Galaxy Builder Results

Many thanks to all the volunteers who helped re-construct all our galaxies in Galaxy Builder, we'd love to share some of our results with you!

For this example, we'll use galaxy UGC 4721, a two-armed barred spiral galaxy.

From citizen scientists we received 32 classifications for this galaxy, containing a total of 28 discs, 24 bulges, 17 bars and 57 drawn spiral arms:

We have a very noisy set of components, but a definite consensus we can extract and use! Using unsupervised clustering techniques we remove components that do not agree well with the same component in other classifications, leaving us with 19 discs, 3 bulges, 4 bars and 29 spiral arms. We then use these to create a single aggregated model for the galaxy.

Determining what constitutes a good cluster of components is difficult; we've tried to find the best trade-off between clustering as many well-drawn components as possible and not including components that don't match the galaxy. The model we get from clustering looks like this:

We can now use this as a starting point to create a fine-tuned, final model of the light from the galaxy. We make use of the scatter in clustered components to penalize large changes from the starting model (i.e. if the clustered bars all have roughly the same axis ratio, we penalize the fit for moving away from that value). The results of this fine-tuning can be seen in this figure:

The galaxy is shown in the top-left panel, the final fitted model in the top middle panel, and the difference between the two in the top right. The bottom panels show how the components have shifted during fitting; the disk has grown slightly larger and more elliptical, the bar has become narrower and the spirals have shifted slightly and now fit the galaxy a bit better. The lack of a clear structure in the residual is a fantastic sign that the fit has progressed well, all without any explicit manual guidance from the researchers (once the method and hyperparameters were chosen)!

We can also double-check that the light from each component has been accounted for correctly, by subtracting them one at a time:

Comparison to other results

It's difficult to compare results between models, especially for the galaxies used in Galaxy Builder (which are all chosen to have spiral structure and are difficult to model using simpler light profiles). The easiest comparison is to check whether the sizes and shapes of components broadly agree with the simpler models. If we compare the aggregate disc, bulge and bar (without fitting) to results from computational fits performed by Kruk (2018), we see great agreement!

What's really important in this plot is that we have a strong distinction between disc, bulge and bar shapes and sizes. As well as being in broad agreement with Kruk (2018), our models contain extra information regarding the disk's spiral arms and don't have the same issues with the computational fit failing to find a realistic result.

This is a fantastic validation of your hard work; we're excited to be able to share a catalogue of complete galaxy models with the community soon!

The Affirmation Set

Right now, we are collecting classifications for a very small number of simulated galaxies we have created from your models, to calibrate our clustering and fitting code and make sure the models we publish are as close as possible to the galaxies they represent. Your help classifying these galaxies is hugely appreciated.

Earlier Posts Are shown below:

Spiral Arms

In our first post over on the Galaxy Zoo blog, we introduced a method by which we could use your classifications to examine a galaxy's spiral arms.

Being able to accurately identify spiral arm regions of a galaxy, and recover parameters for the shape of the spiral arm (which we'll talk about later), allows us to dig into the underlying physical mechanism behind what causes them to form.

Examples of these mechanisms include density waves (think cars in a phantom traffic jam), a mechanism called swing-amplification (which is a little different to what goes on at the local wine bar on a Tuesday), and spiral arms caused by the presence of a bar, or an interaction with a massive neighbour (both scenarios we can empathise with).

Each of these mechanisms has slightly varying predictions as to the number of and features of a galaxy's spirals, though we often need more information to distinguish theories, such as the shape of a galaxy's dark matter halo, or a measure of how its matter is rotating at different radiuses.

How you have helped

Using a similar method to the one described in our Galaxy Zoo blog post (linked above), we can group your drawn spiral arms for each of our galaxies and fit a smooth line to the result, for example:

Scientists often assume that a galaxy's spiral arms follow a certain mathematical profile, known as a logarithmic spiral. We can fit a logarithmic spiral to the data points in each of our clusters and compare the resulting arms to the smooth spiral fitted earlier.

One notable feature of a galaxy's spiral arms is called its "pitch angle". This is a measure of how tightly wound the spiral is; a pitch angle nearer zero means a tighter spiral (negative means the spiral is moving inwards). One nice thing about a logarithmic spiral is that its pitch angle is constant, meaning obtaining a value for an arm's pitch angle is fairly straightforward.

In this image, we've calculated a logarithmic spiral fit for each arm cluster in the above image, as well as the pitch angles for our smooth spiral line and the logarithmic spiral:

Apart from some weirdness where the smoothed spiral line starts misbehaving at small radii, the fits are consistent! What's more, if we compare the results to what is obtained using a rather clever algorithm called SpArcFiRe (Davis & Hayes, 2014), it returns an average pitch angle for all arms in the galaxy of (27.1 ± 4.34)°, which is very much in agreement with our result!

However, not all galaxies seem to contain logarithmic spiral arms. This is where the assumptions used by automated fits will break, and citizen scientists can identify when this is the case.

So how does this tie back to galaxy's physics? The fact that the spiral arm we measured is consistent with a logarithmic spiral and ties in very well with the density wave theory of spiral arm formation. We can also combine our measured pitch angle with the galaxy's rotation curve (obtained through MaNGA or from H-alpha imaging) to test swing-amplification theory (which is high on our homework list at the moment).

The fact that we have obtained results consistent with a leading spiral arm detection and classification algorithm (which occasionally struggles with identifying "real" arms, Hart 2017) is a great sign that citizen scientists can help validate and constrain the increasingly complicated computer algorithms needed to deal with the diversity of gorgeous spiral patterns we see in the night's sky.

Results from the beta of galaxy builder

Over December of 2017 and into early January, 263 volunteers helped beta test Galaxy Builder. The classifications submitted as part of that beta showed promising results for examining spiral arms of galaxies; including their number, tightness and distribution (flocculent and broken up or more regular).

The drawn spiral arms for an example galaxy can be seen in this image:

By grouping the lines drawn by volunteers and fitting a smoothed line to the result, we can recover the underlying spiral arm shape.

This step has not yet been attempted on real data, but the output of a "proof-of-concept" run on noisy test data can be seen here (with many steps missing, the full pipeline can be seen here):

Image showing recovered spiral for test data. The top left panel shows the t vs x(t) for the parametrised spiral, where t runs from 0 to 1. The top right panel is the same for y(t). The bottom panel shows the recovered spiral (orange), the original spiral(green), and the points identified as being part of the spiral, ordered along the spiral (blue).

Once "recovered" spirals have been obtained for a population of galaxies, we can make use of the predictions of various theoretical models of how spiral arm form, as well as the result of simulations.

These comparisons allow us to work towards identifying the driving mechanisms behind the vast number of spiral galaxies present in the sky, a mystery which has eluded us since William Parsons first observed the whirlpool galaxy with the telescope nicknamed the "Leviathan of Parsonstown" over 170 years ago.