Cobaya: error with DES likelihood

[Zhuangfei Wang]

Hello:

I just encountered an error when running Cobaya with mcmc option and CAMB package. For the likelihood, I simply used "planck_2018_lowl.TT" and "des_y1.joint" with the default parameter settings. But after an hour, it stopped and returned an error as:

---
cobaya.log.LoggedError: Not possible to extrapolate to k=3029.872768378632 1/Mpc (maximum k possible is 3000 1/Mpc).
---
I didn't modify the source of Cobaya or the CAMB theory code. Besides, there is no such error when I ran with --test or --minimizer option, so it is quite confusing to me.

Any reply or suggestion will be appreciated.

[Antony Lewis]

Lowl TT is very unconstraining so you are probably exploring very odd parts of parameter space if you have no priors imposed. You can plot the partial chains to see what it is doing.

[Zhuangfei Wang]

Hi Antony. Thanks for your reply. I removed the "planck_2018_lowl.TT" and only keep "des_y1.joint", but I still get the same error. Could it be any other reason?

FYI, the pre-settings for params and sampler in my input yaml file are the same as this example in the middle of the page: https://cobaya.readthedocs.io/en/latest/cosmo_basic_runs.html. Thanks a lot.

[Antony Lewis]

Using just DES is even less constraining. DES only constraints are usually done in a reduced parameter space with some priors.

[Antony Lewis]

You can also set

Code: Select all

stop_at_error=F

to reject points rather than stopping

[Zhuangfei Wang]

That makes sense to me. Thanks a lot.

[Sunanda Dey]

Hi,i am also facing the same issue. I tried to use stop_at_error : False in my yaml file. But i am geeting an error. Any one please help me

[Minh Nguyen]

Hi,

I'm hijacking this thread because I'm running into the same issue. Is there then a way to re-produce fig. 19 and 20 in the Planck 2018 cosmological parameter paper: https://arxiv.org/abs/1807.06209 using cobaya? One probably only need the correct prior ranges in the .yaml file, right? I tried to specify only DES in cobaya-cosmo-generator but the priors do not seem to match what listed in Table 1 of DESY1 paper: https://arxiv.org/abs/1708.01530

According to the paper, the contours were obtained from an implementation of DES likelihood in CosmoMC which I assume to be the virtually same as the one in cobaya. So the .yaml files with the proper priors should be available somewhere, shouldn't they?

[Antony Lewis]

The Cobaya implementation is the same as CosmoMC yes. But it is up to you to set the cosmological parameter priors when you run the chains if what you want is not currently in the cosmo generator. For the Planck results we ran "CMB lensing priors" and "DES priors" which are different. (DES also varied the neutrino mass for most results)

[Minh Nguyen]

Thanks, Antony! I'm indeed asking for some further hints on how the "DES priors" were set in the Planck analysis (Fig. 19+20). Your answer seems to indicate that for constraints that only include DES, neutrino mass is not fixed to 0.06eV? Further, DES put priors on Omega_m, Omega_b, A_s; I'm not entirely sure how to place those prior ranges on them in cobaya as they would be derived parameters...

[Antony Lewis]

You can sample them, and have omch2, ombh2 etc be derived parameters instead. Planck did not vary neutrino mass by default, but many of the DES paper results did.

[Minh Nguyen]

Thank you for the hint, really! Today I eventually got the time to try your suggestion. This is probably trivial, and just me being dense, but if I try to do

Code: Select all

likelihood:
	des_y1.joint: null
params:
	omegab:
		prior:
		[...]
		ref:
		[...]
	ombh2:
		derived: 'lambda omegam, omegab, H0: (omegam-omegab)*(H0/100)**2'

Then I get the error

Code: Select all

*ERROR* Could not find anything to use input parameter(s) {'omegab'}.

As a side note, I found that the DES Y1 flat priors on [math]\Omega_m and [math]\Omega_b is already defined in cosmo_input/input_database.py, and used in

Code: Select all

lensingonly_DES_model

.
How do I make use of that and impose those priors by default for all runs where only DES likelihoods are used? Will I need new preset(s)?

EDIT: Declaring the prior for omegab directly in each .yaml file under cobaya/likelihoods/des_y1/ seems to work for me.

[Antony Lewis]

You probably want "value" rather than "derived" since ombh2 is an indirect input parameter.

You can include snippets of yaml in other yaml using !defaults if you want to share multiple parameters (e.g. see how the Planck likelihoods use it).

[Adam Ormondroyd]

I've also been running into this error when using DES by itself or with Planck, even when using the priors in the DESY1 paper.
```
stop_at_error:False
```
does not seem to have any effect regardless when placed either in the DES likelihoods or CAMB.

I've found replacing the `3000*self.acc` with `6000*self.acc` in likelihoods/base_classes/des.py stops the error, but simply doubling acc does not (the same error just happens at larger k)

[Neha Joshi]

Hi, I'm trying to run the des_y1.shear likelihood and I've been getting the same error--I just wanted to check whether a fix is in the works, or do I need to go into the source code as suggested?