niter Fit image-plane model components to visibilities in the UV plane. EXAMPLE ------- 0>modelfit 4 Partitioning the model into established and variable parts. The fixed established model contains 0 components (0 Jy). The variable part of the model contains 2 components (0.97 Jy). There are 10 variables and 4510 usable visibilities. This gives 2 x 4510 - 10 = 9010 degrees of freedom. Reduced Chi-squared = Chi-squared / 9010. Iteration 00: Reduced Chi-squared=46.315838 ! Flux (Jy) Radius (mas) Theta (deg) Major (mas) Axial ratio Phi (deg) T 0.950000v 0.00000 0.00000 1.00000v 1.00000v 0.00000v 1 0.0200000v 1.00000v -130.000v 1.00000v 1.00000v 0.00000v 1 Iteration 01: Reduced Chi-squared=4.3881679 ! Flux (Jy) Radius (mas) Theta (deg) Major (mas) Axial ratio Phi (deg) T 0.416386v 0.00000 0.00000 0.524179v 0.437841v -6.49258v 1 0.0826878v 1.00780v -137.060v 3.31603v 0.732082v -56.6298v 1 Iteration 02: Reduced Chi-squared=0.89815858 ! Flux (Jy) Radius (mas) Theta (deg) Major (mas) Axial ratio Phi (deg) T 0.493566v 0.00000 0.00000 0.371532v 0.167305v -7.53078v 1 0.0209711v 1.20514v -165.679v 3.93011v 0.331629v -49.3905v 1 Iteration 03: Reduced Chi-squared=1.0050499 (Increased) Iteration 04: Reduced Chi-squared=1.0088361 (Increased) 0> ARGUMENTS --------- niter - The number of iterations to try, or zero to just see the current fit. 10 and 20 are good numbers to try, before the first selfcal. CONTEXT ------- Model fitting directly to the visibilities of interferometric data has two common uses. 1. It is employed to find good starting models for the initial phase self-calibration of phase-unstable data, especially when a point source starting model has proved insufficient. 2. It is commonly used to parameterise the general characteristics of sources with as few variables as possible. The difmap modelfit program fits aggregates of various forms of model components, fitting directly to the real and imaginary parts of the observed visibilities using the powerful Levenberg-Marquardt non-linear least squares minimization technique. This makes it substantially faster and a lesser consumer of memory than the Caltech VLBI package modelfit program, that fitted to an ad hoc combination of closure-phase and amplitude. The disadvantage of the newer algorithm is that it assumes that the phases are well calibrated. The older algorithm was insensitive to this through its use of closure phases. As a result, using modelfit in difmap may require a number of interleaved self-calibration steps to converge on the model that agrees well with the closure phases. SPECIFYING MODELS FOR MODELFIT ------------------------------ There are three ways to specify model components to modelfit. 1. In model files, components with variable parameters are denoted by a 'v' character post-fixing each of the variable parameters. Thus in the following model the first component is an elliptical gaussian component with all parameters variable except its position, while the second component has all parameters variable. ! Flux (Jy) Radius (mas) Theta (deg) Major (mas) Axial ratio Phi (deg) T 0.140490v 0.00000 0.00000 0.704297v 0.147144v -48.4654v 1 0.118984v 1.01917v 137.776v 1.10527v 0.544262v -57.5779v 1 The meaning of each of the above parameters (as originally defined for the Caltech VLBI package), is as follows: Flux - The integrated flux in the component (Jy). Radius - The radial distance of the component center from the center of the map (milli-arcsec). Theta - The position angle of the center of the component (degrees North -> East) wrt an imaginary line drawn vertically through the map center. The following components may be omitted for delta components. Major - The FWHM major axis of the elliptically stretched component (milli-arcsec). Ratio - The ratio of the minor axis to the major axis (0 -> 1). Phi - The Position angle of the major axis (degrees North -> East). T - The type of component. Recognised types are: 0 - Delta function. 1 - Gaussian. 2 - Uniformly bright disk. 3 - Optically thin sphere. 4 - Ring. 5 - Un-supported component type - kept for compatibility. 6 - Sunyeav-Zel'dovich. Internally the positions of components are recorded by their X and Y axis positions wrt the map center and fitting is wrt these parameters. For this reason if either Radius or Theta is made variable, both will be taken as variable. It is not possible to fix one or the other individually. Similarly, except for the special case of a circular component where only Major is marked as variable and the axial ratio is exactly 1.0, it is not possible to individually mark Major,Ratio and Phi to be fixed. Marking one as variable sets them all as variable. More obviously, the model component type T can not be marked to be variable. 2. While model components can be specified via a model file in the above format, it is usually more convenient to incrementally add and remove components interactively via the difmap 'mapplot' command. See help mapplot for further details. 3. The final method is really directed towards changing the trial parameters of an existing variable model, although new components may also be created. This method uses the difmap 'edmodel' command, which invokes an editor on an internally generated model scratch file, initialized with the existing variable model components, appropriately shifted if the 'shift' command has previously been used. See help edmodel for further details. RELATED COMMANDS ---------------- edmodel - Edit the current models with an external editor. rmodel - Read a new model for modelfit. mapplot - Maplot allows interactive addition and removal of components. addcmp - Add a model component by hand to the tentative model.