Text size

Section 44

Fibre refinements and reduction of centres

We retain the notation H,E,FH,E,F of the geometric construction and put L=−KX=H+FL=-K_X=H+F. Thus H3=5H^3=5, F2=0F^2=0, L3=20L^3=20 and L2F=5L^2F=5.

Lemma 44.1. One has SX(F)=11/16S_X(F)=11/16, SX(E)=3/8S_X(E)=3/8, and SX(H)=5/16S_X(H)=5/16. If WFW^F is the refined series obtained by restricting along a fibre, then for every prime divisor TT over FF, S(WF;T)=1516SF(T)+116ordT(C).S(W^F;T)=\frac{15}{16}S_F(T)+\frac1{16}\operatorname{ord}_T(C).

Proof. The nef cone is generated by H,FH,F, and the effective cone by E,FE,F. For the latter assertion, a prime divisor other than EE is the proper transform of a divisor of class aHaH on V5V_5, of multiplicity bb along CC, where a≥1a\ge1 and b≥0b\ge0. Restricting it to a general fibre gives a≥ba\ge b, and its class is aF+(a−b)EaF+(a-b)E. For 0≤u≤10\le u\le1, the divisor L−uF=H+(1−u)FL-uF=H+(1-u)F is nef. For 1≤u≤21\le u\le2, its divisorial decomposition is P(u)=(2−u)H,N(u)=(u−1)E.P(u)=(2-u)H,\qquad N(u)=(u-1)E. Indeed, intersection with a general fibre of E→CE\to C forces the displayed multiplicity along EE, and the remainder is nef. The effective cone gives pseudoeffective threshold two. Therefore SX(F)=120(∫01(20−15u)du+∫125(2−u)3du)=1116.S_X(F)=\frac1{20}\left(\int_0^1(20-15u)\,du+ \int_1^2 5(2-u)^3\,du\right)=\frac{11}{16}. The divisors L−uE=(1−u)H+(1+u)FL-uE=(1-u)H+(1+u)F and L−uH=(1−u)H+FL-uH=(1-u)H+F are nef for 0≤u≤10\le u\le1; both have pseudoeffective threshold one. The intersection numbers give SX(E)=14∫01(4−6u+2u3)du=38,SX(H)=14∫01(4−9u+6u2−u3)du=516.S_X(E)=\frac14\int_0^1(4-6u+2u^3)\,du=\frac38,\qquad S_X(H)=\frac14\int_0^1(4-9u+6u^2-u^3)\,du=\frac5{16}. Finally, P(u)|F=−KFP(u)|_F=-K_F for u≤1u\le1 and P(u)|F=(2−u)(−KF)P(u)|_F=(2-u)(-K_F) for u≥1u\ge1, while N(u)|F=(u−1)CN(u)|_F=(u-1)C on the latter interval. The moving and fixed coefficients in the restriction formula are respectively 1520(∫011du+∫12(2−u)3du)=1516,1520∫12(2−u)2(u−1)du=116.\frac{15}{20}\left(\int_0^1 1\,du+\int_1^2(2-u)^3\,du\right) =\frac{15}{16},\qquad \frac{15}{20}\int_1^2(2-u)^2(u-1)\,du=\frac1{16}. This proves the exact refinement formula; it is also the degree-five instance of the computation in [8, Lemma 2.1]. These decompositions occur on XX itself. No decomposition on a threefold obtained by extracting a point or a line is used in this lemma. ◻

Corollary 44.2. Put dp=δp(F)d_p=\delta_p(F). Then δp(X)≥{min{16/11,16dp/15},p∉E,min{16/11,16dp/(dp+15)},p∈E.\delta_p(X)\ge \begin{cases} \min\{16/11,16d_p/15\},&p\notin E,\\ \min\{16/11,16d_p/(d_p+15)\},&p\in E. \end{cases} In particular, if FF is smooth, these bounds give 16/1316/13 and 8/78/7, respectively.

Proof. Use Lemma 44.1 and Abban–Zhuang along FF. At a point off EE, every valuation entering the local estimate has zero order on CC. At a point of CC, the inequality ordT(C)≤AF(T)\operatorname{ord}_T(C)\le A_F(T) follows from log canonicity of (F,C)(F,C). For a smooth quintic del Pezzo surface, δ(F)=15/13\delta(F)=15/13 by [3, Lemma 2.11]. ◻

Lemma 44.3 (Keeping the coefficient of a fibre). Let FF be a fibre, and let p∈Fp\in F. Suppose that the refined series WFW^F has local stability threshold at least one at pp. If QQ is a prime divisor over XX whose centre contains pp and is contained in FF, then βX(Q)≥516ordQ(F)>0.\beta_X(Q)\ge \frac5{16}\operatorname{ord}_Q(F)>0.

Proof. Choose a basis of H0(X,−mKX)H^0(X,-mK_X) simultaneously adapted to QQ and FF. Write its basis divisor as Dm=amF+ΓmD_m=a_mF+\Gamma_m, where am=Sm(F)a_m=S_m(F) and F⊄Supp(Γm)F\not\subset\operatorname{Supp}(\Gamma_m). Its restriction Γm|F\Gamma_m|_F is a basis divisor of the refined series. Convergence of the finite basis thresholds gives numbers cm→1c_m\to1, with 0<cm≤10<c_m\le1, such that (F,cmΓm|F)(F,c_m\Gamma_m|_F) is log canonical near pp. Inversion of adjunction gives AX(Q)≥ordQ(F)+cmordQ(Γm)=cmSm(Q)+(1−cmam)ordQ(F).A_X(Q)\ge\operatorname{ord}_Q(F)+c_m\operatorname{ord}_Q(\Gamma_m) =c_m S_m(Q)+(1-c_ma_m)\operatorname{ord}_Q(F). Taking the limit and using SX(F)=11/16S_X(F)=11/16 proves the assertion. The last inequality is strict because the centre of QQ is contained in FF. Only prime divisors over XX are being used; this is sufficient for the valuative criterion in this paper. ◻

Let σ:F̃→F\sigma:\widetilde F\to F be the minimal resolution of a singular quintic del Pezzo surface. We use the labels of [12, Figure 33, p. 25]: the exceptional curves over an ArA_r point form the chain E1,…,ErE_1,\ldots,E_r. For type A2+A1A_2+A_1, the chain is E2,E1E_2,E_1 and the isolated exceptional curve is E3E_3. The following table identifies the lines which will remain in the reduction. Incidence means intersection of the proper transform of the line with the indicated exceptional curves. All other incidences with exceptional curves are zero.

Sing(F)\operatorname{Sing}(F) line incidence SF(L)S_F(L) source in [12]
2A12A_1 L12L_{12} E1,E2E_1,E_2 19/1519/15 Lemma 5.8(a), pp. 18–19
A2A_2 L2L_2 E2E_2 17/1517/15 Lemma 5.7, p. 18
A2+A1A_2+A_1 L13L_{13} E1,E3E_1,E_3 23/1523/15 Lemma 5.15, pp. 22–23
A2+A1A_2+A_1 L2L_2 E2E_2 17/1517/15 Lemma 5.7, p. 18
A3A_3 L2L_2 E2E_2 19/1519/15 Lemma 5.8(b), pp. 18–19
A4A_4 L3L_3 E3E_3 5/35/3 Lemma 5.16, p. 23

For type A2A_2, the endpoint E1E_1 meets two further lines L1,L′1L_1,L'_1, while E2E_2 meets L2L_2 only. For type A3A_3, the other line L1L_1 meets E1E_1. For type A4A_4, L3L_3 is the unique line. These conventions fix the choices of endpoints of the chains without ambiguity.

Lemma 44.4. Let p∈C⊂Fp\in C\subset F. Suppose that either δp(F)≥1\delta_p(F)\ge1, or that pp lies on L1L_1 or L′1L'_1 on a fibre of type A2A_2, or on L1L_1 on a fibre of type A3A_3. Then every prime divisor QQ over XX with centre pp has βX(Q)>0\beta_X(Q)>0.

Proof. The exact refinement by a fibre is S(WF;v)=1516SF(v)+116v(C)S(W^F;v)=\frac{15}{16}S_F(v)+\frac1{16}v(C). Since FF and CC are smooth along CC, the pair (F,C)(F,C) is log canonical. Thus δp(F)≥1\delta_p(F)\ge1 implies S(WF;v)≤AF(v)S(W^F;v)\le A_F(v) for every prime divisor vv over FF with centre containing pp. Lemma 44.3 applies.

In the two remaining cases, put L=L1L=L_1 or L′1L'_1 as appropriate. The curve CC meets LL once and transversally: C∼−KFC\sim-K_F, (−KF)⋅L=1(-K_F)\cdot L=1, and FF is smooth along CC. For type A2A_2, the positive part of −KF̃−vL̃-K_{\widetilde F}-v\widetilde L has square 5−2v−v2/35-2v-v^2/3 for 0≤v≤3/20\le v\le3/2 and (2−v)(4−v)(2-v)(4-v) for 3/2≤v≤23/2\le v\le2. Its intersection with L̃\widetilde L is respectively 1+v/31+v/3 and 3−v3-v. Consequently SF(L)=3130,S(WFL;p)=15(∫03/2(1+v/3)2dv+∫3/22(3−v)2dv)=1930.S_F(L)=\frac{31}{30},\qquad S(W_F^L;p)=\frac15\left(\int_0^{3/2}(1+v/3)^2\,dv+ \int_{3/2}^2(3-v)^2\,dv\right)=\frac{19}{30}. For type A3A_3, the corresponding square and intersection are (2−v)(10+v)/4(2-v)(10+v)/4 and 1+v/41+v/4, for 0≤v≤20\le v\le2, giving SF(L)=1615,S(WFL;p)=15∫02(1+v/4)2dv=1930.S_F(L)=\frac{16}{15},\qquad S(W_F^L;p)=\frac15\int_0^2(1+v/4)^2\,dv=\frac{19}{30}. These are the decompositions in [12, Lemmas 5.4–5.5, pp. 16–17]. The denominator in the displayed square in Lemma 5.5 is corrected here from 22 to 44; the stated SS-value in that lemma is unchanged. The negative parts meet L̃\widetilde L only above the singular point of FF, so they contribute nothing at pp.

Refining WFW^F by LL, its first invariant is 31/3231/32 in type A2A_2 and 11 in type A3A_3. The secondary invariant at pp is in both cases 1516⋅1930+116=2132.\frac{15}{16}\cdot\frac{19}{30}+\frac1{16}=\frac{21}{32}. Here the last term is the restriction of the fixed boundary to the reduced point C∩LC\cap L. The different is zero near pp. Abban–Zhuang on FF therefore gives refined threshold at least 32/3132/31 and 11, respectively. Apply Lemma 44.3 once more. ◻

Proposition 44.5 (Reduction of possible centres). Let QQ be a prime divisor over XX with βX(Q)≤0\beta_X(Q)\le0. Then its centre is either one of the lines in the table, a point of one of those lines, or the unique node of a fibre of type A1A_1.

Proof. If the centre is a surface DD, then Q=ordDQ=\operatorname{ord}_D. The cases D=ED=E and D=FD=F are excluded by SX(E)=3/8S_X(E)=3/8 and SX(F)=11/16S_X(F)=11/16. Every other prime surface is the proper transform of a divisor in |aH||aH| on V5V_5, of multiplicity bb along CC, with a≥1a\ge1 and 0≤b≤a0\le b\le a. If a=ba=b, the surface is a fibre, since |aF||aF| is pulled back from the base. Otherwise D∼aF+(a−b)ED\sim aF+(a-b)E, so D−ED-E is linearly equivalent to an effective divisor and SX(D)≤SX(E)<1S_X(D)\le S_X(E)<1.

On a smooth fibre δ(F)=15/13\delta(F)=15/13. The exact fibre refinement gives δp(X)≥16/13\delta_p(X)\ge16/13 off EE and δp(X)≥8/7\delta_p(X)\ge8/7 on EE. Thus any centre meeting a smooth fibre is excluded. In particular, every curve centre is contained in a singular fibre.

At a smooth point of a singular fibre off EE, the local threshold table [12, Tables 10–15, p. 26] gives δp(F)>15/16\delta_p(F)>15/16, except on the listed lines and on L1L_1 of a fibre of type A3A_3. In the latter case δp(F)=15/16\delta_p(F)=15/16, so the refined threshold is at least one and Lemma 44.3 excludes every centre contained in FF. At points on EE, the same tables and Lemma 44.4 exclude all points outside the listed lines. For a curve centre we apply these arguments at a general closed point of the curve.

Finally, the incidence configurations in [12, Figure 33, p. 25] show that every singular point of a fibre of type 2A12A_1, A2A_2, A2+A1A_2+A_1, A3A_3 or A4A_4 lies on a listed line. This leaves precisely the centres in the statement. ◻