Bridgeland stability conditions and skew lines on P 3

Inspired by Schmidt’s work on twisted cubics [20], we study wall crossings in Bridge-land stability, starting with the Hilbert scheme Hilb 2 m +2 ( P 3 ) parametrizing pairs of skew lines and plane conics union a point. We ﬁnd two walls. Each wall crossing corresponds to a contraction of a divisor in the moduli space and the contracted space remains smooth. Building on work by Chen–Coskun–Nollet [8] we moreover prove that the contractions are K -negative extremal in the sense of Mori theory and so the moduli spaces are projective.


Introduction
After the introduction of Bridgeland's manifold of stability conditions on a triangulated category [6], several applications to the study of the birational geometry of moduli spaces have appeared: the moduli space is viewed as parametrizing stable objects in the derived category of some underlying variety X, and the question is how the moduli space changes as the stability condition varies.This is the topic of wall crossing in the stability manifold.We refer to [15] for an overview and in particular for examples of the success of this viewpoint in cases where X is a surface.For threefolds and notably in the case X = P 3 , important progress was made by Schmidt [20], allowing among other things a study of wall crossings for the Hilbert scheme of twisted cubics (see also Xia [22] for further work on this case; additional examples in the same spirit have been investigated by Gallardo-Lozano Huerta-Schmidt [11] and Rezaee [19]).The case considered in the present text is that of pairs of skew lines in P 3 and their deformations.This is analogous to twisted cubics in the sense that a twisted cubic degenerates to a plane nodal curve with an embedded point much as a pair of skew lines degenerates to a pair of lines in a plane together with an embedded point.
More precisely we study wall crossing for the Hilbert scheme Hilb 2m+2 (P 3 ) of subschemes Y ⊂ P 3 with Hilbert polynomial 2m + 2. It has two smooth components C and S: a general point in C is a conic-union-a-point Y = C ∪ {P } and a general point in S is a pair of skew lines Y = L 1 ∪ L 2 .Note that when a line pair is deformed until the two lines meet, the result is a pair of intersecting lines with an embedded point at the intersection, and this can also be viewed as a degenerate case of a conic union a point.
For an appropriately chosen Bridgeland stability condition on the bounded derived category of coherent sheaves D b (P 3 ), the ideal sheaves I Y can be viewed as the stable objects with fixed Chern character, say v = ch(I Y ).When deforming the stability condition, we identify two walls, separating three chambers.Getting slightly ahead of ourselves, the situation is illustrated in Figure 2 in Section 3.2: α and β are parameters for the stability conditions considered and we restrict ourselves to the region to the immediate left in the picture of the hyperbola β 2 − α 2 = 4 (the role of this boundary curve is explained in Section 2.4).In this region we have the two walls W 1 and W 2 separating three chambers, labeled by Roman numerals as in the figure.Let M I , M II and M III be the moduli spaces of Bridgeland stable objects with Chern character v in each chamber, considered as algebraic spaces (for existence see Piyaratne-Toda [18,Corollary 4.23]).
Our first main result contains the set-theoretical description of these moduli spaces: Theorem 1.1.
(I) M I is Hilb 2m+2 (P 3 ) with its two components S and C described above.
(II) M II consists of: (i) Ideal sheaves I Y for Y ∈ Hilb 2m+2 (P 3 ) not contained in a plane.
(ii) Non-split extensions F P,V in 0 → I P/V (−2) → F P,V → O P 3 (−1) → 0 (1.1) for V ⊂ P 3 a plane and P ∈ V .Moreover, F P,V is uniquely determined up to isomorphism by the pair (P, V ).(III) M III consists of: (i) Ideal sheaves I Y for Y ∈ Hilb 2m+2 (P 3 ) a pair of disjoint lines or a pure double line.(ii) Non-split extensions G P,V in 0 → O V (−2) → G P,V → I P (−1) → 0 (1.2) with P and V as above.Moreover, G P,V is uniquely determined by the pair (P, V ).
To locate walls and classify stable objects we employ the method due to Schmidt [20], which involves "lifting" walls from an intermediate notion of tilt stability.Schmidt considers as an application the Hilbert scheme Hilb 3m+1 (P 3 ): it parametrizes twisted cubics and plane cubics union a point.This was our starting point and we can apply many of Schmidt's results directly, although modified or new arguments are needed as well.The end result is closely analogous in the two cases, with two wall crossings of the same nature.In the twisted cubic situation, however, Schmidt also finds an additional "final wall crossing" where all objects are destabilized.This has no direct analogy in our case.
Next we describe the moduli spaces geometrically, guided by the set theoretic classification of objects above; this leads to contractions of the two smooth components C and S of M I = Hilb 2m+2 (P 3 ).First introduce notation for the loci that are destabilized by the two wall crossings according to the above classification: Thus the locus (II)(i) is (C \ E) ∪ S and the locus (III)(i) is S \ F .On the other hand both loci (II)(ii) and (III)(ii) are parametrized by the incidence variety I ⊂ P 3 × P3 consisting of pairs (P, V ) of points P ∈ P 3 inside a plane V ∈ P3 .The process of replacing E and F by I can be realized as contractions of algebraic spaces: E and F may be viewed as projective bundles over I, and in Section 4 we apply Artin's contractibility criterion to obtain smooth algebraic spaces C ′ and S ′ each containing the incidence variety I as a closed subspace, and birational morphisms φ : C → C ′ ψ : S → S ′ which are isomorphisms outside of E, respectively F , and restrict to the natural maps E → I, respectively F → I. Moreover E ⊂ C is disjoint from S, so the union C ′ ∪ S makes sense as the gluing together of (C \ E) ∪ S and C ′ .We can then state our second main result: To prove the theorem it suffices to treat the contracted spaces as algebraic spaces.However, they turn out to be projective varieties: the contractions are in fact Knegative extremal contractions in the sense of Mori theory.The case of S → S ′ can be found in previous work by Chen-Coskun-Nollet [8] and in fact it turns out that S ′ is a Grassmannian; see Section 4.2.Inspired by this work, we exhibit in Section 5 the map C → C ′ as a K-negative extremal contraction.This may be contrasted with Schmidt's approach in the twisted cubic situation [20], where projectivity of the moduli spaces is proved by viewing them as moduli of quiver representations.
In Section 2, we list the background results that we need, in particular, we briefly recall the construction of stability conditions on threefolds, along with the notion of tilt-stability.In Section 3 we apply Schmidt's machinery to prove Theorem 1.1.In Section 4 we study universal families and prove Theorem 1.3.Finally, in Section 5 we work out the Mori cone of C.
We work over C. Throughout and in particular in Section 4, intersections and unions of subschemes are defined by the sum and intersection of ideals, respectively, and inclusions and equalities between subschemes are meant in the scheme theoretic sense.The relative ideal of an inclusion Z ⊂ Y of two closed subschemes of some ambient scheme is the ideal I Z/Y ⊂ O Y defining Z as a subscheme of Y .

Preliminaries
After detailing the two components of Hilb 2m+2 (P 3 ), we collect notions and results from the literature surrounding Bridgeland stability and wall crossings for smooth projective threefolds.There are no original results in this section.

The Hilbert scheme and its two components
It is known that Hilb 2m+2 (P 3 ) has two smooth components C and S, whose general points are conics union a point and pairs of skew lines, respectively.A quick parameter count yields dim C = 11 and dim S = 8.We refer to Lee [13] for an overview, to Chen-Nollet [9] for the smoothness of C and to Chen-Coskun-Nollet [8] for the smoothness of S. In fact, the referenced works show that C is the blowup along the universal conic Z ⊂ P 3 × Hilb 2m+1 (P 3 ) and S is the blowup along the diagonal in the symmetric square of the Grassmannian G(2, 4) of lines in P 3 .In other words, it is the Hilbert scheme Hilb 2 (G(2, 4)) of finite subschemes in G(2, 4) of length two.
We pause to make some preparations regarding embedded points: by a curve C ⊂ P 3 with an embedded point at P ∈ C we mean a subscheme Y ⊂ P 3 such that C ⊂ Y and the relative ideal I C/Y is isomorphic to k(P ).This makes sense even when we allow C to be singular or nonreduced.Embedded points are in bijection with normal directions, i.e. lines in the normal space to C ⊂ P 3 at P : explicitly in our situation, suppose that P = (0, 0, 0) in local affine coordinates x, y, z and C is a conic defined by the ideal I C = (q(x, y), z), where q is a quadric vanishing at the origin.A normal vector to C may be viewed as a k-linear map given by say a = φ(q) and b = φ(z).Thus (a : b) ∈ P 1 parametrizes normal directions and the corresponding scheme Y with an embedded point at P is defined by the kernel of the induced map I C → k, which is Example 2.1.With notation as above, consider the degenerate conic defined by q(x, y) = xy.When (a : b) = (1 : 0) we obtain the planar scheme Y in the xy-plane given by where the final form exhibits Y as the scheme theoretic union of a pair of lines and a thickening of the origin in the xy-plane.At the other extreme (a : b) = (0 : 1) we obtain I Y = (zy, zx, yz, z 2 ) = (xy, z) ∩ (x, y, z) 2   which we label spatial: the subscheme Y ⊂ P 3 is not contained in any nonsingular surface since it contains a full first order infinitesimal neighbourhood around P .It is easy to check that for the remaining values of (a : b) the corresponding scheme Y is neither planar nor spatial.Analogous observations hold for the double line q(x, y) = y 2 .
With these preparations, we next list all elements Y ∈ Hilb 2m+2 (P 3 ), including degenerate cases.We follow Lee [13,Section 3.5], to which we refer for further details and proof that the following list is exhaustive.
The component C parametrizes subschemes Y of the following form: let C be a conic in a plane V ⊂ P 3 , possibly a union of two lines or a planar double line.Then Y is either the disjoint union of C and a point P ∈ P 3 , or C with an embedded point at P ∈ C. If C is nonsingular at P , embedded points correspond to normal directions, parametrized by a P 1 .Since even degenerate conics are complete intersections, also embedded point structures at a singular or nonreduced point P form a P 1 as in Example 2.1, and among these exactly one is planar (Y contained in a plane) and exactly one is spatial (Y contains the first order infinitesimal neighbourhood of P in P 3 ).
The component S parametrizes pairs Y = L 1 ∪ L 2 of skew lines, together with its degenerations.These are of the following three types: (1) a pair of incident lines L 1 ∪L 2 with a spatial embedded point at the intersection point, (2) a planar double line with a spatial embedded point, or (3) a double line in a quadric surface, i.e. there is a line L in a nonsingular quadric surface Q such that Y is the effective divisor 2L ⊂ Q, but viewed as a subscheme of P 3 .We label this case the pure double line, where purity refers to the lack of embedded components.
Clearly, then, C ∩ S consists of the incident or planar double lines with a spatial embedded point.

Stability conditions and walls
Let X be a smooth projective threefold over C and fix a finite rank lattice Λ equipped with a homomorphism K(X) → Λ from the Grothendieck group of coherent sheaves modulo short exact sequences.On P 3 we will take Λ = Z ⊕ Z ⊕ 1 2 Z ⊕ 1 6 Z equipped with the Chern character map ch : K(P 3 ) → Λ.
Recall [6,5,4] that a Bridgeland stability condition σ = (A, Z) on X (with respect to Λ) consists of (i) an abelian subcategory A ⊂ D b (X), which is the heart of a bounded t-structure, and (ii) a stability function Z, which is a group homomorphism whose value on any nonzero object E ∈ A is in the upper half plane This is subject to a list of axioms which we will not give (see [4,Section 8]).We may then partially order the nonzero objects in A by their slope This yields a notion of σ-stability and σ-semistability for objects in A in the usual way by comparing the slope of an object with that of its sub-or quotient objects.These notions extend to D b (X) by shifting in the sense of the t-structure.
Bridgeland's result [6,Theorem 1.2] gives the set Stab Λ (X) of stability conditions the structure of a complex manifold, and for a given u ∈ Λ, it admits a wall and chamber structure: if an object E goes from being stable to being unstable as the stability condition σ varies, σ has to pass through a stability condition for which E is strictly semistable.This observation leads to the definition of a wall: i. Numerical walls: A numerical wall is a nontrivial proper solution set ii. Actual walls: Let W v,w be a numerical wall, defined by classes v, w satisfying When the context is clear, we drop the word "actual" and just say "wall".
Given a union of walls, we refer to each connected component of its complement in Stab Λ (X) as a chamber.By the arguments in [7, Section 9] there is a locally finite collection of (actual) walls in Stab Λ (X), each being a closed codimension one manifold with boundary, such that the set of stable objects in A with class u ∈ Λ remains constant within each chamber, and there are no strictly semistable objects in a chamber.
We say that a short exact sequence as in (ii) above defines the wall.Relaxing this, an unordered pair {F, G} defines the wall if there is a short exact sequence in either direction (i.e.we allow the roles of sub and quotient objects to be swapped) as in (ii).Semistability of F and G is automatic, i.e. it follows from semistability of E and the equality between slopes, but see Remark 2.4.
Remark 2.3.A very weak stability condition (A, Z) is a weakening of the above concept (see Piyaratne-Toda [18]) where Z is allowed to map nonzero objects in A to zero.One may define an associated slope function λ as before, with the convention that λ(E) = +∞ also when Z(E) = 0.An object E ∈ A is declared to be stable if every nontrivial subobject F E satisfies λ(F) < λ(E/F), and semistable when nonstrict inequality is allowed.With this definition one avoids the need to treat cases where Z(F) = 0 or Z(E/F) = 0 separately.We will not need to go into detail.

Construction of stability conditions on threefolds
We next recall the "double tilt" construction of stability conditions by Bayer-Macrì-Toda [5].For this it is necessary to assume that the threefold X satisfies a certain "Bogomolov inequality" type condition [4, Conjecture 4.1]), which is known in several cases including P 3 [14].Fix a polarization H on X; on P 3 this will be a (hyper)plane.

Slope stability
Let β ∈ R. The twisted Chern character of a sheaf or a complex E on X is defined by ch β (E) = e −βH ch(E).Its homogeneous components are The twisted slope stability function on the abelian category Coh(X) of coherent sheaves is given by This is the slope of a very weak stability condition.Notice that µ β (E) = µ(E)−β, where µ(E) is the classical slope stability function.A sheaf E ∈ Coh(X) which is (semi)stable with respect to this very weak stability condition is called µ β -(semi)stable (or slope (semi)stable).

Tilt stability
Next, define the following full subcategories of Coh(X) The pair (T β , T ⊥ β ) is a torsion pair ([7, Definition 3.2]) in Coh(X).Tilt the category Coh(X) with respect to this torsion pair and denote the obtained heart by Coh The associated slope function is (see [15,Section 9.1] for more details).By [4, Proposition B.2 (case B = βH)], the pair (Coh β (X), Z tilt α,β ) is a very weak stability condition continuously parametrized by (α, β) ∈ R >0 × R.An object E ∈ Coh β (X) which is (semi)stable with respect to this very weak stability condition is called ν α,β -(semi)stable (or tilt (semi)stable).Moreover the parameter space R >0 × R admits a wall and chamber structure [4,Proposition B.5], in which walls are nested semicircles centered on the β-axis, or vertical lines (we view α as the vertical axis) [20,Theorem 3.3], and a numerical wall is either an actual wall everywhere or nowhere.We refer to them as "tilt-stability walls" or "ν-walls" interchangeably.
Remark 2.4.Walls in the parameter space for tilt stability are defined analogously to Definition 2.2.With tilt-stability in mind we make the following observation: let E be strictly semistable with respect to a very weak stability condition (A, Z).By definition this means that E is semistable and there exists a short exact sequence 0 → F → E → G → 0 in A such that all three objects share the same slope.This implies that F is semistable, since any destabilizing subobject of F would also destabilize E. When very weak stability conditions are allowed, however, G may not be semistable: this happens exactly when G has finite slope and there is a nontrivial subobject G ′ ⊂ G such that Z(G ′ ) = 0.In this case let G ′ ⊂ G be the maximal such subobject: then G/G ′ is semistable (it is in fact the final factor in the Harder-Narasimhan filtration of G) and has the same slope as G. Moreover the kernel F ′ of the composite E → G → G/G ′ has the same slope as F. Thus in the short exact sequence 0 → F ′ → E → G/G ′ → 0 all objects are semistable and of the same slope.So when looking for walls we may as in Definition 2.2 assume all objects in the defining short exact sequence to be semistable, even in the very weak situation.
The following is the Bogomolov inequality for tilt-stability:

Comparison between ν-stability and λ-stability -after Schmidt
Let E be an object in D b (X).Throughout this section, let (α 0 , β 0 ) ∈ R >0 × R satisfy ν α0,β0 (E) = 0, and fix s > 0. We shall summarize a series of results by Schmidt [20] enabling us to compare walls and chambers with respect to ν-stability with those of λstability.(Looking ahead to our application illustrated in Figures 1 and 2, the dashed hyperbola is the solution set to ν α,β (E) = 0.) Consider the following conditions on E: (1) E is a ν α0,β0 -stable object in Coh β0 (X).
For the proof we refer to Schmidt [20]: the first implication follows from Lemma 6.2 in loc.cit.and the second follows from Lemmas 6.3 and 6.4.
Schmidt furthermore compares walls for ν-stability and λ-stability, for objects E in some fixed class v ∈ Λ.Let (2.5) be a triangle in D b (X) with E in class v.
To control how the set of stable objects changes as a λ-wall is crossed, we take advantage of the fact that the λ-walls we obtain are defined by short exact sequences with stable sub-and quotient objects (in other words, only two Jordan-Hölder factors on the wall) and apply: Proposition 2.9.Suppose F and G are λ α,β,s -stable objects in A α,β (X).Then there is a neighborhood U of (α, β) such that for all (α ′ , β ′ ) ∈ U and all nonsplit extensions This result is stated and proved (for arbitrary Bridgeland stability conditions) in Schmidt [20,Lemma 3.11], and credited there also to Bayer-Macrì [3, Lemma 5.9].

Wall and chamber structure
The starting point for the entire discussion that follows is a simple minded observation.Namely, let V ⊂ P 3 be a plane and let Y be the union of a conic in V and a point P also in V .Then there is a short exact sequence (read O P 3 (−1) as the ideal of V and I P/V (−2) as the relative ideal of Y ⊂ V ).If we instead let Y be the union of a conic in V and a point P outside of V then there is a short exact sequence 0 (read I P (−1) as the ideal of {P } ∪ V and O V (−2) as the relative ideal of a conic in V ).The claim is that in a certain region of the stability manifold of P 3 , there are exactly two walls with respect to the Chern character ch(I Y ) = (1, 0, −2, 2), and they are defined precisely by the two pairs of sub and quotient objects appearing in the short exact sequences (3.1) and (3.2).Mimicking Schmidt's work for twisted cubics (and their deformations), we argue via tilt stability.Since ν α,β -stability only involves Chern classes of codimension at least one, and the above two short exact sequences are indistinguishable in codimension one, they give rise to one and the same wall in the tilt stability parameter space.Making this precise is the content of Section 3.1.Moving on to λ α,β,s -stability, we apply Schmidt's method to see that the single ν α,β -wall "sprouts" two distinct λ α,β,swalls corresponding to (3.1) and (3.2).This is carried out in Section 3.2.
The unique semicircular wall W (solid) for ν-stability and the hyperbola (dashed) from Equation 3.3.
For (α, β) ∈ R >0 × R we have the tilted abelian category Coh β (P 3 ) and the slope function ν α,β .We concentrate on the region β < 0, in which any ideal I Y of a subscheme Y ⊂ P 3 of dimension ≤ 1 satisfies As We begin by establishing that there is exactly one tilt-stability wall in the region β < 0. The result as well as the argument is analogous to the analysis for twisted cubics by Schmidt [20,Theorem 5.3], except that twisted cubics come with a second wall that destabilizes all objects -for our skew lines there is no such final wall.Proposition 3.1.There is exactly one tilt-stability wall for objects with Chern character v = (1, 0, −2, 2) in the region β < 0: it is the semicircle The wall is defined by exactly the unordered pairs of the following two types: (1) {I P (−1), O V (−2)}, where V ⊂ P 3 is a plane and P ∈ V , and (2) O P 3 (−1), I P/V (−2) , where V ⊂ P 3 is a plane and P ∈ V .
Moreover, the four sheaves figuring in the above unordered pairs are ν α,β -stable objects in Coh β (P 3 ) for all (α, β) on W .
We first prove the final claim in the proposition.Here is a slightly more general statement: Lemma 3.2.
(2) Let V ⊂ P 3 be a plane and Z ⊂ V be a finite, possibly empty subscheme.Then the relative ideal sheaf Remark 3.3.The condition on (α, β) in part ( 2) is necessary because of a wall for I Z/V (−2).For simplicity let Z be empty.There is a short exact sequence of coherent sheaves Proof of Lemma 3.2.The sheaf I Z (−1) is µ-stable and satisfies µ β (I Z (−1)) = −1 − β.For all β < −1 it is thus an object in T β and so also in Coh β (P 3 ).Since I Z/V (−2) is a torsion sheaf it too belongs to T β and so to Coh β (P 3 ), for all β.
We reduce to the situation Z = ∅.First consider I Z (−1) and assume β < −1.Note that I Z (−1) is a subobject of O P 3 (−1) also in Coh β (P 3 ) since the torsion sheaf O Z belongs to that category and hence be a proper nonzero subobject in Coh β (P 3 ) with quotient G. View F also as a subobject of O P 3 (−1), with quotient G ′ .Then ν α,β cannot distinguish between G and G ′ .Thus if and so I Z (−1) is ν α,β -stable as well.The reduction from The main task is to establish ν α,β -stability of O V (−2) in the region defined in part (2).By point (3) of [20,Theorem 3.3], the ray β = − 5  2 intersects all potential semicircular ν-walls for ch(O V (−2)) at their top point, meaning they must be centered at (0, − 5  2 ).All such semicircles of radius bigger than 1 2 will intersect the ray β = −2 (as well as β = −3).Thus it suffices to prove that O V (−2) is ν α,β -stable for all α > 0 and all integers β.
Let r F = H 3 ch 0 (F) and c F = H 2 ch 1 (F), i.e. the rank and first Chern class considered as integers.Also let r G = H 3 ch 0 (G) and c G = H 2 ch 1 (G).By the short exact sequence we have The induced long exact cohomology sequence of sheaves shows that H −1 (F) = 0, so from the short exact sequence The map into F cannot be an isomorphism, since β and F is in T β and is nonzero by assumption.Therefore the map in the middle is nonzero and so the rightmost sheaf H 0 (G) is a proper quotient of O V (−2) and so is a torsion sheaf supported in dimension ≤ 1.Thus only H −1 (G) contributes to r G and c G .
Suppose r F = 0.As F ∈ T β and If on the other hand r F = 0 then also H −1 (G) has rank zero and hence must be zero as there are no torsion sheaves in T ⊥ β .Thus also G = H 0 (G) is a sheaf, with vanishing rank and first Chern class.Again ch β 1 (G) = 0 as claimed.This completes the proof.
By explicit computation (see [20,Theorem 3.3]), all numerical tilt walls with respect to v = (1, 0, −2, 2) in the region β < 0 are nested semicircles.More precisely, each is centered on the axis α = 0 and has top point on the curve ν α,β (v) = 0, that is the hyperbola In particular every tilt wall must intersect the ray β = −2.
We establish in the following lemma that there is at most one tilt stability wall intersecting the ray β = −2 for Chern character v and β < 0. We also give the possible Chern characters of sub-and quotient objects that define it.This lemma is tightly analogous to Schmidt [20,Lemma 5.5].We use an asterisk * to denote an unspecified numerical value.Lemma 3.4.Let β 0 = −2 and let α > 0 be arbitrary.Suppose there is a short exact sequence or the other way around.
).Since the (very weak) stability function Z tilt sends effective classes to the upper half plane H ∪ {0} and and so the condition ν α,β0 (F) = ν α,β0 (G) says so this expression must be strictly positive.Suppose r ≥ 1 and apply the Bogomolov inequality (Proposition 2.5) to F: When r ≥ 1 this gives d ≤ 1 2 .On the other hand the positivity of (3.4) gives d > 0 and as d is a half integer this leaves only the possibility d = 1 2 and r = 1.Similarly suppose r ≤ 0 and apply the Bogomolov inequality to G: On the other hand the positivity of (3.4) gives d < 0 and as d is a half integer this leaves only the possibility d = − 1 2 and r = 0. Proof of Proposition 3.1.Assume there is a tilt stability wall for v = (1, 0, −2, 2), i.e. there is a short exact sequence As already pointed out, the same conditions then hold for some (α, β 0 ) with β 0 = −2.Then by Lemma 3.4, up to swapping F and G, we have Given any pair F, G of such objects, write out the condition ν α,β (F) = ν α,β (G) on (α, β) to obtain the equation for the wall in question; this yields the semicircle as claimed.
Thus we have proved that there is at most one tilt-wall and found its equation.
A further result of Schmidt [20,Lemma 5.4] (which requires β to be integral, and so applies for β 0 = −2) says that the only ν α,β0 -semistable objects in Coh β0 (P 3 ) with the invariants (3.5) and (3.6) are for a finite subscheme Z ⊂ P 3 , a plane V ⊂ P 3 and a finite subscheme Z ′ ⊂ V (where Z and Z ′ are allowed to be empty).Let n and n ′ denote the lengths of Z and Z ′ , respectively.Again for and so either Z is empty and Z ′ is a point, or Z is a point and Z ′ is empty.This proves that only the two listed pairs of semistable objects F, G may occur in a short exact sequence defining the wall.
To finish the proof it only remains to show that both pairs of objects listed do in fact realize the wall.By Lemma 3.2, the sheaves I Z (−1) and I Z ′ /V (−2) are in Coh β (P 3 ) and are ν α,β -semistable (in fact ν α,β -stable) for all (α, β) on the semicircle.Also, the ideal E = I Y of any Y ∈ Hilb 2m+2 (P 3 ) is an object in Coh β (P 3 ) (when β < 0) and since any ideal is µ-stable it is ν α,β -stable for α ≫ 0 (by Proposition 2.6).Hence it is ν α,β -stable outside the semicircle and at least ν α,β -semistable on the semicircle.Thus, short exact sequences of the types (3.1) and (3.2) define the wall and we are done.
Write P v ⊂ R >0 × R for the open subset defined by ν α,β (v) > 0 and β < 0; this is the region to the left of the hyperbola (3.3) in Figure 2. Theorem 2.8 addresses walls in P v close to the boundary hyperbola.Proposition 3.5.There are exactly two λ α,β,s -walls with respect to v = (1, 0, −2, 2) in P v whose closure intersect the hyperbola (3.3).They are defined exactly by the two pairs of objects listed in Proposition 3.1.
This means that the two walls are and and the pair of objects defining each wall (close to (α 0 , β 0 )) is unique.We refrain from writing out the (quartic) equations defining them.They do depend on s, but independently of s they both intersect the hyperbola (3.3) 2 ) and as we will show in the following proof, W 1 has negative slope at (α 0 , β 0 ) whereas W 2 has positive slope there.Thus W 1 lies above W 2 (α bigger) in the intersection between P v and a small open neighborhood of (α 0 , β 0 ).
Proof.We apply Schmidt's Theorem 2.8.Firstly, when ch(E) = v we have ch β > 0 so the theorem applies.The first part of the Theorem says that any λ-wall in P v , having a point (α 0 , β 0 ) with ν α0,β0 (v) = 0 in its closure, must be defined by one of the two pairs (F, G) listed in Proposition 3.1.This leaves W 1 and W 2 as the only candidates.Moreover the sub-and quotient objects F and G appearing are ν α0,β0 -stable by Lemma 3.2.Thus the second part of the theorem says that conversely, W 1 and W 2 are indeed λ-walls, provided they contain points (α, β) arbitrarily close to (α 0 , β 0 ) = ( 3 2 , − 5 2 ) such that ν α,β (F) > 0 and ν α,β (G) > 0. It remains to check this last condition.
Proof of Theorem 1.1 (II).Let E be λ α,β,s -stable for (α, β) in chamber II.Since semistability is a closed property, E is semistable on the wall W 1 .If E is stable on the wall, then it is also stable in chamber I hence it is an ideal sheaf in Hilb 2m+2 (P 3 ) by part (I).Such an ideal remains stable on the wall if and only if it is not an extension of the type (3.1), that is if and only if it is the ideal of a nonplanar subscheme.This is case (II)(i) in the Theorem.
If on the other hand E is stable in chamber II, but strictly semistable on W 1 , then by Proposition 3.5 it is a nonsplit extension of the pair and we determine the direction of the extension (which object is the subobject and which is the quotient) as follows: we claim that for all (α, β) in chamber II sufficiently close to (α 0 , β 0 ).Granted this, it follows that for E to be stable in chamber II it must be a nonsplit extension as in case (II)(ii) in the Theorem.Conversely it follows from Proposition 2.9 that every such nonsplit extension is indeed stable in chamber II.To verify (3.13) we let with Z = Z α,β,s , F = O P 3 (−1) and G = I P/V (−2).Thus W 1 is defined by Φ(α, β) = 0 and (3.13) is equivalent to Φ(α, β) < 0. It thus suffices to check that the partial derivative of Φ with respect to α is positive at (α 0 , β 0 ).An explicit computation yields in fact It remains only to show uniqueness of the nonsplit extensions But this space is H 1 (I P/V (−1)), which is isomorphic to H 0 (k(P )) = k via the short exact sequence Proof of Theorem 1.1 (III).Let E be λ α,β,s -stable for (α, β) in chamber III.Since semistability is a closed property, E is semistable on the wall W 2 .If E is stable on W 2 , then it is stable in chamber II.This means two things: first, by part (II) of the Theorem E is either an ideal sheaf of a nonplanar subscheme or a nonsplit extension F P,V as in case (II)(ii).Second, to remain stable on W 2 the object E cannot be in a short exact sequence of the type (3.2) ruling out ideal sheaves of plane conics union a point.Also the sheaves F P,V sit in short exact sequences of this type, as we show in Lemma 3.6 below (the vertical short exact sequence in the middle), and so are ruled out as well.
Hence E is an ideal sheaf of a disjoint pair of lines as claimed in (III)(i).
If on the other hand E is strictly semistable on W 2 , then by Proposition 3.5 E is a nonsplit extension (in either direction) of the pair for all (α, β) in chamber III sufficiently close to (α 0 , β 0 ).We prove this as in part II above, by partial differentiation of Φ defined in (3.14), this time with F = I P (−1) and As before we conclude that E is a nonsplit extension as in (III)(ii) and by Proposition 2.9 all such extensions are stable.It remains to verify uniqueness of the extensions G P,V , i.e.
dim Ext 1 P 3 (I P (−1), O V (−2)) = 1 when P ∈ V .For this first apply Hom(−, O V (−1)) to the short exact sequence to obtain a long exact sequence which together with the vanishing of H 1 (O V (−1)) and ) gives an isomorphism and ignoring twists, as these are not seen by k(P ), the right hand side is Serre dual to Ext 1 (O V , k(P )).This is one dimensional as is seen by applying Hom(−, k(P )) to the sequence

The special sheaves
Let F = F P,V and G = G P,V denote sheaves given by nonsplit extensions of the form (1.1) and (1.2), respectively.The definition through (unique) nonsplit extensions is indirect and it is useful to have alternative constructions available.We give such constructions here and compute the spaces of first order infinitesimal deformations.
Lemma 3.6.There is a commutative diagram with exact rows and columns as follows: 0 0 Proof.Up to identifying the skyscraper sheaf k(P ) with any of its twists, there are canonical short exact sequences as in the bottom row and the rightmost column.The diagram can then be completed by letting F be the fiber product as laid out by the square in the bottom right corner.It remains only to verify that the middle row is nonsplit.But if it were split the middle column twisted by O P 3 (1) would be a short exact sequence of the form 0 → I P → I P/V (−1) Taking global sections this yields a contradictory left exact sequence in which all terms vanish except H 0 (O P 3 ) = k.
Proof.We will actually only prove that the dimension is at most 11.The opposite inequality may be shown by similar techniques, although it follows from viewing Ext 1 (F, F) as a Zariski tangent space to the 11-dimensional moduli space M II studied in the next section.Apply Hom(−, F) to the middle row in the diagram in Lemma 3.6.This yields a long exact sequence and from the middle column of the diagram we compute H 1 (F(1)) = H 2 (F(1)) = 0. Thus we proceed to show that dim Ext 1 (I P/V (−2), F) ≤ 11.
Apply Hom(I P/V (−2), −) to the middle row in the diagram.This yields a long exact sequence: The space on the right is Serre dual to H 2 (I P/V (−5)) ∼ = H 2 (O V (−5)), which again on V is Serre dual to H 0 (O V (2)).This has dimension 6.At least heuristically, the space on the left should have dimension 5, as it may be viewed as a tangent space to the incidence variety I ⊂ P 3 × P3 seen as a moduli space for the sheaves I P/V .More directly we may apply Hom(I P/V , −) to the Koszul complex on where the indicated dimensions may be computed by applying Hom(I P/V (d), −) (for d = 1, 2) to the sequence We skip further details.It follows then that dim Ext 1 (I P/V , I P/V ) ≤ 5 and so dim Ext 1 (I P/V (−2), F) is at most 5 + 6 = 11.Lemma 3.8.There is a commutative diagram with exact rows and columns as follows: on V ∼ = P 2 it follows that Ω 1 V (2) has a unique section (up to scale) vanishing at P ∈ V .This leads to the short exact sequence in the bottom row.Moreover there is a canonical short exact sequence as in the rightmost column.The rest of the diagram can then be formed by taking G to be the fiber product as laid out by the bottom right square.It just remains to verify that the middle row is indeed nonsplit.But if it were split the middle column would be a short exact sequence of the form This sequence implies that H 1 (Ω 1 V (−1)) is isomorphic to H 1 (I P (−2)), which is one dimensional.But the Euler sequence shows that in fact H 1 (Ω 1 V (−1)) = 0.
Proof.We will be using the short exact sequence which sits as the middle column in Lemma 3.8.As preparation we observe that all (dimensions of) H i (Ω 1 V (d)) may be computed from the Euler sequence, and this enables us to compute several H i (G(d)) from (3.16).We use these results freely below without writing out further details.
Apply Hom(−, G) to (3.16) to produce a long exact sequence in which we have indicated some of the dimensions: H i (G(2)) are computed from (3.16) as sketched above, and since G is simple we have Hom(G, G) = k.For the same reason (3.16) is nonsplit, which implies Hom(Ω 1 V , G) = 0.It thus remains to see that the dimension of Ext 1 (Ω 1 V , G) is 11.Next apply Hom(−, G) to the Euler sequence.This gives a long exact sequence where again dimensions have been indicated: the vanishing of Hom(Ω 1 V , G) has already been noted, and there remain several spaces of the form Ext i (O V , G(d)).These may be computed from H i (G(d)) and the long exact sequence resulting from applying Hom(−, G(d)) to V , G) = 11 and we are done.

Moduli spaces and universal families
By the classification of stable objects in chamber II, the moduli space M II is at least obtained as a set from Hilb 2m+2 (P 3 ) = C ∪ S by just replacing the divisor E ⊂ C, parametrizing conics union a point inside a plane, with the incidence variety I, parametrizing just pairs (P, V ) of a point P inside a plane V .Similarly, the moduli space M III of stable objects in chamber III is obtained from M II set-theoretically by removing M II \ S and replacing the divisor F ⊂ S, parametrizing pairs of incident lines with a spatial embedded point at the intersection, with the incidence variety I.
We shall carry out each of these replacements as a contraction, i.e. a blow-down, and prove that this indeed yields M II and M III , essentially by writing down a universal family for each case.

The contraction C → C ′
Recall that C is isomorphic to the blow-up of P 3 × Hilb 2m+1 (P 3 ) along the universal curve Z, where Hilb 2m+1 (P 3 ) is the Hilbert scheme of plane conics in P 3 [13].The exceptional divisor E ′ is comprised of plane conics with an embedded point.
It is helpful to keep an eye at the following diagram where π sends a conic C ∈ Hilb 2m+1 (P 3 ) to the plane V ∈ P3 it spans and b is the blowup along the universal family Z of conics.
Remark 4.1.It will sometimes be useful to resort to explicit computation in local coordinates.For this let U ⊂ P3 be the affine open subset of planes V ⊂ P 3 with equation of the form Furthermore the P 5 of symmetric 3 × 3 matrices (s ij ) parametrizes plane conics 0≤i,j≤2 s ij x i x j = 0 (4.3) so that Hilb 2m+1 (P 3 ) U ∼ = P 5 × U with universal family defined by the two equations (4.2) and (4.3).This is also the center for the blowup b, and we note that it is nonsingular.
Here P(E ∨ ) denotes the projective bundle parametrizing lines in the fibers of E. Starting with the observation that the fiber of E over  Before giving the proof, we harvest our application: It is well known that the condition verified in Proposition 4.3 implies the contractibility of E/I in the above sense.In the category of analytic spaces this is the Moishezon [16] or Fujiki-Nakano [17,10] criterion.In the category of algebraic spaces the contractibility is due to Artin [2,Corollary 6.11], although the statement there lacks the identification with a blowup.Lascu [12, Théorème 1] however shows that once the contracted space C ′ as well as the image I ⊂ C ′ of the contracted locus are both smooth, it does follow that the contracting morphism is a blowup.Strictly speaking Lascu works in the category of varieties, but our C ′ turns out to be a variety anyway: Remark 4.5.The algebraic space C ′ is in fact a projective variety.We prove this in Section 5 using Mori theory.As the arguments there and in the present section are largely independent we separate the statements.
We also point out that the smooth contracted space C ′ is unique once it exists: in general, suppose φ : X → U and ψ : X → V are proper birational morphisms between irreducible separated algebraic spaces (say, of finite type over k) with U and V normal.Moreover assume that φ(x 1 ) = φ(x 2 ) if and only if ψ(x 1 ) = ψ(x 2 ).Let Γ denote the image of (φ, ψ) : X → U × V .Then each of the projections from Γ to U and V is birational and bijective and hence an isomorphism by Zariski's Main Theorem (for this in the language of algebraic spaces we refer to the Stacks Project [21, Tag 05W7]). is a P 5 -bundle, hence its restriction to id P 3 × π −1 (I) = E is a P 5 -bundle over I ⊂ P 3 × P3 .Now, E is the strict transform of E, i.e. its blow-up along Z ⊂ E. But this is a Cartier divisor, since E is smooth, and so E ∼ = E.This proves the first claim.
Again using that E is smooth, its strict transform E satisfies the linear equivalence The term b * (E) = b * (id P 3 × π) * (I) is a pullback from the base of the P 5 -bundle, so its restriction to any fiber is trivial.Thus it suffices to see that E ′ restricted to a fiber P 5 is a hyperplane.Now the isomorphism b : In the local coordinates from Remark 4.1 the divisor E is given by equation (4.2) and Z is given by the additional equation (4.3).Here (s ij ) are the coordinates on the fiber P 5 and clearly (4.3) defines a hyperplane in each fiber -it is the linear condition on the space of plane conics given by passage through a given point.
Remark 4.6.The locus C ∩ S consists of pairs of intersecting lines with a spatial embedded point at the intersection (and, as degenerate cases, planar double lines with a spatial embedded point).On the other hand E consists only of planar objects, so E is disjoint from S. Thus we may extend the contraction φ to a morphism between algebraic spaces which is an isomorphism away from E and restricts to the P 5 -bundle E → I as before.

Moduli in chamber II
In this section we shall modify the universal family on the Hilbert scheme Hilb 2m+2 (P 3 ) = C ∪ S in such a way that we replace its fibers over E ⊂ C with the objects F P,V in Theorem 1.1.This family induces a morphism and we conclude via uniqueness of normal (in this case smooth) contractions that M II coincides with C ′ ∪ S.
Here is the construction: let be the universal family and let be the E-flat family whose fiber V ξ ⊂ P 3 over a point ξ mapping to (P, V ) ∈ I is the plane V .Clearly V can be written down as a pullback of the universal plane over P3 .We view V as a closed subscheme of P 3 × Hilb 2m+2 (P 3 ).Then our modified universal family is the ideal sheaf I Y∪V .
Remark 4.7.We emphasize the (to us, at least) unusual situation that I Y∪V is the ideal of a very nonflat subscheme, yet as we show below it is flat as a coherent sheaf.
Its fibers over points in E are not ideals at all, but rather the objects F P,V .
Theorem 4.8.As above let Y be the universal family over Hilb 2m+2 (P 3 ) and V the family of planes in P 3 parametrized by E.
(i) I Y∪V is flat as a coherent sheaf over Hilb 2m+2 (P 3 ).Its fibers I Y∪V ⊗ k(ξ) over ξ ∈ Hilb 2m+2 (P 3 ) are stable objects for stability conditions in chamber II.(ii) The morphism Hilb 2m+2 (P 3 ) → M II determined by I Y∪V induces an isomorphism We begin by showing that the fibers I Y∪V ⊗ k(ξ) over ξ ∈ E sit in a short exact sequence of the type (1.1).The mechanism producing such a short exact sequence is quite general.Note that when Y = Y ξ is a conic with a (possibly embedded) point P in a plane V = V ξ , we have I Y /V ∼ = I P/V (−2) and I V ∼ = O P 3 (−1).Lemma 4.9.Let X be a projective scheme, Y ⊂ X × S an S-flat subscheme, E ⊂ S a Cartier divisor and V ⊂ E × S an E-flat subscheme such that Y ∩ (E × S) ⊂ V. Let ξ ∈ E. Then there is a short exact sequence In particular if S is integral in a neighborhood of E then I Y∪V is flat over S.
Proof.Observe that the last claim is implied by the first: outside of E the ideal I Y∪V agrees with I Y , which is flat.For ξ ∈ E we have Y ξ ⊂ V ξ and so a short exact sequence 0 → I Vξ → I Yξ → I Yξ/Vξ → 0.
The short exact sequence in the statement has the same sub and quotient objects in opposite roles, so the Hilbert polynomial of I Y∪V ⊗ k(ξ) agrees with that of I Yξ .Thus the Hilbert polynomial of the fibers of I Y∪V is constant; over an integral base this implies flatness.
Begin with the short exact sequence and tensor with O X×E to obtain the exact sequence The kernel of the rightmost map is clearly the ideal I V ⊂ O X×E .To compute the Tor-sheaf on the left use the short exact sequence Pull this back to X × S and tensor with O Y∪V to see that Tor X×S which locally is multiplication by an equation for E. Thus where J ⊂ O Y∪V is the ideal locally consisting of elements annihilated by an equation for E. We compute J in an open affine subset Spec A ⊂ X × S in which Y and V are given by ideals I Y and I V respectively and f ∈ A is (the pullback of) a local equation for E.
Moreover the latter equals I Y , since Y is flat over S, so that multiplication by the non-zero-divisor f remains injective after tensor product with O Y , that is A/I Y .Thus J is locally where we write Y E for the restriction Y ∩ (X × E) = Y ∩ V.This shows and (4.5) gives the short exact sequence Finally restrict to the fiber over a point ξ ∈ E: since Y E and V are both E-flat this yields the short exact sequence in the statement.
Lemma 4.9 does not guarantee that the short exact sequence obtained is nonsplit.Showing this in the case at hand requires some work.Our strategy is to exhibit a certain quotient sheaf I Y∪V ⊗ k(ξ) ։ Q and check that the split extension I Vξ ⊕ I Yξ/Vξ admits no surjection onto Q.In fact Q = O V (−2) will work: Lemma 4.10.Let V ⊂ P 3 be a plane and P ∈ V a point.Then there is no surjection from O P 3 (−1) Proof.Just note that Hom(I P/V (−2), O V (−2)) = k is generated by the (nonsurjective) inclusion, whereas Hom(O P 3 (−1), O V (−2)) = 0.
We will produce the required quotient sheaf by the following construction, which depends on the choice of a tangent direction at ξ in S: Lemma 4.11.With notation as in Lemma 4.9, let T = Spec k[t]/(t 2 ) and let T ⊂ S be a closed embedding such that T ∩E is the reduced point {ξ}.
Define a subscheme Y ′ ⊂ Y by the ideal Then there is a surjection Remark 4.12.Since t 2 = 0 we trivially have t ∈ (I : t).Since also I ⊂ (I : t) we furthermore have Y ′ ⊂ Y .If we extend T to an actual one parameter family of objects Y t , we may think of Y ′ as the limit of Y t ∩ V as t → 0, in other words it is the part of Y that remains in V as we deform along our chosen direction.
Assuming this, there are surjections Restriction to the fiber over ξ gives the surjection in the statement.
To prove the claim, we first observe that for any two subschemes A and B of some ambient scheme, there is an isomorphism between the relative ideal sheaves; this is the identity (I + J)/I ∼ = J/(I ∩ J) between quotients of ideals.Apply this to The claim as stated thus says that I Y ′ /V is isomorphic to I B/A∪B , and we are free to replace the latter by I (A∩B)/A .
Next let Spec R be an affine open subset in V and I ⊂ R[t]/(t 2 ) the ideal defining Y ∩ (V ×T ) there.Locally the ideal The left hand side is precisely Proof of Theorem 4.8 (i).By Lemma 4.9 it suffices to show that the fiber of I Y∪V ⊗ k(ξ) over ξ ∈ E is isomorphic to F P,V .Moreover, for such ξ, the same Lemma yields a short exact sequence and since F P,V is the unique such nonsplit extension it is enough to show that the above extension is nonsplit.In view of Lemma 4.10 this follows once we can show the existence of a surjection For this it suffices, in the notation of Lemma 4.11, to choose T ⊂ Hilb 2m+2 (P 3 ) such that the subscheme Y ′ ⊂ V is a conic.
Nondegenerate case.First assume Y = Y ξ is a disjoint union Y = C ∪ {P } of a conic C ⊂ V and a point P ∈ V .Consider the one parameter family Y t = C ∪ {P t } in which the conic part C is fixed while the point P t travels along a line intersecting V in the point P .In suitable affine coordinates we may take V to be the plane z = 0 in A 3 = Spec k[x, y, z], the point P to be the origin and C to be given by some quadric q = q(x, y) not vanishing at P .Let the one parameter family over Spec k[t] consist of the union of C with the point P t = (0, 0, t).This is given by the ideal (q, z) ∩ (x, y, z − t) = (xq, yq, (z − t)q, xz, yz, (z − t)z).Now restrict to T = Spec k[t]/(t 2 ) and intersect the family with V × T .The resulting subscheme is defined by the ideal I = (xq, yq, (z − t)q, xz, yz, (z − t)z) + (z) = (xq, yq, tq, z) and (I : t)/(t) = (q, z) which defines C ⊂ V .Thus Y ′ = C and we are done.
Embedded point with nonsingular support.Suppose Y is a conic C ⊂ V with an embedded point supported at a point P in which C is nonsingular, where the normal direction corresponding to the embedded point is along V .Then take the one parameter family in which C and the supporting point P is fixed and the embedded structure varies in the P 1 of normal directions.In suitable affine coordinates we may take V to be the plane z = 0 in A 3 = Spec k[x, y, z], P to be the origin and C given by a quadric q = q(x, y) vanishing at P and with, say, linear term y.Take the one parameter family of C with an embedded point given by (q, z) ∩ (x, y 2 , z − ty) = (xq, yq, zq, z − tq) (the equality requires some computation).After intersection with V × T this gives I = (xq, yq, tq, z) and (I : t)/(t) = (q, z).This is C.
Embedded point at a singularity.Let C ⊂ V be the union of two distinct lines intersecting in P and consider a planar embedded point at P .Despite the singularity, there is still a P 1 of embedded points at P .We take this to be our one parameter family, i.e. we deform the embedded point structure away from the planar one.
In local coordinates we take P to be the origin in A 3 and C to be the union of the x-and y-axes in the xy-plane V .Then is our one parameter family of embedded points at the origin, with t = 0 corresponding to the planar embedded point.The intersection with V × T is given by I = (xy 2 , x 2 y, z, txy) and (I : t)/(t) = (xy, z).This is C.
Embedded point in a double line.Let C ⊂ V be a planar double line together with a planar embedded point at P ∈ C and take the one parameter family of embedded points in P .
In local coordinates we take P to be the origin in A 3 and C to be V (z, y 2 ).Then is our one parameter family of embedded points at the origin, with t = 0 corresponding to the planar embedded point.The intersection with V × T is given by and (I : t)/(t) = (z, y 2 ).This is C.
Proof of Theorem 4.8 (ii).The morphism Hilb 2m+2 (P 3 ) → M II is clearly an isomorphism away from E, and it sends ξ ∈ E (lying over (P, V ) ∈ I) to F P,V , which determines and is uniquely determined by (P, V ).Moreover M II is smooth at these points by Proposition 3.7.The claim follows from uniqueness of normal contractions.

Moduli in chamber III
In this section we show that the moduli space M III is a contraction of S. The argument parallels that for M II closely.
Let F ⊂ S be as in Notation 1.2.Thus an element Y ∈ S is either a pair of intersecting lines with a spatial embedded point at the intersection, or as degenerate cases, a planar double line with a spatial embedded point.It is in a natural way a P 2 -bundle over the incidence variety I ⊂ P 3 × P3 via the map F → I that sends Y to the pair (P, V ) consisting of the support P ∈ Y of the embedded point and the plane V containing Y \ {P }.In parallel with Proposition 4.3 one may show that O S (F ) restricts to O P 2 (−1) in the fibers of F/I and so there is a contraction to a smooth algebraic space S ′ , such that ψ is an isomorphism away from F and restricts to the P 2 -bundle F → I.However, in this case we can be much more concrete thanks to the work of Chen-Coskun-Nollet [8], where birational models for S are studied in detail (and in greater generality: moduli spaces for pairs of codimension two linear subspaces of projective spaces in arbitrary dimension).The following proposition is [8, Theorem 1.6 ( 4)]; we sketch a simple and slightly different argument here.It is clearly an isomorphism away from the locus in G(2, n + 1) consisting of lines contained in Q.On the other hand, over every element of G(2, n + 1) defining a line contained in Q, the fiber is the P 2 consisting of length two subschemes of that line.
Apply the above observation to the (Plücker) quadric Q = G(2, 4) in P 5 , so that S ∼ = Hilb 2 (Q) (see Section 2.1).For every plane V ⊂ P 3 and every point P ∈ V , the pencil of lines in V through P defines a line in Q = G(2, 4) and in fact every line is of this form.The fiber of (4.7) above such an element of G(2, 5) consists of all pairs of lines in V intersecting at P .It follows that (4.7) is the required contraction S → S ′ .Remark 4.14.Chen-Coskun-Nollet furthermore shows that (4.6) is a K-negative extremal contraction in the sense of Mori theory.In fact, S is Fano and its Mori cone is spanned by two rays.Either ray is thus contractible; one contraction is (4.6) and the other is the natural map to the symmetric square of the Grassmannian of lines in P 3 .This statement is extracted from Theorem 1.3, Lemma 3.2 and Proposition 3.3 in loc.cit.Inspired by this work we return to the Mori cone of the conics-with-a-point component C in Section 5.
We proceed as for chamber II by modifying the universal family of pairs of lines in order to identify the moduli space M III with the contracted space S ′ .Let Y ⊂ P 3 × S be the restriction of the universal family over Hilb 2m+2 (P 3 ) to the component S.Moreover, there is a flat family over the incidence variety I ⊂ P 3 × P3 whose fiber over (P, V ) is the plane V with an embedded point at P .Pull this back to F to define a family We argue as in Section 4.2 but with the family of planes V replaced by the family of planes with an embedded point W. For ξ ∈ F lying over (P, V ) we have I Yξ/Wξ ∼ = O V (−2) and I Wξ ∼ = I P (−1).Thus Lemma 4.9 yields a short exact sequence 0 → O V (−2) → I Y∪W ⊗ k(ξ) → I P (−1) → 0 and we show that it is nonsplit by exhibiting a certain quotient sheaf of I Y∪W ⊗ k(ξ).This time we use I Q/V (−1) where Q ∈ V is a point distinct from P .Lemma 4.16.Let V ⊂ P 3 be a plane and P, Q ∈ V two distinct points.There is no surjection from O V (−2) ⊕ I P (−1) to I Q/V (−1).

Proof. Every nonzero homomorphism
has image of the form I L/V (−1) where L ⊂ V is a line, whereas every nonzero homomorphism has image I P/V (−1).Thus any nonzero homomorphism from the direct sum of these two sheaves has image I(−1) where I ⊂ O V is one of I L/V , I P/V or their sum Thus the image is never I Q/V (−1) for Q = P .
Proof of Theorem 4.15.The proof for Theorem 4.8 carries over; we only need to detail the construction of quotient sheaves via one parameter families.As before we write down families over A 1 = Spec k[t] and then restrict to T = Spec k[t]/(t 2 ).We then apply Lemma 4.11, with W in the role of the family denoted V in the Lemma.The outcome of Lemma 4.11 will be a quotient sheaf of the form I Y ′ /W , where W = W ξ is a plane V with an embedded point at P .We end by intersecting with V to produce a further quotient of the form I Y ′ ∩V /V .We shall choose one parameter families such that the latter is isomorphic to I Q/V (−1) with Q = P .Distinct lines.Let C = L ∪ L 0 be a pair of distinct lines inside V intersecting at P .Choose another plane V ′ containing L 0 and a point Q ∈ L 0 distinct from P .The pencil of lines L t ⊂ V ′ through Q yields a one parameter family of disjoint pairs of lines for t = 0, with flat limit Z 0 ⊂ W being C with a spatial embedded point at P .
In suitable affine coordinates A 3 let V be V (z), let P be the origin and let C = V (z, xy).Then W = V (xz, yz, z 2 ).Furthermore let Q = (0, 1, 0) and L t = V (x, z − t(y − 1)).This leads to the family Z defined by the ideal (y, z) ∩ (x, z − t(y − 1)) = (xy, xz, (z − t(y − 1))y, (z − t(y − 1))z) and the intersection with W × T is given by I = (xz, yz, z 2 , xy, ty(y − 1), tz) Thus (I : t)/(t) = (z, xy, y(y − 1)), which defines the union of the x-axis and the point Q.This is Y ′ ⊂ W and thus Double lines.Let C ⊂ V be a double line inside the plane V with P ∈ C. We shall define an explicit one parameter family with central fiber Y 0 ⊂ W being C with a spatial embedded point at P .
Geometrically, the family is this: let L ⊂ V be the supporting line of C. Consider a line M ⊂ V not through P and let Q be its intersection point with L. Also let M ′ be a line through P and not contained in V .Let R t be a point on M moving towards Q as t → 0, and let R ′ t be a point on M ′ moving towards P , but much faster than R t moves (quadratic versus linearly).Then let L t be the line through R t and R ′ t and let Y t = L ∪ L t for t = 0.
Reducing this modulo t gives the original Y 0 .The intersection with W × T gives I = (xz, yz, z 2 , (tx − y)y, ty(x − 1)) and so Y ′ ⊂ W is defined by (I : t)/(t) = (xz, yz, z 2 , y 2 , y(x − 1)) = (y, z) ∩ (x − 1, y 2 , z) ∩ (x, y, z 2 ).This is the line L with an embedded point at Q (inside V ) and another embedded point along the z-axis at P .Intersecting with V removes the embedded point at P , leaving the line L with an embedded point at Q.
This establishes part (i) precisely as in the proof of Theorem 4.8 and part (ii) then follows by smoothness of M III (from Proposition 3.9) and by uniqueness of normal (here smooth) contractions.

The Mori cone of C and extremal contractions
In this final section we shall prove that C → C ′ is the contraction of a K-negative extremal ray in the Mori cone.It follows that the contracted space C ′ is projective.
To set the stage we recall the basic mechanism of K-negative extremal contractions.Let X be a projective normal variety and α a curve class (modulo numerical equivalence) which spans an extremal ray in the Mori cone.If also the ray is K-negative, i.e. the intersection number between α and the canonical divisor K X is negative, then there exists a unique projective normal variety Y and a birational morphism f : X → Y which contracts precisely the effective curves in the class α.

Statement
We denote elements in C by the letter Y .It is the union of a (possibly degenerate) conic denoted C and a point denoted P .If the point is embedded, P ∈ C denotes its support.We also write V ⊂ P 3 for the unique plane containing C.
Define four effective curve classes (modulo numerical equivalence) on C. Each is described as a family {Y t }, and we use a subscript t to indicate a parameter on the piece that varies (all choices are to be made general, e.g.C nonsingular unless stated otherwise, etc.): δ: fix a conic C and a point P ∈ C. Let Y t be C with an embedded point at P , varying in the P 1 of normal directions to C ⊂ P 3 at P .ǫ: fix a plane V , a conic C ⊂ V and a line L ⊂ V .Let P t vary along L and let η: fix a line L and a point P ∈ L. Let V t be the pencil of planes containing L and let C t be the planar double structure on L inside V t .Then let Y t be C t with an embedded spatial point at P .
In ǫ there are implicitly two elements with an embedded point, namely where L intersects C. Similarly there is one element in ζ with an embedded point, corresponding to the pencil member C t that contains P .The theorem furthermore reveals a third K-negative extremal ray spanned by ǫ.The corresponding contraction has the effect of forgetting the point part of Y ⊂ V , keeping only the conic; thus the contracted locus in C is the same as for ζ, but the contraction happens in a "different direction".We do not know if the contracted space has an interpretation as a moduli space for Bridgeland stable objects.

The canonical divisor
Use notation as in Diagram 4.1 and Lemma 4.2.We read off that the Picard group of C has rank 4 and is generated by the pullbacks of the following divisor classes: the exceptional divisor for the blowup b.
Moreover numerical and linear equivalence of divisors coincide on C.Here we only use that the Picard group of a projective bundle over some variety X is Pic(X) ⊕ Z, with the added summand generated by O(1), and the Picard group of a blowup of X is Pic(X) ⊕ Z, with the added summand generated by the exceptional divisor.As long as confusion seems unlikely to occur we will continue to use the symbols H, H ′ and A for their pullbacks to C, or to an intermediate variety such as P 3 × Hilb 2m+2 (P 3 ) in Diagram 4.1.
Lemma 5.4.The canonical divisor class of C is Proof.This is a standard computation.First, for the blowup b, with center of codimension two, we have and for the product Now K P 3 = −4H and for the projective bundle Hilb 2m+1 (P 3 ) ∼ = P(E ∨ ) we have Again K P3 = −4H ′ and the short exact sequence Putting this together, the stated expression for K C follows once we have established that c 1 (E) = 4H ′ .
Recall that E = pr 2 * ( pr * 1 O P 3 (2)| I ).We compute its first Chern class by brute force: apply Grothendieck-Riemann-Roch to pr 2 : P 3 × P3 → P3 .Note that all higher direct images vanish, since H p (V, O V (2)) = 0 for all V ∈ P3 and p > 0. Thus by Grothendieck-Riemann-Roch the class is the push forward in the sense of the Chow ring of the degree 4 homogeneous part of ch( pr * 1 O P 3 (2)| I )pr * 1 (td(P 3 )).
We have Moreover I ⊂ P 3 × P3 is a divisor of bidegree (1, 1), so there is a short exact sequence from which we see (suppressing the explicit pullbacks pr * i of cycles in the notation) Now multiply together (5.1) and (5.2) and observe that the H 3 H ′ -coefficient is 4. Since the push forward pr 2 * of any degree 4 monomial H k H ′4−k equals H ′ if k = 3 and 0 otherwise, this shows that c 1 (E) = 4H ′ .

Basis for 1-cycles
We will need a few more effective curves, as before written as families {Y t }: α : fix a conic C and a line L. Let the point P t vary along L and let Y t = C ∪ {P t }. β : fix a quadric surface Q ⊂ P 3 , a line L and a point P .Let V t run through the pencil of planes containing L and let C t = Q ∩ V t .Then take Y t = C t ∪ {P }. γ : fix a plane V and a point P .Let C t ⊂ V run through a pencil of conics and let As before all choices are general, so that in the definition of α, the line L is disjoint from C, etc.
Proof.We need to compute all the intersection numbers and verify that we get 0 or 1 as appropriate.Here it is sometimes useful to explicitly write out the pullbacks to C, e.g.writing b * (pr * 1 (H)) rather than H.We view α, β, γ, δ not just as equivalence classes, but as the effective curves defined above.Only the intersection numbers involving β require some real work, and we will save this for last.
Intersections with α: Since pr 1 * (b * (α))) is the line Similarly pr 2 * (b * (α)) = 0 shows that the intersections with H ′ and A vanish.Finally α has no elements with embedded points, so is disjoint from E ′ .
Intersections with γ: We have A•γ = 1 because γ is a line in a fiber of the projective bundle π, whereas A restricts to a hyperplane in every fiber.The remaining intersection numbers vanish as we can pick disjoint effective representatives.
Intersections with δ: We have E ′ • δ = −1 as δ is a fiber of the blowup b and E ′ is the exceptional divisor.The remaining divisors H, H ′ and A are all pullbacks, i. whose fiber over V ∈ P3 is exactly the restriction of σ to V .This is nowhere zero, so (5.3) is a rank 1 subbundle and it defines a section Since D is defined by a condition on C only, it is the preimage by pr 2 • b (see Diagram 4.1) of the similarly defined divisor in Hilb 2m+1 (P 3 ).Moreover D ′ and E are the strict transforms by b of the similarly defined divisors on P 3 × Hilb 2m+1 (P 3 ).
We will need to control elements of D ′ with an embedded point.
Lemma 5.6.Fix P 0 so that D ′ is defined as an effective divisor.Choose a plane V not containing P 0 , a possibly degenerate conic C ⊂ V and a point P ∈ C. Then there is a unique Y ∈ D ′ with conic part C and an embedded point at P .More precisely: (1) If C is nonsingular at P then the embedded point structure is uniquely determined by the normal direction given by the line through P 0 and P .(2) If C is a pair of lines intersecting at P or a double line, then the embedded point is the spatial one, i.e. the scheme theoretic union of C and the first order infinitesimal neighborhood of P in P 3 .
Proof.Let Q be the cone over C with vertex P 0 .This is a quadratic cone in the usual sense when C is nonsingular, otherwise Q is either a pair of planes or a double plane.
A disjoint union C ∪ {P ′ } with P ′ = P 0 is clearly in D ′ if and only if it is a subscheme of Q.
On the one hand this shows that the subschemes Y listed in (1) and ( 2) are indeed in D ′ , since they are obtained from C ∪ {P ′ } by letting P ′ approach P along the line joining P 0 and P .
On the other hand it follows that if Y ∈ D ′ then Y ⊂ Q, since the latter is a closed condition on Y .In case (1) Q is nonsingular at P and so there is a unique embedded point structure at P ∈ C which is contained in Q.In case (2) the following explicit computation gives the result: suppose in local affine coordinates that C is the pair of lines V (xy, z), the "vertex" P 0 is on the z-axis and P is the origin.Then Q is the pair of planes V (xy).Any C with an embedded point at P has ideal of the form (xy, z)(x, y, z) + (sxy + tz) for (s : t) ∈ P 1 .This contains the defining equation xy of Q if and only if t = 0, which defines the spatial embedded point.The case where C is double line V (x 2 , z) is similar.
Lemma 5.7.We have Proof.The last equality was essentially established in the proof of Proposition 4.3: it follows from the observations (1) E is the strict transform of b(E), and (2) the latter is the pullback of the incidence variety I ⊂ P 3 × P3 which is linearly equivalent to H + H ′ .
The remaining two identities are verified by computing the intersection numbers with the curves in the basis from Lemma 5.5.All curves and divisors involved are concretely defined and it is easy to find and count the intersections directly.Some care is needed to rule out intersection multiplicities, and we often find it most efficient to resort to a computation in local coordinates.We limit ourselves to writing out only two cases.and pr 2 * (b * (β)) is the family of conics C t = V t ∩ Q where Q is a fixed quadric surface and V t runs through the pencil of planes containing a fixed line L.For general choices M ∩Q consists of two points, and each point spans together with L a plane.This yields exactly two planes V 0 and V 1 in the pencil for which C 0 = V 0 ∩ Q and C 1 = V 1 ∩ Q intersects M .It remains to rule out multiplicities.
In the local coordinates in Remark 4.1 let M = V (x 0 , x 1 ).Then the intersection between M and the plane x 3 = c 0 x 0 + c 1 x 1 + c 2 x 2 is the point (0 : 0 : 1 : c 2 ).Now D is the condition that this point is on C, i.e. it satisfies equation (4.3); this gives that D is s 22 = 0. On the other hand, pr 2 * (b * (β)) is a one parameter family in which c i and s ij are functions of degree at most 2 in the parameter.To stay concrete, let Q be i x 2 i = 0 and let V t be x 3 = tx 2 .Substitute x 3 = tx 2 in the equation for Q to find C t = Q ∩ V t .This gives in particular s 22 = 1 + t 2 and so the intersection with D is indeed two distinct points, each of multiplicity 1.
The case D ′ • δ = 1: This is essentially Lemma 5.6, but to ascertain there is no intersection multiplicity to account for we argue differently.D ′ is the strict transform of the divisor b(D ′ ) ⊂ P 3 × Hilb 2m+1 (P 3 ), which contains the center of the blowup.Since b(D ′ ) is nonsingular (pick P 0 = (0 : 0 : 0 : 1) in the definition of D ′ , then in the local coordinates of Remark 4.1 it is simply given by the equation (4.3)) we have and we are done.
The remaining cases are either similar to these or easier.

Nef and Mori cones
It is clear that H, H ′ and D are base point free, hence nef.For instance, consider D: given Y ∈ C, choose a line M ⊂ P 3 disjoint from Y ⊂ P 3 .This defines an effective representative for D not containing Y .
Proof.We begin by narrowing down the base locus of D ′ .First consider an element Y ∈ C without embedded point, that is a disjoint union Y = C ∪ {P }.Then choose P 0 such that the line through P 0 and P is disjoint from C. This defines a representative for D ′ not containing Y , so Y is not in the base locus.Next let Y be a conic C with an embedded point at a point P ∈ C where C is nonsingular.The tangent to C at P together with the normal direction given by the embedded point determines a plane.Pick P 0 such that the line through P and P 0 defines a normal direction to C which is distinct from that defined by the embedded point.This determines a representative for D ′ which by Lemma 5.6(i) does not contain Y , so Y is not in the base locus.
The remaining possibility is that Y is either a pair of intersecting lines with an

Notation 1. 2 .
(a) Let E ⊂ C be the divisor consisting of all planar Y ∈ C. (b) Let F ⊂ S be the divisor consisting of all Y ∈ S having an embedded point.

Figure 2 .
Figure 2. The two walls W 1 and W 2 (solid) for λ-stability separating three chambers, together with the hyperbola (dashed) from Equation 3.3.
0, so base change in cohomology applies) the Lemma is straight forward and we refrain from writing out details.Now let E ⊂ C be the locus of planar Y ∈ C. The condition on a disjoint union Y = C ∪ {P } to be in E is just that P is in the plane V spanned by C. For a conic with an embedded point the condition Y ⊂ V also singles out the scheme structure at the embedded point.View E as a variety over the incidence variety I ⊂ P 3 × P3 via the morphism (id P 3 × π) • b (refer to Figure3for simple illustrations of the types of elements in E and E ′ .In fact, we find it helpful for keeping track of the various divisors introduced in this section.)Proposition 4.3.E is a Zariski locally trivial P 5 -bundle over the incidence variety I ⊂ P 3 × P3 .The restriction O C (E)| P 5 to a fiber is isomorphic to O P 5 (−1).

Figure 3 .
Figure3.A Circle represents a conic contained in a plane that is shown as a parallelogram, and a red dot is a point, possibly embedded in the conic.The arrow is the direction vector at an embedded point.Note that in the left illustration, the arrow is strictly contained in the plane.

Corollary 4 . 4 .
There exist a smooth algebraic space C ′ , a morphism φ : C → C ′ and a closed embedding I ⊂ C ′ , such that φ restricts to an isomorphism from C \ E to C ′ \ I and to the given projective bundle structure E → I. Moreover φ is the blowup of C ′ along I.

Proposition 4 . 13 .
There is a contraction as in (4.6)where S ′ is the Grassmannian G(2, 6) of lines in P 5 .Proof.First consider an arbitrary quadric Q ⊂ P n .Any finite subscheme in Q of length 2, reduced or not, determines a line in P n .This defines a morphism Hilb 2 (Q) → G(2, n + 1).(4.7)

Theorem 4 . 15 .
Let Y and W be as above.(i)I Y∪W is flat as a coherent sheaf over S. Its fibers I Y∪W ⊗ k(ξ) over ξ ∈ S are stable objects for stability conditions in chamber III.(ii) The morphism S → M III determined by I Y∪W induces an isomorphism S ′ ∼ = M III .

Theorem 5 . 1 .Remark 5 . 3 .
The Mori cone of C is the cone over a solid tetrahedron, with extremal rays spanned by the four curve classes δ, ǫ, ζ, η.Of these, the first three are K-negative, whereas η is K-positive.The contraction corresponding to ζ is C → C ′ .Corollary 5.2.C ′ is projective.The last claim in the theorem is clear: by contracting ζ we forget the conic part of Y ⊂ V , keeping only V and the point P ∈ V .By uniqueness of (normal) contractions the contracted variety is C ′ .Also, with reference to Diagram 4.1 (from Section 4.1), the contraction of δ is the blowing down b.