A constructive inversion framework for twisted convolution (Q2385074)

Let \(\pi(x,\omega)=T_xM_\omega\) denote the time-frequency shift for \(x, \omega\in\mathbb R^d\) and define the operator algebra \({\mathcal A}=\{\sum_{k,l\in{\mathbb Z}^d}a_{k,l}\pi(\beta^{-1}k,\alpha^{-1}l)| a=(a_{k,l})\in\ell^1({\mathbb Z}^{2d})\}\) for fixed positive numbers \(\alpha, \beta\). According to the Janssen representation of a frame operator associated to the Gabor system, \textit{A. J. E. Janssen} [J. Fourier Anal. Appl. 1, No.~4, 403--436 (1995; Zbl 0887.42028)] observed that the map \(\kappa:a\to\kappa(a)=\sum_{k,l\in{\mathbb Z}^d}a_{k,l}\pi(\beta^{-1}k,\alpha^{-1}l)\) satisfies \(\kappa(a)\kappa(b)=\kappa(a\natural b)\) and gives an algebra homomorphism from \((\ell^1({\mathbb Z}^{2d}),\natural)\) to \(\mathcal A\). In this paper the authors split the twisted convolution into a finite number of weighted convolution and obtain a constructive method for studying the invertibility of the twisted convolution. For relative prime integers \(p, q\) and \(a,b\in{\mathbb Z}^{2d}\), the twisted convolution \(a\natural b\) with rational parameters is defined by \((a\natural b)_{m,n}=\sum_{k,l\in{\mathbb Z}^d}a_{k,l}b_{m-k,n-l}e^{\omega(m-k)\cdot l}\), where \(\omega=e^{2\pi iq/p}\). Let \(({\mathcal M},\circledast)\) denote an algebra of \(p^d\times p^d\) matrices whose entries belong to \(\ell^1({\mathbb Z}^{2d})\) and multiplication law is given by \((A\circledast B)_{i,j}=\sum_{l\in{\mathcal I}}A_{i,l}*B_{l,j}\), where \(i,j\in{\mathcal I}=\{1,2,\dots,p^d\}\). Then the map \(\phi:a\to\phi(a)\), where \((\phi(a))_{i,j}=\sum_{m\in{\mathbb Z}_p^d}\omega^{m\cdot k_j}a^{m,k_i-k_j}\), is an anti-homomorphism from \((\ell^1({\mathbb Z}^{2d}),\natural)\) to \(({\mathcal M},\circledast)\). Moreover, if \(A\in {\mathcal M}_0=\text{ Im}(\phi)\) is invertible in \(\mathcal M\), then \(A^{-1}\in{\mathcal M}_0\). Hence the invertibility condition is transferred to the Cramer's rule in the matrix algebra.