Correlation function BAO damping

[patricialarsen]

There appears to be under-damping of the BAO signal in the correlation_3d function for halofit. The attached plot shows the comparison between linear, halofit and cosmic emu results from ccl using the code snippet below. (Note that I didn't explictly set m_nu to zero here, but doing this doesn't substantially change the result)

import pyccl
import numpy as np
import matplotlib.pyplot as plt

cosmo = pyccl.Cosmology(transfer_function='boltzmann_camb', matter_power_spectrum='linear', Omega_c=0.26067, Omega_b=0.02242/0.67662,h=0.6766, n_s=0.9665, sigma8=0.8102)
cosmo2 = pyccl.Cosmology(transfer_function='boltzmann_camb', matter_power_spectrum='halofit', Omega_c=0.26067, Omega_b=0.02242/0.67662,h=0.6766, n_s=0.9665, sigma8=0.8102)
cosmo3 = pyccl.Cosmology(transfer_function='boltzmann_camb', matter_power_spectrum='emu', Omega_c=0.26067, Omega_b=0.02242/0.6766**2,h=0.6766, n_s=0.9665, sigma8=0.8102, Neff = 3.04)

r = np.linspace(0.0,200.,1000)
corr_lin = pyccl.correlation_3d(cosmo,1.0,r/0.6766)
corr_halofit = pyccl.correlation_3d(cosmo2,1.0,r/0.6766)
corr_emu = pyccl.correlation_3d(cosmo3,1.0,r/0.6766)

plt.figure()
plt.plot(r,corr_linr**2,label="linear")
plt.plot(r,corr_halofitr2,'--',label="halofit")
plt.plot(r,corr_emu*r2,'-.',label="emu")
plt.legend()
plt.xlabel('r Mpc/h')
plt.ylabel('r^2 zeta_mm (r) Mpc/h ^2')
plt.legend()
plt.show()

[patricialarsen]

Note that this may be a halofit model limitation rather than a CCL issue @beckermr and I are looking into this to confirm

[c-d-leonard]

@patricialarsen - It looks like this is an issue that occurs with halofit in general rather than CCL. I'm not sure we have any immediate fix available.

[beckermr]

This is what we found too. I am going to close this issue for now.

[jablazek]

I'm re-opening this and assigning it to myself. FAST-PT has a resummation method built in (from Ivanov) to do BAO damping at 1loop PT. Exposing this should be straightforward. Generalizing this to damp the halofit BAO is more complicated, but is a good stretch goal.

Note: This issue will require progress on #815

[tilmantroester]

For a more ad-hoc approach, the new HMCode uses a perturbation-theory inspired fitting function (Eq. 14 in https://arxiv.org/pdf/2009.01858.pdf).

[damonge]

@jablazek should this still be open?

[jablazek]

@damonge : Good question. I have not exposed the FAST-PT functionality to PTTracers. However, as @tilmantroester notes, there are now more modern approaches which might be better. Is the HMCode method already available in CCL? If so, maybe that is sufficient?

[damonge]

We can access hmcode via CAMB. It's not ideal, but if we'd rather aim for that than expose the fastpt functionality, we can close this issue and open an HMCode-specific one (or keep both open!)

[jablazek]

Keep this open for now. I have a few fast-pt -> CCL housekeeping tasks to do.

[nikfilippas]

Tagging this as a v3 milestone since it's time we updated the old halofit option with a more modern version like the one implemented in HMCode.

[nikfilippas]

Removed it from v3 milestones because it will work much better after #1063 .