I'm working through a paper and the author states the following result with little justification, please check my working.

Let $h:[y,\infty)\to\mathbb{R}^{>0}$ be decreasing with a continuous derivative, then for all $t\geq y$ we have $$ \sum_{p\leq y}\frac{\log p}{p}h(pt)=\int_{t}^{yt}\frac{h(v)}{v}\ dv+O\big(h(y)\big)\ . $$

(Convention: $\sum_{p\leq y}$ means "sum over primes $p$ less than or equal to $y$")

$\textit{My work:}$

We use two standard results.

  1. $\sum_{p\leq y}\frac{\log p}{p}=\log y+O(1)$ and
  2. $\sum_{n\leq y}f(n)g(n)=F(y)g(y)-\int_{1}^{y}F(u)g'(u)\ du$, where we assume that $g$ has a continuous derivative and $F(x)=\sum_{n\leq x}f(n)$.

Let $f(n)=\frac{\log n}{n}$ for primes $n$ and $0$ otherwise, let $g(x)=h(tx)$ and apply these to 2.:

$$\begin{array}{rcl} \displaystyle{\sum_{p\leq y}\frac{\log p}{p}h(pt)}&=&\displaystyle{\sum_{n\leq y}f(n)h(tn)} \\ &=&\displaystyle{\big(\log y+O(1)\big)h(ty)-\int_{1}^{y}\big(\log u+O(1)\big)\big(th'(ut)\big)\ du}\\ &=&\displaystyle{\log y\big(h(ty)\big)+\underbrace{O\big(h(ty)\big)}_{O(h(y))}-\int_{1}^{y}\log u\big(th'(ut)\big)\ du+\underbrace{O\Big(\int_{1}^{y}\big(th'(ut)\big)\ du\Big)}_{O(h(y))}}\\ &=&\displaystyle{\int_{1}^{y}\frac{h(ut)}{u}\ du+O\big(h(y)\big)} \\ &=&\displaystyle{\int_{t}^{yt}\frac{h(v)}{v}\ dv+O\big(h(y)\big)} \end{array}$$

Is there a nicer way of obtaining the result? I appear to have used integration/summation by parts twice.


Your Answer

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

Browse other questions tagged or ask your own question.