Closed timothydmorton closed 9 years ago
Good, I like this plan.
I will think about this specific one more too. How are you initializing your MCMC?
I can't imagine that there's some weird line blanketing thing going on that makes a mid-M look like an A star in JHKW1-3 color space, can you? I'll look at the models closely and try to see if the results you're getting are in some sense physical.
On Tuesday, March 3, 2015, timothydmorton notifications@github.com wrote:
@benmontet https://github.com/benmontet - I'm focusing on 201912552 to try to understand what's up with my stellar isochrones code. My MCMC stubbornly wants to give it a ~9000K temperature, which we know is absurd.
I'll be working on this by putting together some diagnostic plots showing samples & evolution tracks, but this will take some setup; in the meantime, any thoughts you have on this would be welcome.
— Reply to this email directly or view it on GitHub https://github.com/benmontet/k2-characterization/issues/6.
See fit_mcmc
over here:
I've tried with initial_burn
both off and on. Also tried with just JHK and with just WISE.
The star it's trying to fit is a very low-metallicity star, which isn't a truly terrible fit, but it's not very good relative to what we know the true answer to be. I wonder if somehow your initial fit isn't probing the range of parameters that correspond to the true values, so it's finding this mediocre fit and never steps into the mid-M parameter space?
That doesn't explain why the M4 star 201465501 is well fit by your system though.
On Tue, Mar 3, 2015 at 11:21 AM, timothydmorton notifications@github.com wrote:
See fit_mcmc over here https://github.com/timothydmorton/isochrones/blob/new-dartmouth/isochrones/starmodel.py:
I've tried with initial_burn both off and on. Also tried with just JHK and with just WISE.
— Reply to this email directly or view it on GitHub https://github.com/benmontet/k2-characterization/issues/6#issuecomment-76977937 .
So, I found a couple things-- one, I had a bug in my initial_burn
option--- I wasn't re-randomizing the walkers at the end of the initial burn. Also, I added an option to enforce a max distance, which is a bit of a hack, but does seem to nudge things in the right direction:
There's still a bit of straggling going on into larger masses, but this at least calms the insanity.
I think what might have been going on is that in my initial walker randomization, I am randomizing the distances ~d^2, and for a close distance like this, probably no walkers are landing anywhere close to the right distance. I'll tune this some more to try to avoid this behavior.
That sounds reasonable to me. I think that's probably the explanation, given that it's finding a somewhat okay fit in a region of space really far away from the truth (and at a large distance).
Re-randomizing the walkers is probably also what was causing the issue where some of the stars never had walkers move away from their initial guess, no?
On Tue, Mar 3, 2015 at 12:10 PM, timothydmorton notifications@github.com wrote:
I think what might have been going on is that in my initial walker randomization, I am randomizing the distances ~d^2, and for a close distance like this, probably no walkers are landing anywhere close to the right distance. I'll tune this some more to try to avoid this behavior.
— Reply to this email directly or view it on GitHub https://github.com/benmontet/k2-characterization/issues/6#issuecomment-76988381 .
Yup! I'm futzing around a bit more and then I'll re-run them all.
just for kicks, I added back in BVgri and re-fit with these bug-fixes in place.... how does this compare to the spectroscopic params (in particular, feh)?
Just JHK+WISE prefers a more metal-poor, warmer star (more like ~4000 K).
They are well in agreement! The spectroscopic params are:
SpT = M2.8 Teff = 3503 +/- 60 K [Fe/H] = 0.09 +/- 0.09 R* = 0.394 +/- 0.038 R_sun Mass = 0.413 +/- 0.043 M_sun
On Tue, Mar 3, 2015 at 12:45 PM, timothydmorton notifications@github.com wrote:
just for kicks, I added back in BVgri and re-fit with these bug-fixes in place.... how does this compare to the spectroscopic params (in particular, feh)?
[image: triangle] https://cloud.githubusercontent.com/assets/1895387/6468408/ca79a120-c1a2-11e4-855f-04cbf7aebea0.png
Just JHK+WISE prefers a more metal-poor, warmer star (more like ~4000 K).
— Reply to this email directly or view it on GitHub https://github.com/benmontet/k2-characterization/issues/6#issuecomment-76995628 .
Fantastic. I'm going to put back in all the bands (except W4) and re-run everything, and not apologize for small error bars!
Great! I will add text to remind the dear reader about the difference between statistical and systematic errors.
On Tue, Mar 3, 2015 at 12:50 PM, timothydmorton notifications@github.com wrote:
Fantastic. I'm going to put back in all the bands (except W4) and re-run everything, and not apologize for small error bars!
— Reply to this email directly or view it on GitHub https://github.com/benmontet/k2-characterization/issues/6#issuecomment-76996849 .
@benmontet - I'm focusing on 201912552 to try to understand what's up with my stellar isochrones code. My MCMC stubbornly wants to give it a ~9000K temperature, which we know is absurd.
I'll be working on this by putting together some diagnostic plots showing samples & evolution tracks, but this will take some setup; in the meantime, any thoughts you have on this would be welcome.