Multiband fit in PHOEBE

[kittipongwangnok]

I just wondering about the result from PHOEBE when we use multiband for fitting.
How can we get the PHOEBE model to match each light curve? It seems like only one plot that model fits the data.
Do we have to set up the passband filter?
The result is attached here.

[abokon]

Hi Kittipong Wangnok,

my name is András, and I have worked with multicolour dataset recently, so I have fresh memory about this.

If all the lightcurves are enabled for your compute, and you set the b['passband@lc01'] (and for lc02 etc.):

I think the handling mode of the passband luminosities (e.g. b['pblum_mode@lc01'] ...) are not set up correctly.
(1) You should select one of the datasets to remain as component-coupled.
(2) The other ones (following this example e.g. b['pblum_mode@lc02']) should set up as dataset-coupled, and theoretically the phoebe configures the pblum_dataset to lc01 -- or to the other name of your particular dataset which you have in the first place. For your safety, you should check it. :)

In this situation, all the datasets will be computed based on the informations given in your bundle including the passband luminosity of the first component of binary in dataset lc01. If you would like to optimize it in this new configuration, then you should give 'primary@pblum@lc01' as fit parameter in your solver, at least in this given case.

If you have any questions, do not hesitate to ask. :)

For pblum documentation: here
For some background informations, e.g. how the passband luminosity is computed -- Prsa+ (2016) paper: here

[kittipongwangnok]

Thank you for your suggestion, The real filter that we used is ASAS-SN- g, and ASASSN-V, but Phoebe does not provide that filter. We have to find an acceptable filter that can use instead of that filter.
However, we have tried with two datasets that were taken in ZTF-g and r filters.
By following your code, I still got the same result.
The results are attached here.
Do you have an example of code to fit multi-band?

[abokon]

I think I found out why do you get zero fluxes. I made artificial dataset with some noise and I set a new bundle with large distance, ~few parsec. And voila, I get zero flux, and I can't get to work the optimization, because the pblum is too far away from a good estimate. In this case, the optimization won't work as far as I know.

Please, try to to set manually the pblum with a larger value, and check how large are the model fluxes are (e.g. with print function). I don't know whether are limits are for pblum. (Maybe Kyle or Andrei would know)

Therefore as a first step, before any fitting the dataset, you need to adjust the flux levels between the dataset and the model.

[kittipongwangnok]

Dear András,
Thank you very much again for your suggestions.
Maybe your suggestion might be one of the possible issues that the passband filter cannot handle with the large distance (my object is ~ 3 kpc). However, I am waiting to see any comments/suggestions from Kyle or Andrei to find a way to solve this problem.

[kecnry]

The pblum default of 4pi is defined to give a flux of 1 from each star (ignoring distortion and irradiation) at a distance of 1 meter. If you change the distance to a real physical distance, then the pblum will also need to be set to a real value.

[abokon]

Now I understand why do we have fluxes in order of 1e-12 W/m^2. 😆

I recommend here the strategy of "two step backward and one step forward". 😄

(1) Since you have so large distances, interstellar reddening may play crucial role here, so I would start with only one dataset, with lc01. So disable the lc02.
(2) Calculate the model win b.run_compute, and get the median of the model fluxes. With command e.g.:

median_model = np.median(b['value@fluxes@model@lc01'])

Also, get this for the dataset as well:

median_data = np.median(b['value@fluxes@dataset@lc01'])

Then calculate the ratio of this two number and multiply the pblum by this result. Its only an estimate, but more than nothing.

(3) Calculate a new model and check with plot.

If you have nonzero model and it's close to the dataset, this could be a good starting point for optimizer.

---

All in all a good strategy for fitting multiband set is that you should start with only one dataset. Later, when your model is satisfactory for this one dataset, you can move on by enabling the next one, and so on.

[kittipongwangnok]

Thank you very much for your help.

…

On Wed, Feb 5, 2025 at 12:05 AM abokon ***@***.***> wrote: Now I understand why do we have fluxes in order of 1e-12 W/m^2. 😆 I recommend here the strategy of "two step backward and one step forward". 😄 (1) Since you have so large distances, interstellar reddening may play crucial role here, so I would start *only one* dataset, with lc01. So disable the lc02. (2) Calculate the model win b.run_compute, and get the median of the model fluxes. With command e.g.: median_model = ***@***.***@***@***.***']) Also, get this for the dataset as well: median_data = ***@***.***@***@***.***']) Then calculate the ratio of this two number and multiply the pblum by this result. Its only an estimate, but more than nothing. (3) Calculate a new model and check with plot. If you have nonzero model and it's close to the dataset, this could be a good starting point for optimizer. ------------------------------ All in all a good strategy for fitting multiband set is that you should start with only one dataset. Later, when your model is satisfactory for this one dataset, you can move on the next one, and so on. — Reply to this email directly, view it on GitHub <#999 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ALO4C3AS7KCS7KTJAEN4II32ODXNJAVCNFSM6AAAAABWMM2Y72VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMBVG42TCNA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- Kittipong Wangnok +6684 828 6505 School of Physics Institute of Science Suranaree University of Technology