---
title: Cyclic Subspace Codes
url: https://www.emergentmind.com/topics/cyclic-subspace-codes
type: topic
---

# Cyclic Subspace Codes

Cyclic subspace codes are subspace codes in \(\mathcal{P}_q(n)\) that are closed under the multiplicative action of \(\mathbb{F}_{q^n}^*\) on subspaces of \(\mathbb{F}_{q^n}\), viewed as an \(n\)-dimensional vector space over \(\mathbb{F}_q\). Equivalently, they are unions of multiplicative orbits, and the special case of a single orbit is a single-orbit cyclic subspace code, or orbit code. Within random network coding, this class is valued for the combination of rigid algebraic structure and nontrivial code parameters, and the modern theory connects orbit size, stabilizer subfields, minimum distance, intersection patterns, subspace polynomials, Sidon spaces, and message encoding into a coherent framework [1404.7739].

## 1. Ambient model, cyclic action, and basic terminology

The standard ambient space is the projective space
\[
\mathcal P_q(n)=\bigcup_{k=0}^n \mathcal G_q(n,k),
\]
where \(\mathcal G_q(n,k)\) is the Grassmannian of all \(k\)-dimensional \(\mathbb F_q\)-subspaces of \(\mathbb F_q^n\). In the extension-field model, \(\mathbb F_q^n\) is identified with \(\mathbb F_{q^n}\), and scalar multiplication by \(\alpha\in\mathbb F_{q^n}^*\) acts on a subspace \(U\) by
\[
\alpha U=\{\alpha u:u\in U\}.
\]
A cyclic subspace code is any code closed under this action, so it is a union of sets of the form
\[
\operatorname{Orb}(U)=\{\alpha U:\alpha\in\mathbb F_{q^n}^*\}.
\]
When the code consists of exactly one such orbit, it is a single-orbit cyclic subspace code; in related literature this is also called a cyclic orbit code [1608.03215].

The metric is the subspace distance
\[
d_S(U,V)=\dim(U)+\dim(V)-2\dim(U\cap V).
\]
For constant-dimension codes, where \(U,V\in \mathcal G_q(n,k)\),
\[
d_S(U,V)=2k-2\dim(U\cap V).
\]
Thus the geometry of cyclic constant-dimension codes is controlled entirely by the dimensions of intersections between a subspace and its cyclic shifts. In the extension-field formulation, this simple identity is the basis for orbit-size calculations, minimum-distance estimates, distance distributions, and explicit constructions [1403.1218].

A related but distinct viewpoint uses a cyclic subgroup of \(\mathrm{GL}_n(\mathbb F_q)\). If \(U\in \mathcal G_q(n,k)\) and \(G=\langle P\rangle\le \mathrm{GL}_n(\mathbb F_q)\), then
\[
U\langle P\rangle=\{UP^i\}
\]
is a cyclic orbit code in the linear-group sense. The extension-field model and the matrix model are compatible through companion-matrix realizations of irreducible or primitive polynomials, and much of the finite-field theory of cyclic subspace codes may be regarded as the irreducible cyclic-subgroup case of orbit-code theory [1409.0610].

## 2. Stabilizers, subfields, orbit size, and the “best friend”

For a subspace \(U\subseteq \mathbb F_{q^n}\), the stabilizer is
\[
\operatorname{Stab}(U)=\{\alpha\in\mathbb F_{q^n}^*:\alpha U=U\}.
\]
A standard fact used throughout the subject is that \(\operatorname{Stab}(U)\cup\{0\}\) is a subfield of \(\mathbb F_{q^n}\). Hence
\[
\operatorname{Stab}(U)\cup\{0\}=\mathbb F_{q^t}
\]
for some divisor \(t\mid \gcd(\dim_{\mathbb F_q}(U),n)\). Equivalently, \(U\) is naturally an \(\mathbb F_{q^t}\)-vector space, and orbit-stabilizer gives
\[
|\operatorname{Orb}(U)|=\frac{q^n-1}{q^t-1}.
\]
The case \(t=1\) is the full-length case:
\[
|\operatorname{Orb}(U)|=\frac{q^n-1}{q-1},
\]
while \(t>1\) yields a degenerate orbit [2407.02200].

The same structure is encoded by the largest subfield over which \(U\) is linear. In the primitive setting, this subfield is called the best friend of \(U\), and it coincides with \(\operatorname{Stab}(U)\cup\{0\}\). If the best friend is \(\mathbb F_{q^r}\), then \(\operatorname{Stab}(U)=\mathbb F_{q^r}^*\) and
\[
|\operatorname{Orb}(U)|=\frac{q^n-1}{q^r-1}.
\]
Moreover, if \(k=rt\) and \(s\) is the maximum \(\mathbb F_{q^r}\)-dimension of a nontrivial intersection \(U\cap \alpha U\), then
\[
d_S(\operatorname{Orb}(U))=2r(t-s).
\]
In particular,
\[
2r\le d_S(\operatorname{Orb}(U))\le 2k,
\]
so the best friend simultaneously controls orbit cardinality and the granularity of possible intersection dimensions [1403.1218].

The extreme case \(d_S=2k\) occurs exactly for spread-type subfield orbits. If \(k\mid n\) and \(U=\mathbb F_{q^k}\), then
\[
|\operatorname{Orb}(U)|=\frac{q^n-1}{q^k-1},
\qquad
d_S(\operatorname{Orb}(U))=2k.
\]
Conversely, in the primitive cyclic-orbit setting, maximum distance \(2k\) characterizes this subfield case. This makes spread orbits the canonical high-distance, small-cardinality extreme, while full-length orbits sit at the opposite size extreme [1912.05522].

## 3. Intersection geometry, distance distributions, and arithmetic restrictions

For a constant-dimension orbit code \(\operatorname{Orb}(U)\subseteq \mathcal G_q(n,k)\), the distance distribution relative to \(U\) can be written as
\[
\delta_i=\bigl|\{\alpha U\in \operatorname{Orb}(U): d_S(U,\alpha U)=i\}\bigr|.
\]
The associated intersection distribution is
\[
\mathcal O_i(U)=\{\alpha U\in \operatorname{Orb}(U): \dim(U\cap \alpha U)=i\},
\qquad
\lambda_i=|\mathcal O_i(U)|.
\]
Since
\[
d_S(U,\alpha U)=2(k-\dim(U\cap \alpha U)),
\]
the two distributions are equivalent:
\[
\delta_{2(k-i)}=\lambda_i.
\]
Thus the distance-distribution problem for a single-orbit cyclic subspace code is exactly the problem of counting intersection dimensions between a fixed subspace and its scalar shifts [2407.02200].

The modern study of these distributions begins with a structural reinterpretation via fraction sets and multiplicity counting. For full-length orbit codes of minimum distance \(2k-2\), equivalently for orbits generated by Sidon spaces, the entire distance distribution is universal: it depends only on \(q\), \(n\), and \(k\), not on the specific generating subspace. If
\[
Q=\frac{q^k-1}{q-1}\cdot\frac{q^k-q}{q-1},
\]
then exactly \(Q\) codewords meet a fixed codeword in a line, all remaining codewords meet it trivially, and no other nontrivial intersection dimension occurs. This universality is one of the sharpest general theorems known for cyclic orbit codes [1912.05522].

Outside the optimal Sidon-space regime, the distribution becomes more delicate, but arithmetic regularities persist. For full-length orbit codes, an elementary invariance argument yields that each intersection class is partitioned into packets of size \(q\), hence each \(\lambda_t\) is a multiple of \(q\). If \(n\) is odd, stronger packet decompositions show that each \(\lambda_t\) is a multiple of \(q(q+1)\). When \(n\) is even, degree-\(2\) elements of \(\mathbb F_{q^n}\) create an exceptional phenomenon tied to copies of \(\mathbb F_{q^2}\) contained in \(U\): if \(U\) contains
\[
\frac{q^{2m}-1}{q^2-1}
\]
distinct cyclic shifts of \(\mathbb F_{q^2}\), then
\[
\lambda_{2m}=q+r\,q(q+1)
\]
for some nonnegative integer \(r\), while every other \(\lambda_i\) is a multiple of \(q(q+1)\). In the general stabilizer case \(\operatorname{Stab}(U)=\mathbb F_{q^t}^*\), the same pattern lifts to \(q^t\): if \(n/t\) is odd, then every \(\lambda_i\) is a multiple of \(q^t(q^t+1)\); if \(n/t\) is even and \(U\) contains
\[
\frac{q^{2tm}-1}{q^{2t}-1}
\]
distinct cyclic shifts of \(\mathbb F_{q^{2t}}\), then
\[
\lambda_{2tm}=q^t+s\,q^t(q^t+1),
\]
and all remaining \(\lambda_i\) are multiples of \(q^t(q^t+1)\) [2407.02200].

This arithmetic perspective shows that cyclicity constrains much more than minimum distance. It organizes the full local geometry of an orbit into congruence classes, inverse pairings, and exceptional packets attached to subfield configurations, especially \(\mathbb F_{q^2}\) or \(\mathbb F_{q^{2t}}\).

## 4. Algebraic construction methods: subspace polynomials, Frobenius maps, and quasi-cyclic variants

A principal algebraic tool is the subspace polynomial. For a \(k\)-dimensional \(\mathbb F_q\)-subspace \(V\subseteq \mathbb F_{q^n}\),
\[
P_V(x)=\prod_{v\in V}(x-v)
\]
is a monic linearized polynomial of the form
\[
P_V(x)=x^{[k]}+\sum_{i=0}^{k-1} a_i x^{[i]},
\qquad
x^{[i]}=x^{q^i}.
\]
This representation is faithful: \(U=V\) if and only if \(P_U=P_V\). It converts cyclic shifts, Frobenius shifts, and intersection problems into coefficient computations. The cyclic action transforms coefficients by
\[
P_{\alpha V}(x)=\alpha^{[k]}P_V(\alpha^{-1}x)
= x^{[k]}+\sum_j \alpha^{[k]-[j]}a_jx^{[j]},
\]
while the \(s\)-th Frobenius shift satisfies
\[
P_{F^s(V)}(x)=x^{[k]}+\sum_j F^s(a_j)x^{[j]}.
\]
A simple but powerful invariant is the gap between the largest and second-largest nonzero \(q\)-degrees: if
\[
P_V(x)=x^{[k]}+\sum_{j=0}^{i} a_jx^{[j]},\qquad a_i\neq 0,
\]
then \(\operatorname{gap}(V)=k-i\), and one obtains the distance bound
\[
d(U,V)\ge 2\min(\operatorname{gap}(U),\operatorname{gap}(V)).
\]
This is the basis of many explicit full-length constructions [1404.7739].

The same framework yields explicit cyclic families. If
\[
x^{[k]-1}+x^{[1]-1}+1
\]
is irreducible over \(\mathbb F_q\) and its degree divides \(n\), then
\[
x^{[k]}+x^{[1]}+x
\]
is a subspace polynomial over \(\mathbb F_{q^n}\), and the orbit
\[
C=\{\alpha V:\alpha\in\mathbb F_{q^n}^*\}
\]
has
\[
|C|=\frac{q^n-1}{q-1},
\qquad
d_{\min}\ge 2k-2.
\]
For fixed \(q\) and \(k\), this produces such one-orbit full-length cyclic codes for infinitely many \(n\). Frobenius shifts can also be combined to obtain multiple full-length orbits of total size
\[
n\cdot \frac{q^N-1}{q-1}
\]
and minimum distance \(2k-2\). At the opposite end of the orbit-size spectrum, if \(d\mid \gcd(n,k)\), the family of all \(k\)-subspaces that are also \(\mathbb F_{q^d}\)-subspaces forms a cyclic code
\[
C_d\subseteq \mathcal G_q(n,k)
\]
of size
\[
|C_d|=\binom{n/d}{k/d}_{q^d}
\]
and minimum distance \(2d\) [1404.7739].

A different algebraic line generalizes cyclicity itself. An \(m\)-quasi-cyclic subspace code is invariant under
\[
V\mapsto \alpha^m V,
\qquad
m\mid (q^n-1).
\]
Its \(m\)-quasi orbit has size
\[
\left|\{\alpha^mV:\alpha\in \mathbb F_{q^n}^*\}\right|
=
\frac{1}{m}\left(\frac{q^n-1}{q^t-1}\right)
\]
for some divisor \(t\mid n\). The same two core tools of cyclic-code theory—subspace polynomials and Frobenius mappings—remain effective in this setting, and one obtains one-quasi-orbit and multi-quasi-orbit construction methods parallel to the ordinary cyclic case [1608.03215].

A further extension replaces one seed Sidon space by sums of several Sidon spaces. If \(U\) and \(V\) are Sidon spaces satisfying
\[
(U+V)^2=U^2\oplus UV\oplus V^2
\]
and
\[
\dim(U\cap aV)\le 1
\qquad
\text{for every }a\in \mathbb F_{q^n}^*,
\]
then \(U+V\) is again a Sidon space. More generally, if \(V_1,\dots,V_m\) are Sidon spaces with
\[
\left(\sum_{i=1}^m V_i\right)^2
=
\bigoplus_{1\le i\le j\le m}V_iV_j
\]
and pairwise scalar intersections of dimension at most \(1\), then \(\sum_i V_i\) is Sidon. This criterion produces many new full-length one-orbit cyclic codes of distance \(2K-2\) by turning smaller Sidon building blocks into larger ones [2105.12520].

## 5. Sidon spaces, large families, classification results, and asymptotic constructions

A \(k\)-dimensional subspace \(U\subseteq \mathbb F_{q^n}\) is a Sidon space if for all nonzero \(a,b,c,d\in U\),
\[
ab=cd \implies \{a\mathbb F_q,b\mathbb F_q\}=\{c\mathbb F_q,d\mathbb F_q\}.
\]
This condition is equivalent to the orbit code \(\operatorname{Orb}(U)\) being full-length and having minimum distance \(2k-2\). It is also the central bridge between multiplicative combinatorics and cyclic subspace coding. In the multi-orbit setting, compatible families of Sidon spaces give cyclic subspace codes obtained as unions of full-length orbits while preserving minimum distance \(2k-2\) [1912.05522].

Large families of such unions are now known. One construction yields cyclic \((n,2k-2,k)_q\)-codes of size
\[
r\left(\left\lceil \frac{n}{2rk}\right\rceil -1\right)
\left((q^k-1)^r(q^n-1)+\frac{(q^k-1)^{r-1}(q^n-1)}{q-1}\right)
\]
for \(k\mid n\), \(r\ge 2\), and \(n\ge (2r+1)k\). A second construction, for \(n=2rk\), gives size
\[
\left\lfloor \frac{(r-1)(q^k-2)(q^k-1)^{r-1}(q^n-1)}{2} \right\rfloor.
\]
In the case \(n=4k\), the latter family is within a factor of \(\frac12+o_k(1)\) of the sphere-packing bound as \(k\) goes to infinity. These constructions are built from explicit parameterized families of Sidon spaces and controlled pairwise products among them [2305.15627].

At low dimension, one-orbit cyclic codes admit explicit classifications. In \(\mathcal G_q(n,3)\), three-dimensional one-orbit cyclic subspace codes fall into three families: the spread case \(\operatorname{Orb}(\mathbb F_{q^3})\), the optimum-distance full-length case with minimum distance \(4\), and the full-length minimum-distance-\(2\) case. The classification is organized by \(\dim_{\mathbb F_q}(S^2)\), together with semilinear invariants such as \(\delta_t(S)\) and \(w_t(S)\), and it isolates explicit normal forms including
\[
S=(1,\lambda,\lambda^2)_{\mathbb F_q}
\qquad\text{and}\qquad
S=\mathbb F_{q^2}+(\mu)_{\mathbb F_q}
\]
for the distance-\(2\) families [2405.01652].

A different classification problem concerns full weight spectrum one-orbit cyclic subspace codes, namely those realizing all possible nonzero subspace distances. These are completely classified: every such code is, up to cyclic shift, of exactly one of two types. The first is the polynomial-basis family
\[
S=\langle 1,\lambda,\lambda^2,\dots,\lambda^{k-1}\rangle_{\mathbb F_q},
\]
subject to explicit bounds on \(k\) relative to \([\mathbb F_q(\lambda):\mathbb F_q]\); the second is a quadratic-extension family
\[
S=\langle 1,\lambda,\lambda^2,\dots,\lambda^{\ell-1}\rangle_{\mathbb F_{q^2}\oplus \lambda^\ell \mathbb F_q,
\qquad
k=2\ell+1,
\]
with explicit restrictions on \([\mathbb F_{q^2}(\lambda):\mathbb F_{q^2}]\). This result shows that full weight spectrum is highly rigid in the one-orbit cyclic setting [2405.19911].

Recent asymptotic progress comes from a recursive product construction across towers of field extensions. If
\[
\mathcal C_1\subseteq \mathcal G_q(m,k),
\qquad
\mathcal C_2\subseteq \mathcal G_q(n,m),
\qquad
n=rm,
\]
with
\[
d(\mathcal C_1)=2k-2,
\qquad
d(\mathcal C_2)=2m-2,
\]
then one can form
\[
\mathcal C_2\odot \mathcal C_1\subseteq \mathcal G_q(n,k)
\]
with
\[
d(\mathcal C_2\odot \mathcal C_1)=2k-2,
\qquad
|\mathcal C_2\odot \mathcal C_1|=|\mathcal C_2|\,|\mathcal C_1|.
\]
Iterating this operation produces large multi-orbit cyclic subspace codes for composite \(r=n/k\). In particular, for \(r=3^e\), the resulting families are asymptotically optimal with respect to the Johnson type bound II for infinitely many \(r\); for \(r=2^e\) and \(r=2^e3^{e_1}\), the constructions approach Johnson within factor \(1/2^e\) [2507.09290].

## 6. Encoding, applications, and related generalizations

For cyclic orbit codes, message encoding is naturally indexed by exponents. If
\[
C=U\langle P\rangle,
\]
with orbit size \(\operatorname{ord}_U(P)\), then the message set is
\[
\{0,\dots,\operatorname{ord}_U(P)-1\},
\]
and the encoder is
\[
i\longmapsto UP^i.
\]
In the primitive case, this can be implemented in complexity \(\mathcal O_q(n^3)\). The inverse problem is fundamentally harder: after decoding to \(P^i\), message retrieval becomes a discrete logarithm problem in the multiplicative group generated by the field element corresponding to \(P\). Using Pohlig–Hellman, if
\[
q^n-1=\prod_{i=1}^r p_i^{e_i},
\]
the retrieval complexity is
\[
\mathcal{O}_{q}\Big( n^3\log_2 q  \sum_{i=1}^r e_i +  n^2 \sum_{i=1}^r  e_i\sqrt{p_i} \Big).
\]
A hybrid method avoids this bottleneck in spread-orbit cases by transferring encoding and retrieval through semilinear isometries to Desarguesian spread codes [1409.0610].

Cyclic subspace codes also serve as ingredients in other coding theories. A multi-orbit cyclic subspace code
\[
C=\bigcup_{i=1}^r C_{U_i}\subseteq \mathcal G_q(m,k)
\]
with minimum distance \(d\) yields, via affine translates of the representatives \(U_i\), an optical orthogonal code of parameters
\[
(q^m-1,q^k,q^{k-d/2})
\]
and size
\[
r\frac{q^{m-k}-1}{q-1}.
\]
In the important case \(d=2k-2\), the resulting parameters are
\[
(q^m-1,q^k,q).
\]
This construction gives a multiplicative finite-field route from cyclic subspace codes to OOCs, distinct from additive character-sum and purely combinatorial approaches [2405.18996].

The orbit-code viewpoint also extends from single subspaces to flags. A cyclic orbit flag code is an orbit of a flag
\[
\mathcal F=(\mathcal F_1,\dots,\mathcal F_r)
\]
under a cyclic subgroup, and each projected component
\[
(\operatorname{Orb}_\beta(\mathcal F))_i=\operatorname{Orb}_\beta(\mathcal F_i)
\]
is itself a cyclic orbit subspace code. The stabilizer/best-friend formalism survives at flag level, but new compatibility conditions arise because the flag stabilizer is the intersection of the stabilizers of its coordinates. Thus ordinary cyclic subspace codes reappear as the one-coordinate shadow of a broader orbit theory on nested subspaces [2102.00867].

Taken together, these developments present cyclic subspace codes as a mature algebraic domain: they admit exact orbit-size formulas via stabilizer subfields, refined local invariants via intersection and distance distributions, explicit constructions through subspace polynomials and Sidon spaces, classification theorems in special regimes, recursive asymptotic constructions, concrete encoding maps, and nontrivial applications outside network coding.

Source: https://www.emergentmind.com/topics/cyclic-subspace-codes