Problem whit pointsource model
felavila opened this issue · comments
Hello, i have a problem whit a point source model for a double lens system B0218+357(https://lweb.cfa.harvard.edu/castles/Individual/B0218.html), the problem come whit the resulting Einstein ring because the program estimate a Einstein ring of 6.4arcsec when the distance between image it more or less 0.3arcsec, maybe the problem is just the program need more points to make a good model, but for others doubles it works pretty fine, so i didn't understand the main reason. I also stay making a something like a check of the results using a lenstronomy SIE.
I am using these parameters in the estimation,
shape_native=(25,25), pixel_scales=(0.05, 0.05))
positions_solver = al.PositionsSolver(grid=grid, pixel_scale_precision=0.0025), and using a point_0=al.ps.PointFlux
My other idea is try to reduce the posible extension of the lens like a new constrain, something like maximum 1 in Einstein ring but truly i dont know how to do dat. I hope not beginning rude whit my basic English but i asked these whit the mayor respect i can. Cheers Felipe
Can you send me the positions of the data you are giving autolens? Will experiment myself to see the issue.
Sure, here is the .json file i am using for these system. thank you for your quick response.
#######
[
{
"name": "point_0",
"positions": [
[
-0.046,
-0.178
],
[
0.08,
0.129
]
],
"positions_noise_map": [
0.003,
0.003
],
"fluxes": [
21.83,
19.39
],
"fluxes_noise_map": [
0.5,
0.5
]
}
]
Using your code set up I get the following solution (which is probably similar to yours):
Bayesian Evidence -27.05803404
Maximum Log Likelihood -6.84013730
Maximum Log Posterior 982871.73624649
model CollectionPriorModel (N=8)
galaxies CollectionPriorModel (N=8)
lens Galaxy (N=5)
mass EllIsothermal (N=5)
source Galaxy (N=3)
point_0 PointFlux (N=3)
Maximum Log Likelihood Model:
galaxies
lens
mass
centre
centre_0 0.022
centre_1 -0.013
elliptical_comps
elliptical_comps_0 0.477
elliptical_comps_1 0.540
einstein_radius 7.781
source
point_0
centre
centre_0 -0.377
centre_1 -0.138
flux 1446.910
Summary (3.0 sigma limits):
galaxies
lens
mass
centre
centre_0 0.0230 (-0.0032, 0.0310)
centre_1 -0.0268 (-0.0862, -0.0054)
elliptical_comps
elliptical_comps_0 0.4894 (0.3905, 0.5901)
elliptical_comps_1 0.5309 (0.4167, 0.6288)
einstein_radius 6.9427 (5.1333, 7.9642)
source
point_0
centre
centre_0 -0.0778 (-0.4557, 0.3000)
centre_1 -0.1050 (-0.6086, 0.5265)
flux 1.2657e+03 (9.7967e+02, 1.5164e+03)
Summary (1.0 sigma limits):
galaxies
lens
mass
centre
centre_0 0.0230 (0.0198, 0.0256)
centre_1 -0.0268 (-0.0355, -0.0188)
elliptical_comps
elliptical_comps_0 0.4894 (0.4478, 0.5394)
elliptical_comps_1 0.5309 (0.4871, 0.5786)
einstein_radius 6.9427 (6.0465, 7.5745)
source
point_0
centre
centre_0 -0.0778 (-0.1999, 0.0681)
centre_1 -0.1050 (-0.2327, 0.0470)
flux 1.2657e+03 (1.1086e+03, 1.3834e+03)
instances
galaxies
lens
redshift 0.5
source
redshift 1.0
I can force the einstein_radius to be below 1.0 as you suggest by adding the following line to my code:
lens = af.Model(al.Galaxy, redshift=0.5, mass=al.mp.EllIsothermal)
source = af.Model(al.Galaxy, redshift=1.0, point_0=al.ps.PointFlux)
model = af.Collection(galaxies=af.Collection(lens=lens, source=source))
model.galaxies.lens.mass.einstein_radius = af.UniformPrior(lower_limit=0.0, upper_limit=1.0)
This gives the following solution:
Bayesian Evidence -30.57488783
Maximum Log Likelihood -6.83812181
Maximum Log Posterior 979020.17429459
model CollectionPriorModel (N=8)
galaxies CollectionPriorModel (N=8)
lens Galaxy (N=5)
mass EllIsothermal (N=5)
source Galaxy (N=3)
point_0 PointFlux (N=3)
Maximum Log Likelihood Model:
galaxies
lens
mass
centre
centre_0 0.039
centre_1 -0.041
elliptical_comps
elliptical_comps_0 -0.318
elliptical_comps_1 -0.236
einstein_radius 0.175
source
point_0
centre
centre_0 0.036
centre_1 -0.033
flux 7.457
Summary (3.0 sigma limits):
galaxies
lens
mass
centre
centre_0 -0.0368 (-0.0432, 0.0455)
centre_1 -0.0025 (-0.0455, 0.0056)
elliptical_comps
elliptical_comps_0 0.2182 (-0.3620, 0.2740)
elliptical_comps_1 0.1929 (-0.2597, 0.2636)
einstein_radius 0.2291 (0.1746, 0.2476)
source
point_0
centre
centre_0 -0.0677 (-0.0854, 0.0383)
centre_1 0.0103 (-0.0333, 0.0219)
flux 1.0437e+01 (7.2846e+00, 1.1993e+01)
Summary (1.0 sigma limits):
galaxies
lens
mass
centre
centre_0 -0.0368 (-0.0396, -0.0304)
centre_1 -0.0025 (-0.0060, 0.0006)
elliptical_comps
elliptical_comps_0 0.2182 (0.1595, 0.2372)
elliptical_comps_1 0.1929 (0.1551, 0.2133)
einstein_radius 0.2291 (0.2195, 0.2386)
source
point_0
centre
centre_0 -0.0677 (-0.0769, -0.0568)
centre_1 0.0103 (0.0040, 0.0164)
flux 1.0437e+01 (8.9189e+00, 1.0884e+01)
instances
galaxies
lens
redshift 0.5
source
redshift 1.0
This will therefore give the "correct" solution.
For now, I recommend you adjust priors when the solution clear goes to an incorrect model.
Why does the model with an einstein radius of 7.781 (which is clearly incorrect) produce as good of a fit as the model with 0.2291?
I think this is due to a bug in PyAutoLens, I will look into this in more detail.
Thanks a lot for your answer, i stay alert.
Cheers Felipe Avila