The 23rd (probably?) lecture of algebraic curves by Karl Christ
We return to the case that d>2r. Assume that C⊆Pr is extremal, i.e.
non-degenerate
g(C)=π(d,r), where π(d,r) is the bound from Castelnuova's bound.
Set H to be a general hyperplane and let D=C∩H as always. Then
α1=h0(C,OC(D))=r+1
and on the level of quadrics
α2=h0(C,OC(2D))=(r+1)+(2r−1)=3r.
So for an extremal curve, I exactly know the dimension of the restriction of quadrics to C: we've got the map
φ:H0(Pr,OPr(2))↠H0(C,OC(2))
where the domain has dimension (2r+2) and the codomain has dimension 3r, so
dim(kerφ)=2(r+2)(r+1)−3r=2r2−3r+2=(2r−1).
This is the number of quadrics containing a rational normal scroll.
New questions: What is ρ=⋂Qi, where the Qi are the (2r−1) linearly independent quadrics containing C?
Pick a general hyperplane H and consider D=C∩H. Then ∣D∣=d≥2r+1. The points in D are in linearly general position and impose 2r−1 conditions on quadrics. By Castelnuova's 2r+3 lemma, these is a rational normal curve CD⊆H containing D.
Claim: CD⊆S. Suppose there is Qi such that CD⊆Qi. Then
CD.Qi=2⋅(r−1)≥d=deg(D)
because D⊆CD∩Qi. The count above is simply Bezout's theorem. This is a contradiction however, since d>2r, so we can't have 2⋅(r−1)≥d. Thus CD⊆Qi implies that Cd∩Qi has dimension 0.
Conversely: CD is cut out by (2r−1) quadrics in H≅Pr−1 which implies S∩H=CD. A rational normal curve in Pr is cut out by the 2×2 minors of a matrix of the form
(x0x1x1x2x2x3……xr−1xr),
of which there are (2r). Because CD is a rational normal curve in H≅Pr−1, it is therefore cut out by (2r−1) quadrics. We know already that CD vanishes on the Qi, of which there are (2r−1), so there are no degrees of freedom left and hence S∩H⊃CD. This gives us both inclusions and hence S∩H=CD.
Summary: We get that ⋂Qi=S gives a rational normal curve of degree r−1 when we intersect with a general hyperplane:
CD:=S∩H.
This implies S is a surface of degree r−1, and hence that C lies on a rational normal scroll or on P2→P5 (the Veronese embedding [x:y:z]↦[x2:y2:z2:xy:xz:yz:]).
Suppose S=Xa1,a2 with a1=0 is a smooth, rational normal scroll. We saw (or claimed without proof? Can't remember) last time that Pic(S)=⟨L,H⟩ with L.L=0, L.H=1 and H.H=r−1.
Recall if you have a curve C with bidegree (d1,d2) then 2g−2=KS.C+C.C and C=αL+βH. Then
KS=K1⋅L+K2⋅H.
From L you have
−2=KS⋅L+L2=K2
and for H
−2=KSH+H2=K1+K2(r−1)+(r−1)=K1−(r−1).
This implies K1=r−3. Hence KS=(r−3)L−2H, 2g(C)−2=KS.C+C.C and so
g(C)=2(α−1)(2m−α)(r−1)+ϵ⋅(α−1),
hence d−1=m(r−1)+ϵ. So g is maximal for α=m+1, β=ϵ+2−r for any ϵ or α=m and β=1 and ϵ=0. For these values, equality in Castelneuovo's bound is achieved.
Theorem: Let r≥3 and d≥2r+1. Then extremal curves of degree d exist and are of the following forms:
The image of a smooth degree k curve in P2 under the Veronese embedding P2→P5 (so d=2k and r=5)
A non-singular member of ∣mH+L∣ on a rational normal scroll
A non-singular member of the linear system ∣(m+1)H−(r−ϵ−2)L∣ on a rational normal scroll.
Cor: In case (3),
C admits a
gm+11 cut out by
L.