1
$\begingroup$

I am not sure if this question is too specific on notations (I think the question is intuitive, but basically the only reference I know with this kind of notations is Bingham, Goldie & Teugels book Regular Variation), so let me introduce some things.

Firstly, let $f$ be a measurable positive real function defined in $[a, +\infty)$ for some real $a \geqslant 0$.

- MATUSZEWSKA INDICES:

Say that $f$ is almost increasing when there is $M\geqslant a$ and $m>0$ such that $$ f(y) \geqslant mf(x),\quad \forall y\geqslant x\geqslant M. $$

The definition of almost decreasing is similar. Now define the upper ($\alpha$) and lower ($\beta$) Matuszewska indices of $f$ as

$$ \alpha(f) := \inf\{\alpha \in \mathbb{R} : x^{-\alpha} f(x) \text{ is almost decreasing}\},$$ $$ \beta(f) := \sup\{\beta \in \mathbb{R} : x^{-\beta} f(x) \text{ is almost increasing}\}.$$

These indices basically bound $f$ between two growth orders "neglecting" some slowly varying functions, i.e. $$ \frac{x^{\beta(f)}}{L_1(x)} \ll f(x) \ll x^{\alpha(f)} L_2(x),$$ for some non-decreasing $L_1(x), L_2(x) = o(x^{\varepsilon})$ for all $\varepsilon > 0$.

- $O$-REGULAR VARIATION:

$f$ is said to be $O$-regularly varying when $$ \forall \lambda > 0, f(\lambda x) = \Theta(f(x)),$$ that is, for all $\lambda >0$ there is $M=M(\lambda)>0$ such that: $$ \forall x > M, \exists c_1=c_1(\lambda),c_2 = c_2(\lambda)~: ~c_1 f(x) \leqslant f(\lambda x) \leqslant c_2f(x).$$

This is (un?)surprisingly equivalent to say that $f$ has both Matuszewska indices finite (in $(-\infty, +\infty)$).

My question is:

Q: Let $f$ be a measurable, positive, increasing real function in $[0,\infty)$. If $f$ is $O$-regularly varying and $f(x) \gg x^{1/n}$ for some integer $n \geqslant 1$, then the lower Matuszewska index $\beta(f)$ is $> 1/2n$?

To me it is a somewhat intuitive statement (at a first glance I would even say that $\beta(f) \geqslant 1/n$), but I could not prove it or find it explicitly stated on Bingham's et al. book and other references (Feller's Probability books and some papers). I was looking for some help to develop a better intuition on this statement (being it true or false!). Any tips and/or references would be welcome.

Thanks!


Disclaimer: What I tried: http://i.stack.imgur.com/qGooL.png (These calculations didn't help...)

(Here $A\equiv f$ and $h\equiv n$. Sorry, too lazy to fix!)

$\endgroup$
1
+50
$\begingroup$

The answer is no. Take any real $a>0$. Let, e.g., $y_k:=e^{k^3}$ and $x_k:=y_k/e^k=e^{k^3-k}$, so that $0<x_1<y_1<x_2<y_2<\dots$. Here in this answer $k$ is any natural number, unless further specified. For $k\ge2$, let
\begin{equation} a_k:=a\,\frac{\ln(y_k/y_{k-1})}{\ln(x_k/y_{k-1})}=a\,\frac{k^3-(k-1)^3}{k^3-k-(k-1)^3}>a,\quad c_k:=y_{k-1}^a/y_{k-1}^{a_k}, \end{equation} so that \begin{equation} c_ky_{k-1}^{a_k}=y_{k-1}^a,\quad c_kx_k^{a_k}=y_k^a, \end{equation} $c_k>0$, and $a_k\to a$; any convergence here is as $k\to\infty$. For each $k\ge2$, let next \begin{equation} f(x):=c_kx^{a_k}\quad\text{if}\quad y_{k-1}\le x\le x_k,\quad f(x):=y_k^a\quad\text{if}\quad x_k\le x\le y_k. \end{equation}

This defines a positive nondecreasing function $f\colon[e,\infty)\to\mathbb R$ such that $f(x)\ge x^a$ for all $x\ge e$. Moreover, $f$ is $O$-regular, since it is a piecewise-power function with alternating exponents $a_k$ and $0$, where $a_k\to a$, so that the positive nondecreasing function $f$ grows no faster than the power function with exponent $2a$ (say). Here are the graphs $\{(\ln x,\ln f(x))\colon y_1\le x\le y_4\}$ and $\{(\ln x,\ln(x^a)\colon y_1\le x\le y_4\}$ for $a=1$:

enter image description here

Yet, the lower Matuszewska index of $f$ is $0$ (rather than $a$ or $a/2$), because for any real $b>0$ one has $y_k^{-b}f(y_k)=o(x_k^{-b}f(x_k))$.

$\endgroup$
  • $\begingroup$ Wow, this is a beautiful counterexample! It even rules out some additional hypotheses I've been thinking (like $x^{a-\varepsilon} \ll_{\varepsilon} f(x) \ll_{\varepsilon} x^{a+\varepsilon}$, i.e $\forall \varepsilon >0$). It also reminds me of an answer to a question I asked on M.SE last year, but the fact that $a_k\to a$ makes it even better to this particular case. I'll surely add this to my pocket counterexamples :) Thank you very much! $\endgroup$ – Alufat Mar 23 '16 at 2:13
  • 1
    $\begingroup$ I am glad this was useful. $\endgroup$ – Iosif Pinelis Mar 23 '16 at 2:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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