betaFeedback

WorksMathematics

A Counterexample to Kaplansky's Direct-Finiteness Conjecture in Odd Characteristic

Abstract

We construct a counterexample to Kaplansky's direct-finiteness conjecture in odd characteristic. For one specified odd prime pp, we obtain a field KK of order p4p^4, a finitely generated group GG containing torsion, and finite sums a,b∈K[G]a,b\in K[G] with ab=1ab=1 but ba≠1ba\ne1. The same elements define a cellular automaton on KGK^G that is injective but not surjective.

Introduction

A group algebra K[G]K[G] consists of finite sums ∑gcgg\sum_g c_g g, with cg∈Kc_g \in K, multiplied by the group law and bilinearity. A unital ring is directly finite when ab=1ab=1 implies ba=1ba=1; it is stably finite when this implication holds in every finite square matrix ring. Kaplansky’s direct-finiteness conjecture asks whether K[G]K[G] is directly finite for every field KK and every group GG, including groups with torsion. The result here is a negative answer in one specified odd characteristic.

Theorem 1.1. Put

m=(1200600),p=min⁡{h∈Z:h>1, h∣(m!)2+1}.m=\binom{1200}{600}, \qquad p=\min\{h \in\mathbb{Z}:h>1,\ h\mid(m!)^2+1\}.

Then pp is an odd prime. There are a specified field KK of order p4p^4, a specified finitely generated group GG containing torsion, and specified finite sums a,b∈K[G]a,b \in K[G] satisfying ab=1ab=1 and ba≠1ba \ne1.

The word “specified” refers to finite formulas and choices in ordered finite sets. The field is a quadratic tower, the group is built from finitely many finite groups and explicit identifications, and the sums are defined without solving the word problem of that group. The construction is not claimed to have been evaluated as a practical computation. It concerns the displayed prime, not every odd prime.

The finiteness problem. For finite GG, the algebra and all its finite matrix rings are finite-dimensional over KK. A one-sided inverse between their left multiplication operators is therefore two-sided. This argument works even when the characteristic divides ∣G∣|G|. The difficulty for an infinite group is to obtain a corresponding conclusion from finite support alone.

Kaplansky proved stable finiteness in characteristic zero using finite von Neumann algebras, and asked whether the one-sided-inverse property persists in positive characteristic [19 pp. 122–123]. Ara, O’Meara, and Perera proved stable finiteness for free-by-amenable groups over arbitrary division rings [2]. Elek and Szabó [10 Corollary 4.7] established the same conclusion for all sofic groups. Such groups admit finite permutation models that approximate multiplication while keeping distinct elements separated. These theorems are constraints on the example, not hypotheses about its group. In particular, Elek–Szabó’s theorem implies, after the scalar construction is completed, that GG cannot be sofic.

A separate characteristic-two construction uses incidence structures, polynomial pairings, a planar group-embedding argument, and character packing [20 Theorem 1.1]. The construction here has a different local mechanism. A high-degree subspace of a modular translation representation is annihilated on one side; a weighted pairing gives a compatible annihilation on the other. The characteristic-two theorem is not an input to this odd-prime argument, and the two methods retain distinct coefficient hypotheses.

The two quantities that must be compared. A matrix FF fits a simple graph when Fii=1F_{ii}=1 and Fij=0F_{ij}=0 for distinct nonadjacent vertices. We use a triangle-free graph on mm vertices and a specified factorization

Fij=∑T∈Tαi(T)βj(T),∣T∣=t,tm<1100.F_{ij}=\sum_{T\in\mathcal{T}}\alpha_i(T)\beta_j(T), \qquad|\mathcal{T}|=t, \qquad\frac{t}{m}<\frac{1}{100}.

This is an explicit threshold-Kneser instance of the polynomial constructions of Haviv and Golovnev–Haviv [15 Proposition 13, 11 Theorem 3.2 and Lemma 3.5], whose polynomial ancestry includes Alon–Babai–Suzuki [1]. Low-rank fitting matrices also give storage-code recovery equations. The triangle-free constructions of Barg–Zémor, Huang–Xiang, and Barg–Schwartz–Yohanov provide related context [4, 16, 3]. Their results are not used to supply a group-algebra idempotent.

At every vertex we use a copy of a finite group HH and an idempotent e∈K[H]e\in K[H]. Write w=∣H∣w=|H| and d=dim⁡KeK[H]d=\dim_K eK[H], the rank of left multiplication by ee on the regular representation. The second quantity is

dw>196.\frac{d}{w}>\frac{1}{96}.

The idempotents satisfy a directed condition: after identifying certain cyclic subgroups at adjacent vertices, eiej=0e_i e_j=0 for every edge with i<ji<j. There is no assertion about the reverse product. The asymmetry is what allows a triangular calculation to combine the two strict bounds.

The characteristic-dependent construction appears in Section 3. High augmentation degree gives the annihilated columns, in the elementary abelian setting of Jennings’ theory [17]. A diagonal weighted Gram determinant and Schwartz’s grid bound [23 Lemma 1 and Corollary 1], with related polynomial-identity work of Zippel [25], supply enough nondegenerate pairings. Fourier character orbits turn these projections into actual elements of K[H]K[H]. The full algebra need not be semisimple; only its prime-to-pp normal abelian subgroup is diagonalized, in the character-orbit setting of Clifford [8].

From the comparison to scalar elements. The proof has three transitions that numerical rank inequalities alone cannot justify. First, the quotient identifying the cyclic subgroups must retain the finite vertex groups. Its injectivity proof uses the C(4)C(4)–T(4)T(4) graphical-complex mechanism described by Prytula [22 Section 8]: local short-word restrictions and triangle-freeness rule out a spherical map with exactly one nontrivial face. Ordinary graph-of-groups normal form does not by itself establish this first quotient’s injectivity. Section 4 proves it locally, including the reductions for loops and bridges.

Second, regular ranks must become identities over the same ring. A further group enlargement G1G_1 has a nonzero central idempotent ff in R=K[G1]R = K[G_1]. In the ring S=fRS = fR, whose identity is ff, the elements Pi=feiP_i = fe_i admit rectangular matrices

Ji∈Mat⁡w×d(S),Yi∈Mat⁡d×w(S),YiJi=fId,JiYi=PiIw.J_i \in\operatorname{Mat}_{w \times d}(S), \qquad Y_i \in\operatorname{Mat}_{d \times w}(S), \qquad Y_iJ_i = fI_d,\qquad J_iY_i = P_iI_w.

The finite-group blocks in this step have elementary abelian pp-group stabilizers, possibly nontrivial and nonnormal. Their group algebras have nilpotent augmentation ideals, and a finite change of basis proves exact integer rank identities before coordinates are paired. Section 5 gives the full construction of these matrices, with their entries in explicitly attached finite subgroups.

Third, a matrix counterexample does not yet give two scalar elements of the same group algebra. In Section 6, the fit and the ordered zero products produce a split injection

Smd⟶Swt⟶Smd,wt<md,S^{md} \longrightarrow S^{wt} \longrightarrow S^{md}, \qquad wt < md,

whose composite is the identity. We derive square matrices with a one-sided inverse and certify the reverse defect by a nonzero kernel vector. Section 7 then implements the finite-factor reduction of Dykema–Juschenko [9 Lemma 2.1 and Theorem 2.2]. A finite Heisenberg group gives explicit matrix units, their linear independence makes the encoding injective, and a complementary idempotent restores the full scalar identity. The shift and modulation model is classical [14 Section 1.2]; the composite cyclic order and finite coefficient field used here are checked directly.

The dynamical consequence. A configuration over a finite alphabet AA is any function G→AG \to A. A cellular automaton has one finite memory set E⊂GE \subset G and one rule applied to (x(gu))u∈E(x(gu))_{u \in E} at every gg. A group is surjunctive when every injective such automaton, over every finite alphabet, is surjective. Gottschalk proposed this for all groups [12]; Gromov and Weiss proved it for sofic groups [13][24]. The group-ring connection is recorded by Elek–Szabó and developed in the linear theory of Ceccherini-Silberstein–Coornaert [10][5]. Phung proved stable finiteness over every field for surjunctive groups [21 Theorem B]; subsequent proofs and extensions include [6][7].

Section 8 gives the elementary transfer with its actual multiplication convention. The coefficients of bb define an injective automaton, those of aa give a left inverse, and a nonzero coefficient of ba−1ba - 1 specifies a missing point indicator. The argument excludes preimages of arbitrary support. It is a consequence of the scalar construction, not a further existence theorem.

All matrices below act on columns of free right modules. Matrix entries therefore multiply coordinates on the left. Transpose is ordinary transpose over KK, with no involution. Cardinalities and dimensions in rank comparisons are integers, not elements of the coefficient field. For a positive integer nn, write [n]={1,…,n}[n] = \{1,\ldots,n\}.

The field, the graph, and the local group

We first construct the finite data. This fixes every coefficient field and denominator before the modular argument begins.

Lemma 2.1. For m,pm,p in Theorem 1.1, p>mp > m is an odd prime and −1-1 is a square in Fp\mathbb{F}_p. If η\eta is a nonsquare in Fp∗\mathbb{F}_p^*, then

K0=Fp[ξ]/(ξ2−η),K=K0[ν]/(ν2−ξ)K_0 = \mathbb{F}_p[\xi]/(\xi^2-\eta), \qquad K = K_0[\nu]/(\nu^2-\xi)

are quadratic field extensions, so ∣K∣=p4|K| = p^4. Proof. A least divisor greater than one of an integer is prime: any nontrivial proper divisor of a composite least divisor would be smaller. If p≤mp \le m, then pp divides both m!m! and (m!)2+1(m!)^2+1, which is impossible. Hence p>m>600p>m>600, and the defining divisibility gives (m!)2=−1(m!)^2=-1 in Fp\mathbb{F}_p. There are nonsquares because each nonzero square has exactly two square roots. The first quotient is therefore a quadratic field. The conjugate of ξ\xi is −ξ-\xi, so its norm is −η-\eta. Since −1-1 is square, −η-\eta is nonsquare. A square in K0K_0 would have square norm, proving that ξ\xi is nonsquare and that the second quotient is a field.

Choose the least nonsquare using representatives 0,…,p−10,\ldots,p-1; order KK by its coefficients in (1,ξ,ν,ξν)(1,\xi,\nu,\xi\nu). Put M=p4−1M=p^4-1 and choose the least element ζ∈K∗\zeta\in K^* of order MM. This element exists: if bb is the least common multiple of all orders in K∗K^*, then b∣Mb\mid M, while the root bound for Xb−1X^b-1 gives M≤bM\le b. For each prime-power factor of bb, an appropriate power of one element has that exact order. Multiplying the resulting commuting elements of coprime orders gives order b=Mb=M. In particular, MM and all its powers are invertible in KK.

Use natural orders on integer indices, increasing-list lexicographic orders on subsets, and lexicographic orders on tuples and finite function tables. Order characters of (Z/MZ)D(\mathbb{Z}/M\mathbb{Z})^D by their exponent lists on the ordered coordinate generators. These conventions specify the finite choices below.

Let I1,…,ImI_1,\ldots,I_m be the 600600-subsets of [1200][1200] in order. Join distinct i,ji,j in Γ\Gamma exactly when ∣Ii∩Ij∣<200|I_i\cap I_j|<200. Put

T={T⊆[1200]:∣T∣≤400},t=∑j=0400(1200j).\mathcal{T}=\{T\subseteq[1200]:|T|\le400\},\qquad t=\sum_{j=0}^{400}\binom{1200}{j}.

Proposition 2.2. The graph Γ\Gamma is triangle-free and has a specified fitting factorization Fij=∑T∈Tαi(T)βj(T)F_{ij}=\sum_{T\in\mathcal{T}}\alpha_i(T)\beta_j(T) over Fp\mathbb{F}_p with t/m<1/100t/m<1/100.

Proof. Three pairwise adjacent sets would have union of size at least 3⋅600−3⋅199=12033\cdot600-3\cdot199=1203, exceeding 1200. To construct the fit, let αi(T)=1\alpha_i(T)=1 when T⊆IiT\subseteq I_i, and zero otherwise. Define βj(T)\beta_j(T) as the coefficient of ∏x∈TXx\prod_{x\in T}X_x obtained by expanding

Qj(X)=∏k=200599∑x∈IjXx−k600−kQ_j(X)=\prod_{k=200}^{599}\frac{\sum_{x\in I_j}X_x-k}{600-k}

and replacing every positive exponent in each monomial by one. All denominators are in {1,…,400}\{1,\ldots,400\}, hence invertible. This replacement preserves evaluations on zero-one vectors and does not increase degree. Thus

Fij=∏k=200599∣Ii∩Ij∣−k600−k.F_{ij}=\prod_{k=200}^{599}\frac{|I_i\cap I_j|-k}{600-k}.

It equals one for i=ji=j. For distinct nonneighbors the intersection size lies between 200 and 599, so one factor is zero. This is the threshold-Kneser polynomial specialization of [15 Proposition 13] and [11 Theorem 3.2].

The binomial coefficients increase up to index 400. Consequently

tm≤401∏u=401600u1201−u≤401(5/7)100<1100.\frac{t}{m}\le401\prod_{u=401}^{600}\frac{u}{1201-u}\le401(5/7)^{100}<\frac{1}{100}.

In the second inequality the first hundred factors are at most 500/701<5/7500/701<5/7 and the rest are at most one. The last inequality follows from (5/7)3<1/2(5/7)^3<1/2 and 40100<23340100<2^{33}. □\square

For the local algebra define the additive group P=(Fp)3P = (\mathbb{F}_p)^3, the function group A0=(Z/MZ)PA_0 = (\mathbb{Z}/M\mathbb{Z})^P, and

H=A0⋊P,(a,s)(a′,s′)=(a+s⋅a′,s+s′),(s⋅a)(x)=a(x−s).H = A_0 \rtimes P,\qquad(a,s)(a',s') = (a+s\cdot a',s+s'),\qquad(s\cdot a)(x)=a(x-s).

Its order is w=Mp3p3w=M^{p^3}p^3. Let c∈A0c\in A_0 have value one at zero and zero elsewhere, and put

u(s)=(0,s),U=u(P),v(s)=cu(s)c−1,V=cUc−1.u(s)=(0,s),\qquad U=u(P),\qquad v(s)=cu(s)c^{-1},\qquad V=cUc^{-1}.

The coordinate formula v(s)=(c−s⋅c,s)v(s)=(c-s\cdot c,s) gives U∩V={1}U\cap V=\{1\}, because c−s⋅c≠0c-s\cdot c\ne0 when s≠0s\ne0. We next construct an idempotent with incoming annihilation by UU and outgoing annihilation by VV.

Large idempotents with ordered annihilations

The graph has supplied the ratio t/m<1/100t/m<1/100. Our present objective is an idempotent in the finite group algebra K[H]K[H] whose regular rank fraction exceeds 1/961/96. It must also have two prescribed annihilations, on opposite sides. Here

ℓ=(p−1)/2,D=3(p−1),w=∣H∣.\ell=(p-1)/2,\qquad D=3(p-1),\qquad w=|H|.

The regular rank of e∈K[H]e\in K[H] means dim⁡KeK[H]\dim_K eK[H], the dimension of the image of left multiplication. Although this map multiplies on the left, its image is a right ideal.

Proposition 3.1. There is a specified idempotent e∈K[H]e\in K[H] satisfying, for every s∈Ps\in P,

(u(s)−1)ℓe=0,e(v(s)−1)ℓ=0.(u(s)-1)^\ell e=0,\qquad e(v(s)-1)^\ell=0.

Its regular rank dd satisfies d/w>1/96d/w>1/96. Its coefficients are obtained by finite operations in the specified field KK.

The annihilated column space

On KPK^P, let ex\mathbf{e}_x be the standard basis and Lsex=es+xL_s\mathbf{e}_x=\mathbf{e}_{s+x}. For the three standard generators εj\varepsilon_j of PP set Xj=Lεj−IX_j=L_{\varepsilon_j}-I. The XjX_j commute, and characteristic pp gives Xjp=0X_j^p=0. The vectors

X1n1X2n2X3n3e0,0≤n1,n2,n3<p,X_1^{n_1}X_2^{n_2}X_3^{n_3}\mathbf{e}_0,\qquad0\le n_1,n_2,n_3<p,

are a basis. Indeed, every translation of e0\mathbf{e}_0 is a product of powers of I+XjI+X_j applied to e0\mathbf{e}_0, so the displayed family spans KPK^P; its cardinality is the dimension p3p^3.

This is the elementary abelian augmentation filtration in the setting studied by Jennings [17]. Here the exact degree argument can be given directly. Let BB have as columns the displayed basis vectors with n1+n2+n3≥D−ℓ+1n_1+n_2+n_3\ge D-\ell+1, in order. For every ss, the matrix Ls−IL_s-I is a polynomial in the XjX_j with zero constant term. Multiplication by its ℓ\ellth power raises every column’s total degree to at least D+1D+1. Such a monomial has an exponent at least pp, so vanishes. Applying the same assertion to −s-s and transposing yields

(Ls−I)ℓB=0,BT(Ls−I)ℓ=0.(L_s-I)^\ell B=0,\qquad B^{\mathsf T}(L_s-I)^\ell=0.

The transpose identity uses L−sT=LsL_{-s}^{\mathsf T}=L_s; no scalar conjugation is present.

The columns are independent. Their number hh can be counted by setting aj=p−1−nja_j=p-1-n_j: the new nonnegative indices have sum at most ℓ−1\ell-1. Their individual upper bounds are automatic because ℓ−1<p\ell-1<p. Stars and bars gives

h=(ℓ+23),hp3=(p−1)(p+1)(p+3)48p3>148.(1)h=\binom{\ell+2}{3},\qquad\frac{h}{p^3}=\frac{(p-1)(p+1)(p+3)}{48p^3}>\frac{1}{48}. \tag*{(1)}

For the strict inequality, the numerator minus p3p^3 is 3p2−p−3>03p^2-p-3>0 for p≥3p\geq3.

The same subspace is killed on the two sides in Equation (3.1), but an arbitrary projection onto it need not give the outgoing conjugate annihilation. We therefore choose the row factor by a diagonal weighted pairing.

Many nonsingular diagonal pairings

For a finite function group A∗=(Z/MZ)EA_*=(\mathbb{Z}/M\mathbb{Z})^E, its characters χ:A∗→K∗\chi:A_*\to K^* are all exponent lists of powers of ζ\zeta. Put

pχA∗=∣A∗∣−1∑a∈A∗χ(a)−1a.(2)p_{\chi}^{A_*}=|A_*|^{-1}\sum_{a\in A_*}\chi(a)^{-1}a. \tag*{(2)}

The denominator is invertible. Reindexing the sum gives apχA∗=χ(a)pχA∗ap_{\chi}^{A_*}=\chi(a)p_{\chi}^{A_*}. A nontrivial character has sum zero, since translating its sum by an element on which it is nontrivial multiplies it by a scalar different from one. It follows that

pχA∗pψA∗=δχψpχA∗,∑χpχA∗=1.p_{\chi}^{A_*}p_{\psi}^{A_*}=\delta_{\chi\psi}p_{\chi}^{A_*},\qquad\sum_{\chi}p_{\chi}^{A_*}=1.

For the last identity, sum coefficients over the independent character exponents: every nonzero coordinate of a nonidentity element is detected by a power of ζ\zeta. Each idempotent is nonzero, there are ∣A∗∣|A_*| of them, and orthogonality makes them independent. They form a basis of K[A∗]K[A_*].

Specialize to A0A_0. Its character action is (sλ)(a)=λ((−s)⋅a)(s\lambda)(a)=\lambda((-s)\cdot a). Define

χx=xλ,Dλ=diag⁡(χx(c):x∈P),Nλ=BTDλ−1B.\chi_x=x\lambda,\qquad D_\lambda=\operatorname{diag}(\chi_x(c):x\in P),\qquad N_\lambda=B^{\mathsf T}D_\lambda^{-1}B.

Call λ\lambda admissible if its stabilizer in PP is trivial and NλN_\lambda is invertible. This condition is a finite test.

Lemma 3.2. More than half of the ∣A0∣=Mp3|A_0|=M^{p^3} characters are admissible.

Proof. For each nonzero s∈Ps\in P, translation partitions PP into p2p^2 cycles of length pp. A character fixed by ss has constant exponent on each cycle, so exactly Mp2M^{p^2} characters are fixed. A union bound therefore bounds the fraction with nontrivial stabilizer by

(p3−1)Mp2−p3.(p^3-1)M^{p^2-p^3}.

Let crc_r be the coordinate generator at rr. Since χx(c)=λ(c−x)\chi_x(c)=\lambda(c_{-x}), the diagonal entries of Dλ−1D_\lambda^{-1} range independently over K∗K^* as λ\lambda ranges over all characters. For independent variables τx\tau_x, Cauchy–Binet gives

Q(τ):=det⁡(BTdiag⁡(τx)B)=∑J⊆P∣J∣=hdet⁡(BJ)2∏x∈Jτx.(3)Q(\tau):=\det(B^{\mathsf T}\operatorname{diag}(\tau_x)B)=\sum_{\substack{J\subseteq P\\ |J|=h}}\det(B_J)^2\prod_{x\in J}\tau_x. \tag*{(3)}

Here BJB_J consists of the indicated rows in their fixed order. Some maximal minor is nonzero because BB has full column rank. The monomials for distinct JJ are distinct, so QQ is nonzero of total degree hh, even though the unweighted Gram matrix could be singular.

We use the polynomial grid estimate of Schwartz [23 Lemma 1 and Corollary 1]; see also the related identity-testing method of Zippel [25]. A nonzero polynomial of total degree at most aa vanishes on at most an a/Ma/M fraction of a product grid whose coordinate sets each have size MM. To prove this, induct on the number of variables. If the degree in the last variable is bb, its leading coefficient has degree at most a−ba-b. It vanishes at at most an (a−b)/M(a-b)/M fraction of the earlier choices. For each remaining choice, the last-variable polynomial has at most bb roots, contributing at most b/Mb/M. The sum proves the bound; the constant case starts the induction.

Applying the estimate on (K∗)p3(K^*)^{p^3} to (3), the singular-pairing fraction is at most h/Mh/M. Both excluded fractions are strictly smaller than 1/41/4:

hM≤p3p4−1<14,(p3−1)M−p2(p−1)≤p3−1M<14.\frac{h}{M} \le\frac{p^3}{p^4-1} < \frac{1}{4}, \qquad(p^3-1)M^{-p^2(p-1)} \le\frac{p^3-1}{M} < \frac{1}{4}.

The estimates use p>600p>600; for example, p4−1−4p3=p3(p−4)−1>0p^4-1-4p^3=p^3(p-4)-1>0. Their union therefore excludes fewer than half of the characters.

Select the first admissible character from each PP-orbit containing one, and call the set of selected characters R\mathscr{R}. Each selected orbit is free, of size p3p^3. These orbits cover all admissible characters, so

∣R∣>∣A0∣2p3.(4)|\mathscr{R}|>\frac{|A_0|}{2p^3}. \tag*{(4)}

This argument does not need admissibility to be constant on an orbit.

Realizing the projections in the group algebra

Fix λ∈R\lambda\in\mathscr{R}. For x,y∈Px,y\in P, define

Exyλ=pxλA0u(x−y).E_{xy}^{\lambda}=p_{x\lambda}^{A_0}u(x-y).

Conjugation gives u(s)pχA0u(s)−1=psχA0u(s)p_{\chi}^{A_0}u(s)^{-1}=p_{s\chi}^{A_0}. Fourier orthogonality and freeness of the orbit therefore imply

ExyλEztλ=δyzExtλ.E_{xy}^{\lambda}E_{zt}^{\lambda}=\delta_{yz}E_{xt}^{\lambda}.

Indeed the Fourier product can survive only when xλ=(x−y+z)λx\lambda=(x-y+z)\lambda, which is equivalent to y=zy=z. Products between different orbits vanish. The sum qλ=∑xpxλA0q_\lambda=\sum_xp_{x\lambda}^{A_0} is central: it commutes with A0A_0 and its summands are permuted by PP. The basis pxλA0u(s)p_{x\lambda}^{A_0}u(s) of its ideal is indexed equally by (x,y)(x,y) through s=x−ys=x-y. Thus these units prove

qλK[H]≅Mat⁡p3(K).q_\lambda K[H]\cong\operatorname{Mat}_{p^3}(K).

This is an explicit character-orbit block, in the normal-subgroup representation setting of Clifford [8]. It does not assert semisimplicity of the entire modular algebra.

On this block, the elements u(s)u(s), cc, and v(s)v(s) act respectively by

Ls,Dλ,DλLsDλ−1.L_s,\qquad D_\lambda,\qquad D_\lambda L_sD_\lambda^{-1}.

For example, left multiplication by u(s)u(s) shifts row xx to row s+xs+x, whereas cc multiplies row xx by (xλ)(c)(x\lambda)(c).

Define a matrix and then a finite group-algebra element by

Tλ=BNλ−1BTDλ−1,e=∑λ∈R∑x,y∈P(Tλ)xyExyλ.(5)T_\lambda=BN_\lambda^{-1}B^{\mathsf{T}}D_\lambda^{-1},\qquad e=\sum_{\lambda\in\mathscr{R}}\sum_{x,y\in P}(T_\lambda)_{xy}E_{xy}^{\lambda}. \tag*{(5)}

The inverse exists by the selection criterion. Since BTDλ−1B=NλB^{\mathsf T}D_{\lambda}^{-1}B=N_{\lambda}, we have TλB=BT_{\lambda}B=B and Tλ2=TλT_{\lambda}^{2}=T_{\lambda}. Its image is exactly im⁡B\operatorname{im} B, of dimension hh. Orthogonality of the blocks proves e2=ee^{2}=e; on the complementary ideal its component is zero.

We now compute the regular rank, rather than the smaller rank of a single representation matrix. In Mat⁡p3(K)\operatorname{Mat}_{p^{3}}(K), left multiplication by TλT_{\lambda} has image the matrices whose p3p^{3} columns all lie in im⁡B\operatorname{im} B. Its image dimension is hp3hp^{3}. Summing the selected blocks gives

d=hp3∣R∣,dw=h∣R∣∣A0∣>h2p3>196.(6)d=hp^{3}|\mathscr{R}|,\qquad\frac{d}{w}=\frac{h|\mathscr{R}|}{|A_{0}|}>\frac{h}{2p^{3}}>\frac{1}{96}. \tag*{(6)}

The incoming annihilation follows from (Ls−I)ℓTλ=0(L_{s}-I)^{\ell}T_{\lambda}=0. For the outgoing annihilation, keep the factors in their actual order:

BTDλ−1(DλLsDλ−1−I)ℓ=BT(Ls−I)ℓDλ−1=0.\begin{aligned} B^{\mathsf T}D_{\lambda}^{-1}(D_{\lambda}L_{s}D_{\lambda}^{-1}-I)^{\ell}=B^{\mathsf T}(L_{s}-I)^{\ell}D_{\lambda}^{-1} \\ &=0. \end{aligned}

Equation (3.1) proves both assertions on every selected block. They also hold on the complementary ideal, where ee is zero. This proves Proposition 3.1, including its finite prescription and the strict rank estimate.

From opposite annihilations to an ordered zero product

Take copies HiH_i of HH. For each edge i<ji<j, identify v(1,jˉ,jˉ2)v(1,\bar{j},\bar{j}^{2}) in HiH_i with u(1,iˉ,iˉ2)u(1,\bar{i},\bar{i}^{2}) in HjH_j, where bars denote residues in Fp\mathbb{F}_p. Let G0G_0 be the resulting free-product quotient, and let eie_i denote the image of ee from HiH_i. The injectivity of these copies is proved in Section 4. The following ordered identity already follows from the local calculations and the quotient relations.

Corollary 3.3. For every edge i<ji<j of Γ\Gamma, eiej=0e_i e_j=0 in K[G0]K[G_0].

Proof. Let yy be its generator, initially of order pp on either local side, and put X=y−1X=y-1. In the local cyclic group algebra K[⟨y⟩]=K[X]/(Xp)K[\langle y\rangle]=K[X]/(X^{p}), the annihilator of right multiplication by XℓX^{\ell} is precisely the ideal (Xp−ℓ)(X^{p-\ell}). Choose representatives of the left cosets a⟨y⟩a\langle y\rangle in HiH_i; they give a free right K[⟨y⟩]K[\langle y\rangle]-basis of K[Hi]K[H_i]. Thus the outgoing identity eiXℓ=0e_iX^{\ell}=0, applied coefficientwise in this basis before taking images in the quotient, gives ei=qXp−ℓe_i=qX^{p-\ell} for some local qq. On the incoming side Xℓej=0X^{\ell}e_j=0. Since p−ℓ=ℓ+1p-\ell=\ell+1,

eiej=qXℓ+1ej=qX(Xℓej)=0.e_i e_j=qX^{\ell+1}e_j=qX(X^{\ell}e_j)=0.

No factor is commuted across XX, and no reverse ordered product is asserted.

We have constructed the large regular rank and the edge vanishings. To compare them with the graph factorization, the remaining structural step must put exact rectangular realizations of all those ranks inside one nonzero ring.

Gluing the finite groups without collapsing them

The quotient G0G_0 of Section 3 identifies a cyclic subgroup across each edge of Γ\Gamma. It must retain every vertex group: otherwise a later nonzero element of a finite group algebra could disappear. We first prove a general injectivity criterion and then verify it for the groups used here. The local injectivity argument belongs to the C(4)–T(4) theory of graphical complexes of groups [22 Section 8]. We give a direct proof using only labelled spherical maps and Euler’s formula.

An injectivity criterion

Lemma 4.1 (Triangle-free gluing). Let Δ\Delta be a finite simple triangle-free graph. For each vertex ii, let DiD_i be a finite group. For each edge γ={i,j}\gamma=\{i,j\}, let CγC_\gamma be a nontrivial finite cyclic group, with injective homomorphisms ιγ,i:Cγ→Di\iota_{\gamma,i}:C_\gamma\to D_i and ιγ,j:Cγ→Dj\iota_{\gamma,j}:C_\gamma\to D_j. Assume the following condition at every vertex ii. If 1≤k≤31\le k\le3, the edges γ1,…,γk\gamma_1,\ldots,\gamma_k are incident to ii, and sa∈Cγa∖{1}s_a\in C_{\gamma_a}\setminus\{1\}, then

ιγ1,i(s1)⋯ιγk,i(sk)≠1(7)\iota_{\gamma_1,i}(s_1)\cdots\iota_{\gamma_k,i}(s_k)\ne1 \tag*{(7)}

whenever consecutive edge labels are distinct cyclically. For k=1k=1 the distinctness condition is omitted; for k≥2k\ge2 it includes γk≠γ1\gamma_k\ne\gamma_1. Then every natural homomorphism Di→QD_i\to Q is injective, where

Q=(∗i∈V(Δ)Di)/⟨ ⁣⟨ιγ,i(s)ιγ,j(s)−1:γ={i,j}, s∈Cγ⟩ ⁣⟩.Q=\left(*_{i\in V(\Delta)}D_i\right)\Big/\left\langle\!\left\langle\iota_{\gamma,i}(s)\iota_{\gamma,j}(s)^{-1}:\gamma=\{i,j\},\ s\in C_\gamma\right\rangle\!\right\rangle.

Proof. We first handle the subgroups generated by the incident cyclic groups. The diagram argument will rule out a relation that is nontrivial at one vertex but follows from the relations at all the vertices. At the end, a permutation action extends the conclusion to the full groups DiD_i.

Put

Di0=⟨ιγ,i(Cγ):γ∋i⟩.D_i^0=\langle\iota_{\gamma,i}(C_\gamma):\gamma\ni i\rangle.

Choose a generator of each CγC_\gamma, and let LL have one generator xγx_\gamma for each edge. Its defining relators are all words in the xγ±1x_\gamma^{\pm1} incident to a single vertex that evaluate to 11 in that vertex group. There is a well-defined map Di0→LD_i^0\to L: two words representing the same element of Di0D_i^0 differ by one of these relators.

A failed injection produces a spherical map. Suppose a nonidentity element of Di0D_i^0 maps to 11 in LL, and choose a word ww representing it. There is a finite sequence of words from ww to the empty word, each step inserting or deleting a defining relator, its inverse, or a free inverse pair. Indeed, membership in the normal closure of the relators expresses ww in the free group as a finite product of conjugates of relators; free insertions and cancellations reach such a product, and the conjugated relators can then be erased.

We describe the spherical maps needed to encode this sequence. A map is a finite connected graph drawn tamely on the sphere, allowing loops and parallel edges. An edge is either neutral, with value 11, or is labelled by an edge γ\gamma of Δ\Delta. In the latter case its two oriented traversals, called darts, have mutually inverse values in Cγ∖{1}C_\gamma\setminus\{1\}. Every face has a type i∈V(Δ)i\in V(\Delta) incident to every label in its boundary walk. Its boundary word is evaluated in DiD_i, using the specified embeddings. A face is ordinary when this evaluation is the identity. Call the map exceptional if precisely one face has nonidentity evaluation and all other faces have identity evaluation. Changing the starting point conjugates a boundary product, and reversing the walk inverts it, so identity versus nonidentity is well-defined.

Draw the successive words in horizontal levels of a rectangle, with each letter an oriented edge. Represent an empty level by a single neutral edge. In every intervening strip join corresponding vertices along the unchanged prefix and suffix by noncrossing neutral edges. At an insertion or deletion, join the single insertion vertex on the shorter path to the two endpoints of the inserted block on the longer path. The strip faces then read either a letter followed by its inverse, or the inserted relator, with neutral edges interspersed. If a level is empty, the adjacent nonempty word is the entire inserted or deleted block; one face between these two levels suffices. Assign the relator’s vertex type to its face, and either endpoint type of a letter’s label to an unchanged-letter face. All these faces evaluate to 11. Include the neutral edges along the two vertical sides of the rectangle. The exterior face has word ww or w−1w^{-1}, apart from neutral edges; give it type ii. This connected drawing, viewed on the sphere, is exceptional.

We shall prove that no exceptional map exists. Choose one with the smallest possible number of edges. This number is positive: the only face of an isolated vertex has the empty word, which evaluates to 11. The reductions below will make every vertex degree and every ordinary face length at least four away from a possible loop. Euler’s formula will then rule out both the loop-free case and the remaining one-loop case.

Boundary walks and elementary reductions. Fix an orientation of the sphere. At each vertex let σ\sigma be the cyclic successor of outgoing darts and let α\alpha reverse a dart. Boundary walks are the cycles of σα\sigma\alpha: after traversing a dart, take the successor of its reverse. For a disconnected drawing we use each component’s own boundary walks, counting one empty walk for an isolated vertex. If a drawing has vv vertices, ee edges, FF such walks and cc components, then

v−e+F=2c.v-e+F=2c.

For completeness, in a connected spherical drawing a nonbridge edge lies on a simple circuit. Its two sides are therefore on distinct boundary walks; deleting it merges those walks and preserves connectedness. Repeated deletion leaves a tree, which has one boundary walk and e=v−1e=v-1. This proves the formula for a connected drawing; summing proves the displayed formula. The same walk rules follow by taking small vertex discs and thin edge bands.

A neutral nonloop edge can be contracted, with its neutral darts omitted from the boundary words. This preserves every face evaluation and contradicts minimality. For a neutral loop based at bb, retain the closed side containing the exceptional face, then remove the loop. The retained graph is connected: each side of the loop attaches to the other only at bb. The other dart slots on the retained side form an interval between the loop slots; closing that interval gives precisely the old walks on that side with the neutral loop dart removed. Thus their evaluations are unchanged. If only bb remains, its walk is empty and could not be exceptional. In every possible case a smaller exceptional map results. Hence every edge is nonneutral.

There are no bridges. To see this, delete a bridge with value ss, leaving two components. Its incident face walk splits into words W1,W2W_1,W_2, one in each component, possibly empty. Both receive the old face type. With suitable starting point and orientation, the old word was

W1sW2s−1.W_1sW_2s^{-1}.

If it was exceptional, at least one WiW_i has nonidentity evaluation; retain that component. If it was ordinary, the evaluations of W1W_1 and W2W_2 are either both identity or both nonidentity. In the first case retain the component with the original exceptional face. In the second retain the component without that face: its new boundary walk is now its unique exceptional face. All choices give a smaller exceptional map.

Every edge consequently separates two distinct faces. Their types must also be distinct. Otherwise delete the edge and merge the faces, using their common type. If their words, starting at its opposite darts, are sDsD and s−1Es^{-1}E, the merged word is DEDE, and

DE=s−1(sD)s(s−1E).DE=s^{-1}(sD)s(s^{-1}E).

Thus two ordinary faces merge to an ordinary face, whereas one ordinary and one exceptional face merge to an exceptional face. Again this would reduce the number of edges. The types across an edge labelled γ\gamma are therefore exactly the two endpoints of γ\gamma.

There is at most one loop, and if it exists it bounds the exceptional face as a monogon. In fact, given a loop, keep its closed side without the exceptional face, including the loop and its base vertex, and erase the graph on the other side. All retained old faces are ordinary. The erased side becomes a monogon with the loop’s nonidentity value; either endpoint of its label supplies its type. This is another exceptional map. Minimality forces the erased side to have contained no other edges in the original drawing. That side was consequently already the exceptional monogon. This applies to every loop, and the one exceptional face cannot be a monogon for two different loops.

Consecutive equal labels can be smoothed. We have reduced to nonneutral, nonbridge edges with different face types on their two sides; the sole possible loop bounds the exceptional monogon. It remains to rule out consecutive nonloop darts with equal labels and then compare degrees with Euler’s formula. The remaining reduction ensures that the short-word hypothesis applies to face boundaries. Suppose a face Φ\Phi has consecutive nonloop darts

e:a⟶b,f:b⟶de:a\longrightarrow b,\qquad f:b\longrightarrow d

with a common label γ\gamma and respective values ϵ,φ\epsilon,\varphi. The two darts belong to distinct edges: otherwise the edge would occur on both sides of Φ\Phi and would be a bridge. Let X,YX,Y be the faces containing eˉ,fˉ\bar{e},\bar{f}, respectively. They are distinct from Φ\Phi and have a common type jj, namely the other endpoint of γ\gamma from the type of Φ\Phi. They may coincide with each other.

Remove the two edges and insert an edge g:a→dg:a\to d along their arcs, smoothed through the empty corner of Φ\Phi at bb. Give it value ϵφ\epsilon\varphi and label γ\gamma, or make it neutral when that product is 1. If a=da=d, this produces a loop. Keep all vertices, including any that become isolated. At a,da,d the new darts occupy the old slots of e,fˉe,\bar{f}; at bb delete the consecutive slots eˉ,f\bar{e},f. These instructions specify the spherical embedding even when vertices or other boundary segments coincide. The walk of Φ\Phi replaces efef by gg and has unchanged evaluation.

When X≠YX\ne Y, write

X=eˉD,Y=fˉE,D:a→b,E:b→d.X=\bar{e}D,\qquad Y=\bar{f}E,\qquad D:a\to b,\qquad E:b\to d.

The walks D,ED,E avoid the removed edges, so the new graph remains connected. The two opposite faces merge to the walk gˉDE\bar{g}DE. All other walks, apart from the specified change in Φ\Phi, are unchanged. Its evaluation in DjD_j is

(ϵφ)−1DE=φ−1(ϵ−1D)φ(φ−1E).(\epsilon\varphi)^{-1}DE=\varphi^{-1}(\epsilon^{-1}D)\varphi(\varphi^{-1}E).

Thus the merged face is ordinary when both old faces were ordinary, and exceptional when exactly one was exceptional. This yields a smaller exceptional map.

The case X=YX=Y requires keeping track of components. Write

X=eˉDfˉE,D:a→d,E:b→b.X=\bar{e}D\bar{f}E,\qquad D:a\to d,\qquad E:b\to b.

The slot rules replace this single walk by the two walks

gˉDandE.(8)\bar{g}D\qquad\text{and}\qquad E. \tag*{(8)}

One can read this directly from σα\sigma\alpha: the transition from gˉ\bar{g} is the old transition from eˉ\bar{e}, and removal of the two slots at bb joins the end of EE to its beginning. If EE is empty, the old transitions were e↦fe\mapsto f and fˉ↦eˉ\bar{f}\mapsto\bar{e}. Equivalently, σ(eˉ)=f\sigma(\bar{e})=f and σ(f)=eˉ\sigma(f)=\bar{e}, so bb had degree two and now becomes an isolated vertex with its stipulated empty face. If DD is empty, then a=da=d and gˉD\bar{g}D is the monogon of the new loop. Thus the description includes all empty cases. Figure 1 illustrates the local replacement and the two boundary walks; all other incidences are suppressed.

Smoothing two consecutive nonloop darts of one label

Figure 1. Smoothing two consecutive nonloop darts of one label. All other incidences are omitted. The opposite faces X,YX,Y may be the same face; in that case its boundary splits as shown below the local picture. The resulting drawing has two components, as the boundary-walk calculation shows. If EE is empty, its component is the isolated vertex bb.

Across the components, the total number of boundary walks increases by one, the edge count decreases by one, and the vertex count is unchanged. Formula (4.2) gives exactly two components. The new edge places a,da,d in one of them. Every component must contain an endpoint of a removed edge, because the old graph was connected; hence bb is in the other component. In particular, the two walks in (8) belong to different components.

Give both walks type jj, and denote their evaluations by

A=(ϵφ)−1D,B=E.A=(\epsilon\varphi)^{-1}D,\qquad B=E.

The old word has evaluation ϵ−1Dφ−1B\epsilon^{-1}D\varphi^{-1}B. Moving the first factor ϵ−1D\epsilon^{-1}D to the end is a cyclic change of starting point, so it preserves identity versus nonidentity. The resulting evaluation is

(φ−1Bφ)A.(\varphi^{-1}B\varphi)A.

If XX was exceptional, at least one of A,BA,B is nonidentity; retain its component. If XX was ordinary, AA and BB are either both identity or both nonidentity. In the former case keep the component with the original exceptional face, whose evaluation is unchanged even if it is Φ\Phi. In the latter keep the component without that face; its new boundary is its sole exception. This again gives a smaller exceptional map. We have proved that no face of the minimal map has consecutive nonloop darts with equal labels, including the last and first darts of its cyclic boundary walk.

Euler’s formula rules out the minimal map. Consider a vertex that is not the base of a loop. Reading the adjacent face types around it gives a closed walk in Δ\Delta. Each incident edge crosses between its two different endpoint types, and successive edge labels are distinct by the smoothing reduction. Thus the walk has no immediate reversal. Its length is positive since the connected map has edges. A simple triangle-free graph has no such closed walk of length less than four, so this vertex has degree at least four.

An ordinary face not incident to a loop has a nonempty boundary walk. Indeed every boundary walk of a connected graph with at least one edge is a nonempty dart cycle; an empty walk occurs only for an isolated vertex. Its consecutive labels are distinct, and the short-word condition (7) therefore forces its length to be at least four. Here and below a loop counts twice toward its vertex degree, and boundary lengths count dart occurrences.

For our connected spherical map, Euler’s formula and the two degree sums give

∑v(4−deg⁡v)+∑F(4−∣∂F∣)=8.(9)\sum_v (4-\deg v)+\sum_F (4-\lvert\partial F\rvert)=8. \tag*{(9)}

Without a loop, only the exceptional face can contribute positively, and its nonempty walk contributes at most 3. With a loop, its base vertex contributes at most 2, the exceptional monogon contributes 3, and the face on the other side of the loop contributes at most 2: that face is ordinary and cannot have length one, since its sole letter would be the loop’s nonidentity value. There are no other loops, so every remaining vertex and face contributes at most zero. The total is at most 2+3+2=72+3+2=7, again contradicting (9). Consequently no exceptional map exists, and every Di0→LD_i^0\to L is injective.

Extending to the full vertex groups. Let Ω=L×N\Omega=L\times\mathbb{N}, with LL acting by left multiplication on the first coordinate. The restriction to each embedded Di0D_i^0 is free and has countably infinitely many orbits. The same is true of the left Di0D_i^0-action on Di×ND_i\times\mathbb{N}. Pair their orbits and choose one point in each pair; this determines a Di0D_i^0-equivariant bijection Di×N→ΩD_i\times\mathbb{N}\to\Omega. Transport the free left DiD_i-action along this bijection. Its restriction to Di0D_i^0 is exactly the original LL-action. Make this choice separately for each ii. The resulting actions agree on every edge group because both copies of its chosen generator act as xγ∈Lx_\gamma\in L. They therefore define an action of QQ on Ω\Omega, faithful on every DiD_i. This proves the required injections.

Application to the modular construction

Recall from Section 2 the group H=A0⋊PH=A_0\rtimes P, with P=(Fp)3P=(\mathbb{F}_p)^3 and A0=(Z/MZ)PA_0=(\mathbb{Z}/M\mathbb{Z})^P. Its subgroups U=u(P)U=u(P) and V=cUc−1V=cUc^{-1} satisfy U∩V={1}U\cap V=\{1\}. The quotient G0G_0 identifies, for every edge i<ji<j, the elements

v(1,jˉ,jˉ2)∈Hi,u(1,iˉ,iˉ2)∈Hj.(10)v(1,\bar{j},\bar{j}^{2})\in H_i,\qquad u(1,\bar{i},\bar{i}^{2})\in H_j. \tag*{(10)}

where iˉ\bar{i} denotes the residue of ii in Fp\mathbb{F}_p. Both elements have order pp, so the corresponding maps from CpC_p are embeddings. We now verify the remaining local condition.

Proposition 4.2. Every natural map Hi→G0H_i\to G_0 is injective. In particular, the maps K[Hi]→K[G0]K[H_i]\to K[G_0] induced on group algebras are injective.

Proof. We verify the short-word hypothesis of Lemma 4.1. At a fixed vertex, every incident edge subgroup lies either in UU or in VV. Within either subgroup, the labels supplied by distinct neighbors jj correspond to the vectors (1,jˉ,jˉ2)(1,\bar{j},\bar{j}^{2}); a nonidentity syllable corresponds to a nonzero scalar multiple of its vector. Any three with distinct residues are independent, because their determinant is, up to sign,

(jˉ2−jˉ1)(jˉ3−jˉ1)(jˉ3−jˉ2)≠0.(\bar{j}_2-\bar{j}_1)(\bar{j}_3-\bar{j}_1)(\bar{j}_3-\bar{j}_2)\ne0.

The residues are distinct since p>mp>m. The analogous first two coordinates show independence for two vectors, and a single vector is nonzero. Thus a word of at most three nonidentity syllables with distinct labels, all in UU or all in VV, cannot evaluate to 11.

For a mixed word of length two or three, one of U,VU,V supplies exactly one syllable. After a cyclic permutation, a putative relation would equate that nonidentity syllable to a product of syllables from the other subgroup. This contradicts U∩V={1}U\cap V=\{1\}. For lengths two and three, cyclically distinct consecutive labels are precisely distinct labels, so all required words have been covered. The graph Γ\Gamma is simple and triangle-free, and all groups in question are finite. The lemma proves the injections. The group-algebra assertion follows because distinct elements of an embedded subgroup remain distinct basis elements of K[G0]K[G_0].

Realizing the local ranks in a common central sector

An idempotent in a finite group algebra has a well-defined rank as a linear operator. We now turn each of these numerical ranks into an isomorphism between modules over one common ring. The enlargement of the group is needed because the different simple or modular summands of the original group algebra need not have the same rank fraction.

Here are the precise inputs to this step. Let KK be a field of characteristic pp, let MM be a positive integer prime to pp, and assume that KK contains an element ζ\zeta of order MM. In the application, M=p4−1M = p^4 - 1 and ∣K∣=p4\lvert K\rvert= p^4. Write CM=Z/MZC_M = \mathbb{Z}/M\mathbb{Z} and put

P=(Fp)3,A0=(CM)P,H=A0⋊P,w=∣H∣,P = (\mathbb{F}_p)^3,\qquad A_0 = (C_M)^P,\qquad H = A_0 \rtimes P,\qquad w = \lvert H\rvert,

where PP acts on the functions in A0A_0 by translation. Suppose that a countable group G0G_0 contains specified copies H1,…,HmH_1,\ldots,H_m of HH. For each ii, let ei∈K[Hi]e_i \in K[H_i] be an idempotent, and assume that the positive integer

d=dim⁡KeiK[Hi]d = \dim_K e_iK[H_i]

is independent of ii. All matrices below act on column vectors in free right modules, so their entries multiply coordinates on the left.

Proposition 5.1. Under these assumptions, there is a group G1G_1 containing G0G_0 and a central element z∈G1z \in G_1 of order MM with the following property. Set

R=K[G1],f=1M∑k=0M−1ζ−kzk,S=fR,Pi=fei.R = K[G_1],\qquad f = \frac{1}{M}\sum_{k=0}^{M-1}\zeta^{-k}z^k,\qquad S = fR,\qquad P_i = fe_i.

Then ff is a nonzero central idempotent. There are explicitly specified matrices

Ji∈Mat⁡w×d(S),Yi∈Mat⁡d×w(S)J_i \in\operatorname{Mat}_{w \times d}(S),\qquad Y_i \in\operatorname{Mat}_{d \times w}(S)

satisfying

YiJi=fId,JiYi=PiIw.Y_iJ_i = fI_d,\qquad J_iY_i = P_iI_w.

The entries of JiJ_i and YiY_i lie in the group algebra of a finite subgroup of G1G_1, multiplied by ff. Their construction uses only finite group operations and linear algebra over KK.

We prove the proposition in three stages. First we attach the finite groups while preserving all the given embeddings. Next we describe their ff-parts as matrix algebras over group algebras of elementary abelian pp-groups. Finally, nilpotence of the corresponding augmentation ideals makes the ranks exact and produces the two rectangular matrices.

Attaching the finite groups

Let zz initially generate an independent copy of CMC_M, and write Qbase=⟨z⟩×G0Q_{\mathrm{base}} = \langle z\rangle\times G_0. For each ii, define

Bi=(CM)Hi,Li=Bi⋊Hi,(h⋅b)(x)=b(h−1x).B_i = (C_M)^{H_i},\qquad L_i = B_i \rtimes H_i,\qquad(h \cdot b)(x) = b(h^{-1}x).

The groups BiB_i are written additively. Let δi∈Bi\delta_i \in B_i be the constant function of value 1∈CM1 \in C_M. It has order MM and is fixed by HiH_i, so it is central in LiL_i. Moreover, ⟨δi⟩∩Hi=1\langle\delta_i\rangle\cap H_i = 1. Thus Ci=⟨z⟩×HiC_i = \langle z\rangle\times H_i embeds in both QbaseQ_{\mathrm{base}} and LiL_i, with (zk,h)(z^k,h) sent to (kδi,h)(k\delta_i,h) in the latter. Define G1G_1 to be the quotient of the free product

Qbase∗L1∗⋯∗LmQ_{\mathrm{base}} * L_1 * \cdots* L_m

that identifies these two copies of CiC_i for every ii.

For completeness, the following action proves that all the factors embed in this quotient. Let XX be a countable disjoint union of copies of the left regular QbaseQ_{\mathrm{base}}-set. Its restriction to each CiC_i is free and has countably infinitely many orbits. A countable disjoint union XiX_i of regular LiL_i-sets has the same property when restricted to CiC_i. Pair their CiC_i-orbits and choose one point in each paired orbit; extending those point pairings equivariantly gives a CiC_i-equivariant bijection X→XiX \to X_i. Transport the regular LiL_i-action to XX using this bijection. The resulting actions of Qbase,L1,…,LmQ_{\mathrm{base}}, L_1, \ldots, L_m agree on every prescribed CiC_i, hence define an action of G1G_1 on XX. Each factor acts faithfully, which proves its injectivity in G1G_1. This argument uses the assumed embeddings Hi↪G0H_i \hookrightarrow G_0 in defining the subgroups Ci≤QbaseC_i \leq Q_{\mathrm{base}}.

The element zz is central in every generating factor of G1G_1 and therefore in G1G_1; its order is still MM because QbaseQ_{\mathrm{base}} embeds. In RR, direct multiplication gives

zf=ζf,f2=1M∑k=0M−1ζ−kzkf=f.zf = \zeta f,\qquad f^2 = \frac{1}{M}\sum_{k=0}^{M-1}\zeta^{-k}z^k f = f.

The coefficient of the identity in ff is M−1≠0M^{-1} \ne0, since the elements 1,z,…,zM−11,z,\ldots,z^{M-1} are distinct. Thus ff is a nonzero central idempotent, and S=fRS=fR is a ring with identity ff. Each PiP_i is an idempotent in fK[Li]fK[L_i]. Also, any identity eiej=0e_i e_j=0 already valid in K[G0]K[G_0] remains valid after multiplication by ff in SS.

The matrix blocks of the central sector

Fix ii for the moment. A character of BiB_i means a homomorphism χ:Bi→K∗\chi:B_i \to K^*. If bxb_x denotes the coordinate function with value 11 at x∈Hix \in H_i and 00 elsewhere, then every character is uniquely specified by exponents ax∈Z/MZa_x \in\mathbb{Z}/M\mathbb{Z} through χ(bx)=ζax\chi(b_x)=\zeta^{a_x}. Define

pχ=1∣Bi∣∑b∈Biχ(b)−1bin K[Bi].p_\chi= \frac{1}{|B_i|}\sum_{b\in B_i}\chi(b)^{-1}b \quad\text{in } K[B_i].

Apply the Fourier calculation of (2) to the finite index set HiH_i. The denominator is invertible because p∤Mp \nmid M. The elements pχp_\chi are a basis of pairwise orthogonal idempotents whose sum is 11, and bpχ=χ(b)pχbp_\chi=\chi(b)p_\chi.

Use the action

(hχ)(b)=χ(h−1⋅b).(h\chi)(b)=\chi(h^{-1}\cdot b).

Then hpχh−1=phχhp_\chi h^{-1}=p_{h\chi}. Since z=δiz=\delta_i is fixed by HiH_i, the set

Xi={χ:χ(z)=ζ}\mathcal{X}_i=\{\chi:\chi(z)=\zeta\}

is HiH_i-invariant, and the Fourier formulas give

f=∑χ∈Xipχ.(11)f=\sum_{\chi\in\mathcal{X}_i}p_\chi. \tag*{(11)}

For each orbit in Xi\mathcal{X}_i, choose one representative θ\theta and put Tθ=Stab⁡Hi(θ)T_\theta=\operatorname{Stab}_{H_i}(\theta).

Lemma 5.2. For each such θ\theta, the subgroup TθT_\theta is elementary abelian of exponent pp and rank at most 33, with the trivial subgroup allowed.

Proof. Write θ(bx)=ζax\theta(b_x)=\zeta^{a_x}. Since z=∑x∈Hibxz=\sum_{x\in H_i}b_x, the condition on θ\theta is

∑x∈Hiax=1in Z/MZ.\sum_{x\in H_i}a_x=1\quad\text{in }\mathbb{Z}/M\mathbb{Z}.

Suppose that a nonidentity g∈A0∩Tθg\in A_0\cap T_\theta has order qq. The exponent of A0A_0 divides MM, so q>1q>1 divides MM. Left multiplication by gg partitions HiH_i into cycles of length exactly qq. Invariance of θ\theta says that the exponents axa_x are constant on these cycles. Their sum consequently lies in q(Z/MZ)q(\mathbb{Z}/M\mathbb{Z}). It cannot equal 1: reduction modulo qq would give 0=10=1. This contradiction proves Tθ∩A0=1T_\theta\cap A_0=1. The quotient homomorphism Hi→Hi/A0≅PH_i\to H_i/A_0\cong P is therefore injective on TθT_\theta. Every subgroup of the additive group (Fp)3(\mathbb{F}_p)^3 is an Fp\mathbb{F}_p-linear subspace, giving the assertion.

Eliminating the prime-to-pp part of each stabilizer is what makes its augmentation ideal nilpotent, and will permit the later change of basis. We next give the block isomorphism, including its inverse. This is the character-orbit construction already used in Section 3, now allowing nontrivial stabilizers; compare the Clifford-theoretic orbit picture in [8]. It will identify the underlying KK-linear rank of PiP_i in every block with the same original integer dd. For one orbit representative θ\theta, abbreviate

T=Tθ,s=[Hi:T],D=K[T].T=T_\theta,\qquad s=[H_i:T],\qquad D=K[T].

Choose representatives r1,…,rsr_1,\ldots,r_s of the left cosets hThT. The space K[Hi]K[H_i] is a free right DD-module with this ordered basis. Let g∈Hig\in H_i act on it by left multiplication LgL_g, and let b∈Bib\in B_i act diagonally by

Db(h)=(hθ)(b)h(h∈Hi).D_b(h)=(h\theta)(b)h\qquad(h\in H_i).

Both actions commute with the right action of TT. For the diagonal action this follows from (ht)θ=hθ(ht)\theta=h\theta for t∈Tt\in T. The semidirect-product relation is also respected, because

LgDbLg−1(h)=((g−1h)θ)(b)h=(hθ)(g⋅b)h=Dg⋅b(h).L_gD_bL_g^{-1}(h)=((g^{-1}h)\theta)(b)h=(h\theta)(g\cdot b)h=D_{g\cdot b}(h).

Thus we have an algebra homomorphism ρθ:K[Li]→Mat⁡s(D)\rho_\theta:K[L_i]\to\operatorname{Mat}_s(D). The element zz acts as ζ\zeta, so ρθ(f)=Is\rho_\theta(f)=I_s.

For t0∈Tt_0\in T and 1≤a,b≤s1\leq a,b\leq s, consider

praθrat0rb−1.p_{r_a\theta}r_at_0r_b^{-1}.

It belongs to fK[Li]fK[L_i] by (11). In the θ\theta-representation it sends rbt′r_bt' to rat0t′r_at_0t' for every t′∈Tt'\in T and vanishes on every other basis coset. To check the last assertion, its left group factor sends the coset rcTr_cT to rat0rb−1rcTr_at_0r_b^{-1}r_cT; the Fourier projection keeps this coset precisely when it equals raTr_aT, which is equivalent to c=bc=b. In a representation associated to another orbit, the Fourier projection vanishes identically. Consequently (5.3) maps to the matrix with entry t0t_0 in position (a,b)(a,b) of the θ\theta-block and zero elsewhere.

Taking the product of these representations gives a surjective map

ρ:fK[Li]⟶∏θMat⁡[Hi:Tθ](K[Tθ]).(12)\rho:fK[L_i]\longrightarrow\prod_\theta\operatorname{Mat}_{[H_i:T_\theta]}(K[T_\theta]). \tag*{(12)}

It is injective as well. The elements pχhp_\chi h, for χ∈Xi\chi\in\mathcal{X}_i and h∈Hih\in H_i, form a basis of its source. An orbit of size ss contributes sw=s2∣T∣sw=s^2|T| basis elements, equal to the dimension of the corresponding target block. Hence the source and target have equal finite KK-dimension. Formula (5.3), extended linearly in t0t_0 and summed over the positions and blocks, is an explicit inverse of ρ\rho.

Turning dimensions into rectangular identities

Continue in one block and let E=ρθ(Pi)∈Mat⁡s(D)E=\rho_\theta(P_i)\in\operatorname{Mat}_s(D). As a KK-linear operator on K[Hi]K[H_i], EE is exactly left multiplication by eie_i, since ff acts as the identity. Therefore

dim⁡Kim⁡E=d.\dim_K \operatorname{im} E=d.

This is the link between the original rank calculation and the block decomposition. We will now obtain a basis in which EE is a coordinate projection over DD, not just over KK. The argument is a constructive nilpotent-ideal instance of the freeness of projective modules over local rings, studied by Kaplansky [18 Theorem 2]; no general projective-module theorem is needed here.

Let ε:D→K\varepsilon:D\to K be augmentation, and set a=ker⁡ε\mathfrak{a}=\ker\varepsilon. By Lemma 5.2, we may choose generators t1,…,tqt_1,\ldots,t_q of T≅(Cp)qT\cong(C_p)^q with 0≤q≤30\le q\le3. The elements Xj=tj−1X_j=t_j-1 commute, satisfy Xjp=0X_j^p=0, and generate a\mathfrak{a}. Expanding tj=1+Xjt_j=1+X_j and counting dimensions gives

D≅K[X1,…,Xq]/(X1p,…,Xqp).D\cong K[X_1,\ldots,X_q]/(X_1^p,\ldots,X_q^p).

It follows that

aDmax⁡+1=0,Dmax⁡=3(p−1).\mathfrak{a}^{D_{\max}+1}=0,\qquad D_{\max}=3(p-1).

Indeed, a monomial of total degree greater than q(p−1)q(p-1) has an exponent at least pp. This also covers q=0q=0, when D=KD=K and a=0\mathfrak{a}=0. Matrices with entries in a\mathfrak{a} have the same nilpotence bound under multiplication, because each entry of a product of Dmax⁡+1D_{\max}+1 such matrices lies in aDmax⁡+1\mathfrak{a}^{D_{\max}+1}.

Put E‾=ε(E)\overline{E}=\varepsilon(E), applying augmentation entrywise, and let d0=rank⁡KE‾d_0=\operatorname{rank}_K\overline{E}. As E‾\overline{E} is idempotent, its image and the image of Is−E‾I_s-\overline{E} form complementary subspaces of KsK^s. Scan the columns of E‾\overline{E} in order and keep each column that increases their span; then do the same for Is−E‾I_s-\overline{E}. Build QQ by taking the corresponding d0d_0 columns of EE, followed by the corresponding s−d0s-d_0 columns of Is−EI_s-E. Its augmentation Q0Q_0 is invertible over KK. Regard Q0Q_0 as a scalar matrix over DD, and set N=Q0−1(Q−Q0)N=Q_0^{-1}(Q-Q_0). Its entries belong to a\mathfrak{a}, so NDmax⁡+1=0N^{D_{\max}+1}=0. Since Q=Q0(Is+N)Q=Q_0(I_s+N),

Q−1=(∑k=0Dmax⁡(−N)k)Q0−1.Q^{-1}=\left(\sum_{k=0}^{D_{\max}}(-N)^k\right)Q_0^{-1}.

Each of the first d0d_0 columns of QQ is fixed by EE, and every remaining column is killed by EE. Thus, for Π=diag⁡(Id0,0)\Pi=\operatorname{diag}(I_{d_0},0),

EQ=QΠ,Q−1EQ=Π.EQ=Q\Pi,\qquad Q^{-1}EQ=\Pi.

Conjugation by QQ preserves KK-dimension of images. Comparing with (5.5) gives the integer equalities

d=d0∣T∣,w=s∣T∣,wd0=sd.d=d_0|T|,\qquad w=s|T|,\qquad wd_0=sd.

In particular, these are equalities of ordinary dimensions, not equalities after reduction to KK.

We can now match coordinates exactly. Order pairs lexicographically and choose the increasing bijection between

{(μ,l):1≤μ≤d, 1≤l≤s}and{(λ,k):1≤λ≤w, 1≤k≤d0};\{(\mu,l):1\le\mu\le d,\ 1\le l\le s\}\quad\text{and}\quad\{(\lambda,k):1\le\lambda\le w,\ 1\le k\le d_0\};

the sets have equal size by (5.7). Let VV be the ww-by-dd block matrix with ss-by-ss entries that sends coordinate (μ,l)(\mu,l) to its paired coordinate (λ,k)(\lambda,k). Thus VλμV_{\lambda\mu} is the sum of standard matrix units EklstdE^{\mathrm{std}}_{kl} over all such pairings. Let WW be the dd-by-ww block matrix reversing each pairing, with WμλW_{\mu\lambda} the corresponding sum of ElkstdE_{lk}^{\mathrm{std}}. It sends all unpaired coordinates with k>d0k>d_0 to zero. Directly on these coordinates,

WV=Id⊗Is,VW=Iw⊗Π.WV=I_d\otimes I_s,\qquad VW=I_w\otimes\Pi.

Define block matrices

J(θ)=(Iw⊗Q)V,Y(θ)=W(Iw⊗Q−1).J^{(\theta)}=(I_w\otimes Q)V,\qquad Y^{(\theta)}=W(I_w\otimes Q^{-1}).

They have outer shapes w×dw\times d and d×wd\times w, respectively, with entries in Mat⁡s(D)\operatorname{Mat}_s(D). Their products are

Y(θ)J(θ)=Id⊗Is,J(θ)Y(θ)=Iw⊗E.Y^{(\theta)}J^{(\theta)}=I_d\otimes I_s,\qquad J^{(\theta)}Y^{(\theta)}=I_w\otimes E.

Finally define the (λ,μ)(\lambda,\mu) entry of JiJ_i by applying ρ−1\rho^{-1} to the tuple (Jλμ(θ))θ(J_{\lambda\mu}^{(\theta)})_\theta, and similarly define YiY_i entrywise from the Y(θ)Y^{(\theta)}. The isomorphism (12) carries ff to the tuple of identity matrices and PiP_i to the tuple of matrices EE. The last two identities therefore pull back to (5.1). This proves Proposition 5.1.

All choices in the matrix construction are over finite sets: character orbits, coset representatives, augmentation matrices, and coordinate pairings. Taking the first element or the first independent columns in fixed orders makes the prescription definite. The inverse (5.6) is a finite sum, and the inverse (5.3) is a finite group-algebra formula. Thus the matrices furnished by the proposition are finite sums in the explicitly attached finite groups.

We now apply Proposition 5.1 to the embedded groups of Proposition 4.2 and the identical idempotents of Proposition 3.1. The group G0G_0 is countable because it is finitely generated. All the ranks equal the positive integer dd from Equation (6). Hence the group G1G_1, the nonzero central idempotent ff, and the matrices Ji,YiJ_i,Y_i are now fixed by the constructions above. Corollary 3.3 also gives PiPj=0P_iP_j=0 on every edge with i<ji<j. These are precisely the inputs to the next section.

Compression with a visible nonzero defect

We now use only the graph factorization and the precise common-sector output. The first lemma is independent of the group construction. Its proof explains why ordered edge annihilations, rather than symmetric orthogonality, suffice.

Lemma 6.1 (Triangular compression). Let SS be a nonzero unital algebra over a field KK, and let m,t,w,dm,t,w,d be positive integers. Suppose Pi∈SP_i\in S are idempotents and αi(T),βj(T)∈K\alpha_i(T),\beta_j(T)\in K satisfy

Fij=∑T=1tαi(T)βj(T),Fii=1,FijPiPj=0(i<j).F_{ij}=\sum_{T=1}^{t}\alpha_i(T)\beta_j(T),\qquad F_{ii}=1,\qquad F_{ij}P_iP_j=0\quad(i<j).

Suppose there are Ji∈Mat⁡w×d(S)J_i\in\operatorname{Mat}_{w\times d}(S) and Yi∈Mat⁡d×w(S)Y_i\in\operatorname{Mat}_{d\times w}(S) with YiJi=IdY_iJ_i=I_d, JiYi=PiIwJ_iY_i=P_iI_w. Then explicit matrices

C♯∈Mat⁡wt×md(S),D♯∈Mat⁡md×wt(S)C_{\sharp}\in\operatorname{Mat}_{wt\times md}(S),\qquad D_{\sharp}\in\operatorname{Mat}_{md\times wt}(S)

satisfy D♯C♯=ImdD_{\sharp}C_{\sharp}=I_{md}.

Proof. Define C0C_0 of shape t×mt\times m and D0D_0 of shape m×tm\times t by

(C0)T,j=βj(T)Pj,(D0)i,T=αi(T)Pi.(C_0)_{T,j}=\beta_j(T)P_j,\qquad(D_0)_{i,T}=\alpha_i(T)P_i.

Their product has entries FijPiPjF_{ij}P_iP_j. Therefore D0C0=P+ND_0C_0=\mathcal{P}+\mathcal{N}, where P=diag⁡(P1,…,Pm)\mathcal{P}=\operatorname{diag}(P_1,\ldots,P_m) and N\mathcal{N} is strictly lower triangular. In particular Nm=0\mathcal{N}^m=0. Every off-diagonal entry begins with PiP_i and ends with PjP_j, so idempotence gives

P2=P,PN=N=NP.\mathcal{P}^2=\mathcal{P},\qquad\mathcal{P}\mathcal{N}=\mathcal{N}=\mathcal{N}\mathcal{P}.

This is all the commutation information needed. The finite sum

Q=P+∑k=1m−1(−N)k\mathcal{Q}=\mathcal{P}+\sum_{k=1}^{m-1}(-\mathcal{N})^k

satisfies Q(P+N)=P\mathcal{Q}(\mathcal{P}+\mathcal{N})=\mathcal{P} by cancellation of consecutive powers. Put D1=QD0D_1=\mathcal{Q}D_0; then D1C0=PD_1C_0=\mathcal{P}. For m=1m=1 the sum is empty and the same identity holds.

Take block diagonals CrepC_{\mathrm{rep}} and DrepD_{\mathrm{rep}} of ww copies of C0C_0 and D1D_1. Use ordered coordinates (λ,i)(\lambda,i) on SwmS^{wm} and (λ,T)(\lambda,T) on SwtS^{wt}. Their product is

DrepCrep=Prep:=diag⁡λ=1w(P).D_{\mathrm{rep}}C_{\mathrm{rep}}=\mathcal{P}_{\mathrm{rep}}:=\operatorname{diag}_{\lambda=1}^{w}(\mathcal{P}).

Define matrices J:Smd→Swm\mathcal{J}:S^{md}\to S^{wm} and Y:Swm→Smd\mathcal{Y}:S^{wm}\to S^{md} by

J(λ,i),(j,μ)=δij(Ji)λμ,Y(j,μ),(λ,i)=δij(Yi)μλ,1≤μ≤d.\mathcal{J}_{(\lambda,i),(j,\mu)}=\delta_{ij}(J_i)_{\lambda\mu},\qquad \mathcal{Y}_{(j,\mu),(\lambda,i)}=\delta_{ij}(Y_i)_{\mu\lambda},\qquad1\leq\mu\leq d.

The assumed rectangular identities give YJ=Imd\mathcal{Y}\mathcal{J}=I_{md} and JY=Prep\mathcal{J}\mathcal{Y}=\mathcal{P}_{\mathrm{rep}}. Thus

C♯=CrepJ,D♯=YDrepC_{\sharp}=C_{\mathrm{rep}}\mathcal{J},\qquad D_{\sharp}=\mathcal{Y}D_{\mathrm{rep}}

have the asserted shapes and satisfy

D♯C♯=YPrepJ=YJYJ=Imd.D_{\sharp}C_{\sharp}=\mathcal{Y}\mathcal{P}_{\mathrm{rep}}\mathcal{J}=\mathcal{Y}\mathcal{J}\mathcal{Y}\mathcal{J}=I_{md}.

Only associativity and the stated identities have been used; the PiP_i are not assumed to commute.

Apply the lemma to S=fK[G1]S=fK[G_1], whose identity is ff. Distinct nonedges have Fij=0F_{ij}=0 by Proposition 2.2; edges with i<ji<j have PiPj=0P_iP_j=0. The rectangular identities are exactly those of Proposition 5.1. Consequently

D♯C♯=fImd,C♯:Smd⟶Swt.(13)D_{\sharp}C_{\sharp}=fI_{md},\qquad C_{\sharp}:S^{md}\longrightarrow S^{wt}. \tag*{(13)}

The strict, ordinary rational inequalities are

tm<1100<196<dw,r:=wt<md=:N.\frac{t}{m}<\frac{1}{100}<\frac{1}{96}<\frac{d}{w},\qquad r:=wt<md=:N.

This is more than an abstract rank paradox: it gives a nonzero vector that will certify the reverse-product defect.

Lemma 6.2 (A square inverse and its certificate). Let RR be a unital ring, f≠0f\ne0 a central idempotent, and S=fRS=fR. Suppose N>r≥1N>r\geq1 and matrices C∈Mat⁡r×N(S)C\in\operatorname{Mat}_{r\times N}(S), D∈Mat⁡N×r(S)D\in\operatorname{Mat}_{N\times r}(S) satisfy DC=fINDC=fI_N. Let ι:Sr→SN\iota:S^r\to S^N include the first rr coordinates and π:SN→Sr\pi:S^N\to S^r retain them, with entries ff and zero. Then

AR=πD+(1−f)Ir,BR=Cι+(1−f)IrA_R=\pi D+(1-f)I_r,\qquad B_R=C\iota+(1-f)I_r

satisfy ARBR=IrA_RB_R=I_r and BRAR≠IrB_RA_R\ne I_r. More precisely, if u∗∈SNu_*\in S^N has last coordinate ff and all others zero, then

v∗:=Cu∗≠0,Dv∗=u∗,ARv∗=0,(BRAR−Ir)v∗=−v∗≠0.v_*:=Cu_*\ne0,\qquad Dv_*=u_*,\qquad A_Rv_*=0,\qquad(B_RA_R-I_r)v_*=-v_*\ne0.

Proof. Every entry of C,D,ι,πC,D,\iota,\pi belongs to SS and is annihilated on both sides by 1−f1-f. Since πι=fIr\pi\iota=fI_r,

ARBR=πDCι+(1−f)Ir=fIr+(1−f)Ir=Ir.A_RB_R=\pi DC\iota+(1-f)I_r=fI_r+(1-f)I_r=I_r.

The vector u∗u_* is nonzero because f≠0f\ne0. The equation Dv∗=DCu∗=u∗Dv_*=DCu_*=u_* proves v∗≠0v_*\ne0. But N>rN>r makes πu∗=0\pi u_*=0, so

ARv∗=πDv∗+(1−f)v∗=0.A_Rv_*=\pi Dv_*+(1-f)v_*=0.

Multiplying by BRB_R and subtracting v∗v_* gives the final displayed identity and proves the strict reversal defect.

Using C=C♯C=C_{\sharp}, D=D♯D=D_{\sharp} in Lemma 6.2 gives square matrices over the entire algebra K[G1]K[G_1], not only the ff-summand. The nonzero certificate is rooted in the exact order of zz, which proves f≠0f\ne0, and then in D♯v∗=u∗D_{\sharp}v_*=u_*. It does not depend on identifying a coefficient by deciding equality of group words. The remaining step is to retain this defect when replacing matrices by scalar elements.

An injective finite-group encoding

We have a strict matrix one-sided inverse over K[G1]K[G_1]. The finite-factor reduction of Dykema and Juschenko [9] explains the appropriate next step: put a full matrix algebra inside a finite group algebra and fill the complementary summand with the identity. We prove the actual matrix block needed here, including its injectivity. Its shift and diagonal operators are the finite Heisenberg model; the comparison with [14 Section 1.2] is methodological, since our cyclic order can be composite and our coefficient field is finite.

Proposition 7.1. Let KK be a field, let M≥2M\ge2 be invertible in KK, and let ζ∈K\zeta\in K have order MM. For an integer q≥0q\ge0 put Zq=(Z/MZ)qZ_q=(\mathbb{Z}/M\mathbb{Z})^q and n=Mqn=M^q. There is a finite group Λ\Lambda of order Mn2Mn^2 and a nonzero central idempotent f′∈K[Λ]f'\in K[\Lambda] for which explicitly specified elements Ωuv\Omega_{uv}, u,v∈Zqu,v\in Z_q, form a basis of f′K[Λ]f'K[\Lambda] and obey

ΩuvΩst=δvsΩut,∑uΩuu=f′.\Omega_{uv}\Omega_{st}=\delta_{vs}\Omega_{ut}, \qquad \sum_u\Omega_{uu}=f'.

Thus Mat⁡n(K)≅f′K[Λ]\operatorname{Mat}_n(K)\cong f'K[\Lambda] by its standard matrix units.

Proof. Use the set Λ=Zq×Zq×Z/MZ\Lambda=Z_q\times Z_q\times\mathbb{Z}/M\mathbb{Z} with

(x,y,k)(x′,y′,k′)=(x+x′,y+y′,k+k′+y⋅x′).(x,y,k)(x',y',k')=(x+x',y+y',k+k'+y\cdot x').

All coordinates and the dot product use residues modulo MM. Associativity follows from the bilinear identity

y⋅x′+(y+y′)⋅x′′=y′⋅x′′+y⋅(x′+x′′).y\cdot x'+(y+y')\cdot x''=y'\cdot x''+y\cdot(x'+x'').

The zero tuple is the identity, and the inverse is (−x,−y,−k+y⋅x)(-x,-y,-k+y\cdot x). No primality of MM is involved. The element z′=(0,0,1)z'=(0,0,1) is central of exact order MM. Define

f′=1M∑k=0M−1ζ−k(z′)k.f'=\frac{1}{M}\sum_{k=0}^{M-1}\zeta^{-k}(z')^k.

Reindexing gives z′f′=ζf′z'f'=\zeta f', and convolution gives (f′)2=f′(f')^2=f'. The identity coefficient M−1M^{-1} is nonzero, so f′≠0f'\ne0.

Write Xx=(x,0,0)X_x=(x,0,0) and Yy=(0,y,0)Y_y=(0,y,0), and put

Dv=1n∑y∈Zqζ−y⋅vf′Yy.(14)D_v=\frac{1}{n}\sum_{y\in Z_q}\zeta^{-y\cdot v}f'Y_y. \tag*{(14)}

The denominator is invertible because n=Mqn=M^q. Character orthogonality here means the direct identity

∑y∈Zqζy⋅a={n,a=0,0,a≠0.\sum_{y\in Z_q}\zeta^{y\cdot a}= \begin{cases} n, & a=0,\\ 0, & a\ne0. \end{cases}

When a≠0a\ne0, choose a nonzero coordinate aja_j. Translation of the sum by the jjth coordinate vector multiplies it by ζaj≠1\zeta^{a_j}\ne1. In the field KK, this forces the sum to vanish. The argument still works when aja_j is a zero divisor modulo the composite integer MM.

Convolution in the abelian group of YyY_y gives DvDt=δvtDtD_vD_t=\delta_{vt}D_t; indeed the coefficient of f′Ybf'Y_b in the product is

ζ−b⋅tn2∑y∈Zqζy⋅(t−v).\frac{\zeta^{-b\cdot t}}{n^2}\sum_{y\in Z_q}\zeta^{y\cdot(t-v)}.

Summing (14) over vv also gives ∑vDv=f′\sum_vD_v=f'. The group law gives YyXa=(z′)y⋅aXaYyY_yX_a=(z')^{y\cdot a}X_aY_y, and hence

DvXa=XaDv−a.D_vX_a=X_aD_{v-a}.

Set

Ωuv=Xu−vDv=Xu−v1n∑y∈Zqζ−y⋅vf′Yy.(15)\Omega_{uv}=X_{u-v}D_v=X_{u-v}\frac{1}{n}\sum_{y\in Z_q}\zeta^{-y\cdot v}f'Y_y. \tag*{(15)}

Then

ΩuvΩst=Xu−v+s−tDv−s+tDt=δvsΩut,\Omega_{uv}\Omega_{st}=X_{u-v+s-t}D_{v-s+t}D_t=\delta_{vs}\Omega_{ut},

and the diagonal sum is f′f'.

The multiplication identities alone do not yet prove that this block has no kernel. On KZqK^{Z_q} define

ρ(x,y,k)es=ζk+y⋅ses+x.\rho(x,y,k)\mathbf{e}_s=\zeta^{k+y\cdot s}\mathbf{e}_{s+x}.

The scalar in a composite is ζk+k′+y⋅x′+(y+y′)⋅s\zeta^{k+k'+y\cdot x'+(y+y')\cdot s}, so Equation (7.1) proves that ρ\rho is a representation. It sends f′f' to the identity and DvD_v to the coordinate projection at vv. Therefore

ρ(Ωuv)es=δvseu,\rho(\Omega_{uv})\mathbf{e}_s=\delta_{vs}\mathbf{e}_u,

which proves independence of all n2n^2 elements Ωuv\Omega_{uv}. On the other hand, the n2n^2 elements f′(x,y,0)f'(x,y,0) span f′K[Λ]f'K[\Lambda], because f′(x,y,k)=ζkf′(x,y,0)f'(x,y,k)=\zeta^k f'(x,y,0). Their group-basis supports have distinct (x,y)(x,y) coordinates, so they too are independent. The sector has dimension exactly n2n^2, and the matrix units form a basis. When q=0q=0, the empty tuple set has one element: n=1n=1, Λ=CM\Lambda=C_M, and the sole matrix unit is f′f'. All formulas remain valid. □\square

The elementary identity-completion step is worth recording separately. Its unit need not be central in the target ring.

Lemma 7.2. Let Φ:A→B\Phi:\mathcal{A}\to\mathcal{B} be an injective ring homomorphism between unital rings, not necessarily preserving their identities, and put g=Φ(1A)g=\Phi(1_{\mathcal{A}}). If AB=1AAB=1_{\mathcal{A}} and BA≠1ABA\ne1_{\mathcal{A}}, then

a=1B−g+Φ(A),b=1B−g+Φ(B)a=1_{\mathcal{B}}-g+\Phi(A),\qquad b=1_{\mathcal{B}}-g+\Phi(B)

satisfy ab=1Bab=1_{\mathcal{B}} and ba≠1Bba\ne1_{\mathcal{B}}.

Proof. The element gg is idempotent and is an identity on the image of Φ\Phi on both sides. Thus 1−g1-g annihilates that image on both sides. Expanding the two products gives

ab=1−g+Φ(AB)=1,ba−1=Φ(BA−1)≠0.ab = 1-g+\Phi(AB)=1,\qquad ba-1=\Phi(BA-1)\ne0.

Only the last inequality uses injectivity; no global centrality of gg is required.

Corollary 7.3. Under the field hypotheses of Proposition 7.1, suppose A,B∈Mat⁡r(K[L])A,B\in\operatorname{Mat}_r(K[L]) for a group LL and an integer r≥1r\ge1, with AB=IrAB=I_r and BA≠IrBA\ne I_r. Finite formulas give a finite group Λ\Lambda and a,b∈K[L×Λ]a,b\in K[L\times\Lambda] with ab=1ab=1 and ba≠1ba\ne1. In particular, over any fixed finite field with at least three elements, a failure of stable finiteness transfers to a scalar failure after a finite direct factor, without changing the field.

Proof. Choose the least q≥0q\ge0 with n=Mq≥rn=M^q\ge r. This search is finite, since Mr≥2r≥rM^r\ge2^r\ge r. Pad the matrices by identities:

A~=diag⁡(A,In−r),B~=diag⁡(B,In−r).\widetilde{A}=\operatorname{diag}(A,I_{n-r}),\qquad\widetilde{B}=\operatorname{diag}(B,I_{n-r}).

The padding is empty if n=rn=r. Then A~B~=In\widetilde{A}\widetilde{B}=I_n and B~A~−In=diag⁡(BA−Ir,0)≠0\widetilde{B}\widetilde{A}-I_n=\operatorname{diag}(BA-I_r,0)\ne0. Use Proposition 7.1 and the group-basis identification K[L×Λ]=K[L]⊗KK[Λ]K[L\times\Lambda]=K[L]\otimes_K K[\Lambda]. For an nn-square matrix TT over K[L]K[L], put

Φ(T)=∑u,v∈ZqTuv⊗Ωuv.\Phi(T)=\sum_{u,v\in Z_q}T_{uv}\otimes\Omega_{uv}.

The matrix-unit products prove multiplicativity, with Φ(In)=1⊗f′\Phi(I_n)=1\otimes f'. Independence of the Ωuv\Omega_{uv} gives injectivity: extend their coordinate dual functionals to K[Λ]K[\Lambda] and apply each in the second tensor factor to recover TuvT_{uv}. This works even when K[L]K[L] has zero divisors.

Lemma 7.2 now yields the promised finite sums

a=1⊗(1−f′)+∑u,v∈ZqA~uv⊗Ωuv,a=1\otimes(1-f')+\sum_{u,v\in Z_q}\widetilde{A}_{uv}\otimes\Omega_{uv},
b=1⊗(1−f′)+∑u,v∈ZqB~uv⊗Ωuv.(16)b=1\otimes(1-f')+\sum_{u,v\in Z_q}\widetilde{B}_{uv}\otimes\Omega_{uv}. \tag*{(16)}

Their full scalar product is one, and their reverse defect is Φ(B~A~−In)≠0\Phi(\widetilde{B}\widetilde{A}-I_n)\ne0. For a finite field of order Q≥3Q\ge3, take M=Q−1M=Q-1 and a generator of its multiplicative group. Then M≥2M\ge2 is invertible in the field, so the preceding construction applies.

Proof of Theorem 1.1. Lemma 2.1 supplies the specified odd prime and field of order p4p^4. Propositions 2.2 and 3.1 give the graph fit and the strict regular-rank comparison. Proposition 4.2 embeds the local groups in G0G_0, and Proposition 5.1 places their ranks in the nonzero ring fK[G1]fK[G_1] with the required rectangular identities. Corollary 3.3 supplies every ordered edge product. Lemmas 6.1 and 6.2 produce AR,BR∈Mat⁡r(K[G1])A_R,B_R\in\operatorname{Mat}_r(K[G_1]), with their nonzero certificate. Corollary 7.3, using L=G1L=G_1 and M=p4−1M=p^4-1, gives G=G1×ΛG=G_1\times\Lambda and the sums in (16). Their product is one and their reverse-product defect is nonzero by the injectivity of the matrix encoding.

The group G0G_0 is generated by the finite groups HiH_i. The construction of G1G_1 adds only ⟨z⟩\langle z\rangle and the finitely many finite groups LiL_i, and Λ\Lambda is finite. Thus GG is finitely generated. The embedded copy of any HiH_i supplies nontrivial torsion. All field, group and coefficient choices were given by finite formulas and fixed orders, as required.

Finite prescriptions and scope. Every choice of character, basis, field element, or finite inverse occurs in an explicitly finite set. Products of the resulting group words define finite group-algebra sums even when different words happen to coincide. No word comparisons are needed to define the elements, and the kernel certificate and injective matrix block certify the nonzero reverse defect. The countable permutation actions used to certify group embeddings are not coefficient-search inputs.

Containing torsion does not say that every element has finite order. The group must be infinite, since finite-dimensional left multiplication proves direct finiteness for every finite group algebra, in modular characteristic as well as otherwise. The coefficient field throughout remains the specified KK of order p4p^4.

More generally, the supports of any two finite sums in a group algebra generate a finitely generated subgroup. Its group algebra embeds by the group bases, preserving both a one-sided product and its nonzero reverse defect. Thus a universal direct-finiteness theorem for finitely generated groups would already imply the assertion for all groups.

The cellular automaton on all configurations

The final transfer is short and characteristic-independent. We give it locally to fix the multiplication order and the meaning of a missing configuration. Its group-ring ancestry appears in [10 Section 1] and the linear theory of [5].

For any field kk, group LL, and finite sum c=∑ucuu∈k[L]c=\sum_u c_u u\in k[L], define on all functions x:L→kx:L\to k

(Tcx)(g)=∑ucux(gu).(T_cx)(g)=\sum_u c_u x(gu).

Only the coefficient sum is finite; xx may have infinite support. Expanding the two finite sums gives

(TcTdx)(g)=∑u,vcudvx(guv)=(Tcdx)(g).(T_cT_dx)(g)=\sum_{u,v}c_u d_v x(guv)=(T_{cd}x)(g).

Thus TcTd=TcdT_cT_d=T_{cd} and T1=id⁡T_1=\operatorname{id}. If δh\delta_h is the point indicator at hh, then (Tcδh)(1L)=ch(T_c\delta_h)(1_L)=c_h, which also proves faithfulness.

Proposition 8.1. Let kk be a finite field, LL a group, and a,b∈k[L]a,b\in k[L] with ab=1ab=1. Then Tb:kL→kLT_b:k^L\to k^L is an injective cellular automaton, with

im⁡Tb={y∈kL:Tbay=y}.\operatorname{im}T_b=\{y\in k^L:T_{ba}y=y\}.

For every hh with (ba−1)h≠0(ba-1)_h\ne0, the configuration δh\delta_h is absent from its image.

Proof. The support of bb is a finite memory set with rule (xu)u↦∑ubuxu(x_u)_u\mapsto\sum_u b_u x_u. Equation (8.1) gives TaTb=Tab=id⁡T_aT_b=T_{ab}=\operatorname{id}, proving injectivity. If y=Tbxy=T_bx, then Tbay=TbTaTbx=yT_{ba}y=T_bT_aT_bx=y. Conversely, a fixed point of TbaT_{ba} has y=Tb(Tay)y=T_b(T_ay), hence lies in the image. Finally,

(Tbaδh−δh)(1L)=(ba−1)h≠0,(T_{ba}\delta_h-\delta_h)(1_L)=(ba-1)_h\ne0,

so δh\delta_h is not a fixed point and cannot be in the image.

Apply the proposition to the scalar sums in Equation (16). Their nonzero defect has a nonzero coefficient in the group basis, so the resulting automaton is injective and nonsurjective. The missing point indicator has no preimage even among infinitely supported configurations. If the finite formula for bb displays a group element more than once, its terms simply add in the same local rule; specifying the rule requires no word-problem oracle. Thus the constructed group also gives a negative answer to Gottschalk’s surjunctivity conjecture.

References

  1. [1]Noga Alon, László Babai, and Hiroshi Suzuki. Multilinear polynomials and Frankl–Ray-Chaudhuri–Wilson type intersection theorems. Journal of Combinatorial Theory. Series A, 58(2):165–180, 1991.DOI
  2. [2]Pere Ara, Kevin C. O’Meara, and Francesc Perera. Stable finiteness of group rings in arbitrary characteristic. Advances in Mathematics, 170(2):224–238, 2002.DOI
  3. [3]Alexander Barg, Moshe Schwartz, and Lev Yohananov. Storage codes on coset graphs with asymptotically unit rate. Combinatorica, 44(6):1193–1209, 2024.arxiv.org/abs/2212.12117
  4. [4]Alexander Barg and Gilles Zémor. High-rate storage codes on triangle-free graphs. IEEE Transactions on Information Theory, 68(12):7787–7797, 2022.DOI
  5. [5]Tullio Ceccherini-Silberstein and Michel Coornaert. Linear cellular automata: Garden of Eden Theorem, L-surjunctivity and group rings. Algebra and Discrete Mathematics, pages 22–35, 2006. no. 2.
  6. [6]Tullio Ceccherini-Silberstein, Michel Coornaert, and Xuan Kien Phung. First-order model theory and Kaplansky’s stable finiteness conjecture for surjunctive groups. Groups, Geometry, and Dynamics, 19:495–503, 2025.DOI
  7. [7]Tullio Ceccherini-Silberstein, Michel Coornaert, and Xuan Kien Phung. Stable finiteness of monoid algebras and surjunctivity. Theoretical Computer Science, 1042:115228, 2025.DOI
  8. [8]A. H. Clifford. Representations induced in an invariant subgroup. Proceedings of the National Academy of Sciences of the United States of America, 23(2):89–90, 1937.DOI
  9. [9]Ken Dykema and Kate Juschenko. On stable finiteness of group rings. Algebra and Discrete Mathematics, 19(1):44–47, 2015. https://admjournal.luguniv.edu.ua/index.php/adm/article/download/1174/663.admjournal.luguniv.edu.ua/index.php/adm/article/download/1174/663
  10. [10]Gábor Elek and Endre Szabó. Sofic groups and direct finiteness. Journal of Algebra, 280(2):426–434, 2004.DOI
  11. [11]Alexander Golovnev and Ishay Haviv. The (generalized) orthogonality dimension of (generalized) Kneser graphs: Bounds and applications. Theory of Computing, 18(22):1–22, 2022.arxiv.org/abs/2002.08580
  12. [12]Walter Gottschalk. Some general dynamical notions. In Anatole Beck, editor, Recent Advances in Topological Dynamics, volume 318 of Lecture Notes in Mathematics, pages 120–125. Springer, Berlin, Heidelberg, 1973.DOI
  13. [13]Mikhail Gromov. Endomorphisms of symbolic algebraic varieties. Journal of the European Mathematical Society, 1(2):109–197, 1999.DOI
  14. [14]Shamgar Gurevich and Ronny Hadani. On the diagonalization of the discrete Fourier transform. Applied and Computational Harmonic Analysis, 27:87–99, 2009.arxiv.org/abs/0808.3281
  15. [15]Ishay Haviv. On minrank and the Lovász theta function. In Approximation, Randomization, and Combinatorial Optimization. Algorithms and Techniques (APPROX/RANDOM 2018), volume 116 of Leibniz International Proceedings in Informatics, pages 13:1–13:15. Schloss Dagstuhl – Leibniz-Zentrum für Informatik, 2018.arxiv.org/abs/1802.03920
  16. [16]Hexiang Huang and Qing Xiang. Construction of storage codes of rate approaching one on triangle-free graphs. Designs, Codes and Cryptography, 91:3901–3913, 2023.DOI
  17. [17]S. A. Jennings. The structure of the group ring of a p-group over a modular field. Transactions of the American Mathematical Society, 50(1):175–185, 1941.DOI
  18. [18]Irving Kaplansky. Projective modules. Annals of Mathematics, 68(2):372–377, 1958.DOI
  19. [19]Irving Kaplansky. Fields and Rings. Chicago Lectures in Mathematics. University of Chicago Press, Chicago, second edition, 1972.
  20. [20]OpenAI. A Counterexample to Kaplansky’s Direct-Finiteness Conjecture in Characteristic Two. OpenAI Math Release preprint OAI:A-Counterexample-to-Kaplanskys-Direct-Finiteness-Conjecture-in-Characteristic-Two-September-23-2026, 2026.github.com/openai/math/blob/main/preprints/A-Counterexample-to-Kaplanskys-Direct-Finiteness-Conjecture-in-Characteristic-Two-September-23-2026/paper.pdf
  21. [21]Xuan Kien Phung. A geometric generalization of Kaplansky’s direct finiteness conjecture. Proceedings of the American Mathematical Society, 151(7):2863–2871, 2023.arxiv.org/abs/2111.07930
  22. [22]Tomasz Prytuła. Graphical complexes of groups. Journal of Group Theory, 27(2):223–254, 2024.DOI
  23. [23]J. T. Schwartz. Fast probabilistic algorithms for verification of polynomial identities. Journal of the ACM, 27(4):701–717, 1980.DOI
  24. [24]Benjamin Weiss. Sofic groups and dynamical systems. Sankhyā: The Indian Journal of Statistics, Series A, 62(3):350–359, 2000.
  25. [25]Richard Zippel. Probabilistic algorithms for sparse polynomials. In Edward W. Ng, editor, Symbolic and Algebraic Computation, volume 72 of Lecture Notes in Computer Science, pages 216–226. Springer, Berlin, Heidelberg, 1979.DOI

Paper details

vlnc:W_a4aym8t4
Version
v1 · latest
Published
License
LicenseRef-OpenAI-public
Usage

views · 25

downloads · 6

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