Infinite-dimensional operator ridgelet transform

2. The Gaussian-weighted ridgelet transform🔗

This chapter is Section 3 of the manuscript together with Appendix A (the Gaussian mixture), Appendix G (the finite-dimensional case and the dilation obstruction), Appendix H (abstract weights), and Appendix I (explicit filters). The input measure is the centred Gaussian \mu_Q=\mathcal N(0,Q), the direction measure is the homogeneous Gaussian mixture \nu_\alpha, and the parameter measure is \lambda_\alpha=\nu_\alpha\otimes\mathrm dc.

The formalization states the core theory for an abstract pair (\mu,\nu), as in Appendix H: \mu a Borel probability measure on H and \nu a \sigma-finite Borel measure with full support that is homogeneous of degree \alpha. The Gaussian pair (\mu_Q,\nu_\alpha) is the instance. The Hilbert space \mathcal E_\alpha is represented by the closed subspace \mathcal K_\alpha=\overline{\mathcal G_Q(\mathcal D_\alpha)}\subseteq L^2(\nu_\alpha), to which it is unitarily equivalent by Lemma 2.3.2.

  1. 2.1. The input and direction measures
  2. 2.2. Admissible filters, the transform, and the Fourier slice
  3. 2.3. The Hilbert space
  4. 2.4. Plancherel identity, closed range, and injectivity
  5. 2.5. The finite-dimensional case and the dilation obstruction
  6. 2.6. Explicit admissible filters