Institute for Computational Astrophysics

Previous Images of the Month - 2013

January February March April May June July August September October November

December 2013

click on image to enlarge

Over time, a galaxy will form billions of stars, but exactly how many stars are formed as a function of time is by no means a simple relation. Star formation tends to happen in unpredictable bursts that typically occur in the spiral arms of galaxies. The newly formed stars are much, much brighter than the older stars, meaning that the light from the older stellar populations can often be hidden behind the light from the young stars. This makes disentangling stellar populations challenging, especially at high redshift.

In order to study the different stellar populations separately, PhD student Robert Sorba along with supervisor Dr. Marcin Sawicki fit model spectral energy distributions (SEDs) to each pixel in the images of galaxies. The SED fitting yields best-fit estimates for many parameters, including the stellar age and the stellar mass, both of which are shown above for NGC 628. The figure on the left shows the stellar age map, with blue pixels representing older stellar populations, and orange/red pixels more recently formed stars. The figure on the right shows the stellar mass map, with blue pixels representing higher amounts of mass contained in stars, and red lower. Looking at the maps, one can see the newest stars forming in spiral arms, but note that the majority of the stellar mass is located in the center of the galaxy away from the star forming regions. Although the youngest stars are the brightest, they contain a disproportionately low amount of mass, which can lead to a systematic under-estimation of the total stellar mass of a galaxy unless the relatively faint (but more important by mass) older stellar populations are taken into account.

The SED fitting and Monte Carlo simulations to determine probability density functions of the best-fit parameters were performed using the software SEDfit (Sawicki 2012) and FSPS (Conroy, Gunn, & White 2009). The ability to fit hundreds of galaxies each with thousands of pixels would not be possible without the parallel computing resources provided through the ICA and ACEnet.

November 2013

click on image to enlarge

The mass of supermassive black holes resident in the cores of galaxies appears to follow a fairly tight correlation to the overall velocity dispersion of the stellar component of the galaxy (the so-called M-sigma correlation). While no longer considered a completely universal relationship, understanding the origin of this correlation and more importantly the role of active galactic nuclei in overall galaxy evolution is one of the most important questions in modern galaxy formation.

In work that is an off-shoot of Dr. James Wurster's PhD research using the HYDRA SPH-AP3M code, Dr. Rob Thacker, James and summer student Mr. Chris MacMackin have been examining the correlation between star formation rate (SFR) in both the inner and outer parts of simulated galaxies compared to rate of black hole growth (the so-called "BHAR" for black hole accretion rate).  Observational studies of Seyfert galaxies suggest that the correlation is tightest when considering the star formation rate in the nucleus of the galaxy, and that there is a much weaker correlation when considering the extended or outerlying star formation.

Analysis has shown that while the simulations reproduce the strong correlation for the nuclear region prior to the merger of the two galaxies shown, they actually recover an (unexpected) anti-correlation for the outerparts. Moreover, almost all the different models for AGN feedback examined produce this behaviour. Further research is underway to pin down the cause of the anti-correlation, and thus to understand whether the poor correlation observed for the observations is actually related to a selection effect.

October 2013

click on image to enlarge

This image represents the Stellar Mass Function by galaxy type: Star Forming (blue) and Passively Evolving Galaxies (red) at redshift 2. The stellar mass function for our passively evolving galaxies has a steeply rising massive end, and a declining low-mass end. The peak of the stellar mass function represents the most common passively evolving galaxy at this epoch: galaxies with a stellar mass of ~1011 solar masses.  Our stellar mass function agrees well with previous published results, but our sample contains many more objects and thus has much better statistics than these earlier studies.  A comparison between these two populations of galaxies (star-forming and passive)  shows a "downsizing" scenario in which massive galaxies at high redshifts seem to  shut down star-formation before less massive ones.  This "downsizing" scenario is consistent with the mass quenching mechanism proposed by Peng et al. (2010), which seems to be effective at shutting down star-formation in more massive galaxies at high redshifts.

This stellar mass function was obtained using observations from the Canada France Hawaii Telescope Legacy Survey (specifically their Deep Survey) and the WIRCam Deep Survey. In order to obtain estimates for the luminosities and stellar mass functions for these galaxies we used spectral energy distribution software (SEDfit, Sawicki 2012) and stellar populations from GALAXEV (Bruzual and Charlot 2003).

This work is a result of team work between ICA faculty member Dr. Marcin Sawicki, Dr. Taro Sato and ICA Ph. D. student Ms. Liz Arcila-Osejo.  All processed information was obtained thanks to the capabilities of ACEnet.

For further information please refer to:

Arcila-Osejo & Sawicki 2013

Sato, Sawicki & Arcila-Osejo (in press)

September 2013

click on image to enlarge

The ratio of the deduced effective temperatures between two models of different masses as indicated is plotted as a function of the inclination between the observer and the rotation axis. All models are uniformly rotating on the ZAMS with the surface equatorial velocity of each set so that all models have the same surface shape (in this case the ratio of the polar to the equatorial radius is 0.82, which corresponds to a surface equatorial velocity of about 300 km/s for the masses presented). Having the same surface shape means that the ratio of the surface radii at the same latitude for the two models is independent of latitude.

The deduced effective temperatures are obtained by matching the B-V colour of the spectral energy distribution (SED) of the rotating model with that produced by a nonrotating model. The SEDs of the rotating models are computed by a geometrically weighted integral of the intensity in the direction of the observer over the visible surface of the model. The individual intensities are obtained from NLTE plane parallel PHOENIX model atmospheres, as is the colour effective temperature relation for the nonrotating models.

The important point of the plot is that the ratio of the deduced effective temperatures is virtually independent of the inclination.  This relation is not true for models whose degree of rotation is determined by any other criterion, making the shape as an “orthogonal” variable to the mass. A similar relationship is also true for the deduced luminosity of rotating stars.

This work is being carried out by Dr. Robert Deupree and Ph. D. student Mr. Diego Castañeda, with ICA faculty member Dr. Ian Short providing guidance using the PHOENIX model atmospheres code.

August 2013

click on image to enlarge

This 3D plot shows a grid of 330 simulated stellar surface flux spectra produced by ICA faculty member Dr. Ian Short that span the entire visible band (x-axis: logarthimic wavelength (lambda)) for red giant stars.  The grid spans a range of effective stellar "surface" temperature (y-axis: Teff) and logarithmic surface gravity (log(g) - offset along z-axis).  Note that the z-axis also does double-duty as the brightness (logarthmic surface flux, log f).  Stellar "surface" temperature is also conveyed with colour: Red is cool and violet is hot.

Spectra depicted in bolder lines were computed directly using Version 15 of the PHOENIX atmospheric modelling and spectrum synthesis code running on the ACEnet high performance computing facility, and the remaining spectra were computed by numerical interpolation among these.  These model spectra can be compared to observed spectral energy distributions of real stars to find the closest match, and thus the surface temperature and gravity of the star.

July 2013

click on image to enlarge

The ICA hosted an artist’s exhibit in the ACEnet Data Cave located at Saint Mary’s University in late March and early April. The artist, Lisa Frank, visited Saint Mary’s and gave a lecture about how she took her photographic art and modified it for an immersive environment as part of her MFA project at the University of Wisconsin – Madison. ICA Director Dr. Robert Deupree and Saint Mary’s Communication Manager, Mr. Steve Proctor, hosted the lecture and following reception. The artist’s visit and the exhibit were funded by the Saint Mary’s Office of the Vice President, Academic and Research and by the ICA. The exhibition ran for about three weeks after the lecture, with more than 500 members of the general public viewing it to enthusiastic feedback. After the termination of the exhibition we had to turn away about another 170 people. Given that each tour of the exhibit can only include about five people in the Cave and takes between fifteen and thirty minutes, considerable human resource was required to make the event happen. Ms. Florence Woolaver, ICA Assistant who arranged the tours, and Ph. D. graduate student Mr. Diego Castañeda and ACEnet System Administrator Mr. Phil Romkey, who conducted the tours, are thanked for their efforts to make the exhibit a success. The image, provided by Metro, a Halifax newspaper, shows people viewing one of the artist’s scenes in the Data Cave.

June 2013

click on image to enlarge

To learn about faint, distant galaxies we often observe them through filters that sample their spectra at various wavelengths. This is much more efficient than sampling their entire spectra. Comparing these observations with theoretical models allows us to infer the physical properties of these far away objects. To generate the plots seen here, PhD student Anneya Golob used rest-frame spectra (light as it would be seen by an observer inside the galaxy) created (on ACEnet machines) using the GALAXEV code of Bruzual & Charlot (2003).

To simulate a galaxy much further away, many additional factors must be accounted for including cosmological dimming in an expanding universe, attenuation by gas between galaxies, and obscuring dust within the galaxy itself. These effects were applied to the rest-frame spectra on ACEnet machines using the code SEDfit, created by Anneya's supervisor, Marcin Sawicki. SEDfit also uses the transmission functions of filters from specific telescopes to simulate data that would be obtained from observations of these distant galaxies. 

Many techniques have been developed to classify galaxies into redshift intervals based on observations in a few filters. One of these is the BX method, intended to select objects around z=2.2 based on their position in the (U-G)-(G-R) colour plane, show here as the grey-shaded region. These plots show the redshift evolution of theoretical galaxy spectra created with various star formation histories and dust models. The colour of each point represents increasing redshift from blue at z=0 to red at z=3. The four point sizes in each panel correspond to the amount of dust included in each model, parametrized by E(B-V).

May 2013

click on image to enlarge

When we talk about the temperature of a star, we’re usually referring to what’s known as the star’s effective temperature.  For a star in local thermodynamic equilibrium (LTE), its effective temperature (Teff) is defined as the kinetic temperature (Tkin) of the star's atmosphere at the point where the Rosseland mean optical depth (τR) has a value of 2/3. While this value can be calculated directly for any given star, it takes time and computer resources to do so.

Using the model stellar atmosphere and synthetic spectrum code PHOENIX running on ACEnet’s Mahone cluster, Master’s student Christopher Cooke, in collaboration with ICA faculty member Dr. Ian Short, has created a grid of model stellar atmospheres to use as a reference for estimating the value of Tkin at τR = 2/3 for any arbitrary star with a Teff between= 5600 and 5900 K and a surface gravity (log (g)) between 3.5 and 5.0.

Because the Sun is the best understood star in terms of atmospheric properties, it was used to test how the resolution of the grid of reference models affects the accuracy of estimates calculated from the grid. The above picture shows a comparison of computed values of  Tkin throughout the Sun (Teff = 5780 K, log(g) = 4.44) to estimates of these values using two different sets of reference model atmospheres, plotted as ‘residuals’ – that is, the difference between the actual computed values and the estimated values. The blue line shows the residual values of Tkin that is found when the separation between the reference stars is ΔTeff  = 50 k and Δlog(g) = 0.5, while the red line shows the residual values of Tkin that is found when the separation between the reference stars is ΔTeff  = 300 k and Δlog(g) = 1.5. The vertical green line represents τR = 2/3, and so the points where the blue and red lines intercept the green vertical line are estimates of Tkin = Teff for each of the estimation methods.  Values to the right of the green line become chaotic due to convection in the lower parts of the stellar atmospheres.

As one might expect, choosing a finer grid of references produces more accurate results. The difference between the estimated and modelled temperatures at τR = 2/3 for the finer grid is approximately 0.5 K, whereas for the coarser grid it’s over 2.0 K. These differences are very small relative to the effective temperatures of stars, though for very low mass stars the accuracy to which we can calculate Teff becomes important.  As one looks to larger – and therefore hotter – stars, however, coarser and coarser grids become very effective, and can save many hours or days of computing time.

April 2013

click on image to enlarge

Photometric colours are useful for determining the fundamental parameters of stars: Effective (surface) temperature (Teff), surface gravity (log(g)), and composition. In the above plot, the transmission curves and integrated pass-band fluxes for the UX (purple), BX (dark blue), B (light blue), V (green), R (red) and I (black) filters from the Johnson-Cousins photometric filter set are over-plotted on the original synthetic spectrum from which they are calculated. In collaboration with ICA faculty member Dr. Ian Short, Master's student Mr. Mitchel Young computed the spectrum of a red giant star with the stellar atmospheric modelling and spectrum synthesis code PHOENIX, running on the ACEnet cluster Fundy, and then calculated the integrated pass-band fluxes using his own procedure written in the Python programming language. The model input parameters for the above spectrum were Teff = 3550 K, log(g) = 2.0 (log cm s^-2), and 1/3 solar "metal" abundance, which are representative of the standard star Arcturus (alpha Bootes). This calculation is based on the most realistic treatment of the thermodynamic state of the gas and radiation field in the star's atmosphere (non-local thermodynamic equilibrium, NLTE).  Mitchell is using a library of similar synthetic spectra for a range of red giant stars to study how the NLTE treatment affects our ability to distinguish horizontal temperature variations on the surfaces of such stars. Generally, NLTE effects mimic the brightening of the blue end of the spectrum that can be caused by "hot spots" on the surfaces of stars.

March 2013

click on image to enlarge

The properties of numerical simulations must always be compared to observational results to verify the validity of the simulations. One standard comparison for simulations involving black hole evolution is the M-σ relation, where M is the mass of the black hole, and σ is the velocity dispersion of the stars around it.  In the above top plot, the dotted blue line is the observed M-σ relation given by Gultekin et al. (2009 ApJ, 698, 198); the solid red lines are the one-sigma scatter.

In collaboration with Dr Rob Thacker, Ph. D. student Mr. James Wurster ran a suite of major merger simulations, where the only difference between the models is the accretion and feedback algorithms that govern the activity from active galactic nuclei (AGN). By holding all other aspects of the simulations constant, we were able to determine the effect that each AGN feedback algorithm has on the galaxy merger. Each merger was simulated using the SPH-AP3M code HYDRA, and all simulations were run on Dr Thacker's cluster at the ICA.

After 1.5 Gyr of evolution, each merger yields a triaxial stellar remnant similar to that presented in the bottom images, which show a face on and edge on view of the central 20 kpc of one simulated remnant.  To mimic observational methods, we pick a random line of sight that passes through the centre of the remnant and calculate the stellar velocity dispersion along that line.  As would be expected, the triaxial nature of the stellar remnant yields very different velocity dispersions depending on the line of sight.  Thus, in the top plot, each black dot represents the average velocity dispersion of 1000 random lines of sight for a given model, and the horizontal bars represent the range of velocity dispersions for that model.  Thus, all but three of our models have remnants whose M-σ relation matches the observed relation within the one-sigma scatter.  This indicates that additional analysis is required on these three models to determine why they do not agree with the observational relation.

February 2013

click on image to enlarge

Line profiles for three 2D rotating stellar models simulating the rapidly rotating δ Scuti star α Oph, along with the IUE spectrum for this star. The three models have rotation laws given by the equation.


Ω is the rotation rate, and ϖ is the distance from the rotation axis in units of the surface equatorial radius. Ω0 and a are constants. Uniform rotation corresponds to β=0, and the rotation rate increases as one goes towards the rotation axis more strongly as β increases. The β=0.4 profile is relatively close to rotation profile one would obtain between the center and the surface at the equator for a model starting with uniform rotation on the ZAMS and conserving angular momentum locally during evolution to the late core hydrogen burning state of α Oph. The differential rotation rate from the formula above produces a conservative rotation law, which is unlikely in α Oph itself.

The line profiles have been matched in each of the far wings. Increasing β leads to more absorption in the core, but by far too small an amount to determine which model would be more appropriate for α Oph. This trend is maintained in other lines studied in the ultraviolet and visual spectra. It is not clear, but likely, that the line profiles of the differentially rotating models could be made to resemble that of the uniformly rotating model with only slight modifications to the model properties, suggesting that spectra alone are insufficient to determine the rotation profile for this amount of profile variation.

This work has been carried out by Diego Castañeda as part of his Master’s thesis research under the supervision of Dr. Robert Deupree. The research uses the surface conditions from the 2D stellar evolution and hydrodynamics code, ROTORC, model stellar atmospheres surface intensities from the NLTE stellar atmospheres code, PHOENIX, and an ICA code developed by Catherine Lovekin and Aaron Gillich to determine the spectrum an observer would see from the intensities emitted from the surface of the rotating star.

January 2013

click on image to enlarge

Astronomers have always wanted to know the ages of individual stars. First, this would reveal details about the history of their formation and about the history of the Galaxy as a whole. Secondly, it would pin down the stars on their path through stellar evolution. This, in turn, would allow us to further test and refine the details of our theoretical stellar models. Asteroseismology, which uses stellar pulsation periods (i.e., star quakes) to measure the structure of stars, has been identified as a possible key method for determining stellar ages with unprecedented precision.

The NASA Kepler mission, designed to detect Earth-size planets around Sun-like stars, is able to measure these star quakes, and so researchers (e.g., Mathur et al. 2012) have begun to use asteroseismology to probe stellar ages with asteroseismology. One problem, however, is that there are known systematic difficulties in comparing the pulsation periods to what our models predict. Here at the ICA, PhD candidate Michael Gruberbauer, his supervisor Dr. David B. Guenther, and collaborators, have therefore developed a new statistical method of how to take these problems into account (Gruberbauer et al. 2012).

The picture of the month shows a comparison of their new asteroseismic method ("Bayesian") with the traditional approach ("AMP", Mathur et al. 2012) in terms of the ages of Sun-like stars (note: 1 Gyr = 1 billion years). Each of the 19 data points shown represents one individual Sun-like star for which data was obtained with the Kepler space telescope (Mathur et al. 2012). If both methods agreed, all points should lie on the thick red line. As can be seen, only a few points even come close, and there are large differences between the two methods. These differences are found to be as large as 3.39 Gyr (= 3.39 billion years). More importantly, the error bars indicate that there are also very large differences in the age uncertainties as derived from each method. The error bars in the "Bayesian age" direction are much larger than in the "AMP age" direction. On average, the uncertainties of the Bayesian ages are about 7 times larger than those from the traditional approach. In one case, the uncertainty is even 20 times larger. This does not mean that the results of the Bayesian method are worse but rather that the AMP method (sometimes strongly) underestimates the age uncertainties.

To summarize, these results suggest that the high precision claimed for stellar ages, as determined via traditional asteroseismology ("AMP"), cannot be confirmed using our new method. More work needs to be done in order to remove the systematic difficulties which only the Bayesian method properly takes into account. Only then can we hope to precisely answer the question: How old are the stars?


Cited scientific papers:

Gruberbauer M. et al., 2012, ApJ, 749, 109

Mathur S. et al., 2012, ApJ, 749, 152