---
title: Grassmann Codes Overview
url: https://www.emergentmind.com/topics/grassmann-codes
type: topic
---

# Grassmann Codes Overview

Grassmann codes are linear projective codes attached to finite Grassmannians through the Plücker embedding. For an \(m\)-dimensional vector space \(V\) over \(\mathbb F_q\) and \(1\le \ell<m\), the Grassmannian \(G_{\ell,m}\) consists of the \(\ell\)-dimensional subspaces of \(V\), and its Plücker image inside \(\mathbb P(\bigwedge^\ell V)\) forms a projective system whose associated code is denoted \(C(\ell,m)\). With \(V=\mathbb F_q^m\), one has
\[
n=\binom{m}{\ell}_q,\qquad k=\binom{m}{\ell},\qquad \delta=\ell(m-\ell),
\]
where \(\binom{m}{\ell}_q\) is the Gaussian binomial coefficient, and \(C(\ell,m)\) is a linear \([n,k]_q\)-code with minimum distance
\[
d_1(C(\ell,m))=q^\delta.
\]
The subject lies at the intersection of multilinear algebra, projective geometry, and coding theory, and it has generated a substantial literature on higher weights, automorphisms, decoding, and polar or affine variants [0710.5161].

## 1. Classical construction and basic structure

A nonzero vector \(\omega\in \bigwedge^\ell V\) is decomposable if
\[
\omega=v_1\wedge\cdots\wedge v_\ell
\]
for some \(v_1,\dots,v_\ell\in V\). The Grassmannian \(G_{\ell,m}\) is the variety of \(\ell\)-dimensional subspaces of \(V\), and the Plücker embedding sends an \(\ell\)-plane \(\langle v_1,\dots,v_\ell\rangle\) to the point of \(\mathbb P(\bigwedge^\ell V)\) represented by \(v_1\wedge\cdots\wedge v_\ell\). Accordingly, the image of \(G_{\ell,m}\) is exactly the projectivized locus of decomposable \(\ell\)-vectors [0710.5161].

Over \(V=\mathbb F_q^m\), choosing representatives \(w_1,\dots,w_n\in \bigwedge^\ell V\) of the \(\mathbb F_q\)-rational Plücker points gives the evaluation model
\[
\tau:\bigwedge^{m-\ell}V\longrightarrow \mathbb F_q^n,\qquad \tau(\omega)=\bigl(w_1\wedge \omega,\dots,w_n\wedge \omega\bigr),
\]
using the canonical perfect pairing
\[
\bigwedge^\ell V\times \bigwedge^{m-\ell}V\to \bigwedge^mV\simeq \mathbb F_q.
\]
Since the Plücker embedding is nondegenerate, \(\tau\) is injective, and \(C(\ell,m)\) is defined as its image. In this model, minimum-weight codewords are characterized exactly by decomposability: the minimum distance is \(q^\delta\), codewords \(c\) with \(w_c\) decomposable attain this weight, and
\[
|c|=q^\delta \iff w_c \text{ is decomposable},
\]
while the number of minimum-weight codewords is
\[
A_1(C(\ell,m))=(q-1)n.
\]
Thus the smallest nonzero weights of Grassmann codes are controlled by the simplest multilinear objects in \(\bigwedge^{m-\ell}V\) [0710.5161].

This construction makes Grassmann codes simultaneously projective-system codes and evaluation codes on a homogeneous variety. A plausible implication is that many of their structural properties can be reformulated either in the language of exterior algebra or in the geometry of linear sections of the Grassmannian.

## 2. Higher weights, decomposable subspaces, and linear sections

A central theorem in the subject is the classification of subspaces of \(\bigwedge^\ell V\) all of whose nonzero vectors are decomposable. For \(\omega\in \bigwedge^\ell V\), define
\[
V_\omega:=\{v\in V: v\wedge \omega=0\}.
\]
Then
\[
\omega \text{ is decomposable } \Longleftrightarrow \dim V_\omega=\ell.
\]
For decomposable \(\omega_1,\omega_2\in \bigwedge^\ell V\), the sum \(\omega_1+\omega_2\) is decomposable exactly when the associated \(\ell\)-spaces intersect in dimension \(\ell-1\). This local criterion leads to the structure theorem that a subspace of \(\bigwedge^\ell V\) is decomposable if and only if it is close, where close subspaces are of type I or type II and are given by explicit wedge-product normal forms [0710.5161].

Projectively, this means that the projective linear subspaces contained in the Plücker-embedded Grassmannian are exactly the projectivizations of close subspaces. If
\[
\mu:=\max\{\ell,\;m-\ell\}+1,
\]
then \(\bigwedge^\ell V\) has a decomposable subspace of dimension \(r\) iff \(r\le \mu\), and the maximum possible projective dimension of a linear subspace contained in the Grassmannian is
\[
\mu-1=\max\{\ell,m-\ell\}.
\]
This translates directly into higher-weight formulas for \(C(\ell,m)\) [0710.5161].

If \(\mathcal E\subset \bigwedge^\ell V\) has codimension \(s\), define
\[
g_s(\ell,m):=\max\{g(\mathcal E): \mathcal E\subset \bigwedge^\ell V \text{ of codimension } s\},
\]
where \(g(\mathcal E)\) counts Grassmann points on the section. Then
\[
d_s(C(\ell,m))=n-g_s(\ell,m), \qquad s=1,\dots,k.
\]
For
\[
0\le r\le \mu,
\]
the first and terminal parts of the weight hierarchy are
\[
d_r(C(\ell,m))=q^\delta+q^{\delta-1}+\cdots+q^{\delta-r+1},
\]
and
\[
d_{k-r}(C(\ell,m))=n-(1+q+\cdots+q^{r-1}).
\]
For \(\ell=2\) and \(m\ge 4\), one further higher weight beyond each end is also determined:
\[
d_{k-p-1}(C(2,m))=n-\bigl(1+q+\cdots+q^{p-1}+q^2\bigr),
\]
and
\[
d_{p+1}(C(2,m)) = q^\delta+q^{\delta-1}+2q^{\delta-2}+q^{\delta-3}+\cdots+q^{\delta-p+1},
\]
where \(p=m-1\) and \(\delta=2(m-2)\) [0710.5161].

The same work also introduces a generalized Griesmer-Wei bound. For a linear \([n,k]_q\)-code \(C\), with \(A_r(C)\) the maximal number of minimum-weight codewords in an \(r\)-dimensional subcode and \(e(C)\) the second smallest positive weight, one has
\[
d_r(C)\ge \frac{d(C)A_r(C)+e(C)\bigl(q^r-1-A_r(C)\bigr)}{q^r-q^{r-1}}.
\]
Specialized to Grassmann codes, this refined lower bound is effective because decomposable and near-decomposable subspaces control \(A_r(C)\) in geometric terms [0710.5161].

## 3. Symmetry and automorphism groups

The automorphism theory of Grassmann codes is governed by Chow’s theorem on line-preserving bijections of Grassmannians. For the Plücker projective system \(\mathcal P=G(\ell,m)\subset \mathbb P(\bigwedge^\ell F^m)\),
\[
\Aut(\mathcal P)\simeq \P\Gamma L(m,F)\quad \text{if } m\neq 2\ell,
\]
while in the balanced case
\[
\Aut(\mathcal P)\simeq \P\Gamma L(m,F)\rtimes_{-t}\mathbb Z/2\mathbb Z.
\]
The exceptional factor \(\mathbb Z/2\mathbb Z\) comes from the Hodge-star duality when \(m=2\ell\) [1204.5547].

At the code level, the full semilinear automorphism group is described by an explicit central extension \(\mathcal G\) of \(PGL(m,F)\times \mu_\lambda\). One has
\[
\Aut(C(\ell,m))=
\begin{cases}
\mathcal G\rtimes \Aut(F) & \text{if } m\neq 2\ell,\\[1ex]
\mathcal G\rtimes (\Aut(F)\times \mathbb Z/2\mathbb Z) & \text{if } m=2\ell,
\end{cases}
\]
and likewise
\[
\MAut(C(\ell,m))=
\begin{cases}
\mathcal G & \text{if } m\neq 2\ell,\\[1ex]
\mathcal G\rtimes \mathbb Z/2\mathbb Z & \text{if } m=2\ell.
\end{cases}
\]
Thus the automorphisms of the code are precisely the semilinear symmetries of the ambient Grassmannian, together with the duality involution in the self-complementary case [1204.5547].

The same geometric method yields automorphism groups for affine Grassmann codes and for Schubert divisor codes. For the big cell \(W_0\), the relevant symmetry group is a maximal parabolic subgroup rather than the full \(\P\Gamma L(m,F)\), and the permutation automorphism group of the affine Grassmann code is
\[
\PAut(C^{\mathbb A}(\ell,m))\simeq
\begin{cases}
P_{m-\ell,\ell}/F^\times & \text{if } m\neq 2\ell,\\[1ex]
P_{\ell,\ell}/F^\times\rtimes_{-\tilde t}\mathbb Z/2\mathbb Z & \text{if } m=2\ell.
\end{cases}
\]
This settles the permutation-group question for affine Grassmann codes and shows that Schubert divisor codes have the same full and monomial automorphism groups as affine Grassmann codes [1204.5547].

## 4. Decoding theory and algorithmic constructions

Two distinct decoding paradigms have been developed for Grassmann codes. The first is majority logic decoding based on point-line incidence. For the dual code \(C(\ell,m)^\perp\), the minimum distance is \(3\), and the support of a minimum-weight dual codeword consists of three points on a line in \(G_{\ell,m}\); conversely, any three points on a line support such a codeword. This identifies lines in the Grassmannian with weight-\(3\) parity checks [2001.06278].

Fixing a point \(P\in G_{\ell,m}\) and a complete flag through \(P\), every point \(Q\) of the Grassmannian determines a unique canonical path from \(P\) to \(Q\) among paths with strictly decreasing \(r\)-tuple and strictly increasing \(s\)-tuple. These canonical paths are then used to construct, for each \(1\le i\le \ell\), a family \(J_i\) of parity checks orthogonal on the coordinate \(P\), with
\[
|J_i|=\Bigl\lfloor \frac q2\Bigr\rfloor q^{i^2-i} \binom{\ell}{i}_q\binom{m-\ell}{i}_q.
\]
Taking
\[
J=\sum_{i=1}^{\ell} \Bigl\lfloor \frac q2\Bigr\rfloor^i q^{i^2-i} \binom{\ell}{i}_q\binom{m-\ell}{i}_q,
\]
one obtains one-step majority logic decoding up to
\[
\left\lfloor \frac J2\right\rfloor
\]
errors. The construction gives explicit parity checks orthogonal on every coordinate and leads to a quadratic-time majority logic decoder in the block length [2001.06278].

A second decoder is specific to \(C(2,m)\). Using the action of \(\mathbb F_{q^m}^\times\) on \(G_{2,m}\), the Grassmannian decomposes into multiplicative orbits. Projecting \(C(2,m)\) onto an orbit \(O_\beta\) gives a code \(C_{O_\beta}(2,m)\) that is a subcode of a Reed-Solomon code over \(\mathbb F_{q^m}\). For many such orbits, the projection already contains an information set of the parent Grassmann code. The resulting orbit-projection decoder combines a modified Peterson procedure on the projected code with information-set lifting, and for \(m\ge 4\) it corrects up to
\[
\left\lfloor \frac{d-1}{2}\right\rfloor
\]
errors for \(C(2,m)\), where
\[
d=q^{2(m-2)}.
\]
This reaches the full unique-decoding radius for \(C(2,m)\) [2008.12067].

## 5. Major variants and extensions

Affine Grassmann codes are obtained by evaluating linear combinations of minors of a generic \(\ell\times \ell'\) matrix on the full affine matrix space \(\mathbb A^\delta(\mathbb F_q)\), with
\[
m=\ell+\ell',\qquad \delta=\ell\ell'.
\]
They have parameters
\[
n=q^\delta,\qquad k=\binom{m}{\ell},\qquad d=q^{\delta-\ell^2}\prod_{i=0}^{\ell-1}(q^\ell-q^i),
\]
and may be viewed as variants of generalized Reed-Muller codes that are closely related to Grassmann codes. Their level-\(r\) versions satisfy
\[
d\bigl(C^{\mathbb A}(\ell,m;r)\bigr)=q^\delta\prod_{i=1}^r\left(1-\frac{1}{q^i}\right),
\]
their duals are described explicitly by non-forbidden reduced monomials and binomials, and
\[
d\bigl(C^{\mathbb A}(\ell,m;r)^\perp\bigr)=
\begin{cases}
3,& q>2,\\
4,& q=2\ \text{and}\ \ell'>1,
\end{cases}
\]
apart from trivial edge cases. Both \(C^{\mathbb A}(\ell,m;r)\) and the full dual \(C^{\mathbb A}(\ell,m)^\perp\) are generated by their minimum-weight codewords [0911.1298; 1107.3438].

Symplectic Grassmann codes \(W(n,k)\) arise from the Plücker embedding of the symplectic Grassmannian \(\Lambda_{n,k}\) of totally isotropic \(k\)-subspaces in a \(2n\)-dimensional symplectic space. Their parameters are
\[
N=\prod_{i=0}^{k-1}\frac{q^{2n-2i}-1}{q^{i+1}-1},\qquad
K=\binom{2n}{k}-\binom{2n}{k-2}.
\]
For the line case \(k=2\),
\[
d_{\min}(W(n,2))=q^{4n-5}-q^{2n-3},
\]
and for the Lagrangian case \(W(3,3)\),
\[
d_{\min}(W(3,3))=q^6-q^4.
\]
Lagrangian-Grassmannian codes are therefore a special class of Symplectic Grassmann codes rather than a separate species [1503.05456].

Line Hermitian Grassmann codes are defined from the Plücker image of the totally isotropic lines of a non-degenerate Hermitian polar space over \(\mathbb F_{q^2}\). They are subcodes of the Grassmann codes associated to the \(2\)-Grassmannian. Their dimension is
\[
K=\binom{m}{2},
\]
their length is
\[
N= \frac{(q^m+(-1)^{m-1})(q^{m-1}-(-1)^{m-1})(q^{m-2}+(-1)^{m-3})(q^{m-3}-(-1)^{m-3})}{(q^2-1)^2(q^2+1)},
\]
and the minimum distance is
\[
d_{\min}= \begin{cases}
q^{4m-12}-q^{3m-9}, & \text{if } m \text{ is odd},\\[2mm]
q^{4m-12}-q^{2m-6}, & \text{if } m=4,6,\\[2mm]
q^{4m-12}, & \text{if } m\ge 8 \text{ is even}.
\end{cases}
\]
Minimum-weight codewords are characterized geometrically via radicals of alternating forms and Hermitian cones, except for the exceptional small case \((m,q)=(5,2)\) [1706.10255].

Affine Hermitian Grassmann codes \(C^{\mathbb H}(\ell)\) evaluate minors of a generic Hermitian \(\ell\times\ell\) matrix on the affine space of Hermitian matrices over \(\mathbb F_{q^2}\). They have parameters
\[
\left[q^{\ell^2},\ \binom{2\ell}{\ell},\ q^{\ell^2}-q^{\ell^2-1}-q^{\ell^2-3}\right],
\]
the same length and dimension as the comparable affine Grassmann code, but a larger minimum distance. Their duals have minimum distance
\[
d\bigl(C^{\mathbb H}(\ell)^\perp\bigr)=
\begin{cases}
3,& q>2,\\
4,& q=2.
\end{cases}
\]
The proof introduces Hermitian-specific notions such as spread of a minor and spread-reduction via Hermitian congruence automorphisms [2110.08964].

## 6. Related frameworks and terminological distinctions

The term “Grassmann codes” is frequently used alongside several adjacent but distinct notions. One line of work studies projective or non-degenerate linear \([n,k]_q\)-codes as vertices of the Grassmann graph of \(k\)-subspaces of \(\mathbb F_q^n\). In that setting, the central objects are induced subgraphs such as \(\Pi(n,k)_q\) for projective codes or \(T(n,k)_q\) for non-degenerate codes, and the main questions concern connectedness, geodesics, isometric embedding, and rigidity of graph embeddings. These are Grassmannian coding-theory problems, but they are not classical Grassmann codes in the Plücker-evaluation sense [1712.10198; 2203.06625; 2005.04402; 2304.08397].

A second nearby tradition studies codes in the Grassmannian, meaning constant-dimension subspace codes. Here a Grassmannian code is simply a subset of \(\mathcal G_q(n,k)\), and lifting MRD matrix codes produces families such as
\[
\left(2r+2,\ \frac{p^{2r+2}-1}{p^2-1},\ 4,\ 2\right)_p
\]
that meet the anticode bound when \(p\equiv 2\) or \(3\pmod 5\). Frobenius and cyclic-shift automorphisms have likewise been used to search for parallelisms and Steiner structures in the Grassmannian. This usage belongs to subspace coding rather than to the linear projective codes \(C(\ell,m)\) [1502.04210; 1210.5724].

A third, analytically different meaning appears in coding on complex Grassmann manifolds equipped with the chordal distance. There the codewords are \(p\)-dimensional subspaces of \(\mathbb C^n\), the relevant notions are density, kissing radius, hyperspherical-cap approximations, and Hamming-type bounds, and maximizing minimum distance is not equivalent to maximizing density. This is a manifold-packing theory rather than a finite-field Plücker-evaluation theory [1607.03132].

These distinctions are substantial. In the strict classical sense, Grassmann codes are the finite-field projective codes \(C(\ell,m)\) arising from the Plücker embedding of \(G_{\ell,m}\). The broader literature extends the geometry of Grassmannians in several directions—affine, polar, graph-theoretic, subspace-coding, and asymptotic differential-geometric—but those developments remain anchored in the same basic insight: the Grassmannian is simultaneously a parameter space of subspaces, a projective variety defined by decomposability, and a fertile source of structured codes.

Source: https://www.emergentmind.com/topics/grassmann-codes