Run a few extra test simulations

[andreufont]

Few interesting simulations to run would be:

• simulations with a different random seed (two or three extra sims for the central simulation would be great). This will be very useful to estimate the amount of sample variance due to finite box size.
• simulation with different treecool file (I'd probably run a sim with Puchwein's model). This will be a great test of IGM parameterization of the likelihood.
• simulation with different BAO scale (vary Och^2 and Obh^2 a bit to move BAO by a couple of percent). Useful to confirm that we can fix Obh^2 and Och^2 in the emulator) even within LCDM.
• simulation with non-flat geometry (Ok=-0.04 as favored by Planck? Maybe rescale other params to make sure it is near the center of the box?). This should be an easy test, and it would look good.
• simulation with non-zero running (something like alpha_s=0.02, but compensating A_s / n_s to roughly match linear power at k_star). Same here, easy test to highlight the flexibility of the emulator / likelihood compression.

[andreufont]

We might want to try also different box-size and resolutions. The standard setup is L = 67.5 Mpc and N=768 per side, cell size in IC of d=0.088 Mpc. We could run the following test simulations:

• LOWR: same box size but lower resolution, i.e., L=67.5 Mpc, N=512 and d=0.132 Mpc.
• LARGE: same low-resolution than LOWR but larger box size, i.e., L=101.25 Mpc, N=678 and d=0.088 Mpc.

If we pass the LOWR test (i.e., we recover unbiased cosmo), then the LARGE test will be a good test of box size.

If we do not pass the LOWR test, we could try a different strategy and run:

• SMALL: same standard resolution but smaller box size, i.e., L= 45 Mpc, N=512 and d=0.088 Mpc.
• HIGHR: same box size than SMALL, but higher resolution, i.e., L=45 Mpc, N=768 adn d=0.0586 Mpc.

If we pass the SMALL test (i.e., we recover unbiased cosmo), then the HIGHR test will be a good test of resolution.

If we do not pass either LOWR or SMALL tests, we'll have to bite the bullet and run the following test simulations:

• LARGE_1024: same standard resolution, but larger box size, i.e., L=90 Mpc, N=1024 and d=0.088 Mpc.
• HIGHR_768: same standard box size, but higher resolution, i.e., L=67.5 Mpc, N=1024 and d=0.066 Mpc.

I would start with the LOWR test, and see where it takes us.

[andreufont]

Also potentially interesting tests, at lower priority:

• simulation with initial conditions set with a GLASS file or starting at a different redshift.
• simulation with the same transfer function for both baryons and CDM ?
• simulation without QUICKLYA ? Probably too slow...