MAXIMAL HOLONOMY OF
INFRA-NILMANIFOLDS WITH so(3) e ×R
3-GEOMETRY
Kyung Bai Lee and Joonkook Shin
Abstract. Let so(3) e ×R
3be the 6-dimensional nilpotent Lie group with group operation (s, x)(t, y) = (s + t + xy
t− yx
t, x + y). We prove that the maximal order of the holonomy groups of all infra- nilmanifolds with so(3) e ×R
3-geometry is 16.
1. Introduction
Let so(n) be the additive group of skew-symmetric matrices. This is the Lie algebra of the special orthogonal group SO(n). There is a bilinear map
I : R
n× R
n−→ so(n) defined as follows: For
x =
x
1x
2·
·
· x
n
, y =
y
1y
2·
·
· y
n
,
I(x, y) = xy
t− yx
t,
where ()
tdenotes transpose of the matrix (). With this I, the set so(n)×
R
ngets a group structure given by
(s, x)(t, y) = (s + t + I(x, y), x + y).
Received June 30, 2005.
2000 Mathematics Subject Classification: 20H15, 20F18, 20E99, 53C55.
Key words and phrases: almost Bieberbach group, holonomy group, infra-nil- manifold.
The second author was supported in part by KOSEF grant R01-2000-000-00005-
0(2004).
This group is denoted by
R
n= so(n) e ×R
n.
Then R
nis a simply connected nilpotent Lie group, of nilpotency class 2, with the center so(n). Note that so(n) is viewed as a commutative Lie group isomorphic to R
n(n−1)2(not as a Lie algebra). It fits the short exact sequences of Lie groups
0 → so(n) → R
n→ R
n→ 1.
Let M be an infra-nilmanifold with R
n-geometry; that is, M = Π\R
n, where Π is a torsion free, discrete, cocompact subgroup of R
noC for some compact subgroup C of Aut(R
n). Such a subgroup Π is called an almost Bieberbach group (=AB group). It is well known that Π con- tains a cocompact lattice Γ of R
nof finite index, and the quotient group Π/Γ is called the holonomy group of M .
Our aim is to understand those infra-nilmanifolds modelled on the 6-dimensional nilpotent Lie group
R
3= so(3) e ×R
3.
In particular, we are interested in finding the possible maximal order for the holonomy groups. We shall prove that the holonomy group of maximal order is D
8× Z
2, of order 16.
2. The automorphism group of R
nSince the center, Z(R
n) = so(n), is a characteristic subgroup of R
n, every automorphism of R
nrestricts to an automorphism of so(n). Con- sequently an automorphism of R
ninduces an automorphism on the quo- tient group R
n. Thus there is a natural homomorphism Aut(R
n) → Aut(so(n)) × Aut(R
n), θ 7→ (ˆ θ, ¯ θ). Here Aut(so(n)) is the group of linear automorphisms of the R-vector space so(n), not the Lie algebra automorphisms.
Lemma 2.1. Aut(R
n) → Aut(R
n) = GL(n, R) is surjective. More- over, the exact sequence Aut(R
n) → GL(n, R) → 1 splits.
Proof. First we define
J : GL(n, R) −→ Aut(so(n)) by
(2.1) J(C)(s) = CsC
tfor C ∈ GL(n, R) and s ∈ so(n). Then clearly, J is a homomorphism.
We denote J(C) by b C.
We claim that, for any C ∈ GL(n, R),
(2.2) (s, x) 7→ ( b Cs, Cx)
is an automorphism of R
n. From
( b Cs, Cx) · ( b Ct, Cy) = ( b Cs + b Ct + I(Cx, Cy), Cx + Cy)
= ( b C(s + t) + I(Cx, Cy), C(x + y)) ( b C(s + t + I(x + y)), C(x + y)) = ( b C(s + t) + b CI(x + y), C(x + y)), for the assignment (2.2) to be an automorphism, it is necessary and sufficient that
(2.3) C(I(x, y)) = I(Cx, Cy) b holds for all x, y ∈ R
n. Now this follows from
C(xy
t− yx
t)C
t= Cx(Cy)
t− Cy(Cx)
tfor all x, y ∈ R
n.
Theorem 2.2 (Structure of Aut(R
n)).
Aut(R
n) = Hom(R
n, so(n)) o GL(n, R),
where Hom(R
n, so(n)) is the additive group of linear transformations of vector spaces, and an element (η, C) ∈ Hom(R
n, so(n)) o GL(n, R) acts by
(η, C)(s, x) = ( b Cs + η(x), Cx).
Proof. Let θ ∈ Aut(R
n). Then we have the following commutative diagram of exact sequences
1 −−−−→ so(n) −−−−→ R
n−−−−→ R
n−−−−→ 1
y
θˆ y
θ y
θ¯1 −−−−→ so(n) −−−−→ R
n−−−−→ R
n−−−−→ 1
Thus θ(s, x) = (ˆ θ(s) + η(s, x), ¯ θ(x)) for (s, x) ∈ R
n, where η : R
n→ so(n). Since θ is a homomorphism, one can show that η is a homomor- phism, i.e.,
η((s, x)(t, y)) = η(s, x) + η(t, y).
In particular, (ˆ θ(s), 0) = θ(s, 0) = (ˆ θ(s) + η(s, 0), 0) implies that η(s, 0)
= 0 for all s ∈ so(n), and thus η(s, x) = η((s, 0)(0, x)) = η(s, 0) +
η(0, x) = η(0, x). Hence η ∈ Hom(R
n, so(n)).
Let’s find out the kernel of the surjective homomorphism of Lemma 2.1:
Aut(R
n) → GL(n, R), θ 7→ ¯ θ.
Suppose that θ ∈ Aut(R
n) with ¯ θ = id
Rn. Then ˆ¯ θ = id
so(n)(see the equality (2.1)) and thus θ(s, x) = (s + η(x), x) for some η ∈ Hom(R
n, so(n)). Conversely given η ∈ Hom(R
n, so(n)), define θ ∈ Aut(R
n) by θ(s, x) = (s + η(x), x). Clearly this θ lies in the kernel of the homomor- phism. Hence we have a short exact sequence
1 → Hom(R
n, so(n)) → Aut(R
n) → GL(n, R) → 1.
By Lemma 2.1, this sequence is split.
Note that
Hom(R
n, so(n)) o GL(n, R) ⊂ Hom(R
n, so(n)) o (Aut(so(n)) × GL(n, R)) (η, A) 7→ (η, ( b A, A))
and the action of GL(n, R) on Hom(R
n, so(n)) is
Aη(x) = ˆ A · η(A
−1x).
The group operation on R
no GL(n, R) is given by ((s, x), A)((t, y), B)
= ((s, x) ·
A(t, y), AB)
= ((s, x) · ( ˆ At, Ay), AB)
= ((s + ˆ At + I(x, Ay), x + Ay), AB).
3. The structure of AB-groups for R
nLet Π ⊂ R
no Aut(R
n) be an AB-group. Then it is well known that
Γ = Π ∩R
n, the pure translations in Π, is the maximal normal nilpotent
subgroup, and Φ = Π/Γ, the holonomy group of Π, is finite. Since Γ is
a lattice of R
n, Z = Γ ∩ Z(R
n) is a lattice of Z(R
n) = so(n), and Γ/Z
is a lattice of R
n/Z(R
n) = R
n. Thus Z = Γ ∩ Z(R
n) ∼ = Z
n(n−1)2and
Γ/Γ ∩ Z(R
n) ∼ = Z
n.
Consider the following natural commutative diagram:
1 1
y
y so(n) −−−−→
=
so(n)
y
y
1 −−−−→ R
n−−−−→ R
no Aut(R
n) −−−−→ Aut(R
n) −−−−→ 1
y
y
y
=1 −−−−→ R
n−−−−→ R
no Aut(R
n) −−−−→ Aut(R
n) −−−−→ 1
y
y
1 1
Recall from Theorem 2.2 that an element
(η, A) ∈ Aut(R
n) = Hom(R
n, so(n)) o GL(n, R) acts on (s, x) ∈ R
nby
(η, A)(s, x) = ( ˆ As + η(x), Ax).
Thus GL(n, R) acts on so(n) via the homomorphism b : GL(n, R) → Aut(so(n)), and GL(n, R) acts on R
nby matrix multiplication GL(n, R)
×R
n→ R
n.
Let Q = Π/Z. Then the above diagram induces the following com- mutative diagram:
(3.1)
1 1
y
y Z −−−−→
=
Z
y
y
1 −−−−→ Γ −−−−→ Π −−−−→ Φ −−−−→ 1
y
y
y
=1 −−−−→ Z
n−−−−→ Q −−−−→ Φ −−−−→ 1
y
y
1 1
Here an element (η, A) ∈ Φ ⊂ Aut(R
n) = Hom(R
n, so(n)) o GL(n, R) acts on Z
nby A, and on Z by ˆ A.
Recall from [4, Proposition 2] that a virtually free abelian group 1 → Z
n→ Q → Φ → 1 is a crystallographic group if and only if the centralizer of Z
nin Q has no torsion elements. Suppose an element of Φ acts trivially on Z
n. Then it is of the form (η, A) ∈ Aut(R
n), where A = I ∈ GL(n, R). Then ˆ A = I also, see equation (2.1). Since Hom(R
n, so(n)) is isomorphic to the additive group of n×
n(n−1)2matrices (≈ R
n(n+1)2), it has no elements of finite order. Consequently, (η, A) ∈ Aut(R
n) is trivial. This shows that Φ acts effectively on Z
n. It follows that Q is naturally an n-dimensional crystallographic group.
The finite group Φ must be in a maximal compact subgroup O(n) of Aut(R
n) = Hom(R
n, so(n)) o GL(n, R).
Proposition 3.1. Let Q ⊂ R
no O(n) be a crystallographic group, and Q
0⊂ R
no GL(n, R) be an affine crystallographic group isomorphic to Q. Then there exists an AB-group Π obtained from Q if and only if there exists an AB-group Π
0obtained from Q
0.
Proof. Let
Q = h(v
1, I), (v
2, I), . . . , (v
n, I), (a
1, A
1), (a
2, A
2), . . . , (a
p, A
p)i, Q
0= h(v
10, I), (v
02, I), . . . , (v
0n, I), (a
01, A
01), (a
02, A
02), . . . , (a
0p, A
0p)i and (d, D) ∈ R
no GL(n, R) with
(v
0i, I) = (d, D)
−1(v
i, I)(d, D), i = 1, 2, . . . , n (a
0j, A
0j) = (d, D)
−1(a
j, A
j)(d, D), j = 1, 2, . . . , p.
Suppose there exist t
1, t
2, . . . , t
n, s
1, s
2, . . . , s
p∈ so(n) such that Π = h(t
1, v
1, I), (t
2, v
2, I), . . . , (t
n, v
n, I),
(s
1, a
1, A
1), (s
2, a
2, A
2), . . . , (s
p, a
p, A
p)i
is an AB-group. Then clearly, conjugate of Π by (0, d, D) ∈ R
no Aut(R
n) is an AB-group Π
0obtained from Q
0. The converse is similar.
From this Proposition, we need not put our abstract crystallographic group into R
no O(n), but it is enough to use any convenient affine embedding.
Lemma 3.2. Let v
1, v
2, . . . , v
nbe a basis of R
n. Then the set {I(v
i,
v
j) | i < j} forms a lattice of so(n).
Proof. First we set notation. Let E
ijbe the skew-symmetric matrix whose (i, j) entry is 1, (j, i) entry is −1, and 0 elsewhere. Suppose v
i= e
ifor all i = 1, 2, . . . , n. Then I(e
i, e
j) = e
ie
tj− e
je
ti= E
ij. Therefore {I(e
i, e
j) | i < j} = {E
ij| i < j} is a basis for the vector space so(n). For a general basis {v
1, v
2, . . . , v
n}, let C be the matrix whose jth column is v
j. Then I(v
i, v
j) = J(C)(E
ij), and hence the set {I(v
i, v
j) | i < j} forms a basis of so(n).
This lemma tells us that the lattice Z
nof R
ngenerated by {v
1, v
2, . . . , v
n} yields a lattice generated by {I(v
i, v
j) | i < j}, which must be contained in Z in the above diagram. However, the Z in the diagram can be finer than the lattice generated by {I(v
i, v
j) | i < j}.
Proposition 3.3. Let Q ⊂ R
no SO(n) be a crystallographic group generated by
(v
1, I), (v
2, I), . . . , (v
n, I), (a
1, A
1), (a
2, A
2), . . . , (a
p, A
p),
where the subgroup h(v
1, I), (v
2, I), . . . , (v
n, I)i is the maximal normal free abelian subgroup of Q. If there exists an AB-group obtained from Q, then there is an AB-group Π
(s1,s2,...,sp)generated by
(0, v
1, I), (0, v
2, I), . . . , (0, v
n, I), (s
1, a
1, A
1), (s
2, a
2, A
2), . . . , (s
p, a
p, A
p), for some s
1, s
2, . . . , s
p∈ so(n).
Proof. Let Π ⊂ R
no Aut(R
n) be an AB-group obtained from Q.
Then
Π ∩ so(n) = h(v
1, 0, I), (v
2, 0, I), . . . , (v
n(n−1) 2, 0, I)i for some v
1, v
2, . . . , v
n(n−1)2
∈ so(n). Let us take any pre-images in Π as follows:
(w
1, v
1, I), (w
2, v
2, I), . . . , (w
n, v
n, I), (t
1, a
1, A
1), (t
2, a
2, A
2), . . . , (t
p, a
p, A
p)
for some w
1, w
2, . . . , w
n, t
1, t
2, . . . , t
p∈ so(n). Since [(w
i, v
i, I), (w
j, v
j, I)] = (2I(v
i, v
j), 0, I), the group generated by
(w
1, v
1, I), (w
2, v
2, I), . . . , (w
n, v
n, I)
becomes a lattice of R
nalready (see Lemma 3.2), and hence the group generated by
(w
1, v
1, I), (w
2, v
2, I), . . . , (w
n, v
n, I),
(t
1, a
1, A
1), (t
2, a
2, A
2), . . . , (t
p, a
p, A
p)
(without (v
1, 0, I), (v
2, 0, I), . . . , (v
n(n−1) 2, 0, I)) will be a subgroup of Π of finite index, and is an AB-group (since it is torsion free). Let us call this smaller group Π
0.
Let η ∈ Hom(R
n, so(n)). Consider the product in R
no Aut(R
n):
((0, 0), η)
−1((w
i, v
i), I)((0, 0), η) = ((w
i− η(v
i), v
i), I).
Certainly, there exists η ∈ Hom(R
n, so(n)) such that η(v
i) = w
ifor all i = 1, 2, . . . , n.
Consequently, the group Π
00obtained by conjugating Π
0by (0, 0, η) is another AB-group, satisfying the condition w
1= w
2= · · · = w
n= 0 (without changing the second and third slots). Thus we have obtained an AB-group Π
00= Π
(s1,s2,...,sp)⊂ R
no Aut(R
n) generated by
(0, v
1, I), (0, v
2, I), . . . , (0, v
n, I), (s
1, a
1, A
1), (s
2, a
2, A
2), . . . , (s
p, a
p, A
p) for some s
1, s
2, . . . , s
p∈ so(n).
The Proposition says that: If there is an AB-group Π constructed from Q, then Π
(s1,s2,...,sp)is conjugate to a subgroup of Π of finite in- dex, which is another AB-group constructed from Q. Conversely, if Π
(s1,s2,...,sp)is an AB-group (i.e., is torsion free), then we are done.
Therefore, the existence/non-existence of construction is solely deter- mined by the group of the form Π
(s1,s2,...,sp); that is, whether there exist s
1, s
2, . . . , s
pfor which Π
(s1,s2,...,sp)is torsion free.
4. The group R
3Our group R
3= so(3) e ×R
3has group law
(s, x)(t, y) = (s + t + I(x, y), x + y), where
I(x, y) = xy
t− yx
t=
0 x
1y
2− x
2y
1x
1y
3− x
3y
1x
2y
1− x
1y
20 x
2y
3− x
3y
2x
3y
1− x
1y
3x
3y
2− x
2y
30
.
If we identify so(3) with R
3by
0 z −y
−z 0 x
y −x 0
←→
x y z
, then clearly,
I(x, y) = x × y (cross product)
so that R
3= so(3) × R
3= R
3× R
3has the group operation (s, x)(t, y) = (s + t + x × y, x + y).
In our case, b C = J(C) has a very simple description. Let e
i∈ R
nwhose ith component is 1 and 0 elsewhere. From the condition (2.3), we have
C(I(e b
i, e
j)) = I(Ce
i, Ce
j)
= I(c
i, c
j)
for all i, j, where c
iis the ith column of the matrix C. Note that I(e
i, e
j) is the skew-symmetric matrix whose (i, j) entry is 1, (j, i) entry is −1, and 0 elsewhere.
By the identification of so(3) with R
3as above, the above equality becomes
C(e b
1) = c
2× c
3C(e b
2) = c
3× c
1C(e b
3) = c
1× c
2so that
C = det(C)(C b
−1)
t.
For each 3-dimensional crystallographic group Q, we shall check if there exists an AB-group Π constructed from Q; that is, a torsion free Π ⊂ R
3o O(3) ⊂ R
3o Aut(R
3) fitting the short exact sequence
1 −→ Z −→ Π −→ Q −→ 1.
This is the key notion for our arguments and construction. We have a complete classification of 3-dimensional crystallographic groups (Q’s in the above statement). We shall use the representations of the 3- dimensional crystallographic groups given in the book [1].
Every Q has an explicit representation Q −→ R
3o GL(3, Z) (not into R
3o O(3)) in this book.
Our goal is to determine which Q will give rise to a torsion free Π
that fits the diagram (3.1). When Q is torsion free, then Π will be
automatically torsion free, but when Q contains a torsion subgroup Q
0, we need to check whether the lift Q
0to Π will be torsion free.
Let Π
0be the lift of Q
0to Π. Thus
1 −−−−→ Z −−−−→ Π
0−−−−→ Q
0−−−−→ 1 (4.1)
is exact. If Π is torsion free, then so is Π
0, and is a 3-dimensional Bieberbach group. Therefore,
Proposition 4.1. Any finite subgroup of Q is one of the holonomy groups 1, Z
2, Z
3, Z
4, Z
6or Z
2× Z
2of 3-dimensional crystallographic gro- ups.
Proposition 4.2. There is no AB-group Π ⊂ R
3o Aut(R
3) con- structed from Q given in 7/4/3/02, 7/3/2/02, 7/2/1/03, 7/2/3/02, (all holonomy group of order 24).
Proof. Let
Q = h(e
1, I), (e
2, I), (e
3, I), (a, A), (b, B), (c, C), (d, D)i, and let, for some s, t, u, v ∈ so(3),
t
1= (0, e
1, I), t
2= (0, e
2, I), t
3= (0, e
3, I), α = (s, a, A), β = (t, b, B), γ = (u, c, C), δ = (v, d, D).
By Proposition 3.3, there is no AB-group from Q, if and only if, for any s, t, u, v ∈ so(3), the group Π generated by t
1, t
2, t
3, α, β, γ, δ
(4.2) Π
(s,t,u,v)= ht
1, t
2, t
3, α, β, γ, δi
has a nontrivial torsion element. Now let Z = Π ∩ so(3). Then there is no AB-group from Q if and only if the following holds: for any s, t, u, v ∈ so(3), there exists a nontrivial torsion element of the form
zt
n11t
n22t
n33α
pβ
qγ
rδ
owhere z ∈ Z and n
i, p, q, r, o, k ∈ Z with 0 ≤ p < order of A, etc.
(7/4/3/02). For any choice of s, t, u, v ∈ so(3), α
−1γ
3αγ is a torsion element of order 3.
(7/3/2/02). Let z = (16(1, −1, −1), 0, I). Then z ∈ Z. For any choice of s, t, u, v ∈ so(3), zα
−1γ
3αγ is a torsion element of order 3.
(7/2/1/03). For any choice of s, t, u, v ∈ so(3), α
−1γ
3αγ is a torsion element of order 3.
(7/2/3/02). For any choice of s, t, u, v ∈ so(3), α
−1γ
3αγ is a torsion
element of order 3.
5. A construction of an AB-group with holonomy order 16 Theorem 5.1 (Existence). There exists an almost Bieberbach group Π ⊂ R
3o Aut(R
3) whose holonomy group is D
8× Z
2, of order 16.
Proof. Consider the 3-dimensional crystallographic group Q (Case 4/7/1/10) generated by the following elements:
(e
1, I), (e
2, I), (e
3, I), (a, A), (b, B), (c, C), (d, D), where I is the identity matrix, e
iis the unit vector with 1 in the ith coordinate, and
(a, A) =
12
1 1 0
,
−1 0 0
0 −1 0
0 0 1
,
(b, B) =
12
1 0 0
,
0 −1 0
1 0 0
0 0 1
,
(c, C) =
12
0 1 1
,
−1 0 0
0 1 0
0 0 −1
,
(d, D) =
12
0 0 0
,
−1 0 0
0 −1 0
0 0 −1
.
Note that
A
2= I, B
2= A, C
2= I, D
2= I, [B, C] = I, [B, D] = I, [C, D] = I and that Q has the holonomy group Φ of order 16.
We consider the subgroup Π of (so(3) ˜ ×R
3) o SO(3) = (R
3×R ˜
3) o SO(3) generated by
t
1= (0, e
1, I), t
2= (0, e
2, I), t
3= (0, e
3, I), α =
0 0
14
, a, A
, β =
0 0
18
, b, B
,
γ =
14
0
14
, c, C
, δ =
14 14 14
, d, D
.
Let Z = Π ∩ so(3). We know Z is a lattice of so(3) from Lemma 3.2.
By calculation, we see that Z is actually generated by the vectors
z
1=
14 14
0
, z
2=
−
1414
0
, z
3=
0 0
12