2018 CMB maps
Overview[edit]
This section describes the CMB maps produced from the Planck data. These products are derived from some or all of the nine frequency channel maps using different techniques and, in some cases, using other constraints from external data sets. Here we give a brief description of each product and how it is obtained, followed by a description of the FITS file containing the data and associated information. All the details can be found in Planck2020A4^{[1]} and, for earlier releases, in Planck2013XII^{[2]} and Planck2015A09^{[3]}.
2018 CMB maps[edit]
CMB maps have been produced using four different methods: COMMANDER, NILC, SEVEM, and SMICA, as described in the CMB and foreground separation section and also in Appendices AD of Planck2020A4^{[1]} and references therein.
For each method we provide the following:
 Fullmission CMB intensity map, with corresponding confidence mask and effective beam transfer function.
 Fullmission CMB polarisation map, with corresponding confidence mask and effective beam transfer function.
 Inpainted CMB intensity and polarisation maps, intended for PR purposes.
In addition, and for characterisation purposes, we include four other sets of maps from two data splits: odd/even ring and first/second halfmission. Halfdifference maps can be used to provide an approximate noise estimates for the full mission, but they should be used with caution. Each split has caveats in this regard: there are noise correlations between the odd/even split maps, and unobserved pixels in both splits. Masks flagging unobserved pixels are provided for each split, and we strongly encourage use of these when analysing split maps.
In addition, for SMICA, we also provide a CMB map from which SunyaevZeldovich (SZ) sources have been projected out, while SEVEM provides cleaned singlefrequency maps at 70, 100, 143 and 217 GHz for both intensity and polarization.
All CMB products are provided at an approximate angular resolution of 5 arcmin FWHM, and HEALPix resolution N_{side}=2048. Explicit effective beam profiles are provided for each foreground reduced CMB map.
For a complete description of the above data structures, see below; the content of the first extensions is illustrated and commented in the table below.
The gallery below shows the inpainted fullmission CMB maps (T, Q and U) from each pipeline. The temperature maps are shown at 5 arcmin FWHM resolution, while the polarization maps are shown at 80 arcmin FWHM resolution, in order to suppress instrumental noise.
Product description[edit]
COMMANDER[edit]
 Principle
 COMMANDER is a Planck software code implementing Bayesian parametric component separation. Each astrophysical signal component is modelled in terms of a small number of free parameters per pixel, typically in terms of an amplitude at a given reference frequency and a small set of spectral parameters, and these are fitted to the data with an MCMC Gibbs sampling algorithm. A new feature in the Planck 2018 analysis is support for multiresolution analysis, allowing reconstruction of both CMB and foreground maps at full angular resolution. Only CMB products are provided from Commander in the Planck 2018 release (see Planck2020A4^{[1]} for details), while for polarization both CMB and foreground products are provided. For temperature, a dedicated lowresolution CMB map is also provided as part of the Planck likelihood package.
 Resolution (effective beam)
 The Commander sky maps have different angular resolutions depending on data products:
 CMB temperature and polarization and thermal dust polarization maps are provided at 5 arcmin FWHM resolution
 Synchrotron polarization maps are provided at 40 arcmin FWHM resolution
 The lowresolution CMB likelihood map is provided at an angular resolution of 40 arcmin FWHM.
 Confidence mask
 The Commander temperature confidence mask is produced by thresholding the chisquare map characterizing the global fits, combined with direct CO amplitude thresholding to eliminate known leakage effects. In addition, we exclude all pixels brighter than 10mK in the 30GHz map, in order to remove particularly bright radio sources. Finally, we remove by hand the Virgo and Coma clusters, as well as the Crab nebula. A total of 88% of the sky is admitted for analysis.
 The Commander polarization mask is produced in a similar manner, starting by thresholding the chisquared map. In addition, we exclude all pixels for which the thermal dust polarization amplitude is brighter than 20µK_{RJ} at 353GHz, as well as particularly bright objects in the PCCS2 source catalog. Finally, we remove a small region that is particularly contaminated by cosmic ray glitches. A total of 86% of the sky is admitted for analysis.
 Preprocessing and data selection
 The primary Commander 2018 analysis is carried out at full angular resolution, and no smoothing to a common resolution is applied to the maps, in constrast to the procedure employed in previous releases. The temperature analysis employs all nine Planck frequency maps between 30 and 857 GHz, while the polarization analysis employs the seven frequency maps between 30 and 353 GHz. No external data are used in the 2018 Commander analysis.
 Priors
 The following priors are enforced in the Commander analysis:
 The 30 GHz zerolevel is fixed to zero, while the 44 and 70 GHz zerolevels are fitted freely with uniform priors. HFI zerolevels are fitted with a strong CIB prior.
 Dipoles are fitted only at 70 and 100 GHz; all other are fixed to zero.
 Gaussian priors are enforced on spectral parameters, with values informed by the values derived in the high signaltonoise areas of the sky
 The Jeffreys ignorance prior is enforced on spectral parameters in addition to the informative Gaussian priors
 Fitting procedure
 Given data and priors, Commander either maximizes, or samples from, the Bayesian posterior, P(thetadata). Because this is a highly nonGaussian and correlated distribution, involving millions of parameters, these operations are performed by means of the Gibbs sampling algorithm, in which joint samples from the full distributions are generated by iteratively sampling from the corresponding conditional posterior distributions, P(theta_i data, theta_{j/=i}). All parameters are optimized jointly.
NILC[edit]
 Principle
 Needlet Internal Linear Combination (or NILC in short) is a blind component separation method for the measurement of Cosmic Microwave Background (CMB) from the multifrequency observations of sky. It is an implementation of an Internal Linear Combination (ILC) of the frequency channels under consideration with minimum error variance on a frame of spherical wavelets called needlets, allowing localized filtering in both pixel space and harmonic space. The method includes multipoles up to 4000. Temperature and, Emode and Bmode of polarization maps are produced independently. The Q and U maps of CMB polarization have been reconstructed from the corresponding Emode and Bmode maps.
 Resolution (effective beam)
 The effective beam is equivalent to a Gaussian circular beam with FWHM=5 arcminutes.
 Confidence mask
 For each needlet scale, we identify the frequency channel that contributes the most to the final reconstruction of CMB for that band. Then we scale the sky maps for 30GHz and 353GHz to that frequency channel to obtain the scaledsky map and compute the root mean square (RMS) of full mission CMB map. The mask is obtained by setting a cutoff at each needlet scale. The cutoff values are 500 times the RMS value of CMB for temperature and 1500 times the RMS value of CMB for polarization for each scale. The final mask is reconstructed from the union of all the masks obtained at different needlet scales. The confidence masks cover the most contaminated regions of the sky, leaving approximately 78.6 per cent of useful sky for temperature and 82 per cent for polarization.
 Preprocessing
 All sky maps are convolved/deconvolved in harmonic space, to a common beam resolution corresponding to a Gaussian beam of 5 arcminutes FWHM. A very small preprocessing mask has been used on the temperature sky maps. Prior to implement the pipeline on the sky maps, the masked regions are filled using PSM tools which uses an increasing number of neighboring pixels to fill regions deeper in the hole. At each iteration it uses pixels at up to twice the diameter of the pixel times number of iteration. No preprocessing has been done on polarization sky maps.
 Linear combination
 Needlet ILC weights are computed for each of T, E and B, for each scale and for each pixel of the needlet representation at that scale. For each of T, E and B, a fullsky CMB map, at 5 arcminutes beam resolution, is synthesized from the NILC needlet coefficients.
 Postprocessing
 E and B maps are recombined into Q and U products using standard HEALPix tools.
SEVEM[edit]
 Principle
 SEVEM produces cleaned CMB maps at several frequencies by using a procedure based on template fitting in real space. The templates used in the SEVEM pipeline are typically constructed by subtracting two close Planck frequency channel maps, after first smoothing them to a common resolution to ensure that the CMB signal is properly removed. A linear combination of the templates is then subtracted from (hitherto unused) map d to produce a cleaned CMB map at that frequency. This is done in real space at each position on the sky:
where CMB map.
is the number of templates. The coefficients are obtained by minimising the variance of the cleaned map outside a given mask. Note that the same expression applies for I, Q and U. Although we exclude very contaminated regions during the minimization, the subtraction is performed for all pixels and, therefore, the cleaned maps cover the fullsky (although foreground residuals are expected to be particularly large in those areas excluded by the minimisation). In the cleaning process, no assumptions about the foregrounds or noise levels are needed, rendering the technique very robust. A subset of the cleaned single frequency maps are then combined to obtain the finalThere are several possible configurations of SEVEM with regard to the number of frequency maps which are cleaned or the number of templates that are used in the fitting. Note that the production of cleaned maps at different frequencies is of great interest by itself in order to test the robustness of the results, and these intermediate products (cleaned maps at individual frequencies for intensity and polarization) are also provided in the archive. Therefore, to define the best strategy, one needs to find a compromise between the number of maps that can be cleaned and the number of templates that can be constructed.
 Intensity
For the CMB intensity case, we have cleaned the 70, 100, 143 and 217 GHz maps using a total of five templates. In particular, three templates constructed as the difference of two consecutive Planck channels smoothed to a common resolution [30GHz – 44GHz], [44GHz – 70GHz] and [543GHz – 535GHz] as well as a fourth template given by the 857 GHz channel are used to clean the 100, 143 and 217 GHz maps. Before constructing the templates, the six frequency channels involved in the templates are inpainted at the corresponding point source positions detected at each frequency using the Mexican Hat Wavelet algorithm (these positions are given in the provided point sources masks). The size of the holes to be inpainted is determined taking into account the beam size of the channel as well as the flux of each source. The inpainting algorithm is based on simple diffuse inpainting, which fills one pixel with the mean value of the neighbouring pixels in an iterative way. To avoid inconsistencies when subtracting two channels, each frequency map is inpainted on the sources detected in that map and on the second map (if any) used to construct the template. Then the maps are smoothed to a common resolution, by convolving the first map with the beam of the second one and viceversa. For the fourth template, we simply filter the inpainted 857 GHz map with the 545 GHz beam. The cleaned 70 GHz map is produced similarly by considering two templates, the [30GHz – 44GHz] map and a second template obtained as [353GHz – 143GHz] constructed at the original resolution of the 70 GHz map.
The coefficients to clean the frequency maps are obtained by minimising the variance outside the analysis mask, that covers the 1 per cent brightest emission of the sky as well as point sources detected at all frequency channels. Once the maps are cleaned, each of them is inpainted on the point sources positions detected at that (raw) channel. Then, the Mexican Hat Wavelet algorithm is run again, now on the cleaned maps. A number of new sources are found and are also inpainted at each channel. The resolution of the cleaned map is the same as that of the original data. Our final CMB map is then constructed by combining the 143 and 217 GHz maps by weighting the maps in harmonic space taking into account the noise level, the resolution and a rough estimation of the foreground residuals of each map (obtained from realistic simulations). This final map has a resolution corresponding to a Gaussian beam of fwhm=5 arcminutes and N_{side}=2048 and the maximum considered multipole is . The monopole and dipole over the fullsky have been subtracted from the final CMB map.
In addition, the cleaned CMB maps produced at 70, 100, 143 and 217 GHz frequencies are also provided. The resolution of these maps is the same as that of the uncleaned frequency channels and have been constructed at N_{side}=1024 for 70 GHz and N_{side}=2048 for the rest of the maps. They have been inpainted at the position of the point sources detected in the raw and cleaned maps (these positions are given in the corresponding inpainting masks). The monopole and dipole over the fullsky have also been removed from each of the cleaned maps.
The confidence mask is produced by studying the differences between several SEVEM CMB reconstructions, which correspond to maps cleaned at different frequencies or using different analysis masks. The obtained mask leaves a useful sky fraction of approximately 84 per cent.
 Polarization
To clean the polarization maps, a procedure similar to the one used for intensity data is applied to the Q and U maps independently. Cleaned maps at 70, 100, 143 and 217 GHz are also produced but, given that a smaller number of frequency channels is available for polarization, the templates selected to clean the maps are different. In particular, we clean the 70 GHz map using two templates and the rest of the channels using different combinations of three templates.
Following the same procedure as for the intensity case, those channels involved in the construction of the templates are inpainted in the position of the sources detected in the raw frequency maps. The sources are selected from a nonblind search, based on the Filtered Fusion technique, using as candidates those sources detected in intensity. These inpainted maps are then used to construct a total of six templates, one of them at two different resolutions. To trace the synchrotron emission, we construct a template as the subtraction of the 30 GHz minus the 44 GHz map, after being convolved with the beam of each other. For the dust emission, the following templates are considered: [217GHz – 143GHz], [217GHz – 100GHz] and [143GHz – 100GHz] at 1 degree resolution, [353GHz – 217GHz] and [353GHz – 143GHz] at 10 arcminutes resolution. The last template is also constructed at the resolution of the 70 GHz channel, in order to clean that map.
Different combinations of these templates (see Table C.3 in Planck2020A4^{[1]} for details) are then used to clean the raw 70, 100, 143 and 217 GHz channels (at its native resolution). The corresponding linear coefficients are estimated independently for Q and U by minimising the variance of the cleaned maps outside a mask, that covers the point sources detected in polarization and the 3 per cent brightest Galactic emission. Once the maps have been cleaned, inpainting of the point sources detected at the corresponding raw maps is carried out. Then the nonblind search for point sources is run again on the cleaned maps and the new identified sources are also inpainted. The 100, 143 and 217 GHz cleaned maps are then combined in harmonic space, using E and B decomposition, to produce the final CMB maps for the Q and U components at a resolution of 5' (Gaussian beam) for a HEALPix parameter N_{side}=2048. The maximum considered multipole is . Each map is weighted taking into account its noise and resolution. In addition, the lowest multipoles of the 217 GHz cleaned map are downweighting, since they are expected to be more contaminated by the presence of residual systematics.
The cleaned CMB maps at individual frequency channels produced as intermediate steps of SEVEM are also provided for Q and U, at their native resolution. The four pairs of Q/U maps have been inpainted in the positions of the detected point sources (given by the corresponding inpainting masks).
The confidence mask is constructed as the product of two different masks. One of them is obtained from the 353 GHz data channel and excludes those regions more contaminated by thermal dust. The second mask is constructed by thresholding a map of the ratio between the locally estimated RMS of P in the cleaned CMB map, over the same quantity expected for a map containg CMB plus noise. The combination of these two masks leaves a useful sky fraction of approximately 80 per cent.
 Resolution
 The cleaned CMB maps for intensity and polarization are constructed at N_{side}=2048 and at the standard resolution of 5 arcminutes (Gaussian beam). The maximum considered multipole is for intensity and for polarization.
 Confidence masks
 The confidence masks cover the most contaminated regions of the sky, leaving approximately 84 per cent of useful sky for intensity, and 80 per cent for polarization.
 Point source masks
 The point source masks contain the holes corresponding to the point sources detected at each raw Planck frequency channel in intensity and polarization. The number of sources detected are given in the upper part of Table C.1 of Planck2020A4^{[1]}. There is one mask for intensity and another one for polarization per frequency channel. When using the Planck channels in the construction of the templates, these have been inpainted in the positions of the point sources given in these masks, to reduce the emission from this contaminant in the templates and its propagation to the final cleaned CMB maps.
 Inpainting masks
 The inpainting masks include the positions of the point sources that have been inpainted in the cleaned singlefrequency maps. They contain point sources detected at the original raw data at those frequencies plus the sources detected in the cleaned frequency maps (see Table C.1 of Planck2020A4^{[1]}). There is a mask for intensity and another one for polarization for each of the cleaned frequency maps (70, 100, 143 and 217 GHz) as well as the corresponding masks for the combined map. The latter are constructed as the product of the individual frequency masks of those cleaned channels that are combined in the final CMB map (i.e., the product of 143 and 217 GHz masks for intensity and of 100, 143 and 217 GHz for polarization). Note that the inpainted positions are not excluded by default by the SEVEM confidence mask, but only if they are considered unreliable with the general procedure used to construct the SEVEM confidence mask.
Foregroundsubtracted maps[edit]
In addition to the regular CMB maps, SEVEM provides maps cleaned of the foregrounds for selected frequency channels (categorized as fgsubsevem in the archive). In particular, for both intensity and polarization there are cleaned CMB maps available at 70, 100, 143 and 217 GHz, provided at the original resolution and N_{side} of the uncleaned channel (1024 for 70 GHz and 2048 for the rest of the maps).
SMICA[edit]
 Principle
 SMICA produces CMB maps by linearly combining Planck input channels with multipoledependent weights, including multipoles up to . Temperature and polarization maps are produced independently. In temperature, two distinct CMB renderings are produced and then merged (hybridized) together into a single CMB intensity map. In polarization, the E and B modes are processed independently and the results are combined to produce Q and U maps.
 Resolution (effective beam)
 The SMICA intensity map has an effective beam window function of 5 arcminutes which is truncated at .
 The SMICA Q and U maps are obtained similarly but are produced at =1024 with an effective beam of 10 arcminutes.
 Confidence mask
 A confidence mask is provided which excludes some parts of the Galactic plane, some very bright areas and masked point sources. This mask provides a qualitative (and subjective) indication of the cleanliness of a pixel.
 Intensity.
SMICA operation starts with a preprocessing step to deal with regions of very strong emission (such as the Galactic center) and point sources. The nine preprocessed Planck frequency channels from 30 to 857 GHz are then masked and harmonically transformed up to CMBdominated areas. These two sets of spectral statistics are used to determine two sets of harmonic weights which are thus adapted to two different levels of contamination. Two CMB intensity maps are produced and then merged into a single intensity product. The merging process is devised so that the information at high Galactic latitude and mediumtohigh multipole is provided by the CMB map computed from high Galatic latitude statistics (note that this map does not include the LFI channels) while the remaining information is provided by the other CMB map (which does include all Planck channels). See Planck2020A4^{[1]} for more details.
to form spectral statistics (all auto and cross angular spectra). Two different masks are used to compute the spectral statistics. The first one preserves most of the sky while the second preserves Polarisation.
The SMICA pipeline for polarization uses all the 7 polarized Planck channels. The E and B modes of the frequency maps are processed independently by SMICA to produce E and B modes of the CMB map from which Q and U maps are derived. The foreground model fitted by SMICA is 6dimensional which is the maximal dimension supported by SMICA when operating in blind mode, that is, assuming nothing about the foregrounds except that they can be represented by a superposition of 6 components with unconstrained emission laws, unconstrained angular spectra and unconstrained angular correlation. See Planck2020A4^{[1]} for more details.
Note: in general, any I, Q and U CMB map can be transformed into a T, E and B CMB map using the HEALpix routines "anafast" and "synfast"(See the links below for the details). The "anafast" routine generates harmonic coefficients of T, E and B maps from the full sky I, Q and U maps. Finally, the full sky T, E and B maps in real space are generated using "synfast" routine separately from the corresponding harmonic coefficients obtained using "anafast". Further details about the spherical harmonic transform from HEALPix can be found in https://healpix.jpl.nasa.gov/html/intro.htm, https://healpix.jpl.nasa.gov/html/idlnode25.htm, and https://healpix.jpl.nasa.gov/html/idlnode27.htm". In the particular case of NILC, that works in needlet space, the IQU maps are converted into TEB maps using anafast and synfast, while in the case of SMICA, that works in harmonic space, the IQU maps are converted into TEB harmonic coefficents (alms) using anafast only.
Common Masks[edit]
Common masks have been defined for analysis of the CMB temperature and polarization maps. In previous releases, these were constructed simply as the union of the individual pipeline confidence masks. In the 2018 release, a more direct approach has been adopted, by thresholding the standard deviation map evaluated between each of the four cleaned CMB maps. This standard deviation mask is then augmented with the Commander and SEVEM confidence masks, as well as with the SEVEM and SMICA inpainting masks.
In addition, we provide masks for unobserved pixels for the halfmission and oddeven data splits, as well as an inpainting mask. The latter is not intended for scientific analysis, but for producing visually acceptable CMB representation for PR purposes.
In total, we provide the following masks:
 COM_Mask_CMBcommonMaskInt_2048_R3.00.fits  Temperature confidence mask with f_{sky} = 77.9%. This is the preferred mask for temperature science analysis.
 COM_Mask_CMBcommonMaskPol_2048_R3.00.fits  Polarization confidence mask with f_{sky} = 78.1%. This is the preferred mask for polarization science analysis.
 COM_Mask_CMBHMMisspixMaskInt_2048_R3.00.fits  Temperature halfmission missing pixels mask with f_{sky} = 96.0%. This should be applied in analyses of the halfmission split temperature maps.
 COM_Mask_CMBHMMisspixMaskPol_2048_R3.00.fits  Polarization halfmission missing pixels mask with f_{sky} = 96.1%. This should be applied in analyses of the halfmission split polarization maps.
 COM_Mask_CMBHMMisspixMaskInt_2048_R3.00.fits  Temperature halfmission missing pixels mask with f_{sky} = 98.1%. This should be applied in analyses of the halfmission split temperature maps.
 COM_Mask_CMBHMMisspixMaskPol_2048_R3.00.fits  Polarization halfmission missing pixels mask with f_{sky} = 98.1%. This should be applied in analyses of the halfmission split polarization maps.
 COM_Mask_CMBInpaintingMaskInt_2048_R3.00.fits  Temperature CMB inpainting mask with f_{sky} = 97.9%.
CMBsubtracted frequency maps ("Foreground maps")[edit]
These are the fullsky, fullmission frequency maps in intensity and polarization from which the CMB has been subtracted. The maps contain foregrounds and noise. They are provided for each frequency channel and for each component separation method. They are grouped into 8 files, two for each method of which there is one for each instrument. The maps are are at N_{side} = 1024 for the three LFI channels and at N_{side} = 2048 for the six HFI channels. The filenames are:
 LFI_Foregrounds{method}_1024_Rn.nn.fits (145 MB each)
 HFI_Foregrounds{method}_2048_Rn.nn.fits (1.2 GB each)
To remove the CMB, the respective CMB map was first deconvolved with the 5 arcmin beam, then convolved with the beam of the frequency channel, and finally subtracted from the frequency map. This was done using the in harmonic space, assuming a symmetric beam.
The CMBsubtracted maps have complicated noise properties. The CMB maps contain a noise contribution from each of the frequency maps, depending on the weights with which they were combined. Therefore subtracting the CMB map from a frequency channel contributes additional noise from the other frequency channels. This caveat is particularly important for polarization, for which the noise in the cleaned CMB maps is large. After subtraction this noise term is perfectly correlated between frequency channels, with a perfect blackbody spectrum with T=2.7255K. Caution is therefore warranted when using these maps for scientific analysis.
The frequency maps from which the CMB have been subtracted are:
 LFI_SkyMap_0nn_1024_R3.00_full.fits
 HFI_SkyMap_nnn_2048_R3.00_full.fits
Note that the zodiacal light correction described here was applied to the HFI temperature maps before the CMB subtraction.
Masks[edit]
Summary table with the various masks that have been either been used or produced by the component separation methods to pre or postprocess the CMB maps.
Common mask filename  Field  Description  

COM_Mask_CMBcommonMaskInt_2048_R3.00.fits  TMASK  Common temperature confidence mask.  
COM_Mask_CMBcommonMaskPol_2048_R3.00.fits  PMASK  Common polarization confidence mask.  
COM_Mask_CMBHMMisspixMaskInt_2048_R3.00.fits  TMASK  Missing pixels temperature mask for the halfmission data split.  
COM_Mask_CMBHMMisspixMaskPol_2048_R3.00.fits  PMASK  Missing pixels polarization mask for the halfmission data split.  
COM_Mask_CMBOEMisspixMaskInt_2048_R3.00.fits  TMASK  Missing pixels temperature mask for the oddeven data split.  
COM_Mask_CMBOEMisspixMaskPol_2048_R3.00.fits  PMASK  Missing pixels polarization mask for the oddeven data split.  
COM_Mask_CMBInpaintingMaskInt_2048_R3.00.fits  TMASK  Temperature inpainting mask.  
Pipeline specific mask filename  Field  Description  
COM_CMB_IQUcommander_2048_R3.00_full.fits  TMASK  Commander temperature confidence mask.  
PMASK  Commander polarization confidence mask.  
COM_CMB_IQUnilc_2048_R3.00_full.fits  TMASK  NILC temperature confidence mask.  
PMASK  NILC polarization confidence mask.  
COM_CMB_IQUsevem_2048_R3.00_full.fits  TMASK  SEVEM temperature confidence mask.  
PMASK  SEVEM polarization confidence mask.  
TMASKINP  SEVEM polarization (preprocessing) inpainting mask.  
PMASKINP  SEVEM polarization (preprocessing) inpainting mask.  
COM_CMB_IQUsmica_2048_R3.00_full.fits  TMASK  SMICA temperature confidence mask.  
PMASK  SMICA polarization confidence mask.  
TMASKINP  SMICA polarization (preprocessing) inpainting mask.  
PMASKINP  SMICA polarization (preprocessing) inpainting mask. 
Inputs[edit]
The input maps are the sky temperature maps described in the Sky temperature maps section. All pipelines use all maps between 30 and 857 GHz in temperature, and all maps between 30 and 353 GHz in polarization.
CMB file names[edit]
The CMB products are provided as a set of five files per pipeline, one file covering some part of the entire mission (full mission; first halfmission; second halfmission; odd rings; and even rings), with a filename structure on the form
 COM_CMB_IQU{method}2048R3.00_{full,hm1,hm2,oe1,oe2}.fits
 COM_CMB_IQUSEVEM2048R3.01_{full,hm1,hm2,oe1,oe2}.fits.
UPDATE 17 January 2019: version R3.00 of the SEVEM CMB map has been replaced with version R3.01 because in version R3.00 the temperatue and polarization effective beams were missing.
The first extension contains the fullsky CMB maps in the fields called I_STOKES, Q_STOKES, U_STOKES. The fullmission files additionally contains an ASCII table with the effective beam transfer function in the second extension. The structure of each file is given as follows:
Ext. 1. or 2. EXTNAME = COMPMAP (BINTABLE)  

Column Name  Data Type  Units  Description 
I_STOKES  Real*4  K_cmb  I map 
Q_STOKES  Real*4  K_cmb  Q map 
U_STOKES  Real*4  K_cmb  U map 
TMASK  Int  none  Temperature confidence mask (fullmission only) 
PMASK  Int  none  Polarisation confidence mask (fullmission only) 
I_STOKES  Real*4  K_cmb  I inpainted map 
Q_STOKES  Real*4  K_cmb  Q inpainted map 
U_STOKES  Real*4  K_cmb  U inpainted map 
TMASKINP  Int  none  Temperature confidence mask (fullmission SEVEM, SMICA only) 
PMASKINP  Int  none  Polarisation confidence mask (fullmission SEVEM, SMICA only) 
Keyword  Data Type  Value  Description 
ASTCOMP  String  CMB  Astrophysical compoment name 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
POLCCONV  String  COSMO  Polarization convention 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  2048  Healpix Nside 
METHOD  String  name  Cleaning method (COMMANDER/NILC/SEVEM/SMICA) 
Optional Ext. 2. or 3. EXTNAME = BEAM_TF (BINTABLE). ONLY FULLMISSION DATA FILES  
Column Name  Data Type  Units  Description 
INT_BEAM  Real*4  none  Effective beam transfer function. 
POL_BEAM  Real*4  none  Effective beam transfer function. 
Keyword  Data Type  Value  Description 
LMIN  Int  value  First multipole of beam WF 
LMAX_I  Int  value  Last multipole for Int beam TF 
LMAX_P  Int  value  Last multipole for Pol beam TF 
METHOD  String  name  Cleaning method 
All maps are provided in thermodynamic units (K_{cmb</cmb>), with Nside=2048 and a nominal angular resolution of 5' FWHM.
}
CMB simulations[edit]
Endtoend simulations corresponding to each of the CMB data products are provided in terms of 999 CMB realization and 300 noise realizations individually propagated through each pipeline. These files are called
 dx12_v3_{method}_{cmb,noise,noise_hm1,noise_hm2,noise_oe1,noise_oe2}_mc_?????_raw.fits
 dx12_v3_sevem_{freq}_{cmb,noise,noise_hm1,noise_hm2,noise_oe1,noise_oe2}_mc_?????_raw.fits for SEVEM cleaned cmb maps at single frequencies.
 dx12_v3_smica_nosz_{cmb,noise,noise_hm1,noise_hm2,noise_oe1,noise_oe2}_mc_?????_raw.fits for SMICA SZfree cmb maps.
Note that only 999 CMB realizations are available, as one realization was corrupted during processing.
Previous Releases: (2015) and (2013) CMB Maps[edit]
2015 Release of CMB maps
CMB maps
CMB maps have been produced using four different methods: COMMANDER, NILC, SEVEM, and SMICA, as described in the CMB and foreground separation section and also in Appendices AD of Planck2015A09^{[3]} and references therein.
As discussed extensively in Planck2015A01^{[4]}, Planck2015A06^{[5]}, Planck2015A08^{[6]}, and Planck2015A09^{[3]}, the residual systematics in the Planck 2015 polarization maps have been dramatically reduced compared to 2013, by as much as two orders of magnitude on large angular scales. Nevertheless, on angular scales greater than 10 degrees, correponding to l < 20, systematics are still nonnegligible compared to the expected cosmological signal.
It was not possible, for this data release, to fully characterize the largescale residuals from the data or from simulations. Therefore all results published by the Planck Collaboration in 2015 which are based on CMB polarization have used maps which have been highpass filtered to remove the large angular scales. We warn all users of the CMB polarization maps that they cannot yet be used for cosmological studies at large angular scales.
For convenience, we provide as default polarized CMB maps from which all angular scales at l < 30 have been filtered out.
For each method we provide the following:
 Fullmission CMB intensity map, confidence mask and beam transfer function.
 Fullmission CMB polarisation map,
 A confidence mask.
 A beam transfer function.
In addition, and for characterisation purposes, we include six other sets of maps from three data splits: first/second halfring, odd/even years and first/second halfmission. For the year1,2 and halfmission1,2 data splits we provide halfsum and halfdifference maps which are produced by running the corresponding sums and differences inputs through the pipelines. The halfdifference maps can be used to provide an approximate noise estimates for the full mission, but they should be used with caution. Each split has caveats in this regard: there are noise correlations between the halfring maps, and missing pixels in the other splits. The Intensity maps are provided at Nside = 2048, at 5 arcmin resolution, while the Polarisation ones are provided at Nside = 1024, at 10 arcmin resolution. All maps are in units of K_{cmb}.
In addition, for each method we provide three sets of files, each categorized by the "R2.0X" label as follows:
 R2.02
This set of intensity and polarisation maps are provided at a resolution of Nside=1024. The Stokes Q and U maps are highpass filtered to contain only modes above l > 30, as explained above and as used for analysis by the Planck Collaboration; THESE ARE THE POLARISATION MAPS WHICH SHOULD BE USED FOR COSMOLOGICAL ANALYSIS. Each type of map is packaged into a separate fits file (as for "R2.01"), resulting in file sizes which are easier to download (as opposed to the "R2.00" files), and more convenient to use with commonly used analysis software.
 R2.01
This is the most complete set of 2015 CMB maps, containing Intensity products at a resolution of Nside=2048, and both Intensity and Polarisation at resolution of Nside=1024. For polarisation (Q and U), they contain all angular resolution modes. WE CAUTION USERS ONCE AGAIN THAT THE STOKES Q AND U MAPS ARE NOT CONSIDERED USEABLE FOR COSMOLOGICAL ANALYSIS AT l < 30. The structure of these files is the same as for "R2.02".
 R2.00
This set of files is equivalent to the "R2.01" set, but are packaged into only two large files. Warning: downloading these files could be very lengthy...
For a complete description of the above data structures, see below; the content of the first extensions is illustrated and commented in the table below.
The gallery below shows the Intensity, noise from halfmission, halfdifference, and confidence mask for the four pipelines, in the order COMMANDER, NILC, SEVEM and SMICA, from top to bottom. The Intensity maps' scale is [–500.+500] μK, and the noise spans [–25,+25] μK. We do not show the Q and U maps since they have no significant visible structure to contemplate.
Product description
COMMANDER
 Principle
 COMMANDER is a Planck software code implementing pixel based Bayesian parametric component separation. Each astrophysical signal component is modelled in terms of a small number of free parameters per pixel, typically in terms of an amplitude at a given reference frequency and a small set of spectral parameters, and these are fitted to the data with an MCMC Gibbs sampling algorithm. Instrumental parameters, including calibration, bandpass corrections, monopole and dipoles, are fitted jointly with the astrophysical components. A new feature in the Planck 2015 analysis is that the astrophysical model is derived from a combination of Planck, WMAP and a 408 MHz (Haslam et al. 1982) survey, providing sufficient frequency support to resolve the lowfrequency components into synchrotron, freefree and spinning dust. For full details, see Planck2015A10^{[7]}.
 Resolution (effective beam)
 The Commander sky maps have different angular resolutions depending on data products:
 The components of the full astrophysical sky model derived from the complete data combination (Planck, WMAP, 408 MHz) have a 1 degree FWHM resolution, and are pixelized at N_{side}=256. The corresponding CMB map defines the input map for the lowl Planck 2015 temperature likelihood.
 The Commander CMB temperature map derived from Planckonly observations has an angular resolution of ~5 arcmin and is pixelized at N_{side}=2048. This map is produced by harmonic space hybridiziation, in which independent solutions derived at 40 arcmin (using 30857 GHz data), 7.5 arcmin (using 143857 GHz data), and 5 arcmin (using 217857 GHz data) are coadded into a single map.
 The Commander CMB polarization map has an angular resolution of 10 arcmin and is pixelized at N_{side}=1024. As for the temperature case, this map is produced by harmonic space hybridiziation, in which independent solutions derived at 40 arcmin (using 30353 GHz data) and 10 arcmin (using 100353 GHz data) are coadded into a single map.
 Confidence mask
 The Commander confidence masks are produced by thresholding the chisquare map characterizing the global fits, combined with direct CO amplitude thresholding to eliminate known leakage effects. In addition, we exclude the 9year WMAP point source mask in the temperature mask. For full details, see Sections 5 and 6 in Planck2015A10^{[7]}. A total of 81% of the sky is admitted for highresolution temperature analysis, and 83% for polarization analysis. For lowresolution temperature analysis, for which the additional WMAP and 408 MHz observations improve foreground constraints, a total of 93% of the sky is admitted.
NILC
 Principle
 The NeedletILC (hereafter NILC) CMB map is constructed both in total intensity as well as polarization: Q and U Stokes parameters. For total intensity, all Planck frequency channels are included. For polarization, all polarization sensitive frequency channels are included, from 30 to 353 GHz. The solution, for T, Q and U is obtained by applying the Internal Linear Combination (ILC) technique in needlet space, that is, with combination weights which are allowed to vary over the sky and over the whole multipole range.
 Resolution (effective beam)
 The spectral analysis, and estimation of the NILC coefficients, is performed up to a maximum FWHM=5 arcminutes. . The effective beam is equivalent to a Gaussian circular beam with
 Confidence mask
 The same procedure is followed by SMICA and NILC for producing confidence masks, though with different parametrizations. A low resolution smoothed version of the NILC map, noise subtracted, is thresholded to 73.5 squared microK for T, and 6.75 squared microK for Q and U.
SEVEM
 Principle
 SEVEM produces clean CMB maps at several frequencies by using a procedure based on template fitting in real space. The templates are typically constructed from the lowest and highest Planck frequencies and then subtracted from the CMBdominated channels, with coefficients that are chosen to minimize the variance of the clean map outside a considered mask. In the cleaning process, no assumptions about the foregrounds or noise levels are needed, rendering the technique very robust. Two single frequency clean maps are then combined to obtain the final CMB map.
 Resolution
 For intensity the clean CMB map is constructed up to a maximum at Nside=2048 and at the standard resolution of 5 arcminutes (Gaussian beam).
 For polarization the clean CMB map is produced at Nside=1024 with a resolution of 10 arcminutes (Gaussian beam) and a maximum .
 Confidence masks
 The confidence masks cover the most contaminated regions of the sky, leaving approximately 85 per cent of useful sky for intensity, and 80 per cent for polarization.
Foregroundssubtracted maps
In addition to the regular CMB maps, SEVEM provides maps cleaned of the foregrounds for selected frequency channels (categorized as fgsubsevem in the archive). In particular, for intensity there are clean CMB maps available at 100, 143 and 217 GHz, provided at the original resolution of the uncleaned channel and at Nside=2048. For polarization, there are Q/U clean CMB maps for the 70, 100 and 143 GHz (at Nside=1024). The 70 GHz clean map is provided at its original resolution, whereas the 100 and 143 GHz maps have a resolution given by a Gaussian beam with fwhm=10 arcminutes.
SMICA
 Principle
 SMICA produces CMB maps by linearly combining all Planck input channels with multipoledependent weights. It includes multipoles up to . Temperature and polarization maps are produced independently.
 Resolution (effective beam)
 The SMICA intensity map has an effective beam window function of 5 arcminutes which is truncated at and is not deconvolved from the pixel window function. Thus the delivered beam window function is the product of a Gaussian beam at 5 arcminutes and the pixel window function for =2048.
 The SMICA Q and U maps are obtained similarly but are produced at =1024 with an effective beam of 10 arcminutes (to be multiplied by the pixel window function, as for the intensity map).
 Confidence mask
 A confidence mask is provided which excludes some parts of the Galactic plane, some very bright areas and masked point sources. This mask provides a qualitative (and subjective) indication of the cleanliness of a pixel. See section below detailing the production process.
Common Masks
A number of common masks have been defined for analysis of the CMB temperature and polarization maps. They are based on the confidence masks provided by the component separation methods. One mask for temperature and one mask for polarization have been chosen as the preferred masks based on subsequent analyses.
The common masks for the CMB temperature maps are:
 UT78: union of the Commander, SEVEM, and SMICA temperature confidence masks (the NILC mask was not included since it masks much less of the sky). It has f_{sky} = 77.6%. This is the preferred mask for temperature.
 UTA76: in addition to the UT78 mask, it masks pixels where standard deviation between the four CMB maps is greater than 10 μK. It has f_{sky} = 76.1%.
The common masks for the CMB polarization maps are:
 UP78: the union of the Commander, SEVEM and SMICA polarization confidence masks (the NILC mask was not included since it masks much less of the sky). It has f_{sky} = 77.6%.
 UPA77: In addition to the UP78 mask, it masks pixels where the standard deviation between the four CMB maps, averaged in Q and U, is greater than 4 μK. It has f_{sky} = 76.7%.
 UPB77: in addition to the UP78 mask, it masks polarized point sources detected in the frequency channel maps. It has f_{sky} = 77.4%. This is the preferred mask for polarization.
Additional preprocessing masks used mainly for inpainting of the frequency and/or cmb maps is show below in Masks
CMBsubtracted frequency maps ("Foreground maps")
These are the fullsky, fullmission frequency maps in intensity from which the CMB has been subtracted. The maps contain foregrounds and noise. They are provided for each frequency channel and for each component separation method. They are grouped into 8 files, two for each method of which there is one for each instrument. The maps are are at N_{side} = 1024 for the three LFI channels and at N_{side} = 2048 for the six HFI channels. The filenames are:
 LFI_Foregrounds{method}_1024_Rn.nn.fits (145 MB each)
 HFI_Foregrounds{method}_2048_Rn.nn.fits (1.2 GB each)
To remove the CMB, the respective CMB map was first deconvolved with the 5 arcmin beam, then convolved with the beam of the frequency channel, and finally subtracted from the frequency map. This was done using the in harmonic space, assuming a symmetric beam.
The CMBsubtracted maps have complicated noise properties. The CMB maps contain a noise contribution from each of the frequency maps, depending on the weights with which they were combined. Therefore subtracting the CMB map from a frequency channel contributes additional noise from the other frequency channels.
Quadrupole Residual Maps
The secondorder (kinematic) quadrupole is a frequencydependent effect. During the production of the frequency maps the frequencyindependent part was subtracted, which leaves a frequencydependent residual quadrupole. The residuals in the componentseparated CMB temperature maps have been estimated by simulating the effect in the frequency maps and propagating it through the component separation pipelines. The residuals have an amplitude of around 2 μK peaktopeak. The maps of the estimated residuals can be used to remove the effect by subtracting them from the CMB maps.
Production process
COMMANDER
 Preprocessing
 All sky maps are first convolved to a common resolution that is larger than the largest beam of any frequency channel. For the combined Planck, WMAP and 408 MHz temperature analysis, the common resolution is 1 degree FWHM; for the Planckonly, allfrequency analysis it is 40 arcmin FWHM; and for the intermediateresolution analysis it is 7.5 arcmin; while for the fullresolution analysis, we assume all frequencies between 217 and 857 GHz have a common resolution, and no additional convolution is performed. For polarization, only two smoothing scales are employed, 40 and 10 arcmin, respectively. The instrumental noise rms maps are convolved correspondingly, properly accouting for their matrixlike nature.
 Priors
 The following priors are enforced in the Commander analysis:
 All foreground amplitudes are enforced to be positive definite in the lowresolution analysis, while no amplitude priors are enforced in the highresolution analyses
 Monopoles and dipoles are fixed to nominal values for a small set of reference frequencies
 Gaussian priors are enforced on spectral parameters, with values informed by the values derived in the high signaltonoise areas of the sky
 The Jeffreys ignorance prior is enforced on spectral parameters in addition to the informative Gaussian priors
 Fitting procedure
 Given data and priors, Commander either maximizes, or samples from, the Bayesian posterior, P(thetadata). Because this is a highly nonGaussian and correlated distribution, involving millions of parameters, these operations are performed by means of the Gibbs sampling algorithm, in which joint samples from the full distributions are generated by iteratively sampling from the corresponding conditional posterior distributions, P(theta_i data, theta_{j/=i}). For the lowresolution analysis, all parameters are optimized jointly, while in the highresolution analyses, which employs fewer frequency channels, low signaltonoise parameters are fixed to those derived at low resolution. Examples of such parameters include monopoles and dipoles, calibration and bandpass parameters, thermal dust temperature etc.
NILC
 Preprocessing
 All sky frequency maps are deconvolved using the DPC beam transfer function provided, and reconvolved with a 5 arcminutes FWHM circular Gaussian beam. In polarization, prior to the smoothing process, all sky E and B maps are derived from Q and U using standard HEALPix tools from each individual frequency channels
 Linear combination
 Preprocessed input frequency maps are decomposed in needlet coefficients, specified in the Appendix B of the Planck A11 paper, with shape given by Table B.1. Minimum variance coefficients are then obtained, using all channels for T, from 30 to 353 for E and B.
 Postprocessing
 E and B maps are recombined into Q and U products using standard HEALPix tools.
SEVEM
The templates used in the SEVEM pipeline are typically constructed by subtracting two close Planck frequency channel maps, after first smoothing them to a common resolution to ensure that the CMB signal is properly removed. A linear combination of the templates is then subtracted from (hitherto unused) map d to produce a clean CMB map at that frequency. This is done in real space at each position on the sky: where is the number of templates. The coefficients are obtained by minimising the variance of the clean map outside a given mask. Note that the same expression applies for I, Q and U. Although we exclude very contaminated regions during the minimization, the subtraction is performed for all pixels and, therefore, the cleaned maps cover the fullsky (although we expect that foreground residuals are present in the excluded areas).
There are several possible configurations of SEVEM with regard to the number of frequency maps which are cleaned or the number of templates that are used in the fitting. Note that the production of clean maps at different frequencies is of great interest in order to test the robustness of the results, and these intermediate products (clean maps at individual frequencies for intensity and polarization) are also provided in the archive. Therefore, to define the best strategy, one needs to find a compromise between the number of maps that can be cleaned independently and the number of templates that can be constructed.
 Intensity
For the CMB intensity map, we have cleaned the 100 GHz, 143 GHz and 217 GHz maps using a total of four templates. Three of them are constructed as the difference of two consecutive Planck channels smoothed to a common resolution (3044, 4470 and 545353) while the 857 GHz channel is chosen as the fourth template. First of all, the six frequency channels which are going to be part of the templates are inpainted at the point source positions detected using the Mexican Hat Wavelet algorithm. The size of the holes to be inpainted is determined taking into account the beam size of the channel as well as the flux of each source. The inpainting algorithm is based on simple diffuse inpainting, which fills one pixel with the mean value of the neighbouring pixels in an iterative way. To avoid inconsistencies when subtracting two channels, each frequency map is inpainted on the sources detected in that map and on the second map (if any) used to construct the template. Then the maps are smoothed to a common resolution (the first channel in the subtraction is smoothed with the beam of the second map and viceversa). For the 857 GHz template, we simply filter the inpainted map with the 545 GHz beam.
The coefficients are obtained by minimising the variance outside the analysis mask, that covers the 1 per cent brightest emission of the sky as well as point sources detected at all frequency channels. Once the maps are cleaned, each of them is inpainted on the point sources positions detected at that (raw) channel. Then, the MHW algorithm is run again, now on the clean maps. A relatively small number of new sources are found and are also inpainted at each channel. The resolution of the clean map is the same as that of the original data. Our final CMB map is then constructed by combining the 143 and 217 GHz maps by weighting the maps in harmonic space taking into account the noise level, the resolution and a rough estimation of the foreground residuals of each map (obtained from realistic simulations). This final map has a resolution corresponding to a Gaussian beam of fwhm=5 arcminutes.
In addition, the clean CMB maps produced at 100, 143 and 217 GHz frequencies are also provided. The resolution of these maps is the same as that of the uncleaned frequency channels and have been constructed at Nside=2048. They have been inpainted at the position of the detected point sources. Note that these three clean maps should be close to independent, although some level of correlation will be present since the same templates have been used to clean the maps.
The confidence mask is produced by studying the differences between several SEVEM CMB reconstructions, which correspond to maps cleaned at different frequencies or using different analysis masks. The obtained mask leaves a useful sky fraction of approximately 85 per cent.
 Polarization
To clean the polarization maps, a procedure similar to the one used for intensity data is applied to the Q and U maps independently. However, given that a narrower frequency coverage is available for polarization, the selected templates and maps to be cleaned are different. In particular, we clean the 70, 100 and 143 GHz using three templates for each channel. The first step of the pipeline is to inpaint the positions of the point sources using the MHW, in those channels which are going to be used in the construction of templates, following the same procedure as for the intensity case. The inpainting is performed in the frequency maps at their native resolution. These inpainted maps are then used to construct a total of four templates. To trace the synchrotron emission, we construct a template as the subtraction of the 30 GHz minus the 44 GHz map, after being convolved with the beam of each other. For the dust emission, the following templates are considered: 353217 GHz (smoothed at 10' resolution), 217143 GHz (used to clean 70 and 100 GHz) and 217100 GHz (to clean 143 GHz). These two last templates are constructed at 1 degree resolution since an additional smoothing becomes necessary in order to increase the signaltonoise ratio of the template. Conversely to the intensity case and due to the lower availability of frequency channels, it becomes necessary to use the maps to be cleaned as part of one of the templates. In this way, the 100 GHz map is used to clean the 143 GHz frequency channel and viceversa, making the clean maps less independent between them than in the intensity case.
These templates are then used to clean the noninpainted 70 (at its native resolution), 100 (at 10' resolution) and 143 GHz maps (also at 10'). The corresponding linear coefficients are estimated independently for Q and U by minimising the variance of the clean maps outside a mask, that covers point sources and the 3 per cent brightest Galactic emission. Once the maps have been cleaned, inpainting of the point sources detected at the corresponding raw maps is carried out. The size of the holes to be inpainted takes into account the additional smoothing of the 100 and 143 GHz maps. The 100 and 143 GHz clean maps are then combined in harmonic space, using E and B decomposition, to produce the final CMB maps for the Q and U components at a resolution of 10' (Gaussian beam) for a HEALPix parameter Nside=1024. Each map is weighted taking into account its corresponding noise level at each multipole. Finally, before applying the postprocessing HPF to the clean polarization data, the region with the brightest Galactic residuals is inpainted (5 per cent of the sky) to avoid the introduction of ringing around the Galactic centre in the filtering process.
The clean CMB maps at individual frequency channels produced as intermediate steps of SEVEM are also provided for Q and U, constructed at Nside=1024. The clean 70 GHz map is provided at its native resolution, while the clean maps at 100 and 143 GHz frequencies have a resolution of 10 arcminutes (Gaussian beam). The three maps have been inpainted in the positions of the detected point sources. Note that, due to the availability of a smaller number of templates for polarization than for intensity, these maps are less independent than for the temperature case, since, for instance, the 100 GHz map is used to clean the 143 GHz one and viceversa.
The confidence mask includes all the pixels above a given threshold in a smoothed version of the clean CMB map, the regions more contaminated by the CO emission and those pixels more affected by the highpass filtering, leaving a useful sky fraction of approximately 80 per cent.
SMICA
A) Production of the intensity map.
 1) Preprocessing
 Before computing spherical harmonic coefficients, all input maps undergo a preprocessing step to deal with regions of very strong emission (such as the Galactic center) and point sources. The point sources with SNR > 5 in the PCCS catalogue are fitted in each input map. If the fit is successful, the fitted point source is removed from the map; otherwise it is masked and the hole is filled in by a simple diffusive process to ensure a smooth transition and mitigate spectral leakage. The diffusive inpainting process is also applied to some regions of very strong emissions. This is done at all frequencies but 545 and 857 GHz, here all point sources with SNR > 7.5 are masked and filledin similarly.
 2) Linear combination
 The nine preprocessed Planck frequency channels from 30 to 857 GHz are harmonically transformed up to and coadded with multipoledependent weights as shown in the figure.
 3) Postprocessing
 A confidence mask is determined (see the Planck paper) and all regions which have been masked in the preprocessing step are added to it.
B) Production of the Q and U polarisation maps.
The SMICA pipeline for polarization uses all the 7 polarized Planck channels. The production of the Q and U maps is similar to the production of the intensity map. However, there is no input point source preprocessing of the input maps. The regions of very strong emission are masked out using an apodized mask before computing the E and B modes of the input maps and combining them to produce the E and B modes of the CMB map. Those modes are then used to synthesize the U and Q CMB maps. The E and B parts of the input frequency maps being processed jointly, there are, at each multipole, 2*7=14 coefficients (weights) defined to produce the E modes of the CMB map and as many to produce the B part. The weights are displayed in the figure below. The Q and U maps were originally produced at Nside=2048 with a 5arcminute resolution, but were downgraded to Nside=1024 with a 10 arcminute resolution for this release.
Masks
Summary table with the different masks that have been used by the component separation methods to preprocess and to process the frequency maps and the CMB maps.
Commander 2015 (PR2)  Used for diffuse inpainting of input frequency maps  Used for constrained Gaussian realization inpaiting of CMB map  Description 

T_MASK  NO  NO  T_MASK (the equivalent to PR1 VALMASK) is the confidence mask in temperature that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUcommanderfieldInt_2048_R2.01_full.fits. 
P_MASK  NO  NO  P_MASK is the confidence mask in polarization that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUcommander_1024_R2.02_full.fits. 
INP_MASK_T  NO  YES  Three masks have been used for inpaiting of CMB maps for specific ranges: three different angular resolution maps (40 arcmin, 7.5 arcmin and full resolution), are produced using different data combinations and foreground models. Each of these are inpainted with their own masks with a constrained Gaussian realization before coadding the three maps in harmonic space. 
INP_MASK_P  NO  YES  Mask used for inpainting of the CMB map in polarization. 
SEVEM 2015 (PR2)  Used for Diffuse Inpainting of foregorund subtracted CMB maps (fgsubsevem)  Used for constrained Gaussian realization inpaiting of CMB map  Description 
T_MASK  NO  NO  T_MASK (the equivalent to PR1 VALMASK) is the confidence mask in temperature that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUsevemfieldInt_2048_R2.01_full.fits. 
P_MASK  NO  NO  P_MASK is the confidence mask in polarization that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUsevem_1024_R2.02_full.fits. 
INP_MASK_T  YES  NO  Point source mask for temperature. This mask is the combination of the 143 and 217 T point source masks used for the inpainting of the foreground subtracted CMB maps at those two frequencies. These two maps have been combined to produce the final CMB map. 
INP_MASK_P  YES  NO  Point source mask for polarization. This mask is the combination of the 100 and 143 point source masks used for the inpainting of the foreground subtracted CMB maps at those two frequencies. These two maps have been combined to produce the final CMB map. 
INP_MASK_T for the cleaned 100, 143 and 217 GHz CMB  YES  NO  Three temperature point source masks used for the inpainting of the foreground subtracted CMB maps at the considered frequencies:

INP_MASK_P for the cleaned 70, 100 and 143 GHz CMB  YES  NO  Three polarization point source masks used for the inpainting of the foreground subtracted CMB maps at the considered frequencies: 
NILC 2015 (PR2)  Used for diffuse inpainting of input frequency maps  Used for constrained Gaussian realization inpaiting of CMB map  Description 
T_MASK  NO  NO  T_MASK (the equivalent to PR1 VALMASK) is the confidence mask in temperature that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUnilcfieldInt_2048_R2.01_full.fits. 
P_MASK  NO  NO  P_MASK is the confidence mask in polarization that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUnilc_1024_R2.02_full.fits. 
INP_MASK  YES  NO  The preprocessing involves inpainting of the holes in INP_MASK in the frequency maps prior to applying NILC on them. The first mask (nside 2048) has been used for the preprocessing of sky maps for HFI channels and second one for LFI channels (nside 1024). They can downloaded here:
COM_Mask_PointSrcGalplane_nilc_dx11_preproc_1024_R2.00.fits COM_Mask_PointSrcGalplane_nilc_dx11_preproc_2048_R2.00.fits 
SMICA 2015 (PR2)  Used for diffuse inpainting of input frequency maps  Used for constrained Gaussian realization inpaiting of CMB map  Description 
T_MASK  NO  YES  T_MASK (the equivalent to PR1 VALMASK) is the confidence mask in temperature that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUsmicafieldInt_2048_R2.01_full.fits. 
P_MASK  NO  YES  P_MASK is the confidence mask in polarization that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CMB_IQUsmica_1024_R2.02_full.fits. 
I_MASK  YES  NO  I_MASK, as in PR1, defines the regions over which CMB is not built. It is a combination of point source masks, Galactic plane mask and other bright regions like LMC, SMC, etc. It can downloaded here: COM_Mask_PointSrcGalplane_smica_harmonic_mask_2048_R2.00.fits 
Inputs
The input maps are the sky temperature maps described in the Sky temperature maps section. SMICA and SEVEM use all the maps between 30 and 857 GHz; NILC uses the ones between 44 and 857 GHz. CommanderRuler uses frequency channel maps from 30 to 353 GHz.
File names and structure
Three sets of files FITS files containing the CMB products are available. In the first set all maps (i.e., covering different parts of the mission) and all characterisation products for a given method and a given Stokes parameter are grouped into a single extension, and there are two files per method (smica, nilc, sevem, and commander), one for the high resolution data (I only, Nside=2048) and one for low resolution data (Q and U only, Nside=1024). Each file also contains the associated confidence mask(s) and beam transfer function. These are the R2.00 files which have names like
 COM_CMB_IQU{method}field{Int,Pol}_Nside_R2.00.fits
There are 7 coverage periods:full, halfyear1,2, halfmission1,2, or ringhalf1,2, and 4 characterisation products: halfsum and halfdifference for the year and the halfmission periods.
In the second second set the different coverages are split into different files which in most cases have a single extension with I only (Nside=1024) and I, Q, and U (Nside=1024). This second set was built in order to allow users to use standard codes like spice or anafast on them, directly. So this set contains the I maps at Nside=1024, which are not contained in the R2.00; on the other hand this set does not contain the halfsum and halfdifference maps. These are the 2.01 files which have names like
 COM_CMB_IQU{method}{fieldIntPol}_Nside_R2.01_{coverage}.fits for the regular CMB maps, and
 COM_CMB_IQU{fff}{fgsubsevem}{fieldIntPol}_Nside_R2.01_{coverage}.fits for the sevem frequencydependent, foregroundssubtracted maps,
where fieldIntPol is used to indicate that only Int or only Pol data are contained (at present only fieldInt is used for the highres data), and is not included in the lowres data which contains all three Stokes parameters, and coverage is one of full, halfyear1,2, halfmission1,2, or ringhalf1,2. Also, the coverage=full files contain also the confidence mask(s) and beam transfer function(s) which are valid for all products of the same method (one for Int and one for Pol when both are available).
The third set has the same structure as the Nside=1024 products of R2.01, but the Q and U maps have been highpass filtered to remove modes at l < 30 for the reasons indicated earlier. These are the default products for use in polarisation studies. They are the R2.02 files which have names like:
 COM_CMB_IQU{method}_1024_R2.02_{coverage}.fits
Version 2.00 files
These have names like
 COM_CMB_IQU{method}field{Int,Pol}_Nside_R2.00.fits,
as indicated above. They contain:
 a minimal primary extension with no data;
 one or two BINTABLE data extensions with a table of Npix lines by 14 columns in which the first 13 columns is a CMB maps produced from the full or a subset of the data, as described in the table below, and the last column in a confidence mask. There is a single extension for Int files, and two, for Q and U, for Pol files.
 a BINTABLE extension containing the beam transfer function (mistakenly called window function in the files).
If Nside=1024 the files contain I, Q and U maps, whereas if Nside=2048 only the I map is given.
Ext. 1. or 2. EXTNAME = COMPMAP (BINTABLE)  

Column Name  Data Type  Units  Description 
I or Q or U  Real*4  uK_cmb  I or Q or U map 
HM1  Real*4  uK_cmb  Halfmiss 1 
HM2  Real*4  uK_cmb  Halfmiss 2 
YR1  Real*4  uK_cmb  Year 1 
YR2  Real*4  uK_cmb  Year 2 
HR1  Real*4  uK_cmb  Halfring 1 
HR2  Real*4  uK_cmb  Halfring 2 
HMHS  Real*4  uK_cmb  Halfmiss, half sum 
HMHD  Real*4  uK_cmb  Halfmiss, half diff 
YRHS  Real*4  uK_cmb  Year, half sum 
YRHD  Real*4  uK_cmb  Year, half diff 
HRHS  Real*4  uK_cmb  Halfring half sum 
HRHD  Real*4  uK_cmb  Halfring half diff 
MASK  BYTE  Confidence mask  
Keyword  Data Type  Value  Description 
ASTCOMP  String  CMB  Astrophysical compoment name 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
POLCCONV  String  COSMO  Polarization convention 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  1024 or 2048  Healpix Nside 
METHOD  String  name  Cleaning method (smica/nilc/sevem/commander) 
Ext. 2. or 3. EXTNAME = BEAM_WF (BINTABLE) . See Note 1  
Column Name  Data Type  Units  Description 
BEAMWF  Real*4  none  The effective beam transfer function, including the pixel window function. See Note 2. 
Keyword  Data Type  Value  Description 
LMIN  Int  value  First multipole of beam TF 
LMAX  Int  value  Last multipole of beam TF 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM/COMMANDERRuler) 
Notes:
 Actually this is a beam transfer function, so BEAM_TF would have been more appropriate.
 The beam transfer function given here includes the pixel window function for the Nside=2048 pixelization. It means that, ideally, . The beam Window function is given by
Version 2.01 files
These files have names like:
 COM_CMB_IQU{method}{fieldIntPol}_Nside_R2.01_{coverage}.fits for the regular CMB maps, and
 COM_CMB_IQU{fff}{fgsubsevem}{fieldIntPol}_Nside_R2.01_{coverage}.fits for the sevem frequencydependent, foregroundssubtracted maps,
as indicated above. They contain:
 a minimal primary extension with no data, but with a NUMEXT keyword giving the number of extensions contained.
 one or two BINTABLE data extensions with a table of Npix lines by 15 columns depending on the file, as described above: the minimum begin I only, the maximum begin I, Q, U, and confidence masks for I and P.
 a BINTABLE extension containing the beam transfer function(s): one for I, and a second one that applies to both Q and U, if Nslde=1024.
If Nside=1024 the files contain I, Q and U maps, whereas if Nside=2048 only the I map is given. The basic structure, including information on the most important keywords, is given in the table below. For full details, see the FITS header.
Ext. 1. or 2. EXTNAME = COMPMAP (BINTABLE)  

Column Name  Data Type  Units  Description 
I_Stokes  Real*4  uK_cmb  I map (Nside=1024,2048) 
Q_Stokes  Real*4  uK_cmb  Q map (Nside=1024) 
U_Stokes  Real*4  uK_cmb  U map (Nside=2048) 
TMASK  Int  none  optional Temperature confidence mask 
PMASK  Int  none  optional Polarisation confidence mask 
Keyword  Data Type  Value  Description 
ASTCOMP  String  CMB  Astrophysical compoment name 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
POLCCONV  String  COSMO  Polarization convention 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  1024 or 2048  Healpix Nside 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM) 
Optional Ext. 2. or 3. EXTNAME = BEAM_TF (BINTABLE)  
Column Name  Data Type  Units  Description 
INT_BEAM  Real*4  none  Effective beam transfer function. See Note 1. 
POL_BEAM  Real*4  none  Effective beam transfer function. See Note 1. 
Keyword  Data Type  Value  Description 
LMIN  Int  value  First multipole of beam WF 
LMAX_I  Int  value  Last multipole for Int beam TF 
LMAX_P  Int  value  Last multipole for Pol beam TF 
METHOD  String  name  Cleaning method 
Notes:
 The beam transfer function given here includes the pixel window function for the Nside=2048 pixelization. It means that, ideally, . The beam Window function is given by
Version 2.02 files
For polarisation work, this is the default set of files to be used for cosmological analysis. Their content is identical to the "R2.01" files, except that angular scales above l < 30 have been filtered out of the Q and U maps.
These files have names like:
 COM_CMB_IQU{method}_1024_R2.02_{coverage}.fits
as indicated above. They contain: The files contain
 a minimal primary extension with no data, but with a NUMEXT keyword giving the number of extensions contained.
 one or two BINTABLE data extensions with a table of Npix lines by 15 columns depending on the file, as described above: the minimum begin I only, the maximum begin I, Q, U, and confidence masks for I and P.
 a BINTABLE extension containing 2 beam transfer functions: one for I and one that applies to both Q and U.
If Nside=1024 the files contain I, Q and U maps, whereas if Nside=2048 only the I map is given. The basic structure, including information on the most important keywords, is given in the table below. For full details, see the FITS header.
Ext. 1. or 2. EXTNAME = COMPMAP (BINTABLE)  

Column Name  Data Type  Units  Description 
I_Stokes  Real*4  uK_cmb  I map (Nside=1024) 
Q_Stokes  Real*4  uK_cmb  Q map (Nside=1024) 
U_Stokes  Real*4  uK_cmb  U map (Nside=2048) 
TMASK  Int  none  optional Temperature confidence mask 
PMASK  Int  none  optional Polarisation confidence mask 
Keyword  Data Type  Value  Description 
ASTCOMP  String  CMB  Astrophysical compoment name 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
POLCCONV  String  COSMO  Polarization convention 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  1024 or 2048  Healpix Nside 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM) 
Optional Ext. 2. or 3. EXTNAME = BEAM_TF (BINTABLE)  
Column Name  Data Type  Units  Description 
INT_BEAM  Real*4  none  Effective beam transfer function. See Note 1. 
POL_BEAM  Real*4  none  Effective beam transfer function. See Note 1. 
Keyword  Data Type  Value  Description 
LMIN  Int  value  First multipole of beam WF 
LMAX_I  Int  value  Last multipole for Int beam TF 
LMAX_P  Int  value  Last multipole for Pol beam TF 
METHOD  String  name  Cleaning method 
Notes:
 The beam transfer function given here includes the pixel window function for the Nside=2048 pixelization. It means that, ideally, . The beam Window function is given by
Common masks
The common masks are stored into two different files for Temperature and Polarisation respectively:
 COM_CMB_IQUcommonfieldMaskInt_2048_R2.nn.fits with the UT78 and UTA76 masks
 COM_CMB_IQUcommonfieldMaskPol_1024_R2.nn.fits with the UP78, UPA77, and UPB77 masks
Both files contain also a map of the missing pixels for the half mission and year coverage periods. The 2 (for Temp) or 3 (for Pol) masks and the missing pixels maps are stored in 4 or 5 column a BINTABLE extension 1 of each file, named MASKINT and MASKPOL, respectively. See the FITS file headers for details.
Quadrupole residual maps
The quadrupole residual maps are stored in files called:
 COM_CMB_IQUkqresid{method}fieldInt_2048_R2.02.fits
They contain:
 a minimal primary extension with no data, but with a NUMEXT keyword giving the number of extensions contained.
 a single BINTABLE extension with a single column of Npix lines containing the HEALPIX map indicated
The basic structure of the data extension is shown below. For full details see the extension header.
Ext. 1. EXTNAME = COMPMAP (BINTABLE)  

Column Name  Data Type  Units  Description 
INTENSITY  Real*4  K_cmb  the residual map 
Keyword  Data Type  Value  Description 
ASTCOMP  String  KQRESID  Astrophysical compoment name 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
POLCCONV  String  COSMO  Polarization convention 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  2048  Healpix Nside 
METHOD  String  name  Cleaning method 
2013 Release of CMB maps
CMB Maps
CMB maps have been produced by the SMICA, NILC, SEVEM and COMMANDERRuler pipelines. Of these, the SMICA product is considered the preferred one overall and is labelled Main product in the Planck Legacy Archive, while the other two are labeled as Additional product.
SMICA and NILC also produce inpainted maps, in which the Galactic Plane, some bright regions and masked point sources are replaced with a constrained CMB realization such that the whole map has the same statistical distribution as the observed CMB.
The results of SMICA, NILC and SEVEM pipeline are distributed as a FITS file containing 4 extensions:
 CMB maps and ancillary products (3 or 6 maps)
 CMBcleaned foreground maps from LFI (3 maps)
 CMBcleaned foreground maps from HFI (6 maps)
 Effective beam of the CMB maps (1 vector)
The results of COMMANDERRuler are distributed as two FITS files (the high and low resolution) containing the following extensions: High resolution N$_\rm{side}$=2048 (note that we don't provide the CMBcleaned foregrounds maps for LFI and HFI because the Ruler resolution (~7.4') is lower than the HFI highest channel and and downgrading it will introduce noise correlation).
 CMB maps and ancillary products (4 maps)
 Effective beam of the CMB maps (1 vector)
Low resolution N$_\rm{side}$=256
 CMB maps and ancillary products (3 maps)
 10 example CMB maps used in the montecarlo realization (10 maps)
 Effective beam of the CMB maps (1 vector)
For a complete description of the data structure, see the below; the content of the first extensions is illustrated and commented in the table below.
Col name  SMICA  NILC  SEVEM  COMMANDERRuler H  COMMANDERRuler L  Description / notes 

1: I  Raw CMB anisotropy map. These are the maps used in the component separation paper Planck2013XII^{[2]}.  
2: NOISE  not applicable  Noise map. Obtained by propagating the halfring noise through the CMB cleaning pipelines.  
3: VALMASK  Confidence map. Pixels with an expected low level of foreground contamination. These maps are only indicative and obtained by different ad hoc methods. They cannot be used to rank the CMB maps.  
4: I_MASK  not applicable  not applicable  not applicable  Some areas are masked for the production of the raw CMB maps (for NILC: point sources from 44 GHz to 857 GHz; for SMICA: point sources from 30 GHz to 857 GHz, Galatic region and additional bright regions).  
5: INP_CMB  not applicable  not applicable  not applicable  Inpainted CMB map. The raw CMB maps with some regions (as indicated by INP_MASK) replaced by a constrained Gaussian realization. The inpainted SMICA map was used for PR.  
6: INP_MASK  not applicable  not applicable  not applicable  Mask of the inpainted regions. For SMICA, this is identical to I_MASK. For NILC, it is not. 
The component separation pipelines are described in the CMB and foreground separation section and also in Section 3 and Appendices AD of Planck2013XII^{[2]} and references therein.
The union (or common) mask is defined as the union of the confidence masks from the four component separation pipelines, the three listed above and CommanderRuler. It leaves 73% of the sky available, and so it is denoted as U73.
Product description
SMICA
 Principle
 SMICA produces a CMB map by linearly combining all Planck input channels (from 30 to 857 GHz) with weights which vary with the multipole. It includes multipoles up to .
 Resolution (effective beam)
 The SMICA map has an effective beam window function of 5 arcminutes truncated at CMB and is a 5arcminute Gaussian beam function. Note however that, by convention, the effective beam window function provided in the FITS file does include a pixel window function. Therefore, it is equal to where denotes the pixel window function for an Nside=2048 pixelization. and deconvolved from the pixel window. It means that, ideally, one would have , where is the angular spectrum of the map, where is the angular spectrum of the
 Confidence mask
 A confidence mask is provided which excludes some parts of the Galactic plane, some very bright areas and the masked point sources. This mask provides a qualitative (and subjective) indication of the cleanliness of a pixel.
 Masks and inpainting
 The raw SMICA CMB map has valid pixels except at the location of masked areas: point sources, Galactic plane, some other bright regions. Those invalid pixels are indicated with the mask named 'I_MASK'. The raw SMICA map has been inpainted, producing the map named "INP_CMB". Inpainting consists in replacing some pixels (as indicated by the mask named INP_MASK) by the values of a constrained Gaussian realization which is computed to ensure good statistical properties of the whole map (technically, the inpainted pixels are a sample realisation drawn under the posterior distribution given the unmasked pixels.
NILC
 Principle
 The NeedletILC (hereafter NILC) CMB map is constructed from all Planck channels from 44 to 857 GHz and includes multipoles up to . It is obtained by applying the Internal Linear Combination (ILC) technique in needlet space, that is, with combination weights which are allowed to vary over the sky and over the whole multipole range.
 Resolution (effective beam)
 As in the SMICA product except that there is no abrupt truncation at but a smooth transition to over the range .
 Confidence mask
 A confidence mask is provided which excludes some parts of the Galactic plane, some very bright areas and the masked point sources. This mask provides a qualitative indication of the cleanliness of a pixel. The threshold is somewhat arbitrary.
 Masks and inpainting
 The raw NILC map has valid pixels except at the location of masked point sources. This is indicated with the mask named 'I_MASK'. The raw NILC map has been inpainted, producing the map named "INP_CMB". The inpainting consists in replacing some pixels (as indicated by the mask named INP_MASK) by the values of a constrained Gaussian realization which is computed to ensure good statistical properties of the whole map (technically, the inpainted pixels are a sample realisation drawn under the posterior distribution given the unmasked pixels.
SEVEM
The aim of SEVEM is to produce clean CMB maps at one or several frequencies by using a procedure based on template fitting. The templates are internal, i.e., they are constructed from Planck data, avoiding the need for external data sets, which usually complicates the analyses and may introduce inconsistencies. The method has been successfully applied to Planck simulations^{[8]} and to WMAP polarisation data^{[9]}. In the cleaning process, no assumptions about the foregrounds or noise levels are needed, rendering the technique very robust. Note that unlike the other products, SEVEM does not provide the mask of regions not used in the productions of the CMB ma (I_MASK) nor an inpainted version of the map and its associated mask. On the other hand, it provides channel maps and 100, 143, and 217 GHz which are used as the building blocks of the final map.
COMMANDERRuler
COMMANDERRuler is the Planck software implementing a pixel based parametric component separation. Amplitude of CMB and the main diffuse foregrounds along with the relevant spectral parameters for those (see below in the Astrophysical Foreground Section for the latter) are parametrized and fitted in single MCMC chains conducted at $N_\rm{side}$=256 using COMMANDER, implementing a Gibbs Sampling. The CMB amplitude which is obtained in these runs corresponds to the delivered low resolution CMB component from COMMANDERRuler which has a FWHM of 40 arcminutes. The sampling of the foreground parameters is applied to the data at full resolution for obtaining the high resolution CMB component from Ruler which is available on the PLA. In the Planck Component Separation paper^{[2]} additional material is discussed, specifically concerning the sky region where the solutions are reliable, in terms of chi2 maps. The products mainly consist of:
 Maps of the Amplitudes of the CMB at low resolution, $N_\rm{side}$=256, along with the standard deviations of the outputs, beam profiles derived from the production process.
 Maps of the CMB amplitude, along with the standard deviations, at high resolution, $N_\rm{side}$=2048, beam profiles derived from the production process.
 Mask obtained on the basis of the precision in the fitting procedure; the thresholding is evaluated through the COMMANDERRuler likelihood analysis and excludes 13% of the sky, see Planck2013XII^{[2]}.
Production process
SMICA
 1) Preprocessing
 All input maps undergo a preprocessing step to deal with point sources. The point sources with SNR > 5 in the PCCS catalogue are fitted in each input map. If the fit is successful, the fitted point source is removed from the map; otherwise it is masked and the hole is filled in by a simple diffusive process to ensure a smooth transition and mitigate spectral leakage. This is done at all frequencies but 545 and 857 GHz, here all point sources with SNR > 7.5 are masked and filledin similarly.
 2) Linear combination
 The nine preprocessed Planck frequency channels from 30 to 857 GHzare harmonically transformed up to and coadded with multipoledependent weights as shown in the figure.
 3) Postprocessing
 The areas masked in the preprocessing step are replaced by a constrained Gaussian realization.
Note: The visible power deficit in the raw CMB map around the galactic plane is due to the smooth fillin of the masked areas in the input maps (the result of the preprocessing). It is not to be confused with the postprocessing step of inpainting of the CMB map with a constrained Gaussian realization.
NILC
 1) Preprocessing
 Same preprocessing as SMICA (except the 30 GHz channel is not used).
 2) Linear combination
 The preprocessed Planck frequency channels from 44 to 857 GHz are linearly combined with weights which depend on location on the sky and on the multipole range up to Planck2013XII^{[2]}. . This is achieved using a needlet (redundant spherical wavelet) decomposition. For more details, see
 3) Postprocessing
 The areas masked in the preprocessing plus other bright regions step are replaced by a constrained Gaussian realization as in the SMICA postprocessing step.
SEVEM
The templates are internal, i.e., they are constructed from Planck data, avoiding the need for external data sets, which usually complicates the analyses and may introduce inconsistencies. In the cleaning process, no assumptions about the foregrounds or noise levels are needed, rendering the technique very robust. The fitting can be done in real or wavelet space (using a fast wavelet adapted to the HEALPix pixelization^{[10]}) to properly deal with incomplete sky coverage. By expediency, however, we fill in the small number of unobserved pixels at each channel with the mean value of its neighbouring pixels before applying SEVEM.
We construct our templates by subtracting two close Planck frequency channel maps, after first smoothing them to a common resolution to ensure that the CMB signal is properly removed. A linear combination of the templates is then subtracted from (hitherto unused) map d to produce a clean CMB map at that frequency. This is done either in real or in wavelet space (i.e., scale by scale) at each position on the sky: where is the number of templates. If the cleaning is performed in real space, the coefficients are obtained by minimising the variance of the clean map outside a given mask. When working in wavelet space, the cleaning is done in the same way at each wavelet scale independently (i.e., the linear coefficients depend on the scale). Although we exclude very contaminated regions during the minimization, the subtraction is performed for all pixels and, therefore, the cleaned maps cover the fullsky (although we expect that foreground residuals are present in the excluded areas).
An additional level of flexibility can also be considered: the linear coefficients can be the same for all the sky, or several regions with different sets of coefficients can be considered. The regions are then combined in a smooth way, by weighting the pixels at the boundaries, to avoid discontinuities in the clean maps. Since the method is linear, we may easily propagate the noise properties to the final CMB map. Moreover, it is very fast and permits the generation of thousands of simulations to character ize the statistical properties of the outputs, a critical need for many cosmological applications. The final CMB map retains the angular resolution of the original frequency map.
There are several possible configurations of SEVEM with regard to the number of frequency maps which are cleaned or the number of templates that are used in the fitting. Note that the production of clean maps at different frequencies is of great interest in order to test the robustness of the results. Therefore, to define the best strategy, one needs to find a compromise between the number of maps that can be cleaned independently and the number of templates that can be constructed.
In particular, we have cleaned the 143 GHz and 217 GHz maps using four templates constructed as the difference of the following Planck channels (smoothed to a common resolution): (3044), (4470), (545353) and (857545). For simplicity, the three maps have been cleaned in real space, since there was not a significant improvement when using wavelets (especially at high latitude). In order to take into account the different spectral behaviour of the foregrounds at low and high galactic latitudes, we have considered two independent regions of the sky, for which we have used a different set of coefficients. The first region corresponds to the 3 per cent brightest Galactic emission, whereas the second region is defined by the remaining 97 per cent of the sky. For the first region, the coefficients are actually estimated over the whole sky (we find that this is more optimal than perform the minimisation only on the 3 per cent brightest region, where the CMB emission is very subdominant) while for the second region, we exclude the 3 per cent brightest region of the sky, point sources detected at any frequency and those pixels which have not been observed at all channels. Our final CMB map has then been constructed by combining the 143 and 217 GHz maps by weighting the maps in harmonic space taking into account the noise level, the resolution and a rough estimation of the foreground residuals of each map (obtained from realistic simulations). This final map has a resolution corresponding to a Gaussian beam of fwhm=5 arcminutes.
Moreover, additional CMB clean maps (at frequencies between 44 and 353 GHz) have also been produced using different combinations of templates for some of the analyses carried out in Planck2013XXIII^{[11]} and Planck2013XIX^{[12]}. In particular, clean maps from 44 to 353 GHz have been used for the stacking analysis presented in Planck2013XIX^{[12]}, while frequencies from 70 to 217 GHz were used for consistency tests in Planck2013XXIII^{[11]}.
COMMANDERRuler
The production process consist in low and high resolution runs according to the description above.
 Low Resolution Runs
 Same as the Astrophysics Foregrounds Section below; The CMB amplitude is fitted along with the other foreground parameters and constitutes the CMB Low Resolution Rendering which is in the PLA.
 Ruler Runs
 the sampling at high resolution is used to infer the probability distribution of spectral parameters which is exploited at full resolution in order to obtain the High Resolution CMB Rendering which is in the PLA.
Masks
Summary table with the different masks that have been used by the component separation methods to preprocess and to process the frequency maps and the CMB maps.
Commander 2013 (PR1)  Used for diffuse inpainting of input frequency maps  Used for constrained Gaussian realization inpaiting of CMB map  Description 

VALMASK  NO  NO  VALMASK is the confidence mask that defines the region where the reconstructed CMB is trusted. It can be found inside
COM_CompMap_CMBcommrul_2048_R1.00.fits and COM_CompMap_CMBcommrul_0256_R1.00.fits for low resolution analyses. 
SEVEM 2013 (PR1)  Used diffuse inpainting of input frequency maps  Used for Constrained Gaussian realization inpaiting of CMB map  Description 
VALMASK  NO  NO  VALMASK is the confidence mask that defines the region where the reconstructed CMB is trusted. It can be found inside 
NILC 2013 (PR1)  Used for diffuse inpainting of input frequency maps  Used for constrained Gaussian realization inpaiting of CMB map  Description 
VALMASK  NO  NO  VALMASK is the confidence mask that defines the region where the reconstructed CMB is trusted. It can be found inside COM_CompMap_CMBnilc_2048_R1.20.fits. 
I_MASK  NO  NO  I_MASK defines the regions over which CMB is not built. It is a combination of point source masks, Galactic plane mask and other bright regions like LMC, SMC, etc. It can be found inside COM_CompMap_CMBnilc_2048_R1.20.fits. 
INP_MASK  NO  YES  It can be found inside COM_CompMap_CMBnilc_2048_R1.20.fits. 
SMICA 2013 (PR1)  Used for diffuse inpainting of input frequency maps  Used for constrained Gaussian realization inpaiting of CMB map  Description 
VALMASK  NO  NO  VALMASK is the confidence mask that defines the region where the reconstructed CMB is trusted. It can be found inside 
I_MASK  YES  YES  I_MASK defines the regions over which CMB is not built. It is a combination of point source masks, Galactic plane mask and other bright regions like LMC, SMC, etc. It can be found inside COM_CompMap_CMBsmica_2048_R1.20.fits. 
INP_MASK  YES  YES  INP_MASK for SMICA 2013 release is identical to I_MASK above. 
Inputs
The input maps are the sky temperature maps described in the Sky temperature maps section. SMICA and SEVEM use all the maps between 30 and 857 GHz; NILC uses the ones between 44 and 857 GHz. CommanderRuler uses frequency channel maps from 30 to 353 GHz.
File names and structure
The FITS files corresponding to the three CMB products are the following:
 COM_CompMap_CMBnilc_2048_R1.20.fits
 COM_CompMap_CMBsevem_2048_R1.12.fits
 COM_CompMap_CMBsmica_2048_R1.20.fits
 COM_CompMap_CMBcommrul_2048_R1.00.fits
 COM_CompMap_CMBcommrul_0256_R1.00.fits
The files contain a minimal primary extension with no data and four BINTABLE data extensions. Each column of the BINTABLE is a (Healpix) map; the column names and the most important keywords of each extension are described in the table below; for the remaining keywords, please see the FITS files directly.
Ext. 1. EXTNAME = COMPMAP (BINTABLE)  

Column Name  Data Type  Units  Description 
I  Real*4  uK_cmb  CMB temperature map 
NOISE  Real*4  uK_cmb  Estimated noise map (note 1) 
I_STDEV  Real*4  uK_cmb  Standard deviation, ONLY on COMMANDERRuler products 
VALMASK  Byte  none  Confidence mask (note 2) 
I_MASK  Byte  none  Mask of regions over which CMB map is not built (Optional  see note 3) 
INP_CMB  Real*4  uK_cmb  Inpainted CMB temperature map (Optional  see note 3) 
INP_MASK  Byte  none  mask of inpainted pixels (Optional  see note 3) 
Keyword  Data Type  Value  Description 
ASTCOMP  String  CMB  Astrophysical compoment name 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  2048  Healpix Nside 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM/COMMANDERRuler) 
Ext. 2. EXTNAME = FGDSLFI (BINTABLE)  Note 4  
Column Name  Data Type  Units  Description 
LFI_030  Real*4  K_cmb  30 GHz foregrounds 
LFI_044  Real*4  K_cmb  44 GHz foregrounds 
LFI_070  Real*4  K_cmb  70 GHz foregrounds 
Keyword  Data Type  Value  Description 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  1024  Healpix Nside 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM) 
Ext. 3. EXTNAME = FGDSHFI (BINTABLE)  Note 4  
Column Name  Data Type  Units  Description 
HFI_100  Real*4  K_cmb  100 GHz foregrounds 
HFI_143  Real*4  K_cmb  143 GHz foregrounds 
HFI_217  Real*4  K_cmb  217 GHz foregrounds 
HFI_353  Real*4  K_cmb  353 GHz foregrounds 
HFI_545  Real*4  MJy/sr  545 GHz foregrounds 
HFI_857  Real*4  MJy/sr  857 GHz foregrounds 
Keyword  Data Type  Value  Description 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  2048  Healpix Nside 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM/COMMANDERRuler) 
Ext. 4. EXTNAME = BEAM_WF (BINTABLE)  
Column Name  Data Type  Units  Description 
BEAM_WF  Real*4  none  The effective beam window function, including the pixel window function. See Note 5. 
Keyword  Data Type  Value  Description 
LMIN  Int  value  First multipole of beam WF 
LMAX  Int  value  Lsst multipole of beam WF 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM/COMMANDERRuler) 
Notes:
 The halfring halfdifference (HRHD) map is made by passing the halfring frequency maps independently through the component separation pipeline, then computing half their difference. It approximates a noise realisation, and gives an indication of the uncertainties due to instrumental noise in the corresponding CMB map.
 The confidence mask indicates where the CMB map is considered valid.
 This column is not present in the SEVEM and COMMANDERRuler product file. For SEVEM these three columns give the CMB channel maps at 100, 143, and 217 GHz (columns C100, C143, and C217, in units of K_cmb.
 The subtraction of the CMB from the sky maps in order to produce the foregrounds map is done after convolving the CMB map to the resolution of the given frequency. Those columns are not present in the COMMANDERRuler product file.
 The beam window function given here includes the pixel window function for the Nside=2048 pixelization. It means that, ideally, .
The low resolution COMMANDERRuler CMB product is organized in the following way:
Ext. 1. EXTNAME = COMPMAP (BINTABLE)  

Column Name  Data Type  Units  Description 
I  Real*4  uK_cmb  CMB temperature map obtained as average over 1000 samples 
I_stdev  Real*4  uK_cmb  Corresponding Standard deviation amongst the 1000 samples 
VALMASK  Byte  none  Confidence mask 
Keyword  Data Type  Value  Description 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  2048  Healpix Nside 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM/COMMANDERRuler) 
Ext. 2. EXTNAME = CMBSample (BINTABLE)  
Column Name  Data Type  Units  Description 
I_SIM01  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM02  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM03  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM04  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM05  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM06  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM07  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM08  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM09  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
I_SIM10  Real*4  K_cmb  CMB Sample, smoothed to 40 arcmin 
Keyword  Data Type  Value  Description 
PIXTYPE  String  HEALPIX  
COORDSYS  String  GALACTIC  Coordinate system 
ORDERING  String  NESTED  Healpix ordering 
NSIDE  Int  1024  Healpix Nside 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM/COMMANDERRuler) 
Ext. 4. EXTNAME = BEAM_WF (BINTABLE)  
Column Name  Data Type  Units  Description 
BEAM_WF  Real*4  none  The effective beam window function, including the pixel window function. 
Keyword  Data Type  Value  Description 
LMIN  Int  value  First multipole of beam WF 
LMAX  Int  value  Lsst multipole of beam WF 
METHOD  String  name  Cleaning method (SMICA/NILC/SEVEM/COMMANDERRuler) 
The FITS files containing the union (or common) maks is:
which contains a single BINTABLE extension with a single column (named U73) for the mask, which is boolean (FITS TFORM = B), in GALACTIC coordinates, NESTED ordering, and Nside=2048.
For the benefit of users who are only looking for a small file containing the SMICA cmb map with no additional information (noise or masks) we provide such a file here
This file contains a single extension with a single column containing the SMICA cmb temperature map.
Cautionary notes
 The halfring CMB maps are produced by the pipelines with parameters/weights fixed to the values obtained from the full maps. Therefore the CMB HRHD maps do not capture all of the uncertainties due to foreground modelling on large angular scales.
 The HRHD maps for the HFI frequency channels underestimate the noise power spectrum at high l by typically a few percent. This is caused by correlations induced in the preprocessing to remove cosmic ray hits. The CMB is mostly constrained by the HFI channels at high l, and so the CMB HRHD maps will inherit this deficiency in power.
 The beam transfer functions do not account for uncertainties in the beams of the frequency channel maps.
References[edit]
 ↑ ^{1.0}^{1.1}^{1.2}^{1.3}^{1.4}^{1.5}^{1.6}^{1.7} Planck 2018 results. IV. Diffuse component separation, Planck Collaboration, 2020, A&A, 641, A4.
 ↑ ^{2.0}^{2.1}^{2.2}^{2.3}^{2.4}^{2.5} Planck 2013 results. XI. Component separation, Planck Collaboration, 2014, A&A, 571, A11.
 ↑ ^{3.0}^{3.1}^{3.2} Planck 2015 results. XI. Diffuse component separation: CMB maps, Planck Collaboration, 2016, A&A, 594, A9.
 ↑ Planck 2015 results. I. Overview of products and results, Planck Collaboration, 2016, A&A, 594, A1.
 ↑ Planck 2015 results. VI. LFI mapmaking, Planck Collaboration, 2016, A&A, 594, A6.
 ↑ Planck 2015 results. VIII. High Frequency Instrument data processing: Calibration and maps, Planck Collaboration, 2016, A&A, 594, A8.
 ↑ ^{7.0}^{7.1} Planck 2015 results. X. Diffuse component separation: Foreground maps, Planck Collaboration, 2016, A&A, 594, A10.
 ↑ Component separation methods for the PLANCK mission, S. M. Leach, J.F. Cardoso, C. Baccigalupi, R. B. Barreiro, M. Betoule, J. Bobin, A. Bonaldi, J. Delabrouille, G. de Zotti, C. Dickinson, H. K. Eriksen, J. GonzálezNuevo, F. K. Hansen, D. Herranz, M. Le Jeune, M. LópezCaniego, E. MartínezGonzález, M. Massardi, J.B. Melin, M.A. MivilleDeschênes, G. Patanchon, S. Prunet, S. Ricciardi, E. Salerno, J. L. Sanz, J.L. Starck, F. Stivoli, V. Stolyarov, R. Stompor, P. Vielva, A&A, 491, 597615, (2008).
 ↑ Multiresolution internal template cleaning: an application to the Wilkinson Microwave Anisotropy Probe 7yr polarization data, R. FernándezCobos, P. Vielva, R. B. Barreiro, E. MartínezGonzález, MNRAS, 420, 21622169, (2012).
 ↑ Wilkinson Microwave Anisotropy Probe 7yr constraints on f_{NL} with a fast wavelet estimator, B. Casaponsa, R. B. Barreiro, A. Curto, E. MartínezGonzález, P. Vielva, MNRAS, 411, 20192025, (2011).
 ↑ ^{11.0}^{11.1} Planck 2013 results. XXIII. Isotropy and statistics of the CMB, Planck Collaboration, 2014, A&A, 571, A23.
 ↑ ^{12.0}^{12.1} Planck 2013 results. XIX. The integrated SachsWolfe effect, Planck Collaboration, 2014, A&A, 571, A19.
Cosmic Microwave background
Flexible Image Transfer Specification
SunyaevZel'dovich
FullWidthatHalfMaximum
(Hierarchical Equal Area isoLatitude Pixelation of a sphere, <ref name="Template:Gorski2005">HEALPix: A Framework for HighResolution Discretization and Fast Analysis of Data Distributed on the Sphere, K. M. Górski, E. Hivon, A. J. Banday, B. D. Wandelt, F. K. Hansen, M. Reinecke, M. Bartelmann, ApJ, 622, 759771, (2005).
(Planck) High Frequency Instrument
Planck Sky Model
(Planck) Low Frequency Instrument
Data Processing Center
Planck Legacy Archive