band_shift the absolute magnitudes

[moustakas]

In fastphot, consider band-shifting the ugriz photometry to z=0.1 so the results can be compared with SDSS more easily. The UBVW1 bands can be band-shifted to z=0.

[moustakas]

This ticket ended up being particularly tricky, not least because I hadn't previously figured out how to compute band-shifted magnitudes. But I believe that #69 has this implemented correctly.

As an extra validation step, I took a set of objects and computed K-corrections and absolute magnitudes using both fastphot and the industry standard (IDL version of) K-correct (@blanton144).

Note that these calculations rely on the version of the SDSS filters in speclite which include the atmosphere (see desihub/speclite#76). In addition:

• I use grzW1 as the input photometry, all corrected for Galactic extinction, and the DESI redshift.
• I use the correct North and South filter curves according to the declination of each object.
• I add minimum photometric uncertainties to the input photometry to make the chi2 of the fits a bit more reasonable.

First, carried out a sanity check to make sure the independent model fits were sensible. This first figure shows that fastphot and K-correct agree (in the input magnitudes synthesized from each model) to within a few percent in grzW1 (<2% in all bands except the r-band, which is offset by 3.3%):

The next two plots show the K-corrections and the absolute magnitudes in the rest-frame SDSS ugriz bands, band-shifted to z=0.1. The agreement in i and W1 suggests that I'm handling the band-shifting correctly, but the disagreement in g and r is disturbingly large (0.45 and 0.2 mag, respectively). I'm not sure how to proceed with this.

Finally, I also checked K-correcting to the rest-frame Bessell UBV filters, band-shifted to z=0.0, and these results look pretty good (note the different y-scale):

[blanton144]

I wonder if you can just plot the two K-corrections vs redshift? That might help clarify this for me. You would want to compare to Blanton & Roweis (2007) Figure 16. At these redshifts, the r-band K-corrections are basically about 1*z, and the g-band are about 2*z, so either K-correction being 0.45 mag AWAY FROM ZERO at these redshifts seems large, yet the two methods disagree at that level. So one of these methods is just wrong by a LOT, and looking at the actual K-corrections will say which. In particular, looking at the K-corrections at z=0.1, which should just be about -0.1 mag (-2.5*log10(1+z)) independent of galaxy.

…

On Tue, Jul 12, 2022 at 4:12 PM Moustakas ***@***.***> wrote: This ticket ended up being particularly tricky, not least because I hadn't previously figured out how to compute band-shifted magnitudes. But I believe that #69 <https://urldefense.com/v3/__https://github.com/desihub/fastspecfit/pull/69__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVCzVV9yO$> has this implemented correctly. As an extra validation step, I took a set of objects and computed K-corrections and absolute magnitudes using both fastphot and the industry standard (IDL version of) K-correct ***@***.*** <https://urldefense.com/v3/__https://github.com/blanton144__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVHbUJ8oX$> ). Note that these calculations rely on the version of the SDSS filters in speclite which include the atmosphere (see desihub/speclite#76 <https://urldefense.com/v3/__https://github.com/desihub/speclite/pull/76__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVPJLlThw$>). In addition: - I use *grzW1* as the input photometry, all corrected for Galactic extinction, and the DESI redshift. - I use the correct North and South filter curves according to the declination of each object. - I add minimum photometric uncertainties to the input photometry to make the chi2 of the fits a bit more reasonable. First, carried out a sanity check to make sure the independent model fits were sensible. This first figure shows that fastphot and K-correct agree (in the input magnitudes synthesized from each model) to within a few percent in *grzW1* (<2% in all bands except the r-band, which is offset by 3.3%): [image: Screen Shot 2022-07-12 at 3 55 55 PM] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/1431820/178582671-8c6eed75-460c-4fa1-844b-47104b31d6c9.png__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVIMYYWke$> The next two plots show the K-corrections and the absolute magnitudes in the rest-frame SDSS *ugriz* bands, band-shifted to z=0.1. The agreement in *i* and *W1* suggests that I'm handling the band-shifting correctly, but the disagreement in *g* and *r* is disturbingly large (0.45 and 0.2 mag, respectively). I'm not sure how to proceed with this. [image: Screen Shot 2022-07-12 at 3 59 34 PM] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/1431820/178583291-6059d2e5-9adf-44a4-b75a-929f3d7632a8.png__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVNoRR64_$> [image: Screen Shot 2022-07-12 at 3 59 51 PM] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/1431820/178583341-a54e247f-8c67-40c2-a161-375ac08fac32.png__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVFJ73i2M$> Finally, I also checked K-correcting to the rest-frame Bessell *UBV* filters, band-shifted to z=0.0, and these results look pretty good (note the different y-scale): [image: Screen Shot 2022-07-12 at 4 08 47 PM] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/1431820/178585448-dfbbf65e-740d-40c9-8531-76f59526546d.png__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVA6lfjqj$> — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/desihub/fastspecfit/issues/43*issuecomment-1182455498__;Iw!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVBfzP_0z$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AAODDFRGLOTGZ5JWY723WKTVTXGRDANCNFSM5DL2P64A__;!!BhJSzQqDqA!UWx4QueZL18T_C_aWKneGdy328JpU2ThgsGCteHUKuM_fweQmvIHf3aZYrOMpt9J_EhZ_q8mfbtDRLaIVBOdqNJD$> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- Michael Blanton, Professor Department of Physics New York University

[moustakas]

Thanks for the suggestion, @blanton144, which turned up a couple different bugs in my code. I also ended up digging into my dust attenuation model and decided to switch from a Charlot & Fall power-law (with no UV bump) to a fairly standard Fitzpatrick & Masa Galactic curve (with a UV bump); I'll postpone a deeper dig into how I handle dust attenuation in a future effort.

Below is the K(z) plot you requested and the updated comparison of absolute magnitudes, which now look pretty good (<2% agreement) except in the 0.1u-band. I'm inclined to declare victory because the u-band is quite sensitive to the choice of spectral templates. For example, below is my maximum likelihood fit to an object where we disagree by more than half a magnitude (I'm always brighter); for this object, with fastphot I get M(0.1u)=-18.958 and K-correct yields -18.329 (the open symbols, W2-W4, are bands which are not used during fitting and, of course, I fit to fluxes not magnitudes). The gray curve is the best-fitting K-correct model, and you can see that our UV spectra are very different! (And you can also see why our synthesized r-band photometry doesn't match precisely; there are some features in your templates that are definitely not in my templates.)

I welcome any other thoughts you or others may have.

[blanton144]

Hi John, Yes, I would declare success. Since you don't have an observed bandpass near 0.1u, you don't necessarily expect to be able to recover that magnitude consistently from different SED models! Thanks, Mike

…

On Wed, Jul 13, 2022 at 8:22 AM Moustakas ***@***.***> wrote: Thanks for the suggestion, @blanton144 <https://urldefense.com/v3/__https://github.com/blanton144__;!!BhJSzQqDqA!V1sdKCsyVsNZkafNwBZGShJZ6ZxKW6su9D5wSckLh2Q2W8f4Px9OJnnVx7fRj--mnNaR7yxIpE-sPbMHJQJRO0CL$>, which turned up a couple different bugs in my code. I also ended up digging into my dust attenuation model and decided to switch from a Charlot & Fall power-law (with no UV bump) to a fairly standard Fitzpatrick & Masa Galactic curve (with a UV bump); I'll postpone a deeper dig into how I handle dust attenuation in a future effort. Below is the K(z) plot you requested and the updated comparison of absolute magnitudes, which now look pretty good (<2% agreement) except in the 0.1u-band. I'm inclined to declare victory because the u-band is quite sensitive to the choice of spectral templates. For example, below is my maximum likelihood fit to an object where we disagree by more than half a magnitude (I'm always brighter); for this object, with fastphot I get M(0.1u)=-18.958 and K-correct yields -18.329 (the open symbols, W2-W4, are bands which are not used during fitting and, of course, I fit to fluxes not magnitudes). The gray curve is the best-fitting K-correct model, and you can see that our UV spectra are *very* different! (And you can also see why our synthesized r-band photometry doesn't match precisely; there are some features in your templates that are definitely not in my templates.) I welcome any other thoughts you or others may have. [image: Screen Shot 2022-07-13 at 11 13 56 AM] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/1431820/178768899-78e76aca-4cf9-4cf1-baf6-8b52a1923e35.png__;!!BhJSzQqDqA!V1sdKCsyVsNZkafNwBZGShJZ6ZxKW6su9D5wSckLh2Q2W8f4Px9OJnnVx7fRj--mnNaR7yxIpE-sPbMHJebTFpdK$> [image: Screen Shot 2022-07-13 at 11 15 36 AM] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/1431820/178769262-9210d078-5ef7-4888-8607-e6e40f825611.png__;!!BhJSzQqDqA!V1sdKCsyVsNZkafNwBZGShJZ6ZxKW6su9D5wSckLh2Q2W8f4Px9OJnnVx7fRj--mnNaR7yxIpE-sPbMHJcUNtO3d$> [image: Screen Shot 2022-07-13 at 11 15 56 AM] <https://urldefense.com/v3/__https://user-images.githubusercontent.com/1431820/178769329-8b34c5cb-e58b-448a-ae3f-f158b2f9bc59.png__;!!BhJSzQqDqA!V1sdKCsyVsNZkafNwBZGShJZ6ZxKW6su9D5wSckLh2Q2W8f4Px9OJnnVx7fRj--mnNaR7yxIpE-sPbMHJfPlQrgF$> — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/desihub/fastspecfit/issues/43*issuecomment-1183361353__;Iw!!BhJSzQqDqA!V1sdKCsyVsNZkafNwBZGShJZ6ZxKW6su9D5wSckLh2Q2W8f4Px9OJnnVx7fRj--mnNaR7yxIpE-sPbMHJTgcSDOW$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AAODDFTY5JZCGQWL5OVTQBLVT3NLFANCNFSM5DL2P64A__;!!BhJSzQqDqA!V1sdKCsyVsNZkafNwBZGShJZ6ZxKW6su9D5wSckLh2Q2W8f4Px9OJnnVx7fRj--mnNaR7yxIpE-sPbMHJQycrctn$> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- Michael Blanton, Professor Department of Physics New York University

[moustakas]

Great, thanks, these code changes are now in #69.

And a note to myself: all these plots were made with
https://data.desi.lbl.gov/desi/users/ioannis/fastspecfit/pv-vdisp/fujilupe/pv-vdisp.ipynb