On the source of the diffuse background light in HUDF Tuguldur Sukhbold Dr.Henry C. Ferguson Timothy Dolch Space Telescope Science Institute Aug 19, 2010
HUDF Here is a crop from one of the HUDF images, what do you see in the ultra deep field? Few stars, which are not even in this crop; you basically see bunch of galaxies. But what if you mask out these resolved galaxies?
Then you are left with some diffuse background Then you are left with some diffuse background. It seems like a static noise, but it turns out there is much more than that. This is basically what we call the EBL. In this case of UDF it is the NIR EBL. So the question is why there is this background diffuse light in this wavelength region?
Measurement wise there have been several works done in the past decade Measurement wise there have been several works done in the past decade. Here the flux is shown as a function of wavelength. These points down here are the total galaxy light measurements, defining the lower limit of the flux. And these pink points up here are from IRTS mission and all points around here are measurements from various sources. It seems like there is definitely some excess, actually resembling the lyman-break feature, but we should be aware that all these measurements have different models for zodiacal light calibration, which is something not all scientist have a common solution of. The zodiacal light is a very serious issue for the NIR background. On this plot you see the brightness of the background sky as a function of wavelength. Outside the atmosphere in the solar system, the zodiacal light produces this much, while the sources that we’re trying to determine emit at this level, and remember this is on log-scale so the zodiacal light really produces more than 90% of the brightness. Zodiacal Light >90%
Possible Sources Low Z Galaxies VS Low z Galaxies (e.g. Kashlinsky et al. 2005) (e.g. Thompson et al. 2007) Zodiacal Light Faint Structures Data not high resolution and crowded Detector Issues Method of Analysis In 2005, Kashlinsky and his group at the Goddard space flight center have published a very controversial paper in Nature magazine. In there, they analyzed Spitzer data and claimed that the IR BL is due to an exotic source - first galaxies made out of very metal poor first few generations of stars. Then in later years several other groups including the ones used the NICMOS data have claimed the source to be not those early galaxies but low redshift galaxies that are below the detection limit of the HST. The source is still a subject of a debate and some of these groups have been tossing papers often with not so polite language towards each other. And again all these works are dependent on the zodiacal light model, and beside that there are issues concerning the faint structures of galaxies, crowded nature and low resolution of their data set, problems with the detectors they have used and finally with the method of analysis.
New in this Summer Work New set of data: UDF 05-01 & UDF 05-02 (Illingworth et al. 2010) New method of analysis: P(D) Improved Simulation This summer I continued the project with Dr.Ferguson and grad student Tim Dolch on analyzing HUDF data for NIR background fluctuations. We have made use of the new WFC3 data set: the UDF05-01 and 05-02 fields. We apply the new method of analysis which we call the P(D), I will explain in a minute and we also have improved the galaxy field simulations that are used in our analysis.
Processing Iteratively remove detector blemishes Combined dithered images with some masking Transfer combined image back to original image geometries and subtract Smooth and detect blemishes or persistence Mask and recombine for the final image Create pure noise images in original detector coordinates Gaussian statistics okay for these images; match sky background Predicted RMS matches measured RMS to within a few percent Combine these just as for the real images Mask the detected sources The data are processed in a fairly standard way. The detector issues are cleaned by combining the dithered images with the mask then put back into original frames, subtract, smooth and then mask and combine for the final image. Then we create noise images with gaussian statistics on the original detector space, then combine them as well. These noise images later used in our simulations. Finally we mask out the resolved sources.
Analysis: P(D) Create a shuffled version of the image Unmasked pixels are randomly shuffled, removing correlations For both the shuffled and unshuffled version: Convolve the masked images with kernels of various sizes Compute the histogram of pixel intensities P(D) Subtract the shuffled P(D) from the unshuffled P(D). Excess is amplified when kernel matches the characteristic size of the sources To analyse our data we take an image properly calibrated and all sources masked, then create a copy where we shuffle the pixels. Then we convolve both images (shuffled and unshuffled) with different kernel sizes and compute the histogram of the pixels (that’s why we call pofd). Here is the histogram plot for the shuffled and unshuffled images and you see that the when the pixels are shuffled the histogram is supressed to a gaussian shape. Then we see the difference between the plots and the excess is amplified when the characteristic size of the source matches that of the actual source.
Simulation Tune 2 parameters: Slope of the Galaxy count Inclination Position angle Tune 2 parameters: Slope of the Galaxy count Size of the galaxies The current simulation of galaxies include specific distributions of size, inclination and position angle. We vary 2 main parameters: the slope of the galaxy count and the size distribution.
Sample Simulation A sample simulation result of a specific galaxy count slope and a size distribution. Again we simulate the galaxies below the detection limit of the HST, that are sources have AB magnitudes fainter than ~28.
Validation The galaxy count slope and the sizes are recovered Before going to results we have validated the new method successfully. Here we have made two sets of simulations and fit each element of the one set to the other set. And here you see the resulting bestfit slope and size plots, where it had managed to recover the parameters most of the time.
Preliminary Results New Result on UDF-main: New Results on UDF 05-01: Previously result on UDF-main: F105W: α = 0.7, r = 0.24’’ F125W: α = 0.65, r = 0.24’’ F160W: α = 0.65, r = 0.24’’ New Result on UDF-main: F160W: α = 0.6, r = 0.12’’ New Results on UDF 05-01: F105W: α = 0.6, r = 0.12’’ F125W: α = 0.6, r = 0.30’’ Constrained by the time, today we have only a preliminary result. Previous work done before my arrival here found out that a fairly steep slope of roughly 0.65 and a size corresponding to 0.24” was the best fit to all 3 bands of the UDF-main field. Here we find that….
Future Directions More sophisticated models Galaxy redshift and SED distributions Galaxy Clustering Limiting magnitude variation Change arrays Better calibration of detector issues: hot pixels, persistence and bad cosmic ray rejections
References Abraham et al. 1998 (http://adsabs.harvard.edu/abs/1998AGM....14..E08A) Bock et al. 2006 (http://adsabs.harvard.edu/abs/2006NewAR..50..215B) Illingworth et al. 2010 (http://archive.stsci.edu/proposal_search.php?mission=hst&id=11563) Kashlinksy et al. 2005 (http://adsabs.harvard.edu/abs/2005PhR...409..361K) Thompson et al. 2007 (http://adsabs.harvard.edu/abs/2007ApJ...657..669T)