Text size

Section 120

Anticanonical sections and the distinguished surface

For local calculations write X=XUX=X_U and C=CUC=C_U. Let h=H1h=H_1, f1=H3f_1=H_3, and H=H2H=H_2. The ambient realization of Lemma 117.1 has equation s2q=s1Φ,Φ=uF+vG,s_2q=s_1\Phi,\qquad \Phi=uF+vG, in 𝔽1×ℙ2\mathbb F_1\times\mathbb P^2. The homogeneous coordinates u,v,s1,s2u,v,s_1,s_2 of 𝔽1\mathbb F_1 have classes f1,f1,h−f1,hf_1,f_1,h-f_1,h, respectively. The three plane coordinates x0,x1,x2x_0,x_1,x_2 have class HH. The irrelevant loci on 𝔽1\mathbb F_1 are u=v=0u=v=0 and s1=s2=0s_1=s_2=0. The distinguished surface EE is (s1=0)(s_1=0), of class h−f1h-f_1.

Let E′E' be the exceptional divisor of π\pi and let ee be its defining section. The global sections s1=q/es_1=q/e and s2=Φ/es_2=\Phi/e satisfy es1=q,es2=Φ,es_1=q,\qquad es_2=\Phi, and their divisor classes are [e]=2H−h+f1,[s1]=h−f1,[s2]=h.[e]=2H-h+f_1,\qquad [s_1]=h-f_1,\qquad [s_2]=h. These identities are identities of actual global sections. They will suffice for the inclusions used in the refined series.

Lemma 120.1. For every m≥1m\ge1, ambient sections restrict surjectively onto H0(XU,−mKXU)H^0(X_U,-mK_{X_U}). More explicitly, this space is spanned by the restrictions of s1m−rs2ruiv2m−r−ix0ax1bx2c,0≤r≤m,0≤i≤2m−r,a+b+c=m.s_1^{m-r}s_2^r u^i v^{2m-r-i} x_0^a x_1^b x_2^c, \quad 0\le r\le m,\quad 0\le i\le2m-r,\quad a+b+c=m. Its dimension is (2m+1)(2m2+2m+1)(2m+1)(2m^2+2m+1).

Proof. The anticanonical class is h+f1+Hh+f_1+H, while the hypersurface has class h+2Hh+2H. Its restriction sequence has kernel 𝒪𝔽1×ℙ2((m−1)h+mf1+(m−2)H).\mathcal O_{\mathbb F_1\times\mathbb P^2} \bigl((m-1)h+mf_1+(m-2)H\bigr). For m=1m=1, the factor 𝒪ℙ2(−1)\mathcal O_{\mathbb P^2}(-1) has no cohomology. For m≥2m\ge2, pushing the first factor down to ℙ1\mathbb P^1 gives line bundles of degrees 2m−1,2m−2,…,m2m-1,2m-2,\ldots,m, all nonnegative. Künneth’s formula therefore gives zero first cohomology in either case, proving surjectivity. Comparing the two degrees on 𝔽1\mathbb F_1 gives precisely the displayed monomials.

The dimensions of the ambient space and the kernel are (m+1)2(3m+2)(m+2)4,m2(3m+1)(m−1)4,\frac{(m+1)^2(3m+2)(m+2)}4, \qquad \frac{m^2(3m+1)(m-1)}4, respectively. Their difference is the asserted dimension. This proof uses the homogeneous coordinates of the ambient product and its hypersurface equation; it requires no assertion about generators of the whole Cox ring of XUX_U. ◻

120.1 A flag on E

The surface E≅ℙ1×ℙ1E\cong\mathbb P^1\times\mathbb P^1 has two ruling classes α,β\alpha,\beta, where f1|E=αf_1|_E=\alpha and H|E=2βH|_E=2\beta. Thus h|E=0h|_E=0, E|E=−αE|_E=-\alpha, and E′|E∼α+4βE'|_E\sim\alpha+4\beta.

We also need the following calculation for a particular singular central fibre. It remains valid when the graph curve in EE degenerates to a reduced curve with one vertical component, provided that the flag ruling does not meet that component.

Lemma 120.2. Let XX be a smooth member, or the nodal central member obtained from the net ⟨Y0Y2−Y12,Y1Y2,Y02⟩.\langle Y_0Y_2-Y_1^2,\ Y_1Y_2,\ Y_0^2\rangle. Let p∈Ep\in E and let C∈|β|C\in|\beta| be the ruling through pp. In the nodal case, suppose that CC is disjoint from the vertical component of E′|EE'|_E. Then SX(E)=1724,S(WE;C)=1112,S(WE,C;p)≤78.S_X(E)=\frac{17}{24},\qquad S(W^E;C)=\frac{11}{12},\qquad S(W^{E,C};p)\le\frac78. Consequently δp(X)≥12/11\delta_p(X)\ge12/11.

Proof. For 0≤u≤3/20\le u\le3/2, the decomposition of −KX−uE-K_X-uE is uP(u)N(u)[0,1](1−u)h+H+(1+u)f10[1,3/2](3−2u)H+2f1(u−1)E′.\begin{array}{c|c|c} u&P(u)&N(u)\\ \hline {[0,1]}&(1-u)h+H+(1+u)f_1&0\\ {[1,3/2]}&(3-2u)H+2f_1&(u-1)E'. \end{array} The positive parts are nef, since h,H,f1h,H,f_1 are pullbacks of globally generated divisors. The second row is the fixed-part decomposition: on a general exceptional fibre of π\pi, the original divisor has degree 1−u1-u, so every effective representative contains each component of E′E' with coefficient at least u−1u-1. After this part is removed, the remainder is pulled back from ℙ1×ℙ2\mathbb P^1\times\mathbb P^2 and is nef. The same argument applies to the nodal member, whose centre is a reduced union of two smooth components. Their exceptional divisors are both generically ordinary blow-ups, so the coefficient bound applies to each. For u>3/2u>3/2 the residual plane degree 3−2u3-2u is negative, proving that the threshold is 3/23/2. The common intersection numbers give SX(E)=124(∫01(24−12u−6u2)du+∫13/26(3−2u)2du)=1724.S_X(E)=\frac1{24}\left( \int_0^1(24-12u-6u^2)\,du+ \int_1^{3/2}6(3-2u)^2\,du\right)=\frac{17}{24}.

Write P(u)|E=a(u)α+b(u)βP(u)|_E=a(u)\alpha+b(u)\beta, where (a(u),b(u))={(1+u,2),0≤u≤1,(2,6−4u),1≤u≤3/2.(a(u),b(u))= \begin{cases} (1+u,2),&0\le u\le1,\\ (2,6-4u),&1\le u\le3/2. \end{cases} For 0≤v≤b(u)0\le v\le b(u) the divisor P(u)|E−vC=a(u)α+(b(u)−v)βP(u)|_E-vC=a(u)\alpha+(b(u)-v)\beta is nef; it has volume 2a(u)(b(u)−v)2a(u)(b(u)-v) and intersection a(u)a(u) with CC. Since CC is not a component of E′|EE'|_E, there is no primary fixed contribution. Therefore S(WE;C)=18∫03/2∫0b(u)2a(u)(b(u)−v)dvdu=18(∫014(1+u)du+∫13/22(6−4u)2du)=1112.\begin{aligned} S(W^E;C) &=\frac18\int_0^{3/2}\int_0^{b(u)} 2a(u)(b(u)-v)\,dv\,du\\ &=\frac18\left(\int_0^1 4(1+u)\,du+ \int_1^{3/2}2(6-4u)^2\,du\right)=\frac{11}{12}. \end{aligned} The moving point contribution is 18∫03/2a(u)2b(u)du=18(∫012(1+u)2du+∫13/24(6−4u)du)=56.\frac18\int_0^{3/2}a(u)^2b(u)\,du =\frac18\left(\int_0^1 2(1+u)^2\,du+ \int_1^{3/2}4(6-4u)\,du\right)=\frac56. The curve E′|EE'|_E meets CC transversally at one point. For a smooth member this is the graph property. For the nodal member its other component is disjoint from CC by assumption, and its graph component has the same property. The only possible fixed point contribution is consequently 14∫13/2∫06−4u2(u−1)dvdu=124.\frac14\int_1^{3/2}\int_0^{6-4u}2(u-1)\,dv\,du =\frac1{24}. This proves the three claimed values and bounds. The nodal singularity is disjoint from EE, so the same adjunction inequality applies there. Finally, δp(X)≥min{2417,1211,87}=1211.\delta_p(X)\ge \min\left\{\frac{24}{17},\frac{12}{11},\frac87\right\} =\frac{12}{11}. ◻

120.2 Adjunction for a curve extraction

We recall explicitly the local form of the refinement inequality that we use for curves. A local threshold always takes the infimum over divisors whose centres contain the specified point.

Lemma 120.3. Let π:Y→X\pi\colon Y\to X be a plt extraction of a smooth curve Γ\Gamma in a smooth Fano threefold, with exceptional divisor GG. Write (KY+G)|G=KG+ΔG.(K_Y+G)|_G=K_G+\Delta_G. Put L=−KXL=-K_X, and let Wm,jGW^G_{m,j} be the image of the restriction map from H0(Y,mπ*L−jG)H^0(Y,m\pi^*L-jG) to H0(G,(mπ*L−jG)|G).H^0\bigl(G,(m\pi^*L-jG)|_G\bigr). For p∈Γp\in\Gamma one has δp(X)≥min{AX(G)SX(G),infq∈π−1(p)∩Gδq(G,ΔG;WG)}.\delta_p(X)\ge \min\left\{\frac{A_X(G)}{S_X(G)}, \inf_{q\in\pi^{-1}(p)\cap G} \delta_q(G,\Delta_G;W^G)\right\}. The restriction of π*L\pi^*L to GG is retained in this definition. In particular, it need not be trivial for a curve extraction.

Proof. This is the local refinement inequality of [4, Theorem 3.3]. We recall its basis-divisor argument. Choose an anticanonical basis compatible with the order along GG and with a valuation being tested. Removing the powers of a local equation of GG gives bases of the spaces Wm,jGW^G_{m,j}. The pullback basis divisor has the form Sm(G)G+B′m,S_m(G)G+B'_m, where B′mB'_m does not contain GG and its restriction is a basis divisor of WGW^G with the usual normalization by mh0(X,mL)m h^0(X,mL). If a positive number cc is smaller than both quantities on the right, the proof of [4, Lemma 2.21] gives Sm(WG;v)≤(1+ϵ)S(WG;v)S_m(W^G;v)\le(1+\epsilon)S(W^G;v) for all valuations of finite discrepancy and all sufficiently large mm, with the bound on mm independent of vv. Thus the same strict inequalities hold for the basis thresholds simultaneously over the proper fibre π−1(p)∩G\pi^{-1}(p)\cap G. Then adjunction makes (G,ΔG+cB′m|G)(G,\Delta_G+cB'_m|_G) log canonical over pp. Inversion of adjunction applies to (Y,G+cB′m)(Y,G+cB'_m) there. The coefficient of GG in the pullback of the corresponding pair on XX is 1−AX(G)+cSm(G)<1.1-A_X(G)+cS_m(G)<1. Decreasing the coefficient of GG therefore gives log canonicity of the original basis pair at pp. Taking the infimum over the valuations and the limit of the basis thresholds proves the inequality. Equivalently, this is the same refinement argument as for a point extraction, with the whole fibre over pp retained. ◻

The elementary decompositions along the distinguished surface and the del Pezzo fibres have already been displayed. In all point and curve extractions below, we do not compute a Zariski decomposition on the extracted threefold. We identify the actual refined sections, or embed them in complete series with the same asymptotic total dimension. The remaining decompositions and blow-ups occur on the exceptional surface.