I am trying to fit photo-z through SNANA, which generates some pathologically incorrect fits.
I have developed a Type Ia SN photometric redshift pipeline. I use sncosmo and iminuit to vary the redshift and fit the light curves simultaneously for the parameters 𝑧, 𝑥0, 𝑡0, 𝑥1, and 𝑐, in order to find the value of 𝑧 that produces the lowest reduced 𝜒2. I select the best redshift estimate by considering both the reduced 𝜒2 and the light curve quality, by implementing some flexible selection criteria on the SNe, including:
• |𝑚𝐵(𝑧) − 𝑚_cosmo (𝑧)| ≤ 1 mag,
where 𝑚_cosmo(𝑧) is the apparent magnitude of the SN using the estimated 𝑧, according to our cosmology.
Additionally, I notice that the photo-z algorithm tends to prefer estimates with 𝑧 ≥ 0.9. This bias arises due to the lack of g-band templates at 𝑧 ≥ 0.9, which leads to the discard of g-band data. As a result, the number of degrees of freedom is reduced, which in turn lowers the reduced 𝜒2 value, making these higher 𝑧 fits appear statistically more favourable.
To account for this degeneracy, I first fit the light curves using only the r,i,z bands and retain all redshift estimates with 𝑧 ≥ 0.9. For the remaining cases, I then refit the light curves using all available g,r,i,z bands. Finally, I run a mcmc on the light curves using the estimated photometric redshifts.
I would like to implement these changes that I have made within the SNANA pipeline.
I am trying to fit photo-z through SNANA, which generates some pathologically incorrect fits.
I have developed a Type Ia SN photometric redshift pipeline. I use sncosmo and iminuit to vary the redshift and fit the light curves simultaneously for the parameters 𝑧, 𝑥0, 𝑡0, 𝑥1, and 𝑐, in order to find the value of 𝑧 that produces the lowest reduced 𝜒2. I select the best redshift estimate by considering both the reduced 𝜒2 and the light curve quality, by implementing some flexible selection criteria on the SNe, including:
• |𝑚𝐵(𝑧) − 𝑚_cosmo (𝑧)| ≤ 1 mag,
where 𝑚_cosmo(𝑧) is the apparent magnitude of the SN using the estimated 𝑧, according to our cosmology.
Additionally, I notice that the photo-z algorithm tends to prefer estimates with 𝑧 ≥ 0.9. This bias arises due to the lack of g-band templates at 𝑧 ≥ 0.9, which leads to the discard of g-band data. As a result, the number of degrees of freedom is reduced, which in turn lowers the reduced 𝜒2 value, making these higher 𝑧 fits appear statistically more favourable.
To account for this degeneracy, I first fit the light curves using only the r,i,z bands and retain all redshift estimates with 𝑧 ≥ 0.9. For the remaining cases, I then refit the light curves using all available g,r,i,z bands. Finally, I run a mcmc on the light curves using the estimated photometric redshifts.
I would like to implement these changes that I have made within the SNANA pipeline.