Derived equivalence for stratified Mukai flop on G ( 2 , 4 ) , , November 27, 2006

We consider a conjecture on the derived equivalence for

K

-equivalent varieties in the case of a certain flop between

9

-dimensional varieties. The following is a very general version of this conjecture on the fully faithful derived embedding for log

K

-related varieties ([11] Conjecture 2.2):

Conjecture 1.1. Let

(X, B)

and

(Y, C)

be pairs of quasi-projective varieties with

Q

-divisors such that there exist quasi-finite and surjective morphisms

π : U \to X

and

σ : V \to Y

from smooth varieties, which may be reducible, such that

π^{*} (K_{X} + B) = K_{U}

and

σ^{*} (K_{Y} + C) = K_{V}

. Let

X \to X

and

Y \to Y

be natural morphisms from the associated Deligne-Mumford stacks. Assume that there are proper birational morphisms

μ : W \to X

and

ν : W \to Y

from a third variety

W

such that

μ^{*} (K_{X} + B) \leq ν^{*} (K_{Y} + C)

. Then there exists a fully faithful exact functor

D^{b} (Coh (X)) \to D^{b} (Coh (Y))

The conjecture is proved to be true in some cases in [2] , [5] , [4] , [9] , [6] , [13] , [10] , [11] and [1] .

On the other hand, Namikawa [14] proved that a naturally defined functor between the derived categories for the stratified Mukai flop on

G (2, 4)

is not an equivalence. This is a flop between

9

-dimensional varieties which is the total space of a

3

-parameter degeneration of standard

3

-dimensional flops of

(- 1, - 1)

-curves and the fiber over the most degenerate point is isomorphic to

G (2, 4)

So it is worthwhile to check the conjecture in this special case. We shall prove that there is nevertheless another functor between the same categories which is an equivalence.

2 Stratified Mukai flop

We recall the construction of stratified Mukai flops due to Markman [12] (see also [14] ).

Let

G = G (r, n)

be the Grassmann variety of

r

-dimensional subspaces in an

n

-dimensional vector space

V

. We assume that

2 r \leq n

. Let

S

(resp.

Q

) be the tautological subbundle (resp. quotient) bundle on

G

0 \to S \to V \otimes O_{G} \to Q \to 0 .

The polarization of

G

is given by

det Q = O_{G} (1) .

We have

Ω_{G}^{1} \sim = ℋ o m (Q, S)

. Since

{Ext}^{1} (O_{G}, Ω_{G}^{1}) \sim = {Ext}^{1} (Q, S) \sim = C

there exists a non-trivial extension of vector bundles on

G

induced by the natural homomorphism

O_{G} \to ℋ o m (S, S)

0 - - - \to Ω_{G}^{1} - - - \to {\tilde{Ω}}_{G}^{1} - - - \to O_{G} - - - \to 0 ∥ ↓ ↓ 0 - - - \to ℋ o m (Q, S) - - - \to ℋ o m (V \otimes O_{G}, S) - - - \to ℋ o m (S, S) - - - \to 0 .

Let

X_{0} = T^{*} G

and

X

be the total spaces of

Ω_{G}^{1}

and

{\tilde{Ω}}_{G}^{1}

, respectively, and

π : X \to G

the projection. We write

π^{*} S = S_{X}

and

π^{*} O_{G} (1) = O_{X} (1)

. A point of

X

is given by a pair consisting of a point

p \in G

and a homomorphism

A : V \to S_{p} \subset V

which induces a homothety

t : S_{p} \to S_{p}

(

t \in C

). We have

(p, A) \in X_{0}

if and only if

t = 0

. Let

Z = {A \in End (V) | \exists t \in C s.t. A^{2} = t A, rank A \leq r}

with a morphism

t : Z \to C

, and set

Z_{0} = t^{- 1} (0)

. Then there is a natural morphism

φ : X \to Z

given by

φ (p, A) = A

, which induces a morphism

φ_{0} : X_{0} \to Z_{0}

. We know that

φ_{0}

and

φ

are projective birational morphisms.

We write

W^{(k)} = {(p, A) \in X | rank A \leq k} .

Then the exceptional loci of both

φ_{0}

and

φ

are equal to

W = W^{(r - 1)}

. We have

dim G = r (n - r)

dim X_{0} = 2 r (n - r)

dim X = 2 r (n - r) + 1

, and

dim W = 2 (r - 1) (n - r + 1) + n - 2 r + 1

. We have a dual construction starting from the dual vector space

V^{*}

. Let

G^{+}

be the Grassmann variety of

r

-dimensional subspaces in

V^{*}

S^{+}

(resp.

Q^{+}

) the tautological subbundle (resp. quotient bundle) on

G^{+}

det Q^{+} = O_{G^{+}} (1)

X_{0}^{+}

(resp.

X^{+}

) the total space of

Ω_{G^{+}}^{1}

(resp.

{\tilde{Ω}}_{G^{+}}^{1}

π^{+} : X^{+} \to G^{+}

the projection,

π^{+ *} S^{+} = S_{X^{+}}^{+}

, and

π^{+ *} O_{G^{+}} (1) = O_{X^{+}} (1)

. There is a natural projective birational morphism

φ^{+} : X^{+} \to Z

given by

φ^{+} (q, B) =^{t} B

which induces a projective birational morphism

φ_{0}^{+} : X_{0}^{+} \to Z_{0}

. We write

W^{+ (k)} = {(q, B) \in X^{+} | rank B \leq k} .

The exceptional locus of

φ

W^{+} = W^{+ (r - 1)}

. The diagram

X - - - \to φ Z \leftarrow - - - φ^{+} X^{+}

thus obtained is called a stratified Mukai flop. By restricting to the subspaces defined by

t = 0

, we have a smaller diagram

X_{0} - - - \to φ Z_{0} \leftarrow - - - φ^{+} X_{0}^{+}

which is also called a stratified Mukai flop.

r = 1

, then the above diagrams are reduced to a standard flop and a Mukai flop.

The birational map

(φ^{+})^{- 1} \circ φ : X - \to X^{+}

is decomposed into blow-ups and downs ([12] ). We consider only the case

r = 2

in the following. Let

f_{1} : X_{1} \to X

be the blowing up along the center

G = W^{(0)}

, the

0

-section of the projection

π

. Then the strict transform

W^{'}

W = W^{(1)}

is smooth.

Indeed, the projection

W^{'} \to G

is smooth and its fibers are isomorphic to the cone over

P^{1} \times P^{n - 3} \subset P^{2 n - 5}

. Let

f_{2} : X_{2} \to X_{1}

be the blowing up along the center

W^{'}

. Let

E_{i}

be the exceptional divisor of

f_{i}

for

i = 1, 2

, and

E_{1}^{'} = f_{2 *}^{- 1} E_{1}

, the strict transform of

E_{1}

. We set

f = f_{2} \circ f_{1}

. If

r \geq 3

, then we have similar construction with more blow-ups ([12] ).

On the dual side, let

f_{1}^{+} : X_{1}^{+} \to X^{+}

be the blowing up along the center

G^{+} = W^{+ (0)}

, the

0

-section of the projection

π^{+}

. Then the strict transform

(W^{+})^{'}

W^{+} = W^{+ (1)}

is smooth. Let

f_{2}^{+} : X_{2}^{+} \to X_{1}^{+}

be the blowing up along the center

(W^{+})^{'}

. Let

E_{i}^{+}

be the exceptional divisor of

f_{i}^{+}

for

i = 1, 2

, and

(E_{1}^{+})^{'} = (f_{2 *}^{+})^{- 1} E_{1}^{+}

. We set

f^{+} = f_{2}^{+} \circ f_{1}^{+}

The birational map

(φ^{+})^{- 1} \circ φ : X - \to X^{+}

induces an isomorphism

X_{2} \to X_{2}^{+}

([12] ). We write

Y = X_{2} = X_{2}^{+}

under this identification. Then we have

E_{1}^{'} = (E_{1}^{+})^{'}

E_{2} = E_{2}^{+}

, and a commutative diagram

A \in Z \leftarrow - - - φ X \leftarrow - - - f_{1} X_{1} \leftarrow - - - f_{2} X_{2} ∥ ∥ A^{+} \in Z^{+} \leftarrow - - - φ^{+} X^{+} \leftarrow - - - f_{1}^{+} X_{1}^{+} \leftarrow - - - f_{2}^{+} X_{2}^{+}

Since

dim X = 4 n - 7

dim G = 2 n - 4

and

dim W = 3 n - 5

, we have

K_{Y} = f^{*} K_{X} + (2 n - 4) E_{1}^{'} + (n - 3) E_{2}

The isomorphism

X_{2} \to X_{2}^{+}

can be described set theoretically in the following way.

(1) For a point

(p, A) \in X \ W

, we have

rank A = 2

and

S_{p} = Im A

. The corresponding point

(q,^{t} A) \in X^{+} \ W^{+}

is given by

S_{q}^{+} = {Im}^{t} A = (Ker A)^{⊥}

(2) For a point

{lim}_{ε \to 0} (p, ε A_{1}) \in E_{1} \ W^{'}

, with

rank A_{1} = 2

and

S_{p} = Im A_{1}

, the corresponding point is

{lim}_{ε \to 0} (q, ε^{t} A_{1}) \in E_{1}^{+} \ (W^{+})^{'}

, where

S_{q}^{+} = {Im}^{t} A_{1}

(3) For a point

(p, A) \in W \ G

, we have

rank A = 1

and

S_{p} \supset Im A

. We take

A_{2}

such that

S_{p} = Im (A + A_{2})

. Then a point

{lim}_{ε \to 0} (p, A + ε A_{2}) \in E_{2} \ E_{1}^{'}

corresponds to a point

{lim}_{ε \to 0} (q_{ε},^{t} (A + ε A_{2})) \in E_{2}^{+} \ (E_{1}^{+})^{'}

which is over a point

q \in G^{+}

given by

S_{q}^{+} = {lim}_{ε \to 0} {Im}^{t} (A + ε A_{2})

(4) For a point

{lim}_{ε \to 0} (p, ε A_{1}) \in E_{1} \cap W^{'}

with

rank A_{1} = 1

and

S_{p} \supset Im A_{1}

, we take

A_{2}

such that

S_{p} = Im (A_{1} + A_{2})

. Then a point

{lim}_{ε \to 0} (p, ε A_{1} + ε^{2} A_{2}) \in E_{1}^{'} \cap E_{2}

corresponds to a point

{lim}_{ε \to 0} (q_{ε},^{t} (ε A_{1} + ε^{2} A_{2})) \in (E_{1}^{+})^{'} \cap E_{2}^{+}

which is over a point

q \in G^{+}

given by

S_{q}^{+} = {lim}_{ε \to 0} Im (^{t} (A_{1} + ε A_{2}))

The above description shows that the birational map

X_{2} - \to X_{2}^{+}

induced by

(φ^{+})^{- 1} \circ φ

is an isomorphism by the Zariski main theorem. As a consequence, we have the following lemma:

Lemma 2.1.

S_{X}^{*} |_{X \ W} = S_{X^{+}}^{+} |_{X^{+} \ W^{+}} .

Moreover,

f^{*} S_{X}^{*}

is a locally free subsheaf of

f^{+ *} S_{X^{+}}^{+}

such that a local section of the latter belongs to the former if and only if its value in

S_{q}^{+}

at any point in

(f^{+})^{- 1} (x^{+})

for

x^{+} = (q, B)

is contained in

Im B

Proof. For $(p, A) \in X \ W$ , we have ${Ker}^{t} A = (Im A)^{⊥}$ from (1) above.
Therefore, $^{t} A : V^{*} \to V^{*}$ induces an isomorphism $S_{p}^{*} = V^{*} / {Ker}^{t} A \to S_{q}^{+}$ , hence the first assertion.
Any element $v \in V^{*}$ determines a global section of $f^{*} S_{X}^{*}$ , and the sheaf $f^{*} S_{X}^{*}$ is generated by such sections. The value at the point $(q,^{t} A) \in X^{+} \ W^{+}$ of the corresponding section of $S_{X^{+}}^{+}$ is given by $^{t} A v \subset S_{q}^{+}$ , hence the second assertion.

By taking the determinants, we obtain:

Corollary 2.2.

f^{*} O_{X} (1) = f^{+ *} O_{X^{+}} (- 1) \otimes O_{Y} (- 2 E_{1}^{'} - E_{2}) .

□

Let

P

be a fiber of

f_{1}

above a point in

G

P_{0} = P \cap f_{1 *}^{- 1} X_{0}

and

P_{1} = P \cap W^{'}

Then the sequence

P_{1} \subset P_{0} \subset P

is isomorphic to

P^{1} \times P^{n - 3} \subset P^{2 n - 5} \subset P^{2 n - 4}

, where the first inclusion is the Segre embedding and the second is linear. Let

P^{'} = f_{2 *}^{- 1} P

P_{0}^{'} = f_{2 *}^{- 1} P_{0}

, and

P_{1}^{'} = f_{2}^{- 1} (P_{1})

. Let

l_{1}

be a line on

P_{0}

which meets

W^{'}

2

points,

l_{1}^{'} = f_{2 *}^{- 1} l_{1}

, and

l_{2}

a fiber of

f_{2}

over a point in

W^{'}

Lemma 2.3. The cone of curves

\bar{N E} (Y / X)

is generated by the classes of

l_{1}^{'}

and

l_{2}

. The intersection numbers are given by the following table:

(E_{1}^{'} \cdot l_{1}^{'}) = - 1, (E_{1}^{'} \cdot l_{2}) = 0, (E_{2} \cdot l_{1}^{'}) = 2, (E_{2} \cdot l_{2}) = - 1 .

Proof. $P_{1}$ is a determinantal variety defined by quadratic equations. Therefore we obtain the formula for $\bar{N E} (Y / X)$ . The intersection numbers are obvious.

3 Derived equivalence

Let

X \to Z \leftarrow X^{+}

and

X_{0} \to Z_{0} \leftarrow X_{0}^{+}

be stratified flops defined in the previous section. We shall compare the bounded derived categories of coherent sheaves

D^{b} (Coh (X))

with

D^{b} (Coh (X^{+}))

and

D^{b} (Coh (X_{0}))

with

D^{b} (Coh (X_{0}^{+}))

. By [10] Lemma 5.6 and Corollary 5.7, the derived equivalence for the former pair implies the latter.

We begin with an easy case

r = 1

, i.e., the standard flop and the usual Mukai flop.

Proposition 3.1. Let

l

be an arbitrary integer. Then a functor

Φ_{l} : D^{b} (Coh (X)) \to D^{b} (Coh (X^{+}))

defined by

Φ_{l} (a) = f_{*}^{+} (f^{*} a \otimes O_{Y} (l E))

is an equivalence.

Proof. The adjoint functor

Ψ_{l} : D^{b} (Coh (X^{+})) \to D^{b} (Coh (X))

is given by

Ψ_{l} (b) = f_{*} (f^{+ *} a \otimes O_{Y} ((n - 1 - l) E)) .

The category

D^{b} (Coh (X))

is spanned by the set of sheaves

O_{X} (k)

for

l - n + 1 \leq k \leq l

. By the Kodaira vanishing theorem, we have

Φ_{l} : O_{X} (k) \mapsto O_{Y} (k, 0) (l E) = O_{Y} (0, - k) ((l - k) E) \mapsto O_{X^{+}} (- k)

and

\begin{matrix} Ψ_{l} : & O_{X^{+}} (- k) \mapsto O_{Y} (0, - k) ((n - 1 - l) E) = O_{Y} (k, 0) ((n - 1 - l + k) E) & \mapsto O_{X} (k) . \end{matrix}

There is an adjunction morphism of functors

F : {Id}_{D^{b} (Coh (X))} \to Ψ_{l} Φ_{l}

, which is reduced to the identity when restricted to the open subset

X \ W

. By the above argument, we have isomorphisms

ω \sim = Ψ Φ (ω)

for a spanning class

Ω = {ω}

. Since the

ω

are invertible sheaves, it follows that the morphisms

F (ω)

are isomorphisms. Therefore, the natural homomorphisms

Φ : {Hom}^{p} (ω_{1}, ω^{2}) \to {Hom}^{p} (ω_{1}, Ψ Φ (ω^{2})) \to {Hom}^{p} (Φ (ω_{1}), Φ (ω^{2}))

for any

ω_{1}, ω_{2} \in Ω

and

p \in Z

are isomorphisms. Then

Φ

is an equivalence by [2] and [3] .

We assume that

r = 2

in the rest of the paper. We consider exact functors between bounded derived categories

\begin{matrix} \begin{matrix} Φ : D^{b} (Coh (X)) \to D^{b} (Coh (Y)) & Ψ : D^{b} (Coh (Y)) \to D^{b} (Coh (X)) \end{matrix} \end{matrix}

(3.1)

defined by

\begin{matrix} Φ (a) = R f_{*}^{+} (L f^{*} (a) \otimes O_{Y} ((2 n - 5) E_{1}^{'} + (n - 3) E_{2})) & Ψ (b) = R f_{*} (L f^{+ *} (b) \otimes O_{Y} (E_{1}^{'})) . \end{matrix}

They are adjoints each other because

K_{Y} = f^{*} K_{X} + (2 n - 4) E_{1}^{'} + (n - 3) E_{2}

Lemma 3.2.

D^{b} (Coh (X))

is spanned by the set of the following locally free sheaves:

{Sym}^{i} S_{X}^{*} \otimes O_{X} (j)

for

0 \leq i

0 \leq j

and

i + j \leq n - 2

Proof. By [8] , any point sheaf $O_{p}$ for $p \in G$ has a finite locally free resolution whose terms are direct sums of the sheaves ${Sym}^{i} S^{*} \otimes O_{G} (j)$ for $0 \leq i$ , $0 \leq j$ and $i + j \leq n - 2$ . Hence $O_{π^{- 1} (p)}$ is resolved by our set. Then so is any point sheaf $O_{x}$ for $x \in X$ because $π^{- 1} (p)$ is an affine space.

Let

ℰ^{+}

be a subsheaf of

S_{X^{+}}^{+}

such that a local section of the latter belongs to the former if and only if its value in

S_{q}^{+}

at a point

(q, B)

is contained in

Im B

. We denote by

ℰ_{i}^{+}

for

i > 0

the image of the natural homomorphism

{Sym}^{i - 1} S_{X^{+}}^{+} \otimes ℰ^{+} \to {Sym}^{i} S_{X^{+}}^{+}

Lemma 3.3. (1) For

0 \leq j \leq n - 3

\begin{matrix} Φ (O_{X} (j)) = O_{X^{+}} (- j) & Ψ (O_{X^{+}} (- j)) = O_{X} (j) . \end{matrix}

(2) Let

I_{W^{+}}

be the ideal sheaf of

W^{+} \subset X^{+}

. Then

Φ (O_{X} (n - 2)) = I_{W^{+}} \otimes O_{X^{+}} (- n + 2) .

(3) For

0 < i

0 \leq j

and

i + j \leq n - 3

\begin{matrix} Φ ({Sym}^{i} S_{X}^{*} \otimes O_{X} (j)) = {Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (- j) & Ψ ({Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (- j)) = {Sym}^{i} S_{X}^{*} \otimes O_{X} (j) . \end{matrix}

(4) For

0 < i \leq n - 2

Φ ({Sym}^{i} S_{X}^{*} \otimes O_{X} (n - 2 - i)) = ℰ_{i}^{+} \otimes O_{X^{+}} (- n + 2 + i) .

Proof. (1) By Corollary 2.2

\begin{matrix} f^{*} (O_{X} (j)) \otimes O_{Y} ((2 n - 5) E_{1}^{'} + (n - 3) E_{2}) & = f^{+ *} O_{X^{+}} (- j) \otimes O_{Y} ((2 n - 5 - 2 j) E_{1}^{'} + (n - 3 - j) E_{2}) . \end{matrix}

Since

K_{Y} = f^{*} K_{X} + (2 n - 4) E_{1}^{'} + (n - 3) E_{2}

, the higher direct images for

f^{+}

vanish because

- (2 j + 1) E_{1}^{'} - j E_{2}

is nef. Since

(2 n - 5 - 2 j) E_{1}^{'} + (n - 3 - j) E_{2}

is effective, we have the first formula.

Similarly, we have

f^{+ *} (O_{X^{+}} (- j)) \otimes O_{Y} (E_{1}^{'}) = f^{*} O_{X} (j) \otimes O_{Y} ((2 j + 1) E_{1}^{'} + j E_{2}) .

The higher direct images for

f

vanish because

(2 j - 2 n + 5) E_{1}^{'} + (j - n + 3) E_{2}

is nef.

(2 j + 1) E_{1}^{'} + j E_{2}

is effective, hence the second formula.

(2) We already proved in (1) that the higher direct images vanish. We obtain our formula from

f_{*}^{+} O_{Y} (- E_{1}^{'} - E_{2}) = I_{W^{+}} .

(3) We have an exact sequence

0 \to O_{E_{2}} (- E_{1}^{'}) \to f^{+ *} S_{X^{+}}^{+} \otimes O_{E_{2}} (- E_{1}^{'}) \to O_{E_{2}} (- E_{1}^{'}) \otimes f^{+ *} O_{X^{+}} (- 1) \to 0 .

where the first term is the image of

f^{*} S_{X}^{*}

. We define a decreasing filtration of the sheaf

f^{*} {Sym}^{i} S_{X}^{*}

by locally free subsheaves

ℱ^{k, l}

for

0 \leq l \leq k \leq i

\begin{matrix} ℱ^{k, l} = & f^{*} {Sym}^{i} S_{X}^{*} \cap f^{+ *} {Sym}^{i} S_{X^{+}}^{+} (- i E_{1}^{'} - k E_{2}) & \cap (f^{*} {Sym}^{i} S_{X}^{*} (- l E_{2}) + f^{+ *} {Sym}^{i} S_{X^{+}}^{+} (- i E_{1}^{'} - (k + 1) E_{2})) . \end{matrix}

We have

\begin{matrix} f^{*} {Sym}^{i} S_{X}^{*} = F^{0, 0} \supset F^{1, 0} \supset F^{1, 1} \supset F^{2, 0} \supset & \cdot \cdot \cdot \supset F^{i - 1, i - 1} \supset F^{i, 0} = f^{+ *} {Sym}^{i} S_{X^{+}}^{+} (- i E_{1}^{'} - i E_{2}) . \end{matrix}

and

ℱ^{k, l} / ℱ^{k, l + 1} \sim = O_{E_{2}} (- i E_{1}^{'} - k E_{2}) \otimes f^{+ *} O_{X^{+}} (- k + l)

for

0 \leq l \leq k < i

, where we put

ℱ^{k, k + 1} = ℱ^{k + 1, 0}

. Since

R (f_{2}^{+})_{*} O_{E_{2}} (t E_{2}) = 0

for

0 < t \leq n - 3

, we have

R (f_{2}^{+})_{*} (ℱ^{k, l} / ℱ^{k, l + 1} \otimes O_{Y} ((2 n - 5 - 2 j) E_{1}^{'} + (n - 3 - j) E_{2})) = 0

because

i + j \leq n - 3

. Therefore,

\begin{matrix} Φ ( & {Sym}^{i} S_{X}^{*} \otimes O_{X} (j)) & = & {Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (- j) & \otimes R f_{*}^{+} O_{Y} ((2 n - 5 - i - 2 j) E_{1}^{'} + (n - 3 - i - j) E_{2}) & = & {Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (- j) \otimes R f_{1 *}^{+} O_{X_{1}^{+}} ((2 n - 5 - i - 2 j) E_{1}^{+}) & = & {Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (- j) . \end{matrix}

For the inverse direction, we climb back along the filtration

ℱ^{k, l}

. Since

R f_{2 *} (ℱ^{k, l} / ℱ^{k, l + 1} \otimes O_{Y} ((i + 2 j + 1) E_{1}^{'} + (i + j) E_{2})) = 0

for

0 \leq l \leq k < i

, we have

\begin{matrix} Ψ ({Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (- j)) & = {Sym}^{i} S_{X}^{*} \otimes O_{X} (j) \otimes R f_{*} O_{Y} ((i + 2 j + 1) E_{1}^{'} + (i + j) E_{2}) & = {Sym}^{i} S_{X}^{*} \otimes O_{X} (j) \otimes R f_{1 *} O_{X_{1}} ((i + 2 j + 1) E_{1}) & = {Sym}^{i} S_{X}^{*} \otimes O_{X} (j) . \end{matrix}

(4) We already proved in (3) that

\begin{matrix} R f_{*}^{+} (f^{*} ({Sym}^{i} S_{X}^{*} \otimes O_{X} (n - 2 - i)) \otimes O_{Y} ((2 n - 5) E_{1}^{'} + (n - 3) E_{2})) & \sim = R f_{*}^{+} (F^{i - 1, 0} \otimes f^{+ *} O_{X} (- n + 2 + i) \otimes O_{Y} ((2 i - 1) E_{1}^{'} + (i - 1) E_{2})) . \end{matrix}

Moreover

\begin{matrix} R^{p} f_{*}^{+} (f^{+ *} ({Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (- n + 2 + i)) \otimes O_{Y} ((i - 1) E_{1}^{'} - E_{2})) & \to R^{p} f_{*}^{+} (F^{i - 1, 0} \otimes f^{+ *} O_{X} (- n + 2 + i) \otimes O_{Y} ((2 i - 1) E_{1}^{'} + (i - 1) E_{2})) \end{matrix}

is surjective for

p > 0

. Therefore, there are no higher direct images for

f^{+}

Since

\begin{matrix} 0 \to F^{i - 1, 0} \to f^{+ *} {Sym}^{i} S_{X^{+}}^{+} \otimes O_{Y} (- i E_{1}^{'} - (i - 1) E_{2}) & \to O_{E_{2}} (- i E_{1}^{'} - (i - 1) E_{2}) \otimes f^{+ *} O_{X^{+}} (- i) \to 0 \end{matrix}

we have

\begin{matrix} 0 \to f_{*}^{+} (F^{i - 1, 0} \otimes f^{+ *} O_{X} (- n + 2 + i) \otimes O_{Y} ((2 i - 1) E_{1}^{'} + (i - 1) E_{2})) & \to {Sym}^{i} S_{X^{+}}^{+} \otimes O_{X} (- n + 2 + i) \to O_{W^{+}} \otimes O_{X^{+}} (- n + 2) \to 0 . \end{matrix}

On the other hand, we have

\begin{matrix} f_{*}^{+} (F^{i - 1, 0} \otimes f^{+ *} O_{X} (- n + 2 + i) \otimes O_{Y} ((2 i - 1) E_{1}^{'} + (i - 1) E_{2})) |_{X^{+} \ G^{+}} & \sim = ℰ_{i}^{+} \otimes O_{X^{+}} (- n + 2 + i) |_{X^{+} \ G^{+}} . \end{matrix}

By taking the direct image sheaves of both sides under the inclusion morphism

X^{+} \ G^{+} \to X^{+}

, we obtain

\begin{matrix} f_{*}^{+} (F^{i - 1, 0} \otimes f^{+ *} O_{X} (- n + 2 + i) \otimes O_{Y} ((2 i - 1) E_{1}^{'} + (i - 1) E_{2})) & \sim = ℰ_{i}^{+} \otimes O_{X^{+}} (- n + 2 + i) . \end{matrix}

We assume that

n = 4

besides

r = 2

in the following. Then

dim G = 4

dim X_{0} = 8

dim X = 9

, and

dim W = 7

. In particular,

W

is locally complete intersection. We have

K_{Y} = f^{*} K_{X} + 4 E_{1}^{'} + E_{2}

In this case, Namikawa [14] proved that the functors

Φ^{'} : D^{b} (Coh (X)) \to D^{b} (Coh (Y)), Ψ^{'} : D^{b} (Coh (Y)) \to D^{b} (Coh (X))

defined by

Φ^{'} (a) = R f_{*}^{+} (L f^{*} (a) \otimes O_{Y} (4 E_{1}^{'} + E_{2})), Ψ^{'} (b) = R f_{*} (L f^{+ *} (b) \otimes O_{Y}) .

are not equivalences.

Lemma 3.4. Assume that

n = 4

. Then (1)

Ψ (I_{W^{+}} \otimes O_{X^{+}} (- 2)) = O_{X} (2)

(2)

Ψ (ℰ_{i}^{+} \otimes O_{X^{+}} (i - 2)) = {Sym}^{i} S^{*} \otimes O_{X} (2 - i)

for

0 < i \leq 2

Proof. (1)

W^{+}

is a divisor on

X_{0}^{+}

and

φ^{+}

is the contraction of

(- 2)

-curves along the generic points of

φ (W^{+})

. Hence

O_{X_{0}^{+}} (- W^{+}) = O_{X_{0}^{+}} (2)

. We have an exact sequence

0 \to O_{X^{+}} (- 2 - X_{0}^{+}) \to I_{W^{+}} \otimes O_{X^{+}} (- 2) \to O_{X_{0}^{+}} \to 0 .

Since

X_{0}^{+}

is a Cartier divisor, we have

L_{k} f^{+ *} O_{X_{0}^{+}} = 0

for

k > 0

. Hence

0 \to f^{+ *} O_{X^{+}} (- 2 - X_{0}^{+}) \to f^{+ *} (I_{W^{+}} \otimes O_{X^{+}} (- 2)) \to f^{+ *} O_{X_{0}^{+}} \to 0 .

On the other hand,

0 \to f^{+ *} O_{X^{+}} (- 2 - X_{0}^{+}) \to I_{W^{+}} O_{Y} \otimes f^{+ *} O_{X^{+}} (- 2) \to O_{(X_{0}^{+})^{''}} \to 0

where

(X_{0}^{+})^{''} = (f_{*}^{+})^{- 1} X_{0}^{+}

. Thus

\begin{matrix} 0 \to O_{(E_{1}^{+})^{'} + E_{2}^{+}} (- (X_{0}^{+})^{''}) \to f^{+ *} (I_{W^{+}} \otimes O_{X^{+}} (- 2)) & \to I_{W^{+}} O_{Y} \otimes f^{+ *} O_{X^{+}} (- 2) \to 0 . \end{matrix}

We have

O_{(E_{1}^{+})^{'} + E_{2}^{+}} (- (X_{0}^{+})^{''}) \sim = O_{E_{1}^{'} + E_{2}} (E_{1}^{'} + E_{2}) .

Since

R f_{2 *} O_{E_{2}} (E_{2}) = 0

and

R f_{1 *} O_{E_{1}} (2 E_{1}) = 0

, we have

R f_{*} O_{E_{1}^{'} + E_{2}} (2 E_{1}^{'} + E_{2}) \sim = R f_{1 *} O_{E_{1}} (2 E_{1}) \sim = 0 .

Therefore, we have

\begin{matrix} R f_{*} (f^{+ *} (I_{W^{+}} \otimes O_{X^{+}} (- 2)) \otimes O_{Y} (E_{1}^{'})) & \sim = R f_{*} (I_{W^{+}} O_{Y} \otimes f^{+ *} O_{X^{+}} (- 2) \otimes O_{Y} (E_{1}^{'})) . \end{matrix}

Let

F^{+} = (X_{0}^{+})^{''} \cap (E_{1}^{+})^{'} \subset Y

. The projection

f^{+} : F^{+} \to G^{+}

is a

P^{3}

-bundle whose fiber

(P_{0}^{+})^{'}

is contracted by

f

. We have

I_{W^{+}} O_{Y} = I_{F^{+}} (- E_{1}^{'} - E_{2})

Thus

I_{W^{+}} O_{Y} \otimes f^{+ *} O_{X^{+}} (- 2) \otimes O_{Y} (E_{1}^{'}) = I_{F^{+}} f^{*} O_{X} (2) \otimes O_{Y} (4 E_{1}^{'} + E_{2}) .

On the other hand, there is an exact sequence we have

O_{(P_{0}^{+})^{'}} (E_{1}^{'}) \sim = O_{P^{3}} (- 1)

and

O_{(P_{0}^{+})^{'}} (E_{2}) \sim = O_{P^{3}} (2)

. Hence

O_{(P_{0}^{+})^{'}} (4 E_{1}^{'} + E_{2}) \sim = O_{P^{3}} (- 2)

. It follows that

R f_{*} O_{F^{+}} (4 E_{1}^{'} + E_{2}) = 0

. Therefore,

R f_{*} (I_{F^{+}} O_{Y} \otimes f^{+ *} O_{X^{+}} (- 2) \otimes O_{Y} (E_{1}^{'})) \sim = O_{X} (2) .

(2) We have an exact sequence

0 \to ℰ_{i}^{+} \otimes O_{X^{+}} (i - 2) \to {Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (i - 2) \to O_{W^{+}} \otimes O_{X^{+}} (- 2) \to 0 .

Thus the torsion part of

f^{+ *} ℰ_{i}^{+} \otimes O_{X^{+}} (i - 2)

is isomorphic to that of

f^{+ *} I_{W^{+}} \otimes O_{X^{+}} (- 2)

. Therefore, by the argument in (1), we have

Ψ (ℰ_{i}^{+} \otimes O_{X^{+}} (i - 2)) \sim = R f_{*} (f^{+ *} ℰ_{i}^{+} / torsion \otimes f^{+ *} O_{X^{+}} (i - 2) \otimes O_{Y} (E_{1}^{'})) .

We have an exact sequence

\begin{matrix} 0 \to f^{+ *} ℰ_{i}^{+} / torsion \otimes f^{+ *} O_{X^{+}} (i - 2) \otimes O_{Y} (E_{1}^{'}) & \to f^{+ *} ({Sym}^{i} S_{X^{+}}^{+} \otimes O_{X^{+}} (i - 2)) \otimes O_{Y} (E_{1}^{'}) & \to f^{+ *} (O_{W^{+}} \otimes O_{X^{+}} (- 2)) \otimes O_{Y} (E_{1}^{'}) \to 0 . \end{matrix}

Therefore, we have an exact sequence

\begin{matrix} 0 \to f^{+ *} ℰ_{i}^{+} / torsion \otimes f^{+ *} O_{X^{+}} (i - 2) \otimes O_{Y} (E_{1}^{'}) & \to F^{i - 1, 0} \otimes f^{+ *} O_{X^{+}} (i - 2) \otimes O_{Y} ((i + 1) E_{1}^{'} + (i - 1) E_{2}) \to Q \to 0 \end{matrix}

where the cokernel

Q

has a decomposition as follows

0 \to O_{F^{+}} (- E_{2}) \otimes f^{+ *} O_{X^{+}} (- 2) \to Q \to O_{E_{1}^{'}} (E_{1}^{'} - E_{2}) \otimes f^{+ *} O_{X^{+}} (- 2) \to 0 .

As in (1),

F^{+} = F = X_{0}^{''} \cap E_{1}^{'}

is a

P^{3}

-bundle over

G

with fibers

P_{0}^{'}

such that

O_{P_{0}^{'}} (E_{1}^{'}) \sim = O_{P^{3}} (- 1)

and

O_{P_{0}^{'}} (E_{2}) = O_{P^{3}} (2)

. Thus

O_{F^{+}} (- E_{2}) \otimes f^{+ *} O_{X^{+}} (- 2) \otimes O_{P_{0}^{'}} \sim = O_{P_{0}^{'}} (4 E_{1}^{'} + E_{2}) \sim = O_{P_{0}^{'}} (- 2)

and we have

R f_{*} O_{F^{+}} (- E_{2}) = 0

. We have also

\begin{matrix} R f_{*} O_{E_{1}^{'}} (E_{1}^{'} - E_{2}) \otimes f^{+ *} O_{X^{+}} (- 2) & \sim = R f_{*} O_{E_{1}^{'}} (3 E_{1}^{'} + E_{2}) \otimes O_{X} (2) & \sim = R f_{1 *} O_{E_{1}} (3 E_{1}) \otimes O_{X} (2) \sim = 0 . \end{matrix}

Therefore

\begin{matrix} Ψ (ℰ_{i}^{+} \otimes O_{X^{+}} (i - 2)) & \sim = R f_{*} (F^{i - 1, 0} \otimes f^{+ *} O_{X^{+}} (i - 2) \otimes O_{Y} ((i + 1) E_{1}^{'} + (i - 1) E_{2})) & \sim = R f_{*} (F^{i - 1, 0} \otimes O_{Y} ((5 - i) E_{1}^{'} + E_{2})) \otimes O_{X} (2 - i) . \end{matrix}

i = 1

, then

F^{0, 0} = f^{*} {Sym}^{i} S^{*}

and

R f_{*} O_{Y} (4 E_{1}^{'} + E_{2}) = O_{X}

, hence the result.

i = 2

, then

ℱ^{0, 0} / ℱ^{1, 0} \sim = O_{E_{2}} (- 2 E_{1}^{'})

, hence

R f_{*} (ℱ^{0, 0} / ℱ^{1, 0} \otimes O_{Y} (3 E_{1}^{'} + E_{2})) = 0

and we complete the proof.

Theorem 3.5. If

n = 4

, then

Φ

and

Ψ

in ( 3.1 ) are equivalences.

Proof. There is an adjunction morphism of functors $F : {Id}_{D^{b} (Coh (X))} \to Ψ Φ$ , which is reduced to the identity when restricted to the open subset $X \ W$ .
By Lemmas 3.2 , 3.3 and 3.4 , we have isomorphisms $ω \sim = Ψ Φ (ω)$ for the spanning class $Ω = {ω}$ given by Lemma 3.2 . Since the $ω$ are locally free sheaves, it follows that the morphisms $F (ω)$ are isomorphisms. Therefore, the natural homomorphisms $Φ : {Hom}^{p} (ω_{1}, ω^{2}) \to {Hom}^{p} (ω_{1}, Ψ Φ (ω^{2})) \to {Hom}^{p} (Φ (ω_{1}), Φ (ω^{2}))$ for any $ω_{1}, ω_{2} \in Ω$ and $p \in Z$ are isomorphisms. Then $Φ$ is an equivalence by [2] and [3] .

Remark 3.6. (1) Let

F = \oplus ω

be the sum of the spanning sheaves given in Lemma 3.2 . Then

F

is not an almost exceptional object in the sense of [1] in our case

n = 4

and

r = 2

, because

R^{1} φ_{*} O_{X} (- 2) \neq 0

implies

{Hom}^{1} (F, F) \neq 0

(2) In order to extend our argument to the case

n \geq 5

, the Eagon-Northcott resolution on

X_{0}^{+}

would be useful ([7] ):

\begin{matrix} 0 \to π_{0}^{+ *} ({Sym}^{n - 2} S^{+ 2})^{*} \otimes^{n} ⋀ V^{* n} \overset{B_{*}}{\to} \cdot \cdot \cdot \to π_{0}^{+ *} ({Sym}^{k} S^{+ 2})^{*} \otimes^{k + 2} ⋀ V^{* n} & \overset{B_{*}}{\to} π_{0}^{+ *} ({Sym}^{k - 1} S^{+ 2})^{*} \otimes^{k + 1} ⋀ V^{* n} \overset{B_{*}}{\to} . . . \overset{B_{*}}{\to} π_{0}^{+ *} (S^{+ 2})^{*} \otimes^{3} ⋀ V^{* n} & \overset{B_{*}}{\to} O_{X_{0}^{+}} \otimes^{2} ⋀ V^{* n} \overset{^{2} ⋀ B}{\to} {π_{0}^{+ *}}^{2} ⋀ S^{+ 2} \to O_{W^{+}} (- 1) \to 0 . \end{matrix}

References

R. Bezrukavnikov and D. Kaledin. McKay equivalence for symplectic resolutions of singularities. math.AG/0401002.
A. Bondal, D. Orlov. Semiorthogonal decomposition for algebraic varieties. alg-geom/9506012.
Tom Bridgeland. Equivalences of triangulated categories and Fourier-Mukai transforms. math.AG/9809114. Bull. London Math. Soc. 31 (1999), no. 1, 25–34.
Tom Bridgeland. Flops and derived categories. math.AG/0009053. Invent. Math. 147 (2002), no. 3, 613–632.
Tom Bridgeland, Alastair King and Miles Reid. Mukai implies McKay: the McKay correspondence as an equivalence of derived categories. math.AG/9908027. J. Amer. Math. Soc. 14 (2001), no. 3, 535–554.
J. -C. Chen. Flops and equivalences of derived categories for threefolds with only Gorenstein singularities. math.AG/0202005.
David Eisenbud. Commutative Algebra with a View Toward Algebraic Geometry. Springer-Verlag, 1995.
M. M. Kapranov. Derived category of coherent sheaves on Grassmann manifolds. (Russian) Izv. Akad. Nauk SSSR Ser. Mat. 48 (1984), no. 1, 192–202.
Yujiro Kawamata. Francia's flip and derived categories. math.AG/0111041. Algebraic geometry, 197–215, de Gruyter, Berlin, 2002.
Yujiro Kawamata. D-equivalence and K-equivalence. math.AG/0205287. J. Differential Geom. 61 (2002), no. 1, 147–171.
Yujiro Kawamata. Log Crepant Birational Maps and Derived Categories. math.AG/0311139.
Eyal Markman. Brill-Noether duality for moduli spaces of sheaves on K3 surfaces. math.AG/9901072. Journal of Algebraic Geometry 10 (2001), no. 4, 623-694.
Yoshinori Namikawa. Mukai flops and derived categories. math.AG/0203287. J. Reine Angew. Math. 560 (2003), 65–76.
Yoshinori Namikawa. Mukai flops and derived categories II. math.AG/0305086.

Department of Mathematical Sciences, University of Tokyo, Komaba, Meguro, Tokyo, 153-8914, Japan kawamata@ms.u-tokyo.ac.jp

Derived equivalence for stratified Mukai flop on G ( 2 , 4 )

Yujiro Kawamata

November 27, 2006

Derived equivalence for stratified Mukai flop on $G (2, 4)$