Flag Codes with Longer Type Vectors

• Flag codes with longer type vectors are algebraic network codes defined by nested subspaces that extend classical constructions to maximize flag distance.
• They leverage group actions, MRD-lifting, and spread-based methods to integrate combinatorial and algebraic techniques for optimal code performance.
• The constructions ensure cardinality consistency across projections, achieving precise trade-offs between code size and minimum flag distance.

Flag codes with longer type vectors are algebraic network codes formed from sequences of nested subspaces of vector spaces over finite fields, where the type vector—the list of subspace dimensions in each flag—is extended beyond classical choices to maximize code size and optimize or nearly optimize flag distance. This subject generalizes the paradigm of constant-dimension network coding to flexible multi-hop settings, uniting combinatorial, group-theoretic, and rank-metric constructions. The recent research frontier crystallizes around cardinality-consistent constructions, spread-based and MRD-based liftings, orbital/cyclic orbits, and precise trade-offs as the type vector lengthens.

1. Core Definitions and Flag Distance Framework

Let $\mathbb{F}_q$ be a finite field and $V = \mathbb{F}_q^n$. A flag of type $\vec{t} = (t_1 < t_2 < \cdots < t_r)$ is a strictly ascending sequence of subspaces: $$0 \subsetneq F_1 \subsetneq F_2 \subsetneq \cdots \subsetneq F_r \subsetneq V, \qquad \dim F_i = t_i\ .$$ The flag space $\mathcal{A}_q(\vec{t}, n)$ is the set of all flags of this type. The flag code $\mathcal{C} \subseteq \mathcal{A}_q(\vec{t}, n)$ is a collection of such flags, with the flag distance defined by

$$d_f(\mathcal{F}, \mathcal{F'}) = \sum_{i=1}^r d_S(F_i, F'_i)$$

where $d_S(U, W) = \dim U + \dim W - 2\dim(U \cap W)$ is the classic subspace distance. The maximum possible flag distance for type $\vec{t}$ is

$$D^{(\vec{t}, n)} = 2\left( \sum_{t_i \le \lfloor n/2 \rfloor} t_i + \sum_{t_i > \lfloor n/2 \rfloor} (n - t_i) \right)\ .$$

A code is optimum-distance if $d_f(\mathcal{C}) = D^{(\vec{t}, n)}$. Codes are termed cardinality-consistent if all their projections at every $t_i$ (i.e., the set of all $F_i$ for $F \in \mathcal{C}$) have the same cardinality as $\mathcal{C}$, ensuring structural alignment across projected codes (Jia et al., 13 Jan 2026).

2. Constructions with Extended Type Vectors

Cardinality-consistent flag codes with longer type vectors generalize admissible types to include additional intermediate subspace dimensions, breaking from the restriction to just $\{1,\ldots,k, n-k, \ldots, n-1\}$ and filling in dimensions between $k$ and $n-k$ as permitted by group- and spread-theoretic constraints (Jia et al., 13 Jan 2026, Liu et al., 2023).

• Full-admissible type: $$\vec{t}_\text{full} = (1, 2, \ldots, k, n-k, \ldots, n-1)$$, building on classical results for spread-based codes (Alonso-González et al., 2020, Alonso-González et al., 2020).
• Extended types (longer type vectors):

$$\vec{t}_\text{long} = (1, 2, \ldots, k+h,\, 2k+h,\, 3k+h,\ldots,(s-2)k+h,\, n-k,\ldots, n-1)$$

where $n = sk + h$, $s \ge 2$, $0 \le h < k$.

These longer sequences incorporate additional intermediate dimensionalities while guaranteeing the projected codes maintain sufficient minimum subspace distance, typically at least $2k$ (Jia et al., 13 Jan 2026).

For each $1 \le i \le s-1$, one constructs companion matrices $P_i$ of primitive polynomials of appropriate degree, and defines group actions $$G_i = \langle \operatorname{diag}(I_{(s-i-1)k}, I_k, P_i)\rangle$$ on seed matrices whose row-spaces correspond to the $k$-subspaces. Full flags are built from these orbits by reading off the first $j$ rows ($j=1,\ldots,n-1$). The approach unifies and generalizes cyclic orbit and spread-constructed flag codes.

3. Cardinality and Distance Properties

Both full-admissible and extended-type constructions attain the same code size: $|\mathcal{C}| = \sum_{i=1}^{s-1} q^{ik+h} + 1$ For instance, with $n=sk+h$:

• The $k$-projected code is a partial $k$-spread of size $\sum_{i=1}^{s-1}q^{ik+h} +1$.
• The minimum flag distance for the extended type vector satisfies

$$d_f(\mathcal{C}_\text{long}) = \sum_{t_i\le k} 2t_i + \sum_{k+1\le t_i\le n-k-1}2k + \sum_{t_i\ge n-k} 2(n-t_i)$$

This construction is optimum when $k > (q^h-1)/(q-1)$, attaining the upper bound determined by the maximum size of a $0$-intersecting equidistant code in $G_q(n,k)$. The codes have the property that every projection $\mathcal{C}_i$ (where $t_i\in\vec{t}$) has size $|\mathcal{C}|$, i.e., perfect alignment across dimensions (Jia et al., 13 Jan 2026, Liu et al., 2023).

4. Algebraic and Group-Theoretic Structures

The cyclic orbit and group-action viewpoint is central to these constructions:

• Each orbit of a seed subspace under $G_i$ is a cyclic orbit code in the Grassmannian, and the corresponding flags form a cyclic orbit flag code.
• Codes are constructed as the union of $s-1$ orbits (from matrices $A_i$) and a small number of exceptional flags (from $B_i$ and shift matrices).
• The companion-matrix embedding approach ties the spread-theoretic perspective (planar or partial spreads in $G_q(k, n)$) to the explicit algebraic and group-theoretic context of flag code construction (Alonso-González et al., 2021, Alonso-González et al., 2020).
• In the context of cyclic orbit flag codes, the largest subfield over which all the flag's subspaces are vector spaces ("best friend") determines cardinality and minimal distances, and constrains possible type vectors that yield optimum distance (Alonso-González et al., 2021).

5. Methodologies for Code Construction

Several frameworks are available for constructing such codes:

• MRD-Lifting: Uses Maximum Rank Distance (MRD) codes in block-matrix constructions, ensuring the projected codes remain constant-dimension codes of prescribed distance. The explicit method glues together MRD blocks and uses invertible-matrix gadgets to extract full flags with desired properties (Liu et al., 2023).
• Spread-based field reduction: Uses perfect matchings in bipartite graphs to relate lines and hyperplanes (or higher-dimensional spreads) in extension fields, then applies field reduction to realize the code in the base field, preserving spread properties through dimension collapse (Alonso-González et al., 2020).
• Block-matrix and companion-matrix layers: Sandwiches flag code layers corresponding to partial spreads, a companion matrix (field extension), and further partial spreads, to interpolate between full and extended type vectors while controlling minimum flag distance (Han et al., 18 Jun 2025).
• Orbit and stabilizer methods: Exploits group orbits under general linear or Singer cycles to produce codes of the exact required cardinality and symmetry, specifically handling cyclic orbit codes and their best friend subfields (Alonso-González et al., 2021, Alonso-González et al., 2020).

6. Distance-Vector and Cardinality Bounds for Large Type Vectors

Lengthening the type vector increases the number of distinct projections and imposes complex interlacing conditions on achievable flag distance. The distance vector formalism of (Alonso-González et al., 2021) characterizes, for fixed $(n,q,\vec{t})$, which tuples of subspace distances can be realized simultaneously, leading to explicit combinatorial cardinality bounds. As $r$ increases, achieving larger minimum distances (closer to the upper bound) forces greater code sparsity, but allows finer-grained projections and controls the trade-off between code size and minimum distance. For large $r$, the code size upper bound drops rapidly with increased $d_f$ due to disjointness constraints among flag projections.

Table: Key Relations for Flag Codes with Long Type Vectors

Property	Formula/Statement	Reference
Maximum flag dist.	$$D^{(\vec{t}, n)} = 2(\sum_{t_i \le \lfloor n/2\rfloor} t_i + \sum_{t_i > \lfloor n/2\rfloor} (n-t_i))$$	(Jia et al., 13 Jan 2026)
Code cardinality	$|\mathcal{C}| = \sum_{i=1}^{s-1} q^{ik + h} + 1$	(Jia et al., 13 Jan 2026)
Admissible types	$t_j \le k$ or $t_j \ge n-k$ for all $j$ (with inserted intermediates in longer type cases)	(Jia et al., 13 Jan 2026, Alonso-González et al., 2020)
Optimum size bound	$(q^n - q^{k+h})/(q^k-1) + 1$ under $k > (q^h-1)/(q-1)$	(Jia et al., 13 Jan 2026, Liu et al., 2023)

7. Examples and Parameter Choices

Concrete constructions at small parameters demonstrate the flexibility:

• $s=2$, $n=2k+h$, $t_\text{full} = (1,2,\ldots,k,n-k,\ldots,n-1)$. The code size is $q^{k+h}+1$, with optimum flag distance $2k(k+h)$.
• $s=3$, $h=1$, $n=3k+1$, $$t_\text{long} = (1,2,\ldots,k+1,2k+1,n-k,\ldots,n-1)$$, code size $q^{k+1}+q^{2k+1}+1$, minimum distance $2k(k+2)$ (Jia et al., 13 Jan 2026). MRD-based flag codes likewise support type vectors ranging from all dimensions up to partial choices, with corresponding optimum or nearly optimum distances (Liu et al., 2023).

---

These results illustrate the deep interplay between algebraic construction, combinatorial geometry, and group actions in achieving cardinality-optimal and distance-optimal flag codes as the type vector grows, including for full flags and extended/intermediate flag types. The mathematical infrastructure developed supports application-driven design and precise parameter optimization for robust network coding in multishot and adversarial channels.