[CAMB] Turning on or off the 'get_scalar_cls' parameter

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Minji Oh
Posts: 6
Joined: February 05 2015
Affiliation: UST/KASI

[CAMB] Turning on or off the ''get_scalar_cls'' parameter

Post by Minji Oh » March 06 2015

Hi all,

Thanks to the reply I got here last time, I could go over a small hill :)

This time, I have a question about the parameter, 'get_scalar_cls = T/F'.

I thought that this parameter only determines if I want to get the output for scalar cls or not. But, I found out that when I turn on this parameter, it affects to the matter power spectrum output.
In detail, it affects P(k) on small scales (k ~100) about 1.4% and on scales (k~ 0.1) about 0.1% even when I use quite higher accuracy_boost, 3.
I couldn't find anything in detail about this parameter both in README and in CAMB note.

Could you give me some comments on this?
Just in case, I upload my parameter file.
params.ini

Thanks in advance!

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: [CAMB] Turning on or off the ''''get_scalar_cls'''' para

Post by Antony Lewis » March 06 2015

The lower k change may be mainly due to changes in k sampling. Do you get closer agreement if you use increase transfer_k_per_logint with CMB off?

Minji Oh
Posts: 6
Joined: February 05 2015
Affiliation: UST/KASI

[CAMB] Turning on or off the ''get_scalar_cls'' parameter

Post by Minji Oh » March 06 2015

Thank you for your reply~!

Before, I used transfer_k_per_logint = 0 .

And after your reply, I tried to compare the P(k) with get_scalar_cls = T/F again after changing transfer_k_per_logint = 0 to 5.

Comparison plot: output.ps

It still affects P(k) on small scales (k ~100) about 1.4% (the same as before), and on larger scales (k~0.1), the difference becomes bigger, about 4% (It was 0.1% before changing transfer_k_per_logint = 0 to 5).

(I used accuracy_boost = 3).

Could you give some comments?

Thanks!

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: [CAMB] Turning on or off the ''''get_scalar_cls'''' para

Post by Antony Lewis » March 09 2015

I would increase transfer_k_per_logint until the result converges, then use that as a reference for comparison.

Minji Oh
Posts: 6
Joined: February 05 2015
Affiliation: UST/KASI

[CAMB] Turning on or off the ''get_scalar_cls'' parameter

Post by Minji Oh » March 09 2015

Thank you for your reply~!

As you gave the comments, I increased transfer_k_per_logint from 0, 5, 10, 20, 40 to 80 to find some convergent result.

In summary, the test showed that

transfer_k_per_logint = ? / at around k~0.1 / k~10 / k~100
transfer_k_per_logint = 0 / ~0.1 % difference / ~0.03% / ~1.4%
transfer_k_per_logint = 5 / ~4 % difference / ~0.03% / ~1.4%
transfer_k_per_logint = 10 / ~0.8 % difference/ ~0.03% / ~1.4%
transfer_k_per_logint = 20 / ~0.1 % difference/ ~0.03% / ~1.4%
transfer_k_per_logint = 40 / ~0.1 % difference/ ~0.03% / ~1.4%
transfer_k_per_logint = 80 / ~0.1 % difference/ ~0.03% / ~1.4%

I think... the difference in the last line is the convergent result in comparison of P(k) with get_scalar_cls = T to P(k) with get_scalar_cls = F even though the convergent difference is small and maybe in many cases ignorable.

Even though it may be quite stupid question, but if someone cares about this difference, which one would you recommend?

In addition, I checked that in both get_scalar_cls = T and F case, even though CAMB sampled the same ks, there is some difference. Can I get some idea about what causes this difference?

Thank you very much in advance~!

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: [CAMB] Turning on or off the ''''get_scalar_cls'''' para

Post by Antony Lewis » March 12 2015

I don't think CAMB internally uses the same k with CMB on and off (even if the exported matter function sample points are the same). What exactly causes the 1.4% I'm not sure, would need to track that down (but well out of the domain we usually care about for the linear spectrum).

Minji Oh
Posts: 6
Joined: February 05 2015
Affiliation: UST/KASI

[CAMB] Turning on or off the ''get_scalar_cls'' parameter

Post by Minji Oh » March 12 2015

Thank you for your reply.

As you said, when I mentioned 'sampled the same ks', that was about 'the exported points', not about something internal, which I didn't know about.

Now, I think I can go with the output linear Pk safely by your reply :)

Thank you so much~!

Post Reply