Analogues of Besicovitch-Wiener theorem for the Heisenberg group (Q1386216)

Some Plancherel-type results are obtained for the (operator-valued) Fourier transforms of certain measures on the Heisenberg group \(H^n\cong\mathbb{C}^n\times\mathbb{R}\), which are analogous to results for \(\mathbb{R}^n\). The Besicovitch-Wiener theorem is a generalization of a classical Parseval-type property for the Fourier transform of an infinite combination of Dirac measures on \(\mathbb{R}^n\). We shall describe the first result in some detail. It deals with measures \(\mu\) on \(H^n\) of the form \(\mu=\sum c_j\delta(z_j,t_j)\), with \(C_2:=\sum| c_j|^2<\infty\). Let \(e_k^\lambda\) (\(0\not=\lambda\in\mathbb{R}\), \(0\leq k\in\mathbb{Z}\)) be the elementary spherical functions on \(H^n\). For shortness, let us denote \(\int_{\mathbb{C}^n}|\mu\star e_k^\lambda(z,0)|^2 \text{d}z\) by \(m_k^\lambda\) and \(C_2(2\pi/|\lambda|)^n\) by \(b_\lambda\). Under the extra assumptions that \(C_1 := \sum| c_j| < \infty\) and that the \(z_j\) are different, it is claimed that \(\lim_{N\to\infty} N^{-n} \sum_{k=0}^N m_k^\lambda = b_\lambda/n!\). The second result is of a similar nature; this time \(\mu\) is the product of the normalized surface measure of a sphere in \(\mathbb{C}^n \times \{0\}\) and a function \(g \in L^2(\mathbb{R})\), respectively an infinite linear combination \(\sum c_j \delta(a_j)\) of Dirac measures on \(\mathbb{R}\) satisfying \(\sum | c_j|^2 < \infty\). Finally, a short description is given of some results for Hermite and special Hermite expansions on \(\mathbb{R}^n\). In the proofs repeated use is made of a certain Tauberian theorem. Reviewer's remark: I have not been able to verify the correctness of the last sentence of the proof of the first result above (`By appealing to the Tauberian theorem we get the result.'), and hence am not sure of the correctness of that result. Indeed, setting \(t=1-\varepsilon\), the last formula in the proof becomes \(\lim_{\varepsilon \downarrow 0} \varepsilon^n \sum_{k=0}^\infty (1-\varepsilon)^k m_k^\lambda = b_\lambda\). Now from the Tauberian theorem (with \(l=0\)) it follows that \(\lim_{\varepsilon \downarrow 0} \varepsilon^n \sum_{k=0}^\infty \text{ e}^{-\varepsilon \lambda_k} m_k^\lambda = b_\lambda\) implies \(\lim_{N \rightarrow \infty} N^{-1} \sum_{k=0}^N m_k^\lambda = b_\lambda / n!\) (thus \(N^{-1}\), not \(N^{-n}\)), on condition that \((1-\varepsilon)^k = \text{ e}^{-\varepsilon \lambda_k}\) with \(\lambda_k = k^{1/n}+o(k^{1/n})\). However, we have \(\lambda_k = k \log(1-\varepsilon)^{-1/\varepsilon} \approx k (\varepsilon \downarrow 0)\), thus not \(\approx k^{1/n}\), so that the Tauberian theorem cannot be applied in this way. A more informative indication on how the authors intend to use the Tauberian theorem at this point would have been appropriate.