I am currently reading some Quantum mechanics book which has used Stirling's approximation for complex variables. Could anyone proof me how can Stirling formula be generalized to the complex variable? It has mentioned that in the vicinity of saddle point it can be shown. Any answer is highly appreciated.
1 Answers
In equation $(6)$ of this answer, it is shown that, for non-negative $n\in\mathbb{Z}$, $$ \begin{align} \log(n!) &=\int_0^{n+1/2}\log(x)\,\mathrm{d}x+\tfrac12\log(2\pi)+\sum_{k=n+1}^\infty \int_0^{1/2}\log\left(1-\frac{t^2}{k^2}\right)\,\mathrm{d}t\\ &=\int_0^{n+1/2}\log(x)\,\mathrm{d}x+\tfrac12\log(2\pi)+\sum_{k=1}^\infty \int_0^{1/2}\log\left(1-\frac{t^2}{(n+k)^2}\right)\,\mathrm{d}t\tag{1} \end{align}$$ All we need to show that $(1)$ extends to $\log(\Gamma(n+1))$ on the reals is to show that the right hand side is convex. Take the second derivative with respect to $n$: $$ \begin{align} &\frac1{n+\frac12}+\sum_{k=n+1}^\infty\int_0^{1/2}\frac{-6k^2t^2+2t^4}{\left(k^3-kt^2\right)^2}\,\mathrm{d}t\\ &\ge\frac1{n+\frac12}+\sum_{k=n+1}^\infty\int_0^{1/2}\frac{-6t^2}{\left(k^2-\frac14\right)^2}\,\mathrm{d}t\\ &=\frac1{n+\frac12}-\sum_{k=n+1}^\infty\frac1{4\left(k^2-\frac14\right)^2}\\ &=\frac1{n+\frac12}-\sum_{k=n+1}^\infty\left[\frac1{2k+1}-\frac1{2k-1}+\frac1{(2k+1)^2}+\frac1{(2k-1)^2}\right]\\ &=\frac3{2n+1}-\frac1{(2n+1)^2}-\sum_{k=n+1}^\infty\frac2{(2k+1)^2}\\ &\ge\frac3{2n+1}-\frac1{(2n+1)^2}-\sum_{k=n+1}^\infty\frac2{(2k+1)(2k-1)}\\ &=\frac3{2n+1}-\frac1{(2n+1)^2}-\sum_{k=n+1}^\infty\left[\frac1{2k-1}-\frac1{2k+1}\right]\\ &=\frac2{2n+1}-\frac1{(2n+1)^2}\\[3pt] &=\frac{4n+1}{(2n+1)^2}\tag{2} \end{align} $$ Inequality $(2)$ shows that $(1)$ defines $\log(\Gamma(n+1))$ for $n\in\mathbb{R}$. Since $(1)$ defines an analytic function, $(1)$ defines $\log(\Gamma(n+1))$ for $n\in\mathbb{C}$.
The rest of the estimates in the cited answer are valid for $\operatorname{Re}(n)\gt0$, so we get for $\operatorname{Re}(z)\gt0$, $$ \Gamma(z+1)=\sqrt{2\pi z}\,\frac{z^z}{e^z}\left(1+O\!\left(\frac1{|z|}\right)\right)\tag{3} $$

- 345,667
-
Exactly why does showing the convexity of $\log(\Gamma(n+1))$ suffices to show it extends to the reals? – Dqrksun May 26 '23 at 10:54
-
1The Bohr-Mollerup Theorem says that $\Gamma$ is the unique function on the positive reals that is log-convex and satisfies $\Gamma(x+1)=x\Gamma(x)$ with $\Gamma(1)=1$. – robjohn May 26 '23 at 15:40