Skip to content

Asymptotics of Finite Sums

  • Bounding the tail. Make a rapid decreasing sum infinite

    \[ \begin{aligned} & N ! \sum_{0 \leq k \leq N} \frac{(-1)^k}{k !}=N ! e^{-1}-R_N \text { where } \quad R_N=N ! \sum_{k>N} \frac{(-1)^k}{k !} \\ &=\frac{N !}{e}+O\left(\frac{1}{N}\right), \quad\left|R_N\right|<\frac{1}{N+1}+\frac{1}{(N+1)^2}+\frac{1}{(N+1)^3}+\ldots=\frac{1}{N} \end{aligned} \]
  • Using the tail. The last term of a rapidly increasing sum may dominate.

    \[ \sum_{0 \leq k \leq N} k !=N !\left(1+\frac{1}{N}+\sum_{0 \leq k \leq N-2} \frac{k !}{N !}\right)=N !\left(1+O\left(\frac{1}{N}\right)\right) \]
  • Approx. with an integral

\[ \begin{aligned} & H_N=\sum_{1 \leq i \leq N} \frac{1}{k} \sim \int_1^N \frac{1}{x} d x=\ln N \\ & \ln N !=\sum_{1 \leq k \leq N} \ln k \sim \int_1^N \ln x d x=N \ln N-N+1 \end{aligned} \]

Euler-Maclaurin Summation

  • a classsic formula for estimating sums and integrals

Theorem. (Euler-Maclaurin summation). Let \(f\) be a function defined on \([1, \infty)\) whose derivatives exist and are absolutely integrable. Then $$ \sum_{1 \leq k \leq N} f(k)=\int_1^N f(x) d x+\frac{1}{2} f(N)+C_f+\frac{1}{12} f^{\prime}(N)-\frac{1}{720} f^{\prime \prime \prime}(N)+\ldots $$

Example

\[ H_N=\ln N+\gamma+\frac{1}{2 N}-\frac{1}{12 N^2}+O\left(\frac{1}{N^4}\right) \]
\[ \ln N!=N \ln N-N+\ln \sqrt{2 \pi N}+\frac{1}{12 N}+O\left(\frac{1}{N^3}\right) \]

Exercise: Given Stirling's approximation \(\ln N !=N \ln N-N+\ln \sqrt{2 \pi N}+O\left(\frac{1}{N}\right)\), Develop an asymptotic approximation for \(\binom{2N}N\) to \(O(1 / N)\) (relative error)

\[ \begin{aligned} \binom{2N}N= & \exp (\ln (2 N !)-2 \ln N !) \\ = & \exp (2 N \ln (2 N)-2 N+\ln \sqrt{4 \pi N}+O(1 / N) \\ & \quad-2(N \ln (N)-N+\ln \sqrt{2 \pi N}+O(1 / N))) \\ = & \exp (2 N \ln 2-\ln \sqrt{\pi N}+O(1 / N)) \\ = & \frac{4^N}{\sqrt{\pi N}}\left(1+O\left(\frac{1}{N}\right)\right) \end{aligned} \]

Application: Asymptotics of Catalan numbers

Q: How many bits to represent a binary tree with \(N\) internal nodes?

  • use the approx above.