Text size

Section 69

The classification problem and the results proved here

Family 2.20: the K-polystable classification

A smooth threefold in Family 2.20 is the blowup π:X=BlC0V5→V5\pi:X=\operatorname{Bl}_{C_0}V_5\longrightarrow V_5 of the quintic del Pezzo threefold along a smooth twisted cubic. Here V5⊂ℙ6V_5\subset\mathbb P^6 is a smooth three-dimensional linear section of Gr(2,5)⊂ℙ9\operatorname{Gr}(2,5)\subset\mathbb P^9, and a twisted cubic is a curve isomorphic to ℙ1\mathbb P^1 whose given embedding is defined by the complete series |𝒪ℙ1(3)||\mathcal O_{\mathbb P^1}(3)|. The anticanonical degree of XX is 2626. Projection from the span of C0C_0 resolves to a conic bundle η:X→ℙ2\eta:X\to\mathbb P^2.

This family contains smooth strictly K-semistable members. Thus the assertion that every smooth member is K-polystable is false. The general smooth member is K-stable, and the unique smooth member with infinite automorphism group is K-polystable [3, Proposition 5.43 and Section 7.2]. Section 7.2 of that reference identifies nonclosed parameter orbits and formulates the expected converse on the GIT-stable locus. The later discussion in [23, Introduction and Lemma 3.5] does not settle this converse. The results below prove the converse on the GIT-stable locus and give the complete smooth K-polystable classification.

Let ℋ=Gr(2,Sym4ℂ2)\mathcal H=\operatorname{Gr}(2,\operatorname{Sym}^4\mathbb C^2). Its points are pencils of binary quartics, namely two-dimensional linear spaces of homogeneous polynomials of degree four in two variables. Let ℋ∘\mathcal H^\circ be the open subset of pencils containing no nonzero square of a quadratic. The equivariant description of the Hilbert scheme in [3, Section 7.2] identifies ℋ∘\mathcal H^\circ with smooth twisted cubics on V5V_5. Write XUX_U for the corresponding blowup. The following theorem states exactly the conclusions established in this part.

Theorem 69.1. For U∈ℋ∘U\in\mathcal H^\circ, the threefold XUX_U is K-polystable if and only if UU is GIT-polystable. It is K-stable if and only if UU is GIT-stable. Every smooth member is K-semistable. More explicitly, the following assertions hold.

  1. The pencil UU is GIT-semistable for the Plücker linearization. It fails to be GIT-stable exactly when it is equivalent under PGL2\operatorname{PGL}_2 to Ua,c=⟨x3y,xy3+ax2y2+cx4⟩.U_{a,c}=\langle x^3y,\ xy^3+a x^2y^2+c x^4\rangle. The member XU0,0X_{U_{0,0}} is K-polystable. Every displayed member with (a,c)≠(0,0)(a,c)\ne(0,0) is strictly K-semistable. The latter isomorphism classes are parametrized by ℙ(1,3)\mathbb P(1,3).

  2. If wUw_U is the Wronskian sextic of UU, then all its roots have multiplicity at most three. Two triple roots characterize XU0,0X_{U_{0,0}}. Exactly one triple root characterizes the nonclosed smooth GIT orbits and hence the strictly K-semistable members in (1). All the remaining pencils are GIT-stable.

  3. Put H=π*𝒪V5(1)H=\pi^*\mathcal O_{V_5}(1), let EE be the exceptional divisor, and put B=H−EB=H-E. Then SX(E)=12,SX(B)=63104,SX(H)=37104.S_X(E)=\frac12,\qquad S_X(B)=\frac{63}{104},\qquad S_X(H)=\frac{37}{104}. Every prime divisor on XX has positive beta invariant. If pp lies on a smooth fibre of η\eta, a suitable smooth S∈|B|S\in|B| through that fibre satisfies AX(v)−SX(v)≥41104v(S)>0A_X(v)-S_X(v)\ge\frac{41}{104}v(S)>0 for every divisorial valuation centred at pp. Every valuation with curve centre not contained in the inverse image of the discriminant also has positive beta invariant.

  4. At every double conic fibre 2R2R, let FF be the reduced exceptional divisor of the blowup of the pulled-back maximal ideal of its base point. Then AX(F)=3,SX(F)=297104,AX(F)−SX(F)=15104.A_X(F)=3,\qquad S_X(F)=\frac{297}{104},\qquad A_X(F)-S_X(F)=\frac{15}{104}. The different on F≅ℙ1×ℙ1F\cong\mathbb P^1\times\mathbb P^1 is Γ/2\Gamma/2, where Γ\Gamma has bidegree (2,1)(2,1). The vanishing sequence (1,3)(1,3) occurs exactly when the point over E∩RE\cap R is ramified for Γ→ℙ1\Gamma\to\mathbb P^1 in the direction factor.

  5. If a double fibre corresponds to a common root with vanishing sequence (1,2)(1,2), then δp(X)≥104/99\delta_p(X)\ge104/99 at every point of its support. In particular this holds along every double fibre of a GIT-stable member. The remaining possible centres of nonpositive beta invariant on such a member lie over reduced singular conics.

  6. For the blowup FF of any component AA of a reduced singular fibre, AX(F)=2A_X(F)=2 and SX(F)=43/26S_X(F)=43/26. Its complete refined series is given in Lemma 77.5. If the fibre node is outside EE and π(A)\pi(A) has trivial normal bundle in V5V_5, then δp(X)≥104/99\delta_p(X)\ge104/99 for every p∈Ap\in A.

  7. For a GIT-stable pencil, every valuation of nonpositive beta invariant must have centre at the node on EE of a reduced conic fibre. These points correspond to Δsm∩Q\Delta_{\mathrm{sm}}\cap Q, and there are at most 3−d3-d of them, where dd is the number of nodes of the discriminant. At each of these points the ordinary point extraction satisfies SX(J)=11/4S_X(J)=11/4 and AX(J)=3A_X(J)=3. Its complete refinement gives δp(X)≥104/103>1\delta_p(X)\ge104/103>1 at every such node.

  8. A triangular discriminant gives a K-stable member. There is one such member up to isomorphism, with pencil U△=xy(x−y)⟨x,y⟩U_\triangle=xy(x-y)\langle x,y\rangle.

Consequently, the K-polystable smooth members are exactly those whose Wronskian has no triple root, together with the unique member U0,0U_{0,0}, whose Wronskian has two triple roots. The first collection consists of K-stable members; the latter member is K-polystable but not K-stable. Exactly one triple root gives strict K-semistability. An equivalent criterion using only the conic discriminant and the branch conic is given in Corollary 82.1.

Recall the terminology from geometric invariant theory. For a projective variety with a reductive group action and an ample linearization, a point is GIT-semistable if some invariant section of a positive tensor power does not vanish there. A semistable point is GIT-polystable if its orbit is closed in the semistable locus. It is GIT-stable if this orbit is closed and its stabilizer is finite. We apply the Hilbert–Mumford criterion to the Plücker line bundle 𝒪ℋ(1)\mathcal O_{\mathcal H}(1); see [39]. These definitions concern the parameter space. Identifying them with K-stability of XUX_U requires a separate argument.

The proof is organized as follows. Binary quartics give the normal forms, their actual special degenerations, and the Wronskian test. Smooth conic fibres admit quintic del Pezzo sections with complete refined series. Exact surface decompositions give the strict beta estimates there. We then compute the corresponding estimates for reduced singular fibres and construct the weighted extraction over a double line. Its complete refined series leads to strict estimates along every double fibre on the GIT-stable locus. A discriminant test gives another form of the parameter criterion. We then construct a two-flop model for reduced components, prove their complete refined series, and treat ordinary components away from fibre nodes on EE. Negative-section flags handle special lines, and unmarked rulings handle non-nodal points in the same-ruling configuration. The finite reduction isolates at most three points. At these nodes, an integral initial algebra identifies the complete point refinement. Two explicit surface calculations, including an extra fixed line in the degenerate-conic case, give the strict bound 104/103104/103 and finish the classification.