Institute for Computational Astrophysics

Previous Images of the Month 2015

January February March April May June  July August September October  

November 2015

click on image to enlarge

Galactic Archaeology: Understanding the origin of metals

Benjamin Thompson: visiting PhD student at Saint Marys University from the University of Central Lancashire (UK)

Like how Archaeologists study humanities past from studying artifacts left behind and their chemical traces, Galactic Archaeologists study the history of our own galaxy from the chemical traces of stars. Using data acquired from the modern day sky from the Gaia Satellite (position and kinematics), as well as the Very Large Telescope in Chile (spectrometry).We can learn many things about the formation history of our galaxy and the ongoing processes within it, for example from predicting the age and birth location of stars from a stars metal and kinematic properties and the environment of the birth location since stars with similar chemical properties were born in a similar location.

An aid to the understanding of the evolution of our galaxy is the use of high performance computer simulations, which model the kinematics and chemical evolution of stars within a simulated milky way like galaxy. An example of such model is illustrated on the left hand side of this image, showing the gas disk of a simulated galaxy face on, of which the chemodynamical model was developed by collaborators in the UK (Few et al 2012). On the right hand side if the Metallicity distribution function (MDF), a numerical abundance relation showing the distribution of stars with certain metallicity ([Fe/H]) abundances (Where [Fe/H] is the logarithmic measurement of the Iron to hydrogen fraction in relation to our Sun, where an [Fe/H] of 0 represents the same iron content as the sun). This MDF (entitled Selene-CH after the name of the galaxy) can be compared to that of the Gaia-ESO survey Data Release 2 and along with studies of the local solar neighborhood (Fuhrmann 1998 and Ruchti et al. 2013). But also we can compare both of these with simulated observations - by putting a simulated observer in the plane of the simulated galaxy and projecting the stars onto a synthetic color magnitude diagram, we can compute a metal abundance of a simulated solar neighbourhood (Synthetic-CH), just in a way that an observer would.

This result shows that the latest chemical evolution models have a remarkable success in being able to produce results which are comparable with modern day surveys, but are underproducing metal poor stars, and from simulated observations, we find that we are producing too many red and faint star!

October 2015

click on image to enlarge

Despite most galaxies evolving slowly over cosmic time, Ultra Massive Passive Galaxies (UMPEGs) represent a population of that has already undergone massive star formation by the time the universe is only 1/4 of its current age. Since their star formation is already quenched by the time we observe them (at the peak epoch of cosmic star formation), they must have attained very high rates of star formation to assemble their stellar masses (Mstar > 1011 Msun) in the short time available to them. UMPEGs are also believed to reside in high-density environments, and due to their extremely large stellar masses are guaranteed to be the central galaxies of their host dark matter haloes. Therefore it is important that we constrain this population in an attempt to understand their evolution up until today.

In collaboration with Anneya Golob (SMU), Stephane Arnouts, Thibaud Moutard (Vipers K-band Survey) and Dr. Marcin Sawicki (SMU) we have been able to constrain the stellar mass function of the largest sample (to date) of K-selected passive galaxies at high redshift (z~1.7). This plot combines results from the Deep and Wide Canada-France-Hawaii telescope legacy surveys, allowing us to detect these rare K-selected objects in an effective area of ~27.5 square degrees. The large area meant that ICA and ACE-NET were instrumental in the pre-processing and analysis of the data.

This large population allows us to effectively constrain the bright end of their luminosity and mass functions. As we can see from the image, a simple increase of number density of quiescent galaxies matches the observed z~0 stellar mass function (Peng et al. 2010).  This suggests that the mechanism that quenches star-formation in massive galaxies does not change with time.

Liz M. Arcila-Osejo under the supervision of Dr. Marcin Sawicki

September 2015

click on image to enlarge

Some Cepheid variable stars are more equal than others.

Prof. David Guenther (founding member of the ICA and a member of the MOST Science team).

Classical Cepheid variable stars are well known as distance indicators. Their brightness varies regularly over periods of days to months with the most luminous Cepheids having the longest periods. The relationship between period and the intrinsic luminosity (energy output per unit time) allows astronomers to determine distances to the Cepheids (and any cluster of stars in their vicinity in space) by simply observing their period of variability. Until recently, Cepheids were thought to be very stable, that is the period of their brightness variability repeated cycle to cycle with great accuracy. Recent observations from space, e.g., such as those using Canada’s MOST satellite (Microvariability of Stars) have provided extremely accurate measurements of the periods of several Cepheids that show this is not the case.

N. R. Evans (Harvard), R. Szabo (Budapest), A. Derekas (Budapest), and L. Szaboados (Budapest), along with C. Cameron (a former student at SMU now at Cape Breton), J. M. Matthews (UBC), D. Sasselov (Harvard), R. Kuschnig (Vienna), J. F. Rowe (NASA), D. B. Guenther (SMU), A.F. J. Moffat (Montreal), S. M. Rucinski (Toronto), and W. W. Weiss (Vienna) (2015, MNRAS 446, 4008) found that the overtone Cepheid pulsator SZ Tau shows greater cycle to cycle variation in its period than the fundamental mode pulsator RT Aur. When one plots the brightness of the Cepheid over many periods, one period on top of the other in a phase diagram, one can see the observed slight variations. Comparing the inserts in Fig A and B, which show a magnified view of the variations at the peaks, one can see that the variations in Fig B are greater, i.e., for the first overtone pulsator, than in Fig A, i.e., the fundamental mode pulsator.

Fig A–Phase diagram for the fundamental mode pulsator RT AUR observed by MOST.

Fig B–Phase diagram for the first overtone mode pulsator SZ Tau observed by MOST.

August 2015

click on image to enlarge

A handful of foreground galaxies are seen agains a background of hundreds of more distant, anonymous objects.  This image was taken with the MegaCam imager on the Canada-France-Hawaii Telescope (CFHT) as part of the ongoing CFHT Large Area U-band Deep Survey (CLAUDS).  CLAUDS is a collaboration between Canada, France, and China and, in turn, partners with the HyperSuprimeCam Strategic Survey Program on the Subaru Telescope. This image, the result of 17 hours of integration, shows only a tiny fraction of the area covered by the CLAUDS surveys. CLAUDS aims to provide imaging of ~20 square degrees of the sky to a limiting magnitude of 27 (AB) in the U-band — a hitherto unprecedented combination of area and depth.  SMU professor and ICA member Dr. Marcin Sawicki is the Canadian PI of CLAUDS, while the image shown has been processed by CLAUDS team member Dr. Stephen Gwyn of NRC-Hertzberg.

July 2015

click on image to enlarge

The peak of the global star formation rate (SFR) of the universe and active galactic nuclei activity, coincide at slightly over 10 billion years ago. Observing galaxies at this epoch helps uncover what the universe was like during it’s busiest time and is key to understanding both galactic evolution, and the evolution of the universe as a whole. Determining the dust mass in these galaxies is important because it is a crucial part of understanding the physics of the interstellar medium (ISM) which directly impacts the production of stars. Indeed, studying the evolution of the ISM allows for a more robust understanding of galaxy evolution.

The galaxies in the sample above are all massive star forming galaxies from four different - distant - fields which correspond to epochs between 9 and 10.5 Gyr ago. Shown above is the ratio of molecular gas mass to the submillimetre derived dust mass, which is a key measure of dust evolution in galaxies. The galaxies in the sample were observed with both the James Clerk Maxwell Telescope and the Spitzer space telescope, and the data was analyzed using ICA machines to produce both a star formation rate and dust mass. The gas masses were derived from CO (J=3-2) emission by Tacconi et al. (2013).

The red traces seen on the plot are from simulations by Hayward et al. (2011), who studied submillimetre flux dependence on galaxy properties by using hydrodynamic simulations of both isolated and merging disk galaxies. By performing three-dimensional dust radiative transfer calculations on their simulations, they derived dust masses and star formation rates for their galaxies. Somewhat surprisingly, the simulations show little correlation between submillimetre flux and gas-to-dust ratio, which is mirrored in the 850 micron derived SFR in the plot. It is important however, to note that the simulation data in the upper left corner is highly dependent on initial conditions, and should be interpreted as such. The galaxies in our sample are (for the most part) in agreement with the models, with the exception of a few interesting outliers.

Jennifer Beanlands, MSc. candidate, in collaboration with Dr. Scott Chapman, as well as ICA faculty member Dr. Rob Thacker

June 2015

click on image to enlarge

One of the most important physical processes in galaxy formation is  "stellar feedback" - the mechanism by which stars return energy to the interstellar medium. Early galactic simulations did not include this process and the resulting galaxies were poor analogues of observed disc galaxies: high bulge-to-disc ratios were found, and the simulated galaxies converted nearly all their gas into stars. Adding stellar feedback helps not only to prevent gas being converted too quickly, it also helps produce disc morphologies.

While the overall picture of stellar feedback is understood, the precise approach to modelling it remains a subject of extensive research. ICA MSc student Samantha Pillsworth, supervised by Dr Rob Thacker, is studying the detailed differences between a set of different feedback models. The above figure shows the smoothed tangential velocity of one galactic disc simulated with three different feedback models. The left most plot shows the disc with no feedback. There is a high velocity region close to the center of the disc, consistent with a dominant bulge in the galaxy. The tangential velocities also indicate that the disc is thin and truncated. The middle plot shows the smoothed tangential velocity of the disc simulated with a blastwave (BW) feedback mechanism. BW feedback returns energy to the disc from supernovae (SNe).  Increased energy creates a more turbulent disc (resulting in lower smoothed tangential velocities). There is still a central high velocity region and bulge; BW feedback continues to allow too many bulge stars to form. The left most plot shows the smoothed tangential velocity of a disc with a superbubble (SB) feedback mechanism. Devised by Keller et al. (2014), SB feedback allows many SNe to properly combine into SBs. The tangential velocities shown are further reduced, due to the large amount of turbulence created by these SBs. Note that there is no central high velocity region - SB feedback has prevented the overproduction of bulge stars, and has maintained a realistic, and low, bulge to disc ratio. The raw data for these plots comes from the MUGS2 simulations (Keller et al. 2015). The SPH data was smoothed, analysed, and plotted using the ICA shared-memory server "Orthus".

May 2015

click on image to enlarge

Galaxy formation and evolution are thought to be closely tied to black hole (BH) activity. BH activity is governed by the BH accretion rate and therefore the different processes (i.e. galaxy interactions, gas cooling, gas infall, feedback) that impact the accretion rate over different timescales (ranging from hours to billions of years). The power spectrum (PS) of the Eddington ratio (BH luminosity normalized relative to a maximum luminosity known as the Eddington luminosity) contains valuable information pertaining to the timescales that best describe the variations in accretion and therefore the various processes involved (Novak et al. 2011).  The graph above displays a normalized PS of the BH Eddington ratio: the amount of variation in the Eddington ratio in a given frequency bin. The error bars denote 1σ variations within a given bin.  This graph was produced using simulations of isolated Milky Way like disc galaxy mergers with ~3.8 million particles. The simulations were performed using a parallel version of HYDRA (an adaptive particle-particle, particle-mesh/SPH code; Thacker & Couchman 2006; Wurster & Thacker 2013) on the ICA cluster, Cerberus. The blue and green squares represent the PS of the ONB (Okamoto et al. 2008) and BS (Booth & Schaye 2009) models respectively. The two models are fundamentally different: the ONB model has comparatively non-local feedback while BS injects significant energy into the local environment.

The high frequency peak is caused by the chaotic behavior in the immediate neighborhood of the BH as well as numerical artifacts (i.e. discreteness). On the other hand, the peak in the PS of the BS model at f ~ 0.3/Myr may be attributed to the rapid variations in the gas characteristics around the BH due to feedback outflows and BH motion. Nonetheless, both models show a similar trend in their PS (proportional to 1/f^2 which is, intriguingly, the same as an exponential decay in the Eddington ratio although the raw data does not show this behavior. The ONB model shows minimal variations from the 1/f^2 powerlaw. The origin of such behavior is not fully understood; it may be due to the BH residing in a steady environment (due to the non-local feedback) and therefore accreting at a steady accretion rate.

This work was performed by MSc. candidate Maan H. Hani, in collaboration with ICA faculty member Dr. Rob Thacker.

April 2015

click on image to enlarge

The power of computational astrophysics has grown enormously in the past two decades, and we can now simulate the evolution of a cluster of galaxies with spectacular fidelity using parallel computing. But with this increase in predictive power there is a growing necessity to ensure accuracy. This image, which plots the density of the gas in the region within and surrounding a cluster of galaxies, comes from a comparison project of simulation codes that is being conducted by a group of 25 researchers from Australia, Canada, Germany, Italy, South Africa, Spain, Switzerland, UK and the USA.  Acting ICA Director, Dr. Rob Thacker, has contributed his own simulation code to the project and recently visited the University of Western Australia to work with project leader Dr. Chris Power.  The project is unique in taking a step-by-step approach to increasing the complexity of the physics in the simulation models to determine precisely where differences occur.

A first paper has been submitted, and can be seen here:

In follow-up work, ICA Ph. D. student Nozhan Balafkan will produce a detailed analysis of the accuracy of the different codes when star formation and black hole physics are added.

March 2015

click on image to enlarge

Globular Clusters (GCs) are the largest and brightest groupings of stars within galaxies, easily observed in the Milky Way, and in distant galaxies where individual stars cannot be resolved.  In GCs where the individual stars cannot be resolved, information about the stellar population is not readily observable.  However, this information may still be extracted from the Integrated Light (IL) of a cluster.

In collaboration with ICA faculty member Dr. Ian Short, Doctoral candidate Mr. Mitchell Young constructed a synthetic IL spectral energy distribution (SED) of a GC with parameters similar to Omega Centauri, the brightest GC in the Milky Way.  Starting with an isochrone (a collection of stellar models of common age) obtained from the BaSTi group, and Kroupa's initial mass function, Mr. Young constructed a synthetic population for the GC.  By breaking this population down into sub-populations and using the atmospheric modelling and spectrum synthesis code PHOENIX, running on the ACEnet clusters Fundy and Mahone, Mr. Young produced SEDs representative of these sub-populations that were combined into the cluster's IL SED.

The left plot is a theoretical colour-magnitude diagram of the synthetic stellar population.  The solid blue line represents the cluster's population, with the boxes dividing it up into sub-populations that each emit approximately 3.5% of the total flux emitted by the cluster (the overlapping boxes in the upper half of the diagram do not represent sub-sub-populations, but rather are a consequence of two different sections of the population existing close to each other in the diagram, where the smaller boxes lie along the lower red giant branch, and the larger boxes lie along the upper asymptotic giant branch).  The red dots within the boxes indicate the positions of stars whose parameters are the flux weighted averages of their respective sub-populations.  Inset in the diagram are SEDs of three red dot stars, displaying how the emitted flux is distributed differently over the wavelength range for different sub populations.  The right plot displays the IL SED for the whole population, constructed by adding together the representative SEDs for each sub-population weighted appropriately such that each sub-population contributed equal flux to the IL SED.

February 2015

click on image to enlarge

In the age of big data astronomy, a large survey can yield observations of millions of galaxies. However, accurately calculating distances to the galaxies requires we take spectra, specifically spectral energy distributions (SEDs), which is a comparatively slow and painstaking process. As an alternative, "photometric redshifts" provide a way to overcome the lack of spectroscopic observations by inferring an object’s redshift from photometry in a set of filters.

SED fitting methods compare the photometric observations of objects with those of spectra in a library of template galaxies at known redshifts. Two broad classes of photometric redshift methods have emerged: SED fitting and machine learning. Generating templates requires making many physical assumptions that can affect the results, while machine learning methods are purely numerical and thus are not susceptible to biases caused by assuming incorrect physics. Artificial Neural Networks (ANNs) are a type of machine learning that is ideal for large surveys where spectroscopic redshifts have been measured for a subset of the data, which serves as a ’training set’. Data from the training set is used to relate directly observable properties of the galaxies (like photometry and morphology) to their redshifts.

This density plot shows the quality of redshifts determined using the the ANN code ANNz (Collister & Lahav, 2004.) ANNz uses a network structure consists of a number of layers of nodes. The first layer contains the input observed properties (e.g. galaxy photometry in each filter) and the final layer contains the output redshift. The ‘hidden’ layers in between contain nodes connected to every node in both adjacent layers. These connections carry weights that are varied by the training algorithm to optimize the performance of the network. Here, ANNz was applied to the 555000 galaxies in the COSMOS Mock Catalog (Jouvel et al., 2009) using ACEnet’s computing resources. 10000 objects were used to train the ANN to infer galaxy redshifts from observations in 6 photometric bands. ANNz has proved to be more computationally efficient and consistently produces more precise results than template-based methods. It, and other machine learning codes, will be invaluable for analysis of data from the massive surveys that will be conducted over the next decade.

Anneya Golob, PhD student

January 2015

click on image to enlarge

Total Stellar Mass in Galaxies: Comparing the Whole with the Sum of the Parts

The total stellar mass of a galaxy cannot be measured directly and instead must be inferred from the light we observe. However, young stars are much brighter than old stars and hence may obscure the older population, leading to a possible systematic underestimate of the stellar mass.  In Sorba and Sawicki (2015; submitted), we have attempted to quantify the bias that this outshining effect produces, using a sample of nearby galaxies in which we can disentangle the young and old populations. We measured the stellar mass of each galaxy twice, once by fitting a model spectral energy distribution (SED) to all the light of the galaxy combined together (as happens for very distant galaxies), and a second time by fitting a model SED to each pixel in the galaxy and summing the resulting pixel masses. Plotted in the shown figure is the ratio of the unresolved stellar mass and the pixel-by-pixel (PXP) stellar mass versus the mass-weighted star formation rate (sSFR) for 68 galaxies of various types. A false-colour image of each galaxy is used to quickly show trends with the galaxies' morphological classification and colour.

If there were no discrepancy between the unresolved and spatially resolved stellar mass estimates the galaxies would all be scattered around the horizontal dashed white line at y=1. Instead, we find that the redder (less strongly star-forming) galaxies on the left tend to be centered around the dashed line, while the bluer (more strongly star-forming) galaxies on the right nearly all lie below the dashed line. This indicates that the unresolved stellar mass estimate is lower than that obtained by summing the stellar mass in each pixel. This makes sense if the SED models are preferentially being fit to the young. bright, blue stars and not accounting for the mass contained in the obscured, older stars. We conclude that spatially unresolved stellar mass estimates of star-forming galaxies are underestimated by approximately 15% on average.

Model SEDs have many free parameters, and finding the best-fitting model in the large parameter space using a Chi-squared statistic for the ~750,000 pixels used here would not have been possible without the parallel computing resources provided by the ICA and ACE-net. In total, approximately 600 CPU-hours were used in fitting the models used to make this plot, or about one straight month if only one processor was used. We also thank Tiffany Fields for her great help in the creation of the false-colour images used in this plot.

Robert Sorba, PhD student