Homological projective dualitymath.northwestern.edu/pinsky_lectures/Thomas/HPD.pdfH –Dppt \ptq....
Transcript of Homological projective dualitymath.northwestern.edu/pinsky_lectures/Thomas/HPD.pdfH –Dppt \ptq....
Beilinson
DpPn´1q “@O,Op1q, . . . ,Opn ´ 1q
D
“@Dpptq, Dpptq, . . . , Dpptq
D.
The notation x y means a semi-orthogonal decomposition:
§ They are semi-orthogonal: Ext˚pOpiq,Opjqq “ 0 for i ą j ,
§ Opiq is exceptional: R HompOpiq,Opiqq “ C. id. Equivalently,
Dpptq Ñ DpPn´1q
C ÞÑ Opiq
is an embedding,
§ The sheaves O,Op1q, . . . ,Opn ´ 1q generate DpPn´1q.
Sketch proofThe first two conditions are simple cohomology computations.
Generation: Since O,Op1q, . . . ,Opn ´ 1q are exceptional andsemi-orthogonal the Gram-Schmidt process shows that any objectof DpPn´1q is an extension of a piece in their span and a piece intheir orthogonal.
Thus we want to show their orthogonal is zero.
Use the Koszul resolution associated to the section
Opx0,x1,...,xn´1q // Op1q‘n
with zero locus one point tpu. Shows that Op is in the span ofO,Op1q, . . . ,Opn ´ 1q for any point p P Pn´1.
Thus any object F in the orthogonal has
F |p – R HompOp, F qrn ´ 1s “ 0
for any p P Pn´1. Therefore F “ 0 by the Nakayama lemma.
Family version
Fix
§ X “ PpE q πÝÑ B , a Pn´1-bundle, where
§ E is a rank n vector bundle over B .
Orlov proved that
DpX q “A
DpBq, DpBqp1q, . . . , DpBqpn ´ 1qE
.
Here the ith copy of DpBq is embedded via π˚p ¨ q b Opiq, whereOp´1q ãÑ π˚E is the tautological bundle on X .
Blow ups
Blow up in Z Ă X of codimension-n (both smooth):
E
p
��
� � j // BlZ X
π
��Z
� � i // X
Orlov also proved
DpBlZ X q “A
DpX q, DpZ qp´E q, . . . , DpZ qp´pn ´ 1qE qE
.
Here
§ DpX q is included via π˚ and DpZ q by j˚p˚
§ p : E Ñ Z is a Pn´1-bundle
§ its Op1q line bundle is Op´E q
so this is an analogue of his projective bundle result.
Universal hyperplaneSet up:
§ variety and line bundle pX ,OX p1qq§ basepoint free linear system V Ď H0pOX p1qq
Equivalently
§ map f : X Ñ PpV ˚q such that§ im f not contained in any hyperplane
Over the dual projective space PpV q we have the universal familyof hyperplanes H Ñ PpV q:
H :“␣
px , sq : spxq “ 0(
Ă X ˆ PpV q.
Discriminant locus is classical projective dual X _ Ă PpV q of X .
Given a linear subspace L Ď V with annihilator LK Ď V ˚, set
XLK :“ X ˆPpV ˚q PpLKq,
HL :“ H ˆPpV q PpLq.
Projective duality
Baselocus XLK of L Ď H0pOX p1qq is contained in every fibre ofHL Ñ PpLq, giving a diagram
XLK ˆ PpLq
p
��
� � j // HL
π
��
� � ι // X ˆ PpLq
ρyyttttttttttt
XLK� � i // X
Notice that
§ π has general fibre P`´2 p` :“ dim Lq§ over XLK the fibre is PpLq “ P`´1
E.g. ` “ 2 then HL Ñ X is blow up in codimension-two XLK Ă X .
Homological projective duality I
Suppose that dim XLK “ dim X ´ `.
The above diagram gives an inclusion of the derived category ofXLK into that of the universal hypersurface HL over the linearsystem:
j˚p˚ : D`XLK
˘ãÝÑ DpHLq.
π˚ is also full and faithful. Together these give a semi-orthogonaldecomposition
DpHLq “A
D`XLK
˘, π˚DpX qp0, 1q, . . . , π˚DpX qp0, ` ´ 1q
E.
pi , jq is twist by OX piq b OPpV qpjq (restricted to HL Ă X ˆ PpLq).
The “interesting part” of the derived category
Want to make DpHq smaller; more comparable to DpX q.
Cut it down to its “interesting part” (Kuznetsov).
Standard example: start with DpPn´1q “@O, . . . ,Opn ´ 1q
D.
Restrict to degree-d hypersurface H Ă X , then Opdq, . . . ,Opn ´ 1qremains exceptional semi-orthogonal collection on restriction to H.
Define “interesting part” of DpHq to be its right orthogonal:
CH :“@OHpdq,OHpd ` 1q, . . . ,OHpn ´ 1q
DK
“␣E P DpHq : R HompOpiq, E q “ 0 for i “ d , . . . , n ´ 1
(.
So
DpHq “@CH ,OHpdq,OHpd ` 1q, . . . ,OHpn ´ 1q
D.
Amazingly, CH is always a fractional Calabi-Yau category.
Examples
Interesting examples include pd “ 2, n even) and pd “ 3, n “ 6q:§ Even dimensional quadrics. CH is generated by two spinor
bundles which are exceptional and orthogonal to each other;
CH – Dppt \ ptq.
Families ù double covers of linear systems of quadrics.
§ Cubic fourfolds. CH is the derived category of a K3 surface,noncommutative in general.
2-dimensional CY category RHompE , F q˚ – RHompF , E qr2s,deformation of DpK3q.
“Explains” the Beauville-Donagi holomorphic symplectic form onthe Fano variety F pHq of lines in H. F pHq is a moduli space ofobjects πCH
pILq P CH so inherits Mukai’s symplectic structure.
More later!
Families
Put categories CH Ă DpHq together over the P`H0pOPn´1pdq
˘
family of all Hs. So:
§ Set V “ H0pOPn´1pdqq§ For L Ď V , with universal hypersurface HL Ă Pn´1 ˆ PpLq,
CHL:“
@DpPpLqqpd , 0q, DpPpLqqpd`1, 0q, . . . , DpPpLqqpn´1, 0q
DK
so that DpHLq is
@CHL
, DpPpLqqpd , 0q, DpPpLqqpd ` 1, 0q, . . . , DpPpLqqpn ´ 1, 0qD
§ Putting L “ V gives CH, the HP dual of pPn´1,Opdqq.
Soon we will refine HPD I by replacing DpHLq with CHL.
Lefschetz collections
For general pX ,OX p1qq Kuznetsov replaces the Beilinsondecomposition with a (rectangular) Lefschetz decomposition:
§ an admissible subcategory A Ď DpX q§ generating a semi-orthogonal decomposition
DpX q “@A,Ap1q, . . . ,Api ´ 1q
D.
(Above example is pX ,OX p1qq “ pPn´1,OPn´1pdqq andA “
@O,Op1q, . . . ,Opd ´ 1q
Dwith i “ n{d . If d |n can use
non-rectangular Lefschetz collection.)
Setting CH :“@Ap1q,Ap2q, . . . ,Api ´ 1q
DK Ă DpHq (forH P |OX p1q|) to be the interesting part of DpHq gives
DpHq “@CH ,Ap1q,Ap2q, . . . ,Api ´ 1q
D.
Lefschetz collections and families
Similarly for X Ñ PpV ˚q and L Ď V we set
CHL:“
@Ap1q b DpPpLqq, . . . , Api ´ 1q b DpPpLqq
DK
so that DpHLq is
@CHL
, DpPpLqqpd , 0q, DpPpLqqpd ` 1, 0q, . . . , DpPpLqqpn ´ 1, 0qD
Putting L “ V gives CH, the HP dual of X Ñ PpV q with the(rectangular) Lefschetz decomposition DpX q “
@A, . . . ,Api ´ 1q
D.
(Kuznetsov asks for CH to be geometric: i.e. to have aDpPpV qq-linear equivalence CH – DpY q for some Y Ñ PpV q.)
Picturing HPD I
DpHLq “@D`XLK
˘, π˚DpX qp0, 1q, . . . , π˚DpX qp0, ` ´ 1q
D
Api ´1,1q
Api ´1,
`´1qAp1,`´1q
Ap1, 1q
Ap0,`´1q
Ap0, 1q
D`XLK
˘
Picturing interesting part of DpHLq
CHL“@Ap1q b DpPpLqq, . . . , Api ´ 1q b DpPpLqq
DK
p0, 1q
p1, 0q pi ´2,0q pi ´1,0q
pi ´1,1q
pi ´1,
`´1qp1, `´1q
p1, 1q
p0, `´1q
D`XLK
˘
HPD IIProjecting (“mutating”) the grey boxes into the white boxes infirst column we find
Theorem (Kuznetsov)
§ If ` ą i thenCHL
“@D`XLK
˘, Ap0, 1q, Ap0, 2q, . . . , Ap0, ` ´ iq
D
§ If ` “ i then CHL– D
`XLK
˘
§ If ` ă i thenD`XLK
˘“@CHL
, Ap1, 0q, Ap2, 0q, . . . , Api ´ `, 0qD
Looks like a duality now! X Ñ PpV ˚q and its HP dualCH{DpPpV qq have similar size and are on the same footing.
Passing to a codimension-` linear section,
§ lose the first ` copies of Ap ¨ q§ gain (restriction to XLK of) the PpLq family of categories CH
as H runs through the hyperplanes containing XLK
Example I (with Addington)
Cubic fourfold X Ă P5. Hodge diamond
11
0 1 21 1 011
Generically H2,2pX ,Zq “ xh2y, but special X have an extra classT P H2,2
primpX ,Zq.
Hassett determined when the integral Hodge structurexh2, T yK Ă H4pX ,Zq is isometric to (Tate twist of) H2
primpS ,Zqfor some polarised K3 surface S .
Kuznetsov: DpX q –@CX ,O,Op1q,Op2q
Dwhere CX is a
noncommutative K3 and sometimes really – DpK3q.
We showed (more-or-less) that X is Hassett ðñ Kuznetsov.
Cubic fourfolds containing a planeCubic fourfold X Ă P5 containing a plane P Ă X Ă P5.Defines another plane
P2 :“␣3-planes P Ă P3 Ă P5
(
Such a 3-plane intersects X in a singular cubic surface P Y Q.
ñ P2 family of quadric surfaces Q; in fact
BlP X Ñ P2
is a quadric surface fibration, generic fibre P1 ˆ P1, singular fibres(cone over a conic) over discriminant sextic curve Ă P2.
S :“␣rulings of fibres
(
����P2
is the double cover corresponding to the 2 objects in Cfibre.(Branched over discriminant sextic curve.)
S Q s parameterises the sheaves ι˚IL on X , where L is any line inthe ruling corresponding to the point s P S .
Projecting into CX Ă DpX q, find S is a moduli space of objects inCX . Gives
Theorem (Kuznetsov)
Universal object on S ˆ X gives an equivalenceDpSq „ÝÑ CX Ă DpX q.
(Modulo Brauer class issues.)
We use this description and deformation theory to reach all otherdivisors of special cubic fourfolds (where a different description ofCX – DpK3q holds).
Example II (with Calabrese)
Two ways to get a Calabi-Yau 3-fold from a pencil of cubic 4-folds:
§ X “ X3,3 Ă P5 is baselocus of the pencil, i.e. the intersectionof two cubic fourfolds,
§ Y Ñ P1 is (noncommutative) K3 fibration associated to theuniversal cubic fourfold over the pencil P1.
Choose special cubic fourfolds so Y geometric (commutative).Then HPD predicts
DpX q “ DpY q.
Sheaf on Y ù sheaf on each cubic fourfold fibre ù sheaf onbaselocus X (contained in each fibre)
Derived (auto)equivalences of CY 3-folds
Pencil special ñ X , Y both singular.
For cubic fourfolds containing a plane we were able to resolvecrepantly to give non-birational Calabi-Yau 3-folds pX , pY with
DppX q – DppY q.
For cubic fourfolds with an ODP we were able to resolve crepantlyto give birational Calabi-Yau 3-folds pX , pY with an autoequivalence
DppX q ý .
Example III (with Segal)
The Grassmannian of 2-planes in C2n,
Gr p2, 2nq Ă P`Λ2C2n
˘
has classical projective dual the Pfaffian variety of degenerate2-forms (rank ď 2n ´ 2) on C2n:
Pf p2n ´ 2, 2nq Ă P`Λ2pC2nq˚˘.
Pf is a (singular) degree-n hypersurface␣ω P Λ2pC2nq˚ : ω^n “ 0
(.
Kuznetsov conjectures one can pass to a “categorical crepantresolution” so that Pf Ñ P˚ becomes HP dual to Gr Ñ P.
Taking L Ă Λ2C2n such that PpLKq Ă P`Λ2pC2nq˚
˘misses the
singularities of Pf, would get relation
D`Gr X PpLq
˘ÐÑ D
`Pf X PpLKq
˘
Quintic 3-fold
We prove this under some conditions by different methods(“LG models” or matrix factorisations)
Taking ` “ dim L “ 40, for instance, gives full and faithfulembeddings
D`Pf X P4
˘ãÝÑ D
`Gr X P39
˘.
LHS is a quintic 3-fold.Beauville shows the general quintic 3-fold is Pfaffian.
RHS is Fano 11-fold, linear section of Gr(2,10).