Define $$G(x)=\sum_{n \leq x} T\left(\frac{x}{n}\right)$$ and $G,T: [1,\infty) \to \mathbb R$

And function T satisfies the following conditions:

1) $T(x)=O(x)$

2) $T(x) \sim cx (x \to \infty)$

How to show that $G(x) \sim cx\log{x} (x \to \infty)$? I have tried to use the partial summation, but feel that might not work.

  • $\begingroup$ Generlized Mobius inversion may be helpful here. See Mike Spivey's answer here for a similar problem: math.stackexchange.com/questions/66820/… $\endgroup$
    – JavaMan
    Sep 28, 2011 at 5:48
  • $\begingroup$ I don't see how either of partial summation or Mobius inversion will yield an answer. Israel's hint below suggests a very straightforward solution. $\endgroup$
    – anon
    Oct 6, 2011 at 23:35

2 Answers 2


Actually a little more is needed: $T$ must be bounded on bounded subsets of $[1,\infty)$. Otherwise a counterexample is $T(t) = t + 1/(t-1)^2$ for $t > 1$, $T(1) = 0$.

Hint: If $c_1 x < T(x) < c_2 x$ for $x > N$, and $|T(x)| <= B$ for $x \le N$, what can you say about $G(x)$?

  • $\begingroup$ It's possible the $O(x)$ condition was specified or implied on $[1,\infty)$, otherwise it would be redundant in light of the second condition, but it's hard to tell. $\endgroup$
    – anon
    Oct 6, 2011 at 23:37
  • $\begingroup$ @anon: that is the assumption I made in my answer. $\endgroup$
    – robjohn
    Oct 7, 2011 at 0:05

Since your condition $2$ implies your condition $1$, perhaps you meant that for $x\ge1$ $$ |T(x)|\le c_1x $$ I believe that this would satisfy Robert Israel's concern.

Let $\epsilon>0$. By the second condition, there is a $B_\epsilon$, so that for $x>B_\epsilon$, $$ (c-\epsilon)x<T(x)<(c+\epsilon)x $$ Break up the sum into two parts $$ \sum_{x\ge n>x/B_\epsilon}\left|T\left(\frac{x}{n}\right)\right|\le\sum_{x\ge n>x/B_\epsilon}c_1\frac{x}{n}\tag{1} $$ $$ \sum_{n\le x/B_\epsilon}(c-\epsilon)\frac{x}{n}\le\sum_{n\le x/B_\epsilon}T\left(\frac{x}{n}\right)\le\sum_{n\le x/B_\epsilon}(c+\epsilon)\frac{x}{n}\tag{2} $$ Since $\displaystyle\sum_{n\le x}\frac{1}{n}=\log(x)+\gamma+O\left(\frac{1}{x}\right)$, we get that $$ \begin{align} \left|\sum_{n\le x}T\left(\frac{x}{n}\right)-cx\log(x)\right| &\le\epsilon x(\log(x)-\log(B_\epsilon))+\gamma x+c_1x\log(B_\epsilon)+O\left(1\right)\\ &\le x\log(x)\left(\epsilon+\frac{\gamma+c_1\log(B_\epsilon)+O\left(\frac{1}{x}\right)}{\log(x)}\right) \end{align} $$ which says that $G(x)\sim cx\log(x)$.


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.