Skip to main content
3 of 5
I have simplified the proof of the inequality $II_1\le I_2$.
Iosif Pinelis
  • 127.9k
  • 8
  • 107
  • 229

Let $B(a,b;x):=\int_0^x t^{a-1}(1-t)^{b-1}dt$ denote, as usual, the value of the incomplete Beta function. We have to show that $$ n\binom{n}{k}B(k+1,n-k+1,\tfrac k{n+1}) \le 1/2 \quad\text{for $k=0,\dots,n$}. \tag{1} $$
As suggested by Brendan McKay, it should be easy to use asymptotic analysis to check that $(1)$ holds for any real $c>1/2$ if $n$ is large enough. Thus, we can use the reverse induction in $n$. That is, it is enough to show that, if $(1)$ holds for some natural $n+1\ge2$ in place of $n$ and some real $c>0$ in place of $1/2$, then it holds as written, but with the same real $c>0$ in place of $1/2$. The key here is the simple identity \begin{equation} B(a,b;x)=B(a+1,b;x)+B(a,b+1;x), \end{equation} which yields, by the reverse induction, $$ B(k+1,n-k+1,\tfrac k{n+1})=B(k+2,n-k+1,\tfrac k{n+1})+B(k+1,n-k+2,\tfrac k{n+1}) $$ $$ =B(k+2,n-k+1,\tfrac{k+1}{n+2})+B(k+1,n-k+2,\tfrac k{n+2})+I_1-I_2 $$ $$ \le\frac c{n+1}/\binom{n+1}{k+1} +\frac c{n+1}/\binom{n+1}{k}+I_1-I_2 =\frac{c(n+2)n}{(n+1)^2}\frac1{n\binom nk}+I_1-I_2 <\frac c{n\binom nk}+I_1-I_2, $$ where \begin{equation} I_1:=\int_{t_0}^{t_1}f(t)\,\frac{dt}t,\quad I_2:=\int_{t_1}^{t_2}f(t)\,\frac{dt}{1-t}, \end{equation} \begin{equation} f(t):=t^{k+1}(1-t)^{n-k+1},\quad t_0:=\tfrac k{n+2}, \quad t_1:=\tfrac k{n+1}, \quad t_2:=\tfrac{k+1}{n+2}, \end{equation} so that $0\le t_0<t_1<t_2<1$ and the function $f$ is increasing on the interval $(t_0,t_2)$.

It remains to check the inequality $I_1\le I_2$. This is easy: since $f$ is increasing, \begin{equation} I_1-I_2\le \int_{t_0}^{t_1}f(t_1)\,\frac{dt}t-\int_{t_1}^{t_2}f(t_1)\,\frac{dt}{1-t}=0. \end{equation} $\qed$

Iosif Pinelis
  • 127.9k
  • 8
  • 107
  • 229