betaFeedback

WorksMathematics

The Bass trace conjecture and the characteristic-zero Kaplansky idempotent conjecture

Abstract

We prove the complex group-ring Bass trace conjecture for every discrete group: the Hattori–Stallings trace of a finitely generated projective module over its complex group ring is supported on conjugacy classes of finite-order elements. As a consequence, for every torsion-free group G and every commutative unital domain R of characteristic zero, the only idempotents in RGRG are 0 and 1. This proves Kaplansky's idempotent conjecture in characteristic zero.

Introduction

The rank of a projective module over a group ring contains information about individual conjugacy classes. This refinement of ordinary rank is a basic ingredient of group-ring Euler characteristics: alternating sums of projective-module ranks retain information that a numerical Euler characteristic forgets [3, 4]. The Bass trace conjecture asks how much of that information can occur at elements of infinite order. We first specify the complex version of this question.

For a unital complex algebra AA, let K0(A)K_0(A) be the Grothendieck group of finitely generated projective right AA-modules, and set

HH⁡0(A)=A/[A,A],\operatorname{HH}_0(A) = A/[A,A],

where [A,A][A,A] is the complex linear span of commutators. An idempotent e=(eij)∈Mq(A)e=(e_{ij})\in M_q(A) represents the projective right module eAqeA^q, with matrices acting on columns by left multiplication. Its Hattori–Stallings trace is

HS⁡A([eAq])=[∑i=1qeii]∈HH⁡0(A),\operatorname{HS}_A([eA^q]) = \left[\sum_{i=1}^{q}e_{ii}\right]\in\operatorname{HH}_0(A),

extended additively to virtual projective classes.

Let GG be a discrete group and write con⁡(G)\operatorname{con}(G) for its set of conjugacy classes. The quotient HH⁡0(CG)\operatorname{HH}_0(\mathbb{C}G) identifies with ⨁C∈con⁡(G)CC\bigoplus_{C\in\operatorname{con}(G)}\mathbb{C}C by

[∑h∈Gahh]⟼∑C∈con⁡(G)(∑h∈Cah)C.(1)\left[\sum_{h\in G}a_hh\right]\longmapsto\sum_{C\in\operatorname{con}(G)}\left(\sum_{h\in C}a_h\right)C. \tag*{(1)}

Indeed, a commutator has zero sum on each conjugacy class, and the differences of conjugate group elements span the kernel of the displayed map. Such differences are commutators: tst−1−s=[t,st−1]tst^{-1}-s=[t,st^{-1}]. All sums arising from an element of CG\mathbb{C}G have finite support. We write HS⁡G(x)(C)\operatorname{HS}_G(x)(C) for the coefficient at CC.

The complex group-ring Bass conjecture asks whether these coefficients vanish on every infinite-order conjugacy class, for every group and every class in K0(CG)K_0(\mathbb{C}G); see [5] Conjecture 2.1 and [11] Conjecture 2.

Theorem 1.1. Let GG be any discrete group. For every x∈K0(CG)x \in K_0(\mathbb{C}G) and every infinite-order element g∈Gg \in G,

HS⁡G(x)([g]G)=0.\operatorname{HS}_G(x)([g]_G)=0.

Equivalently, if confin⁡(G)\operatorname{confin}(G) denotes the conjugacy classes of finite-order elements, then

HS⁡G(K0(CG))⊆⨁C∈confin⁡(G)C.\operatorname{HS}_G(K_0(\mathbb{C}G)) \subseteq\bigoplus_{C \in\operatorname{confin}(G)} \mathbb{C} .

This resolves the complex group-ring formulation positively. Finite-order coefficients need not vanish away from the identity: for a finite group FF, the idempotent ∣F∣−1∑h∈Fh|F|^{-1}\sum_{h\in F}h has coefficient ∣C∣/∣F∣|C|/|F| at every conjugacy class CC. The conclusion therefore differs from the identity-only support assertion for integral group rings. That integral assertion follows in Section 9 by combining Theorem 1.1 with the known restriction on its finite-order coefficients.

Trace rank and the idempotent conjecture

For a torsion-free group the identity is the only finite-order element, so Theorem 1.1 also has numerical and ring-theoretic consequences. Define the canonical trace and the augmentation by

τG(∑g∈Gagg)=a1,ϵG(∑g∈Gagg)=∑g∈Gag.\tau_G\left(\sum_{g\in G}a_g g\right)=a_1,\qquad\epsilon_G\left(\sum_{g\in G}a_g g\right)=\sum_{g\in G}a_g.

For a matrix b=(bij)∈Mq(CG)b=(b_{ij})\in M_q(\mathbb{C}G), put τG,q(b)=∑i=1qτG(bii)\tau_{G,q}(b)=\sum_{i=1}^q\tau_G(b_{ii}), and let ϵG,q(b)∈Mq(C)\epsilon_{G,q}(b)\in M_q(\mathbb{C}) be the entrywise augmentation. The coefficient at [1]G[1]_G of the Hattori–Stallings trace defines

τG,∗:K0(CG)⟶C,τG,∗([e(CGq)])=τG,q(e).\tau_{G,*}:K_0(\mathbb{C}G)\longrightarrow\mathbb{C},\qquad\tau_{G,*}([e(\mathbb{C}G^q)])=\tau_{G,q}(e).

The augmentation also defines the integer-valued augmentation rank

rk⁡ϵ:K0(CG)→(ϵG)∗K0(C)→dim⁡CZ.\operatorname{rk}_{\epsilon}:K_0(\mathbb{C}G)\xrightarrow{(\epsilon_G)_*}K_0(\mathbb{C})\xrightarrow{\dim_{\mathbb{C}}}\mathbb{Z}.

The idempotent conjecture commonly attributed to Kaplansky asks whether, for every torsion-free group GG and every field kk, an element e∈kGe\in kG satisfying e2=ee^2=e must be 00 or 11; see [21], Problem 3. It concerns elements of the group ring itself, not arbitrary idempotent matrices over that ring.

Corollary 1.2. Let GG be any torsion-free discrete group.

(i) For every q≥1q\geq1 and every idempotent e∈Mq(CG)e\in M_q(\mathbb{C}G),

τG,q(e)=rank⁡C(ϵG,q(e)).\tau_{G,q}(e)=\operatorname{rank}_{\mathbb{C}}(\epsilon_{G,q}(e)).

Consequently τG,∗=rk⁡ϵ\tau_{G,*}=\operatorname{rk}_{\epsilon}, with Z\mathbb{Z} viewed as a subgroup of C\mathbb{C}, and

τG,∗(K0(CG))=Z.\tau_{G,*}(K_0(\mathbb{C}G))=\mathbb{Z}.

(ii) For every commutative unital domain RR of characteristic zero, every e∈RGe\in RG satisfying

e2=ee^2=e

is either 00 or 11.

Part (ii) resolves the characteristic-zero Kaplansky idempotent conjecture positively, and includes coefficient domains as well as fields. Section 10 derives both parts from the trace theorem. The integer range in part (i) concerns algebraic K0(CG)K_0(\mathbb{C}G); its relation to the reduced C∗C^*-completion is made explicit there.

The integral trace consequence also links the theorem to fixed-point theory. A map ff is a homotopy idempotent if f∘f≃ff \circ f \simeq f. Berrick, Chatterji and Mislin’s equivalences [6] [Theorems 1–2] then imply that every homotopy idempotent on a connected closed smooth oriented manifold of dimension greater than two is freely homotopic to a map with exactly one fixed point. They also identify ordinary and L2L^2-Lefschetz numbers for pointed homotopy idempotents of connected closed smooth oriented manifolds that induce the identity on the fundamental group. Corollary 9.2 states these consequences precisely; Section 9 explains their connection to projective Euler characteristics.

The two parts of the proof

We first describe the algebraic construction to explain the geometric statement it needs. The body of the paper begins with that geometric statement, formulated independently of group rings, and completes its proof before constructing the cycles and cocycles to which it applies.

For a finitely generated group GG, the algebraic construction starts with an idempotent e∈Mq(CG)e \in M_q(\mathbb{C}G) and one infinite-order element g∈Gg \in G. The finitely many nonzero group coefficients of ee determine a finite set of steps. Products of matrix coefficients along paths from hh to ghgh, followed by the ordinary matrix trace, give complex weights on finite lists of group elements. Idempotency says that summing over any intermediate vertex removes that vertex. Cyclicity of matrix trace gives a second identity, a cyclic shift whose last vertex is multiplied by gg.

Project the vertices to the coset set Y=⟨g⟩\GY = \langle g\rangle\backslash G. The group H=CG(g)/⟨g⟩H = C_G(g)/\langle g\rangle, where CG(g)C_G(g) is the centralizer of gg, acts freely on YY. For every positive even dimension 2m2m, the weighted lists define a simplicial cycle modulo this action. Its simplices use connected sets of vertices in one graph of bounded degree, fixed as mm increases. A Pfaffian cocycle evaluates on this cycle as 2−mHSG([e(CG)q])([g]G)2^{-m} HSG([e(\mathbb{C}G)^q])([g]_G). Here a cocycle means an alternating function on vertex lists that vanishes on simplicial boundaries. The explicit construction and the evaluation occupy Sections 7 and 8.

The geometric part, proved first in Sections 2–6, shows that every invariant cocycle evaluates to zero on cycles of this kind in sufficiently high dimension. We cut the supporting simplicial complex successively into pieces of bounded diameter, choosing the separating sets nearly to minimize their areas. A volume estimate and repeated slicing make the expected number of points in the final zero-dimensional set arbitrarily small. A subdivision construction turns this point count into a bound for cocycle evaluations. Equivariant chain homotopy keeps the evaluation unchanged throughout the construction.

The fixed graph degree makes increasing the dimension useful. Through each vertex there are only exponentially many connected vertex sets of a given size. On an nn-simplex, Euclidean structures chosen from spanning trees of its vertex set give volume 1/n!1/n!; repeated slicing produces the same factorial in the comparison with a point of the final separating set. After these factors cancel, radii chosen in terms of the fixed degree bound make the remaining volume ratio tend to zero with nn. The local replacement argument turns this strict volume margin into the small expected point count. No homological-dimension bound on HH is assumed.

Two features permit this argument for arbitrary groups. A compact probability space with a free action separates overlapping translates of local modifications, without any amenability assumption. Also, each top simplex has its own averaged Euclidean area functional. The areas need not match along common faces; the successive cuts keep the relevant dimensional measures away from those faces. The cocycles need not be globally bounded: near each original simplex, the construction uses only a fixed finite set of possible vertices. Bounding the cocycle on those lists suffices for the evaluation estimate.

The final assembly in Section 9 chooses a large even dimension, forcing the individual trace coefficient to vanish. Passing from finitely generated groups to arbitrary groups uses the exact sum over subgroup conjugacy classes contained in an ambient conjugacy class. Section 10 then extracts the augmentation rank from the identity coefficient. In the scalar case this rank is 0 or 1, and the standard faithful-trace argument forces the idempotent itself to be 0 or 1. A finite-coefficient reduction extends that conclusion from complex coefficients to every commutative characteristic-zero domain.

Relation to previous work

The rank construction comes from Hattori and Stallings [16, 24]; Bass developed its relation to Euler characteristics and formulated the integral conjecture in [3]; see also [4], Conjecture (4.5). Bass proved complex trace vanishing at infinite-order elements for linear groups [4], Theorem (6.1)(f). For the integral trace, Linnell established a restriction on the possible nonzero coefficients [19], Lemma 4.1, later recovered through topological Hochschild homology by Berrick and Hesselholt [7], Theorem A. Its finite-order consequence is the input used in Section 9.

The scalar idempotent problem also has a long history of trace methods. Formanek proved the characteristic-zero conclusion for torsion-free groups satisfying the ascending chain condition on subgroups [15], Theorem 1. Burger and Valette revisit the theorems of Kaplansky and Zalesskii: the canonical trace of a complex scalar idempotent lies in [0,1][0,1], its endpoints force the idempotent to be trivial, and its value is rational [9], Theorems 1.3 and 2.1. They also obtain τG,∗(K0(CG))⊆Q\tau_{G,*}(K_0(\mathbb{C}G)) \subseteq\mathbb{Q} [9], Corollary 1.2. Rationality still permits nonintegral values strictly between zero and one. For torsion-free groups, Theorem 1.1 identifies the trace with augmentation rank, which excludes those values. The subsequent faithful-trace argument is classical and is included in Section 10.

Broader group classes have been treated by algebraic and analytic methods. Farrell and Linnell proved the complex Bass statement for elementary amenable groups as part of a result over integral domains [14], Theorem 1.6. Berrick, Chatterji and Mislin obtained it for amenable groups through rational surjectivity of the degree-zero Rost assembly map for countable groups [5], Theorems 1.3–1.4 and Lemma 1.5. Their result concerns the stronger trace assertion on ℓ1(G)\ell^1(G), with finite support on finite-order classes. An algebraic route uses induction from finite subgroups: the Farrell–Jones conjecture with complex coefficients implies the complex Bass statement [2], Theorems 1.5 and 1.9. The present proof uses the finite support of group-ring idempotents throughout; it requires neither an assembly hypothesis nor a statement about a completed group algebra.

For some groups, analytic methods give a stronger idempotent conclusion in the reduced group C∗C^*-algebra Cr∗(G)C^*_r(G), the operator-norm closure of CG\mathbb{C}G acting on ℓ2(G)\ell^2(G). Using Lafforgue’s Banach KKKK-theory, Mineyev and Yu proved this for torsion-free subgroups of hyperbolic groups [20], Theorems 19–21. Puschnigg independently obtained the torsion-free hyperbolic case through local cyclic homology [23]; see also [20], Section 6. Our conclusion for arbitrary torsion-free groups concerns the group ring and the trace on its algebraic K0K_0, as distinguished in Section 10.

The higher-dimensional algebraic construction has antecedents in the cyclic-homological approach to the conjecture. Connes constructed a pairing of even cyclic cocycles with projective classes [12], Part II, §2, Proposition 14. Burghelea’s decomposition of cyclic homology by conjugacy classes identifies the infinite-order components through the reduced centralizers CG(g)/⟨g⟩C_G(g)/\langle g\rangle [10]. Eckmann used higher-degree vanishing to obtain trace vanishing over rational group rings for specified classes of groups of finite rational homological dimension [13]. Sections 7–8 give direct finite-chain constructions in this tradition, including the normalization of the Pfaffian detector; no cyclic-homology theorem is needed as an input. The sparse-chain theorem replaces a homological-dimension hypothesis with connected-support control on the particular cycles.

The geometric proof develops the separating-set method of Papasoglu [22], the separating filtrations and simplex-counting argument of Alpert [1], and the Cantor-parameter approach of Braun and Sauer [8]. Alpert’s equivariant formulations [1] provide the closest model for the local comparisons. Alpert’s filtration results concern manifold geometry and compatible metrics. We give the full piecewise-linear argument needed for simplex-dependent areas and arbitrary cycles in the coinvariant chain complex. The separate Pfaffian calculation converts this geometric vanishing into a statement about each conjugacy-class coefficient.

A vanishing theorem for sparse simplicial cycles

We formulate the geometric statement independently of group rings. Its input is a graph with a free group action and a simplicial complex whose top-dimensional vertex sets are connected in that graph. Its output is the vanishing of every invariant cocycle evaluation in sufficiently high dimension.

For a set YY, let C(Y)C^{(Y)} be the vector space with basis {δy:y∈Y}\{\delta_y : y \in Y\}, consisting of finite linear combinations. The augmented chains of the full simplex on YY are

Ej=⋀j+1C(Y)(j≥0),E−1=C.E_j = \bigwedge^{j+1} C^{(Y)} \quad(j \ge0), \qquad E_{-1} = \mathbb{C}.

The differential is contraction with the functional α(∑ycyδy)=∑ycy\alpha\left(\sum_y c_y\delta_y\right)=\sum_y c_y:

∂(δy0∧⋯∧δyj)=∑i=0j(−1)iδy0∧⋯∧δyi^∧⋯∧δyj.\partial(\delta_{y_0} \wedge\cdots\wedge\delta_{y_j}) = \sum_{i=0}^{j} (-1)^i \delta_{y_0} \wedge\cdots\wedge\widehat{\delta_{y_i}} \wedge\cdots\wedge\delta_{y_j}.

In degree zero this is the augmentation to C\mathbb{C}; the empty wedge is 11. For a simplicial subcomplex KK, its augmented oriented chains C∗(K)C_*(K) form a subcomplex of E∗E_*.

If a group HH acts on YY, it acts on both chain complexes by permuting vertices, with the signs prescribed by the exterior product. For an HH-module VV, its coinvariants are

VH=V/span⁡C{hv−v:h∈H, v∈V}.V_H = V/\operatorname{span}_{\mathbb{C}}\{hv-v : h \in H,\ v \in V\}.

The differentials induce differentials on coinvariants. A cycle in Cn(K)HC_n(K)_H is a class whose boundary is zero there; an arbitrary finite chain representing it need not have zero boundary before passage to coinvariants. An HH-invariant linear functional ψ:En→C\psi:E_n\to\mathbb{C} descends to (En)H(E_n)_H. We call it a cocycle when ψ∂=0\psi\partial=0 on En+1E_{n+1}.

Theorem 2.1 (Sparse-chain vanishing). For each integer D≥2D \ge2 there is an integer n0(D)n_0(D) with the following property. Let YY be the vertex set of a countable connected simple graph of degree at most DD, and let a countable group HH act by graph automorphisms, freely on vertices. Suppose n≥n0(D)n \ge n_0(D) and KK is an HH-invariant subcomplex of the full simplex on YY such that: (i) KK is generated by its nn-simplices and has finitely many HH-orbits of simplices;

(ii) the vertices of each nn-simplex span a connected subgraph of YY.

Then every HH-invariant cocycle ψ:En→C\psi: E_n \to\mathbb{C} evaluates to zero on the image in (En)H(E_n)_H of every cycle in Cn(K)HC_n(K)_H.

The bound n0(D)n_0(D) depends only on the graph degree, not on HH, KK or ψ\psi. There is no boundedness hypothesis on ψ\psi. We will bound it only on a finite set of basis wedges associated with each fixed original simplex. This distinction is necessary in the application.

For the proof assume K≠∅K \ne\varnothing; the empty case has no positive degree chains. Its realization ∣K∣|K| consists of finitely supported probability vectors p:Y→[0,1]p : Y \to[0,1] whose supports are simplices of KK. A simplex will also denote its closed affine realization; σ∘\sigma^\circ denotes its relative interior. We call simplices of KK original to distinguish them from the later subdivisions, and call its nn-simplices top simplices.

The proof has three steps. First, we choose a metric and simplexwise areas for which balls have small volume in high dimension. Second, we construct separating filtrations with very few points in their final stratum, on average over a free probability parameter. Third, we use their subdivisions to construct equivariant chain maps homotopic to inclusion whose cocycle evaluations approach zero. Each approximation uses a finite filtration; no limiting filtration is needed.

A metric and volume bound for connected supports

We work with the graph YY, group HH, and nonempty nn-dimensional complex KK of Theorem 2.1, where n≥1n \ge1. We need two kinds of geometric control: a metric with finite neighborhoods in which to make cuts, and Euclidean area functionals for which those cuts satisfy a slicing inequality. We will use a scaled transportation metric on ∣K∣|K| and, separately inside each top simplex, Euclidean structures determined by its spanning trees. Each top simplex will have volume 1/n!1/n!, while every affine branch of a distance function will have gradient norm at most one in each of these structures. Counting the possible connected vertex supports will then give the ball-volume estimate used in the separating argument.

The metric and its finite local geometry

Represent a point of ∣K∣|K| by a probability vector pp on YY whose support lies in a simplex of KK. Let dYd_Y denote graph distance and define

ρ(p,q)=1nsup⁡Lip⁡dY(ℓ)≤1∑v∈Yℓ(v)(p(v)−q(v)).(2)\rho(p,q)=\frac{1}{\sqrt{n}}\sup_{\operatorname{Lip}_{d_Y}(\ell)\le1}\sum_{v\in Y}\ell(v)(p(v)-q(v)). \tag*{(2)}

This is the scaled Kantorovich–Rubinstein metric on finitely supported probability measures, in its Lipschitz-potential form [18], pp. 1381–1382. We establish its needed properties directly. Only finitely many terms occur in the pairing. To see that the supremum is finite and describe the distance function on each simplex, let V⊂YV \subset Y be a nonempty finite set and put

PV={ℓ∈RV:∑v∈Vℓ(v)=0,∣ℓ(v)−ℓ(w)∣≤dY(v,w)(v,w∈V)}.\mathcal{P}_V=\left\{\ell\in\mathbb{R}^{V}:\sum_{v\in V}\ell(v)=0,\quad\lvert\ell(v)-\ell(w)\rvert\le d_Y(v,w)\quad(v,w\in V)\right\}.

This is a nonempty compact polytope. Indeed, its difference constraints and zero mean bound every coordinate by max⁡v,w∈VdY(v,w)\max_{v,w\in V} d_Y(v,w). Each ℓ∈PV\ell\in\mathcal{P}_V extends to a 1-Lipschitz function on YY by

ℓ~(y)=min⁡v∈V(ℓ(v)+dY(v,y)).\widetilde{\ell}(y)=\min_{v\in V}\bigl(\ell(v)+d_Y(v,y)\bigr).

The difference constraints show that this extension agrees with ℓ\ell on VV, and the triangle inequality proves that it is 1-Lipschitz. Conversely, a global 1-Lipschitz function restricts, after subtraction of its mean, to an element of PV\mathcal{P}_V. Constants pair to zero with p−qp-q. Consequently, when VV contains both supports, the supremum in (2) is the maximum over PV\mathcal{P}_V.

Symmetry follows by replacing ℓ\ell with −ℓ-\ell, and the triangle inequality follows by applying each potential to p−q=(p−u)+(u−q)p-q=(p-u)+(u-q). The indicator of a single vertex is 1-Lipschitz for dYd_Y, so

∣p(v)−q(v)∣≤n ρ(p,q)(v∈Y).(3)\lvert p(v)-q(v)\rvert\le\sqrt{n}\,\rho(p,q) \qquad(v\in Y). \tag*{(3)}

Thus ρ\rho separates points and is a metric. It is HH-invariant because graph automorphisms permute the admissible potentials. On probability vectors supported in a fixed finite set, it is the restriction of a norm on the finite-dimensional space of vectors with coordinate sum zero. It therefore gives the usual topology on every finite subcomplex.

For c∈∣K∣c\in|K|, write

fc(p)=ρ(c,p),Us(c)={p:fc(p)<s},f_c(p)=\rho(c,p), \qquad U_s(c)=\{p:f_c(p)<s\},
Ls(c)={p:fc(p)≤s},Ss(c)={p:fc(p)=s}.L_s(c)=\{p:f_c(p)\le s\}, \qquad S_s(c)=\{p:f_c(p)=s\}.

On a fixed closed simplex σ\sigma, the function fcf_c is a maximum of finitely many affine functions: use the extreme points of PV(σ)∪supp⁡c\mathcal{P}_{V(\sigma)\cup\operatorname{supp}c}. Here V(σ)V(\sigma) denotes the vertex set of σ\sigma. Taking all these extreme points makes the collection of affine branches equivariant under HH.

Lemma 3.1. Every closed ρ\rho-ball meets only finitely many simplices of KK. The complex KK is locally finite, and ρ\rho gives its usual realization topology and is proper. For each c∈∣K∣c\in|K|, s≥0s\ge0, and simplex κ\kappa, the set

{h∈H:Ls(hc)∩κ≠∅}\{h\in H:L_s(hc)\cap\kappa\ne\varnothing\}

is finite. The setwise stabilizer HκH_\kappa of a nonempty simplex is finite as well.

Proof. Let A=supp⁡cA=\operatorname{supp}c. The function dY(⋅,A)d_Y(\mathord{\cdot},A) is 1-Lipschitz and vanishes on AA. Using its negative in (2) gives

min⁡a∈supp⁡cv∈supp⁡pdY(a,v)≤∑vp(v)dY(v,A)≤n ρ(c,p).(4)\min_{\substack{a\in\operatorname{supp}c\\v\in\operatorname{supp}p}}d_Y(a,v) \le\sum_v p(v)d_Y(v,A)\le\sqrt{n}\,\rho(c,p). \tag*{(4)}

If a top simplex σ\sigma meets Ls(c)L_s(c) at pp, it therefore contains a vertex within distance ⌊sn⌋\lfloor s\sqrt{n}\rfloor of AA. The graph on its n+1n+1 vertices is connected, so every vertex of σ\sigma is within another nn steps of that vertex. Thus all such simplices use vertices in the finite graph neighborhood

B⌊sn⌋+nY(A):={v∈Y:dY(v,A)≤⌊sn⌋+n}.B^Y_{\lfloor s\sqrt{n}\rfloor+n}(A):=\{v\in Y:d_Y(v,A)\le\lfloor s\sqrt{n}\rfloor+n\}.

There are only finitely many possibilities for σ\sigma and its faces. Likewise, every top simplex containing a specified vertex vv uses vertices in BnY({v})B^Y_n(\{v\}); hence each vertex has a finite star.

For completeness, the finite-subcomplex topology statement implies the corresponding statement on all of ∣K∣|K|. Given pp, let η=min⁡{p(v):v∈supp⁡p}>0\eta=\min\{p(v):v\in\operatorname{supp}p\}>0. By (3), the ball of radius η/(2n)\eta/(2\sqrt{n}) about pp lies in the finite star of supp⁡p\operatorname{supp}p. Within this star the metric and realization topologies agree. The open star of supp⁡p\operatorname{supp}p is also a neighborhood of pp in the locally finite realization, so the agreement is local in both topologies. A closed metric ball is a closed subset of the finite union of closed simplices it meets, and is therefore compact.

If Ls(hc)L_s(hc) meets κ\kappa, eq:3 supplies a∈Aa \in A and b∈V(κ)b \in V(\kappa) with dY(ha,b)≤snd_Y(ha,b) \le s\sqrt{n}. For each aa, the map h↦hah \mapsto ha is injective because HH acts freely on vertices. The finitely many pairs a,ba,b and finite graph balls therefore allow only finitely many hh. Finally, for v∈V(κ)v \in V(\kappa), the same injection maps HκH_\kappa into V(κ)V(\kappa). In particular, ∣Hκ∣≤∣V(κ)∣|H_\kappa| \le|V(\kappa)|.

We will also use the following consequence. For fixed cc and ss, only finitely many h∈Hh \in H satisfy

Ls(hc)∩Ls(c)≠∅.L_s(hc) \cap L_s(c) \ne\varnothing.

Indeed, an intersection gives ρ(hc,c)≤2s\rho(hc,c) \le2s, and then eq:3 puts some vertex of hsuppch \mathop{\rm supp} c within distance 2sn2s\sqrt{n} of suppc\mathop{\rm supp} c. The same injectivity argument applies.

Simplex areas and distance gradients

The metric now provides finite neighborhoods in which we can make cuts. For area comparisons, we use Euclidean structures separately inside the top simplices. They need not agree on shared faces.

Fix a top simplex σ\sigma. For each spanning tree TT of the connected graph on V(σ)V(\sigma), equip its affine span with the Euclidean structure in which the tree-edge vectors δv−δw\delta_v-\delta_w, with either orientation for each edge, form an orthonormal basis of the translation space. For 0≤j≤n0 \le j \le n, let a0ja_0^j be the average, over these finitely many trees, of Euclidean jj-area on polyhedral subsets of σ∘\sigma^\circ of dimension at most jj. We allow relatively open faces in these sets; pieces of dimension less than jj have zero jj-area, and a00a_0^0 is counting measure. These area functionals are covariant under HH, since the action permutes the spanning trees.

Lemma 3.2. In every tree Euclidean structure on a top simplex σ\sigma, its nn-volume is 1/n!1/n!. Every affine branch of fc∣σf_{c| \sigma} has gradient of norm at most one, for every c∈∣K∣c \in|K|. The same gradient bound holds after restriction to any affine subspace of the span of σ\sigma.

Proof. Root the tree at v0v_0, orient its edges away from the root, and order the remaining vertices so that each parent precedes its children. Put

ti=δvi−δparent⁡(vi),bi=δvi−δv0(1≤i≤n).t_i=\delta_{v_i}-\delta_{\operatorname{parent}(v_i)}, \qquad b_i=\delta_{v_i}-\delta_{v_0}\quad(1 \le i \le n).

Each bib_i is the sum of the tjt_j along its root path. Thus the change-of-basis matrix is triangular with diagonal entries one. The tit_i are an orthonormal basis in the chosen metric, so the parallelepiped spanned by the bib_i has volume one. The simplex has volume 1/n!1/n!.

An affine branch has derivative along tit_i equal, up to sign, to

ℓ(vi)−ℓ(parent⁡(vi))n\frac{\ell(v_i)-\ell(\operatorname{parent}(v_i))}{\sqrt{n}}

for a 1-Lipschitz potential ℓ\ell. Since the two vertices are adjacent, this derivative has absolute value at most 1/n1/\sqrt{n}. The squared gradient norm is the sum of the squares of these nn derivatives and is at most one. Restriction to an affine subspace replaces the gradient by its orthogonal projection, which cannot increase its norm.

The volume margin in high dimension

We next bound the total top-dimensional area of a metric ball by counting its possible connected vertex supports. Fix

R=4D2,r=R/2.R = 4D^2,\qquad r = R/2.

For a specified vertex vv, there are at most D2nD^{2n} connected sets of n+1n+1 vertices containing vv. To prove this, choose a spanning tree of each such set rooted at vv. A traversal using each tree edge once in each direction is a walk of length 2n2n, starting at vv, whose visited set is exactly the prescribed set. Choosing one such walk per set gives an injection into the length-2n2n walks from vv, of which there are at most D2nD^{2n}.

Set k=⌊Rn⌋k=\lfloor R\sqrt{n}\rfloor. A graph ball of radius kk has at most

∑j=0kDj≤Dk+1\sum_{j=0}^{k}D^j\le D^{k+1}

vertices, since D≥2D\ge2. As ∣supp⁡c∣≤n+1|\operatorname{supp} c|\le n+1, there are at most (n+1)Dk+1(n+1)D^{k+1} vertices within distance kk of supp⁡c\operatorname{supp} c. By (4), each top simplex meeting LR(c)L_R(c) contains one of them. The connected-set count and Lemma 3.2 therefore give, for every c∈∣K∣c\in|K|,

∑σ topanσ(LR(c)∩σ∘)≤VR:=(n+1)D⌊Rn⌋+1D2nn!.(5)\sum_{\sigma\ \mathrm{top}}a_n^\sigma\left(L_R(c)\cap\sigma^\circ\right) \le V_R:=\frac{(n+1)D^{\lfloor R\sqrt{n}\rfloor+1}D^{2n}}{n!}. \tag*{(5)}

Counting a simplex more than once only enlarges this upper bound.

With DD fixed, the ratio of this estimate to (R−r)n/n!(R-r)^n/n! is

VR(R−r)n/n!=(n+1)D⌊4D2n⌋+12n⟶0.\frac{V_R}{(R-r)^n/n!} =\frac{(n+1)D^{\lfloor4D^2\sqrt{n}\rfloor+1}}{2^n} \longrightarrow0.

Indeed, its logarithm is −nlog⁡2+4D2(log⁡D)n+OD(log⁡n)-n\log2+4D^2(\log D)\sqrt{n}+O_D(\log n). We may therefore choose n0(D)≥1n_0(D)\ge1 so that, for every n≥n0(D)n\ge n_0(D),

VR<(R−r)nn!.V_R<\frac{(R-r)^n}{n!}.

For the remainder of the geometric proof, fix such an nn. The graph degree bound DD, and hence RR and rr, remain fixed as dimensions increase. The strict margin (3.6) will convert successive area comparisons into a bound on the number of points at the bottom of a separating filtration.

Equivariant separating filtrations

We retain the complex KK, the metric ρ\rho, and the radii r<Rr<R from Section 3. We will repeatedly cut KK into pieces of diameter at most 2R2R, lowering the dimension of the separating set at each step. The cuts vary over a compact probability space on which HH acts freely. This allows local modifications to be separated after taking account of the parameter, even when their translates overlap in ∣K∣|K|.

A free probability parameter

The following construction is a finite-color variant of the coinduction argument in Braun and Sauer [8], who credit the existence theorem to Hjorth and Molberg [17]. We include the construction, including its torsion case, so that no free-action theorem is an input.

Lemma 4.1. Every countable group HH admits a continuous free action on a compact metrizable zero-dimensional space XX, preserving a Borel probability measure μ\mu.

Proof. For a∈H∖{1}a \in H \setminus\{1\}, put ka=ord⁡(a)k_a = \operatorname{ord}(a) if aa has finite order, and ka=2k_a = 2 otherwise. Let

Xa={ξ:H→Z/kaZ:ξ(ta)=ξ(t)+1 for every t∈H}.X_a = \{\xi: H \to\mathbb{Z}/k_a\mathbb{Z} : \xi(ta) = \xi(t) + 1 \text{ for every } t \in H\}.

This is a closed subspace of the product of finite discrete spaces (Z/kaZ)H(\mathbb{Z}/k_a\mathbb{Z})^H. Choose a representative tCt_C for each left coset C=tC⟨a⟩C = t_C\langle a\rangle. A coloring in XaX_a is specified uniquely by offsets ωC∈Z/kaZ\omega_C \in\mathbb{Z}/k_a\mathbb{Z}, through ξ(tCaj)=ωC+j\xi(t_Ca^j) = \omega_C + j. This is well defined also when aa has finite order, because the modulus is then its order. Thus XaX_a is nonempty, compact and zero-dimensional.

Give the offsets independent uniform distributions. The left action

(hξ)(t)=ξ(h−1t)(h\xi)(t) = \xi(h^{-1}t)

preserves XaX_a and its probability measure: it permutes the cosets and adds a fixed element of Z/kaZ\mathbb{Z}/k_a\mathbb{Z} to each offset. These operations preserve every finite cylinder probability. Take the product over a≠1a \ne1 to obtain XX and μ\mu. Countability gives metrizability, and the coordinatewise action is continuous and measure preserving. For any a≠1a \ne1 and any point of XX, its aa-coordinate satisfies

(aξa)(1)=ξa(a−1)=ξa(1)−1≠ξa(1).(a\xi_a)(1) = \xi_a(a^{-1}) = \xi_a(1) - 1 \ne\xi_a(1).

Hence no nonidentity element fixes a point. If HH is trivial, take a one-point probability space. □\square

Fix such (X,μ)(X,\mu) for the rest of the geometric argument. In particular, no amenability assumption on HH is needed.

The allowed filtrations

A triangulation refining KK means a finite affine simplicial subdivision of each closed original simplex, with agreement on shared faces. A family T(x)T(x) of such triangulations, together with specified subcomplexes, is locally constant if, on each original simplex, all its exact affine simplices and subcomplex markings are constant on the pieces of a finite clopen partition of XX. The partition may depend on the original simplex. We require equivariance: T(hx)=hT(x)T(hx) = hT(x), and likewise for every specified subcomplex. Refinement of triangulations always preserves their earlier subcomplexes.

Begin with Tn(x)=KT_n(x) = K and Zn(x)=∣K∣Z_n(x) = |K|. At stage dd, where n≥d≥1n \ge d \ge1, suppose that P(x)=Zd(x)P(x) = Z_d(x) is a subcomplex of Td(x)T_d(x). An admissible next choice consists of a locally constant equivariant refinement Td−1(x)T_{d-1}(x) and its subcomplex Zd−1(x)Z_{d-1}(x) such that

  1. Zd−1(x)⊆P(x)Z_{d-1}(x) \subseteq P(x);

  1. for every simplex FF of Td(x)T_d(x) contained in P(x)P(x),

dim⁡(F∩Zd−1(x))≤dim⁡F−1;\dim(F \cap Z_{d-1}(x)) \le\dim F - 1;

(iii) every connected component of P(x)∖Zd−1(x)P(x) \setminus Z_{d-1}(x) has ρ\rho-diameter at most 2R2R.

A negative dimension bound means that the intersection is empty. Thus the new separator avoids the vertices of TdT_d and meets each higher-dimensional face in lower dimension; its own vertices belong to the finer triangulation Td−1T_{d-1}. Connected components use the subspace topology given by ρ\rho. We call a filtration satisfying these conditions admissible. In particular, these conditions imply dim⁡Zi(x)≤i\dim Z_i(x) \le i.

The local operation behind the construction is easiest to describe at a fixed parameter. For a separator Z⊆PZ \subseteq P, a center c∈∣K∣c \in|K|, and t∈(r,R)t \in(r,R), remove the part of ZZ inside the ball and insert the corresponding slice of PP:

Z′=(Z∖Ut(c))∪(P∩St(c)).Z' = (Z \setminus U_t(c)) \cup(P \cap S_t(c)).

The inserted slice prevents a component of P∖Z′P \setminus Z' from passing between the ball and its exterior. Inside the ball its diameter is at most 2t2t; outside, the old separator remains. The next lemma makes this operation compatible with triangulations, group translates, and clopen parameter choices. It also constructs the level cuts used to start a filtration. Only the affine descriptions of distance levels enter the construction; no Euclidean metrics need to agree across faces.

Lemma 4.2 (Compatible cuts). Fix locally constant equivariant triangulation families TT and SS, where SS refines TT, and subcomplex families P⊆TP \subseteq T and Z⊆SZ \subseteq S with Z⊆PZ \subseteq P. Let {(cα,Bα)}α∈A\{(c_\alpha,B_\alpha)\}_{\alpha\in A} be a countable equivariant family of centers cα∈∣K∣c_\alpha\in|K| and clopen sets Bα⊆XB_\alpha\subseteq X. Assume that only finitely many of the balls LR(cα)L_R(c_\alpha) meet any given original simplex. Set

Ut(x)=⋃α:x∈BαUt(cα),St(x)=⋃α:x∈BαSt(cα).U_t(x) = \bigcup_{\alpha:x\in B_\alpha} U_t(c_\alpha), \qquad S_t(x) = \bigcup_{\alpha:x\in B_\alpha} S_t(c_\alpha).

Outside a countable set of radii t∈(r,R)t \in(r,R), there is a locally constant equivariant triangulation refining SS in which both

P(x)∩St(x),(Z(x)∖Ut(x))∪(P(x)∩St(x))(6)\begin{aligned} P(x) \cap S_t(x),\\ (Z(x) \setminus U_t(x)) \cup(P(x) \cap S_t(x)) \tag*{(6)} \end{aligned}

are subcomplexes. Each inserted level meets every face FF of TT or SS in dimension at most dim⁡F−1\dim F - 1. Consequently, if ZZ satisfies the dimension-drop condition relative to TT, both displayed sets do also. If every component of P∖ZP \setminus Z has diameter at most 2R2R, the second displayed set preserves that property.

Proof. On a fixed original simplex the function fc(p)=ρ(c,p)f_c(p) = \rho(c,p) is a maximum of finitely many affine functions. For a fixed face FF, a level {fc=t}\{f_c = t\} can have full dimension in FF only if one of these affine functions is identically tt on the affine span of FF. Otherwise the level is contained in a finite union of proper affine hyperplanes. Thus only finitely many levels are exceptional for this face and center. There are countably many original simplices and centers, and only finitely many exact triangulation types on each original simplex. Excluding the resulting countable union of exceptional levels gives the stated dimension drop simultaneously for all faces of TT and SS.

We give the compatible refinement explicitly. On each original simplex σ\sigma, cut the simplices of SS by all hyperplanes b=tb=t, where bb is an affine branch for one of the finitely many relevant centers. Use the canonical branch collection from the finite potential polytope in Section 3. On each piece of the finite clopen partition recording the old data and active conditions, this gives a finite polytopal complex Pσ\mathcal{P}_\sigma. It represents the old subcomplexes and the closed conditions defining the two sets in (6). These polytopal complexes may initially disagree on original faces. For each original simplex τ\tau, let QτQ_\tau be the common polytopal refinement of

Pσ∣τ\mathcal{P}_{\sigma| \tau}

over all original cofaces σ⊇τ\sigma\supseteq\tau, including σ=τ\sigma=\tau. There are finitely many cofaces by Lemma 3.1. The common refinement consists of the nonempty intersections of cells and their faces. If τ\tau is a face of σ\sigma, then QτQ_\tau refines Qσ∣τQ_{\sigma| \tau}, since every coface of σ\sigma is also a coface of τ\tau.

Triangulate the original simplices in increasing dimension. When handling σ\sigma, its boundary has already been triangulated by a refinement of the boundary restrictions of QσQ_\sigma. Extend this triangulation over the cells of QσQ_\sigma in increasing cell dimension. Keep cells in the original boundary as already triangulated. For every other positive-dimensional cell, cone its triangulated boundary to the arithmetic mean of its vertices, which lies in its relative interior. Zero-dimensional cells need no operation. The cones agree on common faces and give the desired finite affine subdivision, preserving the prescribed boundary triangulation.

Every operation commutes with the affine maps induced by HH: restriction, common refinement, and the mean of the full vertex set of a polytope are all canonical. This proves equivariance even for nontrivial simplex stabilizers. On each original simplex the procedure uses only finitely many data: those on its faces and their original cofaces, together with finitely many relevant cuts. A common finite clopen partition makes all these data constant. Thus the output has the required local constancy. No manifold condition on KK is involved.

The first set in (4.1) has the dimension bound already proved for the levels. The second is contained in the union of those levels with ZZ, so it has the same bound whenever ZZ does. For its separation property, let CC be a component of its complement in P(x)P(x). If CC meets an active Ut(cα)U_t(c_\alpha), continuity of fcαf_{c_\alpha} and connectedness of CC imply that CC stays in that ball, since it avoids the inserted level. Its diameter is then at most 2t<2R2t<2R. Otherwise CC lies outside all active open balls, where the old set ZZ was retained. It is therefore contained in an old component of P(x)∖Z(x)P(x)\setminus Z(x) and has diameter at most 2R2R. □

The genericity against SS is useful when SS already contains the old separator ZZ: intersections of an inserted level with its faces have one less dimension. The later area comparison will use precisely this conclusion. The lemma itself does not require disjoint active balls; disjointness will be imposed when their areas are counted.

Lemma 4.3. For any allowed choices through stage dd, an allowed choice of (Td−1,Zd−1)(T_{d-1},Z_{d-1}) exists. Hence a filtration can be continued to Z0Z_0.

Proof. Each original simplex is compact in ρ\rho. Cover representatives of the finitely many original simplex orbits by finitely many open balls Ur(c)U_r(c), and take all their HH-translates. These balls cover ∣K∣|K|. By Lemma 3.1, only finitely many translated radius-RR balls meet a given original simplex. Apply Lemma 4.2 with every activation set equal to XX, with T=S=TdT=S=T_d, and with Z=∅Z=\varnothing. For a common allowed radius t∈(r,R)t\in(r,R), take

Zd−1(x)=Zd(x)∩⋃cSt(c).Z_{d-1}(x)=Z_d(x)\cap\bigcup_c S_t(c).

This is a subcomplex of the resulting refinement and has the required dimension drop. If CC is a component of Zd(x)∖Zd−1(x)Z_d(x)\setminus Z_{d-1}(x), choose p∈Cp\in C and a covering ball Ur(c)U_r(c) containing it. The connected set fC(C)f_C(C) avoids tt and contains a value below tt, so C⊆Ut(c)C\subseteq U_t(c). Its diameter is at most 2t<2R2t<2R. The assertion is vacuous for an empty component space. □

Dimension drop along the entire filtration

We record the consequence that will control chains in Section 6. For any simplex FF of Ti(x)T_i(x) contained in Zi(x)Z_i(x) and any 0≤j≤i0 \le j \le i,

dim⁡(F∩Zj(x))≤dim⁡F−(i−j).\dim(F \cap Z_j(x)) \le\dim F - (i-j).

Indeed, the first step is the defining condition. Its intersection is a subcomplex in the next refinement. Applying the next dimension drop to each simplex of that subcomplex decreases its maximal dimension by one. Induction gives (4.2). Applied with i=ni=n, this also shows that a proper original face meets Zj(x)Z_j(x) in dimension strictly less than jj.

Lemma 4.4 (Carriers of the zero-dimensional stratum). Let (Ti,Zi)i=0n(T_i,Z_i)_{i=0}^n be an allowed filtration and let z∈Z0(x)z \in Z_0(x). For every ii, there is a unique ii-simplex of Ti(x)T_i(x) contained in Zi(x)Z_i(x) and containing zz. The point zz lies in its relative interior. In particular, every point of Z0(x)Z_0(x) lies in the interior of an original top simplex.

Proof. Let FF be the unique simplex of Ti(x)T_i(x) whose relative interior contains zz. Since Zi(x)Z_i(x) is a subcomplex containing zz, it contains FF, and dim⁡F≤i\dim F \le i. If dim⁡F<i\dim F < i, then (4.2) with j=0j=0 would make F∩Z0(x)F \cap Z_0(x) empty. Thus dim⁡F=i\dim F=i. Every other simplex containing zz has FF as a face, so no other ii-simplex contains it. Taking i=ni=n proves the last assertion. □\square

We have obtained a nonempty class of equivariant separating filtrations, together with a local cutting operation that stays within that class. The next section chooses separators with area close to the infimum and compares them with these cuts.

Local replacement and the number of zero strata

We now choose the separating filtration so that its zero-dimensional set has small expected cardinality in each original simplex. Throughout this section the graph, complex, parameter space, dimension, and radii are fixed as in Sections 3 and 4. In particular, the strict inequality (3.6) holds. Only the filtration will depend on a positive number ϵ\epsilon.

Proposition 5.1. For every ϵ>0\epsilon> 0 there is an admissible equivariant locally constant filtration (Zj,Tj)j=0n(Z_j,T_j)_{j=0}^n such that, for every original top simplex σ\sigma of KK,

∫X#(Z0(x)∩σ) dμ(x)≤Cσϵ.\int_X \#(Z_0(x) \cap\sigma)\,d\mu(x) \le C_\sigma\epsilon.

Here CσC_\sigma depends on the fixed geometric data but is independent of ϵ\epsilon and of the complexity of the chosen triangulations.

Successive near minimizers

The comparison with a cut through a ball follows the separating-set method of Papasoglu [22], Section 2, especially Lemma 2.5, and its iterated and equivariant forms in Alpert [1], Lemmas 6–7 and 11–12. We prove the comparisons for our simplex-dependent areas, retaining the error from each near minimizer.

Proof of Proposition 5.1. For an equivariant family ZZ of dimension at most jj, define

Ij(Z)=∑σ top mod H1∣Hσ∣∫Xajσ(Z(x)∩σ∘) dμ(x).(7)I_j(Z)=\sum_{\sigma\ \mathrm{top\ mod}\ H}\frac{1}{|H_\sigma|}\int_X a_j^\sigma\left(Z(x)\cap\sigma^\circ\right)\,d\mu(x). \tag*{(7)}

The sum runs over representatives of the original top simplices, and HσH_\sigma is the finite setwise stabilizer of σ\sigma. This quantity is finite for an admissible family. There are finitely many representatives, and, on each, local constancy and compactness of XX give finitely many finite polyhedral configurations, each of finite area.

Start with (Zn,Tn)=(∣K∣,K)(Z_n,T_n)=(|K|,K). At step d=n,…,1d=n,\ldots,1, keep all previously chosen data fixed and choose the admissible pair (Zd−1,Td−1)(Z_{d-1},T_{d-1}) so that Id−1(Zd−1)I_{d-1}(Z_{d-1}) is within ϵ\epsilon of the infimum over the admissible choices at this step. Lemma 4.3 ensures that this class is nonempty. Its infimum is a finite nonnegative number, so such a choice exists. Consequently, for every admissible replacement Z′Z' at this step,

Id−1(Zd−1)≤Id−1(Z′)+ϵ.(8)I_{d-1}(Z_{d-1})\leq I_{d-1}(Z')+\epsilon. \tag*{(8)}

Neither an attained minimum nor a limit of filtrations is used.

Disjoint replacements in the parameter and spatial variables

Fix a center c∈∣K∣c\in|K|. The set

Fc={h∈H∖{1}:LR(hc)∩LR(c)≠∅}\mathcal{F}_c=\{h\in H\setminus\{1\}:L_R(hc)\cap L_R(c)\neq\varnothing\}

is finite. Indeed such an intersection implies ρ(hc,c)≤2R\rho(hc,c)\leq2R; the support-distance bound, the finite graph balls and freeness on vertices then leave only finitely many possibilities for hh.

There is a finite clopen partition of XX into sets AA satisfying A∩hA=∅A\cap hA=\varnothing for every h∈Fch\in\mathcal{F}_c. To construct one, use the free action and the zero-dimensional Hausdorff topology to choose, around each x∈Xx\in X, a clopen neighborhood disjoint from its translates by this finite set. A finite subcover exists by compactness; the atoms of its common refinement have the required property. Thus

{hA×LR(hc):h∈H}\{hA\times L_R(hc):h\in H\}

is a pairwise disjoint family. {#eq:5.4}

This partition is fixed independently of ϵ\epsilon. The assertion concerns the product of parameter and spatial variables; the sets hAhA alone need not be disjoint.

Fix one atom AA and any clopen subset B⊆AB\subseteq A. For r≤s≤Rr\leq s\leq R put

Mj(s)=∫B∑σ topajσ(Zj(x)∩σ∘∩Ls(c)) dμ(x),0≤j≤n.(9)M_j(s)=\int_B\sum_{\sigma\ \mathrm{top}}a_j^\sigma\left(Z_j(x)\cap\sigma^\circ\cap L_s(c)\right)\,d\mu(x),\qquad0\leq j\leq n. \tag*{(9)}

For a generic t∈(r,R)t\in(r,R), also put

Nd−1(t)=∫B∑σ topad−1σ(Zd(x)∩σ∘∩St(c)) dμ(x).N_{d-1}(t)=\int_B\sum_{\sigma\ \mathrm{top}}a_{d-1}^\sigma\left(Z_d(x)\cap\sigma^\circ\cap S_t(c)\right)\,d\mu(x).

These sums involve only finitely many original simplices. Unlike (7), they run over all such simplices and carry no stabilizer weights.

At stage dd, write P=ZdP=Z_d and Z=Zd−1Z=Z_{d-1}. For each xx, make the simultaneous replacement

Z′(x)=(Z(x)∖⋃h:x∈hBUt(hc))∪(P(x)∩⋃h:x∈hBSt(hc)).Z'(x)=\left(Z(x)\setminus\bigcup_{h:x\in hB}U_t(hc)\right)\cup\left(P(x)\cap\bigcup_{h:x\in hB}S_t(hc)\right).

The active closed balls are pairwise disjoint by eq:5.4. The families of balls and levels are locally finite, and their use is conditioned on clopen sets, so the construction is equivariant and locally constant on every original simplex. Apply Lemma 4.2 with T=TdT=T_d, S=Td−1S=T_{d-1}, P=ZdP=Z_d, Z=Zd−1Z=Z_{d-1}, and the centers and activation sets (hC,hB)(h_C,h_B). Outside a countable set of levels, it represents Z′Z' by an admissible common refinement: the inserted levels have the required dimension drop on faces of TdT_d and of the old finer triangulation Td−1T_{d-1}, and the lemma preserves the diameter bound on components of P(x)∖Z′(x)P(x)\setminus Z'(x).

We next compute the change in (7). Genericity against the faces of Td−1T_{d-1} implies that the old set ZZ meets any inserted level in dimension at most d−2d-2. When d=1d=1, that intersection is empty. Hence the removed (d−1)(d-1)-area is unchanged if the open ball is replaced by its closure, and the inserted cut has zero (d−1)(d-1)-area overlap with the retained old set.

The normalization in (7) makes the removed area exactly Md−1(t)M_{d-1}(t) and the inserted area exactly Nd−1(t)N_{d-1}(t). Here is the orbit-counting identity behind this assertion. Let QQ denote the relevant equivariant set, and let W(c)W(c) denote either the ball or the level. Covariance of area and invariance of μ\mu give

∑σ top mod H1∣Hσ∣∑h∈H∫Bajσ(Q(x)∩σ∘∩W(hc)) dμ(x)=∑σ top mod H1∣Hσ∣∑h∈H∫Bajh−1σ(Q(x)∩(h−1σ)∘∩W(c)) dμ(x)=∫B∑τ topajτ(Q(x)∩τ∘∩W(c)) dμ(x).\sum_{\sigma\ \mathrm{top\ mod}\ H}\frac{1}{|H_\sigma|}\sum_{h\in H}\int_{B}a_j^\sigma\bigl(Q(x)\cap\sigma^\circ\cap W(hc)\bigr)\,\mathrm{d}\mu(x) = \sum_{\sigma\ \mathrm{top\ mod}\ H}\frac{1}{|H_\sigma|}\sum_{h\in H}\int_{B}a_j^{h^{-1}\sigma}\bigl(Q(x)\cap(h^{-1}\sigma)^\circ\cap W(c)\bigr)\,\mathrm{d}\mu(x) = \int_{B}\sum_{\tau\ \mathrm{top}}a_j^\tau\bigl(Q(x)\cap\tau^\circ\cap W(c)\bigr)\,\mathrm{d}\mu(x).

The map h↦h−1σh\mapsto h^{-1}\sigma takes each value in the simplex orbit exactly ∣Hσ∣|H_\sigma| times. All local contributions are nonnegative, and only finitely many translates meet any fixed representative simplex. The disjointness and zero-overlap statements above therefore yield

Id−1(Z′)−Id−1(Z)=Nd−1(t)−Md−1(t).I_{d-1}(Z')-I_{d-1}(Z)=N_{d-1}(t)-M_{d-1}(t).

Combining this identity with (8) gives

Md−1(t)≤Nd−1(t)+ϵfor almost every t∈(r,R).(10)M_{d-1}(t)\leq N_{d-1}(t)+\epsilon\quad\text{for almost every }t\in(r,R). \tag*{(10)}

Slicing and the strict volume margin

We have compared the area of a separating set inside a ball with that of a cut through the preceding set. Slicing now connects these comparisons across the successive dimensions.

Work first in one original top simplex σ\sigma and one of its tree metrics. Decompose Zd(x)∩σ∘Z_d(x)\cap\sigma^\circ into affine polyhedral pieces on which fc=ρ(c,⋅)f_c=\rho(c,\cdot) is affine. On a dd-dimensional piece QQ, its tangential gradient has norm α≤1\alpha\leq1 by Lemma 3.2. Euclidean slicing perpendicular to that gradient gives

∫rbvol⁡d−1(Q∩{fc=t}) dt=αvol⁡d(Q∩{r<fc<b})≤vol⁡d(Q∩Lb(c)).\int_r^b\operatorname{vol}_{d-1}(Q\cap\{f_c=t\})\,\mathrm{d}t =\alpha\operatorname{vol}_d(Q\cap\{r<f_c<b\}) \leq\operatorname{vol}_d(Q\cap L_b(c)).

A constant branch contributes at only one level. Pieces of dimension less than dd, including interfaces of this decomposition, have zero (d−1)(d-1)-slice area outside finitely many exceptional levels. Averaging over the tree metrics, summing over original top interiors, and integrating over BB yields

∫rbNd−1(t) dt≤Md(b),r≤b≤R.(11)\int_r^b N_{d-1}(t)\,\mathrm{d}t\leq M_d(b),\qquad r\leq b\leq R. \tag*{(11)}

These operations are legitimate by local finiteness and local constancy; on the fixed ball only finitely many configurations are involved. Every comparison has taken place in a single simplex metric before averaging, so no compatibility of metrics across original faces is needed. Indeed (4.2) bounds ZdZ_d on a proper original face by dimension d−1d-1, and its generic level slice by dimension d−2d-2. Such faces contribute no area to the preceding comparison. The same dimension estimate says that Z0Z_0 misses every proper original face.

From (10) and (11),

Md(b)≥∫rbMd−1(t) dt−ϵ(b−r).M_d(b) \ge\int_r^b M_{d-1}(t)\,\mathrm{d}t-\epsilon(b-r).

Induction, beginning with the monotonicity of M0M_0, gives

Md(b)≥(b−r)dd!M0(r)−ϵ∑i=1d(b−r)ii!.M_d(b) \ge\frac{(b-r)^d}{d!}M_0(r)-\epsilon\sum_{i=1}^{d}\frac{(b-r)^i}{i!}.

Set

L=R−r,An=Lnn!,En=∑i=1nLii!.L=R-r,\qquad A_n=\frac{L^n}{n!},\qquad\mathcal{E}_n=\sum_{i=1}^{n}\frac{L^i}{i!}.

Since Zn=∣K∣Z_n=|K|, the volume bound (5) implies

AnM0(r)≤VRμ(B)+ϵEn.(12)A_nM_0(r)\le V_R\mu(B)+\epsilon\mathcal{E}_n. \tag*{(12)}

We may now choose

B=A∩{x∈X:Z0(x)∩Lr(c)≠∅}.B=A\cap\{x\in X:Z_0(x)\cap L_r(c)\ne\varnothing\}.

This set is clopen: finitely many original simplices meet the ball, and their exact filtration data are constant on a common finite clopen partition. Although BB uses the completed filtration, its use in an earlier comparison is legitimate. Inequality (8) holds for every admissible competitor with the data down to stage dd fixed; it imposes no condition on the later stages. The same BB can therefore be used in every comparison leading to (12).

Each x∈Bx\in B contributes at least one zero stratum, and each such point lies in a unique original top interior. Thus μ(B)≤M0(r)\mu(B)\le M_0(r). Moreover M0(r)M_0(r) is the integral of the point count over all of AA, because that count is zero on A∖BA\setminus B. The strict inequality VR<AnV_R<A_n now gives the direct expected-count bound

∫A#(Z0(x)∩Lr(c)) dμ(x)=M0(r)≤EnAn−VRϵ.(13)\int_A\#(Z_0(x)\cap L_r(c))\,\mathrm{d}\mu(x)=M_0(r)\le\frac{\mathcal{E}_n}{A_n-V_R}\epsilon. \tag*{(13)}

To pass from the ball estimate to (5.1), cover a fixed original top simplex σ\sigma by finitely many open rr-balls with fixed centers c1,…,cqc_1,\ldots,c_q. Such a cover exists by compactness of the simplex. For each center, use the fixed clopen partition constructed above, and write pip_i for its number of atoms. Sum (13) over these atoms and then over the balls. Since the corresponding closed balls also cover σ\sigma, this proves (5.1) with

Cσ=EnAn−VR∑i=1qpi.C_\sigma=\frac{\mathcal{E}_n}{A_n-V_R}\sum_{i=1}^{q}p_i.

All quantities in this constant were fixed independently of ϵ\epsilon. In particular, no bound on the number of subdivision simplices or on the cardinality of an individual zero-dimensional fiber is needed. □\square

From separating filtrations to finite chains

Fix the dimension nn and radii chosen in Section 3. For each ϵ>0\epsilon> 0, use the filtration supplied by Proposition 5.1. We will turn it into an HH-equivariant chain map

Fϵ:C∗(K)⟶E∗F^\epsilon: C_*(K) \longrightarrow E_*

that is equivariantly chain homotopic to inclusion. Its evaluation by a fixed invariant cocycle will tend to zero. These two properties prove Theorem 2.1. Throughout this section nn remains fixed while ϵ\epsilon tends to zero.

Component averages and their finite support

For a fixed parameter xx, take the barycentric subdivision of the final triangulation T0(x)T_0(x). Its vertices are the barycenters bFb_F of nonempty simplices FF of T0(x)T_0(x). Define

ℓ(F)=min⁡{i:bF∈Zi(x)},Z−1(x)=∅.\ell(F) = \min\{i : b_F \in Z_i(x)\}, \qquad Z_{-1}(x) = \varnothing.

All Zi(x)Z_i(x) are subcomplexes of T0(x)T_0(x), so bF∈Zi(x)b_F \in Z_i(x) is equivalent to F⊆Zi(x)F \subseteq Z_i(x). Let UFU_F be the connected component of Zℓ(F)(x)∖Zℓ(F)−1(x)Z_{\ell(F)}(x) \setminus Z_{\ell(F)-1}(x) containing bFb_F. For any such component UU, set

JU=⋃p∈Usupp⁡p,uU=1∣JU∣∑v∈JUδv.(14)J_U = \bigcup_{p \in U} \operatorname{supp} p, \qquad u_U = \frac{1}{|J_U|}\sum_{v \in J_U} \delta_v. \tag*{(14)}

The next lemma proves that JUJ_U is finite and controls these labels as xx varies. The intended image of an oriented barycentric simplex is

[bF0,…,bFj]⟼uUF0∧⋯∧uUFj∈Ej.[b_{F_0}, \ldots, b_{F_j}] \longmapsto u_{U_{F_0}} \wedge\cdots\wedge u_{U_{F_j}} \in E_j.

This choice uses the separating filtration in a specific way. If nested faces F⊂F′F \subset F' have the same level ii, the segment joining their barycenters lies in Zi∖Zi−1Z_i \setminus Z_{i-1}. Except at its endpoint bFb_F, it lies in the relative interior of F′F', which the subcomplex Zi−1Z_{i-1} misses; the endpoint also avoids Zi−1Z_{i-1} by its level. The two barycenters therefore receive the same component vector, making any exterior term containing both zero. Along a complete flag F0⊂⋯⊂FnF_0 \subset\cdots\subset F_n, with dim⁡Fi=i\dim F_i = i, the levels are nondecreasing: if F⊂F′⊆ZiF \subset F' \subseteq Z_i, then F⊆ZiF \subseteq Z_i. A nonzero image must consequently use all levels 0,1,…,n0,1,\ldots,n, beginning at a point of Z0Z_0. We will bound the number of these flags after establishing that the labels define finite chains.

Lemma 6.1. For every original simplex κ\kappa of KK, there is a finite set Wκ⊆YW_\kappa\subseteq Y, independent of ϵ\epsilon and xx, such that JUF⊆WκJ_{U_F} \subseteq W_\kappa whenever bF∈κb_F \in\kappa. The triangulation over κ\kappa, together with all its assigned vectors uUFu_{U_F}, is locally constant in xx. These assignments are equivariant.

Proof. Each component of Zi(x)∖Zi−1(x)Z_i(x) \setminus Z_{i-1}(x) has ρ\rho-diameter at most 2R2R. For i=0i=0 the components are singletons, since Z0(x)Z_0(x) is a locally finite zero-dimensional subcomplex. If bF∈κb_F \in\kappa and p∈UFp \in U_F, the support-distance bound from Lemma 3.1 gives

min⁡dY(supp⁡bF,supp⁡p)≤n ρ(bF,p)≤2Rn.\min d_Y(\operatorname{supp} b_F,\operatorname{supp} p) \le\sqrt{n}\,\rho(b_F,p) \le2R\sqrt{n}.

Choose an original top simplex containing pp. Its n+1n+1 vertices span a connected graph, so every vertex in supp⁡p\operatorname{supp} p is within graph distance nn of the vertex furnished by this inequality. Consequently we may take

Wκ={v∈Y:dY(v,V(κ))≤[2Rn+n]}.W_\kappa= \{v \in Y : d_Y(v,V(\kappa)) \le[2R\sqrt{n}+n]\}.

The bounded degree of YY makes this set finite. Notice that the support of pp need not itself span a connected graph; we used the containing top simplex.

Let K[Wκ]K[W_\kappa] be the finite subcomplex consisting of the simplices whose vertices belong to WκW_\kappa. The whole component UFU_F lies in its realization. It follows that UFU_F is exactly the component containing bFb_F in

(Zℓ(F)(x)∖Zℓ(F)−1(x))∩∣K[Wκ]∣.\left(Z_{\ell(F)}(x) \setminus Z_{\ell(F)-1}(x)\right) \cap|K[W_\kappa]|.

Indeed UFU_F is a connected subset of this restriction, and every connected subset of the restriction containing bFb_F also lies in the original component. Thus restricting to this finite subcomplex does not cut off a part of the component needed to compute JUFJ_{U_F}.

There are finitely many original simplices in K[Wκ]K[W_\kappa]. A common refinement of their finite clopen parameter partitions makes all their triangulation and filtration data constant. On each resulting piece the components and their sets of support vertices are constant as well. This proves local constancy of the assignments over κ\kappa. Finally, JhU=hJUJ_{hU}=hJ_U, and hence uhU=huUu_{hU}=hu_U, proving equivariance.

Subdivide each oriented original simplex into its fundamental subdivision chain and apply the exterior assignment (6.2) to every term. This defines a map fx:C∗(K)→E∗f_x:C_*(K)\to E_*. Both operations commute with the augmented boundary. For subdivision, interior facets cancel in pairs and the boundary facets give the compatible subdivision on each original face. For the second operation, each assigned vector has coefficient sum one, so contraction with the coefficient-sum functional gives

∂(u0∧⋯∧uj)=∑i=0j(−1)iu0∧⋯∧ui^∧⋯∧uj.\partial(u_0\wedge\cdots\wedge u_j)=\sum_{i=0}^{j}(-1)^i u_0\wedge\cdots\wedge\widehat{u_i}\wedge\cdots\wedge u_j.

Thus fxf_x is an augmented chain map.

By Lemma 6.1, its value on any original simplex is constant on a finite clopen partition and is supported in a fixed finite exterior space. We may therefore define

Fϵ(c)=∫Xfx(c) dμ(x).F^\epsilon(c)=\int_X f_x(c)\,d\mu(x).

This integral is an actual finite chain: for every cc it is a finite linear combination of finitely many chain values. It commutes with the augmented boundary. Moreover, fhx(hc)=hfx(c)f_{hx}(hc)=hf_x(c), so invariance of μ\mu shows that FϵF^\epsilon is HH-equivariant.

Counting the surviving flags

We now bound the evaluation of fxf_x on a top simplex. The number of subdivision simplices may depend on ϵ\epsilon; the following bound depends instead on the number of zero-stratum points. The component-labeling and flag-counting method is modeled on Alpert [1], the carrier argument below supplies the bound without a manifold hypothesis.

Lemma 6.2. For every original top simplex σ\sigma and every xx, at most

2n#(Z0(x)∩σ)2^n\#(Z_0(x)\cap\sigma)

terms in its fundamental barycentric subdivision chain have nonzero images under the exterior replacement map.

Proof. Suppress xx. Such a term corresponds to a complete flag

F0⊂F1⊂⋯⊂Fn,dim⁡Fi=i,F_0 \subset F_1 \subset\cdots\subset F_n,\qquad\dim F_i=i,

in the restriction of T0T_0 to σ\sigma. As shown when defining (6.2), a nonzero term has levels exactly 0,1,…,n0,1,\ldots,n. In particular F0={z}F_0=\{z\} for a point z∈Z0∩σz\in Z_0\cap\sigma and Fi⊆ZiF_i\subseteq Z_i for every ii.

By Lemma 4.4, zz lies in the relative interior of a unique ii-simplex SiS_i of TiT_i contained in ZiZ_i. To see why this carrier controls the final refinement, let FiF_i occur in a surviving flag through zz. The TiT_i-carrier of a relative-interior point of FiF_i contains FiF_i by refinement and belongs to ZiZ_i, because ZiZ_i is a subcomplex. That carrier has dimension at least ii, since it contains FiF_i, and at most ii, since dim⁡Zi≤i\dim Z_i\le i. It contains zz, so it equals SiS_i. Hence Fi⊆SiF_i\subseteq S_i.

For a fixed preceding face Fi−1F_{i-1}, the possible FiF_i are thus full-dimensional simplices in a triangulation of the single affine ii-simplex SiS_i, having Fi−1F_{i-1} as a facet. There is at most one on either side of the affine hyperplane spanned by that facet. Otherwise two simplices on the same side would have intersecting relative interiors near an interior point of the facet. There are therefore at most two choices at each stage and at most 2n2^n surviving flags through zz. Summing over zz proves the assertion. This argument concerns one carrier simplex at a time and uses no manifold assumption on KK or on the strata.

Figure 1 illustrates the two choices at each of the two stages when n=2n=2.

The flag count in dimension two, showing shaded triangular choices

Figure 1. The flag count in dimension two. The point z∈Z0z\in Z_0 has two possible incident edges F1F_1 along its one-dimensional carrier. Each edge has at most two incident triangles F2F_2 inside the original two-dimensional carrier. The four shaded triangles give the four possible flags through zz in this example. The unshaded triangle has no one-dimensional-stratum edge through zz.

Let ψ:En→C\psi:E_n\to\mathbb{C} be the invariant cocycle in Theorem 2.1. For an original top simplex σ\sigma, put

Mσ=max⁡v0,…,vn∈Wσ∣ψ(δv0∧⋯∧δvn)∣.M_\sigma=\max_{v_0,\ldots,v_n\in W_\sigma}\left|\psi(\delta v_0\wedge\cdots\wedge\delta v_n)\right|.

This is finite, even though ψ\psi need not be bounded on all basis simplices of the infinite vertex set. Every vector in (14) is a nonnegative convex combination of vertices in WσW_\sigma. Multilinearity therefore bounds the evaluation of each surviving term by MσM_\sigma. The preceding lemma and Proposition 5.1 give

∣ψ(Fϵ(σ))∣≤2nMσ∫X#(Z0(x)∩σ) dμ(x)≤2nMσCσϵ⟶0.(15)\left|\psi(F^\epsilon(\sigma))\right| \le2^nM_\sigma\int_X \#(Z_0(x)\cap\sigma)\,d\mu(x) \le2^nM_\sigma C_\sigma\epsilon\longrightarrow0. \tag*{(15)}

All constants here are independent of ϵ\epsilon. Since chains have finite support, the same convergence holds on every fixed c∈Cn(K)c\in C_n(K).

Equivariant comparison and passage to coinvariants

It remains to show that the small evaluations just obtained equal the original cocycle evaluation on every coinvariant cycle. We do this by constructing a homotopy before taking coinvariants.

Lemma 6.3. Let KK be a simplicial complex with an HH-action and finite setwise simplex stabilizers. Let E∗E_* be the augmented exterior chain complex on a nonempty HH-set. Any two HH-equivariant augmented chain maps A,B:C∗(K)→E∗A,B:C_*(K)\to E_* that agree in degree −1-1 are HH-equivariantly chain homotopic.

Proof. The target is acyclic as an augmented complex. Indeed, for any fixed vertex vv, the operator a↦δv∧aa\mapsto\delta_v\wedge a satisfies

∂(δv∧a)+δv∧∂a=a.\partial(\delta_v\wedge a)+\delta_v\wedge\partial a=a.

This contraction need not be equivariant. We use it only to choose finite fillers and then impose equivariance by finite averaging.

Set h−1=0h_{-1}=0. Suppose the homotopy has been constructed below degree jj. For an oriented jj-simplex ee, the residual

r(e)=A(e)−B(e)−hj−1(∂e)r(e)=A(e)-B(e)-h_{j-1}(\partial e)

is a cycle, by the chain-map identities and the previously constructed homotopy. Choose a finite chain tt with ∂t=r(e)\partial t=r(e). Let SS be the setwise stabilizer of ee and let χ:S→{1,−1}\chi:S\to\{1,-1\} be its orientation character, so that se=χ(s)ese=\chi(s)e. The residual is equivariant, and hence sr(e)=χ(s)r(e)sr(e)=\chi(s)r(e). Define

tˉ=1∣S∣∑s∈Sχ(s) st.\bar{t}=\frac{1}{|S|}\sum_{s\in S}\chi(s)\,st.

Then ∂tˉ=r(e)\partial\bar{t}=r(e) and stˉ=χ(s)tˉs\bar{t}=\chi(s)\bar{t}. Choose such a filler for one oriented representative of every simplex orbit, put hj(e)=tˉh_j(e)=\bar{t}, and extend equivariantly and linearly. The orientation-character identity makes the extension well-defined, including when the stabilizer reverses orientation. This gives A−B=∂h+h∂A-B=\partial h+h\partial in degree jj and completes the induction. □

Proof of Theorem 2.1. The geometric construction applies for every sufficiently large nn depending only on DD, as fixed by (3.6). For nonempty KK, construct FϵF^\epsilon as above. The stabilizers are finite by Lemma 3.1, and FϵF^\epsilon and inclusion ι\iota both induce the identity in augmented degree −1-1. Lemma 6.3 supplies an equivariant homotopy

ι−Fϵ=∂h+h∂.\iota-F^\epsilon=\partial h+h\partial.

Let cˉ\bar{c} be a cycle of Cn(K)HC_n(K)_H and choose a finite chain lift c∈Cn(K)c\in C_n(K). We do not assume that cc is a cycle before taking coinvariants. Its boundary is a finite sum of differences ha−aha-a. Equivariance of the homotopy makes hn−1(∂c)h_{n-1}(\partial c) another such sum, which the invariant functional ψ\psi annihilates. The cocycle identity annihilates ∂hn(c)\partial h_n(c). Thus, for every ϵ>0\epsilon>0,

ψ(ι(c))=ψ(Fϵ(c)).\psi(\iota(c))=\psi(F^\epsilon(c)).

The right-hand side tends to zero by (15), whereas the left-hand side does not depend on ϵ\epsilon. Hence it is zero. The evaluation depends only on the coinvariant class because ψ\psi is invariant. For empty KK there is no nonzero top chain to consider. □

Cycles from an idempotent

We now return to group rings. For an idempotent and one infinite-order conjugacy class, we will construct cycles in every positive even degree. Their vertices follow steps in one fixed graph of bounded degree. In sufficiently high dimension, the supporting complexes will therefore satisfy the hypotheses of Theorem 2.1. We first record why it suffices to perform this construction for finitely generated groups.

The use of cyclic idempotent expressions and reduced centralizers has its conceptual background in Connes’s pairing with projective classes [12] and Burghelea’s conjugacy-class decomposition [10]. Our finite path identities and the detector’s normalization are proved directly; no cyclic-homology identification is assumed.

Lemma 7.1 (Subgroup locality). Let i:L↪Gi:L\hookrightarrow G be an inclusion of discrete groups. Extension of scalars defines i∗:K0(CL)→K0(CG)i_*:K_0(\mathbb{C}L)\to K_0(\mathbb{C}G), and for every y∈K0(CL)y\in K_0(\mathbb{C}L) and C∈con⁡(G)C\in\operatorname{con}(G),

HS⁡G(i∗y)(C)=∑B∈con⁡(L)B⊆L∩CHS⁡L(y)(B).(16)\operatorname{HS}_G(i_*y)(C)=\sum_{\substack{B\in\operatorname{con}(L)\\B\subseteq L\cap C}}\operatorname{HS}_L(y)(B). \tag*{(16)}

Only finitely many terms are nonzero. Moreover, every x∈K0(CG)x\in K_0(\mathbb{C}G) is induced from a finitely generated subgroup, which may be required to contain any prescribed finite subset of GG. Consequently, the finite-order-support assertion of Theorem 1.1 for all finitely generated groups implies that assertion for all discrete groups.

Proof. We regard free right modules as columns: an idempotent e∈Mr(CL)e\in M_r(\mathbb{C}L) acts on the left, with image e(CL)re(\mathbb{C}L)^r. Tensoring the split decomposition defined by ee with CG\mathbb{C}G identifies the extended image with e(CG)re(\mathbb{C}G)^r. The coefficient sum in an ambient class CC is the sum of its coefficient sums over all LL-classes contained in L∩CL\cap C. This proves (7.1) for the image of an idempotent and then, by subtraction, for every virtual class. Finiteness follows from the finite coefficient support of the matrices representing that class.

Write an arbitrary x∈K0(CG)x\in K_0(\mathbb{C}G) as the difference of the classes of two finite idempotent matrices. The group generated by their coefficient supports and the prescribed finite subset is finitely generated. The same matrices are idempotent over its group ring, since that ring embeds in CG\mathbb{C}G, and their difference induces xx. Finally, if an ambient class consists of infinite-order elements, so does every subgroup class occurring in (7.1): subgroup inclusion preserves the order of an element. Vanishing of each subgroup coefficient therefore gives vanishing of the ambient coefficient. This uses all the subgroup classes in the sum, since an ambient conjugacy class can split into several of them. □

A fixed graph and finite path sums

For the rest of this section, fix a finitely generated group GG, an idempotent e∈Mr(CG)e\in M_r(\mathbb{C}G), and an infinite-order element g∈Gg\in G. Write

e=∑s∈GAss,As∈Mr(C),S0={s∈G:As≠0}.e=\sum_{s\in G}A_s s,\qquad A_s\in M_r(\mathbb{C}),\qquad S_0=\{s\in G:A_s\ne0\}.

The set S0S_0 is finite. Idempotence says that ∑uv=tAuAv=At\sum_{uv=t}A_uA_v=A_t for every t∈Gt\in G. Thus the matrix-valued kernel

a(h,h′)=Ah−1h′a(h,h')=A_{h^{-1}h'}

is invariant under simultaneous left translation and satisfies

∑k∈Ga(h,k)a(k,h′)=∑uv=h−1h′AuAv=a(h,h′).\sum_{k\in G} a(h,k)a(k,h')=\sum_{uv=h^{-1}h'} A_uA_v=a(h,h').

All these sums are finite. In particular, the order of the matrix factors agrees with the column convention for right modules.

Put

J=CG(g),H=J/⟨g⟩,Y=⟨g⟩\G,hˉ=⟨g⟩h.J=C_G(g),\qquad H=J/\langle g\rangle,\qquad Y=\langle g\rangle\backslash G,\qquad\bar{h}=\langle g\rangle h.

Choose a finite symmetric generating set S⊂G∖{1}S\subset G\setminus\{1\} containing S0∖{1}S_0\setminus\{1\} and its inverses. Give YY the simple undirected graph whose edges join hˉ\bar{h} to hs‾\overline{hs} for s∈Ss\in S; loops are omitted and repeated edges identified. The graph is connected and has degree at most the fixed integer

D=max⁡{2,∣S∣}.D=\max\{2,|S|\}.

The group HH acts by left multiplication. This is well-defined because JJ centralizes gg, and it preserves right-step edges. It is free on vertices: chˉ=hˉc\bar{h}=\bar{h} with c∈Jc\in J implies ch=gqhch=g^qh for an integer qq, and hence c=gqc=g^q. Both HH and YY are countable. We keep this graph and this value of DD throughout all subsequent choices of dimension.

For k≥1k\geq1, let Qk=J\GkQ_k=J\backslash G^k, with simultaneous left translation on tuples. Define the weight of a tuple by

wk(h0,…,hk−1)=tr⁡(a(h0,h1)⋯a(hk−2,hk−1)a(hk−1,gh0)),(17)w_k(h_0,\ldots,h_{k-1})=\operatorname{tr}(a(h_0,h_1)\cdots a(h_{k-2},h_{k-1})a(h_{k-1},gh_0)), \tag*{(17)}

where w1(h0)=tr⁡(a(h0,gh0))w_1(h_0)=\operatorname{tr}(a(h_0,gh_0)). These weights are JJ-invariant. They define a finite complex linear combination vkv_k of elements of QkQ_k. Indeed, a nonzero weight has increments

si=hi−1hi+1(0≤i<k−1),sk−1=hk−1−1gh0s_i=h_i^{-1}h_{i+1}\quad(0\leq i<k-1),\qquad s_{k-1}=h_{k-1}^{-1}gh_0

in S0S_0, with

s0⋯sk−1=h0−1gh0.s_0\cdots s_{k-1}=h_0^{-1}gh_0.

An increment list determines the tuple from h0h_0. Any two choices of h0h_0 with this conjugate differ by left multiplication in JJ, so the list determines at most one element of QkQ_k. Hence vkv_k has at most ∣S0∣k|S_0|^k nonzero coefficients. For a JJ-invariant function pp on GkG^k, write vk(p)v_k(p) for the finite weighted sum of its values. These weights need not be real or positive.

Lemma 7.2 (Path identities). For the path sums associated to ee and gg, the twisted cyclic shift

T(h0,…,hk−1)=(h1,…,hk−1,gh0)T(h_0,\ldots,h_{k-1})=(h_1,\ldots,h_{k-1},gh_0)

preserves vkv_k. For k≥2k\geq2, deletion of any entry induces a map di:Qk→Qk−1d_i:Q_k\to Q_{k-1} satisfying

(di)∗vk=vk−1(0≤i<k).(d_i)_*v_k=v_{k-1}\qquad(0\leq i<k).

Every vkv_k has the same total weight

λ:=vk(1)=∑s∈[g]Gtr⁡(As)=HS⁡G([ℓ(CG)T])([g]G).(18)\lambda:=v_k(1)=\sum_{s\in[g]_G}\operatorname{tr}(A_s)=\operatorname{HS}_G([\ell(\mathbb{C}G)^T])([g]_G). \tag*{(18)}

Proof. The shift TT is a JJ-equivariant bijection, with inverse

(h0,…,hk−1)↦(g−1hk−1,h0,…,hk−2).(h_0,\ldots,h_{k-1}) \mapsto(g^{-1}h_{k-1},h_0,\ldots,h_{k-2}).

For k=1k=1, the shift is the identity on Q1Q_1, since g∈Jg \in J. For k≥2k \ge2, the shifted weight has the factors of (17) in cyclic order, since a(gh0,gh1)=a(h0,h1)a(gh_0,gh_1)=a(h_0,h_1). Cyclicity of the matrix trace proves T∗vk=vkT_*v_k=v_k.

For deletion, fix a representative of the shorter tuple. Each orbit in its fibre has a unique representative with exactly those retained entries, because simultaneous left translation acts freely on any nonempty tuple. Thus the fibre sum is over the missing entry in GG, without a stabilizer factor. Internal deletions follow from (7.2), as does deletion of the last entry using

∑ua(hk−2,u)a(u,gh0)=a(hk−2,gh0).\sum_u a(h_{k-2},u)a(u,gh_0)=a(h_{k-2},gh_0).

For the first deletion, fix (h1,…,hk−1)(h_1,\ldots,h_{k-1}) and put B=a(h1,h2)⋯a(hk−2,hk−1)B=a(h_1,h_2)\cdots a(h_{k-2},h_{k-1}), with B=IrB=I_r when k=2k=2. Then

∑h0tr⁡(a(h0,h1)Ba(hk−1,gh0))=∑h0tr⁡(Ba(hk−1,gh0)a(gh0,gh1))=tr⁡(Ba(hk−1,gh1)).\begin{aligned} \sum_{h_0}\operatorname{tr}(a(h_0,h_1)Ba(h_{k-1},gh_0)) &=\sum_{h_0}\operatorname{tr}(Ba(h_{k-1},gh_0)a(gh_0,gh_1))\\ &=\operatorname{tr}(Ba(h_{k-1},gh_1)). \end{aligned}

Here the first equality uses trace cyclicity and left invariance; the second uses (7.2) with the substitution u=gh0u=gh_0. This also proves the assertion for k=2k=2. The kernel has finite row and column support, so every fibre sum used in these computations is finite.

The deletion identities show that all total weights equal v1(1)v_1(1). Finally,

J\G⟶[g]G,Jh⟼h−1ghJ\backslash G\longrightarrow[g]_G,\qquad Jh\longmapsto h^{-1}gh

is a bijection: equality of the conjugates is equivalent to h′h−1∈Jh'h^{-1}\in J. Therefore v1(1)=∑s∈[g]Gtr⁡(As)v_1(1)=\sum_{s\in[g]_G}\operatorname{tr}(A_s), which is precisely the indicated Hattori–Stallings coefficient.

Even-dimensional cycles

The path identities now turn the finite sums into cycles. We use the exterior complex E∗E_* on the graph YY. For k≥1k\ge1, define

πk:C(Qk)⟶(Ek−1)H,[h0,…,hk−1]⟼[δhˉ0∧⋯∧δhˉk−1].\pi_k:C(Q_k)\longrightarrow(E_{k-1})_H,\qquad[h_0,\ldots,h_{k-1}]\longmapsto[\delta\bar h_0\wedge\cdots\wedge\delta\bar h_{k-1}].

The map is well-defined because changing a tuple by an element of JJ changes its projected vertices by a simultaneous HH-translation. Set bk=πk(vk)b_k=\pi_k(v_k), and, for n=2m≥2n=2m\ge2, set

zn=bn+1∈(En)H.z_n=b_{n+1}\in(E_n)_H.

Proposition 7.3. For every positive even integer nn, the chain znz_n is a cycle. Moreover, either its defining sum has no nonzero distinct-vertex terms, so zn=0z_n=0, or it is the image of a cycle in Cn(K)HC_n(K)_H for an HH-invariant simplicial complex KK generated by nn-simplices, with finitely many simplex orbits, whose top vertex sets span connected subgraphs of YY. The graph and its degree bound DD are independent of nn.

Proof. Since gh0‾=h0‾\overline{gh_0}=\overline{h_0}, the twisted shift projects to the cyclic permutation of the kk wedge factors. Hence

πkT=(−1)k−1πk.\pi_k T = (-1)^{k-1}\pi_k.

Together with T∗vk=vkT_*v_k=v_k, this shows that bk=0b_k=0 whenever kk is even. The deletion identities give

∂bk=∑i=0k−1(−1)ibk−1(k≥2).\partial b_k=\sum_{i=0}^{k-1}(-1)^i b_{k-1}\qquad(k\ge2).

For even nn, it follows that ∂zn=bn=0\partial z_n=b_n=0.

A tuple with repeated projected vertices contributes zero to the wedge. For every remaining tuple with nonzero weight, consecutive projected vertices are joined by edges from the fixed step set SS. Their path visits all its n+1n+1 vertices, so the subgraph spanned by those vertices is connected. Let KK be generated by these vertex sets and their HH-translates. Finite support of vn+1v_{n+1} gives finitely many orbits of top simplices and of their faces.

The same finite expression for znz_n defines a chain in Cn(K)HC_n(K)_H. To see that its boundary is zero there, decompose EjE_j in every degree as the direct sum of the span of the oriented simplices belonging to KK and the span of all other oriented simplices. Both summands are HH-invariant, including their orientation signs. Thus

Ej=Cj(K)⊕Wj,(Ej)H=Cj(K)H⊕(Wj)H.E_j=C_j(K)\oplus W_j,\qquad(E_j)_H=C_j(K)_H\oplus(W_j)_H.

In particular the inclusion remains injective after taking coinvariants. The boundary of the chain in Cn(K)HC_n(K)_H maps to ∂zn=0\partial z_n=0, so it is zero already in Cn−1(K)HC_{n-1}(K)_H.

We have obtained the required sparse cycles in arbitrarily large even degrees. The underlying path sums all have total weight λ\lambda, the trace coefficient in (18). The next section constructs a cocycle whose value on z2mz_{2m} recovers a nonzero scalar multiple of that coefficient.

A Pfaffian cocycle detecting the trace coefficient

Retain the finitely generated group GG, idempotent, infinite-order element gg, and notation J,H,Y,vk,λ,z2mJ,H,Y,v_k,\lambda,z_{2m} of Section 7. We construct an HH-invariant cocycle whose value on z2mz_{2m} is 2−mλ2^{-m}\lambda. The construction first records the relative powers of gg between two lifts of vertices of YY. An alternating Pfaffian expression will then remove the dependence on those lifts.

Relative exponents between lifts

Lemma 8.1. There is a JJ-invariant function β:G×G→R\beta:G\times G\to\mathbb{R} such that

β(h′,h)=−β(h,h′),β(grh,gsh′)=β(h,h′)+s−r(r,s∈Z).(19)\beta(h',h)=-\beta(h,h'),\qquad\beta(g^r h,g^s h')=\beta(h,h')+s-r\qquad(r,s\in\mathbb{Z}). \tag*{(19)}

Proof. The diagonal HH-action on the set of ordered pairs Y×YY\times Y is free, because its action on the first coordinate is free. This also applies to diagonal pairs (y,y)(y,y). For each orbit choose a representative (y0,y0′)(y_0,y'_0) and lifts (l0,l0′)∈G2(l_0,l'_0)\in G^2. For (h,h′)∈G2(h,h')\in G^2, there is a unique cˉ∈H\bar c\in H carrying the appropriate representative to the projected pair. Choose a lift c∈Jc\in J and write

h=grcl0,h′=gscl0′.h=g^rcl_0,\qquad h'=g^scl'_0.

For fixed cc, the integers r,sr,s are unique because gg has infinite order. Replacing cc by gtcg^{t}c replaces them by r−t,s−tr-t,s-t. Thus B(h,h′)=s−rB(h,h')=s-r is well defined. Commutation with gg shows that BB is JJ-invariant, and independently changing the two lifts gives

B(gah,gbh′)=B(h,h′)+b−a.B(g^{a}h,g^{b}h')=B(h,h')+b-a.

The antisymmetrization

β(h,h′)=B(h,h′)−B(h′,h)2\beta(h,h')=\frac{B(h,h')-B(h',h)}{2}

has all the required properties.

The function β\beta itself need not descend to Y×YY\times Y. In fact, (19) gives β(h,gkh)=k\beta(h,g^{k}h)=k even though these two lifts project to the same vertex. Nor is there a difficulty if an element cˉ∈H\bar c\in H exchanges two distinct projected vertices. If a lift c∈Jc\in J satisfies ch0=gah1ch_{0}=g^{a}h_{1} and ch1=gbh0ch_{1}=g^{b}h_{0}, its invariance merely requires

2β(h0,h1)=b−a,2\beta(h_{0},h_{1})=b-a,

which is precisely what antisymmetrization gives. Real values are useful here: for G=⟨t⟩G=\langle t\rangle and g=t2g=t^{2}, the value β(1,t)=1/2\beta(1,t)=1/2 is forced.

The cyclic Pfaffian identity

In degree two, a lift-independent expression is

c(y0,y1,y2)=β(h1,h2)−β(h0,h2)+β(h0,h1),hˉi=yi.c(y_{0},y_{1},y_{2})=\beta(h_{1},h_{2})-\beta(h_{0},h_{2})+\beta(h_{0},h_{1}),\qquad\bar h_{i}=y_{i}.

Changing hih_{i} to grihig^{r_{i}}h_{i} adds (r2−r1)−(r2−r0)+(r1−r0)=0(r_{2}-r_{1})-(r_{2}-r_{0})+(r_{1}-r_{0})=0, so cc depends only on the projected vertices. In higher even degrees we replace β\beta by a Pfaffian and take the same alternating deletion sum. The Pfaffian still depends on the lifts; its deletion sum will define the cocycle. We first compute the weighted Pfaffian sum, then prove the descent and cocycle properties of its deletion sum.

For a list h=(h0,…,h2m−1)\mathbf{h}=(h_{0},\ldots,h_{2m-1}) with m≥1m\ge1, let

pm(h)=Pf⁡(β(hi,hj))0≤i,j<2m,p0=1.p_{m}(\mathbf{h})=\operatorname{Pf}\bigl(\beta(h_{i},h_{j})\bigr)_{0\le i,j<2m},\qquad p_{0}=1.

We use the convention that pm(h)p_{m}(\mathbf{h}) is the coefficient of ε0∧⋯∧ε2m−1\varepsilon_{0}\wedge\cdots\wedge\varepsilon_{2m-1} in bm/m!b^{m}/m!, where εi\varepsilon_{i} are the coordinate one-forms on the space of list positions and

b=∑i<jβ(hi,hj) εi∧εj.b=\sum_{i<j}\beta(h_{i},h_{j})\,\varepsilon_{i}\wedge\varepsilon_{j}.

In particular, p1(h0,h1)=β(h0,h1)p_{1}(h_{0},h_{1})=\beta(h_{0},h_{1}), and pmp_{m} is alternating in its entries and JJ-invariant. The finite weighted sums

qm=v2m(pm)(m≥1),q0=λq_{m}=v_{2m}(p_{m})\quad(m\ge1),\qquad q_{0}=\lambda

are therefore defined. No boundedness of pmp_{m} on all lists is needed.

Lemma 8.2. For every m≥1m\ge1, one has qm=12qm−1q_{m}=\frac{1}{2}q_{m-1} and hence qm=2−mλq_{m}=2^{-m}\lambda.

Proof. An ordinary cyclic rotation of a list of length 2m2m changes its Pfaffian by the sign (−1)2m−1=−1(-1)^{2m-1}=-1. In the twisted rotation

Th=(h1,…,h2m−1,gh0),T\mathbf{h}=(h_1,\ldots,h_{2m-1},gh_0),

replacing the last entry h0h_0 by gh0gh_0 adds 1 to every matrix entry in the last column above the diagonal. Expansion along that column gives

pm(Th)=−pm(h)+∑i=12m−1(−1)i+1pm−1(h1,…,hi^,…,h2m−1).(20)\begin{aligned} p_m(T\mathbf{h})&=-p_m(\mathbf{h})\\ &\quad+\sum_{i=1}^{2m-1}(-1)^{i+1}p_{m-1}(h_1,\ldots,\widehat{h_i},\ldots,h_{2m-1}). \tag*{(20)} \end{aligned}

Indeed the last-column cofactor in position ii, with positions numbered 1,…,2m1,\ldots,2m, has sign (−1)i+1(-1)^{i+1}; each Pfaffian matching uses exactly one entry from that column.

For m≥2m\ge2, each minor on the second line omits h0h_0 and hih_i from the original list. Applying the deletion identity in Lemma 7.2 twice shows that its weighted sum under v2mv_{2m} is qm−1q_{m-1}. For m=1m=1, the minor is p0=1p_0=1, whose weighted sum is the total weight v2(1)=λ=q0v_2(1)=\lambda=q_0. Thus this endpoint uses no vector v0v_0. The invariance of v2mv_{2m} under TT and ∑i=12m−1(−1)i+1=1\sum_{i=1}^{2m-1}(-1)^{i+1}=1 now turn (20) into

qm=−qm+qm−1.q_m=-q_m+q_{m-1}.

The claimed formulas follow.

Descent to an exterior cocycle

The preceding calculation detects the coefficient on lists of lifts. We next descend it to the exterior complex on YY, where the sparse-chain theorem applies.

Proposition 8.3. For every integer m≥1m\ge1, the formula

ψm(y0,…,y2m)=∑i=02m(−1)ipm(h0,…,hi^,…,h2m),\psi_m(y_0,\ldots,y_{2m})=\sum_{i=0}^{2m}(-1)^i p_m(h_0,\ldots,\widehat{h_i},\ldots,h_{2m}),

where hi∈Gh_i\in G is any lift of yi∈Yy_i\in Y, defines an HH-invariant linear functional ψm:E2m→C\psi_m:E_{2m}\to\mathbb{C} satisfying

ψm∂=0,ψm(z2m)=2−mλ.\psi_m\partial=0,\qquad\psi_m(z_{2m})=2^{-m}\lambda.

Here evaluation on the coinvariant chain z2mz_{2m} is defined by HH-invariance.

Proof. On the 2m+12m+1 list positions put

u=∑i=02mεi,b=∑0≤i<j≤2mβ(hi,hj)εi∧εj.u=\sum_{i=0}^{2m}\varepsilon_i,\qquad b=\sum_{0\le i<j\le2m}\beta(h_i,h_j)\varepsilon_i\wedge\varepsilon_j.

The right side of (8.3) is the coefficient of the ordered top form in u∧bm/m!u\wedge b^m/m!. Moving the factor εi\varepsilon_i past its ii preceding positions gives exactly the sign (−1)i(-1)^i in that formula.

Changing the lifts to grihig^{r_i}h_i replaces β(hi,hj)\beta(h_i,h_j) by β(hi,hj)+rj−ri\beta(h_i,h_j)+r_j-r_i. Thus it replaces bb by b+u∧vb+u\wedge v, with v=∑iriεiv=\sum_i r_i\varepsilon_i. Every correction term in u∧(b+u∧v)mu\wedge(b+u\wedge v)^m contains two factors uu and vanishes. This proves independence of the lifts.

The expression is alternating in the lifted list. After lift independence has been established, it is therefore alternating in the projected vertices as well. In particular, if two projected vertices coincide, choose equal lifts for them to see that the expression vanishes. It consequently defines a linear functional on E2m=⋀2m+1C(Y)E_{2m}=\bigwedge^{2m+1}\mathbb{C}^{(Y)}. For cˉ∈H\bar{c}\in H, choose c∈Jc\in J and use chic h_i as the lifts of the translated vertices. The JJ-invariance of β\beta proves the HH-invariance of ψm\psi_m.

To check the cocycle identity, choose lifts simultaneously for a list of 2m+22m+2 vertices. On every face, (8.3) is the alternating deletion coboundary of pmp_m on these same lifts. A second deletion cancels each twice-omitted list in its two possible orders. Hence ψm∂=0\psi_m\partial=0.

Finally, z2mz_{2m} is the exterior projection of v2m+1v_{2m+1}. Each one-deletion term in (8.3) has weighted sum v2m(pm)=qmv_{2m}(p_m)=q_m, by Lemma 7.2. Since ∑i=02m(−1)i=1\sum_{i=0}^{2m}(-1)^i=1, their signed sum gives

ψm(z2m)=qm=2−mλ,\psi_m(z_{2m})=q_m=2^{-m}\lambda,

using Lemma 8.2. Repeated projected vertices already have zero exterior image and zero value under ψm\psi_m, so they require no separate restriction in this evaluation.

Completion of the trace theorem

We now combine the geometric vanishing with the detector. The order of the choices matters: the graph is fixed by the idempotent before its cycle dimension is chosen.

Proof of Theorem 1.1. First let GG be finitely generated, let e∈Mq(CG)e\in M_q(\mathbb{C}G) be an idempotent, and fix an infinite-order g∈Gg\in G. Construct the graph on Y=⟨g⟩\GY=\langle g\rangle\backslash G and the free action of H=CG(g)/⟨g⟩H=C_G(g)/\langle g\rangle as in Section 7. Let D≥2D\ge2 be a degree bound for this fixed graph, and choose an even integer n=2m≥n0(D)n=2m\ge n_0(D) with m≥1m\ge1.

Proposition 7.3 gives the cycle znz_n, and Proposition 8.3 gives an HH-invariant cocycle ψm:En→C\psi_m:E_n\to\mathbb{C} with

ψm(zn)=2−mλ,λ=HS⁡G([e(CGq)])([g]G).\psi_m(z_n)=2^{-m}\lambda,\qquad\lambda=\operatorname{HS}_G([e(\mathbb{C}G^q)])([g]_G).

If no nonzero simplices occur, zn=0z_n=0 and this identity already gives λ=0\lambda=0. Otherwise Proposition 7.3 realizes znz_n as a cycle in Cn(K)HC_n(K)_H for an HH-invariant complex KK satisfying all hypotheses of Theorem 2.1. That theorem gives ψm(zn)=0\psi_m(z_n)=0, with the same conclusion. Every finitely generated projective right CG\mathbb{C}G-module is represented by an idempotent, and every element of K0(CG)K_0(\mathbb{C}G) is a difference of two such classes. Additivity proves the assertion for all finitely generated groups.

For an arbitrary discrete group GG, choose idempotents representing x∈K0(CG)x\in K_0(\mathbb{C}G) and let G0≤GG_0\le G be generated by gg and all group elements occurring in their entries. Then G0G_0 is finitely generated and x=i∗x0x=i_*x_0 for some x0∈K0(CG0)x_0\in K_0(\mathbb{C}G_0), where i:G0↪Gi:G_0\hookrightarrow G. Lemma 7.1 expresses HS⁡G(x)([g]G)\operatorname{HS}_G(x)([g]_G) as the sum of the coefficients HS⁡G0(x0)(C)\operatorname{HS}_{G_0}(x_0)(C) over those G0G_0-conjugacy classes C⊆G0∩[g]GC\subseteq G_0\cap[g]_G. Only finitely many terms are nonzero, and every element in every such class has infinite order. Each term vanishes by the finitely generated case.

The integral trace

For the integral group ring, the same idempotent formula defines

HS⁡ZG:K0(ZG)⟶ZG/[ZG,ZG]≅⨁C∈con⁡(G)ZC\operatorname{HS}_{\mathbb{Z}G}:K_0(\mathbb{Z}G)\longrightarrow\mathbb{Z}G/[\mathbb{Z}G,\mathbb{Z}G]\cong\bigoplus_{C\in\operatorname{con}(G)}\mathbb{Z}C

where commutators are now taken with their additive span. The integral Bass trace conjecture asks for support only at the identity class.

Corollary 9.1. For every discrete group GG, the Hattori–Stallings trace of every class in K0(ZG)K_0(\mathbb{Z}G) is supported on [1]G[1]_G.

Proof. If PP is a finitely generated projective right ZG\mathbb{Z}G-module, the same idempotent matrix over CG\mathbb{C}G represents P⊗ZGCGP\otimes_{\mathbb{Z}G}\mathbb{C}G. Consequently its trace coefficients are the images under Z↪C\mathbb{Z}\hookrightarrow\mathbb{C} of those of PP. Theorem 1.1 and this coefficientwise injection give vanishing at all infinite-order elements. No injectivity assertion about the map between the two K0K_0 groups is required.

For finite-order elements we use Linnell’s restriction [19], in the form stated and reproved by Berrick and Hesselholt [7]: if the integral trace of a finitely generated projective module is nonzero at hh, there is an integer a≥1a\geq1 such that hh is conjugate to hsah^{s^a} for every integer s≥1s\geq1. Their convention also uses right modules. If hh has finite order t>1t>1, take s=ts=t. Then hta=1h^{t^a}=1, so the restriction would make hh conjugate to 11, a contradiction. Thus every nonidentity finite-order coefficient also vanishes. Taking differences of projective classes finishes the proof. □

Homotopy idempotents and Lefschetz numbers

The integral support statement also has fixed-point consequences. A self-map ff is a homotopy idempotent if f∘f≃ff\circ f\simeq f. Such a map on a connected finite CW complex XX admits maps u:X→Yu:X\to Y and v:Y→Xv:Y\to X with

v∘u≃f,u∘v≃id⁡Yv\circ u\simeq f,\qquad u\circ v\simeq\operatorname{id}_Y

[6], Section 3. The second identity says that YY is finitely dominated by XX. For a pointed homotopy idempotent with f#=id⁡Gf_{\#}=\operatorname{id}_G, where G=π1(X)G=\pi_1(X), the induced maps u#u_{\#} and v#v_{\#} identify π1(Y)\pi_1(Y) with GG. The universal-cover chains of YY are chain homotopy equivalent to a finite complex P∗P_\ast of finitely generated projective ZG\mathbb{Z}G-modules. The sum ∑i(−1)i[Pi]∈K0(ZG)\sum_i(-1)^i[P_i]\in K_0(\mathbb{Z}G) is its projective Euler class. The Hattori–Stallings trace of this class records the fixed-point class indices of ff: it is the Reidemeister trace, with conjugacy classes inverted when passing from the left-module convention of [6], Lemma 4.1, to ours. The augmentation trace and the identity coefficient of this projective class give the ordinary and L2L^2-Lefschetz numbers, respectively [6], Lemmas 8.1–8.2. Berrick, Chatterji and Mislin use these relations to deduce the following consequences of the integral support statement and its numerical trace consequence.

Corollary 9.2 (Berrick–Chatterji–Mislin consequences). Let MM be a connected closed smooth oriented manifold.

(i) If dim⁡M>2\dim M>2 and f:M→Mf:M\to M satisfies f∘f≃ff\circ f\simeq f by free homotopy, then some self-map freely homotopic to ff has exactly one fixed point.

(ii) Choose x0∈Mx_0\in M and put G=π1(M,x0)G=\pi_1(M,x_0). If f:(M,x0)→(M,x0)f:(M,x_0)\to(M,x_0) satisfies f∘f≃∗ff\circ f\simeq_*f by based homotopy and f#=id⁡Gf_{\#}=\operatorname{id}_G, then

L(f)=L(2)(f~).L(f)=L^{(2)}(\widetilde{f}).

Here f~\widetilde{f} is the lift to the universal cover fixing a chosen point above x0x_0; it is GG-equivariant. The left side is the ordinary Lefschetz number, and the right side is the alternating sum of von Neumann traces on the induced L2L^2-homology maps, as in [6], Definition 7.1.

Proof. For x∈K0(ZG)x\in K_0(\mathbb{Z}G) write hC(x)=HS⁡ZG(x)(C)h_C(x)=\operatorname{HS}_{\mathbb{Z}G}(x)(C). The augmentation trace ϵ\epsilon is the sum of these coefficients, while the Kaplansky trace κ\kappa is the identity coefficient. The sum is finite, and Corollary 9.1 gives

ϵ(x)=∑C∈con⁡(G)hC(x)=h[1]G(x)=κ(x).\epsilon(x)=\sum_{C\in\operatorname{con}(G)}h_C(x)=h_{[1]_G}(x)=\kappa(x).

This is the weak Bass conjecture for every group. The cited paper uses projective left modules, whereas we use right modules. The anti-involution g↦g−1g\mapsto g^{-1}, together with transpose on idempotent matrices, transports between these conventions and relabels the coefficient at [g]G[g]_G by [g−1]G[g^{-1}]_G. It preserves identity support and both ϵ\epsilon and κ\kappa. Thus Corollary 9.1 gives the classical Bass conjecture in the cited convention, and the displayed equality gives weak Bass there. The forward implications of [6] now give (i) and (ii), respectively.

Torsion-free groups: trace rank and idempotents

For a torsion-free group, the support theorem leaves only the identity conjugacy class. We first identify its coefficient with augmentation rank, and then use the two possible scalar ranks to determine scalar idempotents.

Proof of Corollary 1.2(i). Let e∈Mq(CG)e\in M_q(\mathbb{C}G) be idempotent and put x=[e(CG)q]∈K0(CG)x=[e(\mathbb{C}G)^q]\in K_0(\mathbb{C}G). Since GG is torsion-free, Theorem 1.1 says that HS⁡G(x)\operatorname{HS}_G(x) is supported on [1]G[1]_G. Therefore

τG,q(e)=HS⁡G(x)([1]G)=∑C∈con⁡(G)HS⁡G(x)(C)=ϵG(∑i=1qeii)=tr⁡(ϵG,q(e))=rank⁡C(ϵG,q(e)).(21)\tau_{G,q}(e)=\operatorname{HS}_G(x)([1]_G) =\sum_{C\in\operatorname{con}(G)}\operatorname{HS}_G(x)(C) =\epsilon_G\left(\sum_{i=1}^{q}e_{ii}\right) =\operatorname{tr}(\epsilon_{G,q}(e)) =\operatorname{rank}_{\mathbb{C}}(\epsilon_{G,q}(e)). \tag*{(21)}

The class sum is finite because the entries of ee have finite group support. The last equality holds because ϵG,q(e)\epsilon_{G,q}(e) is an idempotent complex matrix: its image-kernel decomposition makes its trace equal to its rank. Taking differences of projective classes gives τG,∗=rk⁡ϵ\tau_{G,*}=\operatorname{rk}_{\epsilon}. This map has range exactly Z\mathbb{Z}, since its values are integers and [CG][\mathbb{C}G] has value 1.

For matrices the conclusion is numerical. For example, diag⁡(1,0)∈M2(CG)\operatorname{diag}(1,0)\in M_2(\mathbb{C}G) is an idempotent other than zero and the identity, with augmentation rank one. The scalar case is special because its augmentation rank can only be zero or one.

Proof of Corollary 1.2(ii). First take R=CR=\mathbb{C} and let e∈CGe\in\mathbb{C}G be idempotent. Equation (10.1) with q=1q=1 gives

τG(e)=ϵG(e)∈{0,1}.\tau_G(e)=\epsilon_G(e)\in\{0,1\}.

We use the standard faithful-trace argument, as in [9], spelling out a similarity to a projection. Embed CG\mathbb{C}G by left convolution in the group von Neumann algebra N(G)\mathcal{N}(G), the commutant of the right regular representation on ℓ2(G)\ell^2(G). Its canonical normalized trace is

τG(T)=⟨Tδ1,δ1⟩.\tau_G(T)=\langle T\delta_1,\delta_1\rangle.

Its faithfulness can be seen directly: if T≥0T\geq0 and τG(T)=0\tau_G(T)=0, then T1/2δ1=0T^{1/2}\delta_1=0. Since T1/2T^{1/2} commutes with right translations and their translates of δ1\delta_1 span a dense subspace, T1/2=0T^{1/2}=0, hence T=0T=0.

Set

a=e∗e+(1−e)∗(1−e).a=e^*e+(1-e)^*(1-e).

For every ξ∈ℓ2(G)\xi\in\ell^{2}(G),

⟨aξ,ξ⟩=∥eξ∥2+∥(1−e)ξ∥2≥12∥ξ∥2.\langle a\xi,\xi\rangle= \lVert e\xi\rVert^{2} + \lVert(1-e)\xi\rVert^{2} \ge\frac{1}{2}\lVert\xi\rVert^{2}.

Thus aa is positive and invertible in N(G)\mathcal{N}(G), and a direct multiplication gives ae=e∗aae=e^{*}a. With s=a1/2s=a^{1/2}, the element p=ses−1p=ses^{-1} satisfies p2=pp^{2}=p and p∗=s−1e∗s=ses−1=pp^{*}=s^{-1}e^{*}s=ses^{-1}=p. Hence pp is a projection, and traciality gives τG(p)=τG(e)\tau_G(p)=\tau_G(e). If this value is zero, faithfulness applied to p=p∗pp=p^{*}p gives p=0p=0. If it is one, faithfulness applied to 1−p1-p gives p=1p=1. Similarity then gives e=0e=0 or e=1e=1.

Now let RR be any commutative unital domain of characteristic zero and write an idempotent f∈RGf\in RG as f=∑g∈Srggf=\sum_{g\in S}r_g g with SS finite. Put

R0=Z[rg:g∈S]⊆R,F=Frac⁡(R0).R_0=\mathbb{Z}[r_g:g\in S]\subseteq R,\qquad F=\operatorname{Frac}(R_0).

The field FF is finitely generated over Q\mathbb{Q}, and therefore embeds in C\mathbb{C}. Indeed, send a finite transcendence basis of F/QF/\mathbb{Q} to an algebraically independent tuple of complex numbers; the resulting embedding of the rational function field extends across the finite algebraic extension FF because C\mathbb{C} is algebraically closed. The coefficientwise maps R0G↪RGR_0G\hookrightarrow RG and R0G↪CGR_0G\hookrightarrow\mathbb{C}G are injective. Thus f2=ff^{2}=f already holds in R0GR_0G, and its image in CG\mathbb{C}G is an idempotent. The complex case makes that image 00 or 11, and injectivity gives the same conclusion for ff. Only the finitely generated coefficient field FF was embedded; no embedding of all of RR into C\mathbb{C} is required.

Remark 10.1 (Algebraic and completed trace ranges). Continue to assume that GG is torsion-free. Let j:CG↪Cr∗(G)j:\mathbb{C}G\hookrightarrow C_r^{*}(G) be the inclusion into the reduced group C∗C^{*}-algebra, and let

j∗:K0(CG)⟶K0top(Cr∗(G))j_{*}:K_{0}(\mathbb{C}G)\longrightarrow K_{0}^{\mathrm{top}}(C_r^{*}(G))

be the comparison map to topological KK-theory. Write τ∗top:K0top(Cr∗(G))→R\tau_{*}^{\mathrm{top}}:K_{0}^{\mathrm{top}}(C_r^{*}(G))\to\mathbb{R} for the canonical trace pairing. It agrees with τG,∗\tau_{G,*} on this image: the same positive similarity used above, applied in a matrix algebra over Cr∗(G)C_r^{*}(G), produces a projection representing the image of a matrix idempotent, and traciality preserves its trace. Consequently

τ∗top(j∗K0(CG))=Z.\tau_{*}^{\mathrm{top}}\bigl(j_{*}K_{0}(\mathbb{C}G)\bigr)=\mathbb{Z}.

This is the integer range on the algebraic image, not a statement about every class of K0top(Cr∗(G))K_{0}^{\mathrm{top}}(C_r^{*}(G)). In particular, the argument does not presume that the algebraic augmentation extends to the reduced completion. The stronger completed idempotent question is the Kadison–Kaplansky conjecture; see [21], Section 1.

The coefficient reduction above uses characteristic zero. An idempotent e∉{0,1}e\notin\{0,1\} would give nonzero zero divisors through e(1−e)=0e(1-e)=0, so the zero-divisor conjecture would imply the idempotent conjecture. The reverse implication is not established here.

References

  1. [1]Hannah Alpert. Simplicial volume and 0-strata of separating filtrations, 2024. arXiv:2211.06362v3.arxiv.org/abs/2211.06362
  2. [2]Arthur Bartels, Wolfgang Lück, and Holger Reich. On the Farrell–Jones conjecture and its applications. Journal of Topology, 1(1):57–86, 2008.arxiv.org/abs/math/0703548
  3. [3]Hyman Bass. Euler characteristics and characters of discrete groups. Inventiones Mathematicae, 35:155–196, 1976.DOI
  4. [4]Hyman Bass. Traces and Euler characteristics. In C. T. C. Wall, editor, Homological Group Theory, volume 36 of London Mathematical Society Lecture Note Series, pages 1–26. Cambridge University Press, Cambridge, 1979.DOI
  5. [5]A. J. Berrick, Indira Chatterji, and Guido Mislin. From acyclic groups to the Bass conjecture for amenable groups. Mathematische Annalen, 329(4):597–621, 2004.
  6. [6]A. J. Berrick, Indira Chatterji, and Guido Mislin. Homotopy idempotents on manifolds and Bass’ conjectures. Geometry & Topology Monographs, 10:41–62, 2007.
  7. [7]A. J. Berrick and Lars Hesselholt. Topological Hochschild homology and the Bass trace conjecture. J. reine angew. Math., 704:169–185, 2015.
  8. [8]Sabine Braun and Roman Sauer. Volume and macroscopic scalar curvature. Geometric and Functional Analysis, 31(6):1321–1376, 2021.arxiv.org/abs/2012.08999
  9. [9]Marc Burger and Alain Valette. Idempotents in complex group rings: theorems of Zalesskii and Bass revisited. Journal of Lie Theory, 8(2):219–228, 1998.DOI
  10. [10]Dan Burghelea. The cyclic homology of the group rings. Commentarii Mathematici Helvetici, 60:354–365, 1985.
  11. [11]Indira Chatterji and Guido Mislin. Hattori–Stallings trace and Euler characteristics for groups. In Martin R. Bridson, Peter H. Kropholler, and Ian J. Leary, editors, Geometric and Cohomological Methods in Group Theory, volume 358 of London Mathematical Society Lecture Note Series, pages 256–271. Cambridge University Press, 2009.
  12. [12]Alain Connes. Non-commutative differential geometry. Publications Mathématiques de l’IHÉS, 62:41–144, 1985.DOI
  13. [13]Beno Eckmann. Cyclic homology of groups and the Bass conjecture. Commentarii Mathematici Helvetici, 61:193–202, 1986.
  14. [14]F. Thomas Farrell and Peter A. Linnell. Whitehead groups and the Bass conjecture. Mathematische Annalen, 326:723–757, 2003.
  15. [15]Edward Formanek. Idempotents in Noetherian group rings. Canadian Journal of Mathematics, 25(2):366–369, 1973.DOI
  16. [16]Akira Hattori. Rank element of a projective module. Nagoya Mathematical Journal, 25:113–120, 1965.DOI
  17. [17]Greg Hjorth and Mats Molberg. Free continuous actions on zero-dimensional spaces. Topology and its Applications, 153(7):1116–1131, 2006.
  18. [18]L. V. Kantorovich. On the translocation of masses. Journal of Mathematical Sciences, 133(4):1381–1382, 2006. Translation of the 1942 original.DOI
  19. [19]P. A. Linnell. Decomposition of augmentation ideals and relation modules. Proceedings of the London Mathematical Society, 47(1):83–127, 1983.
  20. [20]Igor Mineyev and Guoliang Yu. The Baum–Connes conjecture for hyperbolic groups. Inventiones Mathematicae, 149(1):97–122, 2002.
  21. [21]Johan Öinert and Stefan Wagner. Complex group rings and group C*-algebras of group extensions. Journal of Algebraic Combinatorics, 58:387–397, 2023.arxiv.org/abs/1811.06424
  22. [22]Panos Papasoglu. Uryson width and volume. Geometric and Functional Analysis, 30(2):574–587, 2020.
  23. [23]Michael Puschnigg. The Kadison–Kaplansky conjecture for word-hyperbolic groups. Inventiones Mathematicae, 149(1):153–194, 2002.DOI
  24. [24]John Stallings. Centerless groups—an algebraic formulation of Gottlieb’s theorem. Topology, 4(2):129–134, 1965.DOI

Paper details

vlnc:W_2wqqkbv8
Version
v1 · latest
Published
License
LicenseRef-OpenAI-public
Usage

views · 40

downloads · 2

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

unreleased internal OpenAI model

View statement
Looking for related papers…
Contents