arXiv++ Combinatorics

We compare ordinary and symmetric variants of two classical measures of pseudorandomness for binary sequences, the $2$-adic complexity and the linear complexity. In the periodic setting, we show that for binary periodic sequences constructed from the binary expansions of non-palindromic primes, the symmetric $2$-adic complexity can be strictly smaller than the ordinary $2$-adic complexity. We also give a direct proof (of the known result) that the linear complexity of a periodic binary sequence is invariant under reversal, and hence coincides with its symmetric version. In the aperiodic setting, we provide explicit families of finite binary sequences for which both the $N$th symmetric 2-adic complexity and the $N$th symmetric linear complexity are substantially smaller than their ordinary counterparts. Furthermore, we show that the expected values of the $N$th rational complexity and of the $N$th exponential linear complexity exceed those of their symmetric analogues by at least a term of order of magnitude $N$. Thus, the effect of symmetrization is clearly visible on an exponential scale. We also establish lower bounds for the expected values of the symmetric rational complexity, symmetric $2$-adic complexity, symmetric linear complexity, and symmetric exponential linear complexity.

Quaternary Legendre pairs are pertinent to the construction of quaternary Hadamard matrices and have many applications, for example in coding theory and communications. In contrast to binary Legendre pairs, quaternary ones can exist for even length $\ell$ as well. It is conjectured that there is a quaternary Legendre pair for any even $\ell$. The smallest open case until now had been $\ell=28$, and $\ell=38$ was the only length $\ell$ with $28\le \ell\le 60$ resolved before. Here we provide constructions for $\ell=28,30,32$, and $34$. In parallel and independently, Jedwab and Pender found a construction of quaternary Legendre pairs of length $\ell=(q-1)/2$ for any prime power $q\equiv 1\bmod 4$, which in particular covers $\ell=30$, $36$, and $40$, so that now $\ell=42$ is the smallest unresolved case. The main new idea of this paper is a way to separate the search for the subsequences along even and odd indices which substantially reduces the complexity of the search algorithm. In addition, we use Galois theory for cyclotomic fields to derive conditions which improve the PSD test.

In the most interesting case of safe prime powers $q$, Gómez and Winterhof showed that a subfamily of the family of Golomb Costas permutations of $\{1,2,\ldots,q-2\}$ of size $\varphi(q-1)$ has maximal cross-correlation of order of magnitude at most $q^{1/2}$. In this paper we study a larger family of Golomb Costas permutations and prove a weaker bound on its maximal cross-correlation. Considering the whole family of Golomb Costas permutations we show that large cross-correlations are very rare. Finally, we collect several conditions for a small cross-correlation of two Costas permutations. Our main tools are the Weil bound and the Szemerédi-Trotter theorem for finite fields.

We introduce quaternary Legendre pairs of length $\ell$. In contrast to binary Legendre pairs they can exist for even $\ell$ as well. First we show that they are pertinent to the construction of quaternary Hadamard matrices of order $2\ell+2$ and thus of binary Hadamard matrices of order $4\ell+4$. Then for a prime $p>2$ we present a construction of a pair of sequences of length $p$ from which we can derive quaternary Legendre pairs of length $\ell=2p$ by decompression for $p=3,5,7,13,19,31,41$. Moreover, we give also constructions of Legendre pairs of length $\ell$ for all remaining even $\ell\le 24$.

Let $q$ be a positive integer and $\mathcal{S}=\left\{x_0,x_1,\ldots,x_{T-1}\right\}\subseteq\mathbb{Z}_q=\{0,1,\ldots,q-1\}$ with $$0\leq x_0<x_1<\ldots< x_{T-1}\leq q-1.$$ We derive from $\mathcal{S}$ three (finite) sequences. 1. For an integer $M\geq 2$ let $(s_n)$ be the $M$-ary sequence defined by \begin{eqnarray*} s_n\equiv x_{n+1}-x_n \bmod M, \qquad n=0,1,\ldots, T-2. \end{eqnarray*} 2. For an integer $m\geq 2$ let $(t_n)$ be the binary sequence defined by \begin{eqnarray*} t_n=\left\{\begin{array}{ll} 1, & \hbox{if } 1\leq x_{n+1}-x_n\leq m-1, \\ 0, & \hbox{otherwise}, \end{array}\right. \qquad n=0,1,\ldots, T-2. \end{eqnarray*} 3. Let $(u_n)$ be the characteristic sequence of $\mathcal{S}$, \begin{eqnarray*} u_n=\left\{\begin{array}{ll} 1, & \hbox{if } n\in \mathcal{S}, \\ 0, & \hbox{otherwise}, \end{array}\right. \qquad n=0,1,\ldots, q-1. \end{eqnarray*} We study the balance and pattern distribution of the sequences $(s_n)$, $(t_n)$ and $(u_n)$. For sets $\mathcal{S}$ with desirable pseudorandom properties, more precisely, sets with low correlation measures, we show the following: 1. The sequence $(s_n)$ is (asymptotically) balanced and has uniform pattern distribution if $T$ is of smaller order of magnitude than $q$. 2. The sequence $(t_n)$ is balanced and has uniform pattern distribution if $T$ is approximately $\left(1-\frac{1}{2^{1/(m-1)}}\right)q$. 3. The sequence $(u_n)$ is balanced and has uniform pattern distribution if $T$ is approximately $\frac{q}{2}$. These results are motivated by earlier results for the sets of quadratic residues and primitive roots modulo a prime. We unify these results and derive many further (asymptotically) balanced sequences with uniform pattern distribution from pseudorandom subsets.

Let ${\mathbb F}_q$ be the finite field of $q$ elements, where $q=p^r$ is a power of the prime $p$, and $\left(β_1, β_2, \dots, β_r \right)$ be an ordered basis of ${\mathbb F}_q$ over ${\mathbb F}_p$. For $$ξ=\sum_{i=1}^rx_iβ_i, \quad \quad x_i\in{\mathbb F}_p,$$ we define the Thue-Morse or sum-of-digits function $T(ξ)$ on ${\mathbb F}_q$ by \[ T(ξ)=\sum_{i=1}^{r}x_i.%,\quad ξ=x_1β_1+\cdots +x_rβ_r\in {\mathbb F}_q. \] For a given pattern length $s$ with $1\le s\le q$, a subset ${\cal A}=\{α_1,\ldots,α_s\}\subset {\mathbb F}_q$, a polynomial $f(X)\in{\mathbb F}_q[X]$ of degree $d$ and a vector $\underline{c}=(c_1,\ldots,c_s)\in{\mathbb F}_p^s$ we put \[ {\cal T}(\underline{c},{\cal A},f)=\{ξ\in{\mathbb F}_q : T(f(ξ+α_i))=c_i,~i=1,\ldots,s\}. \] In this paper we will see that under some natural conditions, the size of~${\cal T}(\underline{c},{\cal A},f)$ is asymptotically the same for all~$\underline{c}$ and ${\cal A}$ in both cases, $p\rightarrow \infty$ and $r\rightarrow \infty$, respectively. More precisely, we have \[ \left||{\cal T}(\underline{c},{\cal A},f)|-p^{r-s}\right|\le (d-1)q^{1/2}\] under certain conditions on $d,q$ and $s$. For monomials of large degree we improve this bound as well as we find conditions on $d,q$ and $s$ for which this bound is not true. In particular, if $1\le d<p$ we have the dichotomy that the bound is valid if $s\le d$ and fails for some $\underline{c}$ and ${\cal A}$ if $s\ge d+1$. The case $s=1$ was studied before by Dartyge and Sárközy.

Let $γ$ be a generator of a cyclic group $G$ of order $n$. The least index of a self-mapping $f$ of $G$ is the index of the largest subgroup $U$ of $G$ such that $f(x)x^{-r}$ is constant on each coset of $U$ for some positive integer~$r$. We determine the index of the univariate Diffie-Hellman mapping $d(γ^a)=γ^{a^2}$, $a=0,1,\ldots,n-1$, and show that any mapping of small index coincides with~$d$ only on a small subset of $G$. Moreover, we prove similar results for the bivariate Diffie-Hellman mapping $D(γ^a,γ^b)=γ^{ab}$, $a,b=0,1,\ldots,n-1$. In the special case that $G$ is a subgroup of the multiplicative group of a finite field we present improvements.

We build on the work of Drakakis et al. (2011) on the maximal cross-correlation of the families of Welch and Golomb Costas permutations. In particular, we settle some of their conjectures. More precisely, we prove two results. First, for a prime $p\ge 5$, the maximal cross-correlation of the family of the $\varphi(p-1)$ different Welch Costas permutations of $\{1,\ldots,p-1\}$ is $(p-1)/t$, where $t$ is the smallest prime divisor of $(p-1)/2$ if $p$ is not a safe prime and at most $1+p^{1/2}$ otherwise. Here $\varphi$ denotes Euler's totient function and a prime $p$ is a safe prime if $(p-1)/2$ is also prime. Second, for a prime power $q\ge 4$ the maximal cross-correlation of a subfamily of Golomb Costas permutations of $\{1,\ldots,q-2\}$ is $(q-1)/t-1$ if $t$ is the smallest prime divisor of $(q-1)/2$ if $q$ is odd and of $q-1$ if $q$ is even provided that $(q-1)/2$ and $q-1$ are not prime, and at most $1+q^{1/2}$ otherwise. Note that we consider a smaller family than Drakakis et al. Our family is of size $\varphi(q-1)$ whereas there are $\varphi(q-1)^2$ different Golomb Costas permutations. The maximal cross-correlation of the larger family given in the tables of Drakakis et al. is larger than our bound (for the smaller family) for some $q$.

We study subsets of the $n$-dimensional vector space over the finite field $\mathbb{F}_q$, for odd $q$, which contain either a sphere for each radius or a sphere for each first coordinate of the center. We call such sets radii spherical Kakeya sets and center spherical Kakeya sets, respectively. For $n\ge 4$ we prove a general lower bound on the size of any set containing $q-1$ different spheres which applies to both kinds of spherical Kakeya sets. We provide constructions which meet the main terms of this lower bound. We also give a construction showing that we cannot get a lower bound of order of magnitude~$q^n$ if we take lower dimensional objects such as circles in $\mathbb{F}_q^3$ instead of spheres, showing that there are significant differences to the line Kakeya problem. Finally, we study the case of dimension $n=1$ which is different and equivalent to the study of sum and difference sets that cover $\mathbb{F}_q$.

Expansion complexity and maximum order complexity are both finer measures of pseudorandomness than the linear complexity which is the most prominent quality measure for cryptographic sequences. The expected value of the $N$th maximum order complexity is of order of magnitude $\log N$ whereas it is easy to find families of sequences with $N$th expansion complexity exponential in $\log N$. This might lead to the conjecture that the maximum order complexity is a finer measure than the expansion complexity. However, in this paper we provide two examples, the Thue-Morse sequence and the Rudin-Shapiro sequence with very small expansion complexity but very large maximum order complexity. More precisely, we prove explicit formulas for their $N$th maximum order complexity which are both of largest possible order of magnitude $N$. We present the result on the Rudin-Shapiro sequence in a more general form as a formula for the maximum order complexity of certain pattern sequences.

Automatic sequences such as the Thue-Morse sequence and the Rudin-Shapiro sequence are highly predictable and thus not suitable in cryptography. In particular, they have small expansion complexity. However, they still have a large maximum order complexity. Certain subsequences of automatic sequences are not automatic anymore and may be attractive candidates for applications in cryptography. In this paper we show that subsequences along the squares of certain pattern sequences including the Thue-Morse sequence and the Rudin-Shapiro sequence have also large maximum order complexity but do not suffer a small expansion complexity anymore.

A Kakeya set contains a line in each direction. Dvir proved a lower bound on the size of any Kakeya set in a finite field using the polynomial method. We prove analogues of Dvir's result for non-degenerate conics, that is, parabolae and hyperbolae (but not ellipses which do not have a direction). We also study so-called conical Nikodym sets where a small variation of the proof provides a lower bound on their sizes. (Here ellipses are included.) Note that the bound on conical Kakeya sets has been known before, however, without an explicitly given constant which is included in our result and close to being best possible.

Balog and Wooley have recently proved that any subset $A$ of either real numbers or of a prime finite field can be decomposed into two parts $U$ and $V$, one of small additive energy and the other of small multiplicative energy. In the case of arbitrary finite fields, we obtain an analogue that under some natural restrictions for a rational function $f$ both the additive energies of $U$ and $f(V)$ are small. Our method is based on bounds of character sums which leads to the restriction $\# A > q^{1/2}$ where $q$ is the field size. The bound is optimal, up to logarithmic factors, when $\# A \geq q^{9/13}$. Using $f(X)=X^{-1}$ we apply this result to estimate some triple additive and multiplicative character sums involving three sets with convolutions $ab+ac+bc$ with variables $a,b,c$ running through three arbitrary subsets of a finite field.

Carlitz rank and index are two important measures for the complexity of a permutation polynomial $f(x)$ over the finite field $\F_q$. In particular, for cryptographic applications we need both, a high Carlitz rank and a high index. In this article we study the relationship between Carlitz rank $Crk(f)$ and index $Ind(f)$. More precisely, if the permutation polynomial is neither close to a polynomial of the form $ax$ nor a rational function of the form $ax^{-1}$, then we show that $Crk(f)>q- \max\{3 Ind(f),(3q)^{1/2}\}$. Moreover we show that the permutation polynomial which represents the discrete logarithm guarantees both a large index and a large Carlitz rank.

For a given subset $A\subseteq \mathbb F_q^*$, we study the problem of finding a large packing set $B$ of $A$, that is, a set $B \subseteq \mathbb F_q^*$ such that $|AB|=|A||B|$. We prove the existence of such a $B$ of size $|B|\ge (q-1)/|A/A|$ and show that this bound is in general optimal. The case that $q=p$ is a prime and $A=\{1,2,\ldots,λ\}$ for some positive integer $λ$ is particularly interesting in view of the construction of limited-magnitude error correcting codes. Here we construct a packing set $B$ of size $|B|\gg p (λ\log p)^{-1}$ for any $λ\le c p^{1/2}$ for some explicitly calcuable constant $c$. This result is optimal up to the logarithmic factor.

The well-known Chowla and Zassenhaus conjecture, proven by Cohen in 1990, states that for any $d\ge 2$ and any prime $p>(d^2-3d+4)^2$ there is no complete mapping polynomial in $\mathbb{F}_{p}[x]$ of degree $d$. For arbitrary finite fields $\mathbb{F}_{q}$, we give a similar result in terms of the Carlitz rank of a permutation polynomial rather than its degree. We prove that if $n<\lfloor q/2\rfloor$, then there is no complete mapping in $\mathbb{F}_{q}[x]$ of Carlitz rank $n$ of small linearity. We also determine how far permutation polynomials $f$ of Carlitz rank $n<\lfloor q/2\rfloor$ are from being complete, by studying value sets of $f+x.$ We provide examples of complete mappings if $n=\lfloor q/2\rfloor$, which shows that the above bound cannot be improved in general.

In this paper we study the non-existence problem of (nearly) perfect (almost) $m$-ary sequences via their connection to (near) Butson-Hadamard (BH) matrices and (near) conference matrices. Firstly, we apply a result on vanishing sums of roots of unity and a result of Brock on the unsolvability of certain equations over a cyclotomic number field to derive non-existence results for near BH matrices and near conference matrices. Secondly, we refine the idea of Brock in the case of cyclotomic number fields whose ring of integers is not a principal ideal domains and get many new non-existence results.