Text size

Section 85

Reduction to the non-reduced conics

In this section we write F1=BF_1=B and F2=AF_2=A, in order to match the notation of [43]. Thus F1F_1 is a sextic del Pezzo fibre, F2F_2 is a cubic del Pezzo fibre, and −KX∼D+F1+2F2-K_X\sim D+F_1+2F_2. We shall use the corrected version of [43], namely arXiv:2403.03700v3, dated 27 June 2026. Its Lemma 6.3 gives a non-strict local estimate. We explain below why this estimate still excludes a divisor with non-positive beta-invariant. The generality condition in [43, Theorem 1.1] is not imposed in this section.

Lemma 85.1. One has SX(D)=1728,SX(F1)=45112,SX(F2)=4556.S_X(D)=\frac{17}{28},\qquad S_X(F_1)=\frac{45}{112},\qquad S_X(F_2)=\frac{45}{56}. Moreover, XX is divisorially stable: βX(B)>0\beta_X(B)>0 for every prime divisor BB on XX.

Proof. The effective cone is generated by D,F1,F2D,F_1,F_2. The intersection numbers are D3=2,D2F1=−2,D2F2=−1,DF1F2=2,F12=F22=0.D^3=2,\qquad D^2F_1=-2,\qquad D^2F_2=-1,\qquad DF_1F_2=2,\qquad F_1^2=F_2^2=0. The divisor −KX−uD-K_X-uD is nef for 0≤u≤10\leq u\leq1, and is not pseudoeffective for u>1u>1. Its cube is 14−6u−6u2−2u314-6u-6u^2-2u^3. Hence SX(D)=114∫01(14−6u−6u2−2u3)du=1728.S_X(D)=\frac1{14}\int_0^1(14-6u-6u^2-2u^3)\,du =\frac{17}{28}. For the two fibres the positive parts are P1(u)={D+(1−u)F1+2F2,0≤u≤12,2(1−u)D+(1−u)F1+2F2,12≤u≤1,P2(u)={D+F1+(2−u)F2,0≤u≤1,(2−u)D+F1+(2−u)F2,1≤u≤2.\begin{aligned} P_1(u)&=\begin{cases} D+(1-u)F_1+2F_2,&0\leq u\leq\frac12,\\ 2(1-u)D+(1-u)F_1+2F_2,&\frac12\leq u\leq1, \end{cases}\\ P_2(u)&=\begin{cases} D+F_1+(2-u)F_2,&0\leq u\leq1,\\ (2-u)D+F_1+(2-u)F_2,&1\leq u\leq2. \end{cases} \end{aligned} The corresponding negative parts are 0,(2u−1)D0,(2u-1)D and 0,(u−1)D0,(u-1)D, respectively. These decompositions follow by intersecting with the two rulings of DD and with a conic fibre; equivalently, they are [43, Propositions 5.1 and 5.2]. Therefore SX(F1)=114(∫01/2(14−18u)du+∫1/21(24(1−u)2−8(1−u)3)du)=45112,SX(F2)=114(∫01(14−9u)du+∫12(6(2−u)2−(2−u)3)du)=4556.\begin{aligned} S_X(F_1)&=\frac1{14}\left( \int_0^{1/2}(14-18u)\,du+ \int_{1/2}^1\bigl(24(1-u)^2-8(1-u)^3\bigr)\,du\right) =\frac{45}{112},\\ S_X(F_2)&=\frac1{14}\left( \int_0^1(14-9u)\,du+ \int_1^2\bigl(6(2-u)^2-(2-u)^3\bigr)\,du\right) =\frac{45}{56}. \end{aligned} For completeness, the divisorial-stability reduction in [43, Section 5] uses the effective cone to reduce the verification to F2F_2, by the criterion of Fujita quoted there. The last computation gives βX(F2)=11/56>0\beta_X(F_2)=11/56>0, and thus proves the stated divisorial stability. This argument does not require any condition on the singular fibres. ◻

Lemma 85.2. For every point p∈Dp\in D, one has δp(X)≥56/45\delta_p(X)\geq56/45.

Proof. Write D≃ℙ1×ℙ1D\simeq\mathbb P^1\times\mathbb P^1, and denote the ruling classes by l1,l2l_1,l_2. Since D|D=−l1−l2D|_D=-l_1-l_2 and (−KX)|D=l1+l2(-K_X)|_D=l_1+l_2, the restriction of −KX−uD-K_X-uD is (1+u)(l1+l2)(1+u)(l_1+l_2). Choose the ruling L∈|l1|L\in|l_1| through pp. For 0≤v≤1+u0\leq v\leq1+u, the divisor (1+u−v)l1+(1+u)l2(1+u-v)l_1+(1+u)l_2 is nef, and beyond this interval it is not pseudoeffective. There are no fixed components in these restrictions. Thus the two successive expected vanishing orders are S(W•,•D;L)=314∫01∫01+u2(1+u−v)(1+u)dvdu=4556,S(W•,•,•D,L;p)=314∫01∫01+u(1+u)2dvdu=4556.\begin{aligned} S(W^D_{\bullet,\bullet};L) &=\frac3{14}\int_0^1\int_0^{1+u} 2(1+u-v)(1+u)\,dv\,du=\frac{45}{56},\\ S(W^{D,L}_{\bullet,\bullet,\bullet};p) &=\frac3{14}\int_0^1\int_0^{1+u} (1+u)^2\,dv\,du=\frac{45}{56}. \end{aligned} Abban–Zhuang adjunction for the flag X⊃D⊃L⊃pX\supset D\supset L\supset p and Lemma 85.1 now give δp(X)≥min{2817,5645,5645}=5645.\delta_p(X)\geq \min\left\{\frac{28}{17},\frac{56}{45},\frac{56}{45}\right\} =\frac{56}{45}. This also reproduces [43, Lemma 6.1]. ◻

Lemma 85.3. Let C0C_0 be a reduced fibre of the conic bundle π:X→ℙ1×ℙ1\pi:X\to\mathbb P^1\times\mathbb P^1. Let FF be a prime divisor over XX whose centre is contained in C0C_0. Then βX(F)>0\beta_X(F)>0.

Proof. Let pp be a general point of the centre of FF. If p∈Dp\in D, apply Lemma 85.2. Hence we may assume that p∉Dp\notin D. Let SiS_i be the fibre of πi\pi_i containing C0C_0; thus Si∼FiS_i\sim F_i. If S2S_2 is smooth along C0C_0, then [43, Lemma 6.2] gives δp(X)>1\delta_p(X)>1.

Suppose that S2S_2 is singular at some point of C0C_0. The restriction π2|D\pi_2|_D is a projection ℙ1×ℙ1→ℙ1\mathbb P^1\times\mathbb P^1\to\mathbb P^1, so S2S_2 is smooth along DD. Thus its singular point on C0C_0 is outside DD. By [43, Lemma 3.5], S1S_1 is a smooth sextic del Pezzo surface. The conic is C0=L3+L4C_0=L_3+L_4, where L3,L4L_3,L_4 are (−1)(-1)-curves meeting transversely. Moreover, D|S1=L1+L2D|_{S_1}=L_1+L_2 is a disjoint union of two (−1)(-1)-curves. We may number them so that L1L3=L2L4=1L_1L_3=L_2L_4=1 and L1L4=L2L3=0L_1L_4=L_2L_3=0.

Let W•,•S1W^{S_1}_{\bullet,\bullet} be the series obtained by refining H0(X,−mKX)H^0(X,-mK_X) with respect to S1S_1. The calculation in the corrected [43, Lemma 6.3] proves δp(S1;W•,•S1)≥1.\delta_p(S_1;W^{S_1}_{\bullet,\bullet})\geq1. We recall the relevant numbers, so that the non-strict inequality is explicit. For p∈L3\(D∪L4)p\in L_3\setminus(D\cup L_4), the flag beginning with L3L_3 has first expected vanishing order 11 and point expected vanishing order 39/5639/56. The same holds with L3,L4L_3,L_4 interchanged. At p=L3∩L4p=L_3\cap L_4, blow up pp on S1S_1 and denote the exceptional curve by ee. Its log discrepancy is 22 and its first expected vanishing order is 22. The next expected vanishing order is 11/2811/28 away from the two points corresponding to L3,L4L_3,L_4, and is 11 at either of these two points. These are all the terms in the surface adjunction estimate. In particular, its lower bound is 11, including at L3∩L4L_3\cap L_4.

It is essential to retain the positive remainder in the first adjunction step. Applied to a valuation centred in S1S_1, the last surface estimate gives SX(F)≤AX(F)−(1−SX(S1))ordF(S1).S_X(F)\leq A_X(F)-\bigl(1-S_X(S_1)\bigr) \operatorname{ord}_F(S_1). Indeed, choose a basis simultaneously compatible with the S1S_1-filtration and ordF\operatorname{ord}_F, and write its basis divisor as amS1+Ωma_mS_1+\Omega_m. Then am→SX(S1)a_m\to S_X(S_1) and Ωm|S1\Omega_m|_{S_1} is the corresponding refined basis divisor. The refined threshold estimate says that (S1,(1−ϵ)Ωm|S1)(S_1,(1-\epsilon)\Omega_m|_{S_1}) is log canonical near pp for all sufficiently divisible mm, with an error tending to zero. Inversion of adjunction gives AX(F)≥ordF(S1)+(1−ϵ)ordF(Ωm).A_X(F)\geq\operatorname{ord}_F(S_1) +(1-\epsilon)\operatorname{ord}_F(\Omega_m). Letting mm tend to infinity and then ϵ\epsilon tend to zero proves the inequality. This is the strictness argument underlying the equality discussion in [43, Section 8].

Here the centre of FF is contained in C0⊂S1C_0\subset S_1, so ordF(S1)>0\operatorname{ord}_F(S_1)>0. Lemma 85.1 therefore gives the strict inequality βX(F)≥67112ordF(S1)>0.\beta_X(F)\geq\frac{67}{112}\operatorname{ord}_F(S_1)>0. Thus the non-strict surface estimate is sufficient. ◻

Proposition 85.4. If a prime divisor FF over XX satisfies βX(F)≤0\beta_X(F)\leq0, then its centre is contained in the reduced support of a non-reduced conic fibre of π\pi and is disjoint from DD.

Proof. The centre is neither a divisor, by Lemma 85.1, nor a subvariety meeting DD, by Lemma 85.2. Suppose first that the centre ZZ is a curve not contracted by π2\pi_2. The general cubic fibre is smooth, and only finitely many fibres of π2\pi_2 are singular. A general point p∈Zp\in Z therefore lies on a smooth cubic fibre. Applying [43, Lemma 6.2] at pp gives βX(F)>0\beta_X(F)>0.

Suppose next that ZZ is contained in a cubic fibre S2S_2 but is not contracted by π\pi. The cubic fibre S2S_2 is normal by [43, Lemma 3.2]. The rationality argument in the proof of [43, Lemma 3.4] then shows that its singularities are Du Val. In particular, its singular locus is finite. There are consequently only finitely many conic fibres in S2S_2 meeting this singular locus. A general point p∈Zp\in Z avoids their union. The cubic surface S2S_2 is smooth along the conic through this point, so [43, Lemma 6.2] again gives a contradiction.

Thus a curve centre is contained in a conic fibre. A point centre is of course contained in its conic fibre as well. If that fibre were reduced, Lemma 85.3 would give the same contradiction. This proves the assertion. ◻

Remark 85.5. The positivity of ordF(S1)\operatorname{ord}_F(S_1) in Lemma 85.3 is used only after the centre has been placed in a conic fibre. A horizontal curve need not have positive order along S1S_1; it is excluded separately in Proposition 85.4. Thus no strict inequality is inferred from a non-strict lower bound for a local delta-invariant alone.