There is a very nice question which arises in the study of the Discrete Logarithm Problem which I wish to present here. The question, in a general setting, is to specify an empirical expression for the "frequency" of fields for which the $p$-adic regulator vanishes $(\bmod p)$. More specifically: If $k$ is a number field and $p$ an odd prime, we let

$$R_{p}(k) \in \operatorname{Rems} :=\{0, 1, ..., p-1\}$$

be the remainder of the p-adic regulator modulo $p$.

One would expect that the values of $R_{p}(k)$ are uniformly distributed within $\operatorname{Rems}$.

I would not expect the proof of such a statement to be within reach with presently known techniques. However, there are some questions concerning the way to even state the uniform distribution.

I think of two possible variants for this:

  1. Arrange the number fields according to some "measure" that may depend on the degree and the discriminant, or only one of these parameters.

Therefore $\|k\| = \operatorname{disc}(k)$ and one takes averages either over all fields or over all fields of fixed degree. Let then

$$N_B = \{ \text{number fields}\ k : \| k \| < B\ \text{and}\ R_{p}(k) = 0 \},$$


$$F_B = \{ \text{number fields}\ k : \| k \| < B \}.$$

$N_B(d)$ and $F_B(d)$ are defined similarly, restricting only to fields of degree $d$.

The first conjectures: Fix $p$. Then $$ \text{Conjecture 1:}\quad \lim_{B\rightarrow +\infty} N_B/F_B = 1/p. $$ $$ \text{Conjecture 2:}\quad \lim_{B\rightarrow +\infty} N_B(d)/F_B(d) = 1/p. $$

  1. The second approach fixes $k$ and lets $p$ vary. Let $P \subset N$ be the set of all natural primes, then let $$\chi(p) = \begin{cases} 0&\text{if}\ R_{p}(k) > 0,\\ 1&\text{if}\ R_{p}(k) = 0.\\ \end{cases} $$

Then the conjectures state: $$ \text{Conjecture 3:}\quad \lim_{B\rightarrow +\infty} \frac{\displaystyle \sum_{ p \in P, p < B} \chi(p) p }{B/\log(B)} = 1. $$ $$ \text{Conjecture 4:}\quad \lim_{B\rightarrow +\infty} \frac{\displaystyle \sum_{ p \in P, p < B} \chi(p) }{\log \log B} = 1 $$


a) Are analogous statements about uniform distribution of, say, the class number modulo $p$ or some other field invariants already discussed in the literature? And if yes, how does one express the uniform distributions?

b) Are any secondary phenomena, that may give raise to some constants in the above conjectures, especially in Conjectures 3 and 4? I have in mind things like the correction factors in the Hardy-Littlewood Conjecture or other similar ones.

c) Do there exist some alternative ways for stating uniform distribution?

  • $\begingroup$ Your formulas would be easier to read if you used MathJax (which is essentially LaTeX) for formatting. There is a good guide here: meta.math.stackexchange.com/q/5020/166535 $\endgroup$ – Joonas Ilmavirta Apr 4 '15 at 20:57
  • 2
    $\begingroup$ A couple of things. First, no need to put the words "Research level" in the title of your post. Readers will judge whether the post is at the right level for this site. Second, MathOverflow uses basic LaTeX formatting. If you know LaTeX, please edit your post so that it's more readable. If not, post a comment asking for help and maybe someone will format it for you. But it really is worthwhile, if you're going to do math or crypto, to learn basic LaTeX. $\endgroup$ – Joe Silverman Apr 4 '15 at 21:30
  • $\begingroup$ I hope people were not downvoting this simply because of tex. $\endgroup$ – Lucia Apr 5 '15 at 4:31
  • $\begingroup$ Let us leave it open for now and let the author do the edits. $\endgroup$ – Dima Pasechnik Apr 5 '15 at 6:37
  • 1
    $\begingroup$ Please indicate a good reference linking the discrete log problem to $p$-adic regulators. Didn't know about that. Unless I'm forgetting something, the $p$-adic regulator is only known to be well-defined for special number fields, e.g., totally real number fields, so do you have in mind all number fields or just some special types of number fields? Does your counting of number fields include each number field once up to isomorphism, or are you counting number fields inside an algebraic closure of $\mathbf Q$ (then a field of degree $d$ occurs $d$ times)? Finally, why do you avoid $p = 2$? $\endgroup$ – KConrad Apr 5 '15 at 14:03

Some "equidistributions $\mod{p}$" are discussed in Washington's "Introduction to Cyclotomic Fields". I adress here only your first question, and have nothing to say about questions b) and c).

The first equidistribution is about irregular primes, and is discussed after Theorem 5.17 (I refer, here and henceforth, to the second edition), on page 62-63. The idea is to assume that the classes of Bernoulli numbers are random $\mod{p}$, in the sense that the probability that $p\mid B_j$ for some $j\in[2,4,6,\dots,p-3]$ is $1/p$. In this case one coops with a Poisson distribution and the deduced heuristic is that approximately $60\%$ of primes are regular, which agrees with computations.

A second distribution argument is briefly touched upon after Theorem 5.37 and concerns the possibility that the Bernoulli number $B_{(p-1)/2}$ be random $\mod{p}$. This would have consequences on the residue $\mod{p}$ of the class number of $\mathbb{Q}(\sqrt{p})$: as it is explained in Exercice 5.9, though, one can show that $B_{(p+1)/2}$ is not random $\mod{p}$, as a consequence of Brauer-Siegel (at least for $p\equiv 3\pmod{4})$.

A third one concerns Vandiver's conjecture and is discussed in the Remark on page 158 ($\S$ 8.4). It shows that the naive approach of assuming that the $i$-th cyclotomic unit be a $p$-th power with probability $1/p$ fails dramatically, while there is a more refined example that combines the above assumption with the needed hypothesis that $p$ be irregular; the very first heuristics I mentioned, about equidistribution of the $j$-th Bernoulli number $B_j\pmod{p}$, implies that possible exceptions to Vandiver's up to $p\leq x$ go like $1/2\log\log(x)$, and is therefore of no significance that we haven't found any counterexample so far. The very same section of Washington's book has a final remark on Vandiver based on the equidistribution assumption that $p\mid h^+$ with probability $1/p$ (where $h^+$ is the class number of the totally real field $\mathbb{Q}(\zeta_p+\zeta_p^{-1})$). In this case again, Vandiver could be false for a prime $p\leq x$ with probability $$ \log\log(x) $$ which is bigger than before but roughly analogous. In both cases, since computations (at the time of Washingtons' writing, namely 1996) can go up to $p\sim 4.000.000$ and $\log\log(4.000.000)=2.72...$, that no exception has been found is "not too strange".

I should perhaps add that it seems that P. Mihailescu has some argument for turning Washington's heuristic "in favor" of Vandiver here.

  • $\begingroup$ Thank you very much for your response. While Bernoulli numbers have a heuristic distribution which is confirmed by intensive computations, in our case numeric evidence is scarce. $\endgroup$ – Michael Th. Rassias Apr 18 '15 at 16:25

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.