5
$\begingroup$

I am currently reading Iwaniec-Kowaleski's book on Analytic Number Theory.

My question is on page 431. Here is what it says:

For any $\chi \mod q$, we have the twisted series $K(s,\chi)=\sum_{n=1}^\infty \left( \sum_{d|n} \lambda_d \right) \left( \sum_{d|n} \theta_b \right) \chi(n) n^{-s}.$

Here, we have $\lambda(d)= \mu(d) \min \left(1, \frac{\log(z/d)}{\log(z/w)} \right)$ for $1 \leq d \leq z$ where $1<w<z$ and we set $\lambda_d=0$ if $d>z$.

Thus, we have $\sum_{d|n} \lambda_d=1$ if $n=1$ and $\sum_{d|n} \lambda_d=0$ if $1<n \leq w.$

Furthermore, we have $\theta_b = \frac{\mu(b) b}{G \phi(b)} \sum_{ab \leq y \atop (a,bq)=1} \frac{\mu^2 (a)}{\phi(a)}$ and $G$ is the normalization factor such that $\theta_1=1.$

Now, we can see that $K(s,\chi)$ factors into $L(s,\chi)M(s,\chi)$, where $M(s,\chi)=\sum_m \left(\sum_{[b,d]} \sum_{=m} \lambda_d \theta_b \right) \chi(m) m^{-s}$.

Now, if we take only a partial sum of $K(s,\chi)$, we can define $K_x(s,\chi)=\sum_{n=1}^x \left( \sum_{d|n} \lambda_d \right) \left( \sum_{d|n} \theta_b \right) \chi(n) n^{-s}$ with $x=(qT)^{23}.$ Let $m=[b,d] \leq bd \leq yz $ and $\chi \not= \chi_0.$

Then, why does this imply $\left|\sum_{n>x/m} \chi(n) n^{-s} \right| \leq 2q |s| (m/x)^{\sigma}$ where $s=\sigma +it$. I do not understand where the $2q|s|$ comes from.

Assuming this, the book states $|K(s,\chi)-K_x(s,\chi)| \leq 2q|s|yz x^{-\sigma}$. I do not understand where the yz term comes from.

Sorry for the length of this post. I wanted to give as much information as possible.

$\endgroup$
1
+50
$\begingroup$

Showing $\left|\sum_{n>x/m} \chi(n) n^{-s} \right| \leq 2q |s| (m/x)^{\sigma}$.

I will assume that $q > 1$ and that $\sigma \geq 1/2$. Then $$\sum_{n \geq N} \frac{\chi(n)}{n^s} = \sum_{n \geq N} S_\chi(n) \left( \frac{1}{n^s} - \frac{1}{(n+1)^s}\right) \tag{1},$$ where $S_\chi(n) = \sum_{N \leq m \leq n} \chi(m)$. This is an application of partial summation.

A trivial bound on $S_\chi(n)$ is $\lvert S_\chi(n) \rvert \leq q$. (It is here that I use that $q > 1$). In fact, this is a very weak bound, but that's ok.

I note that for $\mathrm{Re}(s) > 0$, and $0 < a < b$, we have $$ \left \lvert \frac{1}{a^s} - \frac{1}{b^s}\right \rvert \leq \frac{\lvert s \rvert}{\sigma} \left( \frac{1}{a^\sigma} - \frac{1}{b^\sigma}\right) \leq \frac{\lvert s \rvert(b - a)}{a^{\sigma + 1}}.$$ For a quick proof, look at $$ \int_a^b \frac{dx}{x^{s+1}} = \frac{1}{s} \left( \frac{1}{a^s} - \frac{1}{b^s}\right),$$ and thus $$ \left \lvert \frac{1}{a^s} - \frac{1}{b^s}\right \rvert \leq \lvert s \rvert \int_a^b \frac{dx}{\lvert x^{\sigma+1} \rvert},$$ giving the first inequality. For the second inequality, one can apply the mean value theorem to $1/x^{\sigma}$, or a trivial integral estimate.

In this case, this gives that $$ \left\lvert \frac{1}{n^s} - \frac{1}{(n+1)^s}\right\rvert \leq \frac{\lvert s \rvert}{n^{\sigma + 1}}.$$ Applying to $(1)$ shows that $$ \sum_{n \geq N} \frac{\chi(n)}{n^s} \leq \sum_{n \geq N} q \frac{\lvert s \rvert}{n^{\sigma + 1}} \leq 2 q \lvert s \rvert N^{-\sigma}.$$ In your case, you have $N = x/m$, and this proves the claim. $\diamondsuit$

$|K(s,\chi)-K_x(s,\chi)| \leq 2q|s|yz x^{-\sigma}$

I didn't work out the complete details, but I did work out part of the argument. In short, I didn't want to dive into any of the details concerning $x,y,z$ or $\theta_b, \lambda_d$, and so I use only trivial bounds for those parts.

From IK, we have the following properties:

  • $x = (qT)^{23}$
  • $y = (qT)^2$
  • $z = (qT)^8$
  • $w = (qt)^7$

And also $\lambda_d = 0$ if $d > z$, and $\theta_b = 0$ if $b > y$. Further, $\lvert \lambda_d \rvert \leq 1$ and $\lvert \theta_b \rvert \leq 1$. (IK has a whole section devoted to $\theta_b$ and the Selberg sieve, and restates some relevant bounds here for more precise bounds than I give below).

Then one can write $$ K_x(s, \chi) = \sum_{m \geq 1} \left( \sum_{[b, d] = m} \sum \lambda_d \theta_b \right) \frac{\chi(m)}{m^s} \sum_{n \leq x/m} \frac{\chi(n)}{n^s}$$ (in the same factoring sort-of argument that shows that $K(s, \chi) = M(s, \chi) L(s, \chi)$). Thus $$ \lvert K(s, \chi) - K_x(s, \chi) \rvert \leq \left \lvert \sum_{m \geq 1} \sum_{[b,d] = m} \lambda_d \theta_b \frac{\chi(m)}{m^s}\right \rvert \left \lvert \sum_{n \geq x/m} \chi(n)n^{-s}\right \rvert. \tag{2}$$

The latter part of $(2)$ is exactly what was considered above, and as $m \leq bd \leq yz$, we have that $$ \left \lvert \sum_{n > x/m} \chi(n) n^{-s} \right \rvert \leq 2 q \lvert s \rvert (yz)^\sigma x^{-\sigma}.$$

For the first sum of $(2)$, let us naively and lossfully note that $$ \sum_{[b,d] = n} \lambda_d \theta_b \leq \tau(n),$$ the number of divisors of $n$. Then the first $\lvert \cdot \rvert$ part can then be bounded above by $$ \sum_{n \leq yz} \tau(n) n^{-\sigma} \leq yz \log(yz) \left(1+ \frac{(yz)^{- \sigma}}{\sigma}\right) \leq 2(yz)^{1 - \sigma} \log(yz).$$

Putting these together, we find that $(2)$ is bounded above by $$2(yz)^{1 - \sigma} \log(yz) 2 q \lvert s \rvert (yz)^\sigma x^{-\sigma} = 4 q \lvert s \rvert yz \log(yz) x^{-\sigma}.$$ This is slightly weaker than the stated result in IK, but I think this describes how to get the sort of result as stated.

$\endgroup$

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.