Text size

Section 75

The complete refinement at a double fibre

We now determine the refined series left open above. Let 2R2R be a double fibre, let TT be its weighted extraction, and identify its reduced exceptional surface with F=ℙ1×ℙ1F=\mathbb P^1\times\mathbb P^1. The first factor is RR, and the second parametrizes directions in the base. The different is Γ/2\Gamma/2, with Γ\Gamma a smooth (2,1)(2,1)-curve. Let q∈Γq\in\Gamma be the point above R∩ER\cap E, and let ℓ\ell be the fibre of the second projection through qq. Thus ℓ∈|𝒪F(0,1)|\ell\in|\mathcal O_F(0,1)| and Γ⋅ℓ=2\Gamma\cdot\ell=2. Write ℐq\mathcal I_q for the ideal sheaf of qq.

The distinction between the two pencil types is now simple. For vanishing sequence (1,2)(1,2), the curve Γ\Gamma meets ℓ\ell at two distinct points q,q′q,q'. For sequence (1,3)(1,3) it is tangent to ℓ\ell at qq. This follows from Lemma 73.11.

Lemma 75.1. Let Wm,kW_{m,k} be the leading sections of H0(X,−mKX)H^0(X,-mK_X) of order kk along TT. Choose coordinates [a:b][a:b], [u:v][u:v] on FF such that q=(a=v=0)q=(a=v=0) and ℓ=(v=0)\ell=(v=0), and let Q(a,b,u,v)Q(a,b,u,v) define Γ\Gamma. Use a formal symbol ww satisfying w2=Qw^2=Q to record odd weights. For k=2j+ϵk=2j+\epsilon, where ϵ∈{0,1}\epsilon\in\{0,1\}, put d=max{j−2m,0}d=\max\{j-2m,0\}. Then the actual refined series is Wm,2j+ϵ=wϵQdH0(F,𝒪F(m−ϵ−2d,j−d)⊗ℐqmax{j−m−d,0}).W_{m,2j+\epsilon}= w^\epsilon Q^d H^0\left(F,\mathcal O_F(m-\epsilon-2d,j-d) \otimes\mathcal I_q^{\max\{j-m-d,0\}}\right). A negative first degree gives the zero space. The symbol ww has divisor Γ/2\Gamma/2; equivalently one can read the formula on the index-two cover of the extraction. Its contribution divided by mm vanishes in all asymptotic moments.

Proof. The initial pairs and their graded normalizations were established in Proposition 74.3. In the (1,2)(1,2) model the centre is (A,D−F,E)(A,D-F,E) on V0V_0. Divide a quadratic in its ideal by AA at the generic point of RR. The substitutions F=Bv,E=Av,D=Au+Bv,C2=4ABu+(4B2−A2)vF=Bv,\qquad E=Av,\qquad D=Au+Bv,\qquad C^2=4ABu+(4B^2-A^2)v give the leading sections. In the (1,3)(1,3) model the centre is (B,D,F)(B,D,F); divide by BB and use D=BuD=Bu, F=BvF=Bv, E=AvE=Av, C2=4B2u−A2vC^2=4B^2u-A^2v. After choosing aa to vanish at the marked point, the six spaces in anticanonical degree one are kW1,kdimW1,k0H0(𝒪F(1,0))21wℂ12H0(𝒪F(1,1))43wH0(𝒪F(0,1))24H0(𝒪F(1,2)⊗ℐq)55wvH0(𝒪F(0,1))2.\begin{array}{c|c|c} k&W_{1,k}&\dim W_{1,k}\\ \hline 0&H^0(\mathcal O_F(1,0))&2\\ 1&w\mathbb C&1\\ 2&H^0(\mathcal O_F(1,1))&4\\ 3&wH^0(\mathcal O_F(0,1))&2\\ 4&H^0(\mathcal O_F(1,2)\otimes\mathcal I_q)&5\\ 5&wvH^0(\mathcal O_F(0,1))&2. \end{array} For example, in the first model the weight-four space is generated by the leading terms of D(D−F)D(D-F), F(D−F)F(D-F), E(D−F)E(D-F), E2E^2, and EFEF, divided by AA. These span Au2,Auv,Av2,Buv,Bv2Au^2,Auv,Av^2,Buv,Bv^2. The sixteen displayed sections exhaust H0(X,−KX)H^0(X,-K_X).

Multiplication gives subspaces of the actual associated graded ring. We first construct all the spaces in the statement using these degree-one sections. Suppose j≤2mj\le2m. A monomial aibm−ϵ−iuj−hvha^i b^{m-\epsilon-i}u^{j-h}v^h belongs to the displayed ideal-power space exactly when i+h≥j−mi+h\ge j-m. Distribute the jj symbols u,vu,v among mm factors, with at most two in each factor. There are ii factors carrying aa; these can carry two copies of uu. Every other factor can carry at most one copy of uu. The inequalities j≤2mj\le2m and j−h≤m+ij-h\le m+i are precisely the capacity conditions. For ϵ=1\epsilon=1, one factor carries ww in place of an aa or bb; its possibilities are the three odd rows of the table. The same capacity conditions apply. This constructs every required monomial.

If j>2mj>2m, put r=2d+ϵr=2d+\epsilon and a0=m−ra_0=m-r. When a0<0a_0<0 the claimed space is zero. Otherwise take rr factors from the weight-five row and a0a_0 factors from the weight-four row. Their products span wϵQdvrH0(F,𝒪F(a0,2m−r)⊗ℐqa0).w^\epsilon Q^d v^r H^0\left(F,\mathcal O_F(a_0,2m-r)\otimes\mathcal I_q^{a_0}\right). Indeed multiplication by all degree-rr binary forms in u,vu,v fills the indicated monomial space. A monomial of first degree a0=m−ra_0=m-r vanishing to order mm at qq is necessarily divisible by vrv^r. Thus this is exactly the space asserted in the lemma.

It remains to prove that these subspaces exhaust the refinement. This is a dimension argument on the actual section ring, so it does not assume generation in anticanonical degree one. For 0≤j≤2m0\le j\le2m the constructed dimension is (m−ϵ+1)(j+1)−n(n+1)2,n=max{j−m,0}.(m-\epsilon+1)(j+1) -\frac{n(n+1)}2,\qquad n=\max\{j-m,0\}. For j>2mj>2m it is (a0+1)(m+1+a0/2)(a_0+1)(m+1+a_0/2). Summing the first formula over jj and ϵ\epsilon gives (m+1)(2m+1)2−m(m+1)(m+2)/3(m+1)(2m+1)^2-m(m+1)(m+2)/3. The remaining nonzero terms have a0=0,…,m−2a_0=0,\ldots,m-2 and sum to m(m−1)(4m+1)/6m(m-1)(4m+1)/6. Their total is (2m+1)(13m2+13m+6)6=h0(X,−mKX),\frac{(2m+1)(13m^2+13m+6)}6=h^0(X,-mK_X), by Riemann–Roch and vanishing. The constructed subspaces already have the full dimension. No extra graded sections can occur. ◻

Example 75.2. For m=2m=2, the dimensions in weights 0,…,100,\ldots,10 are 3,2,6,4,9,6,11,7,12,7,33,2,6,4,9,6,11,7,12,7,3. They sum to 7070. The weight-ten space is Qv2H0(𝒪F(0,2))Qv^2H^0(\mathcal O_F(0,2)). In particular the tail is not a complete series of bidegree (2,5)(2,5). Its fixed curve and point condition are essential.

Corollary 75.3. The pseudoeffective threshold of TT is five, and its complete volume profile is vol(f*L−tT)={26−3t2/20≤t≤2,26−3t2/2+(t−2)3/42≤t≤4,(5−t)2(8−t)4≤t≤5,0t≥5.\operatorname{vol}(f^*L-tT)= \begin{cases} 26-3t^2/2&0\le t\le2,\\ 26-3t^2/2+(t-2)^3/4&2\le t\le4,\\ (5-t)^2(8-t)&4\le t\le5,\\ 0&t\ge5. \end{cases}

Proof. The normalized leading dimensions h(t)=limm−2dimWm,⌊mt⌋h(t)=\lim m^{-2}\dim W_{m,\lfloor mt\rfloor} are t/2t/2, t/2−(t−2)2/8t/2-(t-2)^2/8, and (5−t)(7−t)/2(5-t)(7-t)/2 on the three intervals. Lemma 75.1 gives these limits directly and shows that no weight exceeds 5m5m. Summing the leading sections of order at least mtmt gives vol(f*L−tT)=6∫t5h(s)ds\operatorname{vol}(f^*L-tT)=6\int_t^5h(s)\,ds. Integration proves the formula, and a further integration gives SX(T)=297/104S_X(T)=297/104, agreeing with the independent character computation. ◻

75.1 The integration regions

Put ρ=j/m\rho=j/m. After removing the fixed multiple of Γ\Gamma, the asymptotic series consists of polynomials of bidegree (A,B)(A,B) vanishing to order nn at qq, where ρABnfixed part[0,1]1ρ00[1,2]1ρρ−10[2,5/2]5−2ρ21(ρ−2)Γ.\begin{array}{c|c|c|c|c} \rho&A&B&n&\text{fixed part}\\ \hline {[0,1]}&1&\rho&0&0\\ {[1,2]}&1&\rho&\rho-1&0\\ {[2,5/2]}&5-2\rho&2&1&(\rho-2)\Gamma. \end{array} Let x,yx,y record orders at a=0,v=0a=0,v=0, respectively. The monomials fill the polygon 𝒫ρ={(x,y):0≤x≤A,0≤y≤B,x+y≥n}.\mathcal P_\rho=\{(x,y):0\le x\le A,\ 0\le y\le B,\ x+y\ge n\}. Both parities have this limiting polygon. Thus the normalization of a monomial moment is 6/136/13, since ∫05/2area(𝒫ρ)dρ=13/6\int_0^{5/2}\operatorname{area}(\mathcal P_\rho)d\rho=13/6. Equivalently, on BlqF\operatorname{Bl}_qF these are complete divisorial series. In the last interval the additional fixed curve is (2ρ−4)ℓ̃(2\rho-4)\widetilde\ell; the moving part is (5−2ρ)s+(6−2ρ)f−(5−2ρ)e(5-2\rho)s+(6-2\rho)f-(5-2\rho)e, where s,fs,f are the pullbacks of the two rulings and ee is the exceptional curve.

Lemma 75.4. For the series WW of Lemma 75.1, the normalized averages of x,y,A,Bx,y,A,B and of the fixed coefficient of Γ\Gamma are, respectively, 12,83104,99104,7352,5208.\frac12,\qquad \frac{83}{104},\qquad \frac{99}{104},\qquad \frac{73}{52},\qquad \frac5{208}. For the horizontal ruling through qq, one has S(W;ℓ)=83/104S(W;\ell)=83/104. For a different horizontal ruling ℓ0\ell_0, one has S(W;ℓ0)=63/104S(W;\ell_0)=63/104. The further point moments on these rulings are flag and pointpoint momentℓ,p≠q47/104+(5/208)(Γ⋅ℓ)pℓ,q1/2+(5/208)(Γ⋅ℓ)qℓ0,p47/104+(5/104)𝟏p∈(a=0)+(5/208)(Γ⋅ℓ0)p.\begin{array}{c|c} \text{flag and point}&\text{point moment}\\ \hline \ell, p\ne q&47/104+(5/208)(\Gamma\cdot\ell)_p\\ \ell, q&1/2+(5/208)(\Gamma\cdot\ell)_q\\ \ell_0, p&47/104+(5/104)\boldsymbol1_{p\in(a=0)} +(5/208)(\Gamma\cdot\ell_0)_p. \end{array}

Proof. All these statements are integrals over the three displayed polygons. We give the bounds explicitly so that fixed-point corrections can be checked. For each row put b0=max{0,n−A}b_0=\max\{0,n-A\}. Integrate first over b0≤y≤nb_0\le y\le n, with n−y≤x≤An-y\le x\le A, and then over n≤y≤Bn\le y\le B, with 0≤x≤A0\le x\le A. Integrating x,y,A,Bx,y,A,B and (ρ−2)+(\rho-2)_+ over these regions, with factor 6/136/13, gives the five averages. The unnormalized values are 13/1213/12, 83/4883/48, 33/1633/16, 73/2473/24, and 5/965/96.

For the flag ℓ\ell, fix yy. The allowed exponents in the other coordinate form the interval [max{0,n−y},A][\max\{0,n-y\},A]. A general point of ℓ\ell therefore contributes one-half the square of the length of this interval. Its integral, before normalization, is 47/4847/48. At qq, integrate xx instead, obtaining 13/1213/12. The fixed curve Γ\Gamma adds its average coefficient times its local intersection.

For ℓ0≠ℓ\ell_0\ne\ell, subtract a multiple zℓ0z\ell_0 from the series. Restriction to ℓ0\ell_0 has fixed order h=max{0,n−B+z}h=\max\{0,n-B+z\} at a=0a=0 and moving degree A−hA-h. Here 0≤z≤B−n+min{n,A}0\le z\le B-n+\min\{n,A\}. The point moment away from a=0a=0 is the integral of (A−h)2/2(A-h)^2/2. The correction at a=0a=0 is the integral of h(A−h)h(A-h). Their unnormalized values are 47/4847/48 and 5/485/48. The curve moment is obtained by integrating (B−max{0,n−x})2/2(B-\max\{0,n-x\})^2/2 over 0≤x≤A0\le x\le A and then over ρ\rho; it equals 21/1621/16, giving 63/10463/104 after normalization. These computations also follow directly by counting the monomials in Lemma 75.1. ◻

75.2 The marked point

Lemma 75.5. Suppose that Γ\Gamma is transverse to ℓ\ell at qq. For the refinement WW on (F,Γ/2)(F,\Gamma/2) one has δq(F,Γ/2;W)≥312275>1.\delta_q(F,\Gamma/2;W)\ge\frac{312}{275}>1.

Proof. Blow up qq and denote its exceptional curve by JJ. Its log discrepancy for (F,Γ/2)(F,\Gamma/2) is 3/23/2. The polygon moment of x+yx+y, with the fixed-curve contribution included, gives S(W;J)=12+83104+5208=275208.S(W;J)=\frac12+\frac{83}{104}+\frac5{208}=\frac{275}{208}. The different on JJ is one-half of the point determined by the tangent of Γ\Gamma. This point differs from both coordinate directions: transversality excludes the direction of ℓ\ell, and Γ→R\Gamma\to R is an isomorphism, excluding the other ruling.

To compute the second refinement, slice 𝒫ρ\mathcal P_\rho by x+y=zx+y=z. The interval of possible yy is [max{0,z−A},min{B,z}][\max\{0,z-A\},\min\{B,z\}], for n≤z≤A+Bn\le z\le A+B. The integral of one-half the square of its length is 37/4837/48, or 37/10437/104 after normalization. The two coordinate directions have moments 1/21/2 and 83/10483/104. At the tangent direction of Γ\Gamma the moment is 37/104+5/208=79/20837/104+5/208=79/208. Thus the Abban–Zhuang bound is min{312275,2,10483,10479,10437}=312275.\min\left\{\frac{312}{275},\ 2,\ \frac{104}{83},\ \frac{104}{79},\ \frac{104}{37}\right\}=\frac{312}{275}. All restricted spaces used here are complete binary spaces with the displayed fixed endpoint orders, by the exact monomial description. ◻

75.3 Ramification points away from the marked point

Lemma 75.6. Assume that qq is not a ramification point of Γ→ℙ1\Gamma\to\mathbb P^1 in the direction factor. If p∈Γp\in\Gamma is a ramification point of this map, then δp(F,Γ/2;W)≥10489>1.\delta_p(F,\Gamma/2;W)\ge\frac{104}{89}>1.

Proof. The points p,qp,q lie on different fibres of both rulings. Choose coordinates putting them at opposite corners. If x,yx,y are the exponents at qq, the exponents at pp are x′=A−xx'=A-x, y′=B−yy'=B-y. Here y′=0y'=0 denotes the tangent ruling through pp. The curve Γ\Gamma has local equation y′=c(x′)2+higher termsy'=c(x')^2+\text{higher terms}, with c≠0c\ne0. Use the weighted blowup of weights (1,2)(1,2) in these two local coordinates, and denote its exceptional curve by JJ. Then AF,Γ/2(J)=3−2/2=2A_{F,\Gamma/2}(J)=3-2/2=2. The moment is S(W;J)=(99104−12)+2(7352−83104)+25208=8952.S(W;J)=\left(\frac{99}{104}-\frac12\right) +2\left(\frac{73}{52}-\frac{83}{104}\right) +2\frac5{208}=\frac{89}{52}.

The coarse curve JJ is ℙ1\mathbb P^1. Its different consists of two distinct points of coefficient 1/21/2: the quotient point and the point cut out by the initial equation of Γ\Gamma. The other coordinate direction has coefficient zero. For the second refinement slice by z=x′+2y′z=x'+2y'. With K=A+B−nK=A+B-n, the interval of allowed y′y' is [max{0,z−A2,z−K},min{B,z2}],\left[\max\left\{0,\frac{z-A}{2},z-K\right\}, \min\left\{B,\frac z2\right\}\right], when its length is nonnegative. The integrals of one-half the square of its length over the three ρ\rho-intervals are 17/19217/192, 19/6419/64, and 5/965/96. Their sum is 7/167/16. The general point moment on JJ is consequently 21/10421/104. The point belonging to Γ\Gamma has moment 21/104+5/208=47/20821/104+5/208=47/208. The quotient point has moment (99/104−1/2)/2=47/208(99/104-1/2)/2=47/208, and the other coordinate point has moment 73/52−83/104=63/10473/52-83/104=63/104. These give δp(F,Γ/2;W)≥min{10489,10447,10463,10421}=10489.\delta_p(F,\Gamma/2;W)\ge \min\left\{\frac{104}{89},\frac{104}{47}, \frac{104}{63},\frac{104}{21}\right\} =\frac{104}{89}. The division by two at the quotient point is necessary: it is the order on the coarse weighted exceptional curve. ◻

Theorem 75.7. Suppose that a double fibre of a smooth member of Family 2.20 corresponds to a common root with vanishing sequence (1,2)(1,2). Then δp(X)≥10499>1\delta_p(X)\ge\frac{104}{99}>1 at every point of its support. In particular this holds on every double fibre of every smooth member with GIT-stable pencil.

Proof. We first prove δ(F,Γ/2;W)≥104/99\delta(F,\Gamma/2;W)\ge104/99. The marked point is treated by Lemma 75.5. Every ramification point of the direction projection is treated by Lemma 75.6. For another point, use its horizontal ruling. On ℓ\ell, the only point of Γ\Gamma other than qq is q′q', and its refined point moment is 99/20899/208, with discrepancy 1/21/2. All other point moments there are 47/10447/104, with discrepancy one. The first moment is 83/10483/104.

On a different ruling, a transverse intersection with Γ\Gamma has point moment 99/20899/208, again with discrepancy 1/21/2. It cannot also have a=0a=0, since Γ\Gamma meets that vertical ruling only at qq. Points outside Γ\Gamma have moment at most 1/21/2 and discrepancy one. The first moment is 63/10463/104. Thus every remaining flag gives a bound at least 104/99104/99.

Apply Abban–Zhuang refinement on the threefold extraction. Its first ratio is AX(T)SX(T)=3297/104=10499.\frac{A_X(T)}{S_X(T)}=\frac{3}{297/104}=\frac{104}{99}. The pair on its exceptional surface is (F,Γ/2)(F,\Gamma/2), whose refined threshold has just been bounded below by the same number. Inversion of adjunction gives the bound in a neighbourhood of the whole exceptional surface, and hence along RR on XX. This includes all valuations whose centres contain a point of RR. One can equivalently apply the finite-basis proof: choose coefficients tending to 104/99104/99 from below, use the exact refined spaces above, and apply effective adjunction on the extraction before taking limits. ◻

Remark 75.8. The proof uses an ordinary surface blowup at the marked point and a weighted surface blowup at each other ramification point. It computes no Zariski decomposition on an extracted threefold. The complete global refinement is proved by multiplication and a full dimension count; its fixed curve and point condition are retained throughout. The transversality assumption is essential. In the (1,3)(1,3) case the tangent direction of Γ\Gamma is a coordinate endpoint on the blowup of qq, so the two contributions cannot be placed at distinct points. The strictly K-semistable examples are therefore not covered by Theorem 75.7.