This is a package to fit modified blackbodies to photometry data using an affine invariant MCMC.
###Installation The usual
python setup.py build install
###Usage The command line routine is
run_mbb_emcee.py
This has a large number of options which, for example, allow you to compute the IR luminosity or dustmass as part of the fit. Help can be obtained via
run_mbb_emcee.py --help
Carrying out a fit produces a HDF5 output file containing the results, which can either be read directly, or read back into a mbb_results object for analysis. For example, if the output is saved to the file test.h5:
import mbb_emcee
res = mbb_emcee.mbb_results(h5file="test.h5")
T_val = res.par_cen('T')
print("Temperature/(1+z): {:0.2f}+{:0.2f}-{:0.2f} [K]".format(*T_val))
b_val = res.par_cen('beta')
print("Beta: {:0.2f}+{:0.2f}-{:0.2f}".format(*b_val))
res.compute_peaklambda()
p_val = res.peaklambda_cen()
print("Peak Obs wavelength: {:0.1f}+{:0.1f}-{:0.1f} [um]".format(*p_val))
Note that all the blackbody parameters (temperature, etc.) are in the observer frame.
This package is primarily developed in python 3, but should be compatable with python 2.7.
###Input file specification There are two possible input file specifications. The first treats all the instrument responses as delta functions, the second integrates them properly. The latter is slower, and requires that you know the appropriate instrument information.
When using the simple delta function approach, the input file specification are lines of the form: wavelength flux_density flux_density_uncertainty, where the wavelength is specified in microns, and the flux_density and it's uncertainty in mJy. Note that the flux density uncertainties in this file are ignored if the user provides a covariance matrix.
In the second the format is similar, except that the wavlength is replaced with the name of the passband (e.g., SPIRE_250um). The information specifying how the name is used is provided in a filter wheel file. The user can specify their own such file (in a somewhat complicated format -- you will need to read the documentation of the response class), but a default version is also included that should cover most of the standard passbands.
The currently supported passbands in the default set are MIPS_24um, MIPS_70um, MIPS_160um, PACS_70um, PACS_100um, PACS_160um, SPIRE_250um, SPIRE_350um, SPIRE_500um, SABOCA_350um, LABOCA_870um, SCUBA_450um, SCUBA_850um, SCUBA2_450um, SCUBA2_850um, Bolocam_1.1mm, MAMBO2_1.1mm, and GISMO_2mm.
However, this doesn't really work for interferometric observations, where the passbands can usually be tuned in all sorts of complicated ways. One way to handle these is to build your own filter wheel file, but as an alternative a short specification can be provided using some special codes.
For the most common (boxcar) case, the specification is inst_box_cent_width where inst is the instrument name, box specifies a boxcar, cent is the central frequency in GHz (or wavelength, see below), and width is the width (also in GHz). So, for example, PdBI_135_3.6. Delta function passbands are also allowed using inst_delta_cent, where, again, cent is the central frequency.
In addition, a box with a central taken out is also supported. This is meant to represent a dual-sideband or sideband-separating receiver. This is specified in terms of the central frequency, the total width, and the width of the gap (which is centered). So, for example, SMA_dsb_230_8_2 would be a 230 GHz centered band with a total width of 8 GHz but the central 2 GHz with no response.
A specialization of this is also available for the ALMA recievers used in their widest frequency scan mode. Here the specification is inst_alma_cent where, again, cent is the central frequency in GHz. ALMA bands 9 and 10 are not supported, and the assumption is that the standard IF ranges are used. For ALMA band 3, for example, this means 3.75 GHz coverage, a 8 GHz gap, then 3.75 GHz coverage, with the center of the 8 GHz gap at the central frequency. For band 6, on the other hand, the central gap is 12 GHz. More complex tunings require a special built filter file, or maybe the dsb construction used above. Note that this usually means the name of your band is something like ALMA_alma_230, since alma is both a telescope and a type of passband. Yes, this looks kind of stupid, but it's to give you the freedom to name your inst something else if you want. None of the values are case sensitive.
It is also possible to specify the units as microns instead of GHz by appending um to the first numerical argument -- for example, ZSpec_box_1050um_100 is a 100 micron wide box centered at 1050 microns.
###Examining your results You should almost always look at both your actual fit and at histograms of the parameters to make sure there are not long tails of unphysical values, which can happen when your data is not very constraining or some parameter. In particular, be aware of very large values of lambda0, since that parameter is frequently rather poorly constrained by sub-mm data, and which will cause the dustmass computation to produce rather extreme values. This can be done using matplotlib via something like:
import mbb_emcee
import matplotlib.pyplot as plt
import numpy as np
res = mbb_emcee.mbb_results(h5file="test.h5")
wave, flux, flux_unc = res.data
f = plt.figure()
ax1 = f.add_subplot(221)
p_data = ax1.errorbar(wave, flux, yerr=flux_unc, fmt='ro')
p_wave = np.linspace(wave.min() * 0.5, wave.max() * 1.5, 200)
p_fit = ax1.plot(p_wave, res.best_fit_sed(p_wave), color='blue')
ax1.set_xlabel('Wavelength [um]')
ax1.set_ylabel('Flux Density [mJy]')
ax2 = f.add_subplot(222)
h1 = ax2.hist(res.parameter_chain('T/(1+z)'))
ax2.set_xlabel('T / (1 + z)')
ax3 = f.add_subplot(223)
h2 = ax3.hist(res.parameter_chain('beta'))
ax3.set_xlabel(r'$\beta$')
ax4 = f.add_subplot(224)
h3 = ax4.hist(res.lir_chain)
ax4.set_xlabel(r'$L_{IR}$')
f.show()
where the latter (plotting the chain of IR luminosities) requires that you have actually computed them as part of the fit.
###Dependencies This depends on a number of python packages:
- numpy
- scipy
- astropy
- cython
- h5py
- And, perhaps most importantly, emcee
- nose is needed if you want to do the unit tests
- The affine invariant MCMC code is described in Foreman-Mackey et al. (2012)
- This code has been used in a variety of HerMES papers analyzing the SEDs of dusty, star-forming galaxies: