Families of complex Hadamard matrices (Q1940093): Difference between revisions

A complex Hadamard matrix is a unitary \(N\times N\) matrix such that their entries have the same modulus. The Fourier matrix is a canonical example of a Hadamard matrix. The classification of these matrices has been completed for \(N\leq 5\) by \textit{U. Haagerup} [Cambridge, MA: International Press. 296--322 (1997; Zbl 0914.46045)] as well as some partial information in higher dimensions is available in the monograph by \textit{K. J. Horadam} [Hadamard matrices and their applications. Princeton, NJ: Princeton University Press. (2007; Zbl 1145.05014)]. If the rows and columns of a complex Hadamard matrix are multiplied by \(2N-1\) over all phase factors, the Hadamard property is preserved. Hence \(2N-1\) parameters are trivial and it is customary to remove them by presenting complex Hadamard matrices in dephased form, such that the first row and the first column have positive real entries. The number \(d_{1} = \sum _{n=1}^{N-1} (\mathrm{gcd} (n,N) -1)\) is called the defect of the Fourier matrix and gives an upper bound on the dimension of any smooth family of dephased complex Hadamard matrices passing through the Fourier matrix. If \(N= p^{k}\) is a prime power, explicit solutions for dephased Hadamard matrices with \(d_{1}\) free parameters are known in the literature as affine families (see [\textit{W. Tadej} and \textit{K. Życzkowski}, Open Syst. Inf. Dyn. 13, No. 2, 133--177 (2006; Zbl 1105.15020)]). When \(N=6\), for which \(d_{1}= 4\), a \(3\)-dimensional family has been constructed in an explicit way and the existence of a \(4\)-dimensional family has been shown. It has been missing that the \(4\)-dimensional family includes the Fourier matrix and that it is smooth there. Numerical arguments for the truth of this statement appear in the work of \textit{A. J. Skinner, V. A. Newell} and \textit{R. Sanchez} [J. Math. Phys. 50, No. 1, 012107, 7 p. (2009; Zbl 1189.81013)]. Studying all dimensions up to 100, the first order result is misleading unless the dimension is 6 or a power of a prime number. In the paper under review, the confirmation of the above conclusion is proved using a perturbative expansion in which a linear system is solved at each order. At each order, a set of nonlinear consistency conditions on the lower order solution must be satisfied. First, the complexity equations are solved and then they impose unitarity, order by order, on the solution. When the consistency conditions break down, the dimension of the solution space is less than the suggested by the linear defect. In order to see by how much the dimension drops, it is necessary to solve these conditions which are multivariate polynomial equations. The authors find out when the dimension of any smooth family is indeed less than \(d_{1}\). The cases \(N=12, 18,\) and \(20\) are discussed. A conjecture about the solutions for all \(N= p_{1}^{2} p_{2}\), where \(p_{1}\) and \(p_{2}\) are prime numbers, is stated. The conjectured solution consists of two families of dimension \(d= \frac{d_{1} + d_{A}}{2} =3N -3 p_{1}p_{2} - 2 p_{2}^{2} + p_{2} +1\), where \(d_{1}\) is the linear defect and \(d_{A}\) is the maximal dimension of known affine families.