Public repository of the Cosmic Linear Anisotropy Solving System (master for the most recent version of the standard code; GW_CLASS to include Cosmic Gravitational Wave Background anisotropies; classnet branch for acceleration with neutral networks; ExoCLASS branch for exotic energy injection; class_matter branch for FFTlog)
I have been testing the variations on the lensing power spectrum as a function of the precision settings, making use of the precision file pk_ref.pre and comparing it to the default precision settings specified in precisions.h. The differences in the unlensed $C_\ell^{TT}$ and $P(k)$ between these two precision settings are very small:
below 0.15% across all $\ell$ for the unlensed $C_\ell^{TT}$,
below 0.2% for $k\lesssim1 h/\mathrm{Mpc}$ for $P(k)$,
below 2% for $k\gtrsim1 h/\mathrm{Mpc}$ for $P(k)$.
The lensing power spectrum $C\ell^{\phi\phi}$ and, therefore, the lensed temperature spectrum, don't behave so nicely. The differences in the lensed $C\ell^{TT}$ between both precision settings are of up to 10%. For the lensing power spectrum, the default settings overestimate the $C_\ell^{\phi\phi}$ by a factor of up to 4 at $\ell\sim1500$ with respect to the settings fixed by pk_ref.pre.
I tried to get around this by studying the effect of the precision parameters k_step _super, k_step _sub, k_min _tau0, k_max _tau0 _over _l _maxand k_step _super _reduction, among others, but I haven't been able to find appropriate values for them such that the calculation of $C\ell^{\phi\phi}$ converges to a stable result. I have also looked at the effect of the Limber approximation and the inclusion of nonlinearities with HALOFIT, also failing to find a stable solution for the $C\ell^{\phi\phi}$.
Is there a way in which this can be done with a proper choice of precision settings that doesn't take a few minutes to run? Is there an issue with the calculation of the lensing power spectrum from the matter power spectrum?
I have been using class 3.2.0 from the binary (not the Python wrapper). For all precision tests I've been using the best fit parameters from the 2018 Planck results. I'll be happy to share more information at your request.
I have been testing the variations on the lensing power spectrum as a function of the precision settings, making use of the precision file
pk_ref.preand comparing it to the default precision settings specified inprecisions.h. The differences in the unlensed $C_\ell^{TT}$ and $P(k)$ between these two precision settings are very small:The lensing power spectrum $C\ell^{\phi\phi}$ and, therefore, the lensed temperature spectrum, don't behave so nicely. The differences in the lensed $C\ell^{TT}$ between both precision settings are of up to 10%. For the lensing power spectrum, the default settings overestimate the $C_\ell^{\phi\phi}$ by a factor of up to 4 at $\ell\sim1500$ with respect to the settings fixed by
pk_ref.pre.I tried to get around this by studying the effect of the precision parameters
k_step _super,k_step _sub,k_min _tau0,k_max _tau0 _over _l _maxandk_step _super _reduction, among others, but I haven't been able to find appropriate values for them such that the calculation of $C\ell^{\phi\phi}$ converges to a stable result. I have also looked at the effect of the Limber approximation and the inclusion of nonlinearities with HALOFIT, also failing to find a stable solution for the $C\ell^{\phi\phi}$.Is there a way in which this can be done with a proper choice of precision settings that doesn't take a few minutes to run? Is there an issue with the calculation of the lensing power spectrum from the matter power spectrum?
I have been using class 3.2.0 from the binary (not the Python wrapper). For all precision tests I've been using the best fit parameters from the 2018 Planck results. I'll be happy to share more information at your request.