THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL

GALOIS GROUP

FASEEH AHMAD, ROBERT L. BENEDETTO, JENNIFER CAIN, GREGORY CARROLL,

AND LILY FANG
arXiv:1909.00039v1 [math.NT] 30 Aug 2019

Abstract. The arboreal Galois group of a polynomial f over a field K encodes the
action of Galois on the iterated preimages of a root point x0 ∈ K, analogous to the
action of Galois on the ℓ-power torsion of an abelian variety. We compute the arboreal
Galois group of the postcritically finite polynomial f (z) = z 2 − 1 when the field K and
root point x0 satisfy a simple condition. We call the resulting group the arithmetic
basilica group because of its relation to the basilica group associated with the complex
dynamics of f . For K = Q, our condition holds for infinitely many choices of x0 .

Let K be a field with algebraic closure K, let x0 ∈ K, and let f ∈ K[z] be a polynomial
of degree d ≥ 2. For each n ≥ 0, let f n denote the n-th iterate f ◦ · · · ◦ f of f under
composition, with f 0 (z) = z and f 1 (z) = f (z). The backward orbit of x0 under f is
a
Orb− f (x0 ) := f −n (x0 ) ⊆ K,
n≥0

where f −n (y) is the set of roots of the equation f n (z) = y in K.

If f n (z) − x0 is a separable polynomial, then f −n (x0 ) has exactly dn elements, and the
field Kn := K(f −n (x0 )) ⊆ K is a Galois extension of K, with Galois group
Gn := Gal(Kn /K).
n
If f (z) − x0 is separable for all n ≥ 0, then we also define
[
G∞ := Gal(K∞ /K), where K∞ := Kn .
n≥0

The backward orbit Orb−

f (x0 )
has the structure of an infinite d-ary rooted tree Td,∞ ,
formed by connecting each y ∈ f −(n+1) (x0 ) to f (y) ∈ f −n (x0 ) via an edge. Thus, G∞
is isomorphic to a subgroup of Aut(Td,∞ ), and Gn is isomorphic to a subgroup of the
automorphism group Aut(Td,n ), where Td,n is the subtree of just the bottom n levels of
Td,∞ . The resulting action of Galois on the tree is analogous to the action of Galois on
the ℓ-power torsion of an abelian variety A, since the ℓ-power torsion is precisely the
backward orbit of the identity point O under the morphism [ℓ] : A → A.
Odoni introduced the study of such Galois groups in 1985 in [24], and Boston and Jones
in 2007 called them “arboreal” in [6], since they act on trees. These groups have attracted
increasing attention over the years; see [1, 2, 8, 9, 10, 11, 13, 14, 15, 17, 20, 28, 29] for
a (very limited) selection. See also [16] for a survey of the field. Many examples have
been found where G∞ is the full group Aut(Td,∞ ), as in [5, 18, 21, 22, 24, 27, 28]. More
generally, the expectation has emerged that when K is a global field, G∞ should usually
Date: August 30, 2019.
2010 Mathematics Subject Classification. 37P05, 11R32, 14G25.
1
2 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

have finite index in Aut(Td,∞ ); see [16, Conjecture 3.11] for a precise conjecture when
d = 2, and [7, 12, 19] for conditional results for d = 2, 3. By analogy, Serre’s Open
Image Theorem [26] states that for an elliptic curve over a number field, the action of
Galois on the ℓ-power torsion has finite index in the appropriate automorphism group
GL(2, Zℓ ).
However, just as Serre’s Theorem excludes the special case of CM elliptic curves, there
are special situations where G∞ necessarily has infinite index in Aut(Td,∞ ). One such
case is that the map f is postcritically finite, or PCF, meaning that for every ramification
point c of f , the forward orbit {f n (c)|n ≥ 0} is finite; equivalently, every critical point
of f is preperiodic. (See, for example, [16, Theorem 3.1].)
It is natural to ask whether a given PCF map has an associated subgroup of Aut(Td,∞ )
that always contains, and in some cases equals, the arboreal Galois group G∞ . In [4],
this question was answered in the affirmative for the PCF cubic polynomial −2z 3 +
3z 2 , including an explicit computation of the subgroup E∞ ⊆ Aut(T3,∞ ) and a simple
sufficient condition on K and x0 for G∞ to be all of E∞ . In the present paper, we do
the same for the PCF quadratic polynomial z 2 − 1.
For the rest of the paper, then, let f (z) = z 2 − 1, and let T∞ and Tn denote the
binary rooted trees T2,∞ and T2,n , respectively. The two critical points 0, ∞ of f are
both periodic, with ∞ 7→ ∞ and 0 7→ −1 7→ 0. Over the function field K = C(t)
with x0 = t, a setting in which arboreal Galois groups are often known as profinite
iterated monodromy groups, G∞ is isomorphic to the closure B ∞ of a well-understood
subgroup B∞ of Aut(T∞ ) called the basilica group. (See [23, Section 6.12.1], as well as
[3, Section 5], especially Theorem 5.8 and following.) Here and throughout this paper,
when we say that two groups that act on a tree are isomorphic, we mean not only that
they are isomorphic as abstract groups, but that the isomorphism respects the action
on the tree.
More generally, in [25, Theorem 2.5.6], Pink showed for any algebraically closed field
k not of characteristic 2, then with K = k(t) and x0 = t, the arboreal Galois group G∞
is isomorphic to B ∞ . Pink also showed that for function fields K = k(t) where k is not
algebraically closed, the arboreal Galois group G∞ is an extension of B ∞ by a subgroup
of the 2-adic multiplicative group Z× 2 , via the cyclotomic character Gal(k/k) → Z2 .
×

(See [25, Theorem 2.8.4].) We define and discuss B ∞ in Section 2.

However, our interest extends to the case that the field K is a number field, where
Pink’s results in [25] are suggestive but do not apply directly. Instead, we give an explicit
definition of a subgroup M∞ ⊆ Aut(T∞ ) that we call the arithmetic basilica group and
which is an extension of B ∞ by Z× 2 . More specifically, for each σ ∈ Aut(T∞ ) and each
node x of the tree T∞ , we define a quantity P (σ, x) ∈ Z× 2 , which in turn we use to define
M∞ . Our main results can be summarized as follows.
Main Theorem. Let K be a field of characteristic different from 2, and let x0 ∈ K.
Let G∞ be the arboreal Galois group for f (z) = z 2 − 1 over K, rooted at x0 . Then:
(1) G∞ is isomorphic to a subgroup of the arithmetic basilica group M∞ .
(2) The following are equivalent:
(a) G∞ ∼ = M∞ .
(b) G5 =∼
√ M5 . √
(c) [K( −x0 , 1 + x0 , ζ8 ) : K] = 16.
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 3

Here, Mn denotes the quotient of M∞ formed by restricting to its action on the subtree
Tn , and ζ8 denotes a primitive eighth root of unity.
The above theorem shows that, like the map z 7→ −2z 3 + 3z 2 of [4], the PCF map
f (z) = z 2 − 1 has an associated subgroup M∞ ⊆ Aut(T∞ ) that always contains and
sometimes equals the arboreal Galois group G∞ . Condition (2b) shows that this equality
is attained for the entire tree if it is already attained at the fifth level, and condition (2c)
is very easy to check in practice.
We note that if [K(ζ8 ) : K] = 4, then √ by Hilbert’s
√ irreducibility theorem, there are
many choices of x0 ∈ K for which [K( −x0 , 1 + x0 , ζ8) : K] = 16, since the set of
x0 ∈ K failing this condition is a thin set, in the sense of Serre. For example, if K = Q,
then the condition holds for
x0 or − 1 − x0 in {5, 6, 10, 11, 12, 13, 14, 19, 20, 21, 22, 23 . . .}
among (infinitely) many other examples.
√ √
On the other hand, even when [K( −x0 , 1 + x0 , ζ8 ) : K] < 16, our computations
suggest the following conjecture.
Conjecture 1. Let K be a number field. Then for all but finitely many choices of
x0 ∈ K, the associated arboreal Galois group G∞ has finite index in M∞ .
We must allow for finitely many exceptional x0 in Conjecture 1; for example, it is not
hard to see that [M∞ : G∞ ] = ∞ if x0 is periodic. More generally, in light of our main
theorem and the results of [4], as well as [17, Conjecture 1.1] and [8, Theorem 1.1], we
propose the following broader conjecture.
Conjecture 2. Let φ(z) ∈ Q(z) be a rational function of degree d ≥ 2. Then there is a
subgroup G(φ) ⊆ Aut(Td,∞ ) with the following property:
Let K be a number field over which φ is defined, and let x0 ∈ P1 (K). Then the
associated arboreal Galois group G∞ is isomorphic to a subgroup of G(φ). Moreover, it
is possible to choose K and x0 so that G∞ is the full group G(φ).
If Conjecture 2 is true, then one can ask for sufficient conditions that [G(φ) : G∞ ] < ∞.
Besides periodic x0 , we also have [G(φ) : G∞ ] = ∞ if some Orb− φ (x0 ) contains a critical
point of f ; if φ is not PCF, then this can happen for an infinite (but thin) set of x0 ∈ K.
Another example arises for f (z) = z 2 : for x0 = −1, we have K∞ = L, where L = K(ζ2∞ ),
∞
but for x0 = 3, we have K∞ = L(31/2 ), which is an infinite extension of L.
The outline of the paper is as follows. The first three sections are purely group-
theoretic. In Section 1, given a labeling of the tree T∞ := T2,∞ , we define the quantity
P (σ, x) ∈ Z2 mentioned just before our Main Theorem earlier. We then use P to define
the arithmetic basilica group M∞ ⊆ Aut(T∞ ). In Theorem 1.4, we prove that M∞ is
indeed a group, and that the restriction of P to M∞ is a homomorphism. In Section 2, we
recall the definition and some properties of the closed basilica group B ∞ . We also study
the finite groups Mn and Bn formed by restricting M∞ and B ∞ to the finite subtree Tn .
Via a number of technical lemmas, we prove Theorem 2.7, giving sufficient conditions
for generating certain subgroups of Bn . Furthermore, in Corollary 2.12, we show that
the kernel of P : M∞ → Z× 2 is precisely B ∞ . Section 3 concerns the relationships among
B ∞ , M∞ , Bn , and Mn , encapsulated in Theorems 3.1, 3.2, and 3.3.
In the remaining two sections, we discuss the action of Galois on the tree Orb− f (x0 ). In
2
Section 4, we relate M∞ to the arboreal Galois group G∞ of f (z) = z − 1. Specifically,
4 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

aaa aab aba abb baa bab bba bbb

aa ab ba bb

a b

x0

Figure 1. A labeling of T3

Theorem 4.4 shows that G∞ embeds in M∞ after an appropriate labeling of the tree
Orb− Main Theorem. Section 5 is devoted to the proof
f (x0 ), proving statement (1) of our √ √
of Theorem 5.4, that the condition [K( −x0 , 1 + x0 , ζ8 ) : K] = 16 implies G∞ ∼ = M∞ .
Finally, Corollary 5.5 proves statement (2) of our Main Theorem.

1. A special infinite sum on the tree

Let T∞ denote a rooted binary tree, extending infinitely above the root point x0 . For
each n ≥ 0, let Tn denote the finite subtree of T∞ from x0 up to the n-th level above x0 .
Definition 1.1. A labeling of T∞ is a choice of two tree morphisms a, b : T∞ → T∞ such
that a maps T∞ bijectively onto the subtree rooted at one of the two nodes connected
to x0 , and b maps T∞ bijectively onto the subtree rooted at the other.
For any integer n ≥ 1, a labeling of Tn is a choice of two injective tree morphisms
a, b : Tn−1 → Tn with the same property.
To see why the choice of maps a, b in Definition 1.1 should be considered a “labeling”
of each node of the tree, consider a node y at the m-th level of T∞ . By our choice of
the maps a, b, there is a unique ordered m-tuple (s1 , . . . , sm ) ∈ {a, b}m such that y =
s1 ◦ · · · ◦ sm (x0 ). Thus, it makes sense to label the node y with the m-tuple (s1 , . . . , sm ).
The node directly underneath y then has label (s1 , . . . , sm−1 ). We will usually dispense
with the punctuation and write s1 s2 · · · sm instead of (s1 , . . . , sm ). We will also frequently
abuse notation and refer to a node x and its label in {a, b}m interchangeably.
Note that the order we have written the m-tuple (s1 , . . . , sm ) is also the order we trace
up the tree when following the path from x0 to y. That is, s1 tells us whether to go left
(a) or right (b) to get from the root node to level 1; s2 tells us whether to go left or right
from there to level 2; and so on until we arrive at y. See Figure 1.
For any level m ≥ 0, a tree automorphism σ ∈ Aut(T∞ ) or σ ∈ Aut(Tn ) must satisfy
the following properties.
(1) σ permutes the labels in {a, b}m , and
(2) for each (s1 , . . . , sm ) ∈ {a, b}m , we have either
σ(s1 · · · sm a) = σ(s1 · · · sm )a and σ(s1 · · · sm b) = σ(s1 · · · sm )b
or
σ(s1 · · · sm a) = σ(s1 · · · sm )b and σ(s1 · · · sm )b = σ(s1 · · · sm )a.
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 5

For any tree automorphism σ and m-tuple x ∈ {a, b}m , we define the parity Par(σ, x) of
σ at x to be
(
0 if σ(xa) = σ(x)a and σ(xb) = σ(x)b
(1) Par(σ, x) :=
1 if σ(xa) = σ(x)b and σ(xb) = σ(x)a
Observe that any set of choices of Par(σ, x) for each node x of T∞ (respectively, Tn−1 )
determines a unique automorphism σ ∈ Aut(T∞ ) (respectively, σ ∈ Aut(Tn )).
If σ(x) = x, then Par(σ, x) is 0 if σ fixes the two nodes above x, or 1 if it transposes
them. However, Par(σ, x) is defined even when σ(x) 6= x, although in that case its value
depends also on the labeling of the tree.
Definition 1.2. Fix a labeling of T∞ , and let σ ∈ Aut(T∞ ). For any node x of T∞ ,
define
X X
(2) Q(σ, x) := 2i Par(σ, xas1 as2 · · · asi ) ∈ Z2 ,
i≥0 s1 ,...,si ∈{a,b}

and
X X
(3) P (σ, x) := (−1)Par(σ,x) + 2 Q(σ, xbt) − 2 Q(σ, xat) ∈ Z×
2.
t∈{a,b} t∈{a,b}

In addition, for any n ≥ m ≥ 0, any node x at level m of Tn , and any τ ∈ Aut(Tn ), set
j := ⌊(n − m + 1)/2⌋, and define P (τ, x) ∈ (Z/2j Z)× to be
(4) P (τ, x) :≡ P (τ̃ , x) (mod 2j ),
where τ̃ ∈ Aut(T∞ ) is any extension of τ to all of T∞ .
Regarding equation (4), note that every τ ∈ Aut(Tn ) has infinitely many extensions
τ̃ ∈ Aut(T∞ ), since we may choose the parity Par(τ̃ , y) at each node y at levels n+ 1 and
higher to be either 0 or 1 as we please. However, the definition of P (τ, x) in equation (4)
is independent of the extension τ̃ , since the contributions of Par(τ̃ , y) for nodes y at
levels n + 1 and higher from equations (2) and (3) all have coefficients divisible by 2j .
That is, when computing P (τ, x), we may simply truncate the sums in equations (2)
and (3) to include only the contributions from nodes at levels n − 1 and below.
It is immediate from equation (2) that
X
(5) Q(σ, x) = Par(σ, x) + 2 Q(σ, xas),
s∈{a,b}

where we understand this equation to be an equality in Z× 2 in the T∞ case, and a

congruence modulo an appropriate power of 2 in the Tn case.
To help explain Definition 1.2, observe that P (σ, x) is ±1 plus a weighted sum of
Par(σ, y) at certain nodes y, chosen based on the labeling of the tree. For example,
Figure 2 shows the nodes in question up to level 5. To compute P = P (σ, x) in that
case, we count the highlighted nodes as follows:
• count gray circles y for which Par(σ, y) = 1 with weight −2,
• count white circles y for which Par(σ, y) = 1 with weight 2,
• count gray squares y for which Par(σ, y) = 1 with weight −4,
• count white squares y for which Par(σ, y) = 1 with weight 4,
6 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

xa xb
x

Figure 2. P (σ, x) is a weighted sum of Par(σ, y) at the highlighted nodes y.

and so on up the tree. Finally, add 1 if Par(σ, x) = 0 or add −1 if Par(σ, x) = 1.

The strange form of P (σ, x) in Definition 1.2 turns out to be exactly what is needed
to carve out the correct subgroup of Aut(T∞ ) to serve as our arithmetic basilica group.
More precisely, as we will see in Theorem 4.4, if ζ is a primitive 2j -th root of unity, then
an arboreal Galois element σ maps ζ to ζ P (σ,x) . The correct group, then, must consist
only of those tree automorphisms σ for which P (σ, x) is the same for every node x in
the tree, as follows.
Definition 1.3. Fix a labeling a, b of T∞ . Let x0 denote the (empty) label of the
root point of the tree. Define the arithmetic basilica group M∞ to be the set of all
σ ∈ Aut(T∞ ) for which
P (σ, x) = P (σ, x0 ) for every node x ∈ T∞ .
Similarly, given n ≥ 1 and a labeling a, b of Tn , define Mn to be the set of all σ ∈ Aut(Tn )
for which the following condition holds: for every m ≥ 0, we have
(6) P (σ, x) ≡ P (σ, x0 ) (mod 2j ) for every node x ∈ {a, b}m ,
where j := ⌊(n − m + 1)/2⌋.
Theorem 1.4. The following hold.
(1) M∞ is a subgroup of Aut(T∞ ).
(2) For every n ≥ 1, Mn is a subgroup of Aut(Tn ).
(3) The map P : M∞ → Z× 2 given by P : σ 7→ P (σ, x0 ) is a group homomorphism.
(4) For every n ≥ 1, the map P : Mn → (Z/2j Z)× given by P : σ 7→ P (σ, x0 ), where
j := ⌊(n + 1)/2⌋, is a group homomorphism.
Proof. We prove statements (1) and (3); the proofs of statements (2) and (4) are similar.
Step 1. For any σ ∈ Aut(T∞ ) and any node x of T∞ , define
(7) sgn1 (σ, x) := (−1)Par(σ,x) = 1 − 2 Par(σ, x).
It is immediate from equation (1) that for any σ, τ ∈ Aut(T∞ ) and any node x of T∞ ,
we have


(8) sgn1 (στ, x) = sgn1 σ, τ (x) · sgn1 (τ, x).
We also have



(9) Par(στ, x) = Par σ, τ (x) + sgn1 σ, τ (x) Par(τ, x).
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 7

Equation (9) follows from equation (8) by writing Par(·, ·) = (1 − sgn1 (·, ·))/2, or simply
by checking the four possible choices of Par(τ, x) and Par(σ, τ (x)).
Step 2. For any σ ∈ M∞ , any τ ∈ Aut(T∞ ), and any node x of T∞ , define


Zσ,τ (x) := Q σ, τ (x) + P (σ)Q(τ, x) − Q(στ, x),
where P (σ) is the constant value of P (σ, w) for all nodes w of T∞ . We claim that
Zσ,τ (x) = 2Zσ,τ (xaa) + 2Zσ,τ (xab).
To see this, expand each appearance of Q in the definition of Zσ,τ (x) according to
equation (5), yielding


Zσ,τ (x) = Par σ, τ (x) + P (σ) Par(τ, x) − Par(στ, x)
X h
i
+2 Q σ, τ (x)as + P (σ)Q(τ, xas) − Q(στ, xas)
s∈{a,b}



(10) = Par σ, τ (x) + sgn1 σ, τ (x) Par(τ, x) − Par(στ, x)
X h

i
(11) + 2 Par(τ, x) Q σ, τ (x)bs − Q σ, τ (x)as
s∈{a,b}
X h
i
(12) +2 Q σ, τ (x)as + P (σ)Q(τ, xas) − Q(στ, xas) ,
s∈{a,b}

where in the second equality, we expanded the first appearance of P (σ) as P (σ, τ (x)).
The expression on line (10) is zero by equation (9). Next, observe that
(
 τ (xaa), τ (xab) if Par(τ, x) = 0,
(13) τ (x)aa, τ (x)ab = 
τ (xba), τ (xbb) if Par(τ, x) = 1,
and similarly for the set {τ (x)ba, τ (x)bb}. Thus, the expression on line (11) is

0 X h
 if Par(τ, x) = 0,


i
 2 Q σ, τ (xas) − Q σ, τ (xbs) if Par(τ, x) = 1.

s∈{a,b}

and the expression on line (12) is

 X h
i


 2 Q σ, τ (xas) + P (σ)Q(τ, xas) − Q(στ, xas) , if Par(τ, x) = 0,

s∈{a,b} h i
X



 2 Q σ, τ (xbs) + P (σ)Q(τ, xas) − Q(στ, xas) , if Par(τ, x) = 1.

s∈{a,b}

For either possible value of Par(τ, x), then, we have

X h
i
Zσ,τ (x) = 2 Q σ, τ (xas) + P (σ)Q(τ, xas) − Q(στ, xas)
s∈{a,b}

= 2Zσ,τ (xaa) + 2Zσ,τ (xab),

proving our claim.
Step 3. As in Step 2, consider σ ∈ M∞ and τ ∈ Aut(T∞ ). We claim that


(14) Q σ, τ (x) + P (σ)Q(τ, x) = Q(στ, x)
8 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

for every node x of T∞ . That is, we are claiming that Zσ,τ (x) = 0 for every x. To prove
this, it suffices to show, for each j ≥ 0, that for every node x, we have Zσ,τ (x) ∈ 2j Z2 .
We proceed by induction on j. The base case j = 0 is immediate from the fact that
P (·, ·), Q(·, ·) ∈ Z2 . Assuming the statement holds for all x for some particular j ≥ 0,
then for any node x, Step 2 yields
Zσ,τ (x) = 2 Zσ,τ (xaa) + Zσ,τ (xab) ∈ 2 2j Z2 = 2j+1Z2 ,

completing the induction and proving our claim.

Step 4. As in the previous two steps, consider σ ∈ M∞ and τ ∈ Aut(T∞ ), and
consider a node x in T∞ . We claim that
(15) P (σ)P (τ, x) = P (στ, x).
Indeed, expanding P (τ, x) yields
X X
P (σ)P (τ, x) = sgn1 (τ, x)P (σ) + 2 P (σ)Q(τ, xbt) − 2 P (σ)Q(τ, xat)
t∈{a,b} t∈{a,b}
 X X 




= sgn1 (τ, x) sgn1 σ, τ (x) + 2 Q σ, τ (x)bt − 2 Q σ, τ (x)at
t∈{a,b} t∈{a,b}
X X
+2 P (σ)Q(τ, xbt) − 2 P (σ)Q(τ, xat),
t∈{a,b} t∈{a,b}

where we have also expanded the first appearance of P (σ) as P (σ, τ (x)). Applying
equations (8) and (13), then, we have
X h
i
P (σ)P (τ, x) = sgn1 (στ, x) + 2 Q σ, τ (xbt) + P (σ)Q(τ, xbt)
t∈{a,b}
X h
i
−2 Q σ, τ (xat) + P (σ)Q(τ, xat)
t∈{a,b}
X X
= sgn1 (στ, x) + 2 Q(στ, xbt) − 2 Q(στ, xat) = P (στ, x),
t∈{a,b} t∈{a,b}

where we used identity (14) twice in the second equality, thus proving our claim.
Step 5. To prove statement (1), first observe that the identity automorphism e ∈
Aut(T∞ ) belongs to M∞ , since Par(e, y) = 0 for all nodes y, and hence P (e, x) = 1 for
all nodes x of T∞ . Next, given σ, τ ∈ M∞ , it follows from identity (15) that for any node
x of T∞ , we have
P (στ, x) = P (σ)P (τ, x) = P (σ)P (τ, x0) = P (στ, x0 ),
and hence στ ∈ M∞ . Finally, given σ ∈ M∞ , consider σ −1 ∈ Aut(T∞ ). Then for any
node x of T∞ , identity (15) again yields
1 = P (e, x) = P (σσ −1 , x) = P (σ)P (σ −1, x).
Thus,
P σ −1 , x = P (σ)−1 = P σ −1 , x0 ,

and therefore σ −1 ∈ M∞ . That is, M∞ is indeed a subgroup of Aut(T∞ ).

Finally, the fact that P : M∞ → Z× 2 is a homomorphism is immediate from iden-
tity (15), proving statement (2). 
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 9

2. The basilica group and finite subtrees

For any 0 ≤ n ≤ m ≤ ∞, define a function Rm,n : Aut(Tm ) → Aut(Tn ) by restricting
σ ∈ Aut(Tm ) to Tn . Clearly Rm,n is a homomorphism.
The group Aut(T∞ ) has a topological structure, as follows. For each m ≥ 0, let
Wm := ker(R∞,m ). That is, Wm consists of all σ ∈ Aut(T∞ ) that act trivially on levels
0 through n of the tree. The cosets of the normal subgroups Wm form a basis for a
topology on Aut(T∞ ), making Aut(T∞ ) compact and Hausdorff. For n ≥ m ≥ 0, we
will often abuse notation and write Wm for the subgroup R∞,n (Wm ).
For any node x of T∞ , it is immediate from Definition 1.2 that σ 7→ P (σ, x) is a
continuous function from Aut(T∞ ) to Z× 2 . It follows that M∞ is a closed and hence
compact subgroup of Aut(T∞ ).
Fix a labeling of the tree T∞ . Define two particular automorphisms α, β ∈ Aut(T∞ )
by specifying that
• Par(α, bb · · · b) = 1 for any node whose label is a string of an odd number of b’s,
• Par(β, bb · · · b) = 1 for any node whose label is a string of an even number of b’s,
• Par(α, y) = Par(β, y) = 0 for all other nodes y of T∞ .
The maps α and β can be equivalently defined by the recursive relations
α(aw) = aw, α(bw) = bβ(w), β(aw) = bw, β(bw) = aα(w)
for any word w in the symbols a, b.
Definition 2.1. The basilica group is the subgroup B∞ of Aut(T∞ ) generated by α and
β. The closed basilica group is the topological closure B ∞ of B∞ in Aut(T∞ ).
Remark 2.2. Consider σ ∈ W1 , i.e., consider σ ∈ Aut(T∞ ) that fixes level 1 of the tree.
Then σ acts on the subtree T∞,a rooted at a as some automorphism σa ∈ Aut(T∞ ), and
similarly on the subtree T∞,b rooted at b as some σb ∈ Aut(T∞ ). That is, we may write
σ = (σa , σb ).
In this notation, we have β = (e, α), where e is the identity element of Aut(T∞ ).
Similarly, α−1 βα and α2 also belong to W1 , and simple computations show that α−1 βα =
(α, e) and α2 = (β, β).
Consider σ ∈ B∞ ∩ W1 . Then σ must be a finite product of α and β involving an
even number of copies of α. (The parity condition on α is because σ ∈ W1 ). Any
such product can also be written as a product of powers of α2 , β, and α−1 βα. Thus,
writing σ = (σa , σb ), we must have σa , σb ∈ B∞ . Conversely, for any σb ∈ B∞ , there is
some σa ∈ B∞ such that (σa , σb ) ∈ B∞ ∩ W1 . For this reason, the basilica group B∞
is said to be a self-similar group. See [23] for more on self-similar groups, especially
Sections 3.10.2, 5.2.2, and 6.12.1, which specifically concern B∞ .
Definition 2.3. Fix n ≥ 1 and a labeling of T∞ . Define
(1) Bn := R∞,n (B∞ ) = R∞,n (B ∞ ).
(2) Bn′ := R∞,n (ker(P : M∞ → Z× 2 )).
′′ j ×
(3) Bn := ker(P : Mn → (Z/2 Z) ), where j := ⌊(n + 1)/2⌋.
(4) En := Wn−1 ∩ Bn , En′′ := Wn−1 ∩ Bn′′ , and Un := Wn−1 ∩ Mn .
In addition, define B0 = B0′ = B0′′ = E0 = E0′′ = U0 := R∞,0 (Aut(T∞ )), which is the
trivial group acting on the trivial tree.
10 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

Recall from Theorem 1.4 that the maps P used to define Bn′ and Bn′′ above are indeed
homomorphisms, so that all three of Bn , Bn′ , and Bn′′ are subgroups of Aut(Tn ). More-
over, a simple computation shows P (α) = P (β) = 1 for every node x of the tree T∞ .
Thus, we have Bn ⊆ Bn′ ⊆ Bn′′ . (In fact, as we will see in Corollary 2.12, these three
groups coincide.)
Note that En is a normal subgroup of Bn , because Wn−1 is a normal subgroup of
Aut(T∞ ). In addition, once we know that B ∞ = ker(P : M∞ → Z× 2 ), it follows imme-
diately that both En and Bn are normal subgroups of Mn . However, none of Bn , En ,
or Mn is a normal subgroup of Aut(Tn ), because conjugation by an arbitrary element of
Aut(Tn ) has the effect of relabeling the tree, which in turn changes the function P .
Definition 2.4. Let n ≥ m ≥ 1, let σ ∈ En′′ , and let x ∈ {a, b}n−m . If m is odd, write
m = 2i + 1, and define
′
sgnm (σ, x) := (−1)Q (σ,x) ,
where X
Q′ (σ, x) := Par(σ, xas1 as2 · · · asi ) ∈ Z,
s1 ,...,si ∈{a,b}
′
and where we understand Q (σ, x) = Par(σ, x) if m = 1. If m is even, define
(16) sgnm (σ, x) := sgnm−1 (σ, xa) · sgnm−1 (σ, xb).
The quantities Q′ of Definition 2.4 and Q of equation (2) are related as follows.
Suppose σ ∈ En′′ and that x is a node at level n − m where m = 2i + 1. Then for any
extension σ̃ of σ to the full tree T∞ , we have 2−i Q(σ̃, x) ≡ Q′ (σ, x) (mod 2).
When m = 1, the quantity sgn1 (σ, x) above coincides with our previous definition of
sgn1 (σ, x) : 1−2 Par(σ, x) from equation (7). On the other hand, in [25, Section 1.5], Pink
defines a notation sgnn (σ) which is completely different from the quantity sgnm (σ, x) in
Definition 2.4 above.
Lemma 2.5. Let n ≥ m ≥ 1, let σ1 , σ2 ∈ En′′ , let τ ∈ Mn , and let x ∈ {a, b}n−m . Then
(1) sgnm (σ1 σ2 , x) = sgnm (σ1 , x) · sgnm (σ2 , x). That is, sgnm (·, x) : En′′ → {±1} is a
group homomorphism.
(2) If m ≥ 3 is odd, then sgnm−1 (σ1 , xa) = sgnm−1 (σ1 , xb) = sgnm (σ1 , x).
(3) τ σ1 τ −1 ∈ En′′ , and sgnm (τ σ1 τ −1 , τ (x)) = sgnm (σ1 , x).
Proof. (1): By equation (9), we have
Par(σ1 , y) + Par(σ2 , y) ≡ Par(σ1 σ2 , y) (mod 2) for all nodes y ∈ {a, b}n−1 .
For m odd, it follows immediately that Q′ (σ1 , x) + Q′ (σ2 , x) ≡ Q′ (σ1 σ2 , x) (mod 2), and
therefore that sgnm (σ1 σ2 , x) = sgnm (σ1 , x) · sgnm (σ2 , x). For m even, we have
sgnm (σ1 σ2 , x) = sgnm−1 (σ1 σ2 , xa) · sgnm−1 (σ1 σ2 , xb)
= sgnm−1 (σ1 , xa) sgnm−1 (σ2 , xa) sgnm−1 (σ1 , xb) sgnm−1 (σ2 , xb)
= sgnm (σ1 , x) sgnm (σ2 , x).
(2): By definition of Q′ , we have
X X
Q′ (σ1 , x) = Par(σ1 , xaaas2 · · · asi ) + Par(σ1 , xabas2 · · · asi )
s2 ,...,si ∈{a,b} s2 ,...,si ∈{a,b}
′ ′
= Q (σ1 , xaa) + Q (σ1 , xab),
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 11

and hence
sgnm (σ1 , x) = sgnm−2 (σ1 , xaa) sgnm−2 (σ1 , xab) = sgnm−1 (σ1 , xa).
Write m = 2i + 1. Since σ1 ∈ En′′ , we have P (σ1 , x) ≡ 1 (mod 2i+1 ), and therefore
2i Q(σ1 , xaa) + 2i Q(σ1 , xab) ≡ 2i Q(σ1 , xba) + 2i Q(σ1 , xbb) (mod 2i+1 ).
Thus, Q′ (σ1 , xaa) + Q′ (σ1 , xab) ≡ Q′ (σ1 , xba) + Q′ (σ1 , xbb) (mod 2), and hence
sgnm−1 (σ1 , xa) = sgnm−1 (σ1 , xaa) + sgnm−1 (σ1 , xab)
= sgnm−1 (σ1 , xba) + sgnm−1 (σ1 , xbb) = sgnm−1 (σ1 , xb).
(3): The inclusion τ σ1 τ −1 ∈ En′′ is immediate from the facts that Bn′′ ⊆ Mn and
Wn−1 ⊆ Aut(Tn ) are both kernels of homomorphisms. To prove the desired identity, we
proceed by induction on m. For m = 1, we have sgn1 (τ σ1 τ −1 , τ (x)) = sgn1 (σ1 , x) by
equation (8).
For m ≥ 2, assuming the statement is true for m − 1, suppose first that m is even.
By equation (16) of Definition 2.4, we have
sgnm τ σ1 τ −1 , τ (x) = sgnm−1 τ σ1 τ −1 , τ (x)a sgnm−1 τ σ1 τ −1 , τ (x)b

= sgnm−1 τ σ1 τ −1 , τ (xa) sgnm−1 τ σ1 τ −1 , τ (xb)

= sgnm−1 (σ, xa) sgnm−1 (σ, xb) = sgnm (σ, x),

where the second equality is by swapping the order of the two multiplicands if necessary,
and the third is by our inductive hypothesis. On the other hand, if m is odd, then by
part (2), we have
sgnm τ σ1 τ −1 , τ (x) = sgnm−1 τ σ1 τ −1 , τ (x)s for both s = a, b

= sgnm−1 τ σ1 τ −1 , τ (xa) = sgnm−1 (σ1 , xa) = sgnm (σ1 , a) 

i
Remark 2.6. For n = 2i + 1 odd, a simple computation shows that R∞,n (α)2 ∈ Aut(Tn )
is given by
(
i 1 if y = s1 bs2 b · · · si b for some s1 , . . . , si ∈ {a, b},
Par R∞,n (α)2 , y =


0 else,
and hence
i i
R∞,n (α)2 ∈ En , with sgnn R∞,n (α)2 , x0 = −1.

Similarly, for n = 2i even, we have

(
(i−1) 1 if y = bs1 bs2 b · · · si−1 b for some s1 , . . . , si−1 ∈ {a, b},
Par R∞,n (β)2


,y =
0 else,
and hence
(i−1) (i−1)
R∞,n (β)2 sgnn R∞,n (β)2


∈ En , with , x0 = −1.
In particular, sgnn (·, x0 ) is a nontrivial map.
Theorem 2.7. Let n ≥ 1, and let G ⊆ Bn′′ be a subgroup satisfying
(1) Rn,n−1(G ∩ Bn ) ⊇ Bn−1 , and
(2) there is some λ ∈ G ∩ En such that sgnn (λ, x0 ) = −1.
Then En′′ = G ∩ En = En .
12 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

The proof of Theorem 2.7 relies on the following lemmas.

Lemma 2.8. For every n ≥ m ≥ 0, Bn acts transitively on the 2m nodes at level m of
Tn .
Proof. We proceed by induction on m ≥ 0. The statement is trivial for m = 0.
For m ≥ 1, assume the statement holds for m − 1. Given nodes x, y at level m,
after possibly applying α to one or both, we may assume that x = bv and y = bw
for v, w ∈ {a, b}m−1 . By our inductive hypothesis, there is some σb ∈ B∞ such that
σb (v) = w. Therefore, as noted in Remark 2.2, there is some σa ∈ B∞ such that the
automorphism σ := (σa , σb ) ∈ W1 lies in B∞ . We have σ(x) = y, as desired. 
Lemma 2.9. Fix integers n > m ≥ 0. Let σ ∈ En′′ and τ ∈ Mn . Define µ := στ σ −1 τ −1 .
For any node w at level m of the tree for which τ (w) = w, we have sgnn−m−1 (µ, wa) =
sgnn−m−1 (µ, wb), and this common value is
• −1 if sgnn−m−1 (σ, wa) 6= sgnn−m−1 (σ, wb) and Par(τ, w) = 1, or
• +1 otherwise.
Proof. If Par(τ, w) = 0, then τ (wa) = wa and τ (wb) = wb. By Lemma 2.5.(3), then, we
have
sgnn−m−1 (τ σ −1 τ −1 , wa) = sgnn−m−1 (σ −1 , wa) = sgnn−m−1 (σ, wa),
and hence by Lemma 2.5.(1), we have
sgnn−m−1 (µ, wa) = sgnn−m−1 (σ, wa) · sgnn−m−1 (τ σ −1 τ −1 , wa)
= sgnn−m−1 (σ, wa) · sgnn−m−1 (σ, wa) = +1,
and similarly sgnn−m−1 (µ, wb) = +1, as desired.
Assume for the remainder of the proof that Par(τ, w) = 1, and hence that τ (wa) = wb
and τ (wb) = wa. By Lemma 2.5.(3) again, we have
sgnn−m−1 (τ σ −1 τ −1 , wa) = sgnn−m−1 (σ −1 , wb) = sgnn−m−1 (σ, wb).
Therefore, by Lemma 2.5.(1), we have
sgnn−m−1 (µ, wa) = sgnn−m−1 (σ, wa) · sgnn−m−1 (τ σ −1 τ −1 , wa)
= sgnn−m−1 (σ, wa) · sgnn−m−1 (σ, wb),
which is +1 if sgnn−m−1 (σ, wa) = sgnn−m−1 (σ, wb), and −1 otherwise. A similar compu-
tation yields the same result for sgnn−m−1 (µ, wb), completing the proof. 
Lemma 2.10. Fix integers n > m ≥ 0, and let G ⊆ Bn′′ be a subgroup satisfying
(1) Rn,n−1(G ∩ Bn ) ⊇ Bn−1 , and
(2) for every σ ∈ En′′ , there exists τ ∈ G ∩ En such that
(17) sgnn−m (τ, w) = sgnn−m (σ, w) for all nodes w at level m.
Suppose that n − m is even. Then for every node w at level m, there exists µw ∈ G ∩ En
such that for each node y at level m + 1, we have
(
−1 if y = wa or y = wb,
(18) sgnn−m−1 (µw , y) =
+1 otherwise.
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 13

Proof. Case 1: n is odd. Write n = 2ℓ + 1 and m = 2j + 1 for integers ℓ > j ≥ 0. By

ℓ
hypothesis (2) applied to R∞,n (α2 ) ∈ En ⊆ En′′ , there is some γ ∈ G ∩ En such that
ℓ
sgnn−m (γ, w) = sgnn−m α2 , w for all nodes w at level m.

By hypothesis (1), there is some δ ∈ G ∩ Bn such that

j

Rn,n−1 (δ) = R∞,n−1 β 2 .
Let µ := γδγ −1 δ −1 . Note that µ ∈ G ∩ En , since γ ∈ G ∩ En , and En is normal subgroup
of Bn .
A slight generalization of Remark 2.6 shows that for any node y at level m + 1, we
have
(
−1 if y = s0 bs1 b · · · bsj for some s0 , . . . , sj ∈ {a, b},
sgnn−m−1 (γ, y) =
+1 otherwise,

and for any node w at level m, we have

(
1 if w = bt1 bt2 · · · btj for some t1 , . . . , tj ∈ {a, b},
Par(δ, w) =
0 otherwise.

Therefore, by Lemma 2.9, for any node y at level m + 1, we have

(
−1 if y = bb · · · bs for some s ∈ {a, b},
sgnn−m−1 (µ, y) =
+1 otherwise.

Thus, µ is the desired automorphism µx for the node x = bb · · · b at level m of the tree.
By Lemma 2.8 and hypothesis (1), the group G ∩ Bn acts transitively on the nodes of
Tn at level m. Thus, for each node w at level m, there is some ρw ∈ G ∩ Bn such that
ρw (bb · · · b) = w. By Lemma 2.5.(3), the automorphism µw := ρw µρ−1
w ∈ G ∩ En satisfies
equation (18), and we are done.
Case 2: n is even. Write n = 2ℓ and m = 2j for integers ℓ > j ≥ 0. Hypothesis (2)
yields the existence of some γ ∈ G ∩ En such that
ℓ−1
sgnn−m (γ, w) = sgnn−m β 2 , w for all nodes w at level m,

and hypothesis (1) yields some δ ∈ G ∩ Bn such that

j

Rn,n−1 (δ) = R∞,n−1 α2 .
As before, let µ := γδγ −1 δ −1 ∈ G ∩ En . By similar reasoning as in Case 1, it follows
that µ is the desired µx for the node x = bb · · · b, and then that conjugating yields the
desired µw for each node w at level m. 
Lemma 2.11. Fix integers m ≥ 0 and n ≥ m + 2, and let G ⊆ Bn′′ be a subgroup
satisfying hypotheses (1) and (2) of Lemma 2.10. Then for every σ ∈ En′′ , there exists
τ̃ ∈ G ∩ En such that
sgnn−m−1 (τ̃ , y) = sgnn−m−1 (σ, y) for all nodes y at level m + 1.
14 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

Proof. We consider two cases.

Case 1: n − m ≥ 3 is odd. Given σ ∈ En′′ , we may choose τ ∈ G ∩ En satisfying
equation (17), by hypothesis (2). Let τ̃ := τ . For any node y at level m + 1, let w be
the node immediately below y on level m, so that either y = wa or y = wb. Then by
Lemma 2.5.(2),
sgnn−m−1 (τ̃ , y) = sgnn−m (τ̃ , w) = sgnn−m (σ, w) = sgnn−m−1 (σ, y).
Case 2: n − m ≥ 2 is even. Given σ ∈ En′′ , we may choose τ ∈ G ∩ En satisfying
equation (17), by hypothesis (2). For each node w at level m, therefore, according to
equation (16) of Definition 2.4, we have
sgnn−m−1 (τ, wa) sgnn−m−1 (τ, wb) = sgnn−m−1 (σ, wa) sgnn−m−1 (σ, wb).
Thus, for each such w, we have
sgnn−m−1 (τ, wa) sgnn−m−1 (τ, wb)
(19) = ∈ {±1}
sgnn−m−1 (σ, wa) sgnn−m−1 (σ, wb)
Let w1 , . . . , wr be all the nodes w at level m for which the value in equation (19) is −1,
and define
τ̃ := τ µw1 µw2 · · · µwr ∈ G ∩ En ,
where each µw ∈ G ∩ En is the automorphism given by equation (18) of Lemma 2.10.
Then by Lemma 2.5.(1), all the values in equation (19) become +1 when we replace τ
by τ̃ , proving the desired conclusion. 
Proof of Theorem 2.7. Step 1. We claim that for each m = 0, . . . , n − 1, hypothesis (2)
of Lemma 2.10 holds. For m = 0, there is only one node at level 0, and e, λ ∈ G ∩ En
attain the two possible values sgnn (e, x0 ) = +1 and sgnn (λ, x0 ) = −1, where λ is the
element assumed to exist in the statement of Theorem 2.7. Proceeding inductively,
assuming the claim holds for some m ∈ {0, . . . , n − 2}, the hypotheses of Lemma 2.11
hold, since hypothesis (1) is already one of the assumptions of Theorem 2.7. Thus,
Lemma 2.11 verifies that hypothesis (2) holds for m + 1, proving the claim.
Step 2. For any σ ∈ En′′ , the claim of Step 1 for m = n − 1 shows that there is some
τ ∈ G ∩ En such that
sgn1 (τ, w) = sgn1 (σ, w) for all nodes w at level n − 1.
Since σ, τ ∈ Aut(Tn ) fix all nodes at levels below n, it follows that σ = τ ∈ G ∩ En .
Thus,
En′′ ⊆ G ∩ En ⊆ En ⊆ En′′ ,
and hence these three groups coincide, as desired. 
Corollary 2.12. For every n ≥ 0, we have Bn = Bn′ = Bn′′ and En = En′′ . Moreover,
the closed basilica group B ∞ is precisely ker(P : M∞ → Z×
2 ).

Proof. First statement: For n = 0, the relevant groups trivially coincide by Defini-
tion 2.3. Proceeding inductively for n ≥ 1, suppose the desired equalities hold for n − 1,
and let G := Bn′′ . Then
Rn,n−1 (G ∩ Bn ) = Rn,n−1(Bn ) = Bn−1 ,
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 15

so that hypothesis (1) of Theorem 2.7 holds. Hypothesis (2) is immediate from Re-
mark 2.6, and hence Theorem 2.7 yields En′′ = En .
The restriction of Rn,n−1 to Bn′′ is a surjective homomorphism ρ′′ : Bn′′ → Bn−1 ′′
with
′′
kernel En . Similarly, the restriction of Rn,n−1 to Bn is a surjective homomorphism
′′
ρ : Bn → Bn−1 with kernel En . We have Bn−1 = Bn−1 by our inductive hypothesis, and
En = En by the previous paragraph; therefore, since ρ is the restriction of ρ′′ to Bn ⊆ Bn′′ ,
′′

we have Bn = Bn′′ . The first statement now follows from the fact that Bn ⊆ Bn′ ⊆ Bn′′ .
Second statement: As noted near the start of Section 2, the map P : M∞ → Z× 2 is
continuous. Therefore, since α, β ∈ ker(P ), we have B ∞ ⊆ ker(P ).
Conversely, given σ ∈ ker(P ), define σn := R∞,n (σ) ∈ Bn′ for each n ≥ 1. By the first
statement, we have σn ∈ Bn , and hence there exists τn ∈ B∞ such that R∞,n (τn ) = σn .
For any integers n ≥ m ≥ 1, the automorphisms σ, τn ∈ Aut(T∞ ) agree on Tm−1 and
hence belong to the same coset of the subgroup Wm−1 . Thus, we have
σ = lim τn ∈ B ∞ .
n→∞

Corollary 2.13. Let n ≥ 1, and let G be a subgroup of Aut(Tn ). Suppose that
(1) Rn,n−1(G ∩ Bn ) ⊇ Bn−1 , and
(2) there is some λ ∈ G ∩ En such that sgnn (λ, x0 ) = −1.
Then G contains Bn .
Proof. Let G′ := G ∩ Bn . Then G′ satisfies the hypotheses of Theorem 2.7, and hence
En ⊆ G′ ⊆ G. Since En is the kernel of the homomorphism Rn,n−1 : Bn ։ Bn−1 , it
follows that Bn ⊆ G. 
3. From the basilica to the arithmetic basilica
Fix a labeling of T∞ . As in Section 2, we now define two more particular automor-
phisms ε, θ ∈ Aut(T∞ ), as follows. The definition of ε is simple: we specify that:
(20) Par(ε, x) = 1 for all nodes x of T∞ .
It is immediate from Definition 1.2 that Q(ε, x) = 1 + 4 + 42 + · · · = −1/3 ∈ Z2 for every
node x, and hence that P (ε, x) = −1 ∈ Z2 . In particular, ε ∈ M∞ , with P (ε) = −1.
The definition of θ is more involved, proceeding inductively up the tree. First, define
Par(θ, x0 ) := Par(θ, a) := Par(θ, b) := 0,
so that θ acts trivially on T2 . Then, once we have defined Par(θ, x) at a particular node
x, define Par(θ, y) for each node y two levels above x by:
Par(θ, xaa) := Par(θ, xab) := 0,
(21) Par(θ, xba) := 1, and
Par(θ, xbb) := Par(θ, x).
Because Par(θ, yaa) = Par(θ, yab) = 0 for any node y, we have Q(θ, x) = Par(θ, x) for
all nodes x of the tree. Thus, according to Definition 1.2 and equation (21), we have
(
(−1)0 + 2(1 + 0 − 0 − 0) = 3 if Par(θ, x) = 0,
P (θ, x) =
(−1)1 + 2(1 + 1 − 0 − 0) = 3 if Par(θ, x) = 1.
16 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

Thus, θ ∈ M∞ , with P (θ) = 3.

Theorem 3.1. The homomorphism P : M∞ → Z×
2 is surjective, with kernel B ∞ . That
is, we have the short exact sequence
P
0 −→ B ∞ −→ M∞ −→ Z×
2 −→ 0

Proof. By Theorem 1.4, the map P is a homomorphism, and by Corollary 2.12, its kernel
is B ∞ . It remains to show that P is surjective.
Because the automorphisms ε, θ defined in equations (20) and (21) belong to M∞ ,
with P (ε) = −1 and P (θ) = 3, the image of P contains the subgroup h−1, 3i of Z× 2
generated by −1 and 3. In fact, since P is continuous and M∞ is compact, it follows
that the image of P contains the closure of the subgroup h−1, 3i. However, {−1, 3} is a
set of topological generators for Z× ×
2 ; therefore, the image of P is all of Z2 . 
Theorem 3.2. Fix n ≥ 1. Then (
1 if n = 1 or n is even,
(1) En is a subgroup of Un , with [Un : En ] =
2 if n ≥ 3 is odd.
(2) Mn = R∞,n (M∞ ).
(3) If n ≥ 2, then Un ∼
= En−1 × En−1 .
As usual, the isomorphism of Theorem 3.2.(3) is of groups acting on Tn , not just of
abstract groups. For En−1 × En−1 , we mean that the first copy of En−1 acts on the copy
of Tn−1 rooted at node a, and the second acts on the copy of Tn−1 rooted at b.
Proof of Theorem 3.2. (1): From Definitions 1.3 and 2.3, and by Corollary 2.12, we have
En = ker(P : Un → (Z/2j Z)× ), where j := ⌊(n + 1)/2⌋.
In particular, En is a subgroup of Un . Observe that
(22) P (σ, x0 ) ≡ 1 (mod 2ℓ ) for all σ ∈ Un , where ℓ := ⌊n/2⌋.
If n is even, then ℓ = j, and hence P (σ, x0 ) ≡ 1 (mod 2j ) for all σ ∈ Un , whence
Un = En . Similarly, if n = 1, then because both of the conditions P (σ, x0 ) ≡ 1 (mod 20 )
and P (σ, x0 ) ≡ 1 (mod 21 ) are vacuous, we have E1 = U1 = Aut(T1 ).
For n ≥ 3 odd, we have ℓ = j − 1 ≥ 1. Thus, by equation (22), restricting P : Mn →
(Z/2j Z)× to Un yields a homomorphism
P : Un → {1 + 2ℓ + 2j Z, 1 + 2j Z} ∼
= Z/2Z.
By Corollary 2.12, the kernel of this map is precisely En . We claim it is also surjective.
Indeed, by Theorem 3.1, there is some τ ∈ M∞ with P (τ, x0 ) = 1 + 2ℓ . Because
P (τ, x0 ) ≡ 1 (mod 2ℓ ), we have R∞,n−1 (τ ) ∈ Bn−1 , and hence there exists η ∈ B∞ for
which R∞,n−1 (τ ) = R∞,n−1 (η). Let σ := R∞,n (τ η −1 ). Then
σ ∈ Un , and P (σ, x0 ) = 1 + 2ℓ + 2j Z,
proving the claim. Thus, [Un : En ] = |Z/2Z| = 2.
(2): It is immediate from Definition 1.3 that Mn ⊇ R∞,n (M∞ ).
Conversely, given σ ∈ Mn , there is some τ ∈ M∞ such that P (τ, x0 ) ≡ P (σ, x0 )
(mod 2j ), by the surjectivity of P in Theorem 3.1. (As before, we have j := ⌊(n+1)/2⌋.)
Let η := σR∞,n (τ −1 ) ∈ Mn , which satisfies P (η, x0) ≡ 1 (mod 2j ). By Corollary 2.12,
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 17

then, we have η ∈ Bn , and in fact η = R∞,n (η̃) for some η̃ ∈ B∞ . Hence, η̃τ ∈ M∞ , and
σ = R∞,n (η̃τ ), as desired.
(3): Because n ≥ 2, each σ ∈ Un fixes both nodes a and b at level 1 of the tree
Tn . As in Remark 2.2, restricting σ to the subtrees rooted at each of a and b yields
automorphisms σa , σb ∈ Aut(Tn−1 ). In fact, it is immediate from the definition of Mn
that σa , σb ∈ Mn−1 , since σ ∈ Mn−1 . Moreover, we have σa , σb ∈ Un−1 , since σa and σb
act trivially on the subtrees of n − 2 levels above each of a and b. Furthermore, because
σ acts trivially on the Tn−1 rooted at x0 , we have
P (σ, a) ≡ P (σ, b) ≡ P (σ, x0 ) ≡ 1 (mod 2⌊n/2⌋ ),
and hence σa , σb ∈ En−1 . Thus, the function σ 7→ (σa , σb ) maps Un into En−1 × En−1 ,
and it is clearly a homomorphism, with trivial kernel.
It remains to show that this function is onto. Given σa , σb ∈ En−1 , define
σ := (σa , σb ) ∈ Aut(Tn )
in the notation of Remark 2.2. That is, σ fixes the two nodes a and b at level 1, acts as
σa on the subtree rooted at a, and acts as σb on the subtree rooted at b. It suffices to
show that σ ∈ Un . Clearly σ acts trivially on the bottom n − 1 levels of Tn , so it remains
to show that σ satisfies condition (6) of Definition 1.3 for every 0 ≤ m ≤ n − 1.
Let c := P (σ, x0 ), and let ℓ := ⌊n/2⌋. Considering the (trivial) action of σ on the
copy of Tn−1 comprising the bottom n − 1 levels of Tn , we have
(23) c = P (σ, x0 ) ≡ 1 (mod 2ℓ )
by Definition 1.2, because all of the nodes y of Tn−1 for which Par(σ, y) = 1 lie at level
n − 1. If n is even, these nodes do not appear in the formula (3) defining P (σ, x0 ).
If n is odd, their terms show up with a coefficient of ±2ℓ and hence do not affect
equation (23). Thus, for any 0 ≤ m ≤ n − 1 and any node x of Tn at level n, then
setting j := ⌊(n − m + 1)/2⌋, we have three cases, as follows.
First, if x = x0 , then clearly P (σ, x) = c ≡ c (mod 2j ). Second, if x = ay for some
y ∈ {a, b}m−1 , then j ≤ ℓ, and hence
P (σ, x) = P (σa , y) ≡ 1 ≡ c (mod 2j )
by equation (23). Third, if x = by for some y ∈ {a, b}m−1 , then P (σ, x) = c (mod 2j )
by the same reasoning as in the x = ay case, this time applied to σb . Thus, we have
verified condition (6) for σ, and hence σ ∈ Un , as desired. 
Theorem 3.3. Fix n ≥ 2. Then |En | = 2en , |Un | = 2un , and |Mn | = 2mn , where
(
2n 2/3 if n is even,
en = +
3 1/3 if n is odd,

2/3 if n is even,
2n 
(24) un = + 4/3 if n is odd and n ≥ 3,
3 
1/3 if n = 1
(
2n+1 5/3 if n is even,
mn = +n−
3 4/3 if n is odd.
18 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

Proof. For n ≥ 1, let en , un , mn be the integers given by formulas (24), and define
e′n := log2 |En |, u′n := log2 |un |, m′n := log2 |Mn |.
We must show that e′n = en , u′n = un , and m′n = mn for all n ≥ 1. For n = 1,
we have En = Un = Mn = Aut(T1 ) ∼ = Z/2Z, so that e′n = u′n = m′n = 1. Clearly
en = un = mn = 1 as well.
Proceeding inductively, given n ≥ 2, assume the equalities for n − 1. By Theo-
rem 3.2.(3), we have |Un | = |En−1 |2 , and hence
u′n = 2e′n−1 = 2en−1 = un ,
where the identity 2en−1 = un is immediate from formulas (24). Next, Theorem 3.2.(1)
yields u′n = e′n for n even, and u′n = e′n + 1 for n ≥ 3 odd. Thus,
e′n = u′n = un = en if n is even, and
e′n = u′n − 1 = un − 1 = en if n is odd,
where again, each closing equality is by formulas (24). Finally, since Rn,n−1 : Mn → Mn−1
is a surjective homomorphism with kernel Un , we have m′n = m′n−1 + u′n , and hence
m′n = m′n−1 + u′n = mn−1 + un = mn ,
where again, the last equality is by formulas (24). 
To help clarify formulas (24), the following table gives the orders of the groups En ,
Un , Mn , and Aut(Tn ) for some small values of n.
n 1 2 3 4 5 6 7 8 9 10
|En | 21 22 23 26 211 222 243 286 2171 2342
|Un | 21 22 24 26 212 222 244 286 2172 2342
|Mn | 21 23 27 213 225 247 291 2177 2349 2691
| Aut(Tn )| 21 23 27 215 231 263 2127 2255 2511 21023

4. Embedding arboreal Galois groups in the arithmetic basilica

We now return from abstract subgroups of Aut(Tn ) to arboreal Galois groups. We
remind the reader of the following notation, which we set for the remainder of the paper.
K: a field of characteristic different from 2, with algebraic closure K
f: the polynomial f (z) = z 2 − 1
x0 : an element of K, to serve as the root of our preimage tree.
−n
Kn : for each n ≥ 0, the
S extension field Kn := K(f (x0 ))
K∞ : the union K∞ = n≥1 Kn in K
Gn : the Galois group Gn := Gal(Kn /K0 )
G∞ : the Galois group G∞ := Gal(K∞ /K0 )
In this section, we show that the arboreal Galois groups Gn embed in the arithmetic
basilica groups Mn .
Lemma 4.1. Let y ∈ K, and let α1 , α2 ∈ f −2 (y) with f (α2 ) = −f (α1 ). Then
(α1 α2 )2 = −y.
Proof. Write u = f (α1 ). Then α12 = u + 1 and α22 = −u + 1. (See Figure 3.) Thus,
(α1 α2 )2 = (u + 1)(−u + 1) = −(u2 − 1) = −f (u) = −y. 
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 19

α1 α2

u −u

Figure 3. Lemma 4.1: (α1 α2 )2 = −y

α2,1 α2,2 α2,3 α2,4 β2,1 β2,2 β2,3 β2,4

α1,1 α1,2 β1,1 β1,2

α0,1 β0,1
y

Figure 4. Lemma 4.2 for m = 2.

Lemma 4.2. Let m ≥ 0, and let y ∈ K. Let α0,1 ∈ f −1 (y), and define β0,1 := −α0,1 .
For each i = 1, . . . , m, choose points {αi,j , βi,j : 1 ≤ j ≤ 2i } ⊆ f −(2i+1) (y) so that
f −2 αi−1,ℓ = {±αi,2ℓ−1 , ±αi,2ℓ } and f −2 βi−1,ℓ = {±βi,2ℓ−1 , ±βi,2ℓ }

for each ℓ = 1, . . . , 2i−1, as in Figure 4. Define

2 m 2m
Y Y
γm := αm,j and δm := βm,j .
j=1 j=1

2m 2m
Then (−γm ) = β0,1 and (−δm ) = α0,1 . Moreover, γm /δm is a primitive 2m+1 -th root
of unity.
Proof. The conclusion is trivially true for m = 0. Proceeding inductively, consider
m ≥ 1, and assume it holds for m − 1. For each ℓ = 1, . . . , 2m−1 , we have

2
2
αm,2ℓ−1 αm,2ℓ = −αm−1,ℓ and βm,2ℓ−1 βm,2ℓ = −βm−1,ℓ
by Lemma 4.1. It follows immediately that
( (
γ m−1 if m ≥ 2, δm−1 if m ≥ 2,
(−γm )2 = and (−δm )2 =
−γm−1 if m = 1, −δm−1 if m = 1.

Raising each to the power 2m−1 , which is 1 if m = 1 and even for m ≥ 2, we have
m m−1 m m−1
(−γm )2 = (−γm−1 )2 = β0,1 and (−δm )2 = (−δm−1 )2 = α0,1 ,
as desired. The final statement is immediate from the fact that β0,1 /α0,1 = −1. 
20 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

By Lemma 4.2, the field K∞ formed by adjoining all preimages f −n (x0 ) to K0 contains
all 2-power roots of unity. We now use these roots of unity to label the tree T∞ of
preimages Orb−f (x0 ) to be compatible with the action of Galois.

Lemma 4.3. Choose a sequence {ζ2 , ζ4, ζ8 , . . .} of primitive 2-power roots of unity in
K, in such a way that (ζ2m )2 = ζ2m−1 for each m ≥ 1. It is possible to label the tree
T∞ of preimages Orb−
f (x0 ) in such a way that for every node y of the tree and for every
i ≥ 0, we have
! !−1
Y Y
(25) [yas1 as2 · · · asi a] [ybs1 as2 · · · asi a] = ζ2i+1 ,
s1 ,...,si ∈{a,b} s1 ,...,si ∈{a,b}

where [w] denotes the element of K that appears in the tree as a node with label w.
Lemma 4.3 says that it is always possible to choose a labeling of the tree of preimages
Orb−
f (x0 ) so that for the nodes αi,j and βi,j highlighted in such in Figure 4, we have
α2,1 α2,2 α3,1 α3,2 α3,3 α3,4
= ζ4 , = ζ8 , and so on.
β2,1 β2,2 β3,1 β3,2 β3,3 β3,4
In fact, it says that we can label the tree so that these relationships hold for the subtree
based at each node y of the full tree. By contrast, Lemma 4.2 says that even after ap-
plying an arbitrary automorphism τ of T∞ , any such product of elements of f −(2m+1) (y)
is some primitive 2m+1 -root of unity, albeit not necessarily the particular root ζ2m+1 .
Proof of Lemma 4.3. We will label the tree of preimages inductively, starting from the
root point x0 and working our way up. To begin, label the two preimages of x0 arbitrarily
as a and b. Similarly, arbitrarily label the two preimages of a as aa and ab, and the two
preimages of b as ba and bb. Thus, we have a labeling on the copy of T2 rooted at x0 .
For each of the nodes y ∈ {x0 , a, b}, we have (ya)/(yb) = −1 = ζ2 . Thus, the desired
identity (25) holds at every node of T1 for i = 0.
For each successive n ≥ 3, suppose that we have labeled Tn−1 in such a way that for
every node y at every level 0 ≤ ℓ ≤ n − 2 of Tn−1 , and for every 0 ≤ i ≤ ⌊(n − ℓ − 2)/2⌋,
the identity of equation (25) holds. For each node x at level n − 1, label the two points
of f −1 (x) arbitrarily as xa and xb. We will now adjust these labels that we have just
applied at the n-th level of the tree.
If n = 2m + 1 is odd, consider the product on the left side of equation (25) for
y = x0 with i = m; or if n = 2m + 2 is even, consider the product on the left side
of equation (25) for each of y = x0 a and y = x0 b with i = m. As in the proof of
Lemma 4.2, it is immediate from Lemma 4.1 that the square of this product is precisely
the corresponding quantity for y with i = m − 1. (When m = 1 each half has a negative
sign, but the negatives cancel in that case.) By our successful labeling of Tn−1 , this
square is ζ2m . Thus, the original product is ±ζ2m+1 . If it is −ζ2m+1 , exchange the labels
of the two level-n nodes ybaa · · · aa and ybaa · · · ab; otherwise, make no label changes
for now. Since these two points in f −n (x0 ) are negatives of each other, we have ensured
that equation (25) holds for y with i = m.
Next, consider the product on the left side of equation (25) with i = m − 1 for each
node y at level 2 of the tree (if n is odd) or at level 3 (if n is even). By Lemma 4.1 and
our labeling of Tn−1 again, the square of this product is ζ2m−1 , and hence the product
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 21

itself is ±ζ2m . If it is −ζ2m , exchange the labels of the two level-n nodes ybaa · · · aa
and ybaa · · · ab; otherwise, make no label changes for now. Since these two points in
f −n (x0 ) are negatives of each other, we have ensured that equation (25) holds for y with
i = m − 1. In addition, because both of these nodes have labels beginning yb · · · , they
did not show up in the product of equation (25) for nodes lower on the tree than y, so
we have not disrupted our previous arrangements.
Continue in this fashion, considering nodes at successive even levels ℓ of the tree (if n
is odd) or odd levels ℓ of the tree (if n is even). For each such node y, choose whether or
not to switch the labels of ybaa · · · aa and ybaa · · · ab to ensure that equation (25) holds
for y with i = (n − ℓ − 1)/2. Once we have finished working our way up through level
ℓ = n − 1, we have labeled Tn so that for every node y at every level 0 ≤ ℓ ≤ n − 1 of
Tn , and for every 0 ≤ i ≤ ⌊(n − ℓ − 1)/2⌋, the identity of equation (25) holds. Thus, our
inductive construction is complete. 

The following result is a strengthened version of statement (1) of our Main Theorem.

Theorem 4.4. With the notation given at the start of Section 4, choose a sequence
{ζ2 , ζ4 , ζ8, . . .} of primitive 2-power roots of unity in K∞ , with (ζ2m )2 = ζ2m−1 for each
m ≥ 1. Label the tree T∞ of preimages Orb− f (x0 ) as in Lemma 4.3.
Consider the embedding of G∞ in Aut(T∞ ) induced by its action on Orb− f (x0 ) Then
the image of this embedding is contained in the arithmetic basilica group M∞ . Moreover,

σ(ζ) = ζ P (σ) for every 2-power root of unity ζ ∈ K∞ and every σ ∈ G∞ ,

where P (σ) = P (σ, x0 ) is the map of Definition 1.2.

P (σ,y)
Proof. It suffices to show that σ(ζ2m ) = ζ2m for every σ ∈ G∞ , every m ≥ 1, and
−
every y ∈ Orbf (x0 ) Throughout the proof, then, fix such σ, m, and y.
By hypothesis, for any point w ∈ Orb− f (x0 ) and any i ≥ 0, we have

− Par(σ,w)
Y   Y  
(26) σ(wa)t1 at2 · · · ati a = ζ2i+1 σ(w)at1 at2 · · · ati a ,
t1 ,...,ti t1 ,...,ti

by equation (25) of Lemma 4.3 applied to σ(w), and by equation (1) applied to Par(σ, w).
Each product in equation (26) is over t1 , . . . , ti ∈ {a, b}; and for i = 0, we understand it
− Par(σ,w)
to say [σ(wa)[= ζ2 [σ(w)a].
In addition, since the two-element sets {σ(wa), σ(wb)} and {σ(w)a, σ(w)b} always
coincide, we have
Y   Y  
(27) σ(w1 s)w2 = σ(w1 )sw2 .
s∈{a,b} s∈{a,b}

for any strings w1 and w2 of the symbols a, b. Thus, if we define

X
Pj (σ, w) := Par(σ, wt1 at2 a · · · atj ),
t1 ,...,tj ∈{a,b}
22 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

then for any node x of the tree and any m ≥ 1, we have

−P (σ,x)
Y  Y 
σ(xs1 as2 · · · asm−1 a) = ζ2 m−1 σ(xs1 as2 · · · asm−1 )a
−P (σ,x)
Y 
= ζ2 m−1 σ(xs1 as2 · · · a)sm−1 a
−P (σ,x) −Pm−2 (σ,x)
Y 
(28) = ζ2 m−1 ζ4 σ(xs1 as2 · · · sm−2 )asm−1 a
= ···
 m−1
Y −P (σ,x)  Y 
j

= ζ2m−j σ(x)s1 as2 · · · asm−1 a ,
j=1

where each undecorated product is over s1 , . . . , sm−1 ∈ {a, b}. In proving equation (28),
we have alternately applied equations (26) and (27). Specifically, we used equation (26)
with i = 0 and w = xs1 a · · · asm−1 at the first equality, then with i = 1 and w =
xs1 a · · · asm−2 at the third, and so on through i = m − 2 with w = x. We then
used (27) at the second equality with w1 = xs1 a · · · asm−2 a and w2 = a, then with
w1 = xs1 a · · · asm−3 a and w2 = asm−1 a at the fourth, and so on.
Applying equation (28) to both x = ya and x = yb, and substituting the results in
equation (25) with i = m − 1, we obtain
Q 
σ(yas1 as2 . . . asm−1 a)
σ(ζ2m ) = Q  
σ(ybs1 as2 . . . asm−1 a)
m−1 Q 
Y P (σ,yb)−P (σ,ya)
j j
σ(ya)s1 as2 . . . asm−1 a
(29) = ζ2m−j · Q ,
j=1
σ(yb)s 1 as 2 . . . as m−1 a

where each undecorated product is again over s1 , . . . , sm−1 ∈ {a, b}. Since ζ22i = ζ2i−1 for
each i, the first product in expression (29) is ζ2Mm , where
m−1
X
2j Pj (σ, yb) − Pj (σ, ya)


M :=
j=1
X X
≡2 Q(σ, ybt) − 2 Q(σ, yat) (mod 2m ).
t∈{a,b} t∈{a,b}

where Q(σ, x) is as defined in equation (2). On the other hand, by equation (25) applied
to σ(y), the quotient of two products in expression (29) is ζ2m if Par(σ, y) = 0, or ζ2−1
m if

Par(σ, y) = 1. Thus, equation (29) becomes

σ(ζ2m ) = ζ2Pm , where P = (−1)Par(σ,y) + M = P (σ, y). 

Corollary 4.5. Fix notation and a tree labeling as in Theorem 4.4. Let n ≥ 1, and let
m := ⌊(n + 1)/2⌋, so that ζ2m ∈ `Kn . Consider the embedding of Gn = Gal(Kn /K) in
Aut(Tn ) induced by its action on ni=0 f −i (x0 ). Then
(1) The image of Gn under this embedding is contained in Mn .
(2) The image of Gal(Kn /K(ζ2m )) ⊆ Gn under this embedding is contained in Bn .
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 23

Proof. We have the following commutative diagram

G∞ Aut(T∞ )
R∞,n

Gn Aut(Tn )

where the horizontal maps are the embeddings induced by the action of Gn and G∞ on
the tree of preimages, the vertical map on the left is the quotient induced by restricting
to Kn , and the vertical map on the right is the quotient R∞,n induced by restricting to
Tn . By Theorem 4.4, the image of the top map is contained in M∞ ⊆ Aut(T∞ ), and by
Theorem 3.2, we have R∞,n (M∞ ) = Mn . Since the quotient G∞ → Gn is surjective, the
image of the bottom map is contained in Mn , proving statement (1).
Given σ ∈ Gal(Kn /K(ζ2m )), we have
P (σ,x0 )
ζ2m = σ(ζ2m ) = ζ2m ,
where the first equality is by Theorem 4.4, and the second is because σ fixes K(ζ2m ).
Therefore, P (σ, x0 ) ≡ 1 (mod 2m ), and hence the image of σ in Mn lies in the subgroup
Bn′′ from Definition 2.3.(3). Thus, by Corollary 2.12, the image of σ lies in Bn , proving
statement (2). 

5. Surjectivity of the Galois group in the arithmetic basilica

Having shown that the arboreal Galois group G∞ := Gal(K∞ /K) of f (z) = z 2 − 1
embeds in the arithmetic basilica M∞ , we now wish to prove that the embedding is
surjective under certain conditions. The following variant of Lemma 4.2 will prove useful
to that end. As in Corollary 4.5, by restricting to Tn , we may consider Gn := Gal(Kn /K)
to be a subgroup of Mn . In particular, the subgroup Gal(Kn /Kn−1 ) is simply Gn ∩ Un ,
which, for n even, is the same as Gn ∩ En , by Theorem 3.2.(1).
Lemma 5.1. Let m ≥ 1, and let α0,1 ∈ K. For each i = 1, . . . , m, choose points
{αi,j : 1 ≤ j ≤ 2i } ⊆ f −(2i) (α0,1 ) so that
f −2 αi−1,ℓ = {±αi,2ℓ−1 , ±αi,2ℓ } for each ℓ = 1, . . . , 2i−2.

2m
Y
Define γm := αm,j ∈ K2m . Then
j=1
2m
(1) γm = −α0,1 .
(2) Let σ ∈ Gal(K2m /K2m−1 ) = G2m ∩ E2m . Then
σ(γm )
= sgn2m (σ, α0,1 ).
γm
Proof. (1): The statement is true for m = 1 by Lemma 4.1. Proceeding inductively,
consider m ≥ 2, and assume it holds for m − 1. For each ℓ = 1, . . . , 2m−1 , we have
(αm,2ℓ−1 αm,2ℓ )2 = −αm−1,ℓ by Lemma 4.1. Since m − 1 ≥ 1, it follows immediately that
m 2m−1
(γm )2 = γm−1 , and hence (γm )2 = γm−1 = −α0,1 .
24 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

(2): The two preimages of α0,1 are w := f (α1,1 ) and −w = f (α1,2 ). By Definition 2.4
and Lemma 2.5.(2), we have
sgn2m (σ, α0,1 ) = sgn2m−1 (σ, w) sgn2m−1 (σ, −w)
(30) = sgn2m−2 (σ, α1,1 ) sgn2m−2 (σ, α1,2 ).
For m = 1, we have γ1 = α1,1 α1,2 , and hence
σ(γ1 ) σ(α1,1 ) σ(α1,2 )
= · = sgn1 (σ, w) sgn1 (σ, −w) = sgn2 (σ, α0,1 ).
γ1 α1,1 α1,2
Proceeding inductively, assume now that m ≥ 2, and that statement (2) is true for
m − 1. In particular, assume that it holds for the tree with root point α1,i over the field
K(α1,i ), for each of i = 1, 2. Define
m−1
2Y 2m
Y
δm−1,1 := αm,j and δm−1,2 := αm,j ,
j=1 j=2m−1 +1

so that γm = δm−1,1 δm−1,2 . By our inductive hypothesis and equation (30), then,
σ(γm ) σ(δm−1,1 ) σ(δm−1,2 )
= · = sgn2m−2 (σ, α1,1 ) sgn2m−2 (σ, α1,2 ) = sgn2m (σ, α0,1 ). 
γm δm−1,1 δm−1,2
Lemma 5.2. Let n ≥ 2, and suppose that Gn ∼ = Mn ; if n = 2, suppose further that
i 6∈ Kn , where i = ζ4 denotes a primitive fourth root of unity. Let y ∈ Kn−2 with the
property that y has no square root in Kn−2 (i). Then y has no fourth root in Kn (i).
Proof. Case 1. Suppose first that n ≥ 5, so that i ∈ K3 ⊆ Kn−2 . If y has a fourth
root γ ∈ Kn , then Kn−2 (γ)/Kn−2 is a cyclic extension of degree 4, since γ 2 6∈ Kn−2
by hypothesis. Thus, H := Gal(Kn /Kn−2 ) has a quotient J isomorphic to Z/4Z. Let
σ ∈ H be an element such that the image of σ in J has order 4.
Observe that H acts as M2 = Aut(T2 ) on each of the 2n−2 copies of T2 rooted at
points of f −(n−2) (x0 ). Since the image of σ in J has order 4, the resulting composition
H ։ Aut(T2 ) → J
must be surjective for at least one such copy of T2 . Thus, we have a surjective homomor-
phism Aut(T2 ) ։ J. However, Aut(T2 ) is isomorphic to the 8-element dihedral group
D4 , which has no quotients isomorphic to Z/4Z. This contradiction completes the proof
for n ≥ 5.
Case 2. Suppose n = 2. Let H := Gal(K2 (i)/K0 (i)), which is isomorphic to a
subgroup of M2 = Aut(T2 ). Since [K2 (i) : K2 ] = [K0 (i) : K0 ] = 2 and Gal(K2 /K0 ) ∼ =
M2 , we must have H ∼ M
= 2 . If y has a fourth root γ ∈ K 2 (i), then as in Case 1, M ∼
2 = D 4
would have a quotient isomorphic to Z/4Z, a contradiction.
Case 3. Suppose n = 3. Let H := Gal(K3 /K1 (i)), which must be
H = ker(R3,1 : M3 → M1 ) ∩ B3 ,
since its elements fix i and the points of f −1 (x0 ), with no other restrictions. Thus, H
acts as M2 on each of the two copies of T2 rooted at the points of f −1 (x0 ), although any
τ ∈ H must act as an even permutation on the eight points of f −3 (x0 ).
If y has a fourth root γ ∈ K3 , then as in Case 1, at least one of the two copies of
M2 ∼ = D4 would have a quotient isomorphic to Z/4Z, a contradiction.
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 25

Case 4. Suppose n = 4. Let H := Gal(K4 /K2 (i)), which must be

H = ker(R4,2 : M4 → M2 ) ∩ B4 ,
since its elements fix i and the points of f −2 (x0 ), with no other restrictions. Thus, H
again acts as M2 on each of the four copies of T2 rooted at the points of f −2 (x0 ). The
same argument as in Case 1 therefore applies: if y has a fourth root in K4 , then at
least one of the four copies of M2 ∼ = D4 would have a quotient isomorphic to Z/4Z, a
contradiction. 
√ √
Lemma 5.3. Suppose that [K( −x0 , 1 + x0 , ζ8 ) : K] = 16. Then
√
(1) [K(i, 1 + x0 ) : K] = 4, so that K(i) does not contain a square root of 1 + x0 .
(2) K1 (i) does not contain a square root of −x0 .
(3) K2 does not contain a primitive fourth root of 1.
(4) Assuming G3 ∼ = M3 , then K3 does not contain a primitive eighth root of 1.
∼
(5) Assuming G4 = M4 , then K4 does not contain a primitive eighth root of 1.
√ √
Proof. (1): If [K(i, 1 + x0 ) : K] < 4, then 1 + x0 ∈ K(i), and hence
√ √ √
[K( −x0 , 1 + x0 , ζ8 ) : K] = [K( −x0 , ζ8 ) : K] ≤ 8,
√ √
a contradiction. Therefore, [K(i,
√ 1 + x0 ) : K] = 4, and 1 √+ x0 6∈ K(i).
−1
(2):
√ Since √ f (x0 ) = {± 1 + x0 }, we have √ K1√= K( 1 + x0 ). We must have
[K( −x0 , 1 + x0 , i) : K] ≥ 8, or else√[K( √ −x0 , 1 + x0 , ζ8 ) : K] < 16. If K1 (i)
contains a square root of −x0 , then [K( −x0 , 1 + x0 , i) : K] ≤ 4, again a contradic-
tion. Thus, K1 (i) does not contain a square root of −x0 . √
√ By√ Lemma 5.1 for α0,1 = x0 and m = 1, we√have √−x0 ∈ K2 , whence
(3):
K( −x0 , 1 + x0 ) ⊆ K2 . As in √ part√(2), we have [K( −x0 , 1 + x0 , i) : K] ≥ 8.
If i ∈ K2 , then K2 contains K( −x0 , 1 + x0 , i), which is an abelian extension of K
of degree 8. However, Gal(K2 /K) is a subgroup of Aut(T2 ), which is a nonabelian
extension of K of degree 8. By this contradiction, K2 cannot contain i.
(4): Label the tree T3 as in Figure 1. Let C be the commutator subgroup of M3 =
Aut(T3 ). We claim that |C| ≥ 24 , and hence the abelianization M3ab := M3 /C has order
|M3 |/|C| ≤ 23 . (In fact, |C| = 24 , but we only need the inequality here.)
Consider the elements α, β ∈ B3 ⊆ M3 defined early in Section 2. As we saw in the
proof of Lemma 2.10, the commutator λ1 := αβα−1β −1 ∈ C has
Par(λ1 , x0 ) = 0, and Par(λ1 , a) = Par(λ1 , b) = 1.
2 −1 −2
Similarly, the commutator λ2 := βα β α ∈ C acts on T3 by
(
1 if y = ba or bb,
Par(λ2 , y) = for all nodes y of T2 .
0 otherwise
Define σ, τ ∈ M3 by
( (
1 if y = bb, 1 if y = x0 ,
Par(σ, y) = and Par(τ, y) =
0 otherwise, 0 otherwise,
for all nodes y of T2 . Then the commutator λ3 := στ σ −1 τ −1 ∈ C acts on T3 by
(
1 if y = ab or bb,
Par(λ3 , y) = for all nodes y of T2 .
0 otherwise
26 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

− − −
− −
−
σ̃ τ̃

Figure 5. The maps σ̃ and τ̃ in Lemma 5.3.(5)

The commutators λ2 , λ′2 := α−1 λ2 α, and λ3 together generate the 8-element group E3 ,
which does not contain λ1 ∈ C. Thus, |C| ≥ 16, and hence |M3ab | ≤ 8, as claimed.
As noted in the proofs of parts (2) and (3) above, we have
√ √
K( −x0 , 1 + x0 ) ⊆ K2 ⊆ K3 .
√ √
If K√ 3 contains
√ a primitive eighth root of 1, then K( −x 0 , 1 + x0 , ζ8 ) ⊆ K3 . However,
K( −x0 , 1 + x0 , ζ8 )/K is an abelian extension, and by hypothesis, it has degree 16.
Therefore, the abelianization of G3 = Gal(K3 /K) ∼ = M3 must have order at least 16,
contradicting our claim. Thus, K3 cannot contain a primitive eighth root of 1.
(5): Let C be the commutator subgroup of M4 . We claim that |C| ≥ 210 , and hence
the abelianization M4ab := M4 /C has order |M4 |/|C| ≤ 23 . (In fact, |C| = 210 , but as in
the proof of (4), we only need the inequality here.)
Since R4,3 : M4 → M3 is surjective, all of the automorphisms of T3 in the proof of
part (4) can be lifted to M4 , and hence the restriction R4,3 (C) of C to T3 has order at
least 16. Therefore, to prove the claim, it suffices to show that E4 ⊆ C, since E4 is a
26 -element subgroup of ker(R4,3 ).
As in the proof of Lemma 2.10, the commutator µbb := α2 β 2 α−2 β −2 ∈ C has
(
1 if y = bba or bbb,
Par(µbb , y) = for all nodes y of T3 .
0 otherwise
Since M4 acts transitively on the second level of T3 , conjugating µbb yields three more
automorphisms µaa , µab , µba ∈ C, where
(
1 if y = sta or stb,
Par(µst , y) = for all nodes y of T3 .
0 otherwise
Define σ̃, τ̃ ∈ M4 by
( (
1 if y ∈ {aa, abb, bbb} 1 if y ∈ {b, aa, abb},
Par(σ̃, y) = and Par(τ, y) =
0 otherwise, 0 otherwise,
for all nodes y of T3 ; see Figure 5, where the nodes for which Par = 1 are marked. Define
λb := σ̃τ̃ σ̃ −1 τ̃ −1 ∈ C and λa := α−1 λb α ∈ C.
Both σ̃ and τ̃ fix the nodes a and b, and they coincide on the subtree rooted at a, so that
λb acts trivially on this subtree. On the other hand, on the subtree rooted at b, they act
like the automorphisms σ and τ from the proof of part (4). Thus, the commutators λa
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 27

and λb act on T4 by
( (
1 if y = aab or abb, 1 if y = bab or bbb,
Par(λa , y) = and Par(λb , y) =
0 otherwise 0 otherwise
for all nodes y of T3 . Therefore, the six automorphisms
λa , λb , µaa , µab , µba , µbb ∈ C ∩ E4
together generate all 26 elements of E4 . Thus, |C| ≥ 210 , and hence |M4ab | ≤ 8, as
claimed.
As
√ in the
√ proof of part (4) above, if K4 contains a primitive eighth root of 1, then
K( −x0 , 1 + x0 , ζ8 )/K is an abelian subextension of degree 16, by hypothesis. How-
ever, the abelianization of G4 = Gal(K4 /K) ∼= M4 has order at most 8, by the claim.
Therefore, by this contradiction, K4 cannot contain a primitive eighth root of 1. 
Theorem 5.4. Fix notation as at the start of Section √ 4, √ and fix roots of unity ζ2m and
a tree labeling as in Lemma 4.3. Suppose that [K( −x0 , 1 + x0 , ζ8 ) : K] = 16. Then
(1) For every n = 2m + 1 ≥ 1 odd,
(a) Kn contains all the 2m+1 -roots of unity but no primitive 2m+2 -roots of unity.
(b) Kn contains a 2m -root of −x0 but no 2m+1 -root of −x0 .
(c) Kn contains a 2m+1
-root of 1 + x0 but no 2
m+2
-root of 1 + x0 .
∼
(d) Gal Kn /K(ζ2m+1 ) = Bn .
(e) Gal(Kn /K) ∼= Mn
(2) For every n = 2m ≥ 2 even,
(a) Kn contains all the 2m -roots of unity but no primitive 2m+1 -roots of unity.
(b) Kn contains a 2m -root of −x0 but no 2m+1 -root of −x0 .
(c) Kn contains a 2m
-root of 1 + x0 but no 2m+1 -root of 1 + x0 .
(d) Gal Kn /K(ζ2m ) ∼ = Bn .
∼
(e) Gal(Kn /K) = Mn
Proof. We proceed by induction on n ≥ 1. For n = 1, 2, we strengthen the second
statement of (1c) and (2c) to say that K1 (i) and K2 (i) do not contain a fourth root of
1 + x0 . This strengthening will be relevant near the end of Cases 3 and 2, respectively.
Case 1: For n = 1, i.e., n = 2m + 1 with m = 0, clearly K1 contains ζ2 = −1,
which is √a primitive 2-root of 1, and √ −x0 , which is a 20 -root of −x0 . In addition,
K1 = K( 1 + x0 ) contains a 21 -root 1 + x0 of 1 + x0 .
On the other hand, by Lemma 5.3.(3), K2 does not contain a primitive fourth root of
unity, and hence neither does K1 . By Lemma 5.3.(2), K1 (i) does not contain a square
root of −x0 , and hence neither
√ does K1 .
By Lemma
√ 5.3.(1), [K(i, 1 +√ x0 ) : K] = 4, which implies both that [K1 (i) : K(i)] = 2
and that 1 + x0 6∈ K(i). If 4 1 + x0 ∈ K1 (i), then K1 (i)/K(i) would be a cyclic
extension of degree 4, a contradiction. Thus, K1 (i) does not contain a fourth root of
1 + x0 .
We have proven statements (a)–(c), including the strengthened version of (c). Finally,
since [K1 : K] = 2 and ζ2 = −1, we have
Gal(K1 /K) = Gal(K1 /K(ζ2 )) ∼ = Z/2Z ∼ = B2 ∼= M2 ,
proving statements (d) and (e) as well.
28 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

Case 2: Suppose n = 2m ≥ 2 is even, and suppose the theorem holds for all smaller
n. Let ℓ := 2m . By the inductive hypothesis, Kn−1 contains a primitive ℓ-root of unity
ζℓ and an ℓ-root of 1 + x0 , and therefore Kn does as well, proving the first half of
statements (a) and (c).
By Corollary 4.5.(1), Gn = Gal(Kn /K) is isomorphic to a subgroup of Mn , not just as
abstract groups, but also respecting the action on the tree Tn , identified with the tree of
preimages of x0 under f −n . We will therefore abuse notation in the rest of this proof and
view Gn as a subgroup of Mn . Similarly, by Corollary 4.5.(2), G′n := Gal(Kn /K(ζℓ ))
is a subgroup of Bn . Since Kn−1 also contains ζℓ , it follows that Gal(Kn /Kn−1 ) is a
subgroup of En .
By Lemma 5.1.(1) applied to α0,1 := x0 , Kn also contains an ℓ-root γ of −x0 , proving
the first half of statement (b). By the inductive hypothesis again, we have γ 6∈ Kn−1 . On
the other hand, γ 2 is a 2m−1 -root of −x0 and hence lies in Kn−1 . (Indeed, Kn−1 contains
at least one such root, and hence it contains all such roots, as it is a Galois extension of
K.) Therefore, Kn contains the quadratic extension Kn−1 (γ) of Kn−1 , and hence there
exists λ ∈ Gal(Kn /Kn−1 ) = G′n ∩ En such that λ(γ) = −γ. By Lemma 5.1.(2), we have
sgnn (λ, x0 ) = −1. In addition, by our inductive assumption of (d) for n − 1, we have

Rn,n−1 (G′n ∩ Bn ) = Gal Kn−1 /K(ζℓ ) = Bn−1 .

Therefore, by Corollary 2.13, we have G′n ⊇ Bn , and hence G′n = Gal(Kn /K(ζℓ )) = Bn ,
proving statement (d).
Furthermore, it follows that Gal(Kn /Kn−1 ) = En , and hence, by Theorem 3.2.(1), that
Gal(Kn /Kn−1 ) = Un . By our inductive hypothesis, we also have Gal(Kn−1 /K) = Mn−1 .
Thus, we must have Gal(Kn /K) = Mn , proving statement (e).
It remains to show the second half of each of statements (a–c). For (a), Lemma 5.3.(3)
suffices for n = 2, and Lemma 5.3.(5) suffices for n = 4. For n ≥ 6, i.e. m ≥ 3, let
y = ζ2m−1 be a primitive 2m−1 -root of unity. By our inductive hypothesis, we have
y ∈ Kn−2 but y has no square root in Kn−2 = Kn−2 (i). Therefore, by Lemma 5.2, y
has no fourth root in Kn (i) = Kn ; that is, Kn does not contain a primitive 2m+1 -root of
unity.
For (b) and (c), we claim that neither 1 + x0 nor −x0 has a 2m -root in Kn−2 (i). For
n ≥ 6, this is true by our inductive hypothesis and the fact that Kn−2 = Kn−2 (i). For
n = 2, it is true for 1 + x0 by Lemma 5.3.(1), and for −x0 by Lemma 5.3.(2). For n = 4,
our inductive hypothesis says that −x0 has no fourth root in K3 and hence in K2 (i).
Finally, the strengthened version of statement (2c) in our inductive hypothesis says that
1 + x0 has no fourth root in K2 (i), proving our claim.
Thus, letting y ∈ Kn−2 be a 2m−1 -root of −x0 (for (b)), or a 2m−1 -root of 1 + x0
(for (c)), the above claim shows that y has no square root in Kn−2 (i). Therefore, by
Lemma 5.2, y has no fourth root in Kn (i), yielding the desired conclusion that Kn
contains no 2m+1 -roots of −x0 or 1 + x0 , and, for n = 2, that K2 (i) contains no fourth
root of 1 + x0 .
Case 3: Suppose n = 2m + 1 ≥ 3 is odd, and suppose the theorem holds for all
smaller n. Let ℓ := 2m . By the inductive hypothesis, Kn−1 contains an ℓ-root of −x0 ,
and therefore Kn does as well, proving the first half
√ of statement (b). By Lemma 4.2
with y = x0 , which has preimages f −1 (x0 ) = {± 1 + x0 }, we see that Kn contains a
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 29

2m+1 -root of 1 + x0 and a primitive 2m+1 -root of unity ζ2ℓ . Thus, we have proven the
first half of statements (a) and (c) as well.
As in Case 2, by Corollary 4.5, Gn = Gal(Kn /K) is isomorphic to a subgroup of
Mn , and G′n := Gal(Kn /K(ζ2ℓ )) is isomorphic to a subgroup of Bn . We again abuse
notation and view Gn and G′n as subgroups of Mn and Bn , respectively. It follows that
Gal(Kn /Kn−1 (ζ2ℓ )) ⊆ G′n is a subgroup of En .
Consider G′′ := Gal(Kn−1 (ζ2ℓ )/K(ζ2ℓ )). We claim that G′′ ∼ = Bn−1 . To prove this
′′
claim, observe first that G is isomorphic to a subgroup of Bn−1 , by Corollary 4.5.(2).
2
Observe further that [Kn−1 (ζ2ℓ ) : Kn−1 ] = 2, since ζℓ := ζ2ℓ ∈ Kn−1 but ζ2ℓ 6∈ Kn−1 , by
our inductive hypothesis. Therefore,
|Bn−1 | · 2 ≥ [Kn−1 (ζ2ℓ ) : K(ζ2ℓ )][K(ζ2ℓ ) : K(ζℓ )] = [Kn−1 (ζ2ℓ ) : K(ζℓ )]
= [Kn−1 (ζ2ℓ ) : Kn−1 ][Kn−1 : K(ζℓ )] = 2 · |Bn−1 |,
and hence |G′′ | = [Kn−1 (ζ2ℓ ) : K(ζ2ℓ )] = |Bn−1 |. Therefore, G′′ must be isomorphic to
the full group Bn−1 , proving the claim.
Recall from above that Kn contains a 2m+1 -root γ of 1 + x0 . By the inductive hy-
pothesis, we have γ 6∈ Kn−1 . We make a second claim, that γ 6∈ Kn−1 (ζ2ℓ ). To see
this, it suffices to show that the two quadratic extensions L := Kn−1 (γ) and Kn−1 (ζ2ℓ )
of Kn−1 do not coincide. If they did, then Gal(L/Kn−1 ) would be a two-element group
{e, τ } with τ (γ) = −γ and τ (ζ2ℓ ) = −ζ2ℓ . In that case, the product ζ2ℓ γ ∈ L would be
fixed by both e and τ , and hence ζ2ℓ γ ∈ Kn−1 . However, ζ2ℓ γ is a 2m+1 -root of 1 + x0 ,
contradicting the inductive hypothesis and proving our second claim.
Thus, Kn contains the quadratic extension Kn−1 (ζ2ℓ , γ) of Kn−1 (ζ2ℓ ), and hence there
exists λ ∈ Gal(Kn√ /Kn−1 (ζ2ℓ )) = G′n ∩ En such that λ(γ)√ = −γ. By Lemma 5.1.(2)
applied to α0,1 := 1 + x0 ∈ f −1 (x0 ), we have sgnn−1 (λ, 1 + x0 ) = −1, and therefore
sgnn (λ, x0 ) = −1 by Lemma 2.5.(2). In addition,
Rn,n−1 (G′n ∩ Bn ) = Gal Kn−1 (ζ2ℓ )/K(ζ2ℓ ) = G′′ = Bn−1 ,

where the last equality is by our first claim. Therefore, by Corollary 2.13, we have
G′n ⊇ Bn , and hence G′n = Gal(Kn /K(ζ2ℓ )) = Bn , proving statement (d).
It follows that Gal(Kn /Kn−1 (ζ2ℓ )) = En . On the other hand, Gal(Kn /Kn−1 ) ⊆ Un ,
since Gn is a subgroup of Mn . We must therefore have Gal(Kn /Kn−1 ) = Un , because
[Kn−1 (ζ2ℓ ) : Kn−1 ] = 2 = [Un : En ],
where the first equality is by our inductive hypothesis, and the second is by Theo-
rem 3.2.(1). Also by our inductive hypothesis, we have Gal(Kn−1 /K) = Mn−1 . It
follows that Gal(Kn /K) = Mn , proving statement (e).
It remains to show the second half of each of statements (a–c). For (a), Lemma 5.3.(4)
suffices for n = 3. For n ≥ 5, i.e. m ≥ 2, let y = ζ2m be a primitive 2m -root of unity. By
our inductive hypothesis, we have y ∈ Kn−2 but y has no square root in Kn−2 = Kn−2 (i).
Therefore, by Lemma 5.2, y has no fourth root in Kn ; that is, Kn does not contain a
primitive 2m+2 -root of unity.
For (b), let y be a 2m−1 -root of −x0 , or for (c), let y be a 2m -root of 1 + x0 . If
n = 3, then y has no square root in Kn−2 (i), by Lemma 5.3.(2) for −x0 , and by our
strengthened version of (1c) for 1 + x0 . If n ≥ 5, then by our inductive hypothesis, y
has no square root in Kn−2 = Kn−2 (i). Therefore, by Lemma 5.2, y has no fourth root
in Kn . That is, Kn contains no 2m+1 -roots of −x0 or 1 + x0 . 
30 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

We close with the following strengthening of statement (2) of our Main Theorem.
Corollary 5.5. Fix notation as at the start of Section 4, and fix roots of unity ζ2m and
a tree labeling as in Lemma 4.3. The following are equivalent.
√ √
(1) [K( −x0 , 1 + x0 , ζ8) : K] = 16.
(2) [K5 : K] = 225 .
(3) G5 ∼= M5 .
(4) Gn ∼= Mn for all n ≥ 1.
(5) G∞ ∼ = M∞ .
(As always, the isomorphisms of statements (3)–(5) are of groups acting on trees, not
just of abstract groups.)
Proof. We have (1)⇒(4) by Theorem 5.4. Taking inverse limits, we have (4)⇒(5), since
G∞ ∼ = lim G ∼ lim M ∼ M .
←− n = ←− n = ∞
The implications (5)⇒(4)⇒(3) are trivial. Since |G5 | = [K5 : K] and |M5 | = 225 (by
Theorem 3.3), we have (3)⇒(2). In addition, by Corollary 4.5.(1), G5 is isomorphic to
a subgroup of M5 , and hence (2)⇒(3). It suffices to show (3)⇒(1).
Assume that G5 ∼ = M5 , and therefore that Gn ∼ = Mn √
for n = 1, 2, 3, 4, 5.
If K contains a square root of 1 + x0 , then K√ 1 = K( 1 + x0 ) = K, whence |G1 | = 1,
contradicting
√ the fact that |M1 | = 2. Thus, [K( 1 + x0 : K] = 2.
If K( 1 + x0 ) contains a square root of −x0 , then any σ ∈ Gal(K2 /K1 ) fixes −x0 .
By Lemma 5.1, we have sgn2 (σ, x0 ) = +1 for any such σ. However, the isomorphism
G2 ∼= M2 restricts to Gal(K2 /K1 ) ∼ = U2 , and β ∈ U2 satisfies sgn2 (β, x0 ) = −1, a
contradiction. Thus,
√ √ √
[K( 1 + x0 , −x0 ) : K( 1 + x0 )] = 2,
√ √
and hence
√ [K( 1√+ x0 , −x0 ) : K] = 4.
If K( 1 + x0 , −x0 ) ⊆ K2 contains a primitive fourth root of unity i, then every
σ ∈ Gal(K3 /K2 ) fixes i. Via the isomorphism G3 ∼ = M3 , this would mean that every
σ ∈ U3 also lies in E3 . However, [U3 : E3 ] = 2 by Theorem 3.2, a contradiction. Thus,
√ √ √ √
[K( 1 + x0 , −x0 , i) : K( 1 + x0 , −x0 )] = 2,
√ √
and hence [K( √1 + x0 , √−x0 , i) : K] = 8.
Finally, if K( 1 + x0 , −x0 , i) ⊆ K3 contains a primitive eighth root of unity ζ8 , then
every σ ∈ Gal(K5 /K4 ) fixes ζ8 . Via the isomorphism G5 ∼ = M5 , this would mean that
every σ ∈ U5 also lies in E5 . However, [U5 : E5 ] = 2 by Theorem 3.2, a contradiction.
Thus, √ √ √ √
[K( 1 + x0 , −x0 , ζ8 ) : K( 1 + x0 , −x0 , i)] = 2,
√ √
and hence [K( 1 + x0 , −x0 , ζ8) : K] = 16, as desired. 
Acknowledgements. The results of this paper grew out of an REU project at
Amherst College. Our theorems were originally suggested by computations using Magma.
Authors RB, JC, and GC gratefully acknowledge the support of NSF grant DMS-
1501766. Authors FA and LF gratefully acknowledge the support of Amherst College’s
Gregory S. Call student research funding. We thank Harris Daniels and Jamie Juul for
helpful discussions. A particular thanks goes to Rafe Jones, for proposing this problem,
for providing useful background, and for suggestions surrounding Conjectures 1 and 2.
 THE ARITHMETIC BASILICA: A QUADRATIC PCF ARBOREAL GALOIS GROUP 31

References
[1] Wayne Aitken, Farshid Hajir, and Christian Maire, Finitely ramified iterated extensions, Int. Math.
Res. Not. 2005, 855–880.
[2] Jacqueline Anderson, Irene I. Bouw, Ozlem Ejder, Neslihan Girgin, Valentijn Karemaker, and
Michelle Manes, Dynamical Belyi maps, in Women in numbers Europe II, Springer, Cham (2018),
57–82.
[3] Laurent Bartholdi, Rostislav Grigorchuk, and Volodymyr Nekrashevych, From fractal groups to
fractal sets, in Fractals in Graz 2001, 25–118, Birkhäuser, Basel, 2003.
[4] Robert L. Benedetto, Xander Faber, Benjamin Hutz, Jamie Juul, and Yu Yasufuku, A large ar-
boreal Galois representation for a cubic postcritically finite polynomial, Res. Number Theory 3
(2017), Art. 29, 21.
[5] Robert L. Benedetto and Jamie Juul, Odoni’s conjecture for number fields, Bull. Lond. Math. Soc.
51 (2019), 237–350.
[6] Nigel Boston and Rafe Jones, Arboreal Galois representations, Geom. Dedicata 124 (2007), 27–35.
[7] Andrew Bridy and Thomas J. Tucker, Finite index theorems for iterated Galois groups of cubic
polynomials, Math. Ann. 373 (2019), 37–72.
[8] Michael R. Bush, Wade Hindes, and Nicole R. Looper, Galois groups of iterates of some unicritical
polynomials, Acta Arith. 181 (2017), 57–73.
[9] Andrea Ferraguti and Giacomo Micheli, An equivariant isomorphism theorem for mod p reductions
of arboreal Galois representations, preprint, 2019. Available at arXiv:1905.00506.
[10] Andrea Ferraguti, Carlo Pagano, and Daniele Casazza, The inverse problem for arboreal Galois
representations of index two, preprint, 2019. Available at arXiv:1907.08608.
[11] Richard Gottesman and Kwokfung Tang, Quadratic recurrences with a positive density of prime
divisors, Int. J. Number Theory, 6 (2010), 1027–1045.
[12] Chad Gratton, Khoa Nguyen, and Thomas J. Tucker, ABC implies primitive prime divisors in
arithmetic dynamics, Bull. Lond. Math. Soc. 45 (2013), 1194–1208.
[13] Wade Hindes, Average Zsigmondy sets, dynamical Galois groups, and the Kodaira-Spencer map,
Trans. Amer. Math. Soc. 370 (2018), 6391–6410.
[14] Wade Hindes, Classifying Galois groups of small iterates via rational points, Int. J. Number Theory
14 (2018), 1403–1426.
[15] Patrick Ingram, Arboreal Galois representations and uniformization of polynomial dynamics, Bull.
Lond. Math. Soc. 45 (2013), 301–308.
[16] Rafe Jones, Galois representations from pre-image trees: an arboreal survey, in Actes de la
Conférence “Théorie des Nombres et Applications”, Pub. Math. Besançon (2013), 107-136.
[17] Rafe Jones and Michelle Manes, Galois theory of quadratic rational functions, Comment. Math.
Helv. 89 (2014), 173–213.
[18] Jamie Juul, Iterates of generic polynomials and generic rational functions, Trans. Amer. Math.
Soc. 371 (2019), 809–831.
[19] Jamie Juul, Holly Krieger, Nicole Looper, Michelle Manes, Bianca Thompson, and Laura Walton,
Arboreal representations for rational maps with few critical points, preprint, 2018. Available at
arXiv:1804.06053.
[20] Jamie Juul, Pär Kurlberg, Kalyani Madhu, and Tom J. Tucker, Wreath products and proportions
of periodic points, Int. Math. Res. Not. IMRN 2016, 3944–3969.
[21] Borys Kadets, Large arboreal Galois representations, preprint, 2018. Available at
arXiv:1802.09074.
[22] Nicole Looper, Dynamical Galois groups of trinomials and Odoni’s conjecture, Bull. Lond. Math.
Soc. 51 (2019), 278–292.
[23] Volodymyr Nekrashevych, Self-Similar Groups, American Mathematical Society, Providence, 2005.
[24] R. W. K. Odoni, The Galois theory of iterates and composites of polynomials, Proc. London Math.
Soc. (3) 51 (1985), no. 3, 385–414.
[25] Richard Pink, Profinite iterated monodromy groups arising from quadratic polynomials, preprint
2013. Available at arXiv:1307.5678.
32 AHMAD, BENEDETTO, CAIN, CARROLL, AND FANG

[26] Jean-Pierre Serre, Propriétés galoisiennes des points dordre fini des courbes elliptiques, Invent.
Math. 15 (1972), 259–331.
[27] Joel Specter, Polynomials with Surjective Arboreal Galois Representations Exist in Every Degree,
preprint, 2018. Available at arXiv:1803.00434.
[28] Michael Stoll, Galois groups over Q of some iterated polynomials, Arch. Math. (Basel) 59 (1992),
239–244.
[29] Ashvin A. Swaminathan, On arboreal Galois representations of rational functions, J. Algebra 448
(2016), 104–126.

Amherst College, Amherst, MA 01002, USA

E-mail address: faseehirfan@gmail.com

E-mail address: rlbenedetto@amherst.edu

E-mail address: jentcain@gmail.com

E-mail address: gcarroll19@amherst.edu

E-mail address: lilycfang@yahoo.com