3
$\begingroup$

For $p\in (0,1)$ and $\epsilon>0$ a small enough constant, consider the function $f:\mathbb{N}\to\mathbb{R}$ given by

$$f(y)=\int_0^\epsilon \frac{\left(\frac{p-t}{p(1-t)}\right)^y-(1-t)^{y(1-p)/p}}{t\log(1-t)}dt.$$

Is it true that $f(y)=O(1)$ for all $p\in(0,1)$? Numerical experiments seem to support this assertion.

A possible first approach would be to compute the Taylor expansion of the numerator around $t=0$. We have $$ \left(\frac{p-t}{p(1-t)}\right)^y-(1-t)^{y(1-p)/p}=O_y(t^2),$$ where $O_y$ hides terms which depend on $y$. If these terms were independent of $y$ then the result would follow easily, but this is not the case.

$\endgroup$
2
  • $\begingroup$ Is $p$ fixed? That is, are the constants in O allowed to depend on $p$? $\endgroup$ May 4, 2018 at 15:12
  • $\begingroup$ Yes, the constants are allowed to depend on $p$. $\endgroup$ May 4, 2018 at 15:32

2 Answers 2

2
$\begingroup$

$\newcommand{\de}{\delta} \newcommand{\De}{\Delta} \newcommand{\ep}{\epsilon} \newcommand{\ga}{\gamma} \newcommand{\Ga}{\Gamma} \newcommand{\la}{\lambda} \newcommand{\Si}{\Sigma} \renewcommand{\th}{\theta} \newcommand{\R}{\mathbb{R}} \newcommand{\F}{\mathcal{F}} \newcommand{\E}{\operatorname{\mathsf E}} \newcommand{\PP}{\operatorname{\mathsf P}} \newcommand{\ii}[1]{\operatorname{\mathsf I}\{#1\}}$

Let \begin{equation} h_1(t):=\ln\frac{p-t}{p(1-t)},\quad h_2(t):=\frac{1-p}p\,\ln(1-t),\quad \de:=h_2-h_1. \end{equation} Then $\de(0)=0$ and \begin{equation} \de'(t)=\frac{(1-p) t}{p (1-t) (p-t)}>0, \end{equation} so that $\de>0$ and hence $h_1<h_2$. Also, $\de'(t)\sim\frac{(1-p) t}{p^2}$ and hence $\de(t)\sim \frac{(1-p) t^2}{2p^2}$ as $t\to0$. Also, $h_2(t)<-\frac{1-p}p\,t$. So, \begin{multline*} 0< f(y)=\int_0^\ep \frac{e^{h_2(t)y}-e^{h_1(t)y}}{-t\ln(1-t)}dt <\int_0^\ep \frac{\de(t)}{-t\ln(1-t)}\,y\, e^{h_2(t)y}\,dt \\ \ll\int_0^\ep y e^{h_2(t)y}\,dt <\int_0^\infty \exp\{-\tfrac{1-p}p\,ty\}y\,dt=\frac p{1-p}=O(1), \end{multline*} as desired.

$\endgroup$
0
2
$\begingroup$

Just compute.Fix $p$, for $t$ small enough. Let $a=\frac{p-t}{p(1-t)}=1+(1-1/p)t+(1-1/p)t^2+(1-1/p)t^3+O(t^4)$, $b=(1-t)^{\frac{p}{1-p}}=1+(1-1/p)t+(1-3/2p+1/2p^2)t^2+(1-1/p+11/6p^2-1/6p^3)t^3+O(t^4)$,

$b-a=\frac{1-p}{2p^2}t^2+O(t^3)>0$.

Now, let $h(y)=b^y-a^y=(-1)\times $numerator of your integral $>0$, view $y$ as in $\mathbb{R}_{>0}$, this function has a upper bound which is independent of $y$, taking value at $ln(\frac{lnb/lna}{ln(b/a)})$, take this upper bound, substitution and compute the $t$-expension, then you may get the result, which is not hard to you.

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.