Bernoulli-Goss polynomial and class number of cyclotomic function fields (Q920149)

Let \(\zeta_ p\) be a primitive \(p\)-th root of unity and let \({\mathbb{Q}}(\zeta_ p)\) be the associated cyclotomic field. There is a basic, and well-known, theorem of number theory -- due to Kummer -- which presents a relationship between the class number, \(h_ p\), of \({\mathbb{Q}}(\zeta_ p)\) and the \(p\)-divisibility of Bernoulli-numbers \(\{B_ i\}\). One says that the prime \(p\) is regular if \(p\nmid h_ p\); which is then, by the above theorem, translated into the statement \(p\nmid B_ i\) for \(i=2,4,...,p-3\). As is also well-known, the above theorem allows us, conversely, to prove the existence of an infinite number of irregular (i.e., \(p| h_ p)\) primes. Let now \(A=F_ q[T]\) and \(k=F_ q(T)\), \(q=p^ n\). Let \(f\) be a monic prime of \(A\). Through the use of Drinfeld modules, in general, and the Carlitz module, in particular, one is able to associate to \(f\) a ``cyclotomic'' abelian extension \(k(f)\) of \(k\). One can then hope to study the \(p\)-divisibility of its class number, \(h_ f\), in terms of ``Bernoulli-elements'' in \(k\). Indeed, one way to define Bernoulli-numbers classically is through the use of special-values of \(L\)-series and it is known that this method also works for \(A\); thus one produces elements \(\beta(i)\in k\) (in fact, in this case, \(\in A\) itself). The prime \(f\), then, is said to be regular if \(p\nmid h_ f\) which is transformed, in the classical fashion, to a statement about the \(f\)-divisibility of \(\{\beta(i)\}\). In classical theory, one denotes by \(h^+_ p\) the class number of \({\mathbb{Q}}(\zeta_ p+\zeta_ p^{-1})\). One then shows that \(h^-_ p=h_ p/h^+_ p\) is integral. There is, moreover, the basic sequence of implications (also due to Kummer): \[ p| h^+_ p\quad \Rightarrow \quad p| h^-_ p\quad \Leftrightarrow \quad p| h_ p. \] Thus questions about regularity are reduced to questions about \(h^-_ p\) and, in fact, Kummer's theorem on Bernoulli-numbers and class numbers is really a statement about \(h^-_ p.\) In the function field case, one is able in a straightforward fashion to define both \(h_ p\) and \(h^+_ p\). However, there is no obvious analog of Kummer's implications (but see below) and so one needs to handle both \(h^-_ p\) and \(h^+_ p\). Thus, following Feng, we say that \(f\) is irregular of the first (second) class if \(p| h^-_ p\) \((p| h^+_ p)\). The present author, together with M. Rosen, has shown the existence of any infinite number of first class irregular primes when \(q>2\) and the first author (of the paper being reviewed) has shown the (harder!) existence of an infinite number of second class irregular primes in general. In the present paper this is strengthened to the following Theorem. Let \(q>2\). Then there exists an infinite number of irreducible \(f\) in \(A\) with \(p| h^+_ p\) and \(p^{q-2}| h^-_ p\). In particular, for \(q>2\), there exist infinitely many \(f\) which are irregular of both the first and second classes. The authors also present a list of all regular quadratic primes for small \(q\) generalizing earlier work of Ireland and Small [see also \textit{E.-U. Gekeler}, J. Number Theory 34, 114--127 (1990; Zbl 0695.12008)]. Finally one can ask what, if anything, places the role of the classical conjecture of Kummer-Vandiver for \({\mathbb{Q}}(\zeta_ p)\) which states \(p\nmid h^+_ p?\) (Bernoulli numbers only apply to \(h^-_ p\) and give no information on the p-adic properties of \(h^+_ p\); the conjecture ``explains'' this by requiring \(h^+_ p\) to be a \(p\)-adic unit.) For \(A\), as has been known for some time and implied by the table on page 1 of the paper, the naive version of Kummer-Vandiver (i.e., \(p\nmid h^+_ f)\) is indeed false. However, by using Iwasawa theory one can proceed as follows: In Iwasawa theory, the Kummer-Vandiver conjecture is interpreted, via the Reflection Theorem, (``Spiegelungssatz'') to imply the cyclicity of certain Iwasawa modules constructed from \(p\)-torsion of class groups. Thus for \(A\) one can look for conjectures that will also imply cyclicity in the analogous situation. One such is given in [the reviewer, Duke Math. J. 56, No. 1, 163--191 (1988; Zbl 0661.12006)]. This conjecture is a statement not for all even integers, but rather for the subclass of integers called ``magic numbers'' \(=\) the class of all integers of the form \(cq^ j+(q^ j-1)\), \(1\leq c<q\). In fact, for \(c=q- 1\), the conjecture can be verified using an old formula of Carlitz. The form of the Spiegelungssatz can be thought of as \(p\)-adic manifestation of the functional equation of the Riemann \(\zeta\)-function. For \(A\) one has the remarkable result of \textit{D. S. Thakur} [J. Number Theory 35, No. 1, 1--17 (1990; Zbl 0703.11065)] which gives a new interpretation of magic numbers in terms of the divided power series associated to \(\zeta_ A(s)\); however, much is still unclear.