Glossary
========

Key terms used throughout Pyralysis:

Imaging (image formation, image synthesis)
   Forming an image from interferometric visibilities. In radio astronomy this is the
   usual name for the task; CASA and similar packages call it **imaging** or **image
   synthesis**. Pyralysis docs and APIs also say **reconstruction** and **image
   optimization** for the same broad family of methods (dirty mapping, CLEAN-style
   workflows, and regularized / RML optimization). Prefer *imaging* in user-facing
   prose when talking to observatory users; code packages may still use
   ``reconstruction``.
Reconstruction
   Synonym for **imaging** / **image synthesis** in Pyralysis. Often used when the
   image is obtained by minimizing an objective (data fidelity + regularizers) rather
   than by a CLEAN-style minor cycle alone. See also :doc:`optimization`.
Image optimization
   Imaging via gradient-based or proximal algorithms (RML / compressed sensing). Same
   scientific goal as imaging; emphasizes the optimizer stack (:doc:`optimization`,
   :doc:`optimizers`).
Dataset
   The main in-memory container for interferometric data in Pyralysis. It holds metadata (antenna, baseline, field, spectral windows, polarization, observation) and a list of SubMS (``ms_list``) that contain the actual visibility data. Returned by the simulator (e.g. ``sim.simulate(create_dataset=True)``) or by I/O readers such as DaskMS.
SubMS (sub–measurement set)
   A partition of the measurement set by field and spectral window. Each SubMS has an id, field_id, spw_id, polarization_id, and a VisibilitySet with the visibility arrays (DATA, UVW, weights, etc.) for that partition. The Dataset's ``ms_list`` is a list of SubMS; one SubMS for single field/spw, multiple for multi-field or multi–spectral-window data. Used by dask-ms and Pyralysis for efficient I/O and per-partition processing.
RML (Regularized Maximum Likelihood)
   An imaging method that uses regularization to improve solutions.
Measurement operator
   The forward model that maps a model image to model visibilities at irregular UV coordinates. It applies (optionally) primary beam and intensity scaling, 2D FFT, phase shift, then samples the Fourier grid at observed UV positions using nearest neighbor, bilinear interpolation, or degridding.
Gridding
   Interpolating Fourier space data onto a uniform grid.
Degridding
   Visibility estimation from a Fourier grid using a convolution kernel (inverse of gridding). One of the three visibility-estimation methods in the measurement operator.

----

:doc:`faq` | :doc:`usage`