ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III

KEVIN MCGOWN, DANIEL VALLIÈRES

Abstract. Let ℓ be a rational prime. Previously, abelian ℓ-towers of multigraphs were introduced
arXiv:2107.07639v1 [math.CO] 15 Jul 2021

which are analogous to Zℓ -extensions of number fields. It was shown that for towers of bouquets, the
growth of the ℓ-part of the number of spanning trees behaves in a predictable manner (analogous to
a well-known theorem of Iwasawa for Zℓ -extensions of number fields). In this paper, we extend this
result to abelian ℓ-towers over an arbitrary connected multigraph (not necessarily simple and not
necessarily regular). In order to carry this out, we employ integer-valued polynomials to construct
power series with coefficients in Zℓ arising from cyclotomic number fields, different than the power
series appearing in the prequel. This allows us to study the special value at u = 1 of the Artin–Ihara
L-function, when the base multigraph is not necessarily a bouquet.

Contents
1. Introduction 1
2. Integer-valued polynomials 2
3. Some power series arising from cyclotomic number fields 2
4. Voltage assignments and their derived multigraphs 5
5. The special value at u = 1 of Artin-Ihara L-functions 7
6. Abelian ℓ-towers of multigraphs 10
7. Examples 11
8. Iwasawa modules arising from graph theory 14
References 14

1. Introduction
In [17], abelian ℓ-towers of multigraphs, where ℓ is a rational prime, were introduced which can be
viewed as being analogous to Zℓ -extensions of number fields. To every tuple in Ztℓ (with t ∈ N and
not all entries divisible by ℓ) one can associate an abelian ℓ-tower of a bouquet with t loops. It was
proved in [10] that the ℓ-adic valuation of the number of spanning trees behaves similarly to the ℓ-adic
valuation of the class numbers in Zℓ -extensions of number fields. More precisely, if κn denotes the
number of spanning trees of the multigraph at the n-th level, then there exist non-negative integers
µ, λ, n0 and an integer ν such that
ordℓ (κn ) = µℓn + λn + ν,
for n ≥ n0 . Meanwhile, in [5], the same behavior was shown to be true with a different method in the
situation where the base multigraph is an arbitrary simple graph, meaning that it does not contain
loops and parallel edges. In this paper, we extend this result to abelian ℓ-towers over arbitrary base
multigraphs (not necessarily simple and not necessarily regular) by adapting the strategy of [10] and
[17] to the current more general situation.
The paper is organized as follows. In §2, we remind the reader about integer-valued polynomials.
We then use integer-valued polynomials in §3 to construct some power series in Zℓ JT K arising from
cyclotomic number fields which will be used to calculate the invariants µ and λ. In §4, we remind

Date: July 19, 2021.

2010 Mathematics Subject Classification. Primary: 05C25; Secondary: 11R18, 11Z05, 13F20.
1
2 KEVIN MCGOWN, DANIEL VALLIÈRES

the reader about voltage assignments which give a convenient way of constructing abelian ℓ-towers
of multigraphs. One can view the derived multigraphs obtained from voltage assignments as gener-
alizations of the Cayley-Serre multigraphs used in [10] and [17]. In §5, we study the special value at
u = 1 of the Artin-Ihara L-functions arising in abelian ℓ-towers of multigraphs. In §6, we formulate
our main result (see Theorem 6.1), and we present a few examples in §7. At last, in §8 we point out
the connection between our work and the work of Gonet included in [5].

2. Integer-valued polynomials
Our main reference for this section is [1]. Recall that the set of integer-valued polynomials is
Int(Z) = {P (X) ∈ Q[X] | P (Z) ⊆ Z}.
It is a unital commutative ring. As usual, for n ≥ 2, we let
 
X X(X − 1) . . . (X − n + 1)
= ∈ Q[X],
n n!




and we set X0 = 1 and X1 = X. It is known that X n ∈ Int(Z) and that every integer-valued
polynomial P (X) of degree n can be uniquely written as
Xn  
X
P (X) = ck ,
k
k=0

for some ck ∈ Z that can be written down explicitly. We refer to §1 of [1] for details.
Proposition 2.1. Let ℓ be a rational prime and let P (X) ∈ Int(Z). Then the induced function
P : Z −→ Z is Lipschitz when Z is endowed with the ℓ-adic topology. More precisely, we have
|P (m) − P (n)|ℓ ≤ C(ℓ, P ) · |m − n|ℓ ,
for all m, n ∈ Z, where
C(ℓ, P ) = ℓ⌊logℓ (deg(P ))⌋ .
Proof. See Proposition 9 in [1]. 

3. Some power series arising from cyclotomic number fields

In [10], for each a ∈ Zℓ , a power series Pa (T ) ∈ Zℓ JT K obtained as an ℓ-adic limit of some shifted
Chebyshev polynomials was constructed in order to prove the main result. (Theorem 4.1 in [10].)
Using Proposition 2.1, the main idea of §2 and §3 of [10] can be adapted to construct other power
series which we find more convenient to work with in the present paper. (In fact, Proposition 2.1 could
be used to prove Corollary 3.1 of [10] instead of using Lemma 2.2 and Proposition 2.3 of [10].)
Given m ∈ N = {1, 2, . . .}, we let ζm = exp(2πi/m) ∈ C. If a ∈ Z, we set
a
πm (a) = 1 − ζm ∈ Q ⊆ C,
and we write πm rather than πm (1). Note that πm (a) is an element of the cyclotomic number field
Q(ζm ). Let now ℓ be a rational prime, i ∈ N, and consider the cyclotomic number field Q(ζℓi ). Above
ℓ, there is a unique prime ideal which we denote by Li . Furthermore, it is known that Li is a principal
ideal and
(1) Li = (πℓi ).
For a ∈ N, consider
Qa (T ) = 1 − (1 − T )a ∈ Z[T ],
and set Q0 (T ) = 0. Then,
(2) Qa (πm ) = πm (a),
 ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III 3

for all m ∈ N and for all a ∈ Z≥0 . Note that the polynomials Qa (T ) have no constant coefficient and
have degree a. For k, n ∈ N, we let
(

(−1)k−1 nk , if k ≤ n;
ek (n) =
0, otherwise.
Using the binomial expansion of (1 − T )a , we have
Qa (T ) = e1 (a)T + e2 (a)T 2 + . . . + ea (a)T a .
We note in passing that
e1 (n) = n and en (n) = (−1)n−1 .
We now consider the complete local ring Zℓ JT K and the function g : N −→ Zℓ JT K defined via
n 7→ g(n) = Qn (T ).
Proposition 3.1. The function g is uniformly continuous when N is endowed with the ℓ-adic topology.
Proof. Proposition 2.1 implies that if P (X) ∈ Int(Z), then
ordℓ (P (m) − P (n)) ≥ ordℓ (m − n) − ⌊logℓ (deg(P ))⌋,
for all m, n ∈ N. Noting that ⌊logℓ (k)⌋ ≤ k for k ≥ 1, we have in fact
ordℓ (P (m) − P (n)) ≥ ordℓ (m − n) − deg(P ).
Consider now  
X
k−1
ek (X) = (−1) ∈ Int(Z).
k
Note that deg(ek (X)) = k; thus, if m, n ∈ N are such that
m≡n (mod ℓs ),
for some s ∈ N, then
ek (m) ≡ ek (n) (mod ℓs−k )
for k = 1, . . . , s − 1. So given any N ∈ N and m, n ∈ N satisfying
ordℓ (m − n) ≥ N,
we have
Qm (T ) − Qn (T ) ∈ mN ,
where m is the maximal ideal of Zℓ JT K, and this shows the claim. 
Just as in §3 of [10], since N is dense in Zℓ , the function g can be uniquely extended to a continuous
function
g : Zℓ −→ Zℓ JT K,
which we denote by the same symbol. If a ∈ Zℓ , we let
Qa (T ) = g(a) ∈ Zℓ JT K.
We fix an embedding τ : Q(ζℓ∞ ) ֒→ Qℓ and for i = 0, 1, . . . , we let ξℓi = τ (ζℓi ) ∈ Qℓ . Furthermore
for a ∈ Z, we let
ρℓi (a) = τ (πℓi (a)) = 1 − ξℓai ∈ Qℓ ,
and we write ρℓi rather than ρℓi (1). We denote the valuation on Cℓ by vℓ , normalized so that vℓ (ℓ) = 1,
and the absolute value on Cℓ by | · |ℓ , normalized so that vℓ (x) = − logℓ (|x|ℓ ) for all x ∈ Cℓ . The
relationship between vℓ and the valuation ordLi on Q(ζℓi ) associated to the prime ideal Li is
1
(3) vℓ (τ (x)) = ordLi (x),
ϕ(ℓi )
for all x ∈ Q(ζℓi ), where ϕ is the Euler ϕ-function. In particular, by (1), we have
1 i
(4) vℓ (ρℓi ) = i
and |ρℓi |ℓ = ℓ−1/ϕ(ℓ ) .
ϕ(ℓ )
4 KEVIN MCGOWN, DANIEL VALLIÈRES

The same argument as in §3 of [10] shows that for each i ∈ N, we have a continuous function
Zℓ −→ Cℓ defined via
a 7→ ρℓi (a) = 1 − ξℓai ,
and we note that in fact ρℓi (Zℓ ) ⊆ Qℓ . From now on, we let
D = {x ∈ Cℓ : |x|ℓ < 1}.
Lemma 3.2. Let x ∈ D. The function evx : Zℓ JT K −→ Cℓ given by
Q(T ) 7→ evx (Q(T )) = Q(x)
is uniformly continuous.
Proof. The proof is the same as the one for Lemma 3.2 in [10]. 
We now obtain the following result.
Proposition 3.3. Given a ∈ Zℓ and any i ∈ Z≥0 , we have Qa (ρℓi ) = ρℓi (a).
Proof. If i = 0, the equality is clear. If i ≥ 1, then the functions Zℓ −→ Cℓ defined via
a 7→ ρℓi (a) and a 7→ evρℓi ◦ g(a)
are continuous by (4) and Lemma 3.2. They both agree on N by (2). The result follows as once since
N is dense in Zℓ . 
Remark 3.4. We note in passing that we can think about the power series Qa (T ) as follows. Consider
the unital commutative ring Int(Z)JT K. Then, we have
∞
X
QX (T ) = ek (X)T k ∈ Int(Z)JT K.
k=1
By Proposition 2.1, we have that Int(Z) is a subring of the ring C(Zℓ , Zℓ ) of Zℓ -valued continuous
functions on Zℓ for all rational primes ℓ. Thus, we also have
QX (T ) ∈ C(Zℓ , Zℓ )JT K.
The power series Qa (T ) for a ∈ Zℓ are obtained by setting X = a, that is
∞
X
Qa (T ) = ek (a)T k ∈ Zℓ JT K.
k=1

On the other hand, the power series Pa (T ) ∈ Zℓ JT K arising in [10] are related to
∞
X
PX (T ) = dk (X)T k ∈ Inteven (Z)JT K,
k=1
where Inteven (Z) is the subring of Int(Z) consisting of even integer-valued polynomials, and
 
k−1 X + k − 1 X
dk (X) = (−1) .
2k − 1 k


Note that the dk (X) (along with X0 ) form a Z-basis for Inteven (Z). (See problem 89 of Chapter 2,
Part Eight in [11].)
Remark 3.5. For the following discussion, let ℓ be an odd rational prime and recall the identity
ℓ−1
Y
ℓ
1−X = (1 − ξℓk X).
k=0

Thus, if a ∈ Z×
ℓ and n ∈ N, then
ℓ−1
Y
1 − (ξℓan )ℓ = (1 − ξℓk ξℓan ).
k=0
 ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III 5

Let now Fn = Qℓ (ξℓn ) and consider the norm map Nn : Fn −→ Fn−1 for n ≥ 2. We have
ℓ−1
Y
(5) Nn (ρℓn (a)) = (1 − ξℓk ξℓan ) = ρℓn−1 (a).
k=0
Letting
ρℓn (a)
un (a) = ,
ρℓ n
we have un (a) ∈ E(Fn ), where E(Fn ) is the group of units of the ring of integers of Fn . By (5), we get
Nn (un (a)) = un−1 (a).
Therefore, the power series
Qa (T )
Ra (T ) = ∈ Zℓ JT K×
T
satisfies
Ra (ρℓn ) = un (a),
for all n ≥ 1. It follows that the power series Ra (T ), when a ∈ Z×
ℓ , are particular examples of the ones
studied in [2].

4. Voltage assignments and their derived multigraphs

We recall from §2.2 of [17] how we think about multigraphs, but our notation will be slightly
different. Our main reference is [14]. A directed multigraph X consists of a set of vertices VX and a
~ X with a function
set of directed edges E
~ X −→ VX × VX
inc : E
~ X , we write
called the incidence map. For e ∈ E
inc(e) = (o(e), t(e)),
~ X −→ VX called the origin and the terminus maps. An
so that we obtain two functions o, t : E
undirected multigraph X, or simply a multigraph, consists of a directed multigraph as above with
another function
~ X −→ E
inv : E ~X
called the inversion map which satisfies
(1) inv2 = idE~ X ,
~X ,
(2) inv(e) 6= e for all e ∈ E
~X ,
(3) inc(inv(e)) = ι(inc(e)) for all e ∈ E
where ι : VX × VX −→ VX × VX is defined via ι(v, w) = (w, v). If v ∈ VX , we let
~ X | o(e) = v},
Ev = {e ∈ E
and the valency (or degree) of a vertex is defined to be
valX (v) = |Ev |.
The quotient EX = E ~ X /hinvi is the set of undirected edges. A multigraph is called finite if both VX
and EX are finite sets. All multigraphs arising in this paper will be finite and without any vertex of
valency one, unless otherwise stated.
Note that given a directed multigraph, one can always obtain an (undirected) multigraph by “for-
getting” the directions. Formally, one adds for each directed edge another directed edge in the opposite
direction and the map inv sends a directed edge to this new directed edge with opposite direction. The
details are left to the reader.
We now explain the notion of voltage assignment and their derived multigraphs in a way that is
convenient for us. Our main reference for voltage assignments is Chapter 2 of [6]. Let X be a connected
6 KEVIN MCGOWN, DANIEL VALLIÈRES

multigraph and pick a section ω : EX −→ E ~ X of the natural map E

~ X −→ EX . In other words, we fix
a direction for each of the undirected edges of X. Let
~X.
S = ω(EX ) ⊆ E
If G is an additive finite group, a voltage assignment with values in G is a function
α : S −→ G.
To each such voltage assignment, one can associate a derived multigraph which we denote by X(G, S, α).
The multigraph X(G, S, α) is obtained as follows. The vertices of X(G, S, α) consist of VX × G and the
edges consist of S ×G. An edge (s, σ) ∈ S ×G connects the vertex (o(s), σ) to the vertex (t(s), σ+α(s)).
In this way, we get a directed multigraph, but we forget about the directions and think of X(G, S, α)
as a multigraph. Note that the choice of directions for the loops of the base multigraph does not
affect the resulting multigraph X(G, S, α). If X is a bouquet, then the multigraphs X(G, S, α) are the
Cayley-Serre multigraphs used in [10] and [17].
For instance, if we let X be the dumbbell multigraph
,
the section be
s1 s3
s2
,
and the function α : S −→ Z/5Z be defined via α(s1 ) = 1̄, α(s2 ) = 0̄ and α(s3 ) = 2̄, then the graph
X(Z/5Z, S, α) is the Petersen graph

From now on, if X is a connected multigraph and we say “Let α : S −→ G be a function. . . ”, it

will be understood that a section ω has been chosen and that S = ω(EX ). Now, if we start with a
connected multigraph X and a function α : S −→ Zℓ , then for each n ∈ N, we consider the voltage
assignment
αn : S −→ Z/ℓn Z
obtained from the composition
α ≃
S −→ Zℓ −→ Zℓ /ℓn Zℓ −→ Z/ℓn Z.
If we assume that all the multigraphs X(Z/ℓn Z, S, αn ) are connected, then we get an abelian ℓ-tower
of multigraphs
X ←− X(Z/ℓZ, S, α1 ) ←− X(Z/ℓ2 Z, S, α2 ) ←− . . . ←− X(Z/ℓn Z, S, αn ) ←− . . .
in the sense of Definition 4.1 of [17].
For example, if we take ℓ = 5, X and S as above, and the function α : S −→ Z ⊆ Z5 defined by
α(s1 ) = 1, α(s2 ) = 0, and α(s3 ) = 2, then we have
X(Z/5n Z, S, αn ) = G(5n , 2),
where G(5n , 2) is the generalized Petersen graph introduced in [3] and [19]. We then get an abelian
5-tower

←− ←− ←− . . .

Note that since X is 3-regular, then so are G(5n , 2) for all n ≥ 1.

 ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III 7

5. The special value at u = 1 of Artin-Ihara L-functions

We now recall what we need to know about Artin-Ihara L-functions for our present purposes. Our
main reference for this section is [15], but our notation will be the same as in §2 of [17]. Let Y /X be
an abelian cover of connected multigraphs and let G = Gal(Y /X). If ψ ∈ G b = HomZ (G, C× ), then
we denote the corresponding Artin-Ihara L-function by LX (u, ψ). Let us introduce a labeling of the
vertices VX = {v1 , . . . , vg } and for each i = 1, . . . , g, let us fix a vertex wi ∈ VY above vi . Recall the
following definition (Definition 18.13 in [15]).
Definition 5.1. With the same notation as above,
(1) For σ ∈ G, we let A(σ) = (aij (σ)) be the g × g matrix defined via
(
Twice the number of undirected loops at the vertex wi , if i = j and σ = 1;
aij (σ) =
The number of undirected edges connecting wi to wjσ , otherwise.

b then we let
(2) If ψ ∈ G,
X
Aψ = ψ(σ) · A(σ).
σ∈G

If D = (dij ) denotes the valency matrix attached to X, that is D is diagonal and dii = valX (vi ),
then the three-term determinant formula for the Artin-Ihara L-function (Theorem 18.15 in [15]) gives
1
= (1 − u2 )−χ(X) · det(I − Aψ u + (D − I)u2 ),
LX (u, ψ)
where χ(X) is the Euler characteristic of X. As in [17], we let
hX (u, ψ) = det(I − Aψ u + (D − I)u2 ) ∈ Z[ψ][u],
where Z[ψ] is the ring of integers in the cyclotomic number field Q(ψ). Evaluating at u = 1 gives
(6) hX (1, ψ) = det(D − Aψ ) ∈ Z[ψ].
b as usual. Assuming χ(X) 6= 0, equation (7) in
We let the absolute Galois group GQ of Q acts on G
[17] shows that
Y
|G| · κY = κX hX (1, Ψ),
Ψ6=Ψ0

where κ denotes the number of spanning trees of a multigraph, the product is over all non-trivial orbits
b and
Ψ ∈ GQ \G,
Y
hX (1, Ψ) = hX (1, ψ) ∈ Z.
ψ∈Ψ

From now on, we assume that χ(X) 6= 0. (The case where χ(X) = 0 was treated separately. See
the discussion after Definition 4.1 of [17].) If
X = X0 ←− X1 ←− X2 ←− . . . ←− Xn ←− . . .
is an abelian ℓ-tower consisting of connected multigraphs above X, then for each i = 1, 2, . . ., we let
Gi = Gal(Xi /X) ≃ Z/ℓi Z. Furthermore, we let ψi ∈ G b i be the faithful character satisfying ψi (1̄) = ζℓi .
Equation (12) of [17] adapted to the case where κX 6= 1 gives
n
X
(7) ordℓ (κn ) = ordℓ (κX ) − n + ordLi (hX (1, ψi )),
i=1

where κn is the number of spanning trees of Xn .

8 KEVIN MCGOWN, DANIEL VALLIÈRES

Lemma 5.2. Let X be a connected multigraph and G an additive finite group. Let α : S −→ G be
a voltage assignment for which X(G, S, α) is connected. Choose a labeling of the vertices of X, say
VX = {v1 , . . . , vg } and let wi = (vi , 0G ) ∈ VX(G,S,α) , where 0G is the neutral element of G. For each
i, j ∈ {1, . . . , g} and σ ∈ G, let
bij (σ) = |{s ∈ S | inc(s) = (vi , vj ) and α(s) = σ}|
and
cij (σ) = |{s ∈ S | inc(s) = (vj , vi ) and α(s) = −σ}|.
Then, the entries aij (σ) of the matrix A(σ) of Definition 5.1 satisfy
aij (σ) = bij (σ) + cij (σ).
Proof. This follows from the definition of the derived multigraphs X(G, S, α) given in §4. 

Corollary 5.3. With the same notation as in Lemma 5.2, the matrix Aψ of Definition 5.1 satisfies
 
 X X 
Aψ = 
 ψ(α(s)) + ψ(−α(s))
.
s∈S s∈S
inc(s)=(vi ,vj ) inc(s)=(vj ,vi )

Proof. The ij-th entry of Aψ is given by

X X X
ψ(σ)aij (σ) = ψ(σ)bij (σ) + ψ(σ)cij (σ)
σ∈G σ∈G σ∈G
X X
= ψ(α(s)) + ψ(−α(s)),
s∈S s∈S
inc(s)=(vi ,vj ) inc(s)=(vj ,vi )

by Lemma 5.2, and this is what we wanted to show. 

Remark 5.4. If X is a bouquet, then the last corollary specializes to Theorem 5.2 of [17].
We will now apply Lemma 5.2 and Corollary 5.3 to the successive layers in an abelian ℓ-tower of
multigraphs.
Proposition 5.5. Let X be a connected multigraph, ℓ a rational prime, and let α : S −→ Zℓ be
a function for which all of the multigraphs X(Z/ℓn Z, S, αn ) are connected. Label the vertices VX =
{v1 , . . . , vg }. For each n ≥ 1, we let
wi,n = (vi , 0̄) ∈ VX(Z/ℓn Z,S,αn) .
For i, j ∈ {1, . . . , g}, let us define the integers
Bij = |{s ∈ S | inc(s) = (vi , vj )}|
and
Cij = |{s ∈ S | inc(s) = (vj , vi )}|.
Then, for k = 0, 1, 2, . . ., we have
 
 X X 
τ (D − Aψk ) = 
dij − Bij − Cij + ρℓk (α(s)) + ρℓk (−α(s))
.
s∈S s∈S
inc(s)=(vi ,vj ) inc(s)=(vj ,vi )

Proof. For k ∈ N, let pk : Zℓ −→ Z be the function defined by

∞
! k−1
X X
i
pk ai ℓ = ai ℓi , (ai ∈ {0, . . . , ℓ − 1}).
i=0 i=0
 ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III 9

By Corollary 5.3, the ij-th entry of the matrix D − Aψk is

 
 X X 
dij − 
 ψk (αk (s)) + ψk (−αk (s))

s∈S s∈S
inc(s)=(vi ,vj ) inc(s)=(vj ,vi )

which is equal to
X X
dij − Bij − Cij + πℓk (pk (α(s)) + πℓk (−pk (α(s)).
s∈S s∈S
inc(s)=(vi ,vj ) inc(s)=(vj ,vi )

Applying τ to this last equation gives the desired result. If k = 0, we get the Laplacian matrix of X
on both sides, so we still have an equality. 

Corollary 5.6. With the same notation as in Proposition 5.5, write

α(S) = {b1 , . . . , bt }

for some distinct by ∈ Zℓ . For i, j ∈ {1, . . . , g} and y ∈ {1, . . . , t}, define the integers
(y)
γij = |{s ∈ S | inc(s) = (vi , vj ) and α(s) = by }|,

(y)
δij = |{s ∈ S | inc(s) = (vj , vi ) and α(s) = by }|,
and the homogenous linear polynomials
(1) (t) (1) (t)
Pij (X1 , . . . , Xt , Y1 , . . . , Yt ) = γij X1 + . . . + γij Xt + δij Y1 + . . . + δij Yt .

Furthermore, define the g × g matrix

M = (dij − Bij − Cij + Pij (X1 , . . . , Xt , Y1 , . . . , Yt )),

and the polynomial with integer coefficients

P (X1 , . . . , Xt , Y1 , . . . , Yt ) = det(M ).

Then
(1) The polynomial P has no constant term,
(2) τ (hX (1, ψk )) = P (ρℓk (b1 ), . . . , ρℓk (bt ), ρℓk (−b1 ), . . . , ρℓk (−bt )) for all k ≥ 1.

Proof. Indeed, by Proposition 5.5, we have

det(τ (D − Aψk )) = P (ρℓk (b1 ), . . . , ρℓk (bt ), ρℓk (−b1 ), . . . , ρℓk (−bt )),

and thus the second claim follows from (6).

For the first claim, note that

P (0, . . . , 0) = det(dij − Bij − Cij ),

but the matrix (dij − Bij − Cij ) is the Laplacian matrix of X which is singular. 

Remark 5.7. If X is a bouquet, then the polynomial P of Corollary 5.6 has the simple form

P (X1 , . . . , Xt , Y1 , . . . , Yt ) = C1 (X1 + Y1 ) + . . . + Ct (Xt + Yt ),

for some integers Ci (i = 1, . . . , t).

10 KEVIN MCGOWN, DANIEL VALLIÈRES

6. Abelian ℓ-towers of multigraphs

We can now state our main result which can be viewed as a graph theoretical analogue of a theorem
of Iwasawa. (See Theorem 11 in [7] and §4.2 of [8].)
Theorem 6.1. Let X be a connected multigraph satisfying χ(X) 6= 0 and let α : S −→ Zℓ be a function
for which all multigraphs X(Z/ℓn Z, S, αn ) are connected. Write
α(S) = {b1 , . . . , bt },
where the by are distinct ℓ-adic integers. Consider the abelian ℓ-tower
X ←− X(Z/ℓZ, S, α1 ) ←− X(Z/ℓ2 Z, S, α2 ) ←− . . . ←− X(Z/ℓn Z, S, αn ) ←− . . .
and define the ℓ-adic integers cj via
Q(T ) = P (Qb1 (T ), . . . , Qbt (T ), Q−b1 (T ), . . . , Q−bt (T ))
= c1 T + c2 T 2 + . . . ∈ Zℓ JT K,
where P (X1 , . . . , Xt , Y1 , . . . , Yt ) ∈ Z[X1 , . . . , Xt , Y1 , . . . , Yt ] is the polynomial from Corollary 5.6. Let
µ = min{vℓ (cj ) | j = 1, 2, . . .},
and
λ = min{j | j ∈ N and vℓ (cj ) = µ} − 1.
If κn denotes the number of spanning trees of X(Z/ℓn Z, S, αn ), then there exist a nonnegative integer
n0 and a constant ν ∈ Z (depending also on the by ) such that
ordℓ (κn ) = µℓn + λn + ν,
when n ≥ n0 .
Proof. Our starting point is the formula (7). Combining with (3) and Corollary 5.6 give
n
X
ordℓ (κn ) = ordℓ (κX ) − n + ϕ(ℓi )vℓ (τ (hX (1, ψi )))
i=1
n
X
= ordℓ (κX ) − n + ϕ(ℓi )vℓ (P (ρℓi (b1 ), . . . , ρℓi (bt ), ρℓi (−b1 ), . . . , ρℓi (−bt ))) .
i=1

By Proposition 3.3, we have

P (ρℓi (b1 ), . . . , ρℓi (bt ), ρℓi (−b1 ), . . . , ρℓi (−bt )) = P (Qb1 (ρℓi ), . . . , Qbt (ρℓi ), Q−b1 (ρℓi ), . . . , Q−bt (ρℓi ))
= Q(ρℓi ).
The exact same argument as in the proof of Theorem 4.1 of [10] shows that for i large, we have
λ+1
vℓ (Q(ρℓi )) = µ + .
ϕ(ℓi )
Therefore, there exists n0 ≥ 0 and an integer C such that if n ≥ n0 , then
n
X
ordℓ (κn ) = ordℓ (κX ) − n + C + ϕ(ℓi )vℓ (Q(ρℓi ))
i=n0
Xn
= ordℓ (κX ) − n + C + (µ · ϕ(ℓi ) + (λ + 1))
i=n0

= ordℓ (κX ) − n + C + (λ + 1)(n − (n0 − 1)) + µ(ℓn − ℓn0 −1 ),

and this ends the proof. 
 ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III 11

7. Examples
The computations of the number of spanning trees in this section have been performed with the soft-
ware [13]. The computations of the approximated power series Qa (T ) and Q(T ) have been performed
with the software [16]. Examples are numerous and we present only a few.
(1) Let us revisit example (1) of §5.6 of [17]. In this case, ℓ = 2, and we set b1 = 1. The 1 × 1
matrix M is given by

2X1 + 2Y1 .
We have Q1 (T ) = T , and using Remark 3.4, we have
Q−1 (T ) = −T − T 2 − T 3 − . . . ,
since ek (−1) = −1 for all k ≥ 1. Thus, the power series Q is
Q(T ) = −2T 2 − 2T 3 − 2T 4 − . . . ∈ ZJT K ⊆ Z2 JT K,
so we should have µ = 1 and λ = 1 which was indeed the case. We point out in passing that
when working with the power series Qa (T ) of the current paper rather than the power series
Pa (T ) of [10], the power series Q(T ) is not a polynomial anymore. The reason is because the
power series Pa (T ) satisfy Pa (T ) = P−a (T ) and Pa (T ) is a polynomial when a ∈ N, but the
Qa (T ) of the current paper do not satisfy this property.
(2) Let us revisit example (3) of §5.6 of [17]. In this case, ℓ = 3, and we let b1 = 1, b2 = 4, and
b3 = 20. The 1 × 1 matrix M is given by


X1 + X2 + X3 + Y1 + Y2 + Y3 .
The power series Q is given by
−417T 2 − 417T 3 − 13737T 4 − 27057T 5 − 215945T 6 + . . . ∈ ZJT K ⊆ Z3 JT K
The factorizations of the coefficients are
417 = 3 · 139, 13737 = 3 · 19 · 241, 27057 = 3 · 29 · 311, 215945 = 5 · 43189,
so we should have µ = 0 and λ = 5 which was indeed the case.
(3) Let us revisit example (1) of §5 of [10]. In this case, ℓ = 2, and we let b1 = 1/3 and b2 = 3/5.
The 1 × 1 matrix M is given by


X1 + X2 + Y1 + Y2 .
The power series Q(T ) ∈ Z2 JT K is given by
(0.1101 . . .)T 2 + (0.1101 . . .)T 3 + (0.0011 . . .)T 4 + (0.1011 . . .)T 5 + (1.1100 . . .)T 6 + . . . ,
so we should have µ = 0 and λ = 5 which was indeed the case.
(4) Let us revisit (a) of example 7 in §5.1 of [5]. Let X = K10 be the complete graph on ten
vertices and let ℓ = 5. Label the vertices VX = {v1 , . . . , v10 } and pick a section ω so that s1
is the unique edge going from v1 to v2 and all edges are directed using the lexicographic order
on VX . Consider the function α : S −→ Z5 given by α(s1 ) = 1 and α(si ) = 0 if i 6= 1. We let
b1 = 0 and b2 = 1. The 10 × 10 matrix M is given by
 9 −1 + X2 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 
−1 + Y2 9 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 
−1 + Y −1 + Y1 9 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 
 1 
−1 + Y1 −1 + Y1 −1 + Y1 9 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 
 
−1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 9 −1 + X1 −1 + X1 −1 + X1 −1 + X1 −1 + X1 
 
−1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 9 −1 + X1 −1 + X1 −1 + X1 −1 + X1 
 
−1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 9 −1 + X1 −1 + X1 −1 + X1 

−1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 9 −1 + X1 −1 + X1 

−1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 9 −1 + X1
 
−1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 −1 + Y1 9

The power series Q starts as follows

Q(T ) = −210 · 57 · T 2 + T 3 + T 4 + . . . ∈ ZJT K ⊆ Z5 JT K,
so we should have µ = 7 and λ = 1 which is what was obtained in [5] as well.
12 KEVIN MCGOWN, DANIEL VALLIÈRES

(5) Let ℓ = 5 and let X be the dumbbell multigraph. Let α : S −→ Z ⊆ Z5 be as in the example
of §4, where we label the vertices of X from left to right, so the left vertex is v1 and the right
vertex is v2 . We let b1 = 1, b2 = 0, and b3 = 2. Then, we get:
X ←− G(5, 2) ←− G(52 , 2) ←− G(53 , 2) ←− . . . ,
where G(n, k) denotes the generalized Petersen graph. The 2 × 2 matrix M is given by
 
1 + X1 + Y1 −1 + X2
.
−1 + Y2 1 + X3 + Y3
The power series Q starts as follows
Q(T ) = −5T 2 − 5T 3 − 2T 4 + . . . ∈ ZJT K ⊆ Z5 JT K,
so we should have µ = 0 and λ = 3. We calculate
κ0 = 1, κ1 = 24 · 53 , κ2 = 24 · 56 · 4804492,

κ3 = 24 · 59 · 1877512 · 4804492 · 18295012 · 47317512 · 163901034517492, . . .

We have
ord5 (κn ) = 3n,
for all n ≥ 0.
(6) Let ℓ = 5 and let X be the dumbbell multigraph from §4. We take the same section, and
we label the vertices of X from left to right as in the previous example. We change α to be
α : S −→ Z5 , where
α(s1 ) = 2.11111 . . . ∈ Z5 ,
and α(s2 ) = 0, α(s3 ) = 2. We let b1 = 2.11111 . . ., b2 = 0 and b3 = 2. Note that we have
p1 (b1 ) = 2, p2 (b1 ) = 7, p3 (b1 ) = 32, . . .
Then, we get an abelian 5-tower
X ←− I(5, 2, 2) ←− I(52 , 7, 2) ←− I(53 , 32, 2) ←− . . . ,
where I(n, j, k) denotes the I-graph introduced in [4]. For instance, I(5, 2, 2) is

The 2 × 2 matrix M is as in the previous example

 
1 + X1 + Y1 −1 + X2
.
−1 + Y2 1 + X3 + Y3
The power series Q starts as follows
Q(T ) = (2.4321 . . .)T 2 + . . . ∈ Z5 JT K,
so we should have µ = 0 and λ = 1. We calculate
κ0 = 1, κ1 = 5 · 192 , κ2 = 52 · 192 · 53404492,

κ3 = 53 · 192 · 2512 · 2729992 · 53404492 · 601064972512 · 692661252719782512, . . .

We have
ord5 (κn ) = n,
for all n ≥ 0.
 ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III 13

(7) We modify the last example slightly. Add one more loop at v2 and denote it by s4 . In this
case, the base multigraph is not regular anymore. Take α : S −→ Z5 to be defined via
α(s1 ) = 2.1111111 . . . , α(s2 ) = 0, α(s3 ) = 2, α(s4 ) = 1,
and let b1 = 2.11111 . . ., b2 = 0, b3 = 2, and b4 = 1. We get an abelian 5-tower above the
modified dumbbell multigraph. The multigraph at the first layer is

The 2 × 2 matrix M is
 
1 + X1 + Y1 −1 + X2
.
−1 + Y2 1 + X3 + Y3 + X4 + Y4
The power series Q starts as follows
Q(T ) = (1.4321 . . .)T 2 + . . . ∈ Z5 JT K,
so we should have µ = 0 and λ = 1. We calculate
κ0 = 1, κ1 = 5 · 712 , κ2 = 52 · 712 · 1012 · 13992 · 207492,
and κ3 is
53 · 712 · 1012 · 13992 · 97492 · 207492 · 21445012 · 51007512 · 792199660148169081532697559102492

We have
ord5 (κn ) = n,
for all n ≥ 0.
(8) Let ℓ = 3, and consider now the following multigraph X

and the section:

v4

s6 s5
s1 s3
v1 v2 v3
s2 s4
Consider the function α : S −→ Z ⊆ Z3 defined via
α(s3 ) = α(s6 ) = 1, α(s4 ) = 2, α(s1 ) = α(s2 ) = α(s5 ) = 0.
We then obtain an abelian 3-tower over X:
X ←− X(Z/3Z, S, α1 ) ←− X(Z/32 Z, S, α2 ) ←− . . . ←− X(Z/3n Z, S, αn ) ←− . . .
The 4 × 4 matrix M is given by
 
3 −2 + 2X1 0 −1 + Y2
−2 + 2Y1 4 −2 + X2 + X3 0 
 .
 0 −2 + Y2 + Y3 3 −1 + X1 
−1 + X2 0 −1 + Y1 2
14 KEVIN MCGOWN, DANIEL VALLIÈRES

The power series Q starts as follows

−31T 2 − 31T 3 − 45T 4 − . . . ∈ ZJT K ⊆ Z3 JT K,
so we should have µ = 0 and λ = 1. We calculate
κ0 = 22 · 3, κ1 = 22 · 32 · 72 , κ2 = 22 · 33 · 72 · 22512 , . . .
We have
ord3 (κn ) = n + 1,
for all n ≥ 0.

8. Iwasawa modules arising from graph theory

We end this paper with the following comment. Let Γ be a multiplicative topological group isomor-
n
phic to Zℓ . For n ≥ 1, let Γn = Γ/ Γℓ ≃ Z/ℓn Z, and consider the classical Iwasawa algebra
Λ = lim Z [Γ ].
←− ℓ n
n≥1

It was shown in [12] that there is an isomorphism

≃
Zℓ JT K −→ Λ,
given by T 7→ 1 − γ, where γ is a topological generator of Γ. There is a known structure theorem
for finitely generated Λ-modules up to pseudo-isomorphisms (see for instance [18]) and to every such
finitely generated Λ-module N is associated its Iwasawa invariants µ(N ) and λ(N ). In light of Theorem
6.1, it is natural to ask if there would be some finitely generated Λ-modules arising from graph theory.
If
X = X0 ←− X1 ←− X2 ←− . . . ←− Xn ←− . . .
is an abelian ℓ-tower of multigraphs, let An be the Sylow ℓ-subgroup of the Picard group Pic◦ (Xn ).
Recall that the cardinality of Pic◦ (Xn ) is κn and thus
|An | = ℓordℓ (κn ) .
(See the introduction to [9] for an overview of where the finite abelian group Pic◦ (Xn ) arises in
mathematics and under what names.) If the base graph X is a simple graph (meaning it does not
contain loops and parallel edges), Gonet constructed a Λ-module that is finitely generated and for
which one can recapture the An by looking at some of its quotients. This allowed her to obtain the
existence of the invariants µ and λ of Theorem 6.1 when the base multigraph is a simple graph. (See
Theorem 27 of [5].) Theorem 6.1 therefore gives in particular a way to calculate the Iwasawa invariants
introduced in her thesis.

References
[1] Paul-Jean Cahen and Jean-Luc Chabert. What you should know about integer-valued polynomials. Amer. Math.
Monthly, 123(4):311–337, 2016.
[2] Robert F. Coleman. Division values in local fields. Invent. Math., 53(2):91–116, 1979.
[3] H. S. M. Coxeter. Self-dual configurations and regular graphs. Bull. Amer. Math. Soc., 56:413–455, 1950.
[4] Ronald M. Foster. The Foster census. Charles Babbage Research Centre, Winnipeg, MB, 1988. R. M. Foster’s
census of connected symmetric trivalent graphs, With a foreword by H. S. M. Coxeter, With a biographical preface
by Seymour Schuster, With an introduction by I. Z. Bouwer, W. W. Chernoff, B. Monson and Z. Star, Edited and
with a note by Bouwer.
[5] Sophia Rose Gonet. Jacobians of Finite and Infinite Voltage Covers of Graphs. ProQuest LLC, Ann Arbor, MI,
2021. Thesis (Ph.D.)–The University of Vermont and State Agricultural College.
[6] Jonathan L. Gross and Thomas W. Tucker. Topological graph theory. Dover Publications, Inc., Mineola, NY, 2001.
Reprint of the 1987 original [Wiley, New York; MR0898434 (88h:05034)] with a new preface and supplementary
bibliography.
[7] Kenkichi Iwasawa. On Γ-extensions of algebraic number fields. Bull. Amer. Math. Soc., 65:183–226, 1959.
[8] Kenkichi Iwasawa. On Zl -extensions of algebraic number fields. Ann. of Math. (2), 98:246–326, 1973.
[9] Dino Lorenzini. Smith normal form and Laplacians. J. Combin. Theory Ser. B, 98(6):1271–1300, 2008.
[10] Kevin J. McGown and Daniel Vallières. On abelian ℓ-towers of multigraphs II, 2021. arXiv:2105.08661.
 ON ABELIAN ℓ-TOWERS OF MULTIGRAPHS III 15

[11] George Pólya and Gábor Szegő. Problems and theorems in analysis II. Classics in Mathematics. Springer-Verlag,
Berlin, 1998. Theory of functions, zeros, polynomials, determinants, number theory, geometry, Translated from the
German by C. E. Billigheimer, Reprint of the 1976 English translation.
[12] Jean-Pierre Serre. Classes des corps cyclotomiques (d’après K. Iwasawa). In Séminaire Bourbaki, Vol. 5, pages Exp.
No. 174, 83–93. Soc. Math. France, Paris, 1995.
[13] William Stein. Sage: Open Source Mathematical Software (Version 8.5). The Sage Group, 2018. Available from
http://www.sagemath.org.
[14] Toshikazu Sunada. Topological crystallography, volume 6 of Surveys and Tutorials in the Applied Mathematical
Sciences. Springer, Tokyo, 2013. With a view towards discrete geometric analysis.
[15] Audrey Terras. Zeta functions of graphs, volume 128 of Cambridge Studies in Advanced Mathematics. Cambridge
University Press, Cambridge, 2011. A stroll through the garden.
[16] The PARI Group, Bordeaux. PARI/GP, version 2.9.1, 2016. Available from http://pari.math.u-bordeaux.fr/.
[17] Daniel Vallières. On abelian ℓ-towers of multigraphs. To appear in Annales Mathématiques du Québec, 2021.
[18] Lawrence C. Washington. Introduction to cyclotomic fields, volume 83 of Graduate Texts in Mathematics. Springer-
Verlag, New York, second edition, 1997.
[19] Mark E. Watkins. A theorem on Tait colorings with an application to the generalized Petersen graphs. J. Combi-
natorial Theory, 6:152–164, 1969.

Mathematics and Statistics Department, California State University, Chico, CA 95929 USA
Email address: kmcgown@csuchico.edu

Mathematics and Statistics Department, California State University, Chico, CA 95929 USA
Email address: dvallieres@csuchico.edu