Does the differences between Galtab and Halotools acceptable?

[Sherlockids]

In this case, I made a test on results between Galtab and Halotools, all the dHOD parameters are the same in Galtab and Halotools, they fit each other well when 'Acen/sat = 1', but have some differences in 'Acen/sat = 0'.

Does the differences are acceptable?

Here is my code:
test_galtab.ipynb.zip

[Sherlockids]

We found the results are sensitive to the fiducial HOD model, in the figure above, the fiducial model uses parameters of Acen = 1, Asat = 1, which makes the differences between halotools and galtab so huge.

In your paper (Pearl et al., 2023), the fiducial model is set to be the best fit of Wang et al. (2022). What if I want to use the algorithm on a catalog whose dHOD parameters are unkown? Is the best choice of my fiducial model to be Acen = 0, Asat = 0 ? Since we found the results of galtab and halotools align well under this fiducial model.

[AlanPearl]

First off - I checked my code and we actually forced Acen/sat = 0 for the fiducial model (must have forgotten to include that subtle difference in the paper). I think this is probably the best practice.
However, what is more important is that your fiducial model is populating more placeholders than necessary at each halo mass, particularly for satellites. Whenever the precision is not sufficient, the best solution is probably taking your fiducial model and lowering logM1 to increase the number of satellite placeholders. Could also be worth testing lowering logM0, but I think that has less of an effect. Note that increasing the number of placeholders will increase the runtime, so there is a tradeoff here.

[AlanPearl]

they fit each other well when 'Acen/sat = 1', but have some differences in 'Acen/sat = 0'.

Are you still seeing this difference? This is surprising, unless if it was simply an insufficient number of placeholders generated by the fiducial model

[Sherlockids]

First off - I checked my code and we actually forced Acen/sat = 0 for the fiducial model (must have forgotten to include that subtle difference in the paper). I think this is probably the best practice. However, what is more important is that your fiducial model is populating more placeholders than necessary at each halo mass, particularly for satellites. Whenever the precision is not sufficient, the best solution is probably taking your fiducial model and lowering logM1 to increase the number of satellite placeholders. Could also be worth testing lowering logM0, but I think that has less of an effect. Note that increasing the number of placeholders will increase the runtime, so there is a tradeoff here.

Thanks a lot, I'll try it.

[Sherlockids]

they fit each other well when 'Acen/sat = 1', but have some differences in 'Acen/sat = 0'.

Are you still seeing this difference? This is surprising, unless if it was simply an insufficient number of placeholders generated by the fiducial model

I think that's because when I try to populate galaxy tabulation ofAcen = 0, Asat = 0 but using Acen = 1, Asat = 1 as fiducial model, which may cause some problem of insufficient placeholders. Thanks for your attention and I will force the fiducial model to be 'Acen = 0, Asat = 0'.