0
$\begingroup$

I have the following question:

Does there exist a non-negative function $g$ on $(0,1)$ such that $$1\leq F(x):=\dfrac{\displaystyle\sum_{k=0}^{\infty}a_{k}\,(k+1)^{2}\,x^{k}}{\displaystyle\sum_{k=0}^{\infty}(k+1)\,x^{k}}\leq 2,\;\forall\;x\in (0,1),$$ and $$\displaystyle\lim_{x\rightarrow 1^{-}}F(x)\;\text{ does not exist?}$$ Here $$a_{k}:=\int_{0}^{1}g(x)\,x^{2k+1}dx.$$

Note that if $g\equiv2$ then the first condition satisfies, but the second condition doesn't.

$\endgroup$

1 Answer 1

1
$\begingroup$

Such a function $g$ exists.

Indeed, we have \begin{equation*} \sum_{k=0}^\infty(k+1)\,x^k=\frac1{(1-x)^2} \end{equation*} and \begin{align*} \sum_{k=0}^\infty a_k(k+1)^2\,x^k&= \int_0^1 du\,g(u)\,\sum_{k=0}^\infty x^ku^{2k+1} \\ &= \int_0^1 du\,g(u)u\,\frac{1+xu^2}{(1-xu^2)^3} \\ &=\frac12\int_0^1 dv\,h(v)\,\frac{1+xv}{(1-xv)^3}, \end{align*} where $h(v):=g(\sqrt v)\ge0$ for $v\in(0,1)$. So, \begin{equation*} F_h(x):=F(x)=(1-x)^2S(x)/2, \tag{-1} \end{equation*} where \begin{equation*} S(x):=S_h(x):=\int_0^1 dv\,h(v)\,\frac{1+xv}{(1-xv)^3}. \tag{0} \end{equation*} So, for $h(x)=h_2(x):=2$, we have $F_h=1$ and hence the condition \begin{equation*} 1\le F(x)\le2 \tag{1} \end{equation*} for $x\in(0,1)$ holds.

Consider now $\tilde h:=h_2+ph=2+ph$, where $p\in(0,\infty)$ and $h$ is a measurable nonnegative function such that $F_h$ is bounded (on $(0,1)$) and the limit $F_h(1-)$ does not exist. If the $p$ is small enough, then condition (1) will hold for $F_{\tilde h}=1+pF_h$ in place of $F$ and the limit $F_{\tilde h}(1-)$ will not exist.

Thus, it suffices to construct a measurable nonnegative function $h$ on $(0,1)$ such that $F_h$ is bounded and the limit $F_h(1-)$ does not exist. Let us show that the formula \begin{equation*} h(v):=\sum_{n\ge1}h_n1(a_n<v<b_n) \end{equation*} defines such a function, where \begin{equation*} a_n:=1-q_n,\quad 0<b_n-a_n=o(q_n),\quad h_n:=\frac{q_n}{b_n-a_n},\quad q_n:=\frac1{2^{2^n}}, \tag{2} \end{equation*} so that $q_n^2=q_{n+1}$.

Take now any $v_0\in(0,1)$. Since $h$ is bounded on $(0,v_0)$, we have $(1-x)^2\int_0^{v_0} dv\,h(v)\,\frac{1+xv}{(1-xv)^3}\to0$ as $x\uparrow1$. So, by (-1) and (0), \begin{equation*} F_h(x)\sim G(x):=G_h(x):=(1-x)^2\int_0^1 dv\,h(v)\,\frac1{(1-xv)^3} \end{equation*} as $x\uparrow1$. Thus, it suffices to show that the function $G$ is bounded (on $(0,1)$) and the limit $G(1-)$ does not exist. Note that \begin{equation*} G(x)=(1-x)^2\sum_{n\ge1} h_n \int_{a_n}^{b_n} \frac{dv}{(1-xv)^3}. \end{equation*} For any natural $n_0$, the sum $\sum_{n_0\ge n\ge1} h_n \int_{a_n}^{b_n} \frac{dv}{(1-xv)^3}\le\sum_{n_0\ge n\ge1} h_n \int_{a_n}^{b_n} \frac{dv}{(1-v)^3}$ is bounded and hence $(1-x)^2\sum_{n_0\ge n\ge1} h_n \int_{a_n}^{b_n} \frac{dv}{(1-xv)^3}\to0$ as $x\uparrow1$. Also, in view of the first two conditions in (2), it is easy to see that $1-xv\sim1-xa_n$ as $n\to\infty$ uniformly in $x\in(0,1)$ and $v\in(a_n,b_n)$. So, letting \begin{equation*} q:=1-x, \end{equation*} we see that \begin{equation*} G(x)\sim H(q):=\sum_{n\ge1}\frac{q^2 q_n}{(q+q_n)^3} \end{equation*} as $x\uparrow1$ or, equivalently, $q\downarrow0$. Thus, it suffices to show that the function $H$ is bounded (on $(0,1)$) and the limit $H(0+)$ does not exist.

For natural $m\to\infty$ and $q\in[q_{m+1},q_m]$, write \begin{equation*} H(q)=H_{<m}+H_m+H_{m+1}+H_{>m+1}, \end{equation*} where \begin{equation*} H_{<m}:=\sum_{n<m}\frac{q^2 q_n}{(q+q_n)^3}\sim\sum_{n<m}\frac{q^2}{q_n^2} \sim\frac{q^2}{q_{m-1}^2}=\frac{q^2}{q_m}\le q_m\to0, \end{equation*} \begin{equation*} H_{>m+1}:=\sum_{n>m+1}\frac{q^2 q_n}{(q+q_n)^3}\sim\sum_{n>m+1}\frac{q_n}q \sim\frac{q_{m+2}}q\le\frac{q_{m+2}}{q_{m+1}}\to0, \end{equation*} \begin{equation*} H_m:=\frac{q^2 q_m}{(q+q_m)^3}\le\frac{q^2}{q_m^2}\le1, \end{equation*} \begin{equation*} H_{m+1}:=\frac{q^2 q_{m+1}}{(q+q_{m+1})^3}\le\frac{q_{m+1}}q\le1. \end{equation*}
So, $H$ is indeed bounded.

Finally, for $q=q_{m+1}$ \begin{equation*} H_m+H_{m+1}=\frac{q_{m+1}^2q_m}{(q_{m+1}+q_m)^3}+\frac18\to\frac18, \end{equation*} whereas for $q=2q_{m+1}\big[\in[q_{m+1},q_m]\big]$ \begin{equation*} H_m+H_{m+1}=\frac{4q_{m+1}^2q_m}{(2q_{m+1}+q_m)^3}+\frac4{27}\to\frac4{27}, \end{equation*} which yields $H(q_m)\to\frac18$ and $H(2q_m)\to\frac4{27}$. Thus, the limit $H(0+)$ does not exist.

$\endgroup$
1
  • $\begingroup$ @Pinelis Thks, I need time to digest your proof. $\endgroup$
    – Kakalot
    Aug 8, 2020 at 8:49

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.