# Statistics of Number fields

A goal which I have been pursuing is to understand how number fields are distributed with respects to their invariants. To be more precise I was captivated by the following question: Let $N(X,n,G)$ be the number of number fields of dimension $n$ where $G$ is the Galois group of its Galois closure, and their discriminants is bounded by $|X|$ up to isomorphism. This question might be natural to ask: For which group $G\leq S_n$ one might get a positive proportion of number fields when $X\to \infty$.

From Class Field Theory or one can show it more elementary, using Delone-Faddeev correspondence, for $n=3$, $C_3$ has a density 0. And also when $G$ is an abelian group then the answer would be same as $C_3$ by using class field theory. Prof. Manjul Bhargava proved for $n=4$ , $D_4$ has a positive density, and also he showed for $n=5$ no other groups, except $S_5$, could have positive density.

I think, once Manjul told me, one might expect for $n=p$, $p$ is a prime number, the only group which can contribute with positive density is $S_p$. But I could not even find a heuristic that why this should be true.

Is there a heuristic or even a theorem which can support the above expectation? Or more generally what do we know about $N(X,n,G)$?

• How does $G$ enter into your definition? Is it the Galois group? – Qiaochu Yuan Jul 7 '11 at 20:20
• @Yuan: I fixed it, I guess it is OK, now. – M.B Jul 7 '11 at 20:29

There are general conjectures and heuristics; see for example a recent survey by Ellenberg and Venkatesh on this and the function field analog.

Roughly, there is a (modified) conjecture of Malle asserting that given $G$ the number of respective extensions of discriminant at most $X$ is asymptotically $$c x^{1/a} (\log x)^b$$ with a specific dependence of $a,b$ on $G$; and there are some heuristics of Bhargava for $c$. The 'modified' is due to the fact that nowadays conjecture is not exactly the original version of Malle (Klüners gave a counter example to the original some years ago).

This circle of ideas is also linked to the Cohen--Lenstra heuristics on class groups; this is also discussed in this survey.

• The idea which I was trying to convince myself was the following: I said for an irreducible polynomial of degree $p$, since its Galois group is a transitive group on its $p$ roots so its Galois group has a $p$-cycle, so to be $S_p$ it would be enough that its Galois group has a transposition. Then I was trying to find an probabilistic method. But I guess it is unknown how irreducible polynomial distribute with respect to discriminant. I mean we don't have (as far as I know) a version of Van der Waerden's theorem if we want to order polynomials with respects to discriminant. – M.B Jul 7 '11 at 21:46
• Sorry, I do not completely understand your question. But, I should also add that I am not a the right person to comment on this in detail as my knowledge of the subject is superficial. – user9072 Jul 8 '11 at 0:30

A conjecture of Malle (which is known to be false in general, but probably still provides a good heuristic in many cases) implies that if $G$ is a transitive subgroup of $S_n$, then a positive proportion of number fields (ordered by discriminant) of degree $n$ have Galois group $G$ if and only if $G$ contains a transposition. The relevant constant here is $a$ from unknown (google)'s answer: it's the minimum value of $n$- the number of cycles in $g$, where $g$ runs over all non-identity elements of $G$. For example, when $n=4$, the dihedral group $D_4$ contains a transposition, which explains why a positive proportion (roughly 11%) of quartic fields have Galois group $D_4$.

See this paper by Jürgen Klüners for the precise statement of the conjecture, together with a counterexample.

• I'm a fan of any paper with a good conjecture as well as a counterexample to the conjecture. – Frank Thorne Jul 8 '11 at 0:00
• Frank Thorne, I am not sure that you wanted to imply this, but he did not give a counterexample to his own conjecture (formulate in the same paper); the conjecture was already formulated/published before by somebody else. – user9072 Jul 8 '11 at 0:40
• Yeah, yeah, I know, I'm just having fun... – Frank Thorne Jul 8 '11 at 2:46

First of all, the most general conjecture about $N(X,n,G)$ is indeed due to Malle, and there are indeed counterexamples to the original formulation due to Kluners. Seyfi Turkelli has a very nice paper which explains "why" those counterexamples arise by means of a function field analogy, and offers a revised version of Malle's conjecture which seems to me pretty solid.

Why does only $S_p$ give positive density? Because under Malle's conjecture, $N(G,n,X)$ will be asymptotic to $X^{a(G)} \times$(some power of $\log$), where $a(G)$ is at most 1, with equality only when $G$ contains a transposition. So the class of groups expected to have positive density is precisely those containing a transposition. When $p$ is prime, the only transitive subgroup of $S_p$ containing a transposition is $S_p$ itself, so you're done. On the other hand, both $D_4$ and $S_4$ are transitive subgroups of $S_4$ containing a transposition -- indeed, they are the only such, so they're the only two Galois groups that are supposed to arise for a positive density of quartic fields ordered by discriminant.