Why just H0?

[tcollett]

We should get good teams to give us their time-delay distances, and then we can see if any residual inaccuracy leaks into biases on e.g. dark energy parameters.

It doesn't seem any harder than just doing H0?

[drphilmarshall]

Well, it's a higher dimension hyper-parameter space, so that will matter at some point. We're at about 7ms per likelihood evaluation right now, so MCMC is a possibility. Mainly we focused on H0 just to keep things simple and focused on the time delay accuracy itself; Ill think a bit more about how much trouble relaxing the other cosmological parameters would cause.

[tcollett]

Shouldn't be any harder - you just do the importance sampling on a coarser
multidimensional grid - if you get their time delay distances you can still
do the H0 only case by fixing O_M and O_Lambda. But actually MCMC is a
better option than the importance sampling method anyways (and since it'll
be a very smooth likelihood function it's not hard to sample).

There doesn't seem a good reason not to ask for time-delay distances for
each lens rather than H0's.

On 25 July 2016 at 21:42, Phil Marshall notifications@github.com wrote:

Well, it's a higher dimension hyper-parameter space, so that will matter
at some point. We're at about 7ms per likelihood evaluation right now, so
MCMC is a possibility. Mainly we focused on H0 just to keep things simple
and focused on the time delay accuracy itself; Ill think a bit more about
how much trouble relaxing the other cosmological parameters would cause.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#26 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/ABc1BjUE7gClulb7RLFoMlO_Bcr6wW2lks5qZR-9gaJpZM4JTnEc
.

[drphilmarshall]

OK - how about we start by adding an "MCMC" option, and plugging in emcee and inferring H_0? Then we can look at extending the cosmological parameter array to more dimensions (still with uninformative priors). This could be a good hack for someone at Astro Hack Week 2016 next week - I'll pitch it to the group and report back.

[tcollett]

Sounds good!

On 25 August 2016 at 04:38, Phil Marshall notifications@github.com wrote:

OK - how about we start by adding an "MCMC" option, and plugging in emcee
and inferring H_0? Then we can look at extending the cosmological parameter
array to more dimensions (still with uninformative priors). This could be a
good hack for someone at Astro Hack Week 2016 next week - I'll pitch it to
the group and report back.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#26 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/ABc1Bgz0CLBNy83_hnxW-FQUgqt_BAivks5qjQ5SgaJpZM4JTnEc
.