Text size

Section 63

Three refined surface calculations for Family 2.17

We record the surface estimates separately from the construction of the threefold modification. This separation is useful: a numerical estimate for the first exceptional divisor does not supply the estimate for its refined series.

Let F=𝔽1F=\mathbb F_1, let ss be its negative section, and let ff be a fibre of its ruling. Thus s2=−1s^2=-1, s⋅f=1s\cdot f=1, and f2=0f^2=0. Choose a section t∼s+ft\sim s+f disjoint from ss and two fibres f0,f∞f_0,f_\infty. We use Cox coordinates x,y,z,wx,y,z,w defining s,t,f0,f∞s,t,f_0,f_\infty, respectively. A basis of H0(F,𝒪F(as+bf))H^0(F,\mathcal O_F(as+bf)) consists of the monomials (63.1)xiya−izkwb−a+i−k,0≤i≤a,0≤k≤b−a+i.\label{f217:eq:F1-monomials} x^i y^{a-i}z^k w^{b-a+i-k},\qquad 0\leq i\leq a,\quad 0\leq k\leq b-a+i. The statements below concern the series obtained by refining along the exceptional divisor of the blow-up of a fibre of Eℙ3→CE_{\mathbb P^3}\to C. They also apply if the actual refined spaces are only known to be contained in the indicated spaces, with the same asymptotic total dimension.

Lemma 63.1 (The surface spaces and their normalization). Suppose the refined spaces, indexed by u=j/mu=j/m, have the following asymptotic description. Write c=3−uc=3-u in the last interval. uabrfixed divisor/m[0,1]uu+100[1,2]uu+1u−10[2,3]2c3cc(u−2)R\begin{array}{c|c|c|c|c} u& a&b&r&\text{fixed divisor}/m\\ \hline {[0,1]}&u&u+1&0&0\\ {[1,2]}&u&u+1&u-1&0\\ {[2,3]}&2c&3c&c&(u-2)R \end{array} Here the moving sections belong to H0(F,𝒪F(mas+mbf)⊗ℐmr)H^0(F,\mathcal O_F(mas+mbf)\otimes\mathcal I^{mr}), where ℐ\mathcal I is either the ideal of two distinct points or the ideal of a curvilinear subscheme of length two. In the distinct case ℐmr\mathcal I^{mr} means the equal multiplicity conditions at the two points. Rounding the indices changes the formulas below by O(m−1)O(m^{-1}). The divisor RR has class 3s+4f3s+4f and vanishes to order two along ℐ\mathcal I.

In each of the three configurations in Lemma 63.2, the area of the polygon of moving monomials is A(u)={(u2+2u)/2,0≤u≤1,(−u2+6u−2)/2,1≤u≤2,3(3−u)2,2≤u≤3.A(u)= \begin{cases} (u^2+2u)/2,&0\leq u\leq1,\\ (-u^2+6u-2)/2,&1\leq u\leq2,\\ 3(3-u)^2,&2\leq u\leq3. \end{cases} In particular ∫03A(u)du=4\int_0^3 A(u)\,du=4. The normalization is therefore Nm=4m3+O(m2)N_m=4m^3+O(m^2), and the average coefficient of the fixed divisor RR is 1/161/16.

Proof. The complete polygon in (63.1) has area a(b−a)+a2/2a(b-a)+a^2/2. In the second and third intervals the marked-point conditions remove area r2r^2. For two distinct points this is the sum of two triangles of area r2/2r^2/2. For the curvilinear ideal used below it is one triangle with side lengths rr and 2r2r. The displayed formulas follow by substitution. The fixed coefficient is 14∫23(u−2)A(u)du=34∫23(u−2)(3−u)2du=116.\frac14\int_2^3 (u-2)A(u)\,du =\frac34\int_2^3(u-2)(3-u)^2\,du=\frac1{16}. If one has only an inclusion of refined spaces, equality of the leading total dimensions leaves O(m2)O(m^2) omitted sections. All the exponents are O(m)O(m), so their omission changes a normalized first moment by O(m−1)O(m^{-1}). Thus the same limiting first moments give upper bounds for the actual refined series. ◻

Lemma 63.2 (The three toric estimates). Let WW be the series in Lemma 63.1. The following estimates hold.

  1. If the two marked points are distinct, lie off ss, and lie on distinct fibres, then δ(F;W)≥24/23\delta(F;W)\geq24/23.

  2. If one marked point lies on ss and the other lies off ss on a different fibre, and RR contains ss with multiplicity one, then δ(F;W)≥96/95\delta(F;W)\geq96/95.

  3. Suppose the marked scheme is supported off ss, is transverse to its ruling fibre, and has ideal (y,z2)(y,z^2) after a choice of tt and f0f_0 as above. If RR does not contain f0f_0, then δ(F;W)≥16/15\delta(F;W)\geq16/15.

Proof. In the first case an automorphism of FF places the points at t∩f0t\cap f_0 and t∩f∞t\cap f_\infty. Indeed, after choosing coordinates on the base, translations by sections of 𝒪ℙ1(1)\mathcal O_{\mathbb P^1}(1) can prescribe the two fibre coordinates. The extra polygon inequalities are a−i+k≥r,a−i+b−a+i−k≥r.a-i+k\geq r,\qquad a-i+b-a+i-k\geq r. In the second case we place the points at s∩f0s\cap f_0 and t∩f∞t\cap f_\infty. The inequalities become i+k≥r,a−i+b−a+i−k≥r.i+k\geq r,\qquad a-i+b-a+i-k\geq r. In the third case they are replaced by 2(a−i)+k≥2r2(a-i)+k\geq2r.

For completeness, the moving first moments are (63.2)stf0f∞two points off s35/482/35/65/6one point on s79/9655/965/689/96curvilinear point off s35/482/315/1635/48.\label{f217:eq:surface-moment-table} \begin{array}{c|cccc} &s&t&f_0&f_\infty\\ \hline \text{two points off }s&35/48&2/3&5/6&5/6\\ \text{one point on }s&79/96&55/96&5/6&89/96\\ \text{curvilinear point off }s&35/48&2/3&15/16&35/48. \end{array} Here is a direct check of every entry. Without the marked-point conditions the four moments are 23/24,17/24,49/48,49/4823/24,17/24,49/48,49/48. Put 𝒥(ϕ)=14∫12ϕ(u,u+1,u−1)du+14∫23ϕ(2c,3c,c)du.\mathcal J(\phi)=\frac14\int_1^2 \phi(u,u+1,u-1)\,du+ \frac14\int_2^3\phi(2c,3c,c)\,du. The moments removed by the marked conditions are obtained by applying 𝒥\mathcal J to the entries of the following table: stf0f∞two off sr2(a−r/3)r3/3r2(b−r/3)/2r2(b−r/3)/2one on sr2a/2r2a/2r2(b−r/3)/2r2(b−a+r/3)/2curvilinear off sr2(a−r/3)r3/32r3/3r2(b−r).\begin{array}{c|cccc} &s&t&f_0&f_\infty\\ \hline \text{two off }s&r^2(a-r/3)&r^3/3&r^2(b-r/3)/2&r^2(b-r/3)/2\\ \text{one on }s&r^2a/2&r^2a/2&r^2(b-r/3)/2&r^2(b-a+r/3)/2\\ \text{curvilinear off }s&r^2(a-r/3)&r^3/3&2r^3/3&r^2(b-r). \end{array} For example, in the first row the two removed triangles have centroids with ss-coordinate a−r/3a-r/3. Their f0f_0-coordinates are r/3r/3 and b−2r/3b-2r/3. The other entries are obtained in exactly the same way. These are integrals of explicitly displayed polynomials; they give (63.2).

It remains to include the fixed divisor. Its monomials have the form (63.3)xiy3−izkw1+i−k.\label{f217:eq:R-monomial} x^i y^{3-i}z^k w^{1+i-k}. In the first case the double-point conditions imply k≥i−1k\geq i-1 and k≤2k\leq2. Thus the four exponents are at most 3,3,2,23,3,2,2, respectively. After adding the coefficient 1/161/16 of RR, all four coefficients are at most 23/2423/24.

In the second case the smallest xx-exponent of RR is one. Among the terms with this exponent the double-point conditions give 1≤k≤21\leq k\leq2. Choose a torus degeneration selecting one such term. Its initial divisor is s+2t+kf0+(2−k)f∞s+2t+kf_0+(2-k)f_\infty. The four total coefficients are bounded by 85/96,67/96,23/24,95/9685/96,67/96,23/24,95/96, respectively.

In the third case first minimize the zz-exponent. The hypothesis that RR does not contain f0f_0 makes this minimum zero. Since R∈(y,z2)2R\in(y,z^2)^2, a term with k=0k=0 in (63.3) has i≤1i\leq1. The four total coefficients are consequently at most 19/24,41/48,15/16,41/4819/24,41/48,15/16,41/48.

We justify why these monomial estimates control every divisorial valuation. Fix such a valuation vv on FF. In each graded piece choose a basis compatible with vv and with a one-parameter subgroup of the torus. The two-filtration lemma permits this choice. For each mm the subgroup can be chosen in the indicated open cone of weights and can distinguish all the finitely many monomials occurring in the moving spaces. The initial basis is then their monomial basis, and the fixed divisor has the initial monomial just specified. The limiting basis boundary is supported on s,t,f0,f∞s,t,f_0,f_\infty, a simple normal crossings divisor. For every positive number strictly less than 24/2324/23, 96/9596/95, or 16/1516/15, respectively, its multiple is log canonical for all sufficiently large mm. Semicontinuity of log canonical thresholds gives the same assertion before degeneration. The bounds are independent of vv. Letting mm tend to infinity proves the three assertions. ◻

Example 63.3. In the second case suppose the lowest xx-degree part of RR is xy2zwx y^2 z w. The total coefficients of the limiting basis boundary are 85/96,67/96,43/48,95/9685/96,67/96,43/48,95/96. Thus the worst coefficient occurs on f∞f_\infty, although the marked point on the negative section lies on f0f_0. Keeping the marked-point conditions is essential: omitting them would replace each of the two fibre moments by 49/4849/48.

Lemma 63.4 (The remaining curvilinear configuration). Suppose the marked scheme is supported at A∈sA\in s, its infinitesimal direction is transverse both to ss and to the ruling fibre, and RR has multiplicity three at AA. In local coordinates x,zx,z with s=(x=0)s=(x=0) and ruling fibre (z=0)(z=0), write its ideal as (x+z,z2)(x+z,z^2). For the complete surface spaces in Lemma 63.1, the moving expected order of the ordinary blow-up of AA is 89/4889/48. Including the fixed divisor gives S(W;ordA)=8948+316=4924>2.S(W;\operatorname{ord}_A)=\frac{89}{48}+\frac3{16} =\frac{49}{24}>2. Consequently this exceptional-surface refinement does not establish K-stability in this configuration. If the refined spaces have the same leading dimensions as these complete spaces, the displayed expected order is exact.

Proof. Put b=x+zb=x+z and group the monomials in (63.1) by their ordinary degree dd at AA. In the first two intervals their xx-exponents run from i0=max{0,(d−1)/2}i_0=\max\{0,(d-1)/2\} to i1=min{u,d}i_1=\min\{u,d\}. Set n=i1−i0n=i_1-i_0 and q=max{0,2r−d}q=\max\{0,2r-d\}. The corresponding binary space is (b−z)i0zd−i1Symn⟨b,z⟩.(b-z)^{i_0}z^{d-i_1}\operatorname{Sym}^{n}\langle b,z\rangle. Membership in (b,z2)r(b,z^2)^r forces divisibility by bqb^q. The other displayed factors are coprime to bb, so the asymptotic dimension of the degree-dd slice is n−qn-q when n≥qn\geq q and is zero otherwise. Its ordinary first moment is d(n−q)d(n-q).

Here are the integration intervals and the values of n,qn,q; thus the calculation can be checked without an implicit chamber choice: udnq[0,1][0,u]d0[0,1][u,1]u0[0,1][1,1+2u]u−(d−1)/20[1,3/2][u−1,2u−2]d2u−2−d[1,3/2][2u−2,1]d0[1,3/2][1,u](d+1)/20[1,3/2][u,1+2u]u−(d−1)/20[3/2,2][u−1,1]d2u−2−d[3/2,2][1,2u−2](d+1)/22u−2−d[3/2,2][2u−2,u](d+1)/20[3/2,2][u,1+2u]u−(d−1)/20.\begin{array}{c|c|c|c} u&d&n&q\\ \hline {[0,1]}&[0,u]&d&0\\ {[0,1]}&[u,1]&u&0\\ {[0,1]}&[1,1+2u]&u-(d-1)/2&0\\ \hline {[1,3/2]}&[u-1,2u-2]&d&2u-2-d\\ {[1,3/2]}&[2u-2,1]&d&0\\ {[1,3/2]}&[1,u]&(d+1)/2&0\\ {[1,3/2]}&[u,1+2u]&u-(d-1)/2&0\\ \hline {[3/2,2]}&[u-1,1]&d&2u-2-d\\ {[3/2,2]}&[1,2u-2]&(d+1)/2&2u-2-d\\ {[3/2,2]}&[2u-2,u]&(d+1)/2&0\\ {[3/2,2]}&[u,1+2u]&u-(d-1)/2&0. \end{array} Multiplying by the normalization 1/41/4 and integrating d(n−q)d(n-q) over these regions gives 65/4865/48. In the final interval scale the calculation at u=2u=2 by c=3−uc=3-u. At u=2u=2 the degree intervals are 1≤d≤21\leq d\leq2, with n=(d+1)/2,q=2−dn=(d+1)/2,q=2-d, and 2≤d≤52\leq d\leq5, with n=(5−d)/2,q=0n=(5-d)/2,q=0. Their unnormalized first moments sum to 88. The last interval therefore contributes 14∫018c3dc=1/2\frac14\int_0^1 8c^3\,dc=1/2. The moving total is 65/48+1/2=89/4865/48+1/2=89/48. Integrating the slice dimensions (n−q)+(n-q)_+ themselves gives 2/32/3 in the first interval, 7/37/3 in the second, and 11 in the last, for a total of 44. Thus these containing spaces have total dimension 4m3+O(m2)4m^3+O(m^2), the same leading dimension as the actual refinement. The inclusion of Lemma 62.7 therefore omits only O(m2)O(m^2) dimensions. All orders under consideration are O(m)O(m), so the normalized first moment differs by O(m−1)O(m^{-1}). This proves exactness for the actual series in this configuration. The fixed contribution is multA(R)/16=3/16\operatorname{mult}_A(R)/16=3/16. The discrepancy of the ordinary point blow-up of a smooth surface is two, which proves the assertion. ◻

Lemma 63.5. Let p∈D=E∩E′p\in D=E\cap E'. Let ℓ⊂E\ell\subset E be the fibre of π|E\pi|_E through pp, let ℓ′⊂E′\ell'\subset E' be the fibre of q|E′q|_{E'} through pp, and put B=q(ℓ)B=q(\ell). Blow up ℓ\ell, and denote its exceptional surface by F≅𝔽1F\cong\mathbb F_1 and its negative section by ss. The strict transform of ℓ′\ell' meets FF at a unique point AA. Then the following conditions are equivalent:

  1. DD is singular at pp;

  2. B∩C′B\cap C' is the length-two scheme supported at q(p)q(p), and A∈sA\in s;

  3. ℓ\ell is tangent to E′E' at pp, and ℓ′\ell' is tangent to EE at pp.

Consequently, the unhandled configuration consisting of a tangent secant BB and a collision point on ss occurs exactly over Sing(D)\operatorname{Sing}(D).

Proof. The restriction q|ℓq|_\ell is an isomorphism onto the line BB, because q*𝒪Q(1)⋅ℓ=1q^*\mathcal O_Q(1)\cdot\ell=1. The ideal of C′C' pulls back under qq to 𝒪X(−E′)\mathcal O_X(-E'). Its restriction to ℓ\ell therefore identifies the scheme B∩C′B\cap C' with the divisor E′|ℓE'|_\ell. The latter has degree two and contains no one-dimensional component. Thus B∩C′B\cap C' is nonreduced at q(p)q(p) if and only if E′|ℓ=2pE'|_\ell=2p, if and only if Tpℓ⊂TpE′T_p\ell\subset T_pE'. This also proves that the intersection multiplicity is exactly two, rather than merely at least two. The strict transform CZC_Z in BlB(Q)\operatorname{Bl}_B(Q) is isomorphic to C′C': the restricted centre ideal on the smooth curve C′C' is invertible. Its intersection with the exceptional divisor is the same length-two Cartier divisor as B∩C′B\cap C'. Thus nonreducedness here is exactly nonreducedness of the length-two centre scheme Ξ\Xi in the surface calculation.

The tangent lines TpℓT_p\ell and Tpℓ′T_p\ell' are distinct. Indeed, dqdq is nonzero on the first line, since q|ℓq|_\ell is an isomorphism, whereas it vanishes on the second. In particular, ℓ′\ell' is transverse to ℓ\ell as a pair of smooth curves in XX. Its strict transform meets the exceptional surface FF in the point represented by the nonzero normal direction Tpℓ′modTpℓT_p\ell'\bmod T_p\ell in Nℓ/X,pN_{\ell/X,p}.

The normal sequence for ℓ⊂E⊂X\ell\subset E\subset X is 0→𝒪ℓ→Nℓ/X→𝒪ℓ(−1)→0.0\longrightarrow\mathcal O_\ell \longrightarrow N_{\ell/X} \longrightarrow\mathcal O_\ell(-1)\longrightarrow0. It splits, since H1(ℙ1,𝒪(1))=0H^1(\mathbb P^1,\mathcal O(1))=0. The section of ℙ(Nℓ/X)\mathbb P(N_{\ell/X}) corresponding, in the convention of lines, to the subbundle Nℓ/E=𝒪ℓN_{\ell/E}=\mathcal O_\ell has normal bundle Hom(𝒪ℓ,𝒪ℓ(−1))\operatorname{Hom}(\mathcal O_\ell,\mathcal O_\ell(-1)) and hence self-intersection −1-1. It is therefore precisely ss; equivalently it is the intersection of FF with the strict transform of EE. Consequently A∈s⇔Tpℓ′modTpℓ∈Nℓ/E,p⇔Tpℓ′⊂TpE.A\in s\quad\Longleftrightarrow\quad T_p\ell'\bmod T_p\ell\in N_{\ell/E,p} \quad\Longleftrightarrow\quad T_p\ell'\subset T_pE.

Finally, EE and E′E' are smooth divisors in the smooth threefold XX. Their scheme-theoretic intersection is singular at pp precisely when their tangent planes coincide. Each plane already contains its own ruling tangent line. As these two lines are distinct, the two cross-tangency conditions Tpℓ⊂TpE′T_p\ell\subset T_pE' and Tpℓ′⊂TpET_p\ell'\subset T_pE hold precisely when both planes equal their two-dimensional span. This proves all equivalences. ◻

Remark 63.6. The criterion distinguishes three kinds of points on a smooth DD: both incidence intersections can be transverse; E′|ℓE'|_\ell can have a double point while A∉sA\notin s; or A∈sA\in s while E′|ℓE'|_\ell is reduced at pp. The simultaneous tangency case is absent on a smooth DD. No assertion that DD is smooth follows from the smoothness of XX.