Adaptive decomposition by weighted inner functions: a generalization of Fourier series (Q532506)

A generalized Fourier series decomposition of boundary values of analytic functions in the unit disk is presented in terms of a series of inner functions of increasing frequencies. A function in the analytic Hardy space \(\mathcal{H}^2(\mathbb{D})\) can be expressed in terms of the Nevanlinna factorization \(F=OBS\) in which the inner part \(BS\) consists of a Blaschke product \(B\) formed from the zeros of \(F\) in \(\mathbb{D}\) and \(S\) is a singular inner function defined by \(d\mu\), where \(\log (F(e^{it}))\, dt - d\mu\) gives rise to the least harmonic majorant of \(\log |F(z)|\). The authors propose a further recursive decomposition obtained by successively subtracting linear terms from an outer function such that at least two factors \(z\) and \(z-a_j\) can be factored out to obtain \[ \begin{multlined} F(z)=O_1(z)I_1(z)=\cdots=O_{n+1}(z)I_{n+1}(z) z^n\prod_{j=1}^n \Bigl[ \Bigl(\frac{z-a_j}{1-\overline{a_j}z}\Bigr)^{d_j} I_j(z)\Bigr]\, \\ + \Bigl[A_n \frac{B_n z}{1-\overline{a_n}z}\Bigr] I_n(z) z^{n-1} \prod_{j=1}^{n-1} \Bigl[ \Bigl(\frac{z-a_j}{1-\overline{a_j}z}\Bigr)^{d_j} I_j(z)\Bigr]\, \\ +\cdots + \Bigl[A_1 \frac{B_1 z}{1-\overline{a_1}z}\Bigr] I_1(z) R_n(z)+(A_n N_n +B_n M_n)+\cdots + (A_1N_1+B_1M_1). \end{multlined} \] Now, denote \(F=s^{+}\in \mathcal{H}^2(\mathbb{D})\) and denote \[ \rho_k^{(1)}(t) \cos\theta_k^{(1)}(t)={\text{Re}} \, [A_k N_k (e^{it})],\,\, \rho_k^{(2)}(t) \cos\theta_k^{(2)} (t) = {\text{Re}}\,[B_k M_k (e^{it})] \] and \[ s_n (e^{it}) =\sum_{k=1}^n \rho_k^{(1)} (t) \cos\theta_k^{(1)}(t) +\rho_k^{(2)} (t) \cos\theta_k^{(2)}(t). \] Then it is shown that \(s_n\to s\) in \(L^2\) norm on the unit circle. Here, \(s=2 \operatorname{Re} s^+- \int F(e^{it})\). The components of the series \(s_n\) can be thought of as components of a generalized Fourier series in the sense of being mono-components. Such a term can be defined as (1) having a phase \(\theta\) such that the instantaneous frequency \(\theta'(t)\) is strictly increasing and (2) the instantaneous amplitude \(\rho\geq 0\) is such that \(H(\rho\cos\theta)=\rho H(\cos\theta)\), where \(H\) denotes the circular Hilbert transform defined by the principal value average of \(F(e^{it})\) integrated against \(\cot ((t-\cdot)/2)\). This decomposition is in analogy with the analytic phase-amplitude representation \(H(\rho\cos\theta)=\rho\sin\theta\).