ismgcc

ISMGCC (InterStellar Medium Gaussian Component Clustering): Finding gas structures in molecular ISM using Gaussian decomposition and Graph theory

This repository contains the Python code that implements the method described in ISMGCC: Finding Gas Structures in Molecular Interstellar Medium Using Gaussian Decomposition and Graph Theory ( arXiv:2409.01181 ). It is designed to find molecular gas structures from PPV data cubes, especially in the direction with crowded emissions, e.g., the inner galaxy.

Please refer to the tutorial for code examples.

Parameter setting

The parameters of ISMGCC are summarized here, same as Table 1 of the paper.

Symbol	Recommended Value	Tested Values	Description
$R_{\mathrm{ap}}$	$3$	$3, 5, 7$	Radii of the apertures for velocity coherence clustering, in units of pixels (Sect. 2.1);
$\beta$	0.50	$0.25, 0.50, 0.75, 1.0$	\texttt{bandwidth} coefficient that controls the MeanShift clustering, smaller value leads to more velocity slices (Sect. 2.1);
$\delta$	0.5	$0, 0.5, 0.8$	Minimal value of $N_{\mathrm{same}} / N_{\mathrm{share}}$ for two Gaussian components having a chance to be coherent with each other (Sect. 2.1);
$\eta_0$	0	$0, 3$	Amplitude-to-noise threshold for a Gaussian component to have $P_{\mathrm{real}}>0$ (Eq. 6, Sect. 2.2);
$\eta_1$	5	$3, 5, 7$	Amplitude-to-noise value for a Gaussian component to have $P_{\mathrm{real}}=0.5$ (Eq. 6, Sect. 2.2);
$\epsilon$	1.5	$1.5$	Maximum spatial separation between two Gaussian components for being spatially connected, in units of pixels (Sect. 2.2);
$\gamma$	0.01	$0.001, 0.01, 0.1, 1.0$	The resolution parameter of the Clauset-Newman-Moore algorithm, controls the community sizes in the graph, smaller values lead to larger communities (Sect. 2.3);
$N_{\mathrm{pix,min}}$	16	$16$	Minimal Number of pixels for a structure to be considered as a valid one (Sect. 2.4).

The recommended values shown above are based on the MWISP 13CO(1-0) data. For spectral lines with larger line width (e.g. HI 21cm line), smaller $\beta$ values are recommended, such as $\beta=0.25$.

Notes & FAQ

Q: Why does the code stop running when I killed a previous run with Ctrl + C?

A: To prevent the cache files from being manipulated by multiple processes, *.lock files will be created and removed by the code. But when you have mannually killed the program, the *.lock file won't be released properly. The solution is deleteing all .lock files in the ./.cache/ directory.