Correlation functions

Functions to transform CMB angular power spectra into correlation functions (cl2corr) and vice versa (corr2cl), and calculate lensed power spectra from unlensed ones.

The lensed power spectrum functions are not intended to replace those calculated by default when getting CAMB results, but may be useful for tests, e.g. using different lensing potential power spectra, partially-delensed lensing power spectra, etc.

These functions are all pure python/scipy, and operate and return cls including factors \(\ell(\ell+1)/2\pi\) (for CMB) and \([L(L+1)]^2/2\pi\) (for lensing).

  1. Lewis December 2016

camb.correlations.cl2corr(cls, xvals, lmax=None)[source]

Get the correlation function from the power spectra, evaluated at points cos(theta) = xvals. Use roots of Legendre polynomials (np.polynomial.legendre.leggauss) for accurate back integration with corr2cl. Note currently does not work at xvals=1 (can easily calculate that as special case!).

Parameters:

  • cls – 2D array cls(L,ix), with L (\(\equiv \ell\)) starting at zero and ix-0,1,2,3 in order TT, EE, BB, TE. cls should include \(\ell(\ell+1)/2\pi\) factors.

  • xvals – array of \(\cos(\theta)\) values at which to calculate correlation function.

  • lmax – optional maximum L to use from the cls arrays

Returns:

2D array of corrs[i, ix], where ix=0,1,2,3 are T, Q+U, Q-U and cross

camb.correlations.corr2cl(corrs, xvals, weights, lmax)[source]

Transform from correlation functions to power spectra. Note that using cl2corr followed by corr2cl is generally very accurate (< 1e-5 relative error) if xvals, weights = np.polynomial.legendre.leggauss(lmax+1)

Parameters:

  • corrs – 2D array, corrs[i, ix], where ix=0,1,2,3 are T, Q+U, Q-U and cross

  • xvals – values of \(\cos(\theta)\) at which corrs stores values

  • weights – weights for integrating each point in xvals. Typically from np.polynomial.legendre.leggauss

  • lmax – maximum \(\ell\) to calculate \(C_\ell\)

Returns:

array of power spectra, cl[L, ix], where L starts at zero and ix=0,1,2,3 in order TT, EE, BB, TE. They include \(\ell(\ell+1)/2\pi\) factors.

camb.correlations.gauss_legendre_correlation(cls, lmax=None, sampling_factor=1)[source]

Transform power spectrum cls into correlation functions evaluated at the roots of the Legendre polynomials for Gauss-Legendre quadrature. Returns correlation function array, evaluation points and weights. Result can be passed to corr2cl for accurate back transform.

Parameters:

  • cls – 2D array cls(L,ix), with L (\(\equiv \ell\)) starting at zero and ix=0,1,2,3 in order TT, EE, BB, TE. Should include \(\ell(\ell+1)/2\pi\) factors.

  • lmax – optional maximum L to use

  • sampling_factor – uses Gauss-Legendre with degree lmax*sampling_factor+1

Returns:

corrs, xvals, weights; corrs[i, ix] is 2D array where ix=0,1,2,3 are T, Q+U, Q-U and cross

camb.correlations.legendre_funcs(lmax, x, m=(0, 2), lfacs=None, lfacs2=None, lrootfacs=None)[source]

Utility function to return array of Legendre and \(d_{mn}\) functions for all \(\ell\) up to lmax. Note that \(d_{mn}\) arrays start at \(\ell_{\rm min} = \max(m,n)\), so returned arrays are different sizes

Parameters:

  • lmax – maximum \(\ell\)

  • x – scalar value of \(\cos(\theta)\) at which to evaluate

  • m – m values to calculate \(d_{m,n}\), etc. as relevant

  • lfacs – optional pre-computed \(\ell(\ell+1)\) float array

  • lfacs2 – optional pre-computed \((\ell+2)*(\ell-1)\) float array

  • lrootfacs – optional pre-computed sqrt(lfacs*lfacs2) array

Returns:

\((P,P'),(d_{11},d_{-1,1}), (d_{20}, d_{22}, d_{2,-2})\) as requested, where P starts at \(\ell=0\), but spin functions start at \(\ell=\ell_{\rm min}\)

camb.correlations.lensed_cl_derivative_unlensed(clpp, lmax=None, theta_max=0.09817477042468103, apodize_point_width=10, sampling_factor=1.4)[source]

Get derivative dcl of lensed minus unlensed power \(D_\ell \equiv \ell(\ell+1)\Delta C_\ell/2\pi\) with respect to \(\ell(\ell+1)C^{\rm unlens}_\ell/2\pi\)

The difference in power in the lensed spectrum is given by dCL[ix, :, :].dot(cl), where cl is the appropriate \(\ell(\ell+1)C^{\rm unlens}_\ell/2\pi\).

Uses the non-perturbative curved-sky results from Eqs 9.12 and 9.16-9.18 of astro-ph/0601594, to second order in \(C_{{\rm gl},2}\)

Parameters:

  • clpp – array of \([L(L+1)]^2 C_L^{\phi\phi}/2\pi\) lensing potential power spectrum (zero based)

  • lmax – optional maximum L to use from the clpp array

  • theta_max – maximum angle (in radians) to keep in the correlation functions

  • apodize_point_width – if theta_max is set, scale the fixed-angle C2 apodization around the cut; the default corresponds to a transition width of about 48*pi/lmax

  • sampling_factor – npoints = int(sampling_factor*lmax)+1

Returns:

array dCL[ix, ell, L], where ix=0,1,2,3 are TT, EE, BB, TE and result is \(d\left(\Delta D^{\rm ix}_\ell\right) / d D^{{\rm unlens},j}_L\) where j[ix] are TT, EE, EE, TE

camb.correlations.lensed_cl_derivatives(cls, clpp, lmax=None, theta_max=0.09817477042468103, apodize_point_width=10, sampling_factor=1.4)[source]

Get derivative dcl of lensed \(D_\ell\equiv \ell(\ell+1)C_\ell/2\pi\) with respect to \(\log(C^{\phi}_L)\). To leading order (and hence not actually accurate), the lensed correction to power spectrum is is given by dcl[ix,:,:].dot(np.ones(clpp.shape)).

Uses the non-perturbative curved-sky results from Eqs 9.12 and 9.16-9.18 of astro-ph/0601594, to second order in \(C_{{\rm gl},2}\)

Parameters:

  • cls – 2D array of unlensed cls(L,ix), with L starting at zero and ix=0,1,2,3 in order TT, EE, BB, TE. cls should include \(\ell(\ell+1)/2\pi\) factors.

  • clpp – array of \([L(L+1)]^2 C_L^{\phi\phi}/2\pi\) lensing potential power spectrum (zero based)

  • lmax – optional maximum L to use from the cls arrays

  • theta_max – maximum angle (in radians) to keep in the correlation functions

  • apodize_point_width – if theta_max is set, scale the fixed-angle C2 apodization around the cut; the default corresponds to a transition width of about 48*pi/lmax

  • sampling_factor – npoints = int(sampling_factor*lmax)+1

Returns:

array dCL[ix, ell, L], where ix=0,1,2,3 are T, EE, BB, TE and result is \(d[D^{\rm ix}_\ell]/ d (\log C^{\phi}_L)\)

camb.correlations.lensed_cls(cls, clpp, lmax=None, lmax_lensed=None, sampling_factor=1.4, delta_cls=False, theta_max=0.09817477042468103, apodize_point_width=10, leggaus=True, cache=True, use_lensing_template=False, low_l_ee_taper=False)[source]

Get the lensed power spectra from the unlensed power spectra and the lensing potential power. Uses the non-perturbative curved-sky results from Eqs 9.12 and 9.16-9.18 of astro-ph/0601594, to second order in \(C_{{\rm gl},2}\).

Correlations are calculated for Gauss-Legendre integration if leggaus=True; this slows it by several seconds, but will be must faster on subsequent calls with the same lmax*sampling_factor. If Gauss-Legendre is not used, sampling_factor needs to be about 2 times larger for same accuracy.

For a reference implementation with the full integral range and no apodization set theta_max=None.

By default this function does not pad high \(\ell\) with a smooth fit; for accurate results it should be called with lmax high enough that input cls are effectively band limited (lmax >= 2500, or higher for accurate BB to small scales). Set use_lensing_template=True to extend inputs above their supplied maximum using CAMB’s high-L lensing template, matching the Fortran lensing convention. Usually lmax truncation errors are far larger than other numerical errors for lmax<4000. For a faster result use get_lensed_cls_with_spectrum.

Parameters:

  • cls – 2D array of unlensed cls(L,ix), with L starting at zero and ix=0,1,2,3 in order TT, EE, BB, TE. cls should include \(\ell(\ell+1)/2\pi\) factors.

  • clpp – array of \([L(L+1)]^2 C_L^{\phi\phi}/2\pi\) lensing potential power spectrum (zero based)

  • lmax – optional maximum L to use from the cls arrays

  • lmax_lensed – optional maximum L for the returned cl array (lmax_lensed <= lmax)

  • sampling_factor – npoints = int(sampling_factor*lmax)+1

  • delta_cls – if true, return the difference between lensed and unlensed (optional, default False)

  • theta_max – maximum angle (in radians) to keep in the correlation functions; default: pi/32

  • apodize_point_width – if theta_max is set, scale the fixed-angle C2 apodization around the cut; the default corresponds to a transition width of about 48*pi/lmax

  • leggaus – whether to use Gauss-Legendre integration (default True)

  • cache – if leggaus = True, set cache to save the x values and weights between calls (most of the time)

  • use_lensing_template – if True, allow lmax above the input arrays by extrapolating TT, EE, TE and PP with the bundled high-L template scaled at the last supplied multipole

  • low_l_ee_taper – smoothly taper the reionization-bump EE (l<20) out of the lensing kernel, matching the Fortran short-range convention; otherwise its low-L lensed BB is window-sensitive when theta_max is set

Returns:

2D array of cls[L, ix], with L starting at zero and ix=0,1,2,3 in order TT, EE, BB, TE. cls include \(\ell(\ell+1)/2\pi\) factors.

camb.correlations.lensed_correlations(cls, clpp, xvals, weights=None, lmax=None, delta=False, theta_max=None, apodize_point_width=10, low_l_ee_taper=False)[source]

Get the lensed correlation function from the unlensed power spectra, evaluated at points \(\cos(\theta)\) = xvals. Use roots of Legendre polynomials (np.polynomial.legendre.leggauss) for accurate back integration with corr2cl. Note currently does not work at xvals=1 (can easily calculate that as special case!).

To get the lensed cls efficiently, set weights to the integral weights for each x value, then function returns lensed correlations and lensed cls.

Uses the non-perturbative curved-sky results from Eqs 9.12 and 9.16-9.18 of astro-ph/0601594, to second order in \(C_{{\rm gl},2}\)

Parameters:

  • cls – 2D array of unlensed cls(L,ix), with L (\(\equiv\ell\)) starting at zero and ix=0,1,2,3 in order TT, EE, BB, TE. cls should include \(\ell(\ell+1)/2\pi\) factors.

  • clpp – array of \([L(L+1)]^2 C_L^{\phi\phi}/2\pi\) lensing potential power spectrum (zero based)

  • xvals – array of \(\cos(\theta)\) values at which to calculate correlation function.

  • weights – if given also return lensed \(C_\ell\), otherwise just lensed correlations

  • lmax – optional maximum L to use from the cls arrays

  • delta – if true, calculate the difference between lensed and unlensed (default False)

  • theta_max – maximum angle (in radians) to keep in the correlation functions

  • apodize_point_width – scale factor for the fixed-angle C2 apodization at the truncation cut; the default corresponds to a transition width of about 48*pi/lmax

  • low_l_ee_taper – smoothly taper the reionization-bump EE (l<20) out of the lensing kernel, matching the Fortran short-range convention; otherwise its low-L lensed BB is window-sensitive when theta_max is set

Returns:

2D array of corrs[i, ix], where ix=0,1,2,3 are T, Q+U, Q-U and cross; if weights is not None, then return corrs, lensed_cls

camb.correlations.lensing_R(clpp, lmax=None)[source]

Get \(R \equiv \frac{1}{2} \langle |\nabla \phi|^2\rangle\)

Parameters:

  • clpp – array of \([L(L+1)]^2 C_L^{\phi\phi}/2\pi\) lensing potential power spectrum

  • lmax – optional maximum L to use from the cls arrays

Returns:

R

camb.correlations.lensing_correlations(clpp, xvals, lmax=None)[source]

Get the \(\sigma^2(x)\) and \(C_{{\rm gl},2}(x)\) functions from the lensing power spectrum

Parameters:

  • clpp – array of \([L(L+1)]^2 C_L^{\phi\phi}/2\pi\) lensing potential power spectrum (zero based)

  • xvals – array of \(\cos(\theta)\) values at which to calculate correlation function.

  • lmax – optional maximum L to use from the clpp array

Returns:

array of \(\sigma^2(x)\), array of \(C_{{\rm gl},2}(x)\)