A simplification of Laplace's method: applications to the gamma function and Gauss hypergeometric function (Q1040870)

If a function \(f\) has a minimum at \(t_0\in(a, b)\), we have the classical asymptotic expansion \[ F(x):= \int^b_a g(t)\exp(-xf(t))\,dt\sim \exp(- xf(t_0)) \sum^\infty_{n=0} c_{2n}{\Gamma(n+{1\over 2})\over x^{n-{1\over 2}}},\quad x\to+\infty, \] but the coefficients, arising from a change of variable, are in general hard to find. Instead, one may keep \(t\) and use expansions of the functions involved. After introducing \(m\) such that \(f^{(m)}(t_0)\) is the first non-vanishing derivative, and defining \[ f_m(t):= f(t)- f(t_0)- f^{(m)}(t_0)(t- t_0)^m/m!, \] the authors first derive a result of the form \[ F(x)= \exp(- xf(t_0)) \sum^\infty_{n=0} a_n(x)\Phi_n(x), \] where \(a_n(x)\) are coefficients in the expansion of \(\exp(-xf_m(t))g(t)\) around \(t_0\), and \(\Phi_n(x)\) is expressible in terms of incomplete Gamma functions. This is not yet an asymptotic expansion but by a suitable grouping of the terms they finally arrive at such a one, involving inverse powers of the asymptotic variable. The formula is too long to reproduce here. Two examples are considered: \(\Gamma(x)\) for large \(x\), and \(_2F_1(a,b;c;z)\) for large \(b\) and \(c\).