Text size

Section 38

The scroll and the conic bundle

Family 2.13: genus-two sextics on quadrics

Let π:X→Q\pi:X\to Q be the blow-up of a smooth quadric threefold Q⊂ℙ4Q\subset\mathbb P^4 along a smooth curve CC of degree six and genus two. The ideal of CC on QQ is generated by quadrics. Put H=π*𝒪Q(1),E=Exc(π),B=2H−E,L=−KX=H+B.H=\pi^*\mathcal O_Q(1),\qquad E=\operatorname{Exc}(\pi), \qquad B=2H-E,\qquad L=-K_X=H+B. The construction below applies to every member of Family 2.13. In particular, it includes the cone scroll used for the symmetric example in [3, Section 5.5].

Theorem 38.1. Let XX be a smooth member of Family 2.13, and let p∈Xp\in X. According to the conic fibre through pp, one has δp(X)≥{80/77,the fibre is smooth,64/61,the fibre is reducible and p is not its node,160/159,p is the node of a reduced fibre,320/319,the fibre is a double line.\delta_p(X)\ge \begin{cases} 80/77,&\text{the fibre is smooth},\\ 64/61,&\text{the fibre is reducible and $p$ is not its node},\\ 160/159,&\text{$p$ is the node of a reduced fibre},\\ 320/319,&\text{the fibre is a double line}. \end{cases} Consequently δ(X)≥320/319\delta(X)\ge320/319, and XX is K-stable.

We first establish the common geometry and the exact surface refinement. Reduced fibres are treated on smooth quartic del Pezzo sections. Double fibres require a weighted extraction of their reduced line. As in the preceding parts, δp\delta_p includes divisors with positive-dimensional centres containing pp.

Lemma 38.2. The curve CC is the scheme-theoretic intersection of QQ with a cubic scroll T⊂ℙ4T\subset\mathbb P^4. This scroll is either smooth or a cone over a twisted cubic. In suitable coordinates its ideal is generated by the maximal minors of, respectively, (38.1)M=(x0x1x3x1x2x4),M=(x0x1x2x1x2x3).\label{f213:eq:matrices} M=\begin{pmatrix}x_0&x_1&x_3\\x_1&x_2&x_4\end{pmatrix}, \qquad M=\begin{pmatrix}x_0&x_1&x_2\\x_1&x_2&x_3\end{pmatrix}. In the second case the vertex of TT is outside QQ.

Proof. The curve is nondegenerate. Indeed, if it were contained in a hyperplane section T0T_0 of QQ, every quadric section containing CC would contain T0T_0: otherwise its intersection with T0T_0 would have degree four and could not contain CC. This contradicts the stated ideal generation. By Riemann–Roch, ℒ=𝒪C(1)\mathcal L=\mathcal O_C(1) has five sections, so the given embedding is complete.

Let f:C→ℙ1f:C\to\mathbb P^1 be the hyperelliptic double cover. The bundle V=f*ℒV=f_*\mathcal L has rank two and degree three. It is globally generated, since H1(C,ℒ(−f−1(t)))=0H^1(C,\mathcal L(-f^{-1}(t)))=0 for every t∈ℙ1t\in\mathbb P^1: the Serre-dual bundle has degree −2-2. Thus V≃𝒪(2)⊕𝒪(1)orV≃𝒪(3)⊕𝒪.V\simeq\mathcal O(2)\oplus\mathcal O(1) \quad\text{or}\quad V\simeq\mathcal O(3)\oplus\mathcal O. Evaluation embeds CC in ℙ(V)\mathbb P(V), with its map to ℙ4\mathbb P^4 induced by the tautological divisor hh. The image is the corresponding cubic scroll. If rr is a ruling fibre, then h2=3h^2=3, h⋅r=1h\cdot r=1, and the lifted curve has class 2h+dr2h+dr. Its degree six gives d=0d=0. In the cone case the negative section has zero intersection with this curve, and is disjoint from it.

The quadric QQ cannot contain the cubic scroll: every divisor on the smooth quadric threefold has even degree, since Pic(Q)=ℤ[𝒪Q(1)]\operatorname{Pic}(Q)=\mathbb Z[\mathcal O_Q(1)]. Its pullback to ℙ(V)\mathbb P(V) is therefore a nonzero divisor of class 2h2h containing CC, and consequently equals CC. This proves the scheme-theoretic assertion and, in the cone case, excludes the vertex from QQ. The displayed matrices are the usual parametrizations of the two scrolls: the first comes from [x0:x1:x2:x3:x4]=[s2a:sta:t2a:sb:tb],[x_0:x_1:x_2:x_3:x_4] =[s^2a:sta:t^2a:sb:tb], and the second is the twisted-cubic matrix with one unused coordinate. ◻

Lemma 38.3. With homogeneous coordinates [y0:y1:y2][y_0:y_1:y_2] on ℙ2\mathbb P^2, the threefold has the presentation (38.2)X={q(x)=0,M(x)y=0}⊂ℙ4×ℙ2,\label{f213:eq:incidence} X=\{q(x)=0,\ M(x)y=0\}\subset\mathbb P^4\times\mathbb P^2, a complete intersection of types (2,0),(1,1),(1,1)(2,0),(1,1),(1,1). The second projection is a flat conic bundle ψ:X→ℙ2\psi:X\to\mathbb P^2, and B=ψ*𝒪ℙ2(1)B=\psi^*\mathcal O_{\mathbb P^2}(1).

Proof. The rank-zero locus of MM is empty for the first matrix in (38.1); for the second it is the vertex, which misses QQ. Where MM has rank two, the incidence variety is the graph of the map defined by its minors. At a point of CC, invertible local row and column operations put the matrix in the form (1000ab).\begin{pmatrix}1&0&0\\0&a&b\end{pmatrix}. The ideal of CC is (a,b)(a,b), and the incidence equations become y0=0y_0=0 and ay1+by2=0ay_1+by_2=0. These are the equations of the blow-up of this regular codimension-two ideal. Hence (38.2) is the blow-up, scheme-theoretically.

For every yy, the two linear forms comprising M(x)yM(x)y are independent, as is immediate from either matrix in (38.1). They cut out a plane Πy\Pi_y. A smooth quadric threefold contains no plane, so q|Πyq|_{\Pi_y} is nonzero and defines a plane conic. These fibres all have the same Hilbert polynomial, giving flatness. The three minors define |2H−E||2H-E|, which identifies the second polarization with BB. ◻

Lemma 38.4. For all integers a,b≥0a,b\ge0, H0(X,aH+bB)=(ℂ[x0,…,x4,y0,y1,y2]/(q,(My)1,(My)2))(a,b).H^0(X,aH+bB)= \left(\mathbb C[x_0,\ldots,x_4,y_0,y_1,y_2]/ (q,(My)_1,(My)_2)\right)_{(a,b)}. The bigraded Hilbert series and the anticanonical Hilbert function are (38.3)(1−s2)(1−st)2(1−s)5(1−t)3,Nm=h0(X,mL)=(2m+1)(5m2+5m+3)3.\label{f213:eq:hilbert} \frac{(1-s^2)(1-st)^2}{(1-s)^5(1-t)^3}, \qquad N_m=h^0(X,mL)=\frac{(2m+1)(5m^2+5m+3)}3.

Proof. Twist the Koszul resolution of (38.2) by 𝒪(a,b)\mathcal O(a,b). Every term has twists at least −4-4 on ℙ4\mathbb P^4 and at least −2-2 on ℙ2\mathbb P^2, so it has no higher cohomology. Taking sections gives the asserted quotient and its Hilbert series. Its diagonal gives the second formula. ◻

Lemma 38.5. The intersection numbers and the two cones are H3=2,H2B=4,HB2=2,B3=0,L3=20,Nef(X)=ℝ≥0H+ℝ≥0B,Eff¯(X)=ℝ≥0E+ℝ≥0B.\begin{gathered} H^3=2,\qquad H^2B=4,\qquad HB^2=2,\qquad B^3=0, \qquad L^3=20,\\ \mathop{\mathrm{Nef}}(X)=\mathbb R_{\ge0}H+\mathbb R_{\ge0}B, \qquad \overline{\mathop{\mathrm{Eff}}}(X)=\mathbb R_{\ge0}E+\mathbb R_{\ge0}B. \end{gathered} If D≠ED\ne E is a prime divisor and D∼aH+bBD\sim aH+bB, then a,ba,b are nonnegative integers.

Proof. The class of the complete intersection in the product is 2H(H+B)22H(H+B)^2, giving the intersection numbers. The two contractions span the extremal rays, hence give the nef cone. For the last assertion intersect DD with a general conic fibre and with a general exceptional fibre ee of E→CE\to C. These curves are not contained in DD, and D⋅B2=2a≥0,D⋅e=b≥0.D\cdot B^2=2a\ge0,\qquad D\cdot e=b\ge0. The coefficients are integral because H,EH,E form a basis of Pic(X)\operatorname{Pic}(X) and B=2H−EB=2H-E. Since H=(E+B)/2H=(E+B)/2, every such divisor lies in the cone generated by E,BE,B, proving the effective-cone assertion. ◻