NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES
AARON LANDESMAN
CONTENTS 1. Introduction 4 2. 9/2/15 5 2.1. Course Mechanics and Background 5 2.2. The Basics of curves, September 2 5 3. 9/4/15 6 3.1. Outline of Course 6 4. 9/9/15 8 4.1. Riemann Hurwitz 8 4.2. Equivalent characterizations of genus 8 4.3. Consequences of Riemann Roch 9 5. 9/11/15 10 6. 9/14/15 11 6.1. Curves of low genus 12 7. 9/16/15 13 7.1. Geometric Riemann-Roch 13 7.2. Applications of Geometric Riemann-Roch 14 8. 9/18/15 15 8.1. Introduction to parameter spaces 15 8.2. Moduli Spaces 16 9. 9/21/15 18 9.1. Hyperelliptic curves 18 10. 9/23/15 20 10.1. Hyperelliptic Curves 20 10.2. Gonal Curves 21 11. 9/25/15 22 11.1. Curves of genus 5 23 12. 9/28/15 24 12.1. Canonical curves of genus 5 24 12.2. Adjoint linear series 25 13. 9/30/15 27 13.1. Program for the remainder of the semester 27 13.2. Adjoint linear series 27 14. 10/2/15 29 15. 10/5/15 32 15.1. Castelnuovo’s Theorem 33 16. 10/7/15 34 17. 10/9/15 36 18. 10/14/15 38 1 2 AARON LANDESMAN
19. 10/16/15 41 19.1. Review 41 19.2. Minimal Varieties 42 20. 10/19/15 44 21. 10/21/15 47 22. 10/23/15 49 22.1. Agenda and Review 49 22.2. Resolutions of Projective Varieties and Green’s conjecture 50 22.3. The Maximal Rank Conjecture 52 23. 10/26/15 53 24. 10/28/15 55 24.1. Review 55 25. 10/30/15 57 25.1. Review 57 25.2. Today: Brill Noether Theorem in dimension at least 3 58 26. 11/2/15 61 26.1. Review 61 26.2. Hilbert Schemes 61 27. 11/4/15 64 27.1. Logistics 64 27.2. Tangent Spaces in Brill Noether Theory 64 27.3. Martens Theorem 66 28. 11/6/15 67 29. 10/9/15 71 29.1. Agenda and Review 71 30. 11/11/15 75 30.1. Review 75 30.2. The Genus 6 Canonical Model 76 31. 11/13/15 78 32. 11/16/15 81 32.1. Logistics and Review 81 32.2. A continuing study of genus 6 curves 82 33. 11/18/15 84 33.1. Logistics and Overview 84 34. 11/20/15 87 34.1. Plan, conventions, and review. 87 34.2. An Upper bound on the dimension 88 34.3. Proof of Existence for Brill Noether 88 35. 11/23/15 90 35.1. Review 90 35.2. Inflectionary points of linear series 92 36. 11/30/15 93 36.1. Review and overview 93 37. 12/2/15 97 37.1. Overview 97 37.2. Retraction 97 37.3. Finishing the proof of Brill-Noether 97 NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 3
37.4. Further Questions 99 4 AARON LANDESMAN
1. INTRODUCTION Joe Harris taught a course (Math 282) on algebraic curves at Harvard in Fall 2015. These are my “live-TEXed“ notes from the course. Conventions are as follows: Each lecture gets its own “chapter,” and appears in the table of contents with the date. Of course, these notes are not a faithful representation of the course, either in the mathematics itself or in the quotes, jokes, and philosophical musings; in particular, the errors are my fault. By the same token, any virtues in the notes are to be credited to the lecturer and not the scribe. 1 Please email corrections to [email protected].
1This introduction has been adapted from Akhil Matthew’s introduction to his notes, with his permission. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 5
2. 9/2/15 2.1. Course Mechanics and Background. (1) Math 282, Algebraic Curves (2) CA Adrian (3) Text: ACGH, Volume 1 (4) Four years ago, a similar course was taught, following ACGH. The idea was: given a curve, what can we say about it. This is only half the story. Curves can appear in the abstract and in projective space. An important part of understanding curves is how they vary in flat families. The differ- ence between ACGH volumes 1 and 2, is that 1 deals with a fixed curve and 2 deals with families of curves. To learn more on families of curves, look at Moduli of Curves. (5) Two major changes in the language since when the book was written: First, we will use cohomology, and second we will use schemes. (6) Algebraic curves were first studied over the complex numbers. Some peo- ple studied complex analysis of Riemann Surfaces, and others studied polynomials in two variables. Remark 2.1. We will use the language of smooth projective curves and compact Riemann surfaces interchangeably. We will assume all curves are over the complex numbers. The central problem of the course is Question 2.2. What is a curve? In the 19th century, a curve is a subset of Pn for some n. In the 20th century, a curve became an abstract curve, which exists indepen- dently of any particular embedding in projective space. A similar perspective was adapted in group theory. Originally, people viewed groups as subsets of GLn. Now, this is called representation theory. Remark 2.3. In his textbook, Hartshorne says the goal of algebraic geometry is to classify algebraic varieties. In the modern context, we can just specify the genus. However, in the 19th century, you would have to also specify the degree. We can then ask, which pairs of d, g are realized as a curve. This is still not completely known. You can also specify a projective space, and then ask which curves can be realized in that projective space. 2.2. The Basics of curves, September 2. 1 Definition 2.4. Define the genus by g = 2 (1 − χtopX). Definition 2.5. An ordinary singular point of a curve of multiplicity m is a point in which m branches of a curve meet transversely of a point. We can define this more rigorously by saying that the completion of its local ring is isomorphic to k[x, y]/(f1 ··· fm) where fi are distinct linear functions in x, y. Lemma 2.6. Let X be a curve. The following are equivalent: (1) The genus of X. (2) 1 − χOX 1 (3) 2 (deg KX + 2) 6 AARON LANDESMAN
(4) 1 − c, where c is the constant term of the Hilbert polynomial of C ⊂ Pr. ∼ 2 (5) If C = C0 ⊂ P of degree d with ordinary singular points of multiplicity mi, d−1 mi then g(C) = 2 − i 2
Definition 2.7. A divisor isP an formal sum of the form D = i nipi for ni ∈ Z, pi ∈ X. We say D is effective if D ≥ 0. We define the degree by deg D = i ni. For f a rational function on X, we define P P div f = (f) = ordp(f) · p = (f)0 − (f) p X . ∞ Remark 2.8. By the residue formula applied to the logarithmic derivative of a rational function f, we have deg (f) = 0. Definition 2.9. We say D ∼ E, or D is linearly equivalent to E if there exists a rational function f with (f) = D − E. Effective divisors of degree d on X will be d notated as Cd, which is by definition C /Sd. Definition 2.10. Given D ∈ div X, we look at L(D) := {f ∈ K(X)× :(f) + D ≥ 0}. An alternative notation for L is H0. Since we can specify the polar part of such a function, this is a finite dimensional vector space. This uses the fact that there are no nonconstant meromorphic function. We define `(D) = dim L(D) and define r(D) = `(D) − 1. Remark 2.11. If D ∼ E then L(D) =∼ L(E), as given by multiplication by f.
Definition 2.12. We define Picd(X) := Divisors of degree d/ ∼. Remark 2.13. It turns out this set corresponds to the points of a variety. Definition 2.14. Suppose ω is a rational 1-form, which looks locally like f(z) dz. 2g−2 We let (ω) = ordp(ω)p. We define KX := (w) ∈ Pic (X). P 3. 9/4/15 Note: there will be no class Monday or Friday.
3.1. Outline of Course. (1) This week: Basics of linear series (2) Starting 9/14, we’ll talk about curves of low genus and Castelnuovo’s the- orem (gives an upper bound on the genus of a curve of degree d in projec- tive space) (3) Brill-Noether Theory: This addresses the question “What can you say about a general curve?” It makes sense to ask whether there is an open subset of the Hilbert scheme on which there is a uniformity of the corresponding curves. Remark 3.1. For the remainder of the course, we let X be a smooth projective curve. We let D = i niPi and say D ∼ E if D is linearly equivalent to E. We let KX be the divisor class of the canonical sheaf ω, of degree 2g − 2. Use L(D) for P H0(C, D), let `(D) = h0(C, D) and r(D) = `(D) − 1. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 7
The justification of looking at these linear systems is that we only allow poles at specified points. This gives us a finite dimensional vector space, which is some- thing quite manageable. We next explain what this has to do with maps to projective space. r Remark 3.2. Given a map φ : X P and let H = V(x0) be a hyperplane cut r out by the first coordinate of P . Then, the map can be given by [1, f1, ... , fr], and −1 ∞ D = φ H and f1, ... , fr ∈ L(D→).
There are∞ some issues with the above description. In particular, we will need to enforce that these maps are basepoint free. We will come back to this later. Question 3.3. Given X of genus g and D a divisor of degree d on X, what can we say about `(D).
Example 3.4. We know `(KX) = g. Theorem 3.5. We have
`(D) = d − g + 1 + `(KX − D) Proof. Caution: there is a problem with the following proof. We have managed to sweet Serre duality under the rug. The problem is that we assumed D and K − D were effective. Say D = p1 + ··· + pd, where pi are distinct points of X. Choose local coordi- nates z around each p . Any f ∈ L(D) can be written locally as ai + h where h i i zi is holomorphic. So, f is determined up to the addition of a constant function by a1, ... , ad. So, the question is: which d tuples arise as global principal parts of functions. First, note that `(D) ≤ d + 1. Now, we ask, what is the obstruction to having a global rational function with these polar parts. Here, we use that if we have a meromorphic 1-form on a curve then the sum of the residues is 0 by stokes’ theorem. So, if ω ∈ L(KX), and f ∈
L(D), then i Respi (f · ω) = 0. Say ω(pi) = bidzi. So, plugging this in the above formula, we have i aibi = 0. P So, we get g relations, but they might not be independent. The actualP number of relations we are getting is g − `(KX − D). So, `(D) ≤ d + 1 − g + `(KX − D). The reverse inequality is not quite proved correctly, but to do it would be a lot more work, so we give a heuristic argument. Now, we apply the same argument to `(K − D). We have `(K − D) ≤ 2g − 2 − d + 1 − g + `(D) . Adding the two inequalities we have `(D) + `(K − D) ≤ `(D) + `(K − D) . So, we added two inequalities and got an equality, which means they were equal- ities to begin with. Suppose you are give a compact complex Riemann surface, how do you know there are any nonconstant meromorphic functions. This issue underlies the diffi- culty we are encountering in the proof above. In fact, in dimension at least 2, there are compact complex manifolds with no meromorphic functions whatsoever. 8 AARON LANDESMAN
Remark 3.6. How should we define a canonical divisor on a singular curve? We will have to choose a canonical divisor on a singular curve so that it satisfies the condition that sums of residues of f · ω is 0. Now, returning to the issue of obtaining a divisor D by a preimage of a hyper- plane, we look at locally.
Definition 3.7. Given a divisor D on X we define a sheaf OX(D) defined by × OX(D)(U)L := {f ∈ K(U) :(f) + D ≥ 0 in U.} When X is smooth, this corresponds to a line bundle. ∼ Observe that if D ∼ E, OX(D) = OX(E), as given by multiplication by X. Addi- tionally, define Picd to be line bundles of degree d, or equivalently, linear equiva- lence classes of degree d. Finally, define 0 L(D) := H (X, OX(D)) Remark 3.8. This construction is a special case of the fact that for locally factorial schemes which are regular in codimension 1, there is an isomorphism between the vector spaces of Weil and Cartier divisors. Lemma 3.9. The collection of nondegenerate maps (maps whose image is not contained in a hyperplane are in bijection with pairs (V, L) where L ∈ Picd(X), V ⊂ H0(L) with dim V = r + 1, so that V is basepoint free. Proof. See Vakil, chapter 16. 4. 9/9/15 4.1. Riemann Hurwitz. 4.2. Equivalent characterizations of genus. (1) The dimension of the vector space of holomorphic 1 forms. (2) χ(OC) = 1 − g (3) pC(m) = md + 1 − g (4) `(D) = d − g + 1 for D >> 0. (5) Number of handles (6) b1(C) = 2g (7) deg KC = 2g − 2 (8) χ>(C) = 2 − 2g. Remark 4.1. It is easier to prove equivalence of these definitions in the algebraic category because in that setting we already have access to the algebraic functions coming from projective space via an embedding. Suppose π : C B is a nonconstant map of compact Riemann surfaces of degree d. Then, for all but finitely many points in the target, the preimage consists of d points. For all points→ p ∈ C, we can choose local coordinates so that π is of the form z 7 zm, and when m > 1 p is a ramified point.
Definition 4.2. Given a map π : C B as above, set vp(π) = m − 1 and define the ramification→ divisor
(4.1) R =→ vp(π) · p p∈C X NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 9 and the branch divisor by (4.2) B = · q q∈B −1 X p∈πX(q) We define the total ramification index as b := deg R = deg B Theorem 4.3. (Riemann-Hurwitz) Suppose π : C B is a nonconstant map of compact Riemann surfaces of degree d. Then, 2g − 2 = d(2h − 2) + b. ∗ → Proof. We compare the degrees of ωC and π ωB. At a point of B, the pullback of a from on B will pick up zeros equal to the ramification index. So, ∗ (4.3) KC = π KB + R
Then, taking degrees, since deg KC = 2g − 2, we obtain Riemann-Hurwitz. Remark 4.4. It’s also fun to prove this via topological Euler characteristic. You do this by removing the ramification and branch locus. We then obtain an unramified covering space, and use that on this locus, the Euler characteristic is multiplicative. Then, you add the points back in and deduce the formula. 4.3. Consequences of Riemann Roch. Remark 4.5. Recall, nondegenerate maps C Pr of degree d, modulo the action of PGLr+1 are in bijection with pairs (L, V) so that L is a line bundle on C of degree 0 d and V of dimension r + 1 in H (L) with no→ common zeros, i.e., no basepoints. Given a line bundle and sections, we can choose φD(p) = [σ0(p), ... , σr(p)]. We can also describe this intrinsically by Vp = {σ ∈ V : σ(p) = 0} . Then the map φ ∨ sends φD : p 7 Vp ∈ PV . r Definition 4.6. A gd is a line bundle of degree d and a vector space V of dimension r + 1 inside H→0(L). Crucially, we do not assume that V is basepoint free. r r ∼ ∨ Remark 4.7. If D = (L, V) is a basepoint free gd, we get a map C P = PV . The map being injective is equivalent to Vp+q = Vp ∩ Vq has codimension 2. Equivalently, Vp+q = {σ ∈ V : σ ≥ p + q}. The map is an immersion→ if and only if for all p ∈ C, V2p has codimension 2.
Lemma 4.8. A map φ is an embedding if and only if for all effective divisors D ∈ C2 i.e., of degree 2, VD has codimension 2. Proof. This is Remark 4.7 0 0 Corollary 4.9. If deg L ≥ 2g + 1 then φ|L| is an embedding, where |L| = (H (L), H (L)). Proof. By Riemann Roch, deg L ≥ 2g + 1 = dim H0(L) = d − g + 1. But, it’s also true that deg L(−D) ≥ 2g − 1 so H0(L) = d − g − 1. Therefore, any line bundle of degree 2g + 1 or more gives rise to an embedding.⇒ Remark 4.10. We now want to find the best embedding, where it is easiest to deal with the image. What is the lowest degree of an embedding we can find?
Lemma 4.11. Suppose D is a divisor of degree d ≥ g + 3. Then, if D ∈ Cd general then φ|O(D)| = φD is an embedding. 10 AARON LANDESMAN
Proof. Difficulty, we won’t do it now, but it will be helpful to know about the Jacobian. Definition 4.12. A curve C is hyperelliptic if there exists a map π : C P1 of 2 2g+2 degree 2. That is, C can be written as y = i=1 (x − λi) on an open subset. → Exercise 4.13. Not every curve of genus atQ least 3 is hyperelliptic. Example 4.14. In genus 2, the canonical sheaf makes C into a hyperelliptic curve. Definition 4.15. For any curve C of genus g > 0, we have canonical map, defined to be the map associated to the dualizing sheaf. Theorem 4.16. A curve C is hyperelliptic if and only if the canonical map is not an embedding. Proof. Use the criterion for being a closed embedding, and Riemann Roch. Fur- thermore, use the characterization that being hyperelliptic is equivalent to having an effective divisor of degree 2.
5. 9/11/15 Today Joe is out, and Adrian is holding a review. 1 Remark 5.1. Deformations of abstract schemes are classified by H (TC). Embed- 0 ded deformations are classified by H (NY/X). 1 Example 5.2. Let’s show there are non hyperelliptic curves. First, TCMg = H (TC) = 0 0 H (2KC). By Riemann-Roch, h (2KC) = 3g − 3. This gives a proof that dim Mg = 3g − 3.
Exercise 5.3. Compute dim Mg by first calculating the dimension of the space of covers of P1 of degree d. Theorem 5.4. Let X be a smooth projective variety and Y ⊂ X be smooth of codimension 1. Then, KY = (KX + Y)|Y. Exercise 5.5. Let C be a smooth curve on P1 × P1. Then, C is given by a bihomo- geneous polynomial of class (a, b). 1 1 ∼ Solution: We have KC = (KP1×P1 + C)|C. Then, Pic P × P = ZH1 ⊕ ZH2. So,
(5.1) KC = KP1×P1 + C|C (5.2) = (−2H1 − 2H2 + aH1 + bH2)|C
(5.3) = ((a − 2)H1 + (b − 2)H2)|C (5.4) = (a − 2)b + (b − 2)a (5.5) = 2ab − 2a − 2b So, g = (a − 1)(b − 1). Exercise 5.6. Compute the dimension of the space of twisted cubics in P3. (I.e., a component of the Hilbert scheme) Solution: We have a map P1 P3 of degree 3, which is nondegenerate. We can write any map sending [x, y] 7 [fi(x, y)]. where fi is a homogeneous polynomial → → NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 11 of degree 3. This is 16 dimensional, we subtract 1 for scalars and subtract 3 for the PGL2 action. Solution 2: Lemma 5.7. If C is a twisted cubic, then C is contained in a quadric.
Proof. Note that OP3 (2) has 10 dimensional cohomology, while OC(2) is 7 dimen- sional by Riemann-Roch. So, in fact, it lies on 3 quadrics. Now, consider (C, Q) so that C is contained in Q and this gives you what you want. ¡++¿
6. 9/14/15 We will take the approach of interweaving new techniques with applications. Today’s new technique is the adjunction formula. We’ll give two proofs of the adjunction formula. We’ll have the following setting: Let X be smooth varieties of dimension n. Every such scheme has a canonical bundle , which is an invertible sheaf of top n ∨ dimensional forms KX = ∧ TX . Suppose Y ⊂ X is a smooth divisor, i.e., codimension 1 subvariety. Recall we can associated an invertible sheaf to the divisor Y, called L = OX(Y). There is also a section σ ∈ H0(L) corresponding to the regular function 1, with V(σ) = Y. The key fact is
Lemma 6.1. NY/X = L|Y = OX(Y)|Y = OY(Y). Proof 1. Suppose we have a line bundle with total space L over X. Say p ∈ Y and q ∈ L where q is on the zero section over p. Then, TqL = TpX ⊕ Lp. Then, the tangent space to σ, with V(σ) = Y, is the graph of a map TpX Lp. The kernel of this map is TpY, and by the conormal exact sequence, we have (NY/X)p = Lp. Here is an algebraic proof: → Proof 2. We have ∨ ∼ 2 NY/X = IY /IY = OY|Y = OX(−Y)|Y.(6.1)
And so, NY/X = OX(Y)|Y. We are now in a position to prove the adjunction formula. We have an exact sequence
0 TY TX|Y NY/X 0 Recall that in general if we have an exact sequence of vector spaces
0 An−1 Bn C1 0 then there is a natural isomorphism ∧nB =∼ ∧n−1A ⊗ C. Applying this to 6.1 we obtain Theorem 6.2.
(6.2) KY = KX|Y ⊗ NY/X = KX(Y)|Y 12 AARON LANDESMAN
Proof. This is immediate from applying 6.1 to the conormal exact sequence. n Example 6.3. KPn = OPn (−n − 1). To see this, given P , we have homogeneous coordinates z0, ... , zn and on the affine open, we have inhomogeneous coordi- nates zi/z0. Look at dz1 ∧ ··· dzn. This is holomorphic and nonzero in z0 6= 0. This has a pole of order n + 1 along z0 = 0. Alternatively look at
dz1 dzn (6.3) ∧ ··· ∧ z1 zn is a meromorphic differential which has a pole on each of the coordinate hyper- planes, and is otherwise holomorphic. Example 6.4. (1) If C ⊂ P2 is a smooth curve of degree d then by adjunction OP2 (C) = OP2 (d) and so KC = OC(d − 3) which has degree (d − 3) · d. d−1 Therefore, the degree KX = 2g − 2 = (d − 3)d and so g = 2 . (2) If Q ⊂ P3 is a smooth quadric surface. Note Q =∼ P1 × P1 P3 via the Segre embedding. Then, the tangent plane to Q intersects Q at the two lines of the two rulings passing through the point. We say a→ curve C ⊂ Q has bidegree (a, b) if C is the zero locus of a bihomogeneous polynomial of bidegree (a, b), i.e., if C meets lines of one ruling in a points and of the other ruling in b points. (3) If C ⊂ Q ⊂ P3 is a smooth curve of bidegree (a, b). Then, deg(C) = a + b, as can be seen by choosing the hyperplane to be a tangent hyperplane to the quadric, whose intersection with the quadric surface is one line of each ruling. Now, we first compute KQ. We claim, KQ = OQ(−2, −2). First, Q is P1 × P1, so to write a meromorphic 2 form on Q is the same as writing down meromorphic 1 forms on both copies of P1 and the zeros are the preimages of the zeros of the forms on P1. Then, we obtain a meromorphic 2 form with poles on 2 lines in each ruling. So,
(6.4) KQ = KP3 ⊗ OP3 (Q)|Q = OP3 (−2)|Q Then, applying adjunction again,
(6.5) KC = KQ ⊗ OQ(C)|C
(6.6) = OQ(a − 2, b − 2)|C
So, plugging in degrees, 2g − 2 = deg KC = b(a − 2) + a(b − 2). Then, g = (a − 1)(b − 1). Now, if we take a curve with one of the bidegrees 1, we see this is a rational curve, and so g = (a − 1)(b − 1). 6.1. Curves of low genus.
6.1.1. Genus 0. If C has genus 0, then C =∼ P1, as follows from Riemann-Roch (or a conic in P2 over arbitrary fields). Furthermore, there is only one line bundle on the curve of any degree, because any two curves of the same degree are linearly equivalent. Then, we can look at the complete linear system OP1 (d) which embeds C as a rational normal curve of degree d ⊂ Pd. This d is equal to the smallest possible degree of an irreducible nondegenerate curve C ⊂ Pd, and any nondegenerate curve of that degree must be the rational NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 13 normal curve, as follows from Bezout’s theorem: If we had a curve of lower de- gree, d general points would lie in a hyperplane. Conversely, if we an irreducible, nondegenerate curve, d − 1 points of the curve span a codimension 2 linear space, and projection from that space will give an isomorphism of the curve with P1. (1) How many surfaces or hypersurfaces does a curve lie on? (2) What is the normal bundle of a rational curve? (3) Many more open problems about rational normal curves.
6.1.2. Genus 1. (1) If we take d = 3, by Riemann Roch, this gives an embedding C P2 as a plane cubic. Conversely, a smooth plane cubic by adjunction is a genus 1 curve. → (2) Now, take d = 4. This gives an embedding C P3. If we have a cubic plane curve, it is the zero locus of a cubic polynomial. Does this curve lie on a quadric surface? There is a standard way→ of answering this using 0 3 the following technique: We want to know whether H (P , IP3 (2)) van- ish on C. Restricting this to the curve, and to see if the curve lies on a quadric surface, we want to check the kernel of the restriction map. We 0 3 0 know H (P , IP3 (2)) = 10 and H (OC(2)) = 8. Therefore, the kernel is at least two dimensional. In fact, this map has a two dimensional kernel. So this curve lies on a P1 of quadric surfaces. None of the quadrics containing the curve are irreducible because C is irreducible, nondegenerate. Then, by Bezout’s theorem, C is the intersection of two quadrics. We could also do this by adjunction, and so C must be type (2, 2) on a quadric surface. (3) For degree 5, we get an intersection of G(2, 5) with hypersurfaces in the plucker embedding. After this, the equations get even more complicated.
6.1.3. Genus 2. Let C be a curve of geometric genus 4. Take a divisor of degree 4, corresponding to a line bundle L. We get a map C P2, where C has geometric genus 2. By Riemann Roch, h0(L) = 3. Because and the image is of degree 4 in P2, but it cannot be smooth because all plane quartics have→ genus 3. In fact: Exercise 6.5. For D of degree 4 on a curve C of genus 2, we can write p + q = 2 D − KC as an effective divisor for a unique divisor p + q on C. That is φD : C P is either a node if p 6= q or a cusp if p = q. The solution is just Riemann Roch. → 7. 9/16/15 7.1. Geometric Riemann-Roch. New tool: Geometric Riemann-Roch For the moment being, we will assume (1) Assume C is a non hyperelliptic curve of genus g (2) Assume D = pi with pi distinct. g−1 Let φ = φ|K| : CP P be the canonical embedding. In coordinates: if 0 ω1, ... , ωg are a basis for H (K), then φ : p 7 [ω1(p), ... , ωg(p)]. More intrinsically, P→g−1 = PH0(K)∨ and φ : p 7 H0(K − p) , H0(K). → Theorem 7.1. → → r(D) = d − 1 − dim D 14 AARON LANDESMAN
Proof. Suppose p1, ... , pd ∈ C. Note that if n is the number of linear relations on the points pi, so that n − d = dim spanp1, ... , pd − 1
`(K − D) = g − d + n = g − dim spanp1, ... , pd − 1 So, by Riemann-Roch, r(D) is the number of linear relations among the points pi. So, if the pi are linearly independent then there does not exist a nonconstant meromorphic function on C with simple poles at pi. In general, φ : C Pg−1 is some canonical map, which might not be an em- bedding if the curve is hyperelliptic. Let D be any effective divisor. In this case, g−1 define the span D = →∩φ−1(H)⊃DH ⊂ P . If we take D = 2p we have to take the tangent line to the points, that is, the intersection of all planes tangent to the curve at that point.
7.2. Applications of Geometric Riemann-Roch.
7.2.1. Genus 2, degree 5. Suppose D is a divisor class of degree 5 on a genus 2 curve. Then, φ : C P3 embeds C as a quintic curve. Given an embedding, we can ask about the ideal of the embedding. Last time, we asked: → Question 7.2. What surfaces in P3 contain C? 0 To calculate the quadrics on which this surface lies, note that h (OP3 (2)) = 0 10, h (OC(2)) = 10 − 2 + 1. Then, the map
0 0 H (OP3 (2)) H (OC(2)) is in fact surjective, because if C were on two quadrics, being nondegenerate, it → would be degree at most 4. Therefore, the kernel is 1 dimensional, and the curve lies on a unique quadric. This quadric could be either a smooth quadric or a cone, and on the homework we’ll understand the distinction. Now, let’s example the cubic surfaces containing C. We have
0 0 H (OP3 (3)) H (OC(3)) Now, C lies on 6 linearly independent cubics. However, we knew about 4 of the → cubics C lies on from multiples of the quadratics. So, there are at least 2 linearly independent cubics, modulo the ideal generated by the quadratic polynomial Q. Choose S a cubic containing C but not Q. What is S ∩ Q. This is a curve of degree 6. Therefore, it must be the union C ∪ L, since L is degree 1. Here, we use that complete intersections are unmixed (i.e., the resulting scheme structure is reduced). So, C is type (2, 3) or (3, 2) on the quadric. Lemma 7.3. If M ⊂ Q is any line of type (1, 0) then M sup C is S ∩ Q. This gives a bijective correspondence between cubics containing C but not Q and lines of the ruling of S.
Proof. We have Q =∼ P1 × P1 P3 by the Segre map. Homogeneous degree 1 forms on P3 pull back to bidegree (1, 1) forms on Q. Therefore, there is a map → homogeneous polynomials of degree m on P3 bihomogeneous polynomials of bidegree (m,m) on P3
→ NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 15
Furthermore, this map is surjective, which can be proven by writing out the above map in coordinates. There’s another standard way of showing this, without re- sorting to coordinates. This map is the map on global sections associated to
0 OP3 (m − 2) OP3 (m) OQ(m) 0 is a surjection of sheaves. We’re asking whether it is surjective on global sections. Note that H1(L) = 0 for any line bundle on P3. Therefore, we have an associated exact sequence on global sections. So, M ∪ C = S ∩ Q. This shows that in fact S1 ∩ S2 ∩ Q = C. Alternatively, after we knew C was contained in Q, being quintic of genus 2, we would have known it was of type (2, 3), and we could have found which cubics it lies on. For degree 6 line bundles of on genus 2 curves, the geometry gets much more subtle, since in P2, when we found a surface containing the curve, the curve be- came a divisor on that surface.
7.2.2. Genus 4, non-hyperelliptic. Let’s focus on the canonical embedding. Here 3 φK : C P is a sextic curve, which is also the smallest degree embedding of C in any projective space. Being non-hyperelliptic, it’s not expressible as a two 1 sheeted→ cover of P . So, there are no meromorphic functions of degree 2. How- ever, it is trigonal. We’ll see this shortly. Looking at the usual exact sequences, we see 0 0 H (OP3 (2)) H (OC(2)) determines a surjective map, with C lying on a unique quadric surface, uniqueness from Bezout’s theorem. → Next, looking at cubics 0 0 H (OP3 (3)) H (OC(3)) The kernel is at least 5 dimensional, and so there is a cubic containing C but not Q. Then, S ∩ Q is degree 6, hence equal to C→. Conversely, by adjunction, any smooth curve of the form S ∩ Q = C is a canonical curve of genus 4. Now, our curve is realized as a curve of type (3, 3) on Q =∼ P1 × P1. Now, by geometric Riemann Roch, we have three points on the curve which are linearly dependent. So, we’re asking whether the canonical curve contains three colinear points. That is, each line in the quadric meets the line in a divisor of degree 3. So, C is trigonal, and if Q is smooth, there exist 2 such maps, while if Q is singular, the curve is only trigonal in 1 way.
8. 9/18/15 8.1. Introduction to parameter spaces. Today: Jacobians. Application: every curve of genus g admits a nondegenerate embedding C Pn of degree g + 3. One common question is when will there exist a function of degree d. In genus 3 and 4, all curve have functions of degree 3, but this is not→ true in genus 5. Jacobians are examples of parameter spaces. Fix C and let (1) Picard variety Picd(C) = { Line bundles of degree d on C } . 16 AARON LANDESMAN
(2) Hilbert scheme r Hd,g,r = { curves C ⊂ R degree d, genus g }
(3) Moduli space of curves Mg.
Mg = { isomorphism classes of smooth projective curves of genus g } This allows us to talk about the dimension of the family of line bundles on C. This is called the Picard variety or Jacobian. 8.2. Moduli Spaces. Here is a problem Joe got wrong on the quals, which led him to get a conditional in calculus, though it was actually intended by Barry Mazur to be a problem in algebraic geometry. The problem is to computer dx √ x2 + 1 Z The intention of Mazur was to look at y2 = x2 + 1 which is a hyperbola, and 2 t 7 2t t +1 observe this is a genus 0 curve, this curve is parameterized by 1−t2 , 1−t2 1 dx 1 which determines a map P C yielding y and when we pull it back to P R(t) dt → it becomes and then we compute theR integral of this rational function on P1. → R The next integral that was considered was something like dx √ x3 + 1 Z dx 2 3 which we can think of as y on the Riemann surface y = x + 1. This integral isn’t quite well defined, since it depends on the path you take. R Say we have a smooth projective curve C of genus g. Given p, q ∈ C, we want q 0 ∨ to make sense of p ∈ H (K) /H1(C, Z) = J(C). which is called the Jacobian of C. We can look at R 0 ∨ H1(C, Z) H (K)
Z2g Cg
Fix a base point p0 ∈ C. We get a map C J p p 7 → 0 Zp which is non canonical because it depends→ on a basepoint. More generally, for any d we get a map µ.
µ : Cd J pi D = pi →7 0 i p X X Z 0 Theorem 8.1. For D, D ∈ Cd we have → D ∼ D0 µ(D) = µ(D0)
⇐⇒ NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 17
This implies that J =∼ Picd(C). Proof. Idea: The forward direction is quite easy. If we start with D, D0, the con- dition they be linearly equivalent means there is a family interpolating between them. If they are sections of the same line bundle L, then we can take linear com- binations of D, D0 parameterized by the two divisors. That is, there is a family 0 {Dt}t∈P1 where D0 = D, D = D and 1 ∞ µ: P J t 7 µ(Dt) → Now, since J is a complex torus, but the pullback of a form to P1 is 0, because P1 → has no holomorphic 1 forms. Then, since 1 forms generate on J (in some sense, I’m not sure exactly how) and so if they all pull back to 0 the map must be 0. Remark 8.2. The hard part of Theorem 8.1 was proven by Clebsch, even though it is known as Abel’s theorem We have an isomorphism Picd(C) =∼ J noncanonically, which is why we will often want to write Picd(C) instead of J, though we won’t worry about that too much now. Definition 8.3. Define r d 0 Wd(C) = L ∈ Pic (C): h (L) ≥ r + 1
Exercise 8.4. For L ∈ Picd(C) general, then h0(L) = max (0, d − g + 1). Solu- tion: The idea is to equivalently formulate it as follows. If D ∈ Cd is a gen- eral effective divisor, we claim h0(L) = max (1, d − g + 1). The proof of this, d 0 0 D = i=1 pi. We start with h (K) = g, h (K − p1) = g − 1, and repeating we see h0(K − D) = max(0, g − d), supposing we choose the p so that they are as P i independent as possible. Riemann-Roch then implies the statement. So, we only need deal with the noneffective case. If d ≥ g the map µd is surjective. We only 0 need show if 0 ≤ d ≤ g then µd is generically 1-1. In particular, Wd ⊂ J is a closed subvariety of dimension d. The proof of the statement for d ≤ g − 1 follows from Riemann-Roch applied to the statement for d > g − 1.
Lemma 8.5. Say C is an arbitrary curve of genus g. If L ∈ Picg+3(C) is general, so h0(L) = 4, we get a map C P3. We claim this is an embedding.
Proof. If φL is not an embedding, then there exists an effective divisor of degree 2, this is equivalent to the existence→ of an effective divisor of degree 2 D = p + q with 0 0 2 h (L(−D)) = h (L) − 1 = 3. This would imply L(−D) ∈ Wg+1. Note that the set 2 0 0 Wg+1 = K − Wg−3 by Riemann Roch. So, the existence of such a divisor D ∈ W2 0 0 0 with h (L(−D)) = 3 implying L(−D) ∈ K − Wg−3 which implies L ∈ W2 + (K − 0 Wg−3). So, the set of all such sums has dimension at most 2 + (g − 3) = g − 1. So, a general line bundle of degree g + 3 will not lie on this g − 1 dimensional locus, and will therefore determine an embedding, i.e., be very ample. 18 AARON LANDESMAN
Remark 8.6. We showed the above to show the existence of parameter spaces can allow us to prove theorems about the objects we’re concerned with. We’ve been accumulating many instances where we invoked these sorts of facts. On the home- work, we saw a line bundle of degree 5 determines something if and only if D is of the form D = 2K + p, which is a one dimensional space of line bundles inside a 2 dimensional space of line bundles. On Monday we’ll talk about canonical embeddings of curves of genus 5 and 6.
9. 9/21/15 Today: (1) Hyperelliptic (2) Counting moduli (3) canonical curves of genus 5 and 6 9.1. Hyperelliptic curves. Let C be a hyperelliptic curve of genus g. Then, there is 1 a map π : C P of degree 2. We have exactly 2g + 2 branch points α1, ... , α2g+2. Lemma 9.1. There exists a unique curve C with a map π : C P1 branched at → α1, ... , α2g+2. Proof. Uniqueness is shown below via a topological argument. To→ show existence, 2 we can write down C = V(y − i(x − αi)), and specify a transition function to the other chart, as in Vakil’s foundations of algebraic geometry. Alternatively, just Q complete this as a Riemann Surface. Given a Riemann sphere and 2g + 2 points, draw arcs from a given branch point to all other branch points. If we excise these arcs, we obtain two disjoint sheets. We now want to complete it. We can describe the surface complex analytically as follows: Every time we go around a branch point the sheets are exchanged. This constructs the complex analytic structure. The complex analytic structure comes from pulling back from P1. Question 9.2. Suppose we have a three sheeted cover of the Riemann sphere sim- ply branched over 2g + 4 points. How can we describe the structure of that Rie- mann surface? The question is to describe the monodromy group at the branch points. Two of them are exchanged, and there is one sheet containing the simple point. If we are given the branch points, to specify the cover, we only have to topologically de- scribe the cover, which is equivalent to describing the monodromy group around the branch points. That is, the monodromy group is generated by transposition. We also know that loops around all branch points has a product which is trivial in the fundamental group. So, we have transpositions τi ∈ S3, we know i τi = id and hτii is transitive. Hence, hτii = S3, or equivalently, there are at least two Q τi. Furthermore, such τ1, ... , τ2g+4 is determined up to simultaneous conjuga- tion, which corresponds to relabelling the sheets. We can now replace degree 3 by degree d. Now, if the map is of degree d, we would have α1, ... , α2g+2d−2 branch points, and we get a similar description. Here, we get similarly transpositions τi with hτii transitive and i τi = id. This is the only relation because we know the funda- mental group π (S2 \ α , ... , α . Note that the order of this product Q1 1 2g+2d−2 NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 19 is important, and depends on the paths chosen between the points. The number of solutions to this combinatorial problem is finite, and it has been worked out. This was calculated in degree 3 by Riemann and Hurwitz, and these numbers are Riemann-Hurwitz numbers. Question 9.3. With the same setup as in Question 9.2 here is an open question: what is the number of covers with more complicated branching locus (i.e., non- simple branching locus)? Remark 9.4. If d = 2 in the above question, Question 9.2, there is a unique such Riemann surface, since there is a unique collection of such τi, namely τi = (12) for all i. Question 9.5. What if we want to look at covers of curve other than P1. In fact, very few curves appear as branched curves of higher genus curves (as can be seen by counting moduli). We can ask a similar question for covers of elliptic curves: What curves appear as branched covers of elliptic curves? However, going around a branched point is not the only way to get from one sheet to another. One can also move around closed loops on the torus, not ho- mologous to the identity (i.e., around each of the two loops of the torus). So, all together, we have to specify two more generators, of the fundamental group of the torus, in addition to those of the branch points. We will get transpositions for the branch points, and then two more arbitrary permutations corresponding to the two loops. Call these τi, µ1, µ2. We get a relation i τi = [µ1, µ2], since the complement of the µi is a rectangle, and we want to go around the boundary of the rectangle. Q Remark 9.6. The answer to this Question 9.5 is also open, and this question even came up in Dawei Chen’s thesis.
Question 9.7. (Algebra Problem) Given α1, ... , αb, where b = 2g + 4, we want a 3 2 cubic polynomial in y with coefficients in C[x] of the form y + α2(x)y + α1(x)y + α0(x) with discriminant i(x − αi), and here it’s not clear that we can even make such a curve, but we can do so via Riemann surfaces quite easily, and it’s equation will necessarily have thisQ form. So, a hyperelliptic curve is either 2g+2 2g+1 2 2 V(y = (x − αi)) or V(y = (x − αi)) i=1 i=1 Y Y depending on whether the curve has a branch point at or not. So, in the first case, we add point q, r at . To describe this map, we want to write down the canonical map, and we can do this by writing down a differential.∞ We can write down the differential∞ dx. We now ask, what is its divisor? In the plane, the divisor is the sum of the ramification points. Let pi be (αi, 0) ∈ C. Then, (dx) vanishes at the pi, so
(dx) = pi − 2q − 2r i X where we computed the orders at infinity by symmetry, and degree considera- 1 tions. To obtain a holomorphic function, we may note ( y ) = pi + (g + 1)(q + r). P 20 AARON LANDESMAN
We can now take products and note that dx = (g − 1)(q + r) y When g ≥ 2, this is holomorphic. We can generate all differentials by multiplying by multiples of x. We will get g holomorphic differentials dx x dx xg−1 dx , , ... , y y y g−1 g−1 In this case, the canonical map φK : C P is the map given by 1, x, ... , x . In particular, the canonical map factors through the two sheeted cover as → C π P1 (9.1) φK φ|O (g−1) P1 Pg−1
So, the canonical map is 2 : 1 onto the rational normal curve. Suppose we want to embed this curve in projective space. We’ll discuss this next time. Exercise 9.8. Every smooth curve admits an embedding of degree g + 3. If C is hyperelliptic of genus g ≥ 2, then the smallest degree of an embedding C Pr is exactly g + 3. Remark 9.9. In some sense, hyperelliptic curves have the most linear series→ on them, but they have the fewest embeddings.
10. 9/23/15 10.1. Hyperelliptic Curves. Recall from last time that if C is hyperelliptic of genus g, then C is given by y2 = f(x) over A1 with deg(f) = 2g + 2. Note of course that such a curve is affine, but such an affine curve has a unique completion to a smooth projective curve, by adding in two (or one) point at infin- ity, by taking its closure in P2 and resolving the singularities. We can explicitly see that dx xg−1 dx H0(K ) = h , ... , i C y y In other words, the canonical map factors through the hyperelliptic map to P1.
φ C Pg−1 (10.1) π νg−1
P1
Definition 10.1. A line bundle L or divisor D or linear system |L| is special if h0(K − D) 6= 0 NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 21
Now, if we have a special divisor on a hyperelliptic curve, that means the di- visor is contained in the preimage of a hyperplane. So, a divisor D ∈ Pic C, we know that their image under the canonical map is linearly independent unless two points map to the same point. So, by geometric Riemann Roch, the number of linear relations is exactly the number of pairs of points in D which map to the same point. That is, this is pre- 1 cisely the multiple of the g2 that D contains. 1 In conclusion if D is special, we can write |D| = r · g2 + D0, where D0 is a fixed base locus. That is, the map φD given by any special D factors through π, and, in particular, there are no special, very ample linear series. Lemma 10.2. The smallest degree of an embedding of a curve C Pr is g + r. Note that hyperelliptic curves cannot be embedding into P2. But, if r ≥ 3, a general map will be an embedding. → Proof. Proved above. Exercise 10.3. Conversely, it is also true that every nonhyperelliptic curve has an embedding of degree less than g + r. 10.2. Gonal Curves. Definition 10.4. We say C is trigonal if there exists a degree 3 cover f : C P1. Definition 10.5. We say C is tetragonal, pentagonal, hexagonal if there exists a degree 4,5,6 cover f : C P1. → Definition 10.6. Say C is k-gonal if there exists a g1. → k Warning 10.7. Be careful, it’s often unclear whether n-gonal curves are also n − 1 gonal curves. That is, it is unclear whether we allow base points. Remark 10.8. No one knows more about trigonal curves than Anand Patel. Definition 10.9. The moduli space of genus g curves is
Mg = { isomorphism classes of smooth projective genus g curves}
Remark 10.10. The existence of Mg wasn’t proved until 1969, even though people had been working with it a century earlier. For now, we’ll pretend we know what we mean by this.
Question 10.11. How do we calculate the dimension of Mg?
Question 10.12. Is Mg irreducible? Understanding abstract curves is difficult, but we can understand the situation better by examining another moduli space: Definition 10.13. Define the Hurwitz space 1 Hd,g = (C, f): C ∈ Mg, f : C P of degree d with simple branching
The one thing we can see for such a curve in H is the number of branch → d,g points.
Lemma 10.14. dim Mg = 3g − 3. 22 AARON LANDESMAN
Proof. Set b = 2d + 2g − 2 and the branch divisor will consist of an unordered b b tuple of distinct points. So, we obtain a map Hd,g P \ ∆, where the image 1 corresponds to a divisor of degree b on P . We now endow Hd,g with the structure of an algebraic variety via the covering map to Pb.→ We next want to utilize the projection map Hd,g Mg by forgetting the map. Now, the dimension of Hd,g = b = 2d + 2g − 2. The question we now→ want to ask is, what is the fiber dimension of the fiber to the map to Mg? We can’t answer this in general, but when d > 2g − 2. We are now asking, how many simply branched maps f : C P1 of degree d are there? To specify such a map, we have to choose: (1) A line bundle L ∈ Picd(C), which→ is g dimensional, 0 (2) A pair of sections σ0, σ1 ∈ H (L), up to multiplication (which is basepoint free), which has dimension 2(d − g + 1). So, the dimension of the fiber is g + 2(d − g + 1) − 1. We should worry that Hd,g has the correct structure for both the projection maps, but we won’t worry for now. So, the dimension of Mg is 2d + 2g − 2 − (2d − g + 1) = 3g − 3.
Remark 10.15. We can also think of Pb to be polynomials of degree b, and ∆ is the zero locus of the discriminant. Exercise 10.16. Show that unordered b tuples of points without repetition corre- spond to Pb \ ∆. Guess at a Solution: This is precisely symmetric functions in (P1)b, by the fun- damental theorem of symmetric functions.
Remark 10.17. The reason this method for computing dim Mg worked so well was because we introduced some auxiliary information that allows us to get a handle on the curve C.
Lemma 10.18. The dimension of the space of hyperelliptic curves is H2,g, which is 2g − 1 dimensional
Proof. The map H2,g { hyperelliptic curves } have three dimensional fibers. Therefore, the set of hyperelliptic curves is 2g − 1 = 2g + 2 − 3. → Corollary 10.19. Not all genus g curves, for g ≥ 3, are hyperelliptic.
Proof. The dimension of the space of hyperelliptic curves is less than that of Mg.
Next time: curves of genus 5 and 6.
11. 9/25/15 Today: Canonical curves of genus 5 and 6 Monday: Adjoint series Starting Wednesday: Castelnuovo theory, chapter 3 in ACGH Let C be a smooth projective non-hyperelliptic genus 5 curve. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 23
4 11.1. Curves of genus 5. We have a canonical map φ = φK : C P . We first consider which quadrics vanish on C. → 0 0 H (OP4 (2)) H (OC(2)) which is a map from a 15 dimensional vector space to a 12 dimensional vector → space, so the kernel is at least 3 dimensional. That, is C lies on at least three quadrics. There are now two possibilities: (1) So, it might be that curves are an intersection of three quadrics. In this situation, C does not lie on any more quadrics, by the Noether af+bg+ch theorem. (2) It can also be that C is a strict subset of ∩iQi. Let us now make four observations:
(1) Case 1 does occur because if Qi are three general quadrics, then by Bertini, C = ∩iQi is smooth, and by adjunction, we have KC = KP4 (2 + 2 + 2) = OC(1). 1 (2) In case 1, C is not trigonal. If C were trigonal, then there exists a g3, so the curve contains three collinear points. But, if we have three collinear points, every quadric would contain the line through those three points. So, if C is trigonal, then it would not be an intersection of quadrics. (3) If C has genus 0, C is a 1-sheeted cover of P1, but in genus 1 or 2, then C is a two-sheeted cover of P1 (in genus 1, take any degree 2 series, in genus 2, take the canonical linear series). General curves of genus 3 and 4 are expressible as 3 sheeted covers of P1: In genus 3, we can project the plane 1 quartic away from a point, that is, the linear series K − P is a g3. In genus 1 4, the rulings of a quadric cut out the g3 on C. It turns out general curves of genus 5 are expressible as 4 sheeted covers of P1. 1 Note further that in genus 1 there is a 1 dimensional family of g2, in 1 genus 2, there is a finite number of g2. In general, in odd genus there is a 1-dimensional family of such maps, and in even genus there is a 0- dimensional family of such maps. (4) How many curves are complete intersections of quadrics in P4. Such 0 curves arise as an open subset of G(3, H (OP4 (2))) which is 3 · (15 − 3) = 36 dimensional. The fibers of the map
1 U = Λ ∈ G : C = ∩Q∈ΛQ is a smooth curve in P M5
PGL Are 5. We can do a similar count to show the trigonal→ curves form a 10 dimensional subvariety. Note that 3 · g − 3 + g2 − 1 = 36, so indeed, since Mg is irreducible, a general curve will be expressible as such a complete intersection. Quote from Joe: “We’ll answer the question of whether C is tetragonal, or I’m not doing my job.“ We’ll see if he gets to it. Question 11.1. Does the second case, in which C is not equal to an intersection of quadrics, occur? 24 AARON LANDESMAN
Note that trigonal curves exist, and so they must be as in this second case. Start 1 with a trigonal curve of genus 5. Given |D| a g3, let’s look at |K − D|. By Riemann 2 2 Roch, this is a g5. Note that if this curve has a base point, the curve has a g4, so it would either give a two sheeted cover of P1 (impossible since C is not hyper- 2 elliptic) or a plane quartic (impossible since g = 5). Therefore, C C0 ⊂ P , and the image is not smooth, since a smooth plane quintic is genus 6. Let’s now 2 0 look at |2D|, which is a g6. To see h (2D) = 3, note that it is at most→ dimension 3, by Riemann Roch, and at least 3 because the map Sym2H0(D) H0(2D) is injective, because the curve, when embedded in P1 is surjective, so does not lie on any quadrics. → Lemma 11.2. Now, if we take K − 2D = p + q, we claim f(p) = f(q).
Proof. Note that 2(K − D) = K + p + q, so the canonical linear series KC is cut out on C by conics Q ⊂ P2 passing through R. To see this, a conic Q ⊂ P2, the preimage f−1(Q) in C is a divisor of the form K + p + q. If r ∈ V(Q), we can write f−1(Q) = D + p + q where D ∼ K, and D is the residual 8 points of intersection of Q with the curve. Furthermore, we get all canonical divisors in this way because C has genus 5 and we have a 5 dimensional space of conics passing through a point in the plane. So, we now have a very interesting description of C: We can map
2 C C0 ⊂ P (11.1) φK π P4 where the map π is given by quadratic polynomials vanishing at r. Now, we have the C lies on a surface in P4. Let S = im P2 under the right map π. Then, we have that the curve lies on a surface. This turns out to be the projection of the 2 Veronese ν2 surface from a point. We can also say that we can resolve the map P2 S by blowing up at the point r. One way to describe this surface in P4 is to take a line and a plane conic living in complementary spaces spanning P4, take an isomorphism→ between them, and then take the union of the lines joining points on the line and points on the conic. Now, lines through r meets the curve at r and three other points. This deter- mines a trigonal map. So, the lines on the ruling of S meet the curve three times, and the intersection of the quadrics is precisely the surface S.
12. 9/28/15 Today: Adjoint series - appendix A Wednesday: Castelnuovo’s theorem - Chap- ter 3 But first canonical curves of genus 5
12.1. Canonical curves of genus 5. Assume C has genus 5 and is not hyperelliptic. 4 Let C P be a canonical map. Then, C lies on three quadrics Qi. Last time, we saw
(1)→C = ∩iQi. (2) C ⊂ ∩iQi = S, where S is a cubic scroll. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 25
As we saw last time, the second case corresponds to that were C is trigonal, a three sheeted cover of P1. The first case corresponds to that where C is not trigonal, but C is expressible as a 4 sheeted cover of P1. Lemma 12.1. Every canonically embedded curve which is not trigonal admits a map of degree 4 to P1. Proof. Let P2 =∼ { quadrics C ⊂ Q}. We can ask whether there are singular quadrics in this set. Inside P14 which is the space of all quadrics in P4, there is a quintic hypersurface of singular quadrics. Singular quadrics in P4 are cones over quadric surfaces, and so the cone over a quadric has two rulings by 2 planes. Two 2- planes from opposite rulings add up to a hyperplane section on the quadric cone. Altogether they intersect the curve in 8 points, so one of them must intersect in 4 or fewer points. The intersection of the 2 plane and the curve is the intersection with the two other quadrics, which is 4 points. So, each 2 plane meets C in four 1 points. So, we get two g4’s on the curve and they are residual to each other (add to the canonical divisor). In the rank 3 quadric case, it is a cone over a plane conic. In this case, we get a 1 g4 which is semicanonical. In the trigonal case, you can write down the quadrics containing the curve, and they are all singular. 1 Question 12.2. Why do all g4’s come from quadrics in the above proof? Answer: If we have a canonical curve with 4 points D lying in a two plane Λ , look at the restriction map from quadrics in P4 containing C quadrics in Λ ⊂ P2 ⊃ D There is a 1 dimensional family of conics containing D, and the above map is → 1 surjective. Therefore, the 2-planes on the quadric gives us all g4’s. Remark 12.3. The converse, showing that all curves on the Segre cubic surface are trigonal, we’ll see this from Castelnuovo theory. 12.2. Adjoint linear series. We’ll start by talking about plane curves, following three steps: (1) Smooth curves C ⊂ P2. Even when the genus is a binomial coefficients, most curves are not plane curves. E.g., smooth plane quintics have dimen- sion 12 while all curves are genus 6 are 15 dimensional. (2) Nodal curves C C0 ⊂ P2. Given any smooth curve, there exists a birational embedding into the plane so that the image is a nodal plane curve. → (3) Arbitrary plane singularities. 12.2.1. Smooth plane curves. Choose homogeneous coordinates [X, Y, Z] on P2 and X Y let L be the hyperplane Z = 0, and let x = Z , y = Z be affine coordinates on P2 \ L =∼ A2. Say∞ C ⊂ P2 is a smooth curve of degree d with C ∩ A2 the zero locus of f(x, y). Let’s∞ further assume that [0, 1, 0] ∈/ C, so the projection from [0, 1, 0], π : C P1, (x, y) 7 x expresses C as a d sheeted cover of P1. → → 26 AARON LANDESMAN
Question 12.4. How do we write down all holomorphic differentials on C? Let’s start with the differential dx, which is holomorphic in the affine plane. However, this will have poles on L . Now, by pulling back D = L ∩ C. Then, (dx) = π−1( ) = −2D . We have the poles of dx now, we must ask where (dx) is 0. We want to find a function with∞ zeros at and poles at the∞ zeros∞ of dx. These ∂f(x,y) ∞ zeros are where∞ ∂x = 0. We know df|C = 0 since f is zero along the∞ curve. But, we also know ∂f ∂f dx + dy = df| = 0 ∂x ∂y C ∂f ∂f Since C is smooth, we know ∂x , ∂y have no common zeros on C. ∂f ∂f At a point p where dx = 0, we have ∂y (p) = 0, ∂x (p) 6= 0, we have ∂f ord dx = ord p p ∂y ∂f So, we can divide by ∂y without introducing any new poles. So, we consider the differential dx ∂f ω0 = , fy = fy ∂y ∂f Now, ω0 is holomorphic and nonzero in the affine plane. Now, ∂y is holomor- phic of degree d − 1, so the divisor must be (ω0) = (d − 3) · D . This is a very concrete form of the adjunction formula, since KC = OC(d − 3). We have a differential with zeros at infinity. If we multiply∞ by a function, it must be a polynomial in the affine plane, which will be holomorphic (even at ) as long as n ≤ d − 3. So, consider g(x, y) dx ∞ : deg g ≤ d − 3 ⊂ H0(K ) f C y d−1 Both sides have dimension 2 and so the inclusion is an equality, and every holomorphic differential is of this form. ∼ Here is a more modern proof of the above: Equivalently, by adjunction, KC = OC(d − 3). We have an exact sequence
(12.1) 0 OP2 (−3) OP2 (d − 3) KC 0
1 0 0 and since h (OP2 (−3)) = 0, we have that the map H (OP2 (d − 3)) H (KC) is surjective, and in fact defines an isomorphism.
2 → 12.2.2. Nodal Curves. Now, suppose we have C C0 ⊂ P with C0 nodal. For the moment, suppose C0 is chosen so that at the nodes, the branches of C0 do not have vertical tangent. The problem with the previous→ argument for smooth curves ∂f ∂f is based on the assumption that 0 = df|C = ∂x dx + ∂y dy. When C was smooth, these partials could not simultaneously vanish. How- ever, in the singular case, the partials vanish to order 1. We’re assuming that the nodal curve has no vertical tangents by our assumption, meaning that dx, dy are nonzero. (We assume this just for simplicity, we could do the same argument where they do have zeros there.) NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 27
ω = dx ω When we consider 0 fy , we have 0 will have simple poles at the points qi, ri, are the two preimages of each of the nodes pi so we must take g(x, y)ω0 when g(x, y) had degree at most d − 3 and g vanishes at qi, ri. Now, if there are d−1 δ nodes, we impose δ conditions on g, so the dimension is at least 2 − δ, but this is also the genus of the curve. Hence, we obtain all differentials in this way, and we also see that the nodes impose independent conditions on the vanishing of g.
13. 9/30/15 13.1. Program for the remainder of the semester. Through many examples, we have some understanding that linear series are the key to understanding the ge- ometry of curves. We phrase this in terms of three questions, in increasing order of depth and subtlety. r (1) What linear series gd may exist on a curve of genus g? r (2) What very ample or birationally very ample linear series gd may exist on a curve of genus g? r (3) What linear series gd exist on a general curve? Answers to above questions (1) Answered by Clifford’s theorem (2) Answered mostly (in the sense that it only answers the question of bira- r tionally very ample maps) by Castelnuovo’s theorem. For very ample gd’s, we know the answer for r = 3, which is relatively recent, r = 4 might be known, but r ≥ 5 is still open. (3) This is answered by Brill Noether theory We’re quite interested in relations between curves in the abstract and curves in projective space. We’re really interested in embeddings of the curve in projective space, or birational embeddings, and much less so maps to projective space which aren’t generically injective.
13.2. Adjoint linear series. Remark 13.1. There’s a version of the following that applies to arbitrary reduced curves (which is quite similar to the nodal case) and is written up in appendix A in ACGH. ν 2 Recall the set up from last time: Suppose C − C0 ⊂ P is the normaliza- tion map of C, and C0 is a nodal plane curve of degree d. Say C0 has nodes at −1 p1, ... , pδ, and let ν (pi) = qi ∪ ri. Set ∆ = i(→qi + ri) ∈ div C. Choose coordinates [X, Y, Z] on P2 and let L = V(Z), and x = X/Z, y = Y/Z 2 ∼ 2 ∗P 2 be coordinates on A = P \ L . Let D = ν L . In A , let us say that C0 = V(f) where f(x, y) is a polynomial of degree d. Suppose∞ further that at each node of the curve, the tangent lines are not∞ vertical,∞ so that∞ the divisors dx, dy have simple poles at those points. We now have two goals (1) Find an algorithmic way of writing down all differentials on C (2) If I have a divisor on C, how do we find the complete linear series associ- ated to that divisor, algorithmically. 28 AARON LANDESMAN
Lemma 13.2. We can algorithmically write down all regular 1 forms by
∗ g(x, y) dx W := ν , deg g ≤ d − 3, g(pi) = 0. fy Proof 1. Start with dx. This has poles along L . To kill the poles at , we consider dx f = ∂f dx f fy where y ∂y , because wherever has a zero y does as well, at a point on the curve. Therefore, polynomials of the form∞ ∞ g(x, y) dx W := ν∗ , deg g ≤ d − 3 fy so long as we are away from the nodes. But at the nodes dx doesn’t vanish, while fy does vanish. For this, we need g(pi) = 0. Then, by a dimension count, the inclusion 0 W H (KC) is an isomorphism. → We can also present a coordinate free proof. Proof 2. Consider the fiber product
2 C S = Blp1,...,pδ P (13.1) π
2 C0 P .
In S, we have exceptional divisors Ei and E = Ei. Let L be the pullback of the class of a line in P2. Then, the E together with L generate the Picard group of S. i P Now, C ∼ dL − 2E ∈ Pic(S)
KS ∼ −3L + E By adjunction,
KC = (C + KS)|C = (d − 3)L − E|C = (d − 3)D − ∆
We can now show we get all differentials in the modern language∞ of sheaves by using the following exact sequence: (13.2) ∼ ∼ 0 IC((d − 3)L − E) = O(−3L + E) = KS OS((d − 3)L − E) OC((d − 3)L − E) = KC 0
1 1 And we now use that h (KS) = h (OS) = 0 by Kodaira Serre duality, and so the induced map on global sections is surjective. An additional fact is that 0 0 H (OS(mL − E)) H (OC(mD − ∆)) m completeness of adjoint series is surjective for all , and this is referred to as the ∞ . Now, suppose we are given a divisor→ on C. Suppose we’re given a divisor B on C of degree b. Assume that Supp B ∩ ∆ = ∅, for notational convenience. Question 13.3. Describe the complete linear series |B|. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 29
Solution: Choose any plane curve G of degree m so that p1, ... , pδ ∈ G and B ⊂ G. Now, consider ν∗G = B + ∆ + R. Now, the divisor R has degree md − b − 2δ. Consider all plane curves of degree m = deg G, with p1, ... , pδ ∈ H, R ⊂ H. For any such H, we have ν∗H = ∆ + R + S. Here R ∼ m · D − B − ∆ and so S ∼ mD − ∆ − R ∼ B ∞ B B = B+ − B− So, this shows us how to find the complete∞ linear series of . If a difference of effective divisors, we can choose G to contain B+ and H to contain B−, and we still obtain the desired answer. Remark 13.4. (History of genus) What was the genus to the ancients? People looked at lines and conics, and if you have any d points on a line, those points are the intersections of a line and a curve of degree d. Any 2d points on a conic, the 2d points on a conic are in general the intersection of a plane curve of degree d and a conic. However, it is not true that 3d points on a cubic lie on the intersection with a curve of genus g. Rather, we will only find 3d − δ, where δ is some deficiency. Now, people defined genus as this deficiency. The idea that for each node δ, we get a 2 for 1 deal, in the sense that passing through a node drops the degree of the residual intersection by 2 but only imposes 1 linear condition, so the original definition of genus was d − 1 − δ 2
14. 10/2/15 Weierstrass points Let C be a compact Riemann surface of genus g ≥ 2. Application: Prove that automorphism group of such a Riemann surface is fi- nite. Definition 14.1. Define
Sp = {− ordp(f): f holomorphic on C \ p} . This is a semigroup, and is called the Weierstrass semigroup of p ∈ C. The gap sequence Gp is N \ Sp. That is, the gap sequence is the order of poles that don’t occur.
Lemma 14.2. We have #Gp = g. Proof. Observe 0 0 Sp = m : h (mp) > h ((m − 1)p)
0 0 Gp = m : h (mp) = h ((m − 1)p)
Now, let’s compare h0(mp) and h0(K − mp). By Riemann Roch, as we increase m by 1, precisely one of these two quantities changes. Lemma 14.3. For general p ∈ C, we have
Gp = {1, ... , g} Sp = {0, g + 1, g + 2, ...} Proof. Follows from Proposition 14.8. 30 AARON LANDESMAN
TABLE 1. Chart of m and gap sequence
m h0(mp) $hˆ0(K-mp) 0 1 g 1 1 g-1 2 1 or 2 g-2 or g-1 ...... m m-g+1 0
Proposition 14.4. There exists a finite set of points with a function with a pole of order less than g + 1 at p, and no other poles.
Proof. Follows from Proposition 14.8.
Definition 14.5. A point p ∈ C is a Weierstrass point if the gap sequence is not Gp = {1, ... , g}. g Definition 14.6. The weight of p ∈ C is the sum w(p) := i=1(ai − i). Example 14.7. (1) For example, if the weight is 0, theP cap sequence is 1, ... , g, and the point is not a Weierstrass point. (2) If w = 1, then Gp = {1, ... , g − 1, g + 1} (3) if w = 2 we have Gp = {1, ... , g − 1, g + 2} or {1, ... , g − 2, g, g + 1}, and the former cannot happen in genus 3, because it is not a semigroup. Proposition 14.8. For any C, we have
w(p) = g3 − g p∈C X 1 Example 14.9. (1) If g = 2 we have φKC P of degree 2, and it has 6 branch points, which are the 6 Weierstrass points, each with weight 1. (2) In g = 3, if C is hyperelliptic, we have→8 branched points of ΦK where the semigroup is necessarily
Sp = {0, 2, 4, 6, 7, 8, 9, ... , } Gp = {1, 3, 5}
and w = 3, which sums to 24 as claimed. (3) If g = 3 and C is not hyperelliptic. In this case, having a Weierstrass point, we are precisely asking when h0(3p) > 1. Then, there is a line in plane containing the divisor 3p, so we are asking for flex points of C. That is, by Geometric Riemann-Roch, the divisor 3p ⊂ C lies in a line, and p is a flex point of C ⊂ P2. The set of flex points are the points of intersection with the Hessian, and so the Hessian is a sextic plane curve, which meets C of degree 4. Further, the intersection multiplicity of the point with the Hessian is precisely the weight of the point. Overall, we get weight 4 · 6 = 24, as desired.
Proposition 14.8. Let p ∈ C, and choose a local coordinate z around p. Choose a 0 basis ω1, ... , ωg ⊂ H (K). Write ωi = fi(z) dz. Being a Weierstrass point means NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 31 there is a differential vanishing to order g at p. We look at the Wronskian f1 ··· fg 0 0 f1 ··· fg ∆ = det . .. . . . . g−1 g−1 f1 ··· fg This vanishes at p precisely when p is a Weierstrass point: To see this, note that h0(gp) > 1 if and only if h0(K − gp) ≥ 1, which is the same as the existence of a holomorphic differential which vanishes to order g at p. So, Weierstrass points are zeros of ∆. Next, we claim ∆ is a section of a power of the cotangent sheaf. Now, if we choose a different local coordinate, we want to see how the Wron- skian changes over overlaps. Chose a different local coordinate w. The first row of 2 ∂z ∂z ∆ is multiplied by ∂w . The second row is multiplied by is multiplied by ∂w , plus some multiple of the first row. Hence, repeating this, the determinant is mul- g+1 ∂z ( 2 ) tiplied by ∂w . Now, note that these are the transition functions for a power ⊗ g+1 of the cotangent bundle. Hence, ∆ ∈ Γ(K ( 2 )). Hence, the number of zeros of ∆ g+1 3 is the degree of this line bundle. So, this is degree (2g − 2) 2 = g − g. It still remains to show the order of the Weierstrass function is equal to the weight of the Wronskian. This is left as an exercise, as is written in ACGH and 3264, chapter 7. Remark 14.10. We don’t yet know that the Wronskian determinant is nonzero. However, in characteristic p, the Wronskian determinant actually can be 0. How- ever, this doesn’t happen in characteristic 0. Remark 14.11. The maximum possible weight of a Weierstrass point corresponds to the semigroup which starts increasing as soon as possible. That is, Sp = {0, 2, 4, 6, ... , 2g, 2g + 1, 2g + 2, ...} g which has gap sequence Gp = {1, 3, 5, 7, ... , 2g − 1} which has weight 2 . This oc- curs if and only if the curve is hyperelliptic. We further see there are 2g + 2 such points, and so these account for all Weierstrass points Corollary 14.12. There are at least 2g + 2 Weierstrass points, and exactly 2g + 2 if and only if C is hyperelliptic.
Proof. This is immediate from Remark 14.11. Question 14.13. Here are some open problems on Weierstrass points. (1) Which semigroups of Weierstrass points occur? For a long time it was conjectured that every gap sequence occurs. However, Bookveits exhibited a semigroup which cannot occur. (2) Many of the deeper questions come from a variational point of view: Look at Mg,1, the set of pointed curves, up to isomorphism. We have a stratifi- cation of this space, by associating to a point its Weierstrass gap sequence. We know that Weierstrass points occur in codimension 1. We can then ask, for a given semigroup, what is the codimension of the locus of points with that gap sequence? We can ask if it is irreducible. 32 AARON LANDESMAN
Theorem 14.14. If C is a compact Riemann surface of genus g ≥ 2, then Aut(C) is finite. Proof. When C is hyperelliptic, we have an ad hoc argument: We can express C up to automorphism of P2, as a cover of P2 of order 2. Now, the automorphisms of P1 fixing a set of 6 or more points is finite, so we’ll just cover the non-hyperelliptic case. Now, we have strictly more than 2g + 2 Weierstrass points. Lemma 14.15. If φ : C C is any automorphism, the number of fixed points of φ is at most 2g + 2. Proof. By Riemann Roch,→ we can find a meromorphic function f on C of degree g + 1 with g + 1 poles and zeros. Consider f − φ∗f. This has poles where f has poles or the pullback of f has poles. Hence, this difference has at most 2g + 2 poles, hence at most 2g + 2 zeros, which is precisely the number of fixed points. Therefore, the number of fixed points of φ is at most 2g + 2. This suffices, because any automorphism fixing all the Weierstrass points must be the identity.
15. 10/5/15 Here are some questions for the remainder of the course? r (1) What linear series gd may exist on a curve of genus g? r (2) What very ample or birationally very ample gd’s may exist? That is, for which g, r, d does there exist (a) A smooth irreducible nondegenerate C ⊂ Pr of genus g and degree d (b) An irreducible nondegenerate C ⊂ Pr of degree d and geometric genus g? r (3) What gd’s exist on a general curve of genus g? r The problem of which very ample gd’s exist on smooth curves is still open, and the most recent progress was made by a student of Harris in the last 10 years. r r Riemann Roch says that gd is determined once 0 ≤ d ≤ 2g. So, the gd’s are only interesting in the range 0 ≤ d ≤ 2g, at the parallelogram between d = 0, g = 0 and r d = 2g − 1, r = g − 1. Then, Clifford’s theorem tells us all gd’s lie in the lower half of this rectangle, that is, below th e line 2r = d. d Theorem 15.1. Say D has degree d with 0 ≤ d ≤ 2g − 2. Then, r(D) ≤ 2 with equality if and only if (1) d = r = 0, D = 0 (2) d = 2g − 2, r = g − 1, D = K 2 (3) C is hyperelliptic ad D = mg1. Proof. Look at the multiplication map H0(D) ⊗ H0(K − D) H0(K). Now, think of these in terms of projective spaces instead of vector spaces. Think of H0(D) r 0 as a family of divisors parameterized by P , and H (K − D→) as parameterized by Pg−d+r−1. Hence, we obtain |D| × |K − D| |K| Pr × Pg−d+r−1 Pg−1 → → NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 33
This map is finite, because we’re asking how many ways we can express a canon- ical divisor as a sum of effective divisors in |D| and |K − D|. Hence, the dimension of the source is at most the dimension of the target. Therefore, r + g − d + r − 1. So, r + g − d + r − 1 ≤ g − 1 implying r ≤ d/2. The equality will follow from our upcoming discussion of Castelnuovo’s theorem. The strong form will be a way of saying when the multi- plication map is surjective (which will occur if and only if the curve is hyperellip- tic). 1 Remark 15.2. Take C hyperelliptic and take D = r · g2 + D0 and D0 to be general r of degree d − 2r, so that these points are basepoints, and we can get all gd’s on hyperelliptic curves. 15.1. Castelnuovo’s Theorem. Assume C ⊂ Pr is a smooth irreducible, nonde- generate curve of degree d and genus g. First, observe that given r, d, we have that g is bounded. Castelnuovo asks what the largest possible genus is. We won’t answer which g occur. We’ll just find the largest. Theorem 15.3. (This theorem will be stated next time, but it’s essentially a bound on the genus as discussed above) Proof idea. Here is the idea: 0 The plan is to bound from below the dimensions of h (OC(m)). We will do this by bounding from below 0 0 h (OC(m)) − h (OC(m − 1)). 0 When m is large enough, we will have h (OC(m)) = md − g + 1, and by Riemann- Roch, we will know exactly what this dimension is. This will give us a bound from above on the genus. We now carry out the proof. Proof. Let Γ be a general hyperplane section of C ⊂ Pr. Let Γ be a general hyper- plane section of C ⊂ Pr. Consider
(15.1) 0 OC(m − 1) OC(m) OΓ (m) 0 Then, 0 0 ρ 0 (15.2) 0 H (OC(m − 1)) H (OC(m)) H (OΓ (m))
0 0 Therefore, H (OC(m)) − H (OC(m − 1)) = rk ρ, where rk ρ is colloquially the 0 number of conditions imposed by Γ on H (OC(m)). Hence, rk ρ is at least the 0 number of conditions imposed by Γ imposed by H (OPr (m)). as given by the composition of restrictions
0 0 0 (15.3) H (OPr (m)) H (OC(m)) H (OΓ (m))
0 Say Γ imposes at least h conditions on H (OC(m)). This is equivalent to having 0 p1, ... , ph ∈ Γ so that for all i, there is a σ ∈ H (OC(m)) with σ(pj) = 0, j 6= 0 0 i, σ(pi) 6= 0. Now, if we take a subspace im H (OPr (m)) ⊂ H (OC(m)). Now, we’ve moved away from the curve, and the question is: 34 AARON LANDESMAN
Question 15.4. How many conditions do the d points p1, ... , pd ∈ Γ impose on hypersurfaces of degree m?
Now, we have Γ ⊂ H =∼ Pr−1 =∼ Pn, where n = r − 1. This consists of d distinct points. 0 So, the number of conditions imposed by Γ on H (OPn (m)) is hΓ (m). where n hΓ (m) is the Hilbert function of Γ ⊂ P . We know naively that hΓ (m) ≥ min(d, m − 1). Furthermore, this is the best we can do, because if all points of Γ are collinear, we have an equality. We now want to say something about the geometry of a general hyperplane section of a curve. We then obtain precisely the bound that the genus is at most g−1 2 2 . This happens precisely when the curve we start with is a curve in P The question that remains is what conditions we obtain on these lines when the curve is in Pt with t > 2. We then claim that for any curve in P2, a general hyperplane section will not contain three collinear points. The proof will be discussed next time.
16. 10/7/15 Recall from last time that C Pr be irreducible nondegenerate degree d and smooth. → Remark 16.1. We assume C is smooth purely for notational convenience, but the ν following will generalize to singular curves, and in general we can assume C − r C0 ⊂ P is a birational map and C0 may be singular. → Let Γ = C ∩ H be a general hyperplane section, intersecting C in d points. The key inequality
0 0 h (OC(m)) − h (OC(m − 1)) ≥ The number of conditions imposed by Γ on hypersurfaces of degree m in H =∼ Pr−1
= hΓ (m) where hΓ (m) is the Hilbert function, not the Hilbert polynomial, although for large m the two coincide and are equal to d. So, we would like to bound hΓ (m) from below. From this, we deduce a lower 0 bound on h (OC(m)), and so for m large, we will obtain from Riemann Roch, 0 h (OC(m)) = md − g + 1, which yields an upper bound on the genus. Question 16.2. In general, say Γ ⊂ Pn, with n = r − 1, is a collection of d points. What can we say about hΓ ? Lemma 16.3. If the points are all in a line, the points impose min(m + 1, d), which is achieved when Γ consists of d collinear points.
Proof. Let’s cover the case when d ≥ m + 1. For any k ≤ m, we can find a hyper- surface of degree m containing p1, ... , pk ∈ Γ, but no other points of Γ. For each point, we can choose a hyperplane containing 1 points and none of the others. So, intersecting these planes, we get a codimension k hyperplane. If the points are colinear, we obtain equality. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 35
Remark 16.4. We have a converse to the above lemma, which says that if hΓ (m) = m + 1 then the points must all be contained in a line. We can see this by choosing one of the hyperplanes to contain two points but no others. Remark 16.5. If we plug in the minimum Hilbert function into our strategy, we d−1 obtain g = 2 , which occurs precisely for plane curves.
So, if r ≥ 3 a general hyperplane section will not lie in a line, and so hΓ (m) will be bigger than min(m + 1, d). We now want to say something about the geometry of the general hyperplane section of an smooth nondegenerate curve. Definition 16.6. A collection of d points in Pr are in linear general position if no subset of k ≤ r points are linearly dependent. Lemma 16.7. Under the above construction, Γ consists of d points in linear general posi- tion. In the case of curves in P3, we want to show that there is a two parameter family of hyperplanes containing three collinear points. Also, there’s only a two dimen- sional family of secant lines. So, we’re asking whether all secant lines to curves can be three secant lines. This cannot happen because we know that curves can be embedded into P2 with only nodes. The statement of the lemma is, however, much more general. Remark 16.8. The proof we’ll see doesn’t work in characteristic p, and Joe isn’t sure whether it’s known whether this is true in characteristic p. In particular, the uniform position lemma is not true in characteristic p. Proof. We have C ⊂ Pr. Let’s now introduce the family of hyperplane sections, and throw away all hyperplanes which are tangent to the curve, at least for now. Define U = { hyperplanes H transverse to C } ⊂ (Pr)∨ Introduce the incidence correspondence α Φ Φ = {(H, p) ∈ U × C : p ∈ C ∩ H} C (16.1) β
(Pr)∨ U
Note that α is a d sheeted covering space, and is, in fact, a finite proper etale. It’s unramified because we removed the hyperplanes which are tangent. Above, Φ is the closure of Φ, containing hyperplanes with multiple points. It’s still flat, though not necessarily unramified. The next idea is to introduce the monodromy group. Fix a base hyperplane H0, as we vary the hyperplane, avoiding the tangent hyperplane, we can unambigu- ously follow the points. Hence, we obtain a map
π1(U, H0) Sd
Where Sd corresponds to permutations of H0 ∩ C. → Question 16.9. What is the image G ⊂ Sd of π1(U)?
Lemma 16.10. (Uniform Position Lemma) G = Sd. 36 AARON LANDESMAN
Remark 16.11. We can now describe algebraically what this means for G = Sd. Look at the incidence correspondence
Φk = {(H, p1, ... , pk) : pi is distinct in H ∩ C}
We are claiming that Φk is irreducible. Strictly speaking, this only shows it is connected, but since it’s also smooth, being a covering space of a smooth scheme, it is smooth, hence irreducible.
Proof. We’ll come back to this next time. Now, we can complete the proof of the main lemma of linear general position. We’ll just describe this in the case of curves in P3. The question is, can a general hyperplane section of the curve contain 3 collinear points. Suppose every section had three collinear points. We can vary the points so that they remain collinear. 0 More formally, let Φk be the set 0 Φk = {(H, p1, ... , pk) : p1, ... , pk are linearly dependent } 0 0 This is a closed subset. So, Φk 6=⊂ Φk, implying dim Φk < dim Φk = r, where here we are crucially using that the curve is nondegenerate. So the projection map 0 Φk U cannot dominate. → We can now deduce a better lower bound on hΓ .
17. 10/9/15 Let C ⊂ Pr be smooth nondegenerate curve of degree d and genus d. Let Γ = H ∩ C be a general hyperplane section. Today, we’ll (1) prove the uniform position lemma (2) Use this to prove a lower bound on hΓ (3) Use this to derive an upper bound on g. Lemma 17.1. (Uniform position lemma) Define
Φ = (H, p) ∈ (Pr)∨ × C : p ∈ H ∩ C
Let Φ ⊂ Φ be the subset mapping to hyperplanes U ⊂ (Pr)∨ intersecting C in distinct points. Then, Φ U is a d sheeted covering space. Fix a base point H0 ∈ U. We obtain a map → m : π1(U, H0) Aut(H0 ∩ C)
Then, m is surjective, and so G = im (m) = Sd. → Proof. We will show (1) G is twice transitive (2) G contains a transposition If G satisfies these two properties, it is the symmetric group, because it then con- tains all transpositions as given by conjugating the transposition by the twice tran- sitive action. We now show the above claims NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 37
(1) Define
Φ2 = {(H, p, q) ∈ U × C × C : p 6= q, p, q ∈ H ∩ C}
Consider the projection Φ2 C × C \ ∆. We then see that all fibers of this map are all irreducible open subsets of Pr−2 of the same dimension and hence Φ2 is irreducible. This→ means G is twice transitive. (2) Now, we want to show G contains a transposition. Choose a hyperplane r ∨ H0 which is simply tangent, so H0 ∈ (P ) \ U, so
H0 ∩ C = {p1, ... , pd−1} r ∨ and H0 is tangent to C at p1. Look at π : Φ (P ) over an analytic r ∨ (etale) neighborhood of H0 ∈ (P ) , call it V. Note that at this point, the map is not a covering map at this point. → Remark 17.2. The existence of such a hyperplane depends on k having characteristic 0. It is possible in characteristic p that every point is a flex point. (Or, every tangent line is a bitangent) This argument fails in char- acteristic p for this reason, and in this case the monodromy group will be contained in the alternating group, giving a counterexample to this lemma in characteristic p.
Now, when at H0, we have simple branching of π at the points p2, ... , pd−1 and ramification of order 2 at p1. Then, p2, ... , pd−1 have unique analytic 0 continuations q2, ... , qd−1, and there will be two points q, q in a neigh- borhood of p1. Next, note that Φ is smooth. The fibers of the projection map Φ C all have fibers isomorphic to Pr−1. Therefore, Φ is smooth. So, the preimage of U is still integral, because Φ U is a covering map. Hence, we→ can draw an arc contained in Φ the analytic open set. Then, Φ is smooth at (H0, p1), and so Φ ∩ V1, where V1 corresponds→ to the compo- nent containing a neighborhood of (H0, p1), is still integral. In particular, we can join (H, q) and (H, q0), which induces a transposition. The image of this transposition γ in U is an element of π1(U, H0) which induces a transposition on H ∩ C. Remark 17.3. The reference for this proof is Galois groups of enumerative prob- lems, duke math journal, around 1979. Question 17.4. Clearly much of the above discussion was not special to smooth curves. In particular, it generalizes to generically smooth curves. Does it general- ize even further? So, we obtain from the uniform position lemma, proven last time, that for a generic Γ hyperplane section of C ∈ Rr with r ≥ 3, the points of Γ will be in general linear position.
Question 17.5. What sort of lower bound on hΓ can we deduce from the uniform position lemma? The idea will be to take a hyperplane containing 2 of the point, and if the points are in general position, it won’t contain a third point. 38 AARON LANDESMAN
n Lemma 17.6. If Γ is a collection of d point in P in linear general position, then hΓ (m) ≥ min(d, mn + 1). Proof. Take the case d ≥ mn + 1. We claim, there exists a hypersurface X ⊂ Pn of degree m so that X contains p1, ... , p^i, ... , pmn+1 but not pi. This says the points impose independent conditions, so the Hilbert function is at least mn + 1. Now, group the points p1, ... , p^i, ... , pmn+1 into m sets Γα of n points. Take X = ∪αΓ α. This does not contain pi because the points are in general linear position.
Remark 17.7. Is the bound hΓ (m) ≥ min(d, mn + 1) sharp? You might guess we could do better if we use quadrics or cubics. However, this is in fact sharp. Exercise 17.8. Find an example of a configuration of points Γ ⊂ Pn in linear gen- eral position for which hΓ (m) ≥ min(d, mn + 1) is an equality. Question 17.9. What are the curves which achieve the above upper bound on hyperplane sections.
18. 10/14/15 Recall that we have C Pr a degree d genus g smooth nondegenerate curve. Let Γ = H ∩ C be a general hyperplane section. The basic inequality is 0 → 0 h (OC(m)) − h (OC(m − 1)) ≥ hΓ (m) The uniform position lemma tells us that if Γ is in linear general position, then
hΓ (m) ≥ min(d, m(r − 1) + 1) The basic lemma for showing this (letting r − 1 = n) then Lemma 18.1. If Γ ⊂ Pn is a collection of d points in linear general position, then the inequality
hΓ (m) ≥ min(d, m(r − 1) + 1) is sharp, and equality is achieved when Γ ⊂ C for C a rational normal curve. Proof. So, once a hypersurface of degree m contains mn + 1 points of a rational normal curve, it will contain the whole curve by Bezout’s theorem. Let’s now compute the bound on the genus. Assuming d ≥ 2r − 1, we have 0 h (OC) = 1 0 h (OC(1)) ≥ r + 1 0 h (OC(2)) ≥ 3r 0 h (OC(3)) ≥ 6r − 2 . . This continues until we obtain d < m(r − 1) + 1. Set d − 1 m = b c 0 r − 1 so write m0(r − 1) + 1 + ε with 0 ≤ ε ≤ r − 2. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 39
Then, 0 h (OC) = 1 0 h (OC(1)) ≥ r + 1 0 h (OC(2)) ≥ 3r 0 h (OC(3)) ≥ 6r − 2 . . m + 1 h0(O (m )) ≥ 0 (r − 1) + m + 1 C 0 2 0 . . m + 1 h0(O (m + k)) ≥ 0 (r − 1) + m + 1 + kd C 0 2 0
Applying Riemann Roch, for k large, we have m + 1 (m + k)d − g + 1 = h0(O (m + k) ≥ 0 (r − 1) + m + 1 + kd 0 C 0 2 0 Definition 18.2. A divisor D is special if H1(C, D) 6= 0, and is nonspecial other- wise. 0 Remark 18.3. We could have stopped at m0, since we can show h (OC(m0)) is 0 nonspecial. In general, this is the best bound we can give on m so that h (OC(m)) is nonspecial, although it’s an interesting question to ask what the least m is in particular cases. Cancelling kd, we have
m + 1 m d − g + 1 = h0(O (m + k) ≥ 0 (r − 1) + m + 1 0 C 0 2 0 So, m + 1 g ≤ m (m (r − 1) + 1 + ε) − 0 (r − 1) − m 0 0 2 0 m = 0 (r − 1) + εm 2 0 Definition 18.4. Define m0 π(d, r) = (r − 1) + εm 2 0 where again, d = m0(r − 1) + ε, with 0 ≤ ε ≤ r. Theorem 18.5. If C is a curve of genus g and degree d in Pr, for r ≥ 3, then g ≤ π(r, d). Proof. Completed above. The next step is to show there are curves of the maximal genus, called Casteln- uovo curves, and that they are particularly nice. 40 AARON LANDESMAN
TABLE 2. The function π(d, r)
d π m0 r 0 1 r+1 1 1 r+2 2 1 ...... 2r-2 r-2 1 2r-1 r-1 1 2r r+1 2 2r+1 r+3 2 ......
Example 18.6. Take r = 3. k 2 (1) Take d = 2k + 2, m0 = k, ε = 1. So, π(d, 3) = 2 + k = k . This is precisely the genus of a curve of type (k + 1, k + 1) on a quadric surface. Note, it makes sense that the curves achieving a bound can be found on a quadric surface (the only way we can find the sharp equality on hΓ (m)) is when the curve is a rational normal curve, i.e., is a conic in P2. (2) Take d = 2d + 1, m0 = k, ε = 0. Then, π(d, 3) = k(k − 1) which is the genus of a curve of type (k, k + 1) on Q. Let’s make a table for π(d, r). Asymptotically, we have d2 π(3, r) ∼ 2r − 2 We see that there are curves of degree 2r in Pr+1 are canonical curves. We can also see that up to d = 2r, these bounds are a consequence of Clifford’s theorem. We have g ≤ π(d, r). Equality implies two things. First, in order to have equal- ity, we must have equality of 0 0 hC(m) − hC(m − 1) ≤ h (OC(m)) − h (OC(m − 1)) ≥ hΓ (m) throughout. That is, both inequalities must be equalities. So, we must have that C is projectively normal In other words,
0 0 (18.1) H (OPr (m)) H (OC(m)) 0.
Remark 18.7. The word normal in projectively normal is intended. The surjectiv- ity condition is equivalent to saying that the cone over C in Pr+1 has a normal point at the vertex, assuming that C is smooth. If C is not smooth, the correspond- ing condition is called arithmetically Cohen Macaulay. Corollary 18.8. (Noether’s theorem) If C ⊂ Pg−1 is canonical (meaning C is not hyper- elliptic), then the map Sym•H0(K) ⊕H0(K⊗m) is surjective. → NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 41
Proof. This follows from C being extremal and hence the above maps
0 0 (18.2) H (OPr (m)) H (OC(m)) are surjective, which are the same as the maps
Sym•H0(K) ⊕H0(K⊗m).
→ A second consequence is that
hΓ (m) = min(d, m(r − 1) + 1)
Then, with one restriction, every linear combination of points with this minimal Hilbert function hΓ = min(d, mn + 1) and d ≥ 2r − 1, then Γ lies on a rational normal curve B ⊂ Pr−1. In this case, B is the intersection of the quadrics Q ⊂ Pr−1 containing Γ. Now, by our homework, because the curve is projectively normal, (in partic- ular, linearly normal), then every quadric containing the hyperplane section is the restriction of a quadric containing the curve. In this case, we can look at ∩C⊂Q⊂Pr Q. This must be a surface whose general hyperplane section is a rational normal curve.
Question 18.9. What are the surfaces in Pr whose hyperplane section is a rational normal curve?
19. 10/16/15 19.1. Review. Previously, we took C ⊂ Pr irreducible, nondegenerate of degree d and genus g, and found g ≤ π(d, r). We wondered whether this was sharp, and whether we can characterize the curve achieving this maximum possible genus.
Lemma 19.1. (Castelnuovo’s Lemma) Let Γ ⊂ Pn be a configuration of d ≥ 2n + 3 points in linear general position. Then, if hΓ (2) = 2n + 1, then Γ lies on a rational normal curve.
Proof. We may or may not prove this next week.
n Now, suppose Γ ⊂ P be of degree d in linear general position. Then, hΓ (m) ≥ min(d, mn + 1), as we saw before. So, the intersection of the quadrics containing a curve of maximum genus, so
g(C) = π(d, r) d ≥ 2r + 1
Remark 19.2. Caution: The following does not hold when d = 2r.
then, ∩Q⊃CQ is a surface and S ∩ H is a rational normal curve. Today, we’ll talk about something which may seem to have little to do with the above. 42 AARON LANDESMAN
19.2. Minimal Varieties. Say X ⊂ Pn is an irreducible, nondegenerate variety of dimension k. Question 19.3. What is the smallest possible degree of such a variety of such an X? Answer: It’s not hard to obtain a lower bound. Suppose n ≥ 2. A general hyperplane section is going to again be nondegenerate. So, if we continue cutting down the dimension, we will obtain an irreducible nondegenerate set of points: Say dim X = k. Observe X ∩ H1 ∩ · · · ∩ Hk. We then have a nondegenerate set of points in Pn−k, and it takes n − k + 1 points to span Pn−k Therefore, the degree of X must be at least n − k + 1. Remark 19.4. We can also see the above by intersecting with only k − 1 hyper- planes and get a rational normal curve (also of degree n − k + 1). In some sense, minimal varieties are the simplest, smallest varieties. Remark 19.5. There are a lot of interesting open problems about these minimal varieties. Say k = 2, so we’re looking for a surface of minimum possible degree. We will now give a construction: In Pn, choose complimentary linear subspaces. Pk, Pl ⊂ Pn, meaning that we we think of these as corresponding to V, W ⊂ U, then V ⊕ W = U. Caution, the k for Pk has nothing to do with k = dim X = 2. Now, choose a rational normal curve in Pk and Pl, meaning a map φ : P1 Pk φ0 : P1 Pl 0 → where the images are C, C . We’re assuming k + l = n − 1. Now, take X = Xk,l = 0 → ∪t∈P1 φ(t), φ (t). This is an irreducible ruled surface. It’s nondegenerate because it contains the two rational normal curves. Lemma 19.6. deg X = n − 1. Proof. To calculate deg X, we can equivalently find deg X ∩ H, where H is a hyper- plane containing Pk. By Bertini, for a general hyperplane section, X ∩ H is reduced. Set theoretically, the intersection has to be a curve. Now, the surface is a union of lines, all of which contain C. If we have any point not on C, then we contain the complete line. Therefore, the intersection consists of the union of C and a collection of lines of this ruling. The number of such lines is the number of points it meets C0 0 0 l in. It will meet C in l points, because deg C = l. Therefore, X ∩ H = C ∪i=1 Li, where Li are lines. So, deg X = deg X ∩ H = deg C + l = k + l = n − 1. Remark 19.7. Here, we’re implicitly assuming k, l > 0, as if either were 0 we would get a cone over a rational normal curve. When we do a calculation, we can think of X as a curve in the grassmannian of lines. Hence, a map P1 G(1, n). We can take X to be the preimage of a curve in the grassmannian. This will map onto a cone collapsing a curve to the vertex in the cone. → We can also show a general such surface, when k, l > 0 is the blow up of this cone. If k, l > 0, then this surface X is smooth, as can be seen in several ways. For example, we can calculate the tangent space at a given point. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 43
These surfaces are called rational normal surface scrolls. 3 3 Example 19.8. (1) In P , we can take X1,1 ⊂ P . Choose two lines, and draw the lines joining the points. This is a quadric surface. 4 4 (2) Consider X2,1 ⊂ P . This is the union of lines joining a line in P and a conic in a plane. 5 (3) In P , there are X3,1 and X2,2. It’s not obvious these are different, but, in fact, they are, and they have different Hilbert polynomials. ∼ ∼ Fact 19.9. (1) Abstractly, Xa,b = Fa−b = P(OP1 (a) ⊕ OP1 (b)). (2) We have a map P1 G(1, n), so that the image is a rational normal curve of degree n − 1 in G(1, n). n (3) Take general linear→ forms Li, Mi to on P . If we look at the locus where L L ··· L X = rk 1 2 n−1 < 1 M1 M2 ··· Mn−1
(4) We can now generalize this to any dimension k. Then, if we choose Pa1 , ... , Pak ⊂ n k P to be complementary, so that 1 ai = n − k + 1. Then, choose maps 1 ai parameterizing rational normal curves φi : P P and set Xa1,...,ak = P 5 ∪t∈P1 φ1(t), ... , φk(t). The first example is X1,1,1 ∈ P . That is, we choose three spanning lines in P5, and we draw the 2 plane→ spanned by triples of points in P5. This three fold will have the minimum degree 3. Abstractly, this turns out to be the image of P1 × P2 P5 under the Segre embed- ding, but we haven’t checked its a product. Definition 19.10. A variety X ⊂ Pn is a minimal →variety if X is irreducible, non- degenerate of degree n − dim X + 1. Theorem 19.11. If X in Pn is any minimal variety is either (1) X is a scroll (2) X is a quadric hypersurface (redundant in the case of surfaces). 2 5 (3) X is a cone over the Veronese surface ν2(P ) of degree 4 in P . Proof. Not given. Another question we can ask about minimal surfaces is: Question 19.12. What is the smallest possible Hilbert function? (In other words, how many hypersurfaces of a given degree can contain such a variety?) Now, we can come back to the case of surfaces. For now, let’s omit the Veronese surface. So, we’re just trying to find curves on a scroll. We would like to see if we can find a divisor class (i.e. curve) on a minimal surface whose genus achieves Castelnuovo’s bound. 1 Let X = Xk,l be a surface scroll. We will assume X is smooth. X is a P bundle over P1. So, the Picard group is generated by two elements, one which is a line of the ruling, and take the other to be a hyperplane class (which must be independent because it has nonzero intersection with the line class). So, let h be a hyperplane section, f be a fiber. Then, Pic(X) = Zhh, fi Writing out the intersection pairing, we have The next step is to find the canonical class: 44 AARON LANDESMAN
h f h n-1 1 f 1 0
h f h n-1 1 f 1 0
Lemma 19.13. KX ∼ −2h + (n − 3)f. Proof. Perhaps we’ll see next time. 20. 10/19/15 Today and Wednesday, we’ll discuss outgrowths of Castelnuovo theory, and on Friday, we’ll start with Brill Noether Theory. Recall the lemma from last time: Lemma 20.1. Say Γ ⊂ Pn and d ≥ 2n + 3 points in linear general position. Let n hΓ (2) = 2n + 1. Then, Γ is contained in a rational normal curve C in P . Remark 20.2. The generalizations of Lemma 20.1 are what leads to further inves- tigation of the possible degree genus pairs of curves in Pr. We know that if C ⊂ Pr is irreducible smooth nondegenerate of degree d, genus g, we have g ≤ π(d, r). Theorem 20.3. If C ⊂ Pr is irreducible, nondegenerate of degree d and genus g. Then, g ≤ π(d, r). If equality holds and d ≥ 2r + 1 then C lies on a rational normal surface scroll or a Veronese surface. Also, conversely, if S is a rational normal surface scroll and C is a curve of class where d = m(r − 1) + 1 + ε with 0 ≤ ε ≤ r − 2 (m + 1)h − (r − 2 − ε)f or ε = 0 and C ∼ mh + f on such a surface, then equality is achieved (and a similar statement holds for the Veronese surface). Remark 20.4. The key is that when a curve is sitting in Pr, we know very little about its geometry. But, if we get the curve to sit on the surface, we know exactly how to translate the class of the curve into the genus. The key is to introduce Proof. Say S is a rational normal surface scroll in Pr. Recall Pic(S) = Zhh, fi where f is the class of a fiber and h is the hyperplane section. Recall the intersection pairing: A priori, this generates the Picard group over Q. But, since this has determinant of absolute value 1, the Picard group can’t be any larger than that generated by h, f. So, it is generated by h, f. Now, write
KS = ah + bf and we apply adjunction to find a, b. If we apply adjunction to f, we have
−2 = 2g − 2 = f(KS + f) = a so a = −2. Next, to find h, We know the hyperplane section is a rational normal curve. We have
−2 = 2g − 2 = h(KS + h) NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 45 and by solving this, we conclude b = r − 3. So,
KS = −2h + (r − 3)f Remark 20.5. Let’s check this in the quadric surface in P3. By adjunction, the canonical bundle is O(−2), which is precisely what we predicted. So, if C ⊂ S has class ah + bf (different a,b than those above) we see from the intersection table deg C = C · h = a(r − 1) + b and, the genus of C is (ah + bf) ((a − 2)h + (b + r − 3)f) g(C) = + 1 2 a = (r − 1) + (a − 1)(b − 1) 2 Remark 20.6. We can check the above when r = 3, and a curve in the basis h, f is a curve of type (a, a + b), as we’re choosing a basis for the Picard group whose generators are a line of the ruling and a sum of the lines of the two rulings. to find a curve of maximal genus, we write d = m(r − 1) + 1 + ε with 0 ≤ ε ≤ r − 2. We take C ∼ (m + 1)h − (r − 2 − ε)f m g = (r − 1) + mε 2 The choice was unique when r = 3 except for possibly a confusing between two rulings of the quadric surface. The above is also unique except in one case ε = 0, in which case we have another solution given by C ∼ mh + f. Exercise 20.7. Check this on your own. Finally, we should check when r = 5 and S is a Veronese surface. We can do this because every curve on the Veronese surface has even degree, and it is the image of some plane curve, which was can use to find the genus. This also yields equality g = π(d, r). Remark 20.8. If you had a surface and you wanted to bound the geometric genus of curves lying on the surface H0(K), you can use hyperplane sections to get lower bounds on the hyperplane sections, Remark 20.9. To exhibit curves which achieve the bound, we try to use Bertini’s theorem, to show that the linear system (m + 1)h − (r − 2 − ε)f, we take some very singular curves which are unions of hyperplanes on the surface. We exhibit enough of the curves to show it’s basepoint free. Then, Bertini’s theorem tells us there are smooth curves in these linear systems. Remark 20.10. By Clifford’s theorem, you can’t have a curve of degree 2r with genus bigger than r + 1. So, canonical curves are curves of maximum genus. However, canonical curves don’t satisfy d > 2r, so we can’t apply Castelnuovo’s lemma. For instance, if we have 8 points in P3 which impose 7 conditions on quadrics, they could lie on the intersection of three quadrics. So, you really need the 2n + 3rd points for Castelnuovo’s lemma. 46 AARON LANDESMAN
But, now, Enriques made an interesting observation: Canonical curves lie on many quadrics. In genus 5, a general canonical curve is the intersection of three quadrics. Question 20.11. Is a canonical curve C ⊂ Pg−1 cut out by quadrics?
If not: that is, if there is p ∈ ∩Q⊃CQ but p ∈/ C, we claim that C lies on a minimal degree surface. Then, we can go through the whole analysis of the bound on the genus. We can choose a general hyperplane H through p. Since p ∈/ C, by Bertini, such a hyperplane will intersect the curve transversely at {p1, ... , p2r}. The same analysis, by the uniform position lemma, implies Γ ∪ {p} is in linear general position. Essentially, this amounts to applying the uniform position lemma from the projection of the curve from the point C. So, if we have g = π(d, r), which does hold for canonical curves, we obtain, hΓ ∪{p} = 2r − 1, and so Γ ∪ {p} lies on a rational normal curve. So, we obtain that ∩Q⊃CQ is a minimal rational surface. In particular, the in- tersection of the quadrics has to be a surface. The above remark establishes the following, except for trigonality and plane quintics.
Proposition 20.12. Let C be a canonical curve. If there is p ∈ ∩Q⊃CQ but p ∈/ C, we claim that C lies on a minimal degree surface. In particular the curve is trigonal or a plane quintic. Proof. To prove trigonality, if the curve is on a scroll, the fibers meet the curve in three points. In the case that the curve lies on the Veronese surface, it is a plane quintic. Theorem 20.13. If C ⊂ Pg−1 is a canonical curve, then either (1) C is cut out by quadrics (2) C is trigonal (3) g = 6 and C is a plane quintic. Proof. This follows immediately from Proposition 20.12 Remark 20.14. We know trigonal curves will not be cut out by quadrics, because by geometric Riemann Roch, there will be three collinear points, and so it cannot be cut out by quadrics, by Bezout. Remark 20.15. Once you know that the ideal of the curve is generated by quadrics, you can ask what are the relations among the quadrics. That is, you can ask for the next step of the resolution of the curve. Then, the relations are all linear, unless 1 2 C has a g4 or a g6. This was extended by Green and Lazarsfeld to a conjecture on the resolution of the canonical curve, and Voisin proved this in the generic case. Question 20.16. What if the curve does not lie on a rational normal surface scroll? Then, Castelnuovo’s lemma characterizes curves with the smallest Hilbert func- tion. The next question is: Question 20.17. What is the second smallest Hilbert function? If we know this, we can start exhibiting degrees and genera below the bound which don’t occur. NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 47
21. 10/21/15 Recall, C ⊂ Pr is a smooth irreducible nondegenerate curve of degree d genus g, and Γ = H ∩ C. Then, 0 0 h (OC(m)) − h (OC(m − 1)) ≥ hΓ (m) From these inequalities, we have g(C) ≤ π(d, r). Remark 21.1. For today, the Veronese may be (incorrectly) referred to as a rational normal surface scroll. Recall Castelnuovo’s lemma: Lemma 21.2. Γ ⊂ Pn in linear general position of degree d. Then,
(1) hΓ (m) ≥ min(d, mn + 1) = h0(m) (2) If d ≥ 2n + 3 and hΓ (2) = h0(2) then Γ is contained in a rational normal curve. Since we have determined the smallest Hilbert function corresponds to lying on rational normal curve, this motivates the next question. In particular, we’d like to know which curves have genus close to π(d, r) but not equal to π(d, r). Question 21.3. What’s the second smallest Hilbert function? We’ll see, in fact that there are gaps in the maximal genera below where the maximum is achieved, and we still don’t know exactly where those gaps are. The argument for our bound π, we only looked at degree m curves which are unions of hyperplanes. However, in fact, we have a stronger condition. We’ll see what we can deduce for hyperplane sections with this stronger property. Definition 21.4. A set of points Γ are in uniform position if for any two subsets Γ 0, Γ 00 ⊂ Γ so that |Γ 0| = |Γ 00| then
hΓ 0 = hΓ 00 Lemma 21.5. If Γ = C ∩ H is a general hyperplane section of an irreducible C ⊂ Pr, then the points of Γ are in uniform position. Example 21.6. Say r = 3 so we’re looking at curve in P3. Then, if we have a collection of point in a hyperplane H, then either no six points lie on a conic, or else all points lie on a conic. Proof. The idea is similar to that of linear general position: If three points were on a line, the all of Γ would lie on a line, contradicting nondegeneracy. We have a similar situation here. If one set of 6 points lies on a conic, the we can swap out one point at a time, to see that every collection of 6 points lies on a conic. Say U = (Pr)∗ \ C∗, where C∗ means the subset of (Pr)∗ which are tangent of C. The complement is the set of hyperplanes transverse to C. Define the incidence correspondence. Φ = {(H, p): p ∈ H ∩ C} ⊂ U × C We now specialize to the case of curves lying on quadrics in P3. We can similarly introduce the set of sequences of points, which is
ψ = {(H, p1, ... , p6) : pi distinct in H ∩ C} 48 AARON LANDESMAN and let 0 ψ = {(H, p1, ... , p6) : pi lie on a conicH ∩ C} 0 If there exist 6 points p1, ... , p6 that don’t lie on a conic, we have ψ ( ψ im- plies dim ψ0 < dim ψ = r, because ψ is integral, as follows from the monodromy statement. Therefore, general hyperplane sections can’t have some subset of points lying on a conic and not all subsets not lying on a conic. Remark 21.7. So, the second smallest Hilbert function should correspond to the 3 points lying on a cubic in P . So, we currently have hΓ (m) ≥ min(d, 2m + 1) in P3. If we replace the quadrics by a cubic, we would get a bound min(d, 3m + 1), and there is a significant gap between these bounds. Exercise 21.8. If we have some configuration of points in a plane and they don’t lie on a conic, then the next smallest Hilbert function is from those lying on a cubic. First, we recall Castelnuovo’s theorem. Theorem 21.9. (Castelnuovo) If Γ has minimal Hilbert function, then it has that because n it lies on a rational normal curve B ⊂ P so that hΓ (m) = hB(m) = mn + 1, with d m ≤ n . Recall, in general, if B ⊂ Pn is a curve of degree d and genus g, the Hilbert polynomial of B is
pB(m) = d · m − g + 1 To minimize this, we first try to minimize d. The smallest possible degree is n, which is achieved by the rational normal curve. Now, we ask, which curve has the second smallest Hilbert function? The next smallest would have to have degree at least n + 1. Then, the genus of the curve may be either 1, 0 as follows from Castelnuovo’s theorem. We conclude that the second smallest Hilbert function for hΓ of a collection Γ of points in uniform position is achieved by points on an elliptic normal curve. Remark 21.10. We’re guessing this on the basis of a suggestion: Configurations of points with small Hilbert function have that small Hilbert function because they lie on a curve with small Hilbert function. It takes some work to prove this is actually the case, since it might be that they don’t lie on a curve, but it turns out that they do. Proposition 21.11. If Γ is not contained in a rational normal curve, then m(n + 1) if d > m(n + 1) hΓ (m) = m(n + 1) − 1 if d = m(n + 1) d if d < m(n + 1)
Proof. It follows from the comments preceding this, though those comments weren’t really proven. Corollary 21.12. If C 6⊂ a rational normal surface scroll, then we obtain that the genus 2 is bounded by g ≤ π1(d, r) ∼ d /2r NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 49
Proof. This follows from plugging in the inequalities from Proposition 21.11, we 2 obtain g ≤ π1(d, r) ∼ d /2r. We get this by computing the Hilbert function for large values of m, and then using Riemann Roch to bound the genus. When we were analyzing curves on rational normal surface scrolls, we found the Picard group has rank 2, and so we used this to determine the possible genera of curves on these scrolls. Let’s now describe what happens for curves in P3.
3 d2 Proposition 21.13. In P , for g > π1(d, 3) ∼ 6 , we have that g occurs if and only if there exist integers a, b so that a + b = d and (a − 1)(b − 1) = g. If g ≤ π1(d, r), then we next obtain that the curve lies on a cubic or a higher degree curve, and every g occurs.
Proof. In the case that g ≤ π1(d, r) a cubic surface has Picard group of rank 7, and quartic has Picard groups of rank up to 19, and it turns out you can always solve the system for each genus. Question 21.14. What about in higher dimensional space?
r d2 Proposition 21.15. In P , if g > π1(d, r) ∼ 2r , then the curve lies on S, and g occurs if and only if there are a, b so that a(r − 1) + b = d, and 2g − 2 = (ah + bf)((a − 2)h + (b + r − 3)f) a g = (r − 1) + (a − 1)(b − 1) 2 in Pic(S).
Proof. The case g > π1(d, r) follows from prior analysis.
Remark 21.16. In the case g ≤ π1(d, r). However, once r > 1 there are no del Pezzo surfaces, and there are no surfaces of one more than the minimal degree, except for del Pezzo surfaces, but those die off after r = 9 or so, and the rank of the Picard group also decreases. Look in the Montreal notes for more details. Remark 21.17. If we drop the hypothesis of smoothness, and take the geometric genus, we do get curve of every geometric genus up to π(d, r). The reason is that in the linear system on a rational normal surface scroll S, we can find a curve class that ache Ives the maximum, and there are smooth curves of this class. Further, in this linear system, there exist irreducible nodal curves C ⊂ S with δ nodes for any 0 ≤ δ ≤ π(d, r). Exercise 21.18. There exist irreducible plane curves with δ nodes for any δ between d 0 and 2 . 22. 10/23/15 22.1. Agenda and Review. Today (1) Green’s conjecture (2) Voisin’s theorem (3) The maximal rank conjecture (if time allows) On Monday, we’ll move on to Brill Noether Theory Let’s now recap what happened last time. 50 AARON LANDESMAN
Question 22.1. For which d, g, r does there exist a smooth nondegenerate C ⊂ Pr of degree d and genus g? This leads to the following question. Question 22.2. What are the possible Hilbert functions of configurations of points in uniform position? The question of all possible Hilbert functions has been answered by Richard Stanley and others, which is solved by considering monomial ideals. However, the question of uniform position points is more difficult. There are other interesting subschemes whose Hilbert functions we can ask about, such as Hilbert functions of fat points (schemes supported at a point). 22.2. Resolutions of Projective Varieties and Green’s conjecture. Suppose X ⊂ n P is a variety or subscheme. We can look in I(X) ⊂ S = C[x0, ... , xn]. Say I(X) = (f1, ... , fk) and deg fi = ai. Then, we have a resolution of the form
(22.1) ··· ⊕S(−bi) ⊕S(−ai) S S(X) 0 where S(−ai) correspond to the generators. Then, we can ask what relations exist among those generators. As a module the kernel of the map ⊕S(−ai) S, de- termines the module of relations. This resolution terminates for a variety by the Hilbert basis theorem. This resolution also tells you the Hilbert function.→ Hence, understanding the Hilbert function is tantamount to describing the res- olution. Example 22.3. Let X = C ⊂ P4 is a canonical curve of genus 5. We know C lies on three quadrics. So, we have a resolution
(22.2) ··· ⊕S(−4)⊕2 S(−2)⊕3 S S(X) 0 in the case that the quadrics cut out the curve. In the case that the quadrics don’t cut out the curve, there’s an additional cubic in the ideal of the curve. There is also a linear relation among the quadrics in this case, but there’s only a quadratic rela- tion (that corresponding to QiQj − QjQi) in the case that the curve is a complete intersection of quadrics. So, to understand a curve, we could try to understand the resolution of its ideal. The goal is to describe the resolution of canonical curves. Suppose C is a smooth curve of genus g. φ Say C is a smooth curve of genus g. Then, We have C −−K Pg−1 =∼ PH0(K)∨. and S = Sym•H0(K). Then, we introduce the canonical ring 0 m → R = ⊕m=0H (K ) Now, by Noether’s theorem, the map∞ S R or equivalently, Sym•H0(K) 0 m ⊕mH (K ) is surjective if and only if the curve is not hyperelliptic. So, let’s assume C is not hyperelliptic. Then,→ we can ask about the resolution.→ By Enriques’ theorem, we can write
⊕ g−2 (22.3) ··· S(−2) ( 2 ) S S(X) 0 NOTES FOR MATH 282, GEOMETRY OF ALGEBRAIC CURVES 51 in the case that the curve is further not trigonal, nor a plane quintic. 1 2 This is equivalent to saying there does not exist a g3 or a g5. Now, here’s the next step, which will lead to Green’s conjecture. We can now look at the linear relations among these quadrics.
⊕ g−2 (22.4) ··· S(−3)⊕N S(−2) ( 2 ) S S(X) 0
1 2 is exact if and only if C does not have a g4 or g6. Green’s conjecture naturally extends this. If L is a line bundle on C of degree d and dimension r = H0(L) − 1. If we assume H0(L), H1(L) 6= 0 then d − 2r ≥ 0. r Definition 22.4. The Clifford index of a linear series L which is a gd is d − 2r. Then, the Clifford index of a curve is the minimum over all line bundles L with h0(L), h1(L) 6= 0, other than O, K, of the Clifford index of L. Remark 22.5. Observe, that Cliff(C) = 0 C is hyperelliptic Cliff(C) = 1 C is trigonal or a plane quintic ⇐⇒ 1 2 3 Cliff(C) = 2 C has a g3 or g5 or g7 ⇐⇒ 3 3 Note for curves with g7, we get a map C P , then the curve has a 4 secant line. Then, you can project from a⇐⇒ point on the line, and we obtain that C is already trigonal. → Here is a consequence of the Brill Noether Theorem. g−2 Corollary 22.6. For a curve C general of genus g, then Cliff(C) = d 2 e. Proof. Follows from Brill Noether, which we haven’t yet stated. Conjecture 22.7. (Green) This resolution of canonical curves, given by linear rela- tions at each step, is exact for Cliff(C) steps. If you know the resolution remains linear half way through, you can get the remaining Koszul betti numbers from duality. So, you can get all betti numbers. Then, Theorem 22.8. (Voisin) Green’s conjecture is true for a general curve C. Proof. Difficult g−2 Remark 22.9. Voisin does not prove that a curve with Clifford index d 2 e has a maximally linear resolution. She just shows it for some open subset of those curve. Remark 22.10. Here is the idea of proof of Voisin’s theorem. The set of curves which have a maximally linear resolution is an open condition: It’s saying that the ranks of certain maps are as large as they can be. So, to prove Voisin’s theorem, you only have to exhibit a curve somewhere which have that sort of resolution. She does this for curves on a K3 surface. Her approach doesn’t seem like it will yield anything more in the direction of proving Green’s conjecture. So, she doesn’t prove which curves are excluded. 52 AARON LANDESMAN
22.3. The Maximal Rank Conjecture. Now, let’s look at the second graded piece of an ideal of the curve I(C). We have a map
I(C)2 ⊗ S1 I(C)3