Text size

Section 74

The weighted moment at a double fibre in two explicit members

This section concerns two specified members of Family 2.202.20. It does not assert that the calculation is independent of the member. Let V⊂ℙA,B,C,D,E,F,G6V\subset\mathbb P^6_{A,B,C,D,E,F,G} be given by AE−4BD+C2=0,AF−BE+2CD=0,AG−CE+8D2=0,BG−CF+2DE=0,CG−4DF+E2=0.\begin{aligned} AE-4BD+C^2&=0,\qquad AF-BE+2CD=0,\qquad AG-CE+8D^2=0,\\ BG-CF+2DE&=0,\qquad CG-4DF+E^2=0. \end{aligned} Thus VV is the smooth quintic del Pezzo threefold. Consider either of the smooth twisted cubics Cs=V∩{A+2C=D−F=E=0},Cn=V∩{B+2E=D−G=F=0}.C^\mathrm{s}=V\cap\{A+2C=D-F=E=0\},\qquad C^\mathrm{n}=V\cap\{B+2E=D-G=F=0\}. Their corresponding pencils in Sym4ℂ2\operatorname{Sym}^4\mathbb C^2 are ⟨e3,e2+e0⟩\langle e_3,e_2+e_0\rangle and ⟨e1,e3+e0⟩\langle e_1,e_3+e_0\rangle, respectively, where ei=x4−iyie_i=x^{4-i}y^i. The first pencil is GIT-stable, and the second is strictly GIT-semistable. In both cases the conic bundle on X=BlCsVX=\operatorname{Bl}_{C^\mathrm{s}}V or BlCnV\operatorname{Bl}_{C^\mathrm{n}}V has a double fibre supported on the strict transform of R={C=D=E=F=G=0}R=\{C=D=E=F=G=0\}. Write HH for the pullback of a hyperplane of VV, write ECE_C for the exceptional divisor of X→VX\to V, and put L=−KX=2H−ECL=-K_X=2H-E_C. The letter FF occurring in the equations is a homogeneous coordinate; denote the exceptional divisor of the weighted extraction by TT.

Lemma 74.1. For either of these two members, the extraction of the double fibre has AX(T)=3,SX(T)=297104,AX(T)−SX(T)=15104.A_X(T)=3,\qquad S_X(T)=\frac{297}{104},\qquad A_X(T)-S_X(T)=\frac{15}{104}. More precisely, if Nm=h0(X,mL)N_m=h^0(X,mL), the sum of the vanishing orders of a basis adapted to ordT\operatorname{ord}_T is Nm=(2m+1)(13m2+13m+6)6,∑i=1NmordT(si)=m(99m3+146m2+93m+22)8.N_m=\frac{(2m+1)(13m^2+13m+6)}6, \qquad \sum_{i=1}^{N_m}\operatorname{ord}_T(s_i) =\frac{m(99m^3+146m^2+93m+22)}8.

Proof. We give the filtration and the cohomology arguments, since the Hilbert series of an arbitrary initial ideal would not suffice.

Assign the weights (0,0,1,2,2,2,3)(0,0,1,2,2,2,3) to (A,B,C,D,E,F,G)(A,B,C,D,E,F,G). The resulting initial threefold V0V_0 is defined by AE−4BD+C2=AF−BE=AG−CE=BG−CF=CG−4DF+E2=0.AE-4BD+C^2=AF-BE=AG-CE=BG-CF=CG-4DF+E^2=0. These are the Pfaffians of the skew matrix obtained from (0ABCtD−A02C8DE−B−2C02EF−C−8D−2E0G−tD−E−F−G0)\begin{pmatrix} 0&A&B&C&tD\\ -A&0&2C&8D&E\\ -B&-2C&0&2E&F\\ -C&-8D&-2E&0&G\\ -tD&-E&-F&-G&0 \end{pmatrix} by setting t=0t=0. Its general fibre is isomorphic to VV: the substitution (A,B,C,D,E,F,G)↦(A,B,tC,t2D,t2E,t2F,t3G)(A,B,C,D,E,F,G)\mapsto(A,B,tC,t^2D,t^2E,t^2F,t^3G) in the original equations, followed by division by their smallest powers of tt, gives exactly this family.

Here are direct checks on the special fibre. On B=1B=1 it is affine space with coordinates A,C,FA,C,F, since E=AF,D=(A2F+C2)/4,G=CF.E=AF,\qquad D=(A^2F+C^2)/4,\qquad G=CF. On G=1G=1 it is affine space with coordinates D,E,FD,E,F, since C=4DF−E2,A=CE,B=CF.C=4DF-E^2,\qquad A=CE,\qquad B=CF. The subset B=G=0B=G=0 has dimension at most one in projective space. The Pfaffian ideal has codimension three and is Cohen–Macaulay. Consequently there is no extra component or embedded component there, and the ideal is prime. The Hilbert series is that of a quintic del Pezzo threefold; thus these initial equations give the full initial ideal. Equivalently, the Pfaffian resolution gives a flat family over 𝔸1\mathbb A^1. The Jacobian criterion on B=G=0B=G=0 shows that V0V_0 has only one singular point, namely [0:0:0:1:0:0:0][0:0:0:1:0:0:0]. On the chart D=1D=1, after eliminating BB and FF, its equation is AG−CE=0AG-CE=0. This point is an ordinary double point.

We next identify the filtration with the required valuation. At the generic point of RR, use the chart B=1B=1. On the original threefold, A,C,FA,C,F are regular coordinates and D=A2F+C24−2AC,E=AF+2CD,G=CF−2DE.D=\frac{A^2F+C^2}{4-2AC},\qquad E=AF+2CD,\qquad G=CF-2DE. The weighted valuation in the two normal parameters C,FC,F has weights 1,21,2. The displayed initial equations embed in the associated graded ring ℂ(A)[C,F]\mathbb C(A)[C,F], so the global weight filtration on the homogeneous coordinate ring is exactly this valuation filtration. For the stable example the conic-bundle coordinates at the double fibre are (D−F)/(A+2C)(D-F)/(A+2C) and E/(A+2C)E/(A+2C). Their ideal at the generic point of RR is (F,C2)(F,C^2). For the nonstable example use (D−G)/(B+2E)(D-G)/(B+2E) and F/(B+2E)F/(B+2E), whose ideal is again (F,C2)(F,C^2). Thus this valuation is the exceptional valuation of the normalized blowup of the pulled-back base-point ideal. In particular it has log discrepancy 1+2=31+2=3. The exceptional divisor ECE_C does not contain RR, so its valuation is zero. Hence the valuation on H0(X,mL)H^0(X,mL) is the restriction of the weight filtration on H0(V,2mH)H^0(V,2mH).

The centres extend to the flat families {A+2tC=D−F=E=0}⊂Vt,{B+2t2E=D−tG=F=0}⊂Vt.\{A+2tC=D-F=E=0\}\subset V_t, \qquad \{B+2t^2E=D-tG=F=0\}\subset V_t. Each fibre of either family is a smooth twisted cubic. For the first family its coordinates satisfy A=−2tCA=-2tC, D=FD=F, E=0E=0 and C2=4BFC^2=4BF, BG=CFBG=CF, CG=4F2CG=4F^2. For the second family they satisfy B=−2t2EB=-2t^2E, D=tGD=tG, F=0F=0 and, after substituting into the equations of VtV_t, the relations AE+C2+8t3EG=0AE+C^2+8t^3EG=0, AG−CE+8t3G2=0AG-CE+8t^3G^2=0, CG+E2=0CG+E^2=0. Replacing AA by A+8t3GA+8t^3G gives the standard twisted cubic relations. At t=0t=0 the centres are C0s={A=D−F=E=0},C0n={B=D=F=0}.C_0^\mathrm{s}=\{A=D-F=E=0\},\qquad C_0^\mathrm{n}=\{B=D=F=0\}. Both avoid the unique singular point of V0V_0. The total centre is therefore a relative regular immersion of codimension two, contained in the smooth locus of the ambient family. Blowing it up commutes with base change and gives a flat family XtX_t. More explicitly, each quotient by a power of the relative ideal is flat: its successive quotients are the symmetric powers of the relative conormal bundle on the flat centre. Thus every power of the relative ideal is tt-saturated in the ambient Rees algebra.

The ideal sheaf of each centre is generated by the three displayed linear forms. Thus the blowup embeds as a graph in Vt×ℙ2V_t\times\mathbb P^2. If Dt=H−ECD_t=H-E_C, then Lt=H+DtL_t=H+D_t is the restriction of 𝒪(1,1)\mathcal O(1,1), and is ample. Moreover, −KXt=Lt-K_{X_t}=L_t. The special fibre has only the ordinary double point inherited from V0V_0, and is klt. Kawamata–Viehweg vanishing gives Hi(Xt,mLt)=0H^i(X_t,mL_t)=0 for i>0i>0 and m≥0m\geq0. This verifies cohomology and base change for the spaces whose weights we compute.

We calculate the conormal bundle of the special centre, including its grading. For C0sC_0^\mathrm{s}, on B=1B=1 take the conormal parameters a=Aa=A and n=F−C2/4n=F-C^2/4, of weights 0,20,2. On G=1G=1 take E/GE/G and (D−F)/G(D-F)/G. On their overlap, modulo the square of the ideal of the curve, they are a/Ca/C and −4n/C3-4n/C^3. Consequently NC0s/V0*=𝒪ℙ1(−1)[0]⊕𝒪ℙ1(−3)[2].N^*_{C_0^\mathrm{s}/V_0} =\mathcal O_{\mathbb P^1}(-1)[0]\oplus \mathcal O_{\mathbb P^1}(-3)[2]. Here the brackets record the weights of frames at the point with B≠0B\ne0. For C0nC_0^\mathrm{n}, use B,DB,D on A=1A=1 and F/G,D/GF/G,D/G on G=1G=1. Along the curve E=−C2E=-C^2 and G=−C3G=-C^3; the transition functions are B/CB/C and −D/C3-D/C^3, giving the same graded splitting. In both cases H|C0=𝒪ℙ1(3)H|_{C_0}=\mathcal O_{\mathbb P^1}(3), with section weights 0,1,2,30,1,2,3.

For completeness, we perform the character calculation. Let zz record valuation weight, and qq record hyperplane degree. The graded Hilbert series of V0V_0 is 1−q2(2z2+2z3+z4)+q3(2z5+2z4+z3)−q5z7(1−q)2(1−qz)(1−qz2)3(1−qz3).\frac{1-q^2(2z^2+2z^3+z^4)+q^3(2z^5+2z^4+z^3)-q^5z^7} {(1-q)^2(1-qz)(1-qz^2)^3(1-qz^3)}. Differentiating with respect to zz and putting z=1z=1 gives q(10+16q+3q2)/(1−q)5q(10+16q+3q^2)/(1-q)^5. It follows that h0(V0,nH)=(n+1)(5n2+10n+6)6,wtH0(V0,nH)=n(n+1)(29n2+57n+34)24.h^0(V_0,nH)=\frac{(n+1)(5n^2+10n+6)}6, \quad \operatorname{wt}H^0(V_0,nH) =\frac{n(n+1)(29n^2+57n+34)}{24}. The successive quotients of the mmth infinitesimal neighbourhood of C0C_0, twisted by 2mH2mH, are SymjNC0/V0*⊗𝒪C0(6m)=⨁i=0j𝒪ℙ1(6m−j−2i)[2i],0≤j<m.\operatorname{Sym}^jN^*_{C_0/V_0}\otimes\mathcal O_{C_0}(6m) =\bigoplus_{i=0}^{j}\mathcal O_{\mathbb P^1}(6m-j-2i)[2i], \qquad 0\leq j<m. All these line bundles have nonnegative degree. Their total dimensions and weights are, respectively, ∑j=0m−1∑i=0j(6m−j−2i+1)=7m(m+1)(2m+1)6,\sum_{j=0}^{m-1}\sum_{i=0}^j(6m-j-2i+1) =\frac{7m(m+1)(2m+1)}6, ∑j=0m−1∑i=0j(6m−j−2i+1)(3m−j/2+i)=m(m+1)(167m2+83m+2)24.\sum_{j=0}^{m-1}\sum_{i=0}^j (6m-j-2i+1)(3m-j/2+i) =\frac{m(m+1)(167m^2+83m+2)}{24}. The restriction sequence is exact on global sections. Indeed, H1(V0,ℐC0m(2mH))=H1(X0,mL0)=0.H^1(V_0,\mathcal I_{C_0}^m(2mH))=H^1(X_0,mL_0)=0. The higher direct images vanish for the blowup of a regular codimension-two centre. The same assertions hold in the family. Thus the initial of the filtered subspace H0(V,ℐCm(2mH))H^0(V,\mathcal I_C^m(2mH)) is exactly H0(V0,ℐC0m(2mH))H^0(V_0,\mathcal I_{C_0}^m(2mH)); there is no unproved replacement of an initial power by a power of an initial ideal. Subtracting the two jet formulas from the ambient formulas at n=2mn=2m gives the finite identities in the statement. Dividing the weight by mNmmN_m and letting mm tend to infinity proves SX(T)=297/104S_X(T)=297/104. ◻

Remark 74.2. The identical value in a GIT-stable and a strictly K-semistable example is useful. The single inequality AX(T)>SX(T)A_X(T)>S_X(T) does not settle the valuations above TT. Their analysis requires the different and a refined series on TT, including the position of EC∩RE_C\cap R relative to the ramification of the curve of degree (2,1)(2,1).

74.1 The moment at every double fibre

Proposition 74.3. Let XX be any smooth member of Family 2.202.20, and let 2R2R be a double fibre of its conic bundle. Let TT be the exceptional prime divisor obtained by blowing up the pullback of the corresponding base-point ideal. Then AX(T)=3,SX(T)=297104,AX(T)−SX(T)=15104.A_X(T)=3,\qquad S_X(T)=\frac{297}{104},\qquad A_X(T)-S_X(T)=\frac{15}{104}. In fact the finite dimension and weight identities of Lemma 74.1 hold for this valuation.

Proof. By Lemma 73.10, a double fibre corresponds to a common root of the binary quartic pencil. Move this root to x=0x=0. Use the coordinates and weight filtration of the preceding calculation. There are precisely two possible vanishing sequences.

If the sequence is (1,2)(1,2), write the pencil as ⟨e2+be1+ce0,e3+de1+ee0⟩.\langle e_2+b e_1+c e_0,\ e_3+d e_1+e e_0\rangle. Put k=4c−b2k=4c-b^2. The number kk is nonzero, since otherwise the first quartic is a square. By the exterior-square calculation in Lemma 73.11, the ideal of the twisted cubic in V5V_5 is generated by l0=A−bB+(2c−d)C+8eD+2(cd−eb)E,l1=D−cF−eG,l2=E−bF−dG.\begin{aligned} l_0&=A-bB+(2c-d)C+8eD+2(cd-eb)E,\\ l_1&=D-cF-eG,\qquad l_2=E-bF-dG. \end{aligned} Their weighted initial forms define on V0V_0 the curve Zb,c={A−bB=D−cF=E−bF=0}.Z_{b,c}=\{A-bB=D-cF=E-bF=0\}. After eliminating A,D,EA,D,E, its equations are C2=kBF,BG=CF,CG=kF2.C^2=kBF,\qquad BG=CF,\qquad CG=kF^2. It is a smooth twisted cubic, and avoids the node of V0V_0. Both it and the original curve have Hilbert function 3n+13n+1 in every degree n≥0n\geq0. Since the full weighted initial ideal contains the displayed ideal and has the same Hilbert function, the two ideals coincide. This verifies the initial centre, including scheme structure.

We record a graded normalization to make the uniformity of the character calculation explicit. On V0V_0 set A′=A−bB,E′=E−bF,D′=D−b2E+b24F,A'=A-bB,\quad E'=E-bF,\quad D'=D-\frac b2E+\frac{b^2}{4}F, and leave B,C,F,GB,C,F,G unchanged. This is an invertible graded linear change preserving the five equations of V0V_0. It transforms Zb,cZ_{b,c} into {A′=E′=D′−(k/4)F=0}\{A'=E'=D'-(k/4)F=0\}. Choose a∈ℂ*a\in\mathbb C^* with a2k=4a^2k=4. The further graded automorphism (A′,B,C,D′,E′,F,G)↦(aA′,B,aC,a2D′,aE′,F,aG)(A',B,C,D',E',F,G)\longmapsto (aA',B,aC,a^2D',aE',F,aG) transforms this curve into C0sC_0^\mathrm{s} of the preceding lemma. Thus the pair consisting of the weighted initial threefold and its centre is graded-isomorphic to the pair used there.

If the sequence is (1,3)(1,3), write U=⟨e1,e3+ae2+ce0⟩U=\langle e_1,e_3+a e_2+c e_0\rangle. The three generators are B−aC+2cE,D−cG,F−aG.B-aC+2cE,\qquad D-cG,\qquad F-aG. Their initial forms define C0n={B=D=F=0}C_0^\mathrm{n}=\{B=D=F=0\}. This is again a smooth twisted cubic with the same Hilbert function as the original, so the same argument proves equality with the full initial ideal.

In either case the centre in the flat weighted family stays inside the smooth locus of the ambient family. Its infinitesimal neighbourhoods are flat, because the centre is a relative regular immersion and the graded quotients are its symmetric conormal powers. The powers of the relative ideal are therefore tt-saturated. The relative blowup has the same klt Fano special fibre, up to graded isomorphism, as one of the two examples already considered. Kawamata–Viehweg vanishing and cohomology and base change identify its global section module with the saturated Rees submodule of H0(Vt,2mH)H^0(V_t,2mH). In particular no additional initial sections occur.

Finally the filtration is the actual exceptional valuation. At the generic point of RR, the weight-zero generator l0l_0 is a unit after trivializing HH. The ideal (l1/l0,l2/l0)(l_1/l_0,l_2/l_0) has initial ideal (F,C2)(F,C^2) in the two regular normal parameters. Thus its exceptional valuation has weights (1,2)(1,2), exactly the valuation used above. The original exceptional divisor meets RR in a point and has valuation zero, so passing from 2mH2mH to mL=2mH−mECmL=2mH-mE_C introduces no grading shift. The character and the conormal jet calculation of Lemma 74.1 now apply without any change. ◻