1. Introduction
The Cayley graph $\Gamma (G;S)$
of a finitely generated group $G$
equipped with a finite generating subset $S$
allows us to study the group $G$
using both combinatorial and geometric methods. One invariant of the group that arises in this viewpoint is the growth function $V(n;S)$
which was introduced by Schwarz and Milnor independently and is given by the size of a ball of radius $n$
in $\Gamma (G;S)$
. The motivation for this invariant is geometric as seen in case when $G$
is the fundamental group of a Riemannian manifold where the growth function of $G$
gives a discrete approximation of the volume growth of the universal cover of the manifold [Reference Švarc19]. Moreover, it has been shown that the growth function has an exponential growth when the manifold is negatively curved and has polynomial growth when $G$
is virtually nilpotent [Reference Milnor11, Reference Wolf21].
Milnor asked whether the growth function is always either an exponential function or a polynomial function. Moreover, he asked whether one can classify every group whose growth exponent $\lim {\log V(n)}/{\log n}$
exists [Reference Milnor12]. It was shown by Tits [Reference Tits20] that there is no linear group of intermediate growth whereas it is known to be not true in general by work of Grigorchuk [Reference Grigorchuk8]. For the second problem, Bass showed that virtually nilpotent groups always have an integer degree of polynomial growth [Reference Bass1]. Finally, Gromov showed that a group has polynomial growth if and only if it is virtually nilpotent [Reference Gromov9] giving a complete answer to Milnor's question.
For nilpotent groups, one can ask how precisely the growth function $V(n;S)$
behaves like a polynomial over $\mathbb {Q}$
. Towards this end, Pansu showed that the limit $\alpha =\lim _{n\rightarrow \infty }V(n;S)/n^d$
exists for nilpotent groups [Reference Pansu14]. Thus, one may study whether $\alpha$
is a rational number. One approach to this question is to study the growth series $\sum \nolimits V(n;S)t^n$
associated with the growth function and use its analytic properties. If the growth series is a rational function, for instance, one can then show that the limit $\alpha$
is an algebraic integer which is a consequence of the coefficients being linearly recursive.
It turns out the rationality of the growth series has stronger implications to the computability of the group with respect to a finite generating set. As a corollary to the fact that $V(n)$
is linearly recursive when the growth series is rational, one can then see the group $G$
always has a solvable word problem. Cannon showed the growth series for any hyperbolic group is rational by essentially showing that all hyperbolic groups are strongly geodesically automatic [Reference Cannon5]. In fact, Neumann and Shapiro showed that the growth series with respect to $S$
is rational when the full language of geodesics in $G$
is regular [Reference Neumann and Shapiro13].
Little is known about what groups have rational growth with respect to some finite generating subset. In fact, there are only a handful of classes of finitely generated groups that are known to have rational growth with respect to any finite generating subset. Virtually abelian groups were shown to have rational growth by Benson [Reference Benson2]. For Coxeter groups, it was shown by Paris [Reference Paris15], and for solvable Baumslag–Solitar groups, it was shown by Brazil, and independently by Collins, Edjvet, and Gill [Reference Brazil3, Reference Collins, Edjvet and Gill6]. Among these classes of finitely generated groups, hyperbolic groups and virtually abelian groups are also shown to have rational growth with respect to not only some fixed finite generating set, but for all finite generating sets [Reference Benson2, Reference Cannon5]. This property is known as panrationality and the only known example of a panrational group outside of virtually abelian groups and hyperbolic groups is the integral Heisenberg group of dimension 3 [Reference Duchin and Shapiro7]. In general, the rationality of the growth of a group depends on the choice of generating subset. In particular, this was demonstrated by Stoll [Reference Stoll18] in the case of the integral Heisenberg group of dimension 5. Stoll constructed two generating subsets for this group, one for which the generating function associated with the growth function is rational and the other where the generating function is not rational. Thus, a natural direction of research is to investigate when a finitely generated group of interest admits finite generating subsets that are rational, and we may consider the finite generating sets for which the group has rational growth as algorithmically nicer than other generating sets.
Our interests are in the study of rationality of finitely presented solvable groups. Since Kharlampovich [Reference Kharlampovich10] constructed examples of finitely presented solvable groups of derived length $3$
that have unsolvable word problem, we will focus on metabelian, i.e. solvable of derived length 2, groups in this paper. In particular, we will focus on a family of torus bundle groups given by $G = \mathbb {Z}^2 \rtimes \mathbb {Z}$
. Parry showed that when the trace of the action is even and at least 4, these groups have rational growth with respect to some generating set [Reference Parry16]. We extend Parry's result to the traces that are odd and at least 5.
Main Theorem Let $A = \left [\begin {smallmatrix} 0 & -1 \\ 1 & 2k+1\end {smallmatrix}\right ]\in SL(2,\,\mathbb {Z})$
where $2k+1 \ge 5$
. Then $\mathbb {Z}^2\rtimes _A\mathbb {Z}$
has rational growth with respect to the standard generating subset.
It was shown that certain groups of the form $\mathbb {Z}^m\rtimes \mathbb {Z}$
have a finite-index subgroup which have rational growth in some finite generating set [Reference Bregman4, Reference Putman17]. Our work improves on this result for $m = 2$
as we do not require passing to a finite-index subgroup. Moreover, we are able to generalize and to streamline the proof given by Parry. In particular, we define an invariant called the potential which makes the proof cleaner. We conjecture that our argument can be applied to all torus bundle groups to get the following:
Conjecture 1 Any torus bundle group $G = \mathbb {Z}^2 \rtimes _A \mathbb {Z}$
where $A \in SL(2,\,\mathbb {Z})$
has rational growth with respect to some finite generating set.
1.1. Outline of the paper
The general structure of this work is as follows:
Section . In this section, we first recall the definition of a torus bundle group and its structure and give a description of the word length with a specific choice of generators.
Section . Following [Reference Parry16], we reduce the problem of finding the word with the shortest length representing a group element into an optimization problem in the Laurent polynomial ring over the integers. Such minimizing polynomial is called the $n$
-minimal polynomial. We also define a related notion called the $n$
-reduced polynomial given by a list of rules on coefficients.Section . In this section, we give an explicit algorithm for rewriting any polynomial into the $n$
-reduced polynomial. We show that any $n$
-minimal polynomial can be rewritten so that it splits into $n$
-reduced polynomial part and principal part without increasing the length.Section . With rewriting rules established in the previous section, we classify all cases of $n$
-reduced polynomials that require rewriting when you add 1. This will be used to subdivide $n$
-reduced polynomials into $n$
-types and $n$
-classes.Section . Precise definition for the $n$
-types and $n$
-classes are given in this section. The $n$
-types and $n$
-classes are defined recursively over the degree of the polynomial. This recursive definition allows us to inductively compute the growth series for $n$
-reduced polynomials.Section . In this section, we define the successor. We show that any $n$
-reduced polynomial with non-negative leading coefficient can be obtained by applying successor repeatedly on 1 and that it is injective. When counting certain $n$
-classes and $n$
-types is difficult, we will use the successor to count the corresponding $n$
-classes and $n$
-types.Section . Combining the results from § 6, we will find the growth series for the $n$
-reduced polynomial whose leading coefficient is non-negative. We then use this to establish the rationality of the growth of the whole group.
2. Torus bundle groups
In this section, we briefly give the definition of the class of torus bundle groups that we are interested in and a description of their word metrics with respect to a specific choice of finite generating subset. When given a finitely generated group $G$
equipped with a finite generating subset $S$
, we write $\|g\|_S$
to be the minimal length over all representatives of $g$
in the alphabet in $S$
. When the generating subset is clear from context, we will suppress the subscript.
Let $T$
be the automorphism of $\mathbb {Z}^2$
given by the matrix
with respect to the standard basis $a = [\begin {smallmatrix} 1 \\ 0 \end {smallmatrix}],\, \: b = [\begin {smallmatrix} 0 \\ 1 \end {smallmatrix}]$
of $\mathbb {Z}^2$
. Let $\langle t \rangle$
be the infinite cyclic group generated by $t$
. We define the action of $t$
on $\mathbb {Z}^2$
by $t \: x \: t^{-1} = T \cdot x$
for any $x \in \mathbb {Z}^2$
and form the semidirect product $G = \mathbb {Z}^2 \rtimes \langle t \rangle$
. We will view elements of $G$
as ordered pairs $(x,\,t^n)$
where $x \in \mathbb {Z}^2$
and $n \in \mathbb {Z}$
. Thus, we may write $G$
as
where $g^t = t \: g \: t^{-1}$
. In particular, we will investigate the word growth of $G$
with respect to the generating subset $S = \{a,\,b,\,t\}$
.
2.1. Length of elements in $G$
Let $g=(x,\,t^n) \in G$
. We may rewrite $g$
where we collect consecutive factors which are either $a$
's or $b$
's but not collecting consecutive factors which are either $t$
's or their inverses. This allows us to write
where $e \in \mathbb {N}$
, $n_0,\,n_e \in \{-1,\,0,\,1\}$
, $n_1,\,\dotsc,\, n_{e-1} \in \{-1,\,1\}$
, and $x_1,\,\dotsc,\,x_e \in \mathbb {Z}^2$
. We may obtain the word length of $g$
by minimizing the sum

where $|x_i|$
is the length of $x_i$
in $\mathbb {Z}^2$
with respect to the generating subset $\{a,\,b\}$
. We rewrite the given expression for $g$
as
We then note that
for each $0 \leq \ell \leq e-1$
and that $t^{m} \: x \: t^{-m} = T^m \cdot x$
for all $m \in \mathbb {Z}$
and $x \in \mathbb {Z}^2$
. Therefore, we have that
By the above construction, we have that $n = \sum \nolimits _{i=0}^e n_i$
.
For notational simplicity, we let
and
Assume that $n \geq 0$
. By combining terms according to the partial sums $n_0 + \ldots + n_i$
, we see that we will not increase the length of the expression in Equation (1). Thus, we may rewrite Equation (2) as

for elements $y_{-p},\,\dotsc,\, y_q \in \mathbb {Z}^2$
. We furthermore obtain $\|g\|$
by minimizing

We may suppose that $y_i = r_i \: a + s_i \: b$
with $r_i,\,s_i \in \mathbb {Z}$
. Given that $T(a) = b$
, we may write $T^i(y_i)$
as
where $i>0$
. If $i < 0$
, then we may write $T^i(y_i)$
as
We have a similar situation for $n \leq 0$
. Thus, if $g = (x,\,t^n) \in G$
, then $\|g\|$
is the minimal value of

over all representatives of $x$
given by

where $r'_0,\,\dotsc,\, r'_p,\,s'_0,\,\dotsc,\,s'_q$
are integers, $p \geq \text {max}\{0,\,-n\}$
, and $q \geq \text {max}\{0,\,n\}$
. Lastly, we observe that $a = T^{-1}(b)$
. Therefore, we have the following proposition which expresses the value of $\|g\|$
for $g = (x,\,t^n)$
in terms of representatives for $x$
constructed in the above discussion.
Proposition 2 Let $g = (x,\, t^n) \in G$
. We may express $\|g\|$
as the minimal value of

over all representations of $x$
given by

where $p \geq \text {max}\{0,\,-n\},$
$q \geq \text {max}\{0,\,n\},$
and $c_{-p-1},\,\dotsc,\, c_q$
are integers.
3. Polynomial representatives
Following [Reference Parry16, Section 4], we establish a correspondence between Laurent polynomials and elements in the Torus bundle group.
Definition 3 Given a Laurent polynomial $F(X) = \sum \nolimits _{j=-p-1}^q c_j X^j$
, we define the polynomial part of $F(X)$
to be $\sum \nolimits _{j=0}^q c_j X^j$
and the principal part to be $\sum \nolimits _{j=-p-1}^{-1} c_j X^j$
.
For $(x,\,t^n) \in G$
, if $x = F(T)(b)$
, we say $F(X)$
is a representative of $x$
or represents $x$
.
Following the notation, we define $p = \max (\{0,\,-n\}\cup \{p' \mid c_{-p'-1} \neq 0\})$
and $q = \max (\{0,\,n\}\cup \{q' \mid c_{q'} \neq 0\})$
. It follows that if $p > \max (0,\,-n)$
then $c_{-p-1} \neq 0$
, and if $q > \max (0,\,n)$
then $c_q \neq 0$
. We then define the $n$
-length of $F(X)$
to be

Note that $L_n(F(X))$
is always a non-negative integer. For an $F(X)$
representing $x$
, we say $F(X)$
is an $n$
-minimal representative of $x$
if $L_n(F(X)) = \left \|(x,\,t^n)\right \|$
.
By Proposition 2, $\| (x,\,t^n)\|$
is realized by some Laurent polynomial and we have the following theorem:
Theorem 4 Suppose that $g=(x,\,t^n)$
is an element of $G$
. Then $g$
has an $n$
-minimal representative.
In this context, the relation $b^t = (b^{-1})^{t^{-1}}b^{2k+1}$
becomes $T-(2k+1)I+T^{-1}=0$
. Hence, finding $\|g\|$
translates to minimizing $L_n$
over all polynomial representatives $F(X)$
of $x$
up to adding multiples of $(X^2-(2k+1)X+1)$
.
Observe that when given a representative for $x$

where $|c_j| > k+2$
for some $j$
, we can find another polynomial representative by adding polynomial multiples of $(X^2-(2k+1)X+1)$
which yields

where the coefficient $|c_j'| < |c_j|$
. For example, the polynomial $2X^2+(-k+2)X+4$
is preferred to $X^2+(k+3)X+3$
which is obtained by adding $(X^2-(2k+1)X+1)$
. In fact, one can use this cancellation repeatedly to deduce that large coefficients cannot be adjacent to each other too often in an $n$
-minimal polynomial. We will later call this phenomenon an instance of the global rules.
Example 5 Consider the polynomial
By direct computation, one can show that the $7$
-length is given by
By adding
we get the polynomial
Through direct computation, we have
which has a shorter $7$
-length.
More precise conditions on the coefficients will be described below.
3.1. $n$
-reduced polynomials
In this subsection, we define $n$
-reduced polynomials in $\mathbb {Z}[X]$
. Later, we show in Theorem 24 that any $n$
-minimal Laurent polynomial can be written as the sum of two $n$
-reduced polynomials, namely, the polynomial part and the principal part. Since the set of $n$
-reduced polynomials has a natural grading given by the degree of a polynomial, we will be using $n$
-reduced polynomials instead of $n$
-minimal Laurent polynomials for counting purposes. For convenience, we denote a polynomial $F(X)=\sum \nolimits _{i=0}^{m}c_i X^i$
by a string of coefficients $(c_m,\,c_{m-1},\,\dotsc,\,c_{1},\,c_0)$
which we call the word. Here indices are understood as the degree of $X$
and the entries as the coefficient of that power of $X$
. A contiguous substring of this will be called a subword of a word. We adopt the following convention of $\operatorname {sign}$
functions:

with the inequality $\operatorname {sign}^-(x+1)\leq \operatorname {sign}^+(x)$
for all $x\in \mathbb {Z}$
. We start with the following definition.
Definition 6 For a word $(c_{m},\,\dotsc,\, c_{m-i})$
, we define its potential as
In particular, when the string only contains coefficients whose absolute values are $k-2$
, $k-1$
, $k$
or $k+1$
, then the potential can be written as
where $N_t = |\{ j \: | \: c_{j} = |t| \} |$
.
By definition, the potential is additive under concatenating two strings. In order to describe the change in the potential under any rewriting, we analyse the potential change in different cases on a single coefficient which is listed out in the following lemma. Since it follows from definition, we omit the proof.
Lemma 7 Let $a$
be such that $|a|< k+2$
. We then have the following:
• $\operatorname {Pot}(a+1)=\operatorname {Pot}(a)-2\operatorname {sign}^+(a)$

• $\operatorname {Pot}(a-1)=\operatorname {Pot}(a)+2\operatorname {sign}^{-}(a)$

• $\operatorname {Pot}(a-(2k-1))=-\operatorname {Pot}(a) \quad (a>0)$

• $\operatorname {Pot}(a+(2k-1))=-\operatorname {Pot}(a) \quad (a<0)$

• $\operatorname {Pot}(a-2k)=-\operatorname {Pot}(a)-2 \quad (a>0)$

• $\operatorname {Pot}(a+2k)=-\operatorname {Pot}(a)-2 \quad (a<0).$

The potential value of an $n$
-minimal polynomial serves as a test function that determines whether a rewriting is needed. We are now ready to define $n$
-reduced polynomials.
Definition 8 Let $n$
be a fixed integer and $P(X)=(c_m,\,\dotsc,\,c_1,\,c_0)$
be a polynomial of degree $m$
. $P(X)$
is $n$
-reduced if the following rules are satisfied.

Similarly, for $F(X)\in \mathbb {Z}[X^{-1}]$
, we say $F(X)$
is $n$
-reduced if $F(X^{-1})\in \mathbb {Z}[X]$
is $n$
-reduced.
Rule 1, Rule 2 and Rule 3 are called local rules. These rules determine whether a single coefficient requires rewriting. Rule 4, Rule 5 and Rule 6 are called global rules. These determine whether a string of coefficients requires rewriting. The local rules can be considered as the degenerate cases of the corresponding global rules. Note that the local rules for a single coefficient cannot be violated at the same time, and that Rule 4 and Rule 5 cannot be violated at the same time for the same sequence of coefficients. Of course, both rules can be violated at the same time if the violations occur in different locations, and these locations may overlap.
Definition 9 For the global rules, a subword $(c_j,\,c_{j-1},\,\dotsc,\,c_{j-i})$
used in the potential condition will be called the subword associated with the rule. For the local rules, the associated subword will be understood as a single coefficient $\pm (k-1)$
, $\pm k$
, $\pm (k+1)$
, or $\pm (k+2)$
in question.
For convenience, we introduce an alternative definition of global rules.
Lemma 10 Assuming the same conditions for subwords, the potential condition for the global rules can be restated as follows:

Proof. We first consider Rule 4. When $\operatorname {sign}(c_m \cdot c_{m- i-1}) > 0$
, we have that $\operatorname {sign}^- (c_m\cdot c_{m-i-1}) = 1$
, so
Since potentials are integers, this is equivalent to $\operatorname {Pot}(c'_{m-1},\,\dotsc,\,c'_{m-i}) \geq 2$
, the potential condition stipulated in Definition 8 in this case.
Similarly, when $\operatorname {sign}(c_m \cdot c_{m- i-1}) \le 0$
, we have that $\operatorname {sign}^-(c_m\cdot c_{m-i-1}) = -1$
, so
Since potentials are integers, this is equivalent to $\operatorname {Pot}(c'_{m-1},\,\dotsc,\,c'_{m-i}) > 0$
, the potential condition stipulated in Definition 8 in this case.
The proofs for Rule 5 and Rule 6 follow in the same fashion.
Since $\operatorname {Pot}(\cdot )$
is integer valued, there is no distinction between strict inequality and inequality. This allows us to reduce the number of cases one needs to check. We will use this alternative definition when appropriate.
We note that not all $n$
-minimal polynomials are $n$
-reduced which means that we may have to rewrite the polynomial to obtain an $n$
-reduced polynomial representative. Thus, for each violation, we assign rewriting rules that do not increase $n$
-length.
Definition 11 Suppose that $P(X)=(0,\,c_m,\,\dotsc,\,c_0)$
is not $n$
-reduced. For each violation of the rules, we define the rewriting associated with the given rule as the following operations.

Since the rewriting rules may take place at the constant term $c_0$
, the resulting polynomial may have a non-zero coefficient in front of $X^{-1}$
. For example, if $F(X)=k+2$
, the rewriting will give us $F'(X)=X+(-k+1)+X^{-1}$
. However, if we start with $F(X)\in \mathbb {Z}[X]$
, after rewriting, the coefficient for $X^{-1}$
can only be $0$
or $\pm 1$
. While this may seem problematic, as the rules don't apply to a polynomial with $X^{-1}$
, we will show later that once the rewriting is done on the constant coefficient, any subword associated with rules for future violations will not include $X^{-1}$
nor the constant coefficient; thus, $X^{-1}$
part can be ignored in effect.
In this view, Example 5 can be rephrased using the established definitions above as follows: suppose that $n=7$
. For a polynomial
a subword $(k+1,\,k-1,\,k,\,k,\,k)$
is tested with Rule 6. Since
we see that Rule 6 is violated. The rewriting rule for this violation is to add another polynomial
as prescribed in the definition 11, which gives us
As demonstrated in the example, new polynomial has a shorter $7$
-length compared to the polynomial we started with. We show that this is the case with all rewriting rules.
Lemma 12 The rewriting associated with the rules will not increase the $n$
-length of the sequence.
Proof. We split this proof into two sections. The first section is devoted to the local rules and the second is devoted to global rules. For local rules, we check the $n$
-length change for each case.
(1) Local Rules
• Rule 1. The associated rewriting rule is given by
\[ (0,k+3,b,\dotsc) + (1,-2k-1,1) = (1,-k+2,b+1,\dotsc). \]The $n$
-length is changed by $3-5+\operatorname {sign}^+(b) < 0$
.• Rule 2. The associated rewriting rule is given by
\[ (\ldots,b,k+2,c,\dotsc) + (1,-2k-1,1) = (\ldots, b+1,-k+1,c+1,\dotsc). \]The $n$
-length is changed by $\operatorname {sign}^+(b) -3 +\operatorname {sign}^+(c) < 0$
.• Rule 3. The associated rewriting rule is given by
\[ (1,-k,c,\dotsc) + ({-}1,2k+1,-1) = (0, k+1,c-1,\dotsc). \]The $n$
-length is changed by $-3 +1 - \operatorname {sign}^+(c) < 0$
.
(2) Global Rules
Suppose that $c_i<0$
is a coefficient in the subword associated with the rule. We see that adding $(2k-1)$
to $c_i$
for all possible values of $c_i$
yields
\begin{align*} -k-1& \mapsto k-2\\ - k & \mapsto k-1\\ - k+1& \mapsto k\\ - k+2& \mapsto k+1, \end{align*}and for each case, the absolute value of the coefficient changes by $-3$
, $-1$
, $1$
and $3$
, respectively. We will prove the lemma for Rule 6, the proofs for other global rules are similar.
• Rule 6. Assuming $c_j < 0$
, the associated rewriting rule is given by
\begin{align*} & (c_{j+1},c_{j},\ldots,c_{l},c_{l-1})-\operatorname{sign}(c_j)({-}1,2k,2k-1,\dotsc,2k-1,2k,-1)\\ & \quad=(c_{j+1},c_{j},c_{j-1}, \ldots, c_{l+1}, c_{l}, c_{l-1})\\ & \qquad+({-}1,2k,2k-1,\dotsc,2k-1,2k, -1)\\ & \quad=({c_{j+1}-1}, {c_{j}+2k}, {c_{j-1}+2k-1}, \ldots, {c_{l+1}+2k-1}, {c_{l}+2k}, {c_{l-1}-1}) \end{align*}The $n$
-length on the first and last coordinates depend on the sign of $c_{j+1}$
and $c_{l-1}$
. For $l+1 \le s \le j-1$
, we have $c_s = -k$
or $-k+1$
, and from the above discussion, the corresponding $n$
-length change is $-1$
or $1$
, namely $\operatorname {Pot}(c_s)$
. For $s = j$
or $l$
, we have $c_s = -k$
or $-k-1$
, and $c_s+2k = k$
or $k-1$
, so the $n$
-length change is $0$
or $-2$
, namely $\operatorname {Pot}(c_s)+1$
. Thus, the total $n$
-length is changed by
\begin{align*} & -\operatorname{sign}^+(c_{j+1})-\operatorname{sign}^+(c_{l-1})+\operatorname{Pot}(c_j,\ldots, c_l)+2\\ & \quad= \operatorname{sign}^+(c_{j+1}\cdot c_j)+\operatorname{sign}^+(c_{l-1}\cdot c_j)+\operatorname{Pot}(c_j,\ldots, c_l)+2\\ & \quad< \operatorname{sign}^+(c_{j+1}\cdot c_j)+\operatorname{sign}^+(c_{l-1}\cdot c_j)\\ & \qquad+({-}2-\operatorname{sign}^+(c_{j+1}\cdot c_j)-\tfrac12\operatorname{sign}^+(c_{l-1}\cdot c_j))+2\\ & \quad= \tfrac12\operatorname{sign}^+(c_{l-1}\cdot c_j) \end{align*}As everything is an integer here, we have the total $n$
-length change must be non-positive.
In some cases, a polynomial after the rewriting will have the same $n$
-length. We demonstrate this in the example below.
Example 13 Consider the sequence $(2,\, 2,\, k,\, k,\, k,\, k,\, 2)$
. The subsequence $(k,\,k,\,k,\,k)$
has potential $-4$
, and thus violates Rule 6 as $\operatorname {Pot}(k,\,k,\,k,\,k) = -4 < -2-\operatorname {sign}^+ (2\cdot k)-$
$\tfrac 12\operatorname {sign}^+(k\cdot 2) = -3-\tfrac 12$
. The rewriting associated with this rule would be $(2,\,2,\,k,\,k,\,k,\,k,\,2) \,{-}\, (0,\, -1,\, 2k,\, 2k-1,\, 2k-1,\, 2k,\, -1)$
$= (2,\, 3,\, -k,\, -k+1,\, -k+1, k,\, 3)$
. In each coordinate, the $n$
-length is changed by $0,\, 1,\, 0,\, -1,\, -1,\, 0,\, 1$
, so the total $n$
-length is unchanged.
4. Rewriting procedures
In this section, we show how $n$
-reduced polynomials can be obtained by applying the rewriting rules listed above. As noted, while some rules are mutually exclusive, it is possible that a word violates multiple rules at the same time. Moreover, after rewriting, another rule violation may appear somewhere else. For that reason, it is not immediately clear how any polynomial can be uniquely rewritten to an $n$
-reduced polynomial.
We first establish that there is a smallest unit of rewriting called a minimal poison subword, which can be made unique by picking the subword with the lowest degree. The main idea is to first rewrite on this unique subword, then find the minimal poison subword for the new polynomial to rewrite and repeat the process. For this algorithm to work, we will need to show that subsequent minimal poison subwords are eventually away from the minimal poison subword that appeared first.
With the algorithm for $n$
-reduced polynomials established, we investigate the relationship between the $n$
-reduced polynomial and the $n$
-minimal polynomial. While there may be multiple $n$
-minimal representatives for a group element $g$
, we show that by applying the algorithm introduced earlier to both the principal part and the polynomial part, possibly multiple times, that one can get the unique $n$
- minimal representative of the polynomial.
We begin by defining the minimal poison subword.
Definition 14 If a word violates the unique rule, it is called a poison word. The shortest subword that violates the unique rule is called a minimal poison subword. A minimal poison subword that has the smallest leading degree is called the rightmost minimal poison subword.
By the natural ordering by inclusion, we know there has to be a minimal subword, which includes the case when a single coefficient violates one of the local rules. As noted, the local rules are mutually exclusive, and Rule 4 and Rule 5 cannot be tested simultaneously on the same subword. We show that the minimal poison subword can always be obtained.
Lemma 15 If a word violates multiple rules simultaneously, then there is a shorter subword in it that violates only one of the rules.
Proof. We show for Rule 6 first. Suppose that
violates Rule 6. Since a non-top rule applies, the subword can only violate either Rule 2 or Rule 6 on a strict smaller subword. If it violates Rule 2, then we pick the coefficient as the poison subword. If it violates Rule 6 on a smaller subword, we pick this subword and repeat the process until we get the minimal poison subword.
For top rules, we only show for Rule 4. Suppose that the subword
violates Rule 4. If this subword violates some other rules, the violation can only come from Rule 1, Rule 3, Rule 4 and Rule 6. If it violates any of the local rules, then we pick the coefficient in question as the poison subword as before. If it violates Rule 4 on a strictly smaller subword contained in the original, we pick the smaller subword and repeat the same argument again. Hence, it remains to show that if Rule 6 is violated, the violation occurs on a strictly smaller subword. For
we observe that $c_{m-1}$
is either $-k+1$
or $-k+2$
. Since Rule 6 requires the leading coefficient to be $-k$
or $-k-1$
, $c_{m-1}$
is not contained in the subword that violates Rule 6. By repeating this process for a strictly smaller subword, we get a poison subword. The case for Rule 5 can be worked out similarly.
There may be multiple minimal poison subwords, however, by preferring the smaller degree for the leading coefficient, the algorithm above always gives the unique minimal poison subword.
We now have established that the rightmost minimal poison subword exists and is unique. We will repeatedly rewrite the rightmost minimal poison subword to obtain the $n$
-reduced polynomial. Once a rewriting on the rightmost minimal poison subword is done, the consequent poison subword should move left in order for this process to terminate eventually. We start with a definition for two subwords for their relative positions.
Definition 16 Let $(c_i,\,\dotsc,\, c_{i-k})$
and $(c_j,\,\dotsc,\,c_{j-l})$
be two subwords that violate any of the rules. If there is no common subword that is contained in both, i.e., if $i-k > j$
or $j-l > i$
, then the two subwords are said to be disjoint. In the same manner, two rewriting rules are disjoint if their associated subwords are disjoint.
Two disjoint rules do not necessarily commute unless their associated subwords are separated by another non-trivial subword. Thus, the ordering of the rules is important. We are now ready to show this sequence of rewriting progresses to the left without backtracking. See Figure 1.
Theorem 17 Let $(s_i,\,\dotsc,\,s_{i-k})$
be the rightmost minimal poison subword of a given word. Let $(s'_i,\,\dotsc,\,s'_{i-k})$
be the subword obtained from rewriting, and let $(s'_j,\,\dotsc,\,s'_{j-l})$
be the rightmost minimal poison subword for the new word. Then $j>i$
and $j-l>i-k$
. Subsequently, rewriting rules are eventually disjoint from the initial rewriting.
Proof. We rule out all of the other possible cases using the potential of subwords. We have a number of cases based on the inclusion relationship between the subwords $(s'_i,\,\dotsc,\,s'_{i-k})$
and $(s'_j,\,\dotsc,\,s'_{j-l})$
.
(1) The subword $(s'_j,\,\dotsc,\,s'_{j-l})$
is contained in the subword $(s'_i,\,\dotsc,\,s'_{i-k})$
. This case is handled by Lemma 18.(2) The subword $(s'_i,\,\dotsc,\,s'_{i-k})$
is contained in the subword $(s'_j,\,\dotsc,\,s'_{j-l})$
. This case is handled in Lemma 19. See Figure 3.(3) The subword $(s'_j,\,\dotsc,\,s'_{j-l})$
is on the right of the subword $(s'_i,\,\dotsc,\,s'_{i-k})$
. This last case is handled in Lemma 20. See Figure 4.
Given a word (top), we perform a rewriting with respect to the minimal poison subword (in red). After rewriting, a new minimal poison subword always appears on the left (in blue).

By dividing the proof into three different cases, we have control over what rules need to be checked after the initial rewriting. For instance, in Lemma 20, we only need to check Rule 2 and Rule 6 after rewriting as these are the only non-top rules. If followed by a minimal poison subword for any of the top rules, this would fall into the case covered in Lemma 19 instead.
Throughout the proof, we will denote the subword (not necessarily poison) as a whole letter such as $A,\, B$
and $C$
. Once the rewriting has been applied, new subwords will be denoted as $A',\, B'$
and $C'$
, respectively. Instead of writing the sign of the leading coefficient and end coefficient of $A$
, we will be using $\operatorname {sign}^+(A)$
to indicate the sign of the coefficient and $\operatorname {sign}^+(A\cdot B)$
when we need to use the signs of the coefficients from two subwords; since all coefficients in the minimal poison subword have the same sign, there is no ambiguity here.
Lemma 18 Assume that the subword $(s_i,\, \ldots,\, s_{i-k})$
is the rightmost minimal poison subword of a given word. Let $(s'_i,\,\dotsc,\,s'_{i-k})$
be the subword obtained from rewriting, and let $(s'_j,\,\dotsc,\,s'_{j-l})$
be the rightmost minimal poison subword for the new word. Then the subword $(s'_j,\,\dotsc,\,s'_{j-l})$
cannot be contained in the subword $(s'_i,\,\dotsc,\,s'_{i-k})$
. Namely, we cannot have both $i \ge j$
and $j-l \ge i-k$
.
Proof. We first show that the new poison subword $(s'_j,\,\dotsc,\,s'_{j-l})$
cannot be contained in the subword $(s'_i,\,\dotsc,\,s'_{i-k})$
(see Figure 2.)
The new minimal poison subword cannot be contained in where the previous minimal poison subword was. (Lemma 18.).

We first list rules we need to check. (See Table 1.) We note that if the initial minimal poison subword violates a local rule, then the next minimal poison subword that appears after rewriting can only violate a local rule because we assumed containment. More specifically, once $(s_{i+1},\,s_i)$
violates Rule 3, we can only check $(s'_i)$
for Rule 1 and vice versa. Similarly, once $(s_i)$
violates Rule 2, we can only check $(s'_i)$
for Rule 2. However, by Definition 8, we know this is impossible. So assume that the initial minimal poison subword violates a global rule as listed in Table 1. We also point out that, because of containment, if the initial rule violation is from a non-top rule, then the next rule violation cannot be from a top rule.
For Lemma 18, we only need to check these rules.

For simplicity, denote the initial minimal poison subword as
as 3 disjoint subwords concatenated into one word. After rewriting, this subword becomes $(s'_i,\,\dotsc,\,s'_{i-k})$
, which we denote $(A'|B'|C')$
where $(s'_j,\,\dotsc,\,s'_{j-l})=(B')$
. The rewriting rule in this case is
For a top rule, subword $(A)$
is empty. Obviously, the subwords $(A)$
and $(C)$
cannot be empty at the same time, as the rules cannot be applied to the same subword twice by Definition 8.
We argue in the following way: because the subword $(A'|B'|C')$
was obtained from rewriting, there's a bound on the value $\operatorname {Pot}(A'|B'|C')$
from which comes from the value $\operatorname {Pot}(A|B|C)$
. Since the subword $(B')$
violates one of the rules, we can combine with alternate bounds on the values $\operatorname {Pot}(A')$
and $\operatorname {Pot}(C')$
to get another bound on the value $\operatorname {Pot}(A'|B'|C')$
which contradicts the previous bound on the value of $\operatorname {Pot}(A'|B'|C')$
. We proceed through our proof based on the particular rule that the subword $(A|B|C)$
violates. In each case, we assume that either $j \leq i$
or $j-l \leq i -k$
for our contradiction.
1. $(A|B|C)$
violates Rule 4 or Rule 5.
As mentioned, $(A)$
and $(C)$
cannot be both empty. We proceed based on whether $(A)$
, $(C)$
, or $(A)$
and $(C)$
are both empty.
(1) Suppose that both $(A)$
and $(C)$
are nonempty.It then follows that the word $(B')$
can only violate Rule 2 or Rule 6. Since all coefficients in $(B')$
have values either $\pm k$
or $\pm (k-1)$
, and all coefficients in $(A'|B'|C')$
have the same sign, Rule 2 cannot be violated. As for Rule 6, when $(A|B|C)$
violates Rule 4, we then have
\begin{align*} \operatorname{Pot}(A|B|C) & \leq 1+ \tfrac{1}{2}\operatorname{sign}^-(s_{i-k} \cdot s_{i-k-1}),\\ \operatorname{Pot}(A')& ={-}\operatorname{Pot}(A)-2\leq{-}4,\\ \operatorname{Pot}(C')& ={-}\operatorname{Pot}(C)-2\leq 1+\tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}),\\ \operatorname{Pot}(B') & \leq{-}4. \end{align*}This implies that
\[ \operatorname{Pot}(A'|B'|C')\leq{-}7+\tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}). \]But $\operatorname {Pot}(A'|B'|C') \geq -5- \tfrac {1}{2}\operatorname {sign}^-(s_{i-k} \cdot s_{i-k-1})$
which is impossible.
Similarly, when $(A|B|C)$
violates Rule 5 and $(B')$
violates Rule 6, we have
\begin{align*} \operatorname{Pot}(A|B|C) & \leq{-}5- \tfrac{1}{2}\operatorname{sign}^+(s_{i-k} \cdot s_{i-k-1}),\\ \operatorname{Pot}(A')& ={-}\operatorname{Pot}(A)-2\leq 3,\\ \operatorname{Pot}(C')& ={-}\operatorname{Pot}(C)-2\leq 1+\tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}),\\ \operatorname{Pot}(B') & \leq{-}4. \end{align*}Taking this all together, we may write
\[ \operatorname{Pot}(A'|B'|C')\leq \tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}). \]But $\operatorname {Pot}(A'|B'|C') \geq 1+ \tfrac {1}{2}\operatorname {sign}^+(s_{i-k} \cdot s_{i-k-1})$
which is impossible.(2) $(C)$
is empty.The above argument works identically. Assuming that $(B)$
is violating Rule 6, the potential is bounded
\begin{align*} \operatorname{Pot}(A|B) & \leq 1+ \tfrac{1}{2}\operatorname{sign}^-(s_{i-k} \cdot s_{i-k-1}),\\ \operatorname{Pot}(A')& ={-}\operatorname{Pot}(A)-2\leq{-}4,\\ \operatorname{Pot}(B') & \leq{-}3-\tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}),\\ \operatorname{Pot}(A'|B')& \leq{-}7-\tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}),\\ \operatorname{Pot}(A'|B') & \geq{-}5- \tfrac{1}{2}\operatorname{sign}^-(s_{i-k} \cdot s_{i-k-1}) \end{align*}for Rule 4, which is impossible. For Rule 5, we have that
\begin{align*} \operatorname{Pot}(A|B) & \leq{-}5- \tfrac{1}{2}\operatorname{sign}^+(s_{i-k} \cdot s_{i-k-1}),\\ \operatorname{Pot}(A')& ={-}\operatorname{Pot}(A)-2\leq 3,\\ \operatorname{Pot}(B') & \leq{-}3-\tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}),\\ \operatorname{Pot}(A'|B')& \leq{-}\tfrac{1}{2}\operatorname{sign}^+(s_{i-k}\cdot s_{i-k-1}),\\ \operatorname{Pot}(A'|B')& \geq 1+\tfrac{1}{2}\operatorname{sign}^+(s_{i-k} \cdot s_{i-k-1}). \end{align*}which is impossible. For Rule 2, we only need to consider $(B')=s'_{i-k}$
. The only possible coefficients for $s'_{i-k}$
are $\pm k$
and $\pm (k-1)$
, and $s'_{i-k+1}$
has the same sign as $s'_{i-k}$
, so $s'_{i-k}$
cannot violate Rule 2.(3) $(A)$
is empty.If $(B|C)$
violates Rule 4 then $(B')$
violates Rule 1 or Rule 5, and if $(B|C)$
violates Rule 5, then $(B')$
violates Rule 3 or Rule 4. Suppose that $(B|C)$
violates Rule 4. Since $(s'_{i})$
is either $\pm (k+1)$
or $\pm (k+1)$
with $(s'_{i-1})$
having the same sign as $(s'_{i})$
, Rule 1 cannot be violated. As for Rule 5, we check the potential
\begin{align*} \operatorname{Pot}(B|C) & \leq 1+ \tfrac{1}{2}\operatorname{sign}^-(s_{i-k} \cdot s_{i-k-1}),\\ \operatorname{Pot}(C')& \geq{-}3-\tfrac{1}{2}\operatorname{sign}^+(s'_{i-k}\cdot s'_{i-k-1}),\\ \operatorname{Pot}(C)& \geq 1+\tfrac{1}{2}\operatorname{sign}^+(s'_{i-k}\cdot s'_{i-k-1}),\\ \operatorname{Pot}(B') & \leq{-}5,\\ \operatorname{Pot}(B) & \geq 3,\\ \operatorname{Pot}(B|C)& \geq 4+\tfrac{1}{2}\operatorname{sign}^+(s'_{i-k}\cdot s'_{i-k-1}). \end{align*}which is impossible. Now suppose that $(B|C)$
violates Rule 5. For Rule 3, $(B)=s'_{i}$
is either $\pm (k-1)$
or $\pm (k-2)$
, with $s'_{i}$
having the same sign as $s'_{i}$
, so Rule 3 cannot be violated. For Rule 4, We have that
\begin{align*} \operatorname{Pot}(B|C) & \leq{-}5- \tfrac{1}{2}\operatorname{sign}^+(s_{i-k} \cdot s_{i-k-1}),\\ \operatorname{Pot}(C')& \geq{-}3-\tfrac{1}{2}\operatorname{sign}^+(s'_{i-k}\cdot s'_{i-k-1}),\\ \operatorname{Pot}(C)& \geq 1+\tfrac{1}{2}\operatorname{sign}^+(s'_{i-k}\cdot s'_{i-k-1}),\\ \operatorname{Pot}(B') & \leq 1,\\ \operatorname{Pot}(B) & \geq{-}3,\\ \operatorname{Pot}(B|C)& \geq{-}2+\tfrac{1}{2}\operatorname{sign}^+(s'_{i-k}\cdot s'_{i-k-1}) \end{align*}which is impossible.
2. The subword $(A|B|C)$
violates Rule 6
First, we can rule out Rule 2 entirely, as the coefficients appearing in $(A')$
, $(B')$
and $(C')$
are either $\pm k$
or $\pm (k-1)$
, with at least one neighbouring coefficient having the same sign. So suppose that $(B')$
violates Rule 6. We proceed based on whether $A$
, $C$
or $A$
and $C$
are empty.
(1) The subwords $(A)$
and $(C)$
are both nonempty.Recall that the subword $(s_i,\,\dotsc,\,s_{i-k})$
satisfies Rule 6 if
\[ \operatorname{Pot}(s_i,\dotsc,s_{i-k}) \geq{-}2- \operatorname{sign}^+(s_i \cdot s_{i+1})- \tfrac{1}{2}\operatorname{sign}^+(s_i \cdot s_{i-k-1}). \]Since the subword $(s_i,\,\dotsc,\,s_{i-k})=(A|B|C)$
violates Rule 6, we instead have
\[ \operatorname{Pot}(A|B|C) \leq{-}2- \operatorname{sign}^+(s_i \cdot s_{i+1})-\tfrac{1}{2}\operatorname{sign}^+(s_i \cdot s_{i-k-1}). \]Since the subword $(A|B|C)$
is minimal, none of the subwords inside $(A|B|C)$
violate any rules which includes Rule 6. Thus, we have
\begin{align*} \operatorname{Pot}(A) & \geq{-}2- \operatorname{sign}^+(s_i \cdot s_{i+1})- \tfrac{1}{2}\operatorname{sign}^+(A \cdot B),\\ \operatorname{Pot}(A) & \geq{-}2- \operatorname{sign}^+(s_i \cdot s_{i+1}),\\ \operatorname{Pot}(C) & \geq{-}2- \operatorname{sign}^+(B \cdot C)-\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}),\\ \operatorname{Pot}(C) & \geq{-}3-\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}). \end{align*}After rewriting, the subword $(B')$
is the rightmost minimal poison subword violating Rule 6. Therefore, we have
\[ \operatorname{Pot}(B') \leq{-}2- \operatorname{sign}^+(A'\cdot B')- \operatorname{sign}^+(B'\cdot C')={-}4. \]Because $(A'|B'|C')$
was obtained from $(A|B|C)$
by rewriting, we have
\begin{align*} \operatorname{Pot}(A')& ={-}\operatorname{Pot}(A)-2\leq \operatorname{sign}^+(s_i \cdot s_{i+1}),\\ \operatorname{Pot}(C')& ={-}\operatorname{Pot}(C)-2\leq 1+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}). \end{align*}Hence, we may write the value $\operatorname {Pot}(A'|B'|C')$
as
\[ -\operatorname{Pot}(A|B|C)-4\geq{-}2+\operatorname{sign}^+(s_i \cdot s_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i \cdot s_{i-k-1}). \]But $\operatorname {Pot}(B')\leq -4$
. Therefore,
\[ \operatorname{Pot}(A'|B'|C')\leq{-}3+\operatorname{sign}^+(s_i \cdot s_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}) \]which is a contradiction.
(2) $(A)$
is empty.We may write
\begin{align*} \operatorname{Pot}(B')& \leq{-}3-\operatorname{sign}^+(s'_i \cdot s'_{i+1}),\\ \operatorname{Pot}(C')& \leq 1+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}) \end{align*}Therefore, we may express $\operatorname {Pot}(B'|C')$
as
\[ -\operatorname{Pot}(B|C)-4\geq{-}2+\operatorname{sign}^+(s_i \cdot s_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}). \]However, since $\operatorname {sign}^-(s_i \cdot s'_{i+1})\leq \operatorname {sign}^+(s_i \cdot s_{i+1})$
, we have that
\begin{align*} \operatorname{Pot}(B'|C')& \leq{-}2-\operatorname{sign}^+(s'_i \cdot s'_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}),\\ & ={-}2+\operatorname{sign}^-(s_i \cdot s'_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}),\\ & \leq{-}2+\operatorname{sign}^+(s_i \cdot s_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}),\\ & \leq \operatorname{Pot}(B'|C'). \end{align*}Thus, the equality can only happen when $\operatorname {sign}^+(s_i\cdot s_{i-k-1})=0$
which is impossible.(3) $(C)$
is empty.We then have
\begin{align*} \operatorname{Pot}(A')& ={-}\operatorname{Pot}(A)-2\leq \operatorname{sign}^+(s_i \cdot s_{i+1}),\\ \operatorname{Pot}(B')& \leq{-}3+\tfrac{1}{2}\operatorname{sign}^+(s'_i\cdot s'_{i-k-1}). \end{align*}That implies we may write $\operatorname {Pot}(A'|B')$
as
\[ -\operatorname{Pot}(A|B)-4\geq{-}2+\operatorname{sign}^+(s_i \cdot s_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1}). \]Hence, we may further write
\begin{align*} \operatorname{Pot}(A'|B')& \leq{-}3+\operatorname{sign}^+(s_i \cdot s_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s'_i\cdot s'_{i-k-1}),\\ & \leq{-}3+\operatorname{sign}^+(s_i \cdot s_{i+1})+\tfrac{1}{2}\operatorname{sign}^+(s_i\cdot s_{i-k-1})<\operatorname{Pot}(A'|B'). \end{align*}
Lemma 19 Assume that $(s_i,\, \ldots,\, s_{i-k})$
is the rightmost minimal poison subword of a given word. Let $(s'_i,\,\dotsc,\,s'_{i-k})$
be the subword obtained from rewriting, and let $(s'_j,\,\dotsc,\,s'_{j-l})$
be the rightmost minimal poison subword for the new word. Then the subword $(s'_i,\,\dotsc,\,s'_{i-k})$
cannot be contained in the subword $(s'_j,\,\dotsc,\,s'_{j-l})$
. Namely, we cannot have both $j \ge i$
and $i-k \ge j-l$
.
Proof. Again, we begin with listing rules to be checked. (See Table 2.) Because we have already checked the cases when the second violation is from a local rule in Lemma 18, we only need to check global rules for the second violation.
Suppose $(B)$
violates Rule 1. Then $(A)$
is empty and $(B')=s'_{m-1}$
is either $\pm (k-1)$
or $\pm (k-2)$
. Since $(B)$
is assumed to be the rightmost minimal poison, $(C)$
does not violate Rule 6. By Lemma 10, we have
Since $(C')$
is obtained by adding $\operatorname {sign}(B)$
to $s_{i-k-1}$
,
Since $(B'|C')$
violates Rule 4, we have
However, $\operatorname {Pot}(s'_{m-1})>0$
, so this is impossible. The case when $(A'|B'|C')$
violates the Rule 4 can be worked out similarly by using a lower bound for $\operatorname {Pot}(B')$
which is obtained from an upper bound for $\operatorname {Pot}(B)$
.
Now suppose $(B)$
violates Rule 3. Then $(A)$
is empty and $(B')=s'_{m-1}$
is either $\pm k$
or $\pm (k-1)$
. Since $(C)$
does not violate Rule 6, just like before, we have
Since $(B'|C')$
violates Rule 5, we have
However, $\operatorname {Pot}(s'_{m-1})>-2$
, so this is impossible. The case when $(A'|B'|C')$
violates the Rule 5 can be worked out similarly, again by using a lower bound for $\operatorname {Pot}(B')$
which is obtained from an upper bound for $\operatorname {Pot}(B)$
.
Suppose both $(B)$
and $(A'|B'|C')$
violate Rule 6, or $(B)$
violates Rule 2 and $(A'|B'|C')$
violate Rule 6. Since $(B)$
is the rightmost, we have that
Therefore,
On the other hand, since $(A'|B'|C')$
violates Rule 6, we have that
This gives us a new bound $\operatorname {Pot}(A'|B')\leq -3-\operatorname {sign}^+(s_{j+1}\cdot A')$
which implies that $(A'|B')$
violates Rule 6, contradicting that $(A'|B'|C')$
is minimal poison. Using the same bounds for $\operatorname {Pot}(C')$
, we also get a contradiction for $(A'|B'|C')$
violating Rule 4 or Rule 5.
Lemma 20 Assume that $(s_i,\, \ldots,\, s_{i-k})$
is the rightmost minimal poison subword of a given word. Let $(s'_i,\,\dotsc,\,s'_{i-k})$
be the subword obtained from rewriting, and let $(s'_j,\,\dotsc,\,s'_{j-l})$
be the rightmost minimal poison subword for the new word. Assume that $(s'_i,\,\dotsc,\,s'_{i-k})$
and $(s'_j,\,\dotsc,\,s'_{j-l})$
intersect such that the first word does not contain the second word and vice versa. Then $j>i$
and $j-l>i-k$
.
Proof. We begin with listing rules to be checked. (See Table 3.) By Lemma 18 and 19, regardless of the initial rule violation, the next rule violation can only be from Rule 2 or Rule 6.
Base case for types and classes.

Inductive case for types and classes.

Suppose that $(s_i,\, \ldots,\, s_{i-k})$
violates any of the rules. By Lemma 18, the minimal poison subword violating Rule 2 after rewriting must be disjoint from $(s_i,\, \ldots,\, s_{i-k})$
, $(s'_{i-k-1})$
is the minimal poison subword in this case. $(s_{i-k-1})$
is not a poison subword, so by Definition 8,
depending on $\operatorname {sign}(s_{i-k}s_{i-k-1})$
and $\operatorname {sign}(s_{i-k-1}s_{i-k-2})$
. $(s'_{i-k-1})$
violates Rule 2, so
depending on
and
But this is impossible, since $s'_{i-k-1}=s_{i-k-1}+\operatorname {sign}(s_{i-k})$
.
Now we have ruled out all local rules. Suppose $i\geq j\geq i-k\geq j-l$
. Let the intersection be $(B')$
, $(s'_i,\,\dotsc,\,s'_{i-k})=(A'|B')$
, and $(s'_j,\,\dotsc,\,s'_{j-l})=(B'|C')$
is the minimal poison subword violating Rule 6. We show $(B')$
is empty and use the fact that $(A|B)$
is the rightmost minimal poison subword to get a contradiction. In this case, the rewriting is given by
Note that $(A)$
and $(B)$
have the same sign and $(B')$
and $(C')$
have the same sign, since $(A|B)$
and $(B'|C')$
violate some rules. Also, $(B)$
and $(B')$
have opposite signs because of the rewriting. Thus, all coefficients in $(A)$
and $(B)$
have the same sign with all coefficients in $(C)$
the opposite sign.
$(A|B)$
violates Rule 6. Thus,

By combining the inequalities for $\operatorname {Pot}(A')$
and $\operatorname {Pot}(A'|B')$
, we get
Hence, combining the inequalities for $\operatorname {Pot}(B')$
and $\operatorname {Pot}(B'|C')$
allows us to write
However, it then follows that $\operatorname {Pot}(C)\leq -3-\tfrac {1}{2}\operatorname {sign}^+(s'_{j-l}\cdot s_{j-l-1})$
which contradicts the previous bound on $\operatorname {Pot}(C)$
. Thus, $(B)$
is empty. Similarly, we can show that $(B)$
is empty assuming $(A|B)$
violates either Rule 4 or Rule 5.
Since the subword $(B)$
is empty, two consecutive poison subwords are disjoint, and thus, we can directly compare the potentials. $(C')$
is a minimal poison subword. Therefore, we have that $\operatorname {Pot}(C')\leq -2-\operatorname {sign}^+(c_{j}\cdot c_{j+1})-\tfrac {1}{2}\operatorname {sign}^+(c_{j-l}\cdot c_{j-l-1})$
. Because $(A)$
and $(C)$
has opposite sign, the rewriting rule suggests that $\operatorname {Pot}(C)=\operatorname {Pot}(C')-2$
, which means that $(C)$
already violates Rule 6 contradicting the fact $(A)$
is the rightmost minimal poison subword.
We point out that the proof of Lemma 18 does not require the rightmost minimal poison subword. The purpose of choosing the rightmost one is to prevent the next minimal poison subwords from appearing at other places. If there is only one minimal poison subword to start with, Lemma 19 and Lemma 20 can be modified to give us the following corollary.
Corollary 21 If a word has only one minimal poison subword, then all consecutive rewriting rules are disjoint. In particular, if $P$
is an $n$
-reduced polynomial, then all consecutive rewriting rules for $P+1$
are disjoint.
If the only minimal poison subword comes from Rule 4 or Rule 5, then the next minimal poison subword is either contained or contains the initial minimal poison subword. By Lemma 18 and Lemma 19, we know this cannot happen. If the minimal poison subword violates Rule 6, then this follows from Lemma 20 applied to the left side instead of the right side.
We now show that there exists an $n$
-minimal representative of a given element $g\in G$
whose polynomial part and principal part are $n$
-reduced. We start with the following lemma.
Lemma 22 Suppose $F(X) = \sum \nolimits _{i= \ell }^m c_i X^i$
is an $n$
-minimal representative of $g=(x,\,t^n)$
. Then polynomial and principal parts of $F(X)$
satisfy Rule 1, Rule 2, and Rule 3.
Proof. If not, by adding or subtracting $X^d(X^2-(2k+1)X+1)$
with some appropriate choice of $d$
, one can obtain a new Laurent polynomial. By Lemma 12, this new Laurent polynomial has strictly less $n$
-length, contradicting the fact $F(X)$
is an $n$
-minimal representative of $g$
.
With local rules satisfied, we now prescribe how $n$
-minimal Laurent polynomials can be rewritten to have $n$
-reduced polynomial part.
Proposition 23 Let $F(X) = \sum \nolimits _{i=\ell }^m c_i X^i$
be an $n$
-minimal representative of $g=(x,\,t^n)$
. Let $r$
be an integer, and suppose that $\ell < 0 \leq r \leq m$
. Then there exists a deterministic process that takes $F(X)$
to a Laurent polynomial $G(X)$
given by

such that $G(X)$
is an $n$
-minimal representative of $g$
and where $\sum \nolimits _{i=r}^m B_i X^i$
is $n$
-reduced. Moreover, we have that $B_{r-1} \in \{-1,\, 0,\, 1\}$
, and if $B_{r-1} = \pm 1$
, then $B_{r} B_{r-1} < 0$
and $c_{r-1} B_{r} \geq 0$
. We have that $B_r \in \{ c_r,\, c_r \pm 1,\, c_r \pm 2k,\, c_r \pm (2k+1)\}$
. Finally, $M = m$
or $M = m \pm 1$
.
Proof. We induct on $r$
reversely, i.e., we apply the rewriting procedure as in Theorem 17 on $F(X)$
inductively from a higher degree. The base case in our case for the induction is when $r=m$
. Consider $c_m X^m$
. Only local rules apply here, and by Lemma 22, $c_m X^m$
is $n$
-reduced. Thus, $c_m=B_M$
and $m=M$
. For the step case, consider
If it is $n$
-reduced, we are done and $m=M$
. If not, Lemma 22 implies that $F_r(X)$
cannot violate local rules. Thus, we need only consider global rules. As in Theorem 17, we start rewriting from the rightmost minimal poison subword of $(c_m,\,\dotsc,\,c_r)$
. We note that since $F_{r+1}$
is already $n$
-reduced, the rightmost minimal poison subword must contain either $c_r$
or $c_{r+1}$
. If it contains $c_r$
, then $B_{r-1} \in \{-1,\, 1\}$
and $B_r \in \{c_r \pm 2k,\, c_r \pm (2k+1)\}$
, and $B_{r-1}=0$
, $c'_r= c_r \pm 1$
otherwise. Finally, $M=m+1$
or $M=m-1$
depending on whether $c_m$
is contained in the rightmost minimal poison subword in the sequence of rewriting rules. The sequence of rewriting rules terminates as all rewriting rules after the initial rewriting are eventually disjoint from $(c_r)$
.
Theorem 24 Suppose that $g= (x,\,t^n)$
is an element of $G$
. Then $x$
has an $n$
-minimal representative whose polynomial and principal part are both $n$
-reduced.
We show that we can modify any $n$
-minimal representative of $x$
until the desired form is achieved. By Lemma 22, we have any $n$
-minimal representative must satisfy Rule 1, Rule 2 and Rule 3. We will apply Proposition 23 with $r = 0$
to the polynomial part. Furthermore, by the symmetry of the polynomial and principal parts, we can apply Proposition 23 to the principal part by applying it to $X^{-1}\cdot F(X^{-1})$
. We also need the following proposition.
Proposition 25 Let $n \in \mathbb {Z}$
.
(1) If $P(X) = \sum \nolimits _{i=0}^m c_i X^i$
is $n$
-reduced, then so is $\sum \nolimits _{i=\ell }^m c_i X^i$
.(2) Let $P(X) \in \mathbb {Z}[X]$
. Then $P(X)$
is $n$
-reduced if and only if $XP(X)$
is $(n+1)$
-reduced.
Proof. Let $P_\ell (X) = \sum \nolimits _{i=\ell }^m c_i X^k$
. We claim that $P_\ell$
is $n$
-reduced, and towards that end, we may represent $P_\ell$
as the string
where there are $\ell -1$
zero's at the end of the string. Since $P$
is $n$
-reduced and $(c_m,\,\dotsc,\, c_\ell )$
is a substring of $(c_m,\,\dotsc,\, c_0)$
, it is easy to see that Rule 1, Rule 2, and Rule 3 are satisfied for coefficients with indices between $\ell$
and $m$
. Since the coefficients of $P_\ell$
with index less than $\ell$
are zero, we have that the Rule 1, Rule 2, and Rule 3 are all satisfied.
Let $m \geq t \geq i \geq \ell$
, and suppose that $Q = (c_t,\,\dotsc,\, c_i)$
is a poison subword with respect to either Rule 4, Rule 5, or Rule 6. If $i>\ell$
, then $(c_t,\,\dotsc,\, c_i)$
is a poison subword of $P$
with respect to Rule 4, Rule 5, or Rule 6 which is a contradiction. Thus, if $P_\ell$
contains a poison subword, it must contain the coefficient $c_\ell$
.
We proceed by contradiction to demonstrate $(c_m,\, \dotsc,\, c_\ell )$
satisfies Rule 4, Rule 5, and Rule 6. We split this argument based on the rule we are trying to verify.
(1) Rule 4.
If $(c_m,\,\dotsc,\, c_\ell )$
violates Rule 4, we then would have that
\[ (c_m,\dotsc, c_\ell) ={\pm}(1, c_{m-1}^\prime,\dotsc, c^\prime_\ell) \]where $c_{m-j}^\prime \in \{-k+1,\,-k\}$
for $0\leq j< m-\ell$
and $c_\ell ^\prime \in \{-k,\,-k-1\}$
. We have that $\operatorname {sign}(c_m \cdot c_{\ell -1}) = \operatorname {sign}(c_m \cdot 0) = \operatorname {sign}(0)$
. We then have
\[ \text{Pot}(c_m,\dotsc,c_\ell) \leq 0. \]However, since $(c_m,\,\dotsc,\, c_\ell )$
is a substring of $P$
, we have that $(c_m,\,\dotsc,\,c_\ell )$
satisfies Rule 4. In particular, if $\operatorname {sign}(c_m \cdot c_\ell ) > 0$
, then $\text {Pot}(c_m,\,\dotsc,\, c_\ell ) \geq 2$
which is a contradiction. If $\operatorname {sign}(c_m \cdot c_{\ell -1}) \leq 0$
, then $\text {Pot}(c_m,\,\dotsc,\, c_\ell ) > 0$
which is also a contradiction. Thus, $P_\ell$
satisfies Rule 4.(2) Rule 5.
If $(c_m,\,\dotsc,\, c_\ell )$
violates Rule 5, we would have that $c_t \in \{k+1,\, k+2\}$
where $c_s \in \pm \{k-1,\,k\}$
for $s \neq \ell$
and $c_{\ell } \in \{k,\, k+1\}$
. Since $c_{\ell -1} = 0$
, we have that $\operatorname {sign}(c_m \cdot c_{\ell -1}) = 0$
. Thus, we have that
\[ \operatorname{Pot}(c_m,\dotsc, c_\ell) <{-}6. \]Since $(c_m,\,\dotsc,\, c_\ell )$
is a substring of $(c_m,\,\dotsc,\, c_0)$
, we have that $(c_m,\,\dotsc,\, c_\ell )$
satisfies Rule 5. Hence, if $\operatorname {sign}(c_m \cdot c_{i-1}) \geq 0$
, then $\text {Pot}(c_m,\,\dotsc,\, c_\ell ) > -6$
which is a contradiction. If $\operatorname {sign}(c_m \cdot c_{\ell -1}) < 0$
, then $\text {Pot}(c_t,\,\dotsc,\, c_\ell ) > -4$
which is also a contradiction. Thus, $P_\ell$
must satisfy Rule 5.(3) Rule 6.
If $(c_t,\,\dotsc,\, c_\ell )$
violates Rule 6, we would have that $c_s \in \{k,\,k+1\}$
for $s \in \{t,\,\ell \}$
and $c_\ell \in \pm \{k-1,\,k\}$
otherwise. By definition of Rule 6, we have that
\[ \operatorname{Pot}(c_t,\dotsc, c_{\ell}) \leq{-}3 - \operatorname{sign}^+(c_t \cdot c_{t+1}). \]However, $(c_t,\,\dotsc,\, c_\ell )$
is a substring of $P$
which implies that it satisfies Rule 6. If $\operatorname {sign}(c_t \cdot c_{\ell }) \geq 0$
, then $\text {Pot}(c_j,\, \ldots,\, c_\ell ) > - 3 - \operatorname {sign}^+(c_t \cdot c_{t+1})$
which is a contradiction. If $\operatorname {sign}(c_t \cdot c_{\ell -1}) < 0$
, then $\text {Pot}(c_t,\,\dotsc,\, c_{\ell }) \geq -1 - \operatorname {sign}^+(c_t \cdot c_{t+1})$
which is also a contradiction. Thus, $P_\ell$
must satisfy Rule 6.
For the second statement, we may proceed using similar arguments as for the first statement.
Proof of Theorem 24. Suppose that $F(X)$
represents $g=(x,\,t^n)$
with both non-trivial polynomial and principal parts. We first apply Proposition 23 to the principal part of $F(X)$
(namely, apply it to $X^{-1}\cdot F(X^{-1})$
, we obtain $F'(X)=\sum \nolimits _{i=l<0}^{M} B_i X^i$
which represents $g$
. By applying Proposition 23 to the polynomial part, we obtain

If $C_{-1}=0$
, then we are done. If not, we have $C_0 C_{-1}<0$
and $C_0 B_{-1}\geq 0$
. Observe by Proposition 25 that $\sum \nolimits _{i = \ell }^{-2} B_i X^i$
is $n$
-reduced. By Corollary 21, we have that all rewriting rules for the principal parts are disjoint from the polynomial part except $C_0$
. Rewriting using Proposition 23 on the principal part, we obtain a Laurent polynomial that is $n$
-reduced on the principal part, and $C_0 C_{-1}<0$
guarantees that the polynomial part is $n$
-reduced as well.
5. Stability of $n$
-reduced polynomials
In this section, we attempt to count the number of all $n$
-reduced polynomials. Intuitively, we start with $P=0$
which is trivially $n$
-reduced and keep adding 1 repeatedly until we fail to have an $n$
-reduced polynomial. By quantifying this failure, we classify all $n$
-reduced polynomials with a non-negative leading coefficient. We start with a basic lemma.
Lemma 26 Suppose that $P=(\ldots,\,c_1,\,c_0)$
is an $n$
-reduced polynomial. Then $P+1$
is $n$
-reduced or the rightmost minimal poison subword of $P+1$
contains either $c_1$
or $c_0+1$
.
Proof. This is clear from the fact that for any subword not containing $c_1$
and $c_0$
, there is an identical subword in $P$
. Since the subword follows all of the local and global rules applied on $P$
, it never violates any of the rules. If there is no minimal poison subword of $P+1$
containing any of the coefficients, then $P+1$
is $n$
-reduced.
For each case, we define the following.
Definition 27 Suppose that $P=(\ldots,\,c_1,\,c_0)$
is an $n$
-reduced polynomial where $P+1$
is not $n$
-reduced.
• If the rightmost minimal poison subword contains $c_1$
but not $c_0$
, we say $P+1$
fails to be $n$
-reduced by a sign change violation.• If the rightmost minimal poison subword contains $c_0+1$
, we say $P+1$
fails to be $n$
-reduced by a potential value change violation.
Suppose that $P$
is $n$
-reduced, but $P+1$
is not. This next proposition classifies all possibilities for $P$
based on whether we have a sign change violation or a potential value change and what rule $P+1$
violates.
Proposition 28 Suppose that $P$
is an $n$
-reduced polynomial of degree $m$
and $P+1$
is not $n$
-reduced. Then, $P$
falls into one of these categories:
• Sign change violations
(1) $P+1$
violates Rule 1.In this case, $P=(-k-2,\,0)$
and $n\leq 1$
.(2) $P+1$
violates Rule 2.
(i) $P=(-k-1,\,0)$
and $n>1$
.(ii) $P=(\ldots,\,c_2,\,-k-1,\,0)$
and $c_2<0$
.(iii) $P=(\ldots,\,c_2,\,-k,\,0)$
and $c_2\geq 0$
.
(3) $P+1$
violates Rule 3.In this case, $P=(1,\,-k+1,\,0)$
and $n\leq 1$
.(4) $P+1$
violates Rule 4.For $P=(1,\,c_{m-1},\,\dotsc,\,c_1,\,0)$
, $m>n$
, $c_1=-k$
and
\[ \operatorname{Pot}(c_{m-1},\dotsc,c_1)=1. \]
(5) $P+1$
violates Rule 5.For $P = (c_m,\,\dotsc,\, c_1,\, 0)$
, $m\geq n$
, $c_1=-k$
and
\[ \operatorname{Pot}(c_m,\dotsc, c_1)={-}5. \]
(6) If $P+1$
violates Rule 6.Let $P = (\ldots |A|0)$
and $(A)=(c_j,\,\dotsc,\, c_1)$
be the minimal poison subword of $P+1$
. Then, $c_1=-k$
and
\[ \operatorname{Pot}(A)={-}2-\operatorname{sign}^+(c_j \cdot c_{j+1}). \]
• Potential value change violations
(1) $P+1$
violates Rule 1.textrmin this case, $P=k+2$
and $n\leq 0$
.(2) $P+1$
violates Rule 2.
(i) $P=k+1$
, $n>0.$
(ii) $P=(\ldots,\,c_1,\,k+1)$
, $c_1\geq 0.$
(iii) $P=(\ldots,\,c_1,\,k)$
, $c_1<0.$
(3) $P+1$
violates Rule 3.textrmin this case, $n\leq 0$
, $P=(-1,\,k-1)$
.(4) $P+1$
violates Rule 4.For $P=(-1,\,c_{m-1},\,\dotsc,\, c_0)$
, $m>n$
(i) $c_0=k$
and
\[ \operatorname{Pot}(c_{m-1},\dotsc,c_1)\in \{2,3\}. \]
(ii) $c_0=k-1$
and
\[ \operatorname{Pot}(c_{m-1},\dotsc,c_1)=1. \]
(5) If $P+1$
violates Rule 5.For $P = (c_m,\,\dotsc,\, c_1,\, c_0)$
, $m\geq n$
(i) $c_0=k-1$
and
\[ \operatorname{Pot}(c_m,\dotsc, c_1) ={-}5. \]
(ii) $c_0=k$
and
\[ \operatorname{Pot}(c_m,\dotsc, c_1) \in \{{-}4, -3\}. \]
(6) If $P+1$
violates Rule 6.Let $(A|c_0)=(c_j,\,\dotsc,\, c_1,\, c_0)$
be the minimal poison subword of $P+1$
. Then,
(i) $c_0=k-1$
and
\[ \operatorname{Pot}(A) ={-}2 - \operatorname{sign}^+(c_j \cdot c_{j+1}). \]
(ii) $c_0=k$
and
\[ \operatorname{Pot}(A) \in \{{-}1 - \operatorname{sign}^+(c_j \cdot c_{j+1}),-\operatorname{sign}^+(c_j \cdot c_{j+1})\}. \]
Proof. We work out the case when the minimal poison subword for $P+1$
violates Rule 6. Denote the minimal poison subword as $(c_j,\,\dotsc,\,c_{\ell })$
Recall that Rule 6 is violated if
By Lemma 26, either $\ell =0$
or $\ell =1$
.
(1) $\ell =1$
. In this case, $P+1$
contains a subword $(c_j,\,\dotsc,\,c_{1},\,c_0+1)$
. From $P$
, the potential of the subword $(c_j,\,c_{j-1},\,\dotsc,\,c_{1})$
is
\[ \operatorname{Pot}(c_j,c_{j-1},\dotsc,c_{1}) \geq{-}2- \operatorname{sign}^+(c_j\cdot c_{j+1})- \tfrac{1}{2}\operatorname{sign}^+(c_j \cdot (c_{0})) \]but since it is the minimal poison subword of $P+1$
, we also have
\[ \operatorname{Pot}(c_j,c_{j-1},\dotsc,c_{1}) <{-}2- \operatorname{sign}^+(c_j\cdot c_{j+1})- \tfrac{1}{2}\operatorname{sign}^+(c_j \cdot (c_{0}+1)) \]which can happen only when $c_0=0$
and $c_1<0$
. Furthermore, we have $c_1=-k$
or $-k-1$
. (Note $c_1=-k-1$
case is already included in Rule 2 violation.)(2) $\ell =0$
. Again, $P+1$
contains a subword $(c_j,\,\dotsc,\,c_{1},\,c_0+1)$
. From $P$
, the potential of the subword $(c_j,\,c_{j-1},\,\dotsc,\,c_{1},\,c_0)$
is
\[ \operatorname{Pot}(c_j,c_{j-1},\dotsc,c_{1},c_0) \geq{-}3- \operatorname{sign}^+(c_j\cdot c_{j+1}) \]because $c_{-1}=0$
. But since it is minimal poison subword of $P+1$
, we also have
\[ \operatorname{Pot}(c_j,c_{j-1},\dotsc,c_{1},c_0+1) <{-}3- \operatorname{sign}^+(c_j\cdot c_{j+1}) \]which happens only when $c_0>0$
, that is when adding 1 to the polynomial decreases the potential by 2. If $c_0=k-1$
, then the only possible value of the potential for the subword without $c_0$
is
\[ \operatorname{Pot}(c_j,c_{j-1},\dotsc,c_{1}) ={-}2 - \operatorname{sign}^+(c_j \cdot c_{j+1}), \]whereas if $c_0=k$
,
\[ \operatorname{Pot}(c_j,c_{j-1},\dotsc,c_{1}) \in \{{-}1 - \operatorname{sign}^+(c_j \cdot c_{j+1}),-\operatorname{sign}^+(c_j \cdot c_{j+1})\}. \]
Other cases can be worked out similarly.
We note there is a correspondence in these classifications which we shall exploit to reduce the number of cases we consider. We list these by providing a lemma.
Lemma 29 For an $(n+1)$
-reduced polynomial $P=XQ$
, if $P+1$
is not $(n+1)$
-reduced by a sign change violation for some global rule, then $-Q+1$
is not $n$
-reduced by a potential value change violation for the same rule.
Proof. This can be done by simply comparing the conditions for each sign change violation in Proposition 28. Start with an $(n+1)$
-reduced polynomial
and suppose that $P+1$
violates Rule 4. The condition states that
This implies that
which is precisely the condition for potential value change violation by Rule 4. For Rule 5, this case corresponds to 5) (ii) with potential -4, and for Rule 6, this corresponds to 6) (ii) with $Pot(A)=-1 - \operatorname {sign}^+(c_j \cdot c_{j+1})$
.
Therefore, in the later section, we will only select a few cases listed here to generate the entire list of $n$
-reduced polynomials. In addition, once rewriting is triggered by a violation, since all rewriting rules for $P+1$
are disjoint (see Corollary 21), if there were to be another rewriting, the poison subword would fall into one of these categories as well. This allows us to break $P$
into smaller pieces which we call $n$
-types and $n$
-classes.
6. $n$
-Types and $n$
-classes
In this section, we give the precise definition of $n$
-types and $n$
-classes. $n$
-types and $n$
-classes come from a list of cases from Proposition 28 and a list of generic cases, that is, the case when both $P$
and $P+1$
are $n$
-reduced. Intuitively, $n$
-types tells us whether $P+1$
is $n$
-reduced or not when $P$
is $n$
-reduced; if it fails to be $n$
-reduced, it will be subdivided into $n$
-classes using Proposition 28. $n$
-classes serve another purpose here: they capture whether $XP+c$
is $(n+1)$
-reduced or not when $P$
is $n$
-reduced. Since $XP$
is always $(n+1)$
-reduced by Proposition 25, the case when $XP+c$
is $(n+1)$
-reduced can be classified using Proposition 28. More specifically, if $c=1$
, then we can use the sign change violation; if $c=k$
or $k+1$
, then we can use the potential change violation. The main reason we need both $n$
-types and $n$
-classes is that $n$
-length is not the length that we are interested when computing the growth series for the group, rather, we need to count all $n$
-reduced polynomials for all possible integer $n$
. $n$
-classes provides us a way to increment $n$
from $0$
-reduced polynomial.
As mentioned after Proposition 28, there is a correspondence between sign change violations and potential value change violations, hence there is a reduction in cases. Here, we will primarily select polynomials whose leading coefficient is non-negative. Thus, we consider potential value change violation for Rule 5 and Rule 6, and the sign change violations by Rule 4. Rule 2 will be considered as a degenerate case of Rule 6. Rule 1 and Rule 3 have only a finite number of cases, so those cases will appear in the base case, but not in the inductive definition of the $n$
-types and $n$
-classes.
We start with the generic case. If both $P$
and $P+1$
are $n$
-reduced and the constant coefficient of $P$
is non-negative, we say $P$
has $n$
-interior type. If the constant coefficient of $P$
is negative, then it follows that $P+1$
is also $n$
-reduced, and we say $P$
has $n$
-negative type. All other polynomials, i.e. when $P+1$
is not $n$
-reduced, are said to have $n$
-boundary type.
Now we discuss $n$
-classes. We begin with the stable $n$
-class, denoted as $n$
-class $S$
. This class consists of $n$
-reduced polynomials $P(X)$
such that for any positive $c$
, the rightmost minimal poison subword of $X P(X)+c$
is disjoint from the subword corresponding to $X P(X)$
. Necessarily, $P+1$
is $n$
-reduced whenever $P$
has $n$
-class $S$
. These are the most common $n$
-reduced polynomials. The class splits into 3 separate sub-classes $S_{-}$
, $S_{0}$
and $S_+$
depending on the sign of the constant coefficient.

The opposite of the stable $n$
-class is the unstable $n$
-class which we write as $n$
-class $U$
. These are the polynomials $P(X)$
where the rightmost minimal poison subword of $XP(X)+c$
violates some global rule for some positive $c$
. These violations are classified in Proposition 28.
Consider a $n$
-reduced polynomial $P$
. Suppose that $XP(X)+k+1$
is not $(n+1)$
-reduced by violating Rule 6 by a potential value change, but $XP(X)+k$
is ${(n+1)}$
-reduced. By Proposition 28, we know that the rightmost minimal poison subword excluding the constant coefficient will have either potential $- \operatorname {sign}^+(c_j \cdot c_{j+1})$
or $-1- \operatorname {sign}^+(c_j \cdot c_{j+1})$
. We notice that $XP(X)+k$
also satisfies the condition for the unstable $(n+1)$
-class.
For a $n$
-reduced polynomial $P$
, if $XP(X)+k$
is not $(n+1)$
-reduced by violating Rule 6 by a potential value change but $XP(X)+k-1$
is $(n+1)$
-reduced, then by Proposition 28, we know that the rightmost minimal poison subword excluding the constant coefficient will have potential $-2 - \operatorname {sign}^+(c_j \cdot c_{j+1})$
. Again, $XP(X)+k-1$
will be in the unstable $(n+1)$
-class.
We split the unstable $n$
-class into sub-classes $U_{0}$
, $U_{-1}$
, and $U_{-2}$
by their truncated potential of the rightmost minimal poison subword being $- \operatorname {sign}^+(c_j \cdot c_{j+1}),$
$-1- \operatorname {sign}^+(c_j \cdot c_{j+1}),$
and $-2 - \operatorname {sign}^+(c_j \cdot c_{j+1})$
, respectively. As observed, multiplying $X$
and adding either $k$
or $k-1$
to $P$
in the unstable $n$
-class will give us a new polynomial in the unstable $(n+1)$
-class with a different potential value.
We follow a similar kind of logic to classify $n$
-reduced polynomials. For Rule 5 in Proposition 28, We will define $n$
-classes $U^t_{-3}$
, $U^t_{-4}$
and $U^t_{-5}$
where the subscript denotes the potential. For Rule 4, we will define $n$
-classes $E_1$
, $E_2$
, and $E_3$
. Note that the $U^t$
classes and the $E$
classes are both associated with top rules.
Definition 30 Base case
Let $Q$
be an integer linear polynomial with a non-negative leading coefficient. The $n$
-class and $n$
-type of $Q$
are given as follows:
• ($n$
-initial type) $n \leq 0$
, $Q=0$
.• ($n$
-interior type) if $Q=dX+c$
, and
◦ $n > 0$
. Then $Q=0$
has $n$
-class $S_{0}$
.◦ $d = 0,$
$0 < c \leq k$
, and $n \leq 0$
. Then $Q$
has $n$
-class $S_+$
.◦ $d=0$
, $c = k + 1$
, and $n \leq 0$
. Then $Q$
has $n$
-class $U^t_{-3}$
.◦ $d=0$
, $0 < c \leq k -1$
, and $n > 0$
. Then $Q$
has $n$
-class $S_+$
.◦ $d=0$
, $c = k$
, and $n > 0$
. Then $Q$
has $n$
-class $U_{0}$
.
• ($n$
-negative type) if $Q=dX+c$
, and
◦ $d=1$
and $c = -k + 1$
. Then $Q$
has $n$
-class $E_1$
if $n \leq 0$
.◦ $d=1$
and $c= -k +2$
. Then $Q$
has $n$
-class $E_3$
if $n \leq 0$
.
• ($n$
-boundary type)
◦ If $n \leq 0$
. Then $Q=k+2$
has $n$
-class $U^t_{-5}$
.◦ If $n > 0$
. Then $Q=k+1$
has $n$
-class $U_{-2}$
.
We define types and classes for polynomials with a non-negative leading coefficient.
Definition 31 Suppose that $Q(X) = XP(X) + c$
is an $n$
-reduced polynomial whose leading coefficient is positive where $P(X) \neq 0$
and $c$
is constant.
• ($n$
-interior type) if
– For any $P$
and $0< c \leq k - 2$
, $Q$
has $n$
-class $S_+$
.– $P$
has any $(n-1)$
-class except $E_1$
and $-P$
does not have $(n-1)$
-class $U_{-2}$
and $c=0$
. Then $Q$
has $n$
-class $S_0$
.– $P$
has $(n-1)$
-class $S_{0}$
, $S_+$
, $U_0$
, or $U^t_{-3}$
and $c =k-1$
. Then $Q$
has $n$
-class $S_+$
.– $P$
has $(n-1)$
-class $U_{-1}$
and $c = k-1$
. Then $Q$
has $U_{0}$
.– $P$
has $(n-1)$
-class $S_{0}$
or $S_+$
and $c =k$
. Then $Q$
has $n$
-class $U_{0}$
.– $P$
has $(n-1)$
-class $S_{-}$
, $E_1$
, $E_2$
, or $E_3$
and $c= k-1$
. Then $Q$
has $n$
-class $U_{0}$
.– $P$
has $(n-1)$
-class $U^t_{-4}$
and $c = k-1$
. Then $Q$
has $n$
-class $U^t_{-3}$
.
• ($n$
-negative type) if
– $P$
has $(n-1)$
-class $E_1$
and $c = -k+1$
. Then $Q$
has $n$
-class $E_2$
.– $P$
has $(n-1)$
-class $E_2$
and $c = -k+1$
. Then $Q$
has $n$
-class $E_3$
.– $P$
has $(n-1)$
-class $E_3$
and $c = -k+1$
. Then $Q$
has $n$
-class $S_{-}$
.– $P$
has $(n-1)$
-class $E_2$
and $c = -k$
. Then $Q$
has $n$
-class $E_1$
.– $P$
has $(n-1)$
-class $E_3$
and $c = -k$
. Then $Q$
has $n$
-class $E_2$
.– $c < 0$
. Then $Q$
has $n$
-class $S_{-}$
unless otherwise stated above.
• ($n$
-boundary type (P)) if
– $P$
has $(n-1)$
-class $U_{-2}$
and $c = k-1$
. Then $Q$
has $n$
-class $U_{-1}$
.– $P$
has $(n-1)$
-class $U^t_{-5}$
and $c = k-1$
. Then $Q$
has $n$
-class $U_{-4}^t$
.– $P$
has $(n-1)$
-class $U_{-1}$
and $c = k$
. Then $Q$
has $n$
-class $U_{-2}$
.– $P$
has $(n-1)$
-class $U^t_{-4}$
and $c = k$
. Then $Q$
has $n$
-class $U^t_{-5}$
.– $P$
has $(n-1)$
-class $U_{0}$
and $c = k$
. Then $Q$
has $n$
-class $U_{-1}$
.– $P$
has $(n-1)$
-class $U^t_{-3}$
and $c = k$
. Then $Q$
has $n$
-class $U^t_{-4}$
.– $P$
has $(n-1)$
-class $S_{-}$
, $E_1$
, $E_2$
, or $E_3$
and $c = k$
. Then $Q$
has $n$
-class $U_{-2}$
.– $P$
has $(n-1)$
-class $S_+$
or $S_{0}$
and $c = k+1$
. Then $Q$
has $n$
-class $U_{-2}$
.
• ($n$
-boundary type (S)) if
– $P$
has $(n-1)$
-class $S_{-}$
such that $-P$
has $(n-1)$
-class $U_{-2}$
and $c = 0$
. Then $Q$
has $n$
-class $S_{0}$
.– $P$
has $(n-1)$
-class $E_1$
and $c = 0$
. Then $Q$
has $n$
-class $S_0$
.
Remark 32 Although we gave our definitions recursively, it is straightforward to prove inductively that these do correspond to the cases in Proposition 28. For example, if $P$
has $n$
-boundary type (P) and $n$
-class $U_{-1}$
, then:
• $P = (\ldots |A)$
with $A = (c_j,\,\ldots,\,c_0)$
and $\operatorname {Pot}(A) = -1-\operatorname {sign}^+(c_j\cdot c_{j+1})$
,• $P+1$
is not $n$
-reduced and its minimal poison subword is $A' = (c_j,\,\ldots,\,c_1,\,c_0+1)$
,• $XP+k+1$
will violate Rule 6 by a potential violation as described in Proposition 28.
From Proposition 28, it is also easy to check that the $n$
-classes form a partition of all $n$
-reduced polynomials. We record this as lemma.
Lemma 33 For each $n$
, the $n$
-classes form a partition of all $n$
-reduced polynomials with non-negative leading coefficient.
7. Successor function
Previously, we have classified any $n$
-reduced polynomial $P$
into $n$
-types and $n$
-classes using an inductive definition. The definition relied on whether $P+1$
is $n$
-reduced and $XP+C$
is $(n+1)$
-reduced. While the classification is complete, the definitions for $S_{0}$
and $S_{-}$
are not descriptive enough to give the growth series on its own, and we develop a tool to bypass that.
In this section, we will define the successor of a polynomial to generate all $n$
-reduced polynomials with a non-negative leading coefficient. We will add 1 consecutively to the $n$
-reduced polynomial until it is no longer $n$
-reduced. When $P(X)+1\in \mathbb {Z}[X]$
is not $n$
-reduced, Proposition 23 will give us a new polynomial

where $c_{-1}=0$
or $1$
. By Proposition 25, we know that $Q(X) -X^{-1} = \sum \nolimits _{i=0}^d c_i X^i \in \mathbb {Z}[X]$
is also $n$
-reduced. Intuitively, we want to define this to be the successor of $P(X)$
. However, in some cases, $Q(X)-X^{-1}-1$
is also $n$
-reduced, and to make sure that the successor function maps onto the set of $n$
-reduced polynomials, we instead define $Q(X)-X^{-1}-1$
to be the successor of $P(X)$
(see (3) in the formal definition).
Definition 34 Let $P$
be an $n$
-reduced polynomial. The successor of $P$
, denoted $\mathcal {S}(P)$
, is given by the following:
(1) If $P$
has $n$
-initial, $n$
-interior, or $n$
-negative type, then $\mathcal {S}(P) = P+1$
.(2) If $P$
has one of the boundary types in the base case, we define the successor as follows. (Note that this is a special case of 3) below.)

(3) Suppose that $P$
has $n$
-boundary type (P) and $n$
-class $U_{-2}$
or $U_{-5}^t$
. Letting $Q$
be the rewriting of $P+1$
given by Proposition 23, we then set $\mathcal {S}(P) = Q-1-X^{-1}$
.(4) Suppose that $P$
has $n$
-boundary type (P) and $n$
-class $U_{-1}$
or $U_{-4}^t$
. Letting $Q$
be the rewriting of $P+1$
given by Proposition 23, we then set $\mathcal {S}(P) = Q-X^{-1}$
.(5) If $P$
has $n$
-boundary type (S), then $\mathcal {S}(P)$
is defined to be the rewriting of $P+1$
given by Proposition 23.
If $\mathcal {S}(P) = P+1$
, it is called a regular successor. Otherwise, it is called an irregular successor.
It is clear in the definition that $\mathcal {S}(P)$
is $n$
-reduced when it is a regular successor. For irregular successors, as an example, we show that $\mathcal {S}(P)$
is $n$
-reduced when $P$
has $n$
-class $U_{-2}$
. If $P=XR(X)+k$
and $R(X)$
has $n$
-class $S_{-}$
, or $P=XR(X)+k+1$
and $R$
has $n$
-class $S_0$
or $n$
-class $S_+$
, then $P+1$
violates Rule 2. Thus, it can be checked easily that $\mathcal {S}(P)$
is $n$
-reduced. If $P=XR(X)+k$
and $R$
has $n$
-class $U_{-1}$
, there exists $d$
such that
where $B$
has stable $n$
-type and $(A|k+1)$
is the rightmost minimal poison subword of $P+1$
. After rewriting $P+1$
and subtracting $1+X^{-1}$
, we have $\mathcal {S}(P)=X^{d+1}(B+1)+XA'-k.$![]()
Given that

and for any subword adjacent to $(-k)$
not containing $c'_d$
we also have
it then follows that
Thus, $\operatorname {Pot}(c'_j \ldots c'_1)\geq 0$
. Since the potential for $A'$
and of any subword of $A'$
is bounded below, we conclude that $\mathcal {S}(P)$
cannot violate Rule 6.
Denote the set of polynomials with a non-negative leading coefficient $\mathcal {R}^+$
. First, we show that the successor function is a bijection from $\mathcal {R}^+ \cup \{0\}$
onto $\mathcal {R}^+$
. See Figure 5.
Proposition 35 The successor function is a bijection from $\mathcal {R}^+\cup \{0\}$
onto $\mathcal {R}^+$
.
Proof. We show this by showing there is an inverse function $\mathcal {S}^{-1}(P)$
, which will be constructed by applying the successor on $-P$
.
Since $-P$
can have a negative leading coefficient, only for this proof, we add new $n$
-types to Definition 30 and Definition 31. For the base case, suppose that $c$
is a negative integer. We say $c$
has $n$
-class $S_{-}$
when $|c|\leq k+1$
and $n > 0$
, or $|c|\leq k+2$
and $n \leq 0$
. For any $n$
-reduced polynomial, we define $n$
-classes and $n$
-types inductively following Definition 31.
Now we define the extension $\tilde {\mathcal {S}}$
of the successor for any $n$
-reduced polynomial. For polynomials with negative leading coefficients, the successor prescribed in Definition 34 will always give an $n$
-reduced polynomial with few exceptions, which can be observed from the potential change violations by Rule 4 and the sign change violations by Rule 5 in Proposition 28. We add two cases of irregular successors.
(1) If $P$
has $n$
-class $-E_2$
, we let $Q$
be the rewriting of $P+1$
given by Proposition 23, we then set $\tilde {\mathcal {S}}(P) = Q-X^{-1}$
. When $P$
has $n$
-class $-E_1$
, we then define $\tilde {\mathcal {S}}(P) = Q-1-X^{-1}$
(2) If $P=XR$
and $R$
has $(n-1)$
-class $-U^t_{-5}$
, then $\tilde {\mathcal {S}}(P)$
is the rewriting of $P+1$
given by Proposition 23.
We see that this extension now is well defined. We first show that $\S ^{-1}(\mathcal {S}(P))=P$
for each case from Definition 34.
Suppose that $\mathcal {S}$
is a regular successor for $P$
. In order for the identity to hold, we need $-(\mathcal {S}(-P-1))=P$
, or equivalently, $\mathcal {S}(-P-1)=-P$
. When $P$
has $n$
-initial, $n$
-interior type, since $-P-1$
has $n$
-negative type, the identity holds.
Now suppose that $P$
has $n$
-negative type. We show that $-P-1$
always has $n$
-interior type. We begin with the base cases

Thus, the claim is true for the base case. Now, in general, if $P=X R+c$
, then

Working out each case directly shows that $\mathcal {S}(-P-1)=-P$
as desired except when $P$
has $n$
-class $S_-$
and $R$
has $(n-1)$
-class $S_-$
. We note that since $P$
is $n$
-reduced, $-R$
having $n$
-class $U_{-2}$
can be ruled out. Furthermore, $-P$
needs to be $n$
-reduced, significantly reducing the number of cases to be considered. In this case, working out explicitly we see that

Hence, the successor for $-P-1$
is regular.
Now suppose that $\mathcal {S}$
is irregular. For the base case, the identity holds trivially by construction. Let $P$
have $n$
-class $U^t_{-4}$
or $U^t_{-5}$
. It is easy to check that $\mathcal {S}(P)$
has $n$
-class $E_2$
and $E_1$
, respectively. For $-\mathcal {S}(P)$
, by construction of the generalized successor, $\tilde {\mathcal {S}}(-\mathcal {S}(P))=P$
as required.
Now let $P$
have $n$
-class $U$
. For the base case, we check when $P=XR+c_0$
has $n$
-class $U_{-2}$
where $R$
has $n$
-class $S$
or $E$
. For the recursive case, there exists a smallest $d>2$
such that
where $B$
has $(n-d)$
-class $S$
or $E$
and $A$
is a degree $d-2$
polynomial. Denote $P=(B|A|c_0)$
. By choice of $d$
, $XA+c_0+1$
corresponds to the rightmost minimal poison subword of $P+1$
.

We observe that $-\mathcal {S}(P)$
can only have $n$
-class $U$
and that the rightmost minimal poison subword for $-\mathcal {S}(P)$
can only be on the same position as $A$
. The rewriting rule for $-\mathcal {S}(P)$
is identical to the rewriting rule for $P$
, and thus, the identity holds.
For $n$
-boundary (S), let $Q=XP$
where $P$
has $(n-1)$
-class of $E_1$
. $\mathcal {S}(Q)$
is also of the form $XR$
where $R$
has $(n-1)$
-class $-U^t_{-5}$
, and the identity follows from construction of the generalized successor. When $Q=XP$
and $P$
has $(n-1)$
-class $-U_{-2}$
, one can observe $-\mathcal {S}(Q)$
also has $n$
-boundary (S) which completes the proof.
The other direction $\mathcal {S}(\mathcal {S}^{-1}(P))=\mathcal {S}(-\mathcal {S}(-P))$
follows from symmetry.
Proposition 36 Suppose $P$
is an $n$
-reduced polynomial with non-negative leading coefficient that represents $x$
.
(1) If $P$
has $n$
-initial type, then $L_n(\mathcal {S}(P)) = L_n(P) + 1$
and $\mathcal {S}(P)$
represents $x+b$
.(2) If $P$
has $n$
-interior type, then $L_n(\mathcal {S}(P)) = L_n(P) + 1$
and $\mathcal {S}(P)$
represents $x+b$
.(3) If $P$
has $n$
-negative type, then $L_n(\mathcal {S}(P)) = L_n(P) - 1$
and $\mathcal {S}(P)$
represents $x+ b$
.(4) If $P \in U_{-5}^t$
or $P \in U_{-2}$
, then $L_n(\mathcal {S}(P)) = L_n(P)$
and $\mathcal {S}(P)$
represents $x -a$
.(5) If $P \in U_{-4}^t$
or $P \in U_{-1}$
, then $L_n(\mathcal {S}(P)) = L_n(P)$
and $\mathcal {S}(P)$
represents $x +b - a$
.(6) If $P$
has $n$
-boundary type (S), then $L_n(\mathcal {S}(P)) = L_n(P)$
and $\mathcal {S}(P)$
represents $x+b$
.
Proof. We will only prove the case when $P\in U_{-2}$
. The other cases can be proved similarly.
Suppose $P \in U_{-2}$
. We then have $P = (\ldots | A)$
for some $A$
with degree $d$
and potential $-2-\operatorname {sign}^+(c_j\cdot c_{j+1})$
. The rewriting of $P+1$
is by subtracting $(X^{d}+X^{d-1} + \ldots +1) (X-(2k+1)+X^{-1})$
, and thus, we may write $\mathcal {S}(P)$
as

Recall that the potential of a coefficient is exactly the length change when adding $2k-1$
. Therefore, the length difference in the last $d+1$
digits of $P$
and $\mathcal {S}(P)$
is $-2-\operatorname {sign}^+(c_j\cdot c_{j+1})+2 = -\operatorname {sign}^+(c_j\cdot c_{j+1})$
, which with $c_{j+1} > 0$
, is exactly the opposite of the effect $X^{d+1}$
has on the length change. Thus, $L_n(\mathcal {S}(P)) = L_n(P)$
. We also have $\mathcal {S}(P)$
represents $x-a$
which follows from the fact that rewriting does not change the element the polynomial represents and that $X^{-1}$
represents $a$
.
For a Laurent polynomial $F(X)$
, we will write $\overline {F}(X) = X^{-1}\cdot F(X^{-1})$
. We define the successor function on its principal part $Q(X)$
dually to the polynomial part, i.e., the $n$
-successor of $Q(X)$
is $\overline {\mathcal {S}(\overline {Q}(X^{-1})}$
, where $\mathcal {S}$
is the $(-n)$
-successor function.
Our main interest is in counting minimal length group elements, not $n$
-reduced polynomials. Therefore, we need to count the number of $n$
-minimal polynomials that represent any group element and to take care over the overcount. Therefore, suppose $F(X) = \sum \nolimits _{j=-p}^{q} c_j X^j$
represents $x$
, and write its polynomials and principal parts as $P(X)$
and $Q(X)$
, respectively. In the next Lemma, we show that we can write $P(X) = \pm \mathcal {S}^n(0)$
and $\overline {Q}(X) = \pm \mathcal {S}^m(0)$
. Thus, $(\pm \mathcal {S}^n(0)|\pm \overline {\mathcal {S}^m(0)})$
represents $x$
. However, for a fixed $x$
, there may be more than one way to write it as $(\pm \mathcal {S}^n(0)|\pm \overline {\mathcal {S}^m(0)})$
. To quantify this, we will also need the proof of the following lemma.
We view the sequence $\{\mathcal {S}^i(0)\}$
of successive polynomials as a sequence of points $g_i$
which they represent in $\mathbb {Z}^2=\left < a,\,b \right >$
, and the successor corresponds to moving in either $b$
, $-a$
or $b-a$
direction depending on the type. Note that if $\overline {Q}(X)$
represents $ka+tb$
then $Q(X)$
represents $ta+kb$
. We state the lemma in a slightly more general setting.
Lemma 37 Spanning lemma
Let $G = \left < a,\,b \right >$
be a free abelian group. Suppose there exists an infinite sequence $\{g_i \} \subset G$
such that $g_0 = 0,$
$g_1 = b$
and $g_{i+1} \in \{g_{i} + b,\, g_i + b -a,\, g_i-a\}$
. Moreover, assume that if $g_{i+1} - g_i \in \{b - a,\,-a\},$
then $g_{i+2} - g_{i+1} = g_{i+3}-g_{i+2}= b$
. We also define $\{h_i\}$
such that if $g_i = ka+ tb,$
then $h_i = ta+ kb$
. Then for all $\gamma \in G,$
there exist $i,\,j \in \mathbb {N}$
such that $\gamma = \pm g_{i} \pm h_{j}$
.
Proof. We first prove that if some $\gamma$
is a sum of some $g_i$
and $h_j$
, then any $\gamma + ma+nb$
is also for any $m,\,n\in \mathbb {N}$
. We proceed by induction on $n$
and $m$
. Suppose that $\gamma = g_i+h_j$
. We first need to find a pair $(i',\,j')$
such that $\gamma + b = g_{i'}+h_{j'}$
. If $g_{i+1}=g_i+b$
, we are done. If $g_{i+1}=g_i+(b-a)$
, then

On the other hand, if $g_{i+1}=g_i+(-a)$
, then

We observe that by the symmetry between $g_i$
and $h_i$
, the $a$
's can be incremented in the same manner. This proves the lemma for elements in $\{g_i+ma\}\cup \{h_j+nb\}$
, i.e., the upper-right region cut out by the trajectories of $g_i$
and $h_i$
(see Figure 6). By switching signs, any element in $\{-g_i-ma\}\cup \{-h_j-nb\}$
can be expressed as $-g_i-h_j$
as well.
New minimal poison subword cannot contain a subword corresponding to the previous minimal poison subword. (Lemma 19).

New minimal poison subword cannot appear on the right. (Lemma 20).

As we add one to our reduced polynomial, we move from the $n$
-negative type (blue region) to the $n$
-interior type (white region). Finally, as we reach the end of the interior region, we reach the $n$
-boundary types (yellow region).

For point $A$
, there is some $(i,\,j)$
such that $g_i+h_j$
represents $A$
. For point $B$
, the representative to look for is $g_i-h_j$
. Instead of actually finding $(i,\,j)$
, we increment by $\pm a$
and $\pm b$
from a known pair.

By symmetry, this leaves out the case when $\gamma \in \{g_i-ma\}\cap \{-h_j+nb\}$
. We will show that such $\gamma$
can be written as $g_i-h_j$
for some $i,\,j\in N$
. By induction, let $\gamma =g_i-h_j$
for $j>1$
. We need to find a pair $(i',\,j')$
such that $\gamma +b=g_{i'}-h_{j'}$
where $i'>i$
and $j'< j$
. If $g_{i+1}-g_i=b$
, then $\gamma +b=g_i-h_j$
, and we are done. If $g_{i+1}-g_i=b-a$
, then either

where the last case only appears when $j\geq 2$
as $h_1 = a$
. On the other hand, if ${g_{i+1}-g_i=-a}$
, then either

When $j = 0$
, $\gamma = g_i$
and $\gamma +b$
will either be $g_{i+1}$
or is covered by the first case of the proof. Hence, any element in the set $\{g_i-ma\}\cap \{-h_j+nb\}$
can be expressed as a difference $g_i-h_j$
. By symmetry, this can be done for the set $\{-g_j+ma\}\cap \{h_j-nb\}$
.
For counting, we need to know how many representatives of a single $\gamma$
exist. We use the same framework as in Lemma 37 to figure out all the cases and then find the appropriate types and classes in relation to the successor function.
As in the lemma, we first look at the sum of two sequences $g_i$
and $h_j$
. We find all possible pairs of sequences satisfying
Equivalently, we can find sequences with weaker conditions whose sum is 0. Here we only require $-a$
and $b-a$
to be followed by two consecutive $b$
's in the sequence $\{g_i\}$
and a similar requirement for the sequence $\{h_j\}$
. When any of two sequences is long, the $b$
's (or $a$
's) dominate the sequence. Hence, it suffices to look at combinations shorter than 3 symbols:

Now we consider the Laurent polynomial $F(X) = \sum \nolimits _{j=-p}^{q} c_j X^j$
that represents $\gamma$
, and write its polynomials and principal parts as $P(X)$
and $Q(X)$
, respectively. Suppose both $P$
and $\overline {Q}$
have non-negative leading coefficients. For $n\geq 0$
, when $P$
is an $n$
-reduced polynomial and $\overline {Q}$
is a $(-n)$
-reduced polynomial, we have $P$
representing some $g_i$
and $Q$
representing some $h_j$
and $x = g_i+h_j$
. Thus, the overcountings are in the form $g_i + h_j = g_{i'}+h_{j'}$
, and based on the discussion above and Prop 36 can now be classified in terms of types and classes.
$(P|\overline {Q})=(\mathcal {S}(P)|\overline {\mathcal {S}(Q)})$
occurs when

Similarly, $(P|\overline {Q})=(\mathcal {S}(P)|\overline {\mathcal {S}^2(Q)})$
occurs when

And finally, $(P|\overline {Q})=(\mathcal {S}^2(P)|\overline {\mathcal {S}^2(Q)})$
can be classified.

Now we consider the case when $P$
is positive and $\overline {Q}$
is negative, so $P+Q$
represents $x = g_i-h_j$
:
In terms of the successor, this corresponds to
and the possible classes for $P$
and $Q$
are described below.

8. Growth series for $n$
-reduced polynomials
For the reader's convenience, we give a brief summary here. We begin with the main theorem of this article.
Main Theorem Let $A = \left [\begin {smallmatrix} 0 & -1 \\ 1 & 2k+1\end {smallmatrix}\right ]\in SL(2,\,\mathbb {Z})$
where $2k+1 \ge 5$
. Then $\mathbb {Z}^2\rtimes _A\mathbb {Z}$
has rational growth with respect to the standard generating subset.
Recall that for any group element $g=(x,\,t^n)\in \mathbb {Z}^2\rtimes _A\mathbb {Z}$
, we represent this element by a Laurent polynomial $P(X)$
, and $\|g\|$
is calculated by the $n$
-length $L_n$
. If

then $n$
-length $L_n$
is defined as

where $p \geq \text {max}\{0,\,-n\}$
and $q \geq \text {max}\{0,\,n\}$
. Here we assumed the polynomial had zero as its coefficients on both ends if these two conditions are not met. $P(X)$
with minimal $n$
-length is called the $n$
-minimal polynomial. Instead of working with $n$
-minimal polynomials, we defined $n$
-reduced polynomials using a list of rules given in Definition 8.
In Theorem 24, we showed that any $n$
-minimal polynomial can be rewritten as a sum of a polynomial part and a principal part, both of which are $n$
-reduced. So once the growth series for all $n$
-reduced polynomials are set up, the growth series for $n$
-minimal polynomials can be deduced with possible multiplicities.
For the growth series of $n$
-reduced polynomials, we focus on the polynomials with non-negative leading coefficient, as we can easily deduce the series for the polynomial $P$
with negative leading coefficient by looking at $-P$
instead.
For an $n$
-reduced polynomial $P$
, it may be the case that $P+1$
is not $n$
-reduced; we gave a full classification in Proposition 28. Using this, we divided all $n$
-reduced polynomials into $n$
-types and $n$
-classes in Definition 30 and Definition 31. In this section, each $n$
-type and the degree of a polynomial will serve as a state. For any $P$
in that state, we will quantify the state for $XP+C$
; this will play the role of transition between two states. Since we know specifically how much length is changed when the successor is applied, we can explicitly write down the transition as a matrix as well as its impact to the generating series.
Since conditions given in Definition 30 and Definition 31 are not descriptive enough, we define a function called the successor (Definition 34). We show we can obtain all $n$
-reduced polynomials with non-negative leading coefficient by starting from 0 and applying the successor function repeatedly, which is shown to be injective (Proposition 35) and spans all polynomials (Lemma 37). This allows us to count other $n$
-types by looking at the image of the successor instead. In the same lemma, we also listed a finite number of ways to represent the same element where both polynomial and principal parts are $n$
-reduced. From this, we can conclude that once the generating functions for each types are rational, the growth of the group is rational.
We begin by defining the generating function for polynomials having a given $n$
-class of degree $d$
with a non-negative leading coefficient and use the inductive definition given in Definition 30 and Definition 31 to find the recursive relation using the following table.
Definition 38 We denote the generating function having the stable $n$
-classes with given degree as follows

We denote the sum of generating functions for all polynomials of degree $d$
having stable $n$
-classes as
We also introduce the vector notation for the generating function:

Definition 39 We denote the generating functions having the unstable classes associated with potential value change violations associated with Rule 5 and Rule 6 as in Proposition 28 as follows:

We denote the generating function of all polynomials of degree $d$
having unstable $n$
-classes associated with Rule 6 as
and denote the generating function of all polynomials of degree $d$
having unstable $n$
-classes associated with Rule 5 as
As a vector notation, we have

Definition 40 We denote the generating functions having unstable classes associated with potential value changes violations associated with Rule 4 using the following table:

and we denote the generating function of all polynomials of degree $d$
having unstable $n$
-classes as
Finally, as a vector notation, we write

When there is no restriction on the degree, the generating functions will be denoted as

We begin computing these generating functions with small $d$
by computing the $n$
-length for every item in the base case. (Definition 30.)
Proposition 41

Similarly,

Proof. Let $d=0$
. When $n>0$
, we have $n$
-classes $S_0$
, $S_+$
, $U_0$
and $U_{-2}$
, and $L_n(c_0)=n+|c_0|$
. For stable $n$
-types, $0\leq c_0\leq k-1$
. $c_0=k$
and $c_0=k+1$
corresponds to $U_0$
and $U_{-2}$
respectively. On the other hand, if $n\leq 0$
, we have $n$
-classes $S_+$
, $U^t_{-3}$
and $U^t_{-5}$
. In this case, $L_n(c_0)=-n+|c_0|$
. For stable $n$
-types, $0\leq c_0\leq k$
. $c_0=k+1$
and $c_0=k+2$
corresponds to $U^t_{-3}$
and $U^t_{-5}$
, respectively.
Remark 42 We exclude the initial type since it plays no role in the recursive definition of $n$
-types.
Lemma 43

Proof. The $n$
-class $E$
appears only when the degree is greater than 1 and $n< d$
. There are two cases we need to consider: $P(X)=X-k+1$
has $0$
-class $E_1$
and $P(X)=X-k+2$
has $0$
-class $E_3$
.
Now we introduce recursion using Definition 31. We begin with $n$
-class $E$
.
Proposition 44

Proof. Suppose that $P$
has $n$
-class $E_2$
. By Definition 31,
has $(n+1)$
-class $E_3$
, and the difference between the $(n+1)$
-length of $XP+(-k+1)$
and the $n$
-length of $P$
is $k$
. Similarly, if $P$
has $n$
-class $E_1$
, $XP+(-k+1)$
has $(n+1)$
-class $E_2$
and the length increases by $k$
.
On the other hand, if $P$
has $n$
-class $E_3$
, $XP-k$
has $(n+1)$
-class $E_2$
; and if $P$
has $n$
-class $E_2$
, $XP-k$
has $(n+1)$
-class $E_1$
. In both cases, the length increases by $k+1$
.
Note that when $P$
has $n$
-class $E_3$
, $XP+(-k+1)$
has $(n+1)$
-class $S_{-}$
, so this recursive definition will not appear here; this will appear when we consider a transition between $E$
-classes and $S$
-classes later.
For the next proposition, we introduce the following matrices

Proposition 45 $\vec {U}^{(n+1,d+1)}(t) = P_{U,S}\vec {S}^{n,d}(t)+P_{U,U}\vec {U}^{n,d}(t)$![]()
Similarly, for other classes except stable classes, we have
Proposition 46

The proofs for both propositions are similar to that of Proposition 44.
For the next proposition, we define the following matrices:

For stable classes, we set
For the $*$
in these matrices, the recursive definition does not explicitly state what coefficients can be attached at the end to get $n$
-class $S_0$
. Instead of pursuing the exact condition, we note that for any polynomial $P$
having $n$
-class $E$
or $S_{-}$
, we know the $n$
-class of $-P$
by the proof of Proposition 35. Exploiting this symmetry, we have

which can be used to fill the missing entries for $P_{S,S},\, P_{S,U},\, P_{S,T}$
and $P_{S,E}$
.
Denote

by the mentioned symmetry, we have the following identity

where

and we will use this new block matrix instead of $\mathcal {P}$
.
So far, the degree of polynomials having $n$
-classes is fixed. The induction provided above increments $d$
by 1 at the cost of increasing $n$
. To work around this, we will use the following lemma:
Lemma 47 Suppose that $n\leq 0$
, then

Proof. When $n< 0$
, the rules for $n$
-reduced polynomials are identical to that of $n=0$
except the $n$
-length increase by $|n|$
. In other words, for $n<0$
, each $0$
-reduced polynomial with $0$
-length $\ell$
can be considered as an $n$
-reduced polynomial with $n$
-length $\ell +|n|$
.
Now we compute the generating function for each class without restrictions on the degree by summing over all possible degrees. First, observe that

which allows us to write

and for $n$
-classes, we have

which follows from Lemma 47.
Since $\mathcal {P}$
is a $12\times 12$
matrix with polynomial entries, $(1-\mathcal {P})^{-1}$
is a matrix with rational function entries. Thus, $S^n$
, $U^n$
, $T^n$
and $E^n$
are all rational for fixed $n$
. $(1-\mathcal {P}/t)^{-1}$
is also a matrix with rational function entries, hence the sum of $S^n$
, $U^n$
, $T^n$
and $E^n$
over all $n$
are rational. Since every $n$
-minimal polynomial representing a group element can be decomposed into an $n$
-reduced polynomial and $-n$
-reduced polynomial with possibly multiplicities described in Proposition 37, the set of all $n$
-minimal polynomials exhibits a rational growth. Furthermore, taking account of the multiplicities in representing group elements only changes the end function by rational functions because $U^n$
, $T^n$
and $E^n$
are all rational.
Acknowledgements
The authors would like to thank the referee for suggestions that improved the exposition of this article.
The first author is partially supported by NSF grant DMS-1600857 “Von Neumann Algebras: Rigidity, Applications to Measurable Dynamics, and Model Theory”.
The second author acknowledges support from U.S. National Science Foundation grants DMS-1107452, 1107263, 1107367 “RNMS: Geometric Structures and Representation Varieties” (the GEAR Network) and DMS-2054558 “RUI: Randomness, Computability, and Complexity in Groups”.












































