WorksMathematics
Counterexamples to torsion-freeness in the -theory of crystallographic group -algebras
Abstract
We give a negative answer to a question of Sánchez Saldaña–Velásquez and Lück asking whether the topological K-theory of the reduced group C*-algebra of a split crystallographic group with finite cyclic holonomy is always torsion-free. For two explicit split crystallographic groups with translation subgroup of rank four and holonomy of order four, we compute the equivariant K-theory of the classifying space for proper actions and find nontrivial two-torsion. The equivariant universal coefficient theorem transfers this torsion to proper equivariant K-homology, and the Baum–Connes conjecture identifies the resulting groups with the topological K-theory of the corresponding reduced group C*-algebras.
Introduction
The computation of the topological -theory of group -algebras of crystallographic groups has been studied by several authors under various hypotheses on the holonomy action. Davis and Lück [2] considered groups , where is prime and the conjugation action of on is free outside the origin. They computed the real and complex topological -theory of the corresponding group -algebras by computing the equivariant -homology of the classifying spaces for proper actions and using the Baum–Connes conjecture. Langer and Lück [5] extended this calculation from prime-order holonomy to arbitrary finite cyclic holonomy , under the assumption that the action is free outside the origin.
More recently, Sánchez Saldaña and Velásquez [10] removed the freeness hypothesis when the holonomy order is square-free. In particular, they proved that
is a finitely generated torsion-free abelian group for every action of on when is square-free. They asked whether the same torsion-freeness conclusion holds for arbitrary finite cyclic holonomy. Lück formulates the same problem as Question 17.50 on page 548 of [7]. The examples below give a negative answer: for two actions of on , the -theory of the reduced group -algebra contains nontrivial 2-torsion.
Sánchez Saldaña and Velásquez observed that counterexamples to a cohomological conjecture of Adem–Ge–Pan–Petrosyan might provide a source of counterexamples to torsion-freeness. We follow precisely this idea. The two order-four integral representations used here were exhibited by Petrosyan and Putrycz [9] in their study of split crystallographic groups for which the collapsing of the canonical Lyndon–Hochschild–Serre spectral sequence fails.
Our calculation is done through two reductions. First, each four-dimensional group is written as , where is a common three-dimensional crystallographic group. Viewing the corresponding classifying space for proper actions as a mapping torus gives a Mayer–Vietoris sequence, which reduces the calculation for to the equivariant -theory of and the action induced by . Second, the group is expressed as the amalgam
of two planar crystallographic groups over a common planar subgroup. The Mayer–Vietoris sequence for this amalgam reduces the computation of to the restriction map from p4 to p2. We compute this restriction using explicit equivariant line bundles on the square two-torus. The equivariant universal coefficient theorem and Baum–Connes conjecture then identify the above computations with the -theory groups of the reduced group -algebras.
We now state the two examples and fix our conventions. Let
These order-four matrices occur in Remark 3 and equation (3.1), respectively, of [9].
We write for complex equivariant topological -cohomology of a proper -space, and for proper equivariant -homology. In degree zero, on a finite proper -CW-complex, the former is the Grothendieck group of finite-dimensional equivariant complex vector bundles [6].
Theorem 1.1. For the classifying space for proper actions ,
Consequently,
The second table also gives , since the Baum–Connes assembly map is an isomorphism for these groups.
Reduction to planar stabilizers
Put
Let be the three canonical generators of and the generator of expressed as the order four rotation in the first two coordinates and the reflection for the last one. Thus
In order to relate the groups with let us consider the integer changes of lattice basis
give
The last coordinate is fixed by the holonomy and defines a surjection with kernel . Conjugation by the fourth element of the canonical basis yields
and both automorphisms fix , , .
Write points of as , with and . Project the affine action of onto the last coordinate. On this line, , act trivially, acts by , and by . With vertices at , the quotient is an edge . Its stabilizers are
where
There are no edge inversions. The associated graph of groups gives . Since , both edge inclusions are the standard inclusion under these identifications.
Replacing each vertex and edge by its classifying space of proper actions produces , with
and , , acting by translations.
Recall that a proper -CW-complex is a model for if and only if is contractible for every finite subgroup [8 Theorem 1.9]. The affine actions considered here are proper and cocompact. A finite subgroup fixes the average of an orbit, and its fixed-point set is an affine subspace, hence contractible. Thus these are models for the classifying spaces for proper actions. They admit equivariant triangulations with finitely many cell orbits, as does the original -action on .
K-theory of planar groups
Let
With acting on by , for every . Taking the quotient by the freely acting normal subgroup we may identify
The same argument identifies with for the original action of .
The quotient is a sphere with four points of isotropy order two, represented by
The quotient is a sphere with isotropy orders . Its exceptional orbits are
With the actions of , defined above, let , for . We equip with the unique orientation for which is locally orientation-preserving away from the exceptional orbits. Recall that for a complex virtual line bundle over an oriented closed surface , its Chern number is the integer
where is the first Chern class and is the fundamental class determined by the orientation. Let be the dual Hopf line bundle, normalized by , and put
The pullback has the canonical -action for , where . This is well-defined because ; in particular, every stabilizer of acts trivially on its fiber. We give the trivial line bundle in the trivial -action as well, so defines a class in .
The degree will record the topological information that is not detected by the representations on the exceptional fibers. The following Lemma makes this precise for equivariant -theory classes and will be used to compute the restriction map from to .
Lemma 2.1. For these two actions, , and fiber restriction gives exact sequences
where
Moreover, and . Consequently a class in either group is determined by its fiber representations at the exceptional orbits and its degree.
Proof. Choose pairwise disjoint closed disks around the four exceptional points of , or the three exceptional orbits of , and let be the complement of their interiors. Choose slightly larger open disks and let be their union. Let be an open neighborhood of such that is the union of one annular collar for each boundary component. Thus , and deformation retracts onto .
Let , and . The space is free and each component of retracts equivariantly onto the corresponding exceptional orbit. The intersection is a union of free annuli. Since and are free -spaces, passage to orbit spaces identifies their equivariant -theory with ordinary -theory. The Mayer–Vietoris sequences are therefore
Here the initial records the common rank of a bundle over . After ordering the boundary circles compatibly with the exceptional orbits, the degree-zero maps are
Both maps are surjective, and their kernels are precisely and . In suitable integral bases, restriction from to its oriented boundary gives
Both maps are injective and have cokernel . Exactness of (6)–(7) first gives , and then gives
Thus this is (5), once its left-hand generator is identified.
Let and , and consider the summation map
The displayed formulas for show that . Hence induces an explicit isomorphism
In particular, the class of any standard basis vector of is a generator of the cokernel.
It remains to identify the image of this generator under the Mayer–Vietoris connecting homomorphism. Apply non-equivariant -theory to the open cover , formed by a neighborhood of and the union of the slightly enlarged disks around the exceptional points. The degree-one restriction-difference map is . Hence the following Mayer–Vietoris connecting homomorphism factors through the cokernel, giving
With the boundary orientations chosen above, it sends the class of the first standard basis vector to . Concretely, this class is obtained by using a clutching function of winding number one on the first boundary circle and the constant clutching function on the remaining circles; its first Chern number is one, so it is the class of the dual Hopf line bundle with our normalization.
Pullback by is compatible with the two Mayer–Vietoris sequences. On each free annulus, quotient identifies the equivariant boundary group upstairs with the ordinary boundary group downstairs. Writing for the equivariant connecting homomorphism, viewed as a map on the corresponding cokernel, the naturality square is
Therefore
Thus the injection of into identifies it with . To compute its degree, choose a point of away from the exceptional orbits. Its inverse image is one free -orbit and hence consists of exactly points. The map is locally orientation-preserving at each of them, so it has topological degree , or equivalently
Naturality of the first Chern class and of the pairing now gives
Two classes with equal fiber data differ by an integer multiple of ; equality of degrees makes that multiple zero. □
Two convenient bases
We now construct equivariant line bundles realizing the generators below. For each exceptional orbit , perform the following construction and denote the resulting line bundle by . Choose one small disk around every point of , with the disks pairwise disjoint and permuted by the group. Thus, when contains more than one point, the construction is carried out simultaneously on all these disks and produces a single line bundle . In a centered coordinate , the stabilizer generator acts by , with at for , and at every order-two center. On the trivial complex line bundle over each disk, define the stabilizer action by
and extend it to the other disks in the orbit by equivariance. Glue these bundles to the trivial equivariant line bundle on the complement of the centers as follows. Over an annular overlap, use for the fiber coordinate in the trivialization coming from the disk and for the fiber coordinate in the exterior trivialization. Since the center does not belong to the overlap, there, and is a well-defined map from the annulus to . The transition function is
Thus is the reciprocal of the local complex coordinate. The transition is equivariant because the exterior fiber coordinate is fixed and
Finally, the constant exterior section is written in the disk trivialization as
It therefore extends across the center and vanishes there. In the oriented real coordinates underlying the complex coordinate , its local representative is the map . Its derivative at the center is the identity, so the zero is transverse, simple, and has local index . Recall that, for a complex line bundle over an oriented closed surface and a section with isolated zeros,
where is the underlying oriented real plane bundle; see [1 Theorem 11.17]. Repeating the construction equivariantly over every disk in an orbit gives one positive simple zero at each point of and no zeros on the exterior trivialization. Consequently the resulting line bundle has degree . At a center its stabilizer acts on the fiber by the prescribed character , and equivariance transports this character to every point of the orbit. At the other exceptional points the exterior fiber coordinate is fixed, so their fiber characters are trivial.
For call these bundles , and , corresponding to , and ; their degrees are . In particular, is the one bundle obtained by carrying out the construction on the two disks centered at and simultaneously. For write for the degree-one bundle constructed separately at the singleton orbit , with . Set
Then the ordered integral bases are
Indeed give the differences between the character with value and the trivial character at their respective centers; and each give the nontrivial character minus the trivial character at their respective order-two orbits. Together with the common rank-one class, these form bases of . Adding gives the claimed bases by Lemma 2.1. Their ranks are nine and six. All classes except have rank zero, and their degrees are
Computation of the restriction map between the planar groups
Under (4), the inclusion induces restriction of bundles
Proof. At , restriction sends the character to . The classes have degrees , respectively, and the same exceptional fiber representations as , , and . The uniqueness statement in Lemma 2.1 therefore gives
The argument at is identical.
The line bundle has degree two and restricts to the nontrivial fiber character at both and . The K-theory element consequently restricts to . Finally has trivial fiber classes and degree four; it equals .
The displayed formulas give
But we can replace by in this basis, which proves the assertions about image and cokernel. □
Put and denote its inclusion into by . A direct verification gives that has the basis
Equivariant K-theory of EH
Actions on the bundle bases
Let be the automorphism of that fixes , and sends to . Translation by on the plane realizes this automorphism:
To see this on the generators, write , , , and . Conjugation by a translation fixes every translation, since translations commute; in particular, and . For the rotation, direct evaluation at gives
Thus conjugation by has exactly the stated effect on the generators of . On this translation is an involution. The identity says that is equivariant after the action on the target is relabelled by . More explicitly, if is a -equivariant vector bundle, its pullback has fiber . Give this pullback the -action
This is well-defined because sends the fiber over to the fiber over . Thus pulling back and relabelling the action defines a map on equivariant vector bundles, and hence on their K-theory group ; this is the automorphism . The same construction restricted to gives .
Lemma 3.1. On the bases (8), the automorphisms and act by
Proof. To read off these formulas, track the exceptional orbits, their fiber characters, and the degrees. The map sends to . The isotropy generator at is , while the corresponding generator at is . Thus a fiber character is transported to the same character , not to its inverse. Since preserves degree, this gives ; the reverse exchange similarly gives . The order-two orbit is preserved as a set, with its nontrivial fiber character unchanged, so the degree-two class is fixed. For the action, the singleton orbits move as and . Their nontrivial characters are preserved, giving the stated swaps of the . The rank-one trivial class is fixed. Finally, each has zero fiber classes and degree ; orientation preservation keeps that degree unchanged, so is fixed. The uniqueness statement in Lemma 2.1 says that these fiber data and degrees determine the classes and the proof is finished. □
Rotation by normalizes and therefore induces an automorphism
It exchanges , and fixes the other four basis elements. All three operators , , and are involutions, and
Since preserves , its restriction is given by
Mayer–Vietoris for the amalgam decomposition of H
We order the vertex term so that the first copy of corresponds to and the second copy corresponds to . Thus an element records the restrictions to the two vertex spaces in that order.
Since the planar odd K-theory groups vanish, the Mayer–Vietoris sequence in equivariant -theory associated with the decomposition gives
Equation (9) gives
An explicit integral identification for the even group is
Thus
Computation of the -theory of
The cohomological Wang sequence
For , define an affine map by
where and . These maps realize the automorphisms in the sense that
Let denote the generator of the cyclic quotient in . The action on is defined on generators by
The action is proper and cocompact and this is a model for . Finally, the map
is -invariant: fixes and adds one. Cover by two open arcs whose intersection is the disjoint union of two open arcs . Let be their inverse images in .
Choose a lift of each arc , and put for . The preimage of is the union of the integer translates of . The subgroup preserves , and the translates by permute these copies transitively. Consequently, as -spaces,
Since each is contractible and acts trivially on that interval coordinate, is -equivariantly homotopy equivalent to . Induction and homotopy invariance therefore give
The intersection has two components, corresponding to and , and therefore
Choose the lifts so that on one component of , say , the corresponding intervals in overlap in the same translate. The two local fiber coordinates then use the same point of , so the transition map on that component is the identity. Choose them so that on the other component the lift used for is the lift used for shifted by . To pass from the coordinate to the coordinate there, apply ; since , the fiber coordinate changes by . The induced map on equivariant -theory is , because . Thus, after identifying the two intersection components with and choosing the usual signs in the Mayer–Vietoris difference map, a pair restricts to on and to on . Therefore the difference map is
Use as coordinates in the source and subtract the first component of the target from the second. These integral changes of coordinates transform into
The identity summand contributes no kernel or cokernel. The Mayer–Vietoris sequence therefore reduces to the cohomological Wang sequence
Write
Then we must determine the action of on and the kernels and cokernels of . The next subsection computes the two induced actions, and § 5 computes the required kernels and cokernels.
The action of
Extend the automorphism to by fixing , and let be the automorphism of that fixes and sends to . Then
where denotes inner conjugation. Define the affine map
Conjugation by , , on the affine -action gives , , , respectively. Inner conjugations induce the identity after the natural identification of equivariant bundles, so does not change the induced maps on the groups in (14).
To describe the induced maps on the vertex term, put
Since , the plane has stabilizer . In particular, and represent the two vertex orbits, with stabilizers and . The map preserves both planes and restricts on each of them to translation by in the planar coordinate. It therefore induces on each copy of .
On the other hand, and . Thus exchanges the two vertex orbits. Consequently, acts on the vertex term by interchanging the two copies of .
The direct sum is the vertex term in the Mayer–Vietoris sequence (12). Since and differs from by an inner automorphism, the latter first exchanges the two vertex classes and then applies to each. Define the resulting automorphisms by
Thus records the action of on the two vertex restrictions before passing to . The compatibility with the edge map , verified below, shows that each preserves . Under the identification from (12), the restriction is precisely in even degree.
Let us determine the action of for the odd group. Translating by gives . The element folds this interval back by and acts on the plane by . Since acts trivially on the planar coordinate and rotation by induces on , the planar part of this identification induces . The reversal of the edge orientation contributes an additional minus sign. Hence the induced endomorphism of the oriented edge term is . On the other hand, the planar translation defining induces on by Lemma 3.1, and it preserves the orientation of the edge. Thus the induced endomorphism associated with is . Recall that is the generator of coker . Since
the map acts as on , whereas acts as . It follows that
The compatibility of the vertex maps with the Mayer–Vietoris differential can be checked directly:
The second equality uses . It induces the same sign on the cokernel generated by .
Under the identification , the map is the restriction of to . Transporting this restriction through the isomorphism (13), equation (4.1) becomes
Coinvariants and the four-dimensional groups
The Wang sequence reduces the remaining calculation to the kernels and cokernels of . We compute the cokernels first and then recover the kernel ranks.
For an automorphism of an abelian group , write . Since interchanges three pairs in (8),
with basis , where in this quotient. Equation (11) gives
with unrestricted. Therefore
On the other hand, we will divide out the relations defining the cokernel in two steps. First to the first nine columns of we impose . After quotienting by them, the remaining relations necessary to define , indexed by , are
In put
The four kernel generators defined in (10) satisfy
Together with (11), this reduces (20) to exactly two relations in :
The six elements
are an integral basis of : replacing by and by is unimodular. Thus in the relations (21) eliminate and , leaving the eight free generators
In particular,
Both are endomorphisms of . The ranks of their kernels and cokernels therefore agree, so
By (17), the odd-degree middle map is multiplication by two on . Its kernel is zero. Exactness gives
The right-hand term of (23) is free by (22); hence this sequence splits as a sequence of abelian groups. Combining (19)–(24) proves (1).
Reduced group C-algebra -theory
For a finite proper -CW-complex , the equivariant universal coefficient theorem of Joachim–Lück [4 Theorem 0.3, equation (0.5)] gives
The hypotheses hold for .
For a finitely generated group with finite,
Since the right-hand term in (25) is free, the sequence splits as a sequence of abelian groups. Thus the free ranks retain their parity, while the torsion comes from the opposite cohomological parity. Applying (25) to (1) proves (2).
Average an inner product over the finite holonomy group. The original affine action of each on is then proper and isometric. The Higson–Kasparov theorem [3] gives the Baum–Connes assembly isomorphism
This transfers the second table in Theorem 1.1 to the reduced group algebras.
Remark 5.1. The same groups could alternatively be obtained by computing Bredon homology with coefficients in complex representation-ring and then using the equivariant Atiyah–Hirzebruch spectral sequence for equivariant -homology. That approach requires keeping track of the resulting filtration and its additive extensions. We have preferred the more direct -theoretic method above, since the planar restriction map and the two monodromy actions can be described explicitly.
Departamento de Matemáticas, Universidad Nacional de Colombia, sede Bogotá, Cra. 30 cll 45 - Ciudad Universitaria, Bogotá, Colombia
Email address: [email protected]
Departamento de Matemáticas, Universidad Nacional de Colombia, sede Bogotá, Cra. 30 cll 45 - Ciudad Universitaria, Bogotá, Colombia
Email address: [email protected]
References
- [1]R. Bott and L. W. Tu, Differential forms in algebraic topology, Graduate Texts in Mathematics, vol. 82, Springer-Verlag, New York–Berlin, 1982.DOI
- [2]J. F. Davis and W. Lück, The topological K-theory of certain crystallographic groups, J. Noncommut. Geom. 7 (2013), no. 2, 373–431.
- [3]N. Higson and G. Kasparov, E-theory and KK-theory for groups which act properly and isometrically on Hilbert space, Invent. Math. 144 (2001), 23–74.DOI
- [4]M. Joachim and W. Lück, Topological K-(co-)homology of classifying spaces of discrete groups, Algebra. Geom. Topol. 13 (2013), no. 1, 1–34.arxiv.org/abs/1201.4763
- [5]M. Langer and W. Lück, Topological K-theory of the group C*-algebra of a semi-direct product Z^n × Z/m for a free conjugation action, J. Topol. Anal. 4 (2012), no. 2, 121–172.arxiv.org/abs/1109.1440
- [6]W. Lück and B. Oliver, The completion theorem in K-theory for proper actions of a discrete group, Topology 40 (2001), 585–616.DOI
- [7]W. Lück, Isomorphism conjectures in K- and L-theory, Ergebnisse der Mathematik und ihrer Grenzgebiete, 3. Folge, vol. 38, Springer, Cham, 2025.
- [8]W. Lück, Survey on classifying spaces for families of subgroups, in Infinite groups: geometric, combinatorial and dynamical aspects, Progress in Mathematics, vol. 248, Birkhäuser, Basel, 2005, 269–322.
- [9]N. Petrosyan and B. Putrycz, On cohomology of crystallographic groups with cyclic holonomy of split type, J. Algebra 367 (2012), 237–246.arxiv.org/abs/1106.4216
- [10]L. J. Sánchez Saldaña and M. Velásquez, On the K-theory of groups of the form Z^n × Z/m with m square-free, Proc. Amer. Math. Soc. 155 (2027), no. 1, 127–140.