18.
782 Introduction to Arithmetic Geometry Fall 2020
Lecture #37 12/7/2020
Deligne’s original proof of the Riemann hypothesis for arbitrary varieties over Fq is very
difficult. Expositions [2, §6-13; 3, §28-33; 7, §6] all require substantial background and
investment of time. A simplification was given by Katz in 2015 [7, §5], which proves the
Riemann hypothesis for hypersurfaces of arbitrary dimension, then uses a result of Scholl
to deduce the result for general varieties.
Other proofs exist for curves. In this case Weil gave two proofs in his original 1949
paper: a proof using the positivity of the Rosati involution on the Jacobian [4, Thm 1.30]
and a geometric proof using Riemann-Roch and intersections on C × C [5, §3.2]. These
are much more approachable, but still involve a lot of background from algebraic geometry:
endomorphism rings, polarizations and the Weil pairing for abelian varieties in the first
case, and Serre duality for line bundles, the Hodge index theorem and intersection theory
on surfaces for the second. There is a more elementary proof due to Bombieri and Stepanov
[1, 8], but the argument is more intricate.
In this lecture we give a sketch of Katz’ argument, following [7, §5]. We also discuss
the case of curves and the relationship between the Hasse-Weil bound and the Riemann
hypothesis.
37.1 The Riemann hypothesis for hypersurfaces over Fq
Katz’ idea is that we can prove the Riemann hypothesis by deforming from a hypersurface
where it is easy to prove. For any hypersurface with equation G = 0, there is another
hypersurface of the same dimension d and degree n where one can use classical Gauss
sums to compute the number of points (and thus the zeta function). In particular, when
gcd(n, q) = 1 we can take the Fermat hypersurface
d
X
G0 = ai xni = 0
i=0
for any choice of nonzero ai ; when gcd(n, q) 6= 1 we can take
n−1
X
G0 = xn0 + xi xn−1
i+1 = 0.
i=0
We then use the following lemma.
Lemma 37.1. Let U ⊆ P1 be a nonempty open subscheme of the projective line over Fq
and let F be a locally constant sheaf 1 of finite-dimensional Q` -vector spaces on U . Suppose
that, for every closed point x, the characteristic polynomial of Frobenius
Px = det(1 − Fxdeg(x) T |Fx )
has real coefficients and that there is some point x0 ∈ U so that all roots of Px0 have absolute
value 1. Then the same holds for all closed points of U .
The lemma requires a sheaf where the characteristic polynomial of Frobenius has roots
on the unit circle, but the Weil conjectures require roots of absolute value2 q d/2 ; where
A locally constant sheaf is an étale sheaf F on U so that there is some étale cover f : U 0 → U so that
1
∗
f F is constant. The classic example is the sheaf µn on a point Spec(Q): the value µn (K) varies over field
extensions K/k, but once K contains the cyclotomic field Q(ζn ) then µn (K) becomes constant
2
the middle dimension d is the only hard case by the weak Lefschetz theorem
1
does the difference come from? We can rescale using a technique known as a Tate twist
[6, §7.5.4], which is useful since any tensor power of a sheaf with roots on the unit circle
also has roots on the unit circle.
We can then apply the lemma to the dimension d + 1 hypersurface defined by the
equation tG + (1 − t)G0 = 0. In particular we will take U to be the complement of the
singular fibers. One can check that the hypotheses of the lemma are satisfied using the fact
i (X, Q ) are
that the zeta function is rational and that the characteristic polynomials on Hét `
rational for every i 6= d.
37.2 The Hasse-Weil bound
For a smooth projective curve C of genus g over Fq it is an easy consequence of the Weil
conjectures that
√
|1 + q − #C(Fq )| ≤ 2g q. (37.1)
But it turns out that the converse holds as well: the Riemann hypothesis for curves follows
from this bound.
Proposition 37.2. The Hasse-Weil bound (37.1) for a genus g curve C over Fq implies
the Riemann hypothesis for C.
Proof. Note first that, by the functional equation, the roots αj := α1,j are fixed as a set
by the map x 7→ q/x. Thus it is enough to prove only one direction of inequality, that
√
αj ≤ q.
To derive this inequality from (37.1), set am = 1 + q m − #C(Fqm ) = 2g m
P
j=1 αj . Consider
the generating function
2g X 2g
X
m
X X αj T
am T = αjm T m = .
1 − αj T
m≥1 j=1 m≥1 j=1
As a power series around 0, this function has a pole at t = 1/αj . We now show that it
converges for any t ∈ C with |t| < q −1/2 and thus q −1/2 ≤ 1/|αj |.
Given (37.1) we have |am | ≤ 2gq m/2 for all m by base change to Fqm . For |t| < q −1/2 ,
we have √
X X √ 2g q|t|
| am tm | ≤ 2g ( q|t|)m = √ .
1 − q|t|
m≥1 m≥1
The hard work for curves then goes into proving the Hasse-Weil bound. We close by
noting that the proof is much simpler for elliptic curves, where it follows from the following
result about the degree of isogenies of a special form.
Lemma 37.3. Suppose E is an elliptic curve over Fq , F is the Frobenius map of degree q
on E and r, s ∈ Z. Set a = 1 + q − #E(Fq ). Then
deg(rF − s) = r2 q + s2 − rsa.
Proof. See [9, §4.2]
2
Proof of Hasse bound for elliptic curves. Degrees are nonnegative, so dividing the expres-
sion in the lemma by s2 we get
r 2 r
q −a + 1 ≥ 0.
s s
for r, s ∈ Z. Since Q is dense in R, we get that the discriminant a2 −4q must be nonpositive,
√
implying |a| ≤ 2 q.
References
[1] Raymond von Bommel. The Bombieri-Stepanov approach to the Riemann hypothesis for curves over
finite fields [Link]
[2] Uwe Jannsen. Deligne’s Proof of the Weil Conjectures. [Link]
de/Jannsen/home/[Link]
[3] James Milne. Lectures on étale cohomology. [Link]
pdf
[4] James Milne. The Riemann hypothesis over finite fields: from Weil to the present day. In “The legacy of
Bernhard Riemann after one hundred and fifty years,” Lizhen Ji, Frans Oort and Shing-Tung Yau (ed.),
International Press, 2016. arXiv:1509.00797
[5] Mircea Mustaţă. Zeta functions in algebraic geometry. [Link]
˜mmustata/zeta_book.pdf
[6] Bjorn Poonen. Rational points on varieties. Graduate Studies in Mathematics 186. Amer. Math. Soc.,
Providence, 2017. [Link]
[7] Tamás Szamuely. A course on the Weil conjectures. [Link]
omn-view-listing?listingId=110831
[8] Terence Tao. The Bombieri-Stepanov proof of the Hasse-Weil bound. [Link]
com/2014/05/02/the-bombieri-stepanov-proof-of-the-hasse-weil-bound/
[9] Lawrence Washington. Elliptic curves, number theory and cryptography. Taylor & Francis Group, Boca
Raton FL, 2008.