The order of magnitude of $\prod_{p \leq N}(1 + p^{-r})$ as $N \to \infty$

63 Views Asked by At

Let \[ f(N, r) \equiv \prod_{p \leq N} (1 + p^{-r}) \quad 0 < r < 1 \] where $p$ runs through all the primes less than or equal to $N$.

What is the order of magnitude of $f(N, r)$ as $N \to \infty$?

I think this sounds a very easy problem. But I am doubtful of what I did.

The obvious plan is to exponentiate, take the logarithm, and then use the prime number theorem.

So I get \[ f(N, r) = \exp[\log f(N, r) ] = \exp[\sum_{p \leq N}p^{-r} + O(\sum_{p \leq N} p^{-2r}) ] \asymp \exp[\sum_{p \leq N }p^{-r}]. \]

With the prime number theorem and Abel summation formula, I get \[ \sum_{p \leq N}p^{-r} \asymp \pi(N)N^{-r} + r \int_{2}^{N}\pi(u)u^{-r - 1} du \gg N^{1 - r} (\log N)^{-1}. \]

Thus, I get \[ f(N, r) \gg \exp[N^{1 - r} (\log N)^{-1}] \quad ? \]

But my instinct (which is actually not good and often careless) tells me that "$f(N)$ increases exponentially" seems very wrong.

I used Mathematica and got \[ \lfloor f(p_{200}, 2/3) \rfloor = \lfloor f(1223, 2/3) \rfloor = 486, \] which is not of exponential order at all.

Please direct me to the right path.

1

There are 1 best solutions below

1
On

Arguing non-rigorously,

$\begin{array}\\ \sum_{p \leq N}p^{-r} &\approx \sum_{k=1}^{N/\ln(N)} (k \ln k)^{-r}\\ &\approx \int_2^{N/\ln(N)}\dfrac{dx}{(x\ln(x))^r}\\ &\sim \dfrac1{\ln^r(N/\ln(N))}\int_2^{N/\ln(N)}\dfrac{dx}{x^r} \qquad\text{since }\ln^r(x) \text{ doesn't vary much}\\ &\sim \dfrac1{\ln^r(N/\ln(N))}\dfrac{x^{1-r}}{1-r}|_2^{N/\ln(N)}\\ &\sim \dfrac1{\ln^r(N/\ln(N))}\dfrac{(N/\ln(N))^{1-r}}{1-r}\\ &\sim \dfrac{N^{1-r}}{(1-r)(\ln(N))^{1-r}(\ln(N)-\ln\ln(N))^r}\\ &\sim \dfrac{N^{1-r}}{(1-r)\ln(N)}\\ \end{array} $

which is about what you got.

According to Wolfy, for $N = 200\ln(200), r=2/3$, this is about $4.39$ and exp of this is about $80.70$ which is within a factor of 6 of your computation.

Since the approximations are moderately crude with lots of constants being thrown away, I don't think that this is too bad.