In Mean values of multiplicative functions over function fields the mention a proof of Halasz inequality in one of their future pre-prints. In fact here is an a proof from 1999.

I would like to see a proof in any way shape or form, as it might lead to a proof of the prime number theorem.

There are various statements floating around here are two:

[1,3] Let $f : \mathbb{N} \to \mathbb{C}$ be multiplicative function such that $|f(n)| \leq 1$. Then if we establish for all real numbers $|t| < T$ that: $$ \sum_{p < x} \frac{1 - \mathrm{Re}[\, f(p) p^{-it} \,]}{p} \geq M$$ then the average value of $f(n)$ is quite small: $$ \frac{1}{x} \sum_{n < x} f(n) \ll (1+M)e^{-M} + \frac{1}{\sqrt{T}} $$ even the "cheap" version stated in the blog might already imply the prime number theorem.

[2] If the mean value of $f$ is "large" in absolute value, then $f(n)$ pretends to be $n^{it}$ for some "small" real $t$. Here, pretending is defined in terms of Kullback-Liebler distance:

$$ \mathbb{D}(f, g; x)^2 = \sum_{p \leq x} \frac{1 - \mathrm{Re}\big[\,f(p)\overline{g(p)}\,\big]}{p} $$

All of the proofs seem long and technical, and the new proof may not be much simpler. I still might like to know if the weaker version implies PNT

  • 3
    $\begingroup$ I do not see any connection to the Kullback-Liebler . Maybe we should call it "Mochizuki norm" instead? $\endgroup$ – aosjckajsd Jul 6 '16 at 19:22
  • $\begingroup$ @aosjckajsd are you joking? $\endgroup$ – john mangual Jul 6 '16 at 20:27
  • 5
    $\begingroup$ Yes! But you should be up-voting me instead of asking questions. $\endgroup$ – aosjckajsd Jul 6 '16 at 20:30

Heuristically (and in fact rigorously with a bit of work) we have $$ \sum_{p \leq x} \frac{1 + \Re(p^{it})}{p} = \log\log x + \Re \log \zeta(1 + \frac{1}{\log x} + it) + O(1) $$ Therefore if $\zeta$ has a zero at $1 + iu$ then the hyp othesis of Halasz's theorem fails and it doesn't imply the Prime number theorem. So you get the Prime Number Theorem from Halasz + non-vanishing of $\zeta(s)$ on the $1$-line.

  • 3
    $\begingroup$ Can the ignorant masses explain why they are down-voting a correct answer? $\endgroup$ – aosjckajsd Jul 6 '16 at 23:00

It has been over a year since I asked. The rumor it was being worked on (in fact clearly stated by the authors). I note that two papers came out yesterday:

  1. A more intuitive proof of a sharp version of Halász's theorem
  2. A new proof of Halász's Theorem, and its consequences

For both papers, authors are: Andrew Granville, Adam J Harper, Kannan Soundararajan.

I was looking for any proof of Halasz theorem, completing a develoment of the "pretentious" approach to number theory. Here's the lemma they build it from: an averaging formula.

$$ \int_{-T}^T \left|\frac{a_n\,\Lambda(n)}{n^{1+it}} \right|^2 \, dt \ll \sum_{T^2 \leq n \leq x} \frac{|a_n|^2\Lambda(n)}{n}$$

I am not a specialist in analytic number theory and these averages look rather technical and in need of interpretation. They say the proof splits into a combinaorial part and an analytic part.

Combinatorics and Real Analysis are very broad areas with a wide range of techniques available. At least, in 6 short pages, we have the result.

The combinatorial lemma looks like many we have seen before: $$ \sum_{n \leq x} f(n) = \frac{1}{\log x} \sum_{n\leq x} f(n) \big( \log n + \log x/ n \big) = \frac{1}{\log x} \sum_{mp \leq x} f(m)f(p) \log x + O \left( \frac{x}{\log x} \right) $$
This is a double convolution which gets turned into a triple convolution. However, it is a "triple convolution" after discarding many primes and only approximately. $$ (\log x)^4 < p < \frac{x}{2} $$ They do this to make the proof work. For me, an outsider, it takes a bit of thought too appreciate why these substitutions are OK.


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.