NASA - Jet Propulsion Laboratory
    + View the NASA Portal
Search JPL
Jet Propulsion Laboratory Home Earth Solar System Stars & Galaxies Technology
Introduction Background Software Links


This program performs harmonic analysis of the HEALPix maps up to a user specified maximum spherical harmonic order lmax. The integrals are computed on the whole sphere, unless the user chooses a provided option to excise from the input map(s) a simple, constant latitude, symmetric cut, and/or apply an arbitray cut read from an external file. Scalar, or scalar and tensor, spherical harmonic coefficients are evaluated from the map(s) if the input provides, respectively, only the temperature, or temperature and polarisation maps. The total operation count scales as ${\cal {O}}(N_{\rm pix}^{3/2} )$ with a prefactor depending on lmax.

Anafast reads one (two) file(s) containing the map(s) and produces a file containing the temperature auto- (or cross-) power spectrum CTl and, if requested, also the polarisation power spectra CEl, CBl, CT x El, CT x Bl and CE x Bl. The alm coefficients computed during the execution also can be written to a (two) file(s) if requested.

Anafast executes an approximate, discrete point-set quadrature on a sphere sampled at the HEALPix pixel centers. Spherical harmonic transforms are computed using recurrence relations for Legendre polynomials on co-latitude, $\theta$, and Fast Fourier Transforms on longitude, $\phi$.

Anafast is provided with an option to use precomputed Legendre Polynomials to decrease the total execution time.

Anafast permits two execution options which allow a significant improvement of accuracy of the approximate quadrature performed by this facility:

$\bullet$ An improved analysis using the provided ring weights, which correct the quadrature on latitude, and/or

$\bullet$ An iterative scheme using in succession several backward and forward harmonic transforms of the maps.

Location in HEALPix directory tree: src/f90/anafast/anafast.f90


Execution of anafast requires a user to specify the maximum spherical harmonic order lmax up to which the harmonic decomposition of the input maps will be performed. Since there are no formal limits on parameter lmax enforced by anafast, the user should make his/her choices judiciously. Hereafter it is convenient to specify lmax in terms of the HEALPix map resolution parameter nsmax.

If the function to be analysed is strictly band-width limited, or nearly band-width limited (as in the case of a Gaussian beam smoothed signal discretized at a rate of a few pixels per beam area), it is sufficient to run anafast with $l_{max} \approx 2\cdot nsmax$, with a very good Cl error performance already in the raw (i.e. uncorrected quadrature) harmonic transform mode. If quadrature corrections are still desired in this case, it should be sufficient to use, at no extra cost in execution time, the ring-weighted quadrature scheme. This is the recommended mode of operation of anafast for essentially error and worry free typical applications, e.g. CPU-intensive Monte Carlo studies.

If more aggressive attempts are undertaken to extract from a map the spectral coefficients at $l> 2\cdot nsmax$ (for example, as in a possible case of an attempt to analyse an existing map, which was irreversibly binned at a suboptimal resolution) the following should be kept in mind:

$\bullet$ Spherical harmonics discretized using HEALPix (either sampled at pixel centers, or avaraged over pixel areas) form a linearly independent system up to $l_{max} = 3 \cdot nsmax -1$. Hence, the functions which are strictly band-width limited to $l_{max} = 3 \cdot nsmax -1$ can be fully spectrally resolved with anafast, albeit with integration errors in the uncorrected quadrature mode, which grow up to $ \delta C_l\propto \epsilon \cdot C_l$, with $\epsilon <0.1$, at the highest values of l. These integration errors can be efficiently reduced using anafast in the iterative mode. Although this lmax range -- $2 \cdot nsmax < l_{max} < 3 \cdot nsmax - 1$ -- is easily manageable with anafast used on strictly band-width limited functions, it should be used with caution in basic and automated applications, e.g. Monte Carlo simulations.

$\bullet$ As with any discrete Fourier transform, anafast application to functions which are not band-width limited results with aliasing of power, which can not be remedied. If the particular case of interest may result in such a band-width violation (i.e. there is significant power in the function at $l> 3 \cdot nsmax -1$), the function should be smoothed before the application of anafast, or discretized and then analysed, on a refined HEALPix grid (with larger nsmax).

$\bullet$ REMEMBER: A peculiar property of the sphere, which usually surprizes those whose intuition is built on experience with FFTs on a segment, or on a Euclidean multidimensional domains, is the lack of a regular and uniform point-set at arbitrary resolution, and the resulting non-commutativity of the forward and backward discrete Fourier transforms on nearly-uniform point-sets, e.g. HEALPix . Hence, as in any case of attempting an extreme application of an off-the-shelf software, use caution and understand your problem well before executing anafast under such cicumstances!


% anafast [options] [parameter_file]


double precision mode (see Notes on double/single precision modes on page [*])
single precision mode (default)


infile =
Defines the input map file. (default= map.fits)
infile2 =
Defines the 2nd input map file, to be cross-correlated with the first one. The 2 maps should match in resolution (nsmax) and coordinate. (default= `', only the auto-correlation of the first map will be computed)
outfile =
Defines the output file with the power spectrum. If only one input map is provided, outfile will contains its auto-spectra, if 2 maps are provided, outfile will contain their cross-spectra. Note in particular that in the latter case, the CT x El power spectrum will be build from the T field of the 1st (possibly polarized) map, and the E field of the second polarized map. (default= cl_out.fits)
simul_type =
Defines which map(s) to analyse, 1=temperature only, 2=temperature AND polarisation. (default= 1)
nlmax =
Defines the maximum l value to be used. See the Recommendations for Users. (default= 64)
maskfile =
Defines the FITS file containing the pixel mask(s) or weighting scheme(s) by which the map(s) read from infile will be multiplied before analysis. If the file contains several fields, the first one in which at least one pixel is non-zero will be used. This option can be used to, for instance, apply the WMAP Kp intensity mask to the data (see ), but it will not handle the WMAP composite mask correctly. Can be used in conjonction with theta_cut_deg. Masked or weighted pixels will be correctly accounted when performing the monopole and dipole regression. (default= `': no mask, all valid pixels are used)
theta_cut_deg =
Defines the latitude (in degrees) of an optional, straight symmetric cut around the equator. Pixels located within that cut (|b|<theta_cut_deg are ignored. (default= $0^\circ$: no cut)
iter_order =
Defines the maximum order of quadrature iteration to be used. (default=0, no iteration)
outfile_alms =
Defines the name of the file containing the alm coefficients of the first map which can be written optionally. (default= no entry -- alms are not written to a file)
outfile_alms2 =
Defines the name of the file containing the alm coefficients of the second map, if any, which can be written optionally. (default= no entry -- alms are not written to a file)
plmfile =
Defines the name for an input file containing precomputed Legendre polynomials Plm. (default= no entry -- anafast executes the recursive evaluation of Plms)
w8file =
Defines name for an input file containing ring weights in the improved quadrature mode (default= no entry -- the name is assumed to be 'weight_ring_n0xxxx.fits' where xxxx is nsmax)
w8filedir =
Gives the directory where the ring weight files are to be found (default= no entry -- anafast searches the default directories, see introduction)
won =
Set this to 1 if ring weight files are to be used, otherwise set it to 0 (or 2). (default= 0)
regression =
Sets the degree of the regression made on the input map before doing the power spectrum analysis. The regression is a minimal variance fit (assuming a uniform noise) made on valid (unflagged and unmasked) pixels, out of the symmetric cut (if any). In case of cut sky analysis, such a regression reduces the monopole and dipole leakage to higher l's.
0 : no regression, does the a_lm analysis on the raw map
1 : removes the best fit monopole first
2 : removes the best fit monopole and dipole first
default = 0.


Anafast reads one of two binary FITS-files containing a HEALPix map. These files can each contain a temperature map or both temperature and polarisation (Q,U) maps. Anafast analyses the map(s) and makes an output ascii-FITS file containing the angular auto or cross power spectra CTls (and CEl, CBl, CT x El, CT x Bl and CE x Bl if specified). Anafast produces Cls up to a specified maximum l-value (see Recommendations for Users). Note in particular that if 2 maps are provided, the CT x El power spectrum will be build from the T field of the 1st (polarized) map, and the E field of the second polarized map. If requested, the computed alm coefficients can be written to a FITS file. This file can be used in the constrained realisation mode of synfast.

Anafast permits two execution modes that allow to improve the quadrature accuracy: (1) the ring weight corrected quadrature, and (2) the iterative scheme. Using the ring weights does not increase the execution time. The precomputed ring weights to be used for each HEALPix resolution nsmax are provided in the $HEALPIX/data directory. The more sophisticated iterative scheme increases the accuracy more effectively than the weighted ring scheme, but its disadvantage is that the time for the analysis increases, 1 iteration takes 3 times as long, 2 iterations 5 times as long on so forth, since each order of iteration requires one more forward and backward transform.

The spherical harmonics evaluation uses a recurrence on associated Legendre polynomials $P_{lm}(\theta)$. This recurrence consumes most of the CPU time used by anafast. We have therefore included an option to load precomputed values for the $P_{lm}(\theta)$ from a file generated by the HEALPix facility plmgen. This represents a tradeoff between memory usage and execution speed of anafast.


The following datasets are involved in the anafast processing.

Dataset Description
data/weight_ring_n0xxxx.fits Files containing ring weights for the anafast improved quadrature mode.


This section lists those routines and facilities (including those external to the Healpix distribution) which can assist in the utilisation of anafast.

This HEALPix facility can generate a map for analysis by anafast.
This HEALPix Fortran facility can be used to modify the alm extracted by anafast before they are used as constraints on a synfast run.
This HEALPix facility can be used to generate precomputed Legendre polynomials.



Anafast runs in interactive mode -- self-explanatory.


anafast filename  

When 'filename' is present, anafast enters the non-interactive mode and parses its inputs from the file 'filename'. This has the following structure: the first entry is a qualifier which announces to the parser which input immediately follows. If this input is omitted in the input file, the parser assumes the default value. If the equality sign is omitted, then the parser ignores the entry. In this way comments may also be included in the file. In this example, the file contains the following qualifiers:
simul_type = 1
nlmax = 64
theta_cut_deg = 0
iter_order = 0
infile = map.fits
outfile = cl_out.fits
regression = 0
Anafast reads the map from map.fits, makes an analysis and produces CTls up to l=64. This powerspectrum is saved in the file cl_out.fits. No galactic cut is excised and no iterations are performed. As regression is set to 0 (its default value) the map is analyzed as is, without prior best fit removal of the monopole nor the dipole.

were omitted, they take their default values (empty strings). This means that no file for precomputed Legendre polynomials is read, no second map is read, no mask is applied, and anafast does not save the alm values from the analysis.

Also since
is not given, it takes it default value 2, which means that ring weights are not used.


* Initial release (HEALPix 0.90)
* Optional non-interactive operation. Proper FITS file support. Improved reccurence algorithm for $P_{lm}(\theta)$ which can compute to higher l values. New functionality: arbitrary order of iterations, precomputed Plm, dumping of alm. (HEALPix 1.00)
* New functionality: possibility of removing the best fit monopole and dipole. New Parser. Can be linked to FFTW (HEALPix 1.20)
* New functionality: addition of maskfile (HEALPix 2.0)
* Bug correction: correct interaction of iterative scheme with masked pixels (HEALPix 2.01)
* New functionality: cross-correlation of 2 maps; Correction of this documentation: the code expects maskfile and not mask_file (HEALPix 2.1)


This section describes error messages generated by anafast

Message Severity Text
can not allocate memory for array xxx Fatal You do not have sufficient system resources to run this facility at the map resolution you required. Try a lower map resolution.

Eric Hivon 2010-06-18
Privacy / Copyright
FIRST GOV Contact: NASA Home Page Site Manager:

CL 03-2650