After a 12-hour flight, I finally arrived in another continent of the world. I could smell the excitation floating in the air and I could feel the blood pulsing through my veins-- my summer in Madison began!
Emission Line Analysis of an Interacting Galaxy Pair J0754+1648
When two gas rich galaxies approach each other, as long as they are close enough, the two galaxies will begin interacting. During the process, according to the Hopkins models, a large amount of gas will be involved and flows into the center part, which may later result in starbursts and feed the central black hole forming an Active Galactic Nucleus (AGN).
The E+A post-starburst phase is the transitional phase between starburst and early-type elliptical galaxies. The star formation there has already stopped but it is still not a red elliptical galaxy yet. People mainly recognize this kind of galaxy using spectra. The continuum of a post-starburst galaxy is a superposition of the spectrum of an early-type elliptical and the spectrum of a younger stellar population. The post-starburst signature usually shows strong Balmer absorption lines but little or no emission because there is no star formation.
Our object this summer is the galaxy J0754+1648 at a redshift of 0.046 displayed in Fig.1 from Sloan Digital Sky Service (SDSS). The two galaxies have not completely merged but they are interacting with each other, initiating the AGN in one region. It shows a post-starburst signature identified by SDSS, starburst region and a strong radio source.
Fig.1 Galaxy J0754+1648 from SDSS9
Since the post-starburst region is really close to the AGN, We want to understand the interaction between the AGN and the galaxy to see whether it is the AGN outflow that has blown out the gas and affected the star formation or the star formation itself that used up the gas. Therefore, in this research, we would like to check how much gas is ionized by the AGN and, if there is, how fast the AGN outflow is with a greater spatial resolution than we have ever had before.
The data we used was collected by the WIYN observatory using the 3.5 meter telescope in Nov and Dec 2014.
We used a brand-new integral field unit (IFU) called HexPak. IFU is a device used in an integral field spectrograph to divide the 2D spatial plane into a continuous array. This division can be achieved in many ways. Here HexPak uses fibers. So the input image is formed at the entrance to a 2D bundle of optical fibers. And these fibers will transfer the light to the slit of the spectrograph. Due to the distribution of fibers we can have the spectra from difference places of our object.
Our HexPak IFU has 111 fibers. It is specially designed to have small fibers at the center and large fiber surrounding the center. We use this special design because when we focus it on certain galaxy, the center part of the galaxy is usually brighter than the outer part. This way, we can collect more signals in the center part without saturating the fibers. Even the large fibers of HexPak have smaller diameter than the old SparsePak IFU. This will provide us with greater spatial resolution and we are able to look into more details of the galaxies.
Fig.2 HexPak and GradPak
Because HexPak is a newly-built IFU, we do not have much experience reducing the data. This time we have devoted much effort to improving the quality of the spectra we get.
Here is the regular part of the data reduction:
- Overscan correction: to remove the structure along columns that occurs during readout.
- Bias Correction: to correct the bias noise of the detector.
- Dark Correction: to correct the noise caused by dark currents of the detector.
Here are the steps which we have paid extra attention to:
- Fiber tracing
- Throughput Correction
- Sky subtraction
- Flux Calibration
- Noise Map
Due to the instrument intrinsic property, a spectrum is not always along a certain column in CCD. The spectra may be curved to some extent so we have to trace them to find their exact places. For this step we prefer dome flats to twilight flats for that twilight flats are influenced by sky lines. As one can see in Fig.3, this can partly affect the tracing results if there is a high accuracy requirement. The problem here is that there is not enough signal in the blue part and the red part is troubled by the defocus caused by the instrument itself. Although we can do the tracing manually, it is rather time-consuming and we expect something to correct it automatically. So our solution is that we just duplicate the tracing information of the less blue or less red part which is still good enough for tracing. Because the tracing should not change much in such a short wavelength range, it can still provide the correct tracing results while saving plenty of time. Basically, we just copy the spectra of the good parts and overwrite the out-of-signal and out-of-focus regions to produce the proper flat images for tracing. For example, in Fig.4, the left one is the tracing result before the correction and it has trouble with the red part. And in the right one, we can see that this solution works pretty well.
Fig.3 Examples of tracing. Left: Tracing result of a dome flat images. Right: Tracing result of a twilight flat.
Fig.4 Left: Tracing result of a dome flat image before correction. Right: Tracing result of a dome flat image after correction.
Different fibers have different throughput but we can only observe standard stars in certain fibers due to the time limitation of observations, so if we want to get the accurate flux calibration of the fibers without standard stars, we will have to do the throughput correction. For this step, we use twilight flats rather than dome flats because it is more important to collect enough signals in blue in this step than to get rid of the influence of sky lines. Also, twilight sky is more even than any dome lamp on the screen, which is more ideal for throughput correction. The problem here is saturation. At the very beginning of the observation, the twilight sky can be too bright that it may cause some pixels to saturate, which can lead to the inaccuracy of the measurement of the throughput. Fig.5 describes the distribution of the value of the pixels. The left one has an abnormal hump in the high value part, which indicates that some pixels must have been saturated. Our current solution is to combine part of the twilight flats, which have no saturated pixels. However it is still not the ideal one because if we do so, we are still unable to collect enough signals in the small fibers and in the blue part of the spectra. Because when we are doing the flux calibration, we want the reference fiber, which has the standard star, and the object fibers to have the same input signal, that is to say if one pixel is rejected in the object fiber, the one in the same place should be rejected in the reference fiber. So we have discussed about a better solution, which is that we combine all the twilight flats using a mask to reject just the saturated pixels rather than the whole images. The algorithm has already been testified but we are still trying to figure out how to realize that.
Fig.5 The distribution of pixel value. Left: The red circle marks out the part that indicates the saturation. Right: The normal distribution.
Actually there is still problem with sky subtraction but here we just use the “skysub” task in IRAF to deal with it for the current science analysis.
This step is to translate the pixel value to the amount of flux from the object. As we have mentioned before, during the observation, we focus certain fiber on the standard star and once we get the spectra of that fiber, knowing the spectral energy distribution (SED) of the standard star, we can calculate the spectral energy distributions of other fibers after applying the fiber-to-fiber throughput correction.
In later analysis, the algorithm we use for spectra fitting requires noise for chi square minimization method so here we discuss how to produce the proper noise map. We mainly calculate the noise map using the root mean square (RMS) method. In data reduction, we have combined some object images to obtain a higher signal to noise for the spectra. Here, we use the spectra we get from those individual images and the spectra we got from the combined image and calculate the RMS for every pixel. It is simple but it can include both the systematic and random noise.
After all these steps, we can get the spectra and noise maps we want for science analysis.
Fig.6 Spectra. Here are some of the spectra we have got. The horizontal axis shows the wavelength and the vertical axis shows the flux (unit: erg·cm2s-1Å-1)
Emission Line Analysis of an Interacting Galaxy Pair J0754+1648
Once we get the spectra for all fibers, we can do the spectra fitting with simple stellar population (SSP) models. Here we use the IDL codes written by Christy Tremonti to fit the spectral energy distributions with the Bruzual and Charlot models using 10 stellar ages ranging from 5Myr to 10Gyr and 6 metallicities ranging from 0.0001 to 0.05. Fig.7 shows some fitting results.
Fig.7 Examples of some fitting results. Left: Some fitting results and the places of their corresponding fibers on the objects. Right: A single spectral fitting result of fiber 108 (unit: 10-16erg·cm2s-1Å-1).
The white lines are the original spectral energy distributions and the red ones are the fitting results. Those blue lines show the spectral energy distributions of individual simple stellar population models. The red lines are the linear combinations of those blue lines. The program tests the chi square of different linear combinations and choose the one with the lowest number as the fitting result.
After we get the fitting results, we can subtract them from the original spectra and the rest of the part is the contribution of the gas, which is definitely what we want for our science analysis.
Emission Line Analysis
Still, we use the same IDL codes to measure intensities of the four emission lines: Hα Hβ NII OIII.
Fig.8 Measuring emission lines.
Using these four emission lines, we can form BPT plots for different days and different setups. BPT plot is named after J.A.Baldwin, M. M. Philips and Roberto Terlevich. Some combinations of strong emission lines can be used to classify different objects into different types by their primary excitation mechanism. Based on their analysis, they found out that these four lines are good indicators of gas ionization level. In the plot, the vertical axis shows the log of NII over Hα while the horizontal axis shows the log of OIII over Hβ. Since we focused the telescope on different places of the object when observing, the data we have are of three setups. Therefore we form three BPT plots separately.
In Fig.9, I divided the images into several regions and mark fibers as well as their corresponding points in BPT plots with different colors. As we can clearly see in the plots, the pink points, which are corresponding to the starburst region, mainly fall between the orange dot line (Kewley 2001) and black line (SDSS Kewley 2013). The orange dot line and the black line together form the region in the plot where most starburst galaxies would fall. It is consistent with our hypothesis.
Fig.9 Three setup images and their corresponding BPT plots. The red circles mark out the starburst region and the blue circles mark out the highly-ionized region.
If we look more carefully, we can easily find that there is a common structure of the highly-ionized region near the AGN. Almost all of the points are outside the starburst region. The diagonal navy dot line indicates the percentage of the ionization due to the AGN. Star forming has much power to ionize the surrounding hydrogen gas while the AGN is even more powerful that it can ionize oxygen gas and nitrogen gas. Hence, the higher the x value and y value in the plot, the more gas is ionized due to the AGN. Some of these points have really high ionization levels. Once we overlap all three setups, we can have a more general view of the highly-ionized region. In Fig.10, the blue and green points form a bar-like pattern whose corresponding points in the plots all fall into the top right region. Using the cosmology calculator, we can measure the size of this highly-ionized region. The length is about 27 kilo parsec. As a comparison, the diameter of the Galaxy is about 30 kilo parsec. Therefore, there is a large region ionized by a powerful source which is very likely due to the AGN.
Fig.10 Overlapping all the three setups. The red circle marks out the highly-ionized region.
Searching for outflows
As I have mentioned before, we should first subtract the stellar contribution from the spectra and then we are able to measure emission and absorption lines due to the gas. Here we focus on the two sodium lines of 5890Å and 5896Å. The two absorption lines due to the gas may not be at the same wavelength as the two due to the stars because of the gas outflow towards us. They may be blue-shifted and the blue-shifted sodium lines are always considered as the indicator of gas outflows. After measuring how much they have blue shifted, we are able to calculate the outflow velocity. For example in Fig.11, we can see a pair of obvious blue-shifted absorption lines.
Fig.11 Blue-shifted Na D lines due to an outflow in a star forming galaxy (Chen+ 2010). The black line is the observed spectra, the red line is the fit to the stellar contribution, the blue line is the fit to the gas component and the green one is the combination of the red line and the blue line.
We have checked all the possible fibers which may have signs for outflows. Unfortunately, we have not been able to detect any obvious blue-shifted sodium lines. It does not mean that there are no outflows. There can be many factors, low signal to noise and inaccurate stellar population fitting for instance.
- Data Reduction:
- Fiber tracing: Use the tracing information of the nearby regions for the out-of-signal and out-of-focus regions.
- Throughput Correction: Use twilight flat images without any saturated pixels. Still try to use all the twilight flat images applying mask to mask out just the saturated pixels instead of the whole images.
- Noise Map: Keep the individual images, deal with them the same way as the combined images and calculate the RMS. Use the RMS, which includes both the systematic and random noise, as the noise map for stellar population fitting.
- Data Analysis:
- AGN has ionized a broad bar-like area of gas: ~27 kilo parsec.
- Not able to measure the blue-shifted Na D lines —> Not sure if there are outflows.
- Still not sure about the truncation mechanism of starburst but the possibility of AGN outflows being responsible for the exhaustion for gas may be higher.
We have spent plenty of time exploring how to do a better data reduction and how to deal with different data problems.
We have looked into the ionized region due to the AGN and tried to find some clues for outflows.
- Data Reduction
- Spectra Fitting
- Searching for AGN outflows
We want to figure out how to realize the idea of using mask to combine all twilight flat images while clipping out only the saturated pixels instead of the whole images to collect enough signal in small fibers and in blue part so that we can have better fiber-to-fiber throughput correction.
Better sky subtraction is another task. Here we just use the IRAF task to deal with it. Actually, there are still problems. First of all, we only have 2 sky fibers among small fibers. We may not be able to have enough signals to noise if we only use these two sky spectra. Sky conditions can be really tricky and two sky fibers are obviously not enough. We will have to come up with some methods to cope with this. Also, there are several sky lines in the spectra and most of time we cannot have absolute alignment between different images. When we subtract the sky spectra, there will be violet fluctuation around the places of sky lines. Because of the intrinsic properties of the spectrometer, the light signal on the CCD is not uniform.
When we are doing the stellar population fitting, we use the codes written by Christy Tremonti. Because we only have spectra on optical wavelengths, we cannot avoid the age-metallicity degeneracy when doing the fitting. We assume the stellar population has solar oxygen abundance and calculate the metallicity, which may not be so accurate. We are thinking of using the intensities of the emission lines to estimate the metallicity to better select the simple stellar population models to fit the spectra. Hence, we can have a more precise measurement of the ages of stellar populations in the galaxy.
Also, in this step, the codes require an estimated velocity dispersion for each spectrum to set a reasonable line width to fit the emission and absorption lines, which we are supposed to figure out by ourselves. Here I manually give it a range of velocity dispersion, run the code, find the minimum chi square in fitting and use this velocity dispersion as the one to fit the data. However, it can be rather time-consuming if we want to test all the individual images with all the possible best velocity dispersion, such as from 100 km/s to 400 km/s. Sometimes, we just cannot find the best velocity dispersion with the lowest chi square because the chi square keeps going down or up as the velocity increases. There are several kinds of small problems like this. They do not matter that much but once we want really precise stellar population fitting, we will have to deal with them
This time we are not able to detect blue-shifted Na D lines for many reasons. Because the Na D absorption lines due to the gas are much weaker than the ones due to the stars, once we subtract the stellar spectra form the original spectra, the remained ones can be too weak to detect. Therefore, it requires really high signal to noise. We will have to do every step of the data reduction with high precision requirement. Better fiber tracing, better fiber-to-fiber throughput correction, better sky subtraction and better stellar population fitting. We still have a long way to go.
My name is Yu Sijie and I am a senior student from Nanjing University Astronomy and Space Science Department. My hometown is Wuhan, a big city in the central part of China, really famous for a great variety of tasty local breakfast and lots of lakes inside the city.
- Emission Line Analysis of an Interacting Galaxy Pair J0754+1648
- Searching for the exoplanets using light curves
- Study on the influence of the collapse of hypermassive neutron stars
In my spare time, I like to listen to music and play basketball. Music is part of my life and basketball builds up my body. I also enjoy travelling because there is always somthing new waiting for me during the journey.
Since I am a senior student, I am going to apply for the graduate school. Consider my interests in dealing with data and programming, I hope I can continue my study in the related field and learn more.
Never in my life have I been able to learn so much and make so many friends during such a short period of time. I had a really great experience in UW-Madison and I cannot make it without the help from those cute people:
I really want to express my deep appreciation to my advisors, Dr. Eric Hooper and Dr. Marsha Wolf, for their patient guidance, meticulous care and a lot of encouragement.
I also want to thank Dr. Bob Benjamin for supervising the whole REU program and students throughout the summer.
Furthermore, I would like to thank Christy Tremonti for providing the code and the guidance. Talking to her is really helpful and enlightening.
Thanks to Andrea Vang for offering me some really helpful advice on both the research and the life in Madison.
Life in a foreign country would be tough if not for all the cute people I met. A special thank you should be given to my roommate Marybeth Beydler because she was so caring throughout the summer.
Finally, I want to express my gratitude to Nanjing University and University of Wisconsin-Madison for providing such a wonderful opportunity for me.