WorksMathematics
Bloch's conjecture for (anti-)symplectic automorphisms of K3 surfaces
Abstract
We prove Bloch's conjecture for (anti-)symplectic automorphisms of smooth projective complex K3 surfaces of Picard number two. Together with the recent result for abelian surfaces and the known K3 cases, this settles the remaining case of Bloch's conjecture for (anti-)symplectic automorphisms of smooth projective complex surfaces with trivial canonical bundle. More generally, we prove Bloch's conjecture for (anti-)symplectic derived autoequivalences of projective K3 surfaces whose Picard number is at least three, or is two with anisotropic Néron–Severi lattice. The key new result is the congruence property of the subgroup generated by reflective involutions in for anisotropic even lattices of signature . Combined with the local generation theorem and Chow-theoretic criterion of Li–Yu–Zhang, this gives the required reflective decompositions and their consequences for zero-cycles.
Introduction
Throughout, Chow groups are taken with rational coefficients unless otherwise stated. Let be a smooth projective complex K3 surface. The Bloch–Beilinson conjectures predict that the action of a correspondence on zero-cycles is constrained by its action on holomorphic forms. For automorphisms of , this leads to the following formulation of Bloch’s conjecture.
Conjecture 1.1. For any ,
where is the transcendental lattice of .
Equivalently, every symplectic automorphism of satisfies
Here is symplectic if for a nonzero . For projective K3 surfaces, this is equivalent to . An automorphism is anti-symplectic if , equivalently ; the corresponding prediction is on .
Summary of known results
The following results establish the conjecture in several important cases.
Theorem 1.2. Let be a symplectic automorphism of . Then on in each of the following cases:
(1) and [5];
(3) preserves an elliptic fibration on [3];
(4) [8 Theorem 1.3];
(5) and [8 Section 7.2].
Here is the maximal rank of a sublattice whose discriminant is prime to .
We briefly recall the methods underlying these results.
Huybrechts relates the action of spherical twists on Chow groups to reflections in Mukai cohomology. Kneser’s generation theorem then gives the conjecture under the local lattice conditions in Theorem 1.2(1) [5].
Voisin proves the conjecture for symplectic involutions using Prym varieties associated with curves on the quotient surface [13]. Combining Voisin’s theorem with his earlier results, Huybrechts proves the conjecture for symplectic automorphisms of arbitrary finite order [6 Theorem 0.1].
Du–Liu use the relative Jacobian and the group structure of elliptic fibers to treat symplectic automorphisms preserving an elliptic fibration [3]. Their result also applies to automorphisms of infinite order.
Building on Huybrechts’ approach, Li–Yu–Zhang study Bloch’s conjecture for derived autoequivalences. A Fourier–Mukai autoequivalence induces an action
Here is the degree-zero subgroup. The autoequivalence is symplectic (resp. anti-symplectic) if its cohomological action on is (resp. ), equivalently if its restriction to is (resp. ). Li–Yu–Zhang relate its action on zero-cycles to the induced isometry of the algebraic Mukai lattice
Their main criterion is the following.
Theorem 1.3 ([8 Theorem 1.2]). Let be a symplectic (resp. anti-symplectic) derived autoequivalence of . If its action on is a product of reflective involutions, then
The connection with the stable orthogonal group is as follows. Every derived autoequivalence preserves the orientation of the positive two-planes in [7]. Since the Mukai lattice is unimodular, the actions on the discriminant groups of and its orthogonal complement agree under the gluing identification. Consequently,
Here consists of orientation-preserving isometries acting trivially on the discriminant group. For the second implication, preserves the orientation of positive two-planes. Equivalently, is symplectic, and . Thus the anti-symplectic case reduces to the symplectic case by a shift.
For , Li–Yu–Zhang obtain the required decomposition by proving that the reflective subgroup is a congruence subgroup; see Theorem 2.2. When , they also prove Conjecture 1.1 under the assumption [8 Section 7.2].
The remaining automorphism case has . Its Néron–Severi lattice is anisotropic; see Proposition 2.4. We establish the congruence property for every anisotropic even lattice of signature , which also yields a result for derived autoequivalences.
Main result
Our result extends the reflective decomposition method to Picard number two with anisotropic Néron–Severi lattice.
Theorem 1.4. Let be a smooth projective complex K3 surface. Assume that either , or and contains no nonzero isotropic vector. Then every symplectic (resp. anti-symplectic) derived autoequivalence of satisfies
Together with the known cases, this proves the conjecture for (anti-)symplectic automorphisms.
Theorem 1.5. Conjecture 1.1 holds for every smooth projective complex K3 surface. Moreover, if is a symplectic (resp. anti-symplectic) automorphism of , then
Remark 1.6. Theorem 1.5 also holds with integral coefficients. Indeed, Roitman’s theorem [10] implies that the integral degree-zero Chow group of a complex K3 surface is torsion-free, so the rational equalities lift to integral ones.
Remark 1.7. For abelian surfaces, Chen–Li–Zhang [2 Theorem A] prove the corresponding statement for all (anti-)symplectic derived autoequivalences on the Albanese kernel. Together with Theorem 1.5, this establishes Bloch’s conjecture for (anti-)symplectic automorphisms of every smooth projective complex surface with trivial canonical bundle.
The key input is the following statement. We write for the subgroup generated by reflective involutions.
Theorem 1.8. Let be an even lattice of signature , where . If contains no nonzero isotropic vector, then is a congruence subgroup of .
The case follows from [8 Corollary 6.11]. For , we adapt their argument to prove Theorem 1.8 in three steps.
Anisotropy identifies the rational spin group with for a real quadratic field . The corresponding arithmetic lattice is irreducible in .
The inverse image of contains an infinite normal subgroup. Margulis’s normal subgroup theorem gives finite index, and Serre’s theorem gives the congruence property.
Strong approximation for vectors of fixed norm corrects the local signs of spin lifts, giving the congruence property in the orthogonal group.
The local generation theorem of Li–Yu–Zhang then gives the reflective decomposition, and Theorem 1.3 gives the action on zero-cycles. The isotropic example in § 3 explains why the full reflective decomposition need not hold for every lattice of rank two.
Acknowledgements
The author thanks Ruxuan Zhang for helpful discussions, and Zheng Yuan and Zhihong Wang for their kind help. This work was supported by NSFC grant 12425105 and the Shanghai Pilot Program for Basic Research (No. 21TQ001).
History and AI disclosure
Our approach to Bloch’s conjecture was to decompose complicated derived autoequivalences into simpler ones whose actions on zero-cycles are understood. However, carrying out this decomposition directly in the derived category of a K3 surface runs into difficulties at low Picard number. In fact, Example 2.3 gives an obstruction to such a decomposition in general. To address these difficulties, we pursued two approaches over the past two years:
A) Lift the problem to a higher-dimensional hyper-Kähler variety, for instance a variety of -type or O’Grady’s ten-dimensional exceptional hyper-Kähler type. The advantage of this approach is that the increased rank of the relevant lattice ensures that the desired decomposition exists. The difficulty is that the corresponding instances of Bloch’s conjecture are not known for all the reflections occurring in such a decomposition.
B) Restrict the decomposition problem to the actions induced by automorphisms, rather than arbitrary autoequivalences. These actions are represented by certain diagonal matrices, so the aim is to decompose only these matrices into reflective involutions. The difficulty is that we also knew of counterexamples to the desired decomposition for diagonal matrices in general.
The turning point came from an experiment rather than a planned step in the proof. While testing Seed Pro 2.1, I asked whether the reflection group and the diagonal group could have a nontrivial intersection. The model first proposed a counterexample already known to us. I then asked whether it could find one under the additional assumption that the Néron–Severi lattice contains no nonzero vector of square 0, 2, or . For a projective K3 surface of Picard number two, this condition is equivalent to . In response, it drew my attention to Serre’s congruence subgroup theorem.
That suggestion became the starting point of this paper. Building on it, one can easily use our previous approach in [8] to prove that, when the Néron–Severi lattice has rank two and is anisotropic, the associated reflective group is a congruence subgroup. I record this exchange both to acknowledge that contribution and to preserve an unexpected part of the history of the work. LLM were also used to refine and improve the exposition of the manuscript.
Preliminaries
Reflective involutions and Eichler transformations
Let with and . For an even nondegenerate lattice , put and write for its dual lattice and for its discriminant group. Set
For a lattice of signature , let denote the isometries preserving an orientation of the positive two-planes, and put . We recall the reflective involutions of [8 Definition 3.3].
Definition 2.1. Let be the subgroup of generated by the following integral isometries:
for or ;
involutions acting trivially on for which there is a primitive isotropic with and .
The group is normal in . For , it contains the integral Eichler transformations and for ; see [8 Section 6.4]. We use the convention
and define by interchanging and .
Congruence subgroups and reflective decompositions
For an integral lattice , write
A subgroup of is a congruence subgroup if it contains for some integer . We use in the sense of [8 Section 6.4]: it is generated by pairs with and even products of Huybrechts involutions. For , these are and their conjugates by reflections in vectors of square , where ; see [8 Definition 6.6]. These Huybrechts involutions are reflective, so .
The following is a consequence of [8 Lemma 6.1 and Theorem 6.9].
Theorem 2.2. Let be an even lattice of signature with , and put . If is a congruence subgroup of , then
Proof. Every generator of acts trivially on . Conversely, let . Multiplying by if necessary, we may assume that . By [8 Lemma 6.1], has trivial rational spinor norm on . The congruence hypothesis and [8 Theorem 6.9(1′) and (3′)] therefore give . Here changing the sign of the form preserves reflective involutions.
The congruence hypothesis holds when [8 Corollary 6.11], but the resulting equality can fail in Picard number two, as the following example shows.
Example 2.3. Let be a projective complex K3 surface with , and put . In an isotropic basis with and , the isometry
belongs to but not to . The first assertion follows by direct verification. For the second, Appendix A constructs a homomorphism , where contains and , with
Thus has infinite index in . Here is isotropic, so this example lies outside Theorem 1.8.
For anisotropic of rank two, Proposition 3.1 establishes the congruence hypothesis directly.
K3 surfaces of Picard number two
Let be a smooth projective complex K3 surface with , and put . The following criterion is [8 Lemma 7.6].
Proposition 2.4. The group is infinite cyclic if and only if contains no nonzero vector of square , , or .
This identifies the remaining automorphism case in the introduction. For the proof of Theorem 1.5, we use the following consequence of [4 Proposition 3.2].
Proposition 2.5. Let be an infinite-order symplectic automorphism of , and put . Then contains no nonzero isotropic vector or -vector, and
Here denotes the determinant-one isometries preserving a chosen component of the positive cone.
Proof. The assertions concerning and are [4 Proposition 3.2]. Since , the unimodularity of implies that acts trivially on , which is the remaining assertion. □
Normal subgroup and congruence subgroup theorems
We use the following form of Margulis’s normal subgroup theorem; see [9].
Theorem 2.6. Let be an irreducible lattice in . Every infinite normal subgroup of has finite index.
We also use Serre’s congruence subgroup theorem for real quadratic fields [11 p. 499, Corollary 3].
Theorem 2.7. Let be a real quadratic field with ring of integers . Every finite-index subgroup of contains
for some integer $q > 0.
A congruence subgroup theorem in signature (2,2)
Key result
We now establish the congruence property in the remaining case of an anisotropic lattice of rank two.
Proposition 3.1. Let be an anisotropic even lattice of signature , and put . Then is a congruence subgroup of . In particular, is a congruence subgroup of .
Proof. Set and . Since is anisotropic, is not a square. As contains a hyperbolic plane,
Let be the spin covering, and put
Since is a real quadratic field, is an irreducible arithmetic lattice in . The group contains the normal subgroup of generated by lifts of all products with ; conjugation by permutes these products and their lifts. This subgroup is infinite: for with , take
Then has infinite order. Theorem 2.6 gives . Since is commensurable with , Theorem 2.7 implies that is congruence.
It remains to show that contains a principal congruence subgroup of . Since is congruence, choose a finite set of primes and open subgroups such that
Since is finitely generated, its rational spinor norm has finite image, and each nontrivial value is detected at a finite place. By local continuity, we may therefore choose such that every admits a rational lift . Since is an open neighbourhood of the identity in and , increasing further ensures that
It suffices to find with for every tc^{-1}\in\widehat{W}$ by the choice of the , and hence .
To construct , put , so that . The quadric has the rational point , and with isotropic. Strong approximation away from the real place [1 Theorem 1.2], with integrality imposed outside , therefore gives sufficiently close to for every . In the Clifford algebra, set . Then
Since , choosing sufficiently close ensures that for all , as required. Thus . Taking also gives .
Theorem 1.8 follows from Proposition 3.1 when . For , [8 Corollary 6.11] identifies a subgroup of with , which contains a principal congruence subgroup.
Proof of Theorem 1.4
By (2) and the shift argument following it, we may assume that is symplectic. The case is [8 Theorem 1.3]. For , put . Since is again anisotropic of signature , Proposition 3.1 applied to shows that is a congruence subgroup. Theorem 2.2 gives
Theorem 1.3 proves the assertion.
Proof of Theorem 1.5
The cases and with are covered by [8 Theorem 1.3 and Section 7.2]. In Picard number one, is either trivial or . The remaining case has and . Then is anisotropic by Proposition 2.4, and Theorem 1.4 applies to the derived pullback.
Appendix A. A character of an integral orthogonal group
We establish the obstruction asserted in Example 2.3. Realize as the order
Put and .
Lemma A.1. There is a homomorphism such that
Proof. Identify with via the projective class of the bottom row. Use the representatives
For and , assign weight zero to every -edge in the Schreier graph and weight to , where
Reverse edges have opposite weights. At each vertex, the relations in have total weight zero. Hence path weights define a homomorphism with . The weights also satisfy , so and imply that is invariant under conjugation by . Thus defines the required extension. Finally,
has total weight
For with , set and . Let be generated by all such and by . The identity and uniqueness of up to simultaneous sign show that
is a homomorphism.
Proposition A.2. Every reflective involution belongs to , and .
Proof. Put . The normalizer calculation for an Eichler order gives
see [12 Section 23.1, equation (23.1.4)]. Indeed, away from 3 the determinant condition forces the local scalar factor to be a unit; at 3 the additional normalizer coset is represented by .
For a root , put . Then . Now let be an isotropic reflective involution with . By definition, its eigenspace is nondegenerate and intersects in , hence has dimension two. Thus . The left–right description of the special orthogonal group of the determinant form gives
Since , we have for some . Since , the rational line is -invariant, whence with . Neither matrix is scalar: together with and , this would imply . Rescaling both by gives and . The left and right multiplier orders of are both . Applying this to gives . Thus belong to . Since , we have , so they belong to the same coset. If both lie in , then exchanges the two summands of
left–right multiplication by units acts diagonally on , whereas sends to . This contradicts the trivial discriminant action of . Consequently , and . Since is torsion-free, vanishes on every involution in , and consequently on .
The isometry in Example 2.3 is , where
Left multiplication by preserves the orientation of positive two-planes, and the third and fourth columns of are divisible by 9. Thus . Since , we have , and its powers represent infinitely many distinct cosets of in .
References
References
- [1]Yang Cao, Zhizhong Huang, and Runlin Zhang, Effective equidistribution, arithmetic purity of strong approximation, and geometric sieve for affine quadrics, 2025, arXiv:2509.23320v1.arxiv.org/abs/2509.23320
- [2]Zaiyuan Chen, Zhiyuan Li, and Ruxuan Zhang, Bloch’s conjecture for equivalences between twisted abelian surfaces and applications, 2026, arXiv:2606.22323v2.arxiv.org/abs/2606.22323
- [3]Jiabin Du and Wenfei Liu, On symplectic automorphisms of elliptic surfaces acting on CH₀, 2022, arXiv:2103.03052v2.DOI
- [4]Kenji Hashimoto, JongHae Keum, and Kwangwoo Lee, K3 surfaces with Picard number 2, Salem polynomials and Pell equation, J. Pure Appl. Algebra 224 (2020), no. 1, 432–443.arxiv.org/abs/1711.02822
- [5]Daniel Huybrechts, Chow groups and derived categories of K3 surfaces, Current Developments in Algebraic Geometry, Math. Sci. Res. Inst. Publ., vol. 59, Cambridge University Press, Cambridge, 2012, pp. 177–195.
- [6]———, Symplectic automorphisms of K3 surfaces of arbitrary finite order, Math. Res. Lett. 19 (2012), no. 4, 947–951.DOI
- [7]Daniel Huybrechts, Emanuele Macrì, and Paolo Stellari, Derived equivalences of K3 surfaces and orientation, Duke Math. J. 149 (2009), no. 3, 461–507.DOI
- [8]Zhiyuan Li, Xun Yu, and Ruxuan Zhang, Bloch’s conjecture for (anti-)autoequivalences on K3 surfaces, 2024, arXiv:2305.10078v2.arxiv.org/abs/2305.10078
- [9]G. A. Margulis, Discrete subgroups of semisimple lie groups, Ergebnisse der Mathematik und ihrer Grenzgebiete (3), vol. 17, Springer-Verlag, Berlin, 1991.DOI
- [10]A. A. Roitman, The torsion of the group of 0-cycles modulo rational equivalence, Ann. of Math. (2) 111 (1980), no. 3, 553–569.
- [11]Jean-Pierre Serre, Le problème des groupes de congruence pour SL₂, Ann. of Math. (2) 92 (1970), no. 3, 489–527.
- [12]John Voight, Quaternion algebras, Graduate Texts in Mathematics, vol. 288, Springer, Cham, 2021.DOI
- [13]Claire Voisin, Symplectic involutions of K3 surfaces act trivially on CH₀, Doc. Math. 17 (2012), 851–860.