Towards a General Theory of Weierstrass Points on Curves
Sami Douba Department of Mathematics & Statistics McGill University, Montr´eal June 2017
A thesis submitted to McGill University in partial fulfillment of the requirements of the degree of Masters of Science
c Sami Douba 2017 2 Abr´eg´e Nous pr´esentons plusieurs g´en´eralisationsalg´ebro-g´eom´etriquesde la notion classique d’un point de Weierstrass sur une surface de Riemann. Nous montrons qu’une large part de l’intuition acquise par la th´eorie classique est vraie en caract´eristique 0 et en grande caract´eristique.Finalement, nous calculons et examinons les points de Weierstrass de la quartique de Klein en caract´eristiques2 et 3 pour d´emontrer que ces points peuvent faire preuve d’un comportement int´eressant en petite caract´eristique.
Abstract We present various algebro-geometric generalizations of the classical notion of a Weierstrass point on a Riemann surface. We show that much of the intuition gained from the classical theory holds in characteristic 0 and in high characteristic. Finally, we compute and examine the Weierstrass points of the Klein quartic in characteristics 2 and 3 to demonstrate that these points may exhibit interesting behaviour in low characteristic.
3 4 Preface
The definitions and the proofs in Chapter 1 are modelled very closely after Hershel Farkas and Irwin Kra’s exposition of Weierstrass points in [FK80], but the discussion is slightly more general. The theory in Chapter 2 and Section 3.1 is taken entirely from the papers [Lak81], [Lak84] by Dan Laksov, and [LT94] by Laksov and Anders Thorup. Section 3.2 makes extensive use of Noam Elkies’ paper [Elk99] on the Klein quartic, but the results in that section are, to the author’s knowledge, original.
Acknowledgements I would like to express my immense gratitude to my supervisor, Prof. Eyal Goren, for his support, his guidance, and his patience. I would also like to thank Prof. Jacques Hurtubise for agreeing to examine my thesis. Last but not least, I am indebted to my fellow students, with and from whom I gained much of the background knowledge necessary to complete this thesis.
5 6 Contents
1 A na¨ıve approach to Weierstrass points 13 1.1 The gap theorems ...... 13 1.2 Weierstrass points and wronskians ...... 17 1.3 Weierstrass points on hyperelliptic curves ...... 23 1.4 Higher Weierstrass points ...... 26
2 Laksov’s Weierstrass points on curves 29 2.1 The bundle of principal parts ...... 30 2.2 Weierstrass points ...... 35 2.3 The wronskian ...... 36 2.4 Local computations ...... 40 2.5 The classical situation ...... 43
3 Weierstrass points in families 45 3.1 Weierstrass points on schemes ...... 45 3.2 Reductions of the Klein quartic ...... 50
7 8 Introduction
Since elliptic curves have non-trivial automorphisms, one cannot hope to construct a fine moduli space parametrizing isomorphism classes of these objects. However, an elliptic curve E is endowed with distinguished points — the group E[N] of its N-torsion points. We can use this group to endow an elliptic curve with additional structure, namely, a full level N structure, and then stipulate that isomorphisms between elliptic curves preserve this structure. The resulting moduli problem admits a fine moduli space for large enough N; it is the modular curve Y (N). Now what about curves of genus greater than 1? Is there a “non-abelian” analogue of the set E[N]?
In the early 1860s, Weierstrass proved his so-called L¨uckensatz, or “gap theorem”, which can be stated as follows: For any point on a compact Riemann surface of genus g, there are exactly g integers m, called “gaps”, such that there is no meromorphic function on that surface having a pole of order m at that point as its only singularity. For all but a finite number of special points, the surface’s so-called Weierstrass points, these gaps are the integers 1 ≤ m ≤ g. Furthermore, it is possible to attach weights to these points so that the sum of their weights is a constant that depends only on g. From the perspective of classifying curves of a fixed genus, this seems promising. Since compact Riemann surfaces can be thought of as algebraic curves over C, it is natural to ask if one can talk about Weierstrass points on more general algebraic curves, and if one can use these points to “rigidify” spaces of such curves and construct moduli spaces, as is done with torsion points on elliptic curves.
In the first chapter, we provide a na¨ıve theory of Weierstrass points on smooth pro- jective curves over algebraically closed fields. This approach is completely analogous to
9 10
the classical theory of Weierstrass points on compact Riemann surfaces. However, this theory breaks down when the characteristic of the ground field is low relative to the genus. Indeed, in low enough characteristic, there are examples of curves C of genus g with the property that each point on C has a non-classical gap sequence, that is, a gap sequence that differs from 1, . . . , g (see Example 2.0.1). In the second chapter, we resolve this issue by introducing Laksov’s theory of Weier- strass points on curves. Instead of defining a Weierstrass point as a point whose gap sequence is non-classical, Laksov associates to each curve a generic gap sequence, and defines a point on that curve to be Weierstrass if the gap sequence at that point differs from the generic gap sequence. Under this approach, a curve has finitely many Weier- strass points regardless of the characteristic of the ground field. However, the weighted number of Weierstrass points may vary between curves of a fixed genus (see Section 3.2). This is already an indication that Weierstrass points may not be satisfactory candidates for higher-genus analogues of torsion points on elliptic curves. In the final chapter, we expose Laksov and Thorup’s notion of Weierstrass points on families of curves over arbitrary base schemes. This generalizes the theory of the previous chapter. We end with a thorough examination of the Klein quartic. We show that none of the points obtained as reductions mod 3 of Weierstrass points of the classical Klein quartic is a Weierstrass point of the reduction in the sense of Laksov. This shows that Laksov’s Weierstrass points have another drawback: their behaviour in algebraic families appears to be subtle.
Literature Review For the following discussion, let g ≥ 2 be an integer, and denote by X a compact Riemann surface of genus g. Furthermore, let Mg be the moduli space
of Riemann surfaces of genus g, and let Mg be the Mumford-Deligne compactification of
Mg ([DM69]). The study of Weierstrass points dates back to the nineteenth century, when Weier- strass stated and proved the L¨uckensatz (discussed above). The theorem first appeared in the dissertation of Weierstrass’s student Schottky in 1875 (later published as [Sch77]). Through their work on adjoints of plane curves, Brill and Noether implicitly proved in 11
1873 that the set of Weierstrass points on X is nonempty and has size at most g3 − g
([BN74]). In 1882, Noether proved the L¨uckensatz for any sequence of points P1,P2,... on X, generalizing Weierstrass’s result ([Noe84]; see 1.1.4 for a modern — and more general — version of this theorem). He also proved the finiteness of Aut(X) using Weierstrass points ([Noe82]), but the first proof of this fact is probably due to Klein ([dC08, p. 42]).
In his landmark 1893 paper [Hur92], H¨urwitzintroduced the wronskian and the notion of “weights” to the study of Weierstrass points. He even introduced the concept of “higher” Weierstrass points, although the first time Weierstrass points were thus labelled was in a paper by Haure from 1896 ([Hau96]). Segre adopted this label when he discussed “punti di Weierstrass” in his short 1899 publication [Seg99], in which he proved that the maximal weight of a Weierstrass point on a nonhyperelliptic curve of genus g is (g − 1)(g − 2)/2 + 1.
The first half of the twentieth century is considered a period of relative dormancy in the study of Weierstrass points, but it was in this period that a paper of great relevance to this thesis was published. In his 1939 paper [Sch39], Schmidt used the wronskian method to extend the notion of Weierstrass points to curves defined over fields of positive characteristic. The gap sequences and Weierstrass points defined by Laksov in [Lak81], whose modern approach to the topic is the focus of this thesis, are the same as those given by Schmidt.
Rauch ushered in a period of renewed interest in Weierstrass points with the publica- tion of his paper [Rau59] in 1959. Ahlfors had already endowed the set of isomorphism
classes Mg of Riemann surfaces of genus g with the structure of a (3g − 3)-dimensional complex analytic space ([Ahl60]). In his 1959 paper, Rauch proved that the Riemann surfaces of genus g carrying a Weierstrass point with a prescribed first non-gap constitute
a complex-analytic subvariety of Mg. Some years later, expanding on the work of Rauch,
Arbarello examined the closure W n,g in Mg of the set of Riemann surfaces of genus g possessing a Weierstrass point x such that h0(C, nx) ≥ 2, where 2 ≤ n ≤ g (so, for
example, W 2,g is the closure of the hyperelliptic locus, and in general W n,g contains the space of Riemann surfaces of genus g possessing a Weierstrass point whose first non-gap 12
is n). Among other things, Arbarello proved that W 2,g is an irreducible analytic variety of dimension 2g + n − 3 ([Arb74]). From the 1970s onwards, questions about Weierstrass points have tended to involve the existence of Weierstrass points of a given type, or the relationship between Weierstrass points and automorphisms of Riemann surfaces (see, for example [Lax75], [Far73]). There has been relatively little effort to generalize this theory to accommodate algebraic curves over fields of arbitrary characteristic. The work of Laksov and Thorup ([Lak81], [LT94], [LT95]) is exceptional in this regard. Chapter 1
A na¨ıve approach to Weierstrass points
The theory presented in this chapter is “na¨ıve” in the sense that it is modelled very closely after the classical theory of Weierstrass points on Riemann surfaces. For the following discussion, by curve we will mean a projective non-singular algebraic curve over some fixed algebraically closed field k. We fix a curve C of genus g. Let
1 k(C) be its function field, and let OC and ΩC be its structure sheaf and its canonical sheaf, respectively. Given a divisor D on C, denote by O(D) the invertible OC -module associated to D and by H0(C,D) the space of global sections of O(D). This is a finite dimensional k-vector space whose dimension we denote by h0(C,D) or `(D). Denote by
0 1 −1 i(D) the dimension of H (C, ΩC ⊗ O(D) ), that is, i(D) = `(K − D), where K is a canonical divisor on D. For the rest of this chapter, we will assume points are closed unless otherwise stated.
1.1 The gap theorems
The discussion in this section is based on that in [FK80, pp. 78-80].
Lemma 1.1.1. Let P a point on C and D a divisor on C. Let D0 = D + [P ]. Then `(D0) − `(D) is either 0 or 1.
13 Chapter 1. A na¨ıve approach to Weierstrass points 14
Proof. Let K be a canonical divisor on C. By Riemann-Roch, we have
`(D0) − i(D0) = deg D0 − g + 1
`(D) − i(D) = deg D − g + 1
Subtracting the second equation from the first, we obtain
`(D0) − `(D) = 1 + i(D0) − i(D)
Now, since D0 ≥ D, we have H0(C,D) ⊆ H0(C,D0), so `(D0) − `(D) ≥ 0. Furthermore, since D0 ≥ D, we have i(D0) − i(D) ≤ 0. Thus, from the above equation,
0 ≤ `(D0) − `(D) ≤ 1
Remark 1.1.2. Note that this proof also shows that i(D) − i(D + [P ]) is either 0 or 1.
Definition 1.1.3. Let (P1,P2,P3,...) an infinite sequence of points on C. Define a sequence of divisors Dm on C as follows: set D0 = 0, and for m ≥ 1, define
m X Dm = [Pi] i=1
We say that an integer m is a Noether gap w.r.t. the sequence (Pi) if `(Dm)−`(Dm−1) = 0, 0 0 i.e., if H (C,Dm) = H (C,Dm−1). We say m is a Weierstrass gap at a point P on C if it is a Noether gap w.r.t. to the constant sequence P = P1 = P2 = ... = Pi = .... Note that m is a Weierstrass gap at P if and only if there does not exist a morphism ϕ : C → P1 of degree m with ϕ−1(∞) = {P }.
Theorem 1.1.4. Any sequence of points on C gives rise to precisely g Noether gaps, and they are all less than 2g.
Proof. Let (P1,P2,...) be a sequence of points on C,(D0,D1,D2,...) the associated
sequence of divisors, and let AN be the number of Noether gaps ≤ N. For every m ≥ 1, we know from the above lemma that either `(Dm) − `(Dm−1) = 0, in which case m is a 15 1.1. The gap theorems
Noether gap, or `(Dm) − `(Dm−1) = 1. Thus, for any N ≥ 1, the number of integers ≤ N that are not Noether gaps is
N X (`(Dm) − `(Dm−1)) = `(DN ) − `(D0) m=1
= N + `(K − DN ) − `(K) by Riemann-Roch
= N + `(K − DN ) − g
and so AN = N−(N+`(K−DN )−g) = g−`(K−DN ). Thus, for every N ≥ max(1, 2g−1),
we have AN = g (since deg K = 2g − 2). This shows that there are precisely g Noether gaps and they are all less than 2g.
Remark 1.1.5. One consequence of the above theorem is that any sequence of points on a curve of genus 0 gives rise to no Noether gaps. This is straightforward to verify since every such curve is isomorphic to P1. It is also worth noting that for a curve C of positive genus and any sequence of points on C, we have that 1 is a Noether gap; otherwise, we would obtain an isomorphism from C to P1, which is impossible.
Corollary 1.1.6. Given a point P on C, there are precisely g Weierstrass gaps at P , and they are all less than 2g. Furthermore, if g > 0, then 1 is a Weierstrass gap at P .
From this point on, we will refer to Weierstrass gaps simply as gaps, and to integers that are not gaps as non-gaps. We note that the non-gaps at a point P on C are closed under addition; indeed, i, j are non-gaps if and only if ∃ ϕ, ψ ∈ k(C)× whose divisors have polar parts −i[P ], −i[P ], respectively, in which case ϕψ has polar part −(i + j)[P ], and so i + j is a non-gap. From this, we can easily conclude that if n is a gap, then at least half of the positive integers strictly less than n are also gaps, for if n is a gap and m < n is a non-gap, then n − m must be a gap.
Proposition 1.1.7. Suppose g > 0, and let 1 < α1 < . . . < αg = 2g be the g smallest non-gaps at a point P on C. Define α0 := 0. Then
(i) αj + αg−j ≥ αg = 2g for 0 ≤ j ≤ g;
(ii) if α1 = 2, then αj = 2j for 0 ≤ j ≤ g; Chapter 1. A na¨ıve approach to Weierstrass points 16
(iii) αj + αg−j = 2g for 0 ≤ j ≤ g if and only if α1 = 2; Pg (iv) j=1 αj ≥ g(g + 1), with equality if and only if α1 = 2.
Proof. Clearly, α0 + αg−0 = 0 + αg = αg. Now suppose αj + αg−j < αg for some j strictly between 0 and g. Then, since non-gaps are closed under addition, α1 +αg−j, α2 +
αg−j, . . . , αj + αg−j are j non-gaps that are strictly between αg−j and αg. But there are only j − 1 non-gaps strictly between αg−j and αg, namely, αg−j+1, . . . , αg−1. This proves (i).
If α1 = 2, then, by closure of non-gaps under addition, α1 < 2α1 < . . . < gα1 are g non-gaps that are ≤ 2g, hence (ii). In this case, for 0 ≤ j ≤ g, αj +αg−j = 2j +2(g −j) = 2g.
Now suppose αj + αg−j = 2g for 0 ≤ j ≤ g . We prove by strong induction that αj is a multiple of α1 for each j, which is only possible if α1 = 2. Clearly, α0 = 0 and α1 are multiples of α1. Now suppose α0, α1, . . . , αj are all multiples of α1 for some j between 1 and g − 1. We have
αg−(j+1) = 2g − αj+1 < (2g − αj+1) + αj = 2g − (αj+1 − αj) < 2g = αg
Since (2g − αj+1) + αj = αg−(j+1) + αj is a non-gap, by the above inequality it must be one of
αg−j = 2g − αj, αg−(j−1) = 2g − αj−1, . . . , αg−1 = 2g − α1
By the induction hypothesis, this implies that αj+1 is a multiple of α1. It remains to prove (iv). We have
g g g X X X 2 αj = (αj + αg−j) ≥ 2g by (i) j=0 j=0 j=0 = 2g(g + 1)
By (iii), the above inequality is an identity if α1 = 2 and a strict inequality otherwise.
Finally, we note that α0 = 0 to obtain statement (iv) on the nose. 17 1.2. Weierstrass points and wronskians
1.2 Weierstrass points and wronskians
We are now in a position to introduce the notion of a Weierstrass point on C. Throughout this section, we assume g > 0. The discussion in this section is modelled after that in [FK80, pp. 81-86].
Definition 1.2.1. Let 1 = n1 < . . . < ng ≤ 2g − 1 be the g gaps at a point P on C. We define g X τ(P ) := (nj − j) j=1
(Note that τ(P ) ≥ 0.) We call τ(P ) the weight of the point P . If P has positive weight, we call P a classical Weierstrass point or simply a Weierstrass point.
Remark 1.2.2. Note that a curve of genus 1 has no Weierstrass points.
Proposition 1.2.3. Suppose g ≥ 2. For any point P on C, the following are equivalent: (i) P is a Weierstrass point; (ii) at least one of the integers 2, . . . , g is a non-gap; (iii) `(g[P ]) ≥ 2; (iv) i(g[P ]) > 0.
Proof. First, we recall that 1 is necessarily a gap. The equivalence between (i) and (ii) is clear. Now, (ii) holds if and only if there is a function in k(C) with a pole at P of order ≤ g, that is, if and only if there is a non-constant function f ∈ k(C) s.t. div(f) ≥ −g[P ]. This shows that (ii) and (iii) are equivalent. The equivalence of (iii) and (iv) follows directly from the Riemann-Roch Theorem.
Now fix a point P on C and let t ∈ k(C) be a uniformizer at P . Let Ωk(C)/k be the
module of relative differentials of k(C) over k and denote by d : k(C) → Ωk(C)/k the associated derivation. Note that d is simply the map on stalks at the generic point of C
1 induced by the canonical map OC → ΩC . Chapter 1. A na¨ıve approach to Weierstrass points 18
Definition 1.2.4. Given a sequence of functions ϕ1, . . . , ϕn ∈ k(C), we define
ϕ1 . . . ϕn
dϕ1 dϕn ... dt dt Wt(ϕ1, . . . , ϕn) := . . . (1.1) . .. .
dn−1ϕ dn−1ϕ 1 ... n dtn−1 dtn−1 (For any function f ∈ k(C), we write df/dt to denote the unique function h ∈ k(C) s.t. df = h · dt.) We call wt(ϕ1, . . . , ϕn) the wronskian of ϕ1, . . . , ϕn with respect to t. From now on, when the context is clear, we will denote the “derivative” df/dt of a function f ∈ k(C) with respect to t simply as f 0. Note that, from the properties of the determinant, for any f ∈ k(C) we have
n Wt(fϕ1, . . . , fϕn) = f wt(ϕ1, . . . , ϕn) (1.2)
(See [FK80, pp. 82-83] for details.) Also note that for any aij ∈ k,
Wt(a11ϕ1 + ... + an1ϕn, . . . , a1nϕ1 + ... + annϕn) = det(aij)Wt(ϕ1, . . . , ϕn) (1.3)
In particular, if ϕ1, . . . , ϕn and a11ϕ1+...+an1ϕn, . . . , a1nϕ1+...+annϕn are both linearly independent over k, then the wronskian of the latter is a non-zero constant multiple of the wronskian of the former.
× Lemma 1.2.5. Let ϕ1, . . . , ϕn ∈ k(C) . Denote ordP ϕi by µi and suppose
µ1 < µ2 < . . . < µn
Suppose further that we have char(k) - µj − µ1 for j = 1, . . . n. Then the wronskian
Φ := Wt(ϕ1, . . . , ϕn) is non-zero, and n X ordP Φ = (µj − j + 1) j=1 Proof. We proceed by induction on n. The result is clearly true for n = 1. Now assume it is true for n = k. From the above, we have
k+1 Wt(ϕ1, . . . , ϕk+1) = ϕ1 Wt(1, ϕ2/ϕ1, . . . , ϕk+1/ϕ1)
k+1 0 0 = ϕ1 Wt((ϕ2/ϕ1) ,..., (ϕk+1/ϕ1) ) 19 1.2. Weierstrass points and wronskians
where the second identity is obtained by expanding the determinant Wt(1, ϕ2/ϕ1, . . . , ϕk+1/ϕ1) along the first column. Note that ordP (ϕj/ϕ1) = µj − µ1, and so
0 < ordP (ϕ2/ϕ1) < . . . < ordP (ϕk+1/ϕk)
0 In particular, by the condition on char(k), this implies that ordP ((ϕj/ϕ1) ) = µj − µ1 − 1 and
0 0 ordP ((ϕ2/ϕ1) ) < . . . < ordP ((ϕk+1/ϕk) )
Thus, the induction hypothesis gives
k k+1 0 0 X ordP ϕ1 Wt((ϕ2/ϕ1) ,..., (ϕk+1/ϕ1) ) = (k + 1)µ1 + ((µj+1 − µ1 − 1) − j + 1) j=1 k X = µ1 + (µj+1 − (j + 1) + 1) j=1 k+1 X = (µj − j + 1) j=1
Now, following the notation in [FK80], let H(C) ⊂ Ωk(C)/k denote the space of holo- 0 1 1 morphic differentials on C, that is, the space H (C, ΩC ) of global sections of ΩC . This
is a k-vector space of dimension g. We can always construct a basis ω1, . . . , ωg for H(C) s.t.
ordP ω1 < . . . < ordP ωg
If a basis for H(C) has the above property, we say it is adapted to P . In the following lemma, we show that if we have such a basis ω1, . . . , ωg, then ordP ωj = nj − 1, where
1 = n1 < . . . < ng ≤ 2g − 1 are the g Weierstrass gaps at P .
Lemma 1.2.6. Let 1 = n1 < . . . < ng ≤ 2g − 1 be the g gaps at P . Then
{ordP ω : ω ∈ H(C)\{0}} = {n1 − 1, . . . , ng − 1}
In particular, there is a holomorphic differential ω on C that does not vanish at P (i.e., ordP ω = 0). Chapter 1. A na¨ıve approach to Weierstrass points 20
Proof. By Lemma 1.1.1, n is a gap at P if and only if i((n − 1)[P ]) − i(n[P ]) = 1, that
is, if and only if there is a differential ω ∈ H(C)\{0} s.t. ordP ω = n − 1.
Definition 1.2.7. Given a differential ω ∈ Ωk(C)/k, let ω/dt = f where f ∈ k(C) is the unique function satisfying ω = f · dt. For any sequence of differentials ω1, . . . , ωn ∈ ΩC , we define
Wt(ω1, . . . , ωn) := Wt(ω1/dt, . . . , ωn/dt)
We call Wt(ω1, . . . , ωn) the wronskian of ω1, . . . , ωn with respect to t. It is easy to see that the wronskian for differentials inherits properties 1.2 and 1.3 from the wronskian for functions.
Lemma 1.2.8. Let t, t˜ ∈ k(C) uniformizers at points P , P˜ on C, respectively. For any sequence of differentials ω1, . . . , ωn ∈ Ωk(C)/k, we have
dt˜n(n+1)/2 W (ω , . . . , ω ) = W (ω , . . . , ω ) t 1 n t˜ 1 n dt
Proof. Let ϕi = dωi/dt andϕ ˜i = dωi/dt˜ for i = 1, . . . , n. By the properties of the determinant, we have
ϕ˜1 ... ϕ˜n dϕ˜ dϕ˜ Pn j 1 n n(n+1)/2 dt˜ j=1 ... dt˜ dt˜ dt˜ Wt(ω1, . . . , ωn) = . . . = Wt˜(ω1, . . . , ωn) dt . .. . dt
dn−1ϕ˜ dn−1ϕ˜ 1 ... n dt˜n−1 dt˜n−1
Definition 1.2.9. Let C be a curve, and consider the k(C)-vector space
⊗m Ωk(C)/k = Ωk(C)/k ⊗k(C) ... ⊗k(C) Ωk(C)/k | {z } m times
1 ⊗m (Note that this is the stalk of the invertible sheaf (ΩC ) at the generic point of C.) We ⊗m call an element of this space an m-differential on C. Since Ωk(C)/k is one-dimensional 21 1.2. Weierstrass points and wronskians
over k(C), we have that every m-differential ζ on C is of the form ω1 ⊗ ... ⊗ ωm for some
ω1, . . . , ωm ∈ Ωk(C)/k. Assuming ζ is non-zero, we may define
m X ordP ζ := ordP ωi i=1 In fact, for any uniformizer t at P , there is a unique function f ∈ k(C) s.t.
f(dt)m := f dt ⊗ dt ⊗ ... ⊗ dt = ζ | {z } m times and so m X ordP ωi = ordP f i=1 Qm The function f = i=1(dωi/dt).
We associate to ζ a divisor (ζ) on C as follows:
m X X div(ζ) = (ordP ζ)[P ] = div(ωi) P ∈C i=1
Note that deg((ζ)) = m(2g − 2).
Definition 1.2.10. Let ω1, . . . , ωn ∈ Ωk(C)/k, and set m = n(n + 1)/2. We define the
wronskian of ω1, . . . , ωn to be the m-differential
m W (ω1, . . . , ωn) := Wt(ω1, . . . , ωn)(dt)
where t is a uniformizer at any point P on C.
Remark 1.2.11. Note that the wronskian is independent of the choice of point and uniformizer. Indeed if t˜ is a uniformizer at a point P˜ on C, by Lemma 1.2.8, we have
dt˜m W (ω , . . . , ω )(dt)m = W (ω , . . . , ω ) (dt)m = W (ω , . . . , ω )(dt˜)m t 1 n t˜ 1 n dt t˜ 1 n
Lemma 1.2.12. Suppose that char(k) = 0 or char(k) > 2g − 2. Then for any basis
ω1, . . . , ωg of H(C), the wronskian W (ω1, . . . , ωg) is non-zero, and for any point P on C, we have
ordP (W (ω1, . . . , ωg)) = τ(P ) Chapter 1. A na¨ıve approach to Weierstrass points 22
Proof. Let 1 = n1 < . . . < ng ≤ 2g − 1 be the g gaps at P , and let t be any uniformizer
at P . By property (2), we may assume that the basis ω1, . . . , ωg is adapted to P , i.e., that for j = 1, . . . , g, we have
ordP (ωj/dt) = nj − 1
Since nj −n1 = nj −1 ≤ ng −1 ≤ (2g−1)−1 = 2g−2 for j = 1, . . . , g and char(k) > 2g−2, we can apply Lemma 1.2.5 to obtain ordP (W (ω1, . . . , ωg)) = ordP (Wt(ω1, . . . , ωg)) = ordP (Wt(ω1/dt, . . . , ωg/dt)) g X = (ordP (ωj/dt) − j + 1) j=1 g g X X = ((nj − 1) − j + 1) = (nj − j) = τ(P ) j=1 j=1
Proposition 1.2.13. Suppose char(k) = 0 or char(k) > 2g − 2. Then X τ(P ) = (g − 1)g(g + 1) = g3 − g P ∈C In particular, C has at most g3 − g Weierstrass points.
Proof. For each point P on C, let 1 = nP,1 < . . . < nP,g ≤ 2g − 1 be the g gaps at P ,
let ω1, . . . , ωg be any basis for H(C), and let m = g(g + 1)/2. Since W (ω1, . . . , ωg) is an m-differential, we have
3 deg((W (ω1, . . . , ωg))) = m(2g − 2) = g − g
On the other hand, by Lemma 1.2.12, g X X X X deg((W (ω1, . . . , ωg))) = ordP (W (ω1, . . . , ωg)) = (nP,j − j) = τ(P ) P ∈C P ∈C j=1 P ∈C
Theorem 1.2.14. For any point P on C, g(g − 1) τ(P ) ≤ 2 with equality if and only if the smallest non-gap at P is 2. 23 1.3. Weierstrass points on hyperelliptic curves
Proof. Let 1 < α1, . . . , αg = 2g be the g smallest non-gaps and 1 = n1 < . . . < ng ≤ 2g−1 the g gaps at P . Then we have g g g 2g g ! g X X X X X X τ(P ) = (nj − j) = nj − j = j − αj − j j=1 j=1 j=1 j=1 j=1 j=1 2g g X X = j − αj j=g+1 j=1 g g(3g + 1) X = − α 2 j j=1 g(3g + 1) g(g − 1) ≤ − g(g + 1) = 2 2 The inequality follows from Prop. 1.1.7 (iv), and so does the fact that equality holds if
and only if α1 = 2.
Corollary 1.2.15. If g ≥ 2, and char(k) = 0 or char(k) > 2g − 2, then C has at least 2g + 2 and at most g3 − g Weierstrass points.
1.3 Weierstrass points on hyperelliptic curves
In this section, we discuss the situation when C is hyperelliptic, that is, when we have a P1 morphism ϕ : C → k of degree 2. The discussion here is based on [FK80, pp. 99-102] and [SS15, Prop. 11,12]. Throughout the section, we assume g ≥ 2, and that char(k) = 0 or char(k) > 2g − 2.
Proposition 1.3.1. The following are equivalent: (i) C is hyperelliptic. (ii) C has precisely 2g + 2 Weierstrass points. (iii) At every Weierstrass point on C, the gaps are 1, 3,..., 2g − 1. (iv) At some Weierstrass point on C, the gaps are 1, 3,..., 2g − 1. Furthermore, in this case, the ramification points of any double cover C → P1 are precisely the Weierstrass points of C.
Proof. Suppose that C is hyperelliptic, i.e., that there is a map ϕ : C → P1 of degree 2. Since, under the given conditions, char(k) does not divide 2, we can apply Hurwitz’ genus Chapter 1. A na¨ıve approach to Weierstrass points 24
formula (see, for example, [Har77, IV Cor. 2.4]) to obtain that the map ϕ is ramified over precisely 2g + 2 points on C, each with ramification index 2. Take any such point P . Let ψ : P1 → P1 be an automorphism mapping ϕ(P ) to ∞. Thenϕ ˜ = ψ ◦ ϕ is a morphism of degree 2 withϕ ˜−1(∞) = {P }. This shows that the smallest non-gap at any ramification point is 2, and so every ramification point is a Weierstrass point of weight g(g − 1)/2 by Theorem 1.2.14. Thus, if we sum the weights of all 2g + 2 ramification points, we obtain g3 − g. By Prop. 1.2.13, this means that there are no more Weierstrass points, so we have shown that (i) implies (ii). Now suppose C has precisely 2g + 2 Weierstrass points. Then it follows directly from Prop. 1.2.13 and Theorem 1.2.14 that each Weierstrass point P has weight g(g − 1)/2, and so, by Theorem 1.2.14 again, the smallest non-gap at each such P is 2. By Prop. 1.1.7, this is equivalent to saying that the g smallest non-gaps at P are 2, 4,..., 2g, i.e., that the g gaps at P are 1, 3,..., 2g − 1. So we have shown (ii) implies (iii). It is clear that (iii) implies (iv). Now suppose that at some point P on C, the gaps are 1, 3,..., 2g − 1. Then 2 is a non-gap; in particular, there is a morphism ϕ : C → P1 of degree 2. So we have shown (iv) implies (i).
Corollary 1.3.2. Let P be a Weierstrass point on C, and suppose we have a double cover ϕ : C → P1. Let Q, R be the pre-images of a single point in P1 under ϕ (we allow Q = R). Then [Q] + [R] ∼ 2[P ], where ∼ denotes linear equivalence.
Proof. We may assume without loss of generality that Q, R are the pre-images of ∞. By the previous proposition, P is a ramification point of ϕ. Thus, if ϕ(P ) = ∞, then Q = R = P , and we are done. Otherwise, we note that if we define f = ϕ − ϕ(P ), then 2[P ] = [Q] + [R] + (f), again using the fact that P is a ramification point of ϕ.
Lemma 1.3.3. If C is hyperelliptic, then the double cover C → P1 is unique up to automorphisms on P1.
Proof. Let ϕ, ϕ˜ : C → P1 be two double covers. We would like to show thatϕ ˜ = ψ ◦ ϕ, where ψ is an automorphism on P1. Let [P ] + [Q], [P˜] + [Q˜] be the polar parts of (ϕ), (ϕ ˜), respectively. Then, by the preceding corollary, we have [P ]+[Q] ∼ [P˜]+[Q˜]. This implies 25 1.3. Weierstrass points on hyperelliptic curves
that there is some function h ∈ k(C)× such that multiplication by h is a k-vector space isomorphism H0(C, [P ] + [Q]) → H0(C, [P˜] + [Q˜])
By Lemma 1.1.1 (and the fact that 1 is a gap at every point), we have that {1, ϕ} and {1, ϕ˜} are bases for H0(C, [P ] + [Q]),H0(C, [P˜] + [Q˜]), respectively. Thus, we may write
ϕ˜ = aϕh + bh
1 = cϕh + dh
for some a, b, c, d ∈ k s.t. ad 6= bc, and so we have
ϕ˜ aϕh + bh aϕ + b ϕ˜ = = = 1 cϕh + dh cϕ + d at + b Thus,ϕ ˜ = ψ ◦ ϕ, where ψ : P1 → P1 is the automorphism mapping t ∈ P1 to . ct + d Lemma 1.3.4. A non-trivial automorphism of C fixes at most 2g + 2 points.
Proof. Let θ be an automorphism on C that is not the identity. Then there is a point P on C that is not fixed by θ. Let m ∈ {1, 2, . . . , g + 1} be a non-gap at P (there must be a non-gap among these integers by Corollary 1.1.6). This means there is a morphism ϕ : C → P1 with a single pole of order m at P . Now let ψ = ϕ − ϕ ◦ θ. Every fixed point of θ is a zero of ψ. The poles of ψ are P and θ−1(P ), and each pole is of order m. Thus, ψ has at most 2m ≤ 2(g + 1) = 2g + 2 zeroes.
Remark 1.3.5. Note that Lemma 1.3.4 holds regardless of the characteristic of k.
Proposition 1.3.6. Let Aut(C) be the group of automorphisms on C, and Σ the permu- tation group on the set of Weierstrass points of C. The group homomorphism
λ : Aut(C) → Σ
induced by the action of Aut(C) on the Weierstrass points of C is injective if and only if C is not hyperelliptic. If C is hyperelliptic, ker(λ) = {Id, ι}, where ι is the hyperelliptic involution on C. Chapter 1. A na¨ıve approach to Weierstrass points 26
Proof. By Prop. 1.3.1, if C is not hyperelliptic, then it has more than 2g + 2 Weierstrass points, and so no non-identity automorphism on C fixes all of them by the preceding lemma. Thus, in this case, ker(λ) = {Id}. Now suppose C is hyperelliptic. Take a double cover ϕ : C → P1. Since the Weierstrass points of C are the ramification points of ϕ, the hyperelliptic involution fixes them. Now suppose we have an automorphism θ on C that fixes all 2g + 2 Weierstrass points of C. By Lemma 1.3.3, ϕ = ψ ◦ ϕ ◦ θ, where ψ is an automorphism on P1. But, by the given, ψ fixes 2g + 2 ≥ 6 points of P1, and thus it must be the identity. Hence, ϕ = ϕ ◦ θ, and so θ must be the identity or the hyperelliptic involution.
1.4 Higher Weierstrass points
In this section, we briefly discuss higher Weierstrass points. The discussion here is based on [FK80, pp. 84-85]. For the rest of this section, we fix a positive integer q and a canonical divisor K on C, and we suppose g > 0.
Denote by iq(D) the dimension of
0 1 ⊗q −1 ∼ n ⊗q o H (C, (ΩC ) ⊗ O(D) ) = ξ ∈ Ωk(C)/k : div(ξ) ≥ D
Note that
iq(D) = `(qK − D)
Denote by Hq(C) the space of holomorphic q-differentials on C,
q 0 1 ⊗q H (C) = H (C, (ΩC ) )
q and let r = dimk H (C).
Definition 1.4.1. Fix a point P on C. A positive integer n is called a q-gap at P if ` n[P ] − (q − 1)K − ` (n − 1)[P ] − (q − 1)K = 0 (1.4)
Remark 1.4.2. Note that, by Lemma 1.1.1, the left-hand side of 1.4 is either 0 or 1. If it is 1, we say n is a non-q-gap. Furthermore, by the Riemann-Roch Theorem, we have ` n[P ] − (q − 1)K − ` (n − 1)[P ] − (q − 1)K = 1 + iq(n[P ]) − iq((n − 1)[P ]) 27 1.4. Higher Weierstrass points
so n is a q-gap if and only if iq((n − 1)[P ]) − iq(n[P ]) = 1, i.e., if and only if there is a
holomorphic q-differential ζ with ordP ζ = n − 1. A 1-gap at P is, of course, simply a (Weierstrass) gap at P .
Proposition 1.4.3. There are precisely r q-gaps at P . Furthermore, 1 is a q-gap and all q-gaps are less than 2(q(g − 1) + 1).
Proof. By Lemma 1.2.6, there is a holomorphic differential ω ∈ H(C) s.t. ordP ω = 0. ⊗q Now let ζ = ω . ζ is a holomorphic q-differential with ordP ζ = 0. By the above remarks, this means that 0 + 1 = 1 is a q-gap at P .
For N ≥ 1, let AN be the number of q-gaps ≤ N. The number of non-q-gaps ≤ N is
N X ` n[P ] − (q − 1)K − ` (n − 1)[P ] − (q − 1)K = N + iq(N[P ]) − iq(0) n=1
= N + iq(N[P ]) − r
and so
AN = N − (N + iq(N[P ]) − r) = r − iq(N[P ]) = r − `(qK − N[P ])
Now we note that for N ≥ q(2g − 2) + 1, we have deg(qK − N[P ]) < 0, and so for such N we have `(qK − N[P ]) = 0. This shows that there are precisely r q-gaps and they are all less than q(2g − 2) + 2 = 2(q(g − 1) + 1).
Definition 1.4.4. Let 1 = n1 < . . . < nr < 2(q(g − 1) + 1) be the r q-gaps at P . We define r X τq(P ) := (nj − j) j=1
(Note that τq(P ) ≥ 0.) We call τq(P ) the q-weight of the point P . If P has positive q-weight, we call it a q-Weierstrass point.
⊗q Definition 1.4.5. Let t be a uniformizer at P , and ζ1, . . . , ζr ∈ Ωk(C)/k a sequence of q-differentials on C. For i = 1, . . . , r , let ϕi ∈ k(C) be the unique function satisfying q ζi = ϕi(dt) . We define
Wt(ζ1, . . . , ζr) := Wt(ϕ1, . . . , ϕr) Chapter 1. A na¨ıve approach to Weierstrass points 28
We call Wt(ζ1, . . . , ζr) the wronskian of ζ1, . . . , ζn with respect to t. The wronskian for q-differentials inherits properties 1.2 and 1.3 from the wronskian for functions, and so we have the following result.
Proposition 1.4.6. Let each of t, t˜ ∈ k(C) be a uniformizer at some point on C. For ⊗q any sequence ζ1, . . . , ζr ∈ Ωk(C)/k of q-differentials on C, we have
dt˜r(2q−1+r)/2 W (ζ , . . . , ζ ) = W (ζ , . . . , ζ ) t 1 r t˜ 1 r dt
Definition 1.4.7. Let m = r(r + 2q − 1)/2. We define the wronskian W (ζ1, . . . , ζr) of a ⊗q m sequence ζ1, . . . , ζr ∈ Ωk(C)/k of q-differentials to be the m-differential Wt(ζ1, . . . , ζr)(dt) , where t is a uniformizer at some point on C. This definition is independent of the choice of point and uniformizer, since, by the above proposition, we have
dt˜m W (ζ , . . . , ζ )(dt)m = W (ζ , . . . , ζ ) (dt)m = W (ζ , . . . , ζ )(dt˜)m t 1 r t˜ 1 r dt t˜ 1 r for any other uniformizer t˜.
The proof of the following proposition is identical to that of Prop. 1.2.13, so we omit it.
Proposition 1.4.8. Suppose that char(k) = 0 or char(k) > 2q(g − 1). Then for any
q basis ζ1, . . . , ζr of H (C), the Wronskian W (ζ1, . . . , ζr) is non-zero, and
ordP (W (ζ1, . . . , ζr)) = τq(P ) for any point P on C.
Corollary 1.4.9. Let C, g, d, k be as in the above proposition. Then
X τq(P ) = (g − 1)r(r + 2q − 1) P ∈C
In particular, C has at most (g − 1)r(r + 2q − 1) q-Weierstrass points, and the curves that do not have q-Weierstrass points are the curves of genus 1. Chapter 2
Laksov’s Weierstrass points on curves
While the definition of a Weierstrass point that we provided in the previous chapter is perhaps the most natural one from a classical perspective, it presents some problems in low characteristic. Pathologies may occur if the base field of a curve C is of characteristic smaller than 2g − 2, where g is the genus of C. It is possible for such a curve to have infinitely many Weierstrass points; indeed, in the following example, we provide a curve with the property that each of its points is a Weierstrass point in the na¨ıve sense.
Example 2.0.1. A point P on a smooth plane curve C is called a flex if the tangent line to C at P intersects C at P with multiplicity at least 3. In the case that C is a quartic curve, the flexes of C are precisely its Weierstrass points. Indeed, a smooth plane quartic C is a curve of genus 3, so by Prop. 1.2.3, we have that P is a Weierstrass point if and only if there exists a canonical divisor K on C such that K ≥ 3[P ]. As C is canonically embedded, this is equivalent to saying that P is a flex, since the effective P P2 canonical divisors on C are the divisors of the form P iP (L, C) where L is a line in and iP (L, C) is the intersection multiplicity of L with C at P . Now consider the smooth plane quartic C, known as the Klein quartic, given by
X3Y + Y 3Z + Z3X = 0
29 Chapter 2. Laksov’s Weierstrass points on curves 30
over a field k of characteristic 3. This curve has the remarkable property that each of its points is a flex. We verify this. By symmetry, it suffices to show that every point on the
3 3 affine curve C1 given by x y + y + x = 0 is an inflection point in characteristic 3. We note that this curve intersects the line x = 0 solely at the origin with multiplicity 3. Now ˜ let P = (α, β) be a point on C1 with α 6= 0, and let C1 be the affine curve given by
f(x, y) := ((y + x) + α)3 −α−3(y − x) + β + −α−3(y − x) + β3 + ((y + x) + α) = 0
This curve is the image of C1 under a linear change of coordinates. The tangent line to ˜ C1 at the origin is x = 0, and the intersection multiplicity at P of C1 with the tangent ˜ line to C1 at P is equal to the intersection multiplicity at the origin of C1 with the line x = 0. In characteristic 3, we have
f(x, y) = y3 −α−3y + β − α−9 + xg(x, y)
for some g ∈ k[x, y], and so this multiplicity is at least 3.
As the above example illustrates, we will need a different definition of a Weierstrass point if we would like finiteness to be preserved in low characteristic. In this chapter, we expose Laksov’s definition of a Weierstrass point. Under his definition, a curve C has finitely many Weierstrass points regardless of the characteristic of the base field k (and their weighted number is a constant that depends only on the genus of C). Furthermore, Laksov associates Weierstrass points to any effective divisor D on C of positive degree; these are precisely the na¨ıve Weierstrass points defined in the previous chapter when char(k) > 2g − 2 and D is a canonical divisor.
2.1 The bundle of principal parts
Let C be a non-singular curve of genus g over an algebraically closed field k and D an effective divisor on C of positive degree d and projective dimension r, so that h0(C,D) = r + 1. Note that if D is a canonical divisor, then g ≥ 2 by these assumptions. All points in this chapter are assumed to be closed, unless otherwise stated. 31 2.1. The bundle of principal parts
If F is an OC -module and x is a point of C, we denote the fiber of F at x by F(x).
This is the k-vector space Fx ⊗OC,x k, where Fx and OC,x are the stalks of F and OC at x.
If V is a k-vector space, denote by VC the OC -module V ⊗k OC . This is the pullback of V (thought of as an OSpec(k)-module) via the morphism C → Spec(k).
Denote by I the sheaf of ideals defining the diagonal in C ×k C. For m ≥ −1, denote th by ∆m the m infinitesimal neighbourhood of the diagonal; this is the subscheme of m+1 0 C ×k C defined by the sheaf of ideals I . We set I = OC×C .
Denote by p and q the projections of C ×k C onto the first and second coordinates, respectively. The short exact sequence
m+1 0 → I → OC×C → O∆m → 0
∗ of OC×C -modules remains exact after tensoring over OC×C by q O(D). After applying
p∗ we obtain a long exact sequence
m+1 ∗ ∗ ∗ 0 → p∗(I ⊗ q O(D)) → p∗q O(D) → p∗(O∆m ⊗ q O(D))
1 m+1 ∗ 1 ∗ → R p∗(I ⊗ q O(D)) → R p∗q O(D) → 0 (2.1)
Here we have zero to the right because the restriction of p to the closed subscheme ∆m
of C ×k C is affine (topologically, it is a homeomorphism).
∗ th Definition 2.1.1. We call the OC -module p∗(O∆m ⊗ q O(D)) the sheaf of m order principal parts of D and denote it by Pm(D).
For each m ≥ 0, we have an exact sequence
m m+1 0 → I /I → O∆m → O∆m−1 → 0 (2.2)
∗ Applying p∗(· ⊗ q O(D)) to the above exact sequence, we obtain the exact sequence