Text size

Section 31

Comparison with a degree-two surface flag

The following calculation explains why the weighted point extraction above is needed. A small point term on an auxiliary surface does not suffice if the preceding curve term is too large.

Lemma 31.1 (The degree-two surface flag). Let a hyperplane section V0⊂VV_0\subset V be smooth, meet ℓ\ell transversally at qq, and contain no line through qq. Let S∈|H|S\in|H| be its strict transform, let eq⊂Se_q\subset S be the exceptional curve, and put A=H|SA=H|_S. Then SS is a smooth del Pezzo surface of degree two. The strict transform CC of the tangent plane cubic at qq is a (−1)(-1)-curve of class A−2eqA-2e_q. For the flag S⊃eq∋pS\supset e_q\ni p, its constants are SX(S)=38,S(WS;eq)=155144,S(WS,eq;p)=127144+532(Ceq)p.S_X(S)=\frac38,\qquad S(W^S;e_q)=\frac{155}{144},\qquad S(W^{S,e_q};p)=\frac{127}{144}+\frac5{32}(Ce_q)_p.

Proof. Blowing up a point on no line of a smooth cubic surface gives a degree-two del Pezzo surface. The tangent plane cubic is irreducible with a double point at qq, since a reducible tangent section would contain a line through qq. Its strict transform is the stated (−1)(-1)-curve. Contracting it shows that 2A−3eq=−KS+C2A-3e_q=-K_S+C is nef (it is the pullback of the anticanonical class of the resulting smooth cubic surface).

For 0≤u≤10\le u\le1 put a=2−ua=2-u. The first refined series is asymptotically complete of class aA−eqaA-e_q. To refine along eqe_q, put b=1+vb=1+v. The positive parts are bPu,vNu,v[1,3a/2]aA−beq0[3a/2,2a](2a−b)(2A−3eq)(2b−3a)C.\begin{array}{c|cc} b&P_{u,v}&N_{u,v}\\\hline {[1,3a/2]}&aA-be_q&0\\ {[3a/2,2a]}&(2a-b)(2A-3e_q)&(2b-3a)C. \end{array} Nefness in the first interval follows by expressing its classes as nonnegative combinations of AA and 2A−3eq2A-3e_q; in the second the positive part is orthogonal to CC. The threshold is b=2ab=2a, as tested by 2A−3eq2A-3e_q. Their squares are 3a2−b23a^2-b^2 and 3(2a−b)23(2a-b)^2, while their degrees on eqe_q are bb and 3(2a−b)3(2a-b). Consequently S(WS;eq)=16∫12(72a3−3a2+13)da=155144,16∫∫(Pu,veq)2dvdu=16∫12(32a3−13)da=127144,13∫∫(Pu,veq)coeffCNu,vdvdu=13∫12a38da=532.\begin{split} S(W^S;e_q)&=\frac16\int_1^2\left(\frac72a^3-3a^2+\frac13\right)da =\frac{155}{144},\\ \frac16\int\!\int(P_{u,v}e_q)^2\,dv\,du &=\frac16\int_1^2\left(\frac32a^3-\frac13\right)da =\frac{127}{144},\\ \frac13\int\!\int(P_{u,v}e_q)\operatorname{coeff}_CN_{u,v}\,dv\,du &=\frac13\int_1^2\frac{a^3}{8}da=\frac5{32}. \end{split} This proves all three formulas. ◻

Remark 31.2 (The curve term cannot be discarded). Even when p∉Cp\notin C, the preceding flag gives only min{8/3,144/155,144/127}=144/155<1.\min\{8/3,144/155,144/127\}=144/155<1. Likewise, the refined threshold 24/2924/29 for the second-kind exceptional divisor is insufficient by itself. The weight-(1,1,2)(1,1,2) extraction in Lemma 29.7 resolves the remaining point case by computing a different refined series. No negative boundary is introduced to combine these unsuccessful flags: such a correction would require an adjunction statement that does not hold in general.