Functional Analysis and Its Applications, Vol. 38, No. 2, pp.

102–110, 2004
Translated from Funktsional
nyi Analiz i Ego Prilozheniya, Vol. 38, No. 2, pp. 28–37, 2004
Original Russian Text Copyright
c by O. N. Karpenkov

On Tori Triangulations Associated with Two-Dimensional

Continued Fractions of Cubic Irrationalities∗
O. N. Karpenkov
Received February 3, 2003

Abstract. The notion of equivalence of multidimensional continued fractions is introduced. We

consider some properties and state some conjectures related to the structure of the family of equiv-
alence classes of two-dimensional periodic continued fractions. Our approach to the study of the
family of equivalence classes of two-dimensional periodic continued fractions leads to revealing
special subfamilies of continued fractions for which the triangulations of the torus (i.e., the combi-
natorics of their fundamental domains) are subjected to clear rules. Some of these subfamilies are
studied in detail; the way to construct other similar subfamilies is indicated.
Key words: multidimensional continued fractions, convex hulls, integer operators, cubic extensions
of Q.

Introduction
Some properties of ordinary continued fractions have multidimensional analogs. H. Tsu-
chihashi [7] discovered the relationship between periodic multidimensional continued fractions and
multidimensional cusp singularities. The relationship between sails of multidimensional continued
fractions and Hilbert bases is described by J.-O. Moussafir in [6].
In his book [1] devoted to the theory of continued fractions, V. I. Arnold gives various images for
the sails of a two-dimensional continued fraction generalizing the golden section. In [5], E. I. Korkina
investigated the sales for the simplest two-dimensional continued fractions of cubic irrationalities
whose fundamental region consists of two triangles, three edges, and one vertex.
We consider the same model of multidimensional continued fractions. In the present paper we
obtain new examples of triangulations of tori for sails of two-dimensional continued fractions for
cubic irrationalities in some special families whose fundamental regions admit a more complicated
structure.
Definitions and notions we need are given in Sec. 1. In Sec. 2 we study properties of two-
dimensional continued fractions constructed from Frobenius operators and briefly discuss the rela-
tionship between the equivalence classes of triangulations of tori and cubic extensions of the field
of rational numbers. (A detailed study of both the properties of cubic extensions of the rationals
and of the classification of these extensions was carried out by B. N. Delone and D. K. Faddeev
in [3].) Examples of triangulations of tori thus obtained are discussed in Sec 3.
The author is grateful to Professor V. I. Arnold for permanent attention to the research and
for useful remarks.

1. Definitions
By integer points we mean the points of the space Rk (k
1) all of whose coordinates are
integers.
Consider a generic set of n + 1 hyperplanes passing through the origin in the space Rn+1 .
The complement to these hyperplanes consists of 2n+1 open orthants. Let us choose an arbitrary
orthant.
∗
Supported by INTAS-00-0259, NSh-1972.2003.1, and RFBR-01-01-00803 projects.

102 0016–2663/04/3802–0102
2004

c Plenum Publishing Corporation
 The boundary of the convex hull of all integer points but the origin that belong to the closure
of an orthant is called a sail.
By an n-dimensional continued fraction constructed from the given n + 1 hyperplanes in gen-
eral position in (n + 1)-dimensional space we mean the union of the 2n+1 sails defined by these
hyperplanes in the space Rn+1 .
Two n-dimensional sails (continued fractions) are said to be equivalent if there exists a linear
transformation of the (n + 1)-dimensional space preserving the integer lattice and taking one of
them to the other.
To construct the whole continued fraction up to equivalence in the one-dimensional case, it
suffices to know some integer characteristics of a single sail (namely, the integer lengths of the
edges and the integer angles between the consecutive edges of the sail).
Conjecture 1 (Arnold). There is a collection of integer characteristics of the sail from which
the continued fraction can be recovered uniquely up to equivalence.
Let A ∈ GL(n + 1, R) be an operator all of whose eigenvalues are real, distinct, and irrational.
Consider the n-dimensional spaces spanned by all possible subsets of n independent eigenvectors
of the operator A. Since these eigenvectors are linearly independent, the n + 1 hyperspaces thus
obtained are n + 1 hyperspaces in general position. A multidimensional continued fraction is con-
structed just from these hyperspaces.
Proposition 1.1. The continued fractions constructed from operators A and B in the group
GL(n + 1, R) with pairwise distinct real irrational eigenvalues are equivalent if and only if there
exists an integer operator X with unit determinant such that the operator A
obtained from A by
conjugating by the operator X commutes with B.
Proof. Let the continued fractions constructed from operators A and B in GL(n + 1, R) with
pairwise distinct real irrational eigenvalues be equivalent, i.e., there exists a linear transformation
of the space that preserves the integer lattice and takes the continued fraction of A to the continued
fraction of B (and the orthants of the first continued fraction to the orthants of the other). Under
such a transformation of the space, the operator A is conjugated by some integer operator X with
unit determinant. All eigenvalues of the operator A
thus obtained are pairwise distinct and real
(since the characteristic polynomial of the operator is preserved). Since the orthants of the first
continued fraction are taken to the orthants of the other, the sets of the eigendirections of the
operators A
and B coincide. Thus, the operators are simultaneously diagonalizable in some basis,
and hence commute.
Let us prove the converse assertion. Suppose that there exists an integer operator X with unit
determinant such that the operator A
obtained from A by the conjugation by X commutes with
B. Note that the eigenvalues of the operators A and A
coincide. Therefore, all eigenvalues of the
operator A
(as well as those of the operator B) are real, distinct, and irrational. Consider a basis in
which the operator A
is diagonal. A simple verification shows that the operator B is also diagonal in
this basis. Hence, the operators A
and B define the same decomposition of (n+1)-dimensional space
into orthants, and the continued fractions corresponding to these operators coincide. It remains to
note that a conjugation by an integer operator with unit determinant corresponds to a linear
transformation preserving the integer lattice of (n + 1)-dimensional space.
In what follows we consider only continued fractions constructed from some invertible integer
operator with integer inverse on (n + 1)-dimensional space. The set of such operators forms a group
denoted by GL(n + 1, Z). The group consists of the integer operators whose determinants are equal
to ±1.
The n-dimensional continued fraction constructed from an operator A ∈ GL(n + 1, Z) whose
characteristic polynomial over the field of rationals is irreducible and the eigenvalues are real is
referred to as the n-dimensional continued fraction of an (n + 1)-algebraic irrationality. The case
n = 1 (n = 2) corresponds to one-dimensional (two-dimesional ) continued fractions of quadratic
(cubic) irrationalities.

103
 Let the characteristic polynomial of an operator A be irreducible over the field of rationals and
let the roots of this polynomial be real and distinct. Under the action of the integer operators with
unit determinant that preserve a given sail, the sail is taken onto itself in a one-to-one way. These
operators form an Abelian group. It follows from the Dirichlet unit theorem (see [2]) that this group
is isomorphic to Zn and its action is free. The quotient of a sail by this group action is isomorphic to
the n-dimensional torus. (For the converse assertion, see [4] and [7].) The polyhedron decomposition
of the n-dimensional torus is defined in the natural way, and the affine types of the polyhedrons
are also defined (we include the number and mutual arrangement of the integer points on the faces
of the polyhedron into the notion of affine type). For two-dimensional continued fractions of cubic
irrationalities, such a decomposition is usually called a triangulation of the torus.
By a fundamental region of a sail we mean the union of some faces that contains exactly one
face in each equivalence class.

2. Conjugacy Classes of Two-Dimensional Continued Fractions

of Cubic Irrationalities
Two-dimensional continued fractions of cubic irrationalities constructed from operators of the
form A and −A coincide. Thus, the study of continued fractions of integer operators whose deter-
minants are equal to ±1 is reduced to the study of continued fractions of integer operators with
unit determinant (i.e., of the operators in the group SL(3, Z)).
An operator (matrix) with unit determinant of the form
 
0 1 0
Am,n := 0 0 1 ,
1 −m −n
where m and n are arbitrary integers, is called a Frobenius operator (matrix ). Note that, if the
characteristic polynomial χAm,n (x) is irreducible over the field Q, then the matrix of the operator
 
of left multiplication by an element x in the natural basis {1, x, x2 } over the field Q[x] χAm,n (x)
coincides with the matrix Am,n .
Let an operator A ∈ SL(3, Z) have pairwise distinct real irrational eigenvalues. Let e1 be an
integer nonzero vector and let e2 = A(e1 ) and e3 = A2 (e1 ). Then the matrix of the operator in the
basis (e1 , e2 , ce3 ) is a Frobenius matrix for some rational number c. However, the transition matrix
can be noninteger here, and the corresponding continued fraction is not equivalent to the initial
one.
Example 2.1. The continued fraction constructed from the operator
 
1 2 0
A =  0 1 2 ,
−7 0 29
is not equivalent to the continued fraction constructed from any Frobenius operator with unit
determinant.
In this connection, the following question is of interest: How frequent are continued fractions
that do not correspond to Frobenius operators?
In any case, the family of Frobenius operators has useful properties that enable one to simul-
taneously construct large families of nonequivalent two-dimensional periodic continued fractions,
which is very important by itself.
One can readily obtain the following statements.
Proposition 2.1. The set Ω of operators Am,n all of whose eigenvalues are real and pairwise
distinct is defined by the inequality n2 m2 − 4m3 + 4n3 − 18mn − 27  0. For the eigenvalues of the
operators of the set to be irrational, it is necessary to additionally exclude two perpendicular lines
in the integer plain, namely, Aa,−a and Aa,a+2 , a ∈ Z.
104
 Proposition 2.2. The two-dimensional continued fractions of the cubic irrationalities con-
structed from the operators Am,n and A−n,−m are equivalent.
Below we consider all statements modulo this symmetry.
Remark. Example 2.3 below shows that there are equivalent continued fractions among the
periodic continued fractions constructed from operators in the set Ω.
Note that there are nonequivalent two-dimensional periodic continued fractions constructed
from operators in GL(3, Z) whose characteristic polynomials define isomorphic extensions of the
field of rationals. In the following example we present operators with equal characteristic polyno-
mials and distinct continued fractions.
Example 2.2. The operators (A−1,2 )3 and A−4,11 have distinct two-dimensional continued
fractions (although their characteristic polynomials coincide).
On the other hand, equivalent periodic continued fractions can correspond to operators with
distinct characteristic polynomials.
Example 2.3. The operators A20,−a and A−2a,−a2 are conjugate by an operator in GL(3, Z),
and hence the periodic continued fractions (including the triangulations of the torus) corresponding
to the operators A0,−a and A−2a,−a2 are equivalent.
Note that to distinct cubic extensions of the field Q there correspond nonequivalent periodic
continued fractions.

3. Triangulations of the Tori and Fundamental Regions for Some Series

of Operators Am,n
We evaluate here the triangulations of the tori and fundamental domains for several infinite
series of Frobenius operators. In this section we treat only sails whose convex hull contains the
point (0, 0, 1).
By the integer k-dimensional volume of an integer k-dimensional polyhedron in the n-dimensio-
nal space (the integer length of a segment for k = 1 and the integer area of a polygon for k = 2)
one means the ratio of the Euclidean volume of this polyhedron to the Euclidean volume of the
minimal simplex in the same k-dimensional subspace.
By the integer distance from an integer hyperplane (containing an (n − 1)-dimensional integer
sublattice) to an integer point one means the ratio of the corresponding Euclidean distance to the
Euclidean distance from the hyperplane to the nearest integer point outside this hyperplane.
By the integer angle between two integer rays (i.e., rays containing more than one integer point)
with the vertex at the same integer point one means the value S(u, v)/(|u| · |v|), where u and v are
arbitrary integer vectors passing along the rays and S(u, v) is the integer volume of the triangle
with the edges u and v.
Remark. Our integer volume is an integer (the value obtained when measuring the volume
by parallelepipeds in the standard way is k! times less). The integer k-dimensional volume of the
simplex is equal to the index of the subgroup of the lattice generated by the edges of this simplex
that have a common vertex.
Since the integer angles of any triangle all of whose vertices are integer can be recovered uniquely
from the integer side lengths of the triangle and from the integer area, we do not write out the
integer angles of the triangles below.
Conjecture 2. The invariants listed above distinguish all nonequivalent continued fractions of
cubic irrationalities.
In the statements of Propositions 3.1–3.5 one speaks only of the homeomorphic type of tri-
angulations of the tori; however, the description of the fundamental regions that enables one to
evaluate any other invariant, including the affine types of the faces, is given in the proofs. (For
example, we find integer volumes and distances to faces in Propositions 3.1 and 3.2.) Examples of
affine structure of all faces of a triangulation are shown in the figures.
105
 Proposition 3.1. Let m = b − a − 1 and n = (a + 2)(b + 1) (a, b
0). In this case, the torus
triangulation corresponding to the operator Am,n is homeomorphic to a triangulation of the form

(in the figure one has b = 6).

Proof. The operators

Xa,b = A−2
m,n , Ya,b = A−1 −1
m,n (Am,n − (b + 1)I)

commute with the operator Am,n and do not permute the sails (note that the operator Am,n
permutes the sails). Here I is the identity element of the group SL(3, Z).
Let us describe the closure of one of the fundamental regions obtained by factoring a sail by the
operators Xa,b and Ya,b . Consider the points A = (1, 0, a + 2), B = (0, 0, 1), C = (b − a − 1, 1, 0),
and D = ((b + 1)2 , b + 1, 1) of the sail that contains the point (0, 0, 1). Under the action of the
operator Xa,b , the segment AB is taken onto the segment DC (the point A comes to the point D
and B to C ). Under the action of the operator Ya,b , the segment AD is taken onto the segment
BC (the point A comes to the point B and D to C ). The integer points ((b + 1)i, i, 1), where
i ∈ {1, . . . , b}, belong to the segment BD.
One can readily see that the integer lengths of the segments AB, BC , CD, DA, and BD are
equal to 1, 1, 1, 1, and b + 1, respectively; the integer areas of both the triangles ABD and BCD
are equal to b + 1. The integer distances from the origin to the planes containing the triangles ABD
and BCD are equal to 1 and a + 2, respectively.
The operators Xa,b and Ya,b take the sail onto itself because all their eigenvalues are positive
(or, which is equivalent here, the values of their characteristic polynomials on the negative semi-
axis are negative). Moreover, these operators are the generators of the group of integer operators
mapping the sail onto itself because, when passing to the quotient of the sail by the action of these
operators, one obtains the fundamental region having only one vertex (a zero-dimensional face),
and hence the fundamental region has no lesser subperiod.
Let us show that all vertices of a fundamental region of an arbitrary sail of a periodic continued
fraction can always be chosen in the closed convex hull of the following points: the origin, A, X(A),
Y (A), and XY (A), where A is an arbitrary zero-dimensional face of the sail and the operators X
and Y are the generators of the group of integer operators taking the sail onto itself.
Consider a tetrahedral angle with the vertex at the origin and with the edges passing through the
points A, X(A), Y (A), and XY (A). The union of all images of this angle under the transformations
of the form X m Y n , where m and n are integers, covers the whole interior of the orthant. Hence,
all vertices of the sail can be obtained by shifting the vertices of the sail lying in our tetrahedral
angle by using the operators of the form X m Y n . The convex hull of the integer points of the form
X m Y n (A) is contained in the convex hull of all integer points of the given orthant. Therefore, the
boundary of the convex hull of all integer points of the orthant is contained in the complement
to the interior points of the convex hull of the integer points of the form X m Y n (A). In turn, the
complement is contained in the union of all images of the convex hull of the points the origin,
A, X(A), Y (A), and XY (A) under the transformations of the form X m Y n , where m and n are
integers.
It is obvious that all points of the polyhedron thus constructed, except for the origin, belong
to the open orthant under consideration.
Proposition 3.2. Let m = −a and n = 2a + 3 (a
0). In this case, the torus triangulation
corresponding to the operator Am,n is homeomorphic to a triangulation of the form
106
 Proof. Let us choose the following generators of the subgroup of integer operators taking the
sail onto itself:

Xa = A−2
m,n , Ya = (2I − A−1 −1
m,n ) .

As in the previous case, consider the closure of one of the fundamental regions of the sail
(containing the point (0, 0, 1)) that are obtained when factoring by the operators Xa and Ya .
Let A = (0, 0, 1), B = (2, 1, 1), C = (7, 4, 2), and D = (−a, 1, 0). Along with these points, the
fundamental region contains the vertex E = (3, 2, 1). Under the action of the operator Xa , the
segment AB comes onto the segment DC (the point A is taken to the point D and B to C ).
Under the action of the operator Ya , the segment AD comes onto the segment BC (the point A
is taken to the point B and D to C ).
If a = 0, then the integer lengths of the sides AB, BC , CD, and DA are equal to 1, and
the integer areas of the triangles ABD and BCD are equal to 1 and 3, respectively. The integer
distances from the origin to the planes containing the triangles ABD and BCD are equal to 2 and
1, respectively.
If a > 0, then all integer lengths of the sides and the integer areas of all four triangles are equal
to 1. The integer distances from the origin to the planes containing the triangles ABD, BDE ,
BCE , and CED are equal to a + 2, a + 1, 1, and 1, respectively.
Here and below, the proofs of the statements on the generators are similar to those of the
corresponding statements in the proof of Proposition 3.1.
Proposition 3.3. Let m = 2a − 5 and n = 7a − 5 (a
2). In this case, the torus triangulation
corresponding to the operator Am,n is homeomorphic to a triangulation of the form

(in the figure one has a = 5).

Proof. Let us choose the following generators of the subgroup of integer operators taking the
sail onto itself:

Xa = 2A−1
m,n + 7I, Ya = A2m,n .

Let us consider the closure of a fundamental region of the sail (containing the point (0, 0, 1))
among the regions obtained when factoring by the operators Xa and Ya . Let A = (−14, 4, −1),
B = (−1, 1 − a, 7a2 − 10a + 4), C = (1, 5 − 7a, 49a2 − 72a + 30), and D = (0, 0, 1). Under the
action of the operator Xa , the segment AB comes onto the segment DC (the point A is taken
to the point D and B to C ). Under the action of the operator Ya , the segment AD comes onto
the segment BC (the point A is taken to the point B and D to C ). Along with these points, the
fundamental region contains the vertices E = (−1, 0, 2a − 1) and F = (0, −a, 7a2 − 5a + 1). The
interval BE contains a − 2 integer points, the interval DF contains a − 1, and each of the intervals
AD and CB contains one integer point.
Proposition 3.4. Let m = a − 1 and n = 3 + 2a (a
0). In this case, the torus triangulation
corresponding to the operator Am,n is homeomorphic to a triangulation of the form
107
(in the figure one has a = 4).
Proof. Let us choose the following generators of the subgroup of integer operators taking the
sail onto itself:
Xa = (2I + A−1 −2
m,n ) , Ya = A−2
m,n .

Let us take the closure of one of the fundamental regions of the sail (containing the point
(0, 0, 1)) that are obtained when factoring by the operators Xa and Ya . Let A = (1, −2a − 3, 4a2 +
11a + 10), B = (0, 0, 1), C = (−4a − 11, 2a + 5, −a − 2), and D = (−a − 2, 0, a2 + 3a + 3). Along with
these points, the fundamental region contains the vertices E = (−2, 1, 0), F = (−2a − 3, a + 1, 1),
and G = (0, −1 − a, 2a2 + 5a + 4). Each of the intervals BG and DF contains a integer points. In
the interior of the pentagon BEF DG one has (a + 1)2 integer points of the form (−j, −i + j, (2a +
3)i − (a + 2)j + 1), where 1  i  a + 1 and 1  j  2i − 1. Under the action of the operator Xa ,
the segment AB comes onto the segment DC (the point A is taken to the point D and the point
B to C ). Under the action of the operator Ya , the broken line AGD comes onto the broken line
BEC (the point A is taken to the point B, the point G is taken to the point E , and the point D
to the point C ).
Proposition 3.5. Let m = −(a + 2)(b + 2) + 3 and n = (a + 2)(b + 3) − 3 (a
0, b
0). In this
case, the torus triangulation corresponding to the operator Am,n is homeomorphic to a triangulation
of the form

(in the figure one has b = 5).

Proof. Let us choose the following generators of the subgroup of integer operators taking the
sail onto itself:
Xa,b = ((b + 3)I − (b + 2)A−1 −2
m,n )Am,n , Ya,b = A−2
m,n .

Let us take the closure of one of the fundamental regions of the sail (containing the point
(0, 0, 1)) that are obtained when factoring by the operators Xa,b and Ya,b . Let A = (b2 + 3b +
3, b2 + 2b − a + 1, a2 b + 3a2 + 4ab + b2 + 6a + 5b + 4), B = (b2 + 5b + 6, b2 + 4b + 4, b2 + 3b + 3),
C = (−ab − 2a − 2b − 1, 1, 0), and D = (0, 0, 1). The interval BD contains b + 1 integer points.
Along with these points, the vertices E = (b + 4, b + 3, b + 2), F = (b + 2, b + 1, a + b + 2), and
G = (1, 1, 1) also belong to the fundamental region. Under the action of the operator Xa,b , the
segment AB comes onto the segment DC (the point A is taken to the point D and the point B to
the point C ). Under the action of the operator Ya,b , the broken line AF D comes onto the broken
line BEC (the point A is taken to the point B, the point F is taken to the point E , and the point
D to the point C ).
Note that the generators of the subgroup of operators that commute with the operator Am,n
and do not permute the sails can be expressed by using the operators Am,n and αI + βA−1 m,n , where
α and β are nonzero integers.
It turns out that the following assertion holds in the general case: the determinants of the
−1
matrices of the operators αI + βA−1 m,n and αI + βAm+kβ,n+kα are equal. In particular, if the
absolute value of the determinant of the matrix of the operator αI + βA−1 m,n is equal to one, then

108
the absolute value of the determinant of the matrix of the operator αI + βA−1
m+kβ,n+kα is also equal
to one for an arbitrary integer k.
Seemingly, for the other sequences of operators Am0 +βs,n0 +αs with s ∈ N (along with those
considered in Propositions 3.1–3.5), the fundamental domains have much in common (for example,
the number of polygons and their types).
Note that the numbers α and β for such sequences satisfy the following interesting property.
Since
|αI + βA−1
m,n | = α + α βm − αβ n + β ,
3 2 2 3

it follows that integer numbers m and n for which |α3 + α2 βm − αβ 2 n + β 3 | = 1 exist if and only
if either α3 − 1 is divisible by β and β 3 − 1 is divisible by α or α3 + 1 is divisible by β and β 3 + 1
is divisible by α.
For instance, the pairs (α, β) with 10
α
β
−10 having this property (along with those
described in Propositions 3.1–3.5) are (3, 2), (7, −2), (9, −2), (9, 2), (7, −4), (9, 4), (9, 5), and
(9, 7).

Fig. 1. Triangulations of the torus for operators Am,n

In conclusion we present the table whose cells contain triangulations constructed in this paper
for the sails whose convex hull contains the point with the coordinates (0, 0, 1), see Fig. 1. The
triangulation of the torus for the sail of the two-dimensional continued fraction of a cubic irrational-
ity constructed from the operator Am,n is shown in the cell placed at the intersection of the row
with index n and the column with index m. If one of the roots of characteristic polynomial of the
operator is equal to 1 or −1, then we mark the cell (m, n) with the sign ∗ or #, respectively. The
109
cells corresponding to the operators whose characteristic polynomial has two complex-conjugate
roots are painted with the light gray color.

References
1. V. I. Arnold, Continued Fractions [in Russian], MCCME, Moscow, 2002.
2. Z. I. Borevich and I. R. Shafarevich, Number Theory [in Russian], Moscow, 1985; English transl.:
Academic Press, New York–London, 1966.
3. B. N. Delone and D. K. Faddeev, The theory of irrationalities of the third degree, Trudy Mat.
Inst. Steklov., 11, 1940; English transl.: Amer. Math. Soc., Providence, R.I., 1964.
4. E. I. Korkina, “La périodicité des fractions continues multidimensionelles,” C.R. Acad. Sci. Paris,
319, No. 8, 777–780 (1994).
5. E. I. Korkina, “Two-dimensional continued fractions. The simplest examples,” Trudy Mat. Inst.
Steklov, 209, 143–166 (1995).
6. J.-O. Moussafir, “Sails and Hilbert bases,” Funkts. Anal. Prilozhen., 34, No. 2, 43–49 (2000),
7. H. Tsuchihashi, “Higher-dimensional analogues of periodic continued fractions and cusp singu-
larities,” Tohoku Math. J., 35, No. 4, 607–639 (1983).

Moscow State University,

Department of Mathematics and Mechanics
e-mail: karpenk@mccme.ru

Translated by O. N. Karpenkov

110