GPTransmissionSpectra

This code detrends lightcurves and generates transmission spectra given data either (1) in a pickle file in the output format of tepspec or (2) given lightcurves and a set of external parameters. To use it you need:

• MultiNest
• batman
• george

The code fits the lightcurve simultaneously using a transit model, the comparison stars using PCA and a gaussian process with a multi-dimensional squared exponential kernel (or a Matern 3/2; for this, simply run the code adding the flag --matern), which takes time, FWHM, airmass, trace position, sky flux, and shift in wavelength of the trace as inputs in the case of output in the tepspec format. If this is not the input format, the user defines which external paraemters to include instead.

USAGE

To use the code is simple:

1. (a) If the input is going to be given by the user with the tepspec input format, put the .pkl file of outputs in a folder which has the same name as the target in the pickle file. Let's assume the name of the target in my pickle file is WASP19. Then, my pickle file (say, named w19_140322.pkl) should be in the folder WASP19/w19_140322.pkl. (b) If you have a set of lightcurves and external parameters, first create a folder called "outputs". Inside it, create a folder with the name of your target (in our case, WASP19). Inside this folder, create different folders for different datasets (e.g., different nights) of the same target --- in the case of our example, we should create a folder called w19_140322. Inside each of these, put your external parameters in a file called eparams.dat, so that each row is the value of the external parameters at different times, and each (space separated) column is a different external parameter. Inside this folder, create two extra folders: a folder called white-light and a folder called wavelength. Inside the white-light folder, create a file called lc.dat which contains the data for the target lightcurve: in its first column the time, the second the (median-substracted) magnitude (-2.51 x log10(flux)) and the third column contains zeros. Create another file called comps.dat where you input the (median-substracted) magnitude of the comparison stars; one comparison star per column. Insite the wavelength folder, create one different folder for each wavelength bin named as wbin0, wbin1, etc., and inside save the (wavelength-dependant) lightcurves of the target and comparison stars in the same format as was done in the white-light folder.

2. Create an options file for the white-light lightcurve (see the wl_options.dat file for an example with WASP-19, wl_options_h5.dat for HATS-5), which will contain the priors for the fit (assumed to be truncated gaussians). If your input is not in the tespect format (i.e, you followed step (b) in 1.), the datafile parameter should be still filled with the .pkl extension (even though you don't have a pickle file --- this is just to allow backcompatibility of the code). In the case of our example, it should be WASP19/w19_140322.pkl.

3. Run the white-light analysis by doing python run_wl_analysis.py -ofile youroptionsfile.dat; if you don't have a pickle file in the format of tepspec, then add the --nopickle flag (i.e., do python run_wl_analysis.py -ofile youroptionsfile.dat --nopickle). This will run the white-light analysis. Once is done, the results will be outputted in a folder named outputs.

4. When the white-light fit ends, if you used the tepspec pickle mode, the outputs folder will have a white-light folder, inside of which you will find a results.dat file. If you didn't, these same files will be written there. This contain the posterior parameters of the best-fit. Inside white-light there will also be folders named PCA_n, where n is the number of PCA components used for each fit (the code tries them all, and then bayesian-model average the results to obtain the results.dat file); inside each PCA_n folder there will be a detrended_lc.dat file with the detrended lightcurves (first column is time, second detrended lightcurve, third noise on the detrended lightcurve and fourth the best-fit transit model) and a model_lc with the raw magnitude of the target and the full systematic model (i.e., the full model minus the transit). Note you can join the data in these two files to generate the full model fitted to the data.

5. Use the results.dat to create an options file for the wavelength-dependant fits, where every parameter of the transit will be fixed except for the limb-darkening parameters and p=rp/rs (and, of course, the GP and PCA components of the fit). See the wavelength_options_w19.dat file for an example. Note: if analysing multiple nights, consider using results.dat values averaged over all nights.

6. Run the code by doing either python run_wavelength_analysis.py -ofile yourNEWoptionsfile.dat if you want to perform PCA + GP on each wavelength range, or python run_wavelength_cmc_analysis.py -ofile yourNEWoptionsfile.dat -wofile youroptionsfile.dat if you want to run Common Mode Correction (CMC) } GP, where youroptionsfile.dat is the same white-light option file created for step 3 above. Here, if your input format is not the tepspec pickle, you can use the --nopickle flag again. Here, for the common-mode correction, only the first star in the list of your yourNEWoptionsfile.dat options file will be used; the white-light lightcurve of the target and that comparison star will be divided, the best-fit transit lightcurve from step 3 and 4 will be divided to that resulting lightcurve, and this will be the common-mode correction signal. This signal, in turn, will be divided to the resulting lightcurve of the division between the target and the same comparison on every wavelength, and this lightcurve will be fitted directly without the PCA component (but with a zero-point and a GP).

7. Once it runs, generate the transmission spectrum by running python compile_transpec.py -ofile yourNEWoptionsfile.dat if you did a PCA + GP fit, or add --CMC in case you want to compile the results from the common-mode correction fit. This will be saved as transpec.dat inside the outputs folder of your target lightcurve in the case of a PCA + GP fit or transpec_cmc.dat in the case of a common-mode correction fit.

TODO

• Try other kernels?