Steven Sperber

Dwork cohomology, de Rham cohomology, and hypergeometric functions

In the 1960s, Dwork developed a p-adic cohomology theory of de Rham type for varieties over finite fields, based on a trace formula for the action of a Frobenius operator on certain spaces of p-analytic functions. One can consider a purely algebraic analogue of Dworks theory for varieties over a field of characteristic zero and ask what is the connection between this theory and ordinary de Rham cohomology. Katz showed that Dwork cohomology coincides with the primitive part of de Rham cohomology for smooth projective hypersurfaces, but the exact relationship for varieties of higher codimension has been an open question. In this article, we settle the case of smooth, affine, complete intersections.

Computing zeta functions of nondegenerate hypersurfaces with few monomials

Using the cohomology theory of Dwork, as developed by Adolphson and Sperber, we exhibit a deterministic algorithm to compute the zeta function of a nondegenerate hypersurface defined over a finite field. This algorithm is particularly well-suited to work with polynomials in small characteristic that have few monomials (relative to their dimension). Our method covers toric, affine, and projective hypersurfaces and also can be used to compute the L-function of an exponential sum. Let p be prime and let Fq be a finite field with q = p a elements. Let V be a variety defined over Fq, described by the vanishing of a finite set of polynomial equations with coefficients in Fq. We encode the number of points #V (Fqr ) on V over the extensions Fqr of Fq in an exponential generating series, called the zeta function of V : Z(V , T ) = exp ( ∞ ∑ r=1 #V (Fqr ) T r r ) ∈ 1 + TZ[[T ]]. The zeta function Z(V , T ) is a rational function in T , a fact first proved using p-adic methods by Dwork [16, 17]. The algorithmic problem of computing Z(V , T ) efficiently is of significant foundational interest, owing to many practical and theoretical applications (see e.g. Wan [58] for a discussion). From a modern point of view, we consider Z(V , T ) cohomologically: we build a p-adic cohomology theory that functorially associates to V certain vector spaces H over a p-adic field K, each equipped with a (semi-)linear operator Frobi, such that Z(V , T ) is given by an alternating product of the characteristic polynomials of Frobi acting on the spaces H . The theory of l-adic étale cohomology, for example, was used by Deligne to show that Z(V , T ) satisfies a Riemann hypothesis when V is smooth and projective. Parallel developments have followed in the p-adic (de Rham) framework, including the theories of Monsky-Washnitzer, crystalline, and rigid cohomology (see Kedlaya [35] for an introduction). In this paper, for a toric hypersurface V defined by a (nondegenerate) Laurent polynomial f in n variables over Fq, we employ the cohomology theory of Dwork, working with a space H (Ω) obtained as the quotient of a p-adic power series ring over K in n+ 1 variables by the subspace generated by the images of n+ 1 differential operators. Efforts to make these cohomology theories computationally effective have been extensive. Schoof’s algorithm for counting points on an elliptic curve [54] (generalized by Edixhoven and his coauthors [22] to compute coefficients of modular forms) can be viewed in this light, using the theory of mod l étale cohomology. A number of results on the p-adic side have also emerged in recent years. In early work, Wan [59] and Lauder and Wan [47] demonstrated 2000 Mathematics Subject Classification 11Y16 (primary), 11M38, 14D10, 14F30 (secondary). The second author was partially supported by the National Security Agency under Grant Number H9823009-1-0037. Page 2 of 37 STEVEN SPERBER AND JOHN VOIGHT that the p-adic methods of Dwork can be used to efficiently compute zeta functions in small (fixed) characteristic. Lauder and Wan use the Dwork trace formula and calculate the trace of Frobenius acting on a p-adic Banach space, following the original method of Dwork and working on the “chain level”. In this paper, we instead work with the extension of Dwork’s theory due to Adolphson and Sperber [3]; this point of view was also pursued computationally by Lauder and Wan in the special case of Artin-Schreier curves [48, 49]. Under the hypothesis that the Laurent polynomial f is nondegenerate (see below for the precise definition), the zeta function can be recovered from the action of Frobenius on a certain single cohomology space H(Ω). This method works with exponential sums and so extends naturally to the case of toric, affine, or projective hypersurfaces [4]. (It suffices to consider the case of hypersurfaces to compute the zeta function of any variety defined over a finite field using inclusion-exclusion or the Cayley trick.) The method of Dwork takes into account the terms that actually occur in the Laurent polynomial f ; these methods are especially well-suited when the monomial support of f is small, so that certain combinatorial aspects are simple. This condition that f have few monomials in its support, in which case we say (loosely) that f is fewnomial (a term coined by Kouchnirenko [42]), is a natural one to consider. For example, many explicit families of hypersurfaces of interest, including the well-studied (projective) Dwork family x 0 + · · ·+ x n + λx0x1 · · ·xn = 0 of Calabi-Yau hypersurfaces [18] (as well as more general monomial deformations of Fermat hypersurfaces [19]) can be written with few monomials. In cryptographic applications, the condition of fewnomialness also often arises. Finally, the running time of algorithms on fewnomial input are interesting to study from the point of view of complexity theory: see, for example, work of Bates, Bihan, and Sottile [5]. To introduce our result precisely, we now set some notation. Let V be a toric hypersurface, the closed subset of Gm defined by the vanishing of a Laurent polynomial f = ∑ ν∈Zn aνx ν ∈ Fq[x] = Fq[x1 , . . . , xn ]. We use multi-index notation, so x = x1 1 · · ·xνn n . We sometimes write Z(f, T ) = Z(V , T ). Let ∆ = ∆(f) be the Newton polytope of f , the convex hull of its support supp(f) = {ν ∈ Z : aν 6= 0} in R. For simplicity, we assume throughout that dim(∆) = n. For a face τ ⊆ ∆, let f |τ = ∑ ν∈τ aνx ν . Then we say f is (∆-)nondegenerate if for all faces τ ⊆ ∆ (including ∆ itself), the system of equations f |τ = x1 ∂f |τ ∂x1 = · · · = xn ∂f |τ ∂xn = 0 has no solution in F ×n q , where Fq is an algebraic closure of Fq. The set of ∆-nondegenerate polynomials with respect to a polytope ∆ forms an open subset in the affine space parameterizing their coefficients (aν)ν∈∆∩Zn : under mild hypothesis, such as when ∆ contains a unimodular simplex, then this subset is Zariski dense. (See Batyrev and Cox [6] as a reference for this notion as well as the work of Castryck and the second author [11] for a detailed analysis of nondegenerate curves.) We distinguish here between ∆(f) and ∆∞(f) which is the convex closure of ∆(f) ∪ {0}: for the Laurent polynomial wf in n+ 1 variables, f is ∆-nondegenerate if and only if wf is nondegenerate with respect to ∆∞(f) in the sense of Kouchnirenko [41], Adophson and Sperber [3], and others. Nondegenerate hypersurfaces are an attractive class to consider because many of their geometric properties can be deduced from the combinatorics of their Newton polytopes. Let s = #supp(f) and let U be the (n+ 1)× s-matrix with entries in Z whose columns are the vectors (1, ν) ∈ Z for ν ∈ supp(f). Let ρ be the rank of U modulo p. Let v = Vol(∆) = COMPUTING ZETA FUNCTIONS Page 3 of 37 n! vol(∆) be the normalized volume of ∆, so that a unit hypercube [0, 1] has normalized volume n! and the unit simplex σ = {(a1, . . . , an) ∈ R≥0 : ∑ i ai ≤ 1} has normalized volume 1. We say that ∆ is confined if ∆ is contained in an orthotope (box) with side lengths b1, . . . , bn with b1 · · · bn ≤ nv. We say that f is confined if ∆(f) is confined. A slight extension of a theorem of Lagarias and Ziegler [43] shows that every polytope ∆ is GLn(Z)-equivalent to a confined polytope; this existence can also be made effective. (See section 3 for more detail.) In other words, for each Laurent polynomial f there is a computable monomial change of basis of Fq[x ], giving rise to an equality of zeta functions, under which f is confined. (In the theorem below, at the expense of introducing a factor of log δ, where δ = δ(S) = maxν∈S |ν| where |ν| = maxi |νi|, one can remove the assumption that ∆ is confined.) For functions f, g : Z≥0 → R≥0, we say that f = O(g) if there exists c ∈ R>0 and N ∈ Z≥0 such that for every x = (x1, . . . , xm) ∈ Z≥N we have g(x) ≤ cf(x). (The reader is warned that not all properties familiar to big-Oh notation for functions of one variable extend to the multivariable case; see Howell [29]. In fact, our analysis also holds with Howell’s more restrictive definition, but we will not pursue this further here.) We further use the “soft-Oh” notation, where f = Õ(g) if f = O(g log g) for some k ≥ 1. Our main result is as follows. Theorem A. Let n ∈ Z≥1. Then there exists an explicit algorithm that, on input a nondegenerate Laurent polynomial f ∈ Fq[x1 , . . . , xn ] with p ≥ 3 and an integer N ≥ 1, computes as output Z(f, T ) modulo p . If further f is confined, then this algorithm uses Õ ( s + pN log q + p(6N + n)(vN log q) )Using the cohomology theory of Dwork, as developed by Adolphson and Sperber, we exhibit a deterministic algorithm to compute the zeta function of a nondegenerate hypersurface defined over a finite field. This algorithm is particularly well suited to work with polynomials in small characteristic that have few monomials (relative to their dimension). Our method covers toric, affine, and projective hypersurfaces, and also can be used to compute the L -function of an exponential sum.

On twisted de Rham cohomology

Consider the complex of differential forms on an open affine subvariety U of A N with differential where d is the usual exterior derivative and o is a fixed 1-form on U . For certain U and o, we compute the cohomology of this complex.

CONVERGENCE OF POWER SERIES SOLUTIONS OF p-ADIC NONLINEAR DIFFERENTIAL EQUATION

Publisher Summary This chapter discusses convergence of power series solutions of p-adic nonlinear differential equation. It considers Qp, the field of rational p-adic numbers, and Ω, a complete algebraically closed extension of Qp. The Ord denotes the p-adic (additive) valuation in Ω that is normalized by the condition ord p = 1. In this chapter “convergence” means “convergence in the p-adic sense.” The chapter presents the theorem that states that if every root α of the indicial polynomial μq0(m) satisfies the condition ord(m + α) = 0(log m) as m → +∞ in Z, then, for any convergent power series f(z), every formal power series solution of the differential equation L(y) = f(z) is also convergent. An element α of Ω is called a p-adically non-Liouville number if the condition discussed is satisfied. The main purpose of this chapter is to generalize this theorem to a nonlinear differential equation. It also proves two lemmas on which the method of successive approximations is based.

Zeta functions of alternate mirror Calabi–Yau families

We prove that if two Calabi–Yau invertible pencils have the same dual weights, then they share a common factor in their zeta functions. By using Dwork cohomology, we demonstrate that this common factor is related to a hypergeometric Picard–Fuchs differential equation. The factor in the zeta function is defined over the rationals and has degree at least the order of the Picard–Fuchs equation. As an application, we relate several pencils of K3 surfaces to the Dwork pencil, obtaining new cases of arithmetic mirror symmetry.

A-hypergeometric series and the Hasse–Witt matrix of a hypersurface

Abstract We give a short combinatorial proof of the generic invertibility of the Hasse–Witt matrix of a projective hypersurface. We also examine the relationship between the Hasse–Witt matrix and certain A-hypergeometric series, which is what motivated the proof.

A-hypergeometric sustems that come from geometry

In recent work, Beukers characterized A-hypergeometric systems having a full set of algebraic solutions. He accomplished this by (1) determining which A-hypergeometric systems have a full set of solutions modulo p for almost all primes p and (2) showing that these systems come from geometry. He then applied a fundamental theorem of N. Katz, which says that such systems have a full set of algebraic solutions. In this paper we establish some connections between nonresonant A-hypergeometric systems and de Rhamtype complexes, which leads to a determination of which A-hypergeometric systems come from geometry. We do not use the fact that the system is irreducible or find integral formulas for its solutions.

Exponential sums on ⁿ, II

We prove a vanishing theorem for the p-adic cohomology of exponential sums on A n . In particular, we obtain new classes of exponential sums on An that have a single nonvanishing p-adic cohomology group. The dimension of this cohomology group equals a sum of Milnor numbers.

On the L-functions associated with certain exponential sums☆

Abstract By use of p -adic analytic methods, we study the L -functions associated to certain exponential sums defined over a finite field. Estimates for the degree of this L -function as rational function are obtained. In an “asymptotic” sense, these estimates are shown to be best possible. Precise determination of the degree is computed in the one-variable case.

EXPONENTIAL SUMS ON 𝔸n. IV

We find new conditions on a polynomial over a finite field that guarantee that the exponential sum defined by the polynomial has only one nonvanishing p-adic cohomology group, hence the L-function associated to the exponential sum is a polynomial or the reciprocal of a polynomial.