Exploring How Galaxies are Transformed

Fig 1: Spiral galaxy M74. Image
Credit: NASA

When we look at galaxies out in the universe, we find that they come in many different types. Some galaxies have beautiful spiral structure (see Figure 1), while others look like irregular blobs of stars and gas. Still others look like featureless spheres of light (see Figure 2). These galaxies aren't only different in appearance, however. We find that we can separate galaxies into broad classes based not only on their shape (or morphology), but also on their stellar mass and how quickly they are forming stars (their star formation rate, or SFR). We find that galaxies with disky morphologies, such as the spiral galaxies mentioned above, tend to be relatively star-forming compared to galaxies with more elliptical morphologies, which appear smooth, round, and featureless and are often no longer forming stars.

Fig 2: Elliptical galaxy ESO 325-
G004. Image Credit: NASA

Since morphology and star formation rate often appear to be correlated in this way, it has been suggested by many that perhaps the processes responsible for shutting off star formation in galaxies are also associated with the formation of an elliptical component, called a "bulge." One such process for shutting off star formation is AGN feedback, which is the name for when a supermassive black hole at the center of a galaxy affects the galaxy around it. When a supermassive black hole accretes material, large amounts of energy are released from the regions near the black hole, which can then heat up or drive out gas from the surrounding galaxy by launching winds or relativistic jets of plasma. The gas that is driven out or heated up is then no longer available to form stars, so the galaxy becomes "quiescent," which is the term we use for galaxies which have stopped forming stars.

Fig 3: Artist's rendition of a galaxy with
AGN-driven outflows. Image Credit:
ESA/ATG medialab

So how does the morphology of the galaxy change, and what triggers the AGN feedback? Here we rely on galaxy mergers and disk instabilities to drive material toward the center of a galaxy in order to both build a bulge component and feed the central supermassive black hole. During a galaxy merger, gas will be driven toward the center of the merger remnant, whereas a disk instability will lead to material being moved to the center of an isolated disk galaxy. In either case, the result is a galaxy with a significant bulge component that is no longer forming stars.

Fig 4: Galaxies in three different redshift bins being

split into the four quadrants of the specific star formation rate-

morphology plane. On the left are galaxies from our model

and on the right are observed galaxies. The greyscale 2D

histogram and contours indicate the density of galaxies

across the plane.

In order to test these ideas, we implemented a merger and disk instability-based AGN feedback prescription in our semi-analytic model (SAM) of galaxy formation and evolution in order to see how well we could reproduce the fraction of galaxies that are star-forming and disk-dominated (SFD) or quiescent and spheroid-dominated (QS) as compared with data from the CANDELS survey (as well as a local sample of galaxies from the GAMA survey). SAMs are a type of simulation which model large numbers of galaxies over the history of the universe. Our SAM evolves a cosmological sample of galaxies forward in time with relatively simple prescriptions for physical processes like the hierarchical growth of structure formation due to the merging of dark matter halos, the heating and cooling of gas, star formation, stellar evolution, supernovae, chemical enrichment of galactic and intergalactic gas, AGN feedback, and starbursts and morphological transformation due to galaxy mergers and disk instabilities. We divided galaxies based on their specific star formation rates (star formation rate divided by stellar mass) and their Sersic index, which is a measure of morphology. A Sersic index of 1 indicates a pure disk, while a Sersic index of 4 indicates a pure bulge. The distribution of galaxies in this plane, as well as our dividing lines for a few of our redshift bins, can be seen in Figure 4. By focusing on this plane, we also found ourselves studying the more "outlying populations": star-forming and spheroid-dominated (SFS) and quiescent and disk-dominated (QD). These populations are more rare but must still be explained by our evolutionary models.

Fig 5: The fraction of galaxies in each of the four populations.
The solid black line represents the observations, while the dashed
red line represents our primary model which includes AGN feedback
and bulge formation triggered by both mergers and disk instabilities.
The dotted blue line represents our model which only includes mergers.

In Figure 5, we can see the evolution of the fraction of galaxies in each of these four populations for both our model and the observations. Our model, which includes disk instabilities as a driver of bulge formation and AGN feedback, reproduces the fraction of SFDs and QSs much better than our model with a merger-only picture. Meanwhile, we reproduce the rough fractions of SFSs and QDs, although we do not quite match how the fractions evolve.

Our model suggests that SFDs are galaxies which have had very quiet histories; they've avoided major mergers and if they have ever been disturbed, they were able to accrete new gas and continue forming stars. QSs, on the other hand, are very likely to have undergone at least one major merger, or perhaps very many minor mergers, which built up a large bulge component and triggered AGN feedback, eventually leading to the cessation of star formation. SFSs in our model are a very short-lived population, the result of a recent merger which has led to bulge formation and a post-trauma starburst. These are likely soon to experience AGN feedback which will transform them into QSs. Finally, QDs are the result of SFDs which have stopped accreting new gas (perhaps due to environmental effects) or are very large and extended, causing their gas not to be dense enough to form stars.

While we do not match the evolution of these populations exactly, it seems we are beginning to be able to capture the very complicated processes responsible for the diverse galaxy population we see all around us.

Don't Judge A Galaxy By Its Cover

As discussed in previous posts (e.g. here and here), the morphology of a galaxy is a powerful tool when it comes to understanding galaxy evolution. The Hubble Tuning fork diagram, first devised by Edwin Hubble in 1925, shows that galaxies have three different types of visual morphologies: spheroidal (or elliptical), disky (or spiral), and peculiar (or irregular).

The Hubble tuning fork diagram. On the left are spheroidal galaxies, also known as elliptical galaxies. The number which follows the 'E' tell us if the galaxy is circular or flattened. On the right of the diagram are disk galaxies, often called spiral galaxies due to their spiral arm features. The 'B' in the notation for spiral galaxies tell us whether or not a bar feature is present, and the lower case letter tells us how tightly wound the arms of the galaxy are. The final galaxy on the far right is an example of a peculiar or irregular galaxy. 

On the left hand side of the Hubble tuning fork above are the spheroidal galaxies and the disky type galaxies are on the right. An example of a peculiar galaxy is shown on the far right. Not only does this image show us the different types of galaxies, it also shows us clearly that galaxies of different types have different colours. The spheroidal type galaxies are redder, while the disk and peculiar galaxies are blue in colour. If a galaxy is blue it is still young and making new stars. If a galaxy is red it is older and no longer has the fuel to make stars.

Rather helpfully, colour is not the only property that correlates with galaxy morphology. Galaxies that have a greater stellar mass tend to be spheroidal, while the lower mass galaxies have disk and peculiar morphologies. The massive, red, and old spheroidal galaxies are usually found in denser environments than disk and peculiar galaxies. Also, galaxy morphology is often measured using different parameters determined computationally, which in turn relate to the visual morphology of the galaxy.

Galaxies of different visual morphologies. Many of the important features which
allow us to tell these different types of galaxies apart have been revealed
to us thanks to the fantastic CANDELS imaging.

So does this mean that by simply looking at the shape of a galaxy you can tell all these different properties? The answer is not so straight forward. Many of the relations discussed above are what we know to be true in the local Universe, that is to say, they are true for galaxies close to us. As we begin to look at galaxies further away, we are looking at earlier epochs in the life time of the Universe. These galaxies are said to be at higher redshift than their local counterparts and may not have the same properties as those found close to us. Local galaxies have had the entire age of the Universe to form, whereas distant galaxies are younger and therefore are at an earlier stage of formation. Unfortunately, as a galaxy gets further from us, we need increasingly good telescopes to clearly see the visual morphology of a galaxy. A particularly difficult question to answer about distant galaxies has been 'What are the visual morphologies of these galaxies and how do they compare to the Hubble tuning fork?' Now, thanks to the high resolution CANDELS imaging, we can take a look at these distant galaxies and find out about their visual morphologies.

The fraction of galaxies with spheroidal (red), disk (blue) and peculiar (black) morphology as a
function of redshift. As redshift increases the distance away from us increases and the age
of the Universe decreases. This plot shows how the fraction of galaxies changes from
when the Universe was ~6 billion years old to when  the Universe was only 2 billion years old.

We can see from looking at these fractions that as we go further and further out, and the Universe gets younger and younger, the number of galaxies that don't fall nicely into the Hubble tuning fork scheme (the black line) begins to increase. This tells us that at earlier cosmic times the Universe was a messier place than it is today, and that galaxies are busy forming at these times. One puzzling result from this plot is that galaxies which look like Hubble type disk galaxies (the blue line) seem non-existent when the Universe was younger. Other CANDELS work, which has looked at the morphology of galaxies determined by other parameters (e.g. here), have shown some distant galaxies do have properties similar to the disk population in the local Universe. This seems to be telling us there are no galaxies with visual disk morphologies but galaxies which do have similar properties to disks. We can explain this when we consider results from other studies which show that disks in the distant Universe have clumpy structures. These clumps could cause the galaxy to be classed as a peculiar galaxy even though the underlying galaxy structure is that of a disk. In fact, studies have shown, through a process called Integral field spectroscopy (IFS), that such systems do exist. See this publication for a discussion of how IFS has been used to look at the underlying structure of a galaxy.

The results discussed here are important when considering the visual morphology of galaxies. It could be that disk galaxies in the early Universe are simply ‘hiding’ from us because they have different visual morphologies when compared to classic disks from the Hubble tuning fork. This is telling us that judging a distant galaxy based on our ideas regarding visual morphology in the local Universe may not be the best approach. Fortunately the CANDELS dataset has imaging available which is even deeper than what is used in this study, and will have visual classifications for many of these galaxies. These visual classifications come from projects such as Galaxy Zoo and the CANDELS team wide effort to visually classify all the CANDELS galaxies (this will be discussed in a future post). Data from CANDELS will play a key role in the continuing to investigate what visual morphology really means when looking at distant galaxies.

Making Statistical Inferences with Images: Using CANDELS Data to Learn About Our Universe

The data from the CANDELS program have been, unquestionably, a boon for astronomers, providing them with an unprecedented means by which to view galaxy evolution in the early Universe. But these data are a boon for statisticians as well. That's because of their complexity: going from, say, images to making inferences about the Universe requires the use of statistical techniques that can be considerably more advanced than simple linear regression. (And it may require new advanced techniques as well.) The application of the newest, most cutting-edge statistical techniques to CANDELS data: this is where I and my collaborators come in.

I am part of a small but growing class of astrostatisticians. We work at the boundary between astronomy and statistics, understanding the data and the science of the former, and applying the latest methods of the latter.  For my part, my Ph.D. is in astronomy, but for the last decade I have worked in Carnegie Mellon's Department of Statistics and with the International Computational Astrostatistics (or InCA) group, a collection of CMU statisticians and computer scientists as well as astronomers at CMU, Pitt, and places beyond. I work to facilitate communication between statisticians and their astronomical counterparts and, to borrow a phrase from computer science, I try to "abstract away" the complexity of astronomical data to make them more immediately accessible to statisticians, particularly to statistics students.

One major aspect of an astrostatistician's work is working with experts to internalize the nitty-gritty details of the science underlying a particular dataset. But another involves looking beyond these details towards the "big picture."  Are there newer and better ways to analyze a particular dataset, or to answer a particular scientific question? For the specific case of CANDELS, thinking about the big picture has led me to ask the following question: how exactly should we turn galaxy images into constraints on parameter values for cosmological and astrophysical models? Below, I'll describe how one might do this...and long story short: it won't be as easy as performing simple linear regression.

An illustration of the question posed in the text above: how do we turn galaxy images (in this case, the four representative H-band images of high-redshift galaxies from the CANDELS program shown at left) into constraints on parameter values for cosmological and astrophysical models (in this case, the 68- and 95-percent confidence regions for the values of two cosmological parameters shown at right)?  The pixels for each of the galaxies at left are colored according to relative intensity: for instance, we observe less light in a pixel colored green than we do in one colored pink, or red.  Yellow-colored pixels are associated with maximum galaxy intensity.  One might wonder about the two galaxies at left, where the yellow pixels appear offset from the image centers.  These are examples of disturbed galaxies, which are relatively common in the chaotic universe of 10 billion years ago: what we see as a single "galaxy" actually might be two galaxies that are completing the process of merging together.

The basic recipe for constraining the parameters of a given model (like the Lambda CDM model) should be familiar to anyone who has ever tried to fit a curve to a set of data: assume parameter values (like the slope and intercept of a line), construct predicted data based on these values (determine the amplitude of the line at each observed datum), numerically evaluate the difference between predicted and observed data (such as by computing the squares of the differences), and iterate, changing the parameter values until the difference metric is minimized. This recipe seems clear and easy to follow...but with CANDELS data the devil will be in the details!

The first devil-infused detail is "constructing predicted data." Our predicted data are galaxy images, and to map a set of parameter values to images we need state-of-the-art simulation software that takes into account all the intricacies of the physics involved in galaxy formation. This is a detail I leave to the experts; it suffices to say here that members of the CANDELS team including Joel Primack, Avishai Dekel, and others are on the case, combining cosmological simulations with code that models how stars evolve and how photons diffuse through dusty environments to create lovely images of galaxies. So while there is presumably still much to do to  perfect the simulation models, the construction of predicted data is not something that we should have to worry too much about in the long term.

The second devil-infused detail, "numerically evaluating the difference between predicted and observed data," is more problematic. How exactly does one compare (up to hundreds of thousands of) simulated and observed images? A direct image-by-image and pixel-by-pixel comparison is not appropriate, because of the random nature of simulations: we cannot expect a simulation to reproduce the structural details of galaxies that exhibit bars, arms, and/or clumps of star formation. Instead, what we want to do is to compare the entire ensemble of simulated images with that of observed images.  For instance, suppose that each galaxy image is a "postage stamp" measuring 100 by 100 pixels. One can use each of these pixels to define an axis, creating a 10,000-dimensional space; a given galaxy's value along the first axis would be, for instance, the intensity in the lower-leftmost pixel, while the galaxy's value along the second axis would be the intensity in an adjoining pixel, etc., etc. Once all the galaxy images are mapped to this space, we'd have two sets of points, for simulated and observed galaxies, respectively. Then all we have to do it compare them: how "far" is one set from the other? Eventually, we'd want to find the set of model parameters that minimizes that distance.

Before answering that, we should probably admit that working in a 10,000-dimensional space is actually infeasible: we would suffer from the so-called "curse of dimensionality," wherein the amount of data needed to achieve a precise statistical result generally grows exponentially with dimension. But all is not lost, because we wouldn't want to work in such a large space anyway, if we can avoid it.  We expect that galaxies "live" in some subset of that space, perhaps one of much lower dimensionality. For instance, all the galaxies might inhabit a line or curve running through our large space, or perhaps a plane. We don't expect the space in which the data live to be quite so low dimensional, but it may not be much more complicated than that.

The "noisy spiral": these data live along a
one-dimensional spiral embedded in a
two-dimensional space, much like we expect
galaxies to inhabit a low-dimensional subset
of the high-dimensional space defined by
image pixels.  From Lafon & Lee (2006).

There are various flavors of "dimension reduction" techniques that we astrostatisticians use frequently. Some are relatively familiar to astronomers, like principal components analysis, and others less so. Here we'll concentrate on one familiar to galaxy morphologists: using summary statistics. A summary statistic is simply a mapping from data values to a number, with the most familiar one being the mean, or average, of all the values in a dataset. A number of statistics have been developed for summarizing galaxy images in a way that retains important morphological information. (For instance, in Freeman et al.  (2013), we propose three new summary statistics that are useful for detecting disturbed morphologies, which the reader can think of as separating those galaxies with clumpy profiles from those that appear smooth.)

So, let's assume that we have summarized all of our simulated and observed images with some number of these oft-used statistics. Now, instead of two sets of points in a 10,000-dimensional space, we might have two sets in, e.g., an eight-dimensional space. That's more like it...but what do we do now?

One answer to this question involves the statistical concept of density. A probability density function, or pdf, is the probability of sampling a datum, per unit (multi-dimensional) volume. For instance, there is a pdf from which human heights are sampled: we can compute the probability of sampling a height between 66 and 68 inches by integrating this pdf over the range (66,68), and we would find that that probability is much larger than the probability of sampling a height, for instance, between 80 and 82 inches. To compare our sets of points, we could estimate the pdfs for each, then compute some metric that encapsulates how different the two pdfs are. Or even simpler, we could utilize work by my colleagues Ann Lee and Rafael Izbicki and base our metric on the ratio of pdfs at each cluster point, which can be estimated simply without knowing the pdfs themselves. We don't know yet what the optimal method will be, both statistically and in terms of computational feasibility. It's an open question, which is the kind of question statistics researchers like best!

So check back in a few years: maybe by then our group here at CMU, or some other group, will have solved the problem of making statistical inferences in an effective manner using galaxy images. Hopefully it will happen, and if it does, don't worry that astrostatisticians like me will be left with nothing to do. The great thing about astronomy is that there shiny objects everywhere that can grab an astrostatistician's attention. There will always be more data to analyze, and with these data there will always be more statistical problems to solve.

Were Galaxies Really Mature 11 Billion Years Ago?

Nostra Culpa

Yesterday, the European Space Agency put out a press release on CANDELS research that has been widely covered by various media. Sometimes the scientific message gets garbled in press releases, in spite of best intentions all around. In this case, the wording of the press release is pretty good about conveying the key message (although some of the subtleties get lost). But the gorgeous picture that accompanies the release is more wrong than right. Alas, there was a bit more "artistic license" taken with the press-release artwork than there should have been, and we failed to put the brakes on before the press release went out. The pictures of the galaxies marked as "11 billion years" were not taken from the CANDELS images, and are not at the right redshift (redshift z=2.5 corresponds to a lookback time of 11 billion years). If you would like to see what the images really look like, read on.

This is the gorgeous image that accompanied the ESA CANDELS press release. If only it were closer to the truth! The images further down in this blog post show what galaxies 11 billion years ago really look like in the CANDELS images.

What did galaxies look like 11 billion years ago?

Cutouts from Lee et al.
Galaxies on the right are
forming stars rapidly.
Galaxies on the left are
not. 

The basic conclusion of the papers on which this story was based is that we can start to see the dichotomy between star-forming galaxies being "disk like" and non-star-forming galaxies being "spheroid like" already being set into place 11 billion years ago. That's the central message and is what we are finding in the CANDELS survey in the various samples studied by Tao Wang, Eric Bell, BoMee Lee, Alice Mortlock, Victoria Bruce and others.

We wish we could see galaxies 11 billion years ago with the sort of clarity shown on the press-release image, but unfortunately even with Hubble we can't see that level of detail. To the extent that we can distinguish detail they look (a) smaller and (b) generally bluer, and (c) less well-ordered than present-day galaxies.

So -- yes we can see spheroids and disks emerging as separate sequences -- but it is not the case that the galaxies look like their present-day counterparts, which is the impression that you get from looking at the press-release image.  Arjen van der Wel's quote in the press release that they look "remarkably mature" is reasonable when you consider that they might have all looked like train-wreck mergers-in-process.  But some of the news stories based on the press release are taking that to mean they are completely mature, which is certainly not the case.

Illustration of the morphologies of massive galaxies at redshift z~2 using CANDELS cutouts from Wang et al. (2012). The upper-left insert color codes galaxies on the same axis scale, with blue being the star-forming galaxies and red being the less-star-forming galaxies. You can easily see that the less-star-forming ones tend to look more spheroidal.

The light reaching us from galaxies at a redshift z=2.5 took about 11 billion years to get here. So we are looking back in time to when the universe was only about 2.7 billion years old. Below and to the right are some figures from some recent papers by CANDELS and other teams that show images of galaxies at about this redshift. Perhaps this gives a more accurate impression of what we can say and can't say about the comparison between nearby galaxies and their distant progenitors.

A montage of "Milky-Way progenitor" candidates from the CANDELS images put together by Pieter van Dokkum and the 3D-HST team. 

What Did Galaxies Look Like at Cosmic High Noon?

The rate of star formation was highest about 10 billion years ago, a period that CANDELS astronomers call “Cosmic High Noon.” The redshift then was about 2, which means that the universe has expanded by a factor of three since then in each of the three spatial dimensions, so it was 3x3x3 = 27 times denser back then. The universe was also much brighter, with so many galaxies so much closer together forming lots of short-lived massive stars, which shine much more brightly than lower-mass long-lived stars like our sun.

What did those early galaxies look like, and how did they evolve into the galaxies we see around us today? Answering that question is one of the most important goals of the CANDELS survey. The infrared capability of the Wide Field Camera 3 (WFC3), installed in the last astronaut visit to Hubble Space Telescope in 2009, gives us the ability to see galaxies at redshift 2 in the wavelengths of visible light. Visible light, with wavelengths ranging from blue at 0.4 to red at 0.7 microns (a micron is a millionth of a meter), gives us crucial information about the long-lived stars in galaxies. The wavelengths of light emitted at redshift 2 expand by a factor of 3, just as space does, so visible wavelengths expand to 1.2 to 2.1 microns. WFC3 allows us to make images at wavelengths as long as 1.7 microns, while WFC3 and other HST cameras make images at shorter wavelengths that allow us to trace recent star formation because such ultraviolet light is emitted by short-lived massive stars.

Simulated galaxy at redshift 2.1 from a high-resolution cosmological simulation.  Top: rest-frame optical
image from the Sunrise computer code, taking in account stellar evolution and the scattering and absorption
of light by dust and subsequent dust re-radiation.  Bottom: The same simulated galaxy, as seen by Hubble
Space Telescope in V (visual light) and H (1.5-1.7 micron infrared wavelengths) bands.  Because of the
redshift of the radiation from this galaxy, what HST sees as V-band light was emitted as ulraviolet in the
galaxy rest frame, which mainly traces new star formation, while what HST sees as H-band light was emitted
as red light, which traces the older stellar population in the high-redshift galaxy.  Note that the V-band image
is clumpy, which is also often the case for real galaxies at these redshifts. Image Credit: Joel Primack

One of the things that we have found is that star-forming galaxies at redshift 2 were often rather clumpy, unlike the rather smooth Milky Way and other nearby galaxies. My colleagues and I have been simulating the formation and evolution of galaxies, and our simulations often also look rather clumpy, with giant star-forming regions in their disks. The clumps occur partly because the galaxies have so much gas in their disks that the disks become gravitationally unstable and break up into clumps of gas that rapidly form stars. We have been comparing the observed and simulated galaxies systematically, and we have been gratified to find that they appear fairly similar in their sizes and shapes, as well as their clumpiness.

Big galaxies today are a combination of stellar disks and spheroids. The Milky Way is a disk galaxy, with a relatively small spheroidal bulge. The most massive nearby galaxies are giant ellipticals, essentially all spheroidal. Most stars appear to form in galactic disks, but today most stars are in galactic spheroids. How did that transformation occur? The standard answer is that when comparable-size disk galaxies merge, their beautiful disks are turned into spheroids as the disk stars go every which way.

In addition to simulating galaxies, my colleagues and I have also run the Bolshoi simulation, the highest resolution large cosmological simulation yet. Using the Bolshoi simulation, we have found that there are not nearly enough galaxy mergers to account for the large number of galactic spheroids. But we have also found that the clumps can rapidly merge onto the galactic centers, which may well be the most important mechanism for forming galactic spheroids. In order to see if this true, and to discover other secrets of galaxy formation, we are running many more galaxy simulations and comparing them ever more systematically with observations. We have also just finished running a new version of the Bolshoi simulation based on the precise new cosmological parameters released by the Planck cosmic microwave background satellite team on March 21, 2013.

Key collaborators on this work: Most of the galaxy simulations were run by Daniel Ceverino, now a postdoc in Madrid. At UCSC my grad student Chris Moody converted them to realistic images taking into account stellar evolution and dust, using our Sunrise computer code; Priya Kollipara measured the sizes and shapes of these images; and Mark Mozena compared them systematically with CANDELS observations under the direction of David Koo and CANDELS Co-PI Sandra Faber.  Avishai Dekel -- with his postdocs Adi Zolotov and Dylan Tweed, his student Nir Mandelker, and others -- has analyzed these simulations and clarified the nature of the disk instabilities that lead to clumps merging to form spheroids. Anatoly Klypin ran the Bolshoi simulation; Peter Behroozi, Sebastian Trujillo-Gomez, and I helped analyze it; and Lauren Porter and Rachel Somerville worked with me on using the Bolshoi simulation to make galaxies semi-analytically.

Live Fast, Die... Small: The Fall of the First Massive Galaxies

In previous blog entries we talked about how galaxies are found preferentially in two different states. Either they are awake, actively forming a lot of young stars, or they are asleep, having very little or no star-formation activity. This bimodality is reflected in their observed colors, which are either blue or red, but also in their morphology. Galaxies that are awake usually have large spiral structures, just like our own galaxy, whereas asleep galaxies have elliptical or spheroidal shapes.

Hubble Space Telescope color-composite images of 4 clumpy star-forming
galaxies at z~2 in the CANDELS survey. The disturbed appearances are
very common in star-forming galaxies in the adolescent Universe.
Credit: G. Barro

Using data from CANDELS and previous surveys with the Hubble Space Telescope, we have found that these morphological differences were also present when the Universe was much younger. However, the most active galaxies in the adolescent Universe were nothing like our Milky way. Many of them lacked a characteristic spiral structure, showing heavily distorted clumpy appearances instead. Most asleep galaxies, on the other hand were already round, smooth, and remarkably small as soon as 3 billion years after the Big Bang. This implies that, by the time galaxies go to sleep, running out of fuel to keep forming stars, they must have experienced a significant structural transformation, sometimes involving a substantial shrinkage in size.

Interestingly enough, these transformations seem to be taking place in large numbers at the epoch of maximum cosmic star-formation activity. This value is the average of the rate at which galaxies are forming stars in a given epoch of the Universe. So essentially, this tells us that some galaxies that decided to go to sleep when most of their companions were having the most fun of their lives.

Galaxies evolve from large unstable structures to round spheroidal
shapes before they finished their star-formation activity turning into
small red-nuggets. Credit: Barro et al. (2013)

In our recent CANDELS paper, we studied a large sample of massive galaxies at redshift z~2 (when the Universe was only 1/3 of its current age) to investigate how these processes took place. Our results suggest that the structural transformation happens first, meaning that actively star-forming galaxies evolve from large clumpy systems into compact spheroidal shapes and then they doze off. But what phenomenon is responsible for this extreme change? In previous entries we have talked about how galaxy mergers can completely transform the structures of the galaxies involved, producing an elliptical-shaped remnant. Mergers are also important because they can redistribute the gas of the progenitors to feed the black holes living in center of these galaxies, triggering and active galactic nuclei (AGN) which may play an important role in preventing the galaxy from forming new stars. In our analysis we find that many of these compact star-forming galaxies are emitting in the X-rays, a signature for the presence of an AGN, yet we still have to determine if AGN are causing the galaxies to go to sleep or if they are just spectators.

An alternative possibility for the cause of galaxy shrinking is self gravitational collapse. Unlike the galaxy disks observed in the local Universe, the clumpy star-forming galaxies at z~2 are not stable systems. The location and duration of these clumps is determined by the inflow of gas coming into the galaxy from the dark matter halo that they inhabit. Instabilities associated with this accretion process can cause the clumps to collapse and coalesce into small compact remnant.

There are different scenarios to interpret the transformation and subsequent truncation of the star-formation in compact galaxies. The challenge now is to combine observational results with theoretical models to identify the precise nature of this process. So stay tuned as we uncover new pieces of the puzzle.

Supernovae Part II – Supernova Environments

In a previous post, we introduced readers to the exciting field of supernovae. Today we are going to delve a little deeper and discuss the environments where supernovae explode. These environments are interesting because they can provide information about the nature of the stellar systems that produce supernovae.  This can help us understand how and why some of the brightest events in the universe are produced.  

Type Ia supernovae are known as “standardizable candles”. Without going too deep into the science, a "standard candle" is an object whose observed brightness only depends on how far away it is. If you hold a flashlight 5 feet from your eye, it looks much brighter to you than if you hold it 100 feet away. By knowing the difference in brightness, you can estimate the distance to the flashlight. Now picture a flashlight with dying batteries. If you only see the flashlight far away, and have no idea how bright it would be up close, you can't estimate the distance. But if you somehow know how much power is left in the batteries, you can use that information to correct your distance estimate. In a nutshell, this is a "standardizable candle". The distance to a "standardizable candle" is not only based on it's observed brightness, but also on intrinsic properties of the object. For supernovae, what this means is that by measuring certain properties of the explosion through our observations, we can estimate how bright it should be if it were nearby, and therefore we can calculate how far away it is. 

In one of the most astounding tenets of astronomy, looking at objects that are farther away is analogous to looking back in time, because the speed of light is finite. It takes a long time for the light from these supernovae and their host galaxies to reach us. Finding supernovae that are farther away therefore allows us to trace the history of the expansion of the universe. The 2011 Nobel Prize in Physics was awarded for the role of type Ia supernovae in discovering dark energy, the mysterious force that is driving the acceleration of the expansion of the universe. Several CANDELS team members played key roles in the discovery, and team member Adam Riess was one of the prize recipients.

Recent research suggests that the brightness (and therefore our estimated distance) of a type Ia supernova depends in some way on the properties of it’s host galaxy. This suggests that the explosion mechanism for type Ia supernovae depends in some way on the surrounding environment. In the CANDELS Supernova Survey, we are searching for supernovae from a time when the universe was only 3 or 4 billion years old (we have measured the age of the universe to be around 14 billion years old). Galaxies from the early universe do not look like our own Milky Way galaxy; they are smaller, bluer (because young, hot stars are blue), and less polluted with heavier elements such as iron which are produced in the interior of stars and in supernovae. Studying the environments of these very far away supernovae is therefore important for tracing the expansion history, but also in our basic understanding of the systems that produce type Ia supernovae.

In the rest of this post, we will show some of the diverse galaxies that supernovae have been discovered in by the CANDELS supernova team.

CANDELS images of supernova host galaxies: images a), b), c), e) and f) are 0.004 degrees on each side. Image d) is 0.008 degrees on each side. For reference each image is approximately the size of a US quarter dollar viewed at 1/4 of a mile away.

a) This galaxy is one of the most nearby supernova hosts that we’ve discovered in CANDELS. The light that reached the Hubble Space Telescope to produce this image was emitted around 2 billion years ago. It is a fairly typical “edge-on” galaxy; we are viewing the galaxy right along the plane of the disk. If you look closely, this galaxy has a faintly visible dust lane; this material blocks the light emitted behind it and is therefore slightly darker.

b) This is a similar looking galaxy, but slightly farther away. The light in this image was emitted about 4 billion years ago.

c) Likely a similar shape galaxy to the previous two galaxies, however this one is rotated 90 degrees so that we’re viewing it from a very different angle. This galaxy is also quite a bit farther away; the light you are looking at was emitted 7 billion years ago! The blue clumps are likely bright, blue, massive, young stars, indicating significant recent star-formation. The supernova that went off in this galaxy may be a core-collapse supernova (core-collapse supernovae are much more likely to be found in blue regions like the ones in this image) instead of a type Ia supernova. A core-collapse supernova is the death of a massive star; when the star runs out of nuclear fuel, the outer regions of the star collapse onto a very dense, compact core, and the rebound of this material causes the explosion that we observe. The CANDELS supernova team is still working on the classification of all of our discovered supernovae.

d) Don’t be distracted by the galaxies to the left; the compact object in the center is the host galaxy of a supernova that exploded 6 billion years ago. This galaxy is observationally very different from the first 3 hosts. It is likely much less massive, and though it is difficult to draw conclusions about current star formation from just this image, it is not very blue, implying that it contains alot of fairly old stars.

e) Again not to be distracted by surrounding objects, the faint blue galaxy in the center of this image hosted a supernova 7 billion years ago. The size and shape of this galaxy is similar to d), but this galaxy is much more blue. It is highly likely that this galaxy has had more recent star formation.

f) These three galaxies all appear to be at the same distance from Earth, and so we believe they may be interacting, merging together at some point to form a larger galaxy. We are seeing them as they were about 8 billion years ago. A supernova that exploded in one of these galaxies is very likely a type Ia, and the environment appears different from some of the nearby type Ia supernovae. 

The CANDELS supernova team is hard at work classifying and analyzing the supernovae that we are discovering. You can see the diverse nature of galaxies that host supernovae; while star-forming (blue) galaxies are more likely to host a core-collapse supernova than old, red galaxies, this does not mean that type Ia supernovae cannot explode in blue galaxies. There are many questions that remain to be answered regarding the role of the host galaxy environment on type Ia supernova explosions. Understanding the environment of each supernova will be an important tool in constraining the nature of type Ia supernova progenitors. As the supernova environment may evolve as we discover them farther and farther away, understanding the role that the host galaxy plays in our distance estimate is a goal of the CANDELS supernova team.

Exploring Galaxy Evolution with CANDELS Morphologies

The most massive galaxies in the Universe are important probes of galaxy formation as they provide insight into the physical processes which govern the evolution of galaxies at the extreme high mass limit. By studying these most massive systems across cosmic time we can provide rigorous tests for our understanding of how mass is assembled in the Universe.

Example 6x6 arcsecond image stamps of the
bulge+disk decomposition of one of our
objects with significant bulge and disk
components. The residual image illustrates
the goodness of fit of the combined model.

It has recently been discovered that massive galaxies at z>1 have surprisingly smaller sizes than similarly massive galaxies in the local Universe and in the most extreme cases are up to a factor of 4 times smaller. In fact, it has been shown that the most compact of these high redshift galaxies also display the least amount of ongoing star-formation, which raises the question of how such compact systems at z>1 can grow to reach the sizes of comparably massive local galaxies if they have very little continued star-formation. There are several proposed mechanisms which may explain how these galaxies can grown in size, with very little mass increase from high redshift to the present day, such as through minor merger events or from AGN driven gas expulsion, which causes the system to expand. The exact physics responsible for this required growth remains debated, but it is clear that the morphologies of these massive galaxies can provide us with important information about their evolution.

In the local Universe galaxy morphologies can be classified by the Hubble sequence and they display a well-known correlation between colour and morphology, with spheroidals being predominantly red in colour due to the fact that they have very little ongoing star-formation and, conversely, disks being blue, but at higher redshifts the case is more complicated.

The unparalleled high resolution H-band data from CANDELS has allowed us to conduct a detailed study of  ~200 of the most massive galaxies with M>$1x10$¹¹ (i.e. 100 billion) solar masses at 1<z<3 (when the Universe was 1/2 to 1/6 its current age) in the UDS field, where we were able to decompose the rest-frame optical morphologies of galaxies into their separate bulge and disk components for the first time at these high redshifts for such a large sample size (see the CANDELS paper here). By conducting this decomposition we were able to explicitly explore how the sizes of the different components evolve within this redshift range, and compare this to studies in the local Universe. In doing so we found that the bulge components appear to display a more dramatic evolution in size than the disks.  This can be seen in the figure below both from the number of bulges which have sizes significantly smaller than objects in the lower Universe, and in the difference between these sizes, which is more extreme for bulges than disks. 

Size-mass relations for the separate bulge and disk components. Left: bulge components over-plotted by the red solid line with the size-mass relation found in the local Universe for spheroidal galaxies, and by the red-dashed line which is the corresponding error on the local relation. Right: disk components over-plotted in blue by the size-mass relation found in the local Universe for disk galaxies. From these plots we can see that while some bulge components lie on the local relation, the majority of them fall below it, whereas for disks a larger fraction are consistent with the local relation, and for those components which fall below the difference in size with local values is not as great as for the bulge components.

In addition to how massive galaxies evolve in size, decomposing objects into their separate bulges and disks also allows us to explore how the overall morphologies of galaxies evolve with redshift. In the local Universe, the majority of massive galaxies are pure bulge systems, but from this study we found that not only do massive galaxies become increasingly mixed systems with significant bulge and disk components with higher redshift, but that by z~2, they have predominantly disk-dominated morphologies. This suggests that not only is 1<z<3 a crucial era in cosmic time when global star-formation in the Universe peaked, but that is also marks a key phase in morphological evolution, where galaxies undergo a dramatic transformation from disk to eventually bulge-dominated systems.  

The redshift evolution of the morphological fractions in our galaxy sample, after binning into redshift bins of width z = 0:5, using three alternative cuts in morphological classification.

In previous CANDELS posts the issue of what triggers the switching off of star-formation in galaxies to make them passive has been discussed, and it has been suggested that while the presence of a prominent bulge may best correlate with passivity, some passive galaxies with significant disks have also been observed. By utilising our decomposition of objects into their separate bulge and disk components we  directly addressed this question by including star-formation activity for the objects in our sample. When we did this, we found that the majority of star-forming galaxies are disks, and passive galaxies are bulges, but interestingly, a significant fraction (~40%) of passive galaxies have disk-dominated morphologies, i.e. where less than 50% of the total light from the galaxy is contained in the bulge, where the advantage of our decomposition technique allows us to explicitly assess how much of the overall light from the system is associated to the different components. Moreover, we also found that some of our passive galaxies appear to be pure disks. As discussed in previous posts, this suggests that while the traditional star-formation quenching (i.e., shutting down) mechanism of major mergers may indeed be important for some massive galaxies, there may also be additional physical processes which can quench star-formation in a galaxy but leave a massive disk intact.

The next step in our work is to extend our analysis to the CANDELS-COSMOS field to allow greater area coverage, and we are currently implementing a new technique to extend our decomposition of bulge and disk components to the other 3 CANDELS bands (F125W, F814W and F606W) in order to provide photometry for separate components to conduct individual SED fitting, with the aim of generating separate stellar mass and age estimates for the different components. This will add an extra dimension to our morphology decompositions and shed new light on the properties of high redshift massive galaxies.