Text size

Section 99

A weighted extraction of the joining line

Lemma 99.1. The joining line has normal bundle NΓ/X=𝒪(1)⊕𝒪(−2)N_{\Gamma/X}=\mathcal O(1)\oplus\mathcal O(-2). There is a weighted extraction with normal weights (1,2)(1,2), assigning weight one to the positive normal direction. Its exceptional divisor GG has AX(G)=3A_X(G)=3, and its different has coefficient 1/21/2 along a=0a=0. On the open part of Γ\Gamma, its actual refined algebra has the thirteen degree-one generators 1,b,a,z,bz,b2z,h,abz,bz2,b2z2,b3z2,bhz,ab2z2,h=a2+b3z.1,b,a,z,bz,b^2z,h,abz,bz^2,b^2z^2,b^3z^2,bhz,ab^2z^2, \qquad h=a^2+b^3z. Their primary degrees are determined by deg(a)=1\deg(a)=1, deg(z)=2\deg(z)=2, deg(b)=0\deg(b)=0.

Proof. Choose coordinates so that the mixed fibre of the graph is defined by f=x3y2f=x^3y^2. Write g=∑i=05cixiy5−ig=\sum_{i=0}^5 c_i x^i y^{5-i}, where c0c5≠0c_0c_5\ne0. On Y0≠0Y_0\ne0, put a=Y1/Y0a=Y_1/Y_0, b=Y2/Y0b=Y_2/Y_0, q=b−a2q=b-a^2, and let tt be the target coordinate vanishing on the mixed fibre. Set P=c0+c2b+c4b2,Q=c1+c3b+c5b2.P=c_0+c_2b+c_4b^2,\qquad Q=c_1+c_3b+c_5b^2. The two lifted graph sections are A=a3+t(P+aQ)q,B=a2b+t(aP+bQ)q.A=\frac{a^3+t(P+aQ)}{q},\qquad B=\frac{a^2b+t(aP+bQ)}{q}. They satisfy bA−aB=tPbA-aB=tP and B=aA+a2+tQB=aA+a^2+tQ. The anticanonical degree-one sections, in this trivialization, are ti{1,a,b}t^i\{1,a,b\} for 0≤i≤20\le i\le2, together with A,B,tA,tBA,B,tA,tB.

The morphism from the normal bundle of Γ\Gamma to the normal bundle 𝒪(1)⊕𝒪\mathcal O(1)\oplus\mathcal O of its image in ℙ1×ℙ2\mathbb P^1\times\mathbb P^2 is an elementary transform at the two endpoints. At each endpoint its quotient is in the trivial summand: the graph is tangent to the fibre, whereas the joining line is transverse to the conic. Thus its kernel is 𝒪(1)⊕𝒪(−2)\mathcal O(1)\oplus\mathcal O(-2). Equivalently, in local coordinates near an endpoint the differential in the normal directions is (a,w)↦(a,bw)(a,w)\mapsto(a,bw). At the index-two endpoint the equation for tt contains an additional quadratic term, which has zero differential.

There is a coordinate-independent weighted extraction. If II is the ideal of Γ\Gamma, take the inverse image J⊂IJ\subset I of the positive subbundle 𝒪(2)⊂I/I2\mathcal O(2)\subset I/I^2. Locally J=(w,a2)J=(w,a^2). The normalized blow-up of JJ has weights (1,2)(1,2), discrepancy three, and different one half along the section corresponding to a=0a=0. The coarse exceptional ruled surface need not be identified in this argument.

At the generic point of Γ\Gamma, assign weights 1,2,01,2,0 to a,t,ba,t,b. The initial forms of A,BA,B are tP/btP/b and a2+tQa^2+tQ. Subtracting linear combinations of t,tbt,tb, and rescaling, gives t/bt/b and a2+tb2a^2+tb^2. The same operation applies after multiplication by tt. Putting z=t/bz=t/b yields the thirteen generators in the statement. They generate a subalgebra of the primary associated algebra. Lemma 99.2 below gives an exact dimension count after one further refinement. It proves equality of both associated algebras in every degree. ◻

99.1 The fibre refinement

We use the coordinates and initial sections of the preceding extraction calculation. Thus the centre is locally a=t=0a=t=0, the coordinate bb is nonzero, the weights are (1,2)(1,2), and z=t/bz=t/b. Put h=a2+b3zh=a^2+b^3z. The thirteen initial anticanonical sections are 1,b,a,z,bz,b2z,h,abz,bz2,b2z2,b3z2,bhz,ab2z2.1,\ b,\ a,\ z,\ bz,\ b^2z,\ h,\ abz,\ bz^2,\ b^2z^2, \ b^3z^2,\ bhz,\ ab^2z^2. Here the primary degrees of a,b,za,b,z are 1,0,21,0,2, respectively. All statements below concern the open set b≠0,∞b\ne0,\infty. In particular, we make no assertion that the coarse exceptional surface is globally a particular Hirzebruch surface.

Lemma 99.2. Fix c≠0,∞c\ne0,\infty, and refine the primary filtration by order along b=cb=c. The resulting degree-mm vector space has a monomial basis indexed by the triples (a,k,z)(a,k,z) in the semigroup generated in degree one by (0,0,0),(0,1,0),(1,0,0),(0,0,1),(0,1,1),(0,2,1),(2,0,0),(1,0,1),(0,0,2),(0,1,2),(0,2,2),(2,0,1),(1,0,2).\begin{gathered} (0,0,0),(0,1,0),(1,0,0),(0,0,1),(0,1,1),(0,2,1),(2,0,0),\\ (1,0,1),(0,0,2),(0,1,2),(0,2,2),(2,0,1),(1,0,2). \end{gathered} Its degree-mm elements are precisely the integral points satisfying a,k,z≥0,z≤2m,a+k≤2m,a+2k−2z≤2m,2a+k+2z≤6m,\begin{gathered} a,k,z\ge0,\qquad z\le2m,\qquad a+k\le2m,\\ a+2k-2z\le2m,\qquad2a+k+2z\le6m, \end{gathered} except for the points with a+k=2ma+k=2m and both aa and kk odd. The primary degree of such a basis vector is j=a+2zj=a+2z.

Proof. The substitution b↦cbb\mapsto cb, a↦aa\mapsto a, z↦c−3zz\mapsto c^{-3}z preserves hh and scales each of the thirteen sections. We may therefore suppose that c=1c=1. Put d=b−1d=b-1. In the primary degree-two piece use z,dz,d2z,h−3b2z+3bz−z=a2+d3z.z,\quad dz,\quad d^2z,\quad h-3b^2z+3bz-z=a^2+d^3z. In the primary degree-four piece use bz2,bdz2,bd2z2,bhz−3b3z2+3b2z2−bz2=ba2z+bd3z2.bz^2,\quad bdz^2,\quad bd^2z^2,\quad bhz-3b^3z^2+3b^2z^2-bz^2=ba^2z+bd^3z^2. Together with 1,d,a,abz,ab2z21,d,a,abz,ab^2z^2, their leading forms are the thirteen monomials asserted in the statement. Their products give linearly independent leading forms whenever their exponent triples are distinct.

We count these products in every degree. The degree-one generators with first two coordinates (2,0)(2,0) allow third coordinate 00 or 11; those with first two coordinates (0,2)(0,2) allow 11 or 22. The generators with first two coordinates (1,0)(1,0), (0,1)(0,1), or (0,0)(0,0) allow every third coordinate between 00 and 22. Thus a triple (a,k,z)(a,k,z) is a product of mm generators if and only if there exist integers r,s≥0r,s\ge0 such that r≤⌊a2⌋,s≤⌊k2⌋,r+s≥a+k−m,s≤z≤2m−r.r\le\left\lfloor\frac a2\right\rfloor,\quad s\le\left\lfloor\frac k2\right\rfloor,\quad r+s\ge a+k-m,\quad s\le z\le2m-r. Indeed, take rr generators of type (2,0)(2,0), ss of type (0,2)(0,2), a−2ra-2r of type (1,0)(1,0), k−2sk-2s of type (0,1)(0,1), and fill the remaining places with type (0,0)(0,0). Their possible third coordinates form the whole interval [s,2m−r][s,2m-r]. Equivalently, min{⌊a2⌋,2m−z}+min{⌊k2⌋,z}≥a+k−m.\min\left\{\left\lfloor\frac a2\right\rfloor,2m-z\right\} +\min\left\{\left\lfloor\frac k2\right\rfloor,z\right\} \ge a+k-m. Expanding the two minima gives precisely the inequalities and the parity exception in the statement.

For completeness, the count is also explicit. Write a=2r+ϵa=2r+\epsilon, k=2s+ηk=2s+\eta, where ϵ,η∈{0,1}\epsilon,\eta\in\{0,1\}. The condition is r+s+ϵ+η≤mr+s+\epsilon+\eta\le m. Put N=m−r−s−ϵ−ηN=m-r-s-\epsilon-\eta. The number of possible third coordinates is 2m+1−max{0,s−N}−max{0,r−N}.2m+1-\max\{0,s-N\}-\max\{0,r-N\}. If Q(n)=∑r+s≤nmax{0,2r+s−n},Q(n)=\sum_{r+s\le n}\max\{0,2r+s-n\}, then summing first over ss gives Q(n)={n(n+2)(2n+5)/24if n is even,(n+1)(n+3)(2n+1)/24if n is odd.Q(n)=\begin{cases} n(n+2)(2n+5)/24&\text{if $n$ is even},\\ (n+1)(n+3)(2n+1)/24&\text{if $n$ is odd}. \end{cases} Set Q(−1)=Q(−2)=0Q(-1)=Q(-2)=0. Consequently the number of distinct products is (2m+1)((m+22)+2(m+12)+(m2))−2(Q(m)+2Q(m−1)+Q(m−2))=10m3+15m2+11m+33.\begin{aligned} &(2m+1)\left(\binom{m+2}{2}+2\binom{m+1}{2}+\binom m2\right)\\ &\quad -2\bigl(Q(m)+2Q(m-1)+Q(m-2)\bigr) =\frac{10m^3+15m^2+11m+3}{3}. \end{aligned} This is h0(X,−mKX)h^0(X,-mK_X). Thus the products already fill the entire filtered vector space, and there are no further leading forms in any degree. Notice that the semigroup is not normal: (1,1,1)(1,1,1) is the first missing lattice point. Its missing boundary points must not be silently inserted in the finite-degree basis. ◻

Lemma 99.3. Let GG denote the weighted exceptional divisor and let FcF_c denote its fibre over b=c≠0,∞b=c\ne0,\infty. The primary and secondary expected vanishing orders are SX(G)=218,S(WG;Fc)=58.S_X(G)=\frac{21}{8},\qquad S(W^G;F_c)=\frac58. On the coarse fibre Fc≃ℙ1F_c\simeq\mathbb P^1, use weighted coordinates [a:z]∈ℙ(1,2)[a:z]\in\mathbb P(1,2) and ordinary coordinates [a2:z][a^2:z]. For the series refined by FcF_c, the point moments satisfy S(WG,Fc;{z=0})=1,S(WG,Fc;{a=0})=516,S(W^{G,F_c};\{z=0\})=1,\qquad S(W^{G,F_c};\{a=0\})=\frac5{16}, and at every other point they are at most 5/165/16.

Proof. Let PP be the polytope in the preceding lemma with m=1m=1. The missing lattice points belong to a two-dimensional face. There are O(m2)O(m^2) of them, so they do not change the limits of normalized first moments. Over the triangle T={a,k≥0∣a+k≤2}T=\{a,k\ge0\mid a+k\le2\} the interval in the third coordinate is max{0,a+2k−22}≤z≤2−max{0,2a+k−22}.\max\left\{0,\frac{a+2k-2}{2}\right\}\le z\le 2-\max\left\{0,\frac{2a+k-2}{2}\right\}. Each of the two cut-off triangular caps has volume 1/31/3; hence vol(P)=4−2/3=10/3\operatorname{vol}(P)=4-2/3=10/3. The integral of aa over PP is ∫Pa=83−16−512=2512.\int_P a=\frac83-\frac16-\frac5{12}=\frac{25}{12}. Here the first term is its integral over T×[0,2]T\times[0,2]. For the lower cap the base triangle has vertices (0,1),(0,2),(2,0)(0,1),(0,2),(2,0), and the integral of a(a+2k−2)/2a(a+2k-2)/2 is 1/61/6. For the upper cap the base triangle has vertices (1,0),(2,0),(0,2)(1,0),(2,0),(0,2), and the integral of a(2a+k−2)/2a(2a+k-2)/2 is 5/125/12. The involution (a,k,z)↦(k,a,2−z)(a,k,z)\mapsto(k,a,2-z) preserves PP. Therefore 1vol(P)∫P(a,k,z)=(58,58,1).\frac1{\operatorname{vol}(P)}\int_P(a,k,z) =\left(\frac58,\frac58,1\right). The primary order is a+2za+2z, while the secondary order is kk. This proves the first two assertions.

Fix a primary degree jj and a secondary degree kk. In the basis of Lemma 99.2 the remaining exponents satisfy a=j−2za=j-2z. Their allowed values of zz form an interval of consecutive integers. The parity exception can remove only its lower endpoint. Thus this piece, on the weighted fibre, is exactly aa0zz0⟨(a2)d,(a2)d−1z,…,zd⟩a^{a_0}z^{z_0} \langle (a^2)^d,(a^2)^{d-1}z,\ldots,z^d\rangle for some nonnegative integers a0,z0,da_0,z_0,d. At z=0z=0 its vanishing sequence is z0,z0+1,…,z0+dz_0,z_0+1,\ldots,z_0+d. At the coarse point a=0a=0 it is a0/2,a0/2+1,…,a0/2+da_0/2,a_0/2+1,\ldots,a_0/2+d. The rational offset is the order in the induced rational line bundle; it is also obtained by taking an even multiple before passing to the coarse fibre. At any other point the vanishing sequence is 0,1,…,d0,1,\ldots,d. Hence the last sequence has average no larger than the sequence at a=0a=0. Summing over j,kj,k gives the asserted moments. These are complete restricted series, so the estimates hold for bases adapted to any chosen point, rather than only for the displayed monomial basis. ◻

Example 99.4. For m=1m=1, the piece of primary degree 22 and fibre degree 00 is ⟨a2,z⟩\langle a^2,z\rangle, while the pieces of fibre degrees 11 and 22 are both generated by zz. The analogous primary degree-44 pieces are ⟨a2z,z2⟩\langle a^2z,z^2\rangle, ⟨z2⟩\langle z^2\rangle, and ⟨z2⟩\langle z^2\rangle. Over all thirteen basis vectors, the sums of the primary, fibre, and z=0z=0 orders are respectively 33,7,1333,7,13. The sum of the coarse a=0a=0 orders is 7/27/2. This illustrates both the factor 1/21/2 at the quotient point and the fixed factors in the refined series. In every degree the involution (a,k,z)↦(k,a,2m−z)(a,k,z)\mapsto(k,a,2m-z) preserves the semigroup, so the average order at z=0z=0 is exactly mm, even before taking the asymptotic limit.

Proposition 99.5. For every point pp of the joining line with b(p)≠0,∞b(p)\ne0,\infty one has δp(X)≥1\delta_p(X)\ge1.

Proof. The extraction with transverse weights (1,2)(1,2) has AX(G)=3A_X(G)=3. Over the indicated open set its different is one half of the section a=0a=0. The fibre FcF_c is not contained in this different. The different induced on its coarse curve is therefore 12{a=0}\frac12\{a=0\}. Its point discrepancies are 1/21/2 at a=0a=0 and 11 everywhere else. By Lemma 99.3, the ratios on the fibre are 8/58/5 at a=0a=0, exactly 11 at z=0z=0, and at least 16/516/5 elsewhere. The Abban–Zhuang refinement along FcF_c gives a bound min{8/5,1}=1\min\{8/5,1\}=1 on the exceptional surface above pp. Applying the same refinement to GG gives δp(X)≥min{321/8,15/8,1}=1.\delta_p(X)\ge\min\left\{\frac3{21/8},\frac{1}{5/8},1\right\}=1. The equality in the last point estimate only proves semistability. ◻