1: \begin{abstract}

2: 

3: Many of the fastest known algorithms to compute $\pi$ involve generalized hypergeometric series, such as the Ramanujan-Sato series. In this paper, we investigate the rates of convergence for several such series and we give asymptotic expansions for the error of finite approximation.

4: 

5: For example, when using the first $n$ terms of the Chudnovskys' series, we obtain the finite approximation $\pi_n\approx \pi$. It is known that the truncation error satisfies $|\pi_n-\pi|\approx 53360^{-3n}.$ In this paper, we prove that the asymptotic expansion for the truncation error in the Chudnovskys' series is

6: $$\left|\pi_n-\pi\right|=53360^{-3n}\cdot\frac{\numprint{106720}\sqrt{\numprint{10005}\pi}}{\numprint{1672209}\sqrt{n}}\cdot\exp\left(\frac{A_1}{n}+\frac{A_2}{n^2}+\frac{\delta_n}{n^3}\right),$$

7: with $\numprint{0.006907}<\delta_n<\numprint{0.008429}$ and the exact rational values of $A_1$ and $A_2$:

8: \begin{align*}

9:  A_1&= -\nfrac{1781843197433}{7456754505816},\\

10:  A_2&=  -\nfrac{1080096011925710088395}{3475199235000451148614116}.

11: \end{align*}

12: Thus we demonstrate how to establish precise error bounds for the approximations for $\pi$ obtained through Ramanujan-like series for $1/\pi$.

13: 

14: We also give asymptotic expansions for all known rational hypergeometric series for $1/\pi$ in the appendix.

15: \bigskip

16: 

17: \noindent\textbf{Keywords:} Chudnovsky Algorithm $\cdot$ error bounds $\cdot$ inequalities $\cdot$ hypergeometric series~$\cdot$ approximations $\cdot$ asymptotic expansions $\cdot$ Stirling's formula

18: \bigskip

19: 

20: \noindent\textbf{Mathematics Subject Classification (2010):} 26D15 $\cdot$ 33C20 $\cdot$ 33B15

21: 

22: \end{abstract}

23: