\therefore g(n)& =\Theta \left(n^{{\log }_{b}a}\cdot \sum _{i=0}^{{\log }_{b}n-1}{\lg }^k(\frac{n}{b^i})\right)\\ & =\Theta \left(n^{{\log }_{b}a}\cdot \sum _{i=0}^{{\log }_{b}n-1}{\left(\lg n-i\lg b\right)}^k\right)\\ & =\Theta \left(n^{{\log }_{b}a}\cdot \sum _{i=0}^{{\log }_{b}n-1}\left({\lg }^{k}n+O({\lg }^{k}n)\right)\right)\\ & =\Theta \left(n^{{\log }_{b}a}\cdot \left({\log }_{b}n\cdot {\lg }^{k}n+{\log }_{b}n\cdot O({\lg }^{k}n)\right)\right)\\ & =\Theta \left(n^{{\log }_{b}a}\cdot \left({\log }_{b}n\cdot {\lg }^{k}n\right)\right)\\ & =\Theta \left(n^{{\log }_{b}a}\cdot {\lg }^{k+1}n\right)\end{array} \end{gathered}$$
​ (3)由题：
$$\begin{array}{rl}g(n)& =f(n)+\sum _{i=0}^{{\log }_{b}n-1}{a}_{i}f(\frac{n}{b^i})\\ & =\Omega (f(n))\end{array}$$
​ 由题：$\exists c<1,\exists N\in \mathrm{Z},\forall n>N$
$$s.t.af(\frac{n}{b})\le cf(n)$$
​ 递推有$\frac{n}{b^i}>N$​时
$$\begin{gathered} a^{i}f(\frac{n}{b^i})\le c^{i}f(n) \\ \begin{array}{rl}\therefore g(n)& =f(n)+\sum _{i=0}^{{\log }_{b}n-1}{a}^{i}f(\frac{n}{b^i})\\ & \le f(n)+\sum _{i=0}^{{\log }_{b}n-1}{c}^{i}f(n)\\ & =f(n)+\frac{1-c^{{\log }_{b}n}}{1-c}f(n)\\ & <\frac{2-c}{1-c}f(n)=O(f(n))\end{array} \end{gathered}$$
​ 故$g(n)=\Theta (f(n))$​

​ 由$T(n)=\Theta (n^{{\log }_{b}a})+\sum _{i=0}^{{\log }_{b}n-1}{a}^{i}f(\frac{n}{b^i})$

​ 根据上述分析可证得$n=b^k$​时主定理成立

下面推广n至全体正整数

​ 由$T(n)=aT(\lceil \frac{n}{b}\rceil )+f(n)\ge aT(\frac{n}{b})+f(n)$

​ 即 T ( n ) = O ( A ) , A ∈ { n log ⁡ b a , n log ⁡ b a lg ⁡ k + 1 n , f ( n ) ∣ 当 A 满足特定条件 } T(n)=O(A),A\in\{n^{\log_{b}{a}},n^{\log_{b}{a}}\lg^{k+1}{n},f(n)|当A满足特定条件\} T(n)=O(A),A∈{nlogb​a,nlogb​algk+1n,f(n)∣当A满足特定条件}

​ 同理可得$T(n)=aT(\lfloor \frac{n}{b}\rfloor )+f(n)$时，$T(n)=O(A)$​

下证$T(n)=aT(\lceil \frac{n}{b}\rceil )+f(n)$时，$T(n)=O(A)$​

​ 不妨令$t_0=n,t_1=\lceil \frac{t_0}{b}\rceil ,\dots t_k=\lceil \frac{t_{k-1}}{b}\rceil =\Theta (1)$
$$\begin{array}{rl}\because t_1& \le \frac{n}{b}+1[1]\\ t_2& \le \frac{t_1}{b}+1\le \frac{n}{b^2}+\frac{1}{b}+1\\ & \dots \\ t_k& \le \frac{t_{k-1}}{b}+1\le \frac{n}{b^k}+\sum _{i=0}^{k-1}\frac{1}{b^i}\\ \therefore t_k& \le \frac{n}{b^k}+\sum _{i=0}^{\infty }\frac{1}{b^i}\\ & =\frac{n}{b^k}\cdot \frac{b}{b-1}\\ 当k=\lfloor {\log }_{b}n\rfloor 时& \\ \frac{n}{b^k}& \le \frac{n}{b^{{\log }_{b}n-1}}\cdot \frac{b}{b-1}\\ & =\frac{b^2}{b-1}=\Theta (1)\\ \therefore k& \le \lfloor {\log }_{b}n\rfloor \\ \therefore T(n)& =aT(t_1)+f(t_0)\\ & =a^{2}T(t_2)+af(t_1)+f(t_0)\\ & \dots \\ & \le \Theta (n^{{\log }_{b}a})+\sum _{i=0}^{\lfloor {\log }_{b}n\rfloor -1}{a}^{i}f(t_i)\end{array}$$
​ 由前面证明的主定理（部分）有：$T(n)\le \Theta (A)$
$$\therefore T(n)=O(A)$$
​ 同理可得$T(n)=aT(\lfloor \frac{n}{b}\rfloor )+f(n)=\Omega (A)$

​ $Q.E.D$​