Illustris - new images and science

Is the navigable explorer toolbar going to be related to the galaxies we're classifying? I suppose what I'm asking is how do we find out which galaxy we've just classified?

Posted September 22, 2015 11:33 AM

Very, very cool project, thanks for bringing it to GZ! 😃

I have quite a lot of questions, which I'll break up into several different posts. My questions have a rather wide range ...

with image realism as described in Snyder et al., (2014)

That's from the Illustris Galaxy Observatory webpage; there's no link. In the Results webpage there is only one "Snyder et al." paper, and it's dated 2015. Where can one obtain the relevant Snyder et al., (2014) paper?

I am keen to understand the details, not least because the Illustris webpages do not seem to explain many of the things mentioned in the GZ blog post (some bolded):

The Illustris scientists have created images of the galaxies from their simulation that GZ volunteers will classify by their morphology. Our comparison data set for this will be the SDSS results from Galaxy Zoo 2, and the images are designed to match the Sloan images as closely as possible. This includes the same set of filters for the telescope, sizing the images so that the galaxies look like they’re at cosmic distance from the Milky Way, and setting them against backgrounds of stars and other galaxies.

One quick question: are the colors in the simulated images Luptonized? Certainly the examples - both in the blog and on the Illustris site - do not look at all like Luptonized SDSS images ...

ETA: the Hubble UDF image - on the About page - and its Illustris sim both seem to be Luptonized.

Posted September 22, 2015 12:50 PM

Recent News on the main Illustris webpage stops at 2 April, 2015.

Is the latest version of GZ - using Illustris data/sims - going to appear there?

Posted September 22, 2015 12:52 PM

Do the Illustris sims include supernovae in the galaxies simulated? Or, can we expect to find bright point-sources in/near some galaxies, which may surmise are supernovae in those galaxies, and not background stars (in our own - not simulated? - galaxy)?

Ditto, re gravitational lenses (this would be exceptionally cool, if they are there!).

In SDSS, some Local Group galaxies are resolved into individual (bright) stars, star clusters, HII regions, and even dust clouds. Can we expect to see similar-looking objects in the Illustris sims?

Posted September 22, 2015 1:02 PM

by KWillett scientist, admin, translator in response to ChrisMolloy's comment.

Hi @ChrisMolloy,

Absolutely. When you go to the Examine page for an Illustris galaxy, you'll see a lot of parameters that have been measured (in the simulation) for that galaxy (or "subhalo", as it's called in Illustris terms). These include many of the same things we measure or infer for the real SDSS or DECaLS galaxies, like color, brightness, and size.

If you click on the first link below that data ("View data from Illustris Project"), you'll go to the main Illustris page which shows you the full suite of parameters that they measured about the galaxy. It's not as nicely formatted as the SDSS Explorer page right now, but there's lots of really useful stuff there, including many things that are extremely difficult/impossible to directly measure from an image of a real galaxy.

Here's a list describing what all the different Illustris parameters mean: http://www.illustris-project.org/data/docs/specifications/ It's rather technical, but please post questions if you'd like to know more. Thanks for classifying!

Posted September 22, 2015 5:19 PM

by KWillett scientist, admin, translator in response to JeanTate's comment.

Hi @JeanTate,

The Snyder et al. reference you mentioned should be the 2015 paper, I believe. It's not uncommon for scientific references to end up being pushed back by a year, depending on the lag between submission, acceptance, and actual publication. The relevant paper should be this one.

The color balance in the Illustris images is mostly driven by the asinh stretch that you mentioned, originally developed by Robert Lupton. Paul Torrey has made some adjustments to the GZ images, particularly in trying to balance the appearance of the blue and green blobs that many users have already noticed in the spiral galaxies. Those are a result of the way that light from very young stars are rendered in the simulation, and the first iterations didn't resemble most spiral galaxies as they appeared in SDSS. We've made some manual rebalancing of the colors as a result, although you can see that some of the blobs are still there.

Please classify them as you would an SDSS galaxy, including weird blobs if you see them; if those end up affecting your assessment of the morphology, that's important data that we can use to analyze either the image rendering or the physics that produce the sims.

Posted September 22, 2015 5:29 PM

by mlpeck

@KWillett:

I'll throw a few more questions at you.

Are we looking at a z=0 snapshot?
My recollection of ancient history is that it used to be notoriously difficult to produce long lived spiral structure in simulations. Has that problem been solved? I ask in part because I'm not seeing a lot of well organized spiral galaxies in either the Illustris PR material or the few that I've contributed clicks on so far.

I get that comparing the simulated spiral population to the GZ view of reality is undoubtedly part of the experiment, so feel free not to comment on my casual, small sample size observation.

Posted September 22, 2015 5:47 PM

by KWillett scientist, admin, translator in response to JeanTate's comment.

There are algorithms that simulate the effect of stellar winds and supernovae in the Illustris simulation, but it doesn't resolve what you'd think of as individual stars. Each "star" is treated in the simulation as a particle with a total mass of a few million solar masses. So the galaxy should still be subject to overall changes from simulated supernovae (like increasing the amount of metals spread throughout the gas in the galaxy, for example), but you wouldn't expect to see individual supernovae in an image.

For gravitational lenses - the galaxies are placed on real backgrounds, but I believe they've done the full relativistic radiative transfer coding that you'd need to get a lensing effect. If you see anything that looks like an arc or lens, though, please classify it as you would normally and post it in talk - I'd love to see better examples for the Help images!

The question about what you see resolved in the galaxies is a really good one. As I mentioned above, each star in the simulations is really an artificial particle a few million solar masses in size. The images that you see spread the light out from those particles over the area between particles (effectively, the "size"). The brightness of that light (in each of the bands, which determines the color) is set by the spectral energy distribution from the star. That replicates the effect of a large population of stars, and does include emission from HII regions. So the colors and brightness should match what you'd see in the galaxy, but it doesn't explicitly render those regions so that you'd expect to see it. Dust isn't fully implemented in the Illustris images: there's an empirical model applied to the host galaxy, but no individual clouds.

Example of the different ways you could render stars

Example of the different ways you could render stars in Illustris; from Torrey et al. (2014)

The last note is that there aren't any galaxies that appear super close; we've placed all galaxies at the same "virtual" redshift of z=0.05, so there won't be anything like Andromeda that shows up.

Posted September 22, 2015 5:51 PM

by KWillett scientist, admin, translator in response to mlpeck's comment.

Great questions, @mlpeck.

Yes - the galaxies here are from the z=0 snapshot of the simulations. The appearance of the images (apparent size and brightness) make them appear as if they're at a distance corresponding to z=0.05, which is right in the middle of the SDSS main sample. There shouldn't be any strong differences in the evolution of the galaxies over such a short timeframe, though.
That's a really good point - simulations vary strongly on that. In Illustris particularly, I know that I saw many fewer grand-design spiral galaxies than in Sloan (that's an empirical observation, but everyone's classifications will answer that question directly). You see some spiral arms in these galaxies, but they're generated through a mechanism called "swing-amplification". Many other spiral galaxies are thought to be generated through pressure waves in the interstellar medium. Illustris doesn't have the resolution in their simulations to fully replicate this at the scales required (more technically, they need a multiphase ISM and better equation of state); we believe that's the reason why features like star-forming rings are much more common, and spiral arms less so.

Posted September 22, 2015 6:16 PM

Thanks Kyle. I've got the paper, and am reading it; I'll have questions on it later.

It's not uncommon for scientific references to end up being pushed back by a year, depending on the lag between submission, acceptance, and actual publication.

I've noticed this, quite generally; unless there's a full bibliographic reference, or a link, it can be quite difficult - not to mention frustrating - to find the relevant paper (and be sure that it's the one being referred to). I find it, um, odd that the quality control of the Illustris website completely missed the discrepancy re Snyder+ (2014/15). Any way to give feedback, so they can fix it?

The color balance in the Illustris images is mostly driven by the asinh stretch that you mentioned, originally developed by Robert Lupton. Paul Torrey has made some adjustments to the GZ images, particularly in trying to balance the appearance of the blue and green blobs that many users have already noticed in the spiral galaxies. Those are a result of the way that light from very young stars are rendered in the simulation, and the first iterations didn't resemble most spiral galaxies as they appeared in SDSS. We've made some manual rebalancing of the colors as a result, although you can see that some of the blobs are still there.

While I've still classified only a few Illustris sims, I find the lack of green(-ish) objects quite noticable (pinks, strong reds, and purples also seem completely absent, but they are uncommon in standard SDSS images anyway).

Please classify them as you would an SDSS galaxy, including weird blobs if you see them; if those end up affecting your assessment of the morphology, that's important data that we can use to analyze either the image rendering or the physics that produce the sims.

Sure thing.

Like mlpeck, I too have noticed an almost complete absence of typical SDSS spiral structure, in the objects I've classified, in the ones other zooites have Commented on, and in the various Illustris sources (website, Snyder+ paper). Not only no grand design spirals - which are relatively rare anyway - but also nice multi-arm ones, 'loose' spirals, and even flocculent spirals (the 'patterning' of the (face-on) sims seems to me quite unlike that of real SDSS spirals).

Posted September 22, 2015 7:45 PM

Thanks Kyle.

The brightness of that light (in each of the bands, which determines the color) is set by the spectral energy distribution from the star.

I'm still reading Snyder+ (2014/5), but it seems these sims cannot realistically show strong emission-line objects, as they appear in SDSS ... so no Green Peas (or Beans), no voorwerpjes, no EELRs, etc, right? Why? Because the 'stars' do not simulate emission line parts of the IGM. Also, there are no QSOs in the Illustris sims, right (I haven't read enough to comment on AGNs yet)? Similarly, no intensely colored nuclei, due to bright NLRs and BLRs, because emission line regions are not modelled, right?

Dust isn't fully implemented in the Illustris images: there's an empirical model applied to the host galaxy, but no individual clouds.

You're right, the sims do include dust attenuation, but not any fine details of its distribution. That means we are highly unlikely to find any 'dust lane ellipticals', and in foreground spirals/background ETG overlaps no 'silhouettes' of dust lanes/arms, etc.

The last note is that there aren't any galaxies that appear super close; we've placed all galaxies at the same "virtual" redshift of z=0.05, so there won't be anything like Andromeda that shows up.

And that also means we are highly unlikely to see any Illustris sim galaxies with a rich globular cluster (GC) system (like that of M87, for example). Not only because at z=0.05 almost all GCs are invisible in SDSS, but also because Illustris does not (so it seems, based on what I've read so far) attempt to model, or track, GCs.

Posted September 22, 2015 8:01 PM

by JeanTate in response to JeanTate's comment.

Thanks Kyle.

Posted September 23, 2015 1:15 PM

I find the lack of green(-ish) objects quite noticable (pinks, strong reds, and purples also seem completely absent, but they are uncommon in standard SDSS images anyway).

Take AGZ00090cm for example. The SDSS background for this sim contains two objects (likely stars) which have very saturated colors. In the Illustris sims I've classified so far - and a random selection of others I've looked at (mostly in Comments) - I've yet to see any part of any simulated galaxy which has colors other than washed-out blue, yellow (mostly also washed out), or orange (and white, of course). I don't know how much of this is due to a difference in the factors used in the two - quite different? - stretches (i.e. a kind of artifact), and how much due to differences in the two SEDs (i.e. sim doesn't match reality all that well).

Thoughts?

enter image description here

Posted September 24, 2015 3:47 PM

by Peter_Dzwig

Question: in a lot of these images from Illustris there are blue spots very tight which you would normally think of as start formation. Is that the interpretation here or are they an artefact? (guess its for Kyle, this one)

Posted September 25, 2015 11:02 PM

by ElisabethB moderator in response to Peter Dzwig's comment.

Hi Peter

Here's a quote from Kyle from page 2 of this thread 😄

The color balance in the Illustris images is mostly driven by the asinh stretch that you mentioned, originally developed by Robert Lupton. Paul Torrey has made some adjustments to the GZ images, particularly in trying to balance the appearance of the blue and green blobs that many users have already noticed in the spiral galaxies. Those are a result of the way that light from very young stars are rendered in the simulation, and the first iterations didn't resemble most spiral galaxies as they appeared in SDSS. We've made some manual rebalancing of the colors as a result, although you can see that some of the blobs are still there.

Please classify them as you would an SDSS galaxy, including weird blobs if you see them; if those end up affecting your assessment of the morphology, that's important data that we can use to analyze either the image rendering or the physics that produce the sims.

Posted September 26, 2015 9:11 AM

Just checking:

The appearance of the images (apparent size and brightness) make them appear as if they're at a distance corresponding to z=0.05

Is this true of ALL the Illustris sims? If so, then differences in apparent size (in the image) correlate exactly with differences in the 'physical' (simulated) size, right?

Or is the image scale somehow adjusted to reflect changes in the apparent size, as was done in the original GZ (and several later too, IIRC)?

Posted September 27, 2015 2:30 PM

by KWillett scientist, admin, translator

The apparent size does vary depending on the galaxy size, as in previous versions of the project. The effect is lessened by the effect of placing them at the same distance, though, since that controls the pixel size. From Snyder et al. (2015):

For future visual classification projects, we also prepare images for classification by the Galaxy Zoo project (GZ, e.g., Lintott et al. 2008) in SDSS g, r, and i filters. Us- ing our initial radius (rP ) measurements as defined in Sec- tion 2.3, we re-bin our SDSS-like FITS images to a new pixel scale (0.008×rp) and create images with a fixed pixel count (424×424). These choices are such that the galaxy extent de- fined by 2rp always subtends ∼ 2/3 of the linear image size, enabling fair visual classifications as a complement to our fixed-scale non-parametric measurements below. For such visual classification projects, we also add real SDSS back- ground images to create fully synthetic ugriz galaxy images. To accomplish this, we first downloaded mosaics from the SDSS DR10 (Ahn et al. 2014) Science Archive Server with the mosaic web tool (data.sdss3.org/mosaics). From these, we randomly select a region of an appropriate size for each synthetic image, assuming the galaxies are at z = 0.05, and add it to the simulated galaxy image. We demonstrate these steps in Figure 1. This is a simplification from complete image simulations of self-consistent lightcones drawn from the simulation volume (e.g., Overzier et al. 2013; Henriques et al. 2012). We have created several examples of these sim- ulated fields from Illustris, and such techniques (Kitzbichler & White 2007) will become very useful as the volumes of such simulations grow.

Posted September 27, 2015 5:23 PM

Thanks Kyle.

Differences in apparent size, per our human eyes, are thus down to factors which include the 'radial' distribution of light (centered on the nucleus/photocenter?). For example, in a 1D fit using Sérsic profiles (which have been shown to correlate well with halo/DM mass, right?), higher n will result in the simulated galaxy taking up more of the image than lower n, right?

For example, AGZ0009ezp (left; log stellar mass 10.7) to me seems to be much bigger than AGZ000919n (right; log stellar mass 10.9), even though the images both have 2rP (the Petrosian radius) = ~283 pix:

enter image description here

Posted September 27, 2015 6:25 PM

by KWillett scientist, admin, translator

Yep - definitely correct on the concepts, Jean. The two images you posted confuse me a bit, though, since they should have distinctly different Petrosian radii. Via Examine:

AGZ0009EZP: 4.355 kpc
AGZ000919N: 7.540 kpc

Posted September 28, 2015 11:32 PM

by ChrisMolloy

A question Kyle.

Are all the images loaded in The Explorer field? AGZ00090b6 Simulation sub-halo 321411 is not coming up in The Explorer field when you use the search tab. The search icon is absent from the seach results tab.

enter image description here

Posted September 29, 2015 11:17 AM

by KWillett scientist, admin, translator

Hi Chris,

I'm not sure why it doesn't appear in the Explorer. I'd suggest posting this, along with the halo ID, on the Illustris comment form.

http://www.illustris-project.org/people/ (scroll to bottom)

Let us know what they say!

Posted September 29, 2015 6:31 PM

Let us know what they say!

Will do.

Posted September 30, 2015 9:46 AM

like http://talk.galaxyzoo.org/#/subjects/AGZ000acg3

Hi Kyle,

This is the response I got from Dylan Nelson from Illustris.

The Explorer shows a slice of the simulation at z=0, but the 'depth' of this slice is only 15 Mpc/h (a fifth of the total). Therefore, only ~20% of the GZ systems will be immediately visible in this image exploration tool.

Searching for a particular subhalo, you can still find it if you uncheck the "Only search visible slice?" checkbox. Or, you can also find it using the "Group Catalog Search" page or any of the other API tools.

We might someday expand the Explorer to include the entire box, and at multiple redshifts (and other simulations, and...), but the current incarnation was more of a proof of concept of what could be done.

Posted October 2, 2015 3:13 AM

by KWillett scientist, admin, translator

That's useful to know - some galaxies will definitely be visible, then, and we know why it's not available for others.

Posted October 2, 2015 6:57 PM

by firejuggler

the simulated illustris galaxy contain too much blue : most are easy to spot at first glance.

Posted November 3, 2015 12:58 PM

by firejuggler

ther is 2 blue region, at the opposite edge of the galaxy.. come on, really?

Posted November 3, 2015 1:01 PM

by Capella05 moderator in response to firejuggler's comment.

And that is one of the reasons we are classifying them - to see if the models / sims are comparable to real life galaxies 😃

Posted November 3, 2015 4:39 PM

by KWillett scientist, admin, translator

Exactly so - and part of the challenge of classifying these images is to separate the blue star-forming regions from the bulk of the stars that you see.

Posted November 3, 2015 8:00 PM

by dosey in response to KWillett's comment.

Another really good GZ blog post, from yesterday: “Blue stuff” in the Illustris galaxy images! 😃

As with some earlier blog posts, it ends with "Please post here or on Talk if you have more questions!"; here are some from me:

re "plot of the synthetic spectra for galaxies in the Illustris simulation": there are two (perhaps three) quite strong 'breaks'; one seems to be (H) Lyman limit, at ~92nm, the other the 400 nm 'cliff'. Is that so?
Ditto: the observer frame wavelength range, for the Illustris sims, corresponds to the SDSS g, r, and i bands, right? If so, then the Illustris sims will only sample ~400 to 820 nm, (or perhaps a bit bluer, due to the redshift of 0.05), right?
Ditto: for the most massive halos (Log M > 11.3), there seems to be an emission line, at ~280 nm ... is it real? If so, what is it?
Ditto: in the mid-range halos (LogM ~10.0-11.3), there seem to be a couple of absorption lines in the UV ... are they real too?
Re "Stars can only form in “chunks” of about 1 million solar masses, instead of the more typical small clusters and regions that we know exist in the real Universe. Moreover, these chunks have their light spread over a significant fraction of ~1 kpc (which is pretty big, compared to a typical galaxy radius of ~20 kpc), and so they often won’t look much like real star-forming regions": except for globular clusters (see next question), very few star clusters in late-type galaxies (LTG, a.k.a. 'spirals' or disk galaxies) are as massive as 1 million sols, aren't they? And not many are as big as ~1 kpc, right? Given this, is there any particular reason to expect that the distribution of sim star clusters will have much more than a chance correspondence to real ones?
Globular clusters (GCs): I asked about this before, and I don't think it was answered: how, if at all, are GCs modeled in Illustris?
Back to the synthetic spectra: all galaxies but the least massive show emission shortward of 92nm, including some (what look like) lines. Does this mean that the simulated galaxies are ~transparent to such (E)UV radiation? that they are all, to some extent, Lyman continuum sources? If so, that's very unrealistic, isn't it?

Posted November 12, 2015 7:45 PM

by KWillett scientist, admin, translator in response to JeanTate's comment.

Good questions. I'll answer the ones that I think I can, at least for the moment:

re "plot of the synthetic spectra for galaxies in the Illustris
simulation": there are two (perhaps three) quite strong 'breaks'; one
seems to be (H) Lyman limit, at ~92nm, the other the 400 nm 'cliff'.
Is that so?

The three strong vertical features seen in the spectra are, from left to right, the Lyman-limit (912 A), the Balmer break (4000 A), and hydrogen-beta (6563 A). The first two are broad drops in flux caused by bound-free transitions in hydrogen atoms, while the third is a discrete transition from n=4 to n=2.

Ditto: the observer frame wavelength range, for the Illustris sims,
corresponds to the SDSS g, r, and i bands, right? If so, then the
Illustris sims will only sample ~400 to 820 nm, (or perhaps a bit
bluer, due to the redshift of 0.05), right?

That's right. z=0.05 doesn't change very much from rest-frame to observed-frame wavelengths, though - we're still at ~380-780 nm range for our sample.

Ditto: for the most massive halos (Log M > 11.3), there seems to be an
emission line, at ~280 nm ... is it real? If so, what is it?

I'm not positive, although it'd be useful to look at some individual high-mass spectra (these are available on the Illustris site). There's a MgII (singly-ionized magnesium) doublet at around 280 nm, but it's typically seen in absorption for elliptical galaxies against background quasars.

Ditto: in the mid-range halos (LogM ~10.0-11.3), there seem to be a
couple of absorption lines in the UV ... are they real too?

See above. UV absorption features are ubiquitous in real observations, but do require a background source to be detected (usually a quasar). Among the most common are MgII (280 nm) and CIV (155 nm).

Re "Stars can only form in “chunks” of about 1 million solar masses,
instead of the more typical small clusters and regions that we know
exist in the real Universe. Moreover, these chunks have their light
spread over a significant fraction of ~1 kpc (which is pretty big,
compared to a typical galaxy radius of ~20 kpc), and so they often
won’t look much like real star-forming regions": except for globular
clusters (see next question), very few star clusters in late-type
galaxies (LTG, a.k.a. 'spirals' or disk galaxies) are as massive as 1
million sols, aren't they? And not many are as big as ~1 kpc, right?
Given this, is there any particular reason to expect that the
distribution of sim star clusters will have much more than a chance
correspondence to real ones?

Yes - as you quote from the blog post, these are larger sizes than the clusters we typically have seen in real observations. Since the GZ2 decision tree doesn't explicitly track the position of superstar clusters, though, it's not clear how it'll be reflected in the data (unless it decreases the overall likelihood of finding a late-type galaxy, or very large features mimic a small bulge). We'll see what your classifications tell us.

Globular clusters (GCs): I asked about this before, and I don't think
it was answered: how, if at all, are GCs modeled in Illustris?

There are some results on it - you might look at this paper if you're interested. http://arxiv.org/abs/1509.00030

Back to the synthetic spectra: all galaxies but the least massive show
emission shortward of 92nm, including some (what look like) lines.
Does this mean that the simulated galaxies are ~transparent to such
(E)UV radiation? that they are all, to some extent, Lyman continuum
sources? If so, that's very unrealistic, isn't it?

Emission doesn't mean they're transparent to UV radiation; some of that can be produced by massive star clusters near the edges of the galaxy. With nothing to absorb them (remember that Illustris doesn't fully take into account the effects of dust), you'd still expect some emission blueward of the Lyman limit.

Posted November 18, 2015 10:50 PM

Please can you tell me why we are asked to look at Illustris images? They just don't seem to be similar to the real images from DECaLS or SSD. Are they from a program that needs to be improved?

Posted November 19, 2015 6:45 PM