Open andreufont opened 3 years ago
It looks like sim_h also has slightly different file sizes for some skewer files:
ls -l sim_pair_?/sim_plus/skewers/skewers_0_*
-r-xr-xr-x 1 afontrib desi 2705012232 Dec 27 2020 sim_pair_0/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-r-xr-xr-x 1 afontrib desi 2705012232 Dec 28 2020 sim_pair_1/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-r-xr-xr-x 1 afontrib desi 2705012232 Jan 4 2021 sim_pair_2/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-r-xr-xr-x 1 afontrib desi 2705012232 Jan 11 2021 sim_pair_3/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-r-xr-xr-x 1 afontrib desi 2705012232 Jan 12 2021 sim_pair_4/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-r-xr-xr-x 1 afontrib desi 2705012232 Jan 13 2021 sim_pair_5/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-r-xr-xr-x 1 afontrib desi 2705012232 Jan 13 2021 sim_pair_6/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-r-xr-xr-x 1 afontrib desi 2705012232 Jan 14 2021 sim_pair_7/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-rwxr-xr-x 1 afontrib desi 2705012232 Jan 15 2021 sim_pair_8/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-rwxr-xr-x 1 afontrib desi 2705012232 Jan 16 2021 sim_pair_9/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
-rwxr-xr-x 1 afontrib desi 2703012232 Jul 24 2020 sim_pair_h/sim_plus/skewers/skewers_0_Ns500_wM0.05_sT1.0_sg1.0.hdf5
So it is not just rounding errors, I found out that a key contribution to this issues with pixel widths is that fake_spectra ignores the contribution from radiation density in the computation of H(z), and therefore the expansion rates are underestimated by almost 0.1% at the highest redshifts.
I wonder whether we should have a function like extract_skewers.fake_spectra_width_kms that sets the pixel width in velocity units using the same function that fake_spectra uses, to make sure that we end up with the right number of pixels...
Or we can just forget about it, and get used to the fact that the number of pixels (and therefore the pixel width) will vary from simulation to simulation, and snapshot to snapshot, because of this subtle differences in H(z).
Probably harmless, but one of the skewers file (snapshot 10) from one of the simulations (1 plus) in the latin hypercube is slightly larger than it should, probably because rounding errors resulted in a smaller cell size?