Text size

Section 66

The weighted point extraction and its actual section ring

We now treat every point of ΣX\Sigma_X. The line calculation above is insufficient there because its refined surface order is 49/24>249/24>2. We use a weighted point blow-up and calculate its refinement directly. The key is a realization of XX in a four-dimensional symplectic flag variety. All weights below are vanishing weights: the initial form means the part of smallest weight, until the second degeneration is specified.

Lemma 66.1 (The flag realization). Let XX be the blow-up of a smooth elliptic quintic in ℙ3\mathbb P^3. There is a nondegenerate alternating form on a four-dimensional vector space VV such that XX is a divisor of class H+BH+B on Y={(λ,Λ):λ⊂Λ⊂V,dimλ=1,dimΛ=2,Λ isotropic}.Y=\{(\lambda,\Lambda):\lambda\subset\Lambda\subset V, \dim\lambda=1,\ \dim\Lambda=2,\ \Lambda\text{ isotropic}\}. Here HH and BB are the pullbacks of the hyperplane classes of ℙ(V)\mathbb P(V) and LG(2,V)≅Q3\operatorname{LG}(2,V)\cong Q^3, respectively. The two projections restrict to the two contractions of XX.

Proof. Serre’s construction gives a rank-two bundle 𝒩\mathcal N fitting in 0→𝒪→𝒩(2)→ℐC(4)→0.0\longrightarrow\mathcal O\longrightarrow\mathcal N(2) \longrightarrow\mathcal I_C(4)\longrightarrow0. It has c1(𝒩)=0,c2(𝒩)=1c_1(\mathcal N)=0,c_2(\mathcal N)=1. No quadric contains a smooth elliptic quintic in ℙ3\mathbb P^3. Indeed, on a smooth quadric a curve of bidegree (a,b)(a,b) has degree a+ba+b and genus (a−1)(b−1)(a-1)(b-1), which cannot be five and one simultaneously. On a quadric cone its strict transform on 𝔽2\mathbb F_2 is as+5fas+5f and has genus −a2+5a−4-a^2+5a-4, again never one for integral aa. The strict transform is isomorphic to the original smooth curve, since its map to that curve is finite and birational. A reducible quadric would force the curve into a plane, which is also impossible. Consequently H0(𝒩)=0H^0(\mathcal N)=0, and the normalized rank-two bundle 𝒩\mathcal N is stable. By the classification of stable rank-two bundles with these Chern classes [41, Chapter II, Lemma 4.3.2], it is a null-correlation bundle. Its projectivization is YY above. The zero locus of the induced relative hyperplane section is BlCℙ3\operatorname{Bl}_C\mathbb P^3. For clarity, let 𝒮=𝒪ℙ3(−1)\mathcal S=\mathcal O_{\mathbb P^3}(-1). The null-correlation bundle is 𝒮⊥/𝒮\mathcal S^\perp/\mathcal S. On the flag variety let 𝒯=Λ/𝒮\mathcal T=\Lambda/\mathcal S be the tautological line in this quotient, and put ξ=c1(𝒯*)\xi=c_1(\mathcal T^*). Since det(Λ)=𝒮⊗𝒯\det(\Lambda)=\mathcal S\otimes\mathcal T, we have B=H+ξB=H+\xi. The wedge of the section of 𝒩(2)\mathcal N(2) with 𝒯\mathcal T defines a section of 𝒯*⊗𝒪(2H)\mathcal T^*\otimes\mathcal O(2H) (the determinant of 𝒩\mathcal N is trivial). Its zero divisor therefore has class ξ+2H=B+H\xi+2H=B+H. Since −KY=2ξ+4H=2(H+B)-K_Y=2\xi+4H=2(H+B), adjunction gives −KX=H+B-K_X=H+B. ◻

Lemma 66.2 (The initial equation at a residual point). At a point in ΣX\Sigma_X, symplectic coordinates on YY may be chosen so that its big cell has coordinates (x,t,a,b)(x,t,a,b), with (h0,h1,h2,h3)=(1,x,a+xt,xa−b),(b0,b1,b2,b3,b4)=(1,t,a,b,a2+tb).(h_0,h_1,h_2,h_3)=(1,x,a+xt,xa-b),\qquad (b_0,b_1,b_2,b_3,b_4)=(1,t,a,b,a^2+tb). The equation of XX has no terms of weight less than three for wt(x,t,a,b)=(1,1,2,3),\operatorname{wt}(x,t,a,b)=(1,1,2,3), and its weight-three initial form is, after rescaling, f0=b−xa+ν(at+xt2),ν≠0.f_0=b-xa+\nu(at+xt^2),\qquad \nu\ne0. Consequently the induced point valuation on XX is the weight-(1,1,2)(1,1,2) valuation in the regular parameters (x,t,a)(x,t,a).

Proof. The displayed formula is obtained by taking the isotropic plane spanned by (1,x,y,z)(1,x,y,z) and (0,1,t,y−xt)(0,1,t,y-xt), and putting a=y−xta=y-xt, b=xa−zb=xa-z. The bilinear sections of H+BH+B have weight-zero space spanned by 11, weight-one space spanned by x,tx,t, weight-two space spanned by a,xta,xt, and weight-three space spanned by b,xa,at+xt2b,xa,at+xt^2. Both projection fibres pass through the point and lie on XX, so its constant and linear x,tx,t terms vanish. Write the equation as F=F0(t,a,b)+xF1(t,a,b)F=F_0(t,a,b)+xF_1(t,a,b). The exceptional divisor of the second projection is locally F1=0F_1=0. The double intersection with the first projection fibre says (F1)t(0)=0(F_1)_t(0)=0. The triple intersection of the second fibre with the first exceptional divisor is read from Q|t=a=b=0=(F0)t+x((F1)t−(F0)a)−x2((F1)a+(F0)b)−x3(F1)b,Q|_{t=a=b=0}=(F_0)_t+x((F_1)_t-(F_0)_a) -x^2((F_1)_a+(F_0)_b)-x^3(F_1)_b, where all coefficients are evaluated at zero. It forces (F0)a(0)=(F1)t(0)=0(F_0)_a(0)=(F_1)_t(0)=0 and (F1)a(0)=−(F0)b(0)(F_1)_a(0)=-(F_0)_b(0). Smoothness of XX now implies (F0)b(0)≠0(F_0)_b(0)\ne0. Thus all weights below three vanish, and if the weight-three form is αb+βxa+γ(at+xt2)\alpha b+\beta xa+\gamma(at+xt^2) then β=−α\beta=-\alpha and α≠0\alpha\ne0. The first exceptional divisor is cut out by Q=Ft−xFa−x2Fb.Q=F_t-xF_a-x^2F_b. Moreover F1(t,0,0)=γt2F_1(t,0,0)=\gamma t^2: the constant and linear terms vanish, and the restriction has degree at most two. Its zero divisor is E′|ℓ=2pE'|_\ell=2p, so γ≠0\gamma\ne0. Equivalently, the linear term of Q|XQ|_X is γa\gamma a, which is nonzero because EE is smooth. Division by α\alpha proves the assertion. ◻

Lemma 66.3 (Primeness of the initial divisor). In the preceding notation the divisor X0=(f0=0)⊂YX_0=(f_0=0)\subset Y is integral. Its diagonal section ring is the associated graded anticanonical ring of XX for the weight-(1,1,2)(1,1,2) point valuation, with the zero-weight trivializations h0h_0 and b0b_0.

Proof. Retain the nonzero coefficient ν\nu. Write homogeneous coordinates h0,…,h3h_0,\ldots,h_3 and b0,…,b4b_0,\ldots,b_4. The flag incidence equations are −h0b2−h1b1+h2b0=0,h0b3−h1b2+h3b0=0,-h_0b_2-h_1b_1+h_2b_0=0,\quad h_0b_3-h_1b_2+h_3b_0=0, h0b4−h2b2+h3b1=0,h1b4−h2b3−h3b2=0,b0b4=b22+b1b3.h_0b_4-h_2b_2+h_3b_1=0,\quad h_1b_4-h_2b_3-h_3b_2=0, \qquad b_0b_4=b_2^2+b_1b_3. The extra equation is h3b0−νh2b1=0h_3b_0-\nu h_2b_1=0. On h0b0≠0h_0b_0\ne0 it solves uniquely for bb, giving affine three-space. There is no component in the boundary. Indeed, on h0=0,h1≠0h_0=0,h_1\ne0 we have b1=(h2/h1)b0b_1=(h_2/h_1)b_0, b2=(h3/h1)b0b_2=(h_3/h_1)b_0, so the extra equation is (h3−νh22/h1)b0=0(h_3-\nu h_2^2/h_1)b_0=0. The first branch has a one-dimensional base and one-dimensional fibres; the second has a two-dimensional base and a unique fibre point. Both therefore have dimension two. On h0=h1=0h_0=h_1=0 the base is a line and fibres have dimension at most one, again giving dimension at most two. On b0=0,h0≠0b_0=0,h_0\ne0, the extra equation forces h2=0h_2=0: otherwise b1=0b_1=0 and the incidence equations force all bi=0b_i=0. The remaining base is a plane, and the fibre point with b0=0b_0=0 is unique. Thus this boundary also has dimension at most two. As a Cartier divisor on smooth YY, X0X_0 is Cohen–Macaulay; its dense smooth affine three-space and absence of boundary components show that it is integral.

The weights on the fundamental coordinates are (0,1,2,3)and(0,1,2,3,4).(0,1,2,3)\quad\text{and}\quad(0,1,2,3,4). They come from diag(1,λ,λ2,λ3)\operatorname{diag}(1,\lambda,\lambda^2,\lambda^3) in the general symplectic group, of similitude multiplier λ3\lambda^3, with a shift of the Plücker linearization. Its projective action preserves YY. Hence they give a grading on the section ring of YY. Let RYR_Y be the diagonal section ring of YY. For every nonzero g∈RYg\in R_Y, the smallest-weight form of gFgF is in(g)f0\operatorname{in}(g)f_0: the ambient graded ring is a domain. Hence the initial ideal of (F)(F) is exactly (f0)(f_0). This gives a flat degeneration of RY/(F)R_Y/(F) to RY/(f0)R_Y/(f_0), with the same Hilbert function. Restriction and the divisor exact sequence identify the latter quotient with the section ring of the integral divisor X0X_0. It is therefore a domain, so localization at h0b0h_0b_0 is injective. On that chart its weights are precisely the weighted order in the regular parameters x,t,ax,t,a. Therefore there are no further cancellations, and the associated graded ring is the claimed quotient. Restriction identifies its diagonal pieces with H0(X,m(H+B))H^0(X,m(H+B)), by vanishing of the higher cohomology of 𝒪Y((m−1)(H+B))\mathcal O_Y((m-1)(H+B)). ◻

Lemma 66.4 (The exact expected order of the weighted extraction). For the weight-(1,1,2)(1,1,2) valuation GG at a residual point, AX(G)=4,SX(G)=298,βX(G)=38.A_X(G)=4,\qquad S_X(G)=\frac{29}{8},\qquad\beta_X(G)=\frac38.

Proof. The Weyl dimension formula gives dimH0(Y,𝒪Y(aH+bB))=(a+1)(b+1)(a+b+2)(a+2b+3)6.\dim H^0(Y,\mathcal O_Y(aH+bB)) =\frac{(a+1)(b+1)(a+b+2)(a+2b+3)}6. On the diagonal this is (m+1)4(m+1)^4. Weyl symmetry gives average weight 3a/2+2b3a/2+2b under the shifted grading above. The defining equation has weight three. Hence, with Nm=(m+1)4−m4N_m=(m+1)^4-m^4, the sum of the weights on the quotient diagonal piece is 7m2(m+1)4−(3+7(m−1)2)m4=7m2Nm+12m4.\frac{7m}{2}(m+1)^4- \left(3+\frac{7(m-1)}2\right)m^4 =\frac{7m}{2}N_m+\frac12m^4. Divide by mNmmN_m and let mm tend to infinity. This gives SX(G)=7/2+1/8=29/8S_X(G)=7/2+1/8=29/8. The discrepancy is 1+1+2=41+1+2=4. ◻

Lemma 66.5 (The actual weighted section spaces). Let GG be the weight-(1,1,2)(1,1,2) exceptional divisor at p∈ΣXp\in\Sigma_X. Put A=a+xtA=a+xt. For m,j≥0m,j\ge0, put Vm≥j={s∈H0(X,mL):ordG(s)≥j},Wm,j=Vm≥j/Vm≥j+1.V_m^{\ge j}=\{s\in H^0(X,mL):\operatorname{ord}_G(s)\ge j\}, \qquad W_{m,j}=V_m^{\ge j}/V_m^{\ge j+1}. These are the refined section spaces on GG. Using the local trivialization of LL, each Wm,jW_{m,j} is a space of polynomials of weighted degree jj in x,t,Ax,t,A, of weights 1,1,21,1,2. The direct sum of these spaces at level mm is exactly the span of products of degree mm in each list ℋ=(1,x,A,tA),\mathcal H=(1,x,A,tA), ℬ=(1,t,A−xt,(x−νt)A−x2t,A(A−xt−νt2)).\mathcal B= (1,t,A-xt,(x-\nu t)A-x^2t,A(A-xt-\nu t^2)). In particular its dimension is Nm=(m+1)4−m4N_m=(m+1)^4-m^4.

Proof. On the initial divisor, the equation gives b=xa−νt(a+xt)b=xa-\nu t(a+xt). Therefore h3=νtAh_3=\nu tA and b2=A−xt,b3=(x−νt)A−x2t,b4=A(A−xt−νt2).b_2=A-xt,\qquad b_3=(x-\nu t)A-x^2t, \qquad b_4=A(A-xt-\nu t^2). Multiplication of sections of dominant line bundles on the flag variety is surjective: by the Borel–Weil description it is a nonzero equivariant map onto the irreducible highest-weight representation of the sum of the weights. Thus the degree-(m,m)(m,m) sections are spanned by degree-mm products of each set of fundamental coordinates. Restriction to XX is surjective by the exact sequence for its divisor of class H+BH+B and vanishing for (m−1)(H+B)(m-1)(H+B). Lemma 66.3 identifies its associated graded quotient with the integral initial divisor. Substituting the displayed formulas, and dividing h3h_3 by the nonzero constant ν\nu, proves the assertion. The dimension was computed in Lemma 66.4. ◻

Lemma 66.6 (Bounded support and an ample subseries). The series WW has support 0≤j≤7m0\le j\le7m. For 0≤j≤m0\le j\le m, Wm,j=H0(ℙ(1,1,2),𝒪(j)).W_{m,j}=H^0\bigl(\mathbb P(1,1,2),\mathcal O(j)\bigr). In particular it contains a complete ample series on an interior ray.

Proof. The largest weights of the generators in ℋ\mathcal H and ℬ\mathcal B are three and four, respectively. Hence j≤7mj\le7m. There is a nonzero degree-one section of every weight from zero to seven; their products show that every integral jj in this interval occurs at level mm. If i+h+2k=j≤mi+h+2k=j\le m, the monomial xithAkx^i t^h A^k uses i+k≤mi+k\le m factors from ℋ\mathcal H and h≤mh\le m factors from ℬ\mathcal B. Pad each product with the constant generator. Thus every polynomial of weight jj occurs. On the ray (m,j)=(2k,2k)(m,j)=(2k,2k) these are the complete spaces of the ample Cartier bundles 𝒪(2k)\mathcal O(2k). The ray is interior to 0<j<7m0<j<7m, as required for refinement. ◻

Example 66.7 (The first weighted Hilbert function). At m=1m=1, the dimensions of W1,jW_{1,j}, for j=0,…,7j=0,\ldots,7, are 1,2,2,2,3,2,2,1.1,\ 2,\ 2,\ 2,\ 3,\ 2,\ 2,\ 1. They sum to 1515, and their weighted sum is 5353. Thus the first-level expected order is 53/1553/15. The exact character calculation in Lemma 66.4 gives its limiting value 29/829/8.

Definition 66.8 (Normalization of the refined basis boundaries). A basis boundary of WW at level mm is Dm=1mNm∑j≥0∑s∈ℬm,j{s=0},D_m=\frac{1}{mN_m}\sum_{j\ge0}\sum_{s\in\mathcal B_{m,j}}\{s=0\}, where ℬm,j\mathcal B_{m,j} is a basis of Wm,jW_{m,j} and a nonzero constant has zero divisor. For a divisor TT over GG, the maximal value of ordT(Dm)\operatorname{ord}_T(D_m) is its level-mm expected order; its limit is S(W;T)S(W;T). Thus δ(G;W)=infTAG(T)S(W;T).\delta(G;W)=\inf_T\frac{A_G(T)}{S(W;T)}. We also write δm(G;W)\delta_m(G;W) for the infimum of lct(G;Dm)\operatorname{lct}(G;D_m) over all level-mm basis boundaries. Equivalently it is the infimum of AG(T)/Sm(W;T)A_G(T)/S_m(W;T) over prime divisors TT over GG, where Sm(W;T)S_m(W;T) is the maximal order just defined. The normalization is by mNmmN_m, including all weighted-degree pieces. It is not the dimension of an individual binary space appearing below.