On asymptotic and strict monotonicity of a sharper lower bound for Student's \(t\) percentiles (Q607607)

Consider a standard normal random variable \(Z\) and another random variable \(Y\) having Student's \(t\)-distribution with \(\nu\) degrees of freedom, \(\nu=1,2,\cdots\). Denote the customary upper \(100\alpha\)\% points of this two distributions by \(z_{\alpha}\) and \(t_{\nu,\alpha}\), respectively, that is, \[ 1-\alpha ={\mathbf P}(Z\leq z_\alpha)={\mathbf P}(Y\leq t_{\nu,\alpha}),\;\;0<a<1/2. \] It is well-known that for every fixed \(0<\alpha<1/2\) and degree of freedom \(\nu\), one has \(t_{\nu,\alpha}>z_\alpha\) and that \(t_{\nu,\alpha}\) monotonically decreases to \(z_\alpha\) as \(\nu\) increases. Recently, \textit{N. Mukhopadhyay} [Methodol. Comput. Appl. Probab. 12, No. 4, 609--622 (2010; Zbl 1204.60021)] found a new and explicit expression \(b_\nu\;(>1)\): \[ b_\nu=\sqrt{2^{-1}\nu}\Gamma(2^{-1}\nu) \{\Gamma(2^{-1}(\nu+1))\}^{-1}, \] such that \(t_{\nu,\alpha}> b_\nu z_\alpha\) for every \(0<\alpha<2^{-1}\), \(\nu\). Denote \(r_n:=b_{\nu+1}/b_\nu\). The main result of the paper can be stated as follows: \textbf{Prop. 1.1.} We have \[ \log t_\nu =4^{-1}\nu^{-2}+O(\nu^{-3})\;\;as\;\;\nu\to \infty; \] and also \[ \log r_1\sim \log(2\sqrt{2}/\pi)<0\;\;\text{and}\;\;\log r_\nu\leq -4^{-1}\nu^{-1}+(3/2)\nu^{-2}<0\;\;\text{for}\;\;\nu\geq 2. \] That is, the sequence \(\{b_\nu;\;\nu\geq 1\}\) is strictly decreasing in \(\nu\).