Integral homology lattices of matching and circulant complexes

The rational cycle isomorphism identifies \mathcal H_k/\Psi_k(S^{(k)}) with \widehat S^{(k)}/S^{(k)}. The exponent and the sharp classes are given by Theorem 2.2, whose final argument also proves the asserted integral nonisomorphism.

An independent set in C_{3s}^{s-1} has consecutive circular gaps at least s. Consequently the only triangles of \mathcal Y_s are \{i,i+s,i+2s\}, 0\leq i<s, and there are no higher faces. Each triangle edge belongs to exactly one triangle. Choose a directed Hamiltonian edge step d_0 as follows:

Each step is a unit modulo 3s, is 1 modulo three, and belongs to an edge orbit different from the triangle-edge orbit. For the second even case v\geq3; when v is even, v\geq2 suffices. Collapsing each triangle along one of its edges preserves this Hamiltonian cycle, so the resulting graph is connected. There are 3s(s+1)/2 edges before collapsing and s triangles. Thus the collapse gives a wedge of (s-1)(3s-2)/2 circles.

We now work on the original oriented chains, without requiring the collapse to preserve the action. Write e_{d,i}=[i,i+d], with subscripts modulo n=3s, and let u=\sum_{i=0}^{n-1}e_{d_0,i}. For each other directed edge step d, let a_d be the integer in \{1,\ldots,n-1\} congruent to d d_0^{-1} modulo n, and define

These are cycles, with

The full edge orbits give n such cycles apiece. If n is even, use only 0\leq i<n/2 for the half-turn orbit. Together with u, these cycles form an integral basis of graph homology: subtracting their coefficients eliminates all non-Hamiltonian edges from any cycle, and the remainder is an integer multiple of u. Independence follows by inspecting those edges, then one Hamiltonian edge.

For the triangle step d=s, one has a_s=s because d_0\equiv1\pmod3. Put c=c_{s,0}. The triangle boundaries are therefore the translates of

This accounts for all attaching relations. When s is odd, all other edge orbits are full. Eliminating u in (?), and using Tu=u, gives precisely the relation (t-1)N_s c=0. For w=t^s c, (?) gives Rw=-w. There are (s-3)/2 other non-Hamiltonian edge orbits. Their generators have reflection Rc_d=-t^{-d}c_d; a shift t^a c_d replaces the exponent by -2a-d. Since n is odd, each exponent can be made zero modulo n. This proves the odd decomposition over \mathbb Z, with the two specified generators of the group.

Suppose now that s=2v. The half-turn chord h=c_{3v,0} satisfies c_{3v,i+3v}=-c_{3v,i}-u; the two base paths together traverse the Hamiltonian cycle once. Hence, after eliminating u, the complete presentation of the special two-generator part is

The relations t^{6v}c=c and t^{6v}h=h follow from these two relations, so no periodicity relation is omitted. For example, multiplying the second relation by t^{3v}-1 gives the relation for h, since t^{3v}-1 is divisible by t-1.

Set x=t^v, A=1+x+x^2, B=1-x+x^2, and C=1+x. Then N_{2v}=AB, 1+x^3=BC, and A-xC=1. The integral change of generators

is invertible over \mathbb Z[t,t^{-1}]. The two relations in (?) become

Indeed the second transformed relation follows after subtracting a multiple of Bz=0, and conversely it recovers (t-1)ABc=0. This establishes a direct sum, rather than merely annihilating polynomials for two classes. The reflection formulas in the new generators are

Replacing z by z'=t^{-2v}z and w by w'=w-xz gives Rz'=-z' and Rw'=-t^{-v}w'. For the latter equality, the coefficient of z in R(w-xz) is

where x^3=-1 and x^6=1 modulo B. Thus R(w-xz)=-x^{-1}w+z=-x^{-1}(w-xz). These replacements preserve the direct sum because B divides (t-1)ABC. Since ABC=1+x+\cdots+x^5, they give the blocks B_v and W_v with their stated reflections.

The remaining full edge orbits have steps 2v,\ldots,3v-1, excluding the triangle step and the Hamiltonian step. The latter has odd absolute step. The phase in (?) can be shifted to zero for an even step and to one for an odd step, producing \Pi_{6v}^0 and \Pi_{6v}^1. Counting these steps gives (a_v,b_v)=(v/2-1,v/2-1) for even v, and ((v-1)/2,(v-3)/2) for odd v. This proves the even decomposition and completes the full integral action calculation.

The one-skeleton has circular edge distances s,\ldots,\lfloor n/2\rfloor. An independent set has all circular gaps at least s, so it has at most three vertices in the stated range. For a=1, the triangles have gaps (s,s,s+1) and form one full orbit of length n. For a=2, there are two full orbits: type A has gaps (s,s,s+2) and type B has gaps (s,s+1,s+1). The three gaps are not all equal, since 3\nmid n, so none of these orbits has a nontrivial stabilizer. Every triangle belongs to one of the listed orbits.

Choose the Hamiltonian edge step d_0=s for a=1, and d_0=s+1 for a=2. Its inverse modulo n is respectively -3 or 3. Use the cycles c_{d,i} and u constructed in (?), with this new n,d_0. Their integral graph-cycle basis and their translation and reflection formulas have the same proof: elimination of the non-Hamiltonian edge coefficients leaves an integer multiple of u. The half-turn chords, when present, again satisfy

These are graph-cycle identities before any triangle is filled.

For a=1, the base-path length for step s+1 is n-2. The triangle with consecutive steps s,s,s+1 therefore has boundary

Its translates impose precisely c_{s+1,j}=-u for every j. For a=2, the base-path lengths for steps s and s+2 are n-2 and four. The boundaries of type B and type A, respectively, are

For the first boundary, the two Hamiltonian edges complete a base path of total length n. For the second, the total length of the three paths is 2(n-2)+4=2n, giving the coefficient two on u. Thus all attaching relations are equivalent to c_{s,j}=-u and c_{s+2,j}=0 for every j. These substitutions are compatible with T and R and have coefficient one, so they eliminate entire chord orbits integrally. They impose no further relation on the remaining graph-cycle generators.

The ordinary collapse also makes this last assertion transparent at the level of homology groups. For a=1, every edge of distance s+1 is in exactly one triangle; collapse each triangle along that edge. For a=2, each distance-s+2 edge occurs in exactly one type-A triangle and no type-B triangle. Collapse all type-A triangles first; then each distance-s edge is in exactly one remaining type-B triangle, and collapse those. The Hamiltonian edge orbit remains in both cases, so the resulting graph is connected. There are n(s+a+1)/2 edges initially and an triangles. The graph after collapse has first Betti number

It supplies free homology in degree one and no other reduced homology; the preceding chain relations retain the original group action without requiring the collapse to be equivariant.

The remaining full edge-orbit steps are

If n is odd, u generates \mathbb Z_- and each remaining full orbit is a signed regular lattice. Its reflection phase can be centered to zero, as in the odd case of Theorem 1.2. There are (s-a-1)/2 such orbits, proving the odd formula.

If n=2m is even, retain the half-turn orbit and eliminate u using (?). The complete relations on its generator are (t-1)(1+t^m)h=0 and t^{2m}h=h. The second follows from the first, so this is exactly H_m, with the reflection in (?). The other full orbits are independent regular summands; shifting their generators centers the reflection phase to their step parity. The first remaining step s+a+1 is odd, since n is even. The k=(s-a-2)/2 remaining steps therefore contribute \lceil k/2\rceil odd phases and \lfloor k/2\rfloor even phases, as asserted. The four endpoint cases have no regular summands and are included in these formulas.

Fix an oriented (k-1)-edge matching \sigma, leaving three ground vertices a,b,c. Orient each extension by placing its new edge last. Up to the common boundary sign, the coefficient of \sigma in \partial\Psi_k(x) is the pairing with

The last equality follows from \sum_iq_i=0: the sum of the three remaining vectors is minus the sum of the edge-sum vectors already occurring in b_\sigma. This also proves the assertion for the augmented boundary when k=1.

For the stated formula, a coefficient is the determinant whose entries are \langle f_{i_t},q_a+q_b\rangle. An edge with both endpoints in the complement of \{0\}\cup I gives a zero column. If a matching has no such edge but leaves a vertex of that complement unmatched, it has one edge entirely in \{0\}\cup I and uses all vertices of \{0\}\cup I. In this case the sum of its determinant columns is zero, since each f_{i_t} has coordinate sum zero on that set. The only remaining matchings pair every r_s with a distinct element of \{0\}\cup I. They are indexed uniquely by the displayed permutations, with \pi(0) unmatched. In the written edge order their determinant is \operatorname{sgn}(\pi): it is the determinant of \pi on the augmentation lattice of \{0\}\cup I. Thus there are exactly (k+1)! distinct summands with coefficients 1 or -1, proving both the normalization and primitivity.

The upper bound. We prove the upper inclusion by induction in steps of two. It is a statement about alternating forms: if a rational alternating k-form x on Q^{(k)} has integer values on every b_m, then 2^{\lfloor k/2\rfloor}x is integer-valued on \bigwedge^kQ^{(k)}. For k=0 use the one-vertex set and the empty matching, so the assertion is simply x\in\mathbb Z. For k=1, q_a+q_b=-q_c on three labels, so the assertion is immediate.

Suppose k\geq2. Choose two disjoint edges ab,cd and let W be the other 2k-3 ground labels. Contract x with their edge-sum vectors and restrict the remaining arguments to \mathbb Z^W:

This descends to \mathbb Z^W/\mathbb Z\sum_{i\in W}e_i, because in Q^{(k)} the sum over W is -(q_a+q_b)-(q_c+q_d), which contraction annihilates. Its values on maximum matchings of W are integers: append the two fixed edges to obtain a maximum matching of the original ground set. The induction hypothesis therefore gives

whenever all the displayed ground labels are distinct.

Now fix distinct i_1,\ldots,i_{k-2}. There are k+3\geq5 other labels. On their free lattice consider the rational alternating bilinear form obtained by fixing the last k-2 arguments of 2^{\lfloor(k-2)/2\rfloor}x. Its values on all disjoint two-edge generators are integers. Lemma 2.3 shows that twice this form is integer-valued on every wedge of two basis vectors. Thus 2^{\lfloor k/2\rfloor}x is integer-valued on every basis wedge of Q^{(k)}. The perfect pairing proves the upper inclusion in (2.2).

Divided-power cycles. For the lower inclusion, the divided powers \omega^{[j]} are integral: their coefficients are the principal Pfaffians of the alternating matrix with all upper-triangular entries equal to one. Modulo two, \omega(q_a,q_b)=1 for every pair of distinct ground labels, including label zero since q_0=-\sum_{i>0}q_i and 2k is even. Consequently

for disjoint edges. Evaluate \omega^{[j]}\wedge v, where v\in\bigwedge^{k-2j}A^{(k)}, on b_m. The usual exterior product expansion is a sum of Pfaffians on 2j chosen edge-sum arguments, multiplied by the complementary evaluations of v. Each Pfaffian is a sum of products of j even entries. Every term is therefore divisible by 2^j. This proves that 2^{-j}\omega^{[j]}\wedge v lies in \widehat S^{(k)}.

Canonicity. First note that the Gram matrix of the dot product on A^{(k)}/2A^{(k)} in the f_i basis has zero diagonal and ones off the diagonal. Its square is the identity, so \overline\omega is its invariant inverse bivector. Any other integral lift, including g\omega for g\in\mathfrak S_{2k+1}, has the form \omega+2\eta with \eta\in\bigwedge^2A^{(k)}. Its divided powers satisfy

Every \eta^{[b]} is integral, again by the Pfaffian formula. Thus the new lift generates a sublattice of N^{(k)}. Interchanging the two lifts gives equality, proving both lift independence and invariance.

Sharpness. The principal Pfaffian on any ordered set of 2j indices is one. Indeed expansion along its first row reduces this to the alternating sum of 2j-1 ones, starting and ending with a plus sign. Choose any 2j indices disjoint from \{1,\ldots,k-2j\}. The corresponding wedge coefficient of \omega^{[j]}\wedge f_1\wedge\cdots\wedge f_{k-2j} is 1 or -1. Its class after division by 2^j therefore has exactly that order modulo S^{(k)}. Taking j=\lfloor k/2\rfloor, and using the upper inclusion, proves the exact exponent.

Integral nonisomorphism. The rational hook representation is absolutely irreducible, so an integral equivariant isomorphism S^{(k)}\to \mathcal H_k would, under \Psi_{k,\mathbb Q}, identify \widehat S^{(k)} with cS^{(k)} for some nonzero rational c. Since S^{(k)}\subseteq \widehat S^{(k)}, this forces c=\pm1/d for a positive integer d. Lemma 2.1 shows that each \Psi_k(f_{i_1}\wedge\cdots\wedge f_{i_k}) is primitive even in the full chain lattice, hence cannot be divided by any d>1 in \mathcal H_k. Thus d=1. For k\geq2 the order-two class already constructed shows \widehat S^{(k)}\ne S^{(k)}, a contradiction.

In the quotient by this subgroup write w_{ab} for the class of u_a\wedge u_b. Adding the three pairings on four distinct labels, always putting the pair containing a first, gives

For fixed a, comparison of triples among the other m-1\geq4 labels shows that 2w_{ab} is independent of b\ne a; denote it by t_a. The displayed relation gives 3t_a=0. Skew symmetry gives t_a=-t_b for distinct a,b. Using three distinct labels then gives 2t_a=0. Hence t_a=0, proving the assertion.

The class of \frac12\omega\wedge v depends precisely on the reduction of \omega\wedge v modulo 2S^{(k)}, giving the asserted image. On A^{(k)}/2A^{(k)} the dot product is a nondegenerate alternating form: its Gram matrix in the f_i basis has zero diagonal and ones off the diagonal, and its square is the identity. Thus \overline\omega is its invariant inverse bivector. This proves that the image is canonical and equivariant, although the displayed integral lift \omega depends on the basis.

Choose a symplectic basis u_i,v_i, 1\leq i\leq k. Multiplication by \overline\omega=\sum_i u_i\wedge v_i defines a differential of degree two on the exterior algebra, since its square is zero in characteristic two. This complex is the tensor product of the k complexes \bigwedge\langle u_i,v_i\rangle with differential 1\mapsto u_i\wedge v_i. Each factor has homology only in degree one, of dimension two. The tensor product therefore has homology only in degree k. It is exact in degrees below k. Recursively subtracting the rank of the preceding differential from \dim\bigwedge^{k-2}(A^{(k)}/2A^{(k)}) gives the stated alternating sum.

Norm is multiplication by p on A_P, so torsion-freeness gives A_P\cap D_P=0. Since N_P^2=pN_P, the identity px=N_Px+(p-N_P)x puts pL in A_P+D_P. The map [x]\mapsto N_Px\bmod pA_P is surjective onto N_PL/pA_P. If N_Px=pa with a\in A_P, then x-a\in D_P, proving injectivity and (3.1). For (3.1), use [x]\mapsto(\sigma-1)x\bmod(\sigma-1)D_P. If (\sigma-1)x=(\sigma-1)d for d\in D_P, then x-d\in A_P; its cokernel is D_P/(\sigma-1)L. The Tate-group descriptions are the standard cyclic norm formulas [BrownLectures, Sections 6.3–6.4].

The proof uses the classical Tate localization construction [Swan, BrownGroups], with complete resolutions as in [BrownLectures, Section 6.4]. We give its chain argument, including the empty-fixed-set convention. After barycentric subdivision, a stabilized simplex is a chain of faces of distinct dimensions, so its stabilizer fixes it pointwise. Thus F is a subcomplex and, because |P|=p, every simplex outside F has trivial stabilizer. Orient an orbit representative and transport its orientation. The bounded relative augmented chain complex \widetilde C_*(X)/\widetilde C_*(F) is consequently degreewise free over \mathbb Z[P].

Let \widehat P_* be the standard two-periodic complete free resolution of the trivial module. Place chain degree i in cohomological degree -i and form the total complex of \operatorname{Hom}_{\mathbb Z[P]}(\widehat P_*,\widetilde C_*(X)). Each total degree has finitely many chain-degree summands. Filtering by chain degree shows that the relative total complex is acyclic, since free \mathbb Z[P]-modules have zero Tate cohomology. The inclusion of fixed chains therefore induces an isomorphism on total cohomology. Filtering first by chain homology gives a spectral sequence with terms \widehat H^a(P,\widetilde H_i(X;\mathbb Z)) in total degree a-i. There is only the row i=q, so total degree j-q has cohomology \widehat H^j(P,L).

The action on fixed chains is trivial. Its Tate differential alternates between zero and multiplication by p. Pair resolution columns 2a-1 and 2a, whose differential is multiplication by p. Each pair is a translate of the cone of p on the fixed chain complex. Since the chain groups are free abelian, this cone is quasi-isomorphic to the quotient complex modulo p in column 2a. The translates therefore occur in even degrees. Hence, as \mathbb F_p-vector spaces,

In particular \dim\widehat H^0(P,L)=b_0.

Every nonidentity element of P has fixed space F. The simplicial Lefschetz formula gives trace (-1)^q\widetilde\chi(F)=b_0-b_1 on L\otimes\mathbb Q. Averaging this character yields

Subtracting b_0 in (3.1) proves (3.2). If b_0=0, then L^P=N_PL. For a primitive equivariant quotient u, one has u(N_Px)=pu(x) and u(L)=\mathbb Z. Thus u(L^P)=p\mathbb Z, which gives both the lower bound and an invariant element lifting p.

Barycentric subdivision makes F a subcomplex; the relative augmented chain groups outside F are free over \mathbb Z[K]. Use the two-periodic complete resolution \widehat P_* of \mathbb Z over \mathbb Z[K] and the total complex \operatorname{Hom}_{\mathbb Z[K]}(\widehat P_*,\widetilde C_*(X)), placing chain degree i in cohomological degree -i. Since G centralizes K, it acts on coefficients alone, commuting with both resolution differentials and all filtrations. The inclusion of fixed chains induces a G-equivariant isomorphism on total cohomology because the relative complex is degreewise free over \mathbb Z[K] and bounded in chain degree. Filtering by chain homology has just the q row for X and the d row for F; it yields, in total degree -q, an isomorphism \widehat H^0(K,L)\cong\widehat H^{d-q}(K,M) compatible with G. Both coefficient lattices have trivial K action, so the left side is L/pL, while even periodicity identifies the right side with M/pM without any centralizer twist. These coefficient actions factor through G/K, proving the first assertion.

For any lattice V free over \mathbb Z with a P-action, its invariant lattice A=V^P is primitive, since it is the kernel of \sigma-1 on a torsion-free lattice. Therefore A/pA\hookrightarrow V/pV, and the norm model identifies \mathcal G_P(V)=N_PV/pA\cong\operatorname{im}(\overline N_P:V/pV\to V/pV). The comparison L/pL\cong M/pM intertwines the residual norm operators. Taking their images proves the second assertion; the image is stable under the normalizer since conjugation permutes the summands of the norm.

Proposition 23.1 gives h=2s+1, q=1, and the full triangle list \{i,i+s,i+2s\}, 0\leq i<s, with no higher faces. For P_3=\langle T^s\rangle, these triangles are the vertex orbits. A fixed point has invariant support, so the fixed space consists of their s barycenters. Thus b_0=0 and b_1=s-1. Theorem 3.2 gives g=1, \widehat H^0(P_3,L)=0, and the asserted image of every primitive scalar quotient. This also gives a geometric proof of the full dihedral augmentation minimum in Proposition 23.1. Indeed L^{P_3} has rank one and maps injectively to the rank-one target; equivariance forces translation and reflection on it to have the target actions, and its generator maps to \pm3.

If p\ne3 divides s, an invariant simplex must contain a full orbit of p vertices. For p>3 there is none, since all cliques have size at most three. For p=2, the orbit is a half-turn pair with difference 3s/2, which is neither \pm1 nor \pm s modulo 3s. Again there is no invariant simplex, so the fixed space is empty. The augmented convention gives b_0=1, b_1=0, and (3.2) yields g=2s/p and \widehat H^0(P_p,L)\cong\mathbb F_p.

Use the geometric polynomial presentation in Proposition 23.1 and write N_j=1+t^j+t^{2j}. For p\ne3, K=N_s/N_r is monic integral of degree 2r(p-1) and

For example, a root of t^{3r}-1 is also a root of N_s exactly when it is a root of N_r, since p is coprime to three. Monic division therefore gives the entire integral invariant lattice L^P=KL. Put a'=T^{-r(p-1)}K(T)w. Its translates identify L^P with the algebraic module \mathbb Z[t]/((t-1)N_r), also for r=1,2. Reciprocity K(t^{-1})=t^{-2r(p-1)}K(t) gives Ra'=-a'. Thus this identification has translation t and reflection negative inversion.

The polynomial norm factors as

Norm commutes with T and R, since T centralizes P and R conjugates its generator to its inverse. Under the resulting equivariant norm model \mathcal G_P(L)\cong N_PL/pL^P from Lemma 3.1, the quotient is consequently the ideal generated by E in L^P/pL^P; the shift defining a' changes this generator by a unit only. In characteristic p, E=(t^r-1)^{p-1} is coprime to N_r, since N_r is three when t^r=1. Its image in \mathbb F_p[t]/((t-1)N_r) is therefore exactly the augmentation ideal (t-1). Multiplication by t-1 identifies that ideal with \mathbb F_p[t]/N_r. On the generator g=(t-1)a' of this norm-image model one has Rg=t^{-1}g, proving (3.5). Here g is a class in N_PL/pL^P, transported by the norm isomorphism; its literal class in L/(L^P+\ker N_P) would be zero.

Write r=p^a u with p\nmid u. Then N_r=N_u^{p^a} in \mathbb F_p[t], and a primitive 3uth root is a root of N_u of multiplicity one. Thus N_r\mid t^k-1 requires both 3u\mid k and p^a\mid k. Conversely N_r\mid t^{3r}-1, proving the exact rotation order. For P_3, its invariant lattice has rank one, with T=1 and R=-1, and its reduction modulo three is the norm gluing.

For K(14,2), each orbit of \langle T^2\rangle contains an edge of circular difference two, so both orbit vertices are looped and the fixed space is empty. The half-turn orbit graph is C_7^2. Its independence complex has seven vertices and seven edges of circular distance three, forming one seven-cycle, with no triangles. Since (h,q)=(13,2), formula (3.2) gives g=(13-6)/7=1 and g=(13-1)/2=6.

For the three remaining rows the orbit graphs are G(6,4), G(9,5) and G(9,5), respectively, with no loops. The first independence complex consists of the three opposite pairs, hence three disjoint edges. Multiplication by 2 identifies the second orbit graph with the square of the nine-cycle. Its independence complex has nine vertices, eighteen edges and the three triangles \{i,i+3,i+6\}, 0\leq i<3. These triangles have disjoint edges and can each be collapsed along one edge. The resulting graph is connected, with nine vertices and fifteen edges, so it is a wedge of seven circles. The free homology ranks and degrees in Theorem A.1 are (h,q)=(7,3),(11,3),(7,5). Formula (3.2) gives g=1,2,0. In the K(18,4) row, L^P=A_3 and \ker N_P=D_3; in the K(18,5) row, L^P=A_2 and \ker N_P=D_2, by their displayed rational factors and torsion-freeness.

For the three-factor identification write L=L_{14}=\mathbb Z[t]/(\Phi_2\Phi_7\Phi_{14}) and U_i=\ker\Phi_i(T). The actual subgroup invariant lattices are L^{\langle T^2\rangle}=U_2 and L^{\langle T^7\rangle}=U_7. The respective norm kernels are \ker(\Phi_7\Phi_{14})(T) and \ker(\Phi_2\Phi_{14})(T). Thus U_2+U_7+U_{14} is contained in both denominator lattices and gives natural dihedral-equivariant surjections from the full factor quotient Q onto the two norm quotients. Their product is surjective because the two target orders are coprime: multiplying a preimage by an integer that is one modulo one target exponent and zero modulo the other isolates that component. The established order |Q|=7\cdot2^6 in (A.1) equals the product of these orders, so the product map is an isomorphism. It identifies the full primary components through the natural quotient maps, not merely by comparing group orders.

Theorem A.1 gives free reduced homology concentrated in degree five, with \operatorname{rank}L=7 and T^9=1. Corollary 3.6 identifies the orbit graph of K as G(9,5) and gives free reduced homology of rank seven in degree one for F. Multiplication by two on its orbit labels identifies this graph with C_9^2, and takes T to the square of the usual translation. Under this labeling, T^3 becomes the sixth power of the usual translation; it generates the same subgroup as its third power. The P fixed space in F consists of the three barycenters of the independent triples \{i,i+3,i+6\}. Theorem 3.2 applied to F gives \mathcal G_P(H_1(F;\mathbb Z))\cong\mathbb Z/3. Lemma 3.3 transfers this norm gluing to L, because 5-1 is even. Finally the already established rational factors of L are t-1 and \Phi_9. For \sigma=T^3, its invariant rational subspace is the t=1 factor, and its norm is 1+T^3+T^6=\Phi_9(T). Torsion-freeness gives L^P=\ker(T-1) and \ker N_P=\ker\Phi_9(T), identifying the stated quotient.

An equivariant scalar map on each displayed cyclic block is a multiple of augmentation; B_v has no such nonzero map since its annihilator is one at t=1. The other annihilators vanish there, and each reflection has negative augmentation, so all the stated maps exist. Primitivity is exactly \gcd(\alpha,\beta_1,\ldots)=1. The entire T-fixed lattice of M_s is generated by N_s, whose augmentation is three. That of W_v is generated by 1+t^v+\cdots+t^{5v}, whose augmentation is six. The block B_v has no T-fixed element; a regular block has fixed generator 1+t+\cdots+t^{n-1}, of augmentation n. These assertions follow by monic division of (t-1)p by the relevant annihilator. Every such fixed generator has reflection -1. Consequently the full dihedral lifting images are

respectively, where the notation denotes the ideal generated by the entries. Primitivity reduces their positive generators to the asserted gcd formulas; Bezout gives an element attaining each generator. With a regular block present, choose \alpha=e and one \beta_j=1 to realize every stated divisor value. At the two exceptional parameters no regular block is present and \alpha=\pm1.

Every equivariant map on a cyclic summand is a multiple of augmentation; on \mathbb Z_- it is a scalar. All these maps have the required reflection action, and primitivity is exactly \gcd(\alpha,\beta_1,\ldots)=1. For odd n, the entire T-fixed lattice has generators u and the regular norms, so its image is the ideal (\alpha,n\beta_1,\ldots). For even n, monic division gives H_m^T=\mathbb Z(1+t^m), with augmentation two; together with the regular norms it gives image (2\alpha,n\beta_1,\ldots). Every displayed fixed generator has reflection -1. Consequently this image is the full equivariant scalar lifting ideal. Primitivity gives the two gcd formulas, and Bezout supplies lifts attaining them. Choosing one \beta_j=1 and \alpha equal to a chosen divisor realizes all listed minima when a regular summand is present; otherwise \alpha=\pm1.

A face fixed setwise by g is a union of vertex orbits, and the fixed independent faces correspond exactly to independent sets of Q_g. Suppose such a face has m vertices in a orbits. The orientation sign of g on that simplex is (-1)^{m-a}. Its contribution to the alternating simplicial trace is (-1)^{m-1}(-1)^{m-a}=(-1)^{a-1}. The empty face contributes -1 in the augmented complex. Summing gives the reduced Euler characteristic of the quotient independence complex. The alternating chain-trace identity gives the same sum on homology. This identity follows by trace additivity in the cycle and boundary exact sequences: the boundary terms cancel between adjacent degrees.

No nonempty independent face is fixed setwise by one-step translation: the only nonempty invariant vertex subset is the full vertex set, which contains an edge. The reduced alternating chain trace is -1. Hence the trace \epsilon in degree q is (-1)^{q+1}. In rank one this directly gives multiplication by \epsilon.

In rank two let T be the finite-order integral action matrix. Its determinant cannot be -1: a real finite-order matrix of dimension two with negative determinant has eigenvalues 1,-1 and trace zero. Thus \det T=1, and the characteristic polynomial is t^2-\epsilon t+1. It is \Phi_3 or \Phi_6, and is irreducible over \mathbb Q. Cayley–Hamilton makes the lattice a torsion-free rank-one module over the corresponding quadratic ring. Torsion-freeness follows by embedding in the rationalization, where every nonzero ring element acts invertibly. Both rings are the Eisenstein ring \mathbb Z[\zeta_3]=\mathbb Z[\zeta_6], up to the chosen generator. This ring is Euclidean: rounding the two coefficients in the basis 1,\zeta_3 leaves a remainder a+b\zeta_3 with |a|,|b|\le1/2 and norm a^2-ab+b^2\le3/4<1. Consequently a finitely generated torsion-free rank-one module over it is free. The displayed polynomials give the exact orders, which must divide n.

Orient matched incidences upward and all other incidences downward. A directed cycle would have equally many upward and downward edges. Two upward edges cannot be consecutive, since a face belongs to at most one matching pair; hence the edges of such a cycle alternate. Choose the smallest pivot v_i used by an upward edge of the cycle and start just after that edge. Eventually a downward edge removes v_i, say \tau\longrightarrow\tau\setminus\{v_i\}. The upward edge immediately before it has a different pivot, because a matched edge is not oriented downward. That pivot is therefore later than v_i. The upward edge immediately after it likewise has a different pivot, since otherwise it would reverse the same downward edge. Thus both \tau and \tau\setminus\{v_i\} were available at stage i. They would have been paired then, a contradiction.

Discrete Morse theory [Forman] gives a CW model with precisely the critical cells. Under the stated profile it has one vertex and m d-cells. For d=1 this is a wedge of circles. For d>1 every attaching map lands in the unique vertex, so the model is a wedge of d-spheres.

Write a cyclic word as P\mid R\mid Y\mid L, where R,L have length w. The inequalities p\geq w and n\geq p+3w imply |Y|\geq w. There is no edge from R to L: both arcs between these boundary blocks have more than w edges. Every edge meeting P is represented in the displayed local graph, and P has no edge to Y.

Fix Y. Independence inside Y and compatibility of Y with R,L give a Boolean condition A_Y(R,L) independent of P. On a boundary choice with A_Y=0 there are no global faces. On a choice with A_Y=1, the initial global indicator as a function of P is precisely the local indicator with z=1 when Y is nonempty and z=0 otherwise. Since the first p toggles alter only P, the same description holds at every stage of (?). Duplicating P leaves Y,R,L and their compatibility condition unchanged. Equation (6.2) therefore gives a bijection between the global survivors after p old pivots and after 2p new pivots.

After the first pivot, every surviving face containing zero is the singleton \{0\}: a larger such face has already been paired with its nonempty deletion of zero. The singleton has no surviving partner at any subsequent stage. All other survivors omit zero, and the ordinary duplication map commutes with every later toggle:

The inverse on the image deletes the second copy of P and shifts the remaining positions back. Thus two survivors are toggle partners at j if and only if their images are partners at j+p. Induction through the remaining pivots proves the asserted critical-set equality. Each positive critical face acquires exactly t vertices. Lemma 6.1 then gives both wedge descriptions and the suspension equivalence.

Use J=\{2,3\}, w=3 and p=8 in Theorem 6.2. For clarity we specify the whole finite calculation. For any of the local graphs H in that theorem, initialize the indicator on every binary word by

Apply (?) at the processed positions in order. The resulting fourteen-coordinate survivor table is

Each row represents all combinations in its last two columns, for a total of seventeen words. For z=0 add 10000000\mid000\mid000; for z=1 add no word. Applying the same two formulas to the twenty-two-coordinate graph, with pivots 0,\ldots,15, gives precisely this table with P replaced by PP, and the unchanged singleton when z=0. These are equalities of complete survivor sets, not just of their cardinalities. They are the finite identities (6.2).

Here is also the complete base specification. For 6\leq n\leq24, use (Proof) on the cyclic graph with z=0 and apply all n pivots. Besides \{0\}, its survivors are the words in (?) below, with n=8q+r. Put

Using concatenation and word powers, the positive critical words are

The empty power is allowed for n=6,7. Equivalently, their multiplicity–dimension pairs in the full base window are

Equations (Proof) and (?) specify each entry using only Boolean multiplication and subtraction on a fixed finite domain. The complete local survivor sets and the base-word comparisons are the finite part of the proof; they are specified by the displayed graph, initial indicator and toggle recurrence.

The generic threshold is p+3w=17. The eight base cases 17,\ldots,24 meet every residue class modulo eight, and each positive critical word begins with a block having two selected positions. Theorem 6.2 therefore duplicates its initial block, preserves the multiplicity and raises the dimension by two. Induction proves (?) for every n\geq25; the finite base window covers all smaller n\geq6. Lemma 6.1 gives the displayed homotopy types. Comparing the formulas at n and n+8 gives the claimed suspension equivalence, including 6\leq n\leq16 below the generic threshold.

The image of \{\pm1,\pm4\} is \{\pm3,\pm12\}=\{\pm2,\pm3\}. Apply Corollary 6.3 at 14=8+6.

We first include the empty face in the matching. At pivot zero this pairs \varnothing with \{0\}. Removing that one pair instead leaves \{0\} critical, and does not change any other subsequent pair: every two-element partner of \{0\} was already paired at pivot zero, or is not independent. Thus the nonempty-face matching has precisely the same other critical faces as the augmented matching, with the extra singleton \{0\}. This explains the term yx^{3w}/(1-x) in the large-n series.

Assume n\ge3w. At the stage immediately before pivot i, fix the already read prefix b_0\cdots b_{i-1}. For 0\le i\le n-2w, the unprocessed suffix has at least 2w vertices. Let q be its first w bits, and let v be its last w bits. These blocks are disjoint. The survivor indicator on the suffix is

where I_i is the independence indicator of the subgraph induced by the unprocessed suffix and f_i:\{0,1\}^{2w}\to\{0,1\} is a Boolean function. Initially f_0=1.

For i<n-2w, write the actual bit at pivot i as b, and the next w bits as q=(q_0,\ldots,q_{w-1}). For a possible bit c, its compatibility with future vertices is

The second product represents the cyclic edges to the terminal boundary; it is empty for i\ge w. The two evaluations before the toggle use

For w=1, the middle list is empty. The survivor recurrence gives

Indeed, the induced-suffix independence indicator after removing vertex i is common to both toggle variants. If it is zero, both vanish; if it is one, the two variants are exactly h_b and h_{1-b}. This proves (?) and (?) by induction. Edges to earlier vertices have already entered f_i, so none is omitted.

The possible Boolean functions constitute a finite state set of cardinality D=2^{2^{2w}}. The first w pivots use the explicit, time-dependent transitions above. All later pivots before n-2w use the same two transitions, indexed by b=0,1, since the wraparound product is then empty. Let M_J(y) give weight y^b to the transition reading bit b, adding the weights if the two bits yield the same state. Let U_J(y) be the row vector obtained from f_0=1 after the first w transitions, with the same weights.

At i=n-2w, exactly 2w bits remain. Their induced graph has only the ordinary line-distance edges in J: the shortest possible wraparound distance is n-(2w-1)>w. For each state f, initialize the finite table by I_{n-2w}(s)f(s) and perform the final 2w toggles. Let V_J(y)_f be the sum of y^{|s|} over its surviving terminal words. This terminal rule depends on J and f but not on n. Thus the augmented critical-face polynomial is exactly

Summing this expression over n\ge3w proves the displayed rational formula. The omitted lengths are a finite polynomial E_J.

The same construction is a finite automaton: read the first w bits with the initial transitions, read any number of bits with the common transitions, and accept a terminal block of length 2w according to the finite rule. A finite automaton can guess where this terminal block begins; its prescribed length makes the split unique for every accepted word. The extra singleton language and the finitely many short-length words are regular as well. Finally, the characteristic polynomial of \operatorname{diag}(M_J(y),1) gives the stated eventual linear recurrence by Cayley–Hamilton.

Represent a point by nonnegative vertex coordinates (\lambda_v) with sum one and independent support. The point is P_m^{(n)}-fixed precisely when its coordinates are constant on each orbit. Set \mu_a=\sum_{v\equiv a\pmod g}\lambda_v. The inverse map assigns \lambda_v=\mu_a/m on that orbit. An orbit contains an internal edge precisely when g divides a jump; two distinct orbits have an edge between them precisely when their residues differ by a reduced jump. Consequently these two linear maps restrict to mutually inverse homeomorphisms of the stated realizations. A zero reduced jump produces a loop at every residue, so there is no nonempty independent support in that case.

Proposition 8.1 applies to J=\{2,3\}. Its quotient has a loop exactly for g=1,2,3. For g=4 it is K_4, and for g=5 it is a five-cycle whose independence complex is another five-cycle. For g\ge6 it is G_g(\{2,3\}). The action is free if and only if every nontrivial subgroup has empty fixed space, equivalently every proper divisor of n is at most three. For composite n\ge6, let p be its least prime divisor. This condition gives n/p\le3 and p\le n/p, whence p\le3 and n\le9. The composite values satisfying the condition are exactly 6 and 9. Prime n has only the proper divisor one.

The two fixed spaces are Y_{8q} and Y_{8q+b} by Theorem 8.2. The first has total reduced Betti number five. The second has total reduced Betti number one, because b is odd. Taking fixed spaces of an equivariant homotopy equivalence would preserve these numbers, and a representation suspension cannot change them. This contradiction holds for every V, even if it depends on q. Iterating the ordinary eight-periodic equivalence gives the last assertion.

The cyclic gap enumeration gives the thirteen-vertex face vector, and the collapse orbit followed by the displayed ordered matching leaves one vertex and twelve triangles. The two fourteen-vertex enumerations and matchings leave one vertex and one cell in dimension three or two, respectively. Lemma 6.1 gives the three wedge types. Their integral reduced homology is free, with the displayed ranks and degrees.

Translation reduces the assertion to i=0. A fourth vertex compatible with B_0=\{0,6,9\} must avoid

The only remaining residue is 3. Thus B_0 has the unique coface Q_0, and translation proves the claim for every i.

For G_n(\{2,3\}), the proof of Corollary 6.3 identifies the natural-order critical profile as one vertex together with exactly the cells that generate the displayed reduced homology. Thus equality holds in every weak Morse inequality.

For G(13,2), adjoining the thirteen elementary-collapse pairs to the ordered matching on the remaining complex gives an acyclic matching on the original face poset; this is the standard concatenation of a collapse sequence with a Morse matching on the terminal subcomplex. Its critical profile is (1,0,12,0). The two fourteen-vertex profiles are (1,0,0,1,0) and (1,0,1,0,0,0,0).

By Theorem 9.1, these benchmark profiles agree dimension by dimension with the corresponding Betti numbers over every field. The weak Morse inequalities require the number of critical i-cells in any acyclic matching to be at least the ith Betti number. Equality here proves both perfectness and the asserted global optimality.

For 1\ne g\in\Gamma, no oriented simplex is carried to itself as a basis element, even up to sign, so g has trace zero on every simplicial chain group. Since K is connected, \widetilde H_0(K;\mathbb Q)=0. Applying the Hopf trace identity to the ordinary simplicial chain complex gives

Thus the homology character has value \beta at the identity and \varepsilon=(-1)^{d+1} elsewhere.

For odd d this character is \mathbf{1}+((\beta-1)/q)\operatorname{reg}_{\Gamma}; for even d it is -\mathbf{1}+((\beta+1)/q)\operatorname{reg}_{\Gamma}. Taking the inner product with the trivial character shows that the displayed coefficients are integers and, in the even case, that (\beta+1)/q-1\ge0. Maschke semisimplicity and equality of characters identify the rational representations. Finally, \operatorname{reg}_{\Gamma}-\mathbf{1} is the augmentation representation.

Fix a\in\mathbb Z/n\mathbb Z and put g=\gcd(n,a), with \gcd(n,0)=n. Translation by a has g vertex orbits, each of length \ell=n/g. Let B_g(t) count sets of these orbits whose union is an independent face, by number of chosen orbits. If g=1,2,3, one of the jumps 2,3 lies within every orbit, so B_g(t)=1. For g\geq4, a union of orbits is independent precisely when the corresponding residues form an independent set in the loopless quotient with jumps 2 and 3. Thus

because the quotient graphs are K_4 and C_5, while for g\geq6 Corollary 6.3 gives

where m_g and d_g are its sphere multiplicity and dimension.

It remains to relate this orbit count to the character. An invariant face formed from t orbits has t\ell vertices. Translation restricts to one \ell-cycle on each chosen orbit, so its orientation sign is (-1)^{t(\ell-1)}. Including the alternating chain sign, this face contributes

to the ordinary simplicial Lefschetz number. Consequently that number is 1-B_g(-1). Since X_n is connected and has reduced homology only in degree d_n, the Hopf trace identity yields

Substituting the four residue cases from Corollary 6.3 into (?) gives the following characters. For 8\mid n, the value is 5 when 8\mid a, -3 when a\equiv4\pmod8, and 1 otherwise. This is the character of \mathbb Q\oplus\mathbb Q(\zeta_8) inflated from C_8. For n\equiv4\pmod8, the value is 3 when 4\mid a and -1 otherwise, which is the character of \operatorname{Aug}_{\mathbb Q}(C_4) inflated from C_4. In every other residue class the homology has rank one and the generator has trace 1, so the representation is trivial. Rational representations of a finite cyclic group are determined by their characters, completing the proof.

For the thirteen-vertex complex, every nonidentity power of T is transitive because 13 is prime. Since the full vertex set is not independent, Proposition 10.2 applies with q=13 and even d=2. It forces \beta_2\ge12 and \beta_2\equiv-1\pmod {13}; the value 12 from Theorem 9.1 is therefore the smallest possible, and the rational module is the augmentation representation.

It remains to identify the integral lattice. Orient simplices by increasing vertex order and set

Direct boundary calculation gives \partial z=0 and \sum_{j=0}^{12}T^jz=0 at chain level. Order the 65 triangles lexicographically and form the 65\times25 matrix whose first thirteen columns are the tetrahedron boundaries and whose remaining columns are z,T z,\ldots,T^{11}z. On the zero-based rows

the determinant is -1. Hence these columns generate a primitive rank-25 sublattice of the two-cycle lattice. Theorem 9.1 gives \operatorname{rank}\widetilde H_2=12 and \widetilde H_3=0, so the cycle lattice itself has rank 13+12=25. A primitive sublattice of \mathbb Z^{65} cannot lie properly in a rank-equal lattice, so the two lattices are equal. Consequently the classes [z],[T z],\ldots,[T^{11}z] form an integral homology basis, with

Thus the module is \mathbb Z[C_{13}]/(1+T+\cdots+T^{12}), which multiplication by T-1 identifies with \operatorname{Aug}_{\mathbb Z}(C_{13}).

For either fourteen-vertex complex, the generator T is transitive, so it fixes no nonempty simplex setwise. Since each complex is connected, the Hopf trace calculation in the proof of Proposition 10.2 gives trace +1 in degree three and -1 in degree two. Each relevant integral homology group has rank one, so its automorphism is determined by that sign.

A fixed point would lie in the relative interior of a simplex fixed setwise, which does not exist. The face-orbit counts are those of Table ?. Theorem 9.1 makes the covering complex simply connected, so the quotient has fundamental group C_{13}. Over \mathbb Q, the quotient chain complex is obtained by coinvariants. These are exact for the semisimple algebra \mathbb Q[C_{13}], and the augmentation representation has zero coinvariants. The asserted rational homology follows.

These are finite exhaustive matching and signed-boundary computations, specified as follows. The algorithms provide a mathematical reconstruction; Section ? distinguishes this specification from the reported execution records and states their availability. Set F_{-1}=\{\varnothing\} and

where N(v) is the neighborhood of v in G(n,s). This enumerates independent sets uniquely by their largest vertex. Sort F_{n-1}\setminus\{\varnothing\}, construct M, and count its critical complement. Topological sorting visits every face. The f-vectors for K(18,4) and K(18,5), recording numbers of nonempty simplices by dimension, are respectively

For the degree-five differential of K(25,4) in (11.1), sum signed alternating paths from each critical five-cell to the critical four-cell. A downward incidence \beta\to\alpha has weight [\beta:\alpha], and a matched upward incidence has weight -[\beta:\alpha]^{-1}. The sums are -1 and 1. All incidences used are \pm1. An independent recursive enumeration verifies the full face partition and checks every directed Hasse incidence against a topological ordering. The seven additional pure-dimensional cases use the same full-domain check; the five mixed-dimensional cases are treated in Section ?. Execution tests are summarized in Section ?. The finite algorithm, critical lists, and signed path rule determine the data without a stored certificate.

Successive applications of Hurewicz show that \pi_i(X)=0 for i<q and that the Hurewicz map \pi_q(X)\to H_q(X;\mathbb Z) is an isomorphism. Choose sphere maps representing an integral homology basis and combine them into a map \bigvee^r S^q\to X. This induces an isomorphism on all integral homology groups. The homological Whitehead theorem for simply connected CW complexes makes it a homotopy equivalence; these are standard consequences of Hurewicz and Whitehead [Hatcher, Theorems 4.32 and 4.37, and Corollary 4.33].

The induced map \overline P:H_q(C_*)\to\mathbb Z^r has section e_j\mapsto[z_j]. Its kernel is a subgroup of a free group and has rank zero, so is zero. Taking coordinates proves the matrix assertion.

Apply (?)–(?) to the completely enumerated face set and matching. Apply (?) to every resulting chain, and apply P to each chain and each (q+1)-simplex boundary. The former images are the standard basis vectors and all boundaries map to zero, as the chain-map description of P predicts. Cycle support sizes are given in the appendix. An independent check enumerates faces from the graph edge differences and tests the full integer coefficients of P and the lifts, without importing the producer’s recurrence or path construction. Thus the identities hold in the simplicial chain complex itself.

Construct the matrices by signed geometric transport on the verified cycles. The appendix gives the three odd-degree matrices in the fixed critical-face order, and hence specifies their B by their columns. For K(14,2), Proposition 14.4 supplies an explicit cycle, and the same projection and lift recurrences determine B. Successive integer multiplication gives the orbit columns and the indicated annihilating relation. Exact elimination gives the displayed determinants. An independent calculation checks the orbit columns, determinants by rational elimination, and the polynomial identity. Together with Proposition 12.2, these are identities for the geometric homology action.

If ag is divisible by ab in \mathbb Z[t], cancellation gives g=bh. This proves A=bL; the other kernel and the injection in (14.1) follow the same way. The image of the injection is aL, the kernel of reduction modulo a. Over \mathbb Q, multiplication by a is invertible on the b-factor and multiplication by b on the a-factor. Thus their kernels have zero intersection over \mathbb Q. Since L is free over \mathbb Z, A\cap D=0 integrally. The quotient by their sum is precisely the quotient by the ideal (a,b), proving (14.1).

A module section of reduction modulo a has image annihilated by a, hence contained in A. If a section exists, A+D=L, so (a,b)=\mathbb Z[t]. Conversely, this ideal equality gives the integral Chinese remainder decomposition and a section.

The identity f_ip=fh=f_ig_ih implies p=g_ih, proving the kernel formula and the isomorphism by multiplication by g_i. Over \mathbb Q the kernels occupy distinct factors of the Chinese remainder decomposition, so their sum is direct over \mathbb Q, hence over \mathbb Z. The quotient by their sum is the quotient by the displayed ideal. This is an elementary factor-ideal consequence of classical integral Chinese remainder theory [MS].

For a monic h\in\mathbb Z[t] of degree d, let \lambda_h extract the coefficient of t^{d-1} from the representative of degree below d. The integer pairing \beta_h(a,b)=\lambda_h(ab\bmod h) is perfect: its monomial Gram matrix has zero entries for i+j<d-1 and ones for i+j=d-1, so reversing the columns gives a triangular matrix with diagonal ones. Put B=\bigoplus_i\mathbb Z[t]/(f_i) and E:B\to L, (b_i)\mapsto\sum_i(f/f_i)b_i. By Lemma 14.2, its image is the factor-kernel sum. Division by the monic f_i gives

Thus, for the unimodular Gram matrices J_f,J_B, J_fE=P^{\mathsf T}J_B. The embedding and the transpose of the reduction are integrally equivalent, and transposition preserves Smith form. This is an elementary duality within classical integral CRT theory [MS]; the conclusion is about abelian groups, without identifying their geometric actions.

The six vertices in (14.4) induce exactly the three edges \{0,1\},\{4,5\},\{8,9\}. Their independence complex is a join of three zero-spheres, whose fundamental cycle is the displayed eight-term chain. This is a classical cross-cycle construction [SpecialCycles, Section 2]. The chain z equals the first critical-face lift in Appendix ?. The complete matching has one vertex and thirteen two-cells, so gives the ordinary wedge \bigvee^{13}S^2; that type also follows from the prior cycle-power recurrence [CyclePowers]. The face counts are (14,63,98,35). The common projection and lift recurrences verify all thirteen cycles, the entire projection on the 98 triangles, and its vanishing on all 35 tetrahedron boundaries. Thus Lemma 12.1 applies. Signed translation gives a Krylov basis with determinant -1 and the relation \sum_{j=0}^{13}t^jw=0, as in Proposition 13.1. Signed vertex negation gives \rho_*w=t^5w; semilinearity extends this over the orbit basis. Both geometric signed maps are checked on every simplex and on every homology column.

For the gluing, take (f_1,f_2,f_3)=(\Phi_2,\Phi_7,\Phi_{14}) in Lemma 14.2. Their product is 1+t+\cdots+t^{13}. In its monomial basis form the 13 by 13 matrix E whose columns are the coefficients of t^jg_i, in order i=1,2,3 and 0\le j<\deg f_i. These are the embeddings of the three kernel bases. Exact integer elimination gives |\det E|=448, 14E^{-1} integral, and

The determinant and exponent bound restrict the quotient to elementary 2- and 7-primary parts; the two ranks give their multiplicities six and one. This proves (A.1). This quotient is obtained from the kernel-embedding matrix E, independently of a cokernel calculation for the Chinese remainder reduction map L_{14}\to\bigoplus_{d=2,7,14}\mathbb Z[t]/(\Phi_d).

The change w'=t^3w gives \rho_*w'=t^{5-6}w'=t^{-1}w'. Propositions 6.1 and 7.2 of the public nerve preprint [Nerve], with (n,k,\ell)=(28,14,1), give an integral orbit basis \alpha_0,\ldots,\alpha_{12}, relation \sum_{i=0}^{13}\alpha_i=0, and actions \tau\alpha_i=\alpha_{i+1}, \rho\alpha_i=\alpha_{-i-1}. The map \alpha_i\mapsto t^iw' is therefore an integral isomorphism intertwining both generators. The nerve homology rotation has order fourteen, so its D_{28} action factors through D_{14} with those generators. This is an already known abstract dihedral representation, here realized by a cycle-square cross-cycle.

Both defining polynomials have constant term -1, so t is a unit in L_3 and L_2. They satisfy t^r f(t^{-1})=-f(t), for r=7,11 respectively. Thus substitution p(t)\mapsto p(t^{-1}) is a well-defined involution on each quotient.

Apply \rho to every vertex of an oriented simplex and multiply by the sign required to put the image in increasing order, as in (?). Lemma 12.1 gives the matrix with columns P(\rho z_j). For the generators fixed above, the resulting exact identities are

These are obtained using the projection and cycle-lift recurrences (?)–(?) on the critical-face orders in Appendix ?. An independent check computes the signed simplicial reflection, verifies that it commutes with every boundary map, and compares all columns P(\rho z_j). In the respective [z_j] coordinates, the shifted generators are

Multiplication by t^7 is an integral automorphism, so these generators still have unimodular orbit bases. The relation \rho_*t=t^{-1}\rho_* extends (?) over the entire cyclic basis. For w'=t^7w, it gives \rho_*w'=t^{5-14}w'=w' in the first case, \rho_*w'=t^{14-14}w'=w' in the second, and \rho_*w'=-t^{5-14}w'=-w' in the third, using t^9=1 on L_3. This proves the displayed semilinear formulas. The geometric relations also give \rho_*^2=1 and \rho_*t\rho_*=t^{-1}; no invariance of the Morse matching is required.

To compute reflection traces, work over \mathbb Q. On L_3\otimes\mathbb Q=\mathbb Q\oplus\mathbb Q(\zeta_9), inversion has trace one on the first factor and zero on the cyclotomic factor, whose primitive-root eigenspaces are paired by inversion. On L_2\otimes\mathbb Q=\mathbb Q[C_9]\oplus\mathbb Q(\zeta_6), inversion fixes one vector of the usual C_9 basis and pairs the other eight; the cyclotomic contribution again has trace zero. The same calculation applies to inversion composed with any rotation: on the C_9 basis the index map has the form j\mapsto a-j, with exactly one fixed index, and on the nontrivial cyclotomic eigenspaces it exchanges pairs. The negative sign reverses every reflection trace for K(27,5). As an independent check, for every reflection g the reduced signed simplicial Lefschetz sum satisfies \widetilde L(g)=(-1)^q\operatorname{tr}(g_*|\widetilde H_q), for all 18,18,27 reflections respectively. The sign is negative in each of the three odd homological degrees. Since t has order nine on the rank-seven lattices, their homology representations factor through the quotient sending D_{18} or D_{27} to D_9 with the specified generators. Their traces on \rho are 1 and -1, proving rational, hence integral, nonisomorphism.

We first transport the graph and the generators, then use the classical integral basis to compute reflection. For \ell=0, the graph is complete on five vertices and its independence complex consists of five points. Its reduced degree-zero homology is the augmentation ideal of the vertex permutation module. Writing v=e_0-e_1, the four successive translates v,tv,t^2v,t^3v are the edge differences of a spanning path, hence an integral basis, with \Phi_5(t)v=0. Negation gives \rho v=-t^{-1}v; for w'=t^2v, one has \rho w'=-w'. This proves the action formula at n=5; the coinvariant argument below applies unchanged. For the rest of the geometric construction suppose \ell\ge1. Let u=(n+1)/2. Multiplication by u is a permutation of \mathbb Z/n\mathbb Z, with inverse multiplication by 2. It sends the forbidden differences \pm1,\pm3 to the four differences of cyclic distance (n-1)/2 or (n-3)/2. The remaining differences have cyclic distance at most (n-5)/2=5\ell. Hence

The rotation \tau becomes \sigma^u, where \sigma(x)=x+1 in the clique coordinates, and \rho remains vertex negation. In particular u\equiv3\pmod5. The classical classification [Adamaszek, Nerve], as recalled in Theorem 4.3 of [RipsCircle], gives the displayed ordinary homotopy type at winding fraction \ell/(2\ell+1).

We recall the integral basis in a form that also specifies its chain representatives. In clique coordinates, let

Vertices in these displayed lists are reduced modulo n; the displayed order specifies their orientation. Each F_a is a maximal simplex, so the cochain \beta_a=F_a^\vee is a cocycle. For a<b<a+5, set A_j=a+5j, B_j=b+5(j+\ell) modulo n, and define

The antipodal pairs (A_j,B_j) describe the cross-polytope from Lemma 8.2 of [RipsCircle]: the A_j are its even-indexed vertices, and B_j is the antipode of A_j. The all-B list is a cyclic shift of the displayed F_b; a cyclic shift of 2\ell+1 vertices preserves orientation, so its coefficient is -1. To express these displayed orientations in the global increasing-order chain basis, multiply each term by its sorting sign. Every term is a simplex, and cancellation of each omitted pair gives \partial z_{a,b}=0. Its two terms supported in a single residue class modulo 5 are F_a-F_b. Write \alpha_{a,b}=[z_{a,b}]. Proposition 8.4 of [RipsCircle] gives the basis \alpha_{1,5},\ldots,\alpha_{4,5}, dual to \beta_1,\ldots,\beta_4. Equivalently, their pairing is the identity, and the known free rank four makes Lemma 12.1 applicable. This is an integral basis.

Identify these classes with the augmentation ideal

by \alpha_{a,b}\mapsto e_{a\bmod5}-e_{b\bmod5}. The relations in the proof of Proposition 8.4 of [RipsCircle] justify this identification for all permitted a,b. Rotation sends e_i-e_j to e_{i+1}-e_{j+1} in clique coordinates. Negation reverses the displayed ordering of F_a, up to a cyclic shift. Reversal of 2\ell+1 vertices has sign (-1)^{\ell(2\ell+1)}=(-1)^\ell, whereas the cyclic shift has sign +1. Mixed-residue terms remain mixed under both symmetries, so the maximal-face cochains give

Thus this formula computes the geometric reflection, including its orientation sign.

Pull \alpha_{0,1} back to the original graph and call the resulting class w. Since \tau shifts the residue index by u\equiv3\pmod5, the four classes w,tw,t^2w,t^3w correspond to the edge differences on \{0,1\},\{3,4\},\{1,2\},\{4,0\}. These edges form a spanning tree on five vertices; their differences form an integral basis of I_5. Also \Phi_5(t)w=0, and \rho_*w=(-1)^{\ell+1}t^3w. For w'=t^4w, the relation \rho_*t=t^{-1}\rho_* and t^5=1 give \rho_*w'=(-1)^{\ell+1}w'. This proves the cyclic-module identification and (16.1). The factorization through D_5 is at the homology level.

Let L_5^+ and L_5^- denote the two choices of sign in (16.1). Their coinvariant group is

Reflection induces +1 and -1, respectively, on this quotient. No automorphism of \mathbb Z/5\mathbb Z intertwines these two actions, proving integral nonisomorphism. Multiplication by a=t-t^{-1} commutes with t and anticommutes with polynomial inversion. It therefore intertwines L_5^+\otimes\mathbb Q and L_5^-\otimes\mathbb Q. Its determinant is

It is consequently invertible over \mathbb Q, but not over \mathbb Z.

We first identify a natural integral suspension isomorphism for T^2 and R. The known nerve basis then identifies the integer centralizer of T^2; the reduced Lefschetz number chooses the correct square root for T.

The ordered cover and its natural actions. Set A=\{a_i=2i\} and B=\{b_j=2j+1\}, with indices modulo m. Both parts are independent, and the neighbors of a_i are b_{i-2},b_{i-1},b_i,b_{i+1}. Let U=\Delta(B) be the full simplex on B and let

where \operatorname{st} denotes the closed simplicial star. Every face containing a vertex of A belongs to V, so X_m=U\cup V. Both subcomplexes are contractible by the star-cluster lemma [Barmak, Lemma 3.2 of the public preprint]. Indeed, any nonempty intersection of the stars is the closed star of the corresponding face of A, hence a cone, and the cover nerve is a simplex. The maximal faces of U\cap V are the cyclic blocks

Each has m-4 vertices; with b_j labeled j, this is exactly N(m,m-5), whose maximal faces are the cyclic intervals [j,j+m-5]. For m=5 it consists of five isolated vertices. The classical bipartite suspension construction is due to Jonsson; Barmak recalls it in Theorem 3.7 and gives the two-contractible-subcomplex argument in Lemma 3.3 of the public preprint [Barmak]. It yields the ordinary equivalence X_m\simeq\Sigma N(m,m-5). For the fixed ordered cover (U,V), reduced integral Mayer–Vietoris gives an isomorphism

An automorphism preserving both parts preserves each cover member, so naturality makes \delta equivariant for such maps. On the B labels, T^2 is s:j\mapsto j+1 and R is s^{-1}e:j\mapsto-j-1, where e:j\mapsto-j. The one-step translation exchanges the parts, so we recover it separately.

The rank-four lattice and the missing generator. Suppose m=5q. Then (m-5)/m=(q-1)/q. Theorem 3.6 and Propositions 6.1 and 7.2 of the public nerve preprint [Nerve] give a wedge of four (2q-2)-spheres and integral classes \alpha_i, indexed modulo 5, with

Via \delta, identify the homology of X_m with O=\mathbb Z[z]/(1+z+z^2+z^3+z^4), taking \alpha_0 to 1 and T^2 to multiplication by z. This is a free rank-one O-module over the integers. Since T commutes with T^2, it is multiplication by some \lambda\in O satisfying \lambda^2=z. The ring O is an integral domain, z^5=1, and (z^3)^2=z; hence (\lambda-z^3)(\lambda+z^3)=0 forces \lambda=\pm z^3. The actual one-step vertex translation fixes no nonempty independent face setwise: its only nonempty invariant vertex subset is the full vertex set. Thus it has trace zero on every nonnegative simplicial chain group and trace one on the augmentation group, so its reduced Lefschetz number is -1. The homology degree 2q-1 is odd, giving \operatorname{tr}(T)=1. Multiplication by z^3 has cyclotomic trace -1, and therefore

Consequently \mathbb Z[T]=O and the minimal polynomial is \Phi_{10}. For completeness, \Phi_{10}(-z^3)=0 follows by permuting the five powers of z in their norm relation; the resulting surjection \mathbb Z[t]/(\Phi_{10})\to O is an isomorphism between free abelian groups of rank four.

Reflection normalization and the integral distinction. Naturality and the nerve formulas give

Choose w=\delta^{-1}(z^4\alpha_0). Since z^5=1, one has Rw=(-1)^q w. Inversion takes T=-z^3 to T^{-1}, proving the stated formula in t and the integral orbit basis w,Tw,T^2w,T^3w. The order of T is 10, dividing 2m=10q, which gives the D_{10} quotient on homology. Moreover,

Reflection acts there by (-1)^q, distinguishing the two integral types. Over \mathbb Q, multiplication by z-z^{-1} is invertible, commutes with T and anticommutes with inversion, so it intertwines the two signs.

The remaining residues. For 1\le a\le4, the ratio (m-5)/m lies strictly between (q-1)/q and q/(q+1). Theorem 3.6 and Proposition 7.3 of the public nerve preprint [Nerve] give N(m,m-5)\simeq S^{2q-1}, with s=1 and e=(-1)^q on its integral homology. The suspension and naturality give X_m\simeq S^{2q} and R=(-1)^q. On this rank-one group T=\pm1; the same reduced Lefschetz number now gives \operatorname{tr}(T)=-1, so T=-1.

Write I_H(x)=\sum_{\sigma\in\operatorname{Ind}(H)}x^{|\sigma|} for a graph H. The initial values follow from I_{J_0}=1, I_{J_1}=1+x, and J_2=K_{1,1}, J_3=K_{1,2}, J_4=K_{2,2}, using I_{K_{a,b}}(x)=(1+x)^a+(1+x)^b-1.

For L\ge5, independent sets avoiding both 1 and 3 cancel in the signed sum by toggling 0. Every remaining set avoids 0 and either contains 1, or avoids 1 and contains 3. Removing the specified vertex in the first class leaves the induced graph E=J_L[\{3\}\cup\{5,\ldots,L-1\}]; in the second class it leaves D=J_L[\{5\}\cup\{7,\ldots,L-1\}]. All listed vertices outside \{0,\ldots,L-1\} are omitted. Thus F_L=-I_E(-1)-I_D(-1). In E, the only possible neighbor of 3 is 6. Deletion of 3 leaves J_{L-5}, up to translation, and deletion of its closed neighborhood leaves D. The elementary deletion identity

therefore gives I_E(-1)=F_{L-5}-I_D(-1) and proves the recurrence, including L=5,6 when 6 is absent. Since F_5=-F_0=-1, iteration gives the final formula.

The unit relabeling u=(n+1)/2 used in Proposition 16.1 gives

The inequalities in the statement are exactly

The classical clique classification [Adamaszek, Corollary 6.7], also stated as Theorem 8.2 of the public preprint [Nerve] and Theorem 4.3 of [RipsCircle], proves the ordinary sphere type. The integral rotation on its rank-one homology is \pm1; its odd order forces \tau_*=1.

Put n=2h+1. The reflection orbits are \{0\} and \{i,-i\} for 1\le i\le h. The last two orbits, i=h,h-1, contain edges and give looped quotient vertices: their internal differences are n-1 and n-3. There are no other looped vertices. After removing these forbidden vertices, label the remaining orbits by 0,\ldots,h-2. For distinct labels i<j, quotient adjacency means that j-i or i+j is congruent to one of \pm1,\pm3 modulo n. In these ranges the only possibilities are j-i\in\{1,3\} or i+j\in\{1,3\}. The latter produces only (0,1), (0,3) and (1,2), already edges of the former kind whenever their labels occur. Thus the usable quotient graph is J_L, with L=h-1=(n-3)/2.

Write n=10\ell+r with r\in\{7,9,11,13\}. Then L=5\ell+(r-3)/2, and Lemma 18.1 gives F_L=(-1)^{\ell+1}. Proposition 5.1 identifies the reduced Lefschetz number of the actual reflection with

The sole homology degree is odd, so \operatorname{tr}(\rho_*)=(-1)^{\ell+1}, proving the claimed integral action.

Use the face order and vertex-stage matching of Section ?, replacing its vertex order by

The respective face counts, including the empty face, are 309 and 453. The complete ordinary matchings have 150 and 223 pairs, a critical vertex 0, and the following critical faces in lexicographic order:

Topological orders verify all 885 and 1392 directed nonempty Hasse incidences, respectively. Forman’s theorem therefore gives the stated ordinary wedges and free homology ranks. The ordinary types also follow from the earlier cycle-power splitting theorem [CyclePowers].

Let z_1,\ldots,z_r be the critical-face lifts determined by (?)–(?) with these matchings. Their support sizes are (8,8,8,10,8,8,8) and (16,16,57,16,16). The projection is defined on all 140 simplices in degree q in either case and kills all 75 or 16 incoming simplex boundaries. Every lift has zero boundary and projects to its specified standard basis vector. Thus Lemma 12.1 identifies the lifts as an integral homology basis. In their displayed order, choose the coordinate vectors

and put w=[\sum_j v_jz_j] in the corresponding homology group. The signed geometric translation and negation give the following integer identities:

In the orbit bases, the first reflection columns are the coefficients of t^6-t^3-t and t^2, respectively. These finite identities are determined by the graph, matching, critical lists, recurrences and vectors just specified. Independent checks enumerate the entire face set from edge differences, verify every directed Hasse incidence, every projection and lift identity, every signed simplicial chain-map identity and every homology action column, and check the orbit determinants and polynomial relations by rational elimination. Equation (?) then proves the module identifications. The rational cyclotomic factors give the exact action orders. The relation \rho\tau\rho=\tau^{-1} gives the reflection formula on every orbit vector. Both defining polynomials are reciprocal up to sign and have unit constant term, so substitution by t^{-1} is well-defined on their integer quotient rings. The geometric identity \rho_*^2=1 gives h(t)h(t^{-1})=1 in each quotient; thus the displayed multiplier is a unit. For the second case, \rho_*(tw)=t^{-1}t^2w=tw.

For the first gluing quotient use Lemma 14.2 with (f_1,f_2,f_3)=(\Phi_1,\Phi_3,\Phi_5). Put A=\Phi_3 and B=\Phi_5. The identity (1+t^3)A-tB=1 implies that

the left ideal contains t-1, and reduction modulo t-1 takes AB to A(1)B(1)=15. Conversely each left generator lies in (t-1,15). Lemma 14.2 therefore gives \mathbb Z/15\mathbb Z directly. As an independent integer check, the columns t^j f/f_i, 0\le j<\deg f_i, give a kernel-embedding matrix E with |\det E|=15, 15E^{-1} integral, and ranks six over \mathbb F_3 and \mathbb F_5. For the second quotient, Lemma 14.1 gives directly

These are factor-kernel quotients inside the integral homology lattices, rather than just rational factor decompositions.

The proposed module map is well-defined by the annihilator identities and sends the standard polynomial bases to B. Its determinant proves that it is an isomorphism over \mathbb Z. The dihedral relation extends each seed reflection to every polynomial in that seed. For the last assertion, t^n-1 is square-free over \mathbb Q, so the rational module is a direct sum of \mathbb Q[t]/(\Phi_d) vector spaces of dimensions m_d. One cyclic generator contributes at most one dimension over each of these fields. Conversely the rational Chinese remainder theorem combines one basis vector from each constituent into one generator, and repeating this \max_d m_d times generates the module.

Use the ordinary nonempty-face matching of Section ? and the projection/lift conventions of Section ?. For n=17,18 the vertex priority is 0,1,\ldots,n-1; for n=19 it is

The complete face counts, including the empty face, are 664,973,1426. The matchings have respectively 322,473,702 pairs, one critical vertex, and 18,25,20 critical three-faces. The critical vertices are 0,0,2 respectively. Directed Hasse acyclicity is checked on every one of the 2176,3384,5244 nonempty incidences. Forman’s theorem gives the displayed ordinary wedges; these ordinary types are also covered by the prior cycle-power classification [CyclePowers].

Order the critical three-faces lexicographically, and let z_1,\ldots,z_r be their lifts from (?)–(?). The critical faces are the unmatched faces of the fully specified matching, so this also fixes the indexing without an external data file. The projection is defined on all 238,378,570 three-faces and kills all 51,126,266 incoming four-face boundaries. Every lift has zero boundary and projects to its standard basis vector; Lemma 12.1 makes their classes an integral basis.

In each case put u=[z_1]. For n=18, put v=[z_1+z_3+z_8]. For n=17,19, put v=[\sum_j b_jz_j] with the following coordinate vectors:

Let B concatenate the n successive translates of u with the first d translates of v. The signed geometric actions satisfy

Here (t^n-1)u=0 and f_v(t)v=0. These are full integer identities in the lifted-cycle coordinates: the checker derives every column of both actions from the signed vertex maps, rather than merely testing abstract group relations. The concatenated orbit columns and their determinants give Lemma 20.1. The reflection columns lie in their respective blocks. Replacing u by t^2u,t^2u,t^8u for n=17,18,19 respectively changes its reflection multiplier to 1,t,1. Replacing v by t^2v for n=18 changes its multiplier to 1. For a seed with Ru=T^k u, the relation RT=T^{-1}R gives R(T^a u)=T^{k-2a}(T^a u). These shifts are integral automorphisms, giving the stated dihedral formulas. The second summand for n=18 is therefore the same L_3 with inversion already identified on K(18,4).

Finally the three rational factorizations have multiplicities

The last part of Lemma 20.1 excludes a single cyclic generator. The displayed integer orbit bases supply two, proving minimality. Complete signed Lefschetz checks independently recover every rotation and reflection trace.

For this example use the following face-outer matching, rather than the vertex-outer rule of Section ?. List all faces, including the empty face, by cardinality and lexicographic order. For each available face \sigma, inspect vertices 21,20,\ldots,0 and pair it with the first available face \sigma\cup\{x\}, if one exists with x\notin\sigma; mark both unavailable. Remove the first pair (\varnothing,\{21\}) to pass to ordinary chains. There are 4489 faces including the empty face, 2222 ordinary pairs, one critical vertex 21 and forty-three critical four-faces. Acyclicity is checked on all 19162 nonempty Hasse incidences. Forman’s theorem gives the ordinary wedge, already covered by the cycle-power classification [CyclePowers, Theorem 1.1, specialized to r=2].

Use the projection and lift rules of Section ?, with these matched incidences. Order the critical four-faces lexicographically and write their lifts as z_1,\ldots,z_{43}. The projection is defined on all 1452 four-faces and kills every one of the 462 five-face boundaries. Every lift has zero boundary and standard projection, giving an integral basis by Lemma 12.1. In this order z_{24} lifts the critical face \{1,6,10,14,18\}. Set z=[z_{24}].

Let M have columns

in the integral lift basis. The chains defining the two seeds have thirty-two terms each; their boundaries vanish. Applying the signed vertex translations and projection gives all columns of M. These integer calculations give (21.1) and the maximal minors

where the columns are numbered from zero and the other columns retain their written order. The nonzero minors prove rational rank forty-three. The displayed primitive relation therefore generates the entire integer relation group, and the greatest common divisor of these minors is the subgroup index, namely two. There are exactly two circular gap patterns up to translation of a gap word: the two gaps of length five are adjacent or separated. They give the specified seed sets and their translates; the complete family has forty-four sets. This identifies the family being measured without asserting completeness of all cross-cycles of this complex.

Moreover the matrix with ordered columns

has determinant 1 in the same lift basis. It supplies the explicit missing primitive class and proves that u,v,[z_{24}] generate H as a \mathbb Z[t]-module. We now describe the dihedral action on the index-two subgroup L. Signed vertex negation on both seed chains gives Ru=-T^9u and Rv=-T^{13}v. Thus L is D_{22}-stable; the quotient of order two has trivial action. The shifts give multiplier -t by R(T^a u)=-T^{9-2a}(T^a u) and the analogous formula for v. For the shifted seeds u',v', relation (21.1) is unchanged, and the unimodular generator change (u',v')\mapsto(u'+2v',v') puts the relation into N(T)(u'+2v')=0. Its primitive relation group gives exactly (21.1), including the reflection on both factors.

To identify the ambient cyclic module, use the two seeds z,w=u+v. Their forty-four orbit columns have the unique primitive relation

Deleting the initial column z leaves the ordered basis

whose determinant is -1 in the fixed lift basis. With e=z-3w, the relation becomes N(T)e=0. The normalized columns

have determinant 1. Both determinants and all orbit columns are checked from the lifted chains and signed vertex translation. Since T^{22}=1, these columns and their annihilators prove (21.1) by the concatenated orbit-basis criterion of Section ?. One orbit has rank at most twenty-two, whereas H has rank forty-three; the two displayed seeds therefore attain the minimum.

The full summand P=\mathbb Z[t]w lies in L. Thus L=(L\cap A)\oplus P and A/(L\cap A)\cong\mathbb Z/2\mathbb Z. The class of e in this quotient is nonzero: e\equiv z\pmod L, and the determinant in (?) shows that z supplies the missing class. The quotient action is trivial, so its cyclic module map is determined by e\mapsto1 and t\mapsto1. It is well-defined because N(1)=22 is even, and is precisely \epsilon_2. This proves (21.1).

The characteristic polynomial of T on H is

The repeated rational factors also give the same lower bound by Lemma 20.1. Every rotation trace is -1 except the identity trace 43; the traces of RT^k are 1 for even k and -3 for odd k. All these traces are also checked by the signed simplicial Lefschetz sums.

The even-power sum E_n satisfies E_n+B_n=N_n=0 in A_n and tB_n=E_n=-B_n. Inversion permutes the odd residues modulo n, so Tg=-g and Rg=g. The class g is nonintegral: modulo two, the degree-n-1 polynomial B_n is neither zero nor the monic polynomial N_n of that degree, and hence is not a multiple of N_n. Thus the extension has index two and is stable under both generators. The multiplier h is invariant under inversion and therefore intertwines R. Since hg=-B_n and A_nB_n=\mathbb Z B_n, its image is the ideal (h,B_n). Modulo f=1+t^2, the even-power sum is an alternating sum of m terms, equal to 1; consequently B_n=t+fQ for some Q\in\mathbb Z[t]. As t is a unit and h=t^{-1}f, this ideal is all of A_n. Neither i nor -i is an nth root of unity because 4\nmid n; hence h is nonzero on every root of N_n and is injective over \mathbb Q. It is therefore an isomorphism on the indicated integer lattices. Its image hA_n has index two and lies in \ker\epsilon_2, since h(1)=2. The latter kernel also has index two, proving equality.

Use B=B_{22} and the shifted seeds a=u'+2v', b=v' from Proposition 21.1, so L=Aa\oplus Pb with the indicated reflections. The lift projection in the ordered basis (Tu,\ldots,T^{21}u,v,\ldots,T^{21}v) gives

where

This is an equality in rational homology, obtained using the same signed chain projection and integer lift basis as in the preceding proposition. In particular z-B(T)u/2\in L. The even shift satisfies T^4B(T)=B(T) on the 22-periodic modules, and B(T)(2v')/2 is integral; thus B(T)a/2\equiv z\pmod L. Since [H:L]=2 and z\notin L, this identifies the ambient lattice as H=(A^+a)\oplus Pb, where A^+=A+\mathbb Z(B/2) inside A\otimes\mathbb Q. Lemma 21.2 with m=11 now gives the pair map pa+qb\mapsto(hp,q) for p\in A^+ and q\in P, proving the assertion. These dihedral coordinates differ from the particular cyclic coordinates e,w in (?); no reflection-stability of those particular embedded factors is required.

Use the vertex-outer matching with priority 0,1,\ldots,17 from Proposition 20.2. Order its twenty-five critical three-faces lexicographically and let z_1,\ldots,z_{25} be their signed critical-cell lifts. The first critical three-face is \{1,5,9,13\}. The lift and projection recurrences in Section ? give the fixed integral basis of H and the matrices of T,R in it. There are two gap words up to cyclic shift: the two gaps of length five are adjacent or alternating. The former gives eighteen start sets, and the latter gives nine; signed translation of the alternating seed has period nine.

Expand each four-factor join into its sixteen oriented simplices and apply the same projection. Let M be the resulting 25\times27 integer matrix in the written orbit order. Its Smith form has twenty-three unit entries and no other nonzero entry. Equivalently, integer matrices U,V with determinants \pm1 satisfy UMV=(I_{23}\oplus0), with the rectangular zero block understood. Since U,V are unimodular, the last four columns of V form an integral basis of \ker M. The four vectors in (22.1) are killed by M; writing them in this basis gives a 4\times4 coordinate matrix with determinant of absolute value one. Thus they generate the entire integer kernel, not merely its rational span. These finite identities use every orbit column; they prove rank twenty-three and primitivity.

For clarity, the quotient map can be specified independently of the Smith transformations. On the ordered lift basis define Q:H\to\mathbb Z^2 by the following two rows, written in five consecutive blocks of length five:

The full projection and signed action columns give

The last two identities prove surjectivity. The primitive subgroup L\subset\ker Q has the same rank twenty-three as \ker Q, hence they coincide. The other identities give the quotient actions and prove (22.1).

Translation by two on the regular summand has two vertex orbits, giving the first two fixed generators. In the second summand, represent an element by p(t)b with \deg p\leq\deg g. The fixed condition says (t-1)g\mid(t^2-1)p. Since g(-1)\neq0, g and t+1 are relatively prime over \mathbb Q[t], so g\mid p over \mathbb Q[t]. Monic division and the degree bound give p=kg for an integer k. Thus the written generators form the entire integer fixed lattice. Oddness of \epsilon gives R(E_ma)=tE_ma. Reciprocity of g gives R(gb)=t^{-\deg g}gb=gb, because Tgb=gb. This proves R=T on the fixed lattice.

An equivariant map A\to X is determined by x=s(e_0); the necessary and sufficient conditions are T^2x=x and Rx=Tx, with s(e_1)=Tx. The second condition is automatic on the fixed lattice just computed. Its image under q is generated by

The corresponding endomorphisms are m, mt and g(1)(1+t), proving the ideal assertion. If d e_0 lies in this image, its first coefficient minus its second is divisible by m; hence m\mid d. Conversely s(e_0)=E_ma, s(e_1)=tE_ma lifts m\operatorname{id}_A and respects both generators. Rotation equivariance alone has precisely the condition T^2x=x, so it gives the same conclusion.

Retain the lift basis z_1,\ldots,z_{25} used to define Q. The two orbit generators in Proposition 20.2, after its stated normalizations, are

They identify H with \mathcal P_{18}^1\oplus\mathbb Z[t]/((t-1)\Phi_9), with Ra=Ta, Rb=b. The displayed quotient rows and QT=JQ give Qa=e_0 and Qb=e_0+e_1. Lemma 22.2 applies with m=9 and g=\Phi_9=1+t^3+t^6, for which g(1)=3. In particular, the complete fixed-lattice image has matrix

Its entries have greatest common divisor three, and its maximal minors have greatest common divisor twenty-seven. Thus its Smith factors are three and nine, by the standard determinantal-divisor criterion. The lemma excludes a lift of the identity even with only rotation equivariance and supplies a lift of nine times the identity with both generators. Dividing this lift by nine over \mathbb Z[1/3] gives the final assertion.

A closed triangle has three signed steps from \{1,-1,s,-s\} summing to zero modulo 3s. For s\geq3, the only possibility is three equal steps of size s. Consequently the triangles are exactly

Their chord edges are disjoint, and every such edge lies in only one triangle; no four-clique exists. Collapse each triangle along [i,i+s]. The unit-edge cycle remains, so the resulting graph is connected, with 3s vertices and 5s edges. It is therefore homotopy equivalent to a wedge of 2s+1 circles, by the usual maximal-tree argument [Hatcher, Section 1.A]. This collapse is not asserted to be invariant under translation or reflection.

Before filling the triangles, the entire integral cycle group of the graph has basis

Indeed, subtracting the written chord coefficient times c_i eliminates that chord from any cycle. The remaining cycle is supported on the unit-edge circle, so its coefficients are all equal and it is an integer multiple of u. Uniqueness follows first from the chord coefficients, then from the coefficient of any unit edge. Translation sends c_i to c_{i+1} and fixes u. Negation reverses the written edges, giving

The oriented boundary of [i,i+s,i+2s] is

The right side is unchanged when i is replaced by i+s. There are no higher faces, so these boundaries give the complete relations in H_1. Writing c=[c_0], the presentation is

Eliminating u gives (t-1)N_s(t)c=0. The first relation is redundant because (t-1)N_s divides t^{3s}-1. Thus there are no further integer relations, and the monic presentation gives an integral basis c,tc,\ldots,t^{2s}c. Multiplication by t^s is an integral automorphism, giving the asserted basis based at w=t^sc. Since Rc=-t^{-s}c, one obtains Rw=-t^{-2s}c=-w, using t^{3s}=1. This proves both geometric actions.

Augmentation is well-defined and surjective, and its kernel is (t-1)M_s by monic division by t-1. For the whole integral fixed lattice, (t-1)p=0 in M_s holds exactly when N_s\mid p in \mathbb Z[t]. Representatives have degree at most 2s, so

Hence every rotation-equivariant lift of a scalar has scalar divisible by three. The map a(1)=N_sw lifts 3\operatorname{id} and respects reflection as well. Dividing it by three gives a section after inverting three. This is the classical fixed-element lifting calculation [BrownLectures, Section 3.3], applied to the explicit geometric presentation.

One has

Let P_F be reduction modulo all the distinct factors of F=(t-1)N_s, and let P_N be reduction from \mathbb Z[t]/N_s to \bigoplus_{d\mid m}\mathbb Z[t]/\Phi_{qd}. Lemma 14.3 reduces the assertion to \operatorname{coker}P_F. Factor P_F through reduction modulo t-1 and N_s. The coarse reduction has cokernel \mathbb Z[t]/(t-1,N_s)=\mathbb Z/3. Both reductions are injective between free lattices of equal rank, so the cokernel sequence for their composition is

Let

be the complete classical CRT reduction. There are integral ring isomorphisms

using x\mapsto t^m and y\mapsto t^q in every map. The relations vanish by the orders q,m in the source and q,d in each target. A Bezout identity for m,q recovers t from these two invertible images, so the maps are surjective. Their free integer ranks agree, proving injectivity. These are the coprime tensor isomorphisms of [MS, Lemma 2.2 and Remark 2.3 of the public preprint]; the same powers in every component make the reduction square commute. Consequently P_N is integrally equivalent to I_r\otimes\Psi_m, and

The classical prime-power Smith formula and its coprime tensor assembly [MS, Theorems 2.6 and 2.15 of the public preprint] give the p-primary multiplicity (m/p^b)\varphi(p^{b-j}) for each p^j, 1\leq j\leq b. In particular this cokernel has exponent dividing m. Its order is coprime to three, so (Proof) splits as a sequence of abelian groups into its primary parts. This proves (23.2). Each maximal p^b occurs, giving exponent 3m. Finally, \sum_{j=1}^b j\varphi(p^{b-j})=(p^b-1)/(p-1) gives the order.

Eliminate each chord coefficient of any integer cycle by subtracting the corresponding multiple of c_i. The remainder is supported on the unit circle and is an integer multiple of u. Chord coefficients and then any unit-edge coefficient prove independence. This is the same complete cycle-group argument used before triangle filling in Proposition 23.1, now for every n,s in the stated range. Literal vertex translation and negation give the four displayed signed identities, proving the direct sum and retaining the geometric generator labels.

Translation equivariance makes all \ell(c_i) equal; the displayed reflection imposes no further restriction. Thus the image is (\alpha,\beta). Let N_n=1+t+\cdots+t^{n-1}. The entire translation-fixed lattice has basis N_nc_0,u, both of which are negated by R. Its image under \ell is the ideal (n\alpha,\beta). An equivariant lift of a scalar is exactly a vector in this fixed lattice mapping to that scalar. A Bézout combination realizes the positive generator of the image ideal. For a primitive map it equals \gcd(n,\beta); taking \alpha=1,\beta=d realizes every d\mid n. This is the classical fixed-element lifting calculation [BrownLectures, Section 3.3] in the stated geometric coordinates.

A triangle corresponds to three signed steps from \{\pm1,\pm s\} summing to zero modulo n. Two unit steps and one chord can sum to zero only for s=2; two chord steps and one unit give 2s=n-1; three chords give 3s=n. The range excludes every other zero sum, including three unit steps. These conditions are mutually exclusive for n\geq7. In the first case the triangles are \{i,i+1,i+2\} and each length-two chord is a free edge of its unique triangle. Removing these pairs leaves the unit circle. In the second case the triangles are \{i,i+s,i-1\}, each with its unique unit edge. Removing those pairs leaves the chord circle, which is connected since \gcd(s,n)=1. Neither case has a four-clique: its free edge cannot lie in a second triangle. The surviving circle is an invariant subcomplex; its inclusion induces T=1,R=-1 on homology, without requiring an equivariant collapse sequence. The 3s=n case is Proposition 23.1. Otherwise there are no triangles or higher simplices, and a maximal-tree contraction gives n+1 circles; the integer action is the graph action already proved.

Proposition 11.1 and the CW arguments following it give the homotopy types and free integral homology groups. Proposition 5.2 identifies all the rank-one and rank-two modules, as applied in Section ?. Propositions 12.2 and 13.1, with Lemma 12.1 and (?), identify the four higher-rank geometric actions as L_{14},L_3,L_2,L_3 and give their orders. Lemma 14.1 and its two applications prove (A.1) and (A.1); Lemma 14.2 and Proposition 14.4 prove (A.1). Over \mathbb Q the factors are coprime, giving the stated rational decompositions. For each corresponding split integer module, the indicated factor-annihilator kernels sum to the entire lattice. Their index is preserved by a module isomorphism, so none of the three glued lattices is isomorphic to the displayed split module.