Open Access 17-07-2024
Curve-lifted codes for local recovery using lines
Published in: Designs, Codes and Cryptography | Issue 11/2024
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by (Link opens in a new window)
Abstract
1 Introduction
Algebraic geometry codes were introduced in the 1980s [7] and quickly received attention due to the existence of sequences with parameters exceeding the Gilbert-Varshamov bound [19]. These codes are defined by evaluating functions on a smooth projective curve over a finite field at rational points, a construction that is quite flexible allowing for customizations or variants that achieve particular goals. In [2], it was demonstrated how coverings of curves yield locally recoverable codes (also known as LRCs or codes with locality). A code C of length n is said to be locally recoverable if there exist recovery sets \(R_1, \dots , R_n\) such that for every codeword coordinate i, the \(i^{th}\) coordinate \(c_i\) of any codeword \(c \in C\) may be recovered from the codeword symbols \(c\mid _{R_i}\). In this case, the code C is said to have locality R relative to the recovery sets \(R_1, \dots , R_n\), where R is the cardinality of the largest \(R_i\). When the context is clear, we may say that a code C has locality R to mean there exists a recovery set for each coordinate of cardinality at most R. Hence, as we will see a code may have various localities. A code has availability t if each coordinate has t disjoint recovery sets and is called a code with availability if \(t>1\).
Locally recoverable codes, whose study originated in [6, 17], are studied due to their applicability in settings such as distributed storage where vast amounts of data are stored across many servers which may be temporarily offline (and is modeled as an erasure). To limit network traffic involved in recovery, it is desirable that each coordinate (or server) can be recovered using information from a small subset of all other coordinates (or the rest of the network). Moreover, it is useful to have multiple ways in which information can be recovered, so that if a coordinate is lost, the ability to recover does not depend on the availability of some other coordinate which might also be lost.
Advertisement
Algebraic geometry codes can be adapted to define locally recoverable codes, as noted in [2, 9]. In this paper, we define curve-lifted codes, a generalization of the Hermitian-lifted codes [11], for a projective curve \({\mathcal {X}}\) over a finite field \(\mathbb {F}_q\). Curve-lifted codes are evaluation codes in which codewords are determined by evaluating particular functions at affine points on the curve. The functions to evaluate depend on what we will refer to as an intersection number: given a curve \({\mathcal {X}}\) and a collection of lines \({\mathbb {L}}\), the intersection number of the curve relative to \({\mathbb {L}}\) isCodewords arise from the evaluation of rational functions f on X that restrict to low-degree polynomials on \(\left( L \cap {\mathcal {X}} \right) \left( \mathbb {F}_q \right) \). Recovery sets for a position associated with a point P consist of collections of other points of intersection between a line through P and \({\mathcal {X}}\). The parameters of such codes depend on the collection \({\mathbb {L}}\) of lines selected and the number of \(\mathbb {F}_q\)-rational points that lie on both the curve \({\mathcal {X}}\) and a line \(L \in {\mathbb {L}}\). When \({\mathbb {L}}\) is the set of all lines in affine space over \(\mathbb {F}_q\), we refer to the intersection number of the curve relative to \({\mathbb {L}}\) as the intersection number of the curve.
$$\begin{aligned} \min \left\{ \mid \left( L \cap {\mathcal {X}} \right) \left( \mathbb {F}_q \right) \mid : L \in {\mathbb {L}} \right\} . \end{aligned}$$
Intersection numbers of particular curves can be challenging to determine. We demonstrate progress in this direction for the norm-trace curve which is defined byover the field \(\mathbb {F}_{q^r}\) with \(q^r\) elements. Taking \(r=2\) gives the Hermitian curve \({\mathcal {X}}_{q,2}: y^q+y=x^{q+1}\) over \(\mathbb {F}_{q^2}\). This allows us to give important instances of the curve-lifted construction, resulting in norm-trace-lifted codes.
$$\begin{aligned} {\mathcal {X}}_{q,r}: y^{q^{r-1}}+ y^{q^{r-2}}+ \dots +y^q + y = x^{\frac{q^r-1}{q-1}} \end{aligned}$$
Curve-lifted codes and norm-trace-lifted codes are strongly inspired by Hermitian-lifted codes. We note some key distinctions between the binary Hermitian case (\(r=2\)) and the more general norm-trace one in which \(r>2\). Much of the effort in studying the Hermitian case is devoted to finding enough functions to get a positive rate. It should be mentioned that lifted codes are not traditional algebraic geometry codes; while the codewords are of a similar form, obtained by evaluating functions at rational points, the space of functions is not the Riemann-Roch space of a divisor on the curve. Curve-lifted codes are motivated by but should not to be confused with lifted Reed-Solomon codes [8]. Reed-Solomon codes are formed from evaluating polynomials of bounded degree at points on the projective line \({\mathbb {P}}\). It is for this reason that codes defined from positive genus curves, such as the norm-trace curve, can provide codes of same length over smaller fields than Reed-Solomon codes. Lifted Reed-Solomon codes are defined by evaluating multivariate polynomials that restrict to low-degree polynomials on lines. While one may consider a curve-lifted construction relative to \({\mathbb {P}}\), it does not yield interesting locally recoverable codes with large availability. Indeed, recovery sets for a position associated with a point P are found by taking lines through P and determining the other evaluation points that lie on L. However, the only line for which \(\mid L \cap {\mathbb {P}} \mid \ge 2\) is \(L\,=\,{\mathbb {P}}\).
Determining the functions to evaluate to define a curve-lifted code is a crucial task. As we will see in the norm-trace case when \(r>2\), we first bound the number of rational points of intersection between the curve \(\mathcal {X}_{q,r}\) and a line \(\mathbb {F}_{q^r}\). That is a primary contribution of this work. It pays off in that once this bound is determined, we can immediate specify enough functions which yield codewords in the norm-trace-lifted code to prove that this family of codes has positive rate even as the code length grows, for a fixed characteristic. It is an open problem to explicitly describe all functions that give rise to codewords.
Advertisement
This paper is organized as follows. Section 2 reviews the necessary terminology and the Hermitian-lifted codes. We define curve-lifted codes in Sect. 3 and give examples of them. The cardinalities of intersections between the norm-trace curves and lines are found in Sect. 4, laying a foundation to define an array of norm-trace-lifted codes over arbitrary fields in Sect. 5. The paper ends with a conclusion in Sect. 6.
2 Preliminaries
In this section, we review notation to be used throughout the paper as well as the background on other curve-lifted codes in the literature. Throughout, we let q be a power of a prime and \(r\ge 2\) be an integer. The finite field with q elements is denoted \(\mathbb {F}_q\), and the multiplicative group of its nonzero elements is denoted \(\mathbb {F}_q^\times \). The set of nonnegative integers is denoted by \(\mathbb {N}\), and for a positive integer n, \([n]\,:=\,\{1, \dots , n\}\).
An [n, k, d] linear code C over a finite field \(\mathbb {F}_q\) is a k-dimensional \(\mathbb {F}_q\)-subspace of \(\mathbb {F}_q^n\) in which any two distinct elements (called codewords) differ in at least d coordinates. Such a code has length n, dimension k, and minimum distance d. Any \(d-1\) erasures in a received word may be recovered by accessing the remaining \(n-d+1\) coordinates. We are interested in recovering erasures by accessing a small number of coordinates. As is standard in the erasure recovery model, we assume a received word w has the form \(w \in \mathbb {F}_q^n \cup \{ ? \}\) where there exists a codeword \(c \in C\) such that \(w_i \in \{ c_i, ? \}\) for all \(i \in [n]\). All codes considered in this paper are linear, so we use the terms code and linear code interchangeably.
A code C of length n over \(\mathbb {F}_q\) has locality s if and only if for all \(i \in \{ 1,\dots ,n\}\), there existswhere \(|R_i| \le s\) and for all codewords \(c \in C\),for some function \(\varphi _i: \mathbb {F}_q^s \rightarrow \mathbb {F}_q\). In this case, we call C a code with locality s. The set \(R_i\) is called a recovery set for i. The set \(\overline{R_i} \,:=\, R_i \cup \{ i \}\) is called a repair group for i. A code with locality s clearly has locality \(s'\) for all \(s' \ge s\) by taking recovery sets \(S_i = R_i \cup \left\{ j_1, \dots , j_{s'-\mid R_i \mid }: j_1, \dots , j_{s'-\mid R_i \mid } \in [n] {\setminus } \overline{R_i} \right\} \) for all \(i \in [n]\). When particular recovery sets \(R_i\), \(i \in [n]\), are clear from the context, we say that C has locality \(\max \{ |R_i|: i \in [n] \}\). We say that C has availability t if there exist recovery setsfor each index \(i \in [n]\), such thatfor all \(j \ne j'\).
$$\begin{aligned} R_i \subseteq [n] \backslash \{ i \} \end{aligned}$$
$$\begin{aligned} c_i = \varphi _i (c\mid _{R_i}) \end{aligned}$$
$$\begin{aligned} R_{i,1},\dots ,R_{i,t} \subseteq [n] \backslash \{ i \} \end{aligned}$$
$$\begin{aligned} R_{i,j} \cap R_{i,j'} = \emptyset \end{aligned}$$
The codes considered in this paper are reminiscent of algebraic geometry codes, in that they are defined using rational points and functions on curves over finite fields. An algebraic geometry code C(D, G) of length n over a finite field \(\mathbb {F}\) is defined by fixing a curve \({\mathcal {X}}\) over \(\mathbb {F}\) along with divisors G and D on \({\mathcal {X}}\) so that their supports are disjoint and D is the sum of \(\mathbb {F}\)-rational points \(P_1, \dots , P_n\). Each codeword is of the form \((f(P_1), \dots , f(P_n))\) where f is a function in the Riemann-Roch space of G. If D is not specified, it is taken to be the sum of all \(\mathbb {F}\)-rational points other than those in the support of G. For example, Reed-Solomon codes are algebraic geometry codes on projective lines. The next most commonly studied algebraic geometry codes are Hermitian codes. Hermitian codes are special cases of norm-trace codes, meaning algebraic geometry codes defined on norm-trace curves. Norm-trace codes were first defined by Geil [5]. They are defined by evaluating functions from Riemann-Roch spaces at rational points onover the field \(\mathbb {F}_{q^r}\) with \(q^r\) elements; here and throughout, the norm and trace will be considered with respect to the extension \(\mathbb {F}_{q^r}/\mathbb {F}_q\), so that for any \(a \in \mathbb {F}_{q^r}\), they areNote that \({\mathcal {X}}_{q,r}\) has genusLet \({\mathcal {X}}_{q,r}(\mathbb {F}_{q^r})\) denote the set of \(\mathbb {F}_{q^r}\)-rational point on the curve \({\mathcal {X}}_{q,r}\). For each \(a \in \mathbb {F}_{q^r}\), there are \(q^{r-1}\) elements \(b \in \mathbb {F}_{q^r}\) such that \({{\,\textrm{Norm}\,}}(a)={{\,\textrm{Tr}\,}}(b)\), each giving rise to a rational point \((a,b) \in {\mathcal {X}}_{q,r}(\mathbb {F}_{q^r})\), referred to as an affine point of \( {\mathcal {X}}_{q,r}\). In addition, there is a unique point at infinity, \(P_{\infty } \in {\mathcal {X}}_{q,r}(\mathbb {F}_{q^r})\). Hence, \(\left| {\mathcal {X}}_{q,r}(\mathbb {F}_{q^r}) \right| = q^rq^{r-1}+1=q^{2r-1}+1\). Consider the vector space of functions \(V \subseteq \cup _{m=0}^{\infty } {\mathcal {L}}(mP_{\infty })\) with no poles other than at \(P_{\infty }\), where \(P_{\infty }\,:=\,(0:1:0)\) is the unique point at infinity on \(\mathcal {X}_{q,r}\). It is worth noting that the Riemann-Roch space of the divisor \(mP_{\infty }\) isWe denote the set of polynomials with coefficients in \(\mathbb {F}_{q^r}\) and indeterminates x, y of total degree at most k by \(\mathbb {F}_{q^r}[x,y]_{\le k}\). A Hermitian code is an algebraic geometry code over the Hermitian curve. For convenience, we will identify rational functions of the form \(\frac{f(x,y)}{z^{\deg f}}\) where \(f(x,y) \in \mathbb {F}[x,y]\) with the polynomial \(f(x,y) \in \mathbb {F}[x,y]\).
$$\begin{aligned} {\mathcal {X}}_{q,r}: {{\,\textrm{Norm}\,}}(x)={{\,\textrm{Tr}\,}}(y) \end{aligned}$$
$$\begin{aligned} {{\,\textrm{Norm}\,}}(a) \,:=\, a^{\frac{q^r-1}{q-1}}, {{\,\textrm{Tr}\,}}(a)\,:=\,\sum _{i=0}^{r-1} a^{q^{i}} \in \mathbb {F}_q. \end{aligned}$$
$$\begin{aligned} g=\frac{1}{2}\left( \frac{q^r-1}{q-1} -1\right) \left( q^{r-1}-1\right) . \end{aligned}$$
$$\begin{aligned} {\mathcal {L}}(mP_{\infty }) = \bigg \langle x^i y^j: 0 \le i \le \frac{q^r-1}{q-1}-1, iq^{r-1}+j\frac{q^r-1}{q-1} \le m\bigg \rangle \subseteq \mathbb {F}_{q^r}[x,y]. \end{aligned}$$
(1)
A Hermitian code with \(m \le q^2-1\) over \(\mathbb {F}_{q^2}\) has locality q and availability \(q^2-1\) [11]. To see this, note that each affine \(\mathbb {F}_{q^2}\)-rational point P on the Hermitian curve \({\mathcal {X}}_{q,2}\) has the property that any non- tangent line to \({\mathcal {X}}_{q,2}\) intersects the curve in \(q+1\) \(\mathbb {F}_{q^2}\)-rational points and there are \(q^2-1\) such lines. Because any function f in the Riemann-Roch space \({\mathcal {L}}(mP_{\infty })\) may be expressed as \(f(x,y)=\sum _{i=0}^{q-1} \sum _{j=0}^{\lfloor \frac{m-iq}{q+1} \rfloor } c_{ij}x^i y^j\) for some constants \(c_{ij}\), such a function restricted to a non-tangent line can be viewed as a univariate polynomial of degree at most \(q-1\). To recover an erasure at point \(P_{ab}\), one may treat the word corresponding to the \(\mathbb {F}_{q^2}\)-rational points on any line through \(P_{ab}\) as a Reed-Solomon codeword. Thus, the set of \(\mathbb {F}_{q^2}\)-rational points on any non-tangent line through \(P_{ab}\), other than \(P_{ab}\) itself, form a recovery set for the coordinate corresponding to \(P_{ab}\), demonstrating that the Hermitian code \(C(mP_{\infty }, D)\) over \(\mathbb {F}_{q^2}\) has locality q and availability \(q^2-1\). Unfortunately, the rate of these codes approaches 0 as q goes to infinity.
Lifting is a mechanism introduced to increase the rate of codes while maintaining desirable properties. Hermitian-lifted codes over binary fields were introduced in [11] and yield a family of codes with a positive lower bound on the rate as q goes to infinity. More recent work with an improved bound on the rate of Hermitian-lifted codes appears in [1].
3 Curve-lifted codes
In this section, we present curve-lifted codes, which are evaluation codes whose codewords arise from functions that restrict to low degree polynomials on a collection of lines through the curve. In the first subsection, we develop the notion and consider some examples. As we will see, intersection numbers will be a necessary ingredient, and they will be further explored in later sections. In the second subsection, we drill down to better describe the defining sets of functions for particular curves.
3.1 Construction
Consider a projective curve \({\mathcal {X}}\) given by \(F(x,y)=0\) over the finite field \(\mathbb {F}_{q}\). Take an \(\mathbb {F}_{q}\)-rational point \(P \in {\mathcal {X}}(\mathbb {F}_{q})\) and enumerate the other \(\mathbb {F}_{q}\)-rational points on \({\mathcal {X}}\): \(P_1,\dots , P_n\). Set \(D=P_1+\dots +P_n\), and let \(V \subseteq \mathbb {F}_{q}({\mathcal {X}})\) be an \(\mathbb {F}_q\)-vector space of rational function on \({\mathcal {X}}\) with no poles among \(P_1, \dots , P_n\). Consider the mapWe define an evaluation code C(V, D) to be the image of the map \({{\,\textrm{ev}\,}}_D\). Its properties will depend on V and \(\{ P_1, \dots , P_n \}\).
$$\begin{aligned} \begin{array}{lccc} {{\,\textrm{ev}\,}}_D :&{}V &{}\rightarrow &{}\mathbb {F}_{q}^n \\ &{}f &{}\mapsto &{}(f(P_1), \dots , f(P_n)). \end{array}. \end{aligned}$$
We will be interested in non-horizontal lineswith \(\alpha ,\beta \in \mathbb {F}_{q}\), so \(\alpha \ne 0\). LetWe consider which rational functions \(f \in \mathbb {F}_q(\mathcal X)\) restrict to certain polynomials on the intersections of particular lines and the curve \({\mathcal {X}}\). To do so, letWhen F is clear from the context, we may write \(m_{\alpha , \beta }(x)\) for \(m_{\alpha , \beta ,F}(x)\). It is relevant to consider functions f on \({\mathcal {X}}\) modulo the polynomial \(m_{\alpha , \beta }\). More precisely, we use the following notion.
$$\begin{aligned} L_{\alpha ,\beta }(x,y):y=\alpha x + \beta \end{aligned}$$
$$\begin{aligned} {\mathbb {L}}_{q}\,:=\, \left\{ L_{\alpha ,\beta }: \alpha , \beta \in \mathbb {F}_{q}, \alpha \ne 0 \right\} . \end{aligned}$$
$$\begin{aligned} m_{\alpha , \beta , F}(x)\,:=\,F(x,\alpha x + \beta ). \end{aligned}$$
Definition 3.1
For a polynomial \(f(t) \in \mathbb {F}_{q}[t]\), define \({\bar{f}}_{\alpha , \beta }(t)\) to be the remainder resulting upon division of f(t) by \(m_{\alpha , \beta }(t)\); that is,Set
$$\begin{aligned} {\bar{f}}_{\alpha , \beta }(t) \,:=\,f(t) \text { mod } m_{\alpha , \beta }(t). \end{aligned}$$
$$\begin{aligned} \deg _{\alpha , \beta }(f) \,:=\,\deg ({\bar{f}}_{\alpha , \beta }(t)). \end{aligned}$$
Note thatfor all \(f \in \mathbb {F}_{q^r}[t]\). We also write \(f \equiv _{\alpha ,\beta } h\) to mean \(f \equiv h \mod m_{\alpha ,\beta }\), omitting subscripts if they are clear from the context.
$$\begin{aligned} \deg \left( {\bar{f}}_{\alpha , \beta }(t) \right) \le \deg \left( m_{\alpha , \beta } \right) -1 \end{aligned}$$
Consider for each point \(P_i\), \(i \in [n]\), the set of lines \({\mathbb {L}}_i\) containing the point \(P_i\). LetNotice that any line L through \(P_i\) intersects \(\mathcal {X}\) at at least \(B-1\) other \(\mathbb {F}_q\)-points.
$$\begin{aligned} B \le \min \left\{ \mid \left( L \cap \mathcal {X}\right) (\mathbb {F}_{q}) \mid : L \in \mathbb L_i, i \in [n] \right\} . \end{aligned}$$
Definition 3.2
Given a curve \({\mathcal {X}}\) over \(\mathbb {F}_{q}\) with divisor \(D:=P_1+\dots +P_n\) supported by n distinct \(\mathbb {F}_q\)-rational points, a collection of lines \({\mathbb {L}} \subseteq {\mathbb {L}}_q\) with \(\mathbb L \cap {\mathbb {L}}_i \ne \emptyset \) for all \(i \in [n]\), and an integer B, the associated curve-lifted code is \(C(D, \mathcal F_{{\mathbb {L}}, B})\) where
$$\begin{aligned} \mathcal {F}_{{\mathbb {L}}, B} \,:=\, \left\{ f \in \mathbb {F}_{q}\left( {\mathcal {X}} \right) : \exists g \in \mathbb {F}_{q}[t]_{\le B-2} \text { with } f \circ L \equiv g \text { }\forall L \in \mathbb {L} \right\} . \end{aligned}$$
(2)
According to Definition 3.2, the codewords in a curve-lifted code are obtained by evaluating at each point in the support of D functions which restrict on all lines in \({\mathbb {L}}\) to low-degree polynomials. In the next result, we see that this construction provides locality and availability.
Proposition 3.3
The curve-lifted code \(C(D, {\mathcal {F}}_{{\mathbb {L}}, B})\) is a code of length \(n \le |{\mathcal {X}} \left( \mathbb {F}_{q} \right) |\) over \(\mathbb {F}_{q}\) with locality \(B-1\) and availability \(q-1\).
Proof
For \(i \in [n]\), consider the set of lines \({\mathbb {L}}_i\) through the \(\mathbb {F}_{q}\)-rational point point \(P_i\) on \(\mathcal {X}\). By definition, any line \(L \in {\mathbb {L}}\) that contains \(P_i\) intersects \(\mathcal {X}\) in at least \(B-1\) other points among the \(P_j, j \in [n]{\setminus } \{ i \}\). Let \(R_{i,L} \subseteq \left( L \cap \mathcal X\right) \left( \mathbb {F}_{q} \right) {\setminus } \{ P_i \}\) such that \(| {\mathcal {R}}_{i,L} | = B -1\).
Consider a received word w resulting from \({{\,\textrm{ev}\,}}(f)\) in which there is an erasure in the coordinate corresponding to \(P_i\). We claim that \(R_{i,L}\) is a recovery set for position i, for all \(L \in {\mathbb {L}}_i\). To demonstrate this fact, we must determine from \(R_{i,L}\) the value \(f(P_i)\). Observe that for each of the points in the set \(R_{i,L}\,:=\,\left\{ P_{j_1}, \dots P_{j_{B-1}} \right\} \), the value \(f(P_{j_t})\) is known. Since \(f\mid _L = g\), the values \(g \left( P_{j_1}\right) , \dots , g \left( P_{j_{B-1}}\right) \) are known. Because \(\deg g \le B-2\), the polynomial g may be found by interpolation using the \(B-1\) values \(g \left( P_{j_1}\right) , \dots , g \left( P_{j_{B-1}}\right) \). Then \(f(P_i)=g(P_i)\). Hence \(R_{i,L}\) is a recovery set for i. Moreover, the intersection of any two such lines L and \(L'\) satisfiesThus, the sets \(R_{i,L}\), \(L \in {\mathbb {L}}_i\) are disjoint recovery sets indicating that C has availabilty \(q-1\). \(\square \)
$$\begin{aligned} L \cap L' = \{ P_i\}. \end{aligned}$$
Example 3.4
Consider taking \({\mathcal {X}} = {\mathcal {X}}_{q,2}\), the Hermitian curve over \(\mathbb {F}_{q^2}\) where q is even. According to [11], \(B=q+1\). Hence, Proposition 3.3 states such codes have length \(q^3\), locality \(B-1=q\), and availability \(q^2-1\). These are precisely the Hermitian-lifted codes considered in [11].
Example 3.5
In this example, we consider norm-trace-lifted codes over fields \(\mathbb {F}_{3^r}\) for small values of r. When \(r=3\), we have the curvewhich has genus 48 and 243 \(\mathbb {F}_{27}\)-rational points other than \(P_{\infty }\). Using Magma to count points of intersection [3], we see that each line in affine space over \(\mathbb {F}_{27}\) intersects the curve in either 7, 10, or 13 \(\mathbb {F}_{27}\)-rational points. Takingand \({\mathbb {L}} = {\mathbb {L}}_{27}\) givesand the code \(C(D, {\mathcal {F}}_{{\mathbb {L}}, 7})\) of length 243 with locality 6 and availability \(3^3-1=26\). Codewords are of the form \({{\,\textrm{ev}\,}}_D(f)\) where \(\deg f\mid _{(L \cap {\mathcal {X}}_{3,3})(\mathbb {F}_{27})} \le 5\) for all lines L over \(\mathbb {F}_{27}\). As a result, an erasure can be recovered by utilizing only 6 of the other 242 coordinates and in 26 different (disjoint) ways, as each non-horizontal line forms a repair group for each point of intersection with \({\mathcal {X}}_{3,3}\). We may also note thatsoThe containment may be strict as in the Hermitian case; it will be further explored in Sect. 5.
$$\begin{aligned} {\mathcal {X}}_{3,3}: y^9+y^3+y=x^{13} \end{aligned}$$
$$\begin{aligned} B = \min \{ 7, 10, 13 \} =7 \end{aligned}$$
$$\begin{aligned} \mathcal {F}_{{\mathbb {L}}_{27}, 7} \,:=\, \left\{ f \in \mathbb {F}_{27}\left( {\mathcal {X}}_{3,3} \right) : \exists g \in \mathbb {F}_{27}[t]_{\le 5} \text { with } f \circ L \equiv g \text { }\forall L \in \mathbb {L} \right\} \end{aligned}$$
$$\begin{aligned} \bigg \langle x^ay^b: a + b \le 5 \bigg \rangle \subseteq \mathcal {F}_{{\mathbb {L}}_{27}, 7}, \end{aligned}$$
$$\begin{aligned} \bigg \langle {{\,\textrm{ev}\,}}_D(x^ay^b): a + b \le 5 \bigg \rangle \subseteq C(D,\mathcal {F}_{{\mathbb {L}}_{27}, 7}). \end{aligned}$$
We may also consider taking a proper subset of lines. Calculating intersection numbers using Magma [3], we see that for every point \((a,b) \in {\mathcal {X}}_{3,3}(\mathbb {F}_{27})\) with \(a \ne 0\), there are For the points \((0,b) \in {\mathcal {X}}_{3,3}(\mathbb {F}_{27})\), there are Taking \({\mathbb {L}}\) to be the sets of lines in a. and d. above for every point \((a, b) \in {\mathcal {X}}_{3,3}(\mathbb {F}_{27})\) and setting \(B=13\), Proposition 3.3 applies to give a code \(C(D,{\mathcal {F}}_{{\mathcal {L}}, 13})\) which has length 243, locality 12, and availability \(\min \{ 6, 13\}=6\). Notice that the code \(C(D,{\mathcal {F}}_{{\mathcal {L}}, 13})\) includes codewords defined by functions that become polynomials of degree at most 11 when restricted to the lines in a. and d. In particular,This suggests that taking only lines which intersect the curve in more points (13 opposed to 7) yields codes of larger dimension, a fact that will be considered in Sect. 5.
a.
6 lines through (a, b) meeting \({\mathcal {X}}_{3,3}\) in 13 points
b.
10 lines through (a, b) meeting \({\mathcal {X}}_{3,3}\) in 7 points
c.
10 lines through (a, b) meeting \({\mathcal {X}}_{3,3}\) in 10 points.
d.
13 lines through (0, b) meeting \({\mathcal {X}}_{3,3}\) in 13 points
e.
13 lines through (0, b) meeting \({\mathcal {X}}_{3,3}\) in 7 points.
$$\begin{aligned} \bigg \langle {{\,\textrm{ev}\,}}_D(x^ay^b): a + b \le 11 \bigg \rangle \subseteq C(D,\mathcal {F}_{{\mathbb {L}}_{27}, 13}). \end{aligned}$$
Taking instead \(r=4\), we have the curveover \(\mathbb {F}_{81}\) which has genus 507 and 2187 affine \(\mathbb {F}_{81}\)-rational points. Computations in Magma [3] indicate that each line in affine space of \(\mathbb {F}_{81}\) intersects \({\mathcal {X}}_{3,4}\) in 22, 28, or 31 \(\mathbb {F}_{81}\)-rational points. Thus, we takeand note thatHence, \(C(D,{\mathcal {F}}_{{\mathbb {L}}_{81},22})\) is a code over \(\mathbb {F}_{81}\) of length 2187, locality 21, and availability 80. It follows that an erasure can be recovered by utilizing only 20 of the 2186 other coordinates and in 80 different (disjoint) ways.
$$\begin{aligned} {\mathcal {X}}_{3,4}: y^{27}+y^9+y^3+y=x^{40} \end{aligned}$$
$$\begin{aligned} B= \min \left\{ 22, 28, 31 \right\} =22 \end{aligned}$$
$$\begin{aligned} \bigg \langle x^ay^b: a + b \le 20 \bigg \rangle \subseteq \mathcal {F}_{\mathbb L_{81}, 22} = \left\{ f \in \mathbb {F}_{81}\left( {\mathcal {X}}_{3,4} \right) : \exists g \in \mathbb {F}_{81}[t]_{\le 20} \text { with } f \circ L \equiv g \text { }\forall L \in \mathbb {L} \right\} . \end{aligned}$$
Example 3.6
In this example, we consider the curve \({\mathcal {X}}\) given by \(y^8+y=x^3\) over \(\mathbb {F}_{64}\), from the first family of non-classical curves described by Schmidt [18]. One may note that \({\mathcal {X}}\) is maximal [4] and a so-called Castle curve [15]. Note that \({\mathcal {X}}\) which has genus 7 and 176 \(\mathbb {F}_{64}\)-rational points other than \(P_{\infty }\). Using Magma to collect data [3], we see that each line in affine space over \(\mathbb {F}_{64}\) intersects the curve in either 1, 2, 4, or 7 \(\mathbb {F}_{64}\)-rational points. Takingand \({\mathbb {L}} = {\mathbb {L}}_{64}\) givesThus, to obtain curve-lifted codes on \({\mathcal {X}}\), we must take a proper subset of lines in \({\mathbb {L}}_{64}\). Calculating intersection numbers using Magma [3], we see that there are 168 points \(P_1, \dots , P_{168}\) such that for each \(P_i\), \(i \in [168]\), there are In addition, there are 8 points \(P_{169}, \dots , P_{176}\) such that for each \(P_i\), \(i \in \left\{ 169, \dots , 176 \right\} \), there are To design a curve-lifted code on \({\mathcal {X}}\) with locality 3, one could take the set \({\mathbb {L}}\) to consist of those lines in d. and g. above. In this case, the code \(C(P_1+\dots +P_{176}, \mathcal F_{{\mathbb {L}}, 4})\) has length 176 over \(\mathbb {F}_{64}\), locality 3, and availability \(\min \{ 30, 42 \} = 30\). It is formed by taking those functions \(f \in {\mathbb {F}}_{64}({\mathcal {X}})\) that reduce to quadratics on the intersection of each line \(L \in {\mathbb {L}}\) and the curve \({\mathcal {X}}\).
$$\begin{aligned} B = \min \{ 1, 2, 4, 7 \} =1 \end{aligned}$$
$$\begin{aligned} \mathcal {F}_{{\mathbb {L}}_{64}, 1} =\emptyset . \end{aligned}$$
a.
3 lines whose only point of intersection with the curve \({\mathcal {X}}\) in \(P_i\),
b.
12 lines through \(P_i\) meeting \({\mathcal {X}}\) in 2 points,
c.
11 lines through \(P_i\) meeting \({\mathcal {X}}\) in 3 points,
d.
30 lines through \(P_i\) meeting \({\mathcal {X}}\) in 4 points, and
e.
7 lines through \(P_i\) meeting \({\mathcal {X}}\) in 7 points.
f.
21 lines through \(P_i\) meeting \({\mathcal {X}}\) in 3 points and
g.
42 lines meeting the curve \({\mathcal {X}}\) in 4 points.
To increase the dimension, we might consider increasing the sizes of recovery sets. However, there are no lines which contain any of the points \(P_i\), \(i \in \{ 169, \dots , 176 \}\), and intersect the curve in more than 4 points. Consequently, Definition 3.2 does not support the design of a length 176 code over \(\mathbb {F}_{64}\) which has locality greater than 3. Even so, we note that each of the points \(P_i\), \(i \in [168]\), lies on 7 lines that intersect the curve in 7 points as specified in e. above. The Proposition 3.3 implies that \(C(P_1+\dots +P_{168}, \mathcal F_{{\mathbb {L}}, 7})\) has length 168 over \(\mathbb {F}_{64}\), locality 5, and availability 7.
As these examples show, a key component of these curve-lifted constructions is the determination of the number of points of intersection between the curve and lines. In Examples 3.5 and 3.6, computational tools were used to determine them for particular curves over specific fields. In order to determine infinite families of such codes, we need more sophisticated theoretical tools. We will demonstrate that in the next section where we bound intersection numbers of sets of lines on the norm-trace curve.
One may also note that the traditional code parameters dimension and minimum distance are absent in Proposition 3.3. Because locally recoverable codes are designed for erasure recovery using small sets of other coordinates, the minimum distance is not as relevant as in standard error correction or recovery of collections of erasures using the entire received word. However, rates for families of LRCs provide a useful gauge of their capabilities. To examine code rates, we will also need the more information on particular curves. We make some headway on this front in the next subsection.
Definition 3.7
A monomial \(M_{a,b}(x,y)\) is said to be good for \(\mathcal F_{{\mathbb {L}}, B}\) if for all lines \(L_{\alpha , \beta } \in \mathbb {L}\),
$$\begin{aligned} \deg _{\alpha , \beta }(M_{a,b} \circ L_{\alpha , \beta }) \le B-2. \end{aligned}$$
We may simply say that a monomial is good if the set of lines \({\mathbb {L}}\) and integer B are clear from the context. Some monomials are good regardless of the choice of \({\mathbb {L}}\). For instance, \(M_{a,b}(x,y)\) is good for \(\mathcal {F}_{{\mathbb {L}}, B}\) for each \((a,b) \in \mathbb {N}^2\) with \(a+b \le B-2\), for all \({\mathbb {L}} \subseteq {\mathbb {L}}_q\)
3.2 Sporadic monomials
In this subsection, we define monomials \(x^ay^b\) which are good but not simply because \(a+b\) is small enough (as mentioned above). This notion is made precise in the following definition.
Definition 3.8
A monomial \(x^a y^b\) is called sporadic for \(\mathcal {F}_{{\mathbb {L}}, B}\) if it is good for \(\mathcal {F}_{\mathbb L, B}\) and \(a + b \ge B-1\). A monomial good for \(\mathcal {F}_{{\mathbb {L}}, B}\) is called typical if it is not sporadic.
Example 3.9
Recall that Hermitian-lifted codes have locality q and we may think of them as \(C(D, {\mathcal {F}}_{{\mathbb {L}}_{q^2}, q+1})\) where \(D=P_1+\dots +P_{q^3}\) is supported by the \(\mathbb {F}_{q^2}\)-rational points on \(X_{q,2}: y^q+y=x^{q+1}\). The set of typical monomials isLoosely speaking, Hermitian-lifted codes are defined with two sets of monomials \(x^a y^b\): those with \(a + b \le q-1\), which are always good, and some with \(a + b \ge q\) that happen to reduce to small enough degree on all lines. The monomials in the latter set were called sporadic, since their behavior is not yet fully understood, meaning to date, only some of them have been described explicitly. For instance, according to [11, Theorem 10],is a subset of sporadic monomials.
$$\begin{aligned} \left\{ x^ay^b: a+b \le q-1 \right\} . \end{aligned}$$
$$\begin{aligned} \left\{ x^ay^b: \begin{array}{l} a \le q-1, b \le q^2-1, a+b\ge q, \exists l, i \in [l], 0 \le s \le i-1 \text { so that } b=wq+b' \\ \text { with } b'< 2^{l-1}, 2^i \mid w, a < 2^{l-1}, \text { and no } 2^s \text { term in binary expansions of } a \text { and }b' \end{array} \right\} \end{aligned}$$
Identifying explicit sporadic monomials for norm-trace curves and their impact on the code rate as in the Hermitian case remains an open problem.
4 Intersection numbers of norm-trace curves
In this section, we determine the number of points in \(\mathcal X_{q,r}(\mathbb {F}_{q^r})\) on an intersection of a line \(L_{\alpha ,\beta }\) where \(\alpha \not =0\) and \(\alpha ,\beta \in \mathbb {F}_{q^r}\) with the norm-trace curve \({\mathcal {X}}_{q,r}\). We loosely refer to the number of such points as an intersection number. Intersection numbers will be applied in Sect. 5 to construct norm-trace-lifted codes over fields of arbitrary characteristic. In particular, they will be used to set the value B as in Proposition 3.3 and a degree bound which will define an appropriate set of functions to support local recovery with high availability and positive rate. In particular, we will find an integer B, depending only on q and r, such that for all \(L_{\alpha ,\beta }\in \mathbb L_{q^r}\),which will ultimately play a role in the sizes of the recovery sets for the norm-trace-lifted codes.
$$\begin{aligned} B \le \# \left( L_{\alpha , \beta } \cap {\mathcal {X}}_{q,r} \right) (\mathbb {F}_{q^r}) \end{aligned}$$
Given \(\alpha , \beta \in \mathbb {F}_{q^r}\), it will be useful to consider the polynomialor \(m_{\alpha , \beta }\) for short. DefineFirst, we observe that if \(\alpha \) and \(\alpha '\) are nonzero elements with the same norm and \(\beta \) and \(\beta '\) have the same trace, then \(n_{q,r}(\alpha ,\beta )=n_{q,r}(\alpha ',\beta ')\).
$$\begin{aligned} m_{\alpha , \beta ,q,r}(x) \,:=\, x^{(q^r-1)/(q-1)} - {{\,\textrm{Tr}\,}}(\beta )-\sum _{i=0}^{r-1} (\alpha x)^{q^i} \in \mathbb {F}_{q^r}[x]_{\le \frac{q^r-1}{q-1}}, \end{aligned}$$
$$\begin{aligned} n_{q,r}(\alpha ,\beta ) \,:=\,\# (L_{\alpha ,\beta }\cap \mathcal X_{q,r})(\mathbb {F}_{q^r}). \end{aligned}$$
Lemma 4.1
For any \(\alpha \in \mathbb {F}_{q^r}^{\times }\) and \(\beta \in \mathbb {F}_{q^r}\), \(n_{q,r}(\alpha ,\beta )\) depends only on \({{\,\textrm{Tr}\,}}(\beta )\) and \({{\,\textrm{Norm}\,}}(\alpha )\).
Proof
First, plug in \(y=\alpha x + \beta \) into the equation for \(\mathcal X_{q,r}\) to obtainThen notice that \(n_{q,r}(\alpha , \beta )\) is the number of zeros in \(\mathbb {F}_{q^r}\) of the polynomial \(m_{\alpha ,\beta }(x)\). Making the substitution \(t=\alpha ^{-1}x\) yields the polynomialThe number of zeros in \(\mathbb {F}_{q^r}\) of g and \(m_{\alpha ,\beta }\) are the same since the map \(t\mapsto \alpha ^{-1} x\) is a bijection sending zeros of \(m_{\alpha ,\beta }\) to zeros of g, and the number of zeros of g depends only on \({{\,\textrm{Tr}\,}}(\beta )\) and \({{\,\textrm{Norm}\,}}(\alpha )\). \(\square \)
$$\begin{aligned} x^{(q^r-1)/(q-1)} = \sum _{i=0}^{r-1} (\alpha x+\beta )^{q^i} = \sum _{i=0}^{r-1} (\alpha x)^{q^i}+\beta ^{q^i} = {{\,\textrm{Tr}\,}}(\beta )+\sum _{i=0}^{r-1} (\alpha x)^{q^i}. \end{aligned}$$
$$\begin{aligned} g(t)&= m_{\alpha ,\beta }(\alpha ^{-1}x) \\&= (\alpha t)^{(q^r-1)/(q-1)} - {{\,\textrm{Tr}\,}}(\beta ) - \sum _{i=0}^{r-1} t^{q^i} \\&= {{\,\textrm{Norm}\,}}(\alpha )t^{(q^r-1)/(q-1)} - {{\,\textrm{Tr}\,}}(\beta ) - \sum _{i=0}^{r-1} t^{q^i}. \end{aligned}$$
Proposition 4.2
Let \(r \ge 2\) and \(\alpha , \beta \in \mathbb {F}_{2^r}\) with \(\alpha \ne 0\). If \({{\,\textrm{Tr}\,}}(\beta ) \ne 0\), thenIf \({{\,\textrm{Tr}\,}}(\beta ) = 0\), thenThus, for any line \(L_{\alpha , \beta } \in \mathbb {L}_{2,r}\), the cardinality of its intersection with the norm-trace curve \(\mathcal {X}_{2,r}\) over \(\mathbb {F}_{2^r}\) is
$$\begin{aligned} n_{2,r}(\alpha , \beta ) = 2^{r-1} -1. \end{aligned}$$
$$\begin{aligned} n_{2,r}(\alpha , \beta ) = 2^{r-1} +1. \end{aligned}$$
$$\begin{aligned} | L_{\alpha , \beta } \cap \mathcal {X}_{2,r}\left( \mathbb {F}_{2^r} \right) | \in \left\{ 2^{r-1} - 1, 2^{r-1} + 1 \right\} . \end{aligned}$$
Proof
To determine \(n_{2,r}(\alpha ,\beta )\), according to Lemma 4.1, we need only consider the two cases, depending on \({{\,\textrm{Tr}\,}}(\beta )=0\) or \({{\,\textrm{Tr}\,}}(\beta )=1\).
Notice that points in the intersection \(L_{\alpha , \beta } \cap \mathcal {X}_{2,r}\left( \mathbb {F}_{2^r} \right) \) correspond to roots of the polynomialwhich are also elements of \(\mathbb {F}_{2^r}\). Because the roots of \(x^{2^r-1}-x\) are precisely the \(\gamma \in \mathbb {F}_{2^r}\), the problem of determining \(n_{2,r}\) reduces to finding the degree of \(h(x) = \text {gcd}(m_{\alpha , \beta }(x),x^{2^r}-x)\) by Freshman’s Dream.
$$\begin{aligned} x^{2^r-1} - (\alpha x + \beta )^{2^{r-1}} + \cdots + (\alpha x + \beta )^2 + (\alpha x + \beta ) \end{aligned}$$
In the case \({{\,\textrm{Tr}\,}}(\beta ) = 0\), the Euclidean Algorithm revealswhich has degree \(2^{r-1}+1\). In the case \({{\,\textrm{Tr}\,}}(\beta ) = 1\),which has degree \(2^{r-1}-1\). Therefore, \(n_{2,r}(\alpha ,\beta ) = 2^{r-1} -1\) or \(n_{2,r}(\alpha ,\beta ) = 2^{r-1} +1\). \(\square \)
$$\begin{aligned} \text {gcd}(m_{\alpha , \beta }(x),x^{2^r}-x) = \alpha ^{2^{r-1}} x^{2^{r-1}+1} + \cdots + \alpha x^2 + x \end{aligned}$$
$$\begin{aligned} \text {gcd}(m_{\alpha , \beta }(t),x^{2^r}-x) = \alpha ^{2^{r-1}} x^{2^{r-1}-1} + \cdots + \alpha , \end{aligned}$$
For \(q \ne 2\), we observe more intricate behavior. We will use results of Moisio and Moisio-Wan, who build on work of Katz [10], to establish lower and upper bounds on \(n_{q,r}(\alpha ,\beta )\). For an integer \(r\ge 2\) and elements \(a,b\in \mathbb {F}_q\), letIn [10], Katz proves the following result on counting elements of \(\mathbb {F}_{q^r}\) with prescribed norm and trace:
$$\begin{aligned} N_{q,r}(a,b)=\#\{\alpha \in \mathbb {F}_{q^r}: {{\,\textrm{Tr}\,}}(\alpha )=a,{{\,\textrm{Norm}\,}}(\alpha )=b\}. \end{aligned}$$
Lemma 4.3
[10, Theorem 4] If \(r\ge 2\) and \(a,b\in \mathbb {F}_q^\times \), then
$$\begin{aligned} \left| N_{q,r}(a,b) - \frac{q^{r}-1}{q(q-1)}\right| \le rq^{(r-2)/2}. \end{aligned}$$
We will use the following improvement to Katz’ result when \(a\not =0\), due to Moisio and Wan [14].
Lemma 4.4
[14, Theorem 1.2] Let \(a,b\in \mathbb {F}_q^{\times }\) and \(r\ge 2\). Then
$$\begin{aligned} \left| N_{q,r}(a,b)-\frac{q^{r-1}-1}{q-1} \right| \le (r-1)q^{(r-2)/2}. \end{aligned}$$
The previous bound is complemented by the following result of Moisio [13], which provides a bound \(N_{q,r}(a,b)\) when \(a=0\).
Lemma 4.5
[13] Let \(r\ge 2\), \(b\in \mathbb {F}_q^{\times }\), and \(d=\gcd (r,q-1)\). We have
$$\begin{aligned} \left| N_{q,r}(0,b)-\frac{q^{r-1}-1}{q-1}\right| \le (d-1)q^{(r-2)/2}. \end{aligned}$$
Theorem 4.6
Let \(r\ge 2\), \(\alpha ,\beta \in \mathbb {F}_{q^r}\) with \(\alpha \not =0\), and \(d=\gcd (r,q-1)\). If \({{\,\textrm{Tr}\,}}(\beta )\not =0\), thenIf \({{\,\textrm{Tr}\,}}(\beta )=0\), then
$$\begin{aligned} n_{q,r}(\alpha ,\beta ) \ge q^{r-1}-(d-1+(r-1)(q-2))q^{(r-2)/2}-1. \end{aligned}$$
$$\begin{aligned} n_{q,r}(\alpha ,\beta )\ge q^{r-1}-(r-1)(q-1)q^{(r-2)/2}. \end{aligned}$$
Proof
Let \(a={{\,\textrm{Norm}\,}}(\alpha )^{-1}\) and \(b={{\,\textrm{Tr}\,}}(\beta )\). Let \(g(t)=a^{-1}t^{(q^r-1)/(q-1)}-b-\sum _{i=0}^{r-1}t^{q^i}\). We have that \(n_{q,r}(\alpha ,\beta )\) is the number of roots of g in \(\mathbb {F}_{q^r}\). Note that the roots of g are in bijection with the setThus the number of roots of g (and hence \(n_{q,r}(\alpha ,\beta )\)) is given byFirst, assume \(b\not =0\). When \(t=-b\), we have \(N_{q,r}(t,at+ab)=N_{q,r}(-b,0)=0\). If \(t\not =-b\), then \(at+ab\not =0\) so we may apply Lemma 4.4 to bound \(N_{q,r}(t,at+ab)\). Since \(\alpha \not =0\), we may apply Lemma 4.5 to bound \(N_{q,r}(0,ab)\). We therefore obtain the desired lower bound for \(n_{q,r}(\alpha ,\beta )\):Now assume \(b=0\). Then applying Lemma 4.4 to each \(N_{q,r}(t,at)\) for \(t\not =0\) we get\(\square \)
$$\begin{aligned} \bigcup _{t\in \mathbb {F}_q} \{\gamma \in \mathbb {F}_{q^r}: {{\,\textrm{Tr}\,}}(\gamma ) =t \text { and } {{\,\textrm{Norm}\,}}(\gamma ) = at+ab\}. \end{aligned}$$
$$\begin{aligned} n_{q,r}(\alpha ,\beta )&= \#\{\gamma \in \mathbb {F}_{q^r}:g(\gamma )=0\} \\&= \sum _{t\in \mathbb {F}_q} \#\{\gamma \in \mathbb {F}_{q^r}: {{\,\textrm{Tr}\,}}(\gamma ) = t \text { and } {{\,\textrm{Norm}\,}}(\gamma ) = at+ab\} \\&= \sum _{t\in \mathbb {F}_q} N_{r}(t,at+ab). \end{aligned}$$
$$\begin{aligned} n_{q,r}(\alpha ,\beta )&= N_{q,r}(0,ab) + N_{q,r}(-b,0) + \sum _{\begin{array}{c} t\in \mathbb {F}_q \\ t\not =0,-b \end{array}} N_{r}(t,at+ab) \\&= N_{q,r}(0,ab) + \sum _{\begin{array}{c} t\in \mathbb {F}_q \\ t\not =0,-b \end{array}} N_{r}(t,at+ab) \\&\ge \frac{q^{r-1}-1}{q-1} -(d-1)q^{(r-2)/2} + \sum _{\begin{array}{c} t\in \mathbb {F}_q \\ t\not =0,-b \end{array}} \frac{q^{r-1}-1}{q-1} - (r-1)q^{(r-2)/2} \\&= \frac{q^{r-1}-1}{q-1} -(d-1)q^{(r-2)/2} + (q-2) \left( \frac{q^{r-1}-1}{q-1} - (r-1)q^{(r-2)/2}\right) \\&= \frac{q^{r-1}-1}{q-1}(1+q-2) - q^{(r-2)/2}(d-1+(q-2)(r-1)) \\&= q^{r-1}-(d-1+(q-2)(r-1))q^{(r-2)/2}-1. \end{aligned}$$
$$\begin{aligned} n_{q,r}(\alpha , \beta )&= N_{q,r}(0,0) + \sum _{\begin{array}{c} t\in \mathbb {F}_q \\ t\not =0 \end{array}} N_{r}(t,at) \\&\ge 1 + \sum _{\begin{array}{c} t\in \mathbb {F}_q \\ t\not =0 \end{array}} \frac{q^{r-1}-1}{q-1} - (r-1)q^{(r-2)/2} \\&= 1 + (q-1)\left( \frac{q^{r-1}-1}{q-1} - (r-1)q^{(r-2)/2}\right) \\&= q^{r-1}-(r-1)(q-1)q^{(r-2)/2}. \end{aligned}$$
Corollary 4.7
For any line \(L_{\alpha , \beta } \in {\mathbb {L}}_{q^r}\), the cardinality of its intersection with the norm-trace curve \(\mathcal X_{q,r}\) satisfiesThere are \(q^r-1\) lines in \({\mathbb {L}}_{q^r}\). Let \(d=\gcd (r,q-1)\). For those lines \(L_{\alpha , \beta } \in \mathbb L_{q^r}\) with \({{\,\textrm{Tr}\,}}(\beta ) \ne 0\), the cardinality of the intersection of \(L_{\alpha ,\beta }\) with the norm-trace curve \({\mathcal {X}}_{q,r}\) satisfiesif \(r \ne d\). There are \(q^r - q^{r-1}=q^{r-1}(q-1)\) such lines.
$$\begin{aligned} \left| L_{\alpha , \beta } \cap {\mathcal {X}}_{q,r}(\mathbb {F}_{q,r}) \right| \ge q^{r-1}-(r-1)(q-1)q^{(r-2)/2} -1. \end{aligned}$$
$$\begin{aligned} \left| L_{\alpha , \beta } \cap {\mathcal {X}}_{q,r}(\mathbb {F}_{q,r}) \right| \ge q^{r-1}-(r-1)(q-1)q^{(r-2)/2}-1 + q^{\frac{r-2}{2}} \end{aligned}$$
Proof
Notice that \(d=\gcd (r,q-1) \le r\). For convenience, let \(N=q^{r-1}-(q-1)(r-1)q^{(r-2)/2}-1\), which is the right-hand side of the lower bound on the intersection number for lines \(L_{\alpha , \beta }\) with \({{\,\textrm{Tr}\,}}(\beta ) \ne 0\), and \(Z= q^{r-1}-(d-1+(q-2)(r-1))q^{(r-2)/2} - 1\), which is the right-hand side of the lower bound on the intersection number for lines \(L_{\alpha , \beta }\) with \({{\,\textrm{Tr}\,}}(\beta ) =0\). Then for all \(\alpha , \beta \in \mathbb {F}_{q^r}\), \(n_{q,r}(\alpha , \beta ) \ge \min \{ N, Z \}\). Calculating the difference between the values given in the two lower bounds, we see thatTherefore,Consequently, every line \(L_{\alpha , \beta }\in {\mathbb {L}}\) satisfies\(\square \)
$$\begin{aligned} N-Z=(r-d)q^{\frac{r-2}{2}}-1 = {\left\{ \begin{array}{ll} -1 &{} \text {if } d=r \\ q^{\frac{r-2}{2}} - 1 + t, \text { for some } t\in \mathbb {N}&{} \text {otherwise}. \end{array}\right. } \end{aligned}$$
$$\begin{aligned} N = {\left\{ \begin{array}{ll} Z-1 &{} \text {if } d=r \\ Z-1+q^{\frac{r-2}{2}} + t, \text { for some } t\in \mathbb {N}&{} \text {otherwise}. \end{array}\right. } \end{aligned}$$
$$\begin{aligned} n_{q,r}(\alpha ,\beta ) \ge \min \{ N, Z \} \ge Z-1. \end{aligned}$$
We will use the bound in the previous corollary to establish families of norm-trace-lifted codes in arbitrary characteristic.
Remark 4.8
Notice Theorem 4.6 and Corollary 4.7 provide little to no information in the case \(r=2\). Indeed, \(q^{r-1}-(r-1)(q-1)q^{(r-2)/2}-1=q-(q-1)=1\). Moreover, in the case that q is even, \(d=\gcd (2,q-1)=1\) and \( q^{r-1}-\left( d-1+(r-1)(q-2)\right) q^{(r-2)/2}-1=q-\left( 1-1+(2-1)(q-2)\right) q^{(2-2)/2}=2.\) Furthermore, if q is odd, then \(d=2\) and \(q^{r-1}-\left( d-1+(r-1)(q-2)\right) q^{(r-2)/2}-1=q-\left( 2-1+(2-1)((q-2)\right) q^{(2-2)/2}=1\). Hence, these lower bounds are quite poor, since it is known that the actual value for \(n_{q,2}(\alpha , \beta )=q+1\).
5 Norm-trace-lifted codes
In this section, we combine the results from Sects. 3 and 4 to define and study norm-trace-lifted codes defined using the norm-trace curve \({\mathcal {X}}_{q,r}: {{\,\textrm{Tr}\,}}(y)={{\,\textrm{Norm}\,}}(x)\) over \(\mathbb {F}_{q^r}\) where q is any prime power. In light of Remark 4.8 and [11], we restrict our attention to \(r>2\). Now, having found a lower bound to take for B in Proposition 3.3 on intersection numbers as in Corollary 4.7, we may now consider codes defined by sets of rational functions that reduce on all lines to polynomials of degree at most \(B-2\). Also, because the curve itself is specified by a particular equation, we can obtain bounds on the code rates, including some that exceed the comparable Hermitian cases.
5.1 Constructions with highest availability
To obtain codes from the norm-trace curve with highest availability, we use points on every line through a point to form a repair group so as to obtain the largest number of disjoint recovery sets. With that in mind, consider taking \(B=B_{q,r}\) wherewith the goal of forming a repair group \(R_{ab,L}\) for an affine point \(P_{ab}\,:=\,(a,b)\) from each line \(L \in {\mathbb {L}}_{q^r}\) through \(P_{ab}\). We will use the shorthand notation \(\mathcal F_{q,r} \,:=\,\mathcal {F}_{{\mathbb {L}}_{q^r}, B_{q,r}}\) so thatTo do so, we will take a subset \(R_{ab,L} \subseteq L \cap \mathcal {X}_{q,r}(\mathbb {F}_{q^r})\) such that \(P_{ab} \in R_{ab,L}\) and \(|R_{ab,L} | = B_{q,r}\). Such a subset exists by Corollary 4.7. Recall thatNote thatfor all \(f \in \mathbb {F}_{q^r}[t]\), as \(\deg \left( m_{\alpha , \beta }(t) \right) = \frac{q^r-1}{q-1}\).
$$\begin{aligned} B_{q,r} :={\left\{ \begin{array}{ll} q^{r-1}-1 &{} \text {if } q=2 \\ q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}}-1 &{} \text {otherwise} \end{array}\right. } \end{aligned}$$
$$\begin{aligned} {\mathcal {F}}_{q,r}= \left\{ f \in \mathbb {F}_{q^r}[x,y]: \exists g \in \mathbb {F}_{q^r}[t]_{\le B_{q,r}-2} \text { with } f \circ L_{\alpha , \beta } \equiv g \text { for all } L_{\alpha , \beta } \in \mathbb {L}_{q^r} \right\} . \end{aligned}$$
$$\begin{aligned} m_{\alpha , \beta }(t) \,:=\, t^{(q^r-1)/(q-1)} - {{\,\textrm{Tr}\,}}(\beta )-\sum _{i=0}^{r-1} (\alpha t)^{q^i} \in \mathbb {F}_{q^r}[t]_{\le \frac{q^r-1}{q-1}}. \end{aligned}$$
$$\begin{aligned} \deg \left( {\bar{f}}_{\alpha , \beta }(t) \right) \le q^{r-1}+q^{r-2}+\dots +q \end{aligned}$$
According to (1), rational functions on \({\mathcal {X}}_{q,r}\) with no poles at any of the affine points are elements of \(\mathbb {F}_{q^r}[x,y]\), since \(\cup _{m \in \mathbb {N}} {\mathcal {L}}(mP_{\infty }) \subseteq \mathbb {F}_{q^r}[x,y]\).
Definition 5.1
The norm-trace-lifted code defined over \(\mathbb {F}_{q^r}\) is \(C(D, {\mathcal {F}}_{q,r})\), the image of \({\mathcal {F}}_{q,r}\) under the evaluation map \({{\,\textrm{ev}\,}}\); that is,
$$\begin{aligned} C(D, {\mathcal {F}}_{q,r}) \,:=\,\lbrace {{\,\textrm{ev}\,}}_D(f): f \in {\mathcal {F}}_{q,r} \rbrace \subseteq \mathbb {F}_{q^r}^{n}. \end{aligned}$$
Clearly, \(C(D, {\mathcal {F}}_{q,r})\) is a code of length n. To ascertain its dimension, we set out to determine the functions in \({\mathcal {F}}_{q,r}\). Based on the definition of \({\mathcal {F}}_{q,r}\), we are interested in polynomials f(x, y) such thatfor all \(L_{\alpha , \beta } \in {\mathbb {L}}_{q^r}\). Recall that \(M_{a,b}(x,y)\,:=\,x^ay^b\) where \(a, b \in \mathbb {N}\).
$$\begin{aligned} \deg _{\alpha , \beta }(f \circ L_{\alpha , \beta }) \le B_{q,r}-2 \end{aligned}$$
The number of \((a,b) \in \mathbb {N}^2\) with \(a+b \le B\) for some specified positive integer B is \(\sum _{a=0}^B B-a+1 =\frac{1}{2}(B+1)(B+2)\). To ensure that the set of such monomials gives rise to an independent set of codewords, we appeal to a result recorded in [16].
Lemma 5.2
[16, Lemma 2.14] The set of vectorsare linearly independent.
$$\begin{aligned} \left\{ {{\,\textrm{ev}\,}}(M_{a,b}(x,y)): 0 \le a \le \frac{q^r-1}{q-1}-1, 0 \le b \le q^{r-1}-1 \right\} \end{aligned}$$
Proof
Drawing inspiration from the proof of Proposition 5 of [11], we observe that the kernel of the evaluation map \({{\,\textrm{ev}\,}}\) is generated by \(x^{\frac{q^r-1}{q-1}} - y^{q^{r-1}} - \cdots - y^q - y\), \(x^{q^r}-x\), and \(y^{q^r}-y\). Under monomial orderings with \(x^{\frac{q^r-1}{q-1}} < y^{q^{r-1}}\),is a Gröbner basis for the kernel of the evaluation map, and so the evaluations of \(M_{a,b}\) cannot contain any element from the kernel of the evaluation map. Thus, the evaluations of \(M_{a,b}\) are linearly independent. \(\square \)
$$\begin{aligned} \left\{ x^{\frac{q^r-1}{q-1}} - y^{q^{r-1}} - \cdots - y^q - y, x^{q^r}-x \right\} \end{aligned}$$
Theorem 5.3
The norm-trace-lifted code \(C(D, {\mathcal {F}}_{q,r})\) over \(\mathbb {F}_{q^r}\) is a code of length \(q^{2r-1}\), dimension at leastlocality \(q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}}-2\), and availability \(q^r-1\). For fixed q, the rate approaches \(\frac{1}{2q}\) as the extension degree r goes to infinity.
$$\begin{aligned} \frac{1}{2} q^{r/2-1} \left( q^{r-1}-(q-1) (r-1) q^{r/2-1}+1\right) \left( q^{r/2}-q r+q+r-1\right) , \end{aligned}$$
Proof
Given an erasure in the coordinate corresponding to an affine point \(P_{ab}\) on \(\mathcal {X}_{q,r}\), consider the set of lines \(\mathcal L\) through \(P_{ab}\). According to Corollary 4.7, any line \(L \in {\mathcal {L}}\) intersects \(\mathcal {X}_{q,r}\) in at least \(B_{q,r}-1\) other affine points. Moreover, any function \(f \in \mathcal {F}_{q,r}\) has the property that \(f_{\mid L} \equiv g\) where \(\deg g \le B_{q,r}-2\). We claim that the \(B_{q,r}-1\) affine points in the intersection \(L \cap {X}_{q,r} \) other than \(P_{ab}\) may be used to interpolate and find g; that is, we claim that \(L \cap {X}_{q,r} {\setminus } \{ P_{ab} \}\) is a recovery set for the coordinate associated with \(P_{ab}\). Evaluating \(g(P_{ab})\) allows for recovery of the erased coordinate using \(B_{q,r}-1\) points. Hence, \(C(D, {\mathcal {F}}_{q,r})\) with this choice of recovery sets has locality \(B_{q,r}-1= q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}}-2\). Because there are \(q^r-1\) such lines L, \(C(D, {\mathcal {F}}_{q,r})\) has availability \(q^r-1\).
Notice that the set \(S\,:=\,\left\{ x^ay^b: a+b \le B_{q,r}-2 \right\} \) is a set of monomials good for \(\mathcal {F}_{q,r}\) of cardinalityAccording to Lemma 5.2, S is linearly independent, demonstrating that \(C(D, {\mathcal {F}}_{q,r})\) has dimension at leastand rateas \(r \rightarrow \infty \). \(\square \)
$$\begin{aligned} \frac{1}{2} \left( q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}} \right) \left( q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}}+1\right) . \end{aligned}$$
$$\begin{aligned} \frac{1}{2} \left( q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}} \right) \left( q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}}+1\right) . \end{aligned}$$
$$\begin{aligned} \frac{ \frac{1}{2} \left( q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}} \right) \left( q^{r-1}-(r-1)(q-1)q^{\frac{r-2}{2}}+1\right) }{q^{2r-1}} \rightarrow \frac{1}{2q}>0 \end{aligned}$$
Note that the rate is bounded away from 0 for fixed characteristic as the degree of the extension grows. Hence, the codes over fields of small characteristic provide the best asymptotic rates. We also observe that taking functions that reduce to low degree polynomials on all lines (rather than just some) provides the largest availability given by the lifted construction, since the lines are the recovery sets. We may also consider fewer lines with more refined intersection numbers, as in the next subsection.
5.2 Constructions with high availability and larger dimension
As suggested by Proposition 3.3, we may adapt the collections of lines \({\mathcal {L}}\) used in Definition 5.1 and the value B to a more refined bound on the intersection numbers. Considerand the collection of linesThen set \(\mathcal F_{q,r}'\,:=\,\mathcal {F}_{{\mathbb {L}}_{q^r}', B_{q,r}'}\) so thatLet D denote the sum of affine \(\mathbb {F}_{q^r}\)-rational points (a, b) with \(a \ne 0\) (so \({{\,\textrm{Norm}\,}}(a)\ne 0\)). We say the refined norm-trace-lifted code isWe say that \(M_{a,b}(x,y)\) is good for \(\mathcal {F}'_{q,r}\) if for all lines \(L_{\alpha , \beta } \in \mathbb {L}_{q,r}\), \( \deg _{\alpha , \beta }\left( M_{a,b} \circ L_{\alpha , \beta } \right) \le B_{q,r}'.\)
$$\begin{aligned} B_{q,r}'\,:=\, q^{r-1}-\left( \gcd (r,q-1)-1+(r-1)(q-2)\right) q^{(r-2)/2}-1 \end{aligned}$$
$$\begin{aligned} {\mathbb {L}}_{q^r}' = \left\{ L_{\alpha ,\beta }: {{\,\textrm{Tr}\,}}(\beta ) \ne 0 \right\} . \end{aligned}$$
$$\begin{aligned} \mathcal {F}_{q,r}' = \left\{ f \in \mathbb {F}_{q^r}[x,y]: \exists g \in \mathbb {F}_{q^r}[t]_{\le B_{q,r}'-2} \text { with } f \circ L_{\alpha , \beta } \equiv g \text { for all } L_{\alpha , \beta } \in \mathbb {L}_{q^r}', \right\} . \end{aligned}$$
$$\begin{aligned} {C}\left( D, {\mathcal {F}}_{{\mathbb {L}}_{q^r}', B_{q,r}'} \right) = \lbrace {{\,\textrm{ev}\,}}(f): f \in \mathcal {F}'_{q,r} \rbrace \subseteq \mathbb {F}_{q^r}^{n}. \end{aligned}$$
Notice that \(M_{a,b}(x,y)\) is good for \(\mathcal {F}'_{q,r}\) for each \((a,b) \in \mathbb {N}^2\) with \(a+b \le B_{q,r}'\). Consequently, we have the following result.
Proposition 5.4
The refined norm-trace-lifted code \(C'_{q,r}={{\,\textrm{ev}\,}}\left( \mathcal {F}'_{q,r}\right) \) is a code over \(\mathbb {F}_{q^r}\) of length \(q^{2r-1}-q^{r-1}\), locality \(q^{r-1}-\left( \gcd (r,q-1) -1+(r-1)(q-2)\right) q^{\frac{r-2}{2}}-2\), and availability \(q^r-q^{r-1}-2\) with rate bounded away from 0 as the code length grows for a fixed characteristic.
Proof
The proof is similar to that of Theorem 5.3. \(\square \)
In the next subsection, we draw some comparisons between these codes and other families.
5.3 Comparisons
Recall that the norm-trace-lifted codes defined over fields of arbitrary characteristic arise from the set of functions \(\mathcal {F}_{q,r}\) which depend on the degree bounds given in Theorem 4.6. These bounds hold for any q. Tighter bounds may give rise to codes with better parameters.
We begin this subsection with a demonstration of this.
Example 5.5
Consider a curve-lifted code on the norm-trace curve \(\mathcal X_{2,r}\) over \(\mathbb {F}_{2^r}\). First consider \(C(D,\left( \mathcal {F}_{{\mathbb {L}}_{2^r}, 2^{r-1}-(r-1)2^{(r-2)/2}} \right) )\). Recall that Theorem 4.6 givesfor every line \(L_{\alpha ,\beta }\in \mathbb {L}_{2^r}\), whereasoras shown in Proposition 4.2. Hence, a code with larger dimension is obtained by considering the binary norm-trace-lifted code \(C(D,\left( \mathcal {F}_{\mathbb L_{2^r}, B_{2,r}} \right) )\) which is a \([2^{2r-1},(0.25 - \varepsilon _r) \cdot 2^{2r-1},\ge 2^r]\) code with locality \({2^{r-1}-2}\), availability \({2^r-1}\), and asymptotic rate 0.25 [12, Theorem 3].
$$\begin{aligned} n_{2,r}(\alpha ,\beta ) \ge 2^{r-1}-(r-1)2^{(r-2)/2}-1 \end{aligned}$$
$$\begin{aligned} n_{2,r}(\alpha , \beta ) =2^{r-1} - 1 \end{aligned}$$
$$\begin{aligned} n_{2,r}(\alpha , \beta ) =2^{r-1} + 1, \end{aligned}$$
To increase the dimension further, we may take \(\mathbb L_{2^r}''\,:=\,\left\{ L_{\alpha , \beta }: {{\,\textrm{Tr}\,}}(\beta ) = 0 \right\} \). In doing so, according to the proof of Proposition 4.2, we obtain a code \(C (D, \mathcal F_{{\mathbb {L}}_{2^r}'', 2^{r-1}+1})\). We will see thatFurther comparisons are captured in Table 1.
$$\begin{aligned} \bigg \langle {{\,\textrm{ev}\,}}_B(x^a y^b): a+b = 2^{r-1}-1, 2^{r-1}-2 \bigg \rangle \in C (D, {\mathcal {F}}_{{\mathbb {L}}_{2^r}'', 2^{r-1}+1}) \setminus C(D,\left( \mathcal {F}_{{\mathbb {L}}_{2^r}, B_{2,r}} \right) ). \end{aligned}$$
Table 1
Parameters of one-point norm-trace, Hermitian-lifted, binary norm-trace-lifted, and refined binary norm-trace-lifted codes over \(\mathbb {F}_{2^r}\); see also [16]
1-pt norm-trace | HLC | NTLC | RNTLC | |
---|---|---|---|---|
Locality | \(2^{r-1}-2\) | \(2^{r / 2}\) | \(2^{r-1}-2\) | \(2^{r-1}\) |
Availability | \(2^r-1\) | \(2^r-1\) | \(2^r - 1\) | \(2^{r-1} - 1\) |
Length | \(2^{2r-1}\) | \(2^{3r / 2}\) | \(2^{2r-1}-2^{r-1}\) | \(2^{2r-1}\) |
Dimension | \(\le 2^{2r-4}-1\) | \(\ge 0.007 \cdot 2^{3r/2}\) | \((0.25 - \varepsilon _r) \cdot 2^{2r-1}\) | \((0.25 - \varepsilon _r) \cdot \) |
\(\left( 2^{2r-1} -2^{r-1} \right) \) | ||||
Asymptotic | ||||
Rate | \(\le \frac{1}{8} - \frac{1}{2^{2r-1}}\) | \(\ge 0.007\) | 0.25 | 0.25 |
Table 2
Collection of actual intersection numbers \(n_{q,r}\) for all \(\alpha , \beta \) and bounds
p | r | \(\#(L_{\alpha ,\beta }\cap \mathcal {X}_{p,r})(\mathbb {F}_{p^r})\) | \(B_{p,r}\) | \(B_{p,r}'\) |
---|---|---|---|---|
3 | 2 | 1, 4 | 0 | 0 |
3 | 3 | 13, 7, 10 | 1 | 4 |
3 | 4 | 22, 28, 31 | 8 | 14 |
3 | 5 | 73, 76, 85, 91 | 38 | 59 |
3 | 6 | 229, 244, 256 | 152 | 188 |
3 | 7 | 703, 715, 742, 757 | 540 | 634 |
5 | 2 | 1, 6 | 0 | 0 |
5 | 3 | 21, 26,31 | 6 | 10 |
5 | 4 | 111, 121, 126 141 | 64 | 64 |
5 | 5 | 561, 611, 621, 626, 641, 681 | 445 | 489 |
5 | 6 | 3056, 3106, 3126, 3131, 3206 | 2624 | 2724 |
5 | 7 | 15631, 15751, 15731, 15501, 15681, 15456 | 14282 | 14617 |
7 | 2 | 1, 8 | 0 | 0 |
7 | 3 | 43, 50, 57 | 16 | 16 |
7 | 4 | 351, 358, 316, 379, 337 | 216 | 230 |
7 | 5 | 2451, 2325, 2465, 2381, 2437, 2395, 2353, 2402 | 1955 | 2029 |
7 | 6 | 16773, 16738, 17053, 16843, 16801, 16633, 16808 | 15336 | 15336 |
7 | 7 | 117433, 117615, 118693, 117580, 118063, 116173, 117895, 117853, 117685, 117643, 117475 | 112980 | 113758 |
Remark 5.6
We recognize that while the binary norm-trace-lifted codes \(C(D,\left( \mathcal {F}_{{\mathbb {L}}_{2^r}, B_{2,r}} \right) )\) have dimension exceeding their counterparts \(C(D,\left( \mathcal {F}_{{\mathbb {L}}_{2^r}, 2^{r-1}-(r-1)2^{(r-2)/2}} \right) )\) defined from the same curve, their asymptotic rates behave similarly. Moreover, there is a tradeoff in sizes of recovery sets which may be larger for the codes \({{\,\textrm{ev}\,}}\left( \mathcal {F}_{b,2,r} \right) \) than \({{\,\textrm{ev}\,}}\left( \mathcal {F}_{2,r} \right) \), meaning more symbols are needed in order to perform recovery.
Remark 5.7
Consider the code \(C(D, {\mathcal {F}}_{{\mathbb {L}}_{q^r}, n_{q,r}})\) on the norm-trace curve \({\mathcal {X}}_{q,r}\) over \(\mathbb {F}_{q,r}\), whereWhile (to our knowledge) at present there is not a closed form expression for \(n_{q,r}\), values can be computed as in Table 2. Using the precise intersection numbers for odd q produces higher dimensional codes.
$$\begin{aligned} n_{q,r}\,:=\,\min \left\{ n_{q,r}(\alpha , \beta ): \alpha , \beta \in \mathbb {F}_{q^r} \right\} . \end{aligned}$$
We now provide an example to illustrate this fact.
Example 5.8
Consider the curve \({\mathcal {X}}_{7,3}\) over \(\mathbb {F}_{343}\). According to Table 2, \(n_{q,r}=43\). The curve-lifted code \(C(D, {\mathcal {F}}_{{\mathbb {L}}_{343}, 43})\) includes codewords given by evaluating \(x^ay^b\), \(a+b \le 41\), whereas the norm-trace-lifted code given by the bound \(n_{7,3} \ge 16\) only considers those with \(x^ay^b\), \(a+b \le 14\) and possibly some sporadic monomials.
6 Conclusion
In this paper, we defined curve-lifted codes which allow for local recovery by taking as repair groups the points of intersection of the curve with lines through the evaluation points. While inspired by Hermitian-lifted codes, they may exhibit different behaviors depending on the particular defining curve. In addition, we determined bounds on the number of affine points of intersection between a norm-trace curve and a line. We then used them to define norm-trace-lifted codes over fields of arbitrary characteristic. These new codes have high availability and positive rate, bounded away from zero as the code length goes to infinity. We note that codes over fields of small characteristic provide the best asymptotic rates. The opportunity to obtain greater rate by evaluating more functions may motivate one to consider more tailored bounds for the intersection numbers.
Acknowledgements
The work of the first and second authors is supported in part by the Commonwealth Cyber Initiative. The work of the first author is supported by NSF DMS-2201075. This work was performed while the third author was at Virginia Tech. Some results on binary norm-trace codes were presented at ISIT 2022 and in the third author’s dissertation.
Declarations
Conflict of interest
The authors have no financial or proprietary interests in any material discussed in this article.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.