betaFeedback

WorksMathematics

Unsupported Cyclotomic Divisors in Three-Prime Integer Tilings

Abstract

Cyclotomic divisibility imposes strong prime-power structure on integer tiles. We study unsupported cyclotomic divisors: mixed-order divisors for which none of the prime-power components of the order divides the mask, although every prime in the order divides the tile cardinality. Kiss, Łaba, Marshall and Somlai asked whether such a phenomenon can occur in the three-prime setting. We prove that unsupported cyclotomic divisors already occur for periods with three distinct prime factors. For primes p<q<rp<q<r, we characterize the square-period case: an unsupported factor ΦpqrΦ_{pqr} occurs in a tiling of \ZZ_{(pqr)^2} if and only if r∈⟨p,q⟩r\in\langle p,q\rangle, and every such tile lies in a single residue class modulo rr. Among cyclic tilings with the unsupported order dividing the specified modulus, the smallest modulus is 180180; if the order has three distinct prime factors, it is 900900. An Apéry-set construction gives examples for every triple at period p2q2r3p^2q^2r^3.The proof of our results combines Fourier rigidity, a three-cylinder decomposition, and an integer mass obstruction.

2020 Mathematics Subject Classification. Primary 11B75; Secondary 11R18, 52C22.

Introduction and main results

The cyclotomic zeros of an integer tile describe its distribution across residue classes. The Coven–Meyerowitz conditions link zeros at prime-power orders to zeros at their mixed products. A converse implication would impose additional rigidity: a mixed-order zero, whose prime divisors all divide the tile cardinality, would require a zero at one of its prime-power components. We study the failure of this implication for periods with three distinct prime factors, and classify the structure forced when each prime occurs to exponent two.

For a finite set A⊆Z≥0A\subseteq\mathbb{Z}_{\geq0}, write

A(X)=∑a∈AXa,[n]X=1+X+⋯+Xn−1.A(X)=\sum_{a\in A}X^a,\qquad[n]_X=1+X+\cdots+X^{n-1}.

We use the same notation for the mask of a subset of ZM\mathbb{Z}_M, always taking its representatives in {0,…,M−1}\{0,\ldots,M-1\}. Let Φs\Phi_s be the ssth cyclotomic polynomial and put

SA={pα:p prime, α≥1, Φpα∣A(X)}.S_A=\{p^\alpha:p\text{ prime},\ \alpha\geq1,\ \Phi_{p^\alpha}\mid A(X)\}.

The Coven–Meyerowitz conditions are

(T1)∣A∣=∏s∈SAΦs(1),\text{(T1)}\quad|A|=\prod_{s\in S_A}\Phi_s(1),
(T2)Φs1⋯sj∣A(X)whenever s1,…,sj∈SA are powers of distinct primes.\text{(T2)}\quad\Phi_{s_1\cdots s_j}\mid A(X)\quad\text{whenever }s_1,\ldots,s_j\in S_A\text{ are powers of distinct primes.}

Coven and Meyerowitz proved that (T1) and (T2) are sufficient for tiling the integers, that (T1) is necessary, and that (T2) is necessary when the tile cardinality has at most two distinct prime factors [2].

Following Kiss, Łaba, Marshall and Somlai [3], a divisor Φs∣A\Phi_s \mid A, with s>1s > 1, is unsupported if every prime dividing ss divides ∣A∣|A|, but

Φpα∤Afor every pα∥s.\Phi_{p^\alpha} \nmid A \qquad\text{for every } p^\alpha\parallel s.

When AA is considered in a specified cyclic group ZM\mathbb{Z}_M, all unsupported orders discussed below are required to divide MM. This restriction makes the relevant cyclotomic divisibility independent of the choice of representatives. The signature SAS_A remains the full prime-power signature of the ordinary mask polynomial. In [3], the authors ask whether a set satisfying (T1) and (T2) can have an unsupported divisor. They give a four-prime construction and leave the three-prime case undecided in Section 10.2. Their Theorem 1.12 excludes such divisors when the period has at most two distinct prime factors.

We answer the three-prime existence question and determine exactly when the full three-prime order can occur at square period. Write

N=pqr,M=N2.N=pqr,\qquad M=N^2.

where p<q<rp<q<r are distinct primes, and define

E(p,q,r)={A⊂ZM:∣A∣=N, Φd2∣A (1<d∣N), ΦN∣A}.(1)\mathcal{E}(p,q,r)=\{A\subset\mathbb{Z}_M:|A|=N,\ \Phi_{d^2}\mid A\ (1<d\mid N),\ \Phi_N\mid A\}. \tag*{(1)}

The definition is symmetric in p,q,rp,q,r; we retain the same notation when discussing a permutation of the three primes. The three squared prime-power factors exhaust the cardinality at X=1X=1. Thus every member has SA={p2,q2,r2}S_A=\{p^2,q^2,r^2\}, satisfies (T1), (T2), and has unsupported ΦN\Phi_N. Section 3 gives a direct tiling complement.

A set A⊂ZMA\subset\mathbb{Z}_M is confined modulo ss, for a prime s∣Ms\mid M, if it is contained in one residue class modulo ss. This is different from having a nonzero translation period.

Theorem 1.1 (Square-period rigidity and existence). Let p<q<rp<q<r be distinct primes. Every member of E(p,q,r)\mathcal{E}(p,q,r) is confined modulo rr. Moreover,

E(p,q,r)≠∅⟺r∈⟨p,q⟩:={ap+bq:a,b∈Z≥0}.(2)\mathcal{E}(p,q,r)\ne\varnothing\quad\Longleftrightarrow\quad r\in\langle p,q\rangle:=\{ap+bq:a,b\in\mathbb{Z}_{\ge0}\}. \tag*{(2)}

The finite-group formulation does not impose an additional hypothesis on the intended tiling problem. The balanced square-period theorem of Łaba and Londner was proved for odd primes in [6] and for the even case in [7]; the combined statement is also recorded in [8]. Together with prime-power allocation, it gives the following consequence. The reduction is recorded in Section 2.

Corollary 1.2 (Cyclic tilings). There exists a tiling A⊕B=Z(pqr)2A\oplus B=\mathbb{Z}_{(pqr)^2} for which Φpqr\Phi_{pqr} is unsupported for AA if and only if r∈⟨p,q⟩r\in\langle p,q\rangle. In every such tiling,

∣A∣=∣B∣=pqr,SA={p2,q2,r2},SB={p,q,r},|A|=|B|=pqr,\qquad S_A=\{p^2,q^2,r^2\},\qquad S_B=\{p,q,r\},

both factors satisfy (T1) and (T2), and AA is confined modulo rr.

For example, no such tiling exists in Z11025\mathbb{Z}_{11025}, since 7∉⟨3,5⟩7\notin\langle3,5\rangle. The confinement conclusion has an exact form. Define

gcd⁡(M,A−A)=gcd⁡(M,{a−a′:a,a′∈A}),Stab⁡(A)={t∈ZM:A+t=A}.\gcd(M,A-A)=\gcd(M,\{a-a':a,a'\in A\}),\qquad\operatorname{Stab}(A)=\{t\in\mathbb{Z}_M:A+t=A\}.

A set is primitive if the first quantity is 1, and aperiodic if the second set is {0}\{0\}. We prove in Corollary 6.4 and Theorem 6.5 that every A∈E(p,q,r)A \in\mathcal{E}(p,q,r) satisfies

gcd⁡(M,A−A)=r,∣Stab⁡(A)∣∈{1,p,q,pq}.\gcd(M,A-A)=r,\qquad\lvert\operatorname{Stab}(A)\rvert\in\{1,p,q,pq\}.

Each of the four stabilizer orders occurs whenever r∈⟨p,q⟩r \in\langle p,q\rangle, with an aperiodic complement. After translation and division by rr, every such AA becomes a primitive tile of Zp2q2r\mathbb{Z}_{p^2q^2r} with unsupported Φpq\Phi_{pq}.

Existence at a larger period holds for every three-prime set. The construction uses the Apéry mask

[L]XΦpq(X),L=rm∈⟨p,q⟩.[L]_X\Phi_{pq}(X),\qquad L=r^m\in\langle p,q\rangle.

The precise 0–1 criterion follows from the standard numerical-semigroup identity discussed by Moree [10]. An explicit higher-scale factor completes (T1) and (T2) without restoring the absent low prime-power zeros. Taking m≤2m\le2 gives period p2q2rm+1p^2q^2r^{m+1}. Theorem 7.4 classifies every complement of a more general multiscale version and proves that each satisfies (T2).

The least cyclic period for any unsupported order is 180; for an unsupported order containing three distinct primes it is 900. Both bounds are attained with both factors aperiodic. The complete layer description at modulus 900 yields 1 834 800 distinct subsets with unsupported Φ30\Phi_{30}, including 874 800 aperiodic ones. The enumeration is given separately in Appendix A.

Several ingredients have established antecedents. The Apéry identity is classical [10]. The five-point core at (2,3,5)(2,3,5) realizes the CRT configuration in [3]; that core alone is not an unsupported example, since its cardinality is coprime to 6. The higher-scale lift below changes this cardinality constraint while preserving the mixed zero and completing (T1), (T2). Likewise, the spectrum supplied by (T1), (T2) is due to Łaba [4], and standard tiling complements are part of the Coven–Meyerowitz construction; see also [5]. Our direct arguments specialize these constructions to the coordinates used below.

Vanishing sums of roots of unity provide another relevant comparison. Lam and Leung [9] characterize the possible weights of such sums by the additive semigroup generated by the prime divisors of their order. At order and weight pqrpqr, the weight criterion supplies no obstruction. The square-signature conditions give the support capacities needed for our rigidity theorem. The same semigroup threshold also has an elementary mass interpretation, recorded in Remark 6.2.

Outline of the proof

The seven square-order zeros make an N×NN\times N Fourier matrix unitary. Column orthogonality gives one point at each high-digit triple and an exact separation condition on the low digits. The equivalent pairwise difference criterion for integer-box tilings is given in [1]; here we derive it by Fourier orthogonality and express it in high and low digits.

If no low-digit function is constant, the separation condition forces three disjoint cylinder supports. Vanishing at order pqrpqr then gives positive integers n,m,kn,m,k satisfying p2n+q2m+r2k=pqrp^2n+q^2m+r^2k=pqr, together with three support-capacity inequalities. A three-rectangle inequality rules these out for p≥5p\ge5, and short integer arguments treat p=2,3p=2,3. Hence one low-digit function is constant. The remaining character sum is an additive integer matrix, whose row and column divisibilities yield r∈⟨p,q⟩r\in\langle p,q\rangle and exclude confinement at either smaller prime. The Apéry construction gives the converse. The complement classification and the enumeration are subsequent applications.

Polynomial and tiling preliminaries

Throughout, A⊕B=ZMA\oplus B=\mathbb{Z}_M means that every residue has exactly one representation as a+ba+b. Equivalently,

A(X)B(X)≡[M]X(modXM−1).(3)A(X)B(X)\equiv[M]_X\pmod{X^M-1}. \tag*{(3)}

All classifications below refer to this specified cyclic group and its divisors.

We recall the prime-power allocation argument of Coven and Meyerowitz [2]; see also [3]. Its short proof is included to specify the levels relative to the chosen period.

Lemma 2.1 (Prime-power allocation in a cyclic tiling). If A⊕B=ZMA \oplus B = \mathbb{Z}_M, then for every prime p∣Mp \mid M, the factors Φpj\Phi_{p^j}, 1≤j≤vp(M)1 \le j \le v_p(M), are partitioned between AA and BB, and exactly vp(∣A∣)v_p(\lvert A\rvert) of them divide AA. No additional pp-power cyclotomic factor outside these levels divides either mask.

Proof. Every indicated factor divides at least one mask by (3). Distinct monic cyclotomic factors multiply in Z[X]\mathbb{Z}[X], and each of these has value pp at 1. Hence their total allocation, counting any factor assigned twice, is at most vp(∣A∣)+vp(∣B∣)=vp(M)v_p(\lvert A\rvert) + v_p(\lvert B\rvert) = v_p(M). Since all vp(M)v_p(M) levels must be allocated, equality holds, with no repetition or extra level. A prime not dividing MM cannot occur as a prime-power factor of either mask because it would divide its cardinality.

Reduction in Corollary 1.2. Suppose A⊕B=ZN2A \oplus B = \mathbb{Z}_{N^2} and ΦN\Phi_N is unsupported for AA. Then N∣∣A∣N \mid\lvert A\rvert. Each of Φp\Phi_p, Φq\Phi_q, Φr\Phi_r divides BB, so N∣∣B∣N \mid\lvert B\rvert as well. Cardinalities give ∣A∣=∣B∣=N\lvert A\rvert= \lvert B\rvert= N, and Lemma 2.1 assigns precisely the squared levels to AA and the first levels to BB. The odd and even balanced square-period theorems [6, 7], in the combined form [8], give (T2) for both. Hence A∈E(p,q,r)A \in\mathcal{E}(p,q,r), to which Theorem 1.1 applies. Conversely, the square-signature normal form in Theorem 3.1 supplies a cyclic tiling for every member of E(p,q,r)\mathcal{E}(p,q,r).

Numerical semigroups and the Apéry mask

Let p,qp,q be distinct primes, Q=pqQ = pq, and S=⟨p,q⟩S = \langle p,q\rangle. The standard semigroup identity is

ΦQ(X)1−X=1−Xpq(1−Xp)(1−Xq)=∑n∈SXn.(4)\frac{\Phi_Q(X)}{1-X} = \frac{1-X^{pq}}{(1-X^p)(1-X^q)} = \sum_{n\in S} X^n. \tag*{(4)}

For completeness, every element of SS has a unique expression ap+bqap + bq with a≥0a \ge0 and 0≤b<p0 \le b < p: reduce bb modulo pp, transferring multiples of pqpq into apap. Summing these expressions proves the identity. Its connection with binary cyclotomic polynomials, and the Apéry formulation, are classical; see [10].

Lemma 2.2 (Apéry smoothing). For every integer L≥1L \ge1, the polynomial

CL(X)=[L]XΦpq(X)C_L(X) = [L]X\Phi_{pq}(X)

has all coefficients in {0,1}\{0,1\} if and only if L∈SL \in S. When this holds,

supp⁡CL=Ap⁡(S;L):={s∈S:s−L∉S}.\operatorname{supp} C_L = \operatorname{Ap}(S;L) := \{s \in S : s-L \notin S\}.

This support has exactly LL elements and is a complete residue system modulo LL. Its degree is

L−1+(p−1)(q−1)<QL.L - 1 + (p - 1)(q - 1) < QL.

Proof. By (4), its coefficient at XnX^n is

1S(n)−1S(n−L),\mathbf{1}_S(n) - \mathbf{1}_S(n-L),

where membership at negative integers means zero. If L∉SL \notin S, the coefficient at n=Ln=L is −1-1. If L∈SL \in S, closure of SS under addition prevents negative coefficients. The displayed support is then immediate. Each residue class modulo LL has a least member in SS, and all its later members are obtained by adding LL. Hence the least member is the unique member of the Apéry set in that class. The degree assertion follows from deg⁡Φpq=(p−1)(q−1)\deg\Phi_{pq} = (p-1)(q-1) and (p−1)(q−1)<Q(p-1)(q-1) < Q.

Remark 2.3 (General coprime generators). For general coprime integers a,b>1a,b > 1, the right polynomial is

P⟨a,b⟩(X)=(1−X)(1−Xab)(1−Xa)(1−Xb),P_{\langle a,b\rangle}(X) = \frac{(1-X)(1-X^{ab})}{(1-X^a)(1-X^b)},

not necessarily Φab\Phi_{ab}. The same smoothing criterion applies to [L]P⟨a,b⟩[L]P_{\langle a,b\rangle}. The assertion with Φab\Phi_{ab} alone is false for arbitrary coprime integers: 2∈⟨2,9⟩2 \in\langle2,9\rangle, but

[2]XΦ18(X)=1+X−X3−X4+X6+X7.[2]X\Phi_{18}(X) = 1 + X - X^3 - X^4 + X^6 + X^7.

Fourier normalization at square period

Let N=pqrN=pqr and M=N2M=N^2, with distinct primes p<q<rp<q<r. We first impose the seven square-order conditions in (1), without imposing the low-order condition ΦN∣A\Phi_N \mid A. The resulting normal form is necessary and sufficient. The spectrum used in the matrix argument is the square-signature specialization of [4]. The conclusion about the standard complement is also a specialization of the standard-complement criterion [5], and the pairwise difference condition is the specialization of [1] to exponents two. We give both directions explicitly to retain the precise low-digit condition (K).

Use the standard CRT identification

ZM≃Zp2×Zq2×Zr2.\mathbb{Z}_M \simeq\mathbb{Z}_{p^2} \times\mathbb{Z}_{q^2} \times\mathbb{Z}_{r^2}.

In each coordinate, separate the high digit from the low digit, and put I={0,…,p−1}×{0,…,q−1}×{0,…,r−1}I=\{0,\ldots,p-1\}\times\{0,\ldots,q-1\}\times\{0,\ldots,r-1\}.

Theorem 3.1 (Square-signature normal form). A set of cardinality NN has all the square-signature zeros in (1) if and only if it can be written uniquely as

A={(pu+f(v,w),qv+g(u,w),rw+h(u,v)):(u,v,w)∈I},(5)A=\{(pu+f(v,w),qv+g(u,w),rw+h(u,v)):(u,v,w)\in I\}, \tag*{(5)}

where the low-digit functions take values in {0,…,p−1}\{0,\ldots,p-1\}, {0,…,q−1}\{0,\ldots,q-1\}, {0,…,r−1}\{0,\ldots,r-1\}, respectively, and satisfy the following condition:

(K) For every two distinct high-digit triples, at least one coordinate has different high digits and equal low digits.

Every such set tiles with the standard complement

B□={0,…,p−1}×{0,…,q−1}×{0,…,r−1}.(6)B_{\square}=\{0,\ldots,p-1\}\times\{0,\ldots,q-1\}\times\{0,\ldots,r-1\}. \tag*{(6)}

In addition, A∈E(p,q,r)A\in\mathcal{E}(p,q,r) if and only if

∑u,v,wζpf(v,w)ζqg(u,w)ζrh(u,v)=0,ζs=e2πi/s.(7)\sum_{u,v,w}\zeta_p^{f(v,w)}\zeta_q^{g(u,w)}\zeta_r^{h(u,v)}=0,\qquad\zeta_s=e^{2\pi i/s}. \tag*{(7)}

Proof. Form the N×NN\times N matrix with columns indexed by a∈Aa\in A and rows indexed by ω∈{0,…,p−1}×{0,…,q−1}×{0,…,r−1}\omega\in\{0,\ldots,p-1\}\times\{0,\ldots,q-1\}\times\{0,\ldots,r-1\}:

Uω,a=N−1/2exp⁡(2πi(ω1a1p2+ω2a2q2+ω3a3r2)).U_{\omega,a}=N^{-1/2}\exp\left(2\pi i\left(\frac{\omega_1a_1}{p^2}+\frac{\omega_2a_2}{q^2}+\frac{\omega_3a_3}{r^2}\right)\right).

For two distinct rows, each nonzero frequency difference in a coordinate is coprime to that coordinate’s prime. Their inner product therefore evaluates AA at a character whose order is one of the required products of squared primes. Hence the rows are orthonormal and, since UU is square, so are the columns. For two columns, their inner product factors into three geometric sums. For a prime ss,

∑j=0s−1e2πijΔ/s2=0⟺s∣Δ and s2∤Δ.\sum_{j=0}^{s-1} e^{2\pi i j\Delta/s^{2}}=0 \quad\Longleftrightarrow\quad s\mid\Delta\ \text{and}\ s^{2}\nmid\Delta.

Thus any two distinct points have, in some coordinate, a difference divisible by its prime but not by its square. At most one point can have any given triple of high digits, since two such points would have all coordinate differences of absolute value less than the respective primes. As there are NN points, every high-digit triple occurs once. In these coordinates the difference condition is exactly (K). Comparing two triples differing only in uu forces their first low digits to agree. This makes the first low digit independent of uu, and similarly for the other coordinates. This gives (5).

Conversely, (K) makes the columns orthonormal, hence the rows. Taking frequency differences equal to 1 in any chosen nonempty coordinate subset gives a zero of the corresponding square order. Cyclotomic irreducibility gives the full divisibility statement.

To prove tiling with B□B_{\square}, a nonzero difference of two points of AA has some coordinate divisible exactly once by its prime. A difference of two points of B□B_{\square} cannot have that property. Hence the two difference sets intersect only in zero, and their cardinality product is MM. Finally, the character in (7) has order NN, giving the last equivalence.

Lemma 3.2 (Two-dimensional consequence). For a set parametrized by (pu+f(v),qv+g(u))(pu+f(v), qv+g(u)) on a complete p×qp\times q high-digit grid, condition (K) holds if and only if ff or gg is constant.

Proof. If both are nonconstant, choose v≠v′v\ne v' with f(v)≠f(v′)f(v)\ne f(v') and u≠u′u\ne u' with g(u)≠g(u′)g(u)\ne g(u'); the resulting pair violates (K). If one is constant, a pair differing in its corresponding high coordinate is separated there; a pair differing only in the other high coordinate is separated in that other coordinate.

The three-cylinder alternative

The normal form separates the high digits from the low digits, but condition (K) still couples the three low functions. The next lemma describes that coupling when no low function is constant. Its proof is combinatorial and remains valid for arbitrary coordinate sizes at least two.

Lemma 4.1 (Three-cylinder alternative). Suppose that the functions in (5) satisfy (K), and that none of f,g,hf,g,h is constant. After subtracting constants from the low values in their respective residue groups, there are nonempty proper subsets

U0⊂{0,…,p−1},V0⊂{0,…,q−1}U_0\subset\{0,\ldots,p-1\},\qquad V_0\subset\{0,\ldots,q-1\}

and nonempty disjoint subsets Wf,Wg⊂{0,…,r−1}W_f,W_g\subset\{0,\ldots,r-1\} such that

supp⁡f⊂V0×Wf,supp⁡g⊂U0×Wg,supp⁡h⊂U1×V1,U1={0,…,p−1}∖U0,V1={0,…,q−1}∖V0.(8)\begin{aligned} \operatorname{supp} f&\subset V_0\times W_f,\\ \operatorname{supp} g&\subset U_0\times W_g,\\ \operatorname{supp} h&\subset U_1\times V_1,\qquad U_1=\{0,\ldots,p-1\}\setminus U_0,\quad V_1=\{0,\ldots,q-1\}\setminus V_0. \tag*{(8)} \end{aligned}

In particular, at every high-digit triple, at most one of the three low values is nonzero. Conversely, any functions with the support restrictions (8) satisfy (K).

Proof. View hh as a p×qp\times q matrix. A nonconstant row h(u,⋅)h(u,\mathord\cdot) forces g(u,w)g(u,w) to be independent of ww, by Lemma 3.2 applied with uu fixed. A nonconstant column h(⋅,v)h(\mathord\cdot,v) similarly forces f(v,w)f(v,w) to be independent of ww.

We first show that hh has both a constant row and a constant column. If it has no constant row, then g(u,w)=gug(u,w)=g_u. Since gg is not globally constant, Lemma 3.2 on every fixed-ww layer gives f(v,w)=fwf(v,w)=f_w. A nonconstant column of hh would now make fwf_w constant, contrary to the hypothesis. Therefore all columns of hh are constant, so h(u,v)=hvh(u,v)=h_v. Each of fw,gu,hvf_w,g_u,h_v is nonconstant. Choose u≠u′u\ne u', v≠v′v\ne v', and w≠w′w\ne w' on which the corresponding functions have different values. The two triples (u,v,w)(u,v,w) and (u′,v′,w′)(u',v',w') then have different low values in all three coordinates, contradicting (K). Interchanging the first two coordinates proves that hh cannot have no constant column either.

Let U0U_0 be the set of constant rows and V0V_0 the set of constant columns. They are nonempty and proper: if, for example, every row were constant, one constant column would make the whole matrix constant. Their common value is some γ\gamma, because a constant row and a constant column meet. Thus h=γh=\gamma whenever u∈U0u\in U_0 or v∈V0v\in V_0.

For u∈U1u\in U_1, write g(u,w)=gug(u,w)=g_u, and for v∈V1v\in V_1, write f(v,w)=fvf(v,w)=f_v. If the values fvf_v on V1V_1 were not all equal, then every fixed-ww layer would have a nonconstant ff-vector, forcing its gg-vector to be constant. Its value is fixed at any u∈U1u\in U_1, so gg would be globally constant. Hence fv=αf_v=\alpha on V1V_1. The same argument gives gu=βg_u=\beta on U1U_1.

On each fixed-ww layer, at least one of the two low vectors is constant. If the ff-vector is constant, its value is necessarily α\alpha; if the gg-vector is constant, its value is β\beta. Consequently

Wf={w:some f(v,w)≠α},Wg={w:some g(u,w)≠β}W_f=\{w:\text{some }f(v,w)\ne\alpha\},\qquad W_g=\{w:\text{some }g(u,w)\ne\beta\}

are nonempty and disjoint. Subtracting α,β,γ\alpha,\beta,\gamma from the low values gives (8). This normalization preserves equality of low values and multiplies (7) by a nonzero constant; it is used only for these two purposes.

For the converse, label a point by FF, GG, or HH if its nonzero low coordinate is the first, second, or third, and by OO if all are zero. Two points of types F,GF,G have distinct high ww-digits and equal third low digits; types F,HF,H are separated in the second coordinate; types G,HG,H are separated in the first. For two points of type FF, different (v,w)(v,w) are separated in the second or third coordinate; if (v,w)(v,w) agrees, their first low values agree and their high uu-digits differ. The other equal-type cases are identical. A point of type FF and a point of type OO cannot have the same (v,w)(v,w), and are again separated in the second or third coordinate. The remaining cases follow by permutation, or are immediate for two type-OO points. Thus (K) holds. □

The support description reduces the vanishing condition to an integer mass identity and three capacity bounds.

Lemma 4.2 (Axis multiplicities). Under the hypotheses of Lemma 4.1, use the normalized functions and the subsets in (8). Suppose moreover that (7) holds. Set

a=∣U0∣,b=∣V0∣,c=∣Wf∣,d=∣Wg∣.a=|U_0|,\qquad b=|V_0|,\qquad c=|W_f|,\qquad d=|W_g|.

There are positive integers n,m,kn,m,k such that

p2n+q2m+r2k=pqr,(9)p^2n+q^2m+r^2k=pqr, \tag*{(9)}

and

1≤a≤p−1,1≤b≤q−1,c,d≥1,c+d≤r,1\le a\le p-1,\qquad1\le b\le q-1,\qquad c,d\ge1,\qquad c+d\le r,
(p−1)n≤bc,(q−1)m≤ad,(r−1)k≤(p−a)(q−b).(10)(p-1)n\le bc,\qquad(q-1)m\le ad,\qquad(r-1)k\le(p-a)(q-b). \tag*{(10)}

Proof. For 1≤i<p1\le i<p, let nin_i count the pairs (v,w)(v,w) with f(v,w)=if(v,w)=i; define mjm_j and kℓk_\ell similarly. A nonzero first low value occurs at pnipn_i points, because the high digit uu is free. The corresponding counts on the other axes are qmjqm_j and rkℓrk_\ell. All remaining points have zero low values. Thus the low-character sum is

C+p∑i=1p−1niζpi+q∑j=1q−1mjζqj+r∑ℓ=1r−1kℓζrℓ(11)C+p\sum_{i=1}^{p-1}n_i\zeta_p^i+q\sum_{j=1}^{q-1}m_j\zeta_q^j+r\sum_{\ell=1}^{r-1}k_\ell\zeta_r^\ell \tag*{(11)}

where

C=pqr−p∑ini−q∑jmj−r∑ℓkℓ.C = pqr - p\sum_i n_i - q\sum_j m_j - r\sum_\ell k_\ell.

The products ζpiζqjζrℓ\zeta_p^i\zeta_q^j\zeta_r^\ell, with 0≤i<p−10 \le i < p - 1, 0≤j<q−10 \le j < q - 1, and 0≤ℓ<r−10 \le\ell< r - 1, form a basis of Q(ζpqr)\mathbb{Q}(\zeta_{pqr}). Indeed, the three roots generate this field, their individual minimal polynomials reduce every exponent to the indicated range, and the number of spanning monomials is (p−1)(q−1)(r−1)=[Q(ζpqr):Q](p-1)(q-1)(r-1) = [\mathbb{Q}(\zeta_{pqr}):\mathbb{Q}]. Eliminate the last power in each sum in (11). The coefficient of each nonconstant single-coordinate basis vector shows that

n1=⋯=np−1=n,m1=⋯=mq−1=m,k1=⋯=kr−1=k.n_1 = \cdots= n_{p-1} = n,\qquad m_1 = \cdots= m_{q-1} = m,\qquad k_1 = \cdots= k_{r-1} = k.

For a prime equal to 2, the corresponding assertion has just one entry and is automatic. Each common value is positive, since each of the normalized low functions is nonzero somewhere. The constant coefficient of (11) is now pqr−p2n−q2m−r2kpqr - p^2n - q^2m - r^2k. This proves (9). Counting available positions in the three rectangles of (8) gives (10).

Excluding the unconfined alternative

We now show that the mass identity and the capacity bounds are incompatible.

Lemma 5.1 (Three-rectangle inequality). Let x,y,z∈[0,1]x,y,z \in[0,1], and suppose

0≤α≤yz,0≤β≤x(1−z),0≤γ≤(1−x)(1−y).0 \le\alpha\le yz,\qquad0 \le\beta\le x(1-z),\qquad0 \le\gamma\le(1-x)(1-y).

Then

1−α−β−γ≥2αβγ.(12)1-\alpha-\beta-\gamma\ge2\sqrt{\alpha\beta\gamma}. \tag*{(12)}

Proof. Put A=yzA=yz, B=x(1−z)B=x(1-z), and C=(1−x)(1−y)C=(1-x)(1-y). The exact identities

1−A−B−C=D+E,ABC=DE,1-A-B-C=D+E,\qquad ABC=DE,

hold with

D=x(1−y)z,E=(1−x)y(1−z).D=x(1-y)z,\qquad E=(1-x)y(1-z).

Therefore 1−A−B−C≥2ABC1-A-B-C \ge2\sqrt{ABC} by the arithmetic–geometric mean inequality. Decreasing AA, BB, CC to α\alpha, β\beta, γ\gamma increases the left-hand side and decreases the right-hand side.

Lemma 5.2 (Arithmetic obstruction). For distinct primes p<q<rp<q<r, the integer system (9)–(10) has no solution.

Proof. First suppose p≥5p \ge5. Apply Lemma 5.1 with

x=ap,y=bq,z=cr,x=\frac{a}{p},\qquad y=\frac{b}{q},\qquad z=\frac{c}{r},

and

α=(p−1)nqr,β=(q−1)mpr,γ=(r−1)kpq.\alpha=\frac{(p-1)n}{qr},\qquad\beta=\frac{(q-1)m}{pr},\qquad\gamma=\frac{(r-1)k}{pq}.

The capacity bounds apply because d≤r−cd \le r-c. By (9),

1−α−β−γ=pn+qm+rkpqr.1-\alpha-\beta-\gamma=\frac{pn+qm+rk}{pqr}.

Consequently (12) gives

(pn+qm+rk)2nmk≥4(p−1)(q−1)(r−1).(13)\frac{(pn+qm+rk)^2}{nmk}\ge4(p-1)(q-1)(r-1). \tag*{(13)}

On the other hand, n,m,k≥1n,m,k \ge1 and (9) imply

(pn+qm+rk)2nmk=p2nmk+q2mnk+r2knm+2pqk+2prm+2qrn≤pqr+2(pq+pr+qr).(14)\frac{(pn+qm+rk)^2}{nmk}=\frac{p^2n}{mk}+\frac{q^2m}{nk}+\frac{r^2k}{nm}+\frac{2pq}{k}+\frac{2pr}{m}+\frac{2qr}{n} \le pqr+2(pq+pr+qr). \tag*{(14)}

These inequalities are incompatible. Indeed

4(p−1)(q−1)(r−1)−pqr−2(pq+pr+qr)=3pqr−6(pq+pr+qr)+4(p+q+r)−4>0.4(p-1)(q-1)(r-1)-pqr-2(pq+pr+qr) =3pqr-6(pq+pr+qr)+4(p+q+r)-4>0.

To verify positivity, write p=5+sp=5+s, q=7+tq=7+t, and r=11+ur=11+u, where s,t,u≥0s,t,u\ge0. The last expression is

241+127s+73t+37u+27st+15su+9tu+3stu.(15)241+127s+73t+37u+27st+15su+9tu+3stu. \tag*{(15)}

The lower bounds 5,7,115,7,11 hold because the primes are distinct and ordered.

If p=2p=2, then a=1a=1, so the last capacity inequality gives

r−1≤(r−1)k≤q−b≤q−1,r-1\le(r-1)k\le q-b\le q-1,

contrary to r>qr>q.

It remains to treat p=3p=3. If a=2a=2, the last capacity inequality gives r−1≤q−b<r−1r-1\le q-b<r-1, which is impossible. Thus a=1a=1. If k≥2k\ge2, the same inequality gives 2(r−1)≤2(q−b)<2(r−1)2(r-1)\le2(q-b)<2(r-1), again impossible. Hence k=1k=1 and

r≤2q−2b+1.r\le2q-2b+1.

If m≥2m\ge2, the second capacity bound and c+d≤rc+d\le r give

2(q−1)≤d≤r−c≤2q−2b+1−c,2(q-1)\le d\le r-c\le2q-2b+1-c,

so 2b+c≤32b+c\le3. Positivity of b,cb,c forces b=c=1b=c=1, contradicting 2n≤bc2n\le bc. Thus m=1m=1. Equation (9) becomes

9n+q2+r2=3qr.9n+q^2+r^2=3qr.

Modulo 3, the left-hand side is 2, since qq and rr are primes different from 3, whereas the right-hand side is 0. This final contradiction proves the lemma.

Theorem 5.3 (Confinement). Every member of E(p,q,r)\mathcal{E}(p,q,r) is confined modulo at least one of p,q,rp,q,r. Equivalently, at least one of the low functions in (5) is constant.

Proof. If all three low functions were nonconstant, Lemmas 4.1 and 4.2 would give a solution of (9)–(10), contradicting Lemma 5.2.

Remark 5.4 (Why the low-order zero is essential). Condition (K) alone does not force confinement. For (p,q,r)=(3,5,7)(p,q,r)=(3,5,7), take U0={0}U_0=\{0\}, V0={0,1}V_0=\{0,1\}, Wf={0,1,2}W_f=\{0,1,2\}, and Wg={3,4,5,6}W_g=\{3,4,5,6\}. Assign f(0,w)=1f(0,w)=1 and f(1,w)=2f(1,w)=2 on WfW_f; assign g(0,3),…,g(0,6)=1,2,3,4g(0,3),\ldots,g(0,6)=1,2,3,4; and assign the six values 1,…,61,\ldots,6 to hh on {1,2}×{2,3,4}\{1,2\}\times\{2,3,4\}. Put every other low value equal to zero. Lemma 4.1 gives (K), but the low-character sum is

105−32⋅3−52−72=4,105-3^2\cdot3-5^2-7^2=4,

not zero. Thus the square-signature conditions allow unconfined tiles, and the additional low-order zero supplies the obstruction.

The semigroup criterion and the complete layer form

In the CRT normal form, confinement modulo rr means exactly that hh is constant. We now recover a numerical-semigroup condition from that support restriction.

Theorem 6.1 (Confined-support criterion). For any three distinct primes p,q,rp,q,r (without an ordering requirement), there is a member of E(p,q,r)\mathcal{E}(p,q,r) confined modulo rr if and only if

r∈⟨p,q⟩.(16)r \in\langle p,q\rangle. \tag*{(16)}

Proof. Sufficiency is Theorem 7.1 with L=rL=r, e=1e=1. For necessity, translate the third low digit to zero, so h=0h=0. By Lemma 3.2, each fixed-ww layer has f(⋅,w)f(\cdot,w) constant or g(⋅,w)g(\cdot,w) constant. Define

Wij=#{(u,v,w):f(v,w)=i, g(u,w)=j}.W_{ij}=\#\{(u,v,w):f(v,w)=i,\ g(u,w)=j\}.

Every entry belongs to ⟨p,q⟩\langle p,q\rangle: a layer with constant gg contributes multiples of pp to each entry, while one with constant ff contributes multiples of qq. Independently, every row sum is a multiple of pp and every column sum a multiple of qq.

The zero (7) becomes ∑i,jWijζpiζqj=0\sum_{i,j}W_{ij}\zeta_p^i\zeta_q^j=0. The products ζpiζqj\zeta_p^i\zeta_q^j, i<p−1i<p-1, j<q−1j<q-1, form a basis of Q(ζpq)\mathbb{Q}(\zeta_{pq}). Eliminating the last row and column therefore gives

Wij−Wi,q−1−Wp−1,j+Wp−1,q−1=0.W_{ij}-W_{i,q-1}-W_{p-1,j}+W_{p-1,q-1}=0.

Hence WW is an additive integer matrix. Choose a position (i0,j0)(i_0,j_0) of minimum value tt. A row difference is independent of the column; multiplying it by the number qq of columns gives a multiple of pp, so coprimality forces the difference itself to be a multiple of pp. The analogous argument forces each column difference to be a multiple of qq. Thus

Wij=t+pαi+qβj,αi,βj∈Z≥0,min⁡α=min⁡β=0.(17)W_{ij}=t+p\alpha_i+q\beta_j,\qquad\alpha_i,\beta_j\in\mathbb{Z}_{\geq0},\qquad\min\alpha=\min\beta=0. \tag*{(17)}

Put a=∑iαia=\sum_i\alpha_i, b=∑jβjb=\sum_j\beta_j. Total mass and row/column divisibility yield

r=t+a+b,t+b=pv,t+a=qur=t+a+b,\qquad t+b=pv,\qquad t+a=qu

for nonnegative integers u,vu,v. Since the minimum entry is itself in the semigroup, write t=pc+qdt=pc+qd, c,d≥0c,d\geq0. Then

r=p(v−c)+q(u−d).r=p(v-c)+q(u-d).

Both coefficients are nonnegative: pv=t+b≥pcpv=t+b\geq pc and qu=t+a≥qdqu=t+a\geq qd. This proves (16). □\square

Remark 6.2 (An equivalent mass criterion). For any integer L≥1L\geq1 coprime to pqpq, membership L∈⟨p,q⟩L\in\langle p,q\rangle is equivalent to the assertion that

pq=pU+qV+LW,U,V,W∈Z≥0,pq=pU+qV+LW,\qquad U,V,W\in\mathbb{Z}_{\geq0},

has only the solutions (q,0,0)(q,0,0) and (0,p,0)(0,p,0). Indeed, in a representation L=ap+bqL=ap+bq, coprimality forces a,b≥1a,b\geq1. A solution with W>0W>0 would express pqpq as a combination of pp and qq with both coefficients positive, which reduction modulo pp excludes. Conversely, choose b∈{1,…,p−1}b\in\{1,\ldots,p-1\} with bq≡L(modp)bq\equiv L\pmod p. If L∉⟨p,q⟩L\notin\langle p,q\rangle, then a=(bq−L)/p≥1a=(bq-L)/p\geq1, and pq=pa+q(p−b)+Lpq=pa+q(p-b)+L gives a solution with W=1W=1. Thus the admissible Apéry scales are exactly those for which the third term cannot occur in this mass equation.

Proof of Theorem 1.1. By Theorem 5.3, every member of E(p,q,r)\mathcal{E}(p,q,r) is confined modulo some prime dividing MM. Confinement modulo pp would give p∈⟨q,r⟩p\in\langle q,r\rangle by Theorem 6.1, which is impossible. Confinement modulo qq would give q∈⟨p,r⟩q\in\langle p,r\rangle; since q<rq<r and p∤qp\nmid q, this too is impossible. Thus confinement is modulo rr, and Theorem 6.1 proves (2). □\square

Corollary 6.3 (Complete layer form). After translation in the third CRT coordinate, the members of E(p,q,r)\mathcal{E}(p,q,r) are exactly the sets

A={(pu+fw(v),qv+gw(u),rw):u∈{0,…,p−1}, v∈{0,…,q−1}, w∈{0,…,r−1}},(18)A=\{(pu+f_w(v),qv+g_w(u),rw):u\in\{0,\ldots,p-1\},\ v\in\{0,\ldots,q-1\},\ w\in\{0,\ldots,r-1\}\}, \tag*{(18)}

where, for every ww, either fwf_w or gwg_w is constant, and

∑w=0r−1(∑v=0q−1ζpfw(v))(∑u=0p−1ζqgw(u))=0.(19)\sum_{w=0}^{r-1}\left(\sum_{v=0}^{q-1}\zeta_p^{f_w(v)}\right)\left(\sum_{u=0}^{p-1}\zeta_q^{g_w(u)}\right)=0. \tag*{(19)}

Proof. Necessity follows from Theorem 1.1, Theorem 3.1, and Lemma 3.2. For sufficiency, pairs in one ww-layer satisfy (K) by Lemma 3.2; pairs in different layers are separated in the third coordinate because their third low digits are all zero. Theorem 3.1 and (19) therefore give all the required zeros.

Corollary 6.4 (Primitive quotient and complement slices). Let p<q<rp<q<r be distinct primes, N=pqrN=pqr, M=N2M=N^2, and A∈E(p,q,r)A\in\mathcal{E}(p,q,r). Then

gcd⁡(M,A−A)=r,∣Stab⁡(A)∣∣pq.\gcd(M,A-A)=r,\qquad|\operatorname{Stab}(A)|\mid pq.

After translation, write A=rCA=rC, with C⊂ZM/rC\subset\mathbb{Z}_{M/r}. Then

gcd⁡(M/r,C−C)=1,SC={p2,q2,r}.\gcd(M/r,C-C)=1,\qquad S_C=\{p^2,q^2,r\}.

The set CC is a tile of ZM/r\mathbb{Z}_{M/r} and satisfies (T2), and Φpq\Phi_{pq} is unsupported for CC. Conversely, every set C⊂ZM/rC\subset\mathbb{Z}_{M/r} of cardinality NN with this signature, satisfying (T2) and Φpq∣C\Phi_{pq}\mid C, gives rC∈E(p,q,r)rC\in\mathcal{E}(p,q,r). Every complement of rCrC has the unique form

B=⋃j=0r−1(j+rDj),C⊕Dj=ZM/r(0≤j<r).B=\bigcup_{j=0}^{r-1}(j+rD_j),\qquad C\oplus D_j=\mathbb{Z}_{M/r}\quad(0\le j<r).

In particular, ∣Dj∣=pq|D_j|=pq, and the complements DjD_j may be chosen independently.

Proof. Theorem 1.1 gives confinement modulo rr. Confinement modulo pp or qq is excluded by Theorem 6.1 and the ordering of the primes. Confinement modulo r2r^2 would give A(ζr2t)=ζr2t∣A∣≠0A(\zeta_{r^2}^t)=\zeta_{r^2}^t|A|\ne0 for some integer tt, contrary to Φr2∣A\Phi_{r^2}\mid A. Thus gcd⁡(M,A−A)=r\gcd(M,A-A)=r.

The stabilizer acts freely on AA, so its order divides NN. If that order were divisible by rr, the order-rr subgroup of ZM\mathbb{Z}_M would preserve AA. In the layer form of Corollary 6.3, this would make all rr layers identical. The low-character contribution of each individual layer is nonzero: one low vector is constant, and vanishing of the other vector’s sum of prime-order roots would require p∣qp\mid q or q∣pq\mid p. The total low-character sum could therefore not vanish. Hence ∣Stab⁡(A)∣∣pq|\operatorname{Stab}(A)|\mid pq.

Choose the common residue of AA modulo rr in {0,…,r−1}\{0,\ldots,r-1\} and subtract it. This gives the ordinary mask identity A(X)=C(Xr)A(X)=C(X^r) for the translated set. The preceding gcd identity gives primitivity of CC. Evaluating this mask at roots of unity shows that substitution by XrX^r preserves the pp- and qq-power levels and raises every positive rr-power level by one. Consequently SC={p2,q2,r}S_C=\{p^2,q^2,r\}, and the seven square-signature zeros of AA give (T2) for C.Theorder−C. The order-pqrzeroofzero ofAgivestheorder−gives the order-pqzeroofzero ofC$, which is unsupported by its signature. The same root-order calculation proves the converse.

Finally, put Dj={(b−j)/r:b∈B, b≡j(modr)}D_j=\{(b-j)/r:b\in B,\ b\equiv j\pmod r\}, interpreted in ZM/r\mathbb{Z}_{M/r}. Representations of residues congruent to jj in rC+BrC+B correspond exactly to representations in C+DjC+D_j. Thus rC⊕B=ZMrC\oplus B=\mathbb{Z}_M holds precisely when every DjD_j is a complement of CC, proving the last assertion.

All stabilizer orders

The stabilizer restriction in Corollary 6.4 is sharp. All four possibilities can be obtained by modifying the same collection of layers.

Theorem 6.5 (All stabilizer orders). If p<q<rp<q<r are primes and r∈⟨p,q⟩r\in\langle p,q\rangle, the stabilizer orders realized by members of E(p,q,r)\mathcal{E}(p,q,r) are exactly 1,p,q,pq1,p,q,pq. Each occurs in a tiling with the same aperiodic complement B□B_{\square}.

Proof. Write r=ap+bqr=ap+bq, where a,b≥1a,b\ge1; neither coefficient can vanish because rr is a distinct prime. Start with rr layers with h=0h=0. Choose their constant low pairs (f,g)(f,g) as aa copies of the pp-cycle (i,0)(i,0), i∈{0,…,p−1}i\in\{0,\ldots,p-1\}, and bb copies of the qq-cycle (0,j)(0,j), j∈{0,…,q−1}j\in\{0,\ldots,q-1\}. Each layer is a full high-digit rectangle. The low-character sum is pqpq times a sum of full pp- and qq-cycles, hence zero.

Every low vector is constant in this initial construction, so it has both the order-pp and order-qq periods. Corollary 6.4 gives stabilizer order exactly pqpq.

For a first modification, choose the layers (0,0),(1,0)(0,0),(1,0) from one pp-cycle. Replace their constant ff-vectors by

(0,…,0,1),(1,…,1,0),(0,\ldots,0,1),\qquad(1,\ldots,1,0),

respectively, leaving g=0g=0. Their combined low-character contribution is unchanged. For a second, independent modification, choose two layers (0,0),(0,1)(0,0),(0,1) from one qq-cycle, distinct from the two previously selected layers; leave f=0f=0 and replace their constant gg-vectors by the analogous complementary vectors of length pp. Again the combined contribution is unchanged. Every layer satisfies Lemma 3.2, and distinct layers satisfy (K) because h=0h=0.

The order-pp subgroup cycles the high uu-digits, so it preserves AA exactly when every gwg_w is constant. Likewise, the order-qq period is present exactly when every fwf_w is constant. Applying only the first modification therefore gives stabilizer order pp; applying only the second gives order qq; applying both gives order 11. No order-rr period is possible by Corollary 6.4. All four sets tile with B□B_{\square} by Theorem 3.1. Its coordinate intervals are aperiodic, so their product is aperiodic. ∎

In particular, the semigroup criterion also characterizes existence with both factors aperiodic. Every A∈E(p,q,r)A\in\mathcal{E}(p,q,r) has the precise common divisor rr, while its quotient by rr is primitive.

Apéry constructions and all complements

Fix distinct primes p,q,rp,q,r, integers m≥1m\ge1, e≥0e\ge0, and set

Q=pq,L=rm∈⟨p,q⟩,h=re,K=QhL,M=QK.Q=pq,\qquad L=r^m\in\langle p,q\rangle,\qquad h=r^e,\qquad K=QhL,\qquad M=QK.

The parameter ee controls the number of unsupported scales, while mm controls the Apéry core size.

Theorem 7.1 (Construction with arbitrarily many unsupported scales). The polynomials

A(X)=[Q]XK[L]XhΦQ(Xh),B0(X)=[h]X[Q]XhL(20)A(X)=[Q]_{X^K}[L]_{X^h}\Phi_Q(X^h),\qquad B_0(X)=[h]_X[Q]_{X^{hL}} \tag*{(20)}

are 00–11 masks and satisfy

A⊕B0=ZM,∣A∣=QL,∣B0∣=Qh.A\oplus B_0=\mathbb{Z}_M,\qquad|A|=QL,\qquad|B_0|=Qh.

Their exact prime-power signatures are

SA={p2,q2}∪{re+1,…,re+m},S_A=\{p^2,q^2\}\cup\{r^{e+1},\ldots,r^{e+m}\},
SB0={p,q}∪{r,r2,…,re},S_{B_0}=\{p,q\}\cup\{r,r^2,\ldots,r^e\},

with the second union empty for e=0e = 0. Both satisfy (T1)(T1), (T2)(T2). Every factor

ΦQ,ΦQr,…,ΦQre\Phi_Q,\Phi_{Qr},\ldots,\Phi_{Qr^e}

is unsupported for AA.

Proof. The support of [L]XΦQ(Xh)[L]_X\Phi_Q(X^h) is hAp⁡(S;L)h\operatorname{Ap}(S;L) and lies in [0,K)[0,K). Its QQ translates by multiples of KK are disjoint and lie in [0,M)[0,M). The claim for B0B_0 follows directly from its separated intervals. Telescoping geometric series gives the ordinary identity

A(X)B0(X)=[M]XΦQ(Xh).(21)A(X)B_0(X)=[M]_X\Phi_Q(X^h). \tag*{(21)}

For any integer polynomial RR, [M]XR(X)≡R(1)[M]X[M]_XR(X)\equiv R(1)[M]_X modulo XM−1X^M-1. Since ΦQ(1)=1\Phi_Q(1)=1, equation (3) follows.

The complete cyclotomic factorization of AA is

A(X)=∏s∣Ms∤KΦs(X)∏j=e+1e+mΦrj(X)∏j=0eΦQrj(X).(22)A(X)=\prod_{\substack{s\mid M\\s\nmid K}}\Phi_s(X)\prod_{j=e+1}^{e+m}\Phi_{r_j}(X)\prod_{j=0}^{e}\Phi_{Qr_j}(X). \tag*{(22)}

Indeed ΦQ(Xre)=∏j=0eΦQrj(X)\Phi_Q(X^{r^e})=\prod_{j=0}^{e}\Phi_{Qr^j}(X). The first product contains precisely the divisors of MM with a p2p^2 or q2q^2 component. This proves the exact signature and all (T2)(T2) requirements for AA. It also excludes the low components of the last e+1e+1 factors, proving that they are unsupported. The signature and (T2)(T2) of B0B_0 follow either from its factorization or from the complete residue-system argument in Theorem 7.4 below.

The core complement rigidity

Lemma 7.2 (A binary additive matrix). If a real matrix has the form F(x,y)=u(x)+v(y)F(x,y)=u(x)+v(y) and every entry is 0 or 1, it depends on at most one variable.

Proof. If u(x1)≠u(x2)u(x_1)\ne u(x_2), the nonzero difference F(x1,y)−F(x2,y)F(x_1,y)-F(x_2,y) is independent of yy. A nonzero difference between binary entries is 1 or −1-1; the two rows must therefore be constant rows of opposite values. This forces vv to be constant. Otherwise uu was already constant.

Lemma 7.3 (All complements of the Apéry core). In ZQL\mathbb{Z}_{QL}, all complements of CL=Ap⁡(S;L)C_L=\operatorname{Ap}(S;L) are exactly

a+LZQL,0≤a<L.a+L\mathbb{Z}_{QL},\qquad0\le a<L.

Proof. The mask CL(X)=[L]XΦQ(X)C_L(X)=[L]_X\Phi_Q(X) has, among orders dividing QLQL, exactly the zeros of nontrivial order dividing LL, and order QQ. If CL⊕E=ZQLC_L\oplus E=\mathbb{Z}_{QL}, the Fourier support of 1E\mathbf{1}_E is therefore contained in the union of the two coordinate axes under

ZQL≃ZL×ZQ.\mathbb{Z}_{QL}\simeq\mathbb{Z}_L\times\mathbb{Z}_Q.

Indeed the Fourier transform of the convolution is zero at every nontrivial character, so a nonzero Fourier coefficient of EE must occur at a zero of CLC_L. Inverting on the two axes gives 1E(x,y)=u(x)+v(y)\mathbf{1}_E(x,y)=u(x)+v(y), with u,vu,v chosen real because the original function is real. By the preceding lemma, 1E\mathbf{1}_E depends on one variable only. Dependence only on yy would make ∣E∣=Q|E|=Q a multiple of LL, impossible since L>1L>1 and gcd⁡(L,Q)=1\gcd(L,Q)=1. Thus it selects exactly one value of xx. Conversely each indicated coset complements the complete residue system CLC_L mod LL.

Theorem 7.4 (Complete complement parameterization). For the set AA in (20), every complement, and no other set, has the form

B={t+hat+hLj+Kℓt,j:0≤t<h, 0≤j<Q},(23)B=\{t+ha_t+hLj+K\ell_{t,j}:0\le t<h,\ 0\le j<Q\}, \tag*{(23)}

where all parameters are independent and

at∈{0,…,L−1},ℓt,j∈{0,…,Q−1}.a_t\in\{0,\ldots,L-1\},\qquad\ell_{t,j}\in\{0,\ldots,Q-1\}.

Consequently the number of complements as subsets of ZM\mathbb{Z}_M is

LhQQh.(24)L^hQ^{Qh}. \tag*{(24)}

Every such BB satisfies (T1)(T1), (T2)(T2), with the exact signature SB=SB0S_B=S_{B_0}. In the ring Z[X]\mathbb{Z}[X], the monic greatest common divisor of all their masks is

gcd⁡A⊕B=ZMB(X)=[h]X[Q]XhL=B0(X).(25)\gcd_{A\oplus B=\mathbb{Z}_M}B(X)=[h]_X[Q]_{X^{hL}}=B_0(X). \tag*{(25)}

Proof. The set AA is the full preimage of hCL⊂ZKhC_L\subset\mathbb{Z}_K under the quotient by KZMK\mathbb{Z}_M. In a tiling complement, projection to ZK\mathbb{Z}_K is injective: two elements differing by a nonzero member of this period subgroup would give a repeated representation. The projected complement must complement hCLhC_L in ZK\mathbb{Z}_K.

Split that complement according to its residues tt mod hh. In each residue class, division by hh gives a complement of CLC_L in ZQL\mathbb{Z}_{QL}. Lemma 7.3 gives the form at+L{0,…,Q−1}a_t+L\{0,\ldots,Q-1\}. Each projected point then has an arbitrary lift by Kℓt,jK\ell_{t,j}. This proves necessity, sufficiency, and uniqueness of the parameters, and hence (24).

Modulo QhQh, the QQ entries for each tt are all residues with that value modulo hh, since LL is coprime to QQ. Thus BB is a complete residue system modulo QhQh. It vanishes at every nontrivial QhQhth root and contains all prime-power factors in SB0S_{B_0}. These already contribute Qh=∣B∣Qh=|B| at X=1X=1. Any additional prime-power factor would contradict this equality. This proves the exact signature, (T1)(T1), and (T2)(T2).

For an order d∣Kd\mid K, lifting does not affect evaluation at primitive ddth roots, and the projected mask is

[Q]XhL∑t=0h−1Xt+hat.[Q]_{X^{hL}}\sum_{t=0}^{h-1}X^{t+ha_t}.

The first factor vanishes whenever d∣Kd\mid K but d∤hLd\nmid hL. If d∣hd\mid h, d>1d>1, the second factor vanishes. These are exactly the cyclotomic factors of B0B_0. Thus B0B_0 divides every BB. The choice at=ℓt,j=0a_t=\ell_{t,j}=0 gives B=B0B=B_0 itself, so no further factor or higher multiplicity can occur in the gcd. □

Remark 7.5 (Quotient periodicity is not actual periodicity). Formula (23) says that the projection of BB has the indicated fibers. Arbitrary independent lifts usually destroy periods of BB in ZM\mathbb{Z}_M. Universal cyclotomic zeros are inherited, but a subgroup period need not be.

Corollary 7.6 (Three-prime existence at bounded exponents). For every p<q<rp<q<r, there is a tiling with unsupported Φpqr\Phi_{pqr} at period p2q2r3p^2q^2r^3, with both factors satisfying (T1)(T1) and (T2)(T2). The same construction works at period (pqr)2(pqr)^2 exactly when r∈⟨p,q⟩r\in\langle p,q\rangle. For a fixed admissible mm, it also gives arbitrarily many unsupported factors while keeping ∣A∣=pqrm|A|=pqr^m fixed.

Proof. Every integer at least (p−1)(q−1)(p-1)(q-1) belongs to ⟨p,q⟩\langle p,q\rangle, and r2>pq>(p−1)(q−1)r^2>pq>(p-1)(q-1). For completeness, if n≥(p−1)(q−1)n\ge(p-1)(q-1), choose b∈{0,…,p−1}b\in\{0,\ldots,p-1\} with bq≡n(modp)bq\equiv n\pmod p. Then n−bq>−pn-bq>-p and is a multiple of pp, so it is nonnegative. Thus n∈⟨p,q⟩n\in\langle p,q\rangle. Taking L=r2L=r^2 and e=1e=1 proves the first assertion; L=rL=r gives the second. The last assertion follows by increasing ee in Theorem 7.1. □

Sharp small periods and explicit examples

For p,q,r=2,3,5p,q,r=2,3,5, one has

C5=Ap⁡(⟨2,3⟩;5)={0,2,3,4,6}.C_5=\operatorname{Ap}(\langle2,3\rangle;5)=\{0,2,3,4,6\}.

Modulo 66, this has the multiset of residues {0,0,2,3,4}\{0,0,2,3,4\}, the union of a 2-cycle and a 3-cycle, and modulo 55 it is a complete residue system. Thus it is a concrete representative of the configuration in [3, Example 8.1]. The core itself has no unsupported Φ6\Phi_6, since 2,3∤∣C5∣2,3\nmid|C_5|; the following lifts provide the missing cardinality factors.

Taking m=1,e=0m=1,e=0 in (20) gives

M=180,A={30j+c:0≤j<6, c∈C5},B0=5{0,…,6−1},M=180,\qquad A=\{30j+c:0\le j<6,\ c\in C_5\},\qquad B_0=5\{0,\ldots,6-1\},

with SA={4,5,9}S_A=\{4,5,9\} and unsupported Φ6\Phi_6. Taking m=e=1m=e=1 gives

M=900,A={150j+5c:0≤j<6, c∈C5}.(26)M=900,\qquad A=\{150j+5c:0\le j<6,\ c\in C_5\}. \tag*{(26)}

Here SA={4,9,25}S_A=\{4,9,25\} and both Φ6\Phi_6 and Φ30\Phi_{30} are unsupported. The identity

1+Y2+Y3+Y4+Y6=Φ5(Y)Φ6(Y)1+Y^2+Y^3+Y^4+Y^6=\Phi_5(Y)\Phi_6(Y)

gives an immediate direct check of the core. Every complement of (26) is exactly

{t+5at+25j+150ℓt,j:0≤t<5, 0≤j<6},at∈{0,…,5−1},ℓt,j∈{0,…,6−1}.(27)\{t+5a_t+25j+150\ell_{t,j}:0\le t<5,\ 0\le j<6\},\qquad a_t\in\{0,\ldots,5-1\},\quad\ell_{t,j}\in\{0,\ldots,6-1\}. \tag*{(27)}

There are 556305^5 6^{30} such complements.

Proposition 8.1 (Least cyclic periods). Among tilings A⊕B=ZMA\oplus B=\mathbb{Z}_M with AA having an unsupported divisor Φs\Phi_s with s∣Ms\mid M, the least possible MM is 180. If ss is required to contain three distinct primes, the least possible MM is 900, and the least possible ∣A∣|A| is 30.

Proof. An unsupported order contains at least two primes. Each of its primes requires an excluded prime-power level and a different supported level: the first comes from the definition, and the second from (T1), which is necessary for a cyclic tile. By Lemma 2.1, both levels are within the specified period. Hence each such prime occurs in MM to exponent at least two. If MM had at most two distinct prime factors, then ∣A∣|A| would also, so the Coven–Meyerowitz theorem [2] would give (T2). The two-prime exclusion of [3] would then rule out an unsupported divisor. Thus MM has at least one further prime, and M≥22325=180M\ge2^2 3^2 5=180. If the unsupported order has at least three primes, the corresponding bound is M≥223252=900M\ge2^2 3^2 5^2=900, and the cardinality condition gives ∣A∣≥30|A|\ge30. The preceding examples attain all these bounds.

Theorem 1.1 does not make the minimal example unique. Appendix A classifies all members of E(2,3,5)\mathcal{E}(2,3,5) by their layers and counts them exactly. In particular, most do not have the period subgroup of the example (26).

An explicit aperiodic minimal example

In (18), take the following layers:

wwfwf_wgwg_w
0(0,0,1)(0,0,1)(0,0)(0,0)
1(0,1,1)(0,1,1)(0,0)(0,0)
2(0,0,0)(0,0,0)(0,1)(0,1)
3(0,0,0)(0,0,0)(0,1)(0,1)
4(0,0,0)(0,0,0)(2,2)(2,2)

Table 1.

Writing ζ=ζ3\zeta=\zeta_3, their character values sum to 2−2−3ζ2−3ζ2+6ζ2=02-2-3\zeta^2-3\zeta^2+6\zeta^2=0. The standard CRT inverse is

(x,y,z)⟼225x+100y+576z(mod900).(x,y,z)\longmapsto225x+100y+576z \pmod{900}.

It gives

A={0,10,20,60,105,170,180,190,240,255,300,310,320,360,375,450,470,490,540,555,610,620,630,660,705,750,770,790,825,840}.A=\{0,10,20,60,105,170,180,190,240,255,300,310,320,360,375,450,470,490,540,555,610,620,630,660,705,750,770,790,825,840\}.

A complement is the inverse image of {0,…,2−1}×{0,…,3−1}×{0,…,5−1}\{0,\ldots,2-1\}\times\{0,\ldots,3-1\}\times\{0,\ldots,5-1\}:

B□={0,1,28,29,100,101,128,153,200,225,252,253,325,352,353,425,452,477,504,576,577,604,676,677,704,729,776,801,828,829}.B_{\square}=\{0,1,28,29,100,101,128,153,200,225,252,253,325,352,353,425,452,477,504,576,577,604,676,677,704,729,776,801,828,829\}.

Both stabilizers are trivial. For AA, the nonconstant gg in layer 2 breaks the order-2 period; the nonconstant ff in layer 0 breaks the order-3 period; the different layers break the order-5 period. The standard complement is aperiodic by its product-interval form. They satisfy

A⊕B□=Z900,SA={4,9,25},SB□={2,3,5},Φ30∣A.A\oplus B_{\square}=\mathbb{Z}_{900},\qquad S_A=\{4,9,25\},\qquad S_{B_{\square}}=\{2,3,5\},\qquad\Phi_{30}\mid A.

Since affine automorphisms preserve the stabilizer size, this set is not affinely equivalent to the periodic example (26). The aperiodic example is still contained in 5Z9005\mathbb{Z}_{900} and is not primitive.

The lower modulus 180 also allows both factors aperiodic

Divide the preceding AA by 5 and take the zero-modulo-5 slice of B□B_{\square}, also divided by 5. This gives

C={0,2,4,12,21,34,36,38,48,51,60,62,64,72,75,90,94,98,108,111,122,124,126,132,141,150,154,158,165,168},C=\{0,2,4,12,21,34,36,38,48,51,60,62,64,72,75,90,94,98,108,111,122,124,126,132,141,150,154,158,165,168\},
D={0,20,40,45,65,85}.D=\{0,20,40,45,65,85\}.

They tile Z180\mathbb{Z}_{180}, with SC={4,5,9}S_C=\{4,5,9\} and SD={2,3}S_D=\{2,3\}, and Φ6\Phi_6 is unsupported for CC. Any period of CC would lift to a period of AA, so CC is aperiodic. In CRT coordinates Z4×Z9×Z5\mathbb{Z}_4\times\mathbb{Z}_9\times\mathbb{Z}_5, the complement is

D={0,1}×{0,2,4}×{0},D=\{0,1\}\times\{0,2,4\}\times\{0\},

which is aperiodic as well. Moreover CC is primitive, since it contains 0, 2, 21. Thus the least modulus 180 for any unsupported order can be attained with both factors aperiodic and with the unsupported factor’s set primitive. For a three-prime unsupported order itself, the least modulus 900 can be attained with both factors aperiodic. Every AA with the specific unsupported order 30 at this modulus is necessarily nonprimitive.

Acknowledgments

The authors thank the authors of the works cited in this paper for the questions and ideas that motivated the present study. OpenAI’s GPT Astra was used during the preparation of the manuscript for language editing and as a research aid in exploring and identifying useful examples. All mathematical statements, constructions, and proofs were independently verified by the authors, who take full responsibility for the contents of the paper.

The computational checks, including those implemented in several Python files mentioned in Appendix A, supplement the proofs and can be reproduced using the verification code available in the public GitHub repository.

Complete enumeration at modulus 900

For p,q,r=2,3,5p,q,r=2,3,5, Corollary 6.3 is a finite complete classification. For each of the five high-5 layers, choose

f∈{0,1}3,g∈{0,1,2}2,f constant or g constant.f \in\{0,1\}^{3}, \qquad g \in\{0,1,2\}^{2}, \qquad f\text{ constant or }g\text{ constant}.

There are 2⋅32+3⋅23−2⋅3=362\cdot3^{2}+3\cdot2^{3}-2\cdot3=36 distinct layer choices. Their only remaining constraint is the exact zero sum (19). This is a necessary-and-sufficient description of all the sets in question; it does not presuppose periodicity or a particular polynomial factorization.

The constant-term certificate

Put ζ=ζ3\zeta=\zeta_{3}. A layer character value is one of

±6ζj (multiplicity 1),±3ζj (multiplicity 2),±2ζj (multiplicity 3),j=0,1,2.\pm6\zeta^{j}\ \text{(multiplicity 1)}, \qquad\pm3\zeta^{j}\ \text{(multiplicity 2)}, \qquad\pm2\zeta^{j}\ \text{(multiplicity 3)}, \qquad j=0,1,2.

For example, if both low vectors are constant the value is ±6ζj\pm6\zeta^{j}. If only ff is constant, two distinct gg values sum to the negative of the third root. If only gg is constant, the three signs have sum ±1\pm1.

In the integer basis (1,ζ)(1,\zeta), define

Ks(U,V)=Us+U−s+Vs+V−s+(UV)s+(UV)−s,F=K6+2K3+3K2.K_s(U,V)=U^s+U^{-s}+V^s+V^{-s}+(UV)^s+(UV)^{-s}, \qquad F=K_6+2K_3+3K_2.

Then the number of valid ordered five-layer choices for a fixed low-5 residue is

CT⁡F5=366 960.(28)\operatorname{CT}F^5=366\,960. \tag*{(28)}

A period of order 2 is equivalent to every gwg_w being constant; a period of order 3 is equivalent to every fwf_w being constant. Thus the other counts are

CT⁡(K6+3K2)5=146 160,(29)\operatorname{CT}(K_6+3K_2)^5=146\,160, \tag*{(29)}
CT⁡(K6+2K3)5=46 200,(30)\operatorname{CT}(K_6+2K_3)^5=46\,200, \tag*{(30)}
CT⁡K65=360.(31)\operatorname{CT}K_6^5=360. \tag*{(31)}

No order-5 period is possible: it would make all layers identical, while each single-layer character value is nonzero. A stabilizer order divides ∣A∣=30|A|=30, so no other orders need be considered. Inclusion–exclusion, followed by multiplication by the five choices of the common low-5 digit, gives:

Stabilizer orderOne fixed residue modulo 5All five residues
1174 960874 800
2145 800729 000
345 840229 200
63601 800
Total366 9601 834 800

Table 1.

These are counts of distinct subsets of Z900\mathbb{Z}_{900}, not counts modulo affine equivalence.

The integer constants in (28)–(31) are checked by two independent finite expansions in the repository’s verify.py and in the supplementary verification.py: a coefficient-convolution recurrence and an unordered five-term enumeration weighted by multinomial coefficients. Equivalently, for the coefficient function cc of any indicated Laurent polynomial, set D0(0,0)=1D_0(0,0)=1 and D0(x,y)=0D_0(x,y)=0 for (x,y)≠(0,0)(x,y)\ne(0,0), and use

Dk+1(x,y)=∑(a,b)c(a,b)Dk(x−a,y−b).D_{k+1}(x,y)=\sum_{(a,b)}c(a,b)D_k(x-a,y-b).

The desired constant is D5(0,0)D_5(0,0). The self-contained supplementary file verification.py uses only the Python standard library and is run with python3 verification.py. Besides the two constant-term computations, it independently counts actual ordered layers, enumerates all complements of each of the 360 cores below, and checks the explicit tiling examples by exact integer arithmetic. The structural classification and absence of other stabilizer types are proved above, not inferred from enumeration.

The periodic core subclass

For completeness, the order-6-periodic subfamily consists exactly of the sets

Aa,C=a+{150j+5c:0≤j<6, c∈C},0≤a<5,A_{a,C}=a+\{150j+5c:0\leq j<6,\ c\in C\},\qquad0\leq a<5,

where C⊂Z30C\subset\mathbb{Z}_{30} is a five-point complete residue system modulo 5 satisfying Φ6∣C\Phi_6\mid C. Indeed, quotienting by the order-6 period subgroup, then subtracting the common residue modulo 5 and dividing by 5, gives precisely such a core; the converse follows from the layer form. There are 360 such cores.

To see the count directly, let cjc_j be the six residue counts modulo 6. Since ζ6=−ζ32\zeta_6=-\zeta_3^2, the zero condition implies that the difference between the even and odd counts in each of the three ζ3\zeta_3 phases is a common integer kk. The total is five, so k=1k=1 or −1-1. In the majority parity, the three counts are (2,1,1)(2,1,1); in the minority parity they are (1,0,0)(1,0,0), with its nonzero position opposite the doubled residue. There are six such residue patterns. Assigning the five distinct residue labels modulo 5 to a pattern gives 5!/2=605!/2=60 cores. Hence the count is 6⋅60=3606\cdot60=360.

Under affine automorphisms x↦ux+vx\mapsto ux+v of Z30\mathbb{Z}_{30}, with gcd⁡(u,30)=1\gcd(u,30)=1, these cores have two orbits, of sizes 120 and 240, represented by

{0,2,3,4,6},{0,1,3,9,17}.\{0,2,3,4,6\},\qquad\{0,1,3,9,17\}.

One description of the distinction is whether the unique minority-parity point, viewed modulo 5, is the midpoint of the two points sharing the doubled residue modulo 6. Affine transformations preserve this property. After normalizing the doubled residue modulo 6 and its two labels modulo 5, the minority label is either the midpoint or one of the two remaining labels; the available reflections identify the latter two possibilities. The remaining two singletons may be exchanged independently in the modulo-6 coordinate. This proves there are exactly two orbits. Their sizes also follow from the ratio 1:21:2 of midpoint and nonmidpoint choices, and are independently checked in the program.

Both orbit representatives, and hence all 360 cores, have among nontrivial divisor orders of 30 exactly the zeros 5 and 6. To exclude orders 10, 15, 30 without factoring, fix d∈{2,3,6}d\in\{2,3,6\} and index the points of a core as ct≡t(mod5)c_t\equiv t\pmod5. Vanishing at order 5d5d is equivalent to

∑t=04ζdctζ5t=0,\sum_{t=0}^{4}\zeta_d^{c_t}\zeta_5^t=0,

since x↦ζdxζ5xx\mapsto\zeta_d^x\zeta_5^x has order 5d5d. The minimal polynomial of ζ5\zeta_5 over Q(ζd)\mathbb{Q}(\zeta_d) is 1+X+X2+X3+X41+X+X^2+X^3+X^4. The field extension has degree 4 because gcd⁡(d,5)=1\gcd(d,5)=1. Therefore the five coefficients ζdct\zeta_d^{c_t} would all have to be equal, forcing the points to lie in one residue class modulo dd. The five-point residue pattern modulo 6 established above rules this out for each dd. Orders 2, 3 are excluded by the cardinality 5. The Fourier proof of Lemma 7.3 therefore applies to every such core. Consequently all 1 800 order-6-periodic members of E(2,3,5)\mathcal{E}(2,3,5) have exactly the same complement family (27): for fixed aa, slicing a complement by its residues modulo 5 again gives the same coset complements of CC.

References

References

  1. [1]B. Bruce and I. Laba, Keller properties for integer tilings, Electron. J. Combin. 31 (2024), no. 4, Paper No. P4.47.DOI
  2. [2]E. M. Coven and A. Meyerowitz, Tiling the integers with translates of one finite set, J. Algebra 212 (1999), no. 1, 161–174.
  3. [3]G. Kiss, I. Laba, C. Marshall and G. Somlai, Lower bounds for mask polynomials with many cyclotomic divisors, Adv. Math. 494 (2026), 110932.
  4. [4]I. Laba, The spectral set conjecture and multiplicative properties of roots of polynomials, J. London Math. Soc. (2) 65 (2002), no. 3, 661–671.
  5. [5]I. Laba and I. Londner, Combinatorial and harmonic-analytic methods for integer tilings, Forum Math. Pi 10 (2022), e8, 46 pp.
  6. [6]I. Laba and I. Londner, The Coven–Meyerowitz tiling conditions for 3 odd prime factors, Invent. Math. 232 (2023), no. 1, 365–470.
  7. [7]I. Laba and I. Londner, The Coven–Meyerowitz tiling conditions for three prime factors: the even case, Res. Math. Sci. 12 (2025), Article No. 43.DOI
  8. [8]I. Laba and I. Londner, Splitting for integer tilings, Int. Math. Res. Not. IMRN (2025), no. 8, rnaf090.
  9. [9]T. Y. Lam and K. H. Leung, On vanishing sums of roots of unity, J. Algebra 224 (2000), no. 1, 91–109.DOI
  10. [10]P. Moore, Numerical semigroups, cyclotomic polynomials, and Bernoulli numbers, Amer. Math. Monthly 121 (2014), no. 10, 890–902.

Paper details

vlnc:W_asta2uam
Version
v1 · latest
Published
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents