The problem of multi-component fitting

[PaulLiu666]

I fitted an elliptical galaxy using three series models and output a comprehensive best fit model. I calculated the radial brightness distribution of the original data and the best fit model, with label distributions of "data" and "ALL". Then I drew the radial distribution of the three components based on the output parameters, but found that one component was very prominent. I want to know how these three components are combined to obtain the best fitting model? Thank you for watching, and I hope you can help me.

[perwin]

Hi Paul,

For each component, Imfit constructs an image using the component parameters, then adds all the images together -- and finally convolves the result with the PSF, if any. (The latter is mathematically equivalent to doing PSF convolution on the individual-component images and then adding them together.)

If you want to see what the individual model components look like as images, you can run makeimage with the --output-functions option. Be sure to include the PSF image if you were using that in your fit; it will be used to convolve each model-component image appropriately.

Without any more details of your fit, I would guess that either

1. The convolved profile of sersic3 (assuming you used PSF convolution in the fit) will look different from what you plotted;
2. Something is wrong in your generation of the sersic3 profile.

Did you include PSF convolution in your fit? (And: what is the pixel scale?)

[PaulLiu666]

Thank you very much! The pixel scale is 0.05''/pixel. I did use PSF when generating the best fit model. But I did not consider the convolution of PSF when generating individual components using the output parameters. I will check PSF convolution. Thank you for your answer. This is very important to me!

[perwin]

Something I sometimes do is generate the convolved, individual-component images (makeimage --output-functions), then plot major-axis cuts through them, as well as major-axis cuts through the combined model image and the data image. Or, if your data profile happens to come from say ellipse-fitting, generate ellipse-fit profiles for the model images using the same x0,y0,ellipticity,PA parameters. This is perhaps easier and probably better than trying to convolve a 1D Sersic profile with a PSF-based 1D kernel.