Text size

Section 71

Points on smooth fibres of the conic bundle

Let π:X→V5\pi:X\to V_5 be the blowup of a smooth twisted cubic C0C_0, let EE be its exceptional divisor, and put H=π*𝒪V5(1)H=\pi^*\mathcal O_{V_5}(1) and B=H−EB=H-E. Then −KX=2H−E-K_X=2H-E, and |B||B| defines the conic bundle η:X→ℙ2\eta:X\to\mathbb P^2. These constructions apply to every smooth member of Family 2.20.

Lemma 71.1. One has H3=5,H2E=0,HE2=−3,E3=−4,(−KX)3=26,H^3=5,\quad H^2E=0,\quad HE^2=-3,\quad E^3=-4,\quad (-K_X)^3=26, and SX(E)=12,SX(B)=63104,SX(H)=37104.S_X(E)=\frac12,\qquad S_X(B)=\frac{63}{104},\qquad S_X(H)=\frac{37}{104}. Moreover, E≅ℙ1×ℙ1E\cong\mathbb P^1\times\mathbb P^1, and B|EB|_E has bidegree (1,1)(1,1).

Proof. The intersection numbers follow from H⋅C0=3H\cdot C_0=3, g(C0)=0g(C_0)=0 and degNC0/V5=4\deg N_{C_0/V_5}=4. The two nef divisors HH and BB vanish on the fibres of the two extremal contractions, respectively, and hence generate the nef cone. If D≠ED\ne E is a prime divisor, write D∼aH−bED\sim aH-bE. Its pushforward gives a≥1a\geq1, while intersection with a general conic fibre gives 2(a−b)≥02(a-b)\geq0. Thus the effective cone is generated by EE and BB. Therefore −KX−uE-K_X-uE is nef for 0≤u≤10\leq u\leq1 and is not pseudoeffective for u>1u>1. Its volume is 40−18(1+u)2+4(1+u)340-18(1+u)^2+4(1+u)^3. Similarly, −KX−uH-K_X-uH is nef precisely up to its pseudoeffective threshold 11, and its volume is 5(2−u)3−9(2−u)+45(2-u)^3-9(2-u)+4. For BB, the positive and negative parts are uP(u)N(u)[0,1](2−u)H−(1−u)E0[1,2](2−u)H(u−1)E.\begin{array}{c|c|c} u&P(u)&N(u)\\ \hline {[0,1]}&(2-u)H-(1-u)E&0\\ {[1,2]}&(2-u)H&(u-1)E. \end{array} Consequently, SX(B)=126(∫01(5(2−u)3−9(2−u)(1−u)2+4(1−u)3)du+∫125(2−u)3du)=63104.S_X(B)=\frac1{26}\left( \int_0^1\left(5(2-u)^3-9(2-u)(1-u)^2+4(1-u)^3\right)du +\int_1^2 5(2-u)^3du\right)=\frac{63}{104}. The other two expected orders are separately SX(E)=126∫01(40−18(1+u)2+4(1+u)3)du=12,SX(H)=126∫01(5(2−u)3−9(2−u)+4)du=37104.\begin{aligned} S_X(E)&=\frac1{26}\int_0^1\bigl(40-18(1+u)^2+4(1+u)^3\bigr)du=\frac12,\\ S_X(H)&=\frac1{26}\int_0^1\bigl(5(2-u)^3-9(2-u)+4\bigr)du=\frac{37}{104}. \end{aligned}

Write E≅𝔽nE\cong\mathbb F_n, with negative section ss and fibre ff of E→C0E\to C_0. Since the normal bundle has degree 44, nn is even, and −E|E=s+(n−4)f/2-E|_E=s+(n-4)f/2. Hence B|E=s+(n+2)f/2B|_E=s+(n+2)f/2. Its nefness implies n≤2n\leq2. If n=2n=2, then B⋅s=0B\cdot s=0 but −KX⋅s=3-K_X\cdot s=3. This is impossible: ss would be a component of a conic-bundle fibre, whose total anticanonical degree is 22, and −KX-K_X is ample. Therefore n=0n=0. The morphism η|E\eta|_E is projection of the quadric E⊂ℙ3E\subset\mathbb P^3 embedded by |𝒪(1,1)||\mathcal O(1,1)| from a point outside it. It is a finite double cover with smooth branch conic. ◻

Lemma 71.2. Let q∈ℙ2q\in\mathbb P^2 have a smooth fibre Z=η−1(q)Z=\eta^{-1}(q). There is a smooth surface S∈|B|S\in|B| containing ZZ such that SS is a quintic del Pezzo surface and C=E|SC=E|_S is a smooth twisted cubic. One can identify S≅Blp1,…,p4ℙ2S\cong\operatorname{Bl}_{p_1,\ldots,p_4}\mathbb P^2 so that C∼l,Z∼2l−e1−e2−e3−e4,H|S∼3l−e1−e2−e3−e4.C\sim l,\qquad Z\sim2l-e_1-e_2-e_3-e_4,\qquad H|_S\sim3l-e_1-e_2-e_3-e_4. Here ll is the pullback of a line, the eie_i are disjoint exceptional curves, C∩ei=⌀C\cap e_i=\varnothing, and ZZ meets each eie_i transversally in one point.

Proof. Take the inverse image of a general line through qq. Since η\eta is smooth along ZZ, Bertini’s theorem gives a surface smooth along ZZ and away from ZZ. We can also arrange that its intersection with EE is a smooth member of |B|E|=|𝒪(1,1)||B|_E|=|\mathcal O(1,1)|: the lines tangent to the branch conic of the finite double cover η|E\eta|_E form a proper subset of the pencil. Let S0=π(S)S_0=\pi(S). Along C0C_0, choose local coordinates (t,x,y)(t,x,y) with C0=(x=y=0)C_0=(x=y=0) and write the equation of S0S_0 as a(t)x+b(t)y+O((x,y)2)=0a(t)x+b(t)y+O((x,y)^2)=0. If a(t)=b(t)=0a(t)=b(t)=0 at some point, then S∩ES\cap E contains the whole exceptional fibre over that point. This is impossible because S∩ES\cap E is a smooth (1,1)(1,1)-curve. Thus S0S_0 is smooth along C0C_0, and it is smooth elsewhere because SS is. Since C0C_0 is a Cartier divisor on S0S_0, its blowup is the identity. Hence π|S:S→S0\pi|_S:S\to S_0 is an isomorphism, and SS is a quintic del Pezzo surface. On it C2=1C^2=1 and (−KS)⋅C=3(-K_S)\cdot C=3. The plane model associated to |C||C| contracts four disjoint (−1)(-1)-curves and identifies CC with the strict transform of a line avoiding the four points. Finally Z∼B|S=−KS−CZ\sim B|_S=-K_S-C, giving the stated class. Its intersections with the eie_i equal 11; since ZZ is smooth, they are transverse. ◻

Lemma 71.3. Let WSW^S denote the refinement of the anticanonical complete series by SS, and let WS,ZW^{S,Z} be its further refinement by ZZ. For p∈Zp\in Z, put dp=(C⋅Z)pd_p=(C\cdot Z)_p and rp=∑i(ei⋅Z)pr_p=\sum_i(e_i\cdot Z)_p. Then S(WS;Z)=171208,S(WS,Z;p)=4752+5dp104+5rp208≤1.S(W^S;Z)=\frac{171}{208},\qquad S(W^{S,Z};p)=\frac{47}{52}+\frac{5d_p}{104}+\frac{5r_p}{208}\leq1. Equality in the latter bound is possible only when C⋅Z=2pC\cdot Z=2p. If p∉Cp\notin C, the point moment is at most 193/208193/208; if p∈Cp\in C and CC meets ZZ transversally, it equals 99/10499/104.

Proof. Put T=e1+e2+e3+e4T=e_1+e_2+e_3+e_4. On SS, C=lC=l, Z=2l−TZ=2l-T, and C⋅Z=2C\cdot Z=2. The positive part P(u,v)P(u,v) of P(u)|S−vZP(u)|_S-vZ and its negative part N(u,v)N(u,v) are given by uvP(u,v)N(u,v)[0,1][0,2−u](2−u−v)Z+C0[0,1][2−u,52−u](52−u−v)(Z+T)(v+u−2)T[1,2][0,2−u](2−u−v)Z+(2−u)C0[1,2][2−u,32(2−u)](32(2−u)−v)(Z+T)(v+u−2)T.\begin{array}{c|c|c|c} u&v&P(u,v)&N(u,v)\\ \hline {[0,1]}&[0,2-u]&(2-u-v)Z+C&0\\ {[0,1]}&[2-u,\frac52-u]&(\frac52-u-v)(Z+T)&(v+u-2)T\\ {[1,2]}&[0,2-u]&(2-u-v)Z+(2-u)C&0\\ {[1,2]}&[2-u,\frac32(2-u)]&(\frac32(2-u)-v)(Z+T)&(v+u-2)T. \end{array} Indeed, in the second and fourth rows the displayed positive part is orthogonal to every eie_i, while the eie_i are disjoint negative curves. The other displayed divisors are nef, as follows by intersecting the ten (−1)(-1)-curves of the quintic del Pezzo surface. The volume becomes zero at the right endpoint in each row. These observations prove every claimed decomposition and threshold.

We justify that these volume calculations compute the actual refined series. For an integral nef moving part MM on XX, one has M−B=KX+(M+H)M-B=K_X+(M+H). The divisor M+HM+H is nef and big, so Kawamata–Viehweg vanishing gives H1(X,𝒪X(M−B))=0H^1(X,\mathcal O_X(M-B))=0. Consequently restriction to SS is surjective, after removing the fixed multiple of EE. Likewise, for each integral nef part PP in the table, P−Z=KS+(P+C)P-Z=K_S+(P+C), and P+CP+C is nef and big because CC is nef and C2=1C^2=1. Thus restriction from SS to ZZ is surjective after removing the displayed fixed part. The resulting series on Z≅ℙ1Z\cong\mathbb P^1 is complete. The formulas below follow either by summing dimensions of these complete series and taking limits, or by the multigraded Abban–Zhuang volume formulas. Endpoint rounding contributes o(1)o(1) to the normalized moments.

There is no fixed component along ZZ in N(u)|SN(u)|_S. The surface volumes in the four rows are, respectively, 9−4u−4v,4(52−u−v)2,(2−u)(10−5u−4v),(6−3u−2v)2.9-4u-4v,\quad4(\tfrac52-u-v)^2,\quad (2-u)(10-5u-4v),\quad(6-3u-2v)^2. The surface formula is therefore S(WS;Z)=326(∫01∫02−u(9−4u−4v)dvdu+∫01∫2−u5/2−u4(52−u−v)2dvdu+∫12∫02−u(2−u)(10−5u−4v)dvdu+∫12∫2−u3(2−u)/2(6−3u−2v)2dvdu)=171208.\begin{aligned} S(W^S;Z)=\frac3{26}\bigg(&\int_0^1\int_0^{2-u}(9-4u-4v)\,dv\,du\\ &+\int_0^1\int_{2-u}^{5/2-u}4(\tfrac52-u-v)^2\,dv\,du\\ &+\int_1^2\int_0^{2-u}(2-u)(10-5u-4v)\,dv\,du\\ &+\int_1^2\int_{2-u}^{3(2-u)/2}(6-3u-2v)^2\,dv\,du\bigg) =\frac{171}{208}. \end{aligned} Write Ωi\Omega_i for the four integration regions and a(u,v)=P(u,v)⋅Za(u,v)=P(u,v)\cdot Z. Its four expressions are 22, 4(5/2−u−v)4(5/2-u-v), 2(2−u)2(2-u), and 4(3(2−u)/2−v)4(3(2-u)/2-v). The threefold fixed part restricts to (u−1)+C(u-1)_+C. Thus the point formula gives S(WS,Z;p)=326∑i∬Ωia(u,v)2dvdu+626∑i∬Ωia(u,v)((u−1)+dp+n(u,v)rp)dvdu,S(W^{S,Z};p)=\frac3{26}\sum_i\iint_{\Omega_i}a(u,v)^2\,dv\,du +\frac6{26}\sum_i\iint_{\Omega_i}a(u,v) \left((u-1)_+d_p+n(u,v)r_p\right)\,dv\,du, where n(u,v)=v+u−2n(u,v)=v+u-2 in the second and fourth rows, and is zero otherwise. The three integrals, including their normalization factors, are 47/5247/52, 5/1045/104, and 5/2085/208. This proves the formula. Since C∩T=⌀C\cap T=\varnothing, one cannot have both dp>0d_p>0 and rp>0r_p>0. Moreover dp≤2d_p\leq2 and rp≤1r_p\leq1, because the eie_i are disjoint. The assertions follow. ◻

Example 71.4. If p∈Zp\in Z avoids CC and the four exceptional curves, the point moment is 47/5247/52. At a transverse point of C∩ZC\cap Z it is 99/10499/104. At a point where CC is tangent to ZZ it is exactly 11. The last equality does not prevent a strict threefold estimate, as the next lemma shows.

Lemma 71.5. Let pp lie on a smooth fibre of η\eta, and choose SS as above. For every divisorial valuation vv with centre pp on XX, one has AX(v)−SX(v)≥41104v(S)>0.A_X(v)-S_X(v)\geq\frac{41}{104}v(S)>0. The same non-strict inequality holds for a valuation whose centre is ZZ. In particular, no divisor centred at a point on a smooth fibre can destabilize XX or give a non-product zero-Futaki degeneration.

Proof. For each sufficiently divisible mm, choose a basis of H0(X,−mKX)H^0(X,-mK_X) adapted simultaneously to vv and the single lexicographic flag X⊃S⊃Z⊃{p}X\supset S\supset Z\supset\{p\}. This requires compatibility with two filtrations. Let DmD_m be its normalized basis divisor, and write Dm=amS+D′mD_m=a_mS+D'_m. On SS, write D′m|S=bmZ+RmD'_m|_S=b_mZ+R_m, with RmR_m effective and not containing ZZ. Let cm=(Rm⋅Z)pc_m=(R_m\cdot Z)_p. The flag moment formulas give am→63104,bm→171208,cm→S(WS,Z;p)≤1.a_m\longrightarrow\frac{63}{104},\qquad b_m\longrightarrow\frac{171}{208},\qquad c_m\longrightarrow S(W^{S,Z};p)\leq1. Choose positive λm→1\lambda_m\to1 such that λmbm≤1\lambda_mb_m\leq1 and λmcm≤1\lambda_mc_m\leq1. For example, take λm=min{1,1/max(bm,cm)},\lambda_m=\min\{1,1/\max(b_m,c_m)\}, omitting reciprocals of zero. Since ZZ is smooth at pp, inversion of adjunction gives that (S,Z+λmRm)(S,Z+\lambda_mR_m) is log canonical near pp. Therefore (S,λmD′m|S)(S,\lambda_mD'_m|_S) is log canonical there. A second application gives that (X,S+λmD′m)(X,S+\lambda_mD'_m) is log canonical near pp. Both boundaries are effective. Evaluating at vv yields AX(v)−λmv(Dm)≥(1−λmam)v(S).A_X(v)-\lambda_mv(D_m)\geq(1-\lambda_ma_m)v(S). The basis is adapted to vv, so v(Dm)=Sm(v)→SX(v)v(D_m)=S_m(v)\to S_X(v). Taking the limit proves the assertion. For a valuation with centre ZZ, apply the same argument near a general point of ZZ. ◻

Remark 71.6. The preceding estimate uses only the elementary two-chamber decomposition on the original XX and Zariski decompositions on a quintic del Pezzo surface. No Zariski decomposition is computed on a threefold obtained by extracting the valuation vv. It reduces the classification problem to centres above the discriminant of η\eta, together with centres not addressed by this pointwise argument. It does not prove the full classification of Family 2.20.

Lemma 71.7. Every prime divisor DD on XX satisfies SX(D)<1S_X(D)<1. More precisely, SX(E)=1/2S_X(E)=1/2 and SX(D)≤63/104S_X(D)\leq63/104 for D≠ED\ne E.

Proof. For D≠ED\ne E, its pushforward is a nonzero effective divisor of class a𝒪V5(1)a\mathcal O_{V_5}(1) for an integer a≥1a\geq1. Write D∼aH−bE=aB+(a−b)ED\sim aH-bE=aB+(a-b)E. The description of the effective cone gives a−b≥0a-b\geq0. Thus D−BD-B is effective. Monotonicity of volume implies SX(D)≤SX(B)=63/104S_X(D)\leq S_X(B)=63/104. ◻

Lemma 71.8. Let Δ⊂ℙ2\Delta\subset\mathbb P^2 be the discriminant curve of η\eta, and let R⊂ER\subset E be the ramification curve of the finite double cover η|E:E→ℙ2\eta|_E:E\to\mathbb P^2. If p∉η−1(Δ)∪Rp\notin\eta^{-1}(\Delta)\cup R, then δp(X)≥104/99\delta_p(X)\geq104/99. Moreover, δR(X)≥104/63\delta_R(X)\geq104/63. Consequently, every divisorial valuation whose curve centre is not contained in η−1(Δ)\eta^{-1}(\Delta) has positive beta invariant.

Proof. At a point outside RR, the intersection of the smooth conic fibre with EE is either absent or transverse at that point. Lemma 71.3 and the iterated Abban–Zhuang estimate give δp(X)≥min{10463,208171,10499}=10499.\delta_p(X)\geq\min\left\{\frac{104}{63},\frac{208}{171}, \frac{104}{99}\right\}=\frac{104}{99}. For the ramification curve, write E=ℙ1×ℙ1E=\mathbb P^1\times\mathbb P^1, with rulings s,fs,f chosen so that ff is a fibre of π|E\pi|_E. The ramification formula gives R∼s+fR\sim s+f. Since −E|E=s−2f-E|_E=s-2f and H|E=3fH|_E=3f, one has (−KX−uE)|E=(1+u)s+(4−2u)f(0≤u≤1).(-K_X-uE)|_E=(1+u)s+(4-2u)f\quad(0\leq u\leq1). There is no fixed part in this interval. Thus S(WE;R)=326∫01∫01+u2(1+u−v)(4−2u−v)dvdu=63104.S(W^E;R)=\frac3{26}\int_0^1\int_0^{1+u} 2(1+u-v)(4-2u-v)\,dv\,du=\frac{63}{104}. The first-stage moment is SX(E)=1/2S_X(E)=1/2, so δR(X)≥min{2,104/63}=104/63\delta_R(X)\geq\min\{2,104/63\}=104/63. Finally, let TT be a curve not contained in η−1(Δ)\eta^{-1}(\Delta). If T≠RT\ne R, a general point of TT lies outside both RR and η−1(Δ)\eta^{-1}(\Delta), so the first estimate excludes every valuation with centre TT. If T=RT=R, use the second estimate. ◻