1 Introduction
Due to the connection to network coding, the theory of subspace designs has gained a lot of interest recently. Subspace designs are the
q-analogs of combinatorial designs and arise by replacing the subset lattice of the finite ambient set
V by the subspace lattice of a finite ambient vector space
V. Arguably the most important open problem in this field is the question regarding the existence of a
q-analog of the Fano plane, which is a subspace design with the parameters 2-
\((7,3,1)_q\). This problem has already been stated in 1972 by Ray-Chaudhuri [
3, Problem 28]. Despite considerable investigations, its existence remains undecided for every single order
q of the base field.
A
q-analog of the Fano plane would be a
\([7,4;3]_q\) constant dimension subspace code of size
\(q^8 + q^6 + q^5 + q^4 + q^3 + q^2 + 1\). However, the hitherto best known sizes of such constant dimension subspace codes still leave considerable gaps, namely 333 vs. 381 in the binary case [
14] and 6978 vs. 7651 in the ternary case [
16].
1 Furthermore, it has been shown that the smallest instance
\(q=2\), the binary
q-analog of a Fano plane, can have at most a single nontrivial automorphism [
5,
20].
Another approach has been the investigation of the derived designs of a putative
q-analog
D of the Fano plane. A derived design exists for each point
\(P\in {{\,\mathrm{PG}\,}}(6,q)\) and is always a
q-design with the parameters 1-
\((6,2,1)_q\), which is the same as a line spread of
\({{\,\mathrm{PG}\,}}(5,q)\). Following the notation of [
13], a point
P is called an
\(\alpha \)-point of
D if the derived design in
P is the geometric spread, which is the most symmetric and natural one among the line spreads of
\({{\,\mathrm{PG}\,}}(5,q)\). For highest possible regularity, one would expect all points to be
\(\alpha \)-points.
However, this has been shown to be impossible, as there must always be at least one non-
\(\alpha \)-point of
D [
28]. For the binary case
\(q=2\), this result has been improved to the statement that each hyperplane contains at least one non-
\(\alpha \)-point [
13]. In other words, the non-
\(\alpha \)-points of a binary
q-analog of the Fano plane form a blocking set with respect to the hyperplanes.
In this article, \(\alpha \)-points will be investigated for general values of q, which leads to the following theorem.
As a consequence, we get the following generalization of the result of [
13].
2 Preliminaries
Throughout the article, \(q \ne 1\) is a prime power and V is a vector space over \({\mathbb {F}}_q\) of finite dimension v.
2.1 The subspace lattice
For simplicity, a subspace
U of
V of dimension
\(\dim _{{\mathbb {F}}_q}(U) = k\) will be called a
k-
subspace. The set of all
k-subspaces of
V is called the
Graßmannian and will be denoted by
\(\genfrac[]{0.0pt}{}{V}{k}_{q}\). Picking the “best of two worlds”, we will prefer the algebraic dimension
\(\dim _{{\mathbb {F}}_q}(U)\) over the geometric dimension
\(\dim _{{\mathbb {F}}_q}(U) - 1\), but we will otherwise make heavy use of geometric notions, such as calling the 1-subspaces of
V points, the 2-subspaces
lines, the 3-subspaces
planes, the 4-subspaces
solids and the
\((v-1)\)-subspaces
hyperplanes. In fact, the
subspace lattice \({\mathcal {L}}(V)\) consisting of all subspaces of
V ordered by inclusion is nothing else than the finite projective geometry
\({{\,\mathrm{PG}\,}}(v-1,q) = {{\,\mathrm{PG}\,}}(V)\).
2 There are good reasons to consider the subset lattice as a subspace lattice over the unary “field”
\({\mathbb {F}}_1\) [
11].
The number of all
k-subspaces of
V is given by the Gaussian binomial coefficient
$$\begin{aligned} \#\genfrac[]{0.0pt}{}{V}{k}_{q} = \genfrac[]{0.0pt}{}{v}{k}_{q} = {\left\{ \begin{array}{ll} \frac{(q^v-1)\cdots (q^{v-k+1}-1)}{(q^k-1)\cdots (q-1)} &{}\quad \text {if } k\in \{0,\ldots ,v\}\text {;}\\ 0 &{}\quad \text {otherwise.} \end{array}\right. } \end{aligned}$$
The Gaussian binomial coefficient
\(\genfrac[]{0.0pt}{}{v}{1}_{q}\) is also known as the
q-analog of the number
v and will be abbreviated as
\([v]_{q}\).
For
\(S \subseteq {\mathcal {L}}(V)\) and
\(U,W\in {\mathcal {L}}(V)\), we will use the abbreviations
$$\begin{aligned} S|_U&= \{B\in S \mid U \le B\}\text {,} \\ S|^W&= \{B\in S\mid B\le W\}\quad \text {and} \\ S|_U^W&= \{B\in S\mid U \le B\le W\}\text {.} \end{aligned}$$
For a point
P in a plane
E, the set of all lines in
E passing through
P is known as a
line pencil.
The subspace lattice \({\mathcal {L}}(V)\) is isomorphic to its dual, which arises from \({\mathcal {L}}(V)\) by reversing the order. Fixing a non-degenerate bilinear form \(\beta \) on V, a concrete isomorphism is given by \(U \mapsto U^\perp \), where \(U^\perp = \{\mathbf {x}\in V \mid \beta (\mathbf {x},\mathbf {u}) = 0\text { for all }\mathbf {u}\in U\}\). When addressing the dual of some geometric object in \({{\,\mathrm{PG}\,}}(V)\), we mean its (element-wise) image under this map. Up to isomorphism, the image does not depend on the choice of \(\beta \).
2.2 Subspace designs
The earliest reference for subspace designs is [
10]. It is stated that “Several people have observed that the concept of a
t-design can be generalised [...]”, so the idea might been around before. Subspace designs have also been mentioned in a more general context in [
12]. The first nontrivial subspace designs with
\(t \ge 2\) have been constructed in [
27], and the first nontrivial Steiner system with
\(t \ge 2\) in [
4]. An introduction to the theory of subspace designs can be found at [
7], see also [
25, Day 4].
Subspace designs are interlinked to the theory of network coding in various ways. To this effect we mention the recently found
q-analog of the theorem of Assmus and Mattson [
9], and that a
t-
\((v,k,1)_q\) Steiner system provides a
\((v,2(k-t+1);k)_q\) constant dimension network code of maximum possible size.
Classical combinatorial designs can be seen as the limit case
\(q=1\) of subspace designs. Indeed, quite a few statements about combinatorial designs have a generalization to subspace designs, such that the case
\(q = 1\) reproduces the original statement [
6,
18,
19,
22].
One example of such a statement is the following [
26, Lemma 4.1(1)], see also [
18, Lemma 3.6]: If
D is a
t-
\((v, k, \lambda )_q\) subspace design, then
D is also an
s-
\((v,k,\lambda _s)_q\) subspace design for all
\(s\in \{0,\ldots ,t\}\), where
$$\begin{aligned} \lambda _s :=\lambda \frac{\genfrac[]{0.0pt}{}{v-s}{t-s}_{q}}{\genfrac[]{0.0pt}{}{k-s}{t-s}_{q}}. \end{aligned}$$
In particular, the number of blocks in
D equals
$$\begin{aligned} \#D = \lambda _0 = \lambda \frac{\genfrac[]{0.0pt}{}{v}{t}_{q}}{\genfrac[]{0.0pt}{}{k}{t}_{q}}. \end{aligned}$$
So, for a design with parameters
t-
\((v, k, \lambda )_q\), the numbers
\(\lambda _s\) necessarily are integers for all
\(s\in \{0,\ldots ,t\}\) (
integrality conditions). In this case, the parameter set
t-
\((v,k,\lambda )_q\) is called
admissible. It is further called
realizable if a
t-
\((v,k,\lambda )_q\) design actually exists. The smallest admissible parameters of a nontrivial
q-analog of a Steiner system with
\(t\ge 2\) are 2-
\((7,3,1)_q\), which are the parameters of the
q-analog of the Fano plane. This explains the significance of the question of its realizability.
The numbers
\(\lambda _i\) can be refined as follows. Let
i,
j be non-negative integers with
\(i + j \le t\) and let
\(I\in \genfrac[]{0.0pt}{}{V}{i}_{q}\) and
\(J\in \genfrac[]{0.0pt}{}{V}{v-j}_{q}\). By [
26, Lemma 4.1], see also [
7, Lemma 5], the number
$$\begin{aligned} \lambda _{i,j} :=\# D|_I^J = \lambda \frac{\genfrac[]{0.0pt}{}{v-i-j}{k-i}_{q}}{\genfrac[]{0.0pt}{}{v-t}{k-t}_{q}} \end{aligned}$$
only depends on
i and
j, but not on the choice of
I and
J. Apparently,
\(\lambda _{i,0} = \lambda _i\). The numbers
\(\lambda _{i,j}\) are important parameters of a subspace design. A further generalization is given by the intersection numbers in [
19].
A nice way to arrange the numbers
\(\lambda _{i,j}\) is the following triangle form, which may be called the
q-
Pascal triangle of the subspace design
D.
For a
q-analog of the Fano plane, we get:
The proof of the result of this article will make use of the equality
\(\lambda _{1,1} = \lambda _{0,2}\) in the above triangle.
As a consequence of the numbers
\(\lambda _{i,j}\), the
dual design
\(D^\perp = \{B^\perp \mid B\in D\}\) is a subspace design with the parameters
$$\begin{aligned} t\text {-}\left( v,v-k,\frac{\genfrac[]{0.0pt}{}{v-t}{k}_{q}}{\genfrac[]{0.0pt}{}{v-t}{k-t}_{q}}\right) _{\!q}\text {.} \end{aligned}$$
For a point
\(P \le V\), the
derived design of
D in
P is the set of blocks
$$\begin{aligned} {{\,\mathrm{Der}\,}}_P(D) = \{ B/P \mid B \in D|_P\} \end{aligned}$$
in the ambient vector space
V/
P.
3 By [
18],
\({{\,\mathrm{Der}\,}}_P(D)\) is a subspace design with the parameters
\((t-1)\)-
\((v-1,k-1,\lambda )_q\). In the case of a
q-analog of the Fano plane,
\({{\,\mathrm{Der}\,}}_P(D)\) has the parameters 1-
\((6,2,1)_q\).
2.3 Spreads
A 1-\((v,k,1)_q\) Steiner system \({\mathcal {S}}\) is just a partition of the point set of V into k-subspaces. These objects are better known under the name \((k-1)\)-spread and have been investigated in geometry well before the emergence of subspace designs. A 1-spread is also called a line spread.
A set
\({\mathcal {S}}\) of
k-subspaces is called a
partial \((k-1)\)-spread if each point is covered by at most one element of
\({\mathcal {S}}\). The points not covered by any element are called
holes. A recent survey on partial spreads is found in [
17].
The parameters 1-
\((v,k,1)_q\) are admissible if and only
v is divisible by
k. In this case, spreads do always exist [
24, Sect. VI]. An example can be constructed via field reduction: We consider
V as a vector space over
\({\mathbb {F}}_{q^k}\) and set
\({\mathcal {S}} = \genfrac[]{0.0pt}{}{V}{1}_{q^k}\). Switching back to vector spaces over
\({\mathbb {F}}_q\), the set
\({\mathcal {S}}\) is a
\((k-1)\)-spread of
V, known as the
Desarguesian spread.
A
\((k-1)\)-spread
\({\mathcal {S}}\) is called
geometric or
normal if for two distinct blocks
\(B,B'\in {\mathcal {S}}\), the set
\({\mathcal {S}}|^{B + B'}\) is always a
\((k-1)\)-spread of
\(B + B'\). In other words,
\({\mathcal {S}}\) is geometric if every 2
k-subspace of
V contains either 0, 1 or
\([2k]_{q}/[k]_{q} = q^k + 1\) blocks of
\({\mathcal {S}}\). It is not hard to see that the Desarguesian spread is geometric. In fact, it follows from [
2, Theorem 2] that a
\((k-1)\)-spread is geometric if and only if it is isomorphic to a Desarguesian spreads.
The derived designs of a
q-analog of the Fano plane
D are line spreads in
\({{\,\mathrm{PG}\,}}(5,q)\). The most symmetric one among these spreads is the Desarguesian spread. Following the notation of [
13], a point
P is called an
\(\alpha \)-
point of the
q-analog of the Fano plane
D if the derived design in
P is the geometric spread.
4
We remark that in the binary case
\(q=2\), the line spreads of
\({{\,\mathrm{PG}\,}}(5,q)\) have been classified into
\(131\,044\) isomorphism types in [
21].
2.4 Generalized quadrangles
Generalized quadrangles have been introduced in the more general setting of generalized polygons in [
29], as a tool in the theory of finite groups.
A generalized quadrangle \(Q = ({\mathcal {P}},{\mathcal {L}},I)\) is called degenerate if there is a point P such that each point of Q is incident with a line through P. If each line of Q is incident with \(t+1\) points, and each point is incident with \(s+1\) lines, we say that Q is of order (s, t). The dual \(Q^\perp \) arises from Q by interchanging the role of the points and the lines. It is again a generalized quadrangle. Clearly, \((Q^\perp )^\perp = Q\), and Q is of order (s, t) if and only if \(Q^\perp \) is of order (t, s).
Furthermore,
Q is said to be
projective if it is embeddable in some Desarguesian projective geometry in the following sense: There is a Desarguesian projective geometry
\(({{\mathcal {P}}}, {{\mathcal {L}}}, {\bar{I}})\) such that
\({\mathcal {P}}\subseteq \bar{{\mathcal {P}}}\),
\({\mathcal {L}}\subseteq \bar{{\mathcal {L}}}\), for all
\((P,L)\in {{\mathcal {P}}}\times {{\mathcal {L}}}\) we have
\(P\mathrel {I} L\) if and only if
\(P \mathrel {{\bar{I}}} L\), and for each point
\(P\in \bar{{\mathcal {P}}}\) with
\(P \mathrel {{\bar{I}}} L\) for some line
\(L\in {\mathcal {L}}\) we have
\(P\in {\mathcal {P}}\).
5 The non-degenerate finite projective generalized quadrangles have been classified in [
8, Theorem 1], see also [
23, 4.4.8]. These are exactly the so-called
classical generalized quadrangles which are associated to a quadratic form or a symplectic or Hermitian polarity on the ambient geometry, see [
23, 3.1.1].
In this article, two of these classical generalized quadrangles will appear.
(i)
The symplectic generalized quadrangle
W(
q) consisting of the set of points of
\({{\,\mathrm{PG}\,}}(3,q)\) together with the totally isotropic lines with respect to a symplectic polarity. Taking the geometry as
\({{\,\mathrm{PG}\,}}({\mathbb {F}}_q^4)\), the symplectic polarity can be represented by the alternating bilinear form
\(\beta (\mathbf {x},\mathbf {y}) = x_1 y_2 - x_2 y_1 + x_3 y_4 - x_4 y_3\). The configuration of the lines
\({\mathcal {L}}\) in
\({{\,\mathrm{PG}\,}}(3,q)\) is also known as a (general) linear complex of lines, see [
23, 3.1.1 (iii)] or [
15, Theorem 15.2.13]. Under the Klein correspondence,
\({\mathcal {L}}\) is a non-tangent hyperplane section of the Klein quadric.
(ii)
The second one is the parabolic quadric Q(4, q), whose points \({\mathcal {P}}\) are the zeros of a parabolic quadratic form in \({{\,\mathrm{PG}\,}}(4,q)\), and whose lines are all the lines contained in \({\mathcal {P}}\). Taking the geometry as \({{\,\mathrm{PG}\,}}({\mathbb {F}}_q^5)\), the parabolic quadratic form can be represented by \(q(\mathbf {x}) = x_1 x_2 + x_3 x_4 + x_5^2\).
Both
W(
q) and
Q(4,
q) are of order (
q,
q). By [
23, 3.2.1] they are duals of each other, meaning that
\(W(q)^\perp \cong Q(4,q)\).
Let
\(Q = ({\mathcal {P}},{\mathcal {L}},I)\) be a generalized quadrangle. As in projective geometries, a set
\({\mathcal {S}} \subseteq {\mathcal {L}}\) is called a
spread of
Q if each point of
Q is incident with a unique line in
\({\mathcal {S}}\). Dually, a set
\({\mathcal {O}} \subseteq {\mathcal {P}}\) is called an
ovoid of
Q if each line of
Q is incident with a unique point in
\({\mathcal {O}}\). Clearly, the spreads of
Q bijectively correspond to the ovoids of
\(Q^\perp \). This already shows the equivalence of parts (a) and (b) in Theorem
1.
3 Proof of the theorems
For the remainder of the article, we fix
\(v = 7\) and assume that
\(D \subseteq \genfrac[]{0.0pt}{}{V}{3}_{q}\) is a
q-analog of the Fano plane. The numbers
\(\lambda _{i,j}\) are defined as in Sect.
2.2.
By the design property, the intersection dimension of two distinct blocks \(B,B'\in D\) is either 0 or 1. So by the dimension formula, \(\dim (B + B') \in \{5,6\}\). Therefore two distinct blocks contained in a common 5-space always intersect in a point. Moreover, a solid S of V contains either a single block or no block at all. We will call S a rich solid in the former case and a poor solid in the latter.
We will call a 5-subspace F a \(\beta \)-flat with focal point \(P\in \genfrac[]{0.0pt}{}{F}{1}_{q}\) if all the \(\lambda _{0,2} = q^2 + 1\) blocks contained in F pass through P.
Now we fix a hyperplane H of V and assume that all its points are \(\alpha \)-points.
By Lemma
3.6, every 5-subspace
F of
H is a
\(\beta \)-flat. We denote its unique focal point by
\(\alpha (F)\). Moreover by Lemma
3.8, each point
P of
H is the focal point of a unique
\(\beta \)-flat
F in
H. We will denote this
\(\beta \)-flat by
\(\beta (P)\). Clearly, the mappings
$$\begin{aligned} \alpha : \genfrac[]{0.0pt}{}{H}{5}_{q} \rightarrow \genfrac[]{0.0pt}{}{H}{1}_{q} \quad \text {and}\quad \beta : \genfrac[]{0.0pt}{}{H}{1}_{q} \rightarrow \genfrac[]{0.0pt}{}{H}{5}_{q} \end{aligned}$$
are inverse to each other. So they provide a bijective correspondence between the points and the 5-subspaces of
H.
For the remainder of this article, we fix a poor solid
S of
H. Note that by Lemma
3.4(b), every 5-subspace of
H contains a suitable solid
S.
7 The set of
\(\genfrac[]{0.0pt}{}{6-4}{5-4}_{q} = q+1\) intermediate 5-subspaces
F with
\(S< F < H\) will be denoted by
\({\mathcal {F}}\). For each
\(F\in {\mathcal {F}}\), the set
\({\mathcal {L}}_F :=\{B \cap S \mid B\in D|^F\}\) is a line spread of
S by Lemma
3.4(c).
Now let \({\mathcal {L}} = \bigcup _{F\in {\mathcal {F}}} {\mathcal {L}}_F\).
Now we can prove our main result.
Theorem
2 is now a direct consequence.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.