Text size

Section 98

The index-two endpoint of the mixed fibre

We work at the A1A_1 point pp of a fibre of type A2+A1A_2+A_1. Let GG be the exceptional divisor of the ordinary blow-up of pp. Thus AX(G)=3A_X(G)=3 and the exceptional pair is G=ℙ2G=\mathbb P^2 with zero different. The actual refined anticanonical series is denoted by WW. Write Nm=h0(X,−mKX)=(10m3+15m2+11m+3)/3N_m=h^0(X,-mK_X)=(10m^3+15m^2+11m+3)/3.

Theorem 98.1. At this point one has SX(G)=10740,δ(G;W)≥240221,δp(X)≥240221>1.S_X(G)=\frac{107}{40},\qquad \delta(G;W)\ge\frac{240}{221},\qquad \delta_p(X)\ge\frac{240}{221}>1.

Lemma 98.2 (The actual initial sections). In homogeneous coordinates [x:z:w][x:z:w] on GG, put t=zw−x2t=zw-x^2. The algebra of the refined series contains the degree-one sections ti{1,x,z}(0≤i≤2),w,xw,tw,txw.t^i\{1,x,z\}\quad(0\le i\le2),\qquad w,xw,tw,txw. Here the first degree is the anticanonical degree; their second degrees are their usual polynomial degrees in x,z,wx,z,w.

Proof. Put the index-two source point at [0:1][0:1], the index-three source point at [1:0][1:0], and their common image at [1:0][1:0]. The graph equation then has the form ux2y3+vg(x,y)=0,g=y5+∑i=15gixiy5−i,g5≠0.u x^2y^3+v g(x,y)=0,\qquad g=y^5+\sum_{i=1}^5g_i x^iy^{5-i},\qquad g_5\ne0. On the affine chart u=Y0=1u=Y_0=1 of ℙ1×ℙ2\mathbb P^1\times\mathbb P^2, write v=tv=t and Y1=xY_1=x, Y2=cY_2=c. The conic equation is z=c−x2z=c-x^2. The two cubic lifts of the graph equation are c+tG0(x,c),xc+tG1(x,c),c+tG_0(x,c),\qquad xc+tG_1(x,c), where G0=1+g1x+g2c+g3xc+g4c2+g5xc2,G1=x+g1c+g2xc+g3c2+g4xc2+g5c3.\begin{aligned} G_0&=1+g_1x+g_2c+g_3xc+g_4c^2+g_5xc^2,\\ G_1&=x+g_1c+g_2xc+g_3c^2+g_4xc^2+g_5c^3. \end{aligned} In particular G1−xG0=zR(c),R(c)=g1+g3c+g5c2.G_1-xG_0=zR(c),\qquad R(c)=g_1+g_3c+g_5c^2. On the blow-up chart containing pp, set A=c+tG0z,B=xc+tG1z,w=A−1.A=\frac{c+tG_0}{z},\qquad B=\frac{xc+tG_1}{z}, \qquad w=A-1. Then B=xA+tR(c)B=xA+tR(c), c=x2+zc=x^2+z and tG0=zw−x2.tG_0=zw-x^2. Since G0(0,0)=1G_0(0,0)=1, the variables x,z,wx,z,w are regular parameters at pp, and the initial form of tt for their ordinary order is zw−x2zw-x^2.

The thirteen anticanonical sections are ti{1,x,c}t^i\{1,x,c\} for 0≤i≤20\le i\le2 and A,B,tA,tBA,B,tA,tB. Replace the last four by A−1=w,B−x−g1t=xw+t(g3c+g5c2),tA−t=tw,tB−tx−g1t2=txw+t2(g3c+g5c2).\begin{aligned} A-1&=w,& B-x-g_1t&=xw+t(g_3c+g_5c^2),\\ tA-t&=tw,& tB-tx-g_1t^2&=txw+t^2(g_3c+g_5c^2). \end{aligned} Their initial forms are w,xw,tw,txww,xw,tw,txw. The initial form of cc is zz. Thus all thirteen displayed polynomials belong to the actual associated graded series. Products give the asserted algebra inclusion. ◻

Lemma 98.3 (An exact normal basis). Let 𝒜\mathcal A be the algebra generated by these thirteen sections. Its degree-mm space has basis xazbwctsx^az^bw^ct^s, with bc=0bc=0, in the following two disjoint ranges: (A1.1)c=0a,b,s≥0,2a+3b≤3m,s≤2m,s+a+b≤3m,b=0,c≥1a,c,s≥0,c≤m,2a+c≤3m,s+c≤2m,s+a+c≤3m.\begin{array}{c|l} c=0& a,b,s\ge0,\quad 2a+3b\le3m,\quad s\le2m, \quad s+a+b\le3m,\\ b=0,\ c\ge1&a,c,s\ge0,\quad c\le m,\quad2a+c\le3m, \quad s+c\le2m,\quad s+a+c\le3m. \end{array} \tag{A1.1} Its dimension is Nm=10m3+15m2+11m+33.N_m=\frac{10m^3+15m^2+11m+3}{3}. Consequently 𝒜\mathcal A is the entire refined algebra WW.

Proof. The relation is zw=x2+tzw=x^2+t. In addition to the degree-one generators, we may use x3tix^3t^i of degree two for 0≤i≤30\le i\le3: write i=j+ki=j+k, where 0≤j≤20\le j\le2, 0≤k≤10\le k\le1, and use (ztj)(xwtk)=x3ti+xti+1.(zt^j)(xwt^k)=x^3t^i+xt^{i+1}. The last term is a product of two degree-one generators. Straightening a zz-factor against a ww- or xwxw-factor now uses zw=x2+tzw=x^2+t and the displayed identity. Each resulting term remains a product of the generators just listed, of the same anticanonical degree. Repetition eliminates every zwzw pair. Normal monomials with bc=0bc=0 are linearly independent, since zwzw is a leading monomial of the single relation in ℂ[x,z,w,t]/(zw−x2−t)\mathbb C[x,z,w,t]/(zw-x^2-t).

For c=0c=0, use kk factors x3x^3 of degree two. The remaining xx and zz factors have degree one. Filling unused places with 1,t,t21,t,t^2 shows that a prescribed ss is possible if and only if max(0,a+b−m)≤k≤⌊a/3⌋,s≤2m−k.\max(0,a+b-m)\le k\le\lfloor a/3\rfloor, \qquad s\le2m-k. Every integer ss up to the upper bound is possible, since each factor allows a consecutive interval of tt-exponents. Taking the smallest kk gives the first row of (A1.1).

For b=0b=0, after using kk factors x3x^3, pair as many of the remaining xx factors as possible with the cc factors ww. The minimum required anticanonical degree is max(a−k,c+2k)\max(a-k,c+2k), and the maximum tt-exponent in total degree mm is 2m−c−k2m-c-k. Thus the conditions are max(0,a−m)≤k≤min(⌊a/3⌋,⌊(m−c)/2⌋),s≤2m−c−k.\max(0,a-m)\le k\le \min\left(\lfloor a/3\rfloor,\lfloor(m-c)/2\rfloor\right), \qquad s\le2m-c-k. They are equivalent to the second row of (A1.1).

Here is the count, including the parity terms. Put kn=⌊n/2⌋k_n=\lfloor n/2\rfloor. For the first row set n=m−bn=m-b; its contribution for this nn is (n+kn+1)(2m+1)−kn(kn+1)2.(n+k_n+1)(2m+1)-\frac{k_n(k_n+1)}2. For the second row set n=m−cn=m-c; its contribution is (m+kn+1)(m+n+1)−kn(kn+1)2.(m+k_n+1)(m+n+1)-\frac{k_n(k_n+1)}2. Therefore the dimension equals ∑n=0m((n+kn+1)(2m+1)−kn(kn+1)2)+∑n=0m−1((m+kn+1)(m+n+1)−kn(kn+1)2)=Nm.\begin{aligned} &\sum_{n=0}^{m}\left((n+k_n+1)(2m+1) -\frac{k_n(k_n+1)}2\right)\\ &\quad+\sum_{n=0}^{m-1}\left((m+k_n+1)(m+n+1) -\frac{k_n(k_n+1)}2\right)=N_m. \end{aligned} The last identity follows by splitting each sum into even and odd nn; it holds for both parities of mm. This equals h0(X,mL)h^0(X,mL), so Lemma 94.2 proves equality with the actual refined algebra.

The same summation gives the primary weight sum, with no parity restriction. In the first row, for fixed nn, sum ∑a=0n+kn((U+1)(a+m−n)+U(U+1)),U=min(2m,2m+n−a).\sum_{a=0}^{n+k_n}\left((U+1)(a+m-n)+U(U+1)\right), \qquad U=\min(2m,2m+n-a). In the second row replace the upper limit by m+knm+k_n and put U=min(m+n,2m+n−a)U=\min(m+n,2m+n-a). Sum the first row over 0≤n≤m0\le n\le m and the second over 0≤n<m0\le n<m. Splitting at a=na=n and a=ma=m, respectively, and then at the two parities of nn, gives c=0b=0,c≥1m evenm(97m3+233m2+218m+64)24m2(3m+1)(13m+4)8m oddm(m+1)(97m2+136m+55)24m(3m+1)(13m2+4m+3)8.\begin{array}{c|cc} &c=0&b=0,\ c\ge1\\\hline m\text{ even}& \dfrac{m(97m^3+233m^2+218m+64)}{24}& \dfrac{m^2(3m+1)(13m+4)}8\\[2mm] m\text{ odd}& \dfrac{m(m+1)(97m^2+136m+55)}{24}& \dfrac{m(3m+1)(13m^2+4m+3)}8. \end{array} Both rows have total m(107m3+154m2+115m+32)12.\frac{m(107m^3+154m^2+115m+32)}{12}. After division by mNmmN_m this proves SX(G)=107/40S_X(G)=107/40. ◻

Lemma 98.4 (Binary spaces and their moments). Give x,z,wx,z,w the auxiliary torus weights 1,0,21,0,2, respectively. The polynomial t=zw−x2t=zw-x^2 has weight two. In primary degree jj and auxiliary weight qq, put b=max(j−q,0),c=max(q−j,0),k=q−2c.b=\max(j-q,0),\qquad c=\max(q-j,0),\qquad k=q-2c. Put ℓ=max(0,⌈2k+3b+c−3m4⌉,k+b+c−3m),v=min(2m−c,⌊k2⌋).\ell=\max\left(0, \left\lceil\frac{2k+3b+c-3m}{4}\right\rceil, k+b+c-3m\right),\qquad v=\min\left(2m-c,\left\lfloor\frac k2\right\rfloor\right). When 0≤b,c≤m0\le b,c\le m and v≥ℓv\ge\ell, the component is exactly (A1.2)xk−2vzbwctℓSymv−ℓ(x2,zw).x^{k-2v}z^bw^ct^\ell\operatorname{Sym}^{v-\ell}(x^2,zw). \tag{A1.2} Otherwise it is zero. Write d=v−ℓd=v-\ell and a=k−2va=k-2v. Define αm=∑(a+d)(d+1)mNm,bm=∑b(d+1)mNm,cm=∑c(d+1)mNm,sm=∑ℓ(d+1)mNm,hm=∑d(d+1)2mNm.\begin{aligned} \alpha_m&=\frac{\sum(a+d)(d+1)}{mN_m},& b_m&=\frac{\sum b(d+1)}{mN_m},& c_m&=\frac{\sum c(d+1)}{mN_m},\\ s_m&=\frac{\sum\ell(d+1)}{mN_m},& h_m&=\frac{\sum d(d+1)}{2mN_m}. \end{aligned} Their limits are (A1.3)α=91160,b=47320,c=1564,s=193320,h=83320.\alpha=\frac{91}{160},\qquad b=\frac{47}{320},\qquad c=\frac{15}{64},\qquad s=\frac{193}{320},\qquad h=\frac{83}{320}. \tag{A1.3} In particular the fixed xx-coefficient is α−2h=1/20\alpha-2h=1/20, and α+2(c+s+h)=22180,α+2(b+s+h)=20780.\alpha+2(c+s+h)=\frac{221}{80},\qquad \alpha+2(b+s+h)=\frac{207}{80}.

Proof. The equations for the two gradings are j=a+b+c+2sj=a+b+c+2s and q=a+2c+2sq=a+2c+2s. Their difference determines b,cb,c, and (A1.1) gives exactly ℓ≤s≤v\ell\le s\le v. Using t=zw−x2t=zw-x^2 proves (A1.2).

For the asymptotic sums put y=b/my=b/m in the zz-branch and y=c/my=c/m in the ww-branch, and replace k/mk/m by kk. The lower and upper limits of s/ms/m are LUzmax(0,(2k+3y−3)/4,k+y−3)min(2,k/2)wmax(0,(2k+y−3)/4,k+y−3)min(2−y,k/2).\begin{array}{c|c|c} &L&U\\ \hline z&\max(0,(2k+3y-3)/4,k+y-3)&\min(2,k/2)\\ w&\max(0,(2k+y-3)/4,k+y-3)&\min(2-y,k/2). \end{array} Integrate on 0≤y≤10\le y\le1, L≤UL\le U, and put D=U−LD=U-L. Splitting the region at the indicated affine walls gives (A1.4)∫D∫(k−2U+D)D∫yD∫LD∫D2/2z35/2417/2447/96835/768257/768w15/819/1625/32709/768407/768.\begin{array}{c|ccccc} &\int D&\int(k-2U+D)D&\int yD&\int LD&\int D^2/2\\ \hline z&35/24&17/24&47/96&835/768&257/768\\ w&15/8&19/16&25/32&709/768&407/768. \end{array} \tag{A1.4} These are ordinary integrals in dydkdy\,dk; for example, the chamber zz, L=0L=0, U=k/2U=k/2 is the triangle 0≤y≤10\le y\le1, 0≤k≤3(1−y)/20\le k\le3(1-y)/2. Its contributions to the first, second and last columns are 3/163/16, 9/1289/128 and 9/2569/256. Multiplying the summed columns by 3/103/10, and taking the third column separately in each branch, gives (A1.3). Floors and ceilings contribute only O(m3)O(m^3) to the numerators, while mNm=(10/3)m4+O(m3)mN_m=(10/3)m^4+O(m^3).

The average degree on ℙ2\mathbb P^2 is (α−2h)+b+c+2s+4h=10740.(\alpha-2h)+b+c+2s+4h=\frac{107}{40}. By the definition of the primary associated graded series this is SX(G)S_X(G). ◻

Proof of the theorem. The conic pencil on ℙ2\mathbb P^2 is Tλ=(zw−λx2=0),T0=Z0+W0,T∞=2X0.T_\lambda=(zw-\lambda x^2=0),\qquad T_0=Z_0+W_0,\qquad T_\infty=2X_0. Its two base points are [0:1:0][0:1:0] and [0:0:1][0:0:1]. On the respective smooth charts its branches have equations w=λx2w=\lambda x^2 and z=λx2z=\lambda x^2. Two successive ordinary blow-ups at either point separate all distinct branches.

An arbitrary basis divisor obtained by choosing a basis in each space (A1.2) has the form Bm=(αm−2hm)X0+bmZ0+cmW0+smT1+∑νλTλ,B_m=(\alpha_m-2h_m)X_0+b_mZ_0+c_mW_0+s_mT_1 +\sum\nu_\lambda T_\lambda, where ∑νλ=2hm\sum\nu_\lambda=2h_m and 0≤νλ≤hm0\le\nu_\lambda\le h_m. Indeed a basis of H0(ℙ1,𝒪(d))H^0(\mathbb P^1,\mathcal O(d)) has total degree d(d+1)d(d+1), and its total vanishing order at any point is at most d(d+1)/2d(d+1)/2.

The component bounds for this divisor are αm,bm+hm,cm+hm,sm+hm,hm.\alpha_m,\quad b_m+h_m,\quad c_m+h_m,\quad s_m+h_m,\quad h_m. At [0:1:0][0:1:0], the second blow-up has log discrepancy three and order at most αm+2(cm+sm+hm)\alpha_m+2(c_m+s_m+h_m); at [0:0:1][0:0:1] the corresponding order is αm+2(bm+sm+hm)\alpha_m+2(b_m+s_m+h_m). The first blow-up inequalities follow by linear interpolation with the bound on the x=0x=0 component. After these two blow-ups the total transform is a simple normal crossings divisor. Thus these inequalities test all divisorial valuations, not only monomial ones.

The limiting component bounds are 91160,1332,79160,6980,83320;\frac{91}{160},\quad\frac{13}{32},\quad\frac{79}{160}, \quad\frac{69}{80},\quad\frac{83}{320}; the two corner bounds divided by their log discrepancies are 221/240221/240 and 69/8069/80. Their maximum is 221/240221/240. Consequently (G,τBm)(G,\tau B_m) is log canonical for every τ<240/221\tau<240/221 and all sufficiently large mm.

For an arbitrary divisorial valuation over GG, choose in each primary piece a basis compatible with its valuation filtration and the auxiliary torus filtration. Taking least auxiliary weights produces the divisors just considered. Lower semicontinuity of log canonical thresholds under this torus degeneration gives the same bound for the original valuation-adapted basis divisor. Taking mm to infinity proves δ(G;W)≥240/221\delta(G;W)\ge240/221. Abban–Zhuang adjunction now gives δp(X)≥min{3107/40,240221}=240221.\delta_p(X)\ge\min\left\{\frac{3}{107/40}, \frac{240}{221}\right\} =\frac{240}{221}. No Zariski decomposition on the blown-up threefold was used. The entire refined algebra was identified by an exact Hilbert count; the only further modifications are the two explicit surface blow-ups at each base point of the conic pencil. ◻