2
$\begingroup$

I am trying to find smooth functions $f : \mathbb{R}_+ \to \mathbb{R}_+$ such that the quantity $$\Delta_f(x) := 2f(x)-f(2x)$$ is positive for $x$ large enough and has the greatest asymptotic growth.


It seems clear that one cannot go beyond a linear growth, since $\Delta_f$ vanishes for linear functions and will likely be negative for super-linear ones. However, one can construct many examples of almost linear asymptotic growths.

For example, plugging $f(x) := x^a$ for some $a\in(0,1)$ yields $\Delta_f(x) = (2-2^a) x^a$.

Choosing $f(x) := \frac{x}{\ln x}$ yields $\Delta_f(x) \sim 2 \ln 2 \frac{x}{(\ln x)^2}$ which has a larger asymptotic growth.


What would be your candidates for even larger asymptotic growths? And is there a way to prove a (sublinear) a priori upper bound on $\Delta_f$?

$\endgroup$

1 Answer 1

3
$\begingroup$

Let us discretise the problem by setting $a_n=2^{-n}f(2^n)$, $b_n=2^{-n-1}\Delta_f(2^n)$. Then your relation becomes, $$b_n=a_n-a_{n+1}.$$ since $a_n,b_n$ are non-negative, we conclude that $$\sum_{n=1}^\infty b_n<\infty.$$ This is a necessary and sufficient condition. Indeed, take any summable sequence $b_n$ of positive numbers, then we can define $$a_n=\sum_{k=n}^\infty b_k$$ as the sequence of partial sums, and obtain your equation on the sequence $x_n=2^n$. Then you can interpolate by choosing $f(x)$ arbitrarily on the interval $(1,2)$.

Returning to your original notation, the growth condition becomes $$\int\frac{\Delta_f(x)}{x^2}dx<\infty.$$ This is a necessary and sufficient condition. For example, we cannot have $\Delta_f(x)\sim x/\log x$, but can have $$\Delta_f(x)\sim\frac{x}{\log x(\log\log x)^{1+\epsilon}},$$ and so on.

$\endgroup$
3
  • $\begingroup$ Thank you for this nice viewpoint on the question. With this approach, I understand that finding the largest asymptotic growth for $\Delta_f$ is reduced to finding the largest asymptotic behavior for a summable sequence $b_n$. When $b_n$ is decreasing, I think it is known that $b_n = o(1/n)$, so your approach would suggest that one cannot obtain better than $\Delta_f(x) \sim \frac{x}{\ln x}$? $\endgroup$
    – cs89
    Oct 23, 2021 at 14:26
  • $\begingroup$ There is no "largest" growth of $\Delta_f$ like there is no "slowest" decrease of a sequence which ensures that a series is convergent. For every function $g(x)=\Delta_f(x)$ satisfying your condition there is a sequence of larger growth also satisfying this condition. $\endgroup$ Oct 23, 2021 at 14:41
  • $\begingroup$ Of course, the wording "the largest" is inappropriate. Thanks for editing your answer with the continuous version of the condition. I guess it now provides an answer to the sublinear bound I was looking for. $\endgroup$
    – cs89
    Oct 23, 2021 at 14:59

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.