B OLLETTINO
U NIONE M ATEMATICA I TALIANA
Giambattista Marini
Algebraic cycles on abelian varieties and their decomposition
Bollettino dell’Unione Matematica Italiana, serie 8, volume 7-B (2004), n. 1, p. 231-240.
<http://www.bdim.eu/item?id=BUMI_2004_8_7B_1_231_0>
Unione Matematica Italiana, 2004, diritti riservati.c
Articolo digitalizzato nel quadro del programma bdim (Biblioteca Digitale Italiana di Matematica)
SIMAI & UMI http://www.bdim.eu/
Bollettino U. M. I.
(8) 7-B (2004), 231-240
Algebraic Cycles on Abelian Varieties and their Decomposition.
GIAMBATTISTA MARINI
Sunto. – In questo lavoro consideriamo una varietà abeliana X ed il suo anello di Chow CHl(X) dei cicli algebrici modulo equivalenza razionale. Tramite la decom- posizione di Künneth della diagonale D % X 3X è possibile ottenere delle formule esplicite per i proiettori associati alla decomposizione di Beauville (1) di CHl(X), tali formule sono espresse in termini delle immagini dirette e inverse dei morfismi di moltiplicazione per un intero m . Il teorema (4) fornisce delle drastiche semplifi- cazioni di tali formule, la Proposizione (9) ed il Corollario (10) forniscono alcuni risultati ad esse correlati.
Summary. – For an Abelian Variety X , the Künneth decomposition of the rational equivalence class of the diagonal D % X 3X gives rise to explicit formulas for the projectors associated to Beauville’s decomposition (1) of the Chow ring CHl(X), in terms of push-forward and pull-back of m-multiplication. We obtain a few simplifi- cations of such formulas, see theorem (4) below, and some related results, see propo- sition (9) below.
0. – Introduction.
Let X be an abelian variety of dimension n and denote by CHl(X) its Chow group of algebraic cycles modulo rational equivalence. In our notation, CHd(X) is the subgroup of d-dimensional cycles and CHp(X) »4 CHn 2d(X) is the subgroup of p-codimensional cycles. For m Z , let mult (m) denote the multiplication map X KX, x O mx. By the use of Fourier-Mukai transform for abelian varieties (see [M] and [Be]), Beauville has established a decomposition
CHd(X)Q4s 42dn 2d
5
[CHd(X)Q]s(1)
where, by definition, CHd(X)Q4 CHd(X) 7Q is the Chow group with Q-
coefficients and the right-hand-side subgroups are defined as follows:
[CHd(X)Q]s»4 4
]W CHd(X)QN mult (m)xW 4m2 d 1sW , (m Z(
]W CHp(X)QN mult (m)xW 4m2 p 2sW , (m Z( , (2)
where p 4n2d is the codimension of W.
This decomposition is a tool to understand cycles and rational equivalence on abelian varieties and it would give a beautiful answer to many questions concerning the Chow groups of abelian varieties (see [Be], [Bl], [J], [Ku] and [S]), provided that Beauville’s vanishing conjecture [Be] holds. This conjec- ture states that the factors of CHd(X) with s E0 vanish (see B.C. below). As pointed out in the abstract, by the use of Deninger-Murre projectors di, (see [DM], [Ku]), the projections CHd(X) K [CHd(X) ]s with respect to Beauville’s decomposition (1) can be written as linear forms of mult (m)x and mult (m)x. Theorem (4) simplifies such explicit descriptions. A further simplification is given for the case where one works modulo a piece of the decomposition, see proposition (9); see corollary (10) for a reformulation of Beauville’s conjec- ture.
1. – The algebraic set up.
We denote by v(z) the series expansion of log(z 11). Namely,
v(z) »4z2 1 2z21 1
3z3R.
Furthermore, for k and j non-negative integers we define constants ak , j via the formal equality
j 40
!
Q
ak , jzj4 1 k!v(z)k
Let ArMr11, r11(Q) be the matrix (ak , j), where k and j run in [ 0 , R , r].
Let BrMr11, r11(Z) be the matrix (bj , h), where j and h run in [ 0 , R , r] and where, by definition, bj , h4 (21 )j 2h
g
hjh
. It is understood thatg
hjh
4 0 provided that h Dj. For k40, 1, R , r we define linear forms Lk(r)(x0, R , xr) by the following equality:u
L0QQQ(r)Lr(r)
v
4 ArBru
xQQQ0xr
v
,ALGEBRAIC CYCLES ON ABELIAN VARIETIES ETC. 233
namely we define (observe that ak , j4 0 , if j E k and bj , h4 0 , if h D j)
Lk(r)(x0, R , xr) 4
!
j 4k
r
!
h 40 j
ak , j(21)j 2h
g
hjh
xh,and for k Dr we define Lk(r)4 0 .
We now introduce a numerical lemma, the proof of which is very straight- forward (and omitted).
LEMMA 3. – Let j F1 and sF0 be integers. Then
h 40
!
j
(21)j 2h
g
hjh
hs4./´0s!if s Ej if s 4j .
2. – Projections of cycles.
Next, using linear forms Lk(r), we give a criterium to identify components (with respect to Beauville’s decomposition 1) of the algebraic cycles. In the se- quel, X denotes an abelian variety of dimension n ; W CHd(X)Qdenotes a ra- tional algebraic cycle of dimension d and p 4n2d its codimension; further- more, Wsdenotes a component of W with respect to Beauville’s decomposition (1), in particular s is an integer in the range [2d, n2d]. We also consider lin- ear forms Lk(r) as introduced in the previous section. The interpretation, in terms of push-forward and pull-back of multiplication maps, of the decomposi- tion of the diagonal D CHn(X 3X) (see [DM], [Ku]) gives
Ws4
(
[ log (D) ]yrel2 d1siW)
O( 2 d1s) !4(
t[ log (D) ]yrel2 n22d2siW)
O( 2 n22 d2s) ! , where yrel denotes the relative Pontryagin product on CHl(X 3X) with re- spect to projection on the first factor and where, for a CHl(X 3X),ta de- notes its transpose. This equality in turn, in terms of our Lk(r) givesWs4 L2 d 1s(r) (mult ( 0 )x, R , mult (r)x) W
4 L2 p 2s(r)
(
mult ( 0 )x, R , mult (r)x)
W , (r F 2 n .It is worthwhile to stress that the linear forms Lk(r)enter in a natural way (for r 42n) as an explicit version of Deninger-Murre-Künnemann projectors in terms of push-forward and pull-back of multiplication maps. The following theorem (4) goes further, it says that such equalities hold for r that takes smaller values (see ( 4a) and ( 4b) below). We also want to stress that linear forms Lk(r)have an increasing length with respect to r (see the list at the next page).
THEOREM 4. – Let X , W and Ws be as above. Then
Ws4 L2 d 1s(r) (mult ( 0 )x, R , mult (r)x) W , (r F n 1 d ; (4a)
Ws4 L2 p 2s(r) (mult ( 0 )x, R , mult (r)x) W , (r F n 1 p . (4b)
Formulas ( 4a) and ( 4b) are obtained by using lemma (7) below. We shall also see that ( 4b) can be refined: the equality there also holds for r Fn1p2 min ]d, 2(. A similar achievement does not hold for (4a). As an explicit example we want to point out that for a 4-dimensional abelian variety and a 2-cycle W the known formula for projectors would give
W14 8 W 2 14 mult ( 2 )xW 1 56
3 mult ( 3 )xW 2 35
2 mult ( 4 )xW 1 56
5 mult ( 5 )xW 2 14
3 mult ( 6 )xW 1 8
7 mult ( 7 )xW 2 1
8 mult ( 8 )xW meanwhile, by theorem (4), or better by remark (8), one has the simpler ex- pression W14 4 W 2 3 mult ( 2 )xW 14
3mult ( 3 )xW 21
4mult ( 4 )xW . REMARK. – Beauville’s conjecture (see [Be]) states that
[CHd(X)Q]s4 0 , if s E 0 . (B.C.)
As a consequence of theorem (4), proving the conjecture is equivalent to proving that either
L2 d 1s(n 1d)(mult ( 0 )x, R , mult (n 1d)x) or L2 p 2s(n 1p)(mult ( 0 )x, R , mult (n 1p)x) acts trivially on CHd(X)Q, for s E0. Another equivalent formulation for Beauville’s conjecture (B.C.) is that the property ( 4b) holds also for r F2p (this is trivial: since L2 p 2s( 2 p) 4 0 for s E 0 , if ( 4b) holds for r 42p, B.C. holds as well; it is straightforward to check that the converse implication follows from the proof of theorem 4).
REMARK. – Let us look at ( 4a) and ( 4b). The operators L2 d 1s(r) (mult ( 0 )x, R , mult (r)x)
are non-trivial for r Fn1d and the operators L2p2s(r) (mult ( 0 )x, R , mult (r)x) are non-trivial for r Fn1p. Infact, since 2dGsGn2d, then 2d1sGn1d as well as 2 p 2sGn1p.
ALGEBRAIC CYCLES ON ABELIAN VARIETIES ETC. 235
Clearly, one has
mult ( 0 )xW 4. /´ 0 deg W Q o
if d 4dimWD0 ;
if W is a 0 -cycle , where o is the origin of X . mult ( 1 )xW 4W
For n 1d that takes the indicated value, the operators Lk4 Lk(n 1d)(R , mult (i)x, R ) act as follows.
n+d=1
L0W 4 mult(0)xW L1W 42mult(0)xW 1W n+d=2
L0W 4 mult(0)xW L1W 423
2mult ( 0 )xW 12W21
2mult ( 2 )xW L2W 41
2mult ( 0 )xW 2W11
2mult ( 2 )xW n+d=3
L0W 4 mult(0)xW L1W 4211
6 mult ( 0 )xW 13W23
2mult ( 2 )xW 11
3mult ( 3 )xW L2W 4 mult(0)xW 25
2W 12 mult(2)xW 21
2mult ( 3 )xW L3W 421
6mult ( 0 )xW 11
2W 21
2mult ( 2 )xW 11
6mult ( 3 )xW n+d=4
L0W 4 mult(0)xW L1W 4225
12mult ( 0 )xW 14W23 mult(2)xW 14
3mult ( 3 )xW 21
4mult ( 4 )xW L2W435
24mult ( 0 )xW213
3 W119
4 mult ( 2 )xW27
3mult ( 3 )xW111
24mult ( 4 )xW L3W 42 5
12mult ( 0 )xW 13
2W 22 mult(2)xW 17
6mult ( 3 )xW 21
4mult ( 4 )xW L4W 4 1
24mult ( 0 )xW 21
6W 11
4mult ( 2 )xW 21
6mult ( 3 )xW 1 1
24mult ( 4 )xW From Beauville’s conjecture point of view the first interesting case is W214 L5( 8 )(R , mult (i)x, R ) 4L5( 7 )
(
R, mult (i)x, R)
, for W CH2(X)Q and dim X 45, see [Be]. Indeed, we have also W214 L5(r)(
R, mult (i)x, R)
, for r F5 4n1p2min ]d, 2(.Next we prove theorem (4) and some related results. First, we recall that
the Chow group of an abelian variety has two ring structures: the first one is given by the intersection product, the second one is given by the Pontryagin product, which we shall always denote by y. Consider the ring CHl(X 3X) with the natural sum of cycles and the relative Pontryagin product with re- spect to projection on the first factor X 3XKX (in other terms, we consider Pontryagin product on X 3X regarded as an abelian scheme over X via the first-factor-projection). Let D CHn(X 3X) be the diagonal and let E4X3 ]o( CHn(X 3X) be the unit of CHl(X 3X) with respect to the product above, where o is the origin of X . The projectors d0, R , d2 nare defined by (see [Ku], pag. 200)
dj4 1
( 2 n 2j)![ log (D) ]yrel2 n 2j
4 1
( 2 n 2j)!
k
(D 2E)2 12(D 2E)yrel21 13(D 2E)yrel3R
l
yrel2 n2j.Since (D 2E)yrel2 n 114 0 (see [Ku]), the series above are infact finite sums.
Now let Dm denote the graph of mult (m). By Deninger, Murre and Künne- mann theorem (see [DM], [Ku]) we have
[tDm]idj4 mjdj, (m Z , 0 G j G 2 n ;
tdj4 d2 n 2j, ( 0 G j G 2 n ; (5)
where the composition above is the composition of correspondences and where, for s Corr (A , B),ts Corr (B , A) denotes its transpose. As a conse- quence, for W CHd(X)Q and 0 GjG2n, one has
mult (m)x(djiW) 4 [tDm]i(djiW)
4 mj(djiW) , (m Z .
Clearly, one identifies CHl(X) with Corr (Spec C , X) 4CHl(Spec C 3X).
Thus, by the definition (2) one has
djiW [CHd(X)Q]s, s »42n22d2j.
(58)
Since
!
dj4 D acts as the identity map, (5) and (58) give Ws4 d2 n 22d2siW 4td2 d 1siW (59)where, as usual, Ws denotes the component of W with respect to Beauville’s decomposition (1).
For the proof of theorem (4) we need the following.
ALGEBRAIC CYCLES ON ABELIAN VARIETIES ETC. 237
LEMMA 6. – Let W be as in the theorem. Then
[ (D 2E)yrelj]iW 4
!
h 40 j
(21)j 2h
g
hjh
mult (h)xWt[ (D 2E)yrelj]iW 4
!
h 40 j
(21)j 2h
g
hjh
mult (h)xWPROOF. – Since E is the unit for relative Pontryagin product and since DyrelhiW 4 mult(h)xW as well as t[Dyrelh]iW 4 mult(h)xW , the two equali- ties follow by a straightforward computation. r
LEMMA 7. – Let W be as in the theorem. Then
[ (D 2E)yrelj]iW 40, ( j F n 1 d 1 1 ; (7a)
t[ (D 2E)yrelj]iW 40, ( j F n 1 p 1 1 . (7b)
PROOF. – We prove (7b), the proof of (7a) is very similar. By lemma (6), we have to show that for j Fn1p11 one has
h 40
!
j
(21)j 2h
g
hjh
mult (h)xW 40 .By linearity of the left-hand-side operator we are free to assume that W be- longs to one of the factors from Beauville decomposition (1), namely we are free to assume that W [CHd(X)Q]sfor some s [2d, n2d]. Thus (see 2), we assume that mult (m)xW 4m2 p 2sW , (m Z . It follows
h 40
!
j
(21)j 2h
g
hjh
mult (h)xW 4h 40!
j (21)j 2hg
hjh
h2 p 2sW .For s in the range above, the range for 2 p 2s is [p, n1p]; in particular, we have 2 p 2sEj. By lemma (3), the coefficient
!
h 40 j
(21)j 2h
g
hjh
h2 p 2s vanishes.Then we are done. r
PROOF (of theorem 4). – We start with formula ( 4a). Let k 42d1s. Then, we have
Ws4 d2 n 22d2siW 4 1
( 2 d 1s)![ log (D)yrel2 d 1s]iW
4
!
j 4k 2 n
ak , j(D 2E)yreljiW .
Now observe that by lemma (7), we have (D 2E)yreljiW 40 for jFn1d11.
Thus, the summation above can be taken up to r , provided that r Fn1d. It follows that
Ws4
!
j 4k r
ak , j(D 2E)yreljiW , (r F n 1 d .
Looking at the definition of the operators Lk(r)it is then clear that ( 4a) follows from the first equality from lemma (6),
(D 2E)yreljiW 4
!
h 40 j
(21)j 2h
g
hjh
mult (h)xW .The proof of formula ( 4b) is similar. For r Fn1p we have Ws4td2 d 1siW 4 1
( 2 p 2s)!
t[ log (D)xrel2 p 2s]iW
4
!
j 42p2s 2 n
a2 p 2s, jt[ (D 2E)xrelj]iW
4
!
j 42p2s r
a2 p 2s, jt[ (D 2E)xrelj]iW
4
!
j 42p2s r
a2 p 2s, j
!
h 40 j
(21)j 2h
g
hjh
mult (h)xW4 L2 p 2s(r) (mult ( 0 )x, R , mult (r)x) W
where the 4th equality follows by lemma (7), the 5th equality follows by lem- ma (6) and the 6th equality follows by the definition of the operators Lk(r). r
REMARK 8. – The equality (7b) can be improved. We have,
t[ (D 2E)xrelj]iW 40, (j F n 1 p 1 1 2 d (88)
where d4min ]d, 2(. Infact, since [CHd(X)Q]s40 provided that sGmin ]2d11, 21 ( (see [Be]), the actual range for s can be shrinked to min ]2d 1 2 , 0 ( G s Gn2d. Thus in turn, one obtains (88) by the same proof of (7b). As a conse- quence, (4b) can be refined: the equality there also holds for all r Fn1p2d (where d is as above).
Furthermore, for the same reason, if Beauville’s conjecture (B.C.) men- tioned above holds, then
t[ (D 2E)xrelj]iW 40, ( j F 2 p 1 1
In particular, if W is a divisor (hence it satisfies B.C.), then t[(D2E)xrel3]i
ALGEBRAIC CYCLES ON ABELIAN VARIETIES ETC. 239
W 40, namely 3W23 mult(2)xW 1mult(3)xW 40, which is obvious (in the case of divisors, this kind of computations provide trivial results).
Now fix s , working modulo l Fs11
5
[CHd(X)Q]l, or rather modulol Gs21
5
[CHp(X)Q]l, yields simpler formulas than the ones from theorem (4); fur- thermore, it can be used to provide a reformulation for Beauville’s conjecture (B.C.), see corollary (10) and the example below.PROPOSITION 9. – Let W and Ws be as in the theorem. Then
Ws4 1
( 2 d 1s)!
!
h 40 2 d 1s
(21)2 d 1s2h
g
2 d 1shh
mult (h)xW ,modulo
5
l Fs11[CHd(X)Q]l (9a)
Furthermore,
Ws4 1
( 2 p 2s)!
!
h 40 2 p 2s
(21)2 p 2s2h
g
2 p 2shh
mult (h)xW ,modulo
5
l Gs21[CHp(X)Q]l (9b)
PROOF. – We prove (9b). Let K 4 1
( 2 p 2s)!
!
h 40 2 p 2s
(21)2 p 2s2h
g
2 p 2shh
mult (h)x.It suffices to prove that
KW 4. /´ 0 W
if W [CHd(X)Q]l, l Fs11 if W [CHd(X)Q]s.
This is clear by the proof of (7b); as for the case W [CHd(X)Q]s, the equality KW 4W follows since, by lemma (3), the coefficient
!
h 40 s
(21)s 2h
g
shh
hs equalss! (here s 42p2s). The proof of (9a) is similar. r A straightforward consequence of ( 9b) is the following.
COROLLARY 10. – Let X be as in the theorem. Then, it satisfies Beauville’s conjecture for d-dimensional cycles if and only if
h 40
!
k
(21)k 2h
g
khh
mult (h)xacts trivially on CHd(X)Q for k F2p11, where p4n2d as usual.
For 5-dimensional abelian varieties the only bad component that might exist is [CH3(X)Q]21. Then, by the corollary above it follows that a 5-dimen-
sional abelian variety X satisfies Beauville’s conjecture (B.C.) if and only if 5 W 210 mult(2)xW 110 mult(3)xW 25 mult(4)xW 1mult(5)xW 40, for all W CH3(X)Q.
R E F E R E N C E S
[Be] A. BEAUVILLE, Sur l’anneau de Chow d’une varieté abélienne, Math. Ann., 273 (1986), 647-651.
[Bl] S. BLOCH, Some Elementary Theorems about Algebraic cycles on Abelian Va- rieties Inventiones, Math., 37 (1976), 215-228.
[DM] C. DENINGER - J. MURRE, Motivic decomposition of abelian schemes and the Fourier transform, J. Reine Angew. Math., 422 (1991), 201-219.
[J] U. JANNSEN, Equivalence relation on algebraic cycles, NATO Sci. Ser. C Math.
Phys. Sci., 548 (2000). 225-260.
[GMV] M. GREEN- J. MURRE- C. VOISIN, Algebraic Cycles and Hodge Theory, Lecture Notes in Mathematics, 1594 (1993).
[Ku] K. KU¨NNEMANN, On the Chow Motive of an Abelian Scheme, Proceedings of Symposia in pure Mathematics, 55 (1994), 189-205.
[M] S. MUKAI, Duality between D(X) and D(X×) with its applications to Picard Sheaves, Nagoya Math. J., 81 (1981), 153-175.
[S] S. SAITO, Motives and filtrations on Chow groups, II, NATO Sci. Ser. C Math.
Phys. Sci., 548 (2000), 321-346.
Dipartimento di Matematica, Università degli studi di Roma II
«Tor Vergata», Via della Ricerca Scientifica, I-00133 Roma (Italy) E-mail: mariniHaxp.mat.uniroma2.it
Pervenuta in Redazione
il 7 dicembre 2002 e in forma rivista il 31 marzo 2003