Skip to main content
3 of 3
added high level tag, edited latex to be more readable

Asymptotic lower bound for the number of square free with at least two prime factors

In one of Soundararajan's papers, he claims without proof that it is a standard exercise to show that the number $N(X)$ of positive square-free integers $d \equiv 1 \; \bmod \; 8$ less than $X$, with at least two primes factors all of which is congruent to $\pm1 \; \bmod \; 8$ is asymptotically bounded by $\frac{X}{\sqrt{\log(X)}}$ up to a constant. I am new to techniques in analytical number theory, and I find it hard to see this as a standard exercise. Based on my limited knowledge of such techniques I believe this may follow from some kind of Tauberian theorem. But, I couldn't come up with that sort of proof to the above fact. So I would like to see a standard proof of this fact.

Instead, I did some hand-wavy computations that lead me to a much stronger lower bound $\frac{X}{\sqrt{\log(\log X)}}$ (up to a constant). For simplicity, I present my ideas to just counting those $d$ with the stated properties but with prime factors congruent to $1 \; \bmod \; 8$ only, the general case is a simple extension of the same ideas. There are $M = M(x,r) = \frac{X^{1/r}}{8\log(X^{1/r})} = \frac{rX^{1/r}}{8\log(X)}$ number of primes less than $X^{1/r}$ and congruent to $1 \; \bmod \; 8$. So there are $\binom{M}{r}$ number of such $d<X$. Now summing them from $r = 2$ onwards we have $$N(X) = \sum_{r \geq 2} \binom{M(x,r)}{r} \gg \sum_{r \geq 2} {\left(\frac{rX^{1/r}}{8\log(X)}\right)}^r \frac{1}{r!} \gg X \sum_{r \geq 2} \frac{1}{{(8\log(X))}^r} \frac{r^r}{r!}.$$ And now using Stirling's approximation, $\frac{r^r}{r!} \approx \frac{e^r}{\sqrt{2 \pi r}}$. So $N(X) \gg X \sum_{r \geq 2} {\left(\frac{e}{8\log(X)}\right)}^r r^{-1/2}$, approximating the sum as an integral $$N(X) \gg \int_2^{\infty}a^x x^{-1/2} dx \approx \frac{\Gamma(1/2, -2 \log(a))}{\sqrt{-\log(a)}}\gg> \frac{1}{\sqrt{\log(\log X)}},$$ where $a = \frac{e}{8\log(X)}$. Hence we have the result. (I use "$\gg$" above to hide constants in the inequalities)

Since my method gives a somewhat stronger bound, I am skeptical if there is some flaw in my ideas. Please feel free to comment on what I have done.

Melanka
  • 577
  • 2
  • 8