scientific article; zbMATH DE number 1296376

Carlitz-Hayes modules characteristic \(p\) \(L\)-series characteristic \(p\) Dirichlet series Drinfeld modules functional equation Galois packets generalized Riemann hypothesis generalized simplicity conjecture JFM 08.0231.03 local Riemann hypothesis

Mathematics Subject Classification ID

Drinfel'd modules; higher-dimensional motives, etc. (11G09) Zeta and (L)-functions in characteristic (p) (11M38)

Review text

The prototype of all Dirichlet series is Riemann's zeta function defined by \[ \zeta(s):= \sum_{n=1}^\infty n^{-s}= \prod_{p\text{ prime}} (1-p^{-s})^{-1}. \] It is easy to see that this function begins life as an analytic function for those \(s\in \mathbb{C}\) with \(\text{Re }s> 1\). In his fabulous paper ``Ueber die Anzahl der Primzahlen unter einer gegebenen Grösse (``On the number of primes less than a given magnitude, Gesammelte Werke, Teubner, Leipzig (1876; JFM 08.0231.03)), \textit{B. Riemann} showed that his function can be continued to all complex numbers with a simple pole of residue 1 at \(s=1\). Moreover, define \(\xi(s):= \Pi(s/2) (s-1) \pi^{-s/2} \zeta(s)\) and \(\Xi(t):= \xi(1/2+ it)\) (where \(\Pi(s):= \Gamma(s+1))\), which are both readily seen to be entire; then Riemann shows that \[ \xi(s)= \xi(1-s), \] so that \(\Xi(t)= \Xi(-t)\). This is the famous functional equation for \(\zeta(s)\). Riemann also established that \(\xi(s)= \xi(0) \prod_\rho (1-s/\rho)\) where \(\rho\) runs over the zeros of \(\xi(s)\). By taking the logarithm and comparing the two products for \(\xi(s)\), Riemann obtained a formula for the arithmetic function \(\pi(x)\) (= the number of primes \(\leq x\)) in terms of \(\text{Li} (x)\) (where \(\text{Li}'(x)= 1/\log x\)). The famous ``Prime Number Theorem, states that \(\pi(x)\sim \text{Li}(x)\). The ``best result in this direction states that \(\pi(x)= \text{Li} (x)+ O(x^{1/2}\log x)\) which is equivalent to the, still unknown, ``Riemann Hypothesis (RH) that all the zeros of \(\xi(s)\) have real part equal to \(1/2\) (or equivalently, as Riemann actually states, that all the zeros of \(\Xi(t)\in \mathbb{R}\)). All known zeros of \(\Xi(t)\) satisfy the RH and, in fact, are also simple. Stimulated by \(\zeta(s)\), number theorists found a huge collection of similar Dirichlet series; so one has zeta-functions of number fields, \(L\)-functions of characters, \(L\)-functions of automorphic forms and so on. All of these Dirichlet series begin with Euler products (as did \(\zeta(s)\) itself) and then are shown to have analytic continuations and functional equations. The Generalized Riemann Hypothesis (GRH) is also expected to hold for these functions (which can also be easily expressed using the associated \(\Xi\)-function). The analog of the simplicity of zeros is more subtle since ``nonsimple over \(\mathbb{Q}\) \(L\)-series may acquire zeros of high multiplicity for purely formal reasons. Thus one is immediately forced into dealing with irreducible complex characters \(\chi\) associated to a finite Galois extension of \(\mathbb{Q}\). Then the following conjecture, which we state in terms of \(\Xi\)-functions, was told to the present author by J.-P. Serre. Conjecture 1: 1. The zeros of \(\Xi(\chi,t)\) should be simple. 2. \(t=0\) should be a zero of \(\Xi(\chi,t)\) only if \(\chi\) is real and the functional equation of \(\Xi(\chi,t)\) has a minus sign. 3. If \(\chi\) and \(\chi'\) are distinct, then the zeros of \(\Xi(\chi,t)\) not equal to \(t=0\) should be distinct from those of \(\Xi(\chi',t)\). We call the above conjecture the Generalized Simplicity Conjecture (GSC). Beginning with E. Artin and then with H. Hasse and A. Weil, mathematicians discovered another very fertile way to construct zeta functions via the analogy between number fields and function fields in one variable over a finite field. These new Dirichlet series also begin life as an Euler product and are then shown to analytically continue to all of \(\mathbb{C}\) (again with possible poles). Functional equations are established and, now, the analog of the Generalized Riemann Hypothesis was established by H. Hasse and A. Weil. The zeta-functions associated to function fields over \(\mathbb{F}_p\) are actually rational functions in \(p^{-s}\), and so are intrinsically more algebraic than their number field cousins. Thus, for instance, the analog of Euler's formula for \(\zeta(2n)\), \(n\) a positive integer, is lacking for these functions. And, as is well-known, their functional equation, GRH, etc., can all be given algebraic (or algebro-geometric) proofs. The integral spectrum \(\text{Spec} (\mathbb{Z})\) occupies a special place in arithmetic geometry as the final object (obviously the integers \(\mathbb{Z}\) map into any commutative ring). There is, however, a way that one can attempt to escape the integers' attraction. Let \({\mathbf k}\) be a function field over a finite field \(\mathbb{F}_r\) and pick a place \(\infty\) of \({\mathbf k}\) and let \({\mathbf A}\) be the ring of elements of \({\mathbf k}\) which have no poles outside of \(\infty\). As is well-known, \({\mathbf A}\) is a Dedekind domain with a finite class group and unit group \(\mathbb{F}_r^*\). One can by flat simply declare \({\mathbf A}\) to be the base ring (the ``integers for the theory) and develop an arithmetic on schemes and analysis over \({\mathbf A}\). Put \({\mathbf K}= {\mathbf k}_\infty\); so \({\mathbf A}\) is contained in \({\mathbf K}\) discretely and \({\mathbf K}/{\mathbf A}\) is compact, exactly like \(\mathbb{Z}\subset \mathbb{R}\). With a little work, one is then able to develop exponential functions as entire power series over \({\mathbf K}\) (and its finite field extensions). Following L. Carlitz and V. G. Drinfeld, one finds that these exponential functions immediately lead to arithmetic objects called ``Drinfeld modules. Drinfeld modules are highly analogous to elliptic curves and many generalizations of them have also been created (such as shtuka, \(t\)-modules, \(\tau\)-sheaves etc.). (It should be pointed out that the rings of integers in complex quadratic fields also have a unique place at infinity etc. However, using these integer rings as a base ring for an arithmetic does not seem to be very fertile.) The analogy with elliptic curves leads immediately into the study of Drinfeld modules etc., over finite fields where there are Tate modules and Frobenius endomorphisms as one would expect. These Frobenius operators have characteristic polynomials with \({\mathbf A}\)-coefficients and are invariant of the prime used for the Tate construction. Similarly, as Drinfeld showed, the zeros of these polynomials satisfy the Local Riemann Hypothesis in terms of their absolute values as elements of the algebraic closure of \({\mathbf K}\). Thus one is motivated to encode all of this information in characteristic \(p\) \(L\)-series. These are the zeta and \(L\)-functions discussed by Y. Hellegouarch in this Séminaire Bourbaki report. The use of analysis over \({\mathbf K}\) is crucial as, for instance, it allows one to show the analog of Euler's formula for the zeta function where \(\zeta_A(s):= \sum n^{-s}\), where \(n\) runs over all monic polynomials and \(s=(x,y)\in S_\infty:= {\mathbf C}_\infty^*\times \mathbb{Z}_p\) with \({\mathbf C}_\infty\) the completion of the algebraic closure of \({\mathbf K}\). The exact definition of \(n^{-s}\) is not important here, but what is crucial is that \(\zeta_A(s)\) can be analytically continued to a 1-parameter family of entire power series on \(\mathbb{C}_\infty^*\). So, for instance, it makes sense to talk about their zeros. The present author has been studying such characteristic \(p\) Dirichlet series for more than two decades (as of this writing). While very much yet remains quite mysterious, it is clear that the classical complex \(L\)-series of number fields and these characteristic \(p\) functions have many striking similarities in terms of their special values, interpolations at finite primes, etc. Due to the very recent work of \textit{G. Boeckle} and \textit{R. Pink} [A cohomological theory of crystals over function fields, in preparation], the characteristic \(p\) \(L\)-series can now be analytically continued in great generality (as above for \(\zeta_A(s)\)). It is also natural to look for a functional equation along classical lines for these \(L\)-series; however it quickly became evident in the theory that such an invariance did not seem to exist. (However, one can indeed find 1-variable (i.e., with domain \(\mathbb{Z}_p\)) characteristic \(p\) \(L\)-series attached to modular forms on the Drinfeld upper half plane which indeed have classical style functional equations [see \textit{D. Goss}, ``Some integrals attached to modular forms in the theory of function fields, in: The arithmetic of function fields, Ohio State Univ. Math. Res. Inst. Publ. 2, 227-251 (1992; Zbl 0805.11047)]. Recently the present author found, what appears to be, the correct analog of the GRH and the GSC for these functions. This GRH was motivated by the fundamental calculation of the Newton polygons associated to \(\zeta_A(s)\) due to D. Wan, D. Thakur, J. Diaz-Vargas, B. Poonen, and J. Sheats, who found that each side of the Newton polygon always projects to a segment on the \(x\)-axis of unit length; thus the zeros of \(\zeta_A (s)\) are both simple and in \({\mathbf K}\) (the analog of \(\mathbb{R}\)!). This certainly suggests a classical style Riemann Hypothesis of ``lying on a line. However, the correct conjecture actually is a statement about the absolute values of zeros, just as are the local RH's. Let \(L(s)= L(x,y)\), \(s=(x, y)\in S_\infty\), be a characteristic \(p\) valued \(L\)-series coming from arithmetic (such as from a Drinfeld module) etc. Conjecture 2: For each \(y\in \mathbb{Z}_p\), all but finitely many zeros of \(L(s)= L(x,y)\) are uniquely determined by their absolute values. In particular, Conjecture 2 implies that such zeros must be totally inseparable over any field containing the coefficients of the \(L\)-series. To see that Conjecture 2 is the correct ``characteristic \(p\) GRH, let \(L(s)\), \(s\in\mathbb{C}\), now be a classical \(L\)-series of a number field and let \(\Xi(t)\) now denote the associated \(\Xi\)-function (so that the zeroes of \(\Xi(t)\) should be real). Keeping the functional equation in mind leads to the next conjecture. Conjecture 3: All zeros of \(\Xi(t)\) are uniquely determined by their absolute values up to multiplication by \(\pm 1\). It is a very amusing exercise to see that Conjecture 3 implies the classical GRH up to the real critical zeros of \(L(s)\) not \(= 1/2\). As the functional equation implies that there are only finitely such zeros, we see the analogy between Conjectures 3 and 2. Notice however that one obtains a vastly different statement than Conjecture 2 if one just requires the zeros to lie in \({\mathbf K}\) (and which is false in general for all zeros in any case). Due to the huge size of the algebraic closure \({\mathbf K}^{ac}\) over \({\mathbf K}\), notice also that we cannot replace Conjecture 2 with any weaker statement (such as determination up to \(\pm 1\)) and still obtain a conjecture with any consequences; thus, perversely, a functional equation of classical type in the characteristic \(p\) theory would lead to a GRH which lacks any punch. Let \(\chi\) and \(\chi'\) be two \({\mathbf C}_\infty\)-valued characters of \(\text{Gal} ({\mathbf k}^{ac}/{\mathbf k})\) (which factor through a finite Galois extension of \({\mathbf k}\)). We say that \(\chi\) and \(\chi'\) are in the same Galois packet if and only if they differ by the action of some \(\sigma\in \Aut (\mathbb{C}_\infty/{\mathbf K})\) (so classically two characters are in the same packet iff they are complex conjugate of each other). By using the classical GRH and functional equation one can readily restate the number field GSC in terms of absolute values and Galois packets. This leads to the following function field analog. Conjecture 4: 1. For fixed \(y\in \mathbb{Z}_p\) almost all zeros of \(L(\chi,x,y)\) are simple. 2. Let \(\chi\) and \(\chi'\) lie in different Galois packets. Then for fixed \(y\in\mathbb{Z}_p\) almost all zeros of \(L(\chi,x,y)\) and \(L(\chi',x,y)\) have distinct absolute values. In fact, it using Conjecture 2, one sees easily that if \(\chi\) and \(\chi'\) are in the same Galois packet, then for fixed \(y\), and with possibly finitely many exceptions, both \(L\)-series have the same set of zeros. One finds that that the classical and characteristic \(p\) ``absolute value conjectures behave remarkably alike. Moreover, because of the existence of inseparable extensions, the characteristic \(p\) theory seems to give insight into the classical one. Indeed, only by playing off both the characteristic \(p\) GRH and the GSC can we finally conclude that almost all zeros of \(L(\chi,x,y)\) lie on a line; obviously one doesn't need the GSC classically, On the other hand, such a simple object as a logarithm from multiplication to addition does not exist in characteristic \(p\). And so, as of this writing, the ``meaning of the characteristic \(p\) GRH, etc., remains a big mystery. However, there are now good targets to aim for and good reasons for optimism. For the entire collection see [Zbl 0911.00019].

Recommendations

Cited in

(16)

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4248255)