strange binomial coefficient transformation, Equation 41 at MacKay Information theory book

74 Views Asked by At

The top equation is very clear, but how is the first approximation done? The author mentions this line as an alternative to Stirling's approximation. It should be a straightforward equation, but sorry I could not get it

Thank you in advance!

$$ \begin{split} 1 &= \sum_K \binom{N}{K}2^{-N} \\ &\approx 2^{-N} \binom{N}{N/2} \sum_{r=-N/2}^{N/2} e^{-(r/\sigma)^2/2}\\ &\approx 2^{-N} \binom{N}{N/2} \sigma \sqrt{2\pi}. \end{split} $$

1

There are 1 best solutions below

1
On

The last approximation comes because for large $N$, $$ \sum_{r=-N/2}^{N/2} e^{-(r/\sigma)^2/2} \approx \int_\mathbb{R} e^{-(r/\sigma)^2/2} dr = \sigma \int_\mathbb{R} e^{-u^2/2} du = \sigma\sqrt{2\pi} $$ where $u = r/\sigma$ and the last step is a famous Gaussian integral, which can be taken, e.g., by switching to polar coordinates...