5

I just had my first analysis course as an undergraduate, and I'm trying to learn more about analytic number theory. Right now I'm looking at prime numbers in particular--I'm studying (mostly just thinking and scribbling) on my own during my winter break. I've come across an infinite product that I'm trying to evaluate: I'm fairly sure it converges to 0, but I have no idea of how to prove this, since it is a product over the primes and, so far as I know, we have very few techniques for analyzing the behavior of the prime numbers.

I haven't learned how to use LaTex yet, but the infinite product is simple: it is the product over all primes of (p-1)/p.

It seems intuitive to me that the limit of the partial products is 0, but I have no idea how to prove it, mostly because I don't know how to analyze the behavior of the primes in this setting.

Any help would be very much appreciated. Thanks.

Ethan
  • 148
  • @Ethan what you are trying to evaluate is known as merten's result – happymath Dec 25 '15 at 04:59
  • The way a lot of people learn LaTeX is by seeing how others do it. Not the best way to go about it, but it might be the best way for you to go if you don't have time to read a book about it. For example, $$\prod_{p \textrm{ prime}} \frac{p - 1}{p}.$$ I did that with \prod_{p \textrm{ prime}} \frac{p - 1}{p}. – Bob Happ Dec 29 '15 at 21:46

2 Answers2

8

A more elementary way to prove this would be to notice that one may write $$\left(1+\frac{1}p+\frac{1}{p^2}+\frac{1}{p^3}+\ldots\right)=\frac{1}{1-\frac{1}p}=\frac{p}{p-1}.$$ This is the reciprocal of the term you want - which is to say that the product $$S=\prod_p 1+\frac{1}p+\frac{1}{p^2}+\frac{1}{p^3}+\ldots$$ is the reciprocal of the one you're finding (since $\frac{1}x\frac{1}y=\frac{1}{xy}$ we can bring the reciprocal outside the product). We want to show that $S=\infty$. However, this isn't too hard: We can use use a sort of distributive law to turn the above into a sum - in particular, notice that, for instance $$\left(1+\frac{1}2+\frac{1}{2^2}+\ldots\right)\left(1+\frac{1}3+\frac{1}{3^2}+\ldots\right)$$ will be the sum $$1+\frac{1}{2}+\frac{1}{2^2}+\ldots +\frac{1}3+\frac{1}{3\cdot 2}+\frac{1}{3\cdot 2^2}+\ldots + \frac{1}{3^2}+\frac{1}{3^2\cdot 2}+\frac{1}{3^2\cdot 2^2}\ldots$$ where the sum runs over all numbers of the form $\frac{1}{2^a3^b}$. Extending to the infinite case, we can show that, when we distribute out the product for $S$, we get the sum of $\frac{1}n$ over all integers writable as a product of primes - since all positive integers are uniquely representable as such a product, we can conclude that $$S=\sum_{n=1}^{\infty}\frac{1}n$$ meaning $S$ diverges to $\infty$ so $\frac{1}S$ is $0$.

You can make this rigorous by noting the following equality relating partial sums: $$\sum_{n=1}^{k}\frac{1}n\leq \prod_{p\leq k}\left(1+\frac{1}p+\frac{1}{p^2}+\ldots + \frac{1}{p^k}\right)$$ which follows from your run-of-the-mill distributive law and the fact that all $n$ below $k$ are writable as a product of primes below $k$ with exponent less than $k$.

Milo Brandt
  • 61,938
  • 1
    It is worthy of note that this also tells us that $\sum \frac{1}p$ diverges (which says something of the density of the primes) since, as a standard fact of analysis, $\prod 1-s_k$ diverges to $0$ exactly when $\sum s_k$ goes to infinity (for a sequence $s_k\in [0,1)$). – Milo Brandt Dec 25 '15 at 05:47
  • 1
    Awesome, that's an ingenious way of thinking about it! That's really interesting, thanks! – Ethan Dec 25 '15 at 05:48
5

Here is a modified version of proof which I saw in Alan Bakers Book Comprehensive course in Number Theory. In this I have used the prime number theorem but this can be done without it as well as done in the book I mentioned above.

The Prime Number Theorem states that $$\pi (x) = \frac{x}{\log x}+O(\frac{x}{(\log x)^2})$$

First notice that evaluating $\sum\limits_{p \leq x} \ln(1-\frac{1}{p})$ is enough since one has to just take its exponential to obtain the required estimate i.e $\displaystyle\prod \limits_{p \leq x} (\frac{p-1}{p})=\frac{ce^A}{\ln x}$

Mertens Result

$\displaystyle \sum\limits_{p \leq x} \ln(1-\frac{1}{p})= A+\ln(\frac{1}{\ln x})+\ln(c+O(\frac{1}{\ln x}))$

Proof:

We observe that $\displaystyle \sum\limits_{p \leq x} \ln(1-\frac{1}{p})=\sum \limits_{p \leq x} \sum \limits_{m=1}^{\infty} -\frac{1}{mp^m}$ $$\displaystyle \sum \limits_{p \leq x} \sum \limits_{m=1}^{\infty} -\frac{1}{mp^m}=\sum \limits_{p \leq x} \sum \limits_{m=1}^{1} -\frac{1}{mp^m}+\sum \limits_{p \leq x} \sum \limits_{m=2}^{\infty} -\frac{1}{mp^m}$$

But $\displaystyle \sum \limits_{m=2}^{\infty} \frac{1}{mp^m}=O(\frac{1}{p^2})$ this implies that

$$\displaystyle \sum \limits_{p \leq x} \sum \limits_{m=1}^{\infty} -\frac{1}{mp^m}=-\sum \limits_{p \leq x} \frac{1}{p}+\sum \limits_{p} (\ln(1-\frac{1}{p})+\frac{1}{p})-\sum \limits_{p>x} (\ln(1-\frac{1}{p})+\frac{1}{p})$$

$$\displaystyle \sum \limits_{p \leq x} \sum \limits_{m=1}^{\infty} -\frac{1}{mp^m}=-\sum \limits_{p \leq x} \frac{1}{p}+c+O(\sum \limits_{p>x} \frac{1}{p^2})$$ $$\hspace{24mm}=-\sum \limits_{p \leq x} \frac{1}{p}+c+O(\sum \limits_{p>x} \frac{1}{p^2})$$ $$\hspace{18mm}=-\sum \limits_{p \leq x} \frac{1}{p}+c_1+O(\frac{1}{x})$$

Now we will try to estimate the sum $-\sum \frac{1}{p}$

Here I have proved it asssuming the Prime Number Theorem

$\displaystyle \sum\limits_{n \leq x} a_n f(n)=s(x)f(x) - \int\limits_{1}^{x} s(u)f'(u)du$. Now taking $a_n=1$ if $n$ is a prime and $0$ otherwise and taking $f(x)=\frac{1}{x}$ we obtain $$\displaystyle\sum \limits_{p \leq x} \frac{1}{p}= \frac{\pi(x)}{x}+\int \limits_{1}^{x}\frac{\pi(u)}{u^2} du$$ From this after integrating we obtain $$\displaystyle\sum\limits_{p} \frac{1}{p}=\ln \ln x +c_2 + O(\frac{1}{\ln x})$$ Hence Proved.

happymath
  • 6,297
  • But the question is about the product of $1-\frac{1}{p}$ not the sum of $\frac{1}{p}$. – Dan Brumleve Dec 25 '15 at 05:19
  • @DanBrumleve just take the exponential of $\sum\limits_{p \leq x} ln(1-\frac{1}{p})$ – happymath Dec 25 '15 at 05:20
  • 1
    I see now, but is it not much easier to write it as the Euler product valid for $\text{Re}(s) \gt 1$ and take the limit as $s \rightarrow 1$? I realize that's not as formal as what you have so maybe it's not as straightforward as I think. – Dan Brumleve Dec 25 '15 at 05:24
  • Aha. I'm just beginning to study this branch of mathematics, so I'm still struggling to understand a lot of these concepts, but here's another question: what are A and c here? I assume they're constants, but do we have estimated values for them? Are there exact values for them, or do we just care about the fact that they're finite, because that's enough for evaluating the limit?

    By the way, I'm assuming that we can use your result to say that, as x grows without bound, the infinite product goes to 0, because the numerator is finite and the denominator grows without bound.

    – Ethan Dec 25 '15 at 05:32
  • @DanBrumleve I have edited it and I will think about your comment It looks like it should be possible to formalise your idea – happymath Dec 25 '15 at 05:34
  • @Ethan we can estimate $c_1$ but I am unable to estimate $c_2$ – happymath Dec 25 '15 at 06:33