Text size

Section 79

Reduction to finitely many points and a triangle example

79.1 Curve centres in the original exceptional divisor

Lemma 79.1. Every divisorial valuation whose centre is a curve contained in EE has positive beta invariant.

Proof. Write E=ℙ1×ℙ1E=\mathbb P^1\times\mathbb P^1, with ff a fibre of E→C0E\to C_0 and ss the other ruling class. The complete containing series for the EE-refinement has class (1+u)s+(4−2u)f(1+u)s+(4-2u)f, for 0≤u≤10\le u\le1, and no fixed part. Its moment along ss is 326∫01∫01+u2(1+u−v)(4−2u)dvdu=34.\frac3{26}\int_0^1\int_0^{1+u} 2(1+u-v)(4-2u)\,dv\,du=\frac34. For an irreducible curve T∼as+bfT\sim as+bf other than a fibre of E→C0E\to C_0, one has a≥1a\ge1, b≥0b\ge0. Monotonicity therefore gives S(WE;T)≤3/4S(W^E;T)\le3/4. Together with SX(E)=1/2S_X(E)=1/2, adjunction at the generic point of TT gives δT(X)≥4/3\delta_T(X)\ge4/3.

A fibre ff of E→C0E\to C_0 is not contained in η−1(Δ)\eta^{-1}(\Delta). Indeed η|f\eta|_f maps isomorphically onto a line. If this line were contained in Δ\Delta, its general reduced conic fibre would have a component whose image is a line in V5V_5 through the fixed point π(f)\pi(f). This would give infinitely many distinct lines through that point, which is impossible.

Here is an elementary justification of the last fact. Since V5V_5 is cut out by quadrics, the directions of lines through a point are cut out by quadrics in ℙ(TpV5)=ℙ2\mathbb P(T_pV_5)=\mathbb P^2. At least one is nonzero, for otherwise V5V_5 would contain its projective tangent three-space. A positive-dimensional component would thus have degree one or two. The union of the corresponding lines would be a surface of degree one or two in V5V_5. But Pic(V5)=ℤ[H]\operatorname{Pic}(V_5)=\mathbb Z[H] makes the degree of every effective surface a positive multiple of H3=5H^3=5. This contradiction proves finiteness. Now apply Lemma 71.8 to ff. ◻

79.2 An upper bound for the ordinary point extraction

Lemma 79.2. Let AA be a smooth curve in a smooth threefold XX, let p∈Ap\in A, and let FF be the exceptional divisor of its blowup. Write fpf_p for the ruling over pp, and JJ for the divisor of the ordinary blowup of pp. For a big line bundle LL and its actual refinement WFW^F, one has SL(J)≤SL(F)+S(WF;fp).S_L(J)\le S_L(F)+S(W^F;f_p). This is an inequality; identifying the right-hand side with SL(J)S_L(J) requires additional information about the point filtration.

Proof. Choose regular parameters (x,y,z)(x,y,z) at pp with A=(y=z=0)A=(y=z=0). If a local section has order kk along AA, its first nonzero normal term is ∑i+j=kaij(x)yizj.\sum_{i+j=k}a_{ij}(x)y^iz^j. Let q=minordxaijq=\min\operatorname{ord}_x a_{ij}. Then qq is exactly the order of the leading section along fpf_p, whereas the order of the original section at pp is at most k+qk+q. Terms of higher normal order may lower that total order; they cannot invalidate this upper bound.

In each H0(X,mL)H^0(X,mL) choose a basis compatible with the point-order filtration and the single lexicographic refinement by (k,q)(k,q). Two filtrations of a finite-dimensional vector space admit a common adapted basis. The sum of the point orders is therefore at most the sum of the kk-orders plus the sum of the qq-orders. Divide by mh0(X,mL)m h^0(X,mL) and pass to the limit. These three sums give respectively the three moments in the assertion. ◻

Example 79.3. For A=(y=z=0)A=(y=z=0), the local function x2y+z2x^2y+z^2 has normal order k=1k=1 and leading-coefficient order q=2q=2, but its order at the origin is two, strictly smaller than k+q=3k+q=3. Thus one cannot replace the initial-term upper bound by an equality without checking the actual point filtration.

Proposition 79.4. At the node p0∈Ep_0\in E of any reduced fibre, the ordinary point extraction JJ satisfies AX(J)=3,SX(J)≤114,AX(J)−SX(J)≥14.A_X(J)=3,\qquad S_X(J)\le\frac{11}{4},\qquad A_X(J)-S_X(J)\ge\frac14.

Proof. For either component extraction, SX(F)=43/26S_X(F)=43/26. The marked ruling has full moment 57/5257/52 for both normal-bundle types. The ordinary case was proved in Lemma 77.9. For the special case, use q0=[1:0:0]q_0=[1:0:0], qD=[0:1:0]q_D=[0:1:0], and the infinitely near point qEq_E in direction Z=0Z=0. The residual monomials XiYb−i−kZkX^iY^{b-i-k}Z^k satisfy 0≤i≤a,max(0,cE−a+i,cD−i)≤k≤b−i.0\le i\le a,\qquad \max(0,c_E-a+i,c_D-i)\le k\le b-i. Integrating kk over this polygon with factor 3/133/13 gives 11/10411/104, 59/10459/104, and 34/10434/104 on the intervals [0,1][0,1], [1,2][1,2], and [2,3][2,3], respectively. Their sum is one. The fixed ruling in R=s+f0+TR=s+f_0+T adds 5/525/52. Thus its full moment is 57/5257/52 also. The preceding lemma now gives SX(J)≤4326+5752=114.S_X(J)\le\frac{43}{26}+\frac{57}{52}=\frac{11}{4}. The discrepancy of the ordinary point blowup is three. ◻

79.3 The residual point set

Theorem 79.5. Let U∈ℋ∘U\in\mathcal H^\circ be GIT-stable, and put 𝒩E={p∈E:p is the node of a reduced conic fibre}.\mathcal N_E=\{p\in E:\ p\text{ is the node of a reduced conic fibre}\}. Every divisorial valuation of nonpositive beta invariant must have centre a point of 𝒩E\mathcal N_E. The map η\eta identifies this finite set with Δsm∩Q\Delta_{\mathrm{sm}}\cap Q. If dd is the number of nodes of Δ\Delta, then #𝒩E≤3−d.\#\mathcal N_E\le3-d. At every point of this residual set the ordinary point extraction has beta invariant at least 1/41/4.

Proof. The earlier results treat divisor centres, smooth conic fibres, and every double fibre on the GIT-stable locus. Theorem 77.8, Proposition 78.3, and Corollary 78.6 treat closed point centres on every reduced fibre outside the set displayed above. In the possible equality case on a special-line surface, the retained term 9v(F)/269v(F)/26 gives strict positivity.

Curve centres contained in EE are excluded by Lemma 79.1. For a curve not contained in EE, choose a general point outside EE. Over a reduced singular conic the new estimates there have the strict bound 104/99104/99; over a double fibre use Theorem 75.7; otherwise use Lemma 71.8. Thus all curve centres are excluded as well.

To identify the residual set, move a point of QQ to [0:0:1][0:0:1] in the model of Lemma 73.1. The fibre equation is −y12+z(f0y0+f1y1)+h2z2=0.-y_1^2+z(f_0y_0+f_1y_1)+h_2z^2=0. Its discriminant on QQ is f02f_0^2. At an intersection with Δ\Delta, therefore f0=0f_0=0. If the discriminant is smooth there, this is a union of two distinct lines with vertex [y0:y1:z]=[1:0:0][y_0:y_1:z]=[1:0:0], which lies on E=(z=0)E=(z=0). Conversely, a reduced conic whose node lies on EE has a double intersection with EE, so its base point lies on QQ. This proves the claimed bijection.

For a GIT-stable pencil QQ is not a component of Δ\Delta, and every node of Δ\Delta has intersection multiplicity two with QQ, by Proposition 76.1. The restriction Δ|Q\Delta|_Q is the square of a nonzero binary cubic. Each remaining point of intersection also has multiplicity at least two. Since Δ⋅Q=6\Delta\cdot Q=6, there are at most 3−d3-d of them. Finally apply Proposition 79.4. ◻

Corollary 79.6. Every smooth member whose conic discriminant is a triangle is K-stable. There is one such member up to isomorphism. Its pencil can be written U△=xy(x−y)⟨x,y⟩.U_\triangle=xy(x-y)\langle x,y\rangle.

Proof. A triangle has three nodes. All lie on QQ. Bezout’s theorem forces intersection multiplicity two at each and no further intersection. The pencil is GIT-stable by Proposition 76.1. Theorem 79.5 now leaves no possible centre of nonpositive beta invariant. The valuative criterion gives K-stability.

By Lemma 73.10, the three double fibres correspond to three distinct common roots of the pencil. Hence U=g⟨x,y⟩U=g\langle x,y\rangle for a squarefree binary cubic gg. Conversely such a pencil contains no square: multiplying gg by a linear form leaves at least two of its distinct roots with odd multiplicity. It therefore belongs to ℋ∘\mathcal H^\circ and has three double fibres. A reduced nodal plane cubic with three nodes is a triangle. Finally PGL2\operatorname{PGL}_2 is transitive on unordered triples of distinct points, giving the displayed normal form and uniqueness. ◻

Remark 79.7. The finite reduction alone does not finish the classification. There can be three, two, or one points left when the discriminant has zero, one, or two nodes. The bound for the ordinary point extraction controls only that divisor. Sections 80 and 81 identify its entire refined series and control every valuation above these points. Thus the conclusion does not rely on the positivity of a single beta invariant.