Space Warps Talk

ASW00024id modeled results Convincing

  • JasonJason by JasonJason

    Model results for ASW00024id: Convincing by Tom Collett(Scientist) here

    Model results for ASW00024id: Convincing by Me here


    Posted

  • psaha by psaha scientist

    Two more models, one a small variant of the model by @Tom_Collett and the other swapping the spaghetti around.

    @C_cld asked whether opposing models indicate an imposter. Not really, in my opinion. If a system looks clean and yet is resistant to modelling -- ASW00033g4 is such an example -- then I would suspect an imposter. But if there are different plausible-looking models, we can still be optimistic. Of course, of mutually disagreeing models, at most one can be on the right track.

    Posted

  • Tom_Collett by Tom_Collett scientist

    FYI This was my first play with spaghetti lens, so my model shouldn't be treated as the gold standard!

    I agree with @psaha that the presence of multiple plausible models doesn't imply an imposter. Spaghetti lens is quite a simple prescription for lens modelling; if we try to fit a lens in a more quantified way there will be a clear preference for one model over the other. This is a problem for the science team though - really hardcore lens modelling takes a lot of time, and a lot of computer power. With SpagLens we just want a spaghetti model that looks as close as possible to the real light distribution, to check if the lensing hypothesis is reasonable. Be wary though - the predicted light needs to look VERY similar to the observed light for the model to be convincing (or plausible to be honest). Remember lenses are rare; we have a strong prior belief that an image isn't a lens, so we need good data to convince ourselves that a lens is present.

    Posted

  • psaha by psaha scientist

    Here's a summary of what SpaghettiLens does:

    It searches through possible distributions of mass that would give maxima, minima and saddle points of the wavefront at the places specified. The mass distribution needs to be non-negative and look sort of galaxy-like, but otherwise can be fairly arbitrary. Setting the symmetric option (on by default) means the mass distribution will be the same if rotated by 180 degrees. A large number (nModels=200 by default) of such mass distributions are generated, and the average is presented. Algorithm details here.

    All the above is for a single pointlike source, such as the bright nucleus of a galaxy or (better still) a quasar. (If you saw the tutorial, recall that the wavefront originated from a single point.) For generating the synthetic image, SpaghettiLens then assumes the source is circular with a conical light profile. Changing the contrast on the synthetic image amounts to changing to changing the width of the cone.

    Thus, SpaghettiLens works very hard at exploring possible mass maps of the lensing galaxy, but is simplistic about the source and the synthetic lens. After all, it's the lensing galaxy and its dark matter and stuff, that we most want to research. The redshifts we are guessing at present, of course. Redshifts contribute a multiplicative factor to the mass, but won't change the model qualitatively.

    Possible extensions to SpaghettiLens:

    1. Smaller pixels on the synthetic image than the mass map.
    2. Allow multiple point-like sources. This is actually already implemented, but is not in the GUI yet.
    3. Fit a more general source profile. This is computationally straightforward, but needs the user to somehow delineate clearly what part is lensed source and what part is lens itself.

    All of these should improve the synthetic images a lot. They should help improve the models of the lensing galaxy too, but probably will make less of a difference.

    Posted

  • c_cld by c_cld

    Thanks very much @psaha for these explanations, and algorithm details.

    Don't know if you change simplex algorithm of pixelens tool to new Metropolis-Hastings algorithm in SpaghettiLens, but i think the improvement could mainly be on computational resources, right?

    Thanks also for reminding that the source used is a single point, importantly for highlighting lensed source images.

    Does your default choice of a galaxy lens half way to a source at a cosmological distance yield the lens mass without consequence on distance images from center in the reconstructed plot ?

    Thanks again. Cheers c_cld

    Posted

  • psaha by psaha scientist

    Thanks @C_cld for you kind words.

    Using incorrect redshifts changes the whole mass distribution by a constant factor. Since that factor can be trivially corrected at a later stage, I have just been using the default redshifts for the time being.

    Linear programming also plays a role in SpaghettiLens, or rather its modelling engine GLASS (written by Jonathan Coles). But the sampling is done much better in the new algorithm than in PixeLens.

    Picking up now from another thread which I'd missed earlier.

    1. Some form of source-profile fitting is on the todo list for SpaghettiLens. The lens model would still have to based on point sources (one or more). But fitting for the source after the lens model has been done would improve the synthetic images considerably.
    2. Read in and modify a saved model - this is a very good idea. We hadn't thought about it at all, but it sounds like it would be straightforward to implement.

    Posted