memo15.tar

memo15.tex0000644000175000017500000001001213706544126011031 0ustar  njjnjj\documentclass[12pt]{article}
\textwidth 160mm
\textheight 245mm
\oddsidemargin -3mm
\evensidemargin -3mm
\parskip 3mm
\parindent 0mm
\topmargin -20mm
\pagestyle{empty}
\usepackage{graphicx}
\graphicspath{{.}{./figures/}}
\usepackage{color}
\usepackage[utf8]{inputenc}
\usepackage{hyperref}
\usepackage{subcaption}
\usepackage{amssymb}
\usepackage{gensymb}
\newcommand{\memohead}[3]{\clearpage\begin{centering}\mbox{}\hrulefill \vskip 50mm \mbox{}{\Huge \bf LBWG memo #1}\\ \mbox{}\vskip 40mm {\Huge \bf #2}\mbox{}\vskip 40mm{\Large #3}\\\end{centering}\vfill\hrulefill\clearpage}
\usepackage{listings}
\usepackage{rotating}
\begin{document}

\memohead{15}{FACTOR on long-baseline datasets}{Frits Sweijen, 2018.11.22}

\section{FACTOR}
Factor is a facet-based calibration tool for LOFAR, normally used to make direction-dependent calibrated images of observations with the Dutch array. Here we explore the possibilities of Factor when run on a dataset containing international stations, as it should be very similar, in principle.

I have so far tried running it on my dataset of 4C43.15, using all international stations and a subset of 40 sub-bands to make testing less heavy on the disks and time. Next I will try 30 sub-bands with only the German stations.

\section*{Source subtraction}
Before we can run the Factor pipeline, we need an empty dataset. In the prefactor repository, the parset \texttt{Init-Subtract-Fast.parset} takes care of this source subtraction. By default it does a low-resolution subtract with a $25"$ cell size and a high-resolution subtract with a $7.55"$ cell size. The fields are $6.5$ and $2.5$ degrees in size, respectively, at each of these steps. On my test datasets it took about $2$ days.

\subsection*{Required modifications}
A few modifications to the parset are required in order to run the subtraction step. This is because Factor normally starts from the uncalibrated DATA column, and then applies the direction-independent solutions. Here our DATA column contains (fully) calibrated data already, so this is no longer needed. These alterations are listed here.

\begin{enumerate}
    \item Remove line 368
    \item Replace line 415 with \texttt{correct\_high.argument.steps = [avg]}
    \item Remove lines 416-418
    \item Remove line 615
\end{enumerate}

\subsubsection*{Caveats}
Of course this subtraction is not sufficient for long baselines, as the models are very incomplete. It likely helps in suppressing flux on the shorter baselines, however, which may in turn stabilize self-calibration on a larger uv range extending further to the core. It does show in the images that there is something not right, relating to the resolution. When imaging the sources at$0.05"/$px, we see a grid of black holes (not the astrophysical ones) appearing through out the image, spaced by almost exactly $7.5"$, the cell size of the high-resolution subtraction.

\section*{Running Factor}
To run factor the general set up can be followed. There is one thing that needs to be kept in mind: Factor wants to apply the direction independent solutions when putting sources back. Since my DATA column already contains calibrated data, a dummy ParmDB was created with 0 phase and unity amplitude, using \texttt{parmdbm}. The snippet of Python code \texttt{make\_dummy\_parmdb.py} generates the parmdbm config file, which you then use as \texttt{parmdbm < configfile} (adjust the frequency range to your needs).

Next I generate directions for factor by exploiting the Voronoi tesselation, having it generate squares around a few test sources. This code can be found in \texttt{make\_directions.py}. Finally the Factor parset was set up to use a pixel scale of $0.05"/\mathrm{px}$ for imaging and the pipeline can be started with \texttt{runfactor /path/to/factor.parset}.

\section*{Results}
\textit{Some pictures will appear here soon.}

\section*{Conclusion}
I think that the general approach of Factor probably works, i.e. phaseshift to a local calibrator source, calibrate it and image a large(r) field around it, but we most likely cannot directly use it.

\end{document}