Coxeter group symmetries of frieze patterns are investigated through their relation to 2×n matrices and the associated Plücker coordinates. For odd order n, the symmetric group acts on the set of friezes, while for even n, the dihedral subgroup acts. We further show that a larger group acts when n=6.
Introduction
The study of frieze patterns was initiated by Coxeter [6]. In his original formulation, a frieze pattern is represented as an infinite planar array of numbers arranged in staggered rows. A frieze pattern of order n is a grid of numbers with n+1 rows as in Figure 1, where the first and last rows (row 0 and row n) are composed entirely of zeros, while the second and second-to-last rows (row 1 and row n−1) consist entirely of ones. In other words, for any integer i, we have
Figure 1. A frieze pattern of order n arranged in staggered rows.
fi,i=fi,i+n=0,fi,i+1=fi,i+n−1=1.(1)
The internal entries of the frieze are governed by the “unimodular rule”; given any four adjacent elements arranged in a diamond configuration,
fi+1,jfi,jfi+1,j+1fi,j+1
the relation
fi,jfi+1,j+1−fi+1,jfi,j+1=1(2)
must hold for i<j<i+n. By the unimodular rule (2), providing n−3 diagonal entries f1,3,f1,4,…,f1,n−1 is sufficient to determine all other entries [6].
In their foundational work [5], Conway and Coxeter proved that every frieze pattern of order n whose non-trivial entries {fi,i+k;i∈Z,k=2,3,…,n−2} are positive integers corresponds uniquely to a triangulation of a convex n-gon. The entries in the first non-trivial row (“quiddity” [5]) {fi,i+2;i=1,2,…,n} represent the number of triangles incident to each vertex of the polygon. Since then, frieze patterns have been studied in connection with geometry and combinatorics [12, 14]. Recently, the introduction of cluster algebras [7] brought increasing attention to the subject, as Conway-Coxeter friezes give a natural realization of a cluster algebra of type An−3.
As noted by Coxeter [6], if we assume that the entries fi,j are non-zero for i<j<i+n, the following property holds:
Proposition 1 (cf. [6]). For a frieze of order n, let fi,j be its entries. If fi,j=0 for all i<j<i+n, then the Plücker relation
fi,jfk,ℓ−fi,kfj,ℓ+fi,ℓfj,k=0(3)
is satisfied for any integers i≤j≤k≤ℓ≤i+n.
A proof of Proposition 1 can be found in [11] (answer to exercise 2.3.4.6-4). The Plücker relations (3) are fundamental to the geometry of Grassmannians. From this perspective, a frieze pattern can be viewed as a realization of the coordinate ring of the Grassmannian
where vi=(z1,i,z2,i)T (i=1,…,n). In the next section, we introduce an identification of a frieze with a 2×n matrix Z=(v1,…,vn)[14]. This formulation enables us to analyze the symmetries of the frieze as geometric transformations of the vector sequence {vi∣i=1,…,n} induced by the action of the symmetric group Sn. In this paper, we provide an explicit description of these induced transformations. If n is odd, the entire symmetric group Sn acts on the frieze; if n is even, the dihedral group Dn (⊂Sn) acts on it.
This paper is organized as follows. Section 2 establishes the algebraic foundations of frieze patterns, describes their correspondence with the matrix space Mn, and introduces the normal forms. Section 3 then investigates the induced action of the symmetric group Sn. For odd n, we demonstrate that the unique normal form yields a well-defined action of Sn. For even n, we analyze which subgroup can act.
Frieze Patterns and the Matrix Space Mn
In this section, we establish the algebraic foundations for friezes extended to the entire planar lattice Z×Z.
Definition 2 (cf. [15]). Let K be a subset of the field of rational functions with complex coefficients C(X), and let n≥4 be an integer. An array of numbers F=(fi,j)i,j∈Z (fi,j∈K) is called a frieze of order n over K if it satisfies the global Plücker relation,
fi,jfk,ℓ−fi,kfj,ℓ+fi,ℓfj,k=0,(5)
and the boundary conditions,
fi,i=fi,i+n=0,fi,i+1=fi+1,i+n=1.(6)
In what follows, we denote the set of all friezes of order n over C(X) by CFn.
Remark 3. With the initial conditions in (6), Definition 2 is equivalent to the notion of a tame SL2-tiling in [15].
Several noteworthy features of this definition deserve immediate attention, particularly regarding how the global algebraic constraints recover and extend classical properties of friezes. By specializing the indices in the global Plücker relation (5) and applying the boundary conditions (6), we obtain
k=i+1,ℓ=i+n:fi,j=−fi+n,j(anti-periodicity in i),(9)
k=j+1,ℓ=j+n:fi,j=−fi,j+n(anti-periodicity in j).(10)
By combining the skew symmetry (7) with the anti-periodicity (9), we arrive at the identity:
fi,j=fj,i+n.(11)
This relation is called a glide symmetry, a well-known classical property of frieze patterns [12, 14]. Figure 2 illustrates the anti-periodicity (9) and glide symmetry (11) described above.
Henceforth, we let K be the field of rational functions with complex coefficients C(X). Let Mn be the set of 2×n matrices such that all 2×2 minors are non-zero:
Mn:={Z=(z1,1z2,1z1,2z2,2⋯⋯z1,nz2,n)Di,j(Z)=0 for 1≤i,j≤n}.(12)
where
Di,j(Z):=det(z1,iz2,iz1,jz2,j)(13)
for 1≤i,j≤n. We omit the argument Z when there is no risk of confusion. For a matrix Z∈Mn as in (12), we define the coordinate function xi,j by
xi,j(Z):=zi,j.(14)
We define Mn and Mn as subsets of Mn as follows:
Mn:={(0−110∗1∗∗⋯⋯∗∗1∗)∈MnDi,i+1=1 for 1≤i≤n, and D1,n=1}.(15)
Mn:={Z∈Mn∣z1,3=1}.(16)
We remark that Mn⊂Mn⊂Mn. When n=2m and c∈C×, we define a map η(c):M2m→M2m.
η(c)(Z)=diag(c,1)⋅Z⋅diag(1,c−1,…,1,c−1).(17)
For Z∈M2m, z1,3=0 because D1,3(Z)=0. Thus, η(z1,3−1)(Z)∈Mn. We define a map p:M2m→M2m by
Proof. For i=1 and 1≤j≤n, since k=ℓ=0, formula (21) reduces to
f1,j=z1,1z2,j−z1,jz2,1=z1,j.(22)
Similarly, for i=2 and 1≤j≤n, we obtain f2,j=z2,j. Next, we consider the case 1≤i,j≤n. By specializing the indices of the global Plücker relation (5) to k=1 and ℓ=2 and invoking the normalization (6), we can express fi,j as a 2×2 determinant:
The formula (21) follows from the anti-periodicity relations (9) and (10). □
To relate Mn and Mn, we consider the quotient space
GL2(K)\Mn/(K×)n(24)
by the equivalence relation defined as follows: for Z1,Z2∈Mn, we say Z1∼Z2 if and only if Z1=LZ2R for L∈GL2(K) and R=diag(d1,…,dn) (d1,…,dn∈K×). We remark that this quotient space coincides with that used in the study of hypergeometric functions [8, 10].
When n is even, for any matrix Z∈M2m, there exists an invariant mr(Z) with respect to the action of GL2(K)×(K×)2m:
Using this invariant, we define a subset M2m′ of M2m as follows:
M2m′:={Z∈M2m∣mr(Z)=1}.(26)
The following lemma is an immediate consequence of (15).
Lemma 5. For every Z∈M2m, we have mr(Z)=1.
We remark that M2m⊂M2m⊂M2m′.
Remark 6. When we specialize K to C, the quotient space (24) can be interpreted in two different ways, depending on the order in which the quotients are taken. On the one hand, taking the quotient by GL2(C) first yields the quotient of the Grassmannian Gr(2,n) by the right action of (C×)n. On the other hand, taking the quotient by (C×)n first yields the quotient of the configuration space of n points on the complex projective line P1 by the left action of GL2(C) (cf. [2]). The quotient space M2m′/(C×)2m was introduced in [13], where it is denoted by C2m. The invariant mr(Z) in (25) is known as the multi-ratio associated with a configuration of 2m points on the projective line P1. In the six-point case, this multi-ratio was studied by Bobenko et al. in their work on integrable circle patterns [3, 4].
Theorem 7. Let n≥4 be an integer and Z∈Mn.
(i) For odd n, there exists a unique matrix Z∈Mn such that Z∼Z.
(ii) For even n, a unique matrix Z∈Mn such that Z∼Z exists if and only if Z∈Mn′.
We refer to this Z as the normal form of Z.
Proof. We fix a matrix Z=(v1,v2,…,vn)∈Mn, where vi=(ai,bi)T(i=1,…,n). Let L1,L2 and R be given by
where Di,j=Di,j(Z). For simplicity, we write Z:=L2L1ZR. To satisfy the normalization conditions (15) for the first three columns of Mn, the values of λ1, λ2 and λ3 are uniquely determined as λ1=1/(μ2D1,2), λ2=1/(μ1D1,2), and λ3=1/(μ2D2,3). Furthermore, the conditions (15) for adjacent minors,
allow us to recursively solve for λk. For k=4,…,n, the general form of λk is given by:
λk=⎩⎨⎧μ21⋅D2,3D4,5⋯Dk−1,kD3,4D5,6⋯Dk−2,k−1,μ11⋅D1,2D3,4⋯Dk−1,kD2,3D4,5⋯Dk−2,k−1,(if k is odd),(if k is even).(30)
By substituting these values into (28), the matrix Z is explicitly written as:
Z=⎩⎨⎧0−110μ2D2,3μ1D1,31D1,2D3,4D2,3D1,4μ1D1,2D3,4μ2D2,3D2,4⋯⋯μ2D2,3D4,5⋯Dn−3,n−2Dn−1,nμ1D3,4D5,6⋯Dn−2,n−1D1,nD2,3D4,5⋯Dn−3,n−2Dn−1,nD3,4D5,6⋯Dn−2,n−1D2,n,0−110μ2D2,3μ1D1,31D1,2D3,4D2,3D1,4μ1D1,2D3,4μ2D2,3D2,4⋯⋯mr(Z)μ1D1,2D3,4⋯Dn−3,n−2Dn−1,nμ2D2,3D4,5⋯Dn−2,n−1D2,n,(if n is odd),(if n is even).(31)
where both cases satisfy the conditions in (15) except for the (1,n)-entry. If n is odd, we can set
to make the (1,n)-entry of Z equal to 1. If n is even, we require the condition mr(Z)=1, which means that Z∈Mn′. Furthermore, we set
μ2μ1=D1,3D2,3(33)
to make the (1,3)-entry of Z equal to 1, which means Z∈Mn.
We then address the uniqueness of the normal form. Suppose there exist two representatives Z1,Z2∈Mn (resp. Mn) such that Z∼Z1∼Z2. This implies Z2=LZ1R for some L∈GL2(K) and R=diag(λ1,…,λn). From the first and second columns of (15), we see that L must be a diagonal matrix diag(1/λ2,1/λ1). Since Z2∈Mn, we obtain the following n relations:
Moreover, if Z2∈Mn, we obtain the following relation from the condition on the (1,3)-entry:
1=λ2λ3.(36)
The equations (34) imply the recurrence λi+1=λ1λ2/λi, which yields
λ1=λ3=λ5=⋯,λ2=λ4=λ6=⋯.(37)
If n is odd, it follows from (37) that λn=λ1. Then, by (35), we have λ1=λ2=⋯=λn. On the other hand, if Z2∈Mn, then (36) yields λ2=λ3, and again by (37), we obtain λ1=λ2=⋯=λn. Therefore, in either case, L and R can be expressed as L=(1/λ)I and R=λI for some λ∈K×, which implies Z2=LZ1R=Z1. □
Let φodd:M2m+1→M2m+1 (resp. φeven:M2m′→M2m) be the map that assigns Z∈M2m+1 (resp. M2m′) to its normal form in the sense of Theorem 7. More explicitly, for Z∈M2m+1, we can write
z1,k=⎩⎨⎧Dk,k+1Dk+2,k+3⋯D2m−1,2mD1,2m+1Dk+1,k+2Dk+3,k+4⋯D2m,2m+1D1,kD1,2D3,4⋯Dk−3,k−2Dk−1,kD2,3D4,5⋯Dk−2,k−1D1,k(if k is odd),(if k is even),
z2,k=⎩⎨⎧D2,3D4,5⋯Dk−3,k−2Dk−1,kD3,4D5,6⋯Dk−2,k−1D2,kDk,k+1Dk+2,k+3⋯D2m−2,2m−2D2m,2m+1D1,2Dk+1,k+2Dk+3,k+4⋯D2m−1,2mD1,2m+1D2,k(if k is odd),(if k is even).(38)
and for Z′∈M2m′,
φeven(Z′)w1,kw2,k:=(0−11011w1,4w2,4⋯⋯w1,2m−1w2,2m−11w2,2m),=⎩⎨⎧D1,3D4,5⋯Dk−3,k−2Dk−1,kD3,4D4,5⋯Dk−2,k−1D1,kD1,2D3,4⋯Dk−3,k−2Dk−1,kD2,3D4,5⋯Dk−2,k−1D1,k(if k is odd),(if k is even),=⎩⎨⎧D2,3D4,5⋯Dk−3,k−2Dk−1,kD3,4D5,6⋯Dk−2,k−1D2,kD1,2D3,4⋯Dk−3,k−2Dk−1,kD1,3D4,5⋯Dk−2,k−1D2,k(if k is odd),(if k is even).(39)
Remark 8. A formula similar to the expressions for z1,k and z2,k in (38) has also appeared in the study of string theory (see [9], (2.10)).
Action of the symmetric group
Based on Theorem 7, the map φodd:M2m+1→M2m+1, which assigns each matrix Z to its unique normal form Z, is well-defined. For σ∈S2m+1, define ρ(σ):M2m+1→M2m+1 as
ρ(σ)(Z)=φodd(ZPσ),(40)
where Pσ is the permutation matrix corresponding to σ. We define a map ρ(σ):CCF2m+1→CCF2m+1 as
ρ(σ):=α−1∘ρ(σ)∘α.(41)
This action yields a representation of the symmetric group on CCF2m+1. In what follows, let σij denote the transposition (i,j) for i,j=1,…,2m+1.
Example 9. As an example, we consider the following F∈CCF5:
In the case of φeven:M2m′→M2m, the action (40) does not work in general; even if mr(Z′)=1 (i.e. Z′∈M2m′), the condition mr(Z′Pσ)=1 need not hold for every σ∈S2m. Define r and t as follows:
Let i:M2m↪M2m be the natural inclusion map defined by i(Z)=Z. We define an equivalence relation on M2m as follows: for Z1,Z2∈M2m, Z1∼matZ2 if and only if there exists c∈K× such that η(c)(Z1)=Z2. We introduce the quotient space M2m/∼mat and a projection map π:M2m′→M2m/∼mat. The following diagram illustrates these maps:
M2miM2mπM2m/∼mat.(49)
It follows that M2m/∼mat≅M2m. The action of ρ(σ,c) on M2m/∼mat is independent of c∈K×. It follows that ρ(σ,c) induces a representation of D2m on M2m/∼mat.
For m=3, the set CCF6 exhibits enhanced symmetry associated with the Weyl group W(B3) (≅W(C3)):
Proposition 11 (Affolter, George and Ramassamy [1], Lemma 4.2). Let Z be a matrix in M6′. For every permutation σ in S6 such that
σ(i+3)≡σ(i)+3(mod6)(1≤i≤6),(53)
we have mr(ZPσ)=1, i.e. ZPσ∈M6′.
Remark 12. An equivalent discussion can also be found in Proposition 2.1 of [3].
Since g1, g2, and g3 in (51) satisfy the condition (53), Proposition 11 implies that mr(ZPg)=1 holds for all Z∈M2m′ and g∈W(B3). Therefore, for m=3, the Weyl group W(B3), which is larger than the dihedral group D6, acts on M6/∼mat≅M6.
Conjecture 13. If m=3, the symmetry group of M2m coincides with the dihedral group D2m.
Using the computer algebra systems SageMath, Maxima, and GAP, we have verified that Conjecture 13 holds for m=2, 4, 5, and 6.
In what follows, we define the map ρˇ(σ,c):CCF2m→CCF2m by
ρˇ(σ,c):=α−1∘ρ(σ,c)∘α.(54)
The construction of ρˇ can be summarized as follows:
Remark 15. It is a natural direction to extend our matrix realization approach to SLk-frieze patterns by adopting the framework introduced by Surmann [16]. Representing SLk-frieze patterns via k×n matrices (or elements in Gr(k,n)) provides a foundation for analyzing actions of Sn and its subgroups. Generalizing our classification of symmetry groups to the SLk case remains an interesting topic for future research.
Acknowledgments
The authors acknowledge Yasuyuki Hatsuda, Masatoshi Noumi and Yoshihisa Saito for discussions on normal forms of elements in Mn. Thanks are also due to Rei Inoue, Takeyoshi Kogiso and Kyo Nishiyama for providing information on frieze patterns.
During the preparation of this work, the authors used OpenAI Prism to assist in the preparation and editing of this manuscript, and Google Gemini to improve the clarity and style of the English, as well as to double-check the mathematical computations and ensure accuracy. After using these tools, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.
Department of Mathematics, Graduate School of Science, Rikkyo University, Toshima-ku, Tokyo 171-8501, Japan.
[1]N.C. Affolter, T. George and S. Ramassamy, Integrable dynamics in projective geometry via dimers and triple crossing diagram maps on the cylinder, SIGMA 21 (2025), 040.DOI
[2]K. Aomoto, On the structure of integrals of power product linear functions, Sci. Papers. Gen. Univ. Tokyo 27 (1977), 49–61.DOI
[3]A.I. Bobenko and T. Hoffmann, Conformally symmetric circle packings: a generalization of Doyle’s spirals, Experiment. Math. 10 (2001), 141–150.DOI
[4]A.I. Bobenko, T. Hoffmann, and Yu.B. Suris, Hexagonal circle patterns and integrable systems: Patterns with the multi-ratio property and Lax equations on the regular triangular lattice, Internat. Math. Res. Notices 2002, 111–164.
[5]J.H. Conway and H.S.M. Coxeter, Triangulated polygons and frieze patterns, Math. Gaz. 57 (1973), 87–94.
[7]S. Fomin and A. Zelevinsky, Cluster algebras I: foundations, J. Amer. Math. Soc. 15 (2002), 497–529.DOI
[8]I.M. Gel’fand, M.I. Graev and V.S. Retakh, General hypergeometric systems of equations and series of hypergeometric type, Russian Math. Surveys, 47:4 (1992), 1–88.DOI
[9]Y. Hatsuda, K. Ito and Y. Satoh, T-functions and multi-gluon scattering amplitudes, J. High Energy Phys. 1202 (2012) 003.
[10]H. Kimura, Introduction to Hypergeometric Functions, Saieinsu-sha, 2007 [in Japanese].
[11]D.E. Knuth, The Art of computer programming, Vol. 1: fundamental Algorithms, Third Edition, Addison-Wesley Professional, 1997.
[12]S. Morier-Genoud, Coxeter’s frieze patterns at the crossroads of algebra, geometry and combinatorics, Bull. London Math. Soc. 47 (2015), 895–938.
[13]S. Morier-Genoud, Counting Coxeter’s friezes over a finite field via moduli spaces, Algebra. Comb. 4 (2021), 225–240.
[14]K. Nishiyama, Mathematics of frieze patterns — A glimpse of geometry and algebra, Kyoritsu Shuppan Co., Ltd., 2022 [in Japanese].