KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS,

AND ARITHMETICITY

NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

A BSTRACT. In this paper we study crystallographic sphere packings and Kleinian sphere
arXiv:2203.01973v1 [math.GT] 3 Mar 2022

packings, introduced first by Kontorovich and Nakamura in 2017 and then studied further by
Kapovich and Kontorovich in 2021. In particular, we solve the problem of existence of crystal-
lographic sphere packings in certain higher dimensions posed by Kontorovich and Nakamura.
In addition, we present a geometric doubling procedure allowing to obtain sphere packings
from some Coxeter polyhedra without isolated roots, and study “properly integral” packings
(that is, ones which are integral but not superintegral).

§ 1. Introduction
1.1. Sphere packings. For n ≥ 2, let Hn+1 be the ( n + 1)–dimensional hyper-
bolic space. Then the set of all ideal points of hyperbolic space, or its ideal
boundary, is ∂Hn+1 ≈ Sn and can be identified topologically with Rn := Rn ∪{∞}.
One can consider different coordinate systems in Sn , and thus some of the
definitions below depend on this choice of coordinates.
Definition 1.1. By a sphere packing (or just packing) P of Sn we mean an
infinite collection {Bα } of round balls in Sn so that:
(1) The interiors of the balls are disjoint, and
(2) The union of the balls is dense in Sn .
Recall that, given a packing P = {Bα } of Sn ≈ ∂Hn+1 , every sphere S α = ∂Bα
also is the boundary of a hyperplane Hα ∼ = Hn of Hn+1 , i.e., S α = ∂∞ Hα . Any
such hyperplane Hα is associated with a reflection R α ∈ Isom(Hn+1 ), i.e., Hα
is the mirror of the reflection R α . Let ΓR be the group generated by the
reflections in these hyperplanes, ΓR := 〈R α | Bα ∈ P 〉.
Definition 1.2. The superpacking of P is its orbit under the ΓR –action:
f = ΓR · P .
P
Bogachev was partially supported by the Russian Science Foundation, grant no. 21-41-09011.
Kolpakov was partially supported by the Swiss National Science Foundation, project no. PP00P2-202667 .
Kontorovich was partially supported by NSF grant DMS-1802119, BSF grant 2020119, and the 2020-2021
Distinguished Visiting Professorship at the National Museum of Mathematics.
1
2 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

Definition 1.3. The bend of a ball Bα ∈ P is the inverse of its (signed)

radius, which is defined after choosing an identification Sn = Rn ∪ {∞}. A
packing is integral if there is a choice of identification such that all balls Bα
have integer bends. A packing is superintegral if its superpacking has all
integer bends. We say a packing is properly integral if it is integral but not
superintegral.
Definition 1.4. A packing P is Kleinian if its set of accumulation points
is the limit set of some geometrically finite discrete group ΓS < Isom(Hn+1 ).
We call such a ΓS a symmetry group of P .
Definition 1.5. Let P = {Bα } be a Kleinian sphere packing with a symme-
try group ΓS < Isom(Hn+1 ). The supergroup of P is the group Γ e = ΓR o ΓS
generated by the symmetry group ΓS and all reflections R α .
It is a theorem that Γ
e is a lattice, that is, acts on Hn+1 with finite covolume.

Definition 1.6. A packing is crystallographic if it has a reflective symmetry

group. That is, the symmetry group, and hence the supergroup, is generated
by reflections in codimension 1 facets.
1.2. Outline and main theorems. The main focus of this paper is on deter-
mining the dimensions in which superintegral crystallographic sphere pack-
ings exist. It is worth mentioning that Kleinian sphere packings exist in all
dimensions, even those that are superintegral [K2].
The existence of a superintegral crystallographic packing in Sn , n ≥ 2, im-
plies that there exists a reflective, nonuniform, arithmetic lattice of simplest
type defined over Q that acts on Hn+1 . (That it is reflective, and arithmetic
over Q follows directly from the Structure Theorem and Arithmeticity Theo-
rem in [KN]. That it is nonuniform is proved in [K2].) A result of Esselmann
[E] implies that no such lattice is possible for n ≥ 21. Previously, superinte-
gral crystallographic sphere packings were known in dimensions n ≤ 13 and
n = 17 by applying the techniques of [KN]. We also remark that the existence
of such a packing in dimension n = 18 is claimed in [KN], although a further
argument is needed (given below in Lemma 1.8) to substantiate this claim.
The difficulty in producing a crystallographic sphere packing stems from
the Structure Theorem of [KN]. There it is shown that a finite covolume re-
flection group Γe gives rise to a sphere packing of Sn if and only if its Coxeter-
Vinberg diagram (see §2) has at least one “isolated root,” that is, a root which
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 3

is orthogonal to all its neighbors. In terms of the Coxeter-Vinberg diagram,

this means that the corresponding (isolated) vertex is connected to its neigh-
bors only by dotted (common perpendiculars) or bold edges (parallel, i.e. tan-
gent at the ideal boundary ∂Hn+1 ).
Our main result about dimensions of crystallographic sphere packings is
as follows.
Theorem A. There are no crystallographic sphere packings in Sn for n ≥ 19.
The theorem above is implied by the results of Esselman [E], together with
the following fact.
Proposition 1.7. If a superintegral crystallographic sphere packing exists
in Sn , then there exists a non-uniform arithmetic reflective lattice acting on
Hn .
As for the existence of crystallographic sphere packings, we can summarise
our findings together with the previous results of [KN], in order to formulate
the following statement.
Theorem B. Superintegral crystallographic sphere packings in Sn exist
only for n ≤ 18. Examples are known for all 2 ≤ n ≤ 14, and n = 17, 18.
What is new here is the addition of dimensions n = 14 and n = 18. We
construct these additional examples in higher dimensions by studying the
lower–dimensional faces of the 21–dimensional hyperbolic polyhedron dis-
covered by Borcherds [Bor].
The proof of Theorem B relies on the following fact that allows to combine
Coxeter polyhedra with many “even” angles into one that has a facet orthog-
onal to all its neighbours.
Lemma 1.8. Let P be a finite volume Coxeter polyhedron in Hn and P0 be a
facet that meets other facets of P at angles π/2 or π/4. Let P1 , . . . P k be all the
facets of P that meet P0 at π/4. Assume that each P j meets its neighbors at
angles π/2m , with m ≥ 1, an integer. Then the reflection group of P is commen-
surable to the supergroup of a crystallographic packing.
Dimensions n = 15 and 16 are still missing from our consideration: the gap
seems non–trivial to fill, as at present, there are rather few candidate arith-
metic lattices in high dimensions to explore for packings. For example, we
4 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

studied another polyhedron discovered by Borcherds [Bor2] in H17 (with 960

facets), only to discover that it has no isolated roots, and no Coxeter facets
either. The present methods of obtaining sphere packings from polyhedra
([KN] and Lemma 1.8 loc. cit.) cannot be applied to the polyhedron itself. We
would like to thank Daniel Allcock for his invaluable help in our efforts to
transfer this example into a computer–readable format.

Question 1.9. Do there exist crystallographic sphere packings in Sn for

n = 15 and 16?

We should stress the fact that answering the above question in the nega-
tive even in the more narrow superintegral setting will require a complete
classification of reflective quadratic Lorentzian forms over Q, which seems
rather difficult at the moment.
Another interesting direction to pursue is the possible connection between
arithmetic reflection groups and sphere packings. Namely, Kontorovich and
Nakamura [KN] asked the following question:

Question 1.10. Is it true that every arithmetic non–uniform lattice of sim-

plest type defined over Q is commensurable to the supergroup of a superinte-
gral crystallographic packing?

The answer turns out to be affirmative in low dimensions: here we confirm

it for n = 2, 3, 4 by using the classification results of Scharlau, Walhorn, and
Turkalj.

Theorem C. Any reflective arithmetic non–uniform lattice Γ < Isom Hn+1 ,

n = 2, 3, 4, of simplest type defined over Q is commensurable to the supergroup
of a crystallographic packing in Sn .

For n = 2, Kontrovich and Nakamura [KN] confirmed the same for the fi-
nite list of reflective extended Bianchi groups. However, the Borcherds poly-
hedron in dimension 21 provides a clear counterexample.

Acknowledgements. The authors would like to thank Daniel Allcock for

stimulating discussions.
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 5

§ 2. Preliminaries
2.1. Hyperbolic Lobachevsky space and convex polyhedra. Let Rn,1 be
the ( n+1)-dimensional real Minkowski space equipped with the inner product

( x, y) = − x0 y0 + x1 y1 + . . . + xn yn

of signature ( n, 1). A vector model of the n-dimensional hyperbolic Lobachevsky

space Hn is the above component of the standard hyperboloid lying in the fu-
ture light cone:
Hn = { x ∈ Rn,1 | ( x, x) = −1, x0 > 0}.
The points of Hn are called proper points. The points at infinity (or on the
boundary ∂Hn ) in this model correspond to isotropic one-dimensional sub-
spaces of Rn,1 , that is, rays {λ x | λ ≥ 0}, for x ∈ Rn,1 such that ( x, x) = 0.
The hyperbolic metric ρ is given by

cosh ρ ( x, y) = −( x, y).

Let On,1 (R) be the group of orthogonal transformations of the space Rn,1 ,
and let POn,1 (R) be its subgroup of index 2 preserving Hn . The isometry group
of the hyperbolic n-space Hn is Isom(Hn ) ∼= POn,1 (R).
Suppose that e ∈ R is a non–zero vector. Then the set
n,1

H e = { x ∈ Hn | ( x, e) = 0}

is a hyperplane in Hn and it divides the entire space into the half-spaces

H −e = { x ∈ Hn | ( x, e) ≤ 0}, H +e = { x ∈ Hn | ( x, e) ≥ 0}.

The orthogonal transformation given by the formula

( e, x)
R e ( x) = x − 2 e,
( e, e)
is called the reflection in the hyperplane H e , which is called the mirror of R e .
All the facts stated here for the standard Lorentzian form hold also for
any inner product form f ( x, y) of signature ( n, 1), once we put ( x, y) = f ( x, y).
This makes some of our practical computations easier rather than each time
passing to the standard form.
6 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

Definition 2.1. A convex polyhedron in Hn is the intersection of finitely

many half-spaces that has non-empty interior. A generalized convex polyhe-
dron is the intersection (with non-empty interior) of a family (possibly, infi-
nite) of half-spaces such that every ball intersects only finitely many of their
boundary hyperplanes.
Definition 2.2. A generalized convex polyhedron is said to be acute-angled
if all its dihedral angles do not exceed π/2. A generalized convex polyhedron
is called a Coxeter polyhedron if all its dihedral angles are of the form π/ k,
where k ∈ {2, 3, 4, . . . , +∞}.
Definition 2.3. The Coxeter-Vinberg diagram of a Coxeter polyhedron is a
graph having a vertex corresponding to each facet, and vertices attached by
edges of multiplicity k if the corresponding facets meet with dihedral angle
π/( k + 2). This includes the following interpretations:
• k = ∞, in which case the edge is bold and the facets meet at the ideal
boundary,
• k = 0, in which case there is no edge, corresponding to the facets being
orthogonal, and
• k imaginary, in which case the edge is dotted, corresponding to facets
which diverge at infinity.
It is known that the fundamental domains of discrete reflection groups are
generalized Coxeter polyhedra (see [Vin67, Vin85]).
A convex polyhedron has finite volume if and only if it is equal to the convex
hull of finitely many points of the closure Hn = Hn ∪ ∂Hn . If a polyhedron is
compact, then it is a convex hull of finitely many proper points in Hn .
It is also known that compact acute-angled hyperbolic polyhedra, in partic-
ular, compact Coxeter polyhedra in Hn , are simple, i.e. each of their vertices
belongs to exactly n facets (and n edges).
2.2. Kleinian sphere packing and the Structure Theorem. In this sub-
section we discuss some details regarding Kleinian sphere packing and com-
putational aspects of working with them.
In practice this works as follows. We have a quadratic form f of signature
( n + 1, 1) which defines a quadratic space V = Rn+1,1 . This gives us the two–
sheeted convex cone C := { x ∈ V : f ( x, x) < 0}, and its projectivization: PC :=
C /R+ . Then we identify one sheet PC + of this projectivized two–sheeted cone
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 7

with Hn+1 , and use the inner product ( x, y) = f ( x, y) to induce the hyperbolic
metric on it. Here we use the same notation for the quadratic form and the
induced bilinear form as it does not create much ambiguity.
Given a vector e α ∈ {v ∈ V : (v, v) > 0}, we have the hyperplane Hα = e⊥ α ∩P C + ,
( x,e α )
and a reflection R α : x 7→ x − 2 ( e α ,e α ) e α with mirror Hα . This reflection induces
an inversion in the sphere S α = ∂ Hα .
Dual to V is the vector space V ∗ which is a quadratic space under f ∗ . Then
to construct an inversive coordinate system, we simply proceed as follows.
To choose coordinates, we first choose any two f ∗ –isotropic (over R) vectors
b, b̂ ∈ V ∗ , such that f ∗ ( b, b̂) = −2.
Then take any system of orthonormal vectors b j , j ∈ {1, . . . , n} in the space
orthogonal to that generated by b and b̂. The inversive coordinate system of
a sphere S α corresponding to the vector e α ∈ {v ∈ V : (v, v) > 0} is then given
by
vα = ( b( e α ), b̂( e α ), b 1 ( e α ), . . . , b n ( e α )).
This vector vα contains all of the data of the root e α . Indeed, consider the
sphere S α in Sn corresponding to the ideal boundary of the hyperplane Hα
which is f −orthogonal to e α ; the “bend” (inverse radius) of the sphere is b( e α )
and its co-bend is b̂( e α ). Moreover, its center is ( b 1 ( e α ), . . . , b n ( e α ))/ b( e α ). If the
bend b( e α ) = 0, that is, S α is a hyperplane, then this “center” is re-interpreted
to mean the limit as spheres converge to this hyperplane; in that case, this
ratio becomes the (Euclidean) unit normal to this hyperplane in ∂Hn+1 .
Theorem 2.4 (Structure Theorem [KN, K2]).
(1) Let P be a Kleinian packing of Sn ' ∂Hn+1 . Then its supergroup Γ e is a
lattice in Isom(Hn+1 ).
(2) Suppose that Γ e < Isom(Hn+1 ) is a lattice (containing at least one reflec-
tion) with a convex fundamental polyhedron D and a minimal generat-
ing set Se for Γ
e , where elements of Se correspond to face-pairing transfor-
mations of D . Let R ⊂ Se be a non-empty set of generators corresponding
to reflections γα in certain facets Fα , α ∈ A , of D . For each α ∈ A let
S α = ∂∞ Hα ⊂ ∂Hn+1 = Sn be the invariant sphere for the reflection γα
with respect to the supporting hyperplane Hα of the facet Fα (see Fig. 1).
Assume that all the hyperplanes Hα are pairwise disjoint or parallel to
each other, and that, if they meet other supporting hyperplanes of facets
of D , then they do so orthogonally. Let S = Se \ R and let Γ = 〈S 〉.
8 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

F IGURE 1. A facet Fα of a polyhedron P has supporting hyperplane Hα .

Then the orbit of {S α | α ∈ A } under Γ is a Kleinian packing of Sn with

symmetry group Γ and supergroup Γ e.

We remark that there is an even simpler description of the above construc-

tion in the case of crystallographic sphere packings. Suppose as above that
Γ
e is generated by a set Se of reflections, with Coxeter-Vinberg diagram with
vertices Se. Suppose Se can be partitioned into a “cluster” R and “cocluster”
S = Se \ R , with the following properties: (i) Any pair of vertices of R are ei-
ther connected by a dashed edge or a bold edge, and (ii) any vertex of R and
a vertex of S are either connected as above, or disconnected (that is, orthogo-
nal). Then, as in the Structure Theorem, the orbit of the spheres in R under
the action of the group Γ generated by inversions in S is a crystallographic
packing. (And moreover, every such arises in this way.)

2.3. Vinberg’s arithmeticity criterion. In order to understand if a given

lattice generated by reflections is arithmetic or not, we will use a handy cri-
terion developed by Vinberg [Vin67]. Let H −e = { x ∈ Rn,1 | ( x, e) ≤ 0} be the
half-space associated to a hyperplane H e with normal e 6= 0. If P is a Coxeter
polyhedron in Hn , such that
N
H −e j ,
\
P=
j =1
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 9

then let G (P ) = { g i j }N N
i, j =1 = {( e i , e j )} i, j =1 denote its Gram matrix, and let the
field K = Q({ g i j }N
i, j =0 ) be generated by its entries.
The set of all cyclic products of entries of a matrix A (i.e. the set consist-
ing of all possible products of the form a i 1 i 2 a i 2 i 3 . . . a i k i 1 ) will be denoted by
Cyc( A ). Let k = Q(Cyc(G (P ))) ⊂ K , and let O k be the ring of integers of k.
Given the lattice Γ generated by reflections in the facets of a Coxeter poly-
hedron P , we can determine if Γ is arithmetic, quasi–arithmetic, or neither.
Theorem 2.5 (Vinberg’s arithmeticity criterion [Vin67]). Let Γ be a reflec-
tion group acting on Hn with finite volume fundamental Coxeter polyhedron
P . Then Γ is arithmetic if and only if the following conditions hold:
(1) K is a totally real algebraic number field;
(2) for any embedding σ : K → R, such that σ |k 6= id, G σ (P ) is positive semi-
definite;
(3) Cyc(2 · G (P )) ⊂ O k .
The group Γ is quasi–arithmetic if and only if it satisfies conditions (1)–(2),
but not necessarily (3).
Whenever Γ is quasi–arithmetic or arithmetic, k is called its field of defini-
tion (or ground field). The above criterion simplifies greatly if k = Q. Then
the lattice Γ will always be quasi–arithmetic, and it will be arithmetic if and
only if the cyclic products in Theorem 2.5 (3) are integers.
2.4. Vinberg’s algorithm. In addition to his study of arithmeticity of re-
flective lattices, Vinberg also produced an algorithm that produces the set of
generating reflections for a reflective quadratic from. Namely, if f is a qua-
dratic form of signature ( n, 1) over a totally real number field k with the ring
of integers O k .
Moreover, assume that f is admissible, that is for each non–trivial Galois
embedding σ : k → R the form f σ (obtained by applying σ to the coefficients of
f ) is positive definite. Then the group O f (O k ) of integer points of the orthog-
onal group O f (R) is a lattice [AVS, II, Chapter 6].
The form f as above is called reflective if O f (O k ) is generated by a finite
number of reflections, up to a finite index. The algorithm devised by Vinberg
in [Vin72] produces a set of generating reflections for any reflective quadratic
form. We shall refer to it as the Vinberg algorithm. Several computer real-
izations of the algorithm are available [Gug, BP, Bot].
10 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

In contrast, if the form f is not reflective, the algorithm never halts. The
later work by Bugaenko [Bug] also provided a way to certify that a given
Lorentzian form f is non–reflective. This “method of infinite symmetry” has
been also implemented in [Gug, BP, Bot] to various extent.
§ 3. Proofs
3.1. Proof of Lemma 1.8. Let P be a finite volume Coxeter polyhedron in
Hn and P0 be a facet that meets other facets of P at angles either π/4 or π/2.
Let P1 , . . . P k be all the facets of P that meet P0 at π/4. Assume that each P j
meets its neighbors at “even” angles of the form π/2m , with m ≥ 1, an integer.
If P is a polyhedron as above, and F is a facet of P , let R F denote the
reflection in the hyperplane of F .
Then the following inductive steps will produce a polyhedron with an or-
thogonal facet.
1. The double P (1) of P along the facet F1 = P1 has the facet R F1 (P0 ) that
forms ( k − 1) angles π/4 with its neighbours. Observe that P (1) is again a Cox-
eter polyhedron, since doubling of even Coxeter angles produces even Coxeter
angles again or, if we double a right angle, it produces a new facet.
2. The facet F2 = R F1 (P2 ) of P (1) forms even Coxeter angles with its neigh-
bours. Thus the double P (2) of P (1) along F2 has the facet R F2 R F1 (P0 ) that
forms ( k − 2) angles π/4 with its neighbours. All the conclusions of the previ-
ous step apply respectively to P (2) .
3. Continuing in this way, we obtain a polyhedron P (k) from P (k−1) by dou-
bling along F k = R Fk−1 (P k ), and the facet R Fk · · · R F1 (P0 ) of P (k) will be orthog-
onal to all of its neighbours. Such a polyhedron produces a crystallographic
sphere packing by the Structure Theorem (Theorem 2.4).
3.2. Proof of Theorem A. The proof of Theorem A follows from the results
of Esselmann [E] and the following simple lemma which is well known to
experts in hyperbolic reflection groups (and also follows from more general
results of [BK21]).
Lemma 3.1. Let Γ < Isom(Hn+1 ) be the symmetry group of a crystallographic
packing P of Sn , and let Γ0 be the stabilizer of a sphere in P . Then Γ0 is a
reflective lattice acting on Hn . If Γ is arithmetic, then Γ0 is also arithmetic
Proof. Indeed, the sphere corresponds to a facet P0 of the Coxeter polyhe-
dron P associated to the packing P , and P0 is orthogonal to all the adjacent
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 11

facets. Then Γ0 coincides with the finite covolume hyperbolic group generated
by reflections in the facets of P0 . The orthogonality of P0 to its neighbours to-
gether with Vinberg’s criterion (Theorem 2.5) implies that Γ being arithmetic
makes Γ0 arithmetic too. ■

It is easy to see that Lemma 3.1 implies Proposition 1.7 by applying it to

the setting of superintegral crystallographic sphere packings.
In [E], Esselmann showed that Borcherds’ 21–dimensional polyhedron from
[Bor] is a unique, up to commensurablity, arithmetic non–compact Coxeter
polyhedron in H21 and that there are no arithmetic non–compact Coxeter
polyhedra in H20 .
If we would have a superintegral sphere packing of S20 , then it would pro-
duce a superintegral crystallographic packing of S19 , which contradicts the
above mentioned results by Esselmann.

3.3. Proof of Theorem B. Our higher–dimensional examples of superinte-

gral sphere packings come from the Borcherds’ 21–dimensional polyhedron
P 21 [Bor]. By describing its lower–dimensional faces we can fill in some of the
dimensions. For example, in H19 we have a face P 19 (first found by Kaplin-
skaya and Vinberg in [KV]) that satisfies the conditions of Theorem 1.8.
( i)
There are two faces P 18 , i = 1, 2, each with an orthogonal facet, in H18 .
(1)
Here, P 18 denotes the polyhedron described by Kaplinskaya and Vinberg in
(2)
[KV]. The other polyhedron P 18 was found by Allcock in [A].
Be descending to lower–dimensional Coxeter faces (sometimes we have to
pass via non-Coxeter faces of lower codimension in order to find their faces of
higher codimension that turn out to be Coxeter), we obtain another face P 15
satisfying Lemma 1.8 in H15 . However, there are no suitable Coxeter faces of
P 21 in Hn for n = 16 and 17.
The remaining packings in Sn , 2 ≤ n ≤ 13, can be easily obtained from
Vinberg’s examples [Vin67] for the series of quadratic forms f n = −2 x02 + x12 +
. . . x2n , for 3 ≤ n ≤ 14, respectively. Each of the polyhedra in [Vin67, Table 7]
has a facet that is orthogonal to its neighbours1.
All the computations mentioned above (faces of Borcherds’ P 21 and their
properties of being Coxeter and/or having “even” angles) are accessible on
Github [Git1] as a SageMath [Sage] worksheet.
1For packings in Sn , 2 ≤ n ≤ 7, one can also use the totally right–angled polyhedra from [PV].
12 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

3.4. Proof of Theorem C. In [SW] Sharlau and Walhorn provide the list of
maximal arithmetic non–uniform reflective lattices (up to conjugation) acting
on Hn+1 for n = 2, 3. In [Tur] Turkalj performs an analogous classification in
Hn+1 for n = 4.
It is worth mentioning that in the list of lattices acting on H4 has a few
entries where the H direct summand apparently has to be replaced by 2 H.
Such entries are easily identifiable by running the Vinberg algorithm (the
number of roots indicated in the respective tables seems to be correct). The
lattices described by Turkalj require more work for the information to be
extracted from the list provided in [Tur]. We used instead an equivalent list
provided by Scharlau and Kirschmer [SK].
In each case we reduce each list up to rational isometry between the re-
spective lattices. Then the proof amounts to a lengthy computation carried
our by using the Julia [BEKS] implementation of the Vinberg algorithm due
to Bottinelli [Bot].
For each lattice, we consider its reflection subgroup: if the corresponding
polyhedron has an orthogonal facet (or, in term of Vinberg’s work [Vin15],
the lattice has an isolated root), then we are done. Otherwise, we choose
another lattice, up to an appropriate rational isometry, and use it instead: we
get a commensurable lattice having an isolated root. The necessary rational
isometry is usually guessed by trial and error: reflectivity is checked by using
the Vinberg algorithm.
All the data necessary to verify our computation [Git1] and Bottinelli’s im-
plementation of the Vinberg algorithm [Git2] are available on Github.

3.5. Remarks on reflectivity of arithmetic covers. Here we consider an-

other question in the context of properly integral sphere packings.
In the proof of the Arithmeticity Theorem (see [KN, Theorem 19]), given an
integral orbit of spheres O = S 0 · Γ, and choice of n + 2 linearly independent
normals to spheres in the orbit, one constructs an arithmetic lattice Ω which
has (a conjugate of) Γ as a subgroup; therefore Γ is sub–arithmetic, in the
terminology of [K2, p. 3].
One may wonder, since Γ is generated by reflections, whether Ω should
itself be reflective. Here we record an example which answers this question
in the negative.
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 13

p
4 5 8/7 6
1

F IGURE 2. The Coxeter–Vinberg diagram of the prism P .

Let P be the 4–dimensional hyperbolic prism described by the Coxeter-

Vinberg diagram in Figure 2. The facet normals of P are
à p p !
2 p 1 2
e1 = − , 2, 0, , , e 2 = (0, 0, 0, 0, −1),
8 2 2
à p !
1 2 1
e 3 = (0, 0, 0, −1, 0), e 4 = 1, 0, , , ,
2 2 2
Ãp p p !
14 2 14 2 14
e 5 = (0, 0, −1, 0, 0), e 6 = , , , 0, 0 ,
28 7 7
where the isolated vector is e 6 . By applying Vinberg’s criterion (Theorem 2.5)
to the Gram matrix of P
−1 p12 12
 
0 0 0
 p1 1
 2 −1 0 0 0 

 1 2
p1


 2 0 − 1 2
0 0 

G (P ) =  0 1 p1
− 1 1
0 

 2 2 2 q


1 8 
 0 0 0 − 1

 2 q 7 

8
0 0 0 0 7
− 1

we can conclude that P generates a quasi–arithmetic but not arithmetic re-

flection group. Hence the sphere packing corresponding to e 6 cannot be su-
perintegral.
Thus the orbit of e 6 under the group Γ generated by reflections through the
normals e 1 , . . . , e 5 is a crystallographic sphere packing (see Figure 3), which
we claim is properly integral.
14 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

F IGURE 3. Some of the spheres of the packing P shown in blue, and the walls
of the generators e 1 , . . . , e 5 of Γ shown in red.

To see this, we take for a list of n + 2 = 5 linearly independent vectors

the choice V = ( e 6 , e 6 R 5 , e 6 R 5 R 4 , e 6 R 5 R 4 R 2 , e 6 R 5 R 4 R 3 ) as described in p
[KN,
Equation (2)]. These are all in the packing P = e 6 · Γ. After rescaling by 8/7,
these spheres all have bend 1. Then the generators R 1 , . . . , R 5 , conjugated by
V , become the matrices:
   
1 0 0 0 0 1 0 0 0 0
 0 1 0 0 0  0 1 0 0 0
   
V R 1 V −1 =  0 0 1 0 0  , V R 2 V −1 =  0 0 0 1 0  ,
   
 9/4 9/4 3/2 −1/4 −3/4  0 0 1 0 0
   
9/4 9/4 3/2 −5/4 1/4 0 0 −1 1 1

   
1 0 0 0 0 1 0 0 0 0

 0 1 0 0 0 


 0 0 1 0 0 

−1 −1
V R3V =  0 0 0 0 1  , V R4V =  0 1 0 0 0 ,
   
0 0 −1 1 1 0 0 0 1 0
   
   
0 0 1 0 0 0 −1 1 0 1
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 15

and
 
0 1 0 0 0

 1 0 0 0 0 

V R 5 V −1 =  0 0 1 0 0 .
 
0 0 0 1 0
 
 
0 0 0 0 1

These act on the left on the set of bends of the packing; the bends which
arise in the packing are those in the action of the above matrices on the
initial bends vector (1, 1, 1, 1, 1) of all bends equal to 1. All but the first matrix
are themselves integral, and hence evidently preserve integrality. Moreover,
inspection shows that they preserve the congruence b 1 + b 2 + 2 b 3 + 3 b 4 + b 5 ≡
0 (mod 4), which is satisfied by the initial bends vector (1, 1, 1, 1, 1). This
congruence ensures that the action of the first matrix V R 1 V −1 always gives
integer bends in the orbit, thus proving the proper integrality of the packing.
Finally, following the proof of the Arithmeticity Theorem, we exhibit an
arithmetic lattice Ω containing a conjugate of Γ. We find in this case that Ω
is not reflective. To begin, we record that the Gram matrix of V :
 
−7 9 9 9 9
 9 −7 9 9 9 
T 1 
G = G (V ) = V QV = −  9 9 −7 9 25  ,
 
7
 9 9 9 −7 41 

9 9 25 41 −7

where Q = H ⊥ 〈1, 1, 1〉. The inverse Gram matrix equals

 
25 9 0 −9 −9
 9 25 0 −9 −9 
−1 7  
G =− 0 0 12 −8 −4  ,
 

256 
−9 −9 −8 9 1 


−9 −9 −4 1 5

that gives rise to the Lorentzian quadratic form f ( x) = xT G −1 x. The proof

of the Arithmeticity Theorem [KN] shows that V ΓV −1 < Ω := O f (Z). Finally,
16 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

Vinberg’s algorithm applied to Ω produces the following first 31 roots:

e 1 = (1, −1, 0, 0, 0), e 2 = (0, 1, 2, 1, 0), e 3 = (0, 1, 0, 1, 0),
e 4 = (−1, 0, −1, −2, 1), e 5 = (0, 1, 2, 3, 0), e 6 = (2, 4, 9, 15, −1),
e 7 = (1, 2, 8, 11, 0), e 8 = (1, 3, 6, 12, 0), e 9 = (5, 6, 20, 29, −2),
e 10 = (8, 8, 21, 35, −3), e 11 = (7, 10, 23, 36, −3), e 12 = (4, 4, 15, 25, −1),
e 13 = (6, 7, 16, 31, −2), e 14 = (3, 3, 7, 13, −1), e 15 = (5, 6, 13, 28, −1),
e 16 = (5, 6, 18, 29, −2), e 17 = (8, 8, 23, 35, −3), e 18 = (2, 3, 11, 20, 1),
e 19 = (4, 4, 17, 25, −1), e 20 = (1, 1, 2, 5, 0), e 21 = (3, 6, 19, 26, −1),
e 22 = (3, 6, 15, 26, −1), e 23 = (4, 4, 9, 23, 1), e 24 = (3, 3, 14, 22, 0),
e 25 = (3, 3, 11, 17, −1), e 26 = (3, 4, 12, 23, 0), e 27 = (2, 5, 8, 21, 2),
e 28 = (3, 3, 7, 12, −1), e 29 = (11, 12, 34, 59, −4), e 30 = (10, 11, 32, 49, −4),
e 31 = (8, 11, 35, 54, −3)
There is an infinite order symmetry σ mapping the elliptic subdiagram
spanned by the roots { e 13 , e 3 , e 14 , e 1 , e 6 } to another elliptic subdiagram spanned
by the roots { e 12 , e 19 , e 25 , e 31 , e 7 }, exactly in this order
e 13 7→ e 12 , e 3 7→ e 19 , e 14 7→ e 25 , e 1 7→ e 31 , e 6 7→ e 7 .
The matrix from O f (Z) representing σ in the respective bases is
 
54 41 −3 −10 79
 64 48 −3 −12 96 
 
S =  159 122 −8 −30 238  .
 
 276 210 −15 −51 413 
 
−21 −16 1 4 −31
Thus, the arithmetic group Ω containing Γ is not reflective according to
[Vin67].
However, we would like to remark that the group Ω is defined up to com-
mensurability (because the choice of normals described above), and reflectiv-
ity is not commensurability invariant. Indeed, the lattice
 
0 0 49
 0 49 7 
49 7 3
 KLEINIAN SPHERE PACKINGS, REFLECTION GROUPS, AND ARITHMETICITY 17

has no roots [BK22, Col], while it is easily seen to be commensurable to the

diagonal lattice
 
−1 0 0
 0 2 0 ,
0 0 1
that is reflective and corresponds to the (2, 4, ∞) hyperbolic triangle reflection
group.

References
[A] D. Allcock. Infinitely many hyperbolic Coxeter groups through dimension 19. Geom. Topol., Vol. 10,
pp. 737–758, 2006. 11
[AVS] D. Alekseevskij, È. Vinberg, A. Solodovnikov. Geometry of spaces of constant curvature. Encycl.
Math. Sci., Vol. 29, pp. 1–138, 1993. 9
[BEKS] J. Bezanson, A. Edelman, S. Karpinski, V. B. Shah. Julia: A fresh approach to numerical computing.
SIAM Review, Vol. 59, No. 1, pp. 65–98, 2017. 12
[BK21] N. Bogachev, A. Kolpakov. On faces of quasi–arithmetic Coxeter polytopes. IMRN, Vol. 2021, No. 4,
pp. 3078–3096, 2021. 10
[BK22] N. Bogachev, A. Kolpakov. Thin hyperbolic reflection groups. Preprint, arXiv:2112.14642 17
[Git1] N. Bogachev, A. Kolpakov, A. Kontorovich. Crystallographic packings.
https://github.com/sashakolpakov/crystallographic-packings 11, 12
[BP] N. Bogachev, A. Perepechko. Vinberg’s algorithm for hyperbolic lattices. Math. Notes, Vol. 103, No. 5,
pp. 836–840, 2018. 9, 10
[Bor] R. Borcherds. Automorphism groups of Lorentzian lattices. J. Algebra, Vol. 111, pp. 133–153, 1987.
3, 11
[Bor2] R. Borcherds. Reflection groups of Lorentzian lattices. Duke Math J., Vol. 104 no. 2, pp. 319–366,
2000. 4
[Bot] R. Bottinelli. On computational aspects of hyperbolic reflection groups & two homology theories for
graphs. PhD thesis, 2021. 9, 10, 12
[Git2] R. Bottinelli. Vinny the Coxboy: Vinberg’s algorithm over number fields.
https://github.com/bottine/VinbergsAlgorithmNF 12
[Bug] V. O. Bugaenko. Arithmetic crystallographic groups generated by reflections, and reflective hyper-
bolic lattices. In: Lie groups, their discrete subgroups, and invariant theory. Adv. Soviet Math., Vol.
8, Amer. Math. Soc., Providence, RI, 1992, pp. 33–55. 10
[Col] G. Collinet. Personal communication to A. Kolpakov. 10 November 2014 17
[E] F. Esselmann. Über die maximale Dimension von Lorentz–Gittern mit coendlicher Spiegelungs-
gruppe. J. Num. Theory, Vol. 61, pp. 103–144, 1996. 2, 3, 10, 11
[Gug] R. Guglielmetti. Hyperbolic isometries in (in–)finite dimensions and discrete reflection groups: the-
ory and computations. PhD Thesis, Université de Fribourg, 2017. 9, 10
[KN] A. Kontorovich, K. Nakamura. Geometry and arithmetic of crystallographic sphere packings. PNAS,
Vol. 116, No. 2, pp. 436–441, 2019. 2, 3, 4, 7, 12, 14, 15
[KV] I. Kaplinskaya, È. Vinberg. On the groups O18,1 (Z) and O19,1 (Z). Dokl. Akad. Nauk SSSR, Vol. 238,
No. 6, pp. 1273–1275, 1978. 11
[K2] M. Kapovich, A. Kontorovich. On superintegral Kleinian sphere packings, bugs, and arithmetic
groups. Preprint, arXiv:2104.13838 2, 7, 12
[PV] L. Potyagailo, È. Vinberg. On right-angled reflection groups in hyperbolic spaces. Comment. Math.
Helv., Vol. 80, No. 1, pp. 63–73, 2005. 11
18 NIKOLAY BOGACHEV, ALEXANDER KOLPAKOV, AND ALEX KONTOROVICH

[Sage] SageMath, the Sage Mathematics Software System. Version 9.4, The Sage Developers, 2021,
https://www.sagemath.org 11
[SK] R. Scharlau, M. Kirschmer. Personal communication to N. Bogachev. 17 February 2022 12
[SW] R. Scharlau, C. Walhorn. Integral lattices and hyperbolic reflection groups. Astérisque, Tome 209, pp.
279–291, 1992. 12
[Tur] I. Turkalj. Reflective Lorentzian lattices of signature (5,1). PhD Thesis, Universität Dortmund, 2017.
12
[Vin67] È. B. Vinberg. Discrete groups generated by reflections in Lobachevski spaces. Mathematics of the
USSR – Sbornik, Vol. 1, No. 3, pp. 429–444, 1967. 6, 8, 9, 11, 16
[Vin72] È. B. Vinberg. On groups of unit elements of certain quadratic forms. Mathematics of the USSR –
Sbornik, Vol. 16. No. 1, pp. 17–35, 1972. 9
[Vin85] È. B. Vinberg. The absence of crystallographic groups of reflections in Lobachevsky spaces of large
dimension. In: Trans. Moscow Math. Soc., 1985, pp. 75–112. 6
[Vin15] È. B. Vinberg. Non–arithmetic hyperbolic reflection groups in higher dimensions. Moscow Math. J.,
Vol. 15, No. 3, pp. 593–602, 2015. 12

T HE K HARKEVICH I NSTITUTE FOR I NFORMATION T RANSMISSION P ROBLEMS, M OSCOW, R USSIA

M OSCOW I NSTITUTE OF P HYSICS AND T ECHNOLOGY, D OLGOPRUDNY, R USSIA

Email address: nvbogach@mail.ru

I NSTITUT DE M ATHÉMATIQUES, U NIVERSITÉ DE N EUCHÂTEL , CH–2000 N EUCHÂTEL , S WITZERLAND

Email address: kolpakov.alexander@gmail.com

D EPARTMENT OF M ATHEMATICS, R UTGERS U NIVERSITY, 110 F RELINGHUYSEN R D, P ISCATAWAY, NJ

08854, USA
Email address: alex.kontorovich@rutgers.edu