3
$\begingroup$

We know that the number of squarefree integers $\le x$ that are coprime to $A$ is $$ Q_A(x) = x \prod_{p|A} \left(1-\frac{1}{p}\right) \prod_{p \nmid A} \left(1-\frac{1}{p^2}\right) + O(\sqrt{x}). $$ Do we have explicit upper and lower bounds on $Q_A(x)$?

$\endgroup$

1 Answer 1

5
$\begingroup$

I am assuming that explicit refers to the error term? In this case you can write $$Q_A(x)=\sum_{d\mid A}\mu(d)\sum_{k\leq \sqrt x \atop \gcd(A,k)=1}\mu(k)\left[\frac{x}{dk^2}\right],$$ where $[t]$ denotes the largest integer $\leq t $. Note that $[t]=0$ when $0\leq t<1$ hence only the terms with $dk^2 \leq x $ make a contribution. Using $[t]=t+\theta(t)$ for some real number $ \theta(t) $ which satisfies $|\theta(t)|\leq 1$, we get $$\left|Q_A(x)-\sum_{d\mid A}\mu(d)\sum_{k\leq \sqrt x \atop \gcd(A,k)=1}\mu(k) \frac{x}{dk^2}\right| \leq \sum_{d\mid A}|\mu(d)| \sum_{1\leq k\leq \sqrt {x/d} } 1,$$ where we ignored $\gcd(A,k)=1$ in the error term. Now we have $$ \sum_{d\mid A}|\mu(d)| \sum_{1\leq k\leq \sqrt {x/d} } 1\leq \sqrt x \epsilon(A) ,$$ where $$\epsilon(A):=\sum_{d\mid A}\frac{|\mu(d)|}{\sqrt d }=\prod_{p\mid A}\left(1+\frac{1}{\sqrt p }\right).$$ Hence, $$ xM-\epsilon(A) \sqrt x \leq Q_A(x)\leq xM+\epsilon(A) \sqrt x,$$ where $$M=\sum_{d\mid A}\frac{\mu(d)}{d}\sum_{k\leq \sqrt{ x/d} \atop \gcd(A,k)=1} \frac{\mu(k)}{k^2}.$$ We can simplify $M$ to look into an Euler product: first adding the missing terms $k>\sqrt {x/d} $ we introduce an error term that is in absolute value $$\leq \sum_{d\mid A} |\mu(d)|\sum_{k>\sqrt {x/d} } \frac{1}{k^2}\leq \sum_{d\mid A} |\mu(d)|\int_{\sqrt{x/d}/2}^\infty \frac{\mathrm d u }{u^2} =\frac{2\epsilon(A)}{\sqrt{x}}.$$ The full series then becomes $$ \sum_{d\mid A}\frac{\mu(d)}{d}\sum_{k=1 \atop \gcd(k,A)=1}^\infty \frac{\mu(k)}{k^2}=\prod_{p|A} \left(1-\frac{1}{p}\right) \prod_{p \nmid A} \left(1-\frac{1}{p^2}\right),$$ hence, $$\prod_{p|A} \left(1-\frac{1}{p}\right) \prod_{p \nmid A} \left(1-\frac{1}{p^2}\right) -\frac{2\epsilon(A)}{\sqrt{x}}\leq M\leq \prod_{p|A} \left(1-\frac{1}{p}\right) \prod_{p \nmid A} \left(1-\frac{1}{p^2}\right)+\frac{2\epsilon(A)}{\sqrt{x}}.$$ Combining with the previous work we get $$ \left|Q_A(x)-x\prod_{p|A} \left(1-\frac{1}{p}\right) \prod_{p \nmid A} \left(1-\frac{1}{p^2}\right)\right|\leq 3\epsilon(A) \sqrt x . $$ Assuming the Riemann hypothesis one can prove an exponent better than $1/2$ in $x^{1/2}$ and without Riemann hypothesis this can be improved slightly to something like $x^{1/2}/\exp(c\sqrt{\log x } )$ but it will be technical to find the implied constant as it depends on the zero free region of the Riemann zeta.

Remark: A bound for $\epsilon(A)$ may be found as follows: for $y=\tau(A)^{2/3} $ we have $$\epsilon(A) \leq \sum_{d\leq y}1+\sum_{d\mid A}1/\sqrt y \leq y+\tau(A)/\sqrt y=2 \tau(A)^{2/3}.$$

$\endgroup$
6
  • 1
    $\begingroup$ Ideally, what is good for your application? A function of $A$ that grows slower or would you like to have something without $A$ at all? In case that $1\leq A \leq x $ then you could use the inequality $\tau(A) \leq c A^{1/4}$ to deduce that $|Q_A(x)-\prod(\cdots) x| \leq 3cx^{3/4}$. This gives a bound independent of $A$. Such an approach works even when $A\leq x^{100^{100}}$. $\endgroup$
    – Dr. Pi
    May 20, 2021 at 16:30
  • 1
    $\begingroup$ Actually, some dependence on $A$ must always be in the error term. Indeed, assume that for all $A$ and $x$ we have $Q_A(x)=x \prod+O(\sqrt x )$ where the implied constant is absolute. Now take $A$ to be the product of all primes up to $x$. In particular $Q_A(x)=1$ so one would then get $\prod\ll 1/\sqrt{x}.$ But the product can be seen to be essentially of order $1/\log x$ by Mertens' theorem-hence, we proved that $\log x \gg \sqrt x $, contradiction. $\endgroup$
    – Dr. Pi
    May 20, 2021 at 17:44
  • 1
    $\begingroup$ $3\tau(A)$ can be improved a little bit. Namely, the argument in the proof can be modified slightly to give the bound $\sum_{d\mid A\atop d \leq x }\mu^2(d)d^{-1/2}$ in place of $3\tau(A)$. This can be done by replacing the initial condition $k\leq \sqrt x$ by $k\leq \sqrt{x}/d^{1/2}$. On average this is a better error term because the mean value of $\sum_{d\mid A } d^{-1/2}$ is bounded whereas the mean value of $\tau(A)$ is $\log A\to \infty$. $\endgroup$
    – Dr. Pi
    May 20, 2021 at 18:01
  • 1
    $\begingroup$ For my application, $A$ is fixed and quite small, say $A < 10^4$, so I can calculate $\tau(A)$. However, I think your suggestion about using the bound $\sum_{d|A} \mu^2(d)d^{-1/2}$ is a good one because for small values of $A$, I can calculate it easily. $\endgroup$
    – Iguana
    May 21, 2021 at 6:17
  • 1
    $\begingroup$ I edited the first answer to prove the improved bound. $\endgroup$
    – Dr. Pi
    May 21, 2021 at 8:44

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.