Text size

Section 24

Three volume integrals and the reduced-fibre flags

Lemma 24.1. The nef cone is generated by H,BH,B, the effective cone is generated by E,BE,B, and SX(H)=38,SX(B)=4172,SX(E)=59.S_X(H)=\frac38,\qquad S_X(B)=\frac{41}{72},\qquad S_X(E)=\frac59. Every prime divisor on XX has positive beta invariant.

Proof. A fibre of E→ℓE\to\ell and a general conic fibre test the two boundaries of the nef cone. For a prime divisor D≠ED\ne E, write D∼aH−bED\sim aH-bE, with integers a≥1a\ge1 and b≥0b\ge0. A general smooth cubic section as in Lemma 23.4 is not a component of DD. Intersect its effective restriction with the nef conic class FF to obtain 2a−2b≥02a-2b\ge0. Thus D=(a−b)H+bBD=(a-b)H+bB has nonnegative integral coefficients. Since H=E+BH=E+B, this also proves the assertion about the effective cone.

For 0≤t≤10\le t\le1, the three nef divisors in the calculation have volumes vol(L−tH)=3(2−t)3−3(2−t),vol(L−tB)=3+9(1−t)+6(1−t)2,vol(L−tE)=24−6(1+t)2.\begin{split} \mathop{\mathrm{vol}}(L-tH)&=3(2-t)^3-3(2-t),\\ \mathop{\mathrm{vol}}(L-tB)&=3+9(1-t)+6(1-t)^2,\\ \mathop{\mathrm{vol}}(L-tE)&=24-6(1+t)^2. \end{split} The first and third volumes vanish for t≥1t\ge1. For 1≤t≤21\le t\le2, L−tB=(2−t)H+(t−1)EL-tB=(2-t)H+(t-1)E, with fixed part (t−1)E(t-1)E and volume 3(2−t)33(2-t)^3; beyond two the volume is zero. The fixed-part assertion follows by testing a general fibre of E→ℓE\to\ell, or from π*𝒪X(kE)=𝒪V\pi_*\mathcal O_X(kE)=\mathcal O_V for k≥0k\ge0. Hence the three integrals, divided by 1818, are 27/418=38,19/2+3/418=4172,1018=59.\frac{27/4}{18}=\frac38,\qquad \frac{19/2+3/4}{18}=\frac{41}{72},\qquad \frac{10}{18}=\frac59. If D≠ED\ne E, either D−HD-H or D−BD-B is effective, by the integral expression above. Monotonicity of volume gives SX(D)≤max{3/8,41/72}<1=AX(D)S_X(D)\le\max\{3/8,41/72\}<1=A_X(D). The calculation for EE gives the remaining assertion. ◻

Fix a smooth cubic S∈|B|S\in|B| from Lemma 23.4. For u=j/mu=j/m, its refined series has the following moving and fixed parts: (24.1)uDuNu|S[0,1]A+(1−u)F0[1,2](2−u)A(u−1)c.\label{f211:eq:surface} \begin{array}{c|cc} u&D_u&N_u|_S\\\hline {[0,1]}&A+(1-u)F&0\\ {[1,2]}&(2-u)A&(u-1)c. \end{array} These are asymptotically complete. In the first open interval this follows from ampleness and Serre vanishing. In the second, remove the fixed multiple of EE and restrict the complete series on VV to its hyperplane section. Endpoint errors disappear in the limits.

For a smooth rational curve Z⊂SZ\subset S different from cc, write Du−vZ=Pu,v+Nu,vD_u-vZ=P_{u,v}+N_{u,v} for its surface Zariski decomposition and put d=Pu,vZd=P_{u,v}Z. With nu=max{0,u−1}n_u=\max\{0,u-1\}, the flag constants are (24.2)𝒮Z=16∫∫Pu,v2dvdu,𝒯p=16∫∫d2dvdu+13∫∫dordp((nuc+Nu,v)|Z)dvdu.\label{f211:eq:constants} \mathcal S_Z=\frac16\int\!\int P_{u,v}^2\,dv\,du,\qquad \mathcal T_p=\frac16\int\!\int d^2\,dv\,du +\frac13\int\!\int d\,\mathop{\mathrm{ord}}_p((n_uc+N_{u,v})|_Z)\,dv\,du. The inner integral runs to the pseudo-effective threshold. The normalization is h0(X,mL)=3m3+O(m2)h^0(X,mL)=3m^3+O(m^2). The proof of Lemma 12.1, with L3=18L^3=18, gives (24.3)δp(X)≥min{7241,1𝒮Z,1𝒯p}.\label{f211:eq:AZflag} \delta_p(X)\ge\min\left\{\frac{72}{41},\frac1{\mathcal S_Z},\frac1{\mathcal T_p}\right\}. Only decompositions on the cubic surface and the elementary fixed part on XX are used here.

Lemma 24.2 (The smooth-conic calculation). If ZZ is a smooth conic fibre and p∈Zp\in Z, put r=(cZ)pr=(cZ)_p. Then 𝒮Z=2948,𝒯p=8+r9.\mathcal S_Z=\frac{29}{48},\qquad \mathcal T_p=\frac{8+r}{9}. In particular, if p∉Ep\notin E, then δp(X)≥9/8\delta_p(X)\ge9/8.

Proof. For 0≤u≤10\le u\le1, put s=1−us=1-u. The decomposition is vPu,vNu,v[0,s+1/2]A+(s−v)F0[s+1/2,s+1](1+s−v)(A+c)(2v−2s−1)c.\begin{array}{c|cc} v&P_{u,v}&N_{u,v}\\\hline {[0,s+1/2]}&A+(s-v)F&0\\ {[s+1/2,s+1]}&(1+s-v)(A+c)&(2v-2s-1)c. \end{array} For 1≤u≤21\le u\le2, scale the s=0s=0 decomposition by a=2−ua=2-u. Nefness, orthogonality and the threshold follow exactly as in Lemma 12.2: both FF and A+cA+c are nef. The three inner integrals on the first interval are 3s+2s2+7/63s+2s^2+7/6, 4s+8/34s+8/3, and 1/61/6, for P2P^2, d2d^2, and dcoeffcNd\operatorname{coeff}_cN, respectively. On the second interval they are 7a3/67a^3/6, 8a3/38a^3/3, and a3/6a^3/6. Since ∫12a3du=1/4\int_1^2a^3du=1/4, these give 𝒮Z=16(103+724)=2948,16∫∫d2=89,13∫∫dcoeffcN=572.\mathcal S_Z=\frac16\left(\frac{10}{3}+\frac7{24}\right)=\frac{29}{48},\quad \frac16\int\!\int d^2=\frac89,\quad \frac13\int\!\int d\operatorname{coeff}_cN=\frac5{72}. The original fixed part contributes 16∫12(u−1)3(2−u)2du⋅r=r24.\frac16\int_1^2(u-1)\,3(2-u)^2du\cdot r=\frac r{24}. Adding the last two fixed contributions gives r/9r/9. If p∉Ep\notin E, then r=0r=0, and (24.3) proves the assertion. ◻

Lemma 24.3 (The reducible-conic calculation). Suppose the conic fibre is reduced and singular, and the smooth cubic of Lemma 23.4 is available. Then δp(X)≥36/35\delta_p(X)\ge36/35 at every point of that fibre.

Proof. Write the fibre as Z+Z′Z+Z', choosing p∈Zp\in Z. The two curves have square −1-1 and meet transversely once. For 0≤u≤10\le u\le1, set s=1−us=1-u. The positive and negative parts of A+sF−vZA+sF-vZ are vPu,vNu,v[0,1]A+sF−vZ0[1,1+s](A−Z)+(1+s−v)F(v−1)Z′.\begin{array}{c|cc} v&P_{u,v}&N_{u,v}\\\hline {[0,1]}&A+sF-vZ&0\\ {[1,1+s]}&(A-Z)+(1+s-v)F&(v-1)Z'. \end{array} The threshold is 1+s1+s. The classes A−Z,FA-Z,F are conic pencils and are nef, so the displayed decomposition has the required properties. For 1≤u≤21\le u\le2, aA−vZaA-vZ is nef up to its threshold v=av=a, where a=2−ua=2-u.

Put ϵ=1\epsilon=1 if p=Z∩Z′p=Z\cap Z' and zero otherwise, and μ=(cZ)p≤1\mu=(cZ)_p\le1. The inner integrals in the first interval are 5/3+4s+2s25/3+4s+2s^2, 7/3+4s7/3+4s, and s2s^2. In the second they are 5a3/35a^3/3, 7a3/37a^3/3, and zero. Consequently 𝒮Z=16(133+512)=1924,𝒯p=5972+ϵ9+μ24≤3536.\mathcal S_Z=\frac16\left(\frac{13}{3}+\frac5{12}\right)=\frac{19}{24},\qquad \mathcal T_p=\frac{59}{72}+\frac{\epsilon}{9}+\frac{\mu}{24} \le\frac{35}{36}. The final term uses the same fixed-part integral as in Lemma 24.2. Apply (24.3). ◻

Example 24.4 (Why the point term must be retained). If a smooth conic is transverse to cc at pp, the flag gives 𝒯p=1\mathcal T_p=1. If it is tangent, it gives 𝒯p=10/9\mathcal T_p=10/9. Thus the cubic-surface flag alone gives no strict local threshold at these points of EE. A lower bound 9/109/10 does not indicate instability; it indicates the limitation of this flag.