ConceptComplete

Linear Systems |D|

Linear systems are the bridge between divisors and maps to projective space. Every morphism from a curve to projective space arises from a linear system, and the geometry of the curve is reflected in the properties of its linear systems. This page develops the theory systematically, from complete linear systems through the canonical embedding.


Complete linear systems

Definition4.6Complete linear system

Let CC be a smooth projective curve over an algebraically closed field kk, and let DD be a divisor on CC. The complete linear system of DD is the set of all effective divisors linearly equivalent to DD:

∣D∣={Dβ€²β‰₯0∣Dβ€²βˆΌD}.|D| = \{D' \geq 0 \mid D' \sim D\}.

Concretely, each Dβ€²βˆˆβˆ£D∣D' \in |D| has the form Dβ€²=D+div⁑(f)D' = D + \operatorname{div}(f) for some f∈L(D)βˆ–{0}f \in L(D) \setminus \{0\}, and two functions f,gf, g give the same effective divisor if and only if f=Ξ»gf = \lambda g for some λ∈kβˆ—\lambda \in k^*. Therefore:

∣Dβˆ£β‰…P(L(D))=Pr|D| \cong \mathbb{P}(L(D)) = \mathbb{P}^{r}

where r=β„“(D)βˆ’1r = \ell(D) - 1, and β„“(D)=dim⁑kL(D)\ell(D) = \dim_k L(D). If β„“(D)=0\ell(D) = 0, then ∣D∣=βˆ…|D| = \emptyset.

Definition4.7Linear system (sub-linear system)

A linear system on CC is a linear subspace dβŠ†βˆ£D∣\mathfrak{d} \subseteq |D| for some divisor DD, i.e., a projective subspace dβ‰…PrβŠ†P(L(D))\mathfrak{d} \cong \mathbb{P}^r \subseteq \mathbb{P}(L(D)). Equivalently, d=P(V)\mathfrak{d} = \mathbb{P}(V) for some vector subspace VβŠ†L(D)V \subseteq L(D).

The dimension of the linear system is dim⁑d=r\dim \mathfrak{d} = r, and the degree is deg⁑D\deg D.

ExampleComplete linear systems on P^1

On P1\mathbb{P}^1 with D=nβ‹…[∞]D = n \cdot [\infty] for nβ‰₯0n \geq 0:

  • L(D)={f∈k(t)∣div⁑(f)+n[∞]β‰₯0}={L(D) = \{f \in k(t) \mid \operatorname{div}(f) + n[\infty] \geq 0\} = \{polynomials of degree ≀n}\leq n\}.
  • A basis is {1,t,t2,…,tn}\{1, t, t^2, \ldots, t^n\}, so β„“(D)=n+1\ell(D) = n + 1.
  • ∣Dβˆ£β‰…Pn|D| \cong \mathbb{P}^n: each element is a divisor [a1]+[a2]+β‹―+[an][a_1] + [a_2] + \cdots + [a_n] (roots of a degree-nn polynomial, counted with multiplicity).
  • For n=1n = 1: ∣[∞]βˆ£β‰…P1|[\infty]| \cong \mathbb{P}^1 parametrizes single points of P1\mathbb{P}^1.
  • For n=2n = 2: ∣2[∞]βˆ£β‰…P2|2[\infty]| \cong \mathbb{P}^2 parametrizes unordered pairs of points (including coincident pairs).
ExampleComplete linear systems on an elliptic curve

Let EE be an elliptic curve with origin OO.

  • ∣O∣|O|: β„“(O)=1\ell(O) = 1, so ∣O∣={O}|O| = \{O\}, a single point (P0\mathbb{P}^0).
  • ∣2O∣|2O|: β„“(2O)=2\ell(2O) = 2, so ∣2Oβˆ£β‰…P1|2O| \cong \mathbb{P}^1. The basis {1,x}\{1, x\} gives effective divisors div⁑(a+bx)+2O\operatorname{div}(a + bx) + 2O for [a:b]∈P1[a:b] \in \mathbb{P}^1. These are the fibers of the double cover x:Eβ†’P1x: E \to \mathbb{P}^1.
  • ∣3O∣|3O|: β„“(3O)=3\ell(3O) = 3, so ∣3Oβˆ£β‰…P2|3O| \cong \mathbb{P}^2. Each element is a group of 33 points cut out by a line in the Weierstrass embedding Eβ†ͺP2E \hookrightarrow \mathbb{P}^2.
  • ∣nO∣|nO| for nβ‰₯1n \geq 1: β„“(nO)=n\ell(nO) = n, so ∣nOβˆ£β‰…Pnβˆ’1|nO| \cong \mathbb{P}^{n-1}.

Base points and base-point-free systems

Definition4.8Base point and base locus

Let d\mathfrak{d} be a linear system on CC. A point P∈CP \in C is a base point of d\mathfrak{d} if P∈Supp⁑(Dβ€²)P \in \operatorname{Supp}(D') for every Dβ€²βˆˆdD' \in \mathfrak{d}.

The base locus is Bs⁑(d)=β‹‚Dβ€²βˆˆdSupp⁑(Dβ€²)\operatorname{Bs}(\mathfrak{d}) = \bigcap_{D' \in \mathfrak{d}} \operatorname{Supp}(D').

The linear system d\mathfrak{d} is base-point-free (bpf) if Bs⁑(d)=βˆ…\operatorname{Bs}(\mathfrak{d}) = \emptyset, i.e., for every point P∈CP \in C, there exists Dβ€²βˆˆdD' \in \mathfrak{d} with Pβˆ‰Supp⁑(Dβ€²)P \notin \operatorname{Supp}(D').

Equivalently, ∣D∣|D| is base-point-free if and only if β„“(Dβˆ’P)=β„“(D)βˆ’1\ell(D - P) = \ell(D) - 1 for every point P∈CP \in C, i.e., evaluation at each point is surjective.

ExampleBase points on a genus-2 curve

Let CC have genus 22 and let P∈CP \in C be a non-Weierstrass point.

  • ∣P∣|P|: β„“(P)=1\ell(P) = 1 (by Riemann--Roch, since PP is not a Weierstrass point), so ∣P∣={P}|P| = \{P\}. Trivially, PP is a base point.
  • ∣2P∣|2P|: β„“(2P)=1\ell(2P) = 1 for a general PP (since β„“(Kβˆ’2P)=1\ell(K - 2P) = 1 when 2P∼K2P \sim K fails). So ∣2P∣={2P}|2P| = \{2P\} and PP is a base point of multiplicity 22.
  • ∣KC∣|K_C|: β„“(KC)=2\ell(K_C) = 2 and deg⁑KC=2\deg K_C = 2. For any point PP, β„“(KCβˆ’P)=1\ell(K_C - P) = 1 by Riemann--Roch (=β„“(P)+1βˆ’2+1= \ell(P) + 1 - 2 + 1... more directly, since KCK_C is the g21g^1_2, every point appears in some fiber). So ∣KC∣|K_C| is base-point-free.

Now if WW is a Weierstrass point: β„“(2W)=2\ell(2W) = 2, so ∣2Wβˆ£β‰…P1|2W| \cong \mathbb{P}^1 and 2W∼KC2W \sim K_C. In the canonical system, WW still appears in every divisor? No -- ∣KC∣=∣2Wβˆ£β‰…P1|K_C| = |2W| \cong \mathbb{P}^1, and the divisor corresponding to xβˆ’x(W)x - x(W) is 2W2W itself, but other divisors in ∣KC∣|K_C| are pairs Q+Qβ€²Q + Q' not involving WW. So WW is not a base point.

RemarkRemoving base points

If Bs⁑(∣D∣)=βˆ‘PmPβ‹…P\operatorname{Bs}(|D|) = \sum_P m_P \cdot P (where mP=min⁑Dβ€²βˆˆβˆ£D∣vP(Dβ€²)m_P = \min_{D' \in |D|} v_P(D')), then D0=Dβˆ’βˆ‘mPβ‹…PD_0 = D - \sum m_P \cdot P satisfies ∣D0βˆ£β‰…βˆ£D∣|D_0| \cong |D| as projective spaces, and ∣D0∣|D_0| is base-point-free. The "fixed part" βˆ‘mPP\sum m_P P is carried by every divisor in the system.


The gdrg^r_d notation

Definition4.9The notation g^r_d

A gdrg^r_d on a curve CC is a linear system of degree dd and (projective) dimension rr. That is, a gdrg^r_d is a pair (D,V)(D, V) where DD is a divisor of degree dd and VβŠ†L(D)V \subseteq L(D) is a vector subspace of dimension r+1r + 1, giving P(V)β‰…Pr\mathbb{P}(V) \cong \mathbb{P}^r.

If V=L(D)V = L(D) (the full Riemann--Roch space), the gdrg^r_d is complete and we write ∣D∣|D|. In this case r=β„“(D)βˆ’1r = \ell(D) - 1.

Special terminology:

  • A gd0g^0_d is a single effective divisor of degree dd.
  • A gd1g^1_d is a pencil: a 11-parameter family of degree-dd divisors, i.e., a map Cβ†’P1C \to \mathbb{P}^1 of degree dd.
  • A gd2g^2_d is a net: a 22-parameter family giving a map Cβ†’P2C \to \mathbb{P}^2.
ExampleClassical examples of g^r_d
  • P1\mathbb{P}^1: the system ∣nβ‹…[∞]∣|n \cdot [\infty]| is a gnng^n_n. In particular, ∣[∞]∣|[\infty]| is a g11g^1_1 (the identity map P1β†’P1\mathbb{P}^1 \to \mathbb{P}^1).
  • Elliptic curve: ∣nO∣|nO| is a gnnβˆ’1g^{n-1}_n. The system ∣2O∣|2O| is a g21g^1_2 (double cover of P1\mathbb{P}^1). The system ∣3O∣|3O| is a g32g^2_3 (plane cubic embedding).
  • Genus-2 curve: ∣KC∣|K_C| is a g21g^1_2 (the hyperelliptic pencil).
  • Non-hyperelliptic genus-3 curve: ∣KC∣|K_C| is a g42g^2_4 (canonical embedding as plane quartic).
  • Non-hyperelliptic genus-4 curve: ∣KC∣|K_C| is a g63g^3_6 (canonical embedding in P3\mathbb{P}^3, intersection of a quadric and a cubic).
  • General genus-gg curve (gβ‰₯2g \geq 2): ∣KC∣|K_C| is a g2gβˆ’2gβˆ’1g^{g-1}_{2g-2}.
ExampleExistence of g^1_d

A key question in curve theory is: what is the smallest dd such that CC carries a gd1g^1_d?

  • g=0g = 0: d=1d = 1 (isomorphism to P1\mathbb{P}^1).
  • g=1g = 1: d=2d = 2 (every elliptic curve is a double cover of P1\mathbb{P}^1).
  • g=2g = 2: d=2d = 2 (every genus-22 curve is hyperelliptic).
  • g=3g = 3: d=2d = 2 if hyperelliptic; d=3d = 3 if not (project from a point on the canonical quartic).
  • g=4g = 4: d=2d = 2 if hyperelliptic; d=3d = 3 for a general curve (the canonical curve lies on a quadric surface in P3\mathbb{P}^3; rulings give a g31g^1_3).

The Brill--Noether number ρ(g,r,d)=gβˆ’(r+1)(gβˆ’d+r)\rho(g, r, d) = g - (r+1)(g - d + r) governs when gdrg^r_d's exist on a general curve: if ρβ‰₯0\rho \geq 0, a general curve of genus gg carries a gdrg^r_d; if ρ<0\rho < 0, it does not.


Maps to projective space

Definition4.10Map associated to a linear system

Let DD be a divisor on CC with β„“(D)=r+1β‰₯1\ell(D) = r + 1 \geq 1, and let s0,s1,…,srs_0, s_1, \ldots, s_r be a basis of L(D)L(D). The associated rational map is:

Ο•D:Cβ‡’Pr,P↦[s0(P):s1(P):β‹―:sr(P)].\phi_D : C \dashrightarrow \mathbb{P}^r, \quad P \mapsto [s_0(P) : s_1(P) : \cdots : s_r(P)].

This is well-defined (a morphism, not just rational) at points where not all sis_i vanish simultaneously, i.e., at points Pβˆ‰Bs⁑(∣D∣)P \notin \operatorname{Bs}(|D|). If ∣D∣|D| is base-point-free, then Ο•D\phi_D is a morphism everywhere.

More generally, a sub-linear-system d=P(V)βŠ†βˆ£D∣\mathfrak{d} = \mathbb{P}(V) \subseteq |D| with VβŠ†L(D)V \subseteq L(D) of dimension r+1r + 1 gives a rational map Ο•V:Cβ‡’Pr\phi_V : C \dashrightarrow \mathbb{P}^r.

TheoremBase-point-free systems give morphisms

The complete linear system ∣D∣|D| is base-point-free if and only if Ο•D:Cβ†’Pr\phi_D : C \to \mathbb{P}^r is a morphism (defined everywhere). This holds if and only if for every P∈CP \in C:

β„“(Dβˆ’P)=β„“(D)βˆ’1.\ell(D - P) = \ell(D) - 1.

ExampleVeronese maps from P^1

On P1\mathbb{P}^1, the system ∣nβ‹…[∞]∣|n \cdot [\infty]| with basis {1,t,t2,…,tn}\{1, t, t^2, \ldots, t^n\} gives:

Ο•n[∞]:P1β†’Pn,[s:t]↦[sn:snβˆ’1t:β‹―:tn].\phi_{n[\infty]} : \mathbb{P}^1 \to \mathbb{P}^n, \quad [s:t] \mapsto [s^n : s^{n-1}t : \cdots : t^n].

  • n=1n = 1: the identity P1β†’P1\mathbb{P}^1 \to \mathbb{P}^1.
  • n=2n = 2: the Veronese conic P1β†ͺP2\mathbb{P}^1 \hookrightarrow \mathbb{P}^2, image is {xz=y2}\{xz = y^2\}.
  • n=3n = 3: the twisted cubic P1β†ͺP3\mathbb{P}^1 \hookrightarrow \mathbb{P}^3, parametrized by (s3,s2t,st2,t3)(s^3, s^2t, st^2, t^3).

All are embeddings since the gnng^n_n separates points and tangent vectors on P1\mathbb{P}^1.

ExampleMaps from an elliptic curve

Let E:y2=x3+ax+bE: y^2 = x^3 + ax + b with origin OO at infinity.

The g21g^1_2 from ∣2O∣|2O|: basis {1,x}\{1, x\}, giving Ο•:Eβ†’P1\phi : E \to \mathbb{P}^1, P↦[1:x(P)]P \mapsto [1 : x(P)]. This is the double cover (x,y)↦x(x,y) \mapsto x, branched at the four roots of x3+ax+b=0x^3 + ax + b = 0 and at ∞\infty.

The g32g^2_3 from ∣3O∣|3O|: basis {1,x,y}\{1, x, y\}, giving Ο•:Eβ†ͺP2\phi : E \hookrightarrow \mathbb{P}^2, P↦[1:x(P):y(P)]P \mapsto [1 : x(P) : y(P)]. The image is the Weierstrass plane cubic. This is an embedding because 3O3O is very ample on EE.

The g43g^3_4 from ∣4O∣|4O|: basis {1,x,y,x2}\{1, x, y, x^2\}, giving Ο•:Eβ†ͺP3\phi : E \hookrightarrow \mathbb{P}^3, image is a curve of degree 44, the intersection of two quadric surfaces.


Very ample and ample divisors

Definition4.11Very ample and ample divisors on curves

A divisor DD on CC is very ample if Ο•D:Cβ†ͺPr\phi_D : C \hookrightarrow \mathbb{P}^r is a closed embedding. This requires two conditions beyond being base-point-free:

  • Separation of points: for any Pβ‰ QP \neq Q in CC, β„“(Dβˆ’Pβˆ’Q)=β„“(D)βˆ’2\ell(D - P - Q) = \ell(D) - 2, i.e., there exists s∈L(D)s \in L(D) vanishing at PP but not QQ.
  • Separation of tangent vectors: for any P∈CP \in C, β„“(Dβˆ’2P)=β„“(D)βˆ’2\ell(D - 2P) = \ell(D) - 2, i.e., the sections generate the fiber of O(D)\mathcal{O}(D) to first order at PP.

A divisor DD is ample if some positive multiple nDnD is very ample.

On a curve, DD is ample if and only if deg⁑D>0\deg D > 0.

Theorem4.3Criteria for very ampleness on curves

Let CC be a smooth projective curve of genus gg and DD a divisor on CC.

(a) If deg⁑Dβ‰₯2g+1\deg D \geq 2g + 1, then DD is very ample.

(b) If deg⁑Dβ‰₯2g\deg D \geq 2g, then DD is base-point-free.

(c) If deg⁑Dβ‰₯1\deg D \geq 1, then DD is ample.

In particular, for a general divisor of degree dd:

  • d≀0d \leq 0: ∣D∣=βˆ…|D| = \emptyset (no effective divisors).
  • 1≀d≀gβˆ’11 \leq d \leq g - 1: β„“(D)\ell(D) depends on the specific divisor class; may or may not be bpf.
  • g≀d≀2gβˆ’2g \leq d \leq 2g - 2: non-special if general, β„“(D)=dβˆ’g+1\ell(D) = d - g + 1, but may fail to be very ample.
  • dβ‰₯2g+1d \geq 2g + 1: always very ample.
ExampleVery ampleness threshold on an elliptic curve

For an elliptic curve EE (g=1g = 1):

  • D=OD = O (deg⁑=1\deg = 1): β„“(D)=1\ell(D) = 1, not base-point-free (point map to P0\mathbb{P}^0).
  • D=2OD = 2O (deg⁑=2\deg = 2): β„“(D)=2\ell(D) = 2, base-point-free (gives the 2:12:1 cover Eβ†’P1E \to \mathbb{P}^1), but not very ample (does not separate the two preimages of each point).
  • D=3OD = 3O (deg⁑=3=2g+1\deg = 3 = 2g + 1): β„“(D)=3\ell(D) = 3, very ample. Embeds EE as a plane cubic.
  • D=4OD = 4O (deg⁑=4\deg = 4): β„“(D)=4\ell(D) = 4, very ample. Embeds EE in P3\mathbb{P}^3 as the complete intersection of two quadrics.
  • D=nOD = nO for nβ‰₯3n \geq 3: always very ample, embedding EE in Pnβˆ’1\mathbb{P}^{n-1} as a curve of degree nn.
ExampleVery ampleness on a genus-2 curve

For a genus-22 curve CC (g=2g = 2, 2g+1=52g + 1 = 5):

  • ∣KC∣|K_C| is a g21g^1_2 (deg⁑2\deg 2). Base-point-free but not very ample -- gives a 2:12:1 map to P1\mathbb{P}^1.
  • A general divisor DD of degree 33: β„“(D)=2\ell(D) = 2 (since 3βˆ’2+1=23 - 2 + 1 = 2 and DD is non-special for general DD). This g31g^1_3 maps CC to P1\mathbb{P}^1, so not very ample.
  • A general divisor DD of degree 44: β„“(D)=3\ell(D) = 3, giving a map Ο•D:Cβ†’P2\phi_D : C \to \mathbb{P}^2. One can check this is birational onto a plane quartic with one node (since g=2g = 2 but a smooth quartic has g=3g = 3). Not an embedding, so not very ample.
  • DD of degree 55: β„“(D)=4\ell(D) = 4, and DD is very ample. Embeds Cβ†ͺP3C \hookrightarrow \mathbb{P}^3 as a curve of degree 55.

The canonical linear system

Definition4.12Canonical linear system and canonical map

The canonical linear system is ∣KC∣|K_C|, the complete linear system of the canonical divisor. Since β„“(KC)=g\ell(K_C) = g, we have:

∣KCβˆ£β‰…Pgβˆ’1|K_C| \cong \mathbb{P}^{g-1}

which is a g2gβˆ’2gβˆ’1g^{g-1}_{2g-2}. The associated map is the canonical map:

Ο•K:Cβ†’Pgβˆ’1.\phi_K : C \to \mathbb{P}^{g-1}.

If Ο‰1,Ο‰2,…,Ο‰g\omega_1, \omega_2, \ldots, \omega_g is a basis of H0(C,Ο‰C)H^0(C, \omega_C) (regular differentials), then locally Ο‰i=fi dt\omega_i = f_i \, dt for a local coordinate tt, and:

Ο•K(P)=[f1(P):f2(P):β‹―:fg(P)]=[Ο‰1(P):β‹―:Ο‰g(P)].\phi_K(P) = [f_1(P) : f_2(P) : \cdots : f_g(P)] = [\omega_1(P) : \cdots : \omega_g(P)].

Theorem4.4Canonical map: hyperelliptic vs. non-hyperelliptic

Let CC be a smooth projective curve of genus gβ‰₯2g \geq 2. Then:

(a) If CC is non-hyperelliptic, the canonical map Ο•K:Cβ†ͺPgβˆ’1\phi_K : C \hookrightarrow \mathbb{P}^{g-1} is an embedding. The image is called the canonical curve and has degree 2gβˆ’22g - 2 in Pgβˆ’1\mathbb{P}^{g-1}.

(b) If CC is hyperelliptic, the canonical map factors as Cβ†’2:1P1β†’Ξ½gβˆ’1Pgβˆ’1C \xrightarrow{2:1} \mathbb{P}^1 \xrightarrow{\nu_{g-1}} \mathbb{P}^{g-1}, where Ξ½gβˆ’1\nu_{g-1} is the Veronese embedding of degree gβˆ’1g - 1. The image is a rational normal curve of degree gβˆ’1g - 1.

ExampleCanonical map for genus 2

Every genus-22 curve is hyperelliptic. The canonical system ∣KC∣|K_C| is a g21g^1_2:

ϕK:C→2:1P1.\phi_K : C \xrightarrow{2:1} \mathbb{P}^1.

If C:y2=f(x)C : y^2 = f(x) with deg⁑f=5\deg f = 5 or 66, a basis of H0(Ο‰C)H^0(\omega_C) is {dx/y, x dx/y}\{dx/y,\, x\, dx/y\}, and the canonical map is just (x,y)↦[1:x](x, y) \mapsto [1 : x], i.e., the projection to the xx-coordinate. The six branch points are the roots of ff (plus ∞\infty if deg⁑f=5\deg f = 5).

ExampleCanonical embedding for genus 3 (non-hyperelliptic)

Let CC be a non-hyperelliptic curve of genus 33. Then ∣KC∣|K_C| is a g42g^2_4:

Ο•K:Cβ†ͺP2\phi_K : C \hookrightarrow \mathbb{P}^2

embedding CC as a smooth plane quartic. Conversely, every smooth plane quartic is a canonically embedded genus-33 curve, since for a plane curve of degree d=4d = 4, adjunction gives KC∼(4βˆ’3)H∣C=H∣CK_C \sim (4 - 3)H|_C = H|_C, i.e., the canonical class is the hyperplane class.

Explicit example: Let C=V(x4+y4+z4)βŠ‚P2C = V(x^4 + y^4 + z^4) \subset \mathbb{P}^2 (the Fermat quartic).

  • Genus: g=(4βˆ’1)(4βˆ’2)2=3g = \frac{(4-1)(4-2)}{2} = 3.
  • The canonical system is cut out by lines: ∣KC∣={C∩L∣LΒ aΒ lineΒ inΒ P2}|K_C| = \{C \cap L \mid L \text{ a line in } \mathbb{P}^2\}.
  • Each canonical divisor consists of 44 points (the intersection of a line with CC).
  • The differentials Ο‰1=x dzFy,Ο‰2=y dzFy,Ο‰3=z dzFy\omega_1 = x\,\frac{dz}{F_y}, \omega_2 = y\,\frac{dz}{F_y}, \omega_3 = z\,\frac{dz}{F_y} (up to scaling) give the identity embedding Cβ†ͺP2C \hookrightarrow \mathbb{P}^2.
ExampleCanonical map for a hyperelliptic genus-3 curve

Let C:y2=f(x)C: y^2 = f(x) with deg⁑f=7\deg f = 7 or 88 be a hyperelliptic genus-33 curve. A basis of differentials is {dx/y, x dx/y, x2 dx/y}\{dx/y,\, x\, dx/y,\, x^2\, dx/y\}, so:

Ο•K:Cβ†’P2,(x,y)↦[1:x:x2].\phi_K : C \to \mathbb{P}^2, \quad (x, y) \mapsto [1 : x : x^2].

The image is the conic {xz=y2}β‰…P1\{xz = y^2\} \cong \mathbb{P}^1 (the Veronese image of P1\mathbb{P}^1), and Ο•K\phi_K is 2:12:1 onto this conic. The canonical map is not an embedding -- it collapses the pairs of conjugate points (x,y)(x, y) and (x,βˆ’y)(x, -y).

ExampleCanonical embedding for genus 4

Let CC be a non-hyperelliptic curve of genus 44. The canonical map gives:

Ο•K:Cβ†ͺP3\phi_K : C \hookrightarrow \mathbb{P}^3

with image of degree 2gβˆ’2=62g - 2 = 6. By the Enriques--Babbage theorem, the canonical curve CβŠ‚P3C \subset \mathbb{P}^3 lies on a unique quadric surface QQ:

  • If QQ is smooth (Qβ‰…P1Γ—P1Q \cong \mathbb{P}^1 \times \mathbb{P}^1), then CC is a (3,3)(3,3)-curve on QQ, and C=Q∩SC = Q \cap S where SS is a cubic surface.
  • If QQ is a quadric cone, then CC is a divisor of type 3H3H on QQ (where HH is the hyperplane class).

In either case, CC is a complete intersection of a quadric and a cubic in P3\mathbb{P}^3.


The gdrg^r_d and maps to projective space

RemarkRelationship between g^r_d and maps

A gdrg^r_d on CC (with no base points) determines a morphism ϕ:C→Pr\phi : C \to \mathbb{P}^r of degree dd. The dictionary is:

  • gd1g^1_d: a degree-dd map Cβ†’P1C \to \mathbb{P}^1 (a dd-sheeted cover of P1\mathbb{P}^1).
  • gd2g^2_d: a degree-dd map Cβ†’P2C \to \mathbb{P}^2 (a plane model of CC, often with singularities).
  • gdrg^r_d very ample: an embedding Cβ†ͺPrC \hookrightarrow \mathbb{P}^r as a non-degenerate curve of degree dd.

The degree of the image and the degree of the map satisfy: d=(deg⁑ϕ)β‹…(deg⁑ϕ(C))d = (\deg \phi) \cdot (\deg \phi(C)). If Ο•\phi is birational onto its image, the degree of the image equals dd.

ExamplePlane models of curves

Every curve admits a birational plane model (a gd2g^2_d for some dd):

  • Genus 00: P1β†ͺP2\mathbb{P}^1 \hookrightarrow \mathbb{P}^2 as a line (g12g^2_1) or conic (g22g^2_2).
  • Genus 11: plane cubic (g32g^2_3), the Weierstrass embedding.
  • Genus 22: project from P3\mathbb{P}^3 (degree 55 embedding) to P2\mathbb{P}^2: get a plane quintic with (22)=1\binom{2}{2} = 1 node... actually, by the degree-genus formula g=(dβˆ’1)(dβˆ’2)2βˆ’Ξ΄g = \frac{(d-1)(d-2)}{2} - \delta, a genus-22 curve maps birationally to a plane quartic with Ξ΄=1\delta = 1 node. So the minimal plane model has degree 44 with one node.
  • Genus 33, non-hyperelliptic: smooth plane quartic (g42g^2_4, no singularities needed).
  • Genus 33, hyperelliptic: plane model of degree 55 with 22 nodes (since 3=6βˆ’3=4β‹…32βˆ’33 = 6 - 3 = \frac{4 \cdot 3}{2} - 3... more carefully, degree 55 with Ξ΄=3\delta = 3 nodes).

Geometric Riemann--Roch

Theorem4.5Geometric Riemann–Roch

Let Cβ†ͺPgβˆ’1C \hookrightarrow \mathbb{P}^{g-1} be a non-hyperelliptic canonical curve, and let D=P1+P2+β‹―+PdD = P_1 + P_2 + \cdots + P_d be an effective divisor of degree dd on CC. Then:

r(D)=dβˆ’1βˆ’dim⁑P1P2β‹―Pdβ€Ύr(D) = d - 1 - \dim \overline{P_1 P_2 \cdots P_d}

where r(D)=β„“(D)βˆ’1r(D) = \ell(D) - 1 is the projective dimension of ∣D∣|D|, and P1β‹―Pdβ€Ύ\overline{P_1 \cdots P_d} denotes the linear span of the points P1,…,PdP_1, \ldots, P_d in Pgβˆ’1\mathbb{P}^{g-1}.

Equivalently: β„“(D)=dβˆ’dim⁑P1β‹―Pdβ€Ύ\ell(D) = d - \dim \overline{P_1 \cdots P_d}, or rearranging:

dim⁑P1β‹―Pdβ€Ύ=dβˆ’β„“(D)=dβˆ’(deg⁑Dβˆ’g+1+i(D))=gβˆ’1βˆ’i(D).\dim \overline{P_1 \cdots P_d} = d - \ell(D) = d - (\deg D - g + 1 + i(D)) = g - 1 - i(D).

So a divisor is special (i(D)>0i(D) > 0) if and only if its points fail to be in general position in canonical space (the span has dimension less than min⁑(dβˆ’1,gβˆ’1)\min(d - 1, g - 1)).

ExampleGeometric Riemann--Roch: three collinear points on a genus-3 quartic

Let CβŠ‚P2C \subset \mathbb{P}^2 be a smooth plane quartic (canonical genus 33, so gβˆ’1=2g - 1 = 2 and Pgβˆ’1=P2\mathbb{P}^{g-1} = \mathbb{P}^2).

Take D=P1+P2+P3D = P_1 + P_2 + P_3 where P1,P2,P3P_1, P_2, P_3 are collinear (they lie on a line LL). Since L∩CL \cap C consists of 44 points, LL cuts out a canonical divisor K∼P1+P2+P3+P4K \sim P_1 + P_2 + P_3 + P_4.

  • dim⁑P1P2P3β€Ύ=1\dim \overline{P_1 P_2 P_3} = 1 (a line in P2\mathbb{P}^2), not min⁑(2,2)=2\min(2, 2) = 2.
  • Geometric RR: r(D)=3βˆ’1βˆ’1=1r(D) = 3 - 1 - 1 = 1, so β„“(D)=2\ell(D) = 2, i.e., DD moves in a g31g^1_3.
  • Alternatively: i(D)=gβˆ’1βˆ’dim⁑P1P2P3β€Ύ=2βˆ’1=1i(D) = g - 1 - \dim \overline{P_1 P_2 P_3} = 2 - 1 = 1, so β„“(D)=3βˆ’3+1+1=2\ell(D) = 3 - 3 + 1 + 1 = 2. Confirmed.

This g31g^1_3 is cut out by the pencil of lines through P4P_4: each line through P4P_4 meets CC in P4P_4 plus three more points, giving a 11-parameter family of degree-33 divisors.

ExampleGeometric Riemann--Roch: general vs. special divisors

On a non-hyperelliptic curve of genus g=4g = 4 canonically embedded in P3\mathbb{P}^3:

General DD of degree 33: three points P1,P2,P3P_1, P_2, P_3 in general position span a plane (dim⁑=2\dim = 2). So r(D)=3βˆ’1βˆ’2=0r(D) = 3 - 1 - 2 = 0, meaning β„“(D)=1\ell(D) = 1 and DD does not move: it is a g30g^0_3.

Special DD of degree 33: three collinear points P1,P2,P3P_1, P_2, P_3 (lying on a line LβŠ‚P3L \subset \mathbb{P}^3). Then dim⁑P1P2P3β€Ύ=1\dim \overline{P_1 P_2 P_3} = 1, and r(D)=3βˆ’1βˆ’1=1r(D) = 3 - 1 - 1 = 1, so β„“(D)=2\ell(D) = 2. This is a g31g^1_3. Such a line LL meets the canonical curve (degree 66) in at most 66 points, and if it meets in exactly 33 points of DD, the residual intersection gives a g31g^1_3.

ExampleGeometric Riemann--Roch on a genus-5 canonical curve

A non-hyperelliptic genus-55 curve Cβ†ͺP4C \hookrightarrow \mathbb{P}^4 has deg⁑=2gβˆ’2=8\deg = 2g - 2 = 8.

Consider D=P1+P2+P3+P4D = P_1 + P_2 + P_3 + P_4 (four points).

Case 1: the four points are in general position, spanning a P3βŠ‚P4\mathbb{P}^3 \subset \mathbb{P}^4. Then r(D)=4βˆ’1βˆ’3=0r(D) = 4 - 1 - 3 = 0, so β„“(D)=1\ell(D) = 1: the divisor is rigid.

Case 2: the four points lie on a plane Ξ β‰…P2\Pi \cong \mathbb{P}^2. Then dim⁑Dβ€Ύ=2\dim \overline{D} = 2 and r(D)=4βˆ’1βˆ’2=1r(D) = 4 - 1 - 2 = 1, so β„“(D)=2\ell(D) = 2: this is a g41g^1_4.

Case 3 (unlikely for general CC): all four points are collinear. Then dim⁑Dβ€Ύ=1\dim \overline{D} = 1 and r(D)=4βˆ’1βˆ’1=2r(D) = 4 - 1 - 1 = 2, so β„“(D)=3\ell(D) = 3: a g42g^2_4 (plane model of degree 44, giving Cβ†’P2C \to \mathbb{P}^2 with degree 44 image). By Brill--Noether: ρ(5,2,4)=5βˆ’3(5βˆ’4+2)=5βˆ’9=βˆ’4<0\rho(5, 2, 4) = 5 - 3(5 - 4 + 2) = 5 - 9 = -4 < 0, so a general genus-55 curve has no g42g^2_4.


Canonical embedding: detailed examples

ExampleCanonical curve of genus 3: the plane quartic in detail

Let CC be a non-hyperelliptic genus-33 curve, canonically embedded as a quartic CβŠ‚P2C \subset \mathbb{P}^2.

Bitangent lines: a bitangent to CC is a line LL meeting CC at two points P,QP, Q each with multiplicity β‰₯2\geq 2, so Lβ‹…C=2P+2QL \cdot C = 2P + 2Q. Since Lβ‹…C∼KCL \cdot C \sim K_C, we get KC∼2P+2QK_C \sim 2P + 2Q, hence P+QP + Q is a theta-characteristic (2(P+Q)∼KC2(P+Q) \sim K_C). A smooth quartic has exactly 2828 bitangent lines.

Flex points: a flex is a point where Lβ‹…Cβ‰₯3PL \cdot C \geq 3P. If Lβ‹…C=3P+QL \cdot C = 3P + Q, then β„“(3P)β‰₯2\ell(3P) \geq 2. A smooth quartic has 2424 flex points.

Linear systems: ∣KC∣=g42|K_C| = g^2_4 (lines). A pencil of lines through a point Q∈CQ \in C cuts out a g31g^1_3 (after removing QQ from each divisor): the residual system ∣KCβˆ’Q∣|K_C - Q| is a g31g^1_3.

ExampleCanonical curve of genus 4 in detail

A non-hyperelliptic genus-44 curve Cβ†ͺP3C \hookrightarrow \mathbb{P}^3 has degree 66 and lies on a unique quadric QQ.

Case 1: QQ smooth. Then Q≅P1×P1Q \cong \mathbb{P}^1 \times \mathbb{P}^1 (over kˉ\bar{k}), and CC has type (3,3)(3,3) on QQ. The two rulings of QQ cut out two g31g^1_3's on CC. Thus every non-hyperelliptic genus-44 curve with smooth quadric carries exactly two g31g^1_3's.

Case 2: QQ a quadric cone. The rulings through the vertex cut out a single g31g^1_3. Thus CC carries a unique g31g^1_3 and is called trigonal.

Using Brill--Noether: ρ(4,1,3)=4βˆ’2(4βˆ’3+1)=4βˆ’4=0\rho(4, 1, 3) = 4 - 2(4 - 3 + 1) = 4 - 4 = 0, consistent with finitely many g31g^1_3's on a general genus-44 curve.


Ample vs. very ample: summary for curves

RemarkAmpleness hierarchy for divisors on curves

On a smooth projective curve CC of genus gg, we have the following hierarchy for a divisor DD:

Ample (deg⁑D>0\deg D > 0): some multiple nDnD gives an embedding. On a curve, this is purely a degree condition.

Base-point-free (β„“(Dβˆ’P)=β„“(D)βˆ’1\ell(D - P) = \ell(D) - 1 for all PP): gives a morphism Ο•D:Cβ†’Pr\phi_D : C \to \mathbb{P}^r. Guaranteed if deg⁑Dβ‰₯2g\deg D \geq 2g.

Very ample (separates points and tangent vectors): gives an embedding Ο•D:Cβ†ͺPr\phi_D : C \hookrightarrow \mathbb{P}^r. Guaranteed if deg⁑Dβ‰₯2g+1\deg D \geq 2g + 1, but the canonical divisor (deg⁑=2gβˆ’2\deg = 2g - 2) is very ample precisely when CC is non-hyperelliptic.

The gaps between these conditions are where the interesting geometry lies: KCK_C is ample for gβ‰₯2g \geq 2, base-point-free for gβ‰₯2g \geq 2, but very ample only for non-hyperelliptic curves.


Brill--Noether theory (preview)

RemarkBrill--Noether number

For a curve of genus gg, the expected dimension of the variety Gdr(C)G^r_d(C) parametrizing gdrg^r_d's on CC is the Brill--Noether number:

ρ(g,r,d)=gβˆ’(r+1)(gβˆ’d+r).\rho(g, r, d) = g - (r + 1)(g - d + r).

The Brill--Noether theorem (Griffiths--Harris, 1980) states that for a general curve of genus gg:

  • If ρβ‰₯0\rho \geq 0, then Gdr(C)β‰ βˆ…G^r_d(C) \neq \emptyset and dim⁑Gdr(C)=ρ\dim G^r_d(C) = \rho.
  • If ρ<0\rho < 0, then Gdr(C)=βˆ…G^r_d(C) = \emptyset.

This applies to a general curve in the moduli space Mg\mathcal{M}_g. Special curves (hyperelliptic, trigonal, etc.) may carry unexpected linear systems.

ExampleBrill--Noether numbers

Existence of gd1g^1_d on a general genus-gg curve: ρ(g,1,d)=gβˆ’2(gβˆ’d+1)=2dβˆ’gβˆ’2\rho(g, 1, d) = g - 2(g - d + 1) = 2d - g - 2. This is β‰₯0\geq 0 when dβ‰₯(g+2)/2d \geq (g + 2)/2, i.e., dβ‰₯⌈g/2βŒ‰+1d \geq \lceil g/2 \rceil + 1.

  • g=2g = 2: dβ‰₯2d \geq 2. Every genus-22 curve has a g21g^1_2 (hyperelliptic). ρ(2,1,2)=0\rho(2,1,2) = 0: finitely many.
  • g=3g = 3: dβ‰₯3d \geq 3. ρ(3,1,3)=0\rho(3,1,3) = 0: finitely many g31g^1_3's on a general genus-33 curve.
  • g=4g = 4: dβ‰₯3d \geq 3. ρ(4,1,3)=0\rho(4,1,3) = 0: finitely many g31g^1_3's.
  • g=5g = 5: dβ‰₯4d \geq 4. ρ(5,1,4)=0\rho(5,1,4) = 0: finitely many g41g^1_4's.
  • g=6g = 6: dβ‰₯4d \geq 4. ρ(6,1,4)=0\rho(6,1,4) = 0.

Existence of gd2g^2_d: ρ(g,2,d)=gβˆ’3(gβˆ’d+2)=3dβˆ’2gβˆ’6\rho(g, 2, d) = g - 3(g - d + 2) = 3d - 2g - 6. For g=5g = 5: 3dβˆ’16β‰₯03d - 16 \geq 0 requires dβ‰₯6d \geq 6, and ρ(5,2,6)=2\rho(5, 2, 6) = 2. A general genus-55 curve has a 22-dimensional family of plane models of degree 66.


Summary

RemarkKey principles of linear systems on curves
  1. Complete linear systems ∣Dβˆ£β‰…Pβ„“(D)βˆ’1|D| \cong \mathbb{P}^{\ell(D) - 1} parametrize effective divisors in a given linear equivalence class.

  2. Base-point-free linear systems give morphisms to projective space; very ample systems give embeddings.

  3. The gdrg^r_d notation concisely encodes degree and dimension: a gdrg^r_d is a linear system of degree dd and dimension rr.

  4. The canonical system ∣KC∣=g2gβˆ’2gβˆ’1|K_C| = g^{g-1}_{2g-2} is the most important linear system on a curve. It is very ample (giving the canonical embedding) exactly when CC is non-hyperelliptic.

  5. Geometric Riemann--Roch translates the numerical statement r(D)=dβˆ’1βˆ’dim⁑Dβ€Ύr(D) = d - 1 - \dim \overline{D} into a geometric criterion: a divisor is special if and only if its points are in "special position" (linearly dependent) in canonical space.

  6. Brill--Noether theory governs the existence of gdrg^r_d's on a general curve, with the Brill--Noether number ρ(g,r,d)=gβˆ’(r+1)(gβˆ’d+r)\rho(g, r, d) = g - (r+1)(g - d + r) being the key invariant.