arXiv:2103.00426v1 [math.NT] 28 Feb 2021 Let Introduction 1 2020: classification MSC words: Key F ait over variety a evee s´tl hae on ´etale sheaves as viewed be can H eut r ute eeoe n sdb zbli [ Izhboldin by used and developed further are results nowadays. cohomo cohomology machin motivic unramified more the even compute of al.), to development et Voevodsky, the Suslin, With Lichtenbaum, Beilinson, literature. [ the in e.g. studied applications (see important problem many rationality found have they invariants, birational presheaf otadSjtai eiso aes([ papers of series a in Sujatha and Rost h bv naie ooooygop r optdu odegre to up computed are groups cohomology unramified above the hs ruscnas edsrbdi em frsdemp nGli c Galois in ([ maps conjecture residue Gersten’s of on terms theorem in Bloch–Ogus described the be to also thanks can groups These o each For . Zar 0 When F naie ooooyo udisin Quadrics of Cohomology Unramified rm2otie yKh,Rs n uah ntenineteen-nin the in Sujatha and Rost Kahn, by obtained 2 from naie ooooyof cohomology unramified h enladteckre ftentrlmpfo h cohomo the from map natural the of cokernel the and kernel of the coefficients torsion primary fdimension of X, eafil.Let field. a be Let U H F 7→ m j F F a hrceitcdffrn rm2and 2 from different characteristic has ( j udai om,qarc,urmfidchmlg,ccecasmap class cycle cohomology, unramified quadrics, forms, Quadratic j H h naie ooooygroup cohomology unramified The . eafil fcaatrsi n let and 2 characteristic of field a be H ≥ − ´ et j nr j ( ,tetno product tensor the 1, 1) ≥ U, X, over 1 where , Z /m ( m Q hrceitcTwo Characteristic / ( j eapstv nee o iiil ytecaatrsi of characteristic the by divisible not integer positive a be Z F ogH n egSUN Peng and HU Yong − ) 10,1F0 19E15 14F20, 11E04, esuyteurmfidchmlg ruswt 2- with groups cohomology unramified the study We . ′ H ( X ).B aigtedrc ii,w a lodefine also can we limit, direct the taking By 1)). Sal84 j m hsetnsterslsi hrceitcdifferent characteristic in results the extends This . − j ( X j 1) F ,[ ], − shms Let -schemes. ndges2ad3 edtriecompletely determine We 3. and 2 degrees in Abstract =lim := Kah95 )dntsteZrsisefascae othe to associated sheaf Zariski the denotes 1) CTO89 Z char( /m 1 −→ ( ,[ ], F j ,[ ], ) ∤ − m KRS98 X CTP16 H ):= 1) Izh01 nr j easot rjciequadric projective smooth a be X H X nr X, j ,[ ], sasot rjciequadric, projective smooth a is eapoe mohconnected smooth proper a be osleanme fproblems of number a solve to ] µ ) n aebe extensively been have and ]), X, KS00 Z m ⊗ ( /m j − Z 1) /m ( ,[ ], BO74 j of eties. − KS01 ( oyof logy j o ntnet the to instance for , m 1) − r a eapplied be can ery e t ot funity of roots -th ) simportant As ]). 1) j ) oeo their of Some ]). . oyter (by theory logy ≤ F stegroup the is yKahn, by 4 othe to ohomology, on quadratic forms, including a construction of fields of u-invariant 9 in characteristic =6 2. It has been noticed for decades that unramified cohomology theory can be formulated in a more general setting (see [CT95], [CTHK97], [Kah04]). In particular, when F has positive characteristic p, the aforementioned groups have p-primary torsion variants. j r Indeed, the unramified cohomology functors Hnr · , Z/p (j − 1) for all r ≥ 1 and their limit Hj · , Q /Z (j − 1) can be defined by using the Hodge–Witt cohomology (see nr p p (3.1) and (3.10)). In contrast to the prime-to-p case, there has been much less work on these p-primary unramified cohomology groups. In this paper, we are interested in the case of a smooth projective quadric X over a field F of characteristic 2. We investigate the unramified cohomology groups via the natural maps
j j r j r ηr : H F, Z/2 (j − 1) −→ Hnr X, Z/2 (j − 1) , r ≥ 1 and j j j η∞ : H F, Q2/Z2(j − 1) −→ Hnr X, Q2/Z2(j − 1) . j For each j ≥ 1, it is not difficult to see that the maps ηr for different r ≥ 1 have j j essentially the same behavior (Lemma 4.3). So we may focus on the two maps η := η1 j and η∞. They are both isomorphisms if j = 1 (Prop. 3.4) or X is isotropic (Prop. 4.1 (1)). In our main results, we determine completely the kernel and the cokernel of ηj and j η∞ for j =2, 3. The following theorem extends Kahn’s results in [Kah95] to characteristic 2.
Theorem 1.1. Let F be a field of characteristic 2 and let X be the smooth projective quadric defined by a nondegenerate quadratic form ϕ with dim ϕ ≥ 3. Assume that ϕ is anisotropic.
1. (See 5.1, 5.3, 5.6 and 5.7) Suppose dim ϕ =3, so that X is the conic associated to 1 a quaternion division algebra D, or dim ϕ = 4 and e1(ϕ)=0 ∈ H (F, Z/2), that is, ϕ is similar to the reduced norm of a quaternion division algebra D. Then Coker(η2) ∼= Ker(η2)= {0, (D)} ∼= Z/2 , Coker(η3) ∼= Ker(η3)= {(a) ∪ (D) | a ∈ F ∗} ,
where (D) denotes the Brauer class of D.
1 2. (See 5.8 and 6.7) If dim ϕ =4 and e1(ϕ) =06 ∈ H (F, Z/2), then
Coker(η2) = Ker(η2)=0 , Coker(η3) ∼= Ker(η3) , Ker(η3)= {0}∪{(a) ∪ (b) ∪ (c] | ϕ is similar to a subform of hha, b ; c]]} .
3. (See 6.7, 6.11, 6.12 and 8.4) Suppose dim ϕ > 4. Then Coker(η2) = Ker(η2)=0 and
2 (a) if ϕ is an Albert form (i.e. a 6-dimensional form with trivial Arf invariant), then Coker(η3) ∼= Z/2 and Ker(η3)=0; (b) if ϕ is a neighbor of a 3-Pfister form hha, b ; c]], then
Coker(η3) ∼= Ker(η3)= {0, (a) ∪ (b) ∪ (c]} =∼ Z/2 .
(c) in all the other cases (e.g. dim ϕ> 8), Coker(η3) = Ker(η3)=0.
The counterpart in characteristic =6 2 of the next theorem appeared in [KRS98, Thms. 4 and 5].
Theorem 1.2 (See 4.3, 6.8 and 8.5). Let F be a field of characteristic 2 and let X be the smooth projective quadric defined by a nondegenerate quadratic form ϕ with dim ϕ ≥ 3. j j 2 Then Ker(ηr ) = Ker(η∞) for all r ≥ 1, j ≥ 1, and Coker(η∞)=0. 2 ∼ 3 If ϕ is an anisotropic Albert form, then Coker(η ) = Z/2. Otherwise Coker(η∞)=0. Main tools in our proofs include the Bloch–Ogus and the Hochschild–Serre spectral sequences. A key difference between the p-primary torsion cohomology and the prime- to-p case is the lack of homotopy invariance. This results in the phenomenon that our spectral sequences look different from their analogues in characteristic different from 2. Fortunately, due to vanishing theorems for the local cohomology and the fact that the field F has 2-cohomological dimension at most 1, these spectral sequences still have many vanishing terms. The cycle class maps with finite or divisible coefficients are also studied and used in the paper. In this respect we need information about the structure of Chow groups in low codimension. This information can be found in Karpenko’s work [Kar90] in characteristic =6 2, and recently the paper [HLS21] has provided the corresponding results in characteristic 2. In the above two theorems the case of Albert quadrics is more subtle than the others. In that case we have to utilize more techniques from the algebraic theory of quadratic forms, especially residue maps on Witt groups of discrete valuation fields of characteristic 2 ([Ara18]).
Notation and conventions. For any field k, denote by k a separable closure of k. For an algebraic variety Y over k, we write YL = Y ×k L for any field extension L/k, and Y = Y ×k k. We say Y is k-rational if it is integral and birational to the dim Y projective space Pk over k. We say Y is geometrically rational if YL is L-rational for the algebraic closure L of k. M Milnor K-groups of a field k are denoted by Ki (k), i ∈ N. For an abelian group M, we denote by Mtors the subgroup of torsion elements in M. For any positive integer n, we define M[n] and M/n via the exact sequence
× 0 −→ M[n] −→ M −→n M −→ M/n −→ 0 .
2 For any scheme X, let Br(X)= H´et(X, Gm) denote its cohomological Brauer group. In the rest of the paper, F denotes a field of characteristic 2.
3 2 Quadrics and their Chow groups
(2.1) We recall some basic definitions and facts about quadratic forms in characteristic 2. For general reference we refer to [EKM08]. We work over a field F , which has characteristic 2 according to our convention. Let a ∈ F ∗. We denote by hai the 1-dimensional quadratic form x 7→ ax2, and let [1, a] or hha]] denote the binary quadratic form (x, y) 7→ x2 + xy + ay2. A (quadratic) 1-Pfister form is a binary quadratic form isomorphic to hha]] = [1, a] for some a ∈ F ∗. Let h1, aibil denote the binary bilinear form ((x1, x2), (y1, y2)) 7→ x1y1 + ax2y2. For n ≥ 2, a quadratic form is called an n-Pfister form if it is isomorphic to
hha1, ··· , an−1 ; an]] := h1, a1ibil ⊗···⊗h1, an−1ibil ⊗ hhan]]
∗ ∗ for some a1, ··· , an−1 ∈ F and an ∈ F . If λ ∈ F and ϕ is a Pfister form, the scalar multiple λϕ is called a general Pfister form. ∼ For two quadratic forms ϕ and ψ over F , we say ψ is a subform of ϕ if ψ = ϕ|W for some subspace W in the vector space V of ϕ. For n ≥ 2, an n-Pfister neighbor is a subform of dimension > 2n−1 of a general n-Pfister form. 1 We write Iq(F ) or Iq (F ) for the Witt group of even-dimensional nondegenerate n quadratic forms over F . For n ≥ 2, let Iq (F ) denote the subgroup of Iq(F ) gener- n ated by the n-Pfister forms. For a quadratic form ϕ over F , we will write ϕ ∈ Iq (F ) if n ϕ is nondegenerate, of even dimension, and its Witt class lies in Iq (F ). We also have the Witt ring W (F ) of nondegenerate symmetric bilinear forms over F , in which the classes of even-dimensional forms form an ideal I(F ), called the fundamental ideal. For each n ≥ 1, let In(F ) be the n-th power of the ideal I(F ) and put I0(F ) = n n−1 W (F ). The group Iq(F ) has a W (F )-module structure, and we have Iq (F )= I (F ) · Iq(F ) for all n ≥ 1. The Galois cohomology group H1(F, Z/2) can be identified with F/℘(F ) by Artin– Schreier theory, where ℘ denotes the map x 7→ x2 − x. For any b ∈ F , we write (b ] for its canonical image in F/℘(F )= H1(F, Z/2). The map
1 1 e1 : Iq (F ) −→ H (F, Z/2) ; hha]] 7−→ (a ] is a well defined homomorphism, often called the discriminant or Arf invariant. It is 2 2 well known that e1 is surjective with Ker(e1) = Iq (F ). A 6-dimensional form in Iq (F ) (i.e. a 6-dimensional nondegenerate form with trivial Arf invariant) is called an Albert form. For n ≥ 2, by using the Kato–Milne group Hn F, Z/2(n − 1) (cf. [Kat80], [Mil76]), a generalization of which will be discussed in (3.1), one can also define a functorial homomorphism (see [Sah72] for n = 2 and [Kat82b] for general n)
n n en : Iq (F ) −→ H F, Z/2(n − 1)
n+1 which is surjective with Ker(en)= Iq (F ) such that
en hha1, ··· , an−1 ; an]] =(a1) ∪···∪ (an−1) ∪ (an ] , 4 ∗ ∗ ∗2 where for any a ∈ F , (a) denotes its canonical image in F /F . (The maps e2 and e3 are more classical, called the Clifford invariant and the Arason invariant respectively.) Note that for all n ≥ 1,
n ∗ (2.1.1) en(ϕ)= en(c.ϕ) for all ϕ ∈ Iq (F ) and all c ∈ F
n+1 because ϕ − cϕ = hh c iiϕ ∈ Iq (F ) = Ker(en). The group H2 F, Z/2(1) may be identified with the 2-torsion subgroup of the Brauer group Br(F ) of F . For any a ∈ F ∗ and b ∈ F , let (a, b ] be the quaternion algebra generated by two elements i, j subject to the relations
i2 = a , j2 + j = b , ij = ji + i .
Its reduced norm is the 2-Pfister form hha; b]], and its Brauer class is (a)∪(b ], the Clifford invariant of hha; b]]. The plane conic defined by the ternary form hai⊥hhb]] is called the conic associated to the quaternion algebra (a, b ]. For any a, c ∈ F ∗ and b, d ∈ F , the form [1, b+d]⊥a.[1, b]⊥c.[1, d] is an Albert form. Its Clifford invariant is the Brauer class of the biquaternion algebra (a, b] ⊗F (c, d].
(2.2) Now we recall some known facts about Chow groups of projective quadrics (which are valid in arbitrary characteristic). More details can be found in [Kar90, § 2], [EKM08, § 68] and [HLS21, § 5]. Let ϕ be a quadratic form defined on a (finite dimensional) F -vector space V of dimension ≥ 3. Let X = Xϕ be the projective quadric defined by ϕ. It is a closed subvariety in the projective space P(V ). Unless otherwise stated, we always assume ϕ is nondegenerate, which means the projective quadric X = Xϕ is smooth as an algebraic variety over F . In fact, X is a smooth geometrically integral F -variety. It is well known that if K/F is a field extension such that ϕK is isotropic, then XK is K-rational. In particular, X is rational over F . For each i ∈ N, let CHi(X) denote the Chow group of codimension i cycles of X. Let h ∈ CH1(X) be the class of a hyperplane section (i.e., the pullback of the hyperplane class in CH1(P(V ))). Using the intersection pairing as multiplication in the Chow ring ([EKM08, § 57]), we get elements hi ∈ CHi(X) for each i. d−j Set d = dim X. For every integer j ∈ [0, d/2], let ℓj ∈ CH (X) be the class of a j-dimensional linear subspace contained in X. Then, for each 0 ≤ i ≤ d, we have
i d Z.h if 0 ≤ i< 2 , i d CH (X) = CHd−i(X)= Z.ℓd−i if 2 < i ≤ d , i d Z.h ⊕ Z.ℓi if i = 2 . If dim X = 2m is even, there are exactly two different classes of m-dimensional linear ′ m m subspaces ℓm, ℓm in CH (X) and the sum of these two classes is equal to h .
The next two propositions can be proved in arbitrary characteristic.
5 Proposition 2.3 ([Kah99, Lemma 8.2]). Let X = Xϕ be a smooth projective quadric of even dimension dim X =2m ≥ 2. m Then CH (X) is a trivial Galois module if and only if e1(ϕ)=0. When e1(ϕ) =06 , ′ the Galois action permutes the two classes ℓm and ℓm. Proposition 2.4 ([Kar90, (2.4)]). Let X be the smooth projective quadric defined by an anisotropic form ϕ with dim ϕ ≥ 3.
1. If dim X ≥ 3, then
CH1(X) = CH1(X)Gal(k/k) = CH1(X)= Z.h .
2. If dim X =2 and e1(ϕ) =06 , then
1 1 Gal(k/k) 1 CH (X)= Z.h = CH (X) ⊆ CH (X)= Z.h ⊕ Z.ℓ1 .
3. If dim X =2 and e1(ϕ)=0 is trivial, then
1 1 Gal(k/k) 1 CH (X)= Z.2ℓ1 ⊕ Z.h ⊆ CH (X) = CH (X)= Z.ℓ1 ⊕ Z.h .
4. If dim X =1, then
1 1 1 Gal(k/k) CH (X)= Z.h = Z.2ℓ0 ⊆ CH (X) = CH (X) = Z.ℓ0 .
3 Unramified cohomology in positive characteristic
Throughout this section, we fix a positive integer r and a field k of characteristic p> 0. i (3.1) For each i ∈ N, let νr(i) = WrΩlog be the i-th logarithmic Hodge–Witt sheaf on r the big ´etale site of k ([Ill79], [Shi07]). Define Z/p (i) := νr(i)[−i], as an object in the derived category of ´etale sheaves. This object can also be viewed as an ´etale motivic complex ([GL00]). For every integer b, we have the cohomology functor on k-schemes
b r b r b−i H · , Z/p (i) := H´et · , Z/p (i) = H´et · , νr(i) .