The inner kernel theorem for a certain Segal algebra (Q2155978)

The Segal algebra \(\mathbf{S}_0(G)\) consists of all functions \(f\in \mathbf{L}^1(G)\) for which \[ \int_{\widehat{G}}\|E_\omega f * f \|_{\mathbf{L}^1(G)}\,d\omega < \infty, \] \(G\) is a locally compact Abelian group, \(\widehat{G}\) its dual group, \(E_\omega f(t) = \omega(t)f(t)\), \(t\in G\), \(\omega \in \widehat{G}\), and \(d\omega\) is the Haar measure on \(\widehat{G}\). Any non-zero \(g\in\mathbf{S}_0(G)\) defines a Banach space norm on \(\mathbf{S}_0(G)\) via \[ \|f\|_{\mathbf{S}_0(G),g} = \int_{\widehat{G}}\|E_\omega f * g \|_{\mathbf{L}^1(G)}\,d\omega. \] The space \(\mathbf{S}_0(G)\) is a Fourier invariant Banach algebra under convolution and pointwise multiplication. The dual Banach space \(\mathbf{S}'_0(G)\) consists of generalized functions. Spaces \(\mathbf{S}_0(G)\), \(\mathbf{S}'_0(G)\) are non-smooth analogues of Schwartz spaces of rapidly decreasing smooth functions and tempered distributions. Schwartz kernel theorem for pairs \(\mathbf{S}_0(G)\), \(\mathbf{S}'_0(G)\) is well known. It asserts that the spaces of continuous linear operators from \(\mathbf{S}_0(G_1)\) into \(\mathbf{S}'_0(G_2)\) and \(\mathbf{S}'_0(G_1\times G_2)\) are naturally isomorphic. In the current paper the authors show that under the natural isomorphism kernels in \(\mathbf{S}_0(G_1\times G_2)\) correspond to linear operators \(T:\mathbf{S}_0'(G_1)\rightarrow \mathbf{S}_0(G_2)\) mapping bounded weak* convergent nets in \(\mathbf{S}_0'(G_1)\) into norm convergent nets in \(\mathbf{S}_0(G_2)\), and satisfying \[ (T\sigma^{(1)}, \sigma^{(2)})_{\mathbf{S}_0,\mathbf{S}_0'} =\int_{G_1\times \widehat{G_1}\times G_2\times \widehat{G_2}} \mathcal{V}_{g^{(1)}}\sigma^{(1)}\left( \nu^{(1)}\right)\cdot \mathcal{V}_{g^{(2)}}\sigma^{(2)}\left( \nu^{(2)}\right) \cdot \left( \pi \left( \nu^{(2)}\right)g^{(2)}, T \pi \left( \nu^{(1)}\right)g^{(1)}\right)_{\mathbf{S}_0,\mathbf{S}'_0} \,d\left( \nu^{(1)}, \nu^{(2)} \right), \] where for \(i=1,2\) function \(g^{(i)}\in\mathbf{S}_0(G_i)\), \(\|g^{(i)}\|_{\mathbf{L}^2(G_i)}=1\), for \(\nu = (x,\omega)\in G\times \widehat{G}\) the time-frequency shift \(\pi(\nu)=E_\omega T_x\), \(T_xf(t)=f(t-x)\), and the short-time Fourier transform \(\mathcal{V}_g: \mathbf{S}_0'(G)\rightarrow\mathbf{C}_b(G\times \widehat{G})\) with respect to a function \(g \in\mathbf{S}_0(G)\) is defined as \(\mathcal{V}_g \sigma (\nu) = (\overline{\pi(\nu)g}, \sigma)_{\mathbf{S}_0, \mathbf{S}'_0}\). The authors also provide a similar treatment for bilinear forms defined on \(\mathbf{S}_0(G_1)\times\mathbf{S}_0(G_2)\). They derive a multitude of consequences: matrix multiplication type composition rules, nuclearity and trace-class properties, constructions of regularizing approximations of the identity.