Arxiv:1712.05397V3 [Math.AG]

arXiv:1712.05397v3 [math.AG] 3 May 2018 (1.1) and H o h utetof quotient the for Q ie mohprojective smooth a Given ihvle in values with igof ring rsaln rbnu Fr Frobenius crystalline (1.2) ihvle nthe in values with a rjciedaiy eemnna ait,qarc H K quadric, variety, determinantal duality, projective cal vco pc fagbaccce on cycles algebraic of space -vector ter,tplgclproi ylchmlg,noncommut homology, cyclic periodic topological -theory, crys ∗ ie rm number prime a Given Let Conjecture loi h ihis asi se[ (see Parshin eighties, the in Also nteegte,Biisn(e [ (see Beilinson eighties, the In map class cycle the consider vein, same the In Conjecture Conjecture Conjecture e od n phrases. and words Key Date h uhrwsspotdb S AERAward. CAREER NSF a by supported was author The K ( X ∗ k ( p := ) rv hs xeddcnetrsfrcranlow-dimensio certain for of sc conjectures sections from extended Parshin linear these and prove Further Beilinson of Tate, cases of quadrics. conjectures of new original intersections the complete as in and well conjecture) varieties (as Tate conjectures celebrated the these of of As proof (=HPD). a Duality obtain we Projective Homological under invariant Abstract. a ,2018. 7, May : X A E ELNO N ASI CONJECTURES PARSHIN AND BEILINSON TATE, := tpclWt etr,and vectors, Witt -typical ) Q F H q o the for l T P B T crys eafiiefil fcharacteristic of field finite a be ∗ ai ooooy ntesxis ae[ Tate sixties, the In cohomology. -adic p l ( ( 1. ( X X ( X X ( epoeta h ae elno n asi ojcue ar conjectures Parshin and Beilinson Tate, the that prove We X/W ): ): nrdcinadsaeeto results of statement and Introduction ): Q ): Z Q p ehave We h equality The Z h yl ls map class cycle The ∗ vco usaeo hs lmnswihaefie ythe by fixed are which elements those of subspace -vector h yl ls map class cycle The lnaie algebraic -linearized ( P-NAINEO THE OF HPD-INVARIANCE ∗ p X ( ( olwn [ Following . aecnetr,Biisncnetr,Prhnconjectur Parshin conjecture, Beilinson conjecture, Tate k X l ) )) Z Q 6= ) k Q ∗ ⊗ -scheme ( ihrsett h ueia qiaec relation, equivalence numerical the to respect with l p X W GONÇ TABUADA ALO osdrteascae yl ls map class cycle associated the consider , K −→ ) ( i k Q 13 ( ) X p Z H K oj 0)cnetrdtefollowing: the conjectured 50]) Conj. , −→ ∗ ) K 13 26 l 2 Q ( X -adic X ∗ X o h rsaln ooooygop of groups cohomology crystalline the for oj 1)cnetrdtefollowing: the conjectured 51]) Conj. , 0 = ,Tt’ ojcueamt the admits conjecture Tate’s ], := ewl write will we , ) 1 H pt ainlequivalence, rational to up Q ( crys 2 X ( K = ( W ∗ o every for 1.1 1.2 k ter rusof groups -theory , Z ( ( seWi eafnto,obfl,algebraic orbifold, function, zeta asse-Weil X Q ) k p ) ∗ )[1 l ssurjective is ( )( with ssurjective is ( X ∗ tv leri geometry. algebraic ative ∗ /p )) ) ) Q Fr Gal( 38 h rcinfil of field fraction the ] i / q p ∼ ≥ ojcue h following: the conjectured ] = num oe eetn the extend we more, Z k/k a lblorbifolds. global nal and stacks to hemes 1. p ∗ ftesrn form strong the of n ( ) X . holds , . napplication, an W determinantal ) X Q ( k . . o h (graded) the for h associated the ) Z ∗ p ( ,homologi- e, -version: X e ) Q W / ∼ ( num k X ). , 2 GONÇALO TABUADA

All the above conjectures hold whenever dim(X) 1; see [8, 9, 13, 15, 37]. The conjectures T l(X) and T p(X) hold1 moreover for abelian≤ varieties of dimension 3 and for K3-surfaces; see [26, 41]. Besides these cases (and some other scattered≤ cases), the aforementioned important conjectures remain wide open; consult Theo- rems 1.8 and 1.14 below for a proof of the Tate, Beilinson and Parshin conjectures in several new cases. Recall from 2.1 that a differential graded (=dg) category is a category enriched over dg k-vector§ spaces. As explained in 3, given a smoothA proper dg category in the sense of Kontsevich, the conjectures§ of Tate, Beilinson and Parshin ad- A l p mit noncommutative analogues Tnc( ), Tnc( ), Bnc( ), and Pnc( ), respectively. Examples of smooth proper dg categoriesA includeA finiteA dimensionalA k-algebras of finite global dimension A as well as the canonical dg enhancement perfdg(X) of the category of perfect complexes perf(X) of smooth proper k-schemes X (or, more generally, smooth proper algebraic stacks ); consult [16, 23]. X Theorem 1.3. Given a smooth projective k-scheme X, we have the equivalences: T l(X) T l (perf (X)) T p(X) T p (perf (X)) ⇔ nc dg ⇔ nc dg B(X) B (perf (X)) P (X) P (perf (X)) . ⇔ nc dg ⇔ nc dg Theorem 1.3 shows that the conjectures of Tate, Beilinson and Tate belong not only to the realm of algebraic geometry but also to the broad setting of smooth proper dg categories. Making use of this latter noncommutative viewpoint, we now prove that these celebrated conjectures are invariant under Homological Projective Duality (=HPD); for surveys on HPD we invite the reader to consult [22, 39]. Let X be a smooth projective k-scheme equipped with a line bundle X (1); 0 L ∗ we write X P(V ) for the associated morphism, where V := H (X, X (1)) . Assume that→ the triangulated category perf(X) admits a Lefschetz decompositionL A , A (1),..., A (i 1) with respect to (1) in the sense of [21, Def. 4.1]. h 0 1 i−1 − i LX Following [21, Def. 6.1], let Y be the HP-dual of X, Y (1) the HP-dual line bundle, ∗ L and Y P(V ) the morphism associated to Y (1). Given a linear subspace L ∗ → L ⊥ ⊂ V , consider the linear sections X := X P P(L ) and Y := Y P ∗ P(L). L × (V ) L × (V ) Theorem 1.4 (HPD-invariance). Let X and Y be as above. Assume that XL 2 and YL are smooth , that dim(XL) = dim(X) dim(L) and dim(YL) = dim(Y ) ⊥ l dg − p dg dg − dim(L ), and that the conjecture Tnc(A0 ), resp. Tnc(A0 ), resp. Bnc(A0 ), resp. dg dg Pnc(A0 ), holds, where A0 stands for the dg enhancement of A0 induced from perf (X). Under these assumptions, we have the equivalence T l(X ) T l(Y ), dg L ⇔ L resp. T p(X ) T p(Y ), resp. B(X ) B(Y ), resp. P (X ) P (Y ). L ⇔ L L ⇔ L L ⇔ L ∗ Remark 1.5. (i) Given a generic subspace L V , the sections XL and YL are ⊂ ⊥ smooth, and dim(XL) = dim(X) dim(L) and dim(YL) = dim(Y ) dim(L ). l dg p dg− dg dg − (ii) The conjectures Tnc(A0 ), Tnc(A0 ), Bnc(A0 ), and Pnc(A0 ), hold, in particular, whenever the triangulated category A0 admits a full exceptional collection. (iii) Theorem 1.4 holds more generally when Y is singular. In this case we need to replace Y by a noncommutative resolution of singularities perf (Y ; ), where dg F stands for a certain sheaf of noncommutative algebras over Y (consult [22, F2.4] for details), and conjecture T l(Y ), resp. T p(Y ), resp. B(Y ), resp. P (Y ), § 1As explained in §2.4 below, whenever dim(X) ≤ 3, we have T l(X) ⇔ T p(X) (for every l =6 p). 2 The linear section XL is smooth if and only if the linear section YL is smooth; see [22, page 9]. HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 3

by its noncommutative analogue T l (perf (Y ; )), resp. T p (perf (Y ; )), nc dg F nc dg F resp. B (perf (Y ; )), resp. P (perf (Y ; )). nc dg F nc dg F To the best of the author’s knowledge, Theorem 1.4 is new in the literature. In what follows, we illustrate its strength in the case of two important HP-dualities.

Determinantal duality. Let U1 and U2 be two k-vector spaces of dimensions d1 and d2, respectively, with d1 d2, V := U1 U2, and 0

3In [3, Prop. 3.4 and Thm. 3.5] the authors worked over an algebraically closed ﬁeld of characteristic zero. However, the same proof holds mutatis mutandis over k = Fq. Simply replace the reference [14] concerning the existence of a full strong exceptional collection on perf(Gr(r, U1)) by the reference [5, Thm. 1.3] concerning the existence of a tilting bundle on perf(Gr(r, U1)). The author is grateful to Marcello Bernardara for discussions concerning this issue. 4 GONC¸ALO TABUADA

To the best of the author’s knowledge, Theorem 1.8 is new in the literature. It proves the Tate, Beilinson and Parshin conjectures in several new cases. Here are two families of examples: Example 1.9 (Segre varieties). Let r = 1. Thanks to Theorem 1.8(ii), whenever l p d1 d2 2+dim(L) 1, the conjectures T (XL), T (XL), B(XL), and P (XL), hold. − − ≤ 1 In all these cases, XL is a linear section of the Segre variety and the dimension Zd1,d2 of XL is 2(d2 dim(L)) or 2(d2 dim(L)) + 1. Therefore, for example, by letting d and by− keeping dim(L)− ﬁxed, we obtain inﬁnitely many new examples of 2 → ∞ smooth projective k-schemes XL, of arbitrary high dimension, satisfying the Tate, Beilinson and Parshin conjectures.

Subexample 1.10. Let r = 1, d1 = 4, and d2 = 2. In this particular case, the Segre variety 1 P7 agrees with the rational normal 4-fold scroll S ; see [10, Z4,2 ⊂ 1,1,1,1 Ex. 8.27]. Choose a generic linear subspace L V ∗ of dimension 1 such that the ⊥ 7 ⊂ hyperplane P(L ) P does not contain any 3-plane of the rulling of S1,1,1,1. By combining Example⊂1.9 with [6, Prop. 2.5], we conclude that the rational normal 3-fold scroll XL = S1,1,2 satisﬁes the Tate, Beilinson and Parshin conjectures.

Example 1.11 (Square matrices). Let d1 = d2 = d. Thanks to Theorem 1.8(ii), whenever r2 1 + dim(L) 1, the conjectures T l(X ),T p(X ),B(X ), P (X ) − − ≤ L L L L hold. In all these cases the dimension of XL is 2(dr dim(L)) or 2(dr dim(L))+1. Therefore, for example, by letting d and by− keeping r and− dim(L) ﬁxed, → ∞ we obtain inﬁnitely many new examples of smooth projective k-schemes XL, of arbitrary high dimension, satisfying the Tate, Beilinson and Parshin conjectures.

Subexample 1.12. Let d1 = d2 = 3 and r = 2. In this particular case, the determinantal variety 2 P8 has dimension 7 and its singular locus is the 4-dimensional Z3,3 ⊂ Segre variety 1 2 . Given a generic linear subspace L V ∗ of dimension Z3,3 ⊂ Z3,3 ⊂ 5, the associated smooth linear section XL is 2-dimensional and, thanks to Exam- ple 1.11, it satisfies the Tate, Beilinson and Parshin conjectures. Note that since codim(L⊥)=5 > 4 = dim( 1 ), the subspace P(L⊥) P8 does not intersects the Z3,3 ⊂ singular locus 1 of 2 . Therefore, for all the above choices of L, the associated Z3,3 Z3,3 surface X is a linear section of the determinantal variety 2 . L Z3,3 Veronese-Clifford duality. Let W be a k-vector space of dimension d and X the associated projective space P(W ) equipped with the double Veronese embedding P(W ) P(S2W ), [w] [w w]. Consider the Beilinson’s full exceptional collection perf(X→)= ( 1), 7→ , ⊗(1),..., (d 2) (see [2]) and set i := d/2 and hOX − OX OX OX − i ⌈ ⌉ X ( 1), X if d =2i A0 = A1 = = Ai−2 := X ( 1), X Ai−1 := hO − O i ··· hO − O i ( 1) if d =2i 1. (hOX − i − Under these notations, the category perf(X) admits the Lefschetz decomposition A0, A1(1),..., Ai−1(i 1) with respect to the line bundle X (1) = X (2). Remark h − i l dg p dg L dg O dg 1.5(ii) hence implies the conjectures Tnc(A0 ), Tnc(A0 ), Bnc(A0 ), and Pnc(A0 ). 2 ∗ Let := X P 2 X P(S W ) be the universal hyperplane section, H × (S W ) Q ⊂ × where P(S2W ) P(S2W ∗) stands for the incidence quadric. By construction, theQ ⊂ projection ×q : P(S2W ∗) is a flat quadric fibration. As proved in [20, Thm. 5.4] (see alsoH → [1, Thm. 2.3.6]) the HP-dual Y of X is given by 2 ∗ perfdg(P(S W ); l0(q)) (see Remark 1.5(iii)), where l0(q) stands for the sheaf of even Clifford algebrasC associated to q. C HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 5

2 ∗ Let L S W be a generic linear subspace. On the one hand, XL corresponds to the smooth⊂ complete intersection of the dim(L) quadric hypersurfaces in P(W ) parametrized by L. On the other hand, YL is given by perfdg(P(L); l0(q)|L). Theorem 1.4 yields the following result: C Corollary 1.13. We have the following equivalences: T l(X ) T l (perf (P(L); l (q) )) T p(X ) T p (perf (P(L); l (q) )) L ⇔ nc dg C 0 |L L ⇔ nc dg C 0 |L B(X ) B (perf (P(L); l (q) )) P (X ) P (perf (P(L); l (q) )) . L ⇔ nc dg C 0 |L L ⇔ nc dg C 0 |L Recall that the space of quadrics P(S2W ∗) comes equipped with a canonical 2 ∗ ﬁltration ∆d ∆2 ∆1 P(S W ), where ∆i stands for the closed subscheme of those⊂ ··· singular ⊂ ⊂ quadrics⊂ of corank i. ≥ Theorem 1.14 (Intersection of two quadrics). Let XL be as in Corollary 1.13. Assume that dim(L)=2, that P(L) ∆2 = , and that p =2 when d is odd. Under ∩l ∅ p 6 these assumptions, the conjectures T (XL), T (XL), B(XL), and P (XL), hold. Remark 1.15 (Intersection of even-dimensional quadrics). In the case of an intersection XL of (several) even-dimensional quadrics, we prove in Theorem 7.11 below l p that the conjectures T (XL) (for every l = 2), T (XL), B(XL), and P (XL), are equivalent to the corresponding conjectures6 for the discriminant 2-fold cover of the projective space P(L). To the best of the author’s knowledge, this (geometric) result is new in the literature. The proof of Theorem 1.14 is based on the solution of the corresponding noncommutative conjectures of Corollary 1.13; consult 7 for details. In what concerns Tate’s conjecture, an alternative (geometric) proof,§ based on the notion of variety of maximal planes, was obtained by Reid4 in the early seventies; see [29, Thms. 3.14 and 4.14]. Therein, Reid proved the Hodge conjecture but, as Kahn kindly informed me, a similar proof works for the Tate conjecture. In what concerns the Beilinson and Parshin conjectures, Theorem 1.14 is new in the literature.

Strong form of the Tate conjecture. Given a smooth projective k-scheme X, let us write ords=iζ(X,s) for the order of the pole of the Hasse-Weil zeta function ζ(X,s) of X at i 0, 1,..., dim(X) . In the sixties, Tate [38] also conjectured the following strong form∈{ of the Tate conjecture} 5: i Conjecture ST (X): The equality ord ζ(X,s) = dim (X)Q/ holds. s=i Z ∼num ∗ ∗ ∗ Let us write (X)Q/∼l-adic, resp. (X)Q/∼crys, for the quotient of (X)Q with respect toZ the l-adic homologicalZ equivalence relation, resp. crystallineZ homological equivalence relation. Note that Beilinson’s conjecture B(X) implies that ∗ ∗ ∗ (X)Q/ = (X)Q/ = (X)Q/ . Therefore, making use of [37, Z ∼l-adic Z ∼crys Z ∼num Thm. 2.9], resp. [26, Thm. 1.11], we conclude that T l(X)+ B(X) ST (X), resp. T p(X)+ B(X) ST (X). This implies that the strong form of the⇒ Tate conjecture also holds in the⇒ several new cases provided by Theorems 1.8 and 1.14.

4Reid also assumed in loc. cit. that P(L) ∩ ∆2 = ∅; see [29, Def. 1.9]. 5As proved in [37, Thm. 2.9], resp. [26, Thm. 1.11], the strong form of the Tate conjecture ST (X) implies the Tate conjecture T l(X) (for every l =6 p), resp. the p-version of the Tate conjecture T p(X). 6 GONC¸ALO TABUADA

Tate, Beilinson and Parshin conjectures for stacks. Theorem 1.3 allow us to easily extend the original conjectures of Tate, Beilinson and Parshin from smooth projective k-schemes X to smooth proper algebraic k-stacks by setting X ?( ) :=? (perf ( )) with ? T l,T p,B,P . X nc dg X ∈{ } The next result proves these extended conjectures for “low-dimensional” orbifolds: Theorem 1.16. Let G be a finite group of order s, X a smooth projective k-scheme equipped with a G-action, and := [X/G] the associated global orbifold. If p ∤ s, then we have the following implicationsX (1.17) T l(Xσ Spec(k[σ])) T l( ) (forevery l ∤ s) σ⊆G × ⇒ X (1.18) T p(Xσ Spec(k[σ])) T p( ) Pσ⊆G × ⇒ X (1.19) B(Xσ Spec(k[σ])) B( ) P σ⊆G × ⇒ X (1.20) P (Xσ Spec(k[σ])) P ( ) , Pσ⊆G × ⇒ X where σ is an arbitraryP cyclic subgroup of G. Moreover, whenever s (q 1), resp. dim(X) 3, the k-schemes Xσ Spec(k[σ]) in (1.17)-(1.20), resp. in| (1.17− )-(1.18), can be replaced≤ by the k-schemes× Xσ. Note that the assumption s (q 1) implies that p ∤ s. | − Corollary 1.21. (i) Assume that p ∤ s. If dim(X) 1, then the conjectures T l( ) (for every l ∤ s), T p( ), B( ), and P ( ), hold.≤ (ii) AssumeX 6 that s (q 1).X If dim(XX)=2, thenX T l(X) T l( ) (for every l ∤ s), T l(X) T| l( −), B(X) B( ), and P (X) P ( ⇒). X ⇒ X ⇒ X ⇒ X Example 1.22. Let X be an abelian surface equipped with the Z/2-action a a. Since the conjectures T l(X) and T p(X) hold, Corollary 1.21(ii) implies that7→ − the conjectures T l( ) (for every l = 2) and T p( ) also hold. X 6 X We finish this section with the following “twisted” version of Corollary 1.21: Theorem 1.23. Let G be a finite group of order s, X a smooth projective k-scheme equipped with a G-action, := [X/G] the associated global orbifold, and a G- equivariant sheaf of AzumayaX algebras over S of rank r. Assume that s (qF 1). (i) If dim(X) 1, then the conjectures T l( ; ) (for every l ∤ sr), |T p(− ; ), B( ; ), and≤ P ( ; ), hold. X F X F (ii) If theX FG-action isX faithfulF and dim(X)=2, then T l(X) T l( ; ) (for every l ∤ sr), T p(X) T p( ; ), B(X) B( ; ), and P (X⇒) PX( F; ). ⇒ X F ⇒ X F ⇒ X F 2. Preliminaries n Throughout the article, k := Fq is a finite field of characteristic p with q = p . 2.1. Dg categories. For a survey on dg categories, we invite the reader to consult Keller’s ICM address [16]. Let ( (k), , k) be the category of dg k-vector spaces. A differential graded (=dg) categoryC ⊗is a category enriched over (k) and a dg functor F : is a functor enrichedA over (k). In what follows,C we will write dgcat(k) forA→B the category of (essentially small)C dg categories and dg functors. Let be a dg category. The opposite dg category op has the same objects and op(x,A y) := (y, x). A right dg -module is a dg functorA M : op (k) with A A A A → Cdg 6In the particular case of the (p-version of the) Tate conjecture, it suffices to assume that p ∤ s. HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 7 values in the dg category dg(k) of dg k-vector spaces. Following [16, 3.2], the derived category ( ) of C is defined as the localization of the category§ of right dg -modules (D)A with respectA to the objectwise quasi-isomorphisms. In what A C A follows, we will write c( ) for the triangulated subcategory of compact objects. A dg functor F : D Ais called a Morita equivalence if it induces an equivalence on derived categoriesA→B( ) ( ); see [16, 4.6]. As explained in [30, 1.6], the category dgcat(k) admitsD A a Quillen≃ D B model structure§ whose weak equivalences§ are the Morita equivalences. Let us denote by Hmo(k) the associated homotopy category. The tensor product of dg categories is defined as follows: the set of objects is obj( ) obj( ) andA⊗B ( )((x, w), (y,z)) := (x, y) (w,z). As explained in [16,A 2.3],× thisB constructionA⊗B gives rise to a symmetricA ⊗B monoidal structure on dgcat(k)§ which descends to the homotopy category Hmo(k). op A dg - -bimodule is a dg functor B: dg(k) or, equivalently, a right dg ( op A B)-module. A standard exampleA⊗B is the→C dg - -bimodule A ⊗B A B (2.1) : op (k) (x, z) (z, F (x)) F B A⊗B →Cdg 7→ B associated to a dg functor F : . Following Kontsevich [17, 18, 19], a dg A→B category is called smooth if the dg - -bimodule id belongs to the category ( op A ) and proper if dim Hi A(x,A y) < for anyA pair of objects (x, y). Dc A ⊗ A i A ∞ 2.2. Additive invariants.PLet and be two dg categories and B a dg - - bimodule. Consider the followingA dg categoryB T ( , ;B): the set of objectsA B is obj( ) obj( ), the dg k-vector spaces of morphismsA B are given as follows A ∐ B (x, y) if x, y A ∈ A (x, y) if x, y T ( , ; B)(x, y) := B ∈B A B B(x, y) if x and y  ∈ A ∈B 0 if x and y , ∈B ∈ A  and the composition law is induced by the composition law of and and by the dg - -bimodule structure of B. Note that, by construction, weA haveB canonical dg A B functors ιA : T ( , ; B) and ιB : T ( , ; B). Recall fromA [ →30, Def.A B 2.1] that a functorB → E :A dgcat(B k) D, with values in an additive category, is called an additive invariant if it satisfies→ the following conditions: (i) It sends the Morita equivalences to isomorphisms. (ii) Given , , and B, as above, the dg functos ι and ι induce an isomorphism A B A B E( ) E( ) ∼ E(T ( , ; B)) . A ⊕ B −→ A B Let us write rep( , ) for the full triangulated subcategory of ( op ) consisting of those dg - -modulesA B B such that for every object x theD A associated⊗B right dg -module B(Ax,B ) belongs to ( ). As explained in [30∈, A1.6.3], there is a natural B − Dc B § bijection between HomHmo(k)( , ) and the set of isomorphism classes of the category rep( , ). Under this bijection,A B the composition law of Hmo(k) corresponds to the (derived)A B tensor product of bimodules. Therefore, since the dg - -bimodules (2.1) belong to rep( , ), we have the following symmetric monoidalA functor:B A B (2.2) dgcat(k) Hmo(k) ( F ) . −→ A 7→ A A →B 7→ F B The additivization of Hmo(k) is the additive category Hmo0(k) with the same objects as Hmo(k) and with abelian groups of morphisms Hom ( , ) given by Hmo0(k) A B 8 GONÇALO TABUADA the Grothendieck group K0rep( , ) of the triangulated category rep( , ). As explained in [30, 2.3], the followingA B composition A B § (2.2) (2.3) U : dgcat(k) Hmo(k) Hmo (k) ( F ) [ ] −→ −→ 0 A 7→ A A →B 7→ F B is the universal additive invariant. Moreover, the symmetric monoidal structure of Hmo(k) extends to Hmo0(k), making the above functor (2.3) symmetric monoidal. 2.3. Noncommutative motives. For a book, resp. survey, on noncommutative motives, we invite the reader to consult [30], resp. [31]. Given a commutative ring R, recall from [30, 4.1] that the category of noncommutative Chow motives § NChow(k)R (with R-coefficients) is defined as the idempotent completion of the full subcategory of Hmo (k) consisting of those objects U( ) with a smooth 0 R A R A proper dg category. As explained in loc. cit., the category NChow(k)R is R-linear, additive, and rigid symmetric monoidal. Moreover, we have natural isomorphisms: (2.4) Hom (U( ) ,U( ) ) := K (rep( op )) K ( op ) . NChow(k)R A R B R 0 A ⊗B R ≃ 0 A ⊗B R Given a R-linear, additive, rigid symmetric monoidal category ( , , 1), its -ideal is defined as follows (tr(g f) stands for the categorical trace ofC g⊗ f): N ◦ ◦ (a,b) := f Hom (a,b) g Hom (b,a) we have tr(g f)=0 . N { ∈ C | ∀ ∈ C ◦ } Under these notations, recall from [30, 4.6] that the category of noncommutative § numerical motives NNum(k)R (with R-coefficients) is defined as the idempotent completion of the quotient category NChow(k) / . R N 2.4. Tate conjecture for divisors. Let X be a smooth projective k-scheme of dimension d. Given a prime number l = p, consider the Tate conjecture for divisors: 6 Conjecture T l,1(X): The cycle class map (1.1) with =1 is surjective. ∗ As proved in [37, Prop. 4.1], we have the implication T l,1(X) T l,d−1(X). Consequently, whenever dim(X) 3, we conclude that T l(X) T l,1⇒(X). Consider also the p-version of≤ the Tate conjecture for divisors:⇔ Conjecture T p,1(X): The cycle class map (1.2) with =1 is surjective. ∗ As proved in [27, Prop. 4.1], we have T p,1(X) T l,1(X) (for every l = p). Moreover, a proof similar to the one of [37, Prop.⇔ 5.1], with the commutative6 diagram (2.3) of [37] replaced by the commutative diagram of [27, page 25], shows that T p,1(X) T p,d−1(X). Consequently, whenever dim(X) 3, we conclude that T p(X) ⇒T p,1(X). This implies that whenever dim(X) ≤ 3, we have the equivalence T⇔l(X) T p(X) (for every l = p). ≤ ⇔ 6 3. Noncommutative conjectures Throughout this section, denotes a smooth proper dg category. A Noncommutative Tate conjecture. Given a prime number l = p, consider the following abelian groups 6 ∞ (3.1) Hom Z(l ), π L K( F F m ) m 1 , −1 KU A⊗ q q ≥ ∞ where Z(l ) stands for the Prüfer l-group, K( F F m ) for the algebraic K-theory A⊗ q q spectrum of the dg category F F m , and L K( F F m ) for the Bousfield A⊗ q q KU A⊗ q q m localization of K( Fq Fq ) with respect to topological complex K-theory KU. Note that the abelianA⊗ groups (3.1) can, alternatively, be defined as the l-adic Tate HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 9

m module of the abelian groups π−1LKU K( Fq Fq ),m 1. Under these notations, Tate’s conjecture admits the following noncommutativeA⊗ ≥ analogue: Conjecture T l ( ): The abelian groups (3.1) are trivial. nc A Remark 3.2. Note that the conjecture T l ( ) holds, for example, whenever the nc A abelian groups π L K( F F m ),m 1, are finitely generated. −1 KU A⊗ q q ≥ Noncommutative p-version of the Tate conjecture. By construction, the topological Hochschild homology THH( ) of carries a canonical S1-action. This A A leads naturally to the spectrum of homotopy orbits THH( )hS1 , to the spectrum of 1 A homotopy fixed-points T C−( ) := THH( )hS , and also to the Tate construction 1 A A T P ( ) := THH( )tS . As explained in [28, Cor. I.4.3], these spectra are related by theA following cofiberA sequence

N hS1 can tS1 (3.3) Σ THH( ) 1 THH( ) THH( ) , A hS −→ A −→ A where N stands for the norm map. It is well-known that the abelian groups THH∗( ) are k-linear. Hence, after inverting p, we have Σ THH( )hS1 [1/p] . Consequently,A the above coﬁber sequence (3.3) leads to a canonicalA isomorphism:≃∗ (3.4) can: T C−( ) ∼ T P ( ) . 0 A 1/p −→ 0 A 1/p It is also well-known that the spectrum THH( ) is bounded below, i.e. there exists a integer m 0 such that THH ( ) = 0 foreveryA i

Conjecture B ( ): The equality K ( )Q = K ( )Q/ holds. nc A 0 A 0 A ∼num Noncommutative Parshin conjecture. Parshin’s conjecture admits the following noncommutative analogue:

Conjecture P ( ): We have K ( )Q =0 for every i 1. nc A i A ≥ 4. Proof of Theorem 1.3 As proved by Thomason in [40], the Tate conjecture T l(X) is equivalent to ∞ the vanishing of the abelian groups Hom(Z(l ), π L K(X F F m )),m 1. −1 KU × q q ≥ Therefore, the proof of the equivalence T l(X) T l (perf (X)) follows from the ⇔ nc dg canonical Morita equivalence between the dg categories perf (X F F m ) and dg × q q perf (X) F F m ; consult [34, Lem. 4.26]. dg ⊗ q q Let us now prove the equivalence T p(X) T p (perf (X)). Recall that the ring ⇔ nc dg of p-typical Witt vectors W (k) is the unramiﬁed extension of degree m of the ring of p-adic integers Zp. Hence, we have an induced ﬁeld extension Qp K. Note that the cycle class map (1.2) is surjective if and only if the K-linear homomorphism→

∗ 2∗ Frp (X) Q K H (X)( ) Q K Z ⊗ p −→ crys ∗ ⊗ p is surjective. Therefore, making use of the following natural isomorphisms 1 1 n 1 2∗ Frp 2∗ ∗ Frp 2∗ ∗ n Frp 2∗ ∗ Frq H (X)( ) Q K H (X) p Q K H (X) (p ) = H (X) q , crys ∗ ⊗ p ≃ crys ⊗ p ≃ crys crys HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 11 we conclude that the p-version of the Tate conjecture T p(X) is equivalent to the surjectivity of the induced K-linear cycle class map

1 ∗ Fr (4.1) ∗(X) H2∗ (X) q q . Z K −→ crys On the one hand, since char(K) = 0, recall from [7, 18.3] that we have a natu- ∗ § ral isomorphism K0(perfdg(X))K (X)K . On the other hand, recall from [33, ≃ Z 2∗ Thm. 5.2] that we have a natural isomorphism T P0(perfdg(X))1/p Hcrys(X). Un- der these isomorphisms, the endomorphism ϕn corresponds to the≃ endomorphism 1 q∗ Frq (see [11]) and the homomorphism (4.1) corresponds to the K-linear homomorphism (3.8). Consequently, (4.1) is surjective if and only if (3.8) is surjective. Let us now prove the equivalence B(X) B (perf (X)). Note ﬁrst that since ⇔ nc dg (perf (X)) perf(X), the Euler bilinear pairing is given as follows: Dc dg ≃ χ: K (X) K (X) Z ([ ], [ ]) ( 1)idim Hom ( , [ i]) . 0 × 0 −→ F G 7→ − perf(X) F G − i X ∗ Recall from [7, 19] that an algebraic cycle β (X)Q is numerically equivalent to § ∗ ∈ Z zero if α β = 0 for every α (X)Q. Recall also that we have the isomorphism X · ∈ Z ∼ ∗ (4.2) R τ : K (X)Q (X)Q [ ] ch( ) √Td , 0 −→ Z F 7→ F · X where ch( ) stands for the Chern character of and √TdX for the square root of the Todd classF of X; see [7, 18.3]. Given any twoF perfect complexes , perf(X), the Hirzebruch-Riemann-Roch§ theorem (see [7, Cor. 18.3.1]) yieldsF theG equality∈

(4.3) Eu(π ( ∨ )) = τ([ ∨]) τ([ ]) , ∗ F ⊗OX G F · G ZX where Eu denotes the Euler characteristic and π : X Spec(k) denotes the struc- ∨ → tural morphism of X. Since OX Hom( , ), where Hom( , ) stands for the internal Hom of the rigidF symmetric⊗ G ≃ monoidalF G category perf(X−),− we hence conclude that Eu(π∗(Hom( , )))=(4.3) agrees with χ([ ], [ ]). This implies that the above isomorphism (4.2F) descendsG to the numerical quotients:F G

τ / ∗ (4.4) K (X)Q / (X)Q 0 ∼ Z

∼ / ∗ K (X)Q/ / (X)Q/ . 0 ∼num τ Z ∼num Consequently, the proof of the equivalence B(X) B (perf (X)) follows now ⇔ nc dg from the fact that B(X), resp. Bnc(perfdg(X)), is equivalent to the injectivity of the vertical homomorphism on the right-hand side, resp. left hand-side, of (4.4). Finally, the proof of the equivalence P (X) P (perf (X)) is clear. ⇔ nc dg 5. Proof of Theorem 1.4 By deﬁnition of the Lefschetz decomposition A , A (1),..., A (i 1) , we h 0 1 i−1 − i have a chain of admissible triangulated subcategories Ai−1 A1 A0 with A (r) := A (r). Note that A (r) A . Let a be⊆···⊆ the right orthogonal⊆ r r ⊗ LX r ≃ r r complement to Ar+1 in Ar; these are called the primitive subcategories in [21, 4]. By construction, we have the following semi-orthogonal decompositions: § (5.1) A = a , a ,..., a 0 r i 1 . r h r r+1 i−1i ≤ ≤ − 12 GONC¸ALO TABUADA

As proved in [21, Thm. 6.3] (see also [1, Thm. 2.3.4]), the category perf(Y ) admits a HP-dual Lefschetz decomposition Bj−1(1 j), Bj−2(2 j),..., B0 with respect to (1); as above, we have a chainh of admissible− triangulated− subcate-i LY gories Bj−1 Bj−2 B0. Moreover, the primitive subcategories coincide (via a Fourier-Mukai⊆ ⊆ type ··· functor) ⊆ with those of perf(X) and we have the following semi-orthogonal decompositions: (5.2) B = a , a ,..., a 0 r j 1 . r h 0 1 dim(V )−r−2i ≤ ≤ − Furthermore, the assumptions dim(XL) = dim(X) dim(L) and dim(YL) = dim(Y ) dim(L⊥) imply the existence of semi-orthogonal decompositions− − (5.3) perf(X )= C , A (1),..., A (i dim(V )) L h L dim(V ) i−1 − i ⊥ (5.4) perf(Y )= B (dim(L ) j),..., B ⊥ ( 1), C , L h j−1 − dim(L ) − Li dg where CL is a common (triangulated) category. Let us denote, respectively, by CL , dg adg a Ar , and r , the dg enhancement of CL, Ar, and r, induced from perfdg(XL). ′ dg dg Similarly, let us denote by CL and Br the dg enhancement of CL and Br induced from perfdg(YL). Note that since by assumption the k-schemes XL and YL are smooth (and projective), all the above dg categories are smooth (and proper). Let us now prove the equivalence T l(X ) T l(Y ). Consider the functors L ⇔ L (5.5) E : dgcat(k) Mod(Z) π L K( F F m ) m −→ A 7→ −1 KU A⊗ q q with values in the additive category of abelian groups. Proposition 5.6. The functors (5.5) are additive invariants. Proof. Let F : be a Morita equivalence and m 1 an integer. As proved in A→B ≥ m m m [24, Prop. 7.1], the induced dg functor F Fq Fq : Fq Fq Fq Fq is also a Morita equivalence. Therefore, since algebraic⊗ K-theoryA⊗ inverts→B⊗ Morita equivalences

m m m (see [30, 2.2.1]), the homomorphism K(F Fq Fq ): K( Fq Fq ) K( Fq Fq ) is invertible.§ By deﬁnition of the above functors⊗ (5.5),A⊗ we hence conclude→ B⊗ that the induced group homomorphism Em( ) Em( ) is also invertible. Now, let and be two dg categoriesA → andB B a dg - -bimodule. Following 2.2, we needA to showB that the dg functors ι and ι induceA B an isomorphism § A B (5.7) E ( ) E ( ) E (T ( , ; B)) . m A ⊕ m B −→ m A B m m m Consider the dg categories Fq Fq and Fq Fq and the dg bimodule B Fq Fq . Since algebraic K-theory isA⊗ an additive invariantB⊗ of dg categories, the dg⊗ functors

ιA⊗Fq Fqm and ιB⊗Fq Fqm induce an isomorphism ∼ (5.8) K( F F m ) K( F F m ) K(T ( F F m , F F m ;B F F m )) . A⊗ q q ⊕ B⊗ q q −→ A⊗ q q B⊗ q q ⊗ q q Therefore, by deﬁnition of the above functors (5.5), we conclude from (5.8) that the homomorphism (5.7) is also invertible. This ﬁnishes the proof. Thanks to Proposition 5.6, the functors (5.5) are additive invariants. As explained in [30, Prop. 2.2], this implies that the above semi-orthogonal decompositions (5.3)-(5.4) give rise to direct sum decompositions of abelian groups: (5.9) E (perf (X )) E (Cdg) E (Adg ) E (Adg ) m dg L ≃ m L ⊕ m dim(V ) ⊕···⊕ m i−1 ′ dg dg dg (5.10) E (perf (Y )) E (B ) E (B ⊥ ) E (C ) . m dg L ≃ m j−1 ⊕···⊕ m dim(L ) ⊕ m L HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 13

Consequently, by applying the functor Hom(Z(l∞), ) to the direct sum decompositions (5.9)-(5.10), we obtain the following equivalences− of conjectures: (5.11) T l (perf (X )) T l (Cdg)+ T l (Adg )+ + T l (Adg ) nc dg L ⇔ nc L nc dim(V ) ··· nc i−1 ′ l l dg l dg l dg (5.12) T (perf (Y )) T (B )+ + T (B ⊥ )+ T (C ) . nc dg L ⇔ nc j−1 ··· nc dim(L ) nc L l dg On the one hand, since by assumption the conjecture Tnc(A0 ) holds, we conclude from the above semi-orthogonal decompositions (5.1)-(5.2) that the conjectures T l (Adg) and T l (Bdg), with 0 r i 1, also hold. This implies that the nc r nc r ≤ ≤ − right-hand side of (5.11), resp. (5.12), reduces to the conjecture T l (Cdg), resp. ′ nc L l dg T (C ). On the other hand, since the functor perf(XL) CL perf(YL) is of nc L ′ dg dg → → Fourier-Mukai type, the dg categories CL and CL are Morita equivalent. Using the fact that the functors (5.5) invert Morita equivalences, we hence conclude that ′ l dg l dg Tnc(CL ) Tnc(CL ). Consequently, the proof follows now from the equivalences l ⇔ l l l T (XL) Tnc(perfdg(XL)) and T (YL) Tnc(perfdg(YL)) of Theorem 1.4. ⇔ ⇔p p Let us now prove the equivalence T (XL) T (YL). As explained in [30, Prop. 2.2], since the functor (2.3) is an additive invariant,⇔ the above semi-orthogonal decomposition (5.3) gives rise to the following direct sum decomposition (5.13) U(perf (X )) U(Cdg) U(Adg ) U(Adg ) dg L K ≃ L K ⊕ dim(V ) K ⊕···⊕ i−1 K in the K-linearized category Hmo0(k)K . Recall from the proof of Lemma 3.7 that the functor (3.6) comes equipped with the natural transformation ϕn. Therefore, by applying the K-linear functor (3.6) to the above direct sum decomposition (5.13), we conclude that the induced K-linear homomorphism n K (perf (X )) T P (perf (X ))ϕ 0 dg L K −→ 0 dg L 1/p identiﬁes with the induced (diagonal) K-linear homomorphism n n K (Cdg) i−1 K (Adg) T P (Cdg)ϕ i−1 T P (Adg)ϕ . 0 L ⊕⊕r=dim(V ) 0 r K −→ 0 L 1/p ⊕⊕r=dim(V ) 0 r 1/p This implies the following equivalence of conjectures: T p (perf (X )) T p (Cdg)+ T p (Adg )+ + T p (Adg ) . nc dg L ⇔ nc L nc dim(V ) ··· nc i−1 All the above holds mutatis mutandis with XL replaced by YL. Consequently, the above semi-orthogonal decomposition (5.4) leads to the equivalence of conjectures ′ p p dg p dg p dg T (perf (Y )) T (B )+ + T (B ⊥ )+ T (C ) . nc dg L ⇔ nc j−1 ··· nc dim(L ) nc L The remainder of the proof is now similar to the proof of T l(X ) T l(Y ). L ⇔ L Let us now prove the equivalence B(XL) B(YL). As above, the semi-orthogonal decompositions (5.3)-(5.4) give rise to the following⇔ direct sum decompositions dg dg dg (5.14) U(perf (X ))Q U(C )Q U(A )Q U(A )Q dg L ≃ L ⊕ dim(V ) ⊕···⊕ i−1 ′ dg dg dg (5.15) U(perf (Y ))Q U(B )Q U(B ⊥ )Q U(C )Q dg L ≃ j−1 ⊕···⊕ dim(L ) ⊕ L in the Q-linearized category Hmo0(k)Q. As proved in [32, 6], given any smooth proper dg category , we have a natural isomorphism: § A (5.16) Hom (U(k)Q,U( )Q) K ( )Q/ . NNum(k)Q A ≃ 0 A ∼num 14 GONC¸ALO TABUADA

Hence, by applying Hom (U(k)Q, ) and Hom (U(k)Q, ) to the NChow(k)Q − NNum(k)Q − direct sum decompositions (5.14)-(5.15) we obtain the equivalences of conjectures: B (perf (X )) B (Cdg)+ B (Adg )+ + B (Adg ) nc dg L ⇔ nc L nc dim(V ) ··· nc i−1 ′ dg dg dg B (perf (Y )) B (B )+ + B (B ⊥ )+ B (C ) . nc dg L ⇔ nc j−1 ··· nc dim(L ) nc L l l The remainder of the proof is now similar to the proof of T (XL) T (YL). Finally, let us prove the equivalence P (X ) P (Y ). Consider⇔ the functors L ⇔ L (5.17) K ( )Q : dgcat(k) Vect(Q) K ( )Q i − −→ A 7→ i A with values in the category of Q-vector spaces. As explained in [30, 2.2.1], these functors are additive invariants. Therefore, a proof similar to the one§ of the equivalence T l(X ) T l(Y ) allows us to conclude that P (X ) P (Y ). L ⇔ L L ⇔ L 6. Proof of Proposition 1.6

As proved in [5, Thms. 1.3 and 1.7], the dg category perfdg(Gr(r, U1)) is Morita equivalent to a finite dimensional k-algebra of finite global dimension A. Since dg A0 = perfdg(Gr(r, U1)), we hence obtain the following equivalences of conjectures: T l (Adg) T l (A) T p (Adg) T p (A) nc 0 ⇔ nc nc 0 ⇔ nc B (Adg) B (A) P (Adg) P (A) . nc 0 ⇔ nc nc 0 ⇔ nc l We start by proving the conjecture Tnc(A). Recall that every finite field k is perfect. Therefore, using the fact that the above functors (5.5) are additive invariants, [35, Thm. 3.15] implies that Em(A) Em(A/J(A)),m 1, where J(A) stands for the Jacobson radical of A. By≃ applying the functor≥ Hom(Z(l∞), ) to these latter isomorphisms, we hence conclude that T l (A) T l (A/J(A−)). nc ⇔ nc Now, let us write V1,...,Vs for the simple (right) A/J(A)-modules and D1 := EndA/J(A)(V1),...,Ds := EndA/J(A)(Vs) the associated division k-algebras. Thanks to the Artin-Wedderburn theorem, the quotient A/J(A) is Morita equivalent to D D . Moreover, the center Z of D is a finite field extension of k and D 1 ×···× s i i i is a central simple Zi-algebra. Since the Brauer group of a finite field k is trivial, this implies that D1 Ds is Morita equivalent to Z1 Zs. Consequently, we obtain the following×···× equivalences: ×···× T l (A) T l (A/J(A)) T l (D )+ + T l (D ) T l(Z )+ + T l(Z ) . nc ⇔ nc ⇔ nc 1 ··· nc s ⇔ 1 ··· s l The proof of the conjecture Tnc(A) follows now from the fact that the conjectures l T (Zi), 1 i s, hold because dim(Zi) = 0. ≤ ≤ p Let us now prove the conjecture Tnc(A). Since the functor (2.3) is an additive invariant, [35, Thm. 3.15] implies that U(A)K U(A/J(A))K in Hmo0(k)K . Recall from the proof of Lemma 3.7 that the functor (≃3.6) comes equipped with the natural transformation ϕn. Therefore, by applying the K-linear functor (3.6) to the latter isomorphism, we obtain the following identification: n n K (A) T P (A)ϕ K (A/J(A)) T P (A/J(A))ϕ . 0 K −→ 0 1/p ≃ 0 K −→ 0 1/p p p This implies the equivalence of conjectures Tnc(A) Tnc(A/J(A)). The proof of p ⇔ l the conjecture Tnc(A/J(A)) is now similar to the proof of Tnc(A/J(A)). Let us now prove the conjecture Bnc(A). As above, we have an isomorphism U(A)Q U(A/J(A))Q in Hmo (k)Q. Thanks to the natural isomorphisms (2.4) ≃ 0 and (5.16), by applying Hom (U(k)Q, ) and Hom (U(k)Q, ) to NChow(k)Q − NNum(k)Q − HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 15 the latter isomorphism, we hence conclude that Bnc(A) Bnc(A/J(A)). The proof ⇔ l of the conjecture Bnc(A/J(A)) is now similar to the proof of Tnc(A/J(A)). Finally, since the above functors (5.17) are additive invariants, the proof of the l conjecture Pnc(A) is similar to the proof of Tnc(A).

7. Proof of Theorem 1.14 We assume first that d is even. Following [20, 3.5] (see also [1, 1.6]), let § § Z be the center of l (q) and Spec( ) =: P(L) P(L) the discriminant cover C 0 |L Z → of P(L). As explained in loc. cit., P(L) P(L) is a 2-fold cover which is ram- ified over the divisor D := P(L) ∆ .→ Sincee by assumption dim(L) = 2, we ∩ 1 have dim(D) = 0. Consequently, sincee D is smooth, P(L) is also smooth. Let us write for the sheaf of noncommutative algebras l (q) considered as a sheaf F C 0 |L of noncommutative algebras over P(L). As proved ineloc. cit., since by assump- (d/2)−1 tion P(L) ∆2 = , is a sheaf of Azumaya algebras over P(L) of rank 2 . Moreover,∩ the category∅ F perf(P(L); el (q) ) is equivalent (via a Fourier-Mukai type C 0 |L functor) to perf(P(L); ). This leads to a Morita equivalencee between the dg cat- F egories perfdg(P(L); l0(q)|L) and perfdg(P(L); ). Consequently, making use of Corollary 1.13, wee obtainC the following equivalencesF of conjectures: e (7.1) T l(X ) T l (perf (P(L); )) T p(X ) T p (perf (P(L); )) L ⇔ nc dg F L ⇔ nc dg F (7.2) B(X ) B (perf (P(L); )) P (X ) P (perf (P(L); )) . L ⇔ nc dg e F L ⇔ nc dg e F Since by assumption dim(L) = 2,e the 2-fold cover P(L) is a smoothe projective curve. Using the fact that the Brauer group of every smooth curve over a finite field k is trivial (see [25, page 109]), we hence conclude thate the right-hand side conjectures in (7.1)-(7.2) are equivalent to T l(P(L)), T p(P(L)), B(P(L)), and P (P(L)), respectively. The proof follows now from the fact that the Tate, Beilinson and Parshin conjectures hold for smooth projectivee curves.e e e We now assume that d is odd and that p = 2. Following [20, 3.6] (see also 6 § [1, 1.7]), let P(L) be the discriminant stack associated to the pull-back q along § |L P(L) P(S2W ∗) of the flat quadric fibration q : P(S2W ∗). As explained in ⊂ H → loc. cit., sinceb by assumption 1/2 k, P(L) is a smooth Deligne-Mumford stack. ∈ Moreover, using the fact that P(L) is a square root stack and that the critical locus of the flat quadric fibration q|L is the divisorb D, we conclude from [12, Thm. 1.6] that perf(P(L)) = perf(D), perf(b P(L)) . Consequently, an argument similar to the one used in the proofh of Theorem 1.4 yieldsi the following equivalences of conjectures: b (7.3) T l (perf (P(L))) T l(D)+ T l(P(L)) nc dg ⇔ (7.4) T p (perf (P(L))) T p(D)+ T p(P(L)) nc dg b ⇔ (7.5) B (perf (P(L))) B(D)+ B(P(L)) nc dg b ⇔ (7.6) P (perf (P(L))) P (D)+ P (P(L)) . nc dg b ⇔ Let us write for the sheaf of noncommutative algebras l (q) considered as a F b C 0 |L sheaf of noncommutative algebras over P(L). As proved in [20, 3.6] (see also [1, 1.7]), since by assumption P(L) ∆ = , is a sheaf of Azumaya§ algebras over § ∩ 2 ∅ F P(L). Moreover, the category perf(P(L); bl (q) ) is equivalent (via a Fourier-Mukai C 0 |L b 16 GONÇALO TABUADA type functor) to perf(P(L); ). This leads to a Morita equivalence between the F dg categories perfdg(P(L); l0(q)|L) and perfdg(P(L); ). Making use of Corollary 1.13, we hence obtain theb followingC equivalences of conjectures:F b (7.7) T l(X ) T l (perf (P(L); )) T p(X ) T p (perf (P(L); )) L ⇔ nc dg F L ⇔ nc dg F (7.8) B(X ) B (perf (P(L); )) P (X ) P (perf (P(L); )) . L ⇔ nc dg b F L ⇔ nc dg b F Since by assumption dim(L) = 2, we have dim(P(L)) = 1. Using the fact that the b b Brauer group of every smooth curve over a finite field k is trivial, we hence conclude that in (7.7)-(7.8) we can replace perf (P(L); ) by perf (P(L)). Consequently, dg F dg since dim(D) = 0, the proof follows now from the combination of (7.3)-(7.6) with the fact that the Tate, Beilinson and Parshinb conjectures holdb in dimensions 1. ≤ Intersection of even-dimensional quadrics. Given a smooth proper dg category , a prime number l = p, and an integer s 1, consider the Z[1/s]-modules A 6 ≥ ∞ (7.9) Hom(Z(l ), (π L K( F F m )) ) m 1 −1 KU A⊗ q q 1/s ≥ as well as the following variant of the noncommutative Tate conjecture: Conjecture T l ( ;1/s): The Z[1/s]-modules (7.9) are trivial. nc A Lemma 7.10. We have T l ( ) T l ( ;1/s) for every l ∤ s. nc A ⇔ nc A Proof. Since by assumption l ∤ s, the localization homomorphisms

π L K( F F m ) (π L K( F F m )) m 1 −1 KU A⊗ q q −→ −1 KU A⊗ q q 1/s ≥ induce an isomorphism between all the l-power torsion subgroups. Consequently, by passing to the l-adic Tate modules, we conclude that the abelian groups (3.1) are trivial if and only if the Z[1/s]-modules (7.9) are trivial.

Theorem 7.11. Let X be as in Corollary 1.13. Assume that P(L) ∆ = , that L ∩ 2 ∅ the divisor P(L) ∆1 is smooth, and that d is even. Under these assumptions, we have the following∩ equivalences: T l(X ) T l(P(L)) (for every l = 2) L ⇔ 6 p p T (XL) T (P(L)) B(XL) B(P(L)) P (XL) P (P(L)) . ⇔ e ⇔ ⇔ Proof. As explained in the proof of Theorem 1.14, we have the equivalences e e e (7.12) T l(X ) T l (perf (P(L); )) T p(X ) T p (perf (P(L); )) L ⇔ nc dg F L ⇔ nc dg F (7.13) B(X ) B (perf (P(L); )) P (X ) P (perf (P(L); )) , L ⇔ nc dg e F L ⇔ nc dg e F (d/2)−1 where is a certain sheaf of Azumayae algebras over P(L) of ranke 2 . WeF start by proving that the first right-hand side conjecture in (7.12) is equivalent to T l(P(L)) (for every l = 2). Consider the followinge functors 6 E ( ) : dgcat(k) Mod(Z[1/2]) (π L K( F F m )) m −e 1/2 −→ A 7→ −1 KU A⊗ q q 1/2 with values in the additive category of Z[1/2]-modules. Similarly to the proof of Proposition 5.6, these functors are additive invariants. Consequently, using the fact that the rank of the sheaf of Azumaya algebras is a power of 2 and that the category Mod(Z[1/2]) is Z[1/2]-linear, we concludeF from [35, Thm. 2.1] HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 17 that E (perf (P(L))) E (perf (P(L); )) . By applying the functor m dg 1/2 ≃ m dg F 1/2 Hom(Z(l∞), ) to these isomorphisms, we hence obtain the equivalences: − e e (7.14) T l (perf (P(L); ); 1/2) T l (perf (P(L)); 1/2)) T l(P(L); 1/2) . nc dg F ⇔ nc dg ⇔ Thanks to Lemma 7.10, the preceding equivalence (7.14) yields the equivalence e e e T l (perf (P(L); )) T l(P(L)) (for every l = 2), and so the proof is finished. nc dg F ⇔ 6 Let us now prove that the second right-hand side conjecture in (7.12) is equivalent to T p(Pe(L)). Since the functore (2.3) is additive and char(K) = 0, [35, Thm. 2.1] implies that U(perfdg(P(L)))K U(perfdg(P(L); ))K in Hmo0(k)K . Recall from the proof ofe Lemma 3.7 comes equipped≃ with the naturalF transformation ϕn. There- fore, by applying the Ke-linear functor (3.6) toe the latter isomorphism, we conclude that the induced K-linear homomorphism n K (perf (P(L))) T P (perf (P(L)))ϕ 0 dg K −→ 0 dg 1/p identifies with the induced K-linear homomorphism e e n K (perf (P(L); )) T P (perf (P(L); ))ϕ . 0 dg F K −→ 0 dg F 1/p This implies the following equivalences of conjectures: e e T p (perf (P(L); )) T p (perf (P(L))) T p(P(L)) . nc dg F ⇔ nc dg ⇔ Let us now prove that the first right-hand side conjecture in (7.13) is equivalent to e e e B(P(L)). As above, we have U(perf (P(L))Q U(perf (P(L); ))Q in Hmo (k)Q. dg ≃ dg F 0 Thanks to the natural isomorphisms (2.4) and (5.16), by applying the functors

Home NChow(k)Q (U(k)Q, ) and HomNNum(e k)Q (U(k)Q, ) toe the latter isomorphism, we hence obtain the following− equivalences of conjectures:− B (perf (P(L); )) B (perf (P(L))) B(P(L)) . nc dg F ⇔ nc dg ⇔ Finally, since the functors (5.17) are additive invariants (with values in the additive category of Q-vector spaces),e the proof that the seconde right-hande side conjecture in (7.13) is equivalent to P (P(L)) is similar to the proof that the ﬁrst right-hand side conjecture in (7.12) is equivalent to T l(P(L)) (for every l = 2). e 6 Remark 7.15 (Azumaya algebras). Let X be a smooth projective k-scheme and e a sheaf of Azumaya algebras over X of rank r. Note that an argument similar toF the one used in the proof of Theorem 7.11 leads to the following equivalences: T l (perf (X; )) T l(X) (for every l = r) nc dg F ⇔ 6 T p (perf (X; )) T p(X) nc dg F ⇔ B (perf (X; ))) B(X) nc dg F ⇔ P (perf (X; ))) P (X) . nc dg F ⇔ 8. Proof of Theorem 1.16 We start by proving Theorem 1.16 in what regards the Tate conjecture. Consider the following functors with values in the additive category of Z[1/s]-modules:

(8.1) E ( ) : dgcat(k) Mod(Z[1/s]) (π L K( F F m )) . m − 1/s −→ A 7→ −1 KU A⊗ q q 1/s Similarly to the proof of Proposition 5.6, these functors are additive invariants. Consequently, using the fact that 1/s k (since p ∤ s) and that category Mod(Z[1/s]) is Z[1/s]-linear, we conclude from [36,∈ Thm. 1.1 and Rk. 1.4] that E (perf ( )) m dg X 1/s 18 GONÇALO TABUADA is a direct summand of E (perf (Xσ Spec(k[σ]))) . By applying the σ⊆G m dg × 1/s functor Hom(Z(l∞), ), we hence obtain the following implication: − L (8.2) T l(Xσ Spec(k[σ]); 1/s) T l( ;1/s) . × ⇒ X σX⊆G The searched implication (1.17) follows then from the combination of (8.2) with Lemma 7.10. Assume now that s (q 1). Note that s (q 1) if and only if k contains the sth roots of unity. Therefore,| − since 1/s k,| since− the above functors (8.1) are additive invariants, and since the category∈ Mod(Z[1/s]) is Z[1/s]-linear, we conclude from [36, Cor. 1.6(i)] that E (perf ( )) is a direct summand of m dg X 1/s E (perf (Xσ))⊕rσ , where r 1 are certain integers. By applying the σ⊆G m dg 1/s σ ≥ functor Hom(Z(l∞), ), we hence obtain the following implication: L − (8.3) T l(Xσ;1/s) T l( ;1/s) . ⇒ X σX⊆G As above, the searched implication T l(Xσ) T l( ) (for every l ∤ s) fol- σ⊆G ⇒ X lows then from the combination of (8.3) with Lemma 7.10. Assume finally that P dim(X) 3. In this case, the dimension of the smooth projective k-schemes Xσ and Xσ ≤ Spec(k[σ]) is also 3. Therefore, as explained in 2.4, it suffices to consider× the Tate conjecture for≤ divisors T l,1( ). As proved in [§37, Thm. 5.2], we have the following equivalences: − T l,1(Xσ Spec(k[σ])) T l,1(Xσ)+ T l,1(Spec(k[σ])) T l,1(Xσ) . × ⇔ ⇔ This yields the equivalence of conjectures T l(X Spec(k[σ])) T l(Xσ) and hence the searched implication T l(Xσ) T l( ×) (for every l ⇔∤ s). σ⊆G ⇒ X Let us now prove Theorem 1.16 in what regards the p-version of the Tate conjecture. Since the functor (2.3P) is additive, 1/s k, and char(K) = 0, [36, Thm. 1.1 and Rk. 1.4] implies that U(perf ( )) is∈ a direct summand of the direct sum dg X K U(perf (Xσ Spec(k[σ]))) . Recall from the proof of Lemma 3.7 that the σ⊆G dg × K functor (3.6) comes equipped with the natural transformation ϕn. Therefore, since L the dg categories perf ( ) and perf (Xσ Spec(k[σ])) are smooth proper, by dg X dg × applying the functor (3.6) we conclude that the induced K-linear homomorphism n K (perf ( )) T P (perf ( ))ϕ 0 dg X K −→ 0 dg X 1/p is a direct summand of the induced (diagonal) K-linear homomorphism n K (Xσ Spec(k[σ])) T P (Xσ Spec(k[σ]))ϕ . 0 × K −→ 0 × 1/p σM⊆G σM⊆G This yields the searched implication (1.18) Assume now that s (q 1). In this case, th | − since k contains the s roots of unity, [36, Cor. 1.6(i)] implies that U(perfdg( ))K is σ rσ X a direct summand of σ⊆G U(perfdg(X ))K , where rσ are certain integers. Hence, an argument similar to the preceding one shows that T p(Xσ) T p( ). L σ⊆G ⇒ X Assume finally that dim(X) 3. As explained in 2.4, since the dimension of Xσ P and Xσ Spec(k[σ]) is 3, we≤ have T p(Xσ) T l(X§ σ) and T p(Xσ Spec(k[σ])) T l(Xσ ×Spec(k[σ])). Consequently,≤ making⇔ use of the above equivalence× of conjec-⇔ tures T×l(X Spec(k[σ])) T l(Xσ), we conclude that T p(Xσ) T p( ). × ⇔ σ⊆G ⇒ X Let us now prove Theorem 1.16 in what regards the Beilinson conjecture. As P σ above, U(perf ( ))Q is a direct summand of U(perf (X Spec(k[σ])))Q. dg X σ⊆G dg × Therefore, thanks to the natural isomorphisms (2.4) and (5.16), by applying the L HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 19 functors Hom (U(k)Q, ) and Hom (U(k)Q, ), we obtain the im- NChow(k)Q − NNum(k)Q − plication (1.19). Once again as above, whenever m (q 1), U(perfdg( ))Q is a σ ⊕rσ | − X direct summand of σ⊆G U(perfdg(X ))Q . Hence, an argument similar to the preceding one shows that B(Xσ) B( ). L σ⊆G ⇒ X Finally, since the functors (5.17) are additive invariants (with values in the ad- P ditive category of Q-vector spaces), the proof of Theorem 1.16 in what regards the Parshin conjecture is similar to the above proof of Theorem 1.16 in what regards the Tate conjecture.

Proof of Theorem 1.23 We start by proving Theorem 1.23 in what regards the Tate conjecture. Since by assumption s (q 1), k contains the sth roots of unity and 1/s k. There- fore, using the fact| that− the above functors (8.1) (with s replaced by∈sr) are additive invariants and that the category Mod(Z[1/sr]) is Z[1/sr]-linear, we conclude from [36, Cor. 1.29(ii)] that Em(perfdg( ; ))1/sr is a direct summand of X F∨ σ σ⊆G Em(perfdg(Yσ))1/sr, where Yσ is a certain σ -Galois cover of X . By applying the functor Hom(Z(l∞), ), we hence obtain the following implication: L − T l(Y ;1/sr) T l( ; ;1/sr) . σ ⇒ X F σX⊆G Lemma 7.10 yields then the following implication: (8.4) T l(Y ) T l( ; ) (forevery l ∤ sr) . σ ⇒ X F σX⊆G On the one hand, if dim(X) 1, then the dimension of Yσ is also 1 for every σ G. Consequently, the conjecture≤ T l( ; ) follows from the implication≤ (8.4). ⊆ X F On the other hand, if the G-action is faithful and dim(X) = 2, then dim(Yσ) = dim(Xσ) 1 for every non-trivial cyclic subgroup σ G. Consequently, in the ≤ ⊆ above implication (8.4) the k-schemes Yσ can be replaced by the single k-scheme X. Let us now prove Theorem 1.23 in what regards the p-version of the Tate conjecture. Since the functor (2.3) is additive, k contains the sth roots of unity, and char(K) = 0, [36, Cor. 1.29(ii)] implies that U(perf ( ; )) is a direct summand dg X F K of σ⊆G U(perfdg(Yσ))K . Recall from the proof of Lemma 3.7 that the functor (3.6) comes equipped with the natural transformation ϕn. Therefore, since the dg L categories perf ( ; ) and perf (Y ) are smooth proper, by applying the functor dg X F dg σ (3.6) we conclude that the induced K-linear homomorphism n K (perf ( ; )) T P (perf ( ; ))ϕ 0 dg X F K −→ 0 dg X F 1/p is a direct summand of the induced (diagonal) K-linear homomorphism n K (perf (Y )) T P (perf (Y ))ϕ . 0 dg σ K −→ 0 dg σ 1/p σM⊆G σM⊆G This yields the implication T p(Y ) T p( ; ). The remainder of the σ⊆G σ ⇒ X F proof is now similar to the above proof, concerning the Tate conjecture, with the implication (8.4) replaced byP the latter implication. Let us now prove Theorem 1.23 in what regards the Beilinson conjecture. As above, U(perf ( ; ))Q is a direct summand of U(perf (Y ))Q. Therefore, dg X F σ⊆G dg σ thanks to the natural isomorphisms (2.4) and (5.16), by applying the functors L 20 GONC¸ALO TABUADA

Hom (U(k)Q, ) and Hom (U(k)Q, ), we obtain the implication NChow(k)Q − NNum(k)Q − B(Y ) B( ; ). The remainder of the proof is now similar to the above σ⊆G σ ⇒ X F proof, concerning the Tate conjecture, with (8.4) replaced by the latter implication. P Finally, since the functors (5.17) are additive invariants (with values in the additive category of Q-vector spaces), the proof of Theorem 1.23 in what regards the Parshin conjecture is similar to the above proof of Theorem 1.23 in what regards the Tate conjecture. Acknowledgments: I am grateful to Bruno Kahn for his interest in the arguments used in the proof of Theorem 1.14 and for the reference [29], to Kiran Kedlaya for asking me if the p-version of the Tate conjecture would admit a noncommutative analogue, and to Lars Hesselholt and Peter Scholze for useful discussions concerning topological periodic cyclic homology. I also would like to thank the Hausdorﬀ Research Institute for Mathematics in Bonn for its hospitality.

References [1] A. Auel, M. Bernardara and M. Bolognesi, Fibrations in complete intersections of quadrics, Clifford algebras, derived categories, and rationality problems. J. Math. Pures Appl. (9) 102 (2014), no. 1, 249–291. [2] A. Beilinson, Coherent sheaves on Pn and problems in linear algebra. Funktsional. Anal. i Prilozhen. 12 (1978), no. 3, 68–69. [3] M. Bernardara, M. Bolognesi and D. Faenzi, Homological projective duality for determinantal varieties. Adv. Math. 296 (2016), 181–209. [4] B. Bhatt, M. Morrow and P. Scholze, Topological Hochschild homology and integral p-adic Hodge theory. Available at arXiv:1802.03261. [5] R. Buchweitz, G. Leuschke and M. Van den Bergh, On the derived category of Grassmannians in arbitrary characteristic. Compos. Math. 151 (2015), no. 7, 1242–1264. [6] A. Conca and D. Faenzi, A remark on hyperplane sections of rational normal scrolls. Avail- able at arXiv:1709.08332. [7] W. Fulton, Intersection theory. Ergebnisse der Mathematik und ihrer Grenzgebiete (3), 2. Springer-Verlag, Berlin, 1984. [8] D. Grayson, Finite generation of K-groups of a curve over a finite field (after Daniel Quillen). Algebraic K-theory, Part I (Oberwolfach, 1980), pp. 69–90, LNM 966, 1982. [9] G. Harder, Die Kohomologie S-arithmetischer Gruppen über Funktionenkörpern. Invent. Math. 42 (1977), 135–175. [10] J. Harris, Algebraic geometry. A first course. Graduate Texts in Mathematics, 133. Springer- Verlag, New York, 1992. [11] L. Hesselholt, Topological periodic cyclic homology and the Hasse-Weil zeta function. Avail- able at arXiv:1602.01980. [12] A. Ishii and K. Ueda, The special McKay correspondence and exceptional collections. Tohoku Math. J. (2) 67 (2015), no. 4, 585–609. [13] B. Kahn, Algebraic K-theory, algebraic cycles and arithmetic geometry. Handbook of Alge- braic K-theory, Berlin, New York. Springer-Verlag, pp. 351–428, 2005. [14] M. Kapranov, On the derived categories of coherent sheaves on some homogeneous spaces. Invent. Math. 92 (1988), no. 3, 479–508. [15] K. Kato and S. Saito, Unramified class field theory of arithmetical surfaces. Ann. of Math. 118 (1983), 241–275. [16] B. Keller, On differential graded categories. International Congress of Mathematicians (Madrid), Vol. II, 151–190. Eur. Math. Soc., Zürich (2006). [17] M. Kontsevich, Mixed noncommutative motives. Talk at the Workshop on Homological Mirror Symmetry, Miami, 2010. Notes available at www-math.mit.edu/auroux/frg/miami10-notes. [18] , Notes on motives in finite characteristic. Algebra, arithmetic, and geometry: in honor of Yu. I. Manin. Vol. II, 213–247, Progr. Math., 270, Birkhäuser Boston, MA, 2009. [19] , Noncommutative motives. Talk at the IAS on the occasion of the 61st birthday of Pierre Deligne (2005). Available at http://video.ias.edu/Geometry-and-Arithmetic. HPD-INVARIANCE OF THE TATE, BEILINSON AND PARSHIN CONJECTURES 21

[20] A. Kuznetsov, Derived categories of quadric fibrations and intersections of quadrics. Adv. Math. 218 (2008), no. 5, 1340–1369. [21] , Homological projective duality. Publ. Math. IHES,´ no. 105 (2007), 157–220. [22] , Semiorthogonal decompositions in algebraic geometry. Available at arXiv:1404.3143. To appear in Proceedings of the International Congress of Mathematicians (2014), Seoul. [23] V. Lunts and D. Orlov, Uniqueness of enhancement for triangulated categories. J. Amer. Math. Soc. 23 (2010), no. 3, 853–908. [24] M. Marcolli and G. Tabuada, Noncommutative Artin motives. Selecta Math. (N.S.) 20 (2014), no. 1, 315–358. [25] J. Milne, Etale´ cohomology. Princeton Mathematical Series, 33 (1980). [26] , The Tate conjecture over finite fields. AIM talk. Available at Milne’s personal web- page http://www.jmilne.org/math/articles/2007e.pdf. [27] M. Morrow, A variational Tate conjecture in crystalline cohomology. Available at arXiv:1408.6783. [28] T. Nikolaus and P. Scholze, On topological cyclic homology. Available at arXiv:1707.01799 [29] M. Reid, The complete intersection of two or more quadrics. Ph. D. thesis. Available at https://homepages.warwick.ac.uk/~masda/3folds/qu.pdf. [30] G. Tabuada, Noncommutative Motives. With a preface by Yuri I. Manin. University Lecture Series, 63. American Mathematical Society, Providence, RI, 2015. [31] , Recent developments on noncommutative motives. Available at arXiv:1611.05439. To appear in Contemporary Mathematics, AMS. [32] , Noncommutative motives in positive characteristic and their applications. Available at arXiv:1707.04248. [33] , On Grothendieck’s standard conjectures of type C and D in positive characteristic. Available at arXiv:1710.04644. [34] G. Tabuada and M. Van den Bergh, The Gysin triangle via localization and A1-homotopy invariance. Transactions of the American Mathematical Society 370 (2018), no. 1, 421–446. [35] , Noncommutative motives of Azumaya algebras. J. Inst. Math. Jussieu 14 (2015), no. 2, 379–403. [36] , Additive invariants of orbifolds. Available at arXiv:1612.03162. To appear in Geom- etry and Topology. [37] J. Tate, Conjectures on algebraic cycles in l-adic cohomology. Motives (Seattle, WA, 1991), 71–83, Proc. Sympos. Pure Math., 55, Part 1, Amer. Math. Soc., Providence, RI, 1994. [38] , Algebraic cycles and poles of zeta functions. Arithmetical Algebraic Geometry (Proc. Conf. Purdue Univ., 1963) pp. 93–110. Harper & Row, New York 1965. [39] R. Thomas, Notes on HPD. Available at arXiv:1512.08985. To appear in Proceedings of the 2015 AMS Summer Institute, Salt Lake City. [40] R. Thomason, A finiteness condition equivalent to the Tate conjecture over Fq. Algebraic K-theory and algebraic number theory (Honolulu, HI, 1987), 385–392, Contemp. Math., 83, Amer. Math. Soc., Providence, RI, 1989. [41] B. Totaro, Recent progress on the Tate conjecture. Bull. Amer. Math. Soc. 54 (2017), 575–590.

Gonc¸alo Tabuada, Department of Mathematics, MIT, Cambridge, MA 02139, USA E-mail address: [email protected] URL: http://math.mit.edu/~tabuada