I can't see how you get this.

I want to show that

$$\log(\log(N+1)) \leq \sum_{p \leq N} \frac{1}{p}+1$$

Can't see how it follows from this. So you show that $$0 \lt -\log(1-x)-x \lt \frac{x^2}{(1-x)}$$.

Which is fine, but then the lecturer does a jump I don't understand.

He rewrites this $$\sum_{p \leq N} \frac{1}{p} + \sum_{p\leq N} (-\log(1-\frac{1}{p}) - \frac{1}{p})$$

Know that
$$ \sum_{p\leq N} (-\log(1-\frac{1}{p}) - \frac{1}{p}) \leq \sum_{p \leq N} \frac{1}{p(p-1)}$$

Then just claims that this shows

$$\log(\log(N+1)) \leq \sum_{p \leq N} \frac{1}{p} +1$$


Let's start with : $$ \sum_{p\leq N} \left(-\log\left(1-\frac{1}{p}\right) - \frac{1}{p}\right) \leq \sum_{p \leq N} \frac{1}{p(p-1)}$$

The idea is to use the finite version of Euler's product : $$\prod_{p\le N} \frac1{1-p^{-1}} =\prod_{p\le N} \left(1+\frac{1}{p}+\frac{1}{p^2}+\cdots\right)=\sum_{n\in S_N} \frac1n$$
with $S_N$ the set of integers composed of prime factors $p\le N$

Since this set includes all the integers $n\le N$ we have :
$$\prod_{p\le N} \frac1{1-p^{-1}} \ge H_N$$

but the harmonic sum $H_N$ is well known to be greater than $\log(N+1)$ (use for example minoration by $\int_1^{N+1} \frac1n dn$).

So that (taking logarithms) we get : $$\log(\log(N+1)) \le \sum_{p\le N} -\log\left(1-\frac1p\right)$$ that we may replace in your inequality.

UPDATE2: Let's handle with more care this replacement to show that indeed : $$\log(\log(N+1)) \leq \sum_{p \leq N} \frac{1}{p} +1$$

from the convergent : $\ \displaystyle 0 \le \sum_{p\leq N} \left(-\log\left(1-\frac{1}{p}\right) - \frac{1}{p}\right) \leq \sum_{p \leq N} \frac{1}{p(p-1)}$
we get (adding the sum on $\frac1p$) : $$ \sum_{p\leq N} \frac{1}{p}\le \sum_{p\leq N} -\log\left(1-\frac{1}{p}\right) \leq \sum_{p \leq N} \frac{1}{p-1}$$ At this point we may use the previous result to get not your inequality but : $$ \log(\log(N+1)) \le \sum_{p \leq N} \frac{1}{p-1}$$

Let's use the fact that $\frac1{p-1}\le \frac1q$ (if $q$ is the prime before $p$) or $1$ (for $p=2$) to rewrite this as ('shifting' the primes and extracting $1$) : $$ \log(\log(N+1)) \le 1+\left(\sum_{p \le N} \frac1p\right)-\frac1{p_N}\ ,\ \text{with }p_N\ \text{the largest prime}\ \le N$$ this implies indeed : $$ \log(\log(N+1)) \le 1+\sum_{p \le N} \frac1p$$

We may replace the $1+$ term by $\frac12+$ by noticing that your initial inequality could have been (adding $2$ at the denominator) : $$0 \lt -\log(1-x)-x \lt \frac{x^2}{2(1-x)}$$ because the effect of this is to replace the majoration of $\frac1{p(p-1)}+\frac1p=\frac1{p-1}$ by $\frac1{2p(p-1)}+\frac1p=\frac{p+p-1}{2p(p-1)}=\frac12(\frac1{p-1}+\frac1p)$ so that the final inequality is replaced by : $$ \log(\log(N+1)) \le \frac12\left(1+\sum_{p \le N} \frac1p+\sum_{p \le N} \frac1p\right)$$

We could continue this way but better results may be proved (see for example the excellent Hardy&Wright 'An introduction to the theory of numbers')
using the function $\displaystyle C(x)=\sum_{p\le x} \frac{\log p}p$ and the equality $\displaystyle \sum_{p\le x}\frac1p=\frac{C(x)}{\log x}+\int_2^x \frac{C(t)}{t\ \log^2(t)}dt\ $ to get : $$\sum_{p\le x}\frac1p= \log\log x+B_1+o(1)$$ with $B_1$ the Merten's constant : $$B_1=\gamma+\sum_p \log\left(1-\frac1p\right)+\frac1p$$


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.