In the summer of 2019, I began working with Dr. Chris Mihos on studying the star formation and ionized gas in the outskirts of Messier 101 (M101). What you will read below is a summary of my work thus far.

Extragalactic HII Detections

It is well known that the Hα emission of a classical H II region directly traces the star formation rate (SFR) in the local universe. Famously, this has been shown time and time again by Kennicutt (1998) and Kennicutt & Evans (2012). However, those analyses have been biased towards high mass, high luminosity star forming galaxies and tend to fit poorly the gas rich, but low SFR dwarf galaxies. We hope to constrain the faint end of the Hα luminosity function by detecting and investigating any extragalactic H II regions in the M101 Group.

Any H II region in the outskirts of galaxies will necessarily be in low density regions and would thus have low SFRs. They could be found in a wide range of environments, including star-forming dwarf galaxies, low surface brightness galaxies, or isolated OB associations. Extragalactic H II regions have been detected in the past, typically through large Hα surveys which target the brightest Balmer line. Although these surveys have uncovered a large number of H II regions, they tend to use only a single narrowband filter which constrains what they can conclusively say about the likelihood that their sources are proper H II regions and not background high redshift objects.

We have narrowband imaging for the M101 Group taken by Dr. Mihos at Case Western’s Burrell-Schmidt telescope in three bands: Hα (λ6562 Å), Hβ (λ4861 Å), and [OIII]λλ4959,5007. These custom filters target the brightest emission lines of classical H II regions at the M101 Group. Additionally, instead of using a broadband filter for continuum subtraction, we image in the off-band for each filter, going redward for the Hα and [OIII] filters and blueward for the Hβ filter. We also obtained broadband B and V images, and we have access to GALEX FUV and NUV images of only M101 as well.

Originally, we had used a Python implementation of SExtractor called “sep,” (Barbary 2016) but we were unconvinced with the results of that process. In January 2020, we switched our detection routine to another Python module PhotUtils’ “segmentation” routine (Bradley et al. 2019). This program “detects” sources by assigning a label to every pixel in an image such that pixels with the same label are part of the same source. The routine involves detecting on an image, in our case the Hα on-band image to maximize our chances of detecting H II regions. After masking bright sources (stars, M101, its companions), the detection routine finds ~32,000 sources which are further deblended into ~35,000 sources. I will refer any reader to the documentation for further information.

The segmentation map that results from the detection routine. Each colored point is a different detected source. The black voids are masked objects. M101 is the circle at the center of the image.

The segmentation module gives basic information about each detected source after we run it on each image, having defined each region in the Hα on-band image.

At this point, we need to remove potential contaminants such as M stars. For those who are unfamiliar, M stars are cool (~3000 K) stars that are predominantly red (B-V ≳ 1.4) and make up about 76% of all stars in the solar neighborhood, regardless of the initial mass function (IMF) chosen. These M stars show up as bright sources in all of our narrowband filters, since we are creating difference images by subtracting each off-band image from each on-band image. Due to the placement of our filters, our filters often lie on top of absorption troughs in the spectra of M stars created by molecules such as titanium oxide and magnesium hydride.

The spectrum of an M0V star from Pickles (1998). The black solid line is the spectrum. The colored lines are our narrowband filter transmission curves. The dashed lines indicate the locations of our targeted emission lines as well as the locations of a magnesium and sodium absorption troughs.

To remove these M stars, we go through several selection criteria. First, we investigated the structural properties of known resolved and unresolved sources detected in our images by cross-matching with the Sloan Digital Sky Survey (SDSS). We plotted the ellipticity versus Gini index of each source as a function of SDSS g magnitude. Stars follow a nice relation: the brightest stars have low ellipticity and high Gini index, indicating circular, concentrated light. Dimmer stars then move towards more elliptical profiles and more evenly distributed light. However, they are not as elliptical or evenly distributed as unresolved sources. We made the cut that stars must have Gini indices greater than 0.4 and ellipticities less than 0.25.

The ellipticity versus Gini index of our cross-matched sources. Ellipticity ranges from 0 (circular) to 1 (highly elliptical) and Gini index ranges from 0 (evenly distributed light) to 1 (concentrated light). The black lines indicate our selection criterion.

Next, we added in photometric selection criteria. We cross-matched with the AllWISE survey, an infrared survey and looked at optical/IR color-color plots. This creates a very visible separation between resolved/unresolved sources. However, AllWISE reaches 50% completeness before we reach our magnitude limit, so we look at the SDSS optical color-color plot as well. Here, although there is less of a separation between sources, of the resolved sources, there is a plume of M stars that is easy to select for.

After this M star removal, we retain ~80% of our original sources detected.

Now, here is the confusing part. Since we are looking a flux differences, where an object is in emission when it has a net flux, Δf > 0, and an object is in absorption when it has a net flux, Δf < 0, we need to calculate a magnitude that reflects this. A simple magnitude difference might work, but we are detecting very faint objects and at such a low flux difference, the standard magnitudes break down and the errors become larger. Further, if an object is brighter in the off-band than the on-band, then it will have a negative flux difference. Standard magnitudes cannot handle negative fluxes by definition. So we utilize the asinh magnitude system introduced by Lupton et al. (1999), a system that utilizes the inverse hyperbolic sine function. This magnitude system is well behaved at all flux levels, behaving like standard magnitudes at high flux levels, but also behaving well through fluxes of zero and into negative fluxes.

A comparison between standard magnitudes (red boxes with error bars) and asinh magnitudes (black line with errors in dotted black).

We reproduce the Figure 2 of Kellar et al. (2012), replacing their “ratio” quantity with a signal-to-noise measurement. In each filter set, we notice that there are generally two populated regions, an “in emission” region and an “in absorption” region. H II regions only have those three filter sets, maybe only two, in emission, so what are these sources in absorption?

The asinh magnitudes of every source in each band. On the vertical is the asinh magnitude of the flux different. On the horizontal is the signal-to-noise. The dashed line indicates a S/N = 3.

First, they could be faint stars that made it through our selection process described above. We synthesized the Stellar Spectral Flux Library (Pickles 1998) spectra through our filters and investigated how they switched between appearing as emission sources or absorption sources. Early to mid-type stars (OBAFG) mimic Hα and [OIII] emission and Hβ absorption. Meanwhile, like we mentioned before, late-type stars (KM) mimic emission in all three bands, and sometimes mimic absorption in the [OIII] images.

Second, they could be nearby galaxies. We synthesized the SDSS DR5 template spectra through our filters. Regardless of whether a galaxy was early or late-type, they would appear in emission in Hα and [OIII] and in absorption in Hβ.

Finally, we synthesized a toy model H II region, one with only the brightest emission lines, and artificially redshifted it through our filters. At low redshift, all three bands would be marked as in emission, which is what they were designed for. But as we move to higher redshifts, each emission line will move out of the on-band filter, into the off-band filter, and then eventually out of our filters entirely. At a redshift of z ~ 0.3, all three of our bands are in emission once again, this time detecting [OIII] in the Hα filter, Hβ in the [OIII] filter, and [OII] in the Hβ filter.

The response in our filters as each emission line characteristic of an H II region is redshifted.

Understanding all of these potential sources of contaminants is important for the two groups that we are about to select.

We divide our sources into two groups: a Gold Group and a Silver Group. The Gold Group has all three bands in emission and has only sources with a Balmer decrement (Hα/Hβ flux ratio) greater than 1. This appears to weed out any high redshift interlopers. The Silver Group has only sources with Hα and [OIII] in emission since Hβ is a relatively weak line.

At this moment (September 2020) we are still analyzing our potential sources and compiling a catalog of their properties. Watch this space for more info as it becomes available!