arXiv reader

Error-Correcting Codes on Projective Bundles over Deligne-Lusztig varieties

Daniel Camazón Portela and Juan Antonio López Ramos

arXiv
2401.11433v1 · PDF · 21 January 2024
Categories
cs.IT, math.AG, math.IT
Journal
Mathematics 2023, 11(14), 3079;
DOI
10.3390/math11143079
Licence
CC BY 4.0 The site shows the text.

Abstract

The aim of this article is to give lower bounds on the parameters of algebraic geometric error-correcting codes constructed from projective bundles over Deligne–Lusztig surfaces. The methods based on an intensive use of the intersection theory allow us to extend the codes previously constructed from higher-dimensional varieties, as well as those coming from curves. General bounds are obtained for the case of projective bundles of rank 2 over standard Deligne-Lusztig surfaces, and some explicit examples coming from surfaces of type A_{2} and {}^{2}A_{4} are given.

Items 16

Definition

Definition 2.1

HITE

Fix an element w in the Weyl group W, and let w=s_{i_{1}}\cdot \ldots \cdot s_{i_{r}} be a reduced expression of w. Call w a Coxeter element if there in this expression occurs exactly one s_{i} from each of the orbits of F on \left\{s_{1}, \dots, s_{n}\right\}. Denote by \delta the order of F on this set. Then, the Deligne–Lusztig variety X(w) is defined as the image of L^{-1}(B\dot{w}B) in G/B. That is,

X(w)=\pi(L^{-1}(B\dot{w}B)).
Example

Example 2.2

ETNX

Let us consider the Deligne–Lusztig surface \overline{X}(w) of type A_{2}. In this particular case, the connected reductive algebraic group is G=GL(3,k) and the Frobenius map is given by

\xymatrix{F: g\ar[r] & g^{q}}.

Moreover, the Weyl group W of G is isomorphic to the symmetric group of the three vectors in the base of k^{3}, and \overline{X}(w)=\overline{X}(s_{1},s_{2}), where s_{1} corresponds to the permutation of the first and the second vectors of the base and s_{2} corresponds to the permutation of the second and the third vector.

Theorem

Theorem 2.3

S89P

Theorem 3 in [Hansen02] Let \overline{X}(w) be a standard Deligne–Lusztig variety of type {}^{2}A_{n}, B_{n}, C_{n}, D_{n} or {}^{2}D_{n}. Assume that char(k)\neq 2 in the orthogonal cases. Let P be the parabolic subgroup generated by B together with the double cosets Bs_{2}B, Bs_{3}B, \ldots,Bs_{n}B, and let

\pi: (G/B)^{l(w)+1}\rightarrow G/P\subseteq\mathbb{P}(V)\cong\mathbb{P}^{N-1}

be the projection (projection to the first factor, followed by the quotient map). It is worth noting that the inclusion G/P\subseteq\mathbb{P}(V) is an equality in the non-orthogonal cases. Denote by L^{e} the e-dimensional linear subspace of \mathbb{P}^{N-1} obtained by setting the N-1-e last coordinates as equal to zero.

  1. The image Z=\pi(\overline{X}(w)) is a normal, strict complete intersection. In fact, Lusztig shows (see pp. 444–445 in [Lusztig76]) that Z equals the support of the scheme’s theoretic complete intersection Z{'}= \cap_{i=0}^{a_{0}(V)-1}H_{i}, with X(w) mapping isomorphically onto the open subset \left\langle x,F_{V}^{a_{0}(V)}(x)\right\rangle\neq 0 of Z (see Table 2 in [Hansen02]). black In the unitary and orthogonal cases, the singular locus of Z, Z_{sing} consists of the finitely many G^{F}-translates of the closed subscheme Z\cap L^{a_{0}(V)-1}. Hence,

    codim(Z_{sing},Z)=N+1-2a_{0}(V)+a_{0}(L^{a_{0}(V)-1}).

    In the symplectic case, Z_{sing} consists of the G^{F}-translates of the closed subscheme Z\cap L^{a_{0}(V)-2}, and the previous formula becomes

    codim(Z_{sing},Z)=2+a_{0}(L^{a_{0}(V)-2}).
  2. For codim(Z_{sing},Z)\geq 4, Pic(Z)=\mathbb{Z} and consequently

    \begin{split} Pic(\overline{X}(w))=\mathbb{Z}\left[\pi^{*}H\right] & \oplus\mathbb{Z}\left[\left\{\left[V\right]:\enspace V\enspace\text{component of}\enspace D_{1}\right\}\right]\\ & \oplus j_{*} A_{l(w)-1}(\bigcup_{i\in\mathcal{I}-\left\{1\right\}}D_{i}) \end{split}

    where H is the hyperplane section of Z, j is the obvious inclusion and \mathcal{I} is the set of indices \left\{i\right\}, satisfying that some connected component of the Dynkin diagram corresponding to D_{i} occurs as a subgraph of the Dynkin diagram corresponding to D_{1}.

  3. For any Coxeter element w{'}, we have

    Pic(\overline{X}(w{'}))_{p{'}}\cong Pic(\overline{X}(w))_{p{'}}.
Definition

Definition 3.1

OHJD

H-construction p. 272 in [tsfasman2]. black Let X be a normal projective variety over \mathbb{F}_{q}. Let L be a line bundle defined over \mathbb{F}_{q} and let P_{1}, P_{2}, ...P_{n} be distinct \mathbb{F}_{q}-rational points on X. Set \mathcal{P}=\left\{P_{1}, P_{2}, ...P_{n}\right\}. In each P_{i} , choose isomorphisms of the fibers L_{P_{i}} with \mathbb{F}_{q}. The linear code C(L,\mathcal{P}) of length n associated to (X,L,\mathcal{P}) is the image of the germ map

\alpha: \Gamma(X,L)\rightarrow \oplus_{i=1}^{n} L_{P_{i}}\cong\mathbb{F}_{q}.

From now on, we assume that all line bundles considered actually have a non-zero global section.
Suppose L arises as the line bundle associated to a divisor D and that the P_{i} are not in the support of D. Then, we obtain the same code (up to isomorphism) as when evaluating the rational functions

L(D)=\left\{f\in k(X)^{*}: div(f)+D\geq 0\right\}

in the points \mathcal{P}.

Proposition

Proposition 3.2

E0YM

Proposition 3.2 in [shhansen] Let X be a normal projective variety defined over \mathbb{F}_{q} of dimension at least two. Let C_{1}, C_{2}, ..., C_{a} be (irreducible) curves on X with \mathbb{F}_{q}- rational points \mathcal{P}=\left\{P_{1}, P_{2}, \ldots, P_{n}\right\}. Assume the number of \mathbb{F}_{q}-rational points on rational C_{i} is less than N. Let L be a line bundle on X, defined over \mathbb{F}_{q}, such that L\cdot C_{i}\geq 0 for all i. Let

l=sup_{s\in\Gamma(X,L)}\#\left\{i:Z(s)\enspace\text{contains}\enspace C_{i}\right\}.

Then, the code C(L,\mathcal{P}) has length n and minimum distance

d\geq n-lN-\sum_{i=1}^{a}L\cdot C_{i}

If L\cdot C_{i}=\eta\leq N for all i, then d\geq n-lN-(a-l)\eta.

Used by Proof
Corollary

Corollary 3.3

GOOY

Corollary 3.2 in [shhansen]. Assume furthermore that X is a non-singular surface and that H is a nef divisor on X with H\cdot C_{i}>0 for all i. Then,

\mathit{l}\leq\frac{L\cdot H}{min_{i}\left\{C_{i}\cdot H\right\}}

Consequently, if L\cdot H<C_{i}\cdot H for all i, we have \mathit{l}=0 and

d\geq n-m,

where m=\sum_{i=1}^{a}L\cdot C_{i}.

Used by Proof
Theorem

Theorem 4.1

9MXS

Let S be a standard Deligne–Lusztig surface of type A_{2}(\mathbb{F}_{q}), {}^{2}A_{3}(\mathbb{F}_{q^{2}}), {}^{2}A_{4}(\mathbb{F}_{q^{2}}) or C_{2}(\mathbb{F}_{q}), and let V be a vector bundle of rank 2 defined over S. Consider the projective bundle T=P(V) over S, that is p: T\rightarrow S. Then, for some a,b>0, we can construct a code on T over \mathbb{F}_{q^{\delta}} with parameters

  1. n=\#S(\mathbb{F}_{q^{\delta}})\#\mathbb{P}^{1}(\mathbb{F}_{q^{\delta}});

  2. k=h^{0}(S,Symm^{b}(V)\otimes\mathcal{O}_{S}(aD_{j}));

  3. d\geq n-(-bc_{1}(W_{1})\#\left\{B_{i}\right\}+aD_{j}\cdot D_{i})\#\mathbb{P}^{1}(\mathbb{F}_{q^{\delta}})-(\#S(\mathbb{F}_{q^{\delta}})-(-bc_{1}(W_{1})\#\left\{B_{i}\right\}+aD_{j}\cdot D_{i}))b if (-bc_{1}(W_{1})\#\left\{B_{i}\right\}+aD_{j}\cdot D_{i})>0, and d\geq n-(\#S(\mathbb{F}_{q^{\delta}}))b otherwise,

where

\begin{aligned} \delta= & \begin{cases} 1 & \text{if $S$ is of type $A_{2}, C_{2}$,} \\ 2 & \text{if $S$ is of type ${}^{2}A_{3}, {}^{2}A_{4}$,} \\ \end{cases} & D_{i}= & \begin{cases} D_{1} & \text{if $S$ is of type $A_{2}, {}^{2}A_{3}, C_{2}$,} \\ D_{2} & \text{if $S$ is of type ${}^{2}A_{4}$,} \\ \end{cases} \\ D_{j}= & \begin{cases} D_{2} & \text{if $S$ is of type $A_{2}, {}^{2}A_{3}, C_{2}$,} \\ D_{1} & \text{if $S$ is of type ${}^{2}A_{4}$,} \\ \end{cases} & & \end{aligned}

and c_{1}(W_{1}) denotes the first Chern class of the line subbundle of minimum degree of the restricted vector bundle i_{B_{i}}^{*} V over B_{i}.

Proof of Theorem 4.1

We will consider \mathcal{P} to be the \mathbb{F}_{q^{\delta}}-rational points on T. Let C_{1}, C_{2}, ..., C_{a} be the fibers over the \mathbb{F}_{q^{\delta}}-rational points of S. These disjoint lines contain all \mathbb{F}_{q^{\delta}}-rational points of T, that is,

T(\mathbb{F}_{q^{\delta}})=\bigcup_{P\in S(\mathbb{F}_{q^{\delta}})} p^{-1}(P)(\mathbb{F}_{q^{\delta}}).

It follows then that the length n of the code is

n=\#\mathcal{P}=\#S(\mathbb{F}_{q^{\delta}})\cdot\#\mathbb{P}^{1}(\mathbb{F}_{q^{\delta}})

Let L be the line bundle L=\mathcal{O}_{T}(b)\otimes\mathcal{O}_{T}(ap^{*}(D_{j})) over T. From Theorem 9.6 in [EisenbudHarris16] and Proposition II.7.11 in [Hartshorne77], it follows that

\Gamma(T,L)\cong\Gamma(S,p_{*}L)=\Gamma(S,Symm^{b}(V)\otimes\mathcal{O}_{S}(aD_{j})),

so when in the range wherein the bound on the minimum distance ensures the injectivity of the evaluation map, the dimension of the code is

k=h^{0}(S,Symm^{b}(V)\otimes\mathcal{O}_{S}(aD_{j})).

Now, we will apply Proposition 3.2 in order to obtain the bound for the minimum distance. It is worth noting that, in the cases we are interested in, C_{i}=p^{-1}(P) with P\in S(\mathbb{F}_{q^{\delta}}), so the maximum number of rational points on C_{i} will be N=\#\mathbb{P}^{1}(\mathbb{F}_{q^{\delta}}). Moreover, by Lemma 9.7 in [EisenbudHarris16], we have that

L\cdot C_{i}=b.

From Equation Proof, we know that \bigcup_{i=1}^{\#S(\mathbb{F}_{q^{\delta}})} C_{i}\subset p^{*}D_{i}. Now, B_{i} is an irreducible component of D_{i}, it is rational and T_{B_{i}}=p^{-1}(B_{i}) will be isomorphic to the projective bundle p_{T_{B_{i}}}: P(i_{B_{i}}^{*} V)\rightarrow B_{i}, where i_{B_{i}}: B_{i}\rightarrow S denotes the closed embedding. As a consequence, by Theorem 9.6 in [EisenbudHarris16], the Chow ring of T_{B_{i}} is isomorphic to

A^{\bullet}(T_{B_{i}})\cong A^{\bullet}(B_{i})\left[\xi\right]/(\xi^{2}+c_{1}(i_{B_{i}}^{*} V)\xi)

where \xi is the hyperplane section in T_{B_{i}}. It is worth noting that since B_{i} is rational, then by Corollary V.2.14 in [Hartshorne77], i_{B_{i}}^{*} V=W_{1}\oplus W_{2}, and we can always suppose that deg(W_{1})\leq deg(W_{2}).

Now, if we restrict the line bundle L to T_{B_{i}}, then L\vert_{T_{B_{i}}}\cong\mathcal{O}_{T_{B_{i}}}(b)\otimes\mathcal{O}_{T_{B_{i}}}(p_{B_{i}}^{*}i_{B_{i}}^{*}(aD_{j})). Let H=\xi+c_{1}(W_{2})F be a nef divisor on T_{B_{i}} (see Theorem V.2.17 [Hartshorne77]). Then, since H\cdot C_{i}=1, it follows by Corollary 3.3 that

\mathit{l}_{T_{B_{i}}}\leq H\cdot L\vert_{B_{i}}=-bc_{1}(W_{1})+aD_{j}\cdot B_{i},

provided that (-bc_{1}(W_{1})+aD_{j}\cdot B_{i})>0, and \mathit{l}_{T_{B_{i}}}=0 otherwise. As D_{i}=\sqcup B_{i}, we can conclude the following bound for the minimum distance

d\geq n-(-bc_{1}(W_{1})\#\left\{B_{i}\right\}+aD_{j}\cdot D_{i})\#\mathbb{P}^{1}(\mathbb{F}_{q^{\delta}})-(\#S(\mathbb{F}_{q^{\delta}})-(-bc_{1}(W_{1})\#\left\{B_{i}\right\}+aD_{j}\cdot D_{i}))b

if (-bc_{1}(W_{1})\#\left\{B_{i}\right\}+aD_{j}\cdot D_{i})>0 and

d\geq n-(\#S(\mathbb{F}_{q^{\delta}}))b

otherwise.

Lemma

Lemma 4.2

1NV7

Let Y be a variety and V=V_{1}\oplus V_{2} a vector bundle of rank 2 over Y, that is, a direct sum of two line bundles V_{1} and V_{2}. Then, the symmetric algebra of V satisfies

Symm(V)\cong\bigotimes_{i=1}^{2} Symm(V_{i})=\bigoplus_{i_{1},i_{2} \in \mathbb{N}^{2}} Symm^{i_{1}}V_{1} \otimes Symm^{i_{2}}V_{2}\cong \bigoplus_{i_{1},i_{2} \in \mathbb{N}^{2}} V_{1}^{\otimes i_{1}} \otimes V_{2}^{\otimes i_{2}},
Used by Proof, Proof
Proof of Lemma 4.2

The first isomorphism is a consequence of the universal property for symmetric algebras, the second equality is just by definition and the last isomorphism follows from the fact that, for a line bundle V_{i}, we have

V_{i}^{\otimes n}\cong Symm^{n}V_{i}.
Corollary

Corollary 4.3

PTIJ

Let S_{1} be the Deligne–Lusztig surface of type A_{2} defined over the field \mathbb{F}_{q}. Consider some b, such that 0<b<(q+1), and V_{i}=\mathcal{O}_{S_{1}}(n_{i}H-\sum_{j=1}^{q^{2}+q+1} m_{i,j}B_{j}) for i=1,2, where H=\pi^{*}(\mathcal{O}_{\mathbb{P}^{2}}(1)), verifying for any pair i_{1},i_{2}\in\mathbb{N} with i_{1}+i_{2}=b:

  1. (i_{1}n_{1}+i_{2}n_{2})\leq 3(q-1);

  2. (i_{1}n_{1}+i_{2}n_{2})\geq\sum_{j=1}^{3}(i_{1}m_{1,j}+i_{2}m_{2,j});

  3. (i_{1}m_{1,j}+i_{2}m_{2,j})\geq (i_{1}m_{1,j+1}+i_{2}m_{2,j+1});

  4. 3(i_{1}n_{1}+i_{2}n_{2})>\sum_{j=1}^{q^{2}+q+1}(i_{1}m_{1,j+1}+i_{2}m_{2,j+1}).

black

Let T_{1} be the projective bundle P(V_{1}\oplus V_{2}) over S_{1}, p_{1}: T_{1}\rightarrow S_{1}. Then, we can construct a code on T_{1} over \mathbb{F}_{q} with parameters

  1. n=(q^{2}+q+1)(q+1)^{2},;

  2. k=\sum_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\frac{1}{2}((i_{1}n_{1}+i_{2}n_{2})((i_{1}n_{1}+i_{2}n_{2})+3)-\sum_{j=1}^{q^{2}+q+1} (i_{1}m_{1,j}+i_{2}m_{2,j})((i_{1}m_{1,j}+i_{2}m_{2,j})+1))+1,;

  3. d\geq n-(q^{2}+q+1)(q+1)b..

Proof of Corollary 4.3

Since A_{2} is isomorphic to the blow-up of \mathbb{P}^{2} at the set of its rational points over \mathbb{F}_{q} (see Sect. 5 in [Rodier00]), we have that \#S_{1}(\mathbb{F}_{q})=(q^{2}+q+1)(q+1). Moreover, \#\mathbb{P}^{1}(\mathbb{F}_{q})=(q+1), so we can conclude that the length n of the code is

n=\#\mathcal{P}=\#S_{1}(\mathbb{F}_{q})(q+1)=(q^{2}+q+1)(q+1)^{2}

Let L be the line bundle L=\mathcal{O}_{T_{1}}(b). In order to compute the dimension of the code, we have dim \Gamma(T_{1},L)=dim \Gamma(S_{1},p_{1*}L), where p_{1*}L\cong Symm^{b}(V_{1}\oplus V_{2}). As a result of Lemma 4.2 ,

Symm^{b}(V_{1}\oplus V_{2})=\bigoplus_{\substack{i_{1},i_{2} \in \mathbb{N}^{2} \\ i_{1}+i_{2}=b}} \mathcal{O}_{S_{1}}(n_{1}H-\sum_{j=1}^{q^{2}+q+1} m_{1,j}B_{j})^{\otimes i_{1}} \otimes \mathcal{O}_{S_{1}}(n_{2}H-\sum_{j=1}^{q^{2}+q+1} m_{2,j}B_{j})^{\otimes i_{2}}.

Now, global sections of a direct sum of line bundles satisfy

\begin{split} \Gamma(S_{1},\bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}} \mathcal{O}_{S_{1}}(n_{1}H-\sum_{j=1}^{q^{2}+q+1} m_{1,j}B_{j})^{\otimes i_{1}} \otimes \mathcal{O}_{S_{1}}(n_{2}H-\sum_{j=1}^{q^{2}+q+1} m_{2,j}B_{j})^{\otimes i_{2}}) & \cong \\ \bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\Gamma(S_{1},\mathcal{O}_{S_{1}}(n_{1}H-\sum_{j=1}^{q^{2}+q+1} m_{1,j}B_{j})^{\otimes i_{1}} \otimes \mathcal{O}_{S_{1}}(n_{2}H-\sum_{j=1}^{q^{2}+q+1} m_{2,j}B_{j})^{\otimes i_{2}}), & \end{split}

and finally, we can conclude

\begin{split} \bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\Gamma(S_{1},\mathcal{O}_{S_{1}}(n_{1}H-\sum_{j=1}^{q^{2}+q+1} m_{1,j}B_{j})^{\otimes i_{1}} \otimes \mathcal{O}_{S_{1}}(n_{2}H-\sum_{j=1}^{q^{2}+q+1} m_{2,j}B_{j})^{\otimes i_{2}}) & \cong \\ \bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\Gamma(S_{1},\mathcal{O}_{S_{1}}((i_{1}n_{1}+i_{2}n_{2})H-\sum_{j=1}^{q^{2}+q+1} (i_{1}m_{1,j}+i_{2}m_{2,j})B_{j})) & \end{split}

Furthermore, by the hypothesis of the theorem, \mathcal{O}_{S_{1}}((i_{1}n_{1}+i_{2}n_{2})H-\sum_{j=1}^{q^{2}+q+1} (i_{1}m_{1,j}+i_{2}m_{2,j})B_{j}) is excellent with respect to the exceptional configuration of \pi: S_{1}\rightarrow\mathbb{P}^{2} (see p. 215 in [Harbourne85]), so if we denote F=\mathcal{O}_{S_{1}}((i_{1}n_{1}+i_{2}n_{2})H-\sum_{j=1}^{q^{2}+q+1} (i_{1}m_{1,j}+i_{2}m_{2,j})B_{j}), then by Theorem 1.1 in [Harbourne85]:

\begin{split} h^{0}(S_{1}, F) & =\frac{1}{2}(f\cdot f-f\cdot k_{S_{1}})+1 \\ & =\frac{1}{2}((i_{1}n_{1}+i_{2}n_{2})((i_{1}n_{1}+i_{2}n_{2})+3)-\sum_{j=1}^{q^{2}+q+1} (i_{1}m_{1,j}+i_{2}m_{2,j})((i_{1}m_{1,j}+i_{2}m_{2,j})+1))+1. \end{split}

This allows us to conclude that, when in the range wherein the bound on the minimum distance ensures the injectivity of the evaluation map, the dimension of the code is

k=\sum_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\frac{1}{2}((i_{1}n_{1}+i_{2}n_{2})((i_{1}n_{1}+i_{2}n_{2})+3)-\sum_{j=1}^{q^{2}+q+1} (i_{1}m_{1,j}+i_{2}m_{2,j})((i_{1}m_{1,j}+i_{2}m_{2,j})+1))+1.

Finally, by Proposition 9.4 in [EisenbudHarris16], any section of L\vert_{T_{1 B_{i}}} (recall that T_{1 B_{i}}=p_{1}^{-1}(B_{i})) intersects a fiber C_{i}\subset T_{1 B_{i}} in a hyperplane, so \mathit{l}=0 and we can conclude the following bound for the minimum distance:

d\geq n-(q^{2}+q+1)(q+1)b.
Corollary

Corollary 4.4

2S17

Let S_{2} be the Deligne–Lusztig surface of type {}^{2}A_{4} defined over the field \mathbb{F}_{q^{2}}. Consider V_{i}=\pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{i}) for i=1,2, and let T_{2} be the projective bundle P(V_{1}\oplus V_{2}) over S_{2}, p_{2}: T_{2}\rightarrow S_{2}. Then, for some 0<b<(q^{2}+1) and for any t_{1},t_{2}\in\mathbb{N} such that (q+1)<(i_{1}t_{1}+i_{2}t_{2})<(q^{3}+1) for any pair i_{1},i_{2}\in\mathbb{N} with i_{1}+i_{2}=b, we can construct a code on T_{2} over \mathbb{F}_{q^{2}} with parameters

  1. n=(q^{5}+1)(q^{3}+1)(q^{2}+1)^{2};

  2. k=\sum_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\binom{4+i_{1}t_{1}+i_{2}t_{2}}{i_{1}t_{1}+i_{2}t_{2}}-\binom{4+i_{1}t_{1}+i_{2}t_{2}-(q+1)}{i_{1}t_{1}+i_{2}t_{2}-(q+1)};

  3. d\geq n-(q^{5}+1)(q^{3}+1)(q^{2}+1)b.

Proof of Corollary 4.4

Since all the rational points of S_{2} are distributed equally on the disjoint rational curves B_{i} constituting the irreducible components of the divisor D_{2}\subset S_{2}, then \#S_{2}(\mathbb{F}_{q^{2}})=(q^{5}+1)(q^{3}+1)(q^{2}+1). Moreover, \#\mathbb{P}^{1}(\mathbb{F}_{q^{2}})=(q^{2}+1), so we can conclude that the length n of the code is

n=\#\mathcal{P}=\#S_{2}(\mathbb{F}_{q^{2}})(q^{2}+1)=(q^{5}+1)(q^{3}+1)(q^{2}+1)^{2}.

Let L be the line bundle L=\mathcal{O}_{T_{2}}(b). In order to compute the dimension of the code, we have that dim \Gamma(T_{2},L)=dim \Gamma(S_{2},p_{2*}L), with p_{2*}L\cong Symm^{b}(V_{1}\oplus V_{2}). As a result of Lemma 4.2

Symm^{b}(V_{1}\oplus V_{2})=\bigoplus_{\substack{i_{1},i_{2} \in \mathbb{N}^{2} \\ i_{1}+i_{2}=b}} \pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{1})^{\otimes i_{1}} \otimes \pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{2})^{\otimes i_{2}}.

Now, global sections of a direct sum of line bundles satisfy

\Gamma(S_{2},\bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}} \pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{1})^{\otimes i_{1}} \otimes \pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{2})^{\otimes i_{2}})\cong\bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\Gamma(S_{2},\pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{1})^{\otimes i_{1}} \otimes \pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{2})^{\otimes i_{2}}),

and since pull-back commutes with tensor product, it follows then that

\begin{aligned} \bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\Gamma(S_{2},\pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{1})^{\otimes i_{1}} \otimes \pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(t_{2})^{\otimes i_{2}}) & \cong\bigoplus_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\Gamma(S_{2},\pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(i_{1}t_{1}+i_{2}t_{2})) \end{aligned}

Furthermore, as \pi is birational onto Z, we obtain (see Theorem 2.31 in [Iitaka82])

\Gamma(S_{2},\pi^{*}i^{*}\mathcal{O}_{\mathbb{P}^{4}}(i_{1}t_{1}+i_{2}t_{2}))\cong\Gamma(Z,i^{*}\mathcal{O}_{\mathbb{P}^{4}}(i_{1}t_{1}+i_{2}t_{2}))

Since Z is the complete intersection of two hyper-surfaces H_{0}, H_{1}\subset\mathbb{P}^{4} of degrees q+1 and q^{3}+1, respectively, we obtain short exact sequences of line bundles

0\rightarrow\mathcal{O}_{\mathbb{P}^{4}}(t-(q+1))\rightarrow\mathcal{O}_{\mathbb{P}^{4}}(t)\rightarrow\mathcal{O}_{H_{0}}(t)\rightarrow 0,

and

0\rightarrow\mathcal{O}_{H_{0}}(t-(q^{3}+1))\rightarrow\mathcal{O}_{H_{0}}(t)\rightarrow\mathcal{O}_{H_{0}\cap H_{1}}(t)\rightarrow 0,

(see Section 7.3 in [Iitaka82]). These sequences give long exact sequences of cohomology groups,

\begin{split} 0\rightarrow H^{0}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t-(q+1))) & \rightarrow H^{0}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t))\rightarrow H^{0}(H_{0},\mathcal{O}_{H_{0}}(t))\rightarrow \\ & H^{1}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t-(q+1)))\rightarrow H_{1}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t))\rightarrow \\ & H^{1}(H_{0},\mathcal{O}_{H_{0}}(t))\rightarrow H^{2}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{N-1}}(t-(q+1)))\rightarrow \ldots, \end{split}

and

\begin{split} 0\rightarrow H^{0}(H_{0},\mathcal{O}_{H_{0}}(t-(q^{3}+1)) & \rightarrow H^{0}(H_{0},\mathcal{O}_{H_{0}}(t))\rightarrow \\ H^{0}(H_{0}\cap H_{1},\mathcal{O}_{H_{0}\cap H_{1}}(t))& \rightarrow H^{1}(H_{0},\mathcal{O}_{H_{0}}(t-(q^{3}+1))\rightarrow \ldots \end{split}

By the formulas of Theorem III.5.1 in [Hartshorne77] for the cohomology of projective space, the first long exact sequence reduces (for any t) to

\begin{split} 0\rightarrow H^{0}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t-(q+1))) & \rightarrow H^{0}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t))\rightarrow \\ H^{0}(H_{0},\mathcal{O}_{H_{0}}(t))\rightarrow 0\rightarrow 0\rightarrow H^{1}(H_{0},\mathcal{O}_{H_{0}}(t))\rightarrow 0\rightarrow \ldots \end{split}

Hence, for any t, H^{1}(H_{0},\mathcal{O}_{H_{0}}(t))=0 and

\begin{aligned} dim H^{0}(H_{0},\mathcal{O}_{H_{0}}(t)) & =dim H^{0}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t))-dim H^{0}(\mathbb{P}^{4},\mathcal{O}_{\mathbb{P}^{4}}(t-(q+1)) \\ & =\binom{4+t}{t}-\binom{4+t-(q+1)}{t-(q+1)}. \end{aligned}

Consequently, for any q+1<t<q^{3}+1, the last sequence then gives

\Gamma(Z,\mathcal{O}_{Z}(t))\cong H^{0}(H_{0},\mathcal{O}_{H_{0}}(t)),

so, when in the range wherein the bound on the minimum distance ensures the injectivity of the evaluation map, the dimension of the code is

k=\sum_{\substack{i_{1},i_{2}\in\mathbb{N}^{2} \\ i_{1}+i_{2}=b}}\binom{4+i_{1}t_{1}+i_{2}t_{2})}{i_{1}t_{1}+i_{2}t_{2})}-\binom{4+i_{1}t_{1}+i_{2}t_{2})-(q+1)}{i_{1}t_{1}+i_{2}t_{2})-(q+1)}.

Finally, by Proposition 9.4 in [EisenbudHarris16], any section of L\vert_{T_{2 B_{i}}} (recall that T_{2 B_{i}}=p_{2}^{-1}(B_{i})) intersects C_{i}\subset T_{4 B_{i}} in an hyperplane, so \mathit{l}=0 and we can conclude the following bound for the minimum distance:

d\geq n-(q^{5}+1)(q^{3}+1)(q^{2}+1)b.
Example

Example 4.5

Z0V8

Let us consider the particular case q=2, with b=1, n_{1}=n_{2}=3, m_{1,j}=m_{2,j}=1 for j=1,2,3, and m_{1,j}=m_{2,j}=0 otherwise, for the family of codes presented in Corollary 4.3. Then, we obtain codes with the following parameters:

  1. n=63;

  2. k=14;

  3. d\geq 42.

Example

Example 4.6

HP0X

For q=2, let us consider the particular case b=2, t_{1}=t_{2}=4 for the family of codes presented in Corollary 4.4. Then we obtain a code with the following parameters:

  1. n=7425,

  2. k=1107,

  3. and d\geq 4455.

Source and licence

The text on this page is from arXiv:2401.11433v1 by Daniel Camazón Portela and Juan Antonio López Ramos, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2401/2401.11433.jsonl.