---
title: Generalized Cyclotomic Mappings
url: https://www.emergentmind.com/topics/generalized-cyclotomic-mappings
type: topic
---

# Generalized Cyclotomic Mappings

Generalized cyclotomic mappings are finite-field maps that are defined monomially on each coset of a fixed multiplicative subgroup. If \(q\) is a prime power and \(\ell \mid (q-1)\), one fixes an index-\(\ell\) subgroup \(C_0 \le \mathbb F_q^\ast\) and its cosets \(C_i\), and then prescribes \(f(x)=a_i x^{r_i}\) on each \(C_i\), with the normalization \(f(0)=0\). In this sense they interpolate between monomial maps and fully arbitrary piecewise-defined maps, while retaining enough algebraic structure to admit explicit polynomial forms, permutation criteria, inverse formulas, functional-graph descriptions, and, more recently, complete many-to-one classifications in several regimes [2503.06654].

## 1. Definition, cyclotomic classes, and equivalent representations

Let \(q\) be a prime power, let \(\mathbb F_q\) be the finite field with \(q\) elements, and let \(\gamma \in \mathbb F_q^\ast\) be primitive. If \(q-1=\ell s\), the index-\(\ell\) subgroup
\[
C_0=\{\gamma^{k\ell}:0\le k<s\}
\]
has cosets
\[
C_i=\gamma^i C_0=\{\gamma^{k\ell+i}:0\le k<s\}, \qquad 0\le i\le \ell-1,
\]
which partition \(\mathbb F_q^\ast\). A generalized cyclotomic mapping is then the piecewise map
\[
f(x)=
\begin{cases}
0, & x=0,\\
a_i x^{r_i}, & x\in C_i,\ 0\le i\le \ell-1,
\end{cases}
\]
with \(a_i\in \mathbb F_q^\ast\) and \(r_i\in \mathbb N\). Classical cyclotomic mappings appear as special cases: \(r_i=1\) for all \(i\) in the first-order setting, and \(r_i\) constant \(>1\) in the higher-order classical setting [2503.06654].

A central feature of the theory is that the piecewise description admits a single polynomial representative modulo \(x^q-x\). Writing \(\omega=\gamma^s\), a primitive \(\ell\)-th root of unity in \(\mathbb F_q\),
\[
f(x)=\frac{1}{\ell}\sum_{i=0}^{\ell-1} a_i x^{r_i}\sum_{j=0}^{\ell-1}\left(\frac{x^s}{\omega^i}\right)^j
=\frac{1}{\ell}\sum_{i=0}^{\ell-1}\sum_{j=0}^{\ell-1} a_i \omega^{-ij} x^{r_i+js}.
\]
This polynomial form is one side of a broader conversion theory: generalized cyclotomic mappings can be switched between cyclotomic form, polynomial form, and, in the permutation case on \(\mathbb F_q^\ast\), wreath product form \( \mathrm{Hol}(C)\wr \mathrm{Sym}(d)\), where the index is denoted \(d\) rather than \(\ell\) [2103.00664].

This multiplicative-coset description is the standard framework, but it is not the only one used in the literature. In bent-function constructions, an additive-coset analogue is also used, with branch functions specified on additive cosets rather than multiplicative ones. That broader usage suggests that “generalized cyclotomic mapping” now names a family of piecewise algebraic constructions organized by subgroup or subspace partitions, not merely a permutation-polynomial ansatz [2310.01317].

## 2. Cosetwise behavior, target cosets, and the many-to-one formalism

The local behavior of a generalized cyclotomic mapping is governed by two parameters: the exponent collapse inside a coset and the destination coset of its image. For a fixed primitive element \(\gamma\), define
\[
\phi(i,n)=(i r_i+\log_\gamma a_i)\bmod n.
\]
Then \(f\) maps \(C_i\) onto the coset \(C_{\phi(i,\ell)}\). If
\[
d_i=\gcd(r_i,s),
\]
the restriction
\[
f_i:C_i\to C_{\phi(i,\ell)}, \qquad f_i(x)=a_i x^{r_i},
\]
is uniformly \(d_i\)-to-\(1\), and \(E_f(C_i)=\varnothing\). Thus every local branch contributes a completely explicit fiber size: raising to the \(r_i\)-th power on a cyclic group of order \(s\) collapses each fiber by the factor \(\gcd(r_i,s)\) [2503.06654].

The global many-to-one problem is therefore not local but combinatorial: it depends on how the images \(f_i(C_i)\) overlap. For \(0\le i,j\le \ell-1\), with \(d=\gcd(d_i,d_j)\) and \(\bar d=\mathrm{lcm}(d_i,d_j)\), the intersection behavior is characterized by congruences in \(\phi\):
\[
f_i(C_i)\cap f_j(C_j)=\varnothing \iff \phi(i,\ell d)\ne \phi(j,\ell d),
\]
\[
f_i(C_i)\subseteq f_j(C_j)\iff d_j\mid d_i \text{ and } \phi(i,\ell d_j)=\phi(j,\ell d_j),
\]
and partial overlap without containment occurs exactly when \(d_j\nmid d_i\) and \(\phi(i,\ell d)=\phi(j,\ell d)\), in which case the intersection has size \(s/\bar d\) and is an explicit coset progression. The proofs reduce these alternatives to linear Diophantine congruences modulo \(q-1\), and this intersection calculus is the core mechanism behind all global \(m\)-to-\(1\) classifications [2503.06654].

The paper on many-to-one properties also fixes the relevant notion of \(m\)-to-\(1\). If \(A\) is finite and \(|A|=km+r\) with \(0\le r<m\), a map \(f:A\to B\) is called \(m\)-to-\(1\) on \(A\) if there exist \(k\) distinct images each having exactly \(m\) preimages; the remaining \(r\) domain elements form the exceptional set \(E_f(A)\). Over \(\mathbb F_q\), under the normalization \(f(0)=0\) and assuming \(0\) is the only root, one has:
- for \(m=1\), \(f\) is \(1\)-to-\(1\) on \(\mathbb F_q\) iff it is \(1\)-to-\(1\) on \(\mathbb F_q^\ast\);
- for \(m\ge 2\), \(f\) is \(m\)-to-\(1\) on \(\mathbb F_q\) iff \(m\nmid q\) and \(f\) is \(m\)-to-\(1\) on \(\mathbb F_q^\ast\).

A common misconception is that generalized cyclotomic mappings are primarily a permutation-theoretic object. Earlier work indeed concentrated on the one-to-one property, inversion, cycle structure, and functional graphs, but the many-to-one viewpoint shows that the same cosetwise algebra supports a parallel fiber-distribution theory rather than merely a permutation theory [2503.06654].

## 3. Complete many-to-one classifications for \(\ell\le 3\)

For \(\ell=1\), generalized cyclotomic mappings reduce to monomials
\[
f(x)=a x^r.
\]
In that case \(f\) is \(m\)-to-\(1\) on \(\mathbb F_q^\ast\) if and only if
\[
m=\gcd(r,q-1),
\]
and \(E_f(\mathbb F_q^\ast)=\varnothing\). For \(m\ge 2\), \(f\) is \(m\)-to-\(1\) on \(\mathbb F_q\) iff \(m\nmid q\) and the same condition holds on \(\mathbb F_q^\ast\). Thus the monomial case is completely rigid: the fiber size is exactly the exponent’s gcd with the multiplicative group order [2503.06654].

For \(\ell=2\), with \(q\) odd and \(s=(q-1)/2\), the mapping can be written as
\[
f(x)=a_0x^{r_0}(1+x^s)+a_1x^{r_1}(1-x^s),
\]
which agrees with \(a_0x^{r_0}\) on \(C_0\) and \(a_1x^{r_1}\) on \(C_1\). Let \(d_i=\gcd(r_i,s)\). Then for \(1\le m<q\), \(f\) is \(m\)-to-\(1\) on \(\mathbb F_q^\ast\) if and only if one of two situations holds:
1. \(m=d_0=d_1\) and \(\phi(0,2m)\ne \phi(1,2m)\);
2. \(m=d_0+d_1\), \(d=\min\{d_0,d_1\}\) divides \(m\), \(\phi(0,2d)=\phi(1,2d)\), and
\[
s\cdot \frac{m-2d}{m-d}<m.
\]
The paper also isolates low-degree specializations. For \(m=2\), \(f\) is \(2\)-to-\(1\) on \(\mathbb F_q^\ast\) iff either \(d_0=d_1=2\) with \(\phi(0,4)\ne \phi(1,4)\), or \(d_0=d_1=1\) with \(\phi(0,2)=\phi(1,2)\). For \(m=3\) and \(q\ge 13\), \(f\) is \(3\)-to-\(1\) iff \(d_0=d_1=3\) and \(\phi(0,6)\ne \phi(1,6)\) [2503.06654].

For \(\ell=3\), with \(q-1=3s\),
\[
\begin{aligned}
f(x)=&\ a_0x^{r_0}(1+x^s+x^{2s})\\
&+a_1x^{r_1}(1+\omega^2x^s+\omega x^{2s})\\
&+a_2x^{r_2}(1+\omega x^s+\omega^2x^{2s}),
\end{aligned}
\]
where \(\omega\) is a primitive cube root of unity. Writing \(d_t=\gcd(r_t,s)\) and ordering them as \(d_i\le d_j\le d_k\), the paper proves a complete classification: \(f\) is \(m\)-to-\(1\) on \(\mathbb F_q^\ast\) if and only if exactly one of six structural scenarios holds. These six cases are organized by the equalities among the \(d_t\), the equalities or inequalities among the relevant \(\phi(i,\cdot)\), and an exceptional-set inequality. Concretely, the scenarios are:
1. \(m=d_0=d_1=d_2\), with \(\phi(0,3m),\phi(1,3m),\phi(2,3m)\) pairwise distinct;
2. \(m=d_k=d_i+d_j\), \(d_i\mid d_j\), \(\phi(i,3d_i)=\phi(j,3d_i)\ne \phi(k,3d_i)\), and \(s(d_j-d_i)/d_j<m\);
3. \(m=d_i+d_j\), \(d_j=d_k\), \(d_i\mid d_j\), \(\phi(i,3d_i)=\phi(j,3d_i)=\phi(k,3d_i)\), \(\phi(j,3d_j)\ne \phi(k,3d_j)\), and \(s(d_j-2d_i)/d_j<m\);
4. \(m=2s\), \(d_0=d_1=d_2=s\), and \(\phi(i,3s)=\phi(u,3s)\ne \phi(v,3s)\) with \(\{u,v\}=\{j,k\}\);
5. \(m=2s\), \(d_j=d_k=s\), \(\phi(j,3s)=\phi(k,3s)\), and \(\phi(i,3d_i)\ne \phi(j,3d_i)\);
6. \(m=d_0+d_1+d_2\), \(d_i\mid d_k\), \(d_j\mid d_k\), \(\phi(i,3d_i)=\phi(k,3d_i)\), \(\phi(j,3d_j)=\phi(k,3d_j)\), and \(s(2d_k-d_i-d_j)/d_k<m\).

The significance of the \(\ell\le 3\) results is not merely that they settle small indices. They show that many-to-one behavior is controlled by a finite set of local gcd parameters and target-coset congruences, together with explicit exceptional-set bounds. This strongly suggests that the obstruction to a general classification at larger index is combinatorial explosion rather than the absence of structural invariants [2503.06654].

## 4. General \(2\)-to-\(1\) classification and the equal-gcd regime

The strongest all-index theorem in the 2025 work is the complete classification of \(2\)-to-\(1\) generalized cyclotomic mappings for arbitrary \(\ell \mid (q-1)\). Let \(q-1=\ell s\) with \(2\le \ell \le (q-1)/2\), and define
\[
I=\{i\in[\ell]:\gcd(r_i,s)=1\}, \qquad T=\{t\in[\ell]:\gcd(r_t,s)=2\}.
\]
Then \(f\) is \(2\)-to-\(1\) on \(\mathbb F_q^\ast\) if and only if one of three cases holds:
1. \(I=[\ell]\), \(|I|\) is even, and \(i\mapsto \phi(i,\ell)\) is \(2\)-to-\(1\) on \(I\);
2. \(T=[\ell]\) and \(t\mapsto \phi(t,2\ell)\) is injective on \(T\);
3. \(I\cup T=[\ell]\), \(\phi(i,\ell)\ne \phi(t,\ell)\) for all \(i\in I\), \(t\in T\), \(i\mapsto \phi(i,\ell)\) is \(2\)-to-\(1\) on \(I\) with \(|I|\) even, and \(t\mapsto \phi(t,2\ell)\) is injective on \(T\).

This theorem yields a direct decision procedure: compute \(d_i=\gcd(r_i,s)\), reject if any \(d_i\ge 3\), partition the indices into \(I\) and \(T\), compute the appropriate \(\phi(i,n)\), and then check the pairing or injectivity conditions. The proof idea is equally transparent: since each local branch \(f_i\) is \(d_i\)-to-\(1\), only \(d_i\in\{1,2\}\) are compatible with global \(2\)-to-\(1\), and the arithmetic of \(\phi\) determines how distinct coset images collide [2503.06654].

The same paper also records a broader “equal gcd case.” If
\[
(r_i,s)=d \qquad \text{for all } i,
\]
then \(f\) is \(m\)-to-\(1\) on \(\mathbb F_q^\ast\) if and only if
\[
d\mid m,\qquad \phi(x,\ell d)\text{ is }(m/d)\text{-to-1 on }[\ell],\qquad s\big(\ell \bmod (m/d)\big)<m.
\]
This statement does not solve the general \(m\)-to-\(1\) problem for all \(\ell\), but it gives a complete answer when the local collapsing factors are constant across all branches [2503.06654].

The limitation is explicit. Complete many-to-one characterizations are proved for \(\ell\le 3\), while for \(\ell\ge 4\) a complete classification for general \(m\ge 3\) remains open. Enumerative questions—how many parameter tuples \((a_i,r_i)\) yield a prescribed \(m\)-to-\(1\) behavior for fixed \(q\) and \(\ell\)—are also left open, and the paper notes that such counts depend on discrete logarithms and modular constraints [2503.06654].

## 5. Constructions on \(\mathbb F_{q^2}\) via \(U_{q+1}\)

A major extension of the many-to-one theory is the passage from \(\mathbb F_q\) to \(\mathbb F_{q^2}\). Let
\[
U_{q+1}=\{x\in \mathbb F_{q^2}^\ast : x^{q+1}=1\},
\]
a cyclic subgroup of \(\mathbb F_{q^2}^\ast\). If \(q+1=\ell t\), then an index-\(\ell\) subgroup
\[
A_0=\{\zeta^{j\ell}:0\le j<t\}\subset U_{q+1}
\]
has cosets \(A_i=\zeta^i A_0\), where \(\zeta\) generates \(U_{q+1}\). One studies polynomials
\[
f(x)=x^r h(x^{q-1})\in \mathbb F_{q^2}[x]
\]
under the condition that
\[
g(x)=x^r h(x)^{q-1}
\]
induces monomial functions on the cosets \(A_i\), namely
\[
g(x)=\lambda_i x^{e_i}, \qquad x\in A_i.
\]
This recreates the generalized cyclotomic mapping setup over \(U_{q+1}\) [2503.06654].

The bridge lemma is exact. If \((r,q-1)=1\) and \(h\) has no roots in \(U_{q+1}\), then for \(1\le m\le q+1\),
\[
f(x)\text{ is }m\text{-to-1 on }\mathbb F_{q^2}
\iff
g(x)\text{ is }m\text{-to-1 on }U_{q+1}
\]
and
\[
(q-1)\big((q+1)\bmod m\big)<m.
\]
Thus the construction of many-to-one polynomials \(x^r h(x^{q-1})\) reduces to a cyclotomic many-to-one problem on the norm-one subgroup [2503.06654].

For \(\ell=2\), with \(t=(q+1)/2\), the criterion on \(U_{q+1}\) parallels the \(\mathbb F_q^\ast\) case. Writing
\[
g(x)=
\begin{cases}
\lambda_0 x^{e_0}, & x\in A_0,\\
\lambda_1 x^{e_1}, & x\in A_1,
\end{cases}
\qquad
d_i=\gcd(e_i,t),
\]
and
\[
\psi(i,n)=(i e_i+\log_\zeta \lambda_i)\bmod n,
\]
one has that \(g\) is \(m\)-to-\(1\) on \(U_{q+1}\) iff either \(m=d_0=d_1\) with \(\psi(0,2m)\ne \psi(1,2m)\), or \(m=d_0+d_1\), \(d\mid m\), \(\psi(0,2d)=\psi(1,2d)\), and
\[
t\cdot \frac{m-2d}{m-d}<m.
\]
The bridge lemma then transfers this to \(f(x)=x^r h(x^{q-1})\) on \(\mathbb F_{q^2}\) [2503.06654].

The paper derives explicit binomial and trinomial families. For instance, if \(h(x)^{q-1}\) is monomial on each \(A_i\), as in the Hou–Lavorante list, then for
\[
h(x)=1+a x^{(u+vt)}, \qquad a\in U_{q+1},
\]
one obtains explicit many-to-one families. In the case \(v=1\),
\[
f(x)=x^r\big(1+a x^{(u+t)(q-1)}\big)
\]
is \(m\)-to-\(1\) precisely when
\[
m=\gcd(r-u,t)\ \text{ and }\ r-u\not\equiv t \pmod{2m},
\]
or
\[
m=2\gcd(r-u,t)\ \text{ and }\ r-u\equiv t \pmod m,
\]
together with
\[
(q-1)\big((q+1)\bmod m\big)<m
\]
and
\[
(-a)^{(q+1)/(q+1,u+t)}\ne 1.
\]
Analogous trinomial criteria are given for
\[
h(x)=1+a x^t+b x^{u+vt}
\]
under the conditions \(a^{q-1}=-1\), \(b^{q+1}=1-a^2\), \(u\in(0,t)\), and \(v\in\{0,1\}\) [2503.06654].

A plausible implication is that the subgroup \(U_{q+1}\) functions as a transfer device: once generalized cyclotomic many-to-one behavior is understood on this cyclic subgroup, substantial families of structured binomials and trinomials over \(\mathbb F_{q^2}\) become accessible without repeating the entire finite-field analysis from scratch.

## 6. Permutations, inverses, functional graphs, and applications

The many-to-one theory sits inside a larger body of work on generalized cyclotomic mappings. In the permutation direction, the criterion is classical: when the coefficients are nonzero and each local exponent is invertible modulo the subgroup size, bijectivity reduces to distinct coset images. This viewpoint is formalized in several equivalent languages. One paper develops an explicit conversion from cyclotomic form to polynomial form and back, and establishes a concrete permutation-group isomorphism between the restrictions of generalized index-\(d\) cyclotomic permutations to \(\mathbb F_q^\ast\) and the wreath product \(\mathrm{Hol}(C)\wr \mathrm{Sym}(d)\). Within the same framework, one obtains criteria for permutation status, formulas for inverses, and classifications of long cycles and involutions [2103.00664].

A more specialized inverse theory appears in the study of cyclotomic mapping permutation polynomials. For branch monomials \(x^{r_i}\), the permutation criterion can be written as: \((r_i,s)=1\) for all \(i\), and \(\{\mathrm{ind}_\gamma(A_i)+ir_i\mid i=0,\dots,\ell-1\}\) is a complete set of residues modulo \(\ell\). The inverse is again cyclotomic, with exponents \(r_i^{-1}\bmod s\) and explicitly computable coefficients, and the involution condition becomes a system of modular equalities on the exponents and discrete-log parameters. The paper emphasizes that these constructions and tests require only modular operations [1601.04099].

Beyond bijectivity, the iteration theory of generalized cyclotomic mappings has been developed via functional graphs. By identifying each coset \(C_i\) with \(\mathbb Z/s\mathbb Z\), the restriction to a coset becomes an affine map \(A_i(x)=\alpha_i x+\beta_i\), and the global dynamics decomposes into cycles of cosets with affine trees attached. The paper “Functional graphs of generalized cyclotomic mappings of finite fields” provides structural results, parametrizations of connected components by representative vertices, and Las Vegas algorithms whose expected runtime, for fixed index \(d\) and most prime powers \(q\), is polynomial in \(\log q\) on quantum computers and subexponential in \(\log q\) on a classical computer [2304.00181].

The first-order case \(f(x)=a_i x\) on each coset, sometimes treated separately as first-order cyclotomic mappings, has its own cycle-theoretic results. In particular, for \(q\) sufficiently large relative to the index, all cycle types of first-order cyclotomic permutations with only long cycles on \(\mathbb F_q^\ast\) can be achieved through a complete mapping, as can all permutations of the cosets of the underlying subgroup. There are also constructions of complete mappings \(f\) such that both \(f\) and \(f+\mathrm{id}\) permute the nonzero field elements in one cycle [2105.00140].

Applications extend beyond permutation theory. Generalized cyclotomic mappings have been used to construct binary linear codes with few weights and bent functions, and recent bent-function work uses piecewise branch functions of Dillon, Niho, and Kasami type on multiplicative or additive cosets to obtain new explicit infinite families together with duals and EA-inequivalence observations [2503.06654; 2310.01317]. This suggests that fiber-distribution results, such as many-to-one classifications, are relevant not only as standalone algebraic statements but also as structural constraints on code weights, Walsh spectra, and related cryptographic design parameters.

Open problems remain conspicuous. Complete many-to-one characterizations are known for \(\ell\le 3\), complete \(2\)-to-\(1\) classifications are known for arbitrary \(\ell\mid(q-1)\), and the equal-gcd case is settled for general \(m\). For larger index and general \(m\ge 3\), however, a full classification is still unavailable. Further directions explicitly mentioned in the literature include extensions to functional-graph and cycle-structure analysis for many-to-one mappings, more applications in codes and cryptography, and generalizations beyond the subgroup \(U_{q+1}\) [2503.06654].

Source: https://www.emergentmind.com/topics/generalized-cyclotomic-mappings