Vectorial bent functions and partial difference sets (Q2232123)

For a prime \(p\), let \(\mathrm{GF}(q)\) denote the finite field with \(q=p^k\) elements. Let \(V_n\) denote an \(n\)-dimensional vector space over the finite field \(\mathrm{GF}(p)\) provided with a \(\mathrm{GF}(p)\)-valued non-degenerate innter product \(\langle\ ,\ \rangle =\langle\ ,\ \rangle_n\). For example, \(V_n=\mathrm{GF}(p)^n\) with the standard (``Euclidean'') inner product or \(V_n=\mathrm{GF}(q)\) with the trace inner product \(\langle x, y \rangle =\mathrm{Tr}_{\mathrm{GF}(q)/\mathrm{GF}(p)}(xy)\). In general, if an integer \(r\) divides \(s\) then we let \(\mathrm{Tr}^s_r=\mathrm{Tr}_{\mathrm{GF}(p^s)/\mathrm{GF}(p^r)}: \mathrm{GF}(p^s)\to \mathrm{GF}(p^r)\) denote the \(\mathrm{GF}(p^r)\)-linear trace map. The \textit{(two variable) Walsh transform} of a function \(f: V_n\to V_s\) is defined by \[ W_f(u,v)=\sum_{x\in V_n} \varepsilon_p^{\langle v, f(x)\rangle_s -\langle u, x \rangle_n}, \] where \(\varepsilon_p = e^{2\pi i/p}\), \(u\in V_n\), \(v\in V_s^*=V_s-\{0\}\). When \(s=1\), this transform is denoted \[ W_f(u) = W_f(u,1)=\sum_{x\in V_n} \varepsilon_p^{f(x) -\langle u, x \rangle_n}, \] The \textit{Walsh spectrum} is the multiset \(W_f=\{W_f(u,v)\ |\ u\in V_n, v\in V_s^*\}\) of cardinality \(p^n(p^s-1)\). We say \(f\) is \textit{vectorially bent} or \textit{\((n,s)\)-bent} if \(W_f\) is contained in the circle of radius \(p^{n/2}\) in the complex numbers \(\mathbb{C}\). Let \(n,s,\ell\) be integers greater than \(1\) such that (a) \(\gcd(p^s-1,\ell-1)=1\), and (b) \(V_s = \mathrm{GF}(p^s)\) is a subspace of \(V_n\). In this situation, when a function \(f: V_n\to V_s=\mathrm{GF}(p^s)\) satisfies \(f(\alpha x) = \alpha^\ell f(x)\), then we say \(f\) is an \textit{\(\ell\)-form}. The authors remark that a bent function being an \(\ell\)-form ``seems to be crucial for a connection with PDSs,'' where PDS is the commonly used abbreviation for partial different set. See the author's open question at the end of this review for a more precise statement. In this bent case, when \(s=1\) we simply call \(f\) a \textit{\(p\)-ary bent function} (or, when \(p=2\), a \textit{Boolean bent function}). It is known that for each such function \(f\) there is a \textit{dual function} \(f^* : V_n \to \mathrm{GF}(p)\) that satisfies \(W_f(u) = \zeta \varepsilon_p^{f^*(u)}p^{n/2}\), where \(\zeta \in \{\pm 1, \pm i\}\). If \(f:V_n\to \mathrm{GF}(p)\) is such a bent function, it's known that its dual \(f^*:V_n\to \mathrm{GF}(p)\) is also a bent function. If \(f:V_n\to V_s\) is vectorially bent then each of its \textit{component functions} \(f_v(x) = \langle v, f(x)\rangle_s\), \(v\in V_s^*\), is a \(p\)-ary bent function. A vectorially bent function \(f:V_n\to V_s\) is \textit{vectorially dual-bent} if the set of dual functions of the component functions of \(f\) (together with the \(0\) function) forms a vector space of bent functions of dimension \(s\). A \(p\)-ary bent function \(f:V_n\to GF(p)\) with \(W_f(u) = \zeta \varepsilon_p^{f^*(u)}p^{n/2}\), is called \begin{itemize} \item[(a)] \textit{weakly regular} if \(\zeta \in \{\pm 1, \pm i\}\) does not depend on \(u\), \item[(b)] \textit{non-weakly regular} if \(\zeta \in \{\pm 1, \pm i\}\) does depend on \(u\), \item[(c)] \textit{regular} if \(\zeta =1\). \end{itemize} Let \(S\) denote the set of non-zero squares in \(\mathrm{GF}(p^s)\) and let \(N\) denote the set of non-zero non-squares in \(\mathrm{GF}(p^s)\). For any function \(f: V_n\to \mathrm{GF}(p^s)\), define \begin{align*} D_0^f&=\{x\in V_n^*\ |\ f(x)=0\},\\ D_S^f&=\{x\in V_n\ |\ f(x)\in S\},\\ D_N^f&=\{x\in V_n\ |\ f(x)\in N\}. \end{align*} A classic result of Dillon, from his PhD thesis, gives a connection between ordinary bent functions (with \(s=1\)) and Hadamard difference sets. The authors of the paper under review prove the following vectorially bent analogue/generalization of Dillon's result. Proposition. For a prime \(p\) and an even integer \(n\), let \(f:V_n\to \mathrm{GF}(p^2)\) denote a vectorially dual-bent function with \(f(0)=0\) and \(f(-x)=f(x)\). If \(p\) is odd, suppose all component functions are either regular with \(\varepsilon=1\) or weakly regular with \(\varepsilon=-1\). Then \(D_0^f\) is a \((\nu, k, \lambda, \mu)\) partial difference set in \(V_n\) with \(\nu=p^n\), \begin{align*} k &= p^{n-s}+\varepsilon (p^{n/2}-p^{n/2-s})-1,\\ \mu &= p^{n-2s}+\varepsilon p^{n/2-s}),\\ \lambda &= p^{n-2s}+\varepsilon (p^{n/2}-p^{n/2-s})-2\, . \end{align*} Let \(n\) be even, let \(s\) be a divisor of \(n\), and let \(\alpha = \frac{p^n-1}{p^s-1}\). For a primitive element \(\omega\) of \(\mathrm{GF}(p^n)\), denote the cyclotomic classes by \[ C_i = \{\omega^{\alpha t + i}\ |\ 0\leq t \leq p^s-2\}, \ \ \ \ 0\leq i\leq \alpha - 1. \] For \(i,j\) with \(0\leq i, j\leq \alpha - 1\), define the cyclotomic numbers \((i,j)_\alpha\) by \[ (i,j)_\alpha = |\{(x,y)\in C_i\times C_j\ |\ x+1=y\}|. \] Using the above proposition, the authors prove the following result. Theorem. Let \(n,s,\ell\) be integers greater than \(1\) such that (a) \(\gcd(p^s-1,\ell-1)=1\), and (b) \(V_s = \mathrm{GF}(p^s)\) is a subspace of \(V_n\). Let \(f: \mathrm{GF}(p^n) \to \mathrm{GF}(p^s)\) be a vectorially bent \(\ell\)-form for which each of its component functions are either all regular or all weakly regular but not regular. Let \(I\subset \{0,\dots,\alpha-1\}\) be such that \(D_0^f = \bigcup_{i\in I} C_i\). Then \[ |I|=\frac{(p^{n/2}-\varepsilon)(p^{n/2}+\varepsilon)}{p^s-1}, \] and for all \(m\) with \(0\leq m\leq \alpha-1\), we have \[ \sum_{i,j\in I} (m-j, i-j)_\alpha =\begin{cases} p^{n-2s}+\varepsilon (p^{n/2}-p^{n/2-s})-2, & \text{if } m\in I,\\ p^{n-2s}+\varepsilon p^{n/2-s}, & \text{if } m\notin I,\\ \end{cases} \] where \(\varepsilon = 1\) if all component functions of \(f\) are regular, and \(\varepsilon = -1\) if all component functions of \(f\) are weakly regular but not regular. In the context of Maiorana-MacFarland bent functions, the authors prove the following result. Theorem. Let \(n,s,m\) be integers greater than \(1\) such that \(n=2m\) and \(s\) divides \(m\). Let \(L: \mathrm{GF}(p^m)\to \mathrm{GF}(p^m)\) denote a linear permutation and let \(f: V_n=\mathrm{GF}(p^m)\times \mathrm{GF}(p^m)\to \mathrm{GF}(p^s)\) denote the (Maiorana-MacFarland-like, vectorially dual-bent) function \(f(x,y) = Tr^m_s(xL(y))\), for \(x,y\in \mathrm{GF}(p^m)\). Then \(D_N^f\) and \(D_S^f\) are \((\nu, k, \lambda, \mu)\) partial difference sets in \(V_n\) with \(\nu=p^n\), \begin{align*} k &= \frac{p^{s}-1}{2}p^{m-s}(p^m-1),\\ \mu &= \frac{p^{s}-1}{2}p^{m-s}(\frac{p^{s}-1}{2} p^{m-s}-1),\\ \lambda &= p^{m}+\frac{p^{s}-1}{2}p^{m-s}(\frac{p^{s}-1}{2} p^{m-s}-3)\, . \end{align*} The authors close with a number of very interesting open questions. For brevity, only one of them is repeated here. \textbf{Prove or disprove}: Let \(f: V_n\to \mathrm{GF}(p^s)\) be a vectorially dual-bent function. Then \(f\) is linearly equivalent to an \(\ell\)-form for some \(\ell\) satisfying \(\gcd(\ell -1, p^s-1)=1\). Please see this interesting and well-written paper for more details.